fs/btrfs/disk-io.c

   1 #include <linux/module.h>
   2 #include <linux/fs.h>
   3 #include <linux/blkdev.h>
   4 #include <linux/crypto.h>
   5 #include <linux/scatterlist.h>
   6 #include <linux/swap.h>
   7 #include <linux/radix-tree.h>
   8 #include "ctree.h"
   9 #include "disk-io.h"
  10 #include "transaction.h"
  11 #include "btrfs_inode.h"
  12
  13 struct dev_lookup {
  14         u64 block_start;
  15         u64 num_blocks;
  16         u64 device_id;
  17         struct block_device *bdev;
  18 };
  19
  20 int btrfs_insert_dev_radix(struct btrfs_root *root,
  21                            struct block_device *bdev,
  22                            u64 device_id,
  23                            u64 block_start,
  24                            u64 num_blocks)
  25 {
  26         struct dev_lookup *lookup;
  27         int ret;
  28
  29         lookup = kmalloc(sizeof(*lookup), GFP_NOFS);
  30         if (!lookup)
  31                 return -ENOMEM;
  32         lookup->block_start = block_start;
  33         lookup->num_blocks = num_blocks;
  34         lookup->bdev = bdev;
  35         lookup->device_id = device_id;
  36
  37         ret = radix_tree_insert(&root->fs_info->dev_radix, block_start +
  38                                 num_blocks - 1, lookup);
  39         return ret;
  40 }
  41
  42 u64 bh_blocknr(struct buffer_head *bh)
  43 {
  44         int blkbits = bh->b_page->mapping->host->i_blkbits;
  45         u64 blocknr = bh->b_page->index << (PAGE_CACHE_SHIFT - blkbits);
  46         unsigned long offset;
  47
  48         if (PageHighMem(bh->b_page))
  49                 offset = (unsigned long)bh->b_data;
  50         else
  51                 offset = bh->b_data - (char *)page_address(bh->b_page);
  52         blocknr += offset >> (PAGE_CACHE_SHIFT - blkbits);
  53         return blocknr;
  54 }
  55
  56 static int check_tree_block(struct btrfs_root *root, struct buffer_head *buf)
  57 {
  58         struct btrfs_node *node = btrfs_buffer_node(buf);
  59         if (bh_blocknr(buf) != btrfs_header_blocknr(&node->header)) {
  60                 printk(KERN_CRIT "bh_blocknr(buf) is %Lu, header is %Lu\n",
  61                        bh_blocknr(buf), btrfs_header_blocknr(&node->header));
  62                 BUG();
  63         }
  64         return 0;
  65 }
  66
  67 struct buffer_head *btrfs_find_tree_block(struct btrfs_root *root, u64 blocknr)
  68 {
  69         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
  70         int blockbits = root->fs_info->sb->s_blocksize_bits;
  71         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
  72         struct page *page;
  73         struct buffer_head *bh;
  74         struct buffer_head *head;
  75         struct buffer_head *ret = NULL;
  76
  77
  78         page = find_lock_page(mapping, index);
  79         if (!page)
  80                 return NULL;
  81
  82         if (!page_has_buffers(page))
  83                 goto out_unlock;
  84
  85         head = page_buffers(page);
  86         bh = head;
  87         do {
  88                 if (buffer_mapped(bh) && bh_blocknr(bh) == blocknr) {
  89                         ret = bh;
  90                         get_bh(bh);
  91                         goto out_unlock;
  92                 }
  93                 bh = bh->b_this_page;
  94         } while (bh != head);
  95 out_unlock:
  96         unlock_page(page);
  97         if (ret) {
  98                 touch_buffer(ret);
  99         }
 100         page_cache_release(page);
 101         return ret;
 102 }
 103
 104 int btrfs_map_bh_to_logical(struct btrfs_root *root, struct buffer_head *bh,
 105                              u64 logical)
 106 {
 107         struct dev_lookup *lookup[2];
 108
 109         int ret;
 110
 111         if (logical == 0) {
 112                 bh->b_bdev = NULL;
 113                 bh->b_blocknr = 0;
 114                 set_buffer_mapped(bh);
 115                 return 0;
 116         }
 117         root = root->fs_info->dev_root;
