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 <asm/unaligned.h>
  34 #include "compat.h"
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "volumes.h"
  40 #include "print-tree.h"
  41 #include "async-thread.h"
  42 #include "locking.h"
  43 #include "tree-log.h"
  44 #include "free-space-cache.h"
  45 #include "inode-map.h"
  46
  47 static struct extent_io_ops btree_extent_io_ops;
  48 static void end_workqueue_fn(struct btrfs_work *work);
  49 static void free_fs_root(struct btrfs_root *root);
  50 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
  51                                     int read_only);
  52 static int btrfs_destroy_ordered_operations(struct btrfs_root *root);
  53 static int btrfs_destroy_ordered_extents(struct btrfs_root *root);
  54 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
  55                                       struct btrfs_root *root);
  56 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t);
  57 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root);
  58 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
  59                                         struct extent_io_tree *dirty_pages,
  60                                         int mark);
  61 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
  62                                        struct extent_io_tree *pinned_extents);
  63 static int btrfs_cleanup_transaction(struct btrfs_root *root);
  64
  65 /*
  66  * end_io_wq structs are used to do processing in task context when an IO is
  67  * complete.  This is used during reads to verify checksums, and it is used
  68  * by writes to insert metadata for new file extents after IO is complete.
  69  */
  70 struct end_io_wq {
  71         struct bio *bio;
  72         bio_end_io_t *end_io;
  73         void *private;
  74         struct btrfs_fs_info *info;
  75         int error;
  76         int metadata;
  77         struct list_head list;
  78         struct btrfs_work work;
  79 };
  80
  81 /*
  82  * async submit bios are used to offload expensive checksumming
  83  * onto the worker threads.  They checksum file and metadata bios
  84  * just before they are sent down the IO stack.
  85  */
  86 struct async_submit_bio {
  87         struct inode *inode;
  88         struct bio *bio;
  89         struct list_head list;
  90         extent_submit_bio_hook_t *submit_bio_start;
  91         extent_submit_bio_hook_t *submit_bio_done;
  92         int rw;
  93         int mirror_num;
  94         unsigned long bio_flags;
  95         /*
  96          * bio_offset is optional, can be used if the pages in the bio
  97          * can't tell us where in the file the bio should go
  98          */
  99         u64 bio_offset;
 100         struct btrfs_work work;
 101 };
 102
 103 /*
 104  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
 105  * eb, the lockdep key is determined by the btrfs_root it belongs to and
 106  * the level the eb occupies in the tree.
 107  *
 108  * Different roots are used for different purposes and may nest inside each
 109  * other and they require separate keysets.  As lockdep keys should be
 110  * static, assign keysets according to the purpose of the root as indicated
 111  * by btrfs_root->objectid.  This ensures that all special purpose roots
 112  * have separate keysets.
 113  *
 114  * Lock-nesting across peer nodes is always done with the immediate parent
 115  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
 116  * subclass to avoid triggering lockdep warning in such cases.
 117  *
 118  * The key is set by the readpage_end_io_hook after the buffer has passed
 119  * csum validation but before the pages are unlocked.  It is also set by
 120  * btrfs_init_new_buffer on freshly allocated blocks.
 121  *
 122  * We also add a check to make sure the highest level of the tree is the
 123  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
 124  * needs update as well.
 125  */
 126 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 127 # if BTRFS_MAX_LEVEL != 8
 128 #  error
 129 # endif
 130
 131 static struct btrfs_lockdep_keyset {
 132         u64                     id;             /* root objectid */
 133         const char              *name_stem;     /* lock name stem */
 134         char                    names[BTRFS_MAX_LEVEL + 1][20];
 135         struct lock_class_key   keys[BTRFS_MAX_LEVEL + 1];
 136 } btrfs_lockdep_keysets[] = {
 137         { .id = BTRFS_ROOT_TREE_OBJECTID,       .name_stem = "root"     },
 138         { .id = BTRFS_EXTENT_TREE_OBJECTID,     .name_stem = "extent"   },
 139         { .id = BTRFS_CHUNK_TREE_OBJECTID,      .name_stem = "chunk"    },
 140         { .id = BTRFS_DEV_TREE_OBJECTID,        .name_stem = "dev"      },
 141         { .id = BTRFS_FS_TREE_OBJECTID,         .name_stem = "fs"       },
 142         { .id = BTRFS_CSUM_TREE_OBJECTID,       .name_stem = "csum"     },
 143         { .id = BTRFS_ORPHAN_OBJECTID,          .name_stem = "orphan"   },
 144         { .id = BTRFS_TREE_LOG_OBJECTID,        .name_stem = "log"      },
 145         { .id = BTRFS_TREE_RELOC_OBJECTID,      .name_stem = "treloc"   },
 146         { .id = BTRFS_DATA_RELOC_TREE_OBJECTID, .name_stem = "dreloc"   },
 147         { .id = 0,                              .name_stem = "tree"     },
 148 };
 149
 150 void __init btrfs_init_lockdep(void)
 151 {
 152         int i, j;
 153
 154         /* initialize lockdep class names */
 155         for (i = 0; i < ARRAY_SIZE(btrfs_lockdep_keysets); i++) {
 156                 struct btrfs_lockdep_keyset *ks = &btrfs_lockdep_keysets[i];
 157
 158                 for (j = 0; j < ARRAY_SIZE(ks->names); j++)
 159                         snprintf(ks->names[j], sizeof(ks->names[j]),
 160                                  "btrfs-%s-%02d", ks->name_stem, j);
 161         }
 162 }
 163
 164 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb,
 165                                     int level)
 166 {
 167         struct btrfs_lockdep_keyset *ks;
 168
 169         BUG_ON(level >= ARRAY_SIZE(ks->keys));
 170
 171         /* find the matching keyset, id 0 is the default entry */
 172         for (ks = btrfs_lockdep_keysets; ks->id; ks++)
 173                 if (ks->id == objectid)
 174                         break;
 175
 176         lockdep_set_class_and_name(&eb->lock,
 177                                    &ks->keys[level], ks->names[level]);
 178 }
 179
 180 #endif
 181
 182 /*
 183  * extents on the btree inode are pretty simple, there's one extent
 184  * that covers the entire device
 185  */
 186 static struct extent_map *btree_get_extent(struct inode *inode,
 187                 struct page *page, size_t pg_offset, u64 start, u64 len,
 188                 int create)
 189 {
 190         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 191         struct extent_map *em;
 192         int ret;
 193
 194         read_lock(&em_tree->lock);
 195         em = lookup_extent_mapping(em_tree, start, len);
 196         if (em) {
 197                 em->bdev =
 198                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 199                 read_unlock(&em_tree->lock);
 200                 goto out;
 201         }
 202         read_unlock(&em_tree->lock);
 203
 204         em = alloc_extent_map();
 205         if (!em) {
 206                 em = ERR_PTR(-ENOMEM);
 207                 goto out;
 208         }
 209         em->start = 0;
 210         em->len = (u64)-1;
 211         em->block_len = (u64)-1;
 212         em->block_start = 0;
 213         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 214
 215         write_lock(&em_tree->lock);
 216         ret = add_extent_mapping(em_tree, em);
 217         if (ret == -EEXIST) {
 218                 u64 failed_start = em->start;
 219                 u64 failed_len = em->len;
 220
 221                 free_extent_map(em);
 222                 em = lookup_extent_mapping(em_tree, start, len);
 223                 if (em) {
 224                         ret = 0;
 225                 } else {
 226                         em = lookup_extent_mapping(em_tree, failed_start,
 227                                                    failed_len);
 228                         ret = -EIO;
 229                 }
 230         } else if (ret) {
 231                 free_extent_map(em);
 232                 em = NULL;
 233         }
 234         write_unlock(&em_tree->lock);
 235
 236         if (ret)
 237                 em = ERR_PTR(ret);
 238 out:
 239         return em;
 240 }
 241
 242 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 243 {
 244         return crc32c(seed, data, len);
 245 }
 246
 247 void btrfs_csum_final(u32 crc, char *result)
 248 {
 249         put_unaligned_le32(~crc, result);
 250 }
 251
 252 /*
 253  * compute the csum for a btree block, and either verify it or write it
 254  * into the csum field of the block.
 255  */
 256 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 257                            int verify)
 258 {
 259         u16 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
 260         char *result = NULL;
 261         unsigned long len;
 262         unsigned long cur_len;
 263         unsigned long offset = BTRFS_CSUM_SIZE;
 264         char *kaddr;
 265         unsigned long map_start;
 266         unsigned long map_len;
 267         int err;
 268         u32 crc = ~(u32)0;
 269         unsigned long inline_result;
 270
 271         len = buf->len - offset;
 272         while (len > 0) {
 273                 err = map_private_extent_buffer(buf, offset, 32,
 274                                         &kaddr, &map_start, &map_len);
 275                 if (err)
 276                         return 1;
 277                 cur_len = min(len, map_len - (offset - map_start));
 278                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 279                                       crc, cur_len);
 280                 len -= cur_len;
 281                 offset += cur_len;
 282         }
 283         if (csum_size > sizeof(inline_result)) {
 284                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 285                 if (!result)
 286                         return 1;
 287         } else {
 288                 result = (char *)&inline_result;
 289         }
 290
 291         btrfs_csum_final(crc, result);
 292
 293         if (verify) {
 294                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 295                         u32 val;
 296                         u32 found = 0;
 297                         memcpy(&found, result, csum_size);
 298
 299                         read_extent_buffer(buf, &val, 0, csum_size);
 300                         printk_ratelimited(KERN_INFO "btrfs: %s checksum verify "
 301                                        "failed on %llu wanted %X found %X "
 302                                        "level %d\n",
 303                                        root->fs_info->sb->s_id,
 304                                        (unsigned long long)buf->start, val, found,
 305                                        btrfs_header_level(buf));
 306                         if (result != (char *)&inline_result)
 307                                 kfree(result);
 308                         return 1;
 309                 }
 310         } else {
 311                 write_extent_buffer(buf, result, 0, csum_size);
 312         }
 313         if (result != (char *)&inline_result)
 314                 kfree(result);
 315         return 0;
 316 }
 317
 318 /*
 319  * we can't consider a given block up to date unless the transid of the
 320  * block matches the transid in the parent node's pointer.  This is how we
 321  * detect blocks that either didn't get written at all or got written
 322  * in the wrong place.
 323  */
 324 static int verify_parent_transid(struct extent_io_tree *io_tree,
 325                                  struct extent_buffer *eb, u64 parent_transid)
 326 {
 327         struct extent_state *cached_state = NULL;
 328         int ret;
 329
 330         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 331                 return 0;
 332
 333         lock_extent_bits(io_tree, eb->start, eb->start + eb->len - 1,
 334                          0, &cached_state, GFP_NOFS);
 335         if (extent_buffer_uptodate(io_tree, eb, cached_state) &&
 336             btrfs_header_generation(eb) == parent_transid) {
 337                 ret = 0;
 338                 goto out;
 339         }
 340         printk_ratelimited("parent transid verify failed on %llu wanted %llu "
 341                        "found %llu\n",
 342                        (unsigned long long)eb->start,
 343                        (unsigned long long)parent_transid,
 344                        (unsigned long long)btrfs_header_generation(eb));
 345         ret = 1;
 346         clear_extent_buffer_uptodate(io_tree, eb, &cached_state);
 347 out:
 348         unlock_extent_cached(io_tree, eb->start, eb->start + eb->len - 1,
 349                              &cached_state, GFP_NOFS);
 350         return ret;
 351 }
 352
 353 /*
 354  * helper to read a given tree block, doing retries as required when
 355  * the checksums don't match and we have alternate mirrors to try.
 356  */
 357 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 358                                           struct extent_buffer *eb,
 359                                           u64 start, u64 parent_transid)
 360 {
 361         struct extent_io_tree *io_tree;
 362         int ret;
 363         int num_copies = 0;
 364         int mirror_num = 0;
 365
 366         clear_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 367         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 368         while (1) {
 369                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 370                                                btree_get_extent, mirror_num);
 371                 if (!ret &&
 372                     !verify_parent_transid(io_tree, eb, parent_transid))
 373                         return ret;
 374
 375                 /*
 376                  * This buffer's crc is fine, but its contents are corrupted, so
 377                  * there is no reason to read the other copies, they won't be
 378                  * any less wrong.
 379                  */
 380                 if (test_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags))
 381                         return ret;
 382
 383                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 384                                               eb->start, eb->len);
 385                 if (num_copies == 1)
 386                         return ret;
 387
 388                 mirror_num++;
 389                 if (mirror_num > num_copies)
 390                         return ret;
 391         }
 392         return -EIO;
 393 }
 394
 395 /*
 396  * checksum a dirty tree block before IO.  This has extra checks to make sure
 397  * we only fill in the checksum field in the first page of a multi-page block
 398  */
 399
 400 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 401 {
 402         struct extent_io_tree *tree;
 403         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 404         u64 found_start;
 405         unsigned long len;
 406         struct extent_buffer *eb;
 407         int ret;
 408
 409         tree = &BTRFS_I(page->mapping->host)->io_tree;
 410
 411         if (page->private == EXTENT_PAGE_PRIVATE) {
 412                 WARN_ON(1);
 413                 goto out;
 414         }
 415         if (!page->private) {
 416                 WARN_ON(1);
 417                 goto out;
 418         }
 419         len = page->private >> 2;
 420         WARN_ON(len == 0);
 421
 422         eb = alloc_extent_buffer(tree, start, len, page);
 423         if (eb == NULL) {
 424                 WARN_ON(1);
 425                 goto out;
 426         }
 427         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 428                                              btrfs_header_generation(eb));
 429         BUG_ON(ret);
 430         WARN_ON(!btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN));
 431
 432         found_start = btrfs_header_bytenr(eb);
 433         if (found_start != start) {
 434                 WARN_ON(1);
 435                 goto err;
 436         }
 437         if (eb->first_page != page) {
 438                 WARN_ON(1);
 439                 goto err;
 440         }
 441         if (!PageUptodate(page)) {
 442                 WARN_ON(1);
 443                 goto err;
 444         }
 445         csum_tree_block(root, eb, 0);
 446 err:
 447         free_extent_buffer(eb);
 448 out:
 449         return 0;
 450 }
 451
 452 static int check_tree_block_fsid(struct btrfs_root *root,
 453                                  struct extent_buffer *eb)
 454 {
 455         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 456         u8 fsid[BTRFS_UUID_SIZE];
 457         int ret = 1;
 458
 459         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 460                            BTRFS_FSID_SIZE);
 461         while (fs_devices) {
 462                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 463                         ret = 0;
 464                         break;
 465                 }
 466                 fs_devices = fs_devices->seed;
 467         }
 468         return ret;
 469 }
 470
 471 #define CORRUPT(reason, eb, root, slot)                         \
 472         printk(KERN_CRIT "btrfs: corrupt leaf, %s: block=%llu," \
 473                "root=%llu, slot=%d\n", reason,                  \
 474                (unsigned long long)btrfs_header_bytenr(eb),     \
 475                (unsigned long long)root->objectid, slot)
 476
 477 static noinline int check_leaf(struct btrfs_root *root,
 478                                struct extent_buffer *leaf)
 479 {
 480         struct btrfs_key key;
 481         struct btrfs_key leaf_key;
 482         u32 nritems = btrfs_header_nritems(leaf);
 483         int slot;
 484
 485         if (nritems == 0)
 486                 return 0;
 487
 488         /* Check the 0 item */
 489         if (btrfs_item_offset_nr(leaf, 0) + btrfs_item_size_nr(leaf, 0) !=
 490             BTRFS_LEAF_DATA_SIZE(root)) {
 491                 CORRUPT("invalid item offset size pair", leaf, root, 0);
 492                 return -EIO;
 493         }
 494
 495         /*
 496          * Check to make sure each items keys are in the correct order and their
 497          * offsets make sense.  We only have to loop through nritems-1 because
 498          * we check the current slot against the next slot, which verifies the
 499          * next slot's offset+size makes sense and that the current's slot
 500          * offset is correct.
