fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/slab.h>
  30 #include <linux/migrate.h>
  31 #include <linux/ratelimit.h>
  32 #include <linux/uuid.h>
  33 #include <linux/semaphore.h>
  34 #include <asm/unaligned.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "hash.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "volumes.h"
  41 #include "print-tree.h"
  42 #include "async-thread.h"
  43 #include "locking.h"
  44 #include "tree-log.h"
  45 #include "free-space-cache.h"
  46 #include "inode-map.h"
  47 #include "check-integrity.h"
  48 #include "rcu-string.h"
  49 #include "dev-replace.h"
  50 #include "raid56.h"
  51 #include "sysfs.h"
  52
  53 #ifdef CONFIG_X86
  54 #include <asm/cpufeature.h>
  55 #endif
  56
  57 static struct extent_io_ops btree_extent_io_ops;
  58 static void end_workqueue_fn(struct btrfs_work *work);
  59 static void free_fs_root(struct btrfs_root *root);
  60 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  61                                     int read_only);
  62 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
  63                                              struct btrfs_root *root);
  64 static void btrfs_destroy_ordered_extents(struct btrfs_root *root);
  65 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  66                                       struct btrfs_root *root);
  67 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  68 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  69                                         struct extent_io_tree *dirty_pages,
  70                                         int mark);
  71 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  72                                        struct extent_io_tree *pinned_extents);
  73 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  74 static void btrfs_error_commit_super(struct btrfs_root *root);
  75
  76 /*
  77  * end_io_wq structs are used to do processing in task context when an IO is
  78  * complete.  This is used during reads to verify checksums, and it is used
  79  * by writes to insert metadata for new file extents after IO is complete.
  80  */
  81 struct end_io_wq {
  82         struct bio *bio;
  83         bio_end_io_t *end_io;
  84         void *private;
  85         struct btrfs_fs_info *info;
  86         int error;
  87         int metadata;
  88         struct list_head list;
  89         struct btrfs_work work;
  90 };
  91
  92 /*
  93  * async submit bios are used to offload expensive checksumming
  94  * onto the worker threads.  They checksum file and metadata bios
  95  * just before they are sent down the IO stack.
  96  */
  97 struct async_submit_bio {
  98         struct inode *inode;
  99         struct bio *bio;
 100         struct list_head list;
 101         extent_submit_bio_hook_t *submit_bio_start;
 102         extent_submit_bio_hook_t *submit_bio_done;
 103         int rw;
 104         int mirror_num;
 105         unsigned long bio_flags;
 106         /*
 107          * bio_offset is optional, can be used if the pages in the bio
 108          * can't tell us where in the file the bio should go
 109          */
 110         u64 bio_offset;
 111         struct btrfs_work work;
 112         int error;
 113 };
 114
 115 /*
 116  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 117  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 118  * the level the eb occupies in the tree.
 119  *
 120  * Different roots are used for different purposes and may nest inside each
 121  * other and they require separate keysets.  As lockdep keys should be
 122  * static, assign keysets according to the purpose of the root as indicated
 123  * by btrfs_root->objectid.  This ensures that all special purpose roots
 124  * have separate keysets.
 125  *
 126  * Lock-nesting across peer nodes is always done with the immediate parent
 127  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 128  * subclass to avoid triggering lockdep warning in such cases.
 129  *
 130  * The key is set by the readpage_end_io_hook after the buffer has passed
 131  * csum validation but before the pages are unlocked.  It is also set by
 132  * btrfs_init_new_buffer on freshly allocated blocks.
 133  *
 134  * We also add a check to make sure the highest level of the tree is the
 135  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 136  * needs update as well.
 137  */
 138 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 139 # if BTRFS_MAX_LEVEL != 8
 140 #  error
 141 # endif
 142
 143 static struct btrfs_lockdep_keyset {
 144         u64                     id;             /* root objectid */
 145         const char              *name_stem;     /* lock name stem */
 146         char                    names[BTRFS_MAX_LEVEL + 1][20];
 147         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 148 } btrfs_lockdep_keysets[] = {
 149         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 150         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 151         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 152         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 153         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 154         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 155         { .id = BTRFS_QUOTA_TREE_OBJECTID,      .name_stem = "quota"    },
 156         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 157         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 158         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 159         { .id = BTRFS_UUID_TREE_OBJECTID,       .name_stem = "uuid"     },
 160         { .id = 0,                              .name_stem = "tree"     },
 161 };
 162
 163 void __init btrfs_init_lockdep(void)
 164 {
 165         int i, j;
 166
 167         /* initialize lockdep class names */
 168         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 169                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 170
 171                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 172                         snprintf(ks->names[j], sizeof(ks->names[j]),
 173                                  "btrfs-%s-%02d", ks->name_stem, j);
 174         }
 175 }
 176
 177 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 178                                     int level)
 179 {
 180         struct btrfs_lockdep_keyset *ks;
 181
 182         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 183
 184         /* find the matching keyset, id 0 is the default entry */
 185         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 186                 if (ks->id == objectid)
 187                         break;
 188
 189         lockdep_set_class_and_name(&eb->lock,
 190                                    &ks->keys[level], ks->names[level]);
 191 }
 192
 193 #endif
 194
 195 /*
 196  * extents on the btree inode are pretty simple, there's one extent
 197  * that covers the entire device
 198  */
 199 static struct extent_map *btree_get_extent(struct inode *inode,
 200                 struct page *page, size_t pg_offset, u64 start, u64 len,
 201                 int create)
 202 {
 203         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 204         struct extent_map *em;
 205         int ret;
 206
 207         read_lock(&em_tree->lock);
 208         em = lookup_extent_mapping(em_tree, start, len);
 209         if (em) {
 210                 em->bdev =
 211                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 212                 read_unlock(&em_tree->lock);
 213                 goto out;
 214         }
 215         read_unlock(&em_tree->lock);
 216
 217         em = alloc_extent_map();
 218         if (!em) {
 219                 em = ERR_PTR(-ENOMEM);
 220                 goto out;
 221         }
 222         em->start = 0;
 223         em->len = (u64)-1;
 224         em->block_len = (u64)-1;
 225         em->block_start = 0;
 226         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 227
 228         write_lock(&em_tree->lock);
 229         ret = add_extent_mapping(em_tree, em, 0);
 230         if (ret == -EEXIST) {
 231                 free_extent_map(em);
 232                 em = lookup_extent_mapping(em_tree, start, len);
 233                 if (!em)
 234                         em = ERR_PTR(-EIO);
 235         } else if (ret) {
 236                 free_extent_map(em);
 237                 em = ERR_PTR(ret);
 238         }
 239         write_unlock(&em_tree->lock);
 240
 241 out:
 242         return em;
 243 }
 244
 245 u32 btrfs_csum_data(char *data, u32 seed, size_t len)
 246 {
 247         return btrfs_crc32c(seed, data, len);
 248 }
 249
 250 void btrfs_csum_final(u32 crc, char *result)
 251 {
 252         put_unaligned_le32(~crc, result);
 253 }
 254
 255 /*
 256  * compute the csum for a btree block, and either verify it or write it
 257  * into the csum field of the block.
 258  */
 259 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 260                            int verify)
 261 {
 262         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 263         char *result = NULL;
 264         unsigned long len;
 265         unsigned long cur_len;
 266         unsigned long offset = BTRFS_CSUM_SIZE;
 267         char *kaddr;
 268         unsigned long map_start;
 269         unsigned long map_len;
 270         int err;
 271         u32 crc = ~(u32)0;
 272         unsigned long inline_result;
 273
 274         len = buf->len - offset;
 275         while (len > 0) {
 276                 err = map_private_extent_buffer(buf, offset, 32,
 277                                         &kaddr, &map_start, &map_len);
 278                 if (err)
 279                         return 1;
 280                 cur_len = min(len, map_len - (offset - map_start));
 281                 crc = btrfs_csum_data(kaddr + offset - map_start,
 282                                       crc, cur_len);
 283                 len -= cur_len;
 284                 offset += cur_len;
 285         }
 286         if (csum_size > sizeof(inline_result)) {
 287                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 288                 if (!result)
 289                         return 1;
 290         } else {
 291                 result = (char *)&inline_result;
 292         }
 293
 294         btrfs_csum_final(crc, result);
 295
 296         if (verify) {
 297                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 298                         u32 val;
 299                         u32 found = 0;
 300                         memcpy(&found, result, csum_size);
 301
 302                         read_extent_buffer(buf, &val, 0, csum_size);
 303                         printk_ratelimited(KERN_INFO
 304                                 "BTRFS: %s checksum verify failed on %llu wanted %X found %X "
 305                                 "level %d\n",
 306                                 root->fs_info->sb->s_id, buf->start,
 307                                 val, found, btrfs_header_level(buf));
 308                         if (result != (char *)&inline_result)
 309                                 kfree(result);
 310                         return 1;
 311                 }
 312         } else {
 313                 write_extent_buffer(buf, result, 0, csum_size);
 314         }
 315         if (result != (char *)&inline_result)
 316                 kfree(result);
 317         return 0;
 318 }
 319
 320 /*
 321  * we can't consider a given block up to date unless the transid of the
 322  * block matches the transid in the parent node's pointer.  This is how we
 323  * detect blocks that either didn't get written at all or got written
 324  * in the wrong place.
 325  */
 326 static int verify_parent_transid(struct extent_io_tree *io_tree,
 327                                  struct extent_buffer *eb, u64 parent_transid,
 328                                  int atomic)
 329 {
 330         struct extent_state *cached_state = NULL;
 331         int ret;
 332
 333         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 334                 return 0;
 335
 336         if (atomic)
 337                 return -EAGAIN;
 338
 339         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 340                          0, &cached_state);
 341         if (extent_buffer_uptodate(eb) &&
 342             btrfs_header_generation(eb) == parent_transid) {
 343                 ret = 0;
 344                 goto out;
 345         }
 346         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
 347                        "found %llu\n",
 348                        eb->start, parent_transid, btrfs_header_generation(eb));
 349         ret = 1;
 350         clear_extent_buffer_uptodate(eb);
 351 out:
 352         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 353                              &cached_state, GFP_NOFS);
 354         return ret;
 355 }
 356
 357 /*
 358  * Return 0 if the superblock checksum type matches the checksum value of that
 359  * algorithm. Pass the raw disk superblock data.
 360  */
 361 static int btrfs_check_super_csum(char *raw_disk_sb)
 362 {
 363         struct btrfs_super_block *disk_sb =
 364                 (struct btrfs_super_block *)raw_disk_sb;
 365         u16 csum_type = btrfs_super_csum_type(disk_sb);
 366         int ret = 0;
 367
 368         if (csum_type == BTRFS_CSUM_TYPE_CRC32) {
 369                 u32 crc = ~(u32)0;
 370                 const int csum_size = sizeof(crc);
 371                 char result[csum_size];
 372
 373                 /*
 374                  * The super_block structure does not span the whole
 375                  * BTRFS_SUPER_INFO_SIZE range, we expect that the unused space
 376                  * is filled with zeros and is included in the checkum.
 377                  */
 378                 crc = btrfs_csum_data(raw_disk_sb + BTRFS_CSUM_SIZE,
 379                                 crc, BTRFS_SUPER_INFO_SIZE - BTRFS_CSUM_SIZE);
 380                 btrfs_csum_final(crc, result);
 381
 382                 if (memcmp(raw_disk_sb, result, csum_size))
 383                         ret = 1;
 384
 385                 if (ret && btrfs_super_generation(disk_sb) < 10) {
 386                         printk(KERN_WARNING
 387                                 "BTRFS: super block crcs don't match, older mkfs detected\n");
 388                         ret = 0;
 389                 }
 390         }
 391
 392         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 393                 printk(KERN_ERR "BTRFS: unsupported checksum algorithm %u\n",
 394                                 csum_type);
 395                 ret = 1;
 396         }
 397
 398         return ret;
 399 }
 400
 401 /*
 402  * helper to read a given tree block, doing retries as required when
 403  * the checksums don't match and we have alternate mirrors to try.
 404  */
 405 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 406                                           struct extent_buffer *eb,
 407                                           u64 start, u64 parent_transid)
 408 {
 409         struct extent_io_tree *io_tree;
 410         int failed = 0;
 411         int ret;
 412         int num_copies = 0;
 413         int mirror_num = 0;
 414         int failed_mirror = 0;
 415
 416         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 417         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 418         while (1) {
 419                 ret = read_extent_buffer_pages(io_tree, eb, start,
 420                                                WAIT_COMPLETE,
 421                                                btree_get_extent, mirror_num);
 422                 if (!ret) {
 423                         if (!verify_parent_transid(io_tree, eb,
 424                                                    parent_transid, 0))
 425                                 break;
 426                         else
 427                                 ret = -EIO;
 428                 }
 429
 430                 /*
 431                  * This buffer's crc is fine, but its contents are corrupted, so
 432                  * there is no reason to read the other copies, they won't be
 433                  * any less wrong.
 434                  */
 435                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 436                         break;
 437
 438                 num_copies = btrfs_num_copies(root->fs_info,
 439                                               eb->start, eb->len);
 440                 if (num_copies == 1)
 441                         break;
 442
 443                 if (!failed_mirror) {
 444                         failed = 1;
 445                         failed_mirror = eb->read_mirror;
 446                 }
 447
 448                 mirror_num++;
 449                 if (mirror_num == failed_mirror)
 450                         mirror_num++;
 451
 452                 if (mirror_num > num_copies)
 453                         break;
 454         }
 455
 456         if (failed && !ret && failed_mirror)
 457                 repair_eb_io_failure(root, eb, failed_mirror);
 458
 459         return ret;
 460 }
 461
 462 /*
 463  * checksum a dirty tree block before IO.  This has extra checks to make sure
 464  * we only fill in the checksum field in the first page of a multi-page block
 465  */
 466
 467 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 468 {
 469         u64 start = page_offset(page);
 470         u64 found_start;
 471         struct extent_buffer *eb;
 472
 473         eb = (struct extent_buffer *)page->private;
 474         if (page != eb->pages[0])
 475                 return 0;
 476         found_start = btrfs_header_bytenr(eb);
 477         if (WARN_ON(found_start != start || !PageUptodate(page)))
 478                 return 0;
 479         csum_tree_block(root, eb, 0);
 480         return 0;
 481 }
 482
 483 static int check_tree_block_fsid(struct btrfs_root *root,
 484                                  struct extent_buffer *eb)
 485 {
 486         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 487         u8 fsid[BTRFS_UUID_SIZE];
 488         int ret = 1;
 489
 490         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 491         while (fs_devices) {
 492                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 493                         ret = 0;
 494                         break;
 495                 }
 496                 fs_devices = fs_devices->seed;
 497         }
 498         return ret;
 499 }
 500
 501 #define CORRUPT(reason, eb, root, slot)                         \
 502         btrfs_crit(root->fs_info, "corrupt leaf, %s: block=%llu,"       \
 503                    "root=%llu, slot=%d", reason,                        \
 504                btrfs_header_bytenr(eb), root->objectid, slot)
 505
 506 static noinline int check_leaf(struct btrfs_root *root,
 507                                struct extent_buffer *leaf)
 508 {
 509         struct btrfs_key key;
 510         struct btrfs_key leaf_key;
 511         u32 nritems = btrfs_header_nritems(leaf);
 512         int slot;
 513
 514         if (nritems == 0)
 515                 return 0;
 516
 517         /* Check the 0 item */
 518         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 519             BTRFS_LEAF_DATA_SIZE(root)) {
 520                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 521                 return -EIO;
 522         }
 523
 524         /*
 525          * Check to make sure each items keys are in the correct order and their
 526          * offsets make sense.  We only have to loop through nritems-1 because
 527          * we check the current slot against the next slot, which verifies the
 528          * next slot's offset+size makes sense and that the current's slot
 529          * offset is correct.
 530          */
 531         for (slot = 0; slot < nritems - 1; slot++) {
 532                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 533                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 534
 535                 /* Make sure the keys are in the right order */
 536                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 537                         CORRUPT("bad key order", leaf, root, slot);
 538                         return -EIO;
 539                 }
 540
 541                 /*
 542                  * Make sure the offset and ends are right, remember that the
 543                  * item data starts at the end of the leaf and grows towards the
 544                  * front.
 545                  */
 546                 if (btrfs_item_offset_nr(leaf, slot) !=
 547                         btrfs_item_end_nr(leaf, slot + 1)) {
 548                         CORRUPT("slot offset bad", leaf, root, slot);
 549                         return -EIO;
 550                 }
 551
 552                 /*
 553                  * Check to make sure that we don't point outside of the leaf,
 554                  * just incase all the items are consistent to eachother, but
 555                  * all point outside of the leaf.