 118         ret = radix_tree_gang_lookup(&root->fs_info->dev_radix,
 119                                      (void **)lookup,
 120                                      (unsigned long)logical,
 121                                      ARRAY_SIZE(lookup));
 122         if (ret == 0 || lookup[0]->block_start > logical ||
 123             lookup[0]->block_start + lookup[0]->num_blocks <= logical) {
 124                 ret = -ENOENT;
 125                 goto out;
 126         }
 127         bh->b_bdev = lookup[0]->bdev;
 128         bh->b_blocknr = logical - lookup[0]->block_start;
 129         set_buffer_mapped(bh);
 130         ret = 0;
 131 out:
 132         return ret;
 133 }
 134
 135 struct buffer_head *btrfs_find_create_tree_block(struct btrfs_root *root,
 136                                                  u64 blocknr)
 137 {
 138         struct address_space *mapping = root->fs_info->btree_inode->i_mapping;
 139         int blockbits = root->fs_info->sb->s_blocksize_bits;
 140         unsigned long index = blocknr >> (PAGE_CACHE_SHIFT - blockbits);
 141         struct page *page;
 142         struct buffer_head *bh;
 143         struct buffer_head *head;
 144         struct buffer_head *ret = NULL;
 145         int err;
 146         u64 first_block = index << (PAGE_CACHE_SHIFT - blockbits);
 147
 148         page = grab_cache_page(mapping, index);
 149         if (!page)
 150                 return NULL;
 151
 152         if (!page_has_buffers(page))
 153                 create_empty_buffers(page, root->fs_info->sb->s_blocksize, 0);
 154         head = page_buffers(page);
 155         bh = head;
 156         do {
 157                 if (!buffer_mapped(bh)) {
 158                         err = btrfs_map_bh_to_logical(root, bh, first_block);
 159                         BUG_ON(err);
 160                 }
 161                 if (bh_blocknr(bh) == blocknr) {
 162                         ret = bh;
 163                         get_bh(bh);
 164                         goto out_unlock;
 165                 }
 166                 bh = bh->b_this_page;
 167                 first_block++;
 168         } while (bh != head);
 169 out_unlock:
 170         unlock_page(page);
 171         if (ret)
 172                 touch_buffer(ret);
 173         page_cache_release(page);
 174         return ret;
 175 }
 176
 177 static int btree_get_block(struct inode *inode, sector_t iblock,
 178                            struct buffer_head *bh, int create)
 179 {
 180         int err;
 181         struct btrfs_root *root = BTRFS_I(bh->b_page->mapping->host)->root;
 182         err = btrfs_map_bh_to_logical(root, bh, iblock);
 183         return err;
 184 }
 185
 186 int btrfs_csum_data(struct btrfs_root * root, char *data, size_t len,
 187                     char *result)
 188 {
 189         struct scatterlist sg;
 190         struct crypto_hash *tfm = root->fs_info->hash_tfm;
 191         struct hash_desc desc;
 192         int ret;
 193
 194         desc.tfm = tfm;
 195         desc.flags = 0;
 196         sg_init_one(&sg, data, len);
 197         spin_lock(&root->fs_info->hash_lock);
 198         ret = crypto_hash_digest(&desc, &sg, 1, result);
 199         spin_unlock(&root->fs_info->hash_lock);
 200         if (ret) {
 201                 printk("sha256 digest failed\n");
 202         }
 203         return ret;
 204 }
 205 static int csum_tree_block(struct btrfs_root *root, struct buffer_head *bh,
 206                            int verify)
 207 {
 208         char result[BTRFS_CSUM_SIZE];
 209         int ret;
 210         struct btrfs_node *node;
 211
 212         ret = btrfs_csum_data(root, bh->b_data + BTRFS_CSUM_SIZE,
 213                               bh->b_size - BTRFS_CSUM_SIZE, result);
 214         if (ret)
 215                 return ret;
 216         if (verify) {
 217                 if (memcmp(bh->b_data, result, BTRFS_CSUM_SIZE)) {
 218                         printk("checksum verify failed on %Lu\n",
