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/crc32c.h>
  30 #include <linux/slab.h>
  31 #include <linux/migrate.h>
  32 #include <linux/ratelimit.h>
  33 #include <linux/uuid.h>
  34 #include <linux/semaphore.h>
  35 #include <asm/unaligned.h>
  36 #include "ctree.h"
  37 #include "disk-io.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 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 "btrfs: %s checksum verify "
 304                                        "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 "btrfs: super block crcs don't match, older mkfs detected\n");
 387                         ret = 0;
 388                 }
 389         }
 390
 391         if (csum_type >= ARRAY_SIZE(btrfs_csum_sizes)) {
 392                 printk(KERN_ERR "btrfs: unsupported checksum algorithm %u\n",
 393                                 csum_type);
 394                 ret = 1;
 395         }
 396
 397         return ret;
 398 }
 399
 400 /*
 401  * helper to read a given tree block, doing retries as required when
 402  * the checksums don't match and we have alternate mirrors to try.
 403  */
 404 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 405                                           struct extent_buffer *eb,
 406                                           u64 start, u64 parent_transid)
 407 {
 408         struct extent_io_tree *io_tree;
 409         int failed = 0;
 410         int ret;
 411         int num_copies = 0;
 412         int mirror_num = 0;
 413         int failed_mirror = 0;
 414
 415         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 416         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 417         while (1) {
 418                 ret = read_extent_buffer_pages(io_tree, eb, start,
 419                                                WAIT_COMPLETE,
 420                                                btree_get_extent, mirror_num);
 421                 if (!ret) {
 422                         if (!verify_parent_transid(io_tree, eb,
 423                                                    parent_transid, 0))
 424                                 break;
 425                         else
 426                                 ret = -EIO;
 427                 }
 428
 429                 /*
 430                  * This buffer's crc is fine, but its contents are corrupted, so
 431                  * there is no reason to read the other copies, they won't be
 432                  * any less wrong.
 433                  */
 434                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 435                         break;
 436
 437                 num_copies = btrfs_num_copies(root->fs_info,
 438                                               eb->start, eb->len);
 439                 if (num_copies == 1)
 440                         break;
 441
 442                 if (!failed_mirror) {
 443                         failed = 1;
 444                         failed_mirror = eb->read_mirror;
 445                 }
 446
 447                 mirror_num++;
 448                 if (mirror_num == failed_mirror)
 449                         mirror_num++;
 450
 451                 if (mirror_num > num_copies)
 452                         break;
 453         }
 454
 455         if (failed && !ret && failed_mirror)
 456                 repair_eb_io_failure(root, eb, failed_mirror);
 457
 458         return ret;
 459 }
 460
 461 /*
 462  * checksum a dirty tree block before IO.  This has extra checks to make sure
 463  * we only fill in the checksum field in the first page of a multi-page block
 464  */
 465
 466 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 467 {
 468         struct extent_io_tree *tree;
 469         u64 start = page_offset(page);
 470         u64 found_start;
 471         struct extent_buffer *eb;
 472
 473         tree = &BTRFS_I(page->mapping->host)->io_tree;
 474
 475         eb = (struct extent_buffer *)page->private;
 476         if (page != eb->pages[0])
 477                 return 0;
 478         found_start = btrfs_header_bytenr(eb);
 479         if (WARN_ON(found_start != start || !PageUptodate(page)))
 480                 return 0;
 481         csum_tree_block(root, eb, 0);
 482         return 0;
 483 }
 484
 485 static int check_tree_block_fsid(struct btrfs_root *root,
 486                                  struct extent_buffer *eb)
 487 {
 488         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 489         u8 fsid[BTRFS_UUID_SIZE];
 490         int ret = 1;
 491
 492         read_extent_buffer(eb, fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
 493         while (fs_devices) {
 494                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 495                         ret = 0;
 496                         break;
 497                 }
 498                 fs_devices = fs_devices->seed;
 499         }
 500         return ret;
 501 }
 502
 503 #define CORRUPT(reason, eb, root, slot)                         \
 504         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 505                "root=%llu, slot=%d\n", reason,                  \
 506                btrfs_header_bytenr(eb), root->objectid, slot)
 507
 508 static noinline int check_leaf(struct btrfs_root *root,
 509                                struct extent_buffer *leaf)
 510 {
 511         struct btrfs_key key;
 512         struct btrfs_key leaf_key;
 513         u32 nritems = btrfs_header_nritems(leaf);
 514         int slot;
 515
 516         if (nritems == 0)
 517                 return 0;
 518
 519         /* Check the 0 item */
 520         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 521             BTRFS_LEAF_DATA_SIZE(root)) {
 522                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 523                 return -EIO;
 524         }
 525
 526         /*
 527          * Check to make sure each items keys are in the correct order and their
 528          * offsets make sense.  We only have to loop through nritems-1 because
 529          * we check the current slot against the next slot, which verifies the
 530          * next slot's offset+size makes sense and that the current's slot
 531          * offset is correct.
 532          */
 533         for (slot = 0; slot < nritems - 1; slot++) {
 534                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 535                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 536
 537                 /* Make sure the keys are in the right order */
 538                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 539                         CORRUPT("bad key order", leaf, root, slot);
 540                         return -EIO;
 541                 }
 542
 543                 /*
 544                  * Make sure the offset and ends are right, remember that the
 545                  * item data starts at the end of the leaf and grows towards the
 546                  * front.
 547                  */
 548                 if (btrfs_item_offset_nr(leaf, slot) !=
 549                         btrfs_item_end_nr(leaf, slot + 1)) {
 550                         CORRUPT("slot offset bad", leaf, root, slot);
 551                         return -EIO;
 552                 }
 553
 554                 /*
 555                  * Check to make sure that we don't point outside of the leaf,
 556                  * just incase all the items are consistent to eachother, but
 557                  * all point outside of the leaf.
 558                  */
 559                 if (btrfs_item_end_nr(leaf, slot) >
 560                     BTRFS_LEAF_DATA_SIZE(root)) {
 561                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 562                         return -EIO;
 563                 }
 564         }
 565
 566         return 0;
 567 }
 568
 569 static int btree_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
 570                                       u64 phy_offset, struct page *page,
 571                                       u64 start, u64 end, int mirror)
 572 {
 573         struct extent_io_tree *tree;
 574         u64 found_start;
 575         int found_level;
 576         struct extent_buffer *eb;
 577         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 578         int ret = 0;
 579         int reads_done;
 580
 581         if (!page->private)
 582                 goto out;
 583
 584         tree = &BTRFS_I(page->mapping->host)->io_tree;
 585         eb = (struct extent_buffer *)page->private;
 586
 587         /* the pending IO might have been the only thing that kept this buffer
 588          * in memory.  Make sure we have a ref for all this other checks
 589          */
 590         extent_buffer_get(eb);
 591
 592         reads_done = atomic_dec_and_test(&eb->io_pages);
 593         if (!reads_done)
 594                 goto err;
 595
 596         eb->read_mirror = mirror;
 597         if (test_bit(EXTENT_BUFFER_IOERR, &eb->bflags)) {
 598                 ret = -EIO;
 599                 goto err;
 600         }
 601
 602         found_start = btrfs_header_bytenr(eb);
 603         if (found_start != eb->start) {
 604                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
 605                                "%llu %llu\n",
 606                                found_start, eb->start);
 607                 ret = -EIO;
 608                 goto err;
 609         }
 610         if (check_tree_block_fsid(root, eb)) {
 611                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
 612                                eb->start);
 613                 ret = -EIO;
 614                 goto err;
 615         }
 616         found_level = btrfs_header_level(eb);
 617         if (found_level >= BTRFS_MAX_LEVEL) {
 618                 btrfs_info(root->fs_info, "bad tree block level %d\n",
 619                            (int)btrfs_header_level(eb));
 620                 ret = -EIO;
 621                 goto err;
 622         }
 623
 624         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 625                                        eb, found_level);
 626
 627         ret = csum_tree_block(root, eb, 1);
 628         if (ret) {
 629                 ret = -EIO;
 630                 goto err;
 631         }
 632
 633         /*
 634          * If this is a leaf block and it is corrupt, set the corrupt bit so
 635          * that we don't try and read the other copies of this block, just
 636          * return -EIO.
 637          */
 638         if (found_level == 0 && check_leaf(root, eb)) {
 639                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 640                 ret = -EIO;
 641         }
 642
 643         if (!ret)
 644                 set_extent_buffer_uptodate(eb);
 645 err:
 646         if (reads_done &&
 647             test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 648                 btree_readahead_hook(root, eb, eb->start, ret);
 649
 650         if (ret) {
 651                 /*
 652                  * our io error hook is going to dec the io pages
 653                  * again, we have to make sure it has something
 654                  * to decrement
 655                  */
 656                 atomic_inc(&eb->io_pages);
 657                 clear_extent_buffer_uptodate(eb);
 658         }
 659         free_extent_buffer(eb);
 660 out:
 661         return ret;
 662 }
 663
 664 static int btree_io_failed_hook(struct page *page, int failed_mirror)
 665 {
 666         struct extent_buffer *eb;
 667         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 668
 669         eb = (struct extent_buffer *)page->private;
 670         set_bit(EXTENT_BUFFER_IOERR, &eb->bflags);
 671         eb->read_mirror = failed_mirror;
 672         atomic_dec(&eb->io_pages);
 673         if (test_and_clear_bit(EXTENT_BUFFER_READAHEAD, &eb->bflags))
 674                 btree_readahead_hook(root, eb, eb->start, -EIO);
 675         return -EIO;    /* we fixed nothing */
 676 }
 677
 678 static void end_workqueue_bio(struct bio *bio, int err)
 679 {
 680         struct end_io_wq *end_io_wq = bio->bi_private;
 681         struct btrfs_fs_info *fs_info;
 682
 683         fs_info = end_io_wq->info;
 684         end_io_wq->error = err;
 685         end_io_wq->work.func = end_workqueue_fn;
 686         end_io_wq->work.flags = 0;
 687
 688         if (bio->bi_rw & REQ_WRITE) {
 689                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_METADATA)
 690                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 691                                            &end_io_wq->work);
 692                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_FREE_SPACE)
 693                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 694                                            &end_io_wq->work);
 695                 else if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 696                         btrfs_queue_worker(&fs_info->endio_raid56_workers,
 697                                            &end_io_wq->work);
 698                 else
 699                         btrfs_queue_worker(&fs_info->endio_write_workers,
 700                                            &end_io_wq->work);
 701         } else {
 702                 if (end_io_wq->metadata == BTRFS_WQ_ENDIO_RAID56)
 703                         btrfs_queue_worker(&fs_info->endio_raid56_workers,
 704                                            &end_io_wq->work);
 705                 else if (end_io_wq->metadata)
 706                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 707                                            &end_io_wq->work);
 708                 else
 709                         btrfs_queue_worker(&fs_info->endio_workers,
 710                                            &end_io_wq->work);
 711         }
 712 }
 713
 714 /*
 715  * For the metadata arg you want
 716  *
 717  * 0 - if data
 718  * 1 - if normal metadta
 719  * 2 - if writing to the free space cache area
 720  * 3 - raid parity work
 721  */
 722 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 723                         int metadata)
 724 {
 725         struct end_io_wq *end_io_wq;
 726         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 727         if (!end_io_wq)
 728                 return -ENOMEM;
 729
 730         end_io_wq->private = bio->bi_private;
 731         end_io_wq->end_io = bio->bi_end_io;
 732         end_io_wq->info = info;
 733         end_io_wq->error = 0;
 734         end_io_wq->bio = bio;
 735         end_io_wq->metadata = metadata;
 736
 737         bio->bi_private = end_io_wq;
 738         bio->bi_end_io = end_workqueue_bio;
 739         return 0;
 740 }
 741
 742 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 743 {
 744         unsigned long limit = min_t(unsigned long,
 745                                     info->workers.max_workers,
 746                                     info->fs_devices->open_devices);
 747         return 256 * limit;
 748 }
 749
 750 static void run_one_async_start(struct btrfs_work *work)
 751 {
 752         struct async_submit_bio *async;
 753         int ret;
 754
 755         async = container_of(work, struct  async_submit_bio, work);
 756         ret = async->submit_bio_start(async->inode, async->rw, async->bio,
 757                                       async->mirror_num, async->bio_flags,
 758                                       async->bio_offset);
 759         if (ret)
 760                 async->error = ret;
 761 }
 762
 763 static void run_one_async_done(struct btrfs_work *work)
 764 {
 765         struct btrfs_fs_info *fs_info;
 766         struct async_submit_bio *async;
 767         int limit;
 768
 769         async = container_of(work, struct  async_submit_bio, work);
 770         fs_info = BTRFS_I(async->inode)->root->fs_info;
 771
 772         limit = btrfs_async_submit_limit(fs_info);
 773         limit = limit * 2 / 3;
 774
 775         if (atomic_dec_return(&fs_info->nr_async_submits) < limit &&
 776             waitqueue_active(&fs_info->async_submit_wait))
 777                 wake_up(&fs_info->async_submit_wait);
 778
 779         /* If an error occured we just want to clean up the bio and move on */
 780         if (async->error) {
 781                 bio_endio(async->bio, async->error);
 782                 return;
 783         }
 784
 785         async->submit_bio_done(async->inode, async->rw, async->bio,
 786                                async->mirror_num, async->bio_flags,
 787                                async->bio_offset);
 788 }
 789
 790 static void run_one_async_free(struct btrfs_work *work)
 791 {
 792         struct async_submit_bio *async;
 793
 794         async = container_of(work, struct  async_submit_bio, work);
 795         kfree(async);
 796 }
 797
 798 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 799                         int rw, struct bio *bio, int mirror_num,
 800                         unsigned long bio_flags,
 801                         u64 bio_offset,
 802                         extent_submit_bio_hook_t *submit_bio_start,
 803                         extent_submit_bio_hook_t *submit_bio_done)
 804 {
 805         struct async_submit_bio *async;
 806
 807         async = kmalloc(sizeof(*async), GFP_NOFS);
 808         if (!async)
 809                 return -ENOMEM;
 810
 811         async->inode = inode;
 812         async->rw = rw;
 813         async->bio = bio;
 814         async->mirror_num = mirror_num;
 815         async->submit_bio_start = submit_bio_start;
 816         async->submit_bio_done = submit_bio_done;
 817
 818         async->work.func = run_one_async_start;
 819         async->work.ordered_func = run_one_async_done;
 820         async->work.ordered_free = run_one_async_free;
 821
 822         async->work.flags = 0;
 823         async->bio_flags = bio_flags;
 824         async->bio_offset = bio_offset;
 825
 826         async->error = 0;
 827
 828         atomic_inc(&fs_info->nr_async_submits);
 829
 830         if (rw & REQ_SYNC)
 831                 btrfs_set_work_high_prio(&async->work);
 832
 833         btrfs_queue_worker(&fs_info->workers, &async->work);
 834
 835         while (atomic_read(&fs_info->async_submit_draining) &&
 836               atomic_read(&fs_info->nr_async_submits)) {
 837                 wait_event(fs_info->async_submit_wait,
 838                            (atomic_read(&fs_info->nr_async_submits) == 0));
 839         }
 840
 841         return 0;
 842 }
 843
 844 static int btree_csum_one_bio(struct bio *bio)
 845 {
 846         struct bio_vec *bvec = bio->bi_io_vec;
 847         int bio_index = 0;
 848         struct btrfs_root *root;
 849         int ret = 0;
 850
 851         WARN_ON(bio->bi_vcnt <= 0);
 852         while (bio_index < bio->bi_vcnt) {
 853                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 854                 ret = csum_dirty_buffer(root, bvec->bv_page);
 855                 if (ret)
 856                         break;
 857                 bio_index++;
 858                 bvec++;
 859         }
 860         return ret;
 861 }
 862
 863 static int __btree_submit_bio_start(struct inode *inode, int rw,
 864                                     struct bio *bio, int mirror_num,
 865                                     unsigned long bio_flags,
 866                                     u64 bio_offset)
 867 {
 868         /*
 869          * when we're called for a write, we're already in the async
 870          * submission context.  Just jump into btrfs_map_bio
 871          */
 872         return btree_csum_one_bio(bio);
 873 }
 874
 875 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 876                                  int mirror_num, unsigned long bio_flags,
 877                                  u64 bio_offset)
 878 {
 879         int ret;
 880
 881         /*
 882          * when we're called for a write, we're already in the async
 883          * submission context.  Just jump into btrfs_map_bio
 884          */
 885         ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 886         if (ret)
 887                 bio_endio(bio, ret);
 888         return ret;
 889 }
 890
 891 static int check_async_write(struct inode *inode, unsigned long bio_flags)
 892 {
 893         if (bio_flags & EXTENT_BIO_TREE_LOG)
 894                 return 0;
 895 #ifdef CONFIG_X86
 896         if (cpu_has_xmm4_2)
 897                 return 0;
 898 #endif
 899         return 1;
 900 }
 901
 902 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 903                                  int mirror_num, unsigned long bio_flags,
 904                                  u64 bio_offset)
 905 {
 906         int async = check_async_write(inode, bio_flags);
 907         int ret;
 908
 909         if (!(rw & REQ_WRITE)) {
 910                 /*
 911                  * called for a read, do the setup so that checksum validation
 912                  * can happen in the async kernel threads
 913                  */
 914                 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 915                                           bio, 1);
 916                 if (ret)
 917                         goto out_w_error;
 918                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 919                                     mirror_num, 0);
 920         } else if (!async) {
 921                 ret = btree_csum_one_bio(bio);
 922                 if (ret)
 923                         goto out_w_error;
 924                 ret = btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 925                                     mirror_num, 0);
 926         } else {
 927                 /*
 928                  * kthread helpers are used to submit writes so that
 929                  * checksumming can happen in parallel across all CPUs
 930                  */
 931                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 932                                           inode, rw, bio, mirror_num, 0,
 933                                           bio_offset,
 934                                           __btree_submit_bio_start,
 935                                           __btree_submit_bio_done);
 936         }
 937
 938         if (ret) {
 939 out_w_error:
 940                 bio_endio(bio, ret);
 941         }
 942         return ret;
 943 }
 944
 945 #ifdef CONFIG_MIGRATION
 946 static int btree_migratepage(struct address_space *mapping,
 947                         struct page *newpage, struct page *page,
 948                         enum migrate_mode mode)
 949 {
 950         /*
 951          * we can't safely write a btree page from here,
 952          * we haven't done the locking hook
 953          */
 954         if (PageDirty(page))
 955                 return -EAGAIN;
 956         /*
 957          * Buffers may be managed in a filesystem specific way.