 556                  */
 557                 if (btrfs_item_end_nr(leaf, slot) >
 558                     BTRFS_LEAF_DATA_SIZE(root)) {
 559                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 560                         return -EIO;
 561                 }
 562         }
 563
 564         return 0;
 565 }
 566
 567 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 568                                       u64 phy_offset, struct page *page,
 569                                       u64 start, u64 end, int mirror)
 570 {
 571         u64 found_start;
 572         int found_level;
 573         struct extent_buffer *eb;
 574         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 575         int ret = 0;
 576         int reads_done;
 577
 578         if (!page->private)
 579                 goto out;
 580
 581         eb = (struct extent_buffer *)page->private;
 582
 583         /* the pending IO might have been the only thing that kept this buffer
 584          * in memory.  Make sure we have a ref for all this other checks
 585          */
 586         extent_buffer_get(eb);
 587
 588         reads_done = atomic_dec_and_test(&eb->io_pages);
 589         if (!reads_done)
 590                 goto err;
 591
 592         eb->read_mirror = mirror;
 593         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
 594                 ret = -EIO;
 595                 goto err;
 596         }
 597
 598         found_start = btrfs_header_bytenr(eb);
 599         if (found_start != eb->start) {
 600                 printk_ratelimited(KERN_INFO "BTRFS: bad tree block start "
 601                                "%llu %llu\n",
 602                                found_start, eb->start);
 603                 ret = -EIO;
 604                 goto err;
 605         }
 606         if (check_tree_block_fsid(root, eb)) {
 607                 printk_ratelimited(KERN_INFO "BTRFS: bad fsid on block %llu\n",
 608                                eb->start);
 609                 ret = -EIO;
 610                 goto err;
 611         }
 612         found_level = btrfs_header_level(eb);
 613         if (found_level >= BTRFS_MAX_LEVEL) {
 614                 btrfs_info(root->fs_info, "bad tree block level %d",
 615                            (int)btrfs_header_level(eb));
 616                 ret = -EIO;
 617                 goto err;
 618         }
 619
 620         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 621                                        eb, found_level);
 622
 623         ret = csum_tree_block(root, eb, 1);
 624         if (ret) {
 625                 ret = -EIO;
 626                 goto err;
 627         }
 628
 629         /*
 630          * If this is a leaf block and it is corrupt, set the corrupt bit so
 631          * that we don't try and read the other copies of this block, just
 632          * return -EIO.
 633          */
 634         if (found_level == 0 && check_leaf(root, eb)) {
 635                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 636                 ret = -EIO;
 637         }
 638
 639         if (!ret)
 640                 set_extent_buffer_uptodate(eb);
 641 err:
 642         if (reads_done &&
 643             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 644                 btree_readahead_hook(root, eb, eb->start, ret);
 645
 646         if (ret) {
 647                 /*
 648                  * our io error hook is going to dec the io pages
 649                  * again, we have to make sure it has something
 650                  * to decrement
 651                  */
 652                 atomic_inc(&eb->io_pages);
 653                 clear_extent_buffer_uptodate(eb);
 654         }
 655         free_extent_buffer(eb);
 656 out:
 657         return ret;
 658 }
 659
 660 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 661 {
 662         struct extent_buffer *eb;
 663         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 664
 665         eb = (struct extent_buffer *)page->private;
 666         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
 667         eb->read_mirror = failed_mirror;
 668         atomic_dec(&eb->io_pages);
 669         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 670                 btree_readahead_hook(root, eb, eb->start, -EIO);
 671         return -EIO;    /* we fixed nothing */
 672 }
 673
 674 static void end_workqueue_bio(struct bio *bio, int err)
 675 {
 676         struct end_io_wq *end_io_wq = bio->bi_private;
 677         struct btrfs_fs_info *fs_info;
 678
 679         fs_info = end_io_wq->info;
 680         end_io_wq->error = err;
 681         btrfs_init_work(&end_io_wq->work, end_workqueue_fn, NULL, NULL);
 682
 683         if (bio->bi_rw & REQ_WRITE) {
 684                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
 685                         btrfs_queue_work(fs_info->endio_meta_write_workers,
 686                                          &end_io_wq->work);
 687                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
 688                         btrfs_queue_work(fs_info->endio_freespace_worker,
 689                                          &end_io_wq->work);
 690                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 691                         btrfs_queue_work(fs_info->endio_raid56_workers,
 692                                          &end_io_wq->work);
 693                 else
 694                         btrfs_queue_work(fs_info->endio_write_workers,
 695                                          &end_io_wq->work);
 696         } else {
 697                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 698                         btrfs_queue_work(fs_info->endio_raid56_workers,
 699                                          &end_io_wq->work);
 700                 else if (end_io_wq->metadata)
 701                         btrfs_queue_work(fs_info->endio_meta_workers,
 702                                          &end_io_wq->work);
 703                 else
 704                         btrfs_queue_work(fs_info->endio_workers,
 705                                          &end_io_wq->work);
 706         }
 707 }
 708
 709 /*
 710  * For the metadata arg you want
 711  *
 712  * 0 - if data
 713  * 1 - if normal metadta
 714  * 2 - if writing to the free space cache area
 715  * 3 - raid parity work
 716  */
 717 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 718                         int metadata)
 719 {
 720         struct end_io_wq *end_io_wq;
 721         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 722         if (!end_io_wq)
 723                 return -ENOMEM;
 724
 725         end_io_wq->private = bio->bi_private;
 726         end_io_wq->end_io = bio->bi_end_io;
 727         end_io_wq->info = info;
 728         end_io_wq->error = 0;
 729         end_io_wq->bio = bio;
 730         end_io_wq->metadata = metadata;
 731
 732         bio->bi_private = end_io_wq;
 733         bio->bi_end_io = end_workqueue_bio;
 734         return 0;
 735 }
 736
 737 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 738 {
 739         unsigned long limit = min_t(unsigned long,
 740                                     info->thread_pool_size,
 741                                     info->fs_devices->open_devices);
 742         return 256 * limit;
 743 }
 744
 745 static void run_one_async_start(struct btrfs_work *work)
 746 {
 747         struct async_submit_bio *async;
 748         int ret;
 749
 750         async = container_of(work, struct  async_submit_bio, work);
 751         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 752                                       async->mirror_num, async->bio_flags,
 753                                       async->bio_offset);
 754         if (ret)
 755                 async->error = ret;
 756 }
 757
 758 static void run_one_async_done(struct btrfs_work *work)
 759 {
 760         struct btrfs_fs_info *fs_info;
 761         struct async_submit_bio *async;
 762         int limit;
 763
 764         async = container_of(work, struct  async_submit_bio, work);
 765         fs_info = BTRFS_I(async->inode)->root->fs_info;
 766
 767         limit = btrfs_async_submit_limit(fs_info);
 768         limit = limit * 2 / 3;
 769
 770         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 771             waitqueue_active(&fs_info->async_submit_wait))
 772                 wake_up(&fs_info->async_submit_wait);
 773
 774         /* If an error occured we just want to clean up the bio and move on */
 775         if (async->error) {
 776                 bio_endio(async->bio, async->error);
 777                 return;
 778         }
 779
 780         async->submit_bio_done(async->inode, async->rw, async->bio,
 781                                async->mirror_num, async->bio_flags,
 782                                async->bio_offset);
 783 }
 784
 785 static void run_one_async_free(struct btrfs_work *work)
 786 {
 787         struct async_submit_bio *async;
 788
 789         async = container_of(work, struct  async_submit_bio, work);
 790         kfree(async);
 791 }
 792
 793 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 794                         int rw, struct bio *bio, int mirror_num,
 795                         unsigned long bio_flags,
 796                         u64 bio_offset,
 797                         extent_submit_bio_hook_t *submit_bio_start,
 798                         extent_submit_bio_hook_t *submit_bio_done)
 799 {
 800         struct async_submit_bio *async;
 801
 802         async = kmalloc(sizeof(*async), GFP_NOFS);
 803         if (!async)
 804                 return -ENOMEM;
 805
 806         async->inode = inode;
 807         async->rw = rw;
 808         async->bio = bio;
 809         async->mirror_num = mirror_num;
 810         async->submit_bio_start = submit_bio_start;
 811         async->submit_bio_done = submit_bio_done;
 812
 813         btrfs_init_work(&async->work, run_one_async_start,
 814                         run_one_async_done, run_one_async_free);
 815
 816         async->bio_flags = bio_flags;
 817         async->bio_offset = bio_offset;
 818
 819         async->error = 0;
 820
 821         atomic_inc(&fs_info->nr_async_submits);
 822
 823         if (rw & REQ_SYNC)
 824                 btrfs_set_work_high_priority(&async->work);
 825
 826         btrfs_queue_work(fs_info->workers, &async->work);
 827
 828         while (atomic_read(&fs_info->async_submit_draining) &&
 829               atomic_read(&fs_info->nr_async_submits)) {
 830                 wait_event(fs_info->async_submit_wait,
 831                            (atomic_read(&fs_info->nr_async_submits) == 0));
 832         }
 833
 834         return 0;
 835 }
 836
 837 static int btree_csum_one_bio(struct bio *bio)
 838 {
 839         struct bio_vec *bvec;
 840         struct btrfs_root *root;
 841         int i, ret = 0;
 842
 843         bio_for_each_segment_all(bvec, bio, i) {
 844                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 845                 ret = csum_dirty_buffer(root, bvec->bv_page);
 846                 if (ret)
 847                         break;
 848         }
 849
 850         return ret;
 851 }
 852
 853 static int __btree_submit_bio_start(struct inode *inode, int rw,
 854                                     struct bio *bio, int mirror_num,
 855                                     unsigned long bio_flags,
 856                                     u64 bio_offset)
 857 {
 858         /*
 859          * when we're called for a write, we're already in the async
 860          * submission context.  Just jump into btrfs_map_bio
 861          */
 862         return btree_csum_one_bio(bio);
 863 }
 864
 865 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 866                                  int mirror_num, unsigned long bio_flags,
 867                                  u64 bio_offset)
 868 {
 869         int ret;
 870
 871         /*
 872          * when we're called for a write, we're already in the async
 873          * submission context.  Just jump into btrfs_map_bio
 874          */
 875         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 876         if (ret)
 877                 bio_endio(bio, ret);
 878         return ret;
 879 }
 880
 881 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 882 {
 883         if (bio_flags & EXTENT_BIO_TREE_LOG)
 884                 return 0;
 885 #ifdef CONFIG_X86
 886         if (cpu_has_xmm4_2)
 887                 return 0;
 888 #endif
 889         return 1;
 890 }
 891
 892 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 893                                  int mirror_num, unsigned long bio_flags,
 894                                  u64 bio_offset)
 895 {
 896         int async = check_async_write(inode, bio_flags);
 897         int ret;
 898
 899         if (!(rw & REQ_WRITE)) {
 900                 /*
 901                  * called for a read, do the setup so that checksum validation
 902                  * can happen in the async kernel threads
 903                  */
 904                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 905                                           bio, 1);
 906                 if (ret)
 907                         goto out_w_error;
 908                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 909                                     mirror_num, 0);
 910         } else if (!async) {
 911                 ret = btree_csum_one_bio(bio);
 912                 if (ret)
 913                         goto out_w_error;
 914                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 915                                     mirror_num, 0);
 916         } else {
 917                 /*
 918                  * kthread helpers are used to submit writes so that
 919                  * checksumming can happen in parallel across all CPUs
 920                  */
 921                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 922                                           inode, rw, bio, mirror_num, 0,
 923                                           bio_offset,
 924                                           __btree_submit_bio_start,
 925                                           __btree_submit_bio_done);
 926         }
 927
 928         if (ret) {
 929 out_w_error:
 930                 bio_endio(bio, ret);
 931         }
 932         return ret;
 933 }
 934
 935 #ifdef CONFIG_MIGRATION
 936 static int btree_migratepage(struct address_space *mapping,
 937                         struct page *newpage, struct page *page,
 938                         enum migrate_mode mode)
 939 {
 940         /*
 941          * we can't safely write a btree page from here,
 942          * we haven't done the locking hook
 943          */
 944         if (PageDirty(page))
 945                 return -EAGAIN;
 946         /*
 947          * Buffers may be managed in a filesystem specific way.
 948          * We must have no buffers or drop them.