 501          */
 502         for (slot = 0; slot < nritems - 1; slot++) {
 503                 btrfs_item_key_to_cpu(leaf, &leaf_key, slot);
 504                 btrfs_item_key_to_cpu(leaf, &key, slot + 1);
 505
 506                 /* Make sure the keys are in the right order */
 507                 if (btrfs_comp_cpu_keys(&leaf_key, &key) >= 0) {
 508                         CORRUPT("bad key order", leaf, root, slot);
 509                         return -EIO;
 510                 }
 511
 512                 /*
 513                  * Make sure the offset and ends are right, remember that the
 514                  * item data starts at the end of the leaf and grows towards the
 515                  * front.
 516                  */
 517                 if (btrfs_item_offset_nr(leaf, slot) !=
 518                         btrfs_item_end_nr(leaf, slot + 1)) {
 519                         CORRUPT("slot offset bad", leaf, root, slot);
 520                         return -EIO;
 521                 }
 522
 523                 /*
 524                  * Check to make sure that we don't point outside of the leaf,
 525                  * just incase all the items are consistent to eachother, but
 526                  * all point outside of the leaf.
 527                  */
 528                 if (btrfs_item_end_nr(leaf, slot) >
 529                     BTRFS_LEAF_DATA_SIZE(root)) {
 530                         CORRUPT("slot end outside of leaf", leaf, root, slot);
 531                         return -EIO;
 532                 }
 533         }
 534
 535         return 0;
 536 }
 537
 538 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 539                                struct extent_state *state)
 540 {
 541         struct extent_io_tree *tree;
 542         u64 found_start;
 543         int found_level;
 544         unsigned long len;
 545         struct extent_buffer *eb;
 546         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 547         int ret = 0;
 548
 549         tree = &BTRFS_I(page->mapping->host)->io_tree;
 550         if (page->private == EXTENT_PAGE_PRIVATE)
 551                 goto out;
 552         if (!page->private)
 553                 goto out;
 554
 555         len = page->private >> 2;
 556         WARN_ON(len == 0);
 557
 558         eb = alloc_extent_buffer(tree, start, len, page);
 559         if (eb == NULL) {
 560                 ret = -EIO;
 561                 goto out;
 562         }
 563
 564         found_start = btrfs_header_bytenr(eb);
 565         if (found_start != start) {
 566                 printk_ratelimited(KERN_INFO "btrfs bad tree block start "
 567                                "%llu %llu\n",
 568                                (unsigned long long)found_start,
 569                                (unsigned long long)eb->start);
 570                 ret = -EIO;
 571                 goto err;
 572         }
 573         if (eb->first_page != page) {
 574                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 575                        eb->first_page->index, page->index);
 576                 WARN_ON(1);
 577                 ret = -EIO;
 578                 goto err;
 579         }
 580         if (check_tree_block_fsid(root, eb)) {
 581                 printk_ratelimited(KERN_INFO "btrfs bad fsid on block %llu\n",
 582                                (unsigned long long)eb->start);
 583                 ret = -EIO;
 584                 goto err;
 585         }
 586         found_level = btrfs_header_level(eb);
 587
 588         btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb),
 589                                        eb, found_level);
 590
 591         ret = csum_tree_block(root, eb, 1);
 592         if (ret) {
 593                 ret = -EIO;
 594                 goto err;
 595         }
 596
 597         /*
 598          * If this is a leaf block and it is corrupt, set the corrupt bit so
 599          * that we don't try and read the other copies of this block, just
 600          * return -EIO.
 601          */
 602         if (found_level == 0 && check_leaf(root, eb)) {
 603                 set_bit(EXTENT_BUFFER_CORRUPT, &eb->bflags);
 604                 ret = -EIO;
 605         }
 606
 607         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 608         end = eb->start + end - 1;
 609 err:
 610         free_extent_buffer(eb);
 611 out:
 612         return ret;
 613 }
 614
 615 static void end_workqueue_bio(struct bio *bio, int err)
 616 {
 617         struct end_io_wq *end_io_wq = bio->bi_private;
 618         struct btrfs_fs_info *fs_info;
 619
 620         fs_info = end_io_wq->info;
 621         end_io_wq->error = err;
 622         end_io_wq->work.func = end_workqueue_fn;
 623         end_io_wq->work.flags = 0;
 624
 625         if (bio->bi_rw & REQ_WRITE) {
 626                 if (end_io_wq->metadata == 1)
 627                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 628                                            &end_io_wq->work);
 629                 else if (end_io_wq->metadata == 2)
 630                         btrfs_queue_worker(&fs_info->endio_freespace_worker,
 631                                            &end_io_wq->work);
 632                 else
 633                         btrfs_queue_worker(&fs_info->endio_write_workers,
 634                                            &end_io_wq->work);
 635         } else {
 636                 if (end_io_wq->metadata)
 637                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 638                                            &end_io_wq->work);
 639                 else
 640                         btrfs_queue_worker(&fs_info->endio_workers,
 641                                            &end_io_wq->work);
 642         }
 643 }
 644
 645 /*
 646  * For the metadata arg you want
 647  *
 648  * 0 - if data
 649  * 1 - if normal metadta
 650  * 2 - if writing to the free space cache area
 651  */
 652 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 653                         int metadata)
 654 {
 655         struct end_io_wq *end_io_wq;
 656         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 657         if (!end_io_wq)
 658                 return -ENOMEM;
 659
 660         end_io_wq->private = bio->bi_private;
 661         end_io_wq->end_io = bio->bi_end_io;
 662         end_io_wq->info = info;
 663         end_io_wq->error = 0;
 664         end_io_wq->bio = bio;
 665         end_io_wq->metadata = metadata;
 666
 667         bio->bi_private = end_io_wq;
 668         bio->bi_end_io = end_workqueue_bio;
 669         return 0;
 670 }
 671
 672 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 673 {
 674         unsigned long limit = min_t(unsigned long,
 675                                     info->workers.max_workers,
 676                                     info->fs_devices->open_devices);
 677         return 256 * limit;
 678 }
 679
 680 static void run_one_async_start(struct btrfs_work *work)
 681 {
 682         struct async_submit_bio *async;
 683
 684         async = container_of(work, struct  async_submit_bio, work);
 685         async->submit_bio_start(async->inode, async->rw, async->bio,
 686                                async->mirror_num, async->bio_flags,
 687                                async->bio_offset);
 688 }
 689
 690 static void run_one_async_done(struct btrfs_work *work)
 691 {
 692         struct btrfs_fs_info *fs_info;
 693         struct async_submit_bio *async;
 694         int limit;
 695
 696         async = container_of(work, struct  async_submit_bio, work);
 697         fs_info = BTRFS_I(async->inode)->root->fs_info;
 698
 699         limit = btrfs_async_submit_limit(fs_info);
 700         limit = limit * 2 / 3;
 701
 702         atomic_dec(&fs_info->nr_async_submits);
 703
 704         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 705             waitqueue_active(&fs_info->async_submit_wait))
 706                 wake_up(&fs_info->async_submit_wait);
 707
 708         async->submit_bio_done(async->inode, async->rw, async->bio,
 709                                async->mirror_num, async->bio_flags,
 710                                async->bio_offset);
 711 }
 712
 713 static void run_one_async_free(struct btrfs_work *work)
 714 {
 715         struct async_submit_bio *async;
 716
 717         async = container_of(work, struct  async_submit_bio, work);
 718         kfree(async);
 719 }
 720
 721 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 722                         int rw, struct bio *bio, int mirror_num,
 723                         unsigned long bio_flags,
 724                         u64 bio_offset,
 725                         extent_submit_bio_hook_t *submit_bio_start,
 726                         extent_submit_bio_hook_t *submit_bio_done)
 727 {
 728         struct async_submit_bio *async;
 729
 730         async = kmalloc(sizeof(*async), GFP_NOFS);
 731         if (!async)
 732                 return -ENOMEM;
 733
 734         async->inode = inode;
 735         async->rw = rw;
 736         async->bio = bio;
 737         async->mirror_num = mirror_num;
 738         async->submit_bio_start = submit_bio_start;
 739         async->submit_bio_done = submit_bio_done;
 740
 741         async->work.func = run_one_async_start;
 742         async->work.ordered_func = run_one_async_done;
 743         async->work.ordered_free = run_one_async_free;
 744
 745         async->work.flags = 0;
 746         async->bio_flags = bio_flags;
 747         async->bio_offset = bio_offset;
 748
 749         atomic_inc(&fs_info->nr_async_submits);
 750
 751         if (rw & REQ_SYNC)
 752                 btrfs_set_work_high_prio(&async->work);
 753
 754         btrfs_queue_worker(&fs_info->workers, &async->work);
 755
 756         while (atomic_read(&fs_info->async_submit_draining) &&
 757               atomic_read(&fs_info->nr_async_submits)) {
 758                 wait_event(fs_info->async_submit_wait,
 759                            (atomic_read(&fs_info->nr_async_submits) == 0));
 760         }
 761
 762         return 0;
 763 }
 764
 765 static int btree_csum_one_bio(struct bio *bio)
 766 {
 767         struct bio_vec *bvec = bio->bi_io_vec;
 768         int bio_index = 0;
 769         struct btrfs_root *root;
 770
 771         WARN_ON(bio->bi_vcnt <= 0);
 772         while (bio_index < bio->bi_vcnt) {
 773                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 774                 csum_dirty_buffer(root, bvec->bv_page);
 775                 bio_index++;
 776                 bvec++;
 777         }
 778         return 0;
 779 }
 780
 781 static int __btree_submit_bio_start(struct inode *inode, int rw,
 782                                     struct bio *bio, int mirror_num,
 783                                     unsigned long bio_flags,
 784                                     u64 bio_offset)
 785 {
 786         /*
 787          * when we're called for a write, we're already in the async
 788          * submission context.  Just jump into btrfs_map_bio
 789          */
 790         btree_csum_one_bio(bio);
 791         return 0;
 792 }
 793
 794 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 795                                  int mirror_num, unsigned long bio_flags,
 796                                  u64 bio_offset)
 797 {
 798         /*
 799          * when we're called for a write, we're already in the async
 800          * submission context.  Just jump into btrfs_map_bio
 801          */
 802         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 803 }
 804
 805 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 806                                  int mirror_num, unsigned long bio_flags,
 807                                  u64 bio_offset)
 808 {
 809         int ret;
 810
 811         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 812                                           bio, 1);
 813         BUG_ON(ret);
 814
 815         if (!(rw & REQ_WRITE)) {
 816                 /*
 817                  * called for a read, do the setup so that checksum validation
 818                  * can happen in the async kernel threads
 819                  */
 820                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 821                                      mirror_num, 0);
 822         }
 823
 824         /*
 825          * kthread helpers are used to submit writes so that checksumming
 826          * can happen in parallel across all CPUs
 827          */
 828         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 829                                    inode, rw, bio, mirror_num, 0,
 830                                    bio_offset,
 831                                    __btree_submit_bio_start,
 832                                    __btree_submit_bio_done);
 833 }
 834
 835 #ifdef CONFIG_MIGRATION
 836 static int btree_migratepage(struct address_space *mapping,
 837                         struct page *newpage, struct page *page)
 838 {
 839         /*
 840          * we can't safely write a btree page from here,
 841          * we haven't done the locking hook
 842          */
 843         if (PageDirty(page))
 844                 return -EAGAIN;
 845         /*
 846          * Buffers may be managed in a filesystem specific way.
 847          * We must have no buffers or drop them.
