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/crc32c.h>
  22 #include <linux/scatterlist.h>
  23 #include <linux/swap.h>
  24 #include <linux/radix-tree.h>
  25 #include <linux/writeback.h>
  26 #include <linux/buffer_head.h> // for block_sync_page
  27 #include "ctree.h"
  28 #include "disk-io.h"
  29 #include "transaction.h"
  30 #include "btrfs_inode.h"
  31 #include "print-tree.h"
  32
  33 #if 0
  34 static int check_tree_block(struct btrfs_root *root, struct extent_buffer *buf)
  35 {
  36         if (extent_buffer_blocknr(buf) != btrfs_header_blocknr(buf)) {
  37                 printk(KERN_CRIT "buf blocknr(buf) is %llu, header is %llu\n",
  38                        (unsigned long long)extent_buffer_blocknr(buf),
  39                        (unsigned long long)btrfs_header_blocknr(buf));
  40                 return 1;
  41         }
  42         return 0;
  43 }
  44 #endif
  45
  46 static struct extent_map_ops btree_extent_map_ops;
  47
  48 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
  49                                             u64 bytenr, u32 blocksize)
  50 {
  51         struct inode *btree_inode = root->fs_info->btree_inode;
  52         struct extent_buffer *eb;
  53         eb = find_extent_buffer(&BTRFS_I(btree_inode)->extent_tree,
  54                                 bytenr, blocksize, GFP_NOFS);
  55         return eb;
  56 }
  57
  58 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
  59                                                  u64 bytenr, u32 blocksize)
  60 {
  61         struct inode *btree_inode = root->fs_info->btree_inode;
  62         struct extent_buffer *eb;
  63
  64         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->extent_tree,
  65                                  bytenr, blocksize, NULL, GFP_NOFS);
  66         return eb;
  67 }
  68
  69 struct extent_map *btree_get_extent(struct inode *inode, struct page *page,
  70                                     size_t page_offset, u64 start, u64 end,
  71                                     int create)
  72 {
  73         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
  74         struct extent_map *em;
  75         int ret;
  76
  77 again:
  78         em = lookup_extent_mapping(em_tree, start, end);
  79         if (em) {
  80                 goto out;
  81         }
  82         em = alloc_extent_map(GFP_NOFS);
  83         if (!em) {
  84                 em = ERR_PTR(-ENOMEM);
  85                 goto out;
  86         }
  87         em->start = 0;
  88         em->end = (i_size_read(inode) & ~((u64)PAGE_CACHE_SIZE -1)) - 1;
  89         em->block_start = 0;
  90         em->block_end = em->end;
  91         em->bdev = inode->i_sb->s_bdev;
  92         ret = add_extent_mapping(em_tree, em);
  93         if (ret == -EEXIST) {
  94                 free_extent_map(em);
  95                 em = NULL;
  96                 goto again;
  97         } else if (ret) {
  98                 em = ERR_PTR(ret);
  99         }
 100 out:
 101         return em;
 102 }
 103
 104 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 105 {
 106         return crc32c(seed, data, len);
 107 }
 108
 109 void btrfs_csum_final(u32 crc, char *result)
 110 {
 111         *(__le32 *)result = ~cpu_to_le32(crc);
 112 }
 113
 114 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 115                            int verify)
 116 {
 117         char result[BTRFS_CRC32_SIZE];
 118         unsigned long len;
 119         unsigned long cur_len;
 120         unsigned long offset = BTRFS_CSUM_SIZE;
 121         char *map_token = NULL;
 122         char *kaddr;
 123         unsigned long map_start;
 124         unsigned long map_len;
 125         int err;
 126         u32 crc = ~(u32)0;
 127
 128         len = buf->len - offset;
 129         while(len > 0) {
 130                 err = map_private_extent_buffer(buf, offset, 32,
 131                                         &map_token, &kaddr,
 132                                         &map_start, &map_len, KM_USER0);
 133                 if (err) {
 134                         printk("failed to map extent buffer! %lu\n",
 135                                offset);
 136                         return 1;
 137                 }
 138                 cur_len = min(len, map_len - (offset - map_start));
 139                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 140                                       crc, cur_len);
 141                 len -= cur_len;
 142                 offset += cur_len;
 143                 unmap_extent_buffer(buf, map_token, KM_USER0);
 144         }
 145         btrfs_csum_final(crc, result);
 146
 147         if (verify) {