 958          * We must have no buffers or drop them.
 959          */
 960         if (page_has_private(page) &&
 961             !try_to_release_page(page, GFP_KERNEL))
 962                 return -EAGAIN;
 963         return migrate_page(mapping, newpage, page, mode);
 964 }
 965 #endif
 966
 967
 968 static int btree_writepages(struct address_space *mapping,
 969                             struct writeback_control *wbc)
 970 {
 971         struct extent_io_tree *tree;
 972         struct btrfs_fs_info *fs_info;
 973         int ret;
 974
 975         tree = &BTRFS_I(mapping->host)->io_tree;
 976         if (wbc->sync_mode == WB_SYNC_NONE) {
 977
 978                 if (wbc->for_kupdate)
 979                         return 0;
 980
 981                 fs_info = BTRFS_I(mapping->host)->root->fs_info;
 982                 /* this is a bit racy, but that's ok */
 983                 ret = percpu_counter_compare(&fs_info->dirty_metadata_bytes,
 984                                              BTRFS_DIRTY_METADATA_THRESH);
 985                 if (ret < 0)
 986                         return 0;
 987         }
 988         return btree_write_cache_pages(mapping, wbc);
 989 }
 990
 991 static int btree_readpage(struct file *file, struct page *page)
 992 {
 993         struct extent_io_tree *tree;
 994         tree = &BTRFS_I(page->mapping->host)->io_tree;
 995         return extent_read_full_page(tree, page, btree_get_extent, 0);
 996 }
 997
 998 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 999 {
1000         if (PageWriteback(page) || PageDirty(page))
1001                 return 0;
1002
1003         return try_release_extent_buffer(page);
1004 }
1005
1006 static void btree_invalidatepage(struct page *page, unsigned int offset,
1007                                  unsigned int length)
1008 {
1009         struct extent_io_tree *tree;
1010         tree = &BTRFS_I(page->mapping->host)->io_tree;
1011         extent_invalidatepage(tree, page, offset);
1012         btree_releasepage(page, GFP_NOFS);
1013         if (PagePrivate(page)) {
1014                 printk(KERN_WARNING "btrfs warning page private not zero "
1015                        "on page %llu\n", (unsigned long long)page_offset(page));
1016                 ClearPagePrivate(page);
1017                 set_page_private(page, 0);
1018                 page_cache_release(page);
1019         }
1020 }
1021
1022 static int btree_set_page_dirty(struct page *page)
1023 {
1024 #ifdef DEBUG
1025         struct extent_buffer *eb;
1026
1027         BUG_ON(!PagePrivate(page));
1028         eb = (struct extent_buffer *)page->private;
1029         BUG_ON(!eb);
1030         BUG_ON(!test_bit(EXTENT_BUFFER_DIRTY, &eb->bflags));
1031         BUG_ON(!atomic_read(&eb->refs));
1032         btrfs_assert_tree_locked(eb);
1033 #endif
1034         return __set_page_dirty_nobuffers(page);
1035 }
1036
1037 static const struct address_space_operations btree_aops = {
1038         .readpage       = btree_readpage,
1039         .writepages     = btree_writepages,
1040         .releasepage    = btree_releasepage,
1041         .invalidatepage = btree_invalidatepage,
1042 #ifdef CONFIG_MIGRATION
1043         .migratepage    = btree_migratepage,
1044 #endif
1045         .set_page_dirty = btree_set_page_dirty,
1046 };
1047
1048 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1049                          u64 parent_transid)
1050 {
1051         struct extent_buffer *buf = NULL;
1052         struct inode *btree_inode = root->fs_info->btree_inode;
1053         int ret = 0;
1054
1055         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1056         if (!buf)
1057                 return 0;
1058         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
1059                                  buf, 0, WAIT_NONE, btree_get_extent, 0);
1060         free_extent_buffer(buf);
1061         return ret;
1062 }
1063
1064 int reada_tree_block_flagged(struct btrfs_root *root, u64 bytenr, u32 blocksize,
1065                          int mirror_num, struct extent_buffer **eb)
1066 {
1067         struct extent_buffer *buf = NULL;
1068         struct inode *btree_inode = root->fs_info->btree_inode;
1069         struct extent_io_tree *io_tree = &BTRFS_I(btree_inode)->io_tree;
1070         int ret;
1071
1072         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1073         if (!buf)
1074                 return 0;
1075
1076         set_bit(EXTENT_BUFFER_READAHEAD, &buf->bflags);
1077
1078         ret = read_extent_buffer_pages(io_tree, buf, 0, WAIT_PAGE_LOCK,
1079                                        btree_get_extent, mirror_num);
1080         if (ret) {
1081                 free_extent_buffer(buf);
1082                 return ret;
1083         }
1084
1085         if (test_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags)) {
1086                 free_extent_buffer(buf);
1087                 return -EIO;
1088         } else if (extent_buffer_uptodate(buf)) {
1089                 *eb = buf;
1090         } else {
1091                 free_extent_buffer(buf);
1092         }
1093         return 0;
1094 }
1095
1096 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
1097                                             u64 bytenr, u32 blocksize)
1098 {
1099         struct inode *btree_inode = root->fs_info->btree_inode;
1100         struct extent_buffer *eb;
1101         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree, bytenr);
1102         return eb;
1103 }
1104
1105 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
1106                                                  u64 bytenr, u32 blocksize)
1107 {
1108         struct inode *btree_inode = root->fs_info->btree_inode;
1109         struct extent_buffer *eb;
1110
1111         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
1112                                  bytenr, blocksize);
1113         return eb;
1114 }
1115
1116
1117 int btrfs_write_tree_block(struct extent_buffer *buf)
1118 {
1119         return filemap_fdatawrite_range(buf->pages[0]->mapping, buf->start,
1120                                         buf->start + buf->len - 1);
1121 }
1122
1123 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1124 {
1125         return filemap_fdatawait_range(buf->pages[0]->mapping,
1126                                        buf->start, buf->start + buf->len - 1);
1127 }
1128
1129 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1130                                       u32 blocksize, u64 parent_transid)
1131 {
1132         struct extent_buffer *buf = NULL;
1133         int ret;
1134
1135         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1136         if (!buf)
1137                 return NULL;
1138
1139         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1140         if (ret) {
1141                 free_extent_buffer(buf);
1142                 return NULL;
1143         }
1144         return buf;
1145
1146 }
1147
1148 void clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1149                       struct extent_buffer *buf)
1150 {
1151         struct btrfs_fs_info *fs_info = root->fs_info;
1152
1153         if (btrfs_header_generation(buf) ==
1154             fs_info->running_transaction->transid) {
1155                 btrfs_assert_tree_locked(buf);
1156
1157                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1158                         __percpu_counter_add(&fs_info->dirty_metadata_bytes,
1159                                              -buf->len,
1160                                              fs_info->dirty_metadata_batch);
1161                         /* ugh, clear_extent_buffer_dirty needs to lock the page */
1162                         btrfs_set_lock_blocking(buf);
1163                         clear_extent_buffer_dirty(buf);
1164                 }
1165         }
1166 }
1167
1168 static void __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1169                          u32 stripesize, struct btrfs_root *root,
1170                          struct btrfs_fs_info *fs_info,
1171                          u64 objectid)
1172 {
1173         root->node = NULL;
1174         root->commit_root = NULL;
1175         root->sectorsize = sectorsize;
1176         root->nodesize = nodesize;
1177         root->leafsize = leafsize;
1178         root->stripesize = stripesize;
1179         root->ref_cows = 0;
1180         root->track_dirty = 0;
1181         root->in_radix = 0;
1182         root->orphan_item_inserted = 0;
1183         root->orphan_cleanup_state = 0;
1184
1185         root->objectid = objectid;
1186         root->last_trans = 0;
1187         root->highest_objectid = 0;
1188         root->nr_delalloc_inodes = 0;
1189         root->nr_ordered_extents = 0;
1190         root->name = NULL;
1191         root->inode_tree = RB_ROOT;
1192         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1193         root->block_rsv = NULL;
1194         root->orphan_block_rsv = NULL;
1195
1196         INIT_LIST_HEAD(&root->dirty_list);
1197         INIT_LIST_HEAD(&root->root_list);
1198         INIT_LIST_HEAD(&root->delalloc_inodes);
1199         INIT_LIST_HEAD(&root->delalloc_root);
1200         INIT_LIST_HEAD(&root->ordered_extents);
1201         INIT_LIST_HEAD(&root->ordered_root);
1202         INIT_LIST_HEAD(&root->logged_list[0]);
1203         INIT_LIST_HEAD(&root->logged_list[1]);
1204         spin_lock_init(&root->orphan_lock);
1205         spin_lock_init(&root->inode_lock);
1206         spin_lock_init(&root->delalloc_lock);
1207         spin_lock_init(&root->ordered_extent_lock);
1208         spin_lock_init(&root->accounting_lock);
1209         spin_lock_init(&root->log_extents_lock[0]);
1210         spin_lock_init(&root->log_extents_lock[1]);
1211         mutex_init(&root->objectid_mutex);
1212         mutex_init(&root->log_mutex);
1213         init_waitqueue_head(&root->log_writer_wait);
1214         init_waitqueue_head(&root->log_commit_wait[0]);
1215         init_waitqueue_head(&root->log_commit_wait[1]);
1216         atomic_set(&root->log_commit[0], 0);
1217         atomic_set(&root->log_commit[1], 0);
1218         atomic_set(&root->log_writers, 0);
1219         atomic_set(&root->log_batch, 0);
1220         atomic_set(&root->orphan_inodes, 0);
1221         atomic_set(&root->refs, 1);
1222         root->log_transid = 0;
1223         root->last_log_commit = 0;
1224         if (fs_info)
1225                 extent_io_tree_init(&root->dirty_log_pages,
1226                                      fs_info->btree_inode->i_mapping);
1227
1228         memset(&root->root_key, 0, sizeof(root->root_key));
1229         memset(&root->root_item, 0, sizeof(root->root_item));
1230         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1231         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1232         if (fs_info)
1233                 root->defrag_trans_start = fs_info->generation;
1234         else
1235                 root->defrag_trans_start = 0;
1236         init_completion(&root->kobj_unregister);
1237         root->defrag_running = 0;
1238         root->root_key.objectid = objectid;
1239         root->anon_dev = 0;
1240
1241         spin_lock_init(&root->root_item_lock);
1242 }
1243
1244 static struct btrfs_root *btrfs_alloc_root(struct btrfs_fs_info *fs_info)
1245 {
1246         struct btrfs_root *root = kzalloc(sizeof(*root), GFP_NOFS);
1247         if (root)
1248                 root->fs_info = fs_info;
1249         return root;
1250 }
1251
1252 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1253 /* Should only be used by the testing infrastructure */
1254 struct btrfs_root *btrfs_alloc_dummy_root(void)
1255 {
1256         struct btrfs_root *root;
1257
1258         root = btrfs_alloc_root(NULL);
1259         if (!root)
1260                 return ERR_PTR(-ENOMEM);
1261         __setup_root(4096, 4096, 4096, 4096, root, NULL, 1);
1262         root->dummy_root = 1;
1263
1264         return root;
1265 }
1266 #endif
1267
1268 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
1269                                      struct btrfs_fs_info *fs_info,
1270                                      u64 objectid)
1271 {
1272         struct extent_buffer *leaf;
1273         struct btrfs_root *tree_root = fs_info->tree_root;
1274         struct btrfs_root *root;
1275         struct btrfs_key key;
1276         int ret = 0;
1277         u64 bytenr;
1278         uuid_le uuid;
1279
1280         root = btrfs_alloc_root(fs_info);
1281         if (!root)
1282                 return ERR_PTR(-ENOMEM);
1283
1284         __setup_root(tree_root->nodesize, tree_root->leafsize,
1285                      tree_root->sectorsize, tree_root->stripesize,
1286                      root, fs_info, objectid);
1287         root->root_key.objectid = objectid;
1288         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1289         root->root_key.offset = 0;
1290
1291         leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1292                                       0, objectid, NULL, 0, 0, 0);
1293         if (IS_ERR(leaf)) {
1294                 ret = PTR_ERR(leaf);
1295                 leaf = NULL;
1296                 goto fail;
1297         }
1298
1299         bytenr = leaf->start;
1300         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1301         btrfs_set_header_bytenr(leaf, leaf->start);
1302         btrfs_set_header_generation(leaf, trans->transid);
1303         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1304         btrfs_set_header_owner(leaf, objectid);
1305         root->node = leaf;
1306
1307         write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(),
1308                             BTRFS_FSID_SIZE);
1309         write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
1310                             btrfs_header_chunk_tree_uuid(leaf),
1311                             BTRFS_UUID_SIZE);
1312         btrfs_mark_buffer_dirty(leaf);
1313
1314         root->commit_root = btrfs_root_node(root);
1315         root->track_dirty = 1;
1316
1317
1318         root->root_item.flags = 0;
1319         root->root_item.byte_limit = 0;
1320         btrfs_set_root_bytenr(&root->root_item, leaf->start);
1321         btrfs_set_root_generation(&root->root_item, trans->transid);
1322         btrfs_set_root_level(&root->root_item, 0);
1323         btrfs_set_root_refs(&root->root_item, 1);
1324         btrfs_set_root_used(&root->root_item, leaf->len);
1325         btrfs_set_root_last_snapshot(&root->root_item, 0);
1326         btrfs_set_root_dirid(&root->root_item, 0);
1327         uuid_le_gen(&uuid);
1328         memcpy(root->root_item.uuid, uuid.b, BTRFS_UUID_SIZE);
1329         root->root_item.drop_level = 0;
1330
1331         key.objectid = objectid;
1332         key.type = BTRFS_ROOT_ITEM_KEY;
1333         key.offset = 0;
1334         ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
1335         if (ret)
1336                 goto fail;
1337
1338         btrfs_tree_unlock(leaf);
1339
1340         return root;
1341
1342 fail:
1343         if (leaf) {
1344                 btrfs_tree_unlock(leaf);
1345                 free_extent_buffer(leaf);
1346         }
1347         kfree(root);
1348
1349         return ERR_PTR(ret);
1350 }
1351
1352 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1353                                          struct btrfs_fs_info *fs_info)
1354 {
1355         struct btrfs_root *root;
1356         struct btrfs_root *tree_root = fs_info->tree_root;
1357         struct extent_buffer *leaf;
1358
1359         root = btrfs_alloc_root(fs_info);
1360         if (!root)
1361                 return ERR_PTR(-ENOMEM);
1362
1363         __setup_root(tree_root->nodesize, tree_root->leafsize,
1364                      tree_root->sectorsize, tree_root->stripesize,
1365                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1366
1367         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1368         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1369         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1370         /*
1371          * log trees do not get reference counted because they go away
1372          * before a real commit is actually done.  They do store pointers
1373          * to file data extents, and those reference counts still get
1374          * updated (along with back refs to the log tree).