 848          */
 849         if (page_has_private(page) &&
 850             !try_to_release_page(page, GFP_KERNEL))
 851                 return -EAGAIN;
 852         return migrate_page(mapping, newpage, page);
 853 }
 854 #endif
 855
 856 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 857 {
 858         struct extent_io_tree *tree;
 859         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 860         struct extent_buffer *eb;
 861         int was_dirty;
 862
 863         tree = &BTRFS_I(page->mapping->host)->io_tree;
 864         if (!(current->flags & PF_MEMALLOC)) {
 865                 return extent_write_full_page(tree, page,
 866                                               btree_get_extent, wbc);
 867         }
 868
 869         redirty_page_for_writepage(wbc, page);
 870         eb = btrfs_find_tree_block(root, page_offset(page), PAGE_CACHE_SIZE);
 871         WARN_ON(!eb);
 872
 873         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 874         if (!was_dirty) {
 875                 spin_lock(&root->fs_info->delalloc_lock);
 876                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 877                 spin_unlock(&root->fs_info->delalloc_lock);
 878         }
 879         free_extent_buffer(eb);
 880
 881         unlock_page(page);
 882         return 0;
 883 }
 884
 885 static int btree_writepages(struct address_space *mapping,
 886                             struct writeback_control *wbc)
 887 {
 888         struct extent_io_tree *tree;
 889         tree = &BTRFS_I(mapping->host)->io_tree;
 890         if (wbc->sync_mode == WB_SYNC_NONE) {
 891                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 892                 u64 num_dirty;
 893                 unsigned long thresh = 32 * 1024 * 1024;
 894
 895                 if (wbc->for_kupdate)
 896                         return 0;
 897
 898                 /* this is a bit racy, but that's ok */
 899                 num_dirty = root->fs_info->dirty_metadata_bytes;
 900                 if (num_dirty < thresh)
 901                         return 0;
 902         }
 903         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 904 }
 905
 906 static int btree_readpage(struct file *file, struct page *page)
 907 {
 908         struct extent_io_tree *tree;
 909         tree = &BTRFS_I(page->mapping->host)->io_tree;
 910         return extent_read_full_page(tree, page, btree_get_extent);
 911 }
 912
 913 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 914 {
 915         struct extent_io_tree *tree;
 916         struct extent_map_tree *map;
 917         int ret;
 918
 919         if (PageWriteback(page) || PageDirty(page))
 920                 return 0;
 921
 922         tree = &BTRFS_I(page->mapping->host)->io_tree;
 923         map = &BTRFS_I(page->mapping->host)->extent_tree;
 924
 925         ret = try_release_extent_state(map, tree, page, gfp_flags);
 926         if (!ret)
 927                 return 0;
 928
 929         ret = try_release_extent_buffer(tree, page);
 930         if (ret == 1) {
 931                 ClearPagePrivate(page);
 932                 set_page_private(page, 0);
 933                 page_cache_release(page);
 934         }
 935
 936         return ret;
 937 }
 938
 939 static void btree_invalidatepage(struct page *page, unsigned long offset)
 940 {
 941         struct extent_io_tree *tree;
 942         tree = &BTRFS_I(page->mapping->host)->io_tree;
 943         extent_invalidatepage(tree, page, offset);
 944         btree_releasepage(page, GFP_NOFS);
 945         if (PagePrivate(page)) {
 946                 printk(KERN_WARNING "btrfs warning page private not zero "
 947                        "on page %llu\n", (unsigned long long)page_offset(page));
 948                 ClearPagePrivate(page);
 949                 set_page_private(page, 0);
 950                 page_cache_release(page);
 951         }
 952 }
 953
 954 static const struct address_space_operations btree_aops = {
 955         .readpage       = btree_readpage,
 956         .writepage      = btree_writepage,
 957         .writepages     = btree_writepages,
 958         .releasepage    = btree_releasepage,
 959         .invalidatepage = btree_invalidatepage,
 960 #ifdef CONFIG_MIGRATION
 961         .migratepage    = btree_migratepage,
 962 #endif
 963 };
 964
 965 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 966                          u64 parent_transid)
 967 {
 968         struct extent_buffer *buf = NULL;
 969         struct inode *btree_inode = root->fs_info->btree_inode;
 970         int ret = 0;
 971
 972         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 973         if (!buf)
 974                 return 0;
 975         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 976                                  buf, 0, 0, btree_get_extent, 0);
 977         free_extent_buffer(buf);
 978         return ret;
 979 }
 980
 981 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 982                                             u64 bytenr, u32 blocksize)
 983 {
 984         struct inode *btree_inode = root->fs_info->btree_inode;
 985         struct extent_buffer *eb;
 986         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 987                                 bytenr, blocksize);
 988         return eb;
 989 }
 990
 991 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 992                                                  u64 bytenr, u32 blocksize)
 993 {
 994         struct inode *btree_inode = root->fs_info->btree_inode;
 995         struct extent_buffer *eb;
 996
 997         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 998                                  bytenr, blocksize, NULL);
 999         return eb;
1000 }
1001
1002
1003 int btrfs_write_tree_block(struct extent_buffer *buf)
1004 {
1005         return filemap_fdatawrite_range(buf->first_page->mapping, buf->start,
1006                                         buf->start + buf->len - 1);
1007 }
1008
1009 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
1010 {
1011         return filemap_fdatawait_range(buf->first_page->mapping,
1012                                        buf->start, buf->start + buf->len - 1);
1013 }
1014
1015 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
1016                                       u32 blocksize, u64 parent_transid)
1017 {
1018         struct extent_buffer *buf = NULL;
1019         int ret;
1020
1021         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
1022         if (!buf)
1023                 return NULL;
1024
1025         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
1026
1027         if (ret == 0)
1028                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
1029         return buf;
1030
1031 }
1032
1033 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
1034                      struct extent_buffer *buf)
1035 {
1036         struct inode *btree_inode = root->fs_info->btree_inode;
1037         if (btrfs_header_generation(buf) ==
1038             root->fs_info->running_transaction->transid) {
1039                 btrfs_assert_tree_locked(buf);
1040
1041                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
1042                         spin_lock(&root->fs_info->delalloc_lock);
1043                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
1044                                 root->fs_info->dirty_metadata_bytes -= buf->len;
1045                         else
1046                                 WARN_ON(1);
1047                         spin_unlock(&root->fs_info->delalloc_lock);
1048                 }
1049
1050                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
1051                 btrfs_set_lock_blocking(buf);
1052                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
1053                                           buf);
1054         }
1055         return 0;
1056 }
1057
1058 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
1059                         u32 stripesize, struct btrfs_root *root,
1060                         struct btrfs_fs_info *fs_info,
1061                         u64 objectid)
1062 {
1063         root->node = NULL;
1064         root->commit_root = NULL;
1065         root->sectorsize = sectorsize;
1066         root->nodesize = nodesize;
1067         root->leafsize = leafsize;
1068         root->stripesize = stripesize;
1069         root->ref_cows = 0;
1070         root->track_dirty = 0;
1071         root->in_radix = 0;
1072         root->orphan_item_inserted = 0;
1073         root->orphan_cleanup_state = 0;
1074
1075         root->fs_info = fs_info;
1076         root->objectid = objectid;
1077         root->last_trans = 0;
1078         root->highest_objectid = 0;
1079         root->name = NULL;
1080         root->inode_tree = RB_ROOT;
1081         INIT_RADIX_TREE(&root->delayed_nodes_tree, GFP_ATOMIC);
1082         root->block_rsv = NULL;
1083         root->orphan_block_rsv = NULL;
1084
1085         INIT_LIST_HEAD(&root->dirty_list);
1086         INIT_LIST_HEAD(&root->orphan_list);
1087         INIT_LIST_HEAD(&root->root_list);
1088         spin_lock_init(&root->orphan_lock);
1089         spin_lock_init(&root->inode_lock);
1090         spin_lock_init(&root->accounting_lock);
1091         mutex_init(&root->objectid_mutex);
1092         mutex_init(&root->log_mutex);
1093         init_waitqueue_head(&root->log_writer_wait);
1094         init_waitqueue_head(&root->log_commit_wait[0]);
1095         init_waitqueue_head(&root->log_commit_wait[1]);
1096         atomic_set(&root->log_commit[0], 0);
1097         atomic_set(&root->log_commit[1], 0);
1098         atomic_set(&root->log_writers, 0);
1099         root->log_batch = 0;
1100         root->log_transid = 0;
1101         root->last_log_commit = 0;
1102         extent_io_tree_init(&root->dirty_log_pages,
1103                              fs_info->btree_inode->i_mapping);
1104
1105         memset(&root->root_key, 0, sizeof(root->root_key));
1106         memset(&root->root_item, 0, sizeof(root->root_item));
1107         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
1108         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
1109         root->defrag_trans_start = fs_info->generation;
1110         init_completion(&root->kobj_unregister);
1111         root->defrag_running = 0;
1112         root->root_key.objectid = objectid;
1113         root->anon_dev = 0;
1114         return 0;
1115 }
1116
1117 static int find_and_setup_root(struct btrfs_root *tree_root,
1118                                struct btrfs_fs_info *fs_info,
1119                                u64 objectid,
1120                                struct btrfs_root *root)
1121 {
1122         int ret;
1123         u32 blocksize;
1124         u64 generation;
1125
1126         __setup_root(tree_root->nodesize, tree_root->leafsize,
1127                      tree_root->sectorsize, tree_root->stripesize,
1128                      root, fs_info, objectid);
1129         ret = btrfs_find_last_root(tree_root, objectid,
1130                                    &root->root_item, &root->root_key);
1131         if (ret > 0)
1132                 return -ENOENT;
1133         BUG_ON(ret);
1134
1135         generation = btrfs_root_generation(&root->root_item);
1136         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1137         root->commit_root = NULL;
1138         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1139                                      blocksize, generation);
1140         if (!root->node || !btrfs_buffer_uptodate(root->node, generation)) {
1141                 free_extent_buffer(root->node);
1142                 root->node = NULL;
1143                 return -EIO;
1144         }
1145         root->commit_root = btrfs_root_node(root);
1146         return 0;
1147 }
1148
1149 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1150                                          struct btrfs_fs_info *fs_info)
1151 {
1152         struct btrfs_root *root;
1153         struct btrfs_root *tree_root = fs_info->tree_root;
1154         struct extent_buffer *leaf;
1155
1156         root = kzalloc(sizeof(*root), GFP_NOFS);
1157         if (!root)
1158                 return ERR_PTR(-ENOMEM);
1159
1160         __setup_root(tree_root->nodesize, tree_root->leafsize,
1161                      tree_root->sectorsize, tree_root->stripesize,
1162                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1163
1164         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1165         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1166         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1167         /*
1168          * log trees do not get reference counted because they go away
1169          * before a real commit is actually done.  They do store pointers
1170          * to file data extents, and those reference counts still get
1171          * updated (along with back refs to the log tree).
1172          */
1173         root->ref_cows = 0;
1174
1175         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1176                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1177         if (IS_ERR(leaf)) {
1178                 kfree(root);
1179                 return ERR_CAST(leaf);
1180         }
1181
1182         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1183         btrfs_set_header_bytenr(leaf, leaf->start);
1184         btrfs_set_header_generation(leaf, trans->transid);
1185         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1186         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1187         root->node = leaf;
1188
1189         write_extent_buffer(root->node, root->fs_info->fsid,
1190                             (unsigned long)btrfs_header_fsid(root->node),
1191                             BTRFS_FSID_SIZE);
1192         btrfs_mark_buffer_dirty(root->node);
1193         btrfs_tree_unlock(root->node);
1194         return root;
1195 }
1196
1197 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1198                              struct btrfs_fs_info *fs_info)
1199 {
1200         struct btrfs_root *log_root;
1201
1202         log_root = alloc_log_tree(trans, fs_info);
1203         if (IS_ERR(log_root))
1204                 return PTR_ERR(log_root);
1205         WARN_ON(fs_info->log_root_tree);
1206         fs_info->log_root_tree = log_root;
1207         return 0;
1208 }
1209
1210 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1211                        struct btrfs_root *root)
1212 {
1213         struct btrfs_root *log_root;
1214         struct btrfs_inode_item *inode_item;
1215
1216         log_root = alloc_log_tree(trans, root->fs_info);
1217         if (IS_ERR(log_root))
1218                 return PTR_ERR(log_root);
1219
1220         log_root->last_trans = trans->transid;
1221         log_root->root_key.offset = root->root_key.objectid;
1222
1223         inode_item = &log_root->root_item.inode;
1224         inode_item->generation = cpu_to_le64(1);
1225         inode_item->size = cpu_to_le64(3);
1226         inode_item->nlink = cpu_to_le32(1);
1227         inode_item->nbytes = cpu_to_le64(root->leafsize);
1228         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1229
1230         btrfs_set_root_node(&log_root->root_item, log_root->node);
1231
1232         WARN_ON(root->log_root);
1233         root->log_root = log_root;
1234         root->log_transid = 0;
1235         root->last_log_commit = 0;
1236         return 0;
1237 }
1238
1239 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1240                                                struct btrfs_key *location)
1241 {
1242         struct btrfs_root *root;
1243         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1244         struct btrfs_path *path;
1245         struct extent_buffer *l;
1246         u64 generation;
1247         u32 blocksize;
1248         int ret = 0;
1249
1250         root = kzalloc(sizeof(*root), GFP_NOFS);
1251         if (!root)
1252                 return ERR_PTR(-ENOMEM);
1253         if (location->offset == (u64)-1) {
1254                 ret = find_and_setup_root(tree_root, fs_info,
1255                                           location->objectid, root);
1256                 if (ret) {
1257                         kfree(root);
1258                         return ERR_PTR(ret);
1259                 }
1260                 goto out;
1261         }
1262
1263         __setup_root(tree_root->nodesize, tree_root->leafsize,
1264                      tree_root->sectorsize, tree_root->stripesize,
1265                      root, fs_info, location->objectid);
1266
1267         path = btrfs_alloc_path();
1268         if (!path) {
1269                 kfree(root);
1270                 return ERR_PTR(-ENOMEM);
1271         }
1272         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1273         if (ret == 0) {
1274                 l = path->nodes[0];
1275                 read_extent_buffer(l, &root->root_item,
1276                                 btrfs_item_ptr_offset(l, path->slots[0]),
1277                                 sizeof(root->root_item));
1278                 memcpy(&root->root_key, location, sizeof(*location));
1279         }
1280         btrfs_free_path(path);
1281         if (ret) {
1282                 kfree(root);
1283                 if (ret > 0)
1284                         ret = -ENOENT;
1285                 return ERR_PTR(ret);
1286         }
1287
1288         generation = btrfs_root_generation(&root->root_item);
1289         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1290         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1291                                      blocksize, generation);
1292         root->commit_root = btrfs_root_node(root);
1293         BUG_ON(!root->node);
1294 out:
1295         if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1296                 root->ref_cows = 1;
1297                 btrfs_check_and_init_root_item(&root->root_item);
1298         }
1299
1300         return root;
1301 }
1302
1303 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1304                                               struct btrfs_key *location)
1305 {
1306         struct btrfs_root *root;
1307         int ret;
1308
1309         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1310                 return fs_info->tree_root;
1311         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1312                 return fs_info->extent_root;
1313         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1314                 return fs_info->chunk_root;
1315         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1316                 return fs_info->dev_root;
1317         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1318                 return fs_info->csum_root;
1319 again:
1320         spin_lock(&fs_info->fs_roots_radix_lock);
1321         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1322                                  (unsigned long)location->objectid);
1323         spin_unlock(&fs_info->fs_roots_radix_lock);
1324         if (root)
1325                 return root;
1326
1327         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1328         if (IS_ERR(root))
1329                 return root;
1330
1331         root->free_ino_ctl = kzalloc(sizeof(*root->free_ino_ctl), GFP_NOFS);
1332         root->free_ino_pinned = kzalloc(sizeof(*root->free_ino_pinned),
1333                                         GFP_NOFS);
1334         if (!root->free_ino_pinned || !root->free_ino_ctl) {
1335                 ret = -ENOMEM;
1336                 goto fail;
1337         }
1338
1339         btrfs_init_free_ino_ctl(root);
1340         mutex_init(&root->fs_commit_mutex);
1341         spin_lock_init(&root->cache_lock);
1342         init_waitqueue_head(&root->cache_wait);
1343
1344         ret = get_anon_bdev(&root->anon_dev);
1345         if (ret)
1346                 goto fail;
1347
1348         if (btrfs_root_refs(&root->root_item) == 0) {
1349                 ret = -ENOENT;
1350                 goto fail;
1351         }
1352
1353         ret = btrfs_find_orphan_item(fs_info->tree_root, location->objectid);
1354         if (ret < 0)
1355                 goto fail;
1356         if (ret == 0)
1357                 root->orphan_item_inserted = 1;
1358
1359         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
1360         if (ret)
1361                 goto fail;
1362
1363         spin_lock(&fs_info->fs_roots_radix_lock);
1364         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1365                                 (unsigned long)root->root_key.objectid,
1366                                 root);
1367         if (ret == 0)
1368                 root->in_radix = 1;
1369
1370         spin_unlock(&fs_info->fs_roots_radix_lock);
1371         radix_tree_preload_end();
1372         if (ret) {
1373                 if (ret == -EEXIST) {
1374                         free_fs_root(root);
1375                         goto again;
1376                 }
1377                 goto fail;
1378         }
1379
1380         ret = btrfs_find_dead_roots(fs_info->tree_root,
1381                                     root->root_key.objectid);
1382         WARN_ON(ret);
1383         return root;
1384 fail:
1385         free_fs_root(root);
1386         return ERR_PTR(ret);
1387 }
1388
1389 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1390 {
1391         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1392         int ret = 0;
1393         struct btrfs_device *device;
1394         struct backing_dev_info *bdi;
1395
1396         rcu_read_lock();
1397         list_for_each_entry_rcu(device, &info->fs_devices->devices, dev_list) {
1398                 if (!device->bdev)
1399                         continue;
1400                 bdi = blk_get_backing_dev_info(device->bdev);
1401                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1402                         ret = 1;
1403                         break;
1404                 }
1405         }
1406         rcu_read_unlock();
1407         return ret;
1408 }
1409
1410 /*
1411  * If this fails, caller must call bdi_destroy() to get rid of the
1412  * bdi again.