 148                 if (memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 149                         printk("btrfs: %s checksum verify failed on %llu\n",
 150                                root->fs_info->sb->s_id,
 151                                buf->start);
 152                         return 1;
 153                 }
 154         } else {
 155                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 156         }
 157         return 0;
 158 }
 159
 160
 161 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 162 {
 163         struct extent_map_tree *tree;
 164         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 165         u64 found_start;
 166         int found_level;
 167         unsigned long len;
 168         struct extent_buffer *eb;
 169         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 170
 171         if (page->private == EXTENT_PAGE_PRIVATE)
 172                 goto out;
 173         if (!page->private)
 174                 goto out;
 175         len = page->private >> 2;
 176         if (len == 0) {
 177                 WARN_ON(1);
 178         }
 179         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 180         read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1);
 181         found_start = btrfs_header_bytenr(eb);
 182         if (found_start != start) {
 183                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 184                        start, found_start, len);
 185         }
 186         found_level = btrfs_header_level(eb);
 187         csum_tree_block(root, eb, 0);
 188         free_extent_buffer(eb);
 189 out:
 190         return 0;
 191 }
 192
 193 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 194 {
 195         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 196
 197         csum_dirty_buffer(root, page);
 198         return 0;
 199 }
 200
 201 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 202 {
 203         struct extent_map_tree *tree;
 204         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 205         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 206 }
 207
 208 static int btree_writepages(struct address_space *mapping,
 209                             struct writeback_control *wbc)
 210 {
 211         struct extent_map_tree *tree;
 212         tree = &BTRFS_I(mapping->host)->extent_tree;
 213         if (wbc->sync_mode == WB_SYNC_NONE && current_is_pdflush()) {
 214                 u64 num_dirty;
 215                 u64 start = 0;
 216                 unsigned long thresh = 96 * 1024 * 1024;
 217                 num_dirty = count_range_bits(tree, &start, thresh, EXTENT_DIRTY);
 218                 if (num_dirty < thresh) {
 219                         return 0;
 220                 }
 221         }
 222         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 223 }
 224
 225 int btree_readpage(struct file *file, struct page *page)
 226 {
 227         struct extent_map_tree *tree;
 228         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 229         return extent_read_full_page(tree, page, btree_get_extent);
 230 }
 231
 232 static int btree_releasepage(struct page *page, gfp_t unused_gfp_flags)
 233 {
 234         struct extent_map_tree *tree;
 235         int ret;
 236
 237         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 238         ret = try_release_extent_mapping(tree, page);
 239         if (ret == 1) {
 240                 ClearPagePrivate(page);
 241                 set_page_private(page, 0);
 242                 page_cache_release(page);
 243         }
 244         return ret;
 245 }
 246
 247 static void btree_invalidatepage(struct page *page, unsigned long offset)
 248 {
 249         struct extent_map_tree *tree;
 250         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 251         extent_invalidatepage(tree, page, offset);
 252         btree_releasepage(page, GFP_NOFS);
 253 }
 254
 255 #if 0
 256 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 257 {
 258         struct buffer_head *bh;
 259         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 260         struct buffer_head *head;
 261         if (!page_has_buffers(page)) {
 262                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 263                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 264         }
 265         head = page_buffers(page);
 266         bh = head;
 267         do {
 268                 if (buffer_dirty(bh))
 269                         csum_tree_block(root, bh, 0);
 270                 bh = bh->b_this_page;
 271         } while (bh != head);
 272         return block_write_full_page(page, btree_get_block, wbc);
 273 }
 274 #endif
 275
 276 static struct address_space_operations btree_aops = {
 277         .readpage       = btree_readpage,