1375          */
1376         root->ref_cows = 0;
1377
1378         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1379                                       BTRFS_TREE_LOG_OBJECTID, NULL,
1380                                       0, 0, 0);
1381         if (IS_ERR(leaf)) {
1382                 kfree(root);
1383                 return ERR_CAST(leaf);
1384         }
1385
1386         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1387         btrfs_set_header_bytenr(leaf, leaf->start);
1388         btrfs_set_header_generation(leaf, trans->transid);
1389         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1390         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1391         root->node = leaf;
1392
1393         write_extent_buffer(root->node, root->fs_info->fsid,
1394                             btrfs_header_fsid(), BTRFS_FSID_SIZE);
1395         btrfs_mark_buffer_dirty(root->node);
1396         btrfs_tree_unlock(root->node);
1397         return root;
1398 }
1399
1400 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1401                              struct btrfs_fs_info *fs_info)
1402 {
1403         struct btrfs_root *log_root;
1404
1405         log_root = alloc_log_tree(trans, fs_info);
1406         if (IS_ERR(log_root))
1407                 return PTR_ERR(log_root);
1408         WARN_ON(fs_info->log_root_tree);
1409         fs_info->log_root_tree = log_root;
1410         return 0;
1411 }
1412
1413 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1414                        struct btrfs_root *root)
1415 {
1416         struct btrfs_root *log_root;
1417         struct btrfs_inode_item *inode_item;
1418
1419         log_root = alloc_log_tree(trans, root->fs_info);
1420         if (IS_ERR(log_root))
1421                 return PTR_ERR(log_root);
1422
1423         log_root->last_trans = trans->transid;
1424         log_root->root_key.offset = root->root_key.objectid;
1425
1426         inode_item = &log_root->root_item.inode;
1427         btrfs_set_stack_inode_generation(inode_item, 1);
1428         btrfs_set_stack_inode_size(inode_item, 3);
1429         btrfs_set_stack_inode_nlink(inode_item, 1);
1430         btrfs_set_stack_inode_nbytes(inode_item, root->leafsize);
1431         btrfs_set_stack_inode_mode(inode_item, S_IFDIR | 0755);
1432
1433         btrfs_set_root_node(&log_root->root_item, log_root->node);
1434
1435         WARN_ON(root->log_root);
1436         root->log_root = log_root;
1437         root->log_transid = 0;
1438         root->last_log_commit = 0;
1439         return 0;
1440 }
1441
1442 static struct btrfs_root *btrfs_read_tree_root(struct btrfs_root *tree_root,
1443                                                struct btrfs_key *key)
1444 {
1445         struct btrfs_root *root;
1446         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1447         struct btrfs_path *path;
1448         u64 generation;
1449         u32 blocksize;
1450         int ret;
1451
1452         path = btrfs_alloc_path();
1453         if (!path)
1454                 return ERR_PTR(-ENOMEM);
1455
1456         root = btrfs_alloc_root(fs_info);
1457         if (!root) {
1458                 ret = -ENOMEM;
1459                 goto alloc_fail;
1460         }
1461
1462         __setup_root(tree_root->nodesize, tree_root->leafsize,
1463                      tree_root->sectorsize, tree_root->stripesize,
1464                      root, fs_info, key->objectid);
1465
1466         ret = btrfs_find_root(tree_root, key, path,
1467                               &root->root_item, &root->root_key);
1468         if (ret) {
1469                 if (ret > 0)
1470                         ret = -ENOENT;
1471                 goto find_fail;
1472         }
1473
1474         generation = btrfs_root_generation(&root->root_item);
1475         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1476         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1477                                      blocksize, generation);
1478         if (!root->node) {
1479                 ret = -ENOMEM;
1480                 goto find_fail;
1481         } else if (!btrfs_buffer_uptodate(root->node, generation, 0)) {
1482                 ret = -EIO;
1483                 goto read_fail;
1484         }
1485         root->commit_root = btrfs_root_node(root);
1486 out:
1487         btrfs_free_path(path);
1488         return root;
1489
1490 read_fail:
1491         free_extent_buffer(root->node);
1492 find_fail:
1493         kfree(root);
1494 alloc_fail:
1495         root = ERR_PTR(ret);
1496         goto out;
1497 }
1498
1499 struct btrfs_root *btrfs_read_fs_root(struct btrfs_root *tree_root,
1500                                       struct btrfs_key *location)
1501 {
1502         struct btrfs_root *root;
1503
1504         root = btrfs_read_tree_root(tree_root, location);
1505         if (IS_ERR(root))
1506                 return root;
1507
1508         if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
1509                 root->ref_cows = 1;
1510                 btrfs_check_and_init_root_item(&root->root_item);
1511         }
1512
1513         return root;
1514 }
1515
1516 int btrfs_init_fs_root(struct btrfs_root *root)
1517 {
1518         int ret;
1519
1520         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1521         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1522                                         GFP_NOFS);
1523         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1524                 ret = -ENOMEM;
1525                 goto fail;
1526         }
1527
1528         btrfs_init_free_ino_ctl(root);
1529         mutex_init(&root->fs_commit_mutex);
1530         spin_lock_init(&root->cache_lock);
1531         init_waitqueue_head(&root->cache_wait);
1532
1533         ret = get_anon_bdev(&root->anon_dev);
1534         if (ret)
1535                 goto fail;
1536         return 0;
1537 fail:
1538         kfree(root->free_ino_ctl);
1539         kfree(root->free_ino_pinned);
1540         return ret;
1541 }
1542
1543 static struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1544                                                u64 root_id)
1545 {
1546         struct btrfs_root *root;
1547
1548         spin_lock(&fs_info->fs_roots_radix_lock);
1549         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1550                                  (unsigned long)root_id);
1551         spin_unlock(&fs_info->fs_roots_radix_lock);
1552         return root;
1553 }
1554
1555 int btrfs_insert_fs_root(struct btrfs_fs_info *fs_info,
1556                          struct btrfs_root *root)
1557 {
1558         int ret;
1559
1560         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1561         if (ret)
1562                 return ret;
1563
1564         spin_lock(&fs_info->fs_roots_radix_lock);
1565         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1566                                 (unsigned long)root->root_key.objectid,
1567                                 root);
1568         if (ret == 0)
1569                 root->in_radix = 1;
1570         spin_unlock(&fs_info->fs_roots_radix_lock);
1571         radix_tree_preload_end();
1572
1573         return ret;
1574 }
1575
1576 struct btrfs_root *btrfs_get_fs_root(struct btrfs_fs_info *fs_info,
1577                                      struct btrfs_key *location,
1578                                      bool check_ref)
1579 {
1580         struct btrfs_root *root;
1581         int ret;
1582
1583         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1584                 return fs_info->tree_root;
1585         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1586                 return fs_info->extent_root;
1587         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1588                 return fs_info->chunk_root;
1589         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1590                 return fs_info->dev_root;
1591         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1592                 return fs_info->csum_root;
1593         if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
1594                 return fs_info->quota_root ? fs_info->quota_root :
1595                                              ERR_PTR(-ENOENT);
1596         if (location->objectid == BTRFS_UUID_TREE_OBJECTID)
1597                 return fs_info->uuid_root ? fs_info->uuid_root :
1598                                             ERR_PTR(-ENOENT);
1599 again:
1600         root = btrfs_lookup_fs_root(fs_info, location->objectid);
1601         if (root) {
1602                 if (check_ref && btrfs_root_refs(&root->root_item) == 0)
1603                         return ERR_PTR(-ENOENT);
1604                 return root;
1605         }
1606
1607         root = btrfs_read_fs_root(fs_info->tree_root, location);
1608         if (IS_ERR(root))
1609                 return root;
1610
1611         if (check_ref && btrfs_root_refs(&root->root_item) == 0) {
1612                 ret = -ENOENT;
1613                 goto fail;
1614         }
1615
1616         ret = btrfs_init_fs_root(root);
1617         if (ret)
1618                 goto fail;
1619
1620         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1621         if (ret < 0)
1622                 goto fail;
1623         if (ret == 0)
1624                 root->orphan_item_inserted = 1;
1625
1626         ret = btrfs_insert_fs_root(fs_info, root);
1627         if (ret) {
1628                 if (ret == -EEXIST) {
1629                         free_fs_root(root);
1630                         goto again;
1631                 }
1632                 goto fail;
1633         }
1634         return root;
1635 fail:
1636         free_fs_root(root);
1637         return ERR_PTR(ret);
1638 }
1639
1640 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1641 {
1642         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1643         int ret = 0;
1644         struct btrfs_device *device;
1645         struct backing_dev_info *bdi;
1646
1647         rcu_read_lock();
1648         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1649                 if (!device->bdev)
1650                         continue;
1651                 bdi = blk_get_backing_dev_info(device->bdev);
1652                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1653                         ret = 1;
1654                         break;
1655                 }
1656         }
1657         rcu_read_unlock();
1658         return ret;
1659 }
1660
1661 /*
1662  * If this fails, caller must call bdi_destroy() to get rid of the
1663  * bdi again.
1664  */
1665 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1666 {
1667         int err;
1668
1669         bdi->capabilities = BDI_CAP_MAP_COPY;
1670         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1671         if (err)
1672                 return err;
1673
1674         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1675         bdi->congested_fn       = btrfs_congested_fn;
1676         bdi->congested_data     = info;
1677         return 0;
1678 }
1679
1680 /*
1681  * called by the kthread helper functions to finally call the bio end_io
1682  * functions.  This is where read checksum verification actually happens
1683  */
1684 static void end_workqueue_fn(struct btrfs_work *work)
1685 {
1686         struct bio *bio;
1687         struct end_io_wq *end_io_wq;
1688         struct btrfs_fs_info *fs_info;
1689         int error;
1690
1691         end_io_wq = container_of(work, struct end_io_wq, work);
1692         bio = end_io_wq->bio;
1693         fs_info = end_io_wq->info;
1694
1695         error = end_io_wq->error;
1696         bio->bi_private = end_io_wq->private;
1697         bio->bi_end_io = end_io_wq->end_io;
1698         kfree(end_io_wq);
1699         bio_endio(bio, error);
1700 }
1701
1702 static int cleaner_kthread(void *arg)
1703 {
1704         struct btrfs_root *root = arg;
1705         int again;
1706
1707         do {
1708                 again = 0;
1709
1710                 /* Make the cleaner go to sleep early. */
1711                 if (btrfs_need_cleaner_sleep(root))
1712                         goto sleep;
1713
1714                 if (!mutex_trylock(&root->fs_info->cleaner_mutex))
1715                         goto sleep;
1716
1717                 /*
1718                  * Avoid the problem that we change the status of the fs
1719                  * during the above check and trylock.
1720                  */
1721                 if (btrfs_need_cleaner_sleep(root)) {
1722                         mutex_unlock(&root->fs_info->cleaner_mutex);
1723                         goto sleep;
1724                 }
1725
1726                 btrfs_run_delayed_iputs(root);
1727                 again = btrfs_clean_one_deleted_snapshot(root);
1728                 mutex_unlock(&root->fs_info->cleaner_mutex);
1729
1730                 /*
1731                  * The defragger has dealt with the R/O remount and umount,
1732                  * needn't do anything special here.
1733                  */
1734                 btrfs_run_defrag_inodes(root->fs_info);
1735 sleep:
1736                 if (!try_to_freeze() && !again) {
1737                         set_current_state(TASK_INTERRUPTIBLE);
1738                         if (!kthread_should_stop())
1739                                 schedule();
1740                         __set_current_state(TASK_RUNNING);
1741                 }
1742         } while (!kthread_should_stop());
1743         return 0;
1744 }
1745
1746 static int transaction_kthread(void *arg)
1747 {
1748         struct btrfs_root *root = arg;
1749         struct btrfs_trans_handle *trans;
1750         struct btrfs_transaction *cur;
1751         u64 transid;
1752         unsigned long now;
1753         unsigned long delay;
1754         bool cannot_commit;
1755
1756         do {
1757                 cannot_commit = false;
1758                 delay = HZ * root->fs_info->commit_interval;
1759                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1760
1761                 spin_lock(&root->fs_info->trans_lock);
1762                 cur = root->fs_info->running_transaction;
1763                 if (!cur) {
1764                         spin_unlock(&root->fs_info->trans_lock);
1765                         goto sleep;
1766                 }
1767
1768                 now = get_seconds();
1769                 if (cur->state < TRANS_STATE_BLOCKED &&
1770                     (now < cur->start_time ||
1771                      now - cur->start_time < root->fs_info->commit_interval)) {
1772                         spin_unlock(&root->fs_info->trans_lock);
1773                         delay = HZ * 5;
1774                         goto sleep;
1775                 }
1776                 transid = cur->transid;
1777                 spin_unlock(&root->fs_info->trans_lock);
1778
1779                 /* If the file system is aborted, this will always fail. */
1780                 trans = btrfs_attach_transaction(root);
1781                 if (IS_ERR(trans)) {
1782                         if (PTR_ERR(trans) != -ENOENT)
1783                                 cannot_commit = true;
1784                         goto sleep;
1785                 }
1786                 if (transid == trans->transid) {
1787                         btrfs_commit_transaction(trans, root);
1788                 } else {
1789                         btrfs_end_transaction(trans, root);
1790                 }
1791 sleep:
1792                 wake_up_process(root->fs_info->cleaner_kthread);
1793                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1794
1795                 if (unlikely(test_bit(BTRFS_FS_STATE_ERROR,
1796                                       &root->fs_info->fs_state)))
1797                         btrfs_cleanup_transaction(root);
1798                 if (!try_to_freeze()) {
1799                         set_current_state(TASK_INTERRUPTIBLE);
1800                         if (!kthread_should_stop() &&
1801                             (!btrfs_transaction_blocked(root->fs_info) ||
1802                              cannot_commit))
1803                                 schedule_timeout(delay);
1804                         __set_current_state(TASK_RUNNING);
1805                 }
1806         } while (!kthread_should_stop());
1807         return 0;
1808 }
1809
1810 /*
1811  * this will find the highest generation in the array of
1812  * root backups.  The index of the highest array is returned,
1813  * or -1 if we can't find anything.
1814  *
1815  * We check to make sure the array is valid by comparing the
1816  * generation of the latest  root in the array with the generation
1817  * in the super block.  If they don't match we pitch it.
1818  */
1819 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1820 {
1821         u64 cur;
1822         int newest_index = -1;
1823         struct btrfs_root_backup *root_backup;
1824         int i;
1825
1826         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1827                 root_backup = info->super_copy->super_roots + i;
1828                 cur = btrfs_backup_tree_root_gen(root_backup);
1829                 if (cur == newest_gen)
1830                         newest_index = i;
1831         }
1832
1833         /* check to see if we actually wrapped around */
1834         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1835                 root_backup = info->super_copy->super_roots;
1836                 cur = btrfs_backup_tree_root_gen(root_backup);
1837                 if (cur == newest_gen)
1838                         newest_index = 0;
1839         }
1840         return newest_index;
1841 }
1842
1843
1844 /*
1845  * find the oldest backup so we know where to store new entries
1846  * in the backup array.  This will set the backup_root_index
1847  * field in the fs_info struct
1848  */
1849 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1850                                      u64 newest_gen)
1851 {
1852         int newest_index = -1;
1853
1854         newest_index = find_newest_super_backup(info, newest_gen);
1855         /* if there was garbage in there, just move along */
1856         if (newest_index == -1) {
1857                 info->backup_root_index = 0;
1858         } else {
1859                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1860         }
1861 }
1862
1863 /*
1864  * copy all the root pointers into the super backup array.