 219                                bh_blocknr(bh));
 220                         return 1;
 221                 }
 222         } else {
 223                 node = btrfs_buffer_node(bh);
 224                 memcpy(node->header.csum, result, BTRFS_CSUM_SIZE);
 225         }
 226         return 0;
 227 }
 228
 229 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 230 {
 231         struct buffer_head *bh;
 232         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 233         struct buffer_head *head;
 234         if (!page_has_buffers(page)) {
 235                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 236                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 237         }
 238         head = page_buffers(page);
 239         bh = head;
 240         do {
 241                 if (buffer_dirty(bh))
 242                         csum_tree_block(root, bh, 0);
 243                 bh = bh->b_this_page;
 244         } while (bh != head);
 245         return block_write_full_page(page, btree_get_block, wbc);
 246 }
 247
 248 static int btree_readpage(struct file * file, struct page * page)
 249 {
 250         return block_read_full_page(page, btree_get_block);
 251 }
 252
 253 static struct address_space_operations btree_aops = {
 254         .readpage       = btree_readpage,
 255         .writepage      = btree_writepage,
 256         .sync_page      = block_sync_page,
 257 };
 258
 259 struct buffer_head *read_tree_block(struct btrfs_root *root, u64 blocknr)
 260 {
 261         struct buffer_head *bh = NULL;
 262
 263         bh = btrfs_find_create_tree_block(root, blocknr);
 264         if (!bh)
 265                 return bh;
 266         if (buffer_uptodate(bh))
 267                 goto uptodate;
 268         lock_buffer(bh);
 269         if (!buffer_uptodate(bh)) {
 270                 get_bh(bh);
 271                 bh->b_end_io = end_buffer_read_sync;
 272                 submit_bh(READ, bh);
 273                 wait_on_buffer(bh);
 274                 if (!buffer_uptodate(bh))
 275                         goto fail;
 276                 csum_tree_block(root, bh, 1);
 277         } else {
 278                 unlock_buffer(bh);
 279         }
 280 uptodate:
 281         if (check_tree_block(root, bh))
 282                 BUG();
 283         return bh;
 284 fail:
 285         brelse(bh);
 286         return NULL;
 287 }
 288
 289 int dirty_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 290                      struct buffer_head *buf)
 291 {
 292         WARN_ON(atomic_read(&buf->b_count) == 0);
 293         mark_buffer_dirty(buf);
 294         return 0;
 295 }
 296
 297 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 298                      struct buffer_head *buf)
 299 {
 300         WARN_ON(atomic_read(&buf->b_count) == 0);
 301         clear_buffer_dirty(buf);
 302         return 0;
 303 }
 304
 305 static int __setup_root(int blocksize,
 306                         struct btrfs_root *root,
 307                         struct btrfs_fs_info *fs_info,
 308                         u64 objectid)
 309 {
 310         root->node = NULL;
 311         root->inode = NULL;
 312         root->commit_root = NULL;
 313         root->blocksize = blocksize;
 314         root->ref_cows = 0;
 315         root->fs_info = fs_info;
 316         root->objectid = objectid;
 317         root->last_trans = 0;
 318         root->highest_inode = 0;
 319         root->last_inode_alloc = 0;
 320         memset(&root->root_key, 0, sizeof(root->root_key));
 321         memset(&root->root_item, 0, sizeof(root->root_item));
 322         root->root_key.objectid = objectid;
 323         return 0;
 324 }
 325
 326 static int find_and_setup_root(int blocksize,
 327                                struct btrfs_root *tree_root,
 328                                struct btrfs_fs_info *fs_info,
 329                                u64 objectid,
 330                                struct btrfs_root *root)
 331 {
 332         int ret;
 333