 278         .writepage      = btree_writepage,
 279         .writepages     = btree_writepages,
 280         .releasepage    = btree_releasepage,
 281         .invalidatepage = btree_invalidatepage,
 282         .sync_page      = block_sync_page,
 283 };
 284
 285 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 286 {
 287         struct extent_buffer *buf = NULL;
 288         struct inode *btree_inode = root->fs_info->btree_inode;
 289         int ret = 0;
 290
 291         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 292         if (!buf)
 293                 return 0;
 294         read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
 295                                  buf, 0, 0);
 296         free_extent_buffer(buf);
 297         return ret;
 298 }
 299
 300 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 301                                       u32 blocksize)
 302 {
 303         struct extent_buffer *buf = NULL;
 304         struct inode *btree_inode = root->fs_info->btree_inode;
 305         struct extent_map_tree *extent_tree;
 306         int ret;
 307
 308         extent_tree = &BTRFS_I(btree_inode)->extent_tree;
 309
 310         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 311         if (!buf)
 312                 return NULL;
 313         read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
 314                                  buf, 0, 1);
 315         if (buf->flags & EXTENT_CSUM) {
 316                 return buf;
 317         }
 318         if (test_range_bit(extent_tree, buf->start, buf->start + buf->len - 1,
 319                            EXTENT_CSUM, 1)) {
 320                 buf->flags |= EXTENT_CSUM;
 321                 return buf;
 322         }
 323         ret = csum_tree_block(root, buf, 1);
 324         set_extent_bits(extent_tree, buf->start,
 325                         buf->start + buf->len - 1,
 326                         EXTENT_CSUM, GFP_NOFS);
 327         buf->flags |= EXTENT_CSUM;
 328         return buf;
 329 }
 330
 331 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 332                      struct extent_buffer *buf)
 333 {
 334         struct inode *btree_inode = root->fs_info->btree_inode;
 335         clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
 336         return 0;
 337 }
 338
 339 int wait_on_tree_block_writeback(struct btrfs_root *root,
 340                                  struct extent_buffer *buf)
 341 {
 342         struct inode *btree_inode = root->fs_info->btree_inode;
 343         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->extent_tree,
 344                                         buf);
 345         return 0;
 346 }
 347
 348 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 349                         struct btrfs_root *root,
 350                         struct btrfs_fs_info *fs_info,
 351                         u64 objectid)
 352 {
 353         root->node = NULL;
 354         root->inode = NULL;
 355         root->commit_root = NULL;
 356         root->sectorsize = sectorsize;
 357         root->nodesize = nodesize;
 358         root->leafsize = leafsize;
 359         root->ref_cows = 0;
 360         root->fs_info = fs_info;
 361         root->objectid = objectid;
 362         root->last_trans = 0;
 363         root->highest_inode = 0;
 364         root->last_inode_alloc = 0;
 365         root->name = NULL;
 366         memset(&root->root_key, 0, sizeof(root->root_key));
 367         memset(&root->root_item, 0, sizeof(root->root_item));
 368         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 369         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 370         init_completion(&root->kobj_unregister);
 371         init_rwsem(&root->snap_sem);
 372         root->defrag_running = 0;
 373         root->defrag_level = 0;
 374         root->root_key.objectid = objectid;
 375         return 0;
 376 }
 377
 378 static int find_and_setup_root(struct btrfs_root *tree_root,
 379                                struct btrfs_fs_info *fs_info,
 380                                u64 objectid,
 381                                struct btrfs_root *root)
 382 {
 383         int ret;
 384         u32 blocksize;
 385
 386         __setup_root(tree_root->nodesize, tree_root->leafsize,
 387                      tree_root->sectorsize, root, fs_info, objectid);
 388         ret = btrfs_find_last_root(tree_root, objectid,
 389                                    &root->root_item, &root->root_key);
 390         BUG_ON(ret);
 391
 392         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 393         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 394                                      blocksize);
 395         BUG_ON(!root->node);