1865  * this will bump the backup pointer by one when it is
1866  * done
1867  */
1868 static void backup_super_roots(struct btrfs_fs_info *info)
1869 {
1870         int next_backup;
1871         struct btrfs_root_backup *root_backup;
1872         int last_backup;
1873
1874         next_backup = info->backup_root_index;
1875         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1876                 BTRFS_NUM_BACKUP_ROOTS;
1877
1878         /*
1879          * just overwrite the last backup if we're at the same generation
1880          * this happens only at umount
1881          */
1882         root_backup = info->super_for_commit->super_roots + last_backup;
1883         if (btrfs_backup_tree_root_gen(root_backup) ==
1884             btrfs_header_generation(info->tree_root->node))
1885                 next_backup = last_backup;
1886
1887         root_backup = info->super_for_commit->super_roots + next_backup;
1888
1889         /*
1890          * make sure all of our padding and empty slots get zero filled
1891          * regardless of which ones we use today
1892          */
1893         memset(root_backup, 0, sizeof(*root_backup));
1894
1895         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1896
1897         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1898         btrfs_set_backup_tree_root_gen(root_backup,
1899                                btrfs_header_generation(info->tree_root->node));
1900
1901         btrfs_set_backup_tree_root_level(root_backup,
1902                                btrfs_header_level(info->tree_root->node));
1903
1904         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1905         btrfs_set_backup_chunk_root_gen(root_backup,
1906                                btrfs_header_generation(info->chunk_root->node));
1907         btrfs_set_backup_chunk_root_level(root_backup,
1908                                btrfs_header_level(info->chunk_root->node));
1909
1910         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1911         btrfs_set_backup_extent_root_gen(root_backup,
1912                                btrfs_header_generation(info->extent_root->node));
1913         btrfs_set_backup_extent_root_level(root_backup,
1914                                btrfs_header_level(info->extent_root->node));
1915
1916         /*
1917          * we might commit during log recovery, which happens before we set
1918          * the fs_root.  Make sure it is valid before we fill it in.
1919          */
1920         if (info->fs_root && info->fs_root->node) {
1921                 btrfs_set_backup_fs_root(root_backup,
1922                                          info->fs_root->node->start);
1923                 btrfs_set_backup_fs_root_gen(root_backup,
1924                                btrfs_header_generation(info->fs_root->node));
1925                 btrfs_set_backup_fs_root_level(root_backup,
1926                                btrfs_header_level(info->fs_root->node));
1927         }
1928
1929         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1930         btrfs_set_backup_dev_root_gen(root_backup,
1931                                btrfs_header_generation(info->dev_root->node));
1932         btrfs_set_backup_dev_root_level(root_backup,
1933                                        btrfs_header_level(info->dev_root->node));
1934
1935         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1936         btrfs_set_backup_csum_root_gen(root_backup,
1937                                btrfs_header_generation(info->csum_root->node));
1938         btrfs_set_backup_csum_root_level(root_backup,
1939                                btrfs_header_level(info->csum_root->node));
1940
1941         btrfs_set_backup_total_bytes(root_backup,
1942                              btrfs_super_total_bytes(info->super_copy));
1943         btrfs_set_backup_bytes_used(root_backup,
1944                              btrfs_super_bytes_used(info->super_copy));
1945         btrfs_set_backup_num_devices(root_backup,
1946                              btrfs_super_num_devices(info->super_copy));
1947
1948         /*
1949          * if we don't copy this out to the super_copy, it won't get remembered
1950          * for the next commit
1951          */
1952         memcpy(&info->super_copy->super_roots,
1953                &info->super_for_commit->super_roots,
1954                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1955 }
1956
1957 /*
1958  * this copies info out of the root backup array and back into
1959  * the in-memory super block.  It is meant to help iterate through
1960  * the array, so you send it the number of backups you've already
1961  * tried and the last backup index you used.
1962  *
1963  * this returns -1 when it has tried all the backups
1964  */
1965 static noinline int next_root_backup(struct btrfs_fs_info *info,
1966                                      struct btrfs_super_block *super,
1967                                      int *num_backups_tried, int *backup_index)
1968 {
1969         struct btrfs_root_backup *root_backup;
1970         int newest = *backup_index;
1971
1972         if (*num_backups_tried == 0) {
1973                 u64 gen = btrfs_super_generation(super);
1974
1975                 newest = find_newest_super_backup(info, gen);
1976                 if (newest == -1)
1977                         return -1;
1978
1979                 *backup_index = newest;
1980                 *num_backups_tried = 1;
1981         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1982                 /* we've tried all the backups, all done */
1983                 return -1;
1984         } else {
1985                 /* jump to the next oldest backup */
1986                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1987                         BTRFS_NUM_BACKUP_ROOTS;
1988                 *backup_index = newest;
1989                 *num_backups_tried += 1;
1990         }
1991         root_backup = super->super_roots + newest;
1992
1993         btrfs_set_super_generation(super,
1994                                    btrfs_backup_tree_root_gen(root_backup));
1995         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1996         btrfs_set_super_root_level(super,
1997                                    btrfs_backup_tree_root_level(root_backup));
1998         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1999
2000         /*
2001          * fixme: the total bytes and num_devices need to match or we should
2002          * need a fsck
2003          */
2004         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
2005         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
2006         return 0;
2007 }
2008
2009 /* helper to cleanup workers */
2010 static void btrfs_stop_all_workers(struct btrfs_fs_info *fs_info)
2011 {
2012         btrfs_stop_workers(&fs_info->generic_worker);
2013         btrfs_stop_workers(&fs_info->fixup_workers);
2014         btrfs_stop_workers(&fs_info->delalloc_workers);
2015         btrfs_stop_workers(&fs_info->workers);
2016         btrfs_stop_workers(&fs_info->endio_workers);
2017         btrfs_stop_workers(&fs_info->endio_meta_workers);
2018         btrfs_stop_workers(&fs_info->endio_raid56_workers);
2019         btrfs_stop_workers(&fs_info->rmw_workers);
2020         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2021         btrfs_stop_workers(&fs_info->endio_write_workers);
2022         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2023         btrfs_stop_workers(&fs_info->submit_workers);
2024         btrfs_stop_workers(&fs_info->delayed_workers);
2025         btrfs_stop_workers(&fs_info->caching_workers);
2026         btrfs_stop_workers(&fs_info->readahead_workers);
2027         btrfs_stop_workers(&fs_info->flush_workers);
2028         btrfs_stop_workers(&fs_info->qgroup_rescan_workers);
2029 }
2030
2031 static void free_root_extent_buffers(struct btrfs_root *root)
2032 {
2033         if (root) {
2034                 free_extent_buffer(root->node);
2035                 free_extent_buffer(root->commit_root);
2036                 root->node = NULL;
2037                 root->commit_root = NULL;
2038         }
2039 }
2040
2041 /* helper to cleanup tree roots */
2042 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
2043 {
2044         free_root_extent_buffers(info->tree_root);
2045
2046         free_root_extent_buffers(info->dev_root);
2047         free_root_extent_buffers(info->extent_root);
2048         free_root_extent_buffers(info->csum_root);
2049         free_root_extent_buffers(info->quota_root);
2050         free_root_extent_buffers(info->uuid_root);
2051         if (chunk_root)
2052                 free_root_extent_buffers(info->chunk_root);
2053 }
2054
2055 static void del_fs_roots(struct btrfs_fs_info *fs_info)
2056 {
2057         int ret;
2058         struct btrfs_root *gang[8];
2059         int i;
2060
2061         while (!list_empty(&fs_info->dead_roots)) {
2062                 gang[0] = list_entry(fs_info->dead_roots.next,
2063                                      struct btrfs_root, root_list);
2064                 list_del(&gang[0]->root_list);
2065
2066                 if (gang[0]->in_radix) {
2067                         btrfs_drop_and_free_fs_root(fs_info, gang[0]);
2068                 } else {
2069                         free_extent_buffer(gang[0]->node);
2070                         free_extent_buffer(gang[0]->commit_root);
2071                         btrfs_put_fs_root(gang[0]);
2072                 }
2073         }
2074
2075         while (1) {
2076                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2077                                              (void **)gang, 0,
2078                                              ARRAY_SIZE(gang));
2079                 if (!ret)
2080                         break;
2081                 for (i = 0; i < ret; i++)
2082                         btrfs_drop_and_free_fs_root(fs_info, gang[i]);
2083         }
2084 }
2085
2086 int open_ctree(struct super_block *sb,
2087                struct btrfs_fs_devices *fs_devices,
2088                char *options)
2089 {
2090         u32 sectorsize;
2091         u32 nodesize;
2092         u32 leafsize;
2093         u32 blocksize;
2094         u32 stripesize;
2095         u64 generation;
2096         u64 features;
2097         struct btrfs_key location;
2098         struct buffer_head *bh;
2099         struct btrfs_super_block *disk_super;
2100         struct btrfs_fs_info *fs_info = btrfs_sb(sb);
2101         struct btrfs_root *tree_root;
2102         struct btrfs_root *extent_root;
2103         struct btrfs_root *csum_root;
2104         struct btrfs_root *chunk_root;
2105         struct btrfs_root *dev_root;
2106         struct btrfs_root *quota_root;
2107         struct btrfs_root *uuid_root;
2108         struct btrfs_root *log_tree_root;
2109         int ret;
2110         int err = -EINVAL;
2111         int num_backups_tried = 0;
2112         int backup_index = 0;
2113         bool create_uuid_tree;
2114         bool check_uuid_tree;
2115
2116         tree_root = fs_info->tree_root = btrfs_alloc_root(fs_info);
2117         chunk_root = fs_info->chunk_root = btrfs_alloc_root(fs_info);
2118         if (!tree_root || !chunk_root) {
2119                 err = -ENOMEM;
2120                 goto fail;
2121         }
2122
2123         ret = init_srcu_struct(&fs_info->subvol_srcu);
2124         if (ret) {
2125                 err = ret;
2126                 goto fail;
2127         }
2128
2129         ret = setup_bdi(fs_info, &fs_info->bdi);
2130         if (ret) {
2131                 err = ret;
2132                 goto fail_srcu;
2133         }
2134
2135         ret = percpu_counter_init(&fs_info->dirty_metadata_bytes, 0);
2136         if (ret) {
2137                 err = ret;
2138                 goto fail_bdi;
2139         }
2140         fs_info->dirty_metadata_batch = PAGE_CACHE_SIZE *
2141                                         (1 + ilog2(nr_cpu_ids));
2142
2143         ret = percpu_counter_init(&fs_info->delalloc_bytes, 0);
2144         if (ret) {
2145                 err = ret;
2146                 goto fail_dirty_metadata_bytes;
2147         }
2148
2149         fs_info->btree_inode = new_inode(sb);
2150         if (!fs_info->btree_inode) {
2151                 err = -ENOMEM;
2152                 goto fail_delalloc_bytes;
2153         }
2154
2155         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
2156
2157         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
2158         INIT_LIST_HEAD(&fs_info->trans_list);
2159         INIT_LIST_HEAD(&fs_info->dead_roots);
2160         INIT_LIST_HEAD(&fs_info->delayed_iputs);
2161         INIT_LIST_HEAD(&fs_info->delalloc_roots);
2162         INIT_LIST_HEAD(&fs_info->caching_block_groups);
2163         spin_lock_init(&fs_info->delalloc_root_lock);
2164         spin_lock_init(&fs_info->trans_lock);
2165         spin_lock_init(&fs_info->fs_roots_radix_lock);
2166         spin_lock_init(&fs_info->delayed_iput_lock);
2167         spin_lock_init(&fs_info->defrag_inodes_lock);
2168         spin_lock_init(&fs_info->free_chunk_lock);
2169         spin_lock_init(&fs_info->tree_mod_seq_lock);
2170         spin_lock_init(&fs_info->super_lock);
2171         rwlock_init(&fs_info->tree_mod_log_lock);
2172         mutex_init(&fs_info->reloc_mutex);
2173         seqlock_init(&fs_info->profiles_lock);
2174
2175         init_completion(&fs_info->kobj_unregister);
2176         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
2177         INIT_LIST_HEAD(&fs_info->space_info);
2178         INIT_LIST_HEAD(&fs_info->tree_mod_seq_list);
2179         btrfs_mapping_init(&fs_info->mapping_tree);
2180         btrfs_init_block_rsv(&fs_info->global_block_rsv,
2181                              BTRFS_BLOCK_RSV_GLOBAL);
2182         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv,
2183                              BTRFS_BLOCK_RSV_DELALLOC);
2184         btrfs_init_block_rsv(&fs_info->trans_block_rsv, BTRFS_BLOCK_RSV_TRANS);
2185         btrfs_init_block_rsv(&fs_info->chunk_block_rsv, BTRFS_BLOCK_RSV_CHUNK);
2186         btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
2187         btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
2188                              BTRFS_BLOCK_RSV_DELOPS);
2189         atomic_set(&fs_info->nr_async_submits, 0);
2190         atomic_set(&fs_info->async_delalloc_pages, 0);
2191         atomic_set(&fs_info->async_submit_draining, 0);
2192         atomic_set(&fs_info->nr_async_bios, 0);
2193         atomic_set(&fs_info->defrag_running, 0);
2194         atomic64_set(&fs_info->tree_mod_seq, 0);
2195         fs_info->sb = sb;
2196         fs_info->max_inline = 8192 * 1024;
2197         fs_info->metadata_ratio = 0;
2198         fs_info->defrag_inodes = RB_ROOT;
2199         fs_info->free_chunk_space = 0;
2200         fs_info->tree_mod_log = RB_ROOT;
2201         fs_info->commit_interval = BTRFS_DEFAULT_COMMIT_INTERVAL;
2202
2203         /* readahead state */
2204         INIT_RADIX_TREE(&fs_info->reada_tree, GFP_NOFS & ~__GFP_WAIT);
2205         spin_lock_init(&fs_info->reada_lock);
2206
2207         fs_info->thread_pool_size = min_t(unsigned long,
2208                                           num_online_cpus() + 2, 8);
2209
2210         INIT_LIST_HEAD(&fs_info->ordered_roots);
2211         spin_lock_init(&fs_info->ordered_root_lock);
2212         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
2213                                         GFP_NOFS);
2214         if (!fs_info->delayed_root) {
2215                 err = -ENOMEM;
2216                 goto fail_iput;
2217         }
2218         btrfs_init_delayed_root(fs_info->delayed_root);
2219
2220         mutex_init(&fs_info->scrub_lock);
2221         atomic_set(&fs_info->scrubs_running, 0);
2222         atomic_set(&fs_info->scrub_pause_req, 0);
2223         atomic_set(&fs_info->scrubs_paused, 0);
2224         atomic_set(&fs_info->scrub_cancel_req, 0);
2225         init_waitqueue_head(&fs_info->scrub_pause_wait);
2226         fs_info->scrub_workers_refcnt = 0;
2227 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2228         fs_info->check_integrity_print_mask = 0;
2229 #endif
2230
2231         spin_lock_init(&fs_info->balance_lock);
2232         mutex_init(&fs_info->balance_mutex);
2233         atomic_set(&fs_info->balance_running, 0);
2234         atomic_set(&fs_info->balance_pause_req, 0);
2235         atomic_set(&fs_info->balance_cancel_req, 0);
2236         fs_info->balance_ctl = NULL;
2237         init_waitqueue_head(&fs_info->balance_wait_q);
2238
2239         sb->s_blocksize = 4096;
2240         sb->s_blocksize_bits = blksize_bits(4096);
2241         sb->s_bdi = &fs_info->bdi;
2242
2243         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
2244         set_nlink(fs_info->btree_inode, 1);
2245         /*
2246          * we set the i_size on the btree inode to the max possible int.