 949          */
 950         if (page_has_private(page) &&
 951             !try_to_release_page(page, GFP_KERNEL))
 952                 return -EAGAIN;
 953         return migrate_page(mapping, newpage, page, mode);
 954 }
 955 #endif
 956
 957
 958 static int btree_writepages(struct address_space *mapping,
 959                             struct writeback_control *wbc)
 960 {
 961         struct btrfs_fs_info *fs_info;
 962         int ret;
 963
 964         if (wbc->sync_mode == WB_SYNC_NONE) {
 965
 966                 if (wbc->for_kupdate)
 967                         return 0;
 968
 969                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 970                 /* this is a bit racy, but that's ok */
 971                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 972                                              BTRFS_DIRTY_METADATA_THRESH);
 973                 if (ret < 0)
 974                         return 0;
 975         }
 976         return btree_write_cache_pages(mapping, wbc);
 977 }
 978
 979 static int btree_readpage(struct file *file, struct page *page)
 980 {
 981         struct extent_io_tree *tree;
 982         tree = &BTRFS_I(page->mapping->host)->io_tree;
 983         return extent_read_full_page(tree, page, btree_get_extent, 0);
 984 }
 985
 986 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 987 {
 988         if (PageWriteback(page) || PageDirty(page))
 989                 return 0;
 990
 991         return try_release_extent_buffer(page);
 992 }
 993
 994 static void btree_invalidatepage(struct page *page, unsigned int offset,
 995                                  unsigned int length)
 996 {
 997         struct extent_io_tree *tree;
 998         tree = &BTRFS_I(page->mapping->host)->io_tree;
 999         extent_invalidatepage(tree, page, offset);
1000         btree_releasepage(page, GFP_NOFS);
1001         if (PagePrivate(page)) {
1002                 btrfs_warn(BTRFS_I(page->mapping->host)->root->fs_info,
1003                            "page private not zero on page %llu",
1004                            (unsigned long long)page_offset(page));
1005                 ClearPagePrivate(page);
1006                 set_page_private(page, 0);
1007                 page_cache_release(page);
1008         }
1009 }
1010
1011 static int btree_set_page_dirty(struct page *page)
1012 {
1013 #ifdef DEBUG
1014         struct extent_buffer *eb;
1015
1016         BUG_ON(!PagePrivate(page));
1017         eb = (struct extent_buffer *)page->private;
1018         BUG_ON(!eb);
1019         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1020         BUG_ON(!atomic_read(&eb->refs));
1021         btrfs_assert_tree_locked(eb);
1022 #endif
1023         return __set_page_dirty_nobuffers(page);
1024 }
1025
1026 static const struct address_space_operations btree_aops = {
1027         .readpage       = btree_readpage,
1028         .writepages     = btree_writepages,
1029         .releasepage    = btree_releasepage,
1030         .invalidatepage = btree_invalidatepage,
1031 #ifdef CONFIG_MIGRATION
1032         .migratepage    = btree_migratepage,
1033 #endif
1034         .set_page_dirty = btree_set_page_dirty,
1035 };
1036
1037 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1038                          u64 parent_transid)
1039 {
1040         struct extent_buffer *buf = NULL;
1041         struct inode *btree_inode = root->fs_info->btree_inode;
1042         int ret = 0;
1043
1044         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1045         if (!buf)
1046                 return 0;
1047         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1048                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1049         free_extent_buffer(buf);
1050         return ret;
1051 }
1052
1053 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1054                          int mirror_num, struct extent_buffer **eb)
1055 {
1056         struct extent_buffer *buf = NULL;
1057         struct inode *btree_inode = root->fs_info->btree_inode;
1058         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1059         int ret;
1060
1061         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1062         if (!buf)
1063                 return 0;
1064
1065         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1066
1067         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1068                                        btree_get_extent, mirror_num);
1069         if (ret) {
1070                 free_extent_buffer(buf);
1071                 return ret;
1072         }
1073
1074         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1075                 free_extent_buffer(buf);
1076                 return -EIO;
1077         } else if (extent_buffer_uptodate(buf)) {
1078                 *eb = buf;
1079         } else {
1080                 free_extent_buffer(buf);
1081         }
1082         return 0;
1083 }
1084
1085 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1086                                             u64 bytenr, u32 blocksize)
1087 {
1088         return find_extent_buffer(root->fs_info, bytenr);
1089 }
1090
1091 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1092                                                  u64 bytenr, u32 blocksize)
1093 {
1094         return alloc_extent_buffer(root->fs_info, bytenr, blocksize);
1095 }
1096
1097
1098 int btrfs_write_tree_block(struct extent_buffer *buf)
1099 {
1100         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1101                                         buf->start + buf->len - 1);
1102 }
1103
1104 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1105 {
1106         return filemap_fdatawait_range(buf->pages[0]->mapping,
1107                                        buf->start, buf->start + buf->len - 1);
1108 }
1109
1110 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1111                                       u32 blocksize, u64 parent_transid)
1112 {
1113         struct extent_buffer *buf = NULL;
1114         int ret;
1115
1116         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1117         if (!buf)
1118                 return NULL;
1119
1120         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1121         if (ret) {
1122                 free_extent_buffer(buf);
1123                 return NULL;
1124         }
1125         return buf;
1126
1127 }
1128
1129 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1130                       struct extent_buffer *buf)
1131 {
1132         struct btrfs_fs_info *fs_info = root->fs_info;
1133
1134         if (btrfs_header_generation(buf) ==
1135             fs_info->running_transaction->transid) {
1136                 btrfs_assert_tree_locked(buf);
1137
1138                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1139                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1140                                              -buf->len,
1141                                              fs_info->dirty_metadata_batch);
1142                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1143                         btrfs_set_lock_blocking(buf);
1144                         clear_extent_buffer_dirty(buf);
1145                 }
1146         }
1147 }
1148
1149 static struct btrfs_subvolume_writers *btrfs_alloc_subvolume_writers(void)
1150 {
1151         struct btrfs_subvolume_writers *writers;
1152         int ret;
1153
1154         writers = kmalloc(sizeof(*writers), GFP_NOFS);
1155         if (!writers)
1156                 return ERR_PTR(-ENOMEM);
1157
1158         ret = percpu_counter_init(&writers->counter, 0);
1159         if (ret < 0) {
1160                 kfree(writers);
1161                 return ERR_PTR(ret);
1162         }
1163
1164         init_waitqueue_head(&writers->wait);
1165         return writers;
1166 }
1167
1168 static void
1169 btrfs_free_subvolume_writers(struct btrfs_subvolume_writers *writers)
1170 {
1171         percpu_counter_destroy(&writers->counter);
1172         kfree(writers);
1173 }
1174
1175 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1176                          u32 stripesize, struct btrfs_root *root,
1177                          struct btrfs_fs_info *fs_info,
1178                          u64 objectid)
1179 {
1180         root->node = NULL;
1181         root->commit_root = NULL;
1182         root->sectorsize = sectorsize;
1183         root->nodesize = nodesize;
1184         root->leafsize = leafsize;
1185         root->stripesize = stripesize;
1186         root->ref_cows = 0;
1187         root->track_dirty = 0;
1188         root->in_radix = 0;
1189         root->orphan_item_inserted = 0;
1190         root->orphan_cleanup_state = 0;
1191
1192         root->objectid = objectid;
1193         root->last_trans = 0;
1194         root->highest_objectid = 0;
1195         root->nr_delalloc_inodes = 0;
1196         root->nr_ordered_extents = 0;
1197         root->name = NULL;
1198         root->inode_tree = RB_ROOT;
1199         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1200         root->block_rsv = NULL;
1201         root->orphan_block_rsv = NULL;
1202
1203         INIT_LIST_HEAD(&root->dirty_list);
1204         INIT_LIST_HEAD(&root->root_list);
1205         INIT_LIST_HEAD(&root->delalloc_inodes);
1206         INIT_LIST_HEAD(&root->delalloc_root);
1207         INIT_LIST_HEAD(&root->ordered_extents);
1208         INIT_LIST_HEAD(&root->ordered_root);
1209         INIT_LIST_HEAD(&root->logged_list[0]);
1210         INIT_LIST_HEAD(&root->logged_list[1]);
1211         spin_lock_init(&root->orphan_lock);
1212         spin_lock_init(&root->inode_lock);
1213         spin_lock_init(&root->delalloc_lock);
1214         spin_lock_init(&root->ordered_extent_lock);
1215         spin_lock_init(&root->accounting_lock);
1216         spin_lock_init(&root->log_extents_lock[0]);
1217         spin_lock_init(&root->log_extents_lock[1]);
1218         mutex_init(&root->objectid_mutex);
1219         mutex_init(&root->log_mutex);
1220         mutex_init(&root->ordered_extent_mutex);
1221         mutex_init(&root->delalloc_mutex);
1222         init_waitqueue_head(&root->log_writer_wait);
1223         init_waitqueue_head(&root->log_commit_wait[0]);
1224         init_waitqueue_head(&root->log_commit_wait[1]);
1225         INIT_LIST_HEAD(&root->log_ctxs[0]);
1226         INIT_LIST_HEAD(&root->log_ctxs[1]);
1227         atomic_set(&root->log_commit[0], 0);
1228         atomic_set(&root->log_commit[1], 0);
1229         atomic_set(&root->log_writers, 0);
1230         atomic_set(&root->log_batch, 0);
1231         atomic_set(&root->orphan_inodes, 0);
1232         atomic_set(&root->refs, 1);
1233         atomic_set(&root->will_be_snapshoted, 0);
1234         root->log_transid = 0;
1235         root->log_transid_committed = -1;
1236         root->last_log_commit = 0;
1237         if (fs_info)
1238                 extent_io_tree_init(&root->dirty_log_pages,
1239                                      fs_info->btree_inode->i_mapping);
1240
1241         memset(&root->root_key, 0, sizeof(root->root_key));
1242         memset(&root->root_item, 0, sizeof(root->root_item));
1243         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1244         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1245         if (fs_info)
1246                 root->defrag_trans_start = fs_info->generation;
1247         else
1248                 root->defrag_trans_start = 0;
1249         init_completion(&root->kobj_unregister);
1250         root->defrag_running = 0;
1251         root->root_key.objectid = objectid;
1252         root->anon_dev = 0;
1253
1254         spin_lock_init(&root->root_item_lock);
1255 }
1256
1257 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1258 {
1259         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1260         if (root)
1261                 root->fs_info = fs_info;
1262         return root;
1263 }
1264
1265 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1266 /* Should only be used by the testing infrastructure */
1267 struct btrfs_root *btrfs_alloc_dummy_root(void)
1268 {
1269         struct btrfs_root *root;
1270
1271         root = btrfs_alloc_root(NULL);
1272         if (!root)
1273                 return ERR_PTR(-ENOMEM);
1274         __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1275         root->dummy_root = 1;
1276
1277         return root;
1278 }
1279 #endif
1280
1281 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1282                                      struct btrfs_fs_info *fs_info,
1283                                      u64 objectid)
1284 {
1285         struct extent_buffer *leaf;
1286         struct btrfs_root *tree_root = fs_info->tree_root;
1287         struct btrfs_root *root;
1288         struct btrfs_key key;
1289         int ret = 0;
1290         uuid_le uuid;
1291
1292         root = btrfs_alloc_root(fs_info);
1293         if (!root)
1294                 return ERR_PTR(-ENOMEM);
1295
1296         __setup_root(tree_root->nodesize, tree_root->leafsize,
1297                      tree_root->sectorsize, tree_root->stripesize,
1298                      root, fs_info, objectid);
1299         root->root_key.objectid = objectid;
1300         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1301         root->root_key.offset = 0;
1302
1303         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1304                                       0, objectid, NULL, 0, 0, 0);
1305         if (IS_ERR(leaf)) {
1306                 ret = PTR_ERR(leaf);
1307                 leaf = NULL;
1308                 goto fail;
1309         }
1310
1311         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1312         btrfs_set_header_bytenr(leaf, leaf->start);
1313         btrfs_set_header_generation(leaf, trans->transid);
1314         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1315         btrfs_set_header_owner(leaf, objectid);
1316         root->node = leaf;
1317
1318         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1319                             BTRFS_FSID_SIZE);
1320         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1321                             btrfs_header_chunk_tree_uuid(leaf),
1322                             BTRFS_UUID_SIZE);
1323         btrfs_mark_buffer_dirty(leaf);
1324
1325         root->commit_root = btrfs_root_node(root);
1326         root->track_dirty = 1;
1327
1328
1329         root->root_item.flags = 0;
1330         root->root_item.byte_limit = 0;
1331         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1332         btrfs_set_root_generation(&root->root_item, trans->transid);
1333         btrfs_set_root_level(&root->root_item, 0);
1334         btrfs_set_root_refs(&root->root_item, 1);
1335         btrfs_set_root_used(&root->root_item, leaf->len);
1336         btrfs_set_root_last_snapshot(&root->root_item, 0);
1337         btrfs_set_root_dirid(&root->root_item, 0);
1338         uuid_le_gen(&uuid);
1339         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1340         root->root_item.drop_level = 0;
1341
1342         key.objectid = objectid;
1343         key.type = BTRFS_ROOT_ITEM_KEY;
1344         key.offset = 0;
1345         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1346         if (ret)
1347                 goto fail;
1348
1349         btrfs_tree_unlock(leaf);
1350
1351         return root;
1352
1353 fail:
1354         if (leaf) {
1355                 btrfs_tree_unlock(leaf);
1356                 free_extent_buffer(leaf);
1357         }
1358         kfree(root);
1359
1360         return ERR_PTR(ret);
1361 }
1362
1363 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1364                                          struct btrfs_fs_info *fs_info)
1365 {
1366         struct btrfs_root *root;
1367         struct btrfs_root *tree_root = fs_info->tree_root;
1368         struct extent_buffer *leaf;
1369
1370         root = btrfs_alloc_root(fs_info);
1371         if (!root)
1372                 return ERR_PTR(-ENOMEM);
1373
1374         __setup_root(tree_root->nodesize, tree_root->leafsize,
1375                      tree_root->sectorsize, tree_root->stripesize,
1376                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1377
1378         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1379         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1380         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1381         /*
1382          * log trees do not get reference counted because they go away
1383          * before a real commit is actually done.  They do store pointers
1384          * to file data extents, and those reference counts still get
1385          * updated (along with back refs to the log tree).
1386          */
1387         root->ref_cows = 0;
1388
1389         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1390                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1391                                       0, 0, 0);
1392         if (IS_ERR(leaf)) {
1393                 kfree(root);
1394                 return ERR_CAST(leaf);
1395         }
1396
1397         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1398         btrfs_set_header_bytenr(leaf, leaf->start);
1399         btrfs_set_header_generation(leaf, trans->transid);
1400         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1401         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1402         root->node = leaf;
1403
1404         write_extent_buffer(root->node, root->fs_info->fsid,
1405                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1406         btrfs_mark_buffer_dirty(root->node);
1407         btrfs_tree_unlock(root->node);
1408         return root;
1409 }
1410
1411 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1412                              struct btrfs_fs_info *fs_info)
1413 {
1414         struct btrfs_root *log_root;
1415
1416         log_root = alloc_log_tree(trans, fs_info);
1417         if (IS_ERR(log_root))
1418                 return PTR_ERR(log_root);
1419         WARN_ON(fs_info->log_root_tree);
1420         fs_info->log_root_tree = log_root;
1421         return 0;
1422 }
1423
1424 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1425                        struct btrfs_root *root)
1426 {
1427         struct btrfs_root *log_root;
1428         struct btrfs_inode_item *inode_item;
1429
1430         log_root = alloc_log_tree(trans, root->fs_info);
1431         if (IS_ERR(log_root))
1432                 return PTR_ERR(log_root);
1433
1434         log_root->last_trans = trans->transid;
1435         log_root->root_key.offset = root->root_key.objectid;
1436
1437         inode_item = &log_root->root_item.inode;
1438         btrfs_set_stack_inode_generation(inode_item, 1);
1439         btrfs_set_stack_inode_size(inode_item, 3);
1440         btrfs_set_stack_inode_nlink(inode_item, 1);
1441         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1442         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1443
1444         btrfs_set_root_node(&log_root->root_item, log_root->node);
1445
1446         WARN_ON(root->log_root);
1447         root->log_root = log_root;
1448         root->log_transid = 0;
1449         root->log_transid_committed = -1;
1450         root->last_log_commit = 0;
1451         return 0;
1452 }
1453
1454 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1455                                                struct btrfs_key *key)
1456 {
1457         struct btrfs_root *root;
1458         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1459         struct btrfs_path *path;
1460         u64 generation;
1461         u32 blocksize;
1462         int ret;
1463
1464         path = btrfs_alloc_path();
1465         if (!path)
1466                 return ERR_PTR(-ENOMEM);
1467
1468         root = btrfs_alloc_root(fs_info);
1469         if (!root) {
1470                 ret = -ENOMEM;
1471                 goto alloc_fail;
1472         }
1473
1474         __setup_root(tree_root->nodesize, tree_root->leafsize,
1475                      tree_root->sectorsize, tree_root->stripesize,
1476                      root, fs_info, key->objectid);
1477
1478         ret = btrfs_find_root(tree_root, key, path,
1479                               &root->root_item, &root->root_key);
1480         if (ret) {
1481                 if (ret > 0)
1482                         ret = -ENOENT;
1483                 goto find_fail;
1484         }
1485
1486         generation = btrfs_root_generation(&root->root_item);
1487         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1488         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1489                                      blocksize, generation);
1490         if (!root->node) {
1491                 ret = -ENOMEM;
1492                 goto find_fail;
1493         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1494                 ret = -EIO;
1495                 goto read_fail;
1496         }
1497         root->commit_root = btrfs_root_node(root);
1498 out:
1499         btrfs_free_path(path);
1500         return root;
1501
1502 read_fail:
1503         free_extent_buffer(root->node);
1504 find_fail:
1505         kfree(root);
1506 alloc_fail:
1507         root = ERR_PTR(ret);
1508         goto out;
1509 }
1510
1511 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1512                                       struct btrfs_key *location)
1513 {
1514         struct btrfs_root *root;
1515
1516         root = btrfs_read_tree_root(tree_root, location);
1517         if (IS_ERR(root))
1518                 return root;
1519
1520         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1521                 root->ref_cows = 1;
1522                 btrfs_check_and_init_root_item(&root->root_item);
1523         }
1524
1525         return root;
1526 }
1527
1528 int btrfs_init_fs_root(struct btrfs_root *root)
1529 {
1530         int ret;
1531         struct btrfs_subvolume_writers *writers;
1532
1533         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1534         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1535                                         GFP_NOFS);
1536         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1537                 ret = -ENOMEM;
1538                 goto fail;
1539         }
1540
1541         writers = btrfs_alloc_subvolume_writers();
1542         if (IS_ERR(writers)) {
1543                 ret = PTR_ERR(writers);
1544                 goto fail;
1545         }
1546         root->subv_writers = writers;
1547
1548         btrfs_init_free_ino_ctl(root);
1549         mutex_init(&root->fs_commit_mutex);
1550         spin_lock_init(&root->cache_lock);
1551         init_waitqueue_head(&root->cache_wait);
1552
1553         ret = get_anon_bdev(&root->anon_dev);
1554         if (ret)
1555                 goto free_writers;
1556         return 0;
1557
1558 free_writers:
1559         btrfs_free_subvolume_writers(root->subv_writers);
1560 fail:
1561         kfree(root->free_ino_ctl);
1562         kfree(root->free_ino_pinned);
1563         return ret;
1564 }
1565
1566 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1567                                                u64 root_id)
1568 {
1569         struct btrfs_root *root;
1570
1571         spin_lock(&fs_info->fs_roots_radix_lock);
1572         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1573                                  (unsigned long)root_id);
1574         spin_unlock(&fs_info->fs_roots_radix_lock);
1575         return root;
1576 }
1577
1578 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1579                          struct btrfs_root *root)
1580 {
1581         int ret;
1582
1583         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1584         if (ret)
1585                 return ret;
1586
1587         spin_lock(&fs_info->fs_roots_radix_lock);
1588         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1589                                 (unsigned long)root->root_key.objectid,
1590                                 root);
1591         if (ret == 0)
1592                 root->in_radix = 1;
1593         spin_unlock(&fs_info->fs_roots_radix_lock);
1594         radix_tree_preload_end();
1595
1596         return ret;
1597 }
1598
1599 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1600                                      struct btrfs_key *location,
1601                                      bool check_ref)
1602 {
1603         struct btrfs_root *root;
1604         int ret;
1605
1606         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1607                 return fs_info->tree_root;
1608         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1609                 return fs_info->extent_root;
1610         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1611                 return fs_info->chunk_root;
1612         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1613                 return fs_info->dev_root;
1614         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1615                 return fs_info->csum_root;
1616         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1617                 return fs_info->quota_root ? fs_info->quota_root :
1618                                              ERR_PTR(-ENOENT);
1619         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1620                 return fs_info->uuid_root ? fs_info->uuid_root :
1621                                             ERR_PTR(-ENOENT);
1622 again:
1623         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1624         if (root) {
1625                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1626                         return ERR_PTR(-ENOENT);
1627                 return root;
1628         }
1629
1630         root = btrfs_read_fs_root(fs_info->tree_root, location);
1631         if (IS_ERR(root))
1632                 return root;
1633
1634         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1635                 ret = -ENOENT;
1636                 goto fail;
1637         }
1638
1639         ret = btrfs_init_fs_root(root);
1640         if (ret)
1641                 goto fail;
1642
1643         ret = btrfs_find_item(fs_info->tree_root, NULL, BTRFS_ORPHAN_OBJECTID,
1644                         location->objectid, BTRFS_ORPHAN_ITEM_KEY, NULL);
1645         if (ret < 0)
1646                 goto fail;
1647         if (ret == 0)
1648                 root->orphan_item_inserted = 1;
1649
1650         ret = btrfs_insert_fs_root(fs_info, root);
1651         if (ret) {
1652                 if (ret == -EEXIST) {
1653                         free_fs_root(root);
1654                         goto again;
1655                 }
1656                 goto fail;
1657         }
1658         return root;
1659 fail:
1660         free_fs_root(root);
1661         return ERR_PTR(ret);
1662 }
1663
1664 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1665 {
1666         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1667         int ret = 0;
1668         struct btrfs_device *device;
1669         struct backing_dev_info *bdi;
1670
1671         rcu_read_lock();
1672         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1673                 if (!device->bdev)
1674                         continue;
1675                 bdi = blk_get_backing_dev_info(device->bdev);
1676                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1677                         ret = 1;
1678                         break;
1679                 }
1680         }
1681         rcu_read_unlock();
1682         return ret;
1683 }
1684
1685 /*
1686  * If this fails, caller must call bdi_destroy() to get rid of the
1687  * bdi again.