 396         return 0;
 397 }
 398
 399 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 400                                                struct btrfs_key *location)
 401 {
 402         struct btrfs_root *root;
 403         struct btrfs_root *tree_root = fs_info->tree_root;
 404         struct btrfs_path *path;
 405         struct extent_buffer *l;
 406         u64 highest_inode;
 407         u32 blocksize;
 408         int ret = 0;
 409
 410         root = kzalloc(sizeof(*root), GFP_NOFS);
 411         if (!root)
 412                 return ERR_PTR(-ENOMEM);
 413         if (location->offset == (u64)-1) {
 414                 ret = find_and_setup_root(tree_root, fs_info,
 415                                           location->objectid, root);
 416                 if (ret) {
 417                         kfree(root);
 418                         return ERR_PTR(ret);
 419                 }
 420                 goto insert;
 421         }
 422
 423         __setup_root(tree_root->nodesize, tree_root->leafsize,
 424                      tree_root->sectorsize, root, fs_info,
 425                      location->objectid);
 426
 427         path = btrfs_alloc_path();
 428         BUG_ON(!path);
 429         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 430         if (ret != 0) {
 431                 if (ret > 0)
 432                         ret = -ENOENT;
 433                 goto out;
 434         }
 435         l = path->nodes[0];
 436         read_extent_buffer(l, &root->root_item,
 437                btrfs_item_ptr_offset(l, path->slots[0]),
 438                sizeof(root->root_item));
 439         memcpy(&root->root_key, location, sizeof(*location));
 440         ret = 0;
 441 out:
 442         btrfs_release_path(root, path);
 443         btrfs_free_path(path);
 444         if (ret) {
 445                 kfree(root);
 446                 return ERR_PTR(ret);
 447         }
 448         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 449         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 450                                      blocksize);
 451         BUG_ON(!root->node);
 452 insert:
 453         root->ref_cows = 1;
 454         ret = btrfs_find_highest_inode(root, &highest_inode);
 455         if (ret == 0) {
 456                 root->highest_inode = highest_inode;
 457                 root->last_inode_alloc = highest_inode;
 458         }
 459         return root;
 460 }
 461
 462 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 463                                       struct btrfs_key *location,
 464                                       const char *name, int namelen)
 465 {
 466         struct btrfs_root *root;
 467         int ret;
 468
 469         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 470                                  (unsigned long)location->objectid);
 471         if (root)
 472                 return root;
 473
 474         root = btrfs_read_fs_root_no_radix(fs_info, location);
 475         if (IS_ERR(root))
 476                 return root;
 477         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 478                                 (unsigned long)root->root_key.objectid,
 479                                 root);
 480         if (ret) {
 481                 free_extent_buffer(root->node);
 482                 kfree(root);
 483                 return ERR_PTR(ret);
 484         }
 485
 486         ret = btrfs_set_root_name(root, name, namelen);
 487         if (ret) {
 488                 free_extent_buffer(root->node);
 489                 kfree(root);
 490                 return ERR_PTR(ret);
 491         }
 492
 493         ret = btrfs_sysfs_add_root(root);
 494         if (ret) {
 495                 free_extent_buffer(root->node);
 496                 kfree(root->name);
 497                 kfree(root);
 498                 return ERR_PTR(ret);
 499         }
 500
 501         ret = btrfs_find_dead_roots(fs_info->tree_root,
 502                                     root->root_key.objectid, root);
 503         BUG_ON(ret);
 504
 505         return root;
 506 }
 507 #if 0
 508 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 509         struct btrfs_hasher *hasher;
 510
 511         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 512         if (!hasher)
 513                 return -ENOMEM;
 514         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 515         if (!hasher->hash_tfm) {
 516                 kfree(hasher);
 517                 return -EINVAL;
 518         }
 519         spin_lock(&info->hash_lock);
 520         list_add(&hasher->list, &info->hashers);
 521         spin_unlock(&info->hash_lock);
 522         return 0;
 523 }
 524 #endif