 334         __setup_root(blocksize, root, fs_info, objectid);
 335         ret = btrfs_find_last_root(tree_root, objectid,
 336                                    &root->root_item, &root->root_key);
 337         BUG_ON(ret);
 338
 339         root->node = read_tree_block(root,
 340                                      btrfs_root_blocknr(&root->root_item));
 341         BUG_ON(!root->node);
 342         return 0;
 343 }
 344
 345 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 346                                       struct btrfs_key *location)
 347 {
 348         struct btrfs_root *root;
 349         struct btrfs_root *tree_root = fs_info->tree_root;
 350         struct btrfs_path *path;
 351         struct btrfs_leaf *l;
 352         u64 highest_inode;
 353         int ret = 0;
 354
 355 printk("read_fs_root looking for %Lu %Lu %u\n", location->objectid, location->offset, location->flags);
 356         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 357                                  (unsigned long)location->objectid);
 358         if (root) {
 359 printk("found %p in cache\n", root);
 360                 return root;
 361         }
 362         root = kmalloc(sizeof(*root), GFP_NOFS);
 363         if (!root) {
 364 printk("failed1\n");
 365                 return ERR_PTR(-ENOMEM);
 366         }
 367         if (location->offset == (u64)-1) {
 368                 ret = find_and_setup_root(fs_info->sb->s_blocksize,
 369                                           fs_info->tree_root, fs_info,
 370                                           location->objectid, root);
 371                 if (ret) {
 372 printk("failed2\n");
 373                         kfree(root);
 374                         return ERR_PTR(ret);
 375                 }
 376                 goto insert;
 377         }
 378
 379         __setup_root(fs_info->sb->s_blocksize, root, fs_info,
 380                      location->objectid);
 381
 382         path = btrfs_alloc_path();
 383         BUG_ON(!path);
 384         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 385         if (ret != 0) {
 386 printk("internal search_slot gives us %d\n", ret);
 387                 if (ret > 0)
 388                         ret = -ENOENT;
 389                 goto out;
 390         }
 391         l = btrfs_buffer_leaf(path->nodes[0]);
 392         memcpy(&root->root_item,
 393                btrfs_item_ptr(l, path->slots[0], struct btrfs_root_item),
 394                sizeof(root->root_item));
 395         memcpy(&root->root_key, location, sizeof(*location));
 396         ret = 0;
 397 out:
 398         btrfs_release_path(root, path);
 399         btrfs_free_path(path);
 400         if (ret) {
 401                 kfree(root);
 402                 return ERR_PTR(ret);
 403         }
 404         root->node = read_tree_block(root,
 405                                      btrfs_root_blocknr(&root->root_item));
 406         BUG_ON(!root->node);
 407 insert:
 408 printk("inserting %p\n", root);
 409         root->ref_cows = 1;
 410         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 411                                 (unsigned long)root->root_key.objectid,
 412                                 root);
 413         if (ret) {
 414 printk("radix_tree_insert gives us %d\n", ret);
 415                 brelse(root->node);
 416                 kfree(root);
 417                 return ERR_PTR(ret);
 418         }
 419         ret = btrfs_find_highest_inode(root, &highest_inode);
 420         if (ret == 0) {
 421                 root->highest_inode = highest_inode;
 422                 root->last_inode_alloc = highest_inode;
 423 printk("highest inode is %Lu\n", highest_inode);
 424         }
 425 printk("all worked\n");
 426         return root;
 427 }
 428
 429 static int btrfs_open_disk(struct btrfs_root *root, u64 device_id,
 430                            u64 block_start, u64 num_blocks,
 431                            char *filename, int name_len)
 432 {