1413  */
1414 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1415 {
1416         int err;
1417
1418         bdi->capabilities = BDI_CAP_MAP_COPY;
1419         err = bdi_setup_and_register(bdi, "btrfs", BDI_CAP_MAP_COPY);
1420         if (err)
1421                 return err;
1422
1423         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1424         bdi->congested_fn       = btrfs_congested_fn;
1425         bdi->congested_data     = info;
1426         return 0;
1427 }
1428
1429 static int bio_ready_for_csum(struct bio *bio)
1430 {
1431         u64 length = 0;
1432         u64 buf_len = 0;
1433         u64 start = 0;
1434         struct page *page;
1435         struct extent_io_tree *io_tree = NULL;
1436         struct bio_vec *bvec;
1437         int i;
1438         int ret;
1439
1440         bio_for_each_segment(bvec, bio, i) {
1441                 page = bvec->bv_page;
1442                 if (page->private == EXTENT_PAGE_PRIVATE) {
1443                         length += bvec->bv_len;
1444                         continue;
1445                 }
1446                 if (!page->private) {
1447                         length += bvec->bv_len;
1448                         continue;
1449                 }
1450                 length = bvec->bv_len;
1451                 buf_len = page->private >> 2;
1452                 start = page_offset(page) + bvec->bv_offset;
1453                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1454         }
1455         /* are we fully contained in this bio? */
1456         if (buf_len <= length)
1457                 return 1;
1458
1459         ret = extent_range_uptodate(io_tree, start + length,
1460                                     start + buf_len - 1);
1461         return ret;
1462 }
1463
1464 /*
1465  * called by the kthread helper functions to finally call the bio end_io
1466  * functions.  This is where read checksum verification actually happens
1467  */
1468 static void end_workqueue_fn(struct btrfs_work *work)
1469 {
1470         struct bio *bio;
1471         struct end_io_wq *end_io_wq;
1472         struct btrfs_fs_info *fs_info;
1473         int error;
1474
1475         end_io_wq = container_of(work, struct end_io_wq, work);
1476         bio = end_io_wq->bio;
1477         fs_info = end_io_wq->info;
1478
1479         /* metadata bio reads are special because the whole tree block must
1480          * be checksummed at once.  This makes sure the entire block is in
1481          * ram and up to date before trying to verify things.  For
1482          * blocksize <= pagesize, it is basically a noop
1483          */
1484         if (!(bio->bi_rw & REQ_WRITE) && end_io_wq->metadata &&
1485             !bio_ready_for_csum(bio)) {
1486                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1487                                    &end_io_wq->work);
1488                 return;
1489         }
1490         error = end_io_wq->error;
1491         bio->bi_private = end_io_wq->private;
1492         bio->bi_end_io = end_io_wq->end_io;
1493         kfree(end_io_wq);
1494         bio_endio(bio, error);
1495 }
1496
1497 static int cleaner_kthread(void *arg)
1498 {
1499         struct btrfs_root *root = arg;
1500
1501         do {
1502                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1503
1504                 if (!(root->fs_info->sb->s_flags & MS_RDONLY) &&
1505                     mutex_trylock(&root->fs_info->cleaner_mutex)) {
1506                         btrfs_run_delayed_iputs(root);
1507                         btrfs_clean_old_snapshots(root);
1508                         mutex_unlock(&root->fs_info->cleaner_mutex);
1509                         btrfs_run_defrag_inodes(root->fs_info);
1510                 }
1511
1512                 if (freezing(current)) {
1513                         refrigerator();
1514                 } else {
1515                         set_current_state(TASK_INTERRUPTIBLE);
1516                         if (!kthread_should_stop())
1517                                 schedule();
1518                         __set_current_state(TASK_RUNNING);
1519                 }
1520         } while (!kthread_should_stop());
1521         return 0;
1522 }
1523
1524 static int transaction_kthread(void *arg)
1525 {
1526         struct btrfs_root *root = arg;
1527         struct btrfs_trans_handle *trans;
1528         struct btrfs_transaction *cur;
1529         u64 transid;
1530         unsigned long now;
1531         unsigned long delay;
1532         int ret;
1533
1534         do {
1535                 delay = HZ * 30;
1536                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1537                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1538
1539                 spin_lock(&root->fs_info->trans_lock);
1540                 cur = root->fs_info->running_transaction;
1541                 if (!cur) {
1542                         spin_unlock(&root->fs_info->trans_lock);
1543                         goto sleep;
1544                 }
1545
1546                 now = get_seconds();
1547                 if (!cur->blocked &&
1548                     (now < cur->start_time || now - cur->start_time < 30)) {
1549                         spin_unlock(&root->fs_info->trans_lock);
1550                         delay = HZ * 5;
1551                         goto sleep;
1552                 }
1553                 transid = cur->transid;
1554                 spin_unlock(&root->fs_info->trans_lock);
1555
1556                 trans = btrfs_join_transaction(root);
1557                 BUG_ON(IS_ERR(trans));
1558                 if (transid == trans->transid) {
1559                         ret = btrfs_commit_transaction(trans, root);
1560                         BUG_ON(ret);
1561                 } else {
1562                         btrfs_end_transaction(trans, root);
1563                 }
1564 sleep:
1565                 wake_up_process(root->fs_info->cleaner_kthread);
1566                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1567
1568                 if (freezing(current)) {
1569                         refrigerator();
1570                 } else {
1571                         set_current_state(TASK_INTERRUPTIBLE);
1572                         if (!kthread_should_stop() &&
1573                             !btrfs_transaction_blocked(root->fs_info))
1574                                 schedule_timeout(delay);
1575                         __set_current_state(TASK_RUNNING);
1576                 }
1577         } while (!kthread_should_stop());
1578         return 0;
1579 }
1580
1581 /*
1582  * this will find the highest generation in the array of
1583  * root backups.  The index of the highest array is returned,
1584  * or -1 if we can't find anything.
1585  *
1586  * We check to make sure the array is valid by comparing the
1587  * generation of the latest  root in the array with the generation
1588  * in the super block.  If they don't match we pitch it.
1589  */
1590 static int find_newest_super_backup(struct btrfs_fs_info *info, u64 newest_gen)
1591 {
1592         u64 cur;
1593         int newest_index = -1;
1594         struct btrfs_root_backup *root_backup;
1595         int i;
1596
1597         for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
1598                 root_backup = info->super_copy->super_roots + i;
1599                 cur = btrfs_backup_tree_root_gen(root_backup);
1600                 if (cur == newest_gen)
1601                         newest_index = i;
1602         }
1603
1604         /* check to see if we actually wrapped around */
1605         if (newest_index == BTRFS_NUM_BACKUP_ROOTS - 1) {
1606                 root_backup = info->super_copy->super_roots;
1607                 cur = btrfs_backup_tree_root_gen(root_backup);
1608                 if (cur == newest_gen)
1609                         newest_index = 0;
1610         }
1611         return newest_index;
1612 }
1613
1614
1615 /*
1616  * find the oldest backup so we know where to store new entries
1617  * in the backup array.  This will set the backup_root_index
1618  * field in the fs_info struct
1619  */
1620 static void find_oldest_super_backup(struct btrfs_fs_info *info,
1621                                      u64 newest_gen)
1622 {
1623         int newest_index = -1;
1624
1625         newest_index = find_newest_super_backup(info, newest_gen);
1626         /* if there was garbage in there, just move along */
1627         if (newest_index == -1) {
1628                 info->backup_root_index = 0;
1629         } else {
1630                 info->backup_root_index = (newest_index + 1) % BTRFS_NUM_BACKUP_ROOTS;
1631         }
1632 }
1633
1634 /*
1635  * copy all the root pointers into the super backup array.
1636  * this will bump the backup pointer by one when it is
1637  * done
1638  */
1639 static void backup_super_roots(struct btrfs_fs_info *info)
1640 {
1641         int next_backup;
1642         struct btrfs_root_backup *root_backup;
1643         int last_backup;
1644
1645         next_backup = info->backup_root_index;
1646         last_backup = (next_backup + BTRFS_NUM_BACKUP_ROOTS - 1) %
1647                 BTRFS_NUM_BACKUP_ROOTS;
1648
1649         /*
1650          * just overwrite the last backup if we're at the same generation
1651          * this happens only at umount
1652          */
1653         root_backup = info->super_for_commit->super_roots + last_backup;
1654         if (btrfs_backup_tree_root_gen(root_backup) ==
1655             btrfs_header_generation(info->tree_root->node))
1656                 next_backup = last_backup;
1657
1658         root_backup = info->super_for_commit->super_roots + next_backup;
1659
1660         /*
1661          * make sure all of our padding and empty slots get zero filled
1662          * regardless of which ones we use today
1663          */
1664         memset(root_backup, 0, sizeof(*root_backup));
1665
1666         info->backup_root_index = (next_backup + 1) % BTRFS_NUM_BACKUP_ROOTS;
1667
1668         btrfs_set_backup_tree_root(root_backup, info->tree_root->node->start);
1669         btrfs_set_backup_tree_root_gen(root_backup,
1670                                btrfs_header_generation(info->tree_root->node));
1671
1672         btrfs_set_backup_tree_root_level(root_backup,
1673                                btrfs_header_level(info->tree_root->node));
1674
1675         btrfs_set_backup_chunk_root(root_backup, info->chunk_root->node->start);
1676         btrfs_set_backup_chunk_root_gen(root_backup,
1677                                btrfs_header_generation(info->chunk_root->node));
1678         btrfs_set_backup_chunk_root_level(root_backup,
1679                                btrfs_header_level(info->chunk_root->node));
1680
1681         btrfs_set_backup_extent_root(root_backup, info->extent_root->node->start);
1682         btrfs_set_backup_extent_root_gen(root_backup,
1683                                btrfs_header_generation(info->extent_root->node));
1684         btrfs_set_backup_extent_root_level(root_backup,
1685                                btrfs_header_level(info->extent_root->node));
1686
1687         btrfs_set_backup_fs_root(root_backup, info->fs_root->node->start);
1688         btrfs_set_backup_fs_root_gen(root_backup,
1689                                btrfs_header_generation(info->fs_root->node));
1690         btrfs_set_backup_fs_root_level(root_backup,
1691                                btrfs_header_level(info->fs_root->node));
1692
1693         btrfs_set_backup_dev_root(root_backup, info->dev_root->node->start);
1694         btrfs_set_backup_dev_root_gen(root_backup,
1695                                btrfs_header_generation(info->dev_root->node));
1696         btrfs_set_backup_dev_root_level(root_backup,
1697                                        btrfs_header_level(info->dev_root->node));
1698
1699         btrfs_set_backup_csum_root(root_backup, info->csum_root->node->start);
1700         btrfs_set_backup_csum_root_gen(root_backup,
1701                                btrfs_header_generation(info->csum_root->node));
1702         btrfs_set_backup_csum_root_level(root_backup,
1703                                btrfs_header_level(info->csum_root->node));
1704
1705         btrfs_set_backup_total_bytes(root_backup,
1706                              btrfs_super_total_bytes(info->super_copy));
1707         btrfs_set_backup_bytes_used(root_backup,
1708                              btrfs_super_bytes_used(info->super_copy));
1709         btrfs_set_backup_num_devices(root_backup,
1710                              btrfs_super_num_devices(info->super_copy));
1711
1712         /*
1713          * if we don't copy this out to the super_copy, it won't get remembered
1714          * for the next commit
1715          */
1716         memcpy(&info->super_copy->super_roots,
1717                &info->super_for_commit->super_roots,
1718                sizeof(*root_backup) * BTRFS_NUM_BACKUP_ROOTS);
1719 }
1720
1721 /*
1722  * this copies info out of the root backup array and back into
1723  * the in-memory super block.  It is meant to help iterate through
1724  * the array, so you send it the number of backups you've already
1725  * tried and the last backup index you used.