2247          * the real end of the address space is determined by all of
2248          * the devices in the system
2249          */
2250         fs_info->btree_inode->i_size = OFFSET_MAX;
2251         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
2252         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
2253
2254         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
2255         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
2256                              fs_info->btree_inode->i_mapping);
2257         BTRFS_I(fs_info->btree_inode)->io_tree.track_uptodate = 0;
2258         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
2259
2260         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
2261
2262         BTRFS_I(fs_info->btree_inode)->root = tree_root;
2263         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
2264                sizeof(struct btrfs_key));
2265         set_bit(BTRFS_INODE_DUMMY,
2266                 &BTRFS_I(fs_info->btree_inode)->runtime_flags);
2267         btrfs_insert_inode_hash(fs_info->btree_inode);
2268
2269         spin_lock_init(&fs_info->block_group_cache_lock);
2270         fs_info->block_group_cache_tree = RB_ROOT;
2271         fs_info->first_logical_byte = (u64)-1;
2272
2273         extent_io_tree_init(&fs_info->freed_extents[0],
2274                              fs_info->btree_inode->i_mapping);
2275         extent_io_tree_init(&fs_info->freed_extents[1],
2276                              fs_info->btree_inode->i_mapping);
2277         fs_info->pinned_extents = &fs_info->freed_extents[0];
2278         fs_info->do_barriers = 1;
2279
2280
2281         mutex_init(&fs_info->ordered_operations_mutex);
2282         mutex_init(&fs_info->ordered_extent_flush_mutex);
2283         mutex_init(&fs_info->tree_log_mutex);
2284         mutex_init(&fs_info->chunk_mutex);
2285         mutex_init(&fs_info->transaction_kthread_mutex);
2286         mutex_init(&fs_info->cleaner_mutex);
2287         mutex_init(&fs_info->volume_mutex);
2288         init_rwsem(&fs_info->extent_commit_sem);
2289         init_rwsem(&fs_info->cleanup_work_sem);
2290         init_rwsem(&fs_info->subvol_sem);
2291         sema_init(&fs_info->uuid_tree_rescan_sem, 1);
2292         fs_info->dev_replace.lock_owner = 0;
2293         atomic_set(&fs_info->dev_replace.nesting_level, 0);
2294         mutex_init(&fs_info->dev_replace.lock_finishing_cancel_unmount);
2295         mutex_init(&fs_info->dev_replace.lock_management_lock);
2296         mutex_init(&fs_info->dev_replace.lock);
2297
2298         spin_lock_init(&fs_info->qgroup_lock);
2299         mutex_init(&fs_info->qgroup_ioctl_lock);
2300         fs_info->qgroup_tree = RB_ROOT;
2301         INIT_LIST_HEAD(&fs_info->dirty_qgroups);
2302         fs_info->qgroup_seq = 1;
2303         fs_info->quota_enabled = 0;
2304         fs_info->pending_quota_state = 0;
2305         fs_info->qgroup_ulist = NULL;
2306         mutex_init(&fs_info->qgroup_rescan_lock);
2307
2308         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
2309         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
2310
2311         init_waitqueue_head(&fs_info->transaction_throttle);
2312         init_waitqueue_head(&fs_info->transaction_wait);
2313         init_waitqueue_head(&fs_info->transaction_blocked_wait);
2314         init_waitqueue_head(&fs_info->async_submit_wait);
2315
2316         ret = btrfs_alloc_stripe_hash_table(fs_info);
2317         if (ret) {
2318                 err = ret;
2319                 goto fail_alloc;
2320         }
2321
2322         __setup_root(4096, 4096, 4096, 4096, tree_root,
2323                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
2324
2325         invalidate_bdev(fs_devices->latest_bdev);
2326
2327         /*
2328          * Read super block and check the signature bytes only
2329          */
2330         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
2331         if (!bh) {
2332                 err = -EINVAL;
2333                 goto fail_alloc;
2334         }
2335
2336         /*
2337          * We want to check superblock checksum, the type is stored inside.
2338          * Pass the whole disk block of size BTRFS_SUPER_INFO_SIZE (4k).
2339          */
2340         if (btrfs_check_super_csum(bh->b_data)) {
2341                 printk(KERN_ERR "btrfs: superblock checksum mismatch\n");
2342                 err = -EINVAL;
2343                 goto fail_alloc;
2344         }
2345
2346         /*
2347          * super_copy is zeroed at allocation time and we never touch the
2348          * following bytes up to INFO_SIZE, the checksum is calculated from
2349          * the whole block of INFO_SIZE
2350          */
2351         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
2352         memcpy(fs_info->super_for_commit, fs_info->super_copy,
2353                sizeof(*fs_info->super_for_commit));
2354         brelse(bh);
2355
2356         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2357
2358         ret = btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2359         if (ret) {
2360                 printk(KERN_ERR "btrfs: superblock contains fatal errors\n");
2361                 err = -EINVAL;
2362                 goto fail_alloc;
2363         }
2364
2365         disk_super = fs_info->super_copy;
2366         if (!btrfs_super_root(disk_super))
2367                 goto fail_alloc;
2368
2369         /* check FS state, whether FS is broken. */
2370         if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_ERROR)
2371                 set_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state);
2372
2373         /*
2374          * run through our array of backup supers and setup
2375          * our ring pointer to the oldest one
2376          */
2377         generation = btrfs_super_generation(disk_super);
2378         find_oldest_super_backup(fs_info, generation);
2379
2380         /*
2381          * In the long term, we'll store the compression type in the super
2382          * block, and it'll be used for per file compression control.
2383          */
2384         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2385
2386         ret = btrfs_parse_options(tree_root, options);
2387         if (ret) {
2388                 err = ret;
2389                 goto fail_alloc;
2390         }
2391
2392         features = btrfs_super_incompat_flags(disk_super) &
2393                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2394         if (features) {
2395                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2396                        "unsupported optional features (%Lx).\n",
2397                        features);
2398                 err = -EINVAL;
2399                 goto fail_alloc;
2400         }
2401
2402         if (btrfs_super_leafsize(disk_super) !=
2403             btrfs_super_nodesize(disk_super)) {
2404                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2405                        "blocksizes don't match.  node %d leaf %d\n",
2406                        btrfs_super_nodesize(disk_super),
2407                        btrfs_super_leafsize(disk_super));
2408                 err = -EINVAL;
2409                 goto fail_alloc;
2410         }
2411         if (btrfs_super_leafsize(disk_super) > BTRFS_MAX_METADATA_BLOCKSIZE) {
2412                 printk(KERN_ERR "BTRFS: couldn't mount because metadata "
2413                        "blocksize (%d) was too large\n",
2414                        btrfs_super_leafsize(disk_super));
2415                 err = -EINVAL;
2416                 goto fail_alloc;
2417         }
2418
2419         features = btrfs_super_incompat_flags(disk_super);
2420         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2421         if (tree_root->fs_info->compress_type == BTRFS_COMPRESS_LZO)
2422                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2423
2424         if (features & BTRFS_FEATURE_INCOMPAT_SKINNY_METADATA)
2425                 printk(KERN_ERR "btrfs: has skinny extents\n");
2426
2427         /*
2428          * flag our filesystem as having big metadata blocks if
2429          * they are bigger than the page size
2430          */
2431         if (btrfs_super_leafsize(disk_super) > PAGE_CACHE_SIZE) {
2432                 if (!(features & BTRFS_FEATURE_INCOMPAT_BIG_METADATA))
2433                         printk(KERN_INFO "btrfs flagging fs with big metadata feature\n");
2434                 features |= BTRFS_FEATURE_INCOMPAT_BIG_METADATA;
2435         }
2436
2437         nodesize = btrfs_super_nodesize(disk_super);
2438         leafsize = btrfs_super_leafsize(disk_super);
2439         sectorsize = btrfs_super_sectorsize(disk_super);
2440         stripesize = btrfs_super_stripesize(disk_super);
2441         fs_info->dirty_metadata_batch = leafsize * (1 + ilog2(nr_cpu_ids));
2442         fs_info->delalloc_batch = sectorsize * 512 * (1 + ilog2(nr_cpu_ids));
2443
2444         /*
2445          * mixed block groups end up with duplicate but slightly offset
2446          * extent buffers for the same range.  It leads to corruptions
2447          */
2448         if ((features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) &&
2449             (sectorsize != leafsize)) {
2450                 printk(KERN_WARNING "btrfs: unequal leaf/node/sector sizes "
2451                                 "are not allowed for mixed block groups on %s\n",
2452                                 sb->s_id);
2453                 goto fail_alloc;
2454         }
2455
2456         /*
2457          * Needn't use the lock because there is no other task which will
2458          * update the flag.
2459          */
2460         btrfs_set_super_incompat_flags(disk_super, features);
2461
2462         features = btrfs_super_compat_ro_flags(disk_super) &
2463                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2464         if (!(sb->s_flags & MS_RDONLY) && features) {
2465                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2466                        "unsupported option features (%Lx).\n",
2467                        features);
2468                 err = -EINVAL;
2469                 goto fail_alloc;
2470         }
2471
2472         btrfs_init_workers(&fs_info->generic_worker,
2473                            "genwork", 1, NULL);
2474
2475         btrfs_init_workers(&fs_info->workers, "worker",
2476                            fs_info->thread_pool_size,
2477                            &fs_info->generic_worker);
2478
2479         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2480                            fs_info->thread_pool_size, NULL);
2481
2482         btrfs_init_workers(&fs_info->flush_workers, "flush_delalloc",
2483                            fs_info->thread_pool_size, NULL);
2484
2485         btrfs_init_workers(&fs_info->submit_workers, "submit",
2486                            min_t(u64, fs_devices->num_devices,
2487                            fs_info->thread_pool_size), NULL);
2488
2489         btrfs_init_workers(&fs_info->caching_workers, "cache",
2490                            fs_info->thread_pool_size, NULL);
2491
2492         /* a higher idle thresh on the submit workers makes it much more
2493          * likely that bios will be send down in a sane order to the
2494          * devices
2495          */
2496         fs_info->submit_workers.idle_thresh = 64;
2497
2498         fs_info->workers.idle_thresh = 16;
2499         fs_info->workers.ordered = 1;
2500
2501         fs_info->delalloc_workers.idle_thresh = 2;
2502         fs_info->delalloc_workers.ordered = 1;
2503
2504         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2505                            &fs_info->generic_worker);
2506         btrfs_init_workers(&fs_info->endio_workers, "endio",
2507                            fs_info->thread_pool_size,
2508                            &fs_info->generic_worker);
2509         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2510                            fs_info->thread_pool_size,
2511                            &fs_info->generic_worker);
2512         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2513                            "endio-meta-write", fs_info->thread_pool_size,
2514                            &fs_info->generic_worker);
2515         btrfs_init_workers(&fs_info->endio_raid56_workers,
2516                            "endio-raid56", fs_info->thread_pool_size,
2517                            &fs_info->generic_worker);
2518         btrfs_init_workers(&fs_info->rmw_workers,
2519                            "rmw", fs_info->thread_pool_size,
2520                            &fs_info->generic_worker);
2521         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2522                            fs_info->thread_pool_size,
2523                            &fs_info->generic_worker);
2524         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2525                            1, &fs_info->generic_worker);
2526         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2527                            fs_info->thread_pool_size,
2528                            &fs_info->generic_worker);
2529         btrfs_init_workers(&fs_info->readahead_workers, "readahead",
2530                            fs_info->thread_pool_size,
2531                            &fs_info->generic_worker);
2532         btrfs_init_workers(&fs_info->qgroup_rescan_workers, "qgroup-rescan", 1,
2533                            &fs_info->generic_worker);
2534
2535         /*
2536          * endios are largely parallel and should have a very
2537          * low idle thresh
2538          */
2539         fs_info->endio_workers.idle_thresh = 4;
2540         fs_info->endio_meta_workers.idle_thresh = 4;
2541         fs_info->endio_raid56_workers.idle_thresh = 4;
2542         fs_info->rmw_workers.idle_thresh = 2;
2543
2544         fs_info->endio_write_workers.idle_thresh = 2;
2545         fs_info->endio_meta_write_workers.idle_thresh = 2;
2546         fs_info->readahead_workers.idle_thresh = 2;
2547
2548         /*
2549          * btrfs_start_workers can really only fail because of ENOMEM so just
2550          * return -ENOMEM if any of these fail.
2551          */
2552         ret = btrfs_start_workers(&fs_info->workers);
2553         ret |= btrfs_start_workers(&fs_info->generic_worker);
2554         ret |= btrfs_start_workers(&fs_info->submit_workers);
2555         ret |= btrfs_start_workers(&fs_info->delalloc_workers);
2556         ret |= btrfs_start_workers(&fs_info->fixup_workers);
2557         ret |= btrfs_start_workers(&fs_info->endio_workers);
2558         ret |= btrfs_start_workers(&fs_info->endio_meta_workers);
2559         ret |= btrfs_start_workers(&fs_info->rmw_workers);
2560         ret |= btrfs_start_workers(&fs_info->endio_raid56_workers);
2561         ret |= btrfs_start_workers(&fs_info->endio_meta_write_workers);
2562         ret |= btrfs_start_workers(&fs_info->endio_write_workers);
2563         ret |= btrfs_start_workers(&fs_info->endio_freespace_worker);
2564         ret |= btrfs_start_workers(&fs_info->delayed_workers);
2565         ret |= btrfs_start_workers(&fs_info->caching_workers);
2566         ret |= btrfs_start_workers(&fs_info->readahead_workers);
2567         ret |= btrfs_start_workers(&fs_info->flush_workers);
2568         ret |= btrfs_start_workers(&fs_info->qgroup_rescan_workers);
2569         if (ret) {
2570                 err = -ENOMEM;
2571                 goto fail_sb_buffer;
2572         }
2573
2574         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2575         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2576                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2577
2578         tree_root->nodesize = nodesize;
2579         tree_root->leafsize = leafsize;
2580         tree_root->sectorsize = sectorsize;
2581         tree_root->stripesize = stripesize;
2582
2583         sb->s_blocksize = sectorsize;
2584         sb->s_blocksize_bits = blksize_bits(sectorsize);
2585
2586         if (btrfs_super_magic(disk_super) != BTRFS_MAGIC) {
2587                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2588                 goto fail_sb_buffer;
2589         }
2590
2591         if (sectorsize != PAGE_SIZE) {
2592                 printk(KERN_WARNING "btrfs: Incompatible sector size(%lu) "
2593                        "found on %s\n", (unsigned long)sectorsize, sb->s_id);
2594                 goto fail_sb_buffer;
2595         }
2596
2597         mutex_lock(&fs_info->chunk_mutex);
2598         ret = btrfs_read_sys_array(tree_root);
2599         mutex_unlock(&fs_info->chunk_mutex);
2600         if (ret) {
2601                 printk(KERN_WARNING "btrfs: failed to read the system "
2602                        "array on %s\n", sb->s_id);
2603                 goto fail_sb_buffer;
2604         }
2605
2606         blocksize = btrfs_level_size(tree_root,
2607                                      btrfs_super_chunk_root_level(disk_super));
2608         generation = btrfs_super_chunk_root_generation(disk_super);
2609
2610         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2611                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2612
2613         chunk_root->node = read_tree_block(chunk_root,
2614                                            btrfs_super_chunk_root(disk_super),
2615                                            blocksize, generation);
2616         if (!chunk_root->node ||
2617             !test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2618                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2619                        sb->s_id);
2620                 goto fail_tree_roots;
2621         }
2622         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2623         chunk_root->commit_root = btrfs_root_node(chunk_root);
2624
2625         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2626            btrfs_header_chunk_tree_uuid(chunk_root->node), BTRFS_UUID_SIZE);
2627
2628         ret = btrfs_read_chunk_tree(chunk_root);
2629         if (ret) {
2630                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2631                        sb->s_id);
2632                 goto fail_tree_roots;
2633         }
2634
2635         /*
2636          * keep the device that is marked to be the target device for the
2637          * dev_replace procedure
2638          */
2639         btrfs_close_extra_devices(fs_info, fs_devices, 0);
2640
2641         if (!fs_devices->latest_bdev) {
2642                 printk(KERN_CRIT "btrfs: failed to read devices on %s\n",
2643                        sb->s_id);
2644                 goto fail_tree_roots;
2645         }
2646
2647 retry_root_backup:
2648         blocksize = btrfs_level_size(tree_root,
2649                                      btrfs_super_root_level(disk_super));
2650         generation = btrfs_super_generation(disk_super);
2651
2652         tree_root->node = read_tree_block(tree_root,
2653                                           btrfs_super_root(disk_super),
2654                                           blocksize, generation);
2655         if (!tree_root->node ||
2656             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2657                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2658                        sb->s_id);
2659
2660                 goto recovery_tree_root;
2661         }
2662
2663         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2664         tree_root->commit_root = btrfs_root_node(tree_root);
2665         btrfs_set_root_refs(&tree_root->root_item, 1);
2666
2667         location.objectid = BTRFS_EXTENT_TREE_OBJECTID;
2668         location.type = BTRFS_ROOT_ITEM_KEY;
2669         location.offset = 0;
2670
2671         extent_root = btrfs_read_tree_root(tree_root, &location);
2672         if (IS_ERR(extent_root)) {
2673                 ret = PTR_ERR(extent_root);
2674                 goto recovery_tree_root;
2675         }
2676         extent_root->track_dirty = 1;
2677         fs_info->extent_root = extent_root;
2678
2679         location.objectid = BTRFS_DEV_TREE_OBJECTID;
2680         dev_root = btrfs_read_tree_root(tree_root, &location);
2681         if (IS_ERR(dev_root)) {
2682                 ret = PTR_ERR(dev_root);
2683                 goto recovery_tree_root;
2684         }
2685         dev_root->track_dirty = 1;
2686         fs_info->dev_root = dev_root;
2687         btrfs_init_devices_late(fs_info);
2688
2689         location.objectid = BTRFS_CSUM_TREE_OBJECTID;
2690         csum_root = btrfs_read_tree_root(tree_root, &location);
2691         if (IS_ERR(csum_root)) {
2692                 ret = PTR_ERR(csum_root);
2693                 goto recovery_tree_root;
2694         }
2695         csum_root->track_dirty = 1;
2696         fs_info->csum_root = csum_root;
2697
2698         location.objectid = BTRFS_QUOTA_TREE_OBJECTID;
2699         quota_root = btrfs_read_tree_root(tree_root, &location);
2700         if (!IS_ERR(quota_root)) {
2701                 quota_root->track_dirty = 1;
2702                 fs_info->quota_enabled = 1;
2703                 fs_info->pending_quota_state = 1;
2704                 fs_info->quota_root = quota_root;
2705         }
2706
2707         location.objectid = BTRFS_UUID_TREE_OBJECTID;
2708         uuid_root = btrfs_read_tree_root(tree_root, &location);
2709         if (IS_ERR(uuid_root)) {
2710                 ret = PTR_ERR(uuid_root);
2711                 if (ret != -ENOENT)
2712                         goto recovery_tree_root;
2713                 create_uuid_tree = true;
2714                 check_uuid_tree = false;
2715         } else {
2716                 uuid_root->track_dirty = 1;
2717                 fs_info->uuid_root = uuid_root;
2718                 create_uuid_tree = false;
2719                 check_uuid_tree =
2720                     generation != btrfs_super_uuid_tree_generation(disk_super);
2721         }
2722
2723         fs_info->generation = generation;
2724         fs_info->last_trans_committed = generation;
2725
2726         ret = btrfs_recover_balance(fs_info);
2727         if (ret) {
2728                 printk(KERN_WARNING "btrfs: failed to recover balance\n");
2729                 goto fail_block_groups;
2730         }
2731
2732         ret = btrfs_init_dev_stats(fs_info);
2733         if (ret) {
2734                 printk(KERN_ERR "btrfs: failed to init dev_stats: %d\n",
2735                        ret);
2736                 goto fail_block_groups;
2737         }
2738
2739         ret = btrfs_init_dev_replace(fs_info);
2740         if (ret) {
2741                 pr_err("btrfs: failed to init dev_replace: %d\n", ret);
2742                 goto fail_block_groups;
2743         }
2744
2745         btrfs_close_extra_devices(fs_info, fs_devices, 1);
2746
2747         ret = btrfs_sysfs_add_one(fs_info);
2748         if (ret) {
2749                 pr_err("btrfs: failed to init sysfs interface: %d\n", ret);
2750                 goto fail_block_groups;
2751         }
2752
2753         ret = btrfs_init_space_info(fs_info);
2754         if (ret) {
2755                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2756                 goto fail_block_groups;
2757         }
2758
2759         ret = btrfs_read_block_groups(extent_root);
2760         if (ret) {
2761                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2762                 goto fail_block_groups;
2763         }
2764         fs_info->num_tolerated_disk_barrier_failures =
2765                 btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
2766         if (fs_info->fs_devices->missing_devices >
2767              fs_info->num_tolerated_disk_barrier_failures &&
2768             !(sb->s_flags & MS_RDONLY)) {
2769                 printk(KERN_WARNING
2770                        "Btrfs: too many missing devices, writeable mount is not allowed\n");
2771                 goto fail_block_groups;
2772         }
2773
2774         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2775                                                "btrfs-cleaner");
2776         if (IS_ERR(fs_info->cleaner_kthread))
2777                 goto fail_block_groups;
2778
2779         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2780                                                    tree_root,
2781                                                    "btrfs-transaction");
2782         if (IS_ERR(fs_info->transaction_kthread))
2783                 goto fail_cleaner;
2784
2785         if (!btrfs_test_opt(tree_root, SSD) &&
2786             !btrfs_test_opt(tree_root, NOSSD) &&
2787             !fs_info->fs_devices->rotating) {
2788                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2789                        "mode\n");
2790                 btrfs_set_opt(fs_info->mount_opt, SSD);