 525 struct btrfs_root *open_ctree(struct super_block *sb)
 526 {
 527         u32 sectorsize;
 528         u32 nodesize;
 529         u32 leafsize;
 530         u32 blocksize;
 531         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 532                                                  GFP_NOFS);
 533         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 534                                                GFP_NOFS);
 535         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 536                                                 GFP_NOFS);
 537         int ret;
 538         int err = -EIO;
 539         struct btrfs_super_block *disk_super;
 540
 541         if (!extent_root || !tree_root || !fs_info) {
 542                 err = -ENOMEM;
 543                 goto fail;
 544         }
 545         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 546         INIT_LIST_HEAD(&fs_info->trans_list);
 547         INIT_LIST_HEAD(&fs_info->dead_roots);
 548         INIT_LIST_HEAD(&fs_info->hashers);
 549         spin_lock_init(&fs_info->hash_lock);
 550
 551         memset(&fs_info->super_kobj, 0, sizeof(fs_info->super_kobj));
 552         init_completion(&fs_info->kobj_unregister);
 553         sb_set_blocksize(sb, 4096);
 554         fs_info->running_transaction = NULL;
 555         fs_info->last_trans_committed = 0;
 556         fs_info->tree_root = tree_root;
 557         fs_info->extent_root = extent_root;
 558         fs_info->sb = sb;
 559         fs_info->btree_inode = new_inode(sb);
 560         fs_info->btree_inode->i_ino = 1;
 561         fs_info->btree_inode->i_nlink = 1;
 562         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 563         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 564         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
 565                              fs_info->btree_inode->i_mapping,
 566                              GFP_NOFS);
 567         BTRFS_I(fs_info->btree_inode)->extent_tree.ops = &btree_extent_map_ops;
 568
 569         extent_map_tree_init(&fs_info->free_space_cache,
 570                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 571         extent_map_tree_init(&fs_info->block_group_cache,
 572                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 573         extent_map_tree_init(&fs_info->pinned_extents,
 574                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 575         extent_map_tree_init(&fs_info->pending_del,
 576                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 577         extent_map_tree_init(&fs_info->extent_ins,
 578                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 579         fs_info->do_barriers = 1;
 580         fs_info->closing = 0;
 581         fs_info->total_pinned = 0;
 582
 583         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
 584         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 585         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 586                sizeof(struct btrfs_key));
 587         insert_inode_hash(fs_info->btree_inode);
 588         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 589
 590         mutex_init(&fs_info->trans_mutex);
 591         mutex_init(&fs_info->fs_mutex);
 592
 593 #if 0
 594         ret = add_hasher(fs_info, "crc32c");
 595         if (ret) {
 596                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
 597                 err = -ENOMEM;
 598                 goto fail_iput;
 599         }
 600 #endif
 601         __setup_root(512, 512, 512, tree_root,
 602                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 603
 604         fs_info->sb_buffer = read_tree_block(tree_root,
 605                                              BTRFS_SUPER_INFO_OFFSET,
 606                                              512);
 607
 608         if (!fs_info->sb_buffer)
 609                 goto fail_iput;
 610
 611         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
 612                            sizeof(fs_info->super_copy));
 613
 614         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
 615                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
 616                            BTRFS_FSID_SIZE);
 617         disk_super = &fs_info->super_copy;
 618         if (!btrfs_super_root(disk_super))
 619                 goto fail_sb_buffer;
 620
 621         nodesize = btrfs_super_nodesize(disk_super);
 622         leafsize = btrfs_super_leafsize(disk_super);
 623         sectorsize = btrfs_super_sectorsize(disk_super);
 624         tree_root->nodesize = nodesize;
 625         tree_root->leafsize = leafsize;
 626         tree_root->sectorsize = sectorsize;
 627         sb_set_blocksize(sb, sectorsize);