1726  *
1727  * this returns -1 when it has tried all the backups
1728  */
1729 static noinline int next_root_backup(struct btrfs_fs_info *info,
1730                                      struct btrfs_super_block *super,
1731                                      int *num_backups_tried, int *backup_index)
1732 {
1733         struct btrfs_root_backup *root_backup;
1734         int newest = *backup_index;
1735
1736         if (*num_backups_tried == 0) {
1737                 u64 gen = btrfs_super_generation(super);
1738
1739                 newest = find_newest_super_backup(info, gen);
1740                 if (newest == -1)
1741                         return -1;
1742
1743                 *backup_index = newest;
1744                 *num_backups_tried = 1;
1745         } else if (*num_backups_tried == BTRFS_NUM_BACKUP_ROOTS) {
1746                 /* we've tried all the backups, all done */
1747                 return -1;
1748         } else {
1749                 /* jump to the next oldest backup */
1750                 newest = (*backup_index + BTRFS_NUM_BACKUP_ROOTS - 1) %
1751                         BTRFS_NUM_BACKUP_ROOTS;
1752                 *backup_index = newest;
1753                 *num_backups_tried += 1;
1754         }
1755         root_backup = super->super_roots + newest;
1756
1757         btrfs_set_super_generation(super,
1758                                    btrfs_backup_tree_root_gen(root_backup));
1759         btrfs_set_super_root(super, btrfs_backup_tree_root(root_backup));
1760         btrfs_set_super_root_level(super,
1761                                    btrfs_backup_tree_root_level(root_backup));
1762         btrfs_set_super_bytes_used(super, btrfs_backup_bytes_used(root_backup));
1763
1764         /*
1765          * fixme: the total bytes and num_devices need to match or we should
1766          * need a fsck
1767          */
1768         btrfs_set_super_total_bytes(super, btrfs_backup_total_bytes(root_backup));
1769         btrfs_set_super_num_devices(super, btrfs_backup_num_devices(root_backup));
1770         return 0;
1771 }
1772
1773 /* helper to cleanup tree roots */
1774 static void free_root_pointers(struct btrfs_fs_info *info, int chunk_root)
1775 {
1776         free_extent_buffer(info->tree_root->node);
1777         free_extent_buffer(info->tree_root->commit_root);
1778         free_extent_buffer(info->dev_root->node);
1779         free_extent_buffer(info->dev_root->commit_root);
1780         free_extent_buffer(info->extent_root->node);
1781         free_extent_buffer(info->extent_root->commit_root);
1782         free_extent_buffer(info->csum_root->node);
1783         free_extent_buffer(info->csum_root->commit_root);
1784
1785         info->tree_root->node = NULL;
1786         info->tree_root->commit_root = NULL;
1787         info->dev_root->node = NULL;
1788         info->dev_root->commit_root = NULL;
1789         info->extent_root->node = NULL;
1790         info->extent_root->commit_root = NULL;
1791         info->csum_root->node = NULL;
1792         info->csum_root->commit_root = NULL;
1793
1794         if (chunk_root) {
1795                 free_extent_buffer(info->chunk_root->node);
1796                 free_extent_buffer(info->chunk_root->commit_root);
1797                 info->chunk_root->node = NULL;
1798                 info->chunk_root->commit_root = NULL;
1799         }
1800 }
1801
1802
1803 struct btrfs_root *open_ctree(struct super_block *sb,
1804                               struct btrfs_fs_devices *fs_devices,
1805                               char *options)
1806 {
1807         u32 sectorsize;
1808         u32 nodesize;
1809         u32 leafsize;
1810         u32 blocksize;
1811         u32 stripesize;
1812         u64 generation;
1813         u64 features;
1814         struct btrfs_key location;
1815         struct buffer_head *bh;
1816         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1817                                                  GFP_NOFS);
1818         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1819                                                  GFP_NOFS);
1820         struct btrfs_root *tree_root = btrfs_sb(sb);
1821         struct btrfs_fs_info *fs_info = NULL;
1822         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1823                                                 GFP_NOFS);
1824         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1825                                               GFP_NOFS);
1826         struct btrfs_root *log_tree_root;
1827
1828         int ret;
1829         int err = -EINVAL;
1830         int num_backups_tried = 0;
1831         int backup_index = 0;
1832
1833         struct btrfs_super_block *disk_super;
1834
1835         if (!extent_root || !tree_root || !tree_root->fs_info ||
1836             !chunk_root || !dev_root || !csum_root) {
1837                 err = -ENOMEM;
1838                 goto fail;
1839         }
1840         fs_info = tree_root->fs_info;
1841
1842         ret = init_srcu_struct(&fs_info->subvol_srcu);
1843         if (ret) {
1844                 err = ret;
1845                 goto fail;
1846         }
1847
1848         ret = setup_bdi(fs_info, &fs_info->bdi);
1849         if (ret) {
1850                 err = ret;
1851                 goto fail_srcu;
1852         }
1853
1854         fs_info->btree_inode = new_inode(sb);
1855         if (!fs_info->btree_inode) {
1856                 err = -ENOMEM;
1857                 goto fail_bdi;
1858         }
1859
1860         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
1861
1862         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_ATOMIC);
1863         INIT_LIST_HEAD(&fs_info->trans_list);
1864         INIT_LIST_HEAD(&fs_info->dead_roots);
1865         INIT_LIST_HEAD(&fs_info->delayed_iputs);
1866         INIT_LIST_HEAD(&fs_info->hashers);
1867         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1868         INIT_LIST_HEAD(&fs_info->ordered_operations);
1869         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1870         spin_lock_init(&fs_info->delalloc_lock);
1871         spin_lock_init(&fs_info->trans_lock);
1872         spin_lock_init(&fs_info->ref_cache_lock);
1873         spin_lock_init(&fs_info->fs_roots_radix_lock);
1874         spin_lock_init(&fs_info->delayed_iput_lock);
1875         spin_lock_init(&fs_info->defrag_inodes_lock);
1876         spin_lock_init(&fs_info->free_chunk_lock);
1877         mutex_init(&fs_info->reloc_mutex);
1878
1879         init_completion(&fs_info->kobj_unregister);
1880         fs_info->tree_root = tree_root;
1881         fs_info->extent_root = extent_root;
1882         fs_info->csum_root = csum_root;
1883         fs_info->chunk_root = chunk_root;
1884         fs_info->dev_root = dev_root;
1885         fs_info->fs_devices = fs_devices;
1886         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1887         INIT_LIST_HEAD(&fs_info->space_info);
1888         btrfs_mapping_init(&fs_info->mapping_tree);
1889         btrfs_init_block_rsv(&fs_info->global_block_rsv);
1890         btrfs_init_block_rsv(&fs_info->delalloc_block_rsv);
1891         btrfs_init_block_rsv(&fs_info->trans_block_rsv);
1892         btrfs_init_block_rsv(&fs_info->chunk_block_rsv);
1893         btrfs_init_block_rsv(&fs_info->empty_block_rsv);
1894         atomic_set(&fs_info->nr_async_submits, 0);
1895         atomic_set(&fs_info->async_delalloc_pages, 0);
1896         atomic_set(&fs_info->async_submit_draining, 0);
1897         atomic_set(&fs_info->nr_async_bios, 0);
1898         atomic_set(&fs_info->defrag_running, 0);
1899         fs_info->sb = sb;
1900         fs_info->max_inline = 8192 * 1024;
1901         fs_info->metadata_ratio = 0;
1902         fs_info->defrag_inodes = RB_ROOT;
1903         fs_info->trans_no_join = 0;
1904         fs_info->free_chunk_space = 0;
1905
1906         fs_info->thread_pool_size = min_t(unsigned long,
1907                                           num_online_cpus() + 2, 8);
1908
1909         INIT_LIST_HEAD(&fs_info->ordered_extents);
1910         spin_lock_init(&fs_info->ordered_extent_lock);
1911         fs_info->delayed_root = kmalloc(sizeof(struct btrfs_delayed_root),
1912                                         GFP_NOFS);
1913         if (!fs_info->delayed_root) {
1914                 err = -ENOMEM;
1915                 goto fail_iput;
1916         }
1917         btrfs_init_delayed_root(fs_info->delayed_root);
1918
1919         mutex_init(&fs_info->scrub_lock);
1920         atomic_set(&fs_info->scrubs_running, 0);
1921         atomic_set(&fs_info->scrub_pause_req, 0);
1922         atomic_set(&fs_info->scrubs_paused, 0);
1923         atomic_set(&fs_info->scrub_cancel_req, 0);
1924         init_waitqueue_head(&fs_info->scrub_pause_wait);
1925         init_rwsem(&fs_info->scrub_super_lock);
1926         fs_info->scrub_workers_refcnt = 0;
1927
1928         sb->s_blocksize = 4096;
1929         sb->s_blocksize_bits = blksize_bits(4096);
1930         sb->s_bdi = &fs_info->bdi;
1931
1932         fs_info->btree_inode->i_ino = BTRFS_BTREE_INODE_OBJECTID;
1933         fs_info->btree_inode->i_nlink = 1;
1934         /*
1935          * we set the i_size on the btree inode to the max possible int.
1936          * the real end of the address space is determined by all of
1937          * the devices in the system
1938          */
1939         fs_info->btree_inode->i_size = OFFSET_MAX;
1940         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1941         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1942
1943         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1944         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1945                              fs_info->btree_inode->i_mapping);
1946         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree);
1947
1948         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1949
1950         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1951         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1952                sizeof(struct btrfs_key));
1953         BTRFS_I(fs_info->btree_inode)->dummy_inode = 1;
1954         insert_inode_hash(fs_info->btree_inode);
1955
1956         spin_lock_init(&fs_info->block_group_cache_lock);
1957         fs_info->block_group_cache_tree = RB_ROOT;
1958
1959         extent_io_tree_init(&fs_info->freed_extents[0],
1960                              fs_info->btree_inode->i_mapping);
1961         extent_io_tree_init(&fs_info->freed_extents[1],
1962                              fs_info->btree_inode->i_mapping);
1963         fs_info->pinned_extents = &fs_info->freed_extents[0];
1964         fs_info->do_barriers = 1;
1965
1966
1967         mutex_init(&fs_info->ordered_operations_mutex);
1968         mutex_init(&fs_info->tree_log_mutex);
1969         mutex_init(&fs_info->chunk_mutex);
1970         mutex_init(&fs_info->transaction_kthread_mutex);
1971         mutex_init(&fs_info->cleaner_mutex);
1972         mutex_init(&fs_info->volume_mutex);
1973         init_rwsem(&fs_info->extent_commit_sem);
1974         init_rwsem(&fs_info->cleanup_work_sem);
1975         init_rwsem(&fs_info->subvol_sem);
1976
1977         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1978         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1979
1980         init_waitqueue_head(&fs_info->transaction_throttle);
1981         init_waitqueue_head(&fs_info->transaction_wait);
1982         init_waitqueue_head(&fs_info->transaction_blocked_wait);
1983         init_waitqueue_head(&fs_info->async_submit_wait);
1984
1985         __setup_root(4096, 4096, 4096, 4096, tree_root,
1986                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1987
1988         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1989         if (!bh) {
1990                 err = -EINVAL;
1991                 goto fail_alloc;
1992         }
1993
1994         memcpy(fs_info->super_copy, bh->b_data, sizeof(*fs_info->super_copy));
1995         memcpy(fs_info->super_for_commit, fs_info->super_copy,
1996                sizeof(*fs_info->super_for_commit));
1997         brelse(bh);
1998
1999         memcpy(fs_info->fsid, fs_info->super_copy->fsid, BTRFS_FSID_SIZE);
2000
2001         disk_super = fs_info->super_copy;
2002         if (!btrfs_super_root(disk_super))
2003                 goto fail_alloc;
2004
2005         /* check FS state, whether FS is broken. */
2006         fs_info->fs_state |= btrfs_super_flags(disk_super);
2007
2008         btrfs_check_super_valid(fs_info, sb->s_flags & MS_RDONLY);
2009
2010         /*
2011          * run through our array of backup supers and setup
2012          * our ring pointer to the oldest one
2013          */
2014         generation = btrfs_super_generation(disk_super);
2015         find_oldest_super_backup(fs_info, generation);
2016
2017         /*
2018          * In the long term, we'll store the compression type in the super
2019          * block, and it'll be used for per file compression control.