1688  */
1689 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1690 {
1691         int err;
1692
1693         bdi->capabilities = BDI_CAP_MAP_COPY;
1694         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1695         if (err)
1696                 return err;
1697
1698         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1699         bdi->congested_fn       = btrfs_congested_fn;
1700         bdi->congested_data     = info;
1701         return 0;
1702 }
1703
1704 /*
1705  * called by the kthread helper functions to finally call the bio end_io
1706  * functions.  This is where read checksum verification actually happens
1707  */
1708 static void end_workqueue_fn(struct btrfs_work *work)
1709 {
1710         struct bio *bio;
1711         struct end_io_wq *end_io_wq;
1712         int error;
1713
1714         end_io_wq = container_of(work, struct end_io_wq, work);
1715         bio = end_io_wq->bio;
1716
1717         error = end_io_wq->error;
1718         bio->bi_private = end_io_wq->private;
1719         bio->bi_end_io = end_io_wq->end_io;
1720         kfree(end_io_wq);
1721         bio_endio_nodec(bio, error);
1722 }
1723
1724 static int cleaner_kthread(void *arg)
1725 {
1726         struct btrfs_root *root = arg;
1727         int again;
1728
1729         do {
1730                 again = 0;
1731
1732                 /* Make the cleaner go to sleep early. */
1733                 if (btrfs_need_cleaner_sleep(root))
1734                         goto sleep;
1735
1736                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1737                         goto sleep;
1738
1739                 /*
1740                  * Avoid the problem that we change the status of the fs
1741                  * during the above check and trylock.
1742                  */
1743                 if (btrfs_need_cleaner_sleep(root)) {
1744                         mutex_unlock(&root->fs_info->cleaner_mutex);
1745                         goto sleep;
1746                 }
1747
1748                 btrfs_run_delayed_iputs(root);
1749                 again = btrfs_clean_one_deleted_snapshot(root);
1750                 mutex_unlock(&root->fs_info->cleaner_mutex);
1751
1752                 /*
1753                  * The defragger has dealt with the R/O remount and umount,
1754                  * needn't do anything special here.
1755                  */
1756                 btrfs_run_defrag_inodes(root->fs_info);
1757 sleep:
1758                 if (!try_to_freeze() && !again) {
1759                         set_current_state(TASK_INTERRUPTIBLE);
1760                         if (!kthread_should_stop())
1761                                 schedule();
1762                         __set_current_state(TASK_RUNNING);
1763                 }
1764         } while (!kthread_should_stop());
1765         return 0;
1766 }
1767
1768 static int transaction_kthread(void *arg)
1769 {
1770         struct btrfs_root *root = arg;
1771         struct btrfs_trans_handle *trans;
1772         struct btrfs_transaction *cur;
1773         u64 transid;
1774         unsigned long now;
1775         unsigned long delay;
1776         bool cannot_commit;
1777
1778         do {
1779                 cannot_commit = false;
1780                 delay = HZ * root->fs_info->commit_interval;
1781                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1782
1783                 spin_lock(&root->fs_info->trans_lock);
1784                 cur = root->fs_info->running_transaction;
1785                 if (!cur) {
1786                         spin_unlock(&root->fs_info->trans_lock);
1787                         goto sleep;
1788                 }
1789
1790                 now = get_seconds();
1791                 if (cur->state < TRANS_STATE_BLOCKED &&
1792                     (now < cur->start_time ||
1793                      now - cur->start_time < root->fs_info->commit_interval)) {
1794                         spin_unlock(&root->fs_info->trans_lock);
1795                         delay = HZ * 5;
1796                         goto sleep;
1797                 }
1798                 transid = cur->transid;
1799                 spin_unlock(&root->fs_info->trans_lock);
1800
1801                 /* If the file system is aborted, this will always fail. */
1802                 trans = btrfs_attach_transaction(root);
1803                 if (IS_ERR(trans)) {
1804                         if (PTR_ERR(trans) != -ENOENT)
1805                                 cannot_commit = true;
1806                         goto sleep;
1807                 }
1808                 if (transid == trans->transid) {
1809                         btrfs_commit_transaction(trans, root);
1810                 } else {
1811                         btrfs_end_transaction(trans, root);
1812                 }
1813 sleep:
1814                 wake_up_process(root->fs_info->cleaner_kthread);
1815                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1816
1817                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1818                                       &root->fs_info->fs_state)))
1819                         btrfs_cleanup_transaction(root);
1820                 if (!try_to_freeze()) {
1821                         set_current_state(TASK_INTERRUPTIBLE);
1822                         if (!kthread_should_stop() &&
1823                             (!btrfs_transaction_blocked(root->fs_info) ||
1824                              cannot_commit))
1825                                 schedule_timeout(delay);
1826                         __set_current_state(TASK_RUNNING);
1827                 }
1828         } while (!kthread_should_stop());
1829         return 0;
1830 }
1831
1832 /*
1833  * this will find the highest generation in the array of
1834  * root backups.  The index of the highest array is returned,
1835  * or -1 if we can't find anything.
1836  *
1837  * We check to make sure the array is valid by comparing the
1838  * generation of the latest  root in the array with the generation
1839  * in the super block.  If they don't match we pitch it.
1840  */
1841 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1842 {
1843         u64 cur;
1844         int newest_index = -1;
1845         struct btrfs_root_backup *root_backup;
1846         int i;
1847
1848         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1849                 root_backup = info->super_copy->super_roots + i;
1850                 cur = btrfs_backup_tree_root_gen(root_backup);
1851                 if (cur == newest_gen)
1852                         newest_index = i;
1853         }
1854
1855         /* check to see if we actually wrapped around */
1856         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1857                 root_backup = info->super_copy->super_roots;
1858                 cur = btrfs_backup_tree_root_gen(root_backup);
1859                 if (cur == newest_gen)
1860                         newest_index = 0;
1861         }
1862         return newest_index;
1863 }
1864
1865
1866 /*
1867  * find the oldest backup so we know where to store new entries
1868  * in the backup array.  This will set the backup_root_index
1869  * field in the fs_info struct
1870  */
1871 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1872                                      u64 newest_gen)
1873 {
1874         int newest_index = -1;
1875
1876         newest_index = find_newest_super_backup(info, newest_gen);
1877         /* if there was garbage in there, just move along */
1878         if (newest_index == -1) {
1879                 info->backup_root_index = 0;
1880         } else {
1881                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1882         }
1883 }
1884
1885 /*
1886  * copy all the root pointers into the super backup array.
1887  * this will bump the backup pointer by one when it is
1888  * done
1889  */
1890 static void backup_super_roots(struct btrfs_fs_info *info)
1891 {
1892         int next_backup;
1893         struct btrfs_root_backup *root_backup;
1894         int last_backup;
1895
1896         next_backup = info->backup_root_index;
1897         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1898                 BTRFS_NUM_BACKUP_ROOTS;
1899
1900         /*
1901          * just overwrite the last backup if we're at the same generation
1902          * this happens only at umount
1903          */
1904         root_backup = info->super_for_commit->super_roots + last_backup;
1905         if (btrfs_backup_tree_root_gen(root_backup) ==
1906             btrfs_header_generation(info->tree_root->node))
1907                 next_backup = last_backup;
1908
1909         root_backup = info->super_for_commit->super_roots + next_backup;
1910
1911         /*
1912          * make sure all of our padding and empty slots get zero filled
1913          * regardless of which ones we use today
1914          */
1915         memset(root_backup, 0, sizeof(*root_backup));
1916
1917         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1918
1919         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1920         btrfs_set_backup_tree_root_gen(root_backup,
1921                                btrfs_header_generation(info->tree_root->node));
1922
1923         btrfs_set_backup_tree_root_level(root_backup,
1924                                btrfs_header_level(info->tree_root->node));
1925
1926         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1927         btrfs_set_backup_chunk_root_gen(root_backup,
1928                                btrfs_header_generation(info->chunk_root->node));
1929         btrfs_set_backup_chunk_root_level(root_backup,
1930                                btrfs_header_level(info->chunk_root->node));
1931
1932         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1933         btrfs_set_backup_extent_root_gen(root_backup,
1934                                btrfs_header_generation(info->extent_root->node));
1935         btrfs_set_backup_extent_root_level(root_backup,
1936                                btrfs_header_level(info->extent_root->node));
1937
1938         /*
1939          * we might commit during log recovery, which happens before we set
1940          * the fs_root.  Make sure it is valid before we fill it in.
1941          */
1942         if (info->fs_root && info->fs_root->node) {
1943                 btrfs_set_backup_fs_root(root_backup,
1944                                          info->fs_root->node->start);
1945                 btrfs_set_backup_fs_root_gen(root_backup,
1946                                btrfs_header_generation(info->fs_root->node));
1947                 btrfs_set_backup_fs_root_level(root_backup,
1948                                btrfs_header_level(info->fs_root->node));
1949         }
1950
1951         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1952         btrfs_set_backup_dev_root_gen(root_backup,
1953                                btrfs_header_generation(info->dev_root->node));
1954         btrfs_set_backup_dev_root_level(root_backup,
1955                                        btrfs_header_level(info->dev_root->node));
1956
1957         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1958         btrfs_set_backup_csum_root_gen(root_backup,
1959                                btrfs_header_generation(info->csum_root->node));
1960         btrfs_set_backup_csum_root_level(root_backup,
1961                                btrfs_header_level(info->csum_root->node));
1962
1963         btrfs_set_backup_total_bytes(root_backup,
1964                              btrfs_super_total_bytes(info->super_copy));
1965         btrfs_set_backup_bytes_used(root_backup,
1966                              btrfs_super_bytes_used(info->super_copy));
1967         btrfs_set_backup_num_devices(root_backup,
1968                              btrfs_super_num_devices(info->super_copy));
1969
1970         /*
1971          * if we don't copy this out to the super_copy, it won't get remembered
1972          * for the next commit
1973          */
1974         memcpy(&info->super_copy->super_roots,
1975                &info->super_for_commit->super_roots,
1976                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1977 }
1978
1979 /*
1980  * this copies info out of the root backup array and back into
1981  * the in-memory super block.  It is meant to help iterate through
1982  * the array, so you send it the number of backups you've already
1983  * tried and the last backup index you used.
1984  *
1985  * this returns -1 when it has tried all the backups
1986  */
1987 static noinline int next_root_backup(struct btrfs_fs_info *info,
1988                                      struct btrfs_super_block *super,
1989                                      int *num_backups_tried, int *backup_index)
1990 {
1991         struct btrfs_root_backup *root_backup;
1992         int newest = *backup_index;
1993
1994         if (*num_backups_tried == 0) {
1995                 u64 gen = btrfs_super_generation(super);
1996
1997                 newest = find_newest_super_backup(info, gen);
1998                 if (newest == -1)
1999                         return -1;
2000
2001                 *backup_index = newest;
2002                 *num_backups_tried = 1;
2003         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
2004                 /* we've tried all the backups, all done */
2005                 return -1;
2006         } else {
2007                 /* jump to the next oldest backup */
2008                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
2009                         BTRFS_NUM_BACKUP_ROOTS;
2010                 *backup_index = newest;
2011                 *num_backups_tried += 1;
2012         }
2013         root_backup = super->super_roots + newest;
2014
2015         btrfs_set_super_generation(super,
2016                                    btrfs_backup_tree_root_gen(root_backup));
2017         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
2018         btrfs_set_super_root_level(super,
2019                                    btrfs_backup_tree_root_level(root_backup));
2020         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
2021
2022         /*
2023          * fixme: the total bytes and num_devices need to match or we should
2024          * need a fsck
2025          */
2026         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2027         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2028         return 0;
2029 }
2030
2031 /* helper to cleanup workers */
2032 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2033 {
2034         btrfs_destroy_workqueue(fs_info->fixup_workers);
2035         btrfs_destroy_workqueue(fs_info->delalloc_workers);
2036         btrfs_destroy_workqueue(fs_info->workers);
2037         btrfs_destroy_workqueue(fs_info->endio_workers);
2038         btrfs_destroy_workqueue(fs_info->endio_meta_workers);
2039         btrfs_destroy_workqueue(fs_info->endio_raid56_workers);
2040         btrfs_destroy_workqueue(fs_info->rmw_workers);
2041         btrfs_destroy_workqueue(fs_info->endio_meta_write_workers);
2042         btrfs_destroy_workqueue(fs_info->endio_write_workers);
2043         btrfs_destroy_workqueue(fs_info->endio_freespace_worker);
2044         btrfs_destroy_workqueue(fs_info->submit_workers);
2045         btrfs_destroy_workqueue(fs_info->delayed_workers);
2046         btrfs_destroy_workqueue(fs_info->caching_workers);
2047         btrfs_destroy_workqueue(fs_info->readahead_workers);
2048         btrfs_destroy_workqueue(fs_info->flush_workers);
2049         btrfs_destroy_workqueue(fs_info->qgroup_rescan_workers);
2050 }
2051
2052 static void free_root_extent_buffers(struct btrfs_root *root)
2053 {
2054         if (root) {
2055                 free_extent_buffer(root->node);
2056                 free_extent_buffer(root->commit_root);
2057                 root->node = NULL;
2058                 root->commit_root = NULL;
2059         }
2060 }
2061
2062 /* helper to cleanup tree roots */
2063 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2064 {
2065         free_root_extent_buffers(info->tree_root);
2066
2067         free_root_extent_buffers(info->dev_root);
2068         free_root_extent_buffers(info->extent_root);
2069         free_root_extent_buffers(info->csum_root);
2070         free_root_extent_buffers(info->quota_root);
2071         free_root_extent_buffers(info->uuid_root);
2072         if (chunk_root)
2073                 free_root_extent_buffers(info->chunk_root);
2074 }
2075
2076 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2077 {
2078         int ret;
2079         struct btrfs_root *gang[8];
2080         int i;
2081
2082         while (!list_empty(&fs_info->dead_roots)) {
2083                 gang[0] = list_entry(fs_info->dead_roots.next,
2084                                      struct btrfs_root, root_list);
2085                 list_del(&gang[0]->root_list);
2086
2087                 if (gang[0]->in_radix) {
2088                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2089                 } else {
2090                         free_extent_buffer(gang[0]->node);
2091                         free_extent_buffer(gang[0]->commit_root);
2092                         btrfs_put_fs_root(gang[0]);
2093                 }
2094         }
2095
2096         while (1) {
2097                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2098                                              (void **)gang, 0,
2099                                              ARRAY_SIZE(gang));