2791         }
2792
2793 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
2794         if (btrfs_test_opt(tree_root, CHECK_INTEGRITY)) {
2795                 ret = btrfsic_mount(tree_root, fs_devices,
2796                                     btrfs_test_opt(tree_root,
2797                                         CHECK_INTEGRITY_INCLUDING_EXTENT_DATA) ?
2798                                     1 : 0,
2799                                     fs_info->check_integrity_print_mask);
2800                 if (ret)
2801                         printk(KERN_WARNING "btrfs: failed to initialize"
2802                                " integrity check module %s\n", sb->s_id);
2803         }
2804 #endif
2805         ret = btrfs_read_qgroup_config(fs_info);
2806         if (ret)
2807                 goto fail_trans_kthread;
2808
2809         /* do not make disk changes in broken FS */
2810         if (btrfs_super_log_root(disk_super) != 0) {
2811                 u64 bytenr = btrfs_super_log_root(disk_super);
2812
2813                 if (fs_devices->rw_devices == 0) {
2814                         printk(KERN_WARNING "Btrfs log replay required "
2815                                "on RO media\n");
2816                         err = -EIO;
2817                         goto fail_qgroup;
2818                 }
2819                 blocksize =
2820                      btrfs_level_size(tree_root,
2821                                       btrfs_super_log_root_level(disk_super));
2822
2823                 log_tree_root = btrfs_alloc_root(fs_info);
2824                 if (!log_tree_root) {
2825                         err = -ENOMEM;
2826                         goto fail_qgroup;
2827                 }
2828
2829                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2830                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2831
2832                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2833                                                       blocksize,
2834                                                       generation + 1);
2835                 if (!log_tree_root->node ||
2836                     !extent_buffer_uptodate(log_tree_root->node)) {
2837                         printk(KERN_ERR "btrfs: failed to read log tree\n");
2838                         free_extent_buffer(log_tree_root->node);
2839                         kfree(log_tree_root);
2840                         goto fail_trans_kthread;
2841                 }
2842                 /* returns with log_tree_root freed on success */
2843                 ret = btrfs_recover_log_trees(log_tree_root);
2844                 if (ret) {
2845                         btrfs_error(tree_root->fs_info, ret,
2846                                     "Failed to recover log tree");
2847                         free_extent_buffer(log_tree_root->node);
2848                         kfree(log_tree_root);
2849                         goto fail_trans_kthread;
2850                 }
2851
2852                 if (sb->s_flags & MS_RDONLY) {
2853                         ret = btrfs_commit_super(tree_root);
2854                         if (ret)
2855                                 goto fail_trans_kthread;
2856                 }
2857         }
2858
2859         ret = btrfs_find_orphan_roots(tree_root);
2860         if (ret)
2861                 goto fail_trans_kthread;
2862
2863         if (!(sb->s_flags & MS_RDONLY)) {
2864                 ret = btrfs_cleanup_fs_roots(fs_info);
2865                 if (ret)
2866                         goto fail_trans_kthread;
2867
2868                 ret = btrfs_recover_relocation(tree_root);
2869                 if (ret < 0) {
2870                         printk(KERN_WARNING
2871                                "btrfs: failed to recover relocation\n");
2872                         err = -EINVAL;
2873                         goto fail_qgroup;
2874                 }
2875         }
2876
2877         location.objectid = BTRFS_FS_TREE_OBJECTID;
2878         location.type = BTRFS_ROOT_ITEM_KEY;
2879         location.offset = 0;
2880
2881         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2882         if (IS_ERR(fs_info->fs_root)) {
2883                 err = PTR_ERR(fs_info->fs_root);
2884                 goto fail_qgroup;
2885         }
2886
2887         if (sb->s_flags & MS_RDONLY)
2888                 return 0;
2889
2890         down_read(&fs_info->cleanup_work_sem);
2891         if ((ret = btrfs_orphan_cleanup(fs_info->fs_root)) ||
2892             (ret = btrfs_orphan_cleanup(fs_info->tree_root))) {
2893                 up_read(&fs_info->cleanup_work_sem);
2894                 close_ctree(tree_root);
2895                 return ret;
2896         }
2897         up_read(&fs_info->cleanup_work_sem);
2898
2899         ret = btrfs_resume_balance_async(fs_info);
2900         if (ret) {
2901                 printk(KERN_WARNING "btrfs: failed to resume balance\n");
2902                 close_ctree(tree_root);
2903                 return ret;
2904         }
2905
2906         ret = btrfs_resume_dev_replace_async(fs_info);
2907         if (ret) {
2908                 pr_warn("btrfs: failed to resume dev_replace\n");
2909                 close_ctree(tree_root);
2910                 return ret;
2911         }
2912
2913         btrfs_qgroup_rescan_resume(fs_info);
2914
2915         if (create_uuid_tree) {
2916                 pr_info("btrfs: creating UUID tree\n");
2917                 ret = btrfs_create_uuid_tree(fs_info);
2918                 if (ret) {
2919                         pr_warn("btrfs: failed to create the UUID tree %d\n",
2920                                 ret);
2921                         close_ctree(tree_root);
2922                         return ret;
2923                 }
2924         } else if (check_uuid_tree ||
2925                    btrfs_test_opt(tree_root, RESCAN_UUID_TREE)) {
2926                 pr_info("btrfs: checking UUID tree\n");
2927                 ret = btrfs_check_uuid_tree(fs_info);
2928                 if (ret) {
2929                         pr_warn("btrfs: failed to check the UUID tree %d\n",
2930                                 ret);
2931                         close_ctree(tree_root);
2932                         return ret;
2933                 }
2934         } else {
2935                 fs_info->update_uuid_tree_gen = 1;
2936         }
2937
2938         return 0;
2939
2940 fail_qgroup:
2941         btrfs_free_qgroup_config(fs_info);
2942 fail_trans_kthread:
2943         kthread_stop(fs_info->transaction_kthread);
2944         btrfs_cleanup_transaction(fs_info->tree_root);
2945         del_fs_roots(fs_info);
2946 fail_cleaner:
2947         kthread_stop(fs_info->cleaner_kthread);
2948
2949         /*
2950          * make sure we're done with the btree inode before we stop our
2951          * kthreads
2952          */
2953         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2954
2955 fail_block_groups:
2956         btrfs_put_block_group_cache(fs_info);
2957         btrfs_free_block_groups(fs_info);
2958
2959 fail_tree_roots:
2960         free_root_pointers(fs_info, 1);
2961         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2962
2963 fail_sb_buffer:
2964         btrfs_stop_all_workers(fs_info);
2965 fail_alloc:
2966 fail_iput:
2967         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2968
2969         iput(fs_info->btree_inode);
2970 fail_delalloc_bytes:
2971         percpu_counter_destroy(&fs_info->delalloc_bytes);
2972 fail_dirty_metadata_bytes:
2973         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
2974 fail_bdi:
2975         bdi_destroy(&fs_info->bdi);
2976 fail_srcu:
2977         cleanup_srcu_struct(&fs_info->subvol_srcu);
2978 fail:
2979         btrfs_free_stripe_hash_table(fs_info);
2980         btrfs_close_devices(fs_info->fs_devices);
2981         return err;
2982
2983 recovery_tree_root:
2984         if (!btrfs_test_opt(tree_root, RECOVERY))
2985                 goto fail_tree_roots;
2986
2987         free_root_pointers(fs_info, 0);
2988
2989         /* don't use the log in recovery mode, it won't be valid */
2990         btrfs_set_super_log_root(disk_super, 0);
2991
2992         /* we can't trust the free space cache either */
2993         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2994
2995         ret = next_root_backup(fs_info, fs_info->super_copy,
2996                                &num_backups_tried, &backup_index);
2997         if (ret == -1)
2998                 goto fail_block_groups;
2999         goto retry_root_backup;
3000 }
3001
3002 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
3003 {
3004         if (uptodate) {
3005                 set_buffer_uptodate(bh);
3006         } else {
3007                 struct btrfs_device *device = (struct btrfs_device *)
3008                         bh->b_private;
3009
3010                 printk_ratelimited_in_rcu(KERN_WARNING "lost page write due to "
3011                                           "I/O error on %s\n",
3012                                           rcu_str_deref(device->name));
3013                 /* note, we dont' set_buffer_write_io_error because we have
3014                  * our own ways of dealing with the IO errors
3015                  */
3016                 clear_buffer_uptodate(bh);
3017                 btrfs_dev_stat_inc_and_print(device, BTRFS_DEV_STAT_WRITE_ERRS);
3018         }
3019         unlock_buffer(bh);
3020         put_bh(bh);
3021 }
3022
3023 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
3024 {
3025         struct buffer_head *bh;
3026         struct buffer_head *latest = NULL;
3027         struct btrfs_super_block *super;
3028         int i;
3029         u64 transid = 0;
3030         u64 bytenr;
3031
3032         /* we would like to check all the supers, but that would make
3033          * a btrfs mount succeed after a mkfs from a different FS.
3034          * So, we need to add a special mount option to scan for
3035          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
3036          */
3037         for (i = 0; i < 1; i++) {
3038                 bytenr = btrfs_sb_offset(i);
3039                 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
3040                                         i_size_read(bdev->bd_inode))
3041                         break;
3042                 bh = __bread(bdev, bytenr / 4096,
3043                                         BTRFS_SUPER_INFO_SIZE);
3044                 if (!bh)
3045                         continue;
3046
3047                 super = (struct btrfs_super_block *)bh->b_data;
3048                 if (btrfs_super_bytenr(super) != bytenr ||
3049                     btrfs_super_magic(super) != BTRFS_MAGIC) {
3050                         brelse(bh);
3051                         continue;
3052                 }
3053
3054                 if (!latest || btrfs_super_generation(super) > transid) {
3055                         brelse(latest);
3056                         latest = bh;
3057                         transid = btrfs_super_generation(super);
3058                 } else {
3059                         brelse(bh);
3060                 }
3061         }
3062         return latest;
3063 }
3064
3065 /*
3066  * this should be called twice, once with wait == 0 and
3067  * once with wait == 1.  When wait == 0 is done, all the buffer heads
3068  * we write are pinned.
3069  *
3070  * They are released when wait == 1 is done.
3071  * max_mirrors must be the same for both runs, and it indicates how
3072  * many supers on this one device should be written.
3073  *
3074  * max_mirrors == 0 means to write them all.
3075  */
3076 static int write_dev_supers(struct btrfs_device *device,
3077                             struct btrfs_super_block *sb,
3078                             int do_barriers, int wait, int max_mirrors)
3079 {
3080         struct buffer_head *bh;
3081         int i;
3082         int ret;
3083         int errors = 0;
3084         u32 crc;
3085         u64 bytenr;
3086
3087         if (max_mirrors == 0)
3088                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
3089
3090         for (i = 0; i < max_mirrors; i++) {
3091                 bytenr = btrfs_sb_offset(i);
3092                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
3093                         break;
3094
3095                 if (wait) {
3096                         bh = __find_get_block(device->bdev, bytenr / 4096,
3097                                               BTRFS_SUPER_INFO_SIZE);
3098                         if (!bh) {
3099                                 errors++;
3100                                 continue;
3101                         }
3102                         wait_on_buffer(bh);
3103                         if (!buffer_uptodate(bh))
3104                                 errors++;
3105
3106                         /* drop our reference */
3107                         brelse(bh);
3108
3109                         /* drop the reference from the wait == 0 run */
3110                         brelse(bh);
3111                         continue;
3112                 } else {
3113                         btrfs_set_super_bytenr(sb, bytenr);
3114
3115                         crc = ~(u32)0;
3116                         crc = btrfs_csum_data((char *)sb +
3117                                               BTRFS_CSUM_SIZE, crc,
3118                                               BTRFS_SUPER_INFO_SIZE -
3119                                               BTRFS_CSUM_SIZE);
3120                         btrfs_csum_final(crc, sb->csum);
3121
3122                         /*
3123                          * one reference for us, and we leave it for the
3124                          * caller
3125                          */
3126                         bh = __getblk(device->bdev, bytenr / 4096,
3127                                       BTRFS_SUPER_INFO_SIZE);
3128                         if (!bh) {
3129                                 printk(KERN_ERR "btrfs: couldn't get super "
3130                                        "buffer head for bytenr %Lu\n", bytenr);
3131                                 errors++;
3132                                 continue;
3133                         }
3134
3135                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
3136
3137                         /* one reference for submit_bh */
3138                         get_bh(bh);
3139
3140                         set_buffer_uptodate(bh);
3141                         lock_buffer(bh);
3142                         bh->b_end_io = btrfs_end_buffer_write_sync;
3143                         bh->b_private = device;
3144                 }
3145
3146                 /*
3147                  * we fua the first super.  The others we allow
3148                  * to go down lazy.