2020          */
2021         fs_info->compress_type = BTRFS_COMPRESS_ZLIB;
2022
2023         ret = btrfs_parse_options(tree_root, options);
2024         if (ret) {
2025                 err = ret;
2026                 goto fail_alloc;
2027         }
2028
2029         features = btrfs_super_incompat_flags(disk_super) &
2030                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
2031         if (features) {
2032                 printk(KERN_ERR "BTRFS: couldn't mount because of "
2033                        "unsupported optional features (%Lx).\n",
2034                        (unsigned long long)features);
2035                 err = -EINVAL;
2036                 goto fail_alloc;
2037         }
2038
2039         features = btrfs_super_incompat_flags(disk_super);
2040         features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
2041         if (tree_root->fs_info->compress_type & BTRFS_COMPRESS_LZO)
2042                 features |= BTRFS_FEATURE_INCOMPAT_COMPRESS_LZO;
2043         btrfs_set_super_incompat_flags(disk_super, features);
2044
2045         features = btrfs_super_compat_ro_flags(disk_super) &
2046                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
2047         if (!(sb->s_flags & MS_RDONLY) && features) {
2048                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
2049                        "unsupported option features (%Lx).\n",
2050                        (unsigned long long)features);
2051                 err = -EINVAL;
2052                 goto fail_alloc;
2053         }
2054
2055         btrfs_init_workers(&fs_info->generic_worker,
2056                            "genwork", 1, NULL);
2057
2058         btrfs_init_workers(&fs_info->workers, "worker",
2059                            fs_info->thread_pool_size,
2060                            &fs_info->generic_worker);
2061
2062         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
2063                            fs_info->thread_pool_size,
2064                            &fs_info->generic_worker);
2065
2066         btrfs_init_workers(&fs_info->submit_workers, "submit",
2067                            min_t(u64, fs_devices->num_devices,
2068                            fs_info->thread_pool_size),
2069                            &fs_info->generic_worker);
2070
2071         btrfs_init_workers(&fs_info->caching_workers, "cache",
2072                            2, &fs_info->generic_worker);
2073
2074         /* a higher idle thresh on the submit workers makes it much more
2075          * likely that bios will be send down in a sane order to the
2076          * devices
2077          */
2078         fs_info->submit_workers.idle_thresh = 64;
2079
2080         fs_info->workers.idle_thresh = 16;
2081         fs_info->workers.ordered = 1;
2082
2083         fs_info->delalloc_workers.idle_thresh = 2;
2084         fs_info->delalloc_workers.ordered = 1;
2085
2086         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1,
2087                            &fs_info->generic_worker);
2088         btrfs_init_workers(&fs_info->endio_workers, "endio",
2089                            fs_info->thread_pool_size,
2090                            &fs_info->generic_worker);
2091         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
2092                            fs_info->thread_pool_size,
2093                            &fs_info->generic_worker);
2094         btrfs_init_workers(&fs_info->endio_meta_write_workers,
2095                            "endio-meta-write", fs_info->thread_pool_size,
2096                            &fs_info->generic_worker);
2097         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
2098                            fs_info->thread_pool_size,
2099                            &fs_info->generic_worker);
2100         btrfs_init_workers(&fs_info->endio_freespace_worker, "freespace-write",
2101                            1, &fs_info->generic_worker);
2102         btrfs_init_workers(&fs_info->delayed_workers, "delayed-meta",
2103                            fs_info->thread_pool_size,
2104                            &fs_info->generic_worker);
2105
2106         /*
2107          * endios are largely parallel and should have a very
2108          * low idle thresh
2109          */
2110         fs_info->endio_workers.idle_thresh = 4;
2111         fs_info->endio_meta_workers.idle_thresh = 4;
2112
2113         fs_info->endio_write_workers.idle_thresh = 2;
2114         fs_info->endio_meta_write_workers.idle_thresh = 2;
2115
2116         btrfs_start_workers(&fs_info->workers, 1);
2117         btrfs_start_workers(&fs_info->generic_worker, 1);
2118         btrfs_start_workers(&fs_info->submit_workers, 1);
2119         btrfs_start_workers(&fs_info->delalloc_workers, 1);
2120         btrfs_start_workers(&fs_info->fixup_workers, 1);
2121         btrfs_start_workers(&fs_info->endio_workers, 1);
2122         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
2123         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
2124         btrfs_start_workers(&fs_info->endio_write_workers, 1);
2125         btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
2126         btrfs_start_workers(&fs_info->delayed_workers, 1);
2127         btrfs_start_workers(&fs_info->caching_workers, 1);
2128
2129         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
2130         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
2131                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
2132
2133         nodesize = btrfs_super_nodesize(disk_super);
2134         leafsize = btrfs_super_leafsize(disk_super);
2135         sectorsize = btrfs_super_sectorsize(disk_super);
2136         stripesize = btrfs_super_stripesize(disk_super);
2137         tree_root->nodesize = nodesize;
2138         tree_root->leafsize = leafsize;
2139         tree_root->sectorsize = sectorsize;
2140         tree_root->stripesize = stripesize;
2141
2142         sb->s_blocksize = sectorsize;
2143         sb->s_blocksize_bits = blksize_bits(sectorsize);
2144
2145         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
2146                     sizeof(disk_super->magic))) {
2147                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
2148                 goto fail_sb_buffer;
2149         }
2150
2151         mutex_lock(&fs_info->chunk_mutex);
2152         ret = btrfs_read_sys_array(tree_root);
2153         mutex_unlock(&fs_info->chunk_mutex);
2154         if (ret) {
2155                 printk(KERN_WARNING "btrfs: failed to read the system "
2156                        "array on %s\n", sb->s_id);
2157                 goto fail_sb_buffer;
2158         }
2159
2160         blocksize = btrfs_level_size(tree_root,
2161                                      btrfs_super_chunk_root_level(disk_super));
2162         generation = btrfs_super_chunk_root_generation(disk_super);
2163
2164         __setup_root(nodesize, leafsize, sectorsize, stripesize,
2165                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
2166
2167         chunk_root->node = read_tree_block(chunk_root,
2168                                            btrfs_super_chunk_root(disk_super),
2169                                            blocksize, generation);
2170         BUG_ON(!chunk_root->node);
2171         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
2172                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
2173                        sb->s_id);
2174                 goto fail_tree_roots;
2175         }
2176         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
2177         chunk_root->commit_root = btrfs_root_node(chunk_root);
2178
2179         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
2180            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
2181            BTRFS_UUID_SIZE);
2182
2183         mutex_lock(&fs_info->chunk_mutex);
2184         ret = btrfs_read_chunk_tree(chunk_root);
2185         mutex_unlock(&fs_info->chunk_mutex);
2186         if (ret) {
2187                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
2188                        sb->s_id);
2189                 goto fail_tree_roots;
2190         }
2191
2192         btrfs_close_extra_devices(fs_devices);
2193
2194 retry_root_backup:
2195         blocksize = btrfs_level_size(tree_root,
2196                                      btrfs_super_root_level(disk_super));
2197         generation = btrfs_super_generation(disk_super);
2198
2199         tree_root->node = read_tree_block(tree_root,
2200                                           btrfs_super_root(disk_super),
2201                                           blocksize, generation);
2202         if (!tree_root->node ||
2203             !test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
2204                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
2205                        sb->s_id);
2206
2207                 goto recovery_tree_root;
2208         }
2209
2210         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
2211         tree_root->commit_root = btrfs_root_node(tree_root);
2212
2213         ret = find_and_setup_root(tree_root, fs_info,
2214                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
2215         if (ret)
2216                 goto recovery_tree_root;
2217         extent_root->track_dirty = 1;
2218
2219         ret = find_and_setup_root(tree_root, fs_info,
2220                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
2221         if (ret)
2222                 goto recovery_tree_root;
2223         dev_root->track_dirty = 1;
2224
2225         ret = find_and_setup_root(tree_root, fs_info,
2226                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
2227         if (ret)
2228                 goto recovery_tree_root;
2229
2230         csum_root->track_dirty = 1;
2231
2232         fs_info->generation = generation;
2233         fs_info->last_trans_committed = generation;
2234         fs_info->data_alloc_profile = (u64)-1;
2235         fs_info->metadata_alloc_profile = (u64)-1;
2236         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
2237
2238         ret = btrfs_init_space_info(fs_info);
2239         if (ret) {
2240                 printk(KERN_ERR "Failed to initial space info: %d\n", ret);
2241                 goto fail_block_groups;
2242         }
2243
2244         ret = btrfs_read_block_groups(extent_root);
2245         if (ret) {
2246                 printk(KERN_ERR "Failed to read block groups: %d\n", ret);
2247                 goto fail_block_groups;
2248         }
2249
2250         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
2251                                                "btrfs-cleaner");
2252         if (IS_ERR(fs_info->cleaner_kthread))
2253                 goto fail_block_groups;
2254
2255         fs_info->transaction_kthread = kthread_run(transaction_kthread,
2256                                                    tree_root,
2257                                                    "btrfs-transaction");
2258         if (IS_ERR(fs_info->transaction_kthread))
2259                 goto fail_cleaner;
2260
2261         if (!btrfs_test_opt(tree_root, SSD) &&
2262             !btrfs_test_opt(tree_root, NOSSD) &&
2263             !fs_info->fs_devices->rotating) {
2264                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
2265                        "mode\n");
2266                 btrfs_set_opt(fs_info->mount_opt, SSD);
2267         }
2268
2269         /* do not make disk changes in broken FS */
2270         if (btrfs_super_log_root(disk_super) != 0 &&
2271             !(fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)) {
2272                 u64 bytenr = btrfs_super_log_root(disk_super);
2273
2274                 if (fs_devices->rw_devices == 0) {
2275                         printk(KERN_WARNING "Btrfs log replay required "
2276                                "on RO media\n");
2277                         err = -EIO;
2278                         goto fail_trans_kthread;
2279                 }
2280                 blocksize =
2281                      btrfs_level_size(tree_root,
2282                                       btrfs_super_log_root_level(disk_super));
2283
2284                 log_tree_root = kzalloc(sizeof(struct btrfs_root), GFP_NOFS);
2285                 if (!log_tree_root) {
2286                         err = -ENOMEM;
2287                         goto fail_trans_kthread;
2288                 }
2289
2290                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
2291                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
2292
2293                 log_tree_root->node = read_tree_block(tree_root, bytenr,
2294                                                       blocksize,
2295                                                       generation + 1);
2296                 ret = btrfs_recover_log_trees(log_tree_root);
2297                 BUG_ON(ret);
2298
2299                 if (sb->s_flags & MS_RDONLY) {
2300                         ret =  btrfs_commit_super(tree_root);
2301                         BUG_ON(ret);
2302                 }
2303         }
2304
2305         ret = btrfs_find_orphan_roots(tree_root);
2306         BUG_ON(ret);
2307
2308         if (!(sb->s_flags & MS_RDONLY)) {
2309                 ret = btrfs_cleanup_fs_roots(fs_info);
2310                 BUG_ON(ret);
2311
2312                 ret = btrfs_recover_relocation(tree_root);
2313                 if (ret < 0) {
2314                         printk(KERN_WARNING
2315                                "btrfs: failed to recover relocation\n");
2316                         err = -EINVAL;
2317                         goto fail_trans_kthread;
2318                 }
2319         }
2320
2321         location.objectid = BTRFS_FS_TREE_OBJECTID;
2322         location.type = BTRFS_ROOT_ITEM_KEY;
2323         location.offset = (u64)-1;
2324
2325         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
2326         if (!fs_info->fs_root)
2327                 goto fail_trans_kthread;
2328         if (IS_ERR(fs_info->fs_root)) {
2329                 err = PTR_ERR(fs_info->fs_root);
2330                 goto fail_trans_kthread;
2331         }
2332
2333         if (!(sb->s_flags & MS_RDONLY)) {
2334                 down_read(&fs_info->cleanup_work_sem);
2335                 err = btrfs_orphan_cleanup(fs_info->fs_root);
2336                 if (!err)
2337                         err = btrfs_orphan_cleanup(fs_info->tree_root);
2338                 up_read(&fs_info->cleanup_work_sem);
2339                 if (err) {
2340                         close_ctree(tree_root);
2341                         return ERR_PTR(err);
2342                 }
2343         }
2344
2345         return tree_root;
2346
2347 fail_trans_kthread:
2348         kthread_stop(fs_info->transaction_kthread);
2349 fail_cleaner:
2350         kthread_stop(fs_info->cleaner_kthread);
2351
2352         /*
2353          * make sure we're done with the btree inode before we stop our
2354          * kthreads
2355          */
2356         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
2357         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2358
2359 fail_block_groups:
2360         btrfs_free_block_groups(fs_info);
2361
2362 fail_tree_roots:
2363         free_root_pointers(fs_info, 1);
2364
2365 fail_sb_buffer:
2366         btrfs_stop_workers(&fs_info->generic_worker);
2367         btrfs_stop_workers(&fs_info->fixup_workers);
2368         btrfs_stop_workers(&fs_info->delalloc_workers);
2369         btrfs_stop_workers(&fs_info->workers);
2370         btrfs_stop_workers(&fs_info->endio_workers);
2371         btrfs_stop_workers(&fs_info->endio_meta_workers);
2372         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2373         btrfs_stop_workers(&fs_info->endio_write_workers);
2374         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2375         btrfs_stop_workers(&fs_info->submit_workers);
2376         btrfs_stop_workers(&fs_info->delayed_workers);
2377         btrfs_stop_workers(&fs_info->caching_workers);
2378 fail_alloc:
2379 fail_iput:
2380         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
2381         iput(fs_info->btree_inode);
2382
2383         btrfs_close_devices(fs_info->fs_devices);
2384         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2385 fail_bdi:
2386         bdi_destroy(&fs_info->bdi);
2387 fail_srcu:
2388         cleanup_srcu_struct(&fs_info->subvol_srcu);
2389 fail:
2390         free_fs_info(fs_info);
2391         return ERR_PTR(err);
2392
2393 recovery_tree_root:
2394
2395         if (!btrfs_test_opt(tree_root, RECOVERY))
2396                 goto fail_tree_roots;
2397
2398         free_root_pointers(fs_info, 0);
2399
2400         /* don't use the log in recovery mode, it won't be valid */
2401         btrfs_set_super_log_root(disk_super, 0);
2402
2403         /* we can't trust the free space cache either */
2404         btrfs_set_opt(fs_info->mount_opt, CLEAR_CACHE);
2405
2406         ret = next_root_backup(fs_info, fs_info->super_copy,
2407                                &num_backups_tried, &backup_index);
2408         if (ret == -1)
2409                 goto fail_block_groups;
2410         goto retry_root_backup;
2411 }
2412
2413 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
2414 {
2415         char b[BDEVNAME_SIZE];
2416
2417         if (uptodate) {
2418                 set_buffer_uptodate(bh);
2419         } else {
2420                 printk_ratelimited(KERN_WARNING "lost page write due to "
2421                                         "I/O error on %s\n",
2422                                        bdevname(bh->b_bdev, b));
2423                 /* note, we dont' set_buffer_write_io_error because we have
2424                  * our own ways of dealing with the IO errors
2425                  */
2426                 clear_buffer_uptodate(bh);
2427         }
2428         unlock_buffer(bh);
2429         put_bh(bh);
2430 }
2431
2432 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2433 {
2434         struct buffer_head *bh;
2435         struct buffer_head *latest = NULL;
2436         struct btrfs_super_block *super;
2437         int i;
2438         u64 transid = 0;
2439         u64 bytenr;
2440
2441         /* we would like to check all the supers, but that would make
2442          * a btrfs mount succeed after a mkfs from a different FS.
2443          * So, we need to add a special mount option to scan for
2444          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2445          */
2446         for (i = 0; i < 1; i++) {
2447                 bytenr = btrfs_sb_offset(i);
2448                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2449                         break;
2450                 bh = __bread(bdev, bytenr / 4096, 4096);
2451                 if (!bh)
2452                         continue;
2453
2454                 super = (struct btrfs_super_block *)bh->b_data;
2455                 if (btrfs_super_bytenr(super) != bytenr ||
2456                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2457                             sizeof(super->magic))) {
2458                         brelse(bh);
2459                         continue;
2460                 }
2461
2462                 if (!latest || btrfs_super_generation(super) > transid) {
2463                         brelse(latest);
2464                         latest = bh;
2465                         transid = btrfs_super_generation(super);
2466                 } else {
2467                         brelse(bh);
2468                 }
2469         }
2470         return latest;
2471 }
2472
2473 /*
2474  * this should be called twice, once with wait == 0 and
2475  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2476  * we write are pinned.
2477  *
2478  * They are released when wait == 1 is done.
2479  * max_mirrors must be the same for both runs, and it indicates how
2480  * many supers on this one device should be written.
2481  *
2482  * max_mirrors == 0 means to write them all.