2100                 if (!ret)
2101                         break;
2102                 for (i = 0; i < ret; i++)
2103                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2104         }
2105
2106         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
2107                 btrfs_free_log_root_tree(NULL, fs_info);
2108                 btrfs_destroy_pinned_extent(fs_info->tree_root,
2109                                             fs_info->pinned_extents);
2110         }
2111 }
2112
2113 int open_ctree(struct super_block *sb,
2114                struct btrfs_fs_devices *fs_devices,
2115                char *options)
2116 {
2117         u32 sectorsize;
2118         u32 nodesize;
2119         u32 leafsize;
2120         u32 blocksize;
2121         u32 stripesize;
2122         u64 generation;
2123         u64 features;
2124         struct btrfs_key location;
2125         struct buffer_head *bh;
2126         struct btrfs_super_block *disk_super;
2127         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2128         struct btrfs_root *tree_root;
2129         struct btrfs_root *extent_root;
2130         struct btrfs_root *csum_root;
2131         struct btrfs_root *chunk_root;
2132         struct btrfs_root *dev_root;
2133         struct btrfs_root *quota_root;
2134         struct btrfs_root *uuid_root;
2135         struct btrfs_root *log_tree_root;
2136         int ret;
2137         int err = -EINVAL;
2138         int num_backups_tried = 0;
2139         int backup_index = 0;
2140         int max_active;
2141         int flags = WQ_MEM_RECLAIM | WQ_FREEZABLE | WQ_UNBOUND;
2142         bool create_uuid_tree;
2143         bool check_uuid_tree;
2144
2145         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2146         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2147         if (!tree_root || !chunk_root) {
2148                 err = -ENOMEM;
2149                 goto fail;
2150         }
2151
2152         ret = init_srcu_struct(&fs_info->subvol_srcu);
2153         if (ret) {
2154                 err = ret;
2155                 goto fail;
2156         }
2157
2158         ret = setup_bdi(fs_info, &fs_info->bdi);
2159         if (ret) {
2160                 err = ret;
2161                 goto fail_srcu;
2162         }
2163
2164         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2165         if (ret) {
2166                 err = ret;
2167                 goto fail_bdi;
2168         }
2169         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2170                                         (1 + ilog2(nr_cpu_ids));
2171
2172         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2173         if (ret) {
2174                 err = ret;
2175                 goto fail_dirty_metadata_bytes;
2176         }
2177
2178         ret = percpu_counter_init(&fs_info->bio_counter, 0);
2179         if (ret) {
2180                 err = ret;
2181                 goto fail_delalloc_bytes;
2182         }
2183
2184         fs_info->btree_inode = new_inode(sb);
2185         if (!fs_info->btree_inode) {
2186                 err = -ENOMEM;
2187                 goto fail_bio_counter;
2188         }
2189
2190         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2191
2192         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2193         INIT_RADIX_TREE(&fs_info->buffer_radix, GFP_ATOMIC);
2194         INIT_LIST_HEAD(&fs_info->trans_list);
2195         INIT_LIST_HEAD(&fs_info->dead_roots);
2196         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2197         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2198         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2199         spin_lock_init(&fs_info->delalloc_root_lock);
2200         spin_lock_init(&fs_info->trans_lock);
2201         spin_lock_init(&fs_info->fs_roots_radix_lock);
2202         spin_lock_init(&fs_info->delayed_iput_lock);
2203         spin_lock_init(&fs_info->defrag_inodes_lock);
2204         spin_lock_init(&fs_info->free_chunk_lock);
2205         spin_lock_init(&fs_info->tree_mod_seq_lock);
2206         spin_lock_init(&fs_info->super_lock);
2207         spin_lock_init(&fs_info->buffer_lock);
2208         rwlock_init(&fs_info->tree_mod_log_lock);
2209         mutex_init(&fs_info->reloc_mutex);
2210         mutex_init(&fs_info->delalloc_root_mutex);
2211         seqlock_init(&fs_info->profiles_lock);
2212
2213         init_completion(&fs_info->kobj_unregister);
2214         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2215         INIT_LIST_HEAD(&fs_info->space_info);
2216         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2217         btrfs_mapping_init(&fs_info->mapping_tree);
2218         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2219                              BTRFS_BLOCK_RSV_GLOBAL);
2220         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2221                              BTRFS_BLOCK_RSV_DELALLOC);
2222         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2223         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2224         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2225         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2226                              BTRFS_BLOCK_RSV_DELOPS);
2227         atomic_set(&fs_info->nr_async_submits, 0);
2228         atomic_set(&fs_info->async_delalloc_pages, 0);
2229         atomic_set(&fs_info->async_submit_draining, 0);
2230         atomic_set(&fs_info->nr_async_bios, 0);
2231         atomic_set(&fs_info->defrag_running, 0);
2232         atomic64_set(&fs_info->tree_mod_seq, 0);
2233         fs_info->sb = sb;
2234         fs_info->max_inline = 8192 * 1024;
2235         fs_info->metadata_ratio = 0;
2236         fs_info->defrag_inodes = RB_ROOT;
2237         fs_info->free_chunk_space = 0;
2238         fs_info->tree_mod_log = RB_ROOT;
2239         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2240         fs_info->avg_delayed_ref_runtime = div64_u64(NSEC_PER_SEC, 64);
2241         /* readahead state */
2242         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2243         spin_lock_init(&fs_info->reada_lock);
2244
2245         fs_info->thread_pool_size = min_t(unsigned long,
2246                                           num_online_cpus() + 2, 8);
2247
2248         INIT_LIST_HEAD(&fs_info->ordered_roots);
2249         spin_lock_init(&fs_info->ordered_root_lock);
2250         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2251                                         GFP_NOFS);
2252         if (!fs_info->delayed_root) {
2253                 err = -ENOMEM;
2254                 goto fail_iput;
2255         }
2256         btrfs_init_delayed_root(fs_info->delayed_root);
2257
2258         mutex_init(&fs_info->scrub_lock);
2259         atomic_set(&fs_info->scrubs_running, 0);
2260         atomic_set(&fs_info->scrub_pause_req, 0);
2261         atomic_set(&fs_info->scrubs_paused, 0);
2262         atomic_set(&fs_info->scrub_cancel_req, 0);
2263         init_waitqueue_head(&fs_info->replace_wait);
2264         init_waitqueue_head(&fs_info->scrub_pause_wait);
2265         fs_info->scrub_workers_refcnt = 0;
2266 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2267         fs_info->check_integrity_print_mask = 0;
2268 #endif
2269
2270         spin_lock_init(&fs_info->balance_lock);
2271         mutex_init(&fs_info->balance_mutex);
2272         atomic_set(&fs_info->balance_running, 0);
2273         atomic_set(&fs_info->balance_pause_req, 0);
2274         atomic_set(&fs_info->balance_cancel_req, 0);
2275         fs_info->balance_ctl = NULL;
2276         init_waitqueue_head(&fs_info->balance_wait_q);
2277
2278         sb->s_blocksize = 4096;
2279         sb->s_blocksize_bits = blksize_bits(4096);
2280         sb->s_bdi = &fs_info->bdi;
2281
2282         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2283         set_nlink(fs_info->btree_inode, 1);
2284         /*
2285          * we set the i_size on the btree inode to the max possible int.
2286          * the real end of the address space is determined by all of
2287          * the devices in the system
2288          */
2289         fs_info->btree_inode->i_size = OFFSET_MAX;
2290         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2291         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2292
2293         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2294         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2295                              fs_info->btree_inode->i_mapping);
2296         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2297         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2298
2299         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2300
2301         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2302         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2303                sizeof(struct btrfs_key));
2304         set_bit(BTRFS_INODE_DUMMY,
2305                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2306         btrfs_insert_inode_hash(fs_info->btree_inode);
2307
2308         spin_lock_init(&fs_info->block_group_cache_lock);
2309         fs_info->block_group_cache_tree = RB_ROOT;
2310         fs_info->first_logical_byte = (u64)-1;
2311
2312         extent_io_tree_init(&fs_info->freed_extents[0],
2313                              fs_info->btree_inode->i_mapping);
2314         extent_io_tree_init(&fs_info->freed_extents[1],
2315                              fs_info->btree_inode->i_mapping);
2316         fs_info->pinned_extents = &fs_info->freed_extents[0];
2317         fs_info->do_barriers = 1;
2318
2319
2320         mutex_init(&fs_info->ordered_operations_mutex);
2321         mutex_init(&fs_info->ordered_extent_flush_mutex);
2322         mutex_init(&fs_info->tree_log_mutex);
2323         mutex_init(&fs_info->chunk_mutex);
2324         mutex_init(&fs_info->transaction_kthread_mutex);
2325         mutex_init(&fs_info->cleaner_mutex);
2326         mutex_init(&fs_info->volume_mutex);
2327         init_rwsem(&fs_info->extent_commit_sem);
2328         init_rwsem(&fs_info->cleanup_work_sem);
2329         init_rwsem(&fs_info->subvol_sem);
2330         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2331         fs_info->dev_replace.lock_owner = 0;
2332         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2333         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2334         mutex_init(&fs_info->dev_replace.lock_management_lock);
2335         mutex_init(&fs_info->dev_replace.lock);
2336
2337         spin_lock_init(&fs_info->qgroup_lock);
2338         mutex_init(&fs_info->qgroup_ioctl_lock);
2339         fs_info->qgroup_tree = RB_ROOT;
2340         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2341         fs_info->qgroup_seq = 1;
2342         fs_info->quota_enabled = 0;
2343         fs_info->pending_quota_state = 0;
2344         fs_info->qgroup_ulist = NULL;
2345         mutex_init(&fs_info->qgroup_rescan_lock);
2346
2347         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2348         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2349
2350         init_waitqueue_head(&fs_info->transaction_throttle);
2351         init_waitqueue_head(&fs_info->transaction_wait);
2352         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2353         init_waitqueue_head(&fs_info->async_submit_wait);
2354
2355         ret = btrfs_alloc_stripe_hash_table(fs_info);
2356         if (ret) {
2357                 err = ret;
2358                 goto fail_alloc;
2359         }
2360
2361         __setup_root(4096, 4096, 4096, 4096, tree_root,
2362                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2363
2364         invalidate_bdev(fs_devices->latest_bdev);
2365
2366         /*
2367          * Read super block and check the signature bytes only
2368          */
2369         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2370         if (!bh) {
2371                 err = -EINVAL;
2372                 goto fail_alloc;
2373         }
2374
2375         /*
2376          * We want to check superblock checksum, the type is stored inside.
2377          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2378          */
2379         if (btrfs_check_super_csum(bh->b_data)) {
2380                 printk(KERN_ERR "BTRFS: superblock checksum mismatch\n");
2381                 err = -EINVAL;
2382                 goto fail_alloc;
2383         }
2384
2385         /*
2386          * super_copy is zeroed at allocation time and we never touch the
2387          * following bytes up to INFO_SIZE, the checksum is calculated from
2388          * the whole block of INFO_SIZE
2389          */
2390         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2391         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2392                sizeof(*fs_info->super_for_commit));
2393         brelse(bh);
2394
2395         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2396
2397         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2398         if (ret) {
2399                 printk(KERN_ERR "BTRFS: superblock contains fatal errors\n");
2400                 err = -EINVAL;
2401                 goto fail_alloc;
2402         }
2403
2404         disk_super = fs_info->super_copy;
2405         if (!btrfs_super_root(disk_super))
2406                 goto fail_alloc;
2407
2408         /* check FS state, whether FS is broken. */
2409         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2410                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2411
2412         /*
2413          * run through our array of backup supers and setup
2414          * our ring pointer to the oldest one
2415          */
2416         generation = btrfs_super_generation(disk_super);
2417         find_oldest_super_backup(fs_info, generation);
2418
2419         /*
2420          * In the long term, we'll store the compression type in the super
2421          * block, and it'll be used for per file compression control.
2422          */
2423         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2424
2425         ret = btrfs_parse_options(tree_root, options);
2426         if (ret) {
2427                 err = ret;
2428                 goto fail_alloc;
2429         }
2430
2431         features = btrfs_super_incompat_flags(disk_super) &
2432                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2433         if (features) {
2434                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2435                        "unsupported optional features (%Lx).\n",
2436                        features);
2437                 err = -EINVAL;
2438                 goto fail_alloc;
2439         }
2440
2441         if (btrfs_super_leafsize(disk_super) !=
2442             btrfs_super_nodesize(disk_super)) {
2443                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2444                        "blocksizes don't match.  node %d leaf %d\n",
2445                        btrfs_super_nodesize(disk_super),
2446                        btrfs_super_leafsize(disk_super));
2447                 err = -EINVAL;
2448                 goto fail_alloc;
2449         }
2450         if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2451                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2452                        "blocksize (%d) was too large\n",
2453                        btrfs_super_leafsize(disk_super));
2454                 err = -EINVAL;
2455                 goto fail_alloc;
2456         }
2457
2458         features = btrfs_super_incompat_flags(disk_super);
2459         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2460         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2461                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2462
2463         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2464                 printk(KERN_ERR "BTRFS: has skinny extents\n");
2465
2466         /*
2467          * flag our filesystem as having big metadata blocks if
2468          * they are bigger than the page size
2469          */
2470         if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2471                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2472                         printk(KERN_INFO "BTRFS: flagging fs with big metadata feature\n");
2473                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2474         }
2475
2476         nodesize = btrfs_super_nodesize(disk_super);
2477         leafsize = btrfs_super_leafsize(disk_super);
2478         sectorsize = btrfs_super_sectorsize(disk_super);
2479         stripesize = btrfs_super_stripesize(disk_super);
2480         fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2481         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2482
2483         /*
2484          * mixed block groups end up with duplicate but slightly offset
2485          * extent buffers for the same range.  It leads to corruptions
2486          */
2487         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2488             (sectorsize != leafsize)) {
2489                 printk(KERN_WARNING "BTRFS: unequal leaf/node/sector sizes "
2490                                 "are not allowed for mixed block groups on %s\n",
2491                                 sb->s_id);
2492                 goto fail_alloc;
2493         }
2494
2495         /*
2496          * Needn't use the lock because there is no other task which will
2497          * update the flag.