 628
 629         i_size_write(fs_info->btree_inode,
 630                      btrfs_super_total_bytes(disk_super));
 631
 632         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 633                     sizeof(disk_super->magic))) {
 634                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
 635                 goto fail_sb_buffer;
 636         }
 637
 638         blocksize = btrfs_level_size(tree_root,
 639                                      btrfs_super_root_level(disk_super));
 640
 641         tree_root->node = read_tree_block(tree_root,
 642                                           btrfs_super_root(disk_super),
 643                                           blocksize);
 644         if (!tree_root->node)
 645                 goto fail_sb_buffer;
 646
 647         mutex_lock(&fs_info->fs_mutex);
 648
 649         ret = find_and_setup_root(tree_root, fs_info,
 650                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 651         if (ret) {
 652                 mutex_unlock(&fs_info->fs_mutex);
 653                 goto fail_tree_root;
 654         }
 655
 656         btrfs_read_block_groups(extent_root);
 657
 658         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 659         mutex_unlock(&fs_info->fs_mutex);
 660         return tree_root;
 661
 662 fail_tree_root:
 663         free_extent_buffer(tree_root->node);
 664 fail_sb_buffer:
 665         free_extent_buffer(fs_info->sb_buffer);
 666 fail_iput:
 667         iput(fs_info->btree_inode);
 668 fail:
 669         kfree(extent_root);
 670         kfree(tree_root);
 671         kfree(fs_info);
 672         return ERR_PTR(err);
 673 }
 674
 675 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 676                       *root)
 677 {
 678         int ret;
 679         struct extent_buffer *super = root->fs_info->sb_buffer;
 680         struct inode *btree_inode = root->fs_info->btree_inode;
 681
 682         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, super);
 683         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
 684                                      super->start, super->len);
 685         return ret;
 686 }
 687
 688 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 689 {
 690         radix_tree_delete(&fs_info->fs_roots_radix,
 691                           (unsigned long)root->root_key.objectid);
 692         btrfs_sysfs_del_root(root);
 693         if (root->inode)
 694                 iput(root->inode);
 695         if (root->node)
 696                 free_extent_buffer(root->node);
 697         if (root->commit_root)
 698                 free_extent_buffer(root->commit_root);
 699         if (root->name)
 700                 kfree(root->name);
 701         kfree(root);
 702         return 0;
 703 }
 704
 705 static int del_fs_roots(struct btrfs_fs_info *fs_info)
 706 {
 707         int ret;
 708         struct btrfs_root *gang[8];
 709         int i;
 710
 711         while(1) {
 712                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 713                                              (void **)gang, 0,
 714                                              ARRAY_SIZE(gang));
 715                 if (!ret)
 716                         break;
 717                 for (i = 0; i < ret; i++)
 718                         btrfs_free_fs_root(fs_info, gang[i]);
 719         }
 720         return 0;
 721 }
 722
 723 int close_ctree(struct btrfs_root *root)
 724 {
 725         int ret;
 726         struct btrfs_trans_handle *trans;
 727         struct btrfs_fs_info *fs_info = root->fs_info;
 728
 729         fs_info->closing = 1;
 730         btrfs_transaction_flush_work(root);
 731         mutex_lock(&fs_info->fs_mutex);
 732         btrfs_defrag_dirty_roots(root->fs_info);
 733         trans = btrfs_start_transaction(root, 1);
 734         ret = btrfs_commit_transaction(trans, root);
 735         /* run commit again to  drop the original snapshot */
 736         trans = btrfs_start_transaction(root, 1);
 737         btrfs_commit_transaction(trans, root);
 738         ret = btrfs_write_and_wait_transaction(NULL, root);
 739         BUG_ON(ret);
 740         write_ctree_super(NULL, root);
 741         mutex_unlock(&fs_info->fs_mutex);
 742
 743         if (fs_info->extent_root->node)
 744                 free_extent_buffer(fs_info->extent_root->node);
 745
 746         if (fs_info->tree_root->node)
 747                 free_extent_buffer(fs_info->tree_root->node);
 748
 749         free_extent_buffer(fs_info->sb_buffer);
 750
 751         btrfs_free_block_groups(root->fs_info);
 752         del_fs_roots(fs_info);
 753         extent_map_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->extent_tree);