 433         char *null_filename;
 434         struct block_device *bdev;
 435         int ret;
 436
 437         null_filename = kmalloc(name_len + 1, GFP_NOFS);
 438         if (!null_filename)
 439                 return -ENOMEM;
 440         memcpy(null_filename, filename, name_len);
 441         null_filename[name_len] = '\0';
 442
 443         bdev = open_bdev_excl(null_filename, O_RDWR, root->fs_info->sb);
 444         if (IS_ERR(bdev)) {
 445                 ret = PTR_ERR(bdev);
 446                 goto out;
 447         }
 448         set_blocksize(bdev, root->fs_info->sb->s_blocksize);
 449         ret = btrfs_insert_dev_radix(root, bdev, device_id,
 450                                      block_start, num_blocks);
 451         BUG_ON(ret);
 452         ret = 0;
 453 out:
 454         kfree(null_filename);
 455         return ret;
 456 }
 457
 458 static int read_device_info(struct btrfs_root *root)
 459 {
 460         struct btrfs_path *path;
 461         int ret;
 462         struct btrfs_key key;
 463         struct btrfs_leaf *leaf;
 464         struct btrfs_device_item *dev_item;
 465         int nritems;
 466         int slot;
 467
 468         root = root->fs_info->dev_root;
 469
 470         path = btrfs_alloc_path();
 471         if (!path)
 472                 return -ENOMEM;
 473         key.objectid = 0;
 474         key.offset = 0;
 475         key.flags = 0;
 476         btrfs_set_key_type(&key, BTRFS_DEV_ITEM_KEY);
 477
 478         mutex_lock(&root->fs_info->fs_mutex);
 479         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 480         leaf = btrfs_buffer_leaf(path->nodes[0]);
 481         nritems = btrfs_header_nritems(&leaf->header);
 482         while(1) {
 483                 slot = path->slots[0];
 484                 if (slot >= nritems) {
 485                         ret = btrfs_next_leaf(root, path);
 486                         if (ret)
 487                                 break;
 488                         leaf = btrfs_buffer_leaf(path->nodes[0]);
 489                         nritems = btrfs_header_nritems(&leaf->header);
 490                         slot = path->slots[0];
 491                 }
 492                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
 493                 if (btrfs_key_type(&key) != BTRFS_DEV_ITEM_KEY) {
 494                         path->slots[0]++;
 495                         continue;
 496                 }
 497                 dev_item = btrfs_item_ptr(leaf, slot, struct btrfs_device_item);
 498 printk("found key %Lu %Lu\n", key.objectid, key.offset);
 499                 if (btrfs_device_id(dev_item) !=
 500                     btrfs_super_device_id(root->fs_info->disk_super)) {
 501                         ret = btrfs_open_disk(root, btrfs_device_id(dev_item),
 502                                               key.objectid, key.offset,
 503                                               (char *)(dev_item + 1),
 504                                               btrfs_device_pathlen(dev_item));
 505                         BUG_ON(ret);
 506                 }
 507                 path->slots[0]++;
 508         }
 509         btrfs_free_path(path);
 510         mutex_unlock(&root->fs_info->fs_mutex);
 511         return 0;
 512 }
 513
 514 struct btrfs_root *open_ctree(struct super_block *sb)
 515 {
 516         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 517                                                  GFP_NOFS);
 518         struct btrfs_root *dev_root = kmalloc(sizeof(struct btrfs_root),
 519                                                  GFP_NOFS);
 520         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 521                                                GFP_NOFS);
 522         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 523                                                 GFP_NOFS);
 524         int ret;
 525         struct btrfs_super_block *disk_super;
 526         struct dev_lookup *dev_lookup;
 527