2498          */
2499         btrfs_set_super_incompat_flags(disk_super, features);
2500
2501         features = btrfs_super_compat_ro_flags(disk_super) &
2502                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2503         if (!(sb->s_flags & MS_RDONLY) && features) {
2504                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2505                        "unsupported option features (%Lx).\n",
2506                        features);
2507                 err = -EINVAL;
2508                 goto fail_alloc;
2509         }
2510
2511         max_active = fs_info->thread_pool_size;
2512
2513         fs_info->workers =
2514                 btrfs_alloc_workqueue("worker", flags | WQ_HIGHPRI,
2515                                       max_active, 16);
2516
2517         fs_info->delalloc_workers =
2518                 btrfs_alloc_workqueue("delalloc", flags, max_active, 2);
2519
2520         fs_info->flush_workers =
2521                 btrfs_alloc_workqueue("flush_delalloc", flags, max_active, 0);
2522
2523         fs_info->caching_workers =
2524                 btrfs_alloc_workqueue("cache", flags, max_active, 0);
2525
2526         /*
2527          * a higher idle thresh on the submit workers makes it much more
2528          * likely that bios will be send down in a sane order to the
2529          * devices
2530          */
2531         fs_info->submit_workers =
2532                 btrfs_alloc_workqueue("submit", flags,
2533                                       min_t(u64, fs_devices->num_devices,
2534                                             max_active), 64);
2535
2536         fs_info->fixup_workers =
2537                 btrfs_alloc_workqueue("fixup", flags, 1, 0);
2538
2539         /*
2540          * endios are largely parallel and should have a very
2541          * low idle thresh
2542          */
2543         fs_info->endio_workers =
2544                 btrfs_alloc_workqueue("endio", flags, max_active, 4);
2545         fs_info->endio_meta_workers =
2546                 btrfs_alloc_workqueue("endio-meta", flags, max_active, 4);
2547         fs_info->endio_meta_write_workers =
2548                 btrfs_alloc_workqueue("endio-meta-write", flags, max_active, 2);
2549         fs_info->endio_raid56_workers =
2550                 btrfs_alloc_workqueue("endio-raid56", flags, max_active, 4);
2551         fs_info->rmw_workers =
2552                 btrfs_alloc_workqueue("rmw", flags, max_active, 2);
2553         fs_info->endio_write_workers =
2554                 btrfs_alloc_workqueue("endio-write", flags, max_active, 2);
2555         fs_info->endio_freespace_worker =
2556                 btrfs_alloc_workqueue("freespace-write", flags, max_active, 0);
2557         fs_info->delayed_workers =
2558                 btrfs_alloc_workqueue("delayed-meta", flags, max_active, 0);
2559         fs_info->readahead_workers =
2560                 btrfs_alloc_workqueue("readahead", flags, max_active, 2);
2561         fs_info->qgroup_rescan_workers =
2562                 btrfs_alloc_workqueue("qgroup-rescan", flags, 1, 0);
2563
2564         if (!(fs_info->workers && fs_info->delalloc_workers &&
2565               fs_info->submit_workers && fs_info->flush_workers &&
2566               fs_info->endio_workers && fs_info->endio_meta_workers &&
2567               fs_info->endio_meta_write_workers &&
2568               fs_info->endio_write_workers && fs_info->endio_raid56_workers &&
2569               fs_info->endio_freespace_worker && fs_info->rmw_workers &&
2570               fs_info->caching_workers && fs_info->readahead_workers &&
2571               fs_info->fixup_workers && fs_info->delayed_workers &&
2572               fs_info->qgroup_rescan_workers)) {
2573                 err = -ENOMEM;
2574                 goto fail_sb_buffer;
2575         }
2576
2577         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2578         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2579                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2580
2581         tree_root->nodesize = nodesize;
2582         tree_root->leafsize = leafsize;
2583         tree_root->sectorsize = sectorsize;
2584         tree_root->stripesize = stripesize;
2585
2586         sb->s_blocksize = sectorsize;
2587         sb->s_blocksize_bits = blksize_bits(sectorsize);
2588
2589         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2590                 printk(KERN_INFO "BTRFS: valid FS not found on %s\n", sb->s_id);
2591                 goto fail_sb_buffer;
2592         }
2593
2594         if (sectorsize != PAGE_SIZE) {
2595                 printk(KERN_WARNING "BTRFS: Incompatible sector size(%lu) "
2596                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2597                 goto fail_sb_buffer;
2598         }
2599
2600         mutex_lock(&fs_info->chunk_mutex);
2601         ret = btrfs_read_sys_array(tree_root);
2602         mutex_unlock(&fs_info->chunk_mutex);
2603         if (ret) {
2604                 printk(KERN_WARNING "BTRFS: failed to read the system "
2605                        "array on %s\n", sb->s_id);
2606                 goto fail_sb_buffer;
2607         }
2608
2609         blocksize = btrfs_level_size(tree_root,
2610                                      btrfs_super_chunk_root_level(disk_super));
2611         generation = btrfs_super_chunk_root_generation(disk_super);
2612
2613         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2614                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2615
2616         chunk_root->node = read_tree_block(chunk_root,
2617                                            btrfs_super_chunk_root(disk_super),
2618                                            blocksize, generation);
2619         if (!chunk_root->node ||
2620             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2621                 printk(KERN_WARNING "BTRFS: failed to read chunk root on %s\n",
2622                        sb->s_id);
2623                 goto fail_tree_roots;
2624         }
2625         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2626         chunk_root->commit_root = btrfs_root_node(chunk_root);
2627
2628         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2629            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2630
2631         ret = btrfs_read_chunk_tree(chunk_root);
2632         if (ret) {
2633                 printk(KERN_WARNING "BTRFS: failed to read chunk tree on %s\n",
2634                        sb->s_id);
2635                 goto fail_tree_roots;
2636         }
2637
2638         /*
2639          * keep the device that is marked to be the target device for the
2640          * dev_replace procedure
2641          */
2642         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2643
2644         if (!fs_devices->latest_bdev) {
2645                 printk(KERN_CRIT "BTRFS: failed to read devices on %s\n",
2646                        sb->s_id);
2647                 goto fail_tree_roots;
2648         }
2649
2650 retry_root_backup:
2651         blocksize = btrfs_level_size(tree_root,
2652                                      btrfs_super_root_level(disk_super));
2653         generation = btrfs_super_generation(disk_super);
2654
2655         tree_root->node = read_tree_block(tree_root,
2656                                           btrfs_super_root(disk_super),
2657                                           blocksize, generation);
2658         if (!tree_root->node ||
2659             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2660                 printk(KERN_WARNING "BTRFS: failed to read tree root on %s\n",
2661                        sb->s_id);
2662
2663                 goto recovery_tree_root;
2664         }
2665
2666         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2667         tree_root->commit_root = btrfs_root_node(tree_root);
2668         btrfs_set_root_refs(&tree_root->root_item, 1);
2669
2670         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2671         location.type = BTRFS_ROOT_ITEM_KEY;
2672         location.offset = 0;
2673
2674         extent_root = btrfs_read_tree_root(tree_root, &location);
2675         if (IS_ERR(extent_root)) {
2676                 ret = PTR_ERR(extent_root);
2677                 goto recovery_tree_root;
2678         }
2679         extent_root->track_dirty = 1;
2680         fs_info->extent_root = extent_root;
2681
2682         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2683         dev_root = btrfs_read_tree_root(tree_root, &location);
2684         if (IS_ERR(dev_root)) {
2685                 ret = PTR_ERR(dev_root);
2686                 goto recovery_tree_root;
2687         }
2688         dev_root->track_dirty = 1;
2689         fs_info->dev_root = dev_root;
2690         btrfs_init_devices_late(fs_info);
2691
2692         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2693         csum_root = btrfs_read_tree_root(tree_root, &location);
2694         if (IS_ERR(csum_root)) {
2695                 ret = PTR_ERR(csum_root);
2696                 goto recovery_tree_root;
2697         }
2698         csum_root->track_dirty = 1;
2699         fs_info->csum_root = csum_root;
2700
2701         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2702         quota_root = btrfs_read_tree_root(tree_root, &location);
2703         if (!IS_ERR(quota_root)) {
2704                 quota_root->track_dirty = 1;
2705                 fs_info->quota_enabled = 1;
2706                 fs_info->pending_quota_state = 1;
2707                 fs_info->quota_root = quota_root;
2708         }
2709
2710         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2711         uuid_root = btrfs_read_tree_root(tree_root, &location);
2712         if (IS_ERR(uuid_root)) {
2713                 ret = PTR_ERR(uuid_root);
2714                 if (ret != -ENOENT)
2715                         goto recovery_tree_root;
2716                 create_uuid_tree = true;
2717                 check_uuid_tree = false;
2718         } else {
2719                 uuid_root->track_dirty = 1;
2720                 fs_info->uuid_root = uuid_root;
2721                 create_uuid_tree = false;
2722                 check_uuid_tree =
2723                     generation != btrfs_super_uuid_tree_generation(disk_super);
2724         }
2725
2726         fs_info->generation = generation;
2727         fs_info->last_trans_committed = generation;
2728
2729         ret = btrfs_recover_balance(fs_info);
2730         if (ret) {
2731                 printk(KERN_WARNING "BTRFS: failed to recover balance\n");
2732                 goto fail_block_groups;
2733         }
2734
2735         ret = btrfs_init_dev_stats(fs_info);
2736         if (ret) {
2737                 printk(KERN_ERR "BTRFS: failed to init dev_stats: %d\n",
2738                        ret);
2739                 goto fail_block_groups;
2740         }
2741
2742         ret = btrfs_init_dev_replace(fs_info);
2743         if (ret) {
2744                 pr_err("BTRFS: failed to init dev_replace: %d\n", ret);
2745                 goto fail_block_groups;
2746         }
2747
2748         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2749
2750         ret = btrfs_sysfs_add_one(fs_info);
2751         if (ret) {
2752                 pr_err("BTRFS: failed to init sysfs interface: %d\n", ret);
2753                 goto fail_block_groups;
2754         }
2755
2756         ret = btrfs_init_space_info(fs_info);
2757         if (ret) {
2758                 printk(KERN_ERR "BTRFS: Failed to initial space info: %d\n", ret);
2759                 goto fail_sysfs;
2760         }
2761
2762         ret = btrfs_read_block_groups(extent_root);
2763         if (ret) {
2764                 printk(KERN_ERR "BTRFS: Failed to read block groups: %d\n", ret);
2765                 goto fail_sysfs;
2766         }
2767         fs_info->num_tolerated_disk_barrier_failures =
2768                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2769         if (fs_info->fs_devices->missing_devices >
2770              fs_info->num_tolerated_disk_barrier_failures &&
2771             !(sb->s_flags & MS_RDONLY)) {
2772                 printk(KERN_WARNING "BTRFS: "
2773                         "too many missing devices, writeable mount is not allowed\n");
2774                 goto fail_sysfs;
2775         }
2776
2777         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2778                                                "btrfs-cleaner");
2779         if (IS_ERR(fs_info->cleaner_kthread))
2780                 goto fail_sysfs;
2781
2782         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2783                                                    tree_root,
2784                                                    "btrfs-transaction");
2785         if (IS_ERR(fs_info->transaction_kthread))
2786                 goto fail_cleaner;
2787
2788         if (!btrfs_test_opt(tree_root, SSD) &&
2789             !btrfs_test_opt(tree_root, NOSSD) &&
2790             !fs_info->fs_devices->rotating) {
2791                 printk(KERN_INFO "BTRFS: detected SSD devices, enabling SSD "
2792                        "mode\n");
2793                 btrfs_set_opt(fs_info->mount_opt, SSD);
2794         }
2795
2796         /* Set the real inode map cache flag */
2797         if (btrfs_test_opt(tree_root, CHANGE_INODE_CACHE))
2798                 btrfs_set_opt(tree_root->fs_info->mount_opt, INODE_MAP_CACHE);
2799
2800 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2801         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2802                 ret = btrfsic_mount(tree_root, fs_devices,
2803                                     btrfs_test_opt(tree_root,
2804                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2805                                     1 : 0,
2806                                     fs_info->check_integrity_print_mask);
2807                 if (ret)
2808                         printk(KERN_WARNING "BTRFS: failed to initialize"
2809                                " integrity check module %s\n", sb->s_id);
2810         }
2811 #endif
2812         ret = btrfs_read_qgroup_config(fs_info);
2813         if (ret)
2814                 goto fail_trans_kthread;
2815
2816         /* do not make disk changes in broken FS */
2817         if (btrfs_super_log_root(disk_super) != 0) {
2818                 u64 bytenr = btrfs_super_log_root(disk_super);
2819
2820                 if (fs_devices->rw_devices == 0) {
2821                         printk(KERN_WARNING "BTRFS: log replay required "
2822                                "on RO media\n");
2823                         err = -EIO;
2824                         goto fail_qgroup;
2825                 }
2826                 blocksize =
2827                      btrfs_level_size(tree_root,
2828                                       btrfs_super_log_root_level(disk_super));
2829
2830                 log_tree_root = btrfs_alloc_root(fs_info);
2831                 if (!log_tree_root) {
2832                         err = -ENOMEM;
2833                         goto fail_qgroup;
2834                 }
2835
2836                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2837                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2838
2839                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2840                                                       blocksize,
2841                                                       generation + 1);
2842                 if (!log_tree_root->node ||
2843                     !extent_buffer_uptodate(log_tree_root->node)) {
2844                         printk(KERN_ERR "BTRFS: failed to read log tree\n");
2845                         free_extent_buffer(log_tree_root->node);
2846                         kfree(log_tree_root);
2847                         goto fail_trans_kthread;
2848                 }
2849                 /* returns with log_tree_root freed on success */
2850                 ret = btrfs_recover_log_trees(log_tree_root);
2851                 if (ret) {
2852                         btrfs_error(tree_root->fs_info, ret,
2853                                     "Failed to recover log tree");
2854                         free_extent_buffer(log_tree_root->node);
2855                         kfree(log_tree_root);
2856                         goto fail_trans_kthread;
2857                 }
2858
2859                 if (sb->s_flags & MS_RDONLY) {
2860                         ret = btrfs_commit_super(tree_root);
2861                         if (ret)
2862                                 goto fail_trans_kthread;
2863                 }
2864         }
2865
2866         ret = btrfs_find_orphan_roots(tree_root);
2867         if (ret)
2868                 goto fail_trans_kthread;
2869
2870         if (!(sb->s_flags & MS_RDONLY)) {
2871                 ret = btrfs_cleanup_fs_roots(fs_info);
2872                 if (ret)
2873                         goto fail_trans_kthread;
2874
2875                 ret = btrfs_recover_relocation(tree_root);
2876                 if (ret < 0) {
2877                         printk(KERN_WARNING
2878                                "BTRFS: failed to recover relocation\n");
2879                         err = -EINVAL;
2880                         goto fail_qgroup;
2881                 }
2882         }
2883
2884         location.objectid = BTRFS_FS_TREE_OBJECTID;
2885         location.type = BTRFS_ROOT_ITEM_KEY;
2886         location.offset = 0;
2887
2888         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2889         if (IS_ERR(fs_info->fs_root)) {
2890                 err = PTR_ERR(fs_info->fs_root);
2891                 goto fail_qgroup;
2892         }
2893
2894         if (sb->s_flags & MS_RDONLY)
2895                 return 0;
2896
2897         down_read(&fs_info->cleanup_work_sem);
2898         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2899             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2900                 up_read(&fs_info->cleanup_work_sem);
2901                 close_ctree(tree_root);
2902                 return ret;
2903         }
2904         up_read(&fs_info->cleanup_work_sem);
2905
2906         ret = btrfs_resume_balance_async(fs_info);
2907         if (ret) {
2908                 printk(KERN_WARNING "BTRFS: failed to resume balance\n");
2909                 close_ctree(tree_root);
2910                 return ret;
2911         }
2912
2913         ret = btrfs_resume_dev_replace_async(fs_info);
2914         if (ret) {
2915                 pr_warn("BTRFS: failed to resume dev_replace\n");
2916                 close_ctree(tree_root);
2917                 return ret;
2918         }
2919
2920         btrfs_qgroup_rescan_resume(fs_info);
2921
2922         if (create_uuid_tree) {
2923                 pr_info("BTRFS: creating UUID tree\n");
2924                 ret = btrfs_create_uuid_tree(fs_info);
2925                 if (ret) {
2926                         pr_warn("BTRFS: failed to create the UUID tree %d\n",
2927                                 ret);
2928                         close_ctree(tree_root);
2929                         return ret;
2930                 }
2931         } else if (check_uuid_tree ||
2932                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2933                 pr_info("BTRFS: checking UUID tree\n");
2934                 ret = btrfs_check_uuid_tree(fs_info);
2935                 if (ret) {
2936                         pr_warn("BTRFS: failed to check the UUID tree %d\n",
2937                                 ret);
2938                         close_ctree(tree_root);
2939                         return ret;
2940                 }
2941         } else {
2942                 fs_info->update_uuid_tree_gen = 1;
2943         }
2944
2945         return 0;
2946
2947 fail_qgroup:
2948         btrfs_free_qgroup_config(fs_info);
2949 fail_trans_kthread:
2950         kthread_stop(fs_info->transaction_kthread);
2951         btrfs_cleanup_transaction(fs_info->tree_root);
2952         del_fs_roots(fs_info);
2953 fail_cleaner:
2954         kthread_stop(fs_info->cleaner_kthread);
2955
2956         /*
2957          * make sure we're done with the btree inode before we stop our
2958          * kthreads
2959          */
2960         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2961
2962 fail_sysfs:
2963         btrfs_sysfs_remove_one(fs_info);
2964
2965 fail_block_groups:
2966         btrfs_put_block_group_cache(fs_info);
2967         btrfs_free_block_groups(fs_info);
2968
2969 fail_tree_roots:
2970         free_root_pointers(fs_info, 1);
2971         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2972
2973 fail_sb_buffer:
2974         btrfs_stop_all_workers(fs_info);
2975 fail_alloc:
2976 fail_iput:
2977         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2978
2979         iput(fs_info->btree_inode);
2980 fail_bio_counter:
2981         percpu_counter_destroy(&fs_info->bio_counter);
2982 fail_delalloc_bytes:
2983         percpu_counter_destroy(&fs_info->delalloc_bytes);
2984 fail_dirty_metadata_bytes:
2985         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2986 fail_bdi:
2987         bdi_destroy(&fs_info->bdi);
2988 fail_srcu:
2989         cleanup_srcu_struct(&fs_info->subvol_srcu);
2990 fail:
2991         btrfs_free_stripe_hash_table(fs_info);
2992         btrfs_close_devices(fs_info->fs_devices);
2993         return err;
2994
2995 recovery_tree_root:
2996         if (!btrfs_test_opt(tree_root, RECOVERY))
2997                 goto fail_tree_roots;
2998
2999         free_root_pointers(fs_info, 0);
3000
3001         /* don't use the log in recovery mode, it won't be valid */
3002         btrfs_set_super_log_root(disk_super, 0);
3003
3004         /* we can't trust the free space cache either */
3005         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
3006
3007         ret = next_root_backup(fs_info, fs_info->super_copy,
3008                                &num_backups_tried, &backup_index);
3009         if (ret == -1)
3010                 goto fail_block_groups;
3011         goto retry_root_backup;
3012 }
3013
3014 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3015 {
3016         if (uptodate) {
3017                 set_buffer_uptodate(bh);
3018         } else {
3019                 struct btrfs_device *device = (struct btrfs_device *)
3020                         bh->b_private;
3021
3022                 printk_ratelimited_in_rcu(KERN_WARNING "BTRFS: lost page write due to "
3023                                           "I/O error on %s\n",
3024                                           rcu_str_deref(device->name));
3025                 /* note, we dont' set_buffer_write_io_error because we have
3026                  * our own ways of dealing with the IO errors
3027                  */
3028                 clear_buffer_uptodate(bh);
3029                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3030         }
3031         unlock_buffer(bh);
3032         put_bh(bh);
3033 }
3034
3035 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3036 {
3037         struct buffer_head *bh;
3038         struct buffer_head *latest = NULL;
3039         struct btrfs_super_block *super;
3040         int i;
3041         u64 transid = 0;
3042         u64 bytenr;
3043
3044         /* we would like to check all the supers, but that would make
3045          * a btrfs mount succeed after a mkfs from a different FS.