 754         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
 755         iput(fs_info->btree_inode);
 756 #if 0
 757         while(!list_empty(&fs_info->hashers)) {
 758                 struct btrfs_hasher *hasher;
 759                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
 760                                     hashers);
 761                 list_del(&hasher->hashers);
 762                 crypto_free_hash(&fs_info->hash_tfm);
 763                 kfree(hasher);
 764         }
 765 #endif
 766         kfree(fs_info->extent_root);
 767         kfree(fs_info->tree_root);
 768         return 0;
 769 }
 770
 771 int btrfs_buffer_uptodate(struct extent_buffer *buf)
 772 {
 773         struct inode *btree_inode = buf->first_page->mapping->host;
 774         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree, buf);
 775 }
 776
 777 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
 778 {
 779         struct inode *btree_inode = buf->first_page->mapping->host;
 780         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree,
 781                                           buf);
 782 }
 783
 784 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 785 {
 786         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 787         u64 transid = btrfs_header_generation(buf);
 788         struct inode *btree_inode = root->fs_info->btree_inode;
 789
 790         if (transid != root->fs_info->generation) {
 791                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
 792                         (unsigned long long)buf->start,
 793                         transid, root->fs_info->generation);
 794                 WARN_ON(1);
 795         }
 796         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
 797 }
 798
 799 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
 800 {
 801         balance_dirty_pages_ratelimited_nr(
 802                         root->fs_info->btree_inode->i_mapping, 1);
 803 }
 804
 805 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
 806 {
 807         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 808         struct inode *btree_inode = root->fs_info->btree_inode;
 809         set_extent_bits(&BTRFS_I(btree_inode)->extent_tree, buf->start,
 810                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
 811 }
 812
 813 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
 814 {
 815         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 816         struct inode *btree_inode = root->fs_info->btree_inode;
 817         set_extent_bits(&BTRFS_I(btree_inode)->extent_tree, buf->start,
 818                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
 819                         GFP_NOFS);
 820 }
 821
 822 int btrfs_buffer_defrag(struct extent_buffer *buf)
 823 {
 824         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 825         struct inode *btree_inode = root->fs_info->btree_inode;
 826         return test_range_bit(&BTRFS_I(btree_inode)->extent_tree,
 827                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
 828 }
 829
 830 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
 831 {
 832         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 833         struct inode *btree_inode = root->fs_info->btree_inode;
 834         return test_range_bit(&BTRFS_I(btree_inode)->extent_tree,
 835                      buf->start, buf->start + buf->len - 1,
 836                      EXTENT_DEFRAG_DONE, 0);
 837 }
 838
 839 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
 840 {
 841         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 842         struct inode *btree_inode = root->fs_info->btree_inode;
 843         return clear_extent_bits(&BTRFS_I(btree_inode)->extent_tree,
 844                      buf->start, buf->start + buf->len - 1,
 845                      EXTENT_DEFRAG_DONE, GFP_NOFS);
 846 }
 847
 848 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
 849 {
 850         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 851         struct inode *btree_inode = root->fs_info->btree_inode;
 852         return clear_extent_bits(&BTRFS_I(btree_inode)->extent_tree,
 853                      buf->start, buf->start + buf->len - 1,
 854                      EXTENT_DEFRAG, GFP_NOFS);
 855 }
 856
 857 int btrfs_read_buffer(struct extent_buffer *buf)
 858 {
 859         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 860         struct inode *btree_inode = root->fs_info->btree_inode;
 861         return read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
 862                                         buf, 0, 1);
 863 }
 864
 865 static struct extent_map_ops btree_extent_map_ops = {
 866         .writepage_io_hook = btree_writepage_io_hook,
 867 };