3149                  */
3150                 ret = btrfsic_submit_bh(WRITE_FUA, bh);
3151                 if (ret)
3152                         errors++;
3153         }
3154         return errors < i ? 0 : -1;
3155 }
3156
3157 /*
3158  * endio for the write_dev_flush, this will wake anyone waiting
3159  * for the barrier when it is done
3160  */
3161 static void btrfs_end_empty_barrier(struct bio *bio, int err)
3162 {
3163         if (err) {
3164                 if (err == -EOPNOTSUPP)
3165                         set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
3166                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
3167         }
3168         if (bio->bi_private)
3169                 complete(bio->bi_private);
3170         bio_put(bio);
3171 }
3172
3173 /*
3174  * trigger flushes for one the devices.  If you pass wait == 0, the flushes are
3175  * sent down.  With wait == 1, it waits for the previous flush.
3176  *
3177  * any device where the flush fails with eopnotsupp are flagged as not-barrier
3178  * capable
3179  */
3180 static int write_dev_flush(struct btrfs_device *device, int wait)
3181 {
3182         struct bio *bio;
3183         int ret = 0;
3184
3185         if (device->nobarriers)
3186                 return 0;
3187
3188         if (wait) {
3189                 bio = device->flush_bio;
3190                 if (!bio)
3191                         return 0;
3192
3193                 wait_for_completion(&device->flush_wait);
3194
3195                 if (bio_flagged(bio, BIO_EOPNOTSUPP)) {
3196                         printk_in_rcu("btrfs: disabling barriers on dev %s\n",
3197                                       rcu_str_deref(device->name));
3198                         device->nobarriers = 1;
3199                 } else if (!bio_flagged(bio, BIO_UPTODATE)) {
3200                         ret = -EIO;
3201                         btrfs_dev_stat_inc_and_print(device,
3202                                 BTRFS_DEV_STAT_FLUSH_ERRS);
3203                 }
3204
3205                 /* drop the reference from the wait == 0 run */
3206                 bio_put(bio);
3207                 device->flush_bio = NULL;
3208
3209                 return ret;
3210         }
3211
3212         /*
3213          * one reference for us, and we leave it for the
3214          * caller
3215          */
3216         device->flush_bio = NULL;
3217         bio = btrfs_io_bio_alloc(GFP_NOFS, 0);
3218         if (!bio)
3219                 return -ENOMEM;
3220
3221         bio->bi_end_io = btrfs_end_empty_barrier;
3222         bio->bi_bdev = device->bdev;
3223         init_completion(&device->flush_wait);
3224         bio->bi_private = &device->flush_wait;
3225         device->flush_bio = bio;
3226
3227         bio_get(bio);
3228         btrfsic_submit_bio(WRITE_FLUSH, bio);
3229
3230         return 0;
3231 }
3232
3233 /*
3234  * send an empty flush down to each device in parallel,
3235  * then wait for them
3236  */
3237 static int barrier_all_devices(struct btrfs_fs_info *info)
3238 {
3239         struct list_head *head;
3240         struct btrfs_device *dev;
3241         int errors_send = 0;
3242         int errors_wait = 0;
3243         int ret;
3244
3245         /* send down all the barriers */
3246         head = &info->fs_devices->devices;
3247         list_for_each_entry_rcu(dev, head, dev_list) {
3248                 if (!dev->bdev) {
3249                         errors_send++;
3250                         continue;
3251                 }
3252                 if (!dev->in_fs_metadata || !dev->writeable)
3253                         continue;
3254
3255                 ret = write_dev_flush(dev, 0);
3256                 if (ret)
3257                         errors_send++;
3258         }
3259
3260         /* wait for all the barriers */
3261         list_for_each_entry_rcu(dev, head, dev_list) {
3262                 if (!dev->bdev) {
3263                         errors_wait++;
3264                         continue;
3265                 }
3266                 if (!dev->in_fs_metadata || !dev->writeable)
3267                         continue;
3268
3269                 ret = write_dev_flush(dev, 1);
3270                 if (ret)
3271                         errors_wait++;
3272         }
3273         if (errors_send > info->num_tolerated_disk_barrier_failures ||
3274             errors_wait > info->num_tolerated_disk_barrier_failures)
3275                 return -EIO;
3276         return 0;
3277 }
3278
3279 int btrfs_calc_num_tolerated_disk_barrier_failures(
3280         struct btrfs_fs_info *fs_info)
3281 {
3282         struct btrfs_ioctl_space_info space;
3283         struct btrfs_space_info *sinfo;
3284         u64 types[] = {BTRFS_BLOCK_GROUP_DATA,
3285                        BTRFS_BLOCK_GROUP_SYSTEM,
3286                        BTRFS_BLOCK_GROUP_METADATA,
3287                        BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA};
3288         int num_types = 4;
3289         int i;
3290         int c;
3291         int num_tolerated_disk_barrier_failures =
3292                 (int)fs_info->fs_devices->num_devices;
3293
3294         for (i = 0; i < num_types; i++) {
3295                 struct btrfs_space_info *tmp;
3296
3297                 sinfo = NULL;
3298                 rcu_read_lock();
3299                 list_for_each_entry_rcu(tmp, &fs_info->space_info, list) {
3300                         if (tmp->flags == types[i]) {
3301                                 sinfo = tmp;
3302                                 break;
3303                         }
3304                 }
3305                 rcu_read_unlock();
3306
3307                 if (!sinfo)
3308                         continue;
3309
3310                 down_read(&sinfo->groups_sem);
3311                 for (c = 0; c < BTRFS_NR_RAID_TYPES; c++) {
3312                         if (!list_empty(&sinfo->block_groups[c])) {
3313                                 u64 flags;
3314
3315                                 btrfs_get_block_group_info(
3316                                         &sinfo->block_groups[c], &space);
3317                                 if (space.total_bytes == 0 ||
3318                                     space.used_bytes == 0)
3319                                         continue;
3320                                 flags = space.flags;
3321                                 /*
3322                                  * return
3323                                  * 0: if dup, single or RAID0 is configured for
3324                                  *    any of metadata, system or data, else
3325                                  * 1: if RAID5 is configured, or if RAID1 or
3326                                  *    RAID10 is configured and only two mirrors
3327                                  *    are used, else
3328                                  * 2: if RAID6 is configured, else
3329                                  * num_mirrors - 1: if RAID1 or RAID10 is
3330                                  *                  configured and more than
3331                                  *                  2 mirrors are used.
3332                                  */
3333                                 if (num_tolerated_disk_barrier_failures > 0 &&
3334                                     ((flags & (BTRFS_BLOCK_GROUP_DUP |
3335                                                BTRFS_BLOCK_GROUP_RAID0)) ||
3336                                      ((flags & BTRFS_BLOCK_GROUP_PROFILE_MASK)
3337                                       == 0)))
3338                                         num_tolerated_disk_barrier_failures = 0;
3339                                 else if (num_tolerated_disk_barrier_failures > 1) {
3340                                         if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
3341                                             BTRFS_BLOCK_GROUP_RAID5 |
3342                                             BTRFS_BLOCK_GROUP_RAID10)) {
3343                                                 num_tolerated_disk_barrier_failures = 1;
3344                                         } else if (flags &
3345                                                    BTRFS_BLOCK_GROUP_RAID6) {
3346                                                 num_tolerated_disk_barrier_failures = 2;
3347                                         }
3348                                 }
3349                         }
3350                 }
3351                 up_read(&sinfo->groups_sem);
3352         }
3353
3354         return num_tolerated_disk_barrier_failures;
3355 }
3356
3357 static int write_all_supers(struct btrfs_root *root, int max_mirrors)
3358 {
3359         struct list_head *head;
3360         struct btrfs_device *dev;
3361         struct btrfs_super_block *sb;
3362         struct btrfs_dev_item *dev_item;
3363         int ret;
3364         int do_barriers;
3365         int max_errors;
3366         int total_errors = 0;
3367         u64 flags;
3368
3369         do_barriers = !btrfs_test_opt(root, NOBARRIER);
3370         backup_super_roots(root->fs_info);
3371
3372         sb = root->fs_info->super_for_commit;
3373         dev_item = &sb->dev_item;
3374
3375         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
3376         head = &root->fs_info->fs_devices->devices;
3377         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
3378
3379         if (do_barriers) {
3380                 ret = barrier_all_devices(root->fs_info);
3381                 if (ret) {
3382                         mutex_unlock(
3383                                 &root->fs_info->fs_devices->device_list_mutex);
3384                         btrfs_error(root->fs_info, ret,
3385                                     "errors while submitting device barriers.");
3386                         return ret;
3387                 }
3388         }
3389
3390         list_for_each_entry_rcu(dev, head, dev_list) {
3391                 if (!dev->bdev) {
3392                         total_errors++;
3393                         continue;
3394                 }
3395                 if (!dev->in_fs_metadata || !dev->writeable)
3396                         continue;
3397
3398                 btrfs_set_stack_device_generation(dev_item, 0);
3399                 btrfs_set_stack_device_type(dev_item, dev->type);
3400                 btrfs_set_stack_device_id(dev_item, dev->devid);
3401                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
3402                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
3403                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
3404                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
3405                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
3406                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
3407                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
3408
3409                 flags = btrfs_super_flags(sb);
3410                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
3411
3412                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
3413                 if (ret)
3414                         total_errors++;
3415         }
3416         if (total_errors > max_errors) {
3417                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
3418                        total_errors);
3419                 mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3420
3421                 /* FUA is masked off if unsupported and can't be the reason */
3422                 btrfs_error(root->fs_info, -EIO,
3423                             "%d errors while writing supers", total_errors);
3424                 return -EIO;
3425         }
3426
3427         total_errors = 0;
3428         list_for_each_entry_rcu(dev, head, dev_list) {
3429                 if (!dev->bdev)
3430                         continue;
3431                 if (!dev->in_fs_metadata || !dev->writeable)
3432                         continue;
3433
3434                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
3435                 if (ret)
3436                         total_errors++;
3437         }
3438         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
3439         if (total_errors > max_errors) {
3440                 btrfs_error(root->fs_info, -EIO,
3441                             "%d errors while writing supers", total_errors);
3442                 return -EIO;
3443         }
3444         return 0;
3445 }
3446
3447 int write_ctree_super(struct btrfs_trans_handle *trans,
3448                       struct btrfs_root *root, int max_mirrors)
3449 {
3450         return write_all_supers(root, max_mirrors);
3451 }
3452
3453 /* Drop a fs root from the radix tree and free it. */
3454 void btrfs_drop_and_free_fs_root(struct btrfs_fs_info *fs_info,
3455                                   struct btrfs_root *root)
3456 {
3457         spin_lock(&fs_info->fs_roots_radix_lock);
3458         radix_tree_delete(&fs_info->fs_roots_radix,
3459                           (unsigned long)root->root_key.objectid);
3460         spin_unlock(&fs_info->fs_roots_radix_lock);
3461
3462         if (btrfs_root_refs(&root->root_item) == 0)
3463                 synchronize_srcu(&fs_info->subvol_srcu);
3464
3465         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) {
3466                 btrfs_free_log(NULL, root);
3467                 btrfs_free_log_root_tree(NULL, fs_info);
3468         }
3469
3470         __btrfs_remove_free_space_cache(root->free_ino_pinned);
3471         __btrfs_remove_free_space_cache(root->free_ino_ctl);
3472         free_fs_root(root);
3473 }
3474
3475 static void free_fs_root(struct btrfs_root *root)
3476 {
3477         iput(root->cache_inode);
3478         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
3479         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3480         root->orphan_block_rsv = NULL;
3481         if (root->anon_dev)
3482                 free_anon_bdev(root->anon_dev);
3483         free_extent_buffer(root->node);
3484         free_extent_buffer(root->commit_root);
3485         kfree(root->free_ino_ctl);
3486         kfree(root->free_ino_pinned);
3487         kfree(root->name);
3488         btrfs_put_fs_root(root);
3489 }
3490
3491 void btrfs_free_fs_root(struct btrfs_root *root)
3492 {
3493         free_fs_root(root);
3494 }
3495
3496 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
3497 {
3498         u64 root_objectid = 0;
3499         struct btrfs_root *gang[8];
3500         int i;
3501         int ret;
3502
3503         while (1) {
3504                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
3505                                              (void **)gang, root_objectid,
3506                                              ARRAY_SIZE(gang));
3507                 if (!ret)
3508                         break;
3509
3510                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
3511                 for (i = 0; i < ret; i++) {
3512                         int err;
3513
3514                         root_objectid = gang[i]->root_key.objectid;
3515                         err = btrfs_orphan_cleanup(gang[i]);
3516                         if (err)
3517                                 return err;
3518                 }
3519                 root_objectid++;
3520         }
3521         return 0;
3522 }
3523
3524 int btrfs_commit_super(struct btrfs_root *root)
3525 {
3526         struct btrfs_trans_handle *trans;
3527
3528         mutex_lock(&root->fs_info->cleaner_mutex);
3529         btrfs_run_delayed_iputs(root);
3530         mutex_unlock(&root->fs_info->cleaner_mutex);
3531         wake_up_process(root->fs_info->cleaner_kthread);
3532
3533         /* wait until ongoing cleanup work done */
3534         down_write(&root->fs_info->cleanup_work_sem);
3535         up_write(&root->fs_info->cleanup_work_sem);
3536
3537         trans = btrfs_join_transaction(root);
3538         if (IS_ERR(trans))
3539                 return PTR_ERR(trans);
3540         return btrfs_commit_transaction(trans, root);
3541 }
3542
3543 int close_ctree(struct btrfs_root *root)
3544 {
3545         struct btrfs_fs_info *fs_info = root->fs_info;
3546         int ret;
3547
3548         fs_info->closing = 1;
3549         smp_mb();
3550
3551         /* wait for the uuid_scan task to finish */
3552         down(&fs_info->uuid_tree_rescan_sem);
3553         /* avoid complains from lockdep et al., set sem back to initial state */
3554         up(&fs_info->uuid_tree_rescan_sem);
3555
3556         /* pause restriper - we want to resume on mount */
3557         btrfs_pause_balance(fs_info);
3558
3559         btrfs_dev_replace_suspend_for_unmount(fs_info);
3560
3561         btrfs_scrub_cancel(fs_info);
3562
3563         /* wait for any defraggers to finish */
3564         wait_event(fs_info->transaction_wait,
3565                    (atomic_read(&fs_info->defrag_running) == 0));
3566
3567         /* clear out the rbtree of defraggable inodes */
3568         btrfs_cleanup_defrag_inodes(fs_info);
3569
3570         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
3571                 ret = btrfs_commit_super(root);
3572                 if (ret)
3573                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
3574         }
3575
3576         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
3577                 btrfs_error_commit_super(root);
3578
3579         btrfs_put_block_group_cache(fs_info);
3580
3581         kthread_stop(fs_info->transaction_kthread);
3582         kthread_stop(fs_info->cleaner_kthread);
3583
3584         fs_info->closing = 2;
3585         smp_mb();
3586
3587         btrfs_free_qgroup_config(root->fs_info);
3588
3589         if (percpu_counter_sum(&fs_info->delalloc_bytes)) {
3590                 printk(KERN_INFO "btrfs: at unmount delalloc count %lld\n",
3591                        percpu_counter_sum(&fs_info->delalloc_bytes));
3592         }
3593
3594         btrfs_sysfs_remove_one(fs_info);
3595
3596         del_fs_roots(fs_info);
3597
3598         btrfs_free_block_groups(fs_info);
3599
3600         btrfs_stop_all_workers(fs_info);
3601
3602         free_root_pointers(fs_info, 1);
3603
3604         iput(fs_info->btree_inode);
3605
3606 #ifdef CONFIG_BTRFS_FS_CHECK_INTEGRITY
3607         if (btrfs_test_opt(root, CHECK_INTEGRITY))
3608                 btrfsic_unmount(root, fs_info->fs_devices);
3609 #endif
3610
3611         btrfs_close_devices(fs_info->fs_devices);
3612         btrfs_mapping_tree_free(&fs_info->mapping_tree);
3613
3614         percpu_counter_destroy(&fs_info->dirty_metadata_bytes);
3615         percpu_counter_destroy(&fs_info->delalloc_bytes);
3616         bdi_destroy(&fs_info->bdi);
3617         cleanup_srcu_struct(&fs_info->subvol_srcu);
3618
3619         btrfs_free_stripe_hash_table(fs_info);
3620
3621         btrfs_free_block_rsv(root, root->orphan_block_rsv);
3622         root->orphan_block_rsv = NULL;
3623
3624         return 0;
3625 }
3626
3627 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid,
3628                           int atomic)
3629 {
3630         int ret;
3631         struct inode *btree_inode = buf->pages[0]->mapping->host;
3632
3633         ret = extent_buffer_uptodate(buf);
3634         if (!ret)
3635                 return ret;
3636
3637         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
3638                                     parent_transid, atomic);
3639         if (ret == -EAGAIN)
3640                 return ret;
3641         return !ret;
3642 }
3643
3644 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
3645 {
3646         return set_extent_buffer_uptodate(buf);
3647 }
3648
3649 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
3650 {
3651         struct btrfs_root *root;
3652         u64 transid = btrfs_header_generation(buf);
3653         int was_dirty;
3654
3655 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
3656         /*
3657          * This is a fast path so only do this check if we have sanity tests
3658          * enabled.  Normal people shouldn't be marking dummy buffers as dirty
3659          * outside of the sanity tests.