3046          * So, we need to add a special mount option to scan for
3047          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3048          */
3049         for (i = 0; i < 1; i++) {
3050                 bytenr = btrfs_sb_offset(i);
3051                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3052                                         i_size_read(bdev->bd_inode))
3053                         break;
3054                 bh = __bread(bdev, bytenr / 4096,
3055                                         BTRFS_SUPER_INFO_SIZE);
3056                 if (!bh)
3057                         continue;
3058
3059                 super = (struct btrfs_super_block *)bh->b_data;
3060                 if (btrfs_super_bytenr(super) != bytenr ||
3061                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3062                         brelse(bh);
3063                         continue;
3064                 }
3065
3066                 if (!latest || btrfs_super_generation(super) > transid) {
3067                         brelse(latest);
3068                         latest = bh;
3069                         transid = btrfs_super_generation(super);
3070                 } else {
3071                         brelse(bh);
3072                 }
3073         }
3074         return latest;
3075 }
3076
3077 /*
3078  * this should be called twice, once with wait == 0 and
3079  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3080  * we write are pinned.
3081  *
3082  * They are released when wait == 1 is done.
3083  * max_mirrors must be the same for both runs, and it indicates how
3084  * many supers on this one device should be written.
3085  *
3086  * max_mirrors == 0 means to write them all.
3087  */
3088 static int write_dev_supers(struct btrfs_device *device,
3089                             struct btrfs_super_block *sb,
3090                             int do_barriers, int wait, int max_mirrors)
3091 {
3092         struct buffer_head *bh;
3093         int i;
3094         int ret;
3095         int errors = 0;
3096         u32 crc;
3097         u64 bytenr;
3098
3099         if (max_mirrors == 0)
3100                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3101
3102         for (i = 0; i < max_mirrors; i++) {
3103                 bytenr = btrfs_sb_offset(i);
3104                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3105                         break;
3106
3107                 if (wait) {
3108                         bh = __find_get_block(device->bdev, bytenr / 4096,
3109                                               BTRFS_SUPER_INFO_SIZE);
3110                         if (!bh) {
3111                                 errors++;
3112                                 continue;
3113                         }
3114                         wait_on_buffer(bh);
3115                         if (!buffer_uptodate(bh))
3116                                 errors++;
3117
3118                         /* drop our reference */
3119                         brelse(bh);
3120
3121                         /* drop the reference from the wait == 0 run */
3122                         brelse(bh);
3123                         continue;
3124                 } else {
3125                         btrfs_set_super_bytenr(sb, bytenr);
3126
3127                         crc = ~(u32)0;
3128                         crc = btrfs_csum_data((char *)sb +
3129                                               BTRFS_CSUM_SIZE, crc,
3130                                               BTRFS_SUPER_INFO_SIZE -
3131                                               BTRFS_CSUM_SIZE);
3132                         btrfs_csum_final(crc, sb->csum);
3133
3134                         /*
3135                          * one reference for us, and we leave it for the
3136                          * caller
3137                          */
3138                         bh = __getblk(device->bdev, bytenr / 4096,
3139                                       BTRFS_SUPER_INFO_SIZE);
3140                         if (!bh) {
3141                                 printk(KERN_ERR "BTRFS: couldn't get super "
3142                                        "buffer head for bytenr %Lu\n", bytenr);
3143                                 errors++;
3144                                 continue;
3145                         }
3146
3147                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3148
3149                         /* one reference for submit_bh */
3150                         get_bh(bh);
3151
3152                         set_buffer_uptodate(bh);
3153                         lock_buffer(bh);
3154                         bh->b_end_io = btrfs_end_buffer_write_sync;
3155                         bh->b_private = device;
3156                 }
3157
3158                 /*
3159                  * we fua the first super.  The others we allow
3160                  * to go down lazy.
3161                  */
3162                 if (i == 0)
3163                         ret = btrfsic_submit_bh(WRITE_FUA, bh);
3164                 else
3165                         ret = btrfsic_submit_bh(WRITE_SYNC, bh);
3166                 if (ret)
3167                         errors++;
3168         }
3169         return errors < i ? 0 : -1;
3170 }
3171
3172 /*
3173  * endio for the write_dev_flush, this will wake anyone waiting
3174  * for the barrier when it is done
3175  */
3176 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3177 {
3178         if (err) {
3179                 if (err == -EOPNOTSUPP)
3180                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3181                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3182         }
3183         if (bio->bi_private)
3184                 complete(bio->bi_private);
3185         bio_put(bio);
3186 }
3187
3188 /*
3189  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3190  * sent down.  With wait == 1, it waits for the previous flush.
3191  *
3192  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3193  * capable
3194  */
3195 static int write_dev_flush(struct btrfs_device *device, int wait)
3196 {
3197         struct bio *bio;
3198         int ret = 0;
3199
3200         if (device->nobarriers)
3201                 return 0;
3202
3203         if (wait) {
3204                 bio = device->flush_bio;
3205                 if (!bio)
3206                         return 0;
3207
3208                 wait_for_completion(&device->flush_wait);
3209
3210                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3211                         printk_in_rcu("BTRFS: disabling barriers on dev %s\n",
3212                                       rcu_str_deref(device->name));
3213                         device->nobarriers = 1;
3214                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3215                         ret = -EIO;
3216                         btrfs_dev_stat_inc_and_print(device,
3217                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3218                 }
3219
3220                 /* drop the reference from the wait == 0 run */
3221                 bio_put(bio);
3222                 device->flush_bio = NULL;
3223
3224                 return ret;
3225         }
3226
3227         /*
3228          * one reference for us, and we leave it for the
3229          * caller
3230          */
3231         device->flush_bio = NULL;
3232         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3233         if (!bio)
3234                 return -ENOMEM;
3235
3236         bio->bi_end_io = btrfs_end_empty_barrier;
3237         bio->bi_bdev = device->bdev;
3238         init_completion(&device->flush_wait);
3239         bio->bi_private = &device->flush_wait;
3240         device->flush_bio = bio;
3241
3242         bio_get(bio);
3243         btrfsic_submit_bio(WRITE_FLUSH, bio);
3244
3245         return 0;
3246 }
3247
3248 /*
3249  * send an empty flush down to each device in parallel,
3250  * then wait for them
3251  */
3252 static int barrier_all_devices(struct btrfs_fs_info *info)
3253 {
3254         struct list_head *head;
3255         struct btrfs_device *dev;
3256         int errors_send = 0;
3257         int errors_wait = 0;
3258         int ret;
3259
3260         /* send down all the barriers */
3261         head = &info->fs_devices->devices;
3262         list_for_each_entry_rcu(dev, head, dev_list) {
3263                 if (dev->missing)
3264                         continue;
3265                 if (!dev->bdev) {
3266                         errors_send++;
3267                         continue;
3268                 }
3269                 if (!dev->in_fs_metadata || !dev->writeable)
3270                         continue;
3271
3272                 ret = write_dev_flush(dev, 0);
3273                 if (ret)
3274                         errors_send++;
3275         }
3276
3277         /* wait for all the barriers */
3278         list_for_each_entry_rcu(dev, head, dev_list) {
3279                 if (dev->missing)
3280                         continue;
3281                 if (!dev->bdev) {
3282                         errors_wait++;
3283                         continue;
3284                 }
3285                 if (!dev->in_fs_metadata || !dev->writeable)
3286                         continue;
3287
3288                 ret = write_dev_flush(dev, 1);
3289                 if (ret)
3290                         errors_wait++;
3291         }
3292         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3293             errors_wait > info->num_tolerated_disk_barrier_failures)
3294                 return -EIO;
3295         return 0;
3296 }
3297
3298 int btrfs_calc_num_tolerated_disk_barrier_failures(
3299         struct btrfs_fs_info *fs_info)
3300 {
3301         struct btrfs_ioctl_space_info space;
3302         struct btrfs_space_info *sinfo;
3303         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3304                        BTRFS_BLOCK_GROUP_SYSTEM,
3305                        BTRFS_BLOCK_GROUP_METADATA,
3306                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3307         int num_types = 4;
3308         int i;
3309         int c;
3310         int num_tolerated_disk_barrier_failures =
3311                 (int)fs_info->fs_devices->num_devices;
3312
3313         for (i = 0; i < num_types; i++) {
3314                 struct btrfs_space_info *tmp;
3315
3316                 sinfo = NULL;
3317                 rcu_read_lock();
3318                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3319                         if (tmp->flags == types[i]) {
3320                                 sinfo = tmp;
3321                                 break;
3322                         }
3323                 }
3324                 rcu_read_unlock();
3325
3326                 if (!sinfo)
3327                         continue;
3328
3329                 down_read(&sinfo->groups_sem);
3330                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3331                         if (!list_empty(&sinfo->block_groups[c])) {
3332                                 u64 flags;
3333
3334                                 btrfs_get_block_group_info(
3335                                         &sinfo->block_groups[c], &space);
3336                                 if (space.total_bytes == 0 ||
3337                                     space.used_bytes == 0)
3338                                         continue;
3339                                 flags = space.flags;
3340                                 /*
3341                                  * return
3342                                  * 0: if dup, single or RAID0 is configured for
3343                                  *    any of metadata, system or data, else
3344                                  * 1: if RAID5 is configured, or if RAID1 or
3345                                  *    RAID10 is configured and only two mirrors
3346                                  *    are used, else
3347                                  * 2: if RAID6 is configured, else
3348                                  * num_mirrors - 1: if RAID1 or RAID10 is
3349                                  *                  configured and more than
3350                                  *                  2 mirrors are used.
3351                                  */
3352                                 if (num_tolerated_disk_barrier_failures > 0 &&
3353                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3354                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3355                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3356                                       == 0)))
3357                                         num_tolerated_disk_barrier_failures = 0;
3358                                 else if (num_tolerated_disk_barrier_failures > 1) {
3359                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3360                                             BTRFS_BLOCK_GROUP_RAID5 |
3361                                             BTRFS_BLOCK_GROUP_RAID10)) {
3362                                                 num_tolerated_disk_barrier_failures = 1;
3363                                         } else if (flags &
3364                                                    BTRFS_BLOCK_GROUP_RAID6) {
3365                                                 num_tolerated_disk_barrier_failures = 2;
3366                                         }
3367                                 }
3368                         }
3369                 }
3370                 up_read(&sinfo->groups_sem);
3371         }
3372
3373         return num_tolerated_disk_barrier_failures;
3374 }
3375
3376 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3377 {
3378         struct list_head *head;
3379         struct btrfs_device *dev;
3380         struct btrfs_super_block *sb;
3381         struct btrfs_dev_item *dev_item;
3382         int ret;
3383         int do_barriers;
3384         int max_errors;
3385         int total_errors = 0;
3386         u64 flags;
3387
3388         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3389         backup_super_roots(root->fs_info);
3390
3391         sb = root->fs_info->super_for_commit;
3392         dev_item = &sb->dev_item;
3393
3394         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3395         head = &root->fs_info->fs_devices->devices;
3396         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3397
3398         if (do_barriers) {
3399                 ret = barrier_all_devices(root->fs_info);
3400                 if (ret) {
3401                         mutex_unlock(
3402                                 &root->fs_info->fs_devices->device_list_mutex);
3403                         btrfs_error(root->fs_info, ret,
3404                                     "errors while submitting device barriers.");
3405                         return ret;
3406                 }
3407         }
3408
3409         list_for_each_entry_rcu(dev, head, dev_list) {
3410                 if (!dev->bdev) {
3411                         total_errors++;
3412                         continue;
3413                 }
3414                 if (!dev->in_fs_metadata || !dev->writeable)
3415                         continue;
3416
3417                 btrfs_set_stack_device_generation(dev_item, 0);
3418                 btrfs_set_stack_device_type(dev_item, dev->type);
3419                 btrfs_set_stack_device_id(dev_item, dev->devid);
3420                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3421                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3422                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3423                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3424                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3425                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3426                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3427
3428                 flags = btrfs_super_flags(sb);
3429                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3430
3431                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3432                 if (ret)
3433                         total_errors++;
3434         }
3435         if (total_errors > max_errors) {
3436                 btrfs_err(root->fs_info, "%d errors while writing supers",
3437                        total_errors);
3438                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3439
3440                 /* FUA is masked off if unsupported and can't be the reason */
3441                 btrfs_error(root->fs_info, -EIO,
3442                             "%d errors while writing supers", total_errors);
3443                 return -EIO;
3444         }
3445
3446         total_errors = 0;
3447         list_for_each_entry_rcu(dev, head, dev_list) {
3448                 if (!dev->bdev)
3449                         continue;
3450                 if (!dev->in_fs_metadata || !dev->writeable)
3451                         continue;
3452
3453                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3454                 if (ret)
3455                         total_errors++;
3456         }
3457         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3458         if (total_errors > max_errors) {
3459                 btrfs_error(root->fs_info, -EIO,
3460                             "%d errors while writing supers", total_errors);
3461                 return -EIO;
3462         }
3463         return 0;
3464 }
3465
3466 int write_ctree_super(struct btrfs_trans_handle *trans,
3467                       struct btrfs_root *root, int max_mirrors)
3468 {
3469         return write_all_supers(root, max_mirrors);
3470 }
3471
3472 /* Drop a fs root from the radix tree and free it. */
3473 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3474                                   struct btrfs_root *root)
3475 {
3476         spin_lock(&fs_info->fs_roots_radix_lock);
3477         radix_tree_delete(&fs_info->fs_roots_radix,
3478                           (unsigned long)root->root_key.objectid);
3479         spin_unlock(&fs_info->fs_roots_radix_lock);
3480
3481         if (btrfs_root_refs(&root->root_item) == 0)
3482                 synchronize_srcu(&fs_info->subvol_srcu);
3483
3484         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3485                 btrfs_free_log(NULL, root);
3486
3487         __btrfs_remove_free_space_cache(root->free_ino_pinned);
3488         __btrfs_remove_free_space_cache(root->free_ino_ctl);
3489         free_fs_root(root);
3490 }
3491
3492 static void free_fs_root(struct btrfs_root *root)
3493 {
3494         iput(root->cache_inode);
3495         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3496         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3497         root->orphan_block_rsv = NULL;
3498         if (root->anon_dev)
3499                 free_anon_bdev(root->anon_dev);
3500         if (root->subv_writers)
3501                 btrfs_free_subvolume_writers(root->subv_writers);
3502         free_extent_buffer(root->node);
3503         free_extent_buffer(root->commit_root);
3504         kfree(root->free_ino_ctl);
3505         kfree(root->free_ino_pinned);
3506         kfree(root->name);
3507         btrfs_put_fs_root(root);
3508 }
3509
3510 void btrfs_free_fs_root(struct btrfs_root *root)
3511 {
3512         free_fs_root(root);
3513 }
3514
3515 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3516 {
3517         u64 root_objectid = 0;
3518         struct btrfs_root *gang[8];
3519         int i;
3520         int ret;
3521
3522         while (1) {
3523                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3524                                              (void **)gang, root_objectid,
3525                                              ARRAY_SIZE(gang));
3526                 if (!ret)
3527                         break;
3528
3529                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3530                 for (i = 0; i < ret; i++) {
3531                         int err;
3532
3533                         root_objectid = gang[i]->root_key.objectid;
3534                         err = btrfs_orphan_cleanup(gang[i]);
3535                         if (err)
3536                                 return err;
3537                 }
3538                 root_objectid++;
3539         }
3540         return 0;
3541 }
3542
3543 int btrfs_commit_super(struct btrfs_root *root)
3544 {
3545         struct btrfs_trans_handle *trans;
3546
3547         mutex_lock(&root->fs_info->cleaner_mutex);
3548         btrfs_run_delayed_iputs(root);
3549         mutex_unlock(&root->fs_info->cleaner_mutex);
3550         wake_up_process(root->fs_info->cleaner_kthread);
3551
3552         /* wait until ongoing cleanup work done */
3553         down_write(&root->fs_info->cleanup_work_sem);
3554         up_write(&root->fs_info->cleanup_work_sem);
3555
3556         trans = btrfs_join_transaction(root);
3557         if (IS_ERR(trans))
3558                 return PTR_ERR(trans);
3559         return btrfs_commit_transaction(trans, root);
3560 }
3561
3562 int close_ctree(struct btrfs_root *root)
3563 {
3564         struct btrfs_fs_info *fs_info = root->fs_info;
3565         int ret;
3566
3567         fs_info->closing = 1;
3568         smp_mb();
3569
3570         /* wait for the uuid_scan task to finish */
3571         down(&fs_info->uuid_tree_rescan_sem);
3572         /* avoid complains from lockdep et al., set sem back to initial state */
3573         up(&fs_info->uuid_tree_rescan_sem);
3574
3575         /* pause restriper - we want to resume on mount */
3576         btrfs_pause_balance(fs_info);
3577
3578         btrfs_dev_replace_suspend_for_unmount(fs_info);
3579
3580         btrfs_scrub_cancel(fs_info);
3581
3582         /* wait for any defraggers to finish */
3583         wait_event(fs_info->transaction_wait,
3584                    (atomic_read(&fs_info->defrag_running) == 0));
3585
3586         /* clear out the rbtree of defraggable inodes */
3587         btrfs_cleanup_defrag_inodes(fs_info);
3588
3589         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3590                 ret = btrfs_commit_super(root);
3591                 if (ret)
3592                         btrfs_err(root->fs_info, "commit super ret %d", ret);
3593         }
3594
3595         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3596                 btrfs_error_commit_super(root);
3597
3598         kthread_stop(fs_info->transaction_kthread);
3599         kthread_stop(fs_info->cleaner_kthread);
3600
3601         fs_info->closing = 2;
3602         smp_mb();
3603
3604         btrfs_free_qgroup_config(root->fs_info);
3605
3606         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3607                 btrfs_info(root->fs_info, "at unmount delalloc count %lld",
3608                        percpu_counter_sum(&fs_info->delalloc_bytes));
3609         }
3610
3611         btrfs_sysfs_remove_one(fs_info);
3612
3613         del_fs_roots(fs_info);
3614
3615         btrfs_put_block_group_cache(fs_info);
3616
3617         btrfs_free_block_groups(fs_info);
3618
3619         btrfs_stop_all_workers(fs_info);
3620
3621         free_root_pointers(fs_info, 1);
3622
3623         iput(fs_info->btree_inode);
3624
3625 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3626         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3627                 btrfsic_unmount(root, fs_info->fs_devices);
3628 #endif
3629
3630         btrfs_close_devices(fs_info->fs_devices);
3631         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3632
3633         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3634         percpu_counter_destroy(&fs_info->delalloc_bytes);
3635         percpu_counter_destroy(&fs_info->bio_counter);
3636         bdi_destroy(&fs_info->bdi);
3637         cleanup_srcu_struct(&fs_info->subvol_srcu);
3638
3639         btrfs_free_stripe_hash_table(fs_info);
3640
3641         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3642         root->orphan_block_rsv = NULL;
3643
3644         return 0;
3645 }
3646
3647 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3648                           int atomic)
3649 {
3650         int ret;
3651         struct inode *btree_inode = buf->pages[0]->mapping->host;
3652
3653         ret = extent_buffer_uptodate(buf);
3654         if (!ret)
3655                 return ret;
3656
3657         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3658                                     parent_transid, atomic);
3659         if (ret == -EAGAIN)
3660                 return ret;
3661         return !ret;
3662 }
3663
3664 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3665 {
3666         return set_extent_buffer_uptodate(buf);
3667 }
3668
3669 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3670 {
3671         struct btrfs_root *root;
3672         u64 transid = btrfs_header_generation(buf);
3673         int was_dirty;
3674
3675 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3676         /*
3677          * This is a fast path so only do this check if we have sanity tests
3678          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3679          * outside of the sanity tests.