2483  */
2484 static int write_dev_supers(struct btrfs_device *device,
2485                             struct btrfs_super_block *sb,
2486                             int do_barriers, int wait, int max_mirrors)
2487 {
2488         struct buffer_head *bh;
2489         int i;
2490         int ret;
2491         int errors = 0;
2492         u32 crc;
2493         u64 bytenr;
2494         int last_barrier = 0;
2495
2496         if (max_mirrors == 0)
2497                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2498
2499         /* make sure only the last submit_bh does a barrier */
2500         if (do_barriers) {
2501                 for (i = 0; i < max_mirrors; i++) {
2502                         bytenr = btrfs_sb_offset(i);
2503                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2504                             device->total_bytes)
2505                                 break;
2506                         last_barrier = i;
2507                 }
2508         }
2509
2510         for (i = 0; i < max_mirrors; i++) {
2511                 bytenr = btrfs_sb_offset(i);
2512                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2513                         break;
2514
2515                 if (wait) {
2516                         bh = __find_get_block(device->bdev, bytenr / 4096,
2517                                               BTRFS_SUPER_INFO_SIZE);
2518                         BUG_ON(!bh);
2519                         wait_on_buffer(bh);
2520                         if (!buffer_uptodate(bh))
2521                                 errors++;
2522
2523                         /* drop our reference */
2524                         brelse(bh);
2525
2526                         /* drop the reference from the wait == 0 run */
2527                         brelse(bh);
2528                         continue;
2529                 } else {
2530                         btrfs_set_super_bytenr(sb, bytenr);
2531
2532                         crc = ~(u32)0;
2533                         crc = btrfs_csum_data(NULL, (char *)sb +
2534                                               BTRFS_CSUM_SIZE, crc,
2535                                               BTRFS_SUPER_INFO_SIZE -
2536                                               BTRFS_CSUM_SIZE);
2537                         btrfs_csum_final(crc, sb->csum);
2538
2539                         /*
2540                          * one reference for us, and we leave it for the
2541                          * caller
2542                          */
2543                         bh = __getblk(device->bdev, bytenr / 4096,
2544                                       BTRFS_SUPER_INFO_SIZE);
2545                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2546
2547                         /* one reference for submit_bh */
2548                         get_bh(bh);
2549
2550                         set_buffer_uptodate(bh);
2551                         lock_buffer(bh);
2552                         bh->b_end_io = btrfs_end_buffer_write_sync;
2553                 }
2554
2555                 if (i == last_barrier && do_barriers)
2556                         ret = submit_bh(WRITE_FLUSH_FUA, bh);
2557                 else
2558                         ret = submit_bh(WRITE_SYNC, bh);
2559
2560                 if (ret)
2561                         errors++;
2562         }
2563         return errors < i ? 0 : -1;
2564 }
2565
2566 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2567 {
2568         struct list_head *head;
2569         struct btrfs_device *dev;
2570         struct btrfs_super_block *sb;
2571         struct btrfs_dev_item *dev_item;
2572         int ret;
2573         int do_barriers;
2574         int max_errors;
2575         int total_errors = 0;
2576         u64 flags;
2577
2578         max_errors = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
2579         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2580         backup_super_roots(root->fs_info);
2581
2582         sb = root->fs_info->super_for_commit;
2583         dev_item = &sb->dev_item;
2584
2585         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2586         head = &root->fs_info->fs_devices->devices;
2587         list_for_each_entry_rcu(dev, head, dev_list) {
2588                 if (!dev->bdev) {
2589                         total_errors++;
2590                         continue;
2591                 }
2592                 if (!dev->in_fs_metadata || !dev->writeable)
2593                         continue;
2594
2595                 btrfs_set_stack_device_generation(dev_item, 0);
2596                 btrfs_set_stack_device_type(dev_item, dev->type);
2597                 btrfs_set_stack_device_id(dev_item, dev->devid);
2598                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2599                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2600                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2601                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2602                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2603                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2604                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2605
2606                 flags = btrfs_super_flags(sb);
2607                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2608
2609                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2610                 if (ret)
2611                         total_errors++;
2612         }
2613         if (total_errors > max_errors) {
2614                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2615                        total_errors);
2616                 BUG();
2617         }
2618
2619         total_errors = 0;
2620         list_for_each_entry_rcu(dev, head, dev_list) {
2621                 if (!dev->bdev)
2622                         continue;
2623                 if (!dev->in_fs_metadata || !dev->writeable)
2624                         continue;
2625
2626                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2627                 if (ret)
2628                         total_errors++;
2629         }
2630         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2631         if (total_errors > max_errors) {
2632                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2633                        total_errors);
2634                 BUG();
2635         }
2636         return 0;
2637 }
2638
2639 int write_ctree_super(struct btrfs_trans_handle *trans,
2640                       struct btrfs_root *root, int max_mirrors)
2641 {
2642         int ret;
2643
2644         ret = write_all_supers(root, max_mirrors);
2645         return ret;
2646 }
2647
2648 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2649 {
2650         spin_lock(&fs_info->fs_roots_radix_lock);
2651         radix_tree_delete(&fs_info->fs_roots_radix,
2652                           (unsigned long)root->root_key.objectid);
2653         spin_unlock(&fs_info->fs_roots_radix_lock);
2654
2655         if (btrfs_root_refs(&root->root_item) == 0)
2656                 synchronize_srcu(&fs_info->subvol_srcu);
2657
2658         __btrfs_remove_free_space_cache(root->free_ino_pinned);
2659         __btrfs_remove_free_space_cache(root->free_ino_ctl);
2660         free_fs_root(root);
2661         return 0;
2662 }
2663
2664 static void free_fs_root(struct btrfs_root *root)
2665 {
2666         iput(root->cache_inode);
2667         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2668         if (root->anon_dev)
2669                 free_anon_bdev(root->anon_dev);
2670         free_extent_buffer(root->node);
2671         free_extent_buffer(root->commit_root);
2672         kfree(root->free_ino_ctl);
2673         kfree(root->free_ino_pinned);
2674         kfree(root->name);
2675         kfree(root);
2676 }
2677
2678 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2679 {
2680         int ret;
2681         struct btrfs_root *gang[8];
2682         int i;
2683
2684         while (!list_empty(&fs_info->dead_roots)) {
2685                 gang[0] = list_entry(fs_info->dead_roots.next,
2686                                      struct btrfs_root, root_list);
2687                 list_del(&gang[0]->root_list);
2688
2689                 if (gang[0]->in_radix) {
2690                         btrfs_free_fs_root(fs_info, gang[0]);
2691                 } else {
2692                         free_extent_buffer(gang[0]->node);
2693                         free_extent_buffer(gang[0]->commit_root);
2694                         kfree(gang[0]);
2695                 }
2696         }
2697
2698         while (1) {
2699                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2700                                              (void **)gang, 0,
2701                                              ARRAY_SIZE(gang));
2702                 if (!ret)
2703                         break;
2704                 for (i = 0; i < ret; i++)
2705                         btrfs_free_fs_root(fs_info, gang[i]);
2706         }
2707         return 0;
2708 }
2709
2710 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2711 {
2712         u64 root_objectid = 0;
2713         struct btrfs_root *gang[8];
2714         int i;
2715         int ret;
2716
2717         while (1) {
2718                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2719                                              (void **)gang, root_objectid,
2720                                              ARRAY_SIZE(gang));
2721                 if (!ret)
2722                         break;
2723
2724                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2725                 for (i = 0; i < ret; i++) {
2726                         int err;
2727
2728                         root_objectid = gang[i]->root_key.objectid;
2729                         err = btrfs_orphan_cleanup(gang[i]);
2730                         if (err)
2731                                 return err;
2732                 }
2733                 root_objectid++;
2734         }
2735         return 0;
2736 }
2737
2738 int btrfs_commit_super(struct btrfs_root *root)
2739 {
2740         struct btrfs_trans_handle *trans;
2741         int ret;
2742
2743         mutex_lock(&root->fs_info->cleaner_mutex);
2744         btrfs_run_delayed_iputs(root);
2745         btrfs_clean_old_snapshots(root);
2746         mutex_unlock(&root->fs_info->cleaner_mutex);
2747
2748         /* wait until ongoing cleanup work done */
2749         down_write(&root->fs_info->cleanup_work_sem);
2750         up_write(&root->fs_info->cleanup_work_sem);
2751
2752         trans = btrfs_join_transaction(root);
2753         if (IS_ERR(trans))
2754                 return PTR_ERR(trans);
2755         ret = btrfs_commit_transaction(trans, root);
2756         BUG_ON(ret);
2757         /* run commit again to drop the original snapshot */
2758         trans = btrfs_join_transaction(root);
2759         if (IS_ERR(trans))
2760                 return PTR_ERR(trans);
2761         btrfs_commit_transaction(trans, root);
2762         ret = btrfs_write_and_wait_transaction(NULL, root);
2763         BUG_ON(ret);
2764
2765         ret = write_ctree_super(NULL, root, 0);
2766         return ret;
2767 }
2768
2769 int close_ctree(struct btrfs_root *root)
2770 {
2771         struct btrfs_fs_info *fs_info = root->fs_info;
2772         int ret;
2773
2774         fs_info->closing = 1;
2775         smp_mb();
2776
2777         btrfs_scrub_cancel(root);
2778
2779         /* wait for any defraggers to finish */
2780         wait_event(fs_info->transaction_wait,
2781                    (atomic_read(&fs_info->defrag_running) == 0));
2782
2783         /* clear out the rbtree of defraggable inodes */
2784         btrfs_run_defrag_inodes(root->fs_info);
2785
2786         /*
2787          * Here come 2 situations when btrfs is broken to flip readonly:
2788          *
2789          * 1. when btrfs flips readonly somewhere else before
2790          * btrfs_commit_super, sb->s_flags has MS_RDONLY flag,
2791          * and btrfs will skip to write sb directly to keep
2792          * ERROR state on disk.
2793          *
2794          * 2. when btrfs flips readonly just in btrfs_commit_super,
2795          * and in such case, btrfs cannot write sb via btrfs_commit_super,
2796          * and since fs_state has been set BTRFS_SUPER_FLAG_ERROR flag,
2797          * btrfs will cleanup all FS resources first and write sb then.
2798          */
2799         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2800                 ret = btrfs_commit_super(root);
2801                 if (ret)
2802                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2803         }
2804
2805         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) {
2806                 ret = btrfs_error_commit_super(root);
2807                 if (ret)
2808                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2809         }
2810
2811         btrfs_put_block_group_cache(fs_info);
2812
2813         kthread_stop(root->fs_info->transaction_kthread);
2814         kthread_stop(root->fs_info->cleaner_kthread);
2815
2816         fs_info->closing = 2;
2817         smp_mb();
2818
2819         if (fs_info->delalloc_bytes) {
2820                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2821                        (unsigned long long)fs_info->delalloc_bytes);
2822         }
2823         if (fs_info->total_ref_cache_size) {
2824                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2825                        (unsigned long long)fs_info->total_ref_cache_size);
2826         }
2827
2828         free_extent_buffer(fs_info->extent_root->node);
2829         free_extent_buffer(fs_info->extent_root->commit_root);
2830         free_extent_buffer(fs_info->tree_root->node);
2831         free_extent_buffer(fs_info->tree_root->commit_root);
2832         free_extent_buffer(root->fs_info->chunk_root->node);
2833         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2834         free_extent_buffer(root->fs_info->dev_root->node);
2835         free_extent_buffer(root->fs_info->dev_root->commit_root);
2836         free_extent_buffer(root->fs_info->csum_root->node);
2837         free_extent_buffer(root->fs_info->csum_root->commit_root);
2838
2839         btrfs_free_block_groups(root->fs_info);
2840
2841         del_fs_roots(fs_info);
2842
2843         iput(fs_info->btree_inode);
2844
2845         btrfs_stop_workers(&fs_info->generic_worker);
2846         btrfs_stop_workers(&fs_info->fixup_workers);
2847         btrfs_stop_workers(&fs_info->delalloc_workers);
2848         btrfs_stop_workers(&fs_info->workers);
2849         btrfs_stop_workers(&fs_info->endio_workers);
2850         btrfs_stop_workers(&fs_info->endio_meta_workers);
2851         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2852         btrfs_stop_workers(&fs_info->endio_write_workers);
2853         btrfs_stop_workers(&fs_info->endio_freespace_worker);
2854         btrfs_stop_workers(&fs_info->submit_workers);
2855         btrfs_stop_workers(&fs_info->delayed_workers);
2856         btrfs_stop_workers(&fs_info->caching_workers);
2857
2858         btrfs_close_devices(fs_info->fs_devices);
2859         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2860
2861         bdi_destroy(&fs_info->bdi);
2862         cleanup_srcu_struct(&fs_info->subvol_srcu);
2863
2864         free_fs_info(fs_info);
2865
2866         return 0;
2867 }
2868
2869 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2870 {
2871         int ret;
2872         struct inode *btree_inode = buf->first_page->mapping->host;
2873
2874         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf,
2875                                      NULL);
2876         if (!ret)
2877                 return ret;
2878
2879         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2880                                     parent_transid);
2881         return !ret;
2882 }
2883
2884 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2885 {
2886         struct inode *btree_inode = buf->first_page->mapping->host;
2887         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2888                                           buf);
2889 }
2890
2891 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2892 {
2893         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2894         u64 transid = btrfs_header_generation(buf);
2895         struct inode *btree_inode = root->fs_info->btree_inode;
2896         int was_dirty;
2897
2898         btrfs_assert_tree_locked(buf);
2899         if (transid != root->fs_info->generation) {
2900                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2901                        "found %llu running %llu\n",
2902                         (unsigned long long)buf->start,
2903                         (unsigned long long)transid,
2904                         (unsigned long long)root->fs_info->generation);
2905                 WARN_ON(1);
2906         }
2907         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2908                                             buf);
2909         if (!was_dirty) {
2910                 spin_lock(&root->fs_info->delalloc_lock);