3660          */
3661         if (unlikely(test_bit(EXTENT_BUFFER_DUMMY, &buf->bflags)))
3662                 return;
3663 #endif
3664         root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3665         btrfs_assert_tree_locked(buf);
3666         if (transid != root->fs_info->generation)
3667                 WARN(1, KERN_CRIT "btrfs transid mismatch buffer %llu, "
3668                        "found %llu running %llu\n",
3669                         buf->start, transid, root->fs_info->generation);
3670         was_dirty = set_extent_buffer_dirty(buf);
3671         if (!was_dirty)
3672                 __percpu_counter_add(&root->fs_info->dirty_metadata_bytes,
3673                                      buf->len,
3674                                      root->fs_info->dirty_metadata_batch);
3675 }
3676
3677 static void __btrfs_btree_balance_dirty(struct btrfs_root *root,
3678                                         int flush_delayed)
3679 {
3680         /*
3681          * looks as though older kernels can get into trouble with
3682          * this code, they end up stuck in balance_dirty_pages forever
3683          */
3684         int ret;
3685
3686         if (current->flags & PF_MEMALLOC)
3687                 return;
3688
3689         if (flush_delayed)
3690                 btrfs_balance_delayed_items(root);
3691
3692         ret = percpu_counter_compare(&root->fs_info->dirty_metadata_bytes,
3693                                      BTRFS_DIRTY_METADATA_THRESH);
3694         if (ret > 0) {
3695                 balance_dirty_pages_ratelimited(
3696                                    root->fs_info->btree_inode->i_mapping);
3697         }
3698         return;
3699 }
3700
3701 void btrfs_btree_balance_dirty(struct btrfs_root *root)
3702 {
3703         __btrfs_btree_balance_dirty(root, 1);
3704 }
3705
3706 void btrfs_btree_balance_dirty_nodelay(struct btrfs_root *root)
3707 {
3708         __btrfs_btree_balance_dirty(root, 0);
3709 }
3710
3711 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
3712 {
3713         struct btrfs_root *root = BTRFS_I(buf->pages[0]->mapping->host)->root;
3714         return btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
3715 }
3716
3717 static int btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3718                               int read_only)
3719 {
3720         /*
3721          * Placeholder for checks
3722          */
3723         return 0;
3724 }
3725
3726 static void btrfs_error_commit_super(struct btrfs_root *root)
3727 {
3728         mutex_lock(&root->fs_info->cleaner_mutex);
3729         btrfs_run_delayed_iputs(root);
3730         mutex_unlock(&root->fs_info->cleaner_mutex);
3731
3732         down_write(&root->fs_info->cleanup_work_sem);
3733         up_write(&root->fs_info->cleanup_work_sem);
3734
3735         /* cleanup FS via transaction */
3736         btrfs_cleanup_transaction(root);
3737 }
3738
3739 static void btrfs_destroy_ordered_operations(struct btrfs_transaction *t,
3740                                              struct btrfs_root *root)
3741 {
3742         struct btrfs_inode *btrfs_inode;
3743         struct list_head splice;
3744
3745         INIT_LIST_HEAD(&splice);
3746
3747         mutex_lock(&root->fs_info->ordered_operations_mutex);
3748         spin_lock(&root->fs_info->ordered_root_lock);
3749
3750         list_splice_init(&t->ordered_operations, &splice);
3751         while (!list_empty(&splice)) {
3752                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3753                                          ordered_operations);
3754
3755                 list_del_init(&btrfs_inode->ordered_operations);
3756                 spin_unlock(&root->fs_info->ordered_root_lock);
3757
3758                 btrfs_invalidate_inodes(btrfs_inode->root);
3759
3760                 spin_lock(&root->fs_info->ordered_root_lock);
3761         }
3762
3763         spin_unlock(&root->fs_info->ordered_root_lock);
3764         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3765 }
3766
3767 static void btrfs_destroy_ordered_extents(struct btrfs_root *root)
3768 {
3769         struct btrfs_ordered_extent *ordered;
3770
3771         spin_lock(&root->ordered_extent_lock);
3772         /*
3773          * This will just short circuit the ordered completion stuff which will
3774          * make sure the ordered extent gets properly cleaned up.
3775          */
3776         list_for_each_entry(ordered, &root->ordered_extents,
3777                             root_extent_list)
3778                 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags);
3779         spin_unlock(&root->ordered_extent_lock);
3780 }
3781
3782 static void btrfs_destroy_all_ordered_extents(struct btrfs_fs_info *fs_info)
3783 {
3784         struct btrfs_root *root;
3785         struct list_head splice;
3786
3787         INIT_LIST_HEAD(&splice);
3788
3789         spin_lock(&fs_info->ordered_root_lock);
3790         list_splice_init(&fs_info->ordered_roots, &splice);
3791         while (!list_empty(&splice)) {
3792                 root = list_first_entry(&splice, struct btrfs_root,
3793                                         ordered_root);
3794                 list_move_tail(&root->ordered_root,
3795                                &fs_info->ordered_roots);
3796
3797                 btrfs_destroy_ordered_extents(root);
3798
3799                 cond_resched_lock(&fs_info->ordered_root_lock);
3800         }
3801         spin_unlock(&fs_info->ordered_root_lock);
3802 }
3803
3804 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3805                                       struct btrfs_root *root)
3806 {
3807         struct rb_node *node;
3808         struct btrfs_delayed_ref_root *delayed_refs;
3809         struct btrfs_delayed_ref_node *ref;
3810         int ret = 0;
3811
3812         delayed_refs = &trans->delayed_refs;
3813
3814         spin_lock(&delayed_refs->lock);
3815         if (delayed_refs->num_entries == 0) {
3816                 spin_unlock(&delayed_refs->lock);
3817                 printk(KERN_INFO "delayed_refs has NO entry\n");
3818                 return ret;
3819         }
3820
3821         while ((node = rb_first(&delayed_refs->root)) != NULL) {
3822                 struct btrfs_delayed_ref_head *head = NULL;
3823                 bool pin_bytes = false;
3824
3825                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3826                 atomic_set(&ref->refs, 1);
3827                 if (btrfs_delayed_ref_is_head(ref)) {
3828
3829                         head = btrfs_delayed_node_to_head(ref);
3830                         if (!mutex_trylock(&head->mutex)) {
3831                                 atomic_inc(&ref->refs);
3832                                 spin_unlock(&delayed_refs->lock);
3833
3834                                 /* Need to wait for the delayed ref to run */
3835                                 mutex_lock(&head->mutex);
3836                                 mutex_unlock(&head->mutex);
3837                                 btrfs_put_delayed_ref(ref);
3838
3839                                 spin_lock(&delayed_refs->lock);
3840                                 continue;
3841                         }
3842
3843                         if (head->must_insert_reserved)
3844                                 pin_bytes = true;
3845                         btrfs_free_delayed_extent_op(head->extent_op);
3846                         delayed_refs->num_heads--;
3847                         if (list_empty(&head->cluster))
3848                                 delayed_refs->num_heads_ready--;
3849                         list_del_init(&head->cluster);
3850                 }
3851
3852                 ref->in_tree = 0;
3853                 rb_erase(&ref->rb_node, &delayed_refs->root);
3854                 if (head)
3855                         rb_erase(&head->href_node, &delayed_refs->href_root);
3856
3857                 delayed_refs->num_entries--;
3858                 spin_unlock(&delayed_refs->lock);
3859                 if (head) {
3860                         if (pin_bytes)
3861                                 btrfs_pin_extent(root, ref->bytenr,
3862                                                  ref->num_bytes, 1);
3863                         mutex_unlock(&head->mutex);
3864                 }
3865                 btrfs_put_delayed_ref(ref);
3866
3867                 cond_resched();
3868                 spin_lock(&delayed_refs->lock);
3869         }
3870
3871         spin_unlock(&delayed_refs->lock);
3872
3873         return ret;
3874 }
3875
3876 static void btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3877 {
3878         struct btrfs_inode *btrfs_inode;
3879         struct list_head splice;
3880
3881         INIT_LIST_HEAD(&splice);
3882
3883         spin_lock(&root->delalloc_lock);
3884         list_splice_init(&root->delalloc_inodes, &splice);
3885
3886         while (!list_empty(&splice)) {
3887                 btrfs_inode = list_first_entry(&splice, struct btrfs_inode,
3888                                                delalloc_inodes);
3889
3890                 list_del_init(&btrfs_inode->delalloc_inodes);
3891                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
3892                           &btrfs_inode->runtime_flags);
3893                 spin_unlock(&root->delalloc_lock);
3894
3895                 btrfs_invalidate_inodes(btrfs_inode->root);
3896
3897                 spin_lock(&root->delalloc_lock);
3898         }
3899
3900         spin_unlock(&root->delalloc_lock);
3901 }
3902
3903 static void btrfs_destroy_all_delalloc_inodes(struct btrfs_fs_info *fs_info)
3904 {
3905         struct btrfs_root *root;
3906         struct list_head splice;
3907
3908         INIT_LIST_HEAD(&splice);
3909
3910         spin_lock(&fs_info->delalloc_root_lock);
3911         list_splice_init(&fs_info->delalloc_roots, &splice);
3912         while (!list_empty(&splice)) {
3913                 root = list_first_entry(&splice, struct btrfs_root,
3914                                          delalloc_root);
3915                 list_del_init(&root->delalloc_root);
3916                 root = btrfs_grab_fs_root(root);
3917                 BUG_ON(!root);
3918                 spin_unlock(&fs_info->delalloc_root_lock);
3919
3920                 btrfs_destroy_delalloc_inodes(root);
3921                 btrfs_put_fs_root(root);
3922
3923                 spin_lock(&fs_info->delalloc_root_lock);
3924         }
3925         spin_unlock(&fs_info->delalloc_root_lock);
3926 }
3927
3928 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3929                                         struct extent_io_tree *dirty_pages,
3930                                         int mark)
3931 {
3932         int ret;
3933         struct extent_buffer *eb;
3934         u64 start = 0;
3935         u64 end;
3936
3937         while (1) {
3938                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3939                                             mark, NULL);
3940                 if (ret)
3941                         break;
3942
3943                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3944                 while (start <= end) {
3945                         eb = btrfs_find_tree_block(root, start,
3946                                                    root->leafsize);
3947                         start += root->leafsize;
3948                         if (!eb)
3949                                 continue;
3950                         wait_on_extent_buffer_writeback(eb);
3951
3952                         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3953                                                &eb->bflags))
3954                                 clear_extent_buffer_dirty(eb);
3955                         free_extent_buffer_stale(eb);
3956                 }
3957         }
3958
3959         return ret;
3960 }
3961
3962 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3963                                        struct extent_io_tree *pinned_extents)
3964 {
3965         struct extent_io_tree *unpin;
3966         u64 start;
3967         u64 end;
3968         int ret;
3969         bool loop = true;
3970
3971         unpin = pinned_extents;
3972 again:
3973         while (1) {
3974                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3975                                             EXTENT_DIRTY, NULL);
3976                 if (ret)
3977                         break;
3978
3979                 /* opt_discard */
3980                 if (btrfs_test_opt(root, DISCARD))
3981                         ret = btrfs_error_discard_extent(root, start,
3982                                                          end + 1 - start,
3983                                                          NULL);
3984
3985                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3986                 btrfs_error_unpin_extent_range(root, start, end);
3987                 cond_resched();
3988         }
3989
3990         if (loop) {
3991                 if (unpin == &root->fs_info->freed_extents[0])
3992                         unpin = &root->fs_info->freed_extents[1];
3993                 else
3994                         unpin = &root->fs_info->freed_extents[0];
3995                 loop = false;
3996                 goto again;
3997         }
3998
3999         return 0;
4000 }
4001
4002 void btrfs_cleanup_one_transaction(struct btrfs_transaction *cur_trans,
4003                                    struct btrfs_root *root)
4004 {
4005         btrfs_destroy_ordered_operations(cur_trans, root);
4006
4007         btrfs_destroy_delayed_refs(cur_trans, root);
4008
4009         cur_trans->state = TRANS_STATE_COMMIT_START;
4010         wake_up(&root->fs_info->transaction_blocked_wait);
4011
4012         cur_trans->state = TRANS_STATE_UNBLOCKED;
4013         wake_up(&root->fs_info->transaction_wait);
4014
4015         btrfs_destroy_delayed_inodes(root);
4016         btrfs_assert_delayed_root_empty(root);
4017
4018         btrfs_destroy_marked_extents(root, &cur_trans->dirty_pages,
4019                                      EXTENT_DIRTY);
4020         btrfs_destroy_pinned_extent(root,
4021                                     root->fs_info->pinned_extents);
4022
4023         cur_trans->state =TRANS_STATE_COMPLETED;
4024         wake_up(&cur_trans->commit_wait);
4025
4026         /*
4027         memset(cur_trans, 0, sizeof(*cur_trans));
4028         kmem_cache_free(btrfs_transaction_cachep, cur_trans);
4029         */
4030 }
4031
4032 static int btrfs_cleanup_transaction(struct btrfs_root *root)
4033 {
4034         struct btrfs_transaction *t;
4035
4036         mutex_lock(&root->fs_info->transaction_kthread_mutex);
4037
4038         spin_lock(&root->fs_info->trans_lock);
4039         while (!list_empty(&root->fs_info->trans_list)) {
4040                 t = list_first_entry(&root->fs_info->trans_list,
4041                                      struct btrfs_transaction, list);
4042                 if (t->state >= TRANS_STATE_COMMIT_START) {
4043                         atomic_inc(&t->use_count);
4044                         spin_unlock(&root->fs_info->trans_lock);
4045                         btrfs_wait_for_commit(root, t->transid);
4046                         btrfs_put_transaction(t);
4047                         spin_lock(&root->fs_info->trans_lock);
4048                         continue;
4049                 }
4050                 if (t == root->fs_info->running_transaction) {
4051                         t->state = TRANS_STATE_COMMIT_DOING;
4052                         spin_unlock(&root->fs_info->trans_lock);
4053                         /*
4054                          * We wait for 0 num_writers since we don't hold a trans
4055                          * handle open currently for this transaction.
4056                          */
4057                         wait_event(t->writer_wait,
4058                                    atomic_read(&t->num_writers) == 0);
4059                 } else {
4060                         spin_unlock(&root->fs_info->trans_lock);
4061                 }
4062                 btrfs_cleanup_one_transaction(t, root);
4063
4064                 spin_lock(&root->fs_info->trans_lock);
4065                 if (t == root->fs_info->running_transaction)
4066                         root->fs_info->running_transaction = NULL;
4067                 list_del_init(&t->list);
4068                 spin_unlock(&root->fs_info->trans_lock);
4069
4070                 btrfs_put_transaction(t);
4071                 trace_btrfs_transaction_commit(root);
4072                 spin_lock(&root->fs_info->trans_lock);
4073         }
4074         spin_unlock(&root->fs_info->trans_lock);
4075         btrfs_destroy_all_ordered_extents(root->fs_info);
4076         btrfs_destroy_delayed_inodes(root);
4077         btrfs_assert_delayed_root_empty(root);
4078         btrfs_destroy_pinned_extent(root, root->fs_info->pinned_extents);
4079         btrfs_destroy_all_delalloc_inodes(root->fs_info);
4080         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
4081
4082         return 0;
4083 }
4084
4085 static struct extent_io_ops btree_extent_io_ops = {
4086         .readpage_end_io_hook = btree_readpage_end_io_hook,
4087         .readpage_io_failed_hook = btree_io_failed_hook,
4088         .submit_bio_hook = btree_submit_bio_hook,
4089         /* note we're sharing with inode.c for the merge bio hook */
4090         .merge_bio_hook = btrfs_merge_bio_hook,
4091 };