3680          */
3681         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3682                 return;
3683 #endif
3684         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3685         btrfs_assert_tree_locked(buf);
3686         if (transid != root->fs_info->generation)
3687                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3688                        "found %llu running %llu\n",
3689                         buf->start, transid, root->fs_info->generation);
3690         was_dirty = set_extent_buffer_dirty(buf);
3691         if (!was_dirty)
3692                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3693                                      buf->len,
3694                                      root->fs_info->dirty_metadata_batch);
3695 }
3696
3697 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3698                                         int flush_delayed)
3699 {
3700         /*
3701          * looks as though older kernels can get into trouble with
3702          * this code, they end up stuck in balance_dirty_pages forever
3703          */
3704         int ret;
3705
3706         if (current->flags & PF_MEMALLOC)
3707                 return;
3708
3709         if (flush_delayed)
3710                 btrfs_balance_delayed_items(root);
3711
3712         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3713                                      BTRFS_DIRTY_METADATA_THRESH);
3714         if (ret > 0) {
3715                 balance_dirty_pages_ratelimited(
3716                                    root->fs_info->btree_inode->i_mapping);
3717         }
3718         return;
3719 }
3720
3721 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3722 {
3723         __btrfs_btree_balance_dirty(root, 1);
3724 }
3725
3726 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3727 {
3728         __btrfs_btree_balance_dirty(root, 0);
3729 }
3730
3731 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3732 {
3733         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3734         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3735 }
3736
3737 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3738                               int read_only)
3739 {
3740         /*
3741          * Placeholder for checks
3742          */
3743         return 0;
3744 }
3745
3746 static void btrfs_error_commit_super(struct btrfs_root *root)
3747 {
3748         mutex_lock(&root->fs_info->cleaner_mutex);
3749         btrfs_run_delayed_iputs(root);
3750         mutex_unlock(&root->fs_info->cleaner_mutex);
3751
3752         down_write(&root->fs_info->cleanup_work_sem);
3753         up_write(&root->fs_info->cleanup_work_sem);
3754
3755         /* cleanup FS via transaction */
3756         btrfs_cleanup_transaction(root);
3757 }
3758
3759 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3760                                              struct btrfs_root *root)
3761 {
3762         struct btrfs_inode *btrfs_inode;
3763         struct list_head splice;
3764
3765         INIT_LIST_HEAD(&splice);
3766
3767         mutex_lock(&root->fs_info->ordered_operations_mutex);
3768         spin_lock(&root->fs_info->ordered_root_lock);
3769
3770         list_splice_init(&t->ordered_operations, &splice);
3771         while (!list_empty(&splice)) {
3772                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3773                                          ordered_operations);
3774
3775                 list_del_init(&btrfs_inode->ordered_operations);
3776                 spin_unlock(&root->fs_info->ordered_root_lock);
3777
3778                 btrfs_invalidate_inodes(btrfs_inode->root);
3779
3780                 spin_lock(&root->fs_info->ordered_root_lock);
3781         }
3782
3783         spin_unlock(&root->fs_info->ordered_root_lock);
3784         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3785 }
3786
3787 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3788 {
3789         struct btrfs_ordered_extent *ordered;
3790
3791         spin_lock(&root->ordered_extent_lock);
3792         /*
3793          * This will just short circuit the ordered completion stuff which will
3794          * make sure the ordered extent gets properly cleaned up.
3795          */
3796         list_for_each_entry(ordered, &root->ordered_extents,
3797                             root_extent_list)
3798                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3799         spin_unlock(&root->ordered_extent_lock);
3800 }
3801
3802 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3803 {
3804         struct btrfs_root *root;
3805         struct list_head splice;
3806
3807         INIT_LIST_HEAD(&splice);
3808
3809         spin_lock(&fs_info->ordered_root_lock);
3810         list_splice_init(&fs_info->ordered_roots, &splice);
3811         while (!list_empty(&splice)) {
3812                 root = list_first_entry(&splice, struct btrfs_root,
3813                                         ordered_root);
3814                 list_move_tail(&root->ordered_root,
3815                                &fs_info->ordered_roots);
3816
3817                 spin_unlock(&fs_info->ordered_root_lock);
3818                 btrfs_destroy_ordered_extents(root);
3819
3820                 cond_resched();
3821                 spin_lock(&fs_info->ordered_root_lock);
3822         }
3823         spin_unlock(&fs_info->ordered_root_lock);
3824 }
3825
3826 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3827                                       struct btrfs_root *root)
3828 {
3829         struct rb_node *node;
3830         struct btrfs_delayed_ref_root *delayed_refs;
3831         struct btrfs_delayed_ref_node *ref;
3832         int ret = 0;
3833
3834         delayed_refs = &trans->delayed_refs;
3835
3836         spin_lock(&delayed_refs->lock);
3837         if (atomic_read(&delayed_refs->num_entries) == 0) {
3838                 spin_unlock(&delayed_refs->lock);
3839                 btrfs_info(root->fs_info, "delayed_refs has NO entry");
3840                 return ret;
3841         }
3842
3843         while ((node = rb_first(&delayed_refs->href_root)) != NULL) {
3844                 struct btrfs_delayed_ref_head *head;
3845                 bool pin_bytes = false;
3846
3847                 head = rb_entry(node, struct btrfs_delayed_ref_head,
3848                                 href_node);
3849                 if (!mutex_trylock(&head->mutex)) {
3850                         atomic_inc(&head->node.refs);
3851                         spin_unlock(&delayed_refs->lock);
3852
3853                         mutex_lock(&head->mutex);
3854                         mutex_unlock(&head->mutex);
3855                         btrfs_put_delayed_ref(&head->node);
3856                         spin_lock(&delayed_refs->lock);
3857                         continue;
3858                 }
3859                 spin_lock(&head->lock);
3860                 while ((node = rb_first(&head->ref_root)) != NULL) {
3861                         ref = rb_entry(node, struct btrfs_delayed_ref_node,
3862                                        rb_node);
3863                         ref->in_tree = 0;
3864                         rb_erase(&ref->rb_node, &head->ref_root);
3865                         atomic_dec(&delayed_refs->num_entries);
3866                         btrfs_put_delayed_ref(ref);
3867                 }
3868                 if (head->must_insert_reserved)
3869                         pin_bytes = true;
3870                 btrfs_free_delayed_extent_op(head->extent_op);
3871                 delayed_refs->num_heads--;
3872                 if (head->processing == 0)
3873                         delayed_refs->num_heads_ready--;
3874                 atomic_dec(&delayed_refs->num_entries);
3875                 head->node.in_tree = 0;
3876                 rb_erase(&head->href_node, &delayed_refs->href_root);
3877                 spin_unlock(&head->lock);
3878                 spin_unlock(&delayed_refs->lock);
3879                 mutex_unlock(&head->mutex);
3880
3881                 if (pin_bytes)
3882                         btrfs_pin_extent(root, head->node.bytenr,
3883                                          head->node.num_bytes, 1);
3884                 btrfs_put_delayed_ref(&head->node);
3885                 cond_resched();
3886                 spin_lock(&delayed_refs->lock);
3887         }
3888
3889         spin_unlock(&delayed_refs->lock);
3890
3891         return ret;
3892 }
3893
3894 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3895 {
3896         struct btrfs_inode *btrfs_inode;
3897         struct list_head splice;
3898
3899         INIT_LIST_HEAD(&splice);
3900
3901         spin_lock(&root->delalloc_lock);
3902         list_splice_init(&root->delalloc_inodes, &splice);
3903
3904         while (!list_empty(&splice)) {
3905                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3906                                                delalloc_inodes);
3907
3908                 list_del_init(&btrfs_inode->delalloc_inodes);
3909                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3910                           &btrfs_inode->runtime_flags);
3911                 spin_unlock(&root->delalloc_lock);
3912
3913                 btrfs_invalidate_inodes(btrfs_inode->root);
3914
3915                 spin_lock(&root->delalloc_lock);
3916         }
3917
3918         spin_unlock(&root->delalloc_lock);
3919 }
3920
3921 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3922 {
3923         struct btrfs_root *root;
3924         struct list_head splice;
3925
3926         INIT_LIST_HEAD(&splice);
3927
3928         spin_lock(&fs_info->delalloc_root_lock);
3929         list_splice_init(&fs_info->delalloc_roots, &splice);
3930         while (!list_empty(&splice)) {
3931                 root = list_first_entry(&splice, struct btrfs_root,
3932                                          delalloc_root);
3933                 list_del_init(&root->delalloc_root);
3934                 root = btrfs_grab_fs_root(root);
3935                 BUG_ON(!root);
3936                 spin_unlock(&fs_info->delalloc_root_lock);
3937
3938                 btrfs_destroy_delalloc_inodes(root);
3939                 btrfs_put_fs_root(root);
3940
3941                 spin_lock(&fs_info->delalloc_root_lock);
3942         }
3943         spin_unlock(&fs_info->delalloc_root_lock);
3944 }
3945
3946 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3947                                         struct extent_io_tree *dirty_pages,
3948                                         int mark)
3949 {
3950         int ret;
3951         struct extent_buffer *eb;
3952         u64 start = 0;
3953         u64 end;
3954
3955         while (1) {
3956                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3957                                             mark, NULL);
3958                 if (ret)
3959                         break;
3960
3961                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3962                 while (start <= end) {
3963                         eb = btrfs_find_tree_block(root, start,
3964                                                    root->leafsize);
3965                         start += root->leafsize;
3966                         if (!eb)
3967                                 continue;
3968                         wait_on_extent_buffer_writeback(eb);
3969
3970                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3971                                                &eb->bflags))
3972                                 clear_extent_buffer_dirty(eb);
3973                         free_extent_buffer_stale(eb);
3974                 }
3975         }
3976
3977         return ret;
3978 }
3979
3980 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3981                                        struct extent_io_tree *pinned_extents)
3982 {
3983         struct extent_io_tree *unpin;
3984         u64 start;
3985         u64 end;
3986         int ret;
3987         bool loop = true;
3988
3989         unpin = pinned_extents;
3990 again:
3991         while (1) {
3992                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3993                                             EXTENT_DIRTY, NULL);
3994                 if (ret)
3995                         break;
3996
3997                 /* opt_discard */
3998                 if (btrfs_test_opt(root, DISCARD))
3999                         ret = btrfs_error_discard_extent(root, start,
4000                                                          end + 1 - start,
4001                                                          NULL);
4002
4003                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
4004                 btrfs_error_unpin_extent_range(root, start, end);
4005                 cond_resched();
4006         }
4007
4008         if (loop) {
4009                 if (unpin == &root->fs_info->freed_extents[0])
4010                         unpin = &root->fs_info->freed_extents[1];
4011                 else
4012                         unpin = &root->fs_info->freed_extents[0];
4013                 loop = false;
4014                 goto again;
4015         }
4016
4017         return 0;
4018 }
4019
4020 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4021                                    struct btrfs_root *root)
4022 {
4023         btrfs_destroy_ordered_operations(cur_trans, root);
4024
4025         btrfs_destroy_delayed_refs(cur_trans, root);
4026
4027         cur_trans->state = TRANS_STATE_COMMIT_START;
4028         wake_up(&root->fs_info->transaction_blocked_wait);
4029
4030         cur_trans->state = TRANS_STATE_UNBLOCKED;
4031         wake_up(&root->fs_info->transaction_wait);
4032
4033         btrfs_destroy_delayed_inodes(root);
4034         btrfs_assert_delayed_root_empty(root);
4035
4036         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4037                                      EXTENT_DIRTY);
4038         btrfs_destroy_pinned_extent(root,
4039                                     root->fs_info->pinned_extents);
4040
4041         cur_trans->state =TRANS_STATE_COMPLETED;
4042         wake_up(&cur_trans->commit_wait);
4043
4044         /*
4045         memset(cur_trans, 0, sizeof(*cur_trans));
4046         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4047         */
4048 }
4049
4050 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4051 {
4052         struct btrfs_transaction *t;
4053
4054         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4055
4056         spin_lock(&root->fs_info->trans_lock);
4057         while (!list_empty(&root->fs_info->trans_list)) {
4058                 t = list_first_entry(&root->fs_info->trans_list,
4059                                      struct btrfs_transaction, list);
4060                 if (t->state >= TRANS_STATE_COMMIT_START) {
4061                         atomic_inc(&t->use_count);
4062                         spin_unlock(&root->fs_info->trans_lock);
4063                         btrfs_wait_for_commit(root, t->transid);
4064                         btrfs_put_transaction(t);
4065                         spin_lock(&root->fs_info->trans_lock);
4066                         continue;
4067                 }
4068                 if (t == root->fs_info->running_transaction) {
4069                         t->state = TRANS_STATE_COMMIT_DOING;
4070                         spin_unlock(&root->fs_info->trans_lock);
4071                         /*
4072                          * We wait for 0 num_writers since we don't hold a trans
4073                          * handle open currently for this transaction.
4074                          */
4075                         wait_event(t->writer_wait,
4076                                    atomic_read(&t->num_writers) == 0);
4077                 } else {
4078                         spin_unlock(&root->fs_info->trans_lock);
4079                 }
4080                 btrfs_cleanup_one_transaction(t, root);
4081
4082                 spin_lock(&root->fs_info->trans_lock);
4083                 if (t == root->fs_info->running_transaction)
4084                         root->fs_info->running_transaction = NULL;
4085                 list_del_init(&t->list);
4086                 spin_unlock(&root->fs_info->trans_lock);
4087
4088                 btrfs_put_transaction(t);
4089                 trace_btrfs_transaction_commit(root);
4090                 spin_lock(&root->fs_info->trans_lock);
4091         }
4092         spin_unlock(&root->fs_info->trans_lock);
4093         btrfs_destroy_all_ordered_extents(root->fs_info);
4094         btrfs_destroy_delayed_inodes(root);
4095         btrfs_assert_delayed_root_empty(root);
4096         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4097         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4098         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4099
4100         return 0;
4101 }
4102
4103 static struct extent_io_ops btree_extent_io_ops = {
4104         .readpage_end_io_hook = btree_readpage_end_io_hook,
4105         .readpage_io_failed_hook = btree_io_failed_hook,
4106         .submit_bio_hook = btree_submit_bio_hook,
4107         /* note we're sharing with inode.c for the merge bio hook */
4108         .merge_bio_hook = btrfs_merge_bio_hook,
4109 };