2911                 root->fs_info->dirty_metadata_bytes += buf->len;
2912                 spin_unlock(&root->fs_info->delalloc_lock);
2913         }
2914 }
2915
2916 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2917 {
2918         /*
2919          * looks as though older kernels can get into trouble with
2920          * this code, they end up stuck in balance_dirty_pages forever
2921          */
2922         u64 num_dirty;
2923         unsigned long thresh = 32 * 1024 * 1024;
2924
2925         if (current->flags & PF_MEMALLOC)
2926                 return;
2927
2928         btrfs_balance_delayed_items(root);
2929
2930         num_dirty = root->fs_info->dirty_metadata_bytes;
2931
2932         if (num_dirty > thresh) {
2933                 balance_dirty_pages_ratelimited_nr(
2934                                    root->fs_info->btree_inode->i_mapping, 1);
2935         }
2936         return;
2937 }
2938
2939 void __btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2940 {
2941         /*
2942          * looks as though older kernels can get into trouble with
2943          * this code, they end up stuck in balance_dirty_pages forever
2944          */
2945         u64 num_dirty;
2946         unsigned long thresh = 32 * 1024 * 1024;
2947
2948         if (current->flags & PF_MEMALLOC)
2949                 return;
2950
2951         num_dirty = root->fs_info->dirty_metadata_bytes;
2952
2953         if (num_dirty > thresh) {
2954                 balance_dirty_pages_ratelimited_nr(
2955                                    root->fs_info->btree_inode->i_mapping, 1);
2956         }
2957         return;
2958 }
2959
2960 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2961 {
2962         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2963         int ret;
2964         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2965         if (ret == 0)
2966                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2967         return ret;
2968 }
2969
2970 static int btree_lock_page_hook(struct page *page, void *data,
2971                                 void (*flush_fn)(void *))
2972 {
2973         struct inode *inode = page->mapping->host;
2974         struct btrfs_root *root = BTRFS_I(inode)->root;
2975         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2976         struct extent_buffer *eb;
2977         unsigned long len;
2978         u64 bytenr = page_offset(page);
2979
2980         if (page->private == EXTENT_PAGE_PRIVATE)
2981                 goto out;
2982
2983         len = page->private >> 2;
2984         eb = find_extent_buffer(io_tree, bytenr, len);
2985         if (!eb)
2986                 goto out;
2987
2988         if (!btrfs_try_tree_write_lock(eb)) {
2989                 flush_fn(data);
2990                 btrfs_tree_lock(eb);
2991         }
2992         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2993
2994         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2995                 spin_lock(&root->fs_info->delalloc_lock);
2996                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2997                         root->fs_info->dirty_metadata_bytes -= eb->len;
2998                 else
2999                         WARN_ON(1);
3000                 spin_unlock(&root->fs_info->delalloc_lock);
3001         }
3002
3003         btrfs_tree_unlock(eb);
3004         free_extent_buffer(eb);
3005 out:
3006         if (!trylock_page(page)) {
3007                 flush_fn(data);
3008                 lock_page(page);
3009         }
3010         return 0;
3011 }
3012
3013 static void btrfs_check_super_valid(struct btrfs_fs_info *fs_info,
3014                               int read_only)
3015 {
3016         if (read_only)
3017                 return;
3018
3019         if (fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR)
3020                 printk(KERN_WARNING "warning: mount fs with errors, "
3021                        "running btrfsck is recommended\n");
3022 }
3023
3024 int btrfs_error_commit_super(struct btrfs_root *root)
3025 {
3026         int ret;
3027
3028         mutex_lock(&root->fs_info->cleaner_mutex);
3029         btrfs_run_delayed_iputs(root);
3030         mutex_unlock(&root->fs_info->cleaner_mutex);
3031
3032         down_write(&root->fs_info->cleanup_work_sem);
3033         up_write(&root->fs_info->cleanup_work_sem);
3034
3035         /* cleanup FS via transaction */
3036         btrfs_cleanup_transaction(root);
3037
3038         ret = write_ctree_super(NULL, root, 0);
3039
3040         return ret;
3041 }
3042
3043 static int btrfs_destroy_ordered_operations(struct btrfs_root *root)
3044 {
3045         struct btrfs_inode *btrfs_inode;
3046         struct list_head splice;
3047
3048         INIT_LIST_HEAD(&splice);
3049
3050         mutex_lock(&root->fs_info->ordered_operations_mutex);
3051         spin_lock(&root->fs_info->ordered_extent_lock);
3052
3053         list_splice_init(&root->fs_info->ordered_operations, &splice);
3054         while (!list_empty(&splice)) {
3055                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3056                                          ordered_operations);
3057
3058                 list_del_init(&btrfs_inode->ordered_operations);
3059
3060                 btrfs_invalidate_inodes(btrfs_inode->root);
3061         }
3062
3063         spin_unlock(&root->fs_info->ordered_extent_lock);
3064         mutex_unlock(&root->fs_info->ordered_operations_mutex);
3065
3066         return 0;
3067 }
3068
3069 static int btrfs_destroy_ordered_extents(struct btrfs_root *root)
3070 {
3071         struct list_head splice;
3072         struct btrfs_ordered_extent *ordered;
3073         struct inode *inode;
3074
3075         INIT_LIST_HEAD(&splice);
3076
3077         spin_lock(&root->fs_info->ordered_extent_lock);
3078
3079         list_splice_init(&root->fs_info->ordered_extents, &splice);
3080         while (!list_empty(&splice)) {
3081                 ordered = list_entry(splice.next, struct btrfs_ordered_extent,
3082                                      root_extent_list);
3083
3084                 list_del_init(&ordered->root_extent_list);
3085                 atomic_inc(&ordered->refs);
3086
3087                 /* the inode may be getting freed (in sys_unlink path). */
3088                 inode = igrab(ordered->inode);
3089
3090                 spin_unlock(&root->fs_info->ordered_extent_lock);
3091                 if (inode)
3092                         iput(inode);
3093
3094                 atomic_set(&ordered->refs, 1);
3095                 btrfs_put_ordered_extent(ordered);
3096
3097                 spin_lock(&root->fs_info->ordered_extent_lock);
3098         }
3099
3100         spin_unlock(&root->fs_info->ordered_extent_lock);
3101
3102         return 0;
3103 }
3104
3105 static int btrfs_destroy_delayed_refs(struct btrfs_transaction *trans,
3106                                       struct btrfs_root *root)
3107 {
3108         struct rb_node *node;
3109         struct btrfs_delayed_ref_root *delayed_refs;
3110         struct btrfs_delayed_ref_node *ref;
3111         int ret = 0;
3112
3113         delayed_refs = &trans->delayed_refs;
3114
3115         spin_lock(&delayed_refs->lock);
3116         if (delayed_refs->num_entries == 0) {
3117                 spin_unlock(&delayed_refs->lock);
3118                 printk(KERN_INFO "delayed_refs has NO entry\n");
3119                 return ret;
3120         }
3121
3122         node = rb_first(&delayed_refs->root);
3123         while (node) {
3124                 ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node);
3125                 node = rb_next(node);
3126
3127                 ref->in_tree = 0;
3128                 rb_erase(&ref->rb_node, &delayed_refs->root);
3129                 delayed_refs->num_entries--;
3130
3131                 atomic_set(&ref->refs, 1);
3132                 if (btrfs_delayed_ref_is_head(ref)) {
3133                         struct btrfs_delayed_ref_head *head;
3134
3135                         head = btrfs_delayed_node_to_head(ref);
3136                         mutex_lock(&head->mutex);
3137                         kfree(head->extent_op);
3138                         delayed_refs->num_heads--;
3139                         if (list_empty(&head->cluster))
3140                                 delayed_refs->num_heads_ready--;
3141                         list_del_init(&head->cluster);
3142                         mutex_unlock(&head->mutex);
3143                 }
3144
3145                 spin_unlock(&delayed_refs->lock);
3146                 btrfs_put_delayed_ref(ref);
3147
3148                 cond_resched();
3149                 spin_lock(&delayed_refs->lock);
3150         }
3151
3152         spin_unlock(&delayed_refs->lock);
3153
3154         return ret;
3155 }
3156
3157 static int btrfs_destroy_pending_snapshots(struct btrfs_transaction *t)
3158 {
3159         struct btrfs_pending_snapshot *snapshot;
3160         struct list_head splice;
3161
3162         INIT_LIST_HEAD(&splice);
3163
3164         list_splice_init(&t->pending_snapshots, &splice);
3165
3166         while (!list_empty(&splice)) {
3167                 snapshot = list_entry(splice.next,
3168                                       struct btrfs_pending_snapshot,
3169                                       list);
3170
3171                 list_del_init(&snapshot->list);
3172
3173                 kfree(snapshot);
3174         }
3175
3176         return 0;
3177 }
3178
3179 static int btrfs_destroy_delalloc_inodes(struct btrfs_root *root)
3180 {
3181         struct btrfs_inode *btrfs_inode;
3182         struct list_head splice;
3183
3184         INIT_LIST_HEAD(&splice);
3185
3186         spin_lock(&root->fs_info->delalloc_lock);
3187         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
3188
3189         while (!list_empty(&splice)) {
3190                 btrfs_inode = list_entry(splice.next, struct btrfs_inode,
3191                                     delalloc_inodes);
3192
3193                 list_del_init(&btrfs_inode->delalloc_inodes);
3194
3195                 btrfs_invalidate_inodes(btrfs_inode->root);
3196         }
3197
3198         spin_unlock(&root->fs_info->delalloc_lock);
3199
3200         return 0;
3201 }
3202
3203 static int btrfs_destroy_marked_extents(struct btrfs_root *root,
3204                                         struct extent_io_tree *dirty_pages,
3205                                         int mark)
3206 {
3207         int ret;
3208         struct page *page;
3209         struct inode *btree_inode = root->fs_info->btree_inode;
3210         struct extent_buffer *eb;
3211         u64 start = 0;
3212         u64 end;
3213         u64 offset;
3214         unsigned long index;
3215
3216         while (1) {
3217                 ret = find_first_extent_bit(dirty_pages, start, &start, &end,
3218                                             mark);
3219                 if (ret)
3220                         break;
3221
3222                 clear_extent_bits(dirty_pages, start, end, mark, GFP_NOFS);
3223                 while (start <= end) {
3224                         index = start >> PAGE_CACHE_SHIFT;
3225                         start = (u64)(index + 1) << PAGE_CACHE_SHIFT;
3226                         page = find_get_page(btree_inode->i_mapping, index);
3227                         if (!page)
3228                                 continue;
3229                         offset = page_offset(page);
3230
3231                         spin_lock(&dirty_pages->buffer_lock);
3232                         eb = radix_tree_lookup(
3233                              &(&BTRFS_I(page->mapping->host)->io_tree)->buffer,
3234                                                offset >> PAGE_CACHE_SHIFT);
3235                         spin_unlock(&dirty_pages->buffer_lock);
3236                         if (eb) {
3237                                 ret = test_and_clear_bit(EXTENT_BUFFER_DIRTY,
3238                                                          &eb->bflags);
3239                                 atomic_set(&eb->refs, 1);
3240                         }
3241                         if (PageWriteback(page))
3242                                 end_page_writeback(page);
3243
3244                         lock_page(page);
3245                         if (PageDirty(page)) {
3246                                 clear_page_dirty_for_io(page);
3247                                 spin_lock_irq(&page->mapping->tree_lock);
3248                                 radix_tree_tag_clear(&page->mapping->page_tree,
3249                                                         page_index(page),
3250                                                         PAGECACHE_TAG_DIRTY);
3251                                 spin_unlock_irq(&page->mapping->tree_lock);
3252                         }
3253
3254                         page->mapping->a_ops->invalidatepage(page, 0);
3255                         unlock_page(page);
3256                 }
3257         }
3258
3259         return ret;
3260 }
3261
3262 static int btrfs_destroy_pinned_extent(struct btrfs_root *root,
3263                                        struct extent_io_tree *pinned_extents)
3264 {
3265         struct extent_io_tree *unpin;
3266         u64 start;
3267         u64 end;
3268         int ret;
3269
3270         unpin = pinned_extents;
3271         while (1) {
3272                 ret = find_first_extent_bit(unpin, 0, &start, &end,
3273                                             EXTENT_DIRTY);
3274                 if (ret)
3275                         break;
3276
3277                 /* opt_discard */
3278                 if (btrfs_test_opt(root, DISCARD))
3279                         ret = btrfs_error_discard_extent(root, start,
3280                                                          end + 1 - start,
3281                                                          NULL);
3282
3283                 clear_extent_dirty(unpin, start, end, GFP_NOFS);
3284                 btrfs_error_unpin_extent_range(root, start, end);
3285                 cond_resched();
3286         }
3287
3288         return 0;
3289 }
3290
3291 static int btrfs_cleanup_transaction(struct btrfs_root *root)
3292 {
3293         struct btrfs_transaction *t;
3294         LIST_HEAD(list);
3295
3296         WARN_ON(1);
3297
3298         mutex_lock(&root->fs_info->transaction_kthread_mutex);
3299
3300         spin_lock(&root->fs_info->trans_lock);
3301         list_splice_init(&root->fs_info->trans_list, &list);
3302         root->fs_info->trans_no_join = 1;
3303         spin_unlock(&root->fs_info->trans_lock);
3304
3305         while (!list_empty(&list)) {
3306                 t = list_entry(list.next, struct btrfs_transaction, list);
3307                 if (!t)
3308                         break;
3309
3310                 btrfs_destroy_ordered_operations(root);
3311
3312                 btrfs_destroy_ordered_extents(root);
3313
3314                 btrfs_destroy_delayed_refs(t, root);
3315
3316                 btrfs_block_rsv_release(root,
3317                                         &root->fs_info->trans_block_rsv,
3318                                         t->dirty_pages.dirty_bytes);
3319
3320                 /* FIXME: cleanup wait for commit */
3321                 t->in_commit = 1;
3322                 t->blocked = 1;
3323                 if (waitqueue_active(&root->fs_info->transaction_blocked_wait))
3324                         wake_up(&root->fs_info->transaction_blocked_wait);
3325
3326                 t->blocked = 0;
3327                 if (waitqueue_active(&root->fs_info->transaction_wait))
3328                         wake_up(&root->fs_info->transaction_wait);
3329
3330                 t->commit_done = 1;
3331                 if (waitqueue_active(&t->commit_wait))
3332                         wake_up(&t->commit_wait);
3333
3334                 btrfs_destroy_pending_snapshots(t);
3335
3336                 btrfs_destroy_delalloc_inodes(root);
3337
3338                 spin_lock(&root->fs_info->trans_lock);
3339                 root->fs_info->running_transaction = NULL;
3340                 spin_unlock(&root->fs_info->trans_lock);
3341
3342                 btrfs_destroy_marked_extents(root, &t->dirty_pages,
3343                                              EXTENT_DIRTY);
3344
3345                 btrfs_destroy_pinned_extent(root,
3346                                             root->fs_info->pinned_extents);
3347
3348                 atomic_set(&t->use_count, 0);
3349                 list_del_init(&t->list);
3350                 memset(t, 0, sizeof(*t));
3351                 kmem_cache_free(btrfs_transaction_cachep, t);
3352         }
3353
3354         spin_lock(&root->fs_info->trans_lock);
3355         root->fs_info->trans_no_join = 0;
3356         spin_unlock(&root->fs_info->trans_lock);
3357         mutex_unlock(&root->fs_info->transaction_kthread_mutex);
3358
3359         return 0;
3360 }
3361
3362 static struct extent_io_ops btree_extent_io_ops = {
3363         .write_cache_pages_lock_hook = btree_lock_page_hook,
3364         .readpage_end_io_hook = btree_readpage_end_io_hook,
3365         .submit_bio_hook = btree_submit_bio_hook,
3366         /* note we're sharing with inode.c for the merge bio hook */
3367         .merge_bio_hook = btrfs_merge_bio_hook,
3368 };