 528         init_bit_radix(&fs_info->pinned_radix);
 529         init_bit_radix(&fs_info->pending_del_radix);
 530         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 531         INIT_RADIX_TREE(&fs_info->dev_radix, GFP_NOFS);
 532         INIT_LIST_HEAD(&fs_info->trans_list);
 533         sb_set_blocksize(sb, 4096);
 534         fs_info->running_transaction = NULL;
 535         fs_info->tree_root = tree_root;
 536         fs_info->extent_root = extent_root;
 537         fs_info->dev_root = dev_root;
 538         fs_info->sb = sb;
 539         fs_info->btree_inode = new_inode(sb);
 540         fs_info->btree_inode->i_ino = 1;
 541         fs_info->btree_inode->i_nlink = 1;
 542         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 543         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 544         fs_info->do_barriers = 1;
 545         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 546         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 547                sizeof(struct btrfs_key));
 548         insert_inode_hash(fs_info->btree_inode);
 549         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 550         fs_info->hash_tfm = crypto_alloc_hash("sha256", 0, CRYPTO_ALG_ASYNC);
 551         spin_lock_init(&fs_info->hash_lock);
 552         if (!fs_info->hash_tfm || IS_ERR(fs_info->hash_tfm)) {
 553                 printk("failed to allocate sha256 hash\n");
 554                 return NULL;
 555         }
 556         mutex_init(&fs_info->trans_mutex);
 557         mutex_init(&fs_info->fs_mutex);
 558         memset(&fs_info->current_insert, 0, sizeof(fs_info->current_insert));
 559         memset(&fs_info->last_insert, 0, sizeof(fs_info->last_insert));
 560
 561         __setup_root(sb->s_blocksize, dev_root,
 562                      fs_info, BTRFS_DEV_TREE_OBJECTID);
 563
 564         __setup_root(sb->s_blocksize, tree_root,
 565                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 566
 567         dev_lookup = kmalloc(sizeof(*dev_lookup), GFP_NOFS);
 568         dev_lookup->block_start = 0;
 569         dev_lookup->num_blocks = (u32)-2;
 570         dev_lookup->bdev = sb->s_bdev;
 571         dev_lookup->device_id = 0;
 572         ret = radix_tree_insert(&fs_info->dev_radix, (u32)-2, dev_lookup);
 573         BUG_ON(ret);
 574         fs_info->sb_buffer = read_tree_block(tree_root,
 575                                              BTRFS_SUPER_INFO_OFFSET /
 576                                              sb->s_blocksize);
 577
 578         if (!fs_info->sb_buffer)
 579                 return NULL;
 580         disk_super = (struct btrfs_super_block *)fs_info->sb_buffer->b_data;
 581         if (!btrfs_super_root(disk_super))
 582                 return NULL;
 583
 584         i_size_write(fs_info->btree_inode,
 585                      btrfs_super_total_blocks(disk_super) <<
 586                      fs_info->btree_inode->i_blkbits);
 587
 588         radix_tree_delete(&fs_info->dev_radix, (u32)-2);
 589         dev_lookup->block_start = btrfs_super_device_block_start(disk_super);
 590         dev_lookup->num_blocks = btrfs_super_device_num_blocks(disk_super);
 591         dev_lookup->device_id = btrfs_super_device_id(disk_super);
 592
 593         ret = radix_tree_insert(&fs_info->dev_radix,
 594                                 dev_lookup->block_start +
 595                                 dev_lookup->num_blocks - 1, dev_lookup);
 596         BUG_ON(ret);
 597
 598         fs_info->disk_super = disk_super;
 599
 600         dev_root->node = read_tree_block(tree_root,
 601                                           btrfs_super_device_root(disk_super));
 602
 603         ret = read_device_info(dev_root);
 604         BUG_ON(ret);
 605
 606         tree_root->node = read_tree_block(tree_root,
 607                                           btrfs_super_root(disk_super));
 608         BUG_ON(!tree_root->node);
 609
 610         mutex_lock(&fs_info->fs_mutex);
 611         ret = find_and_setup_root(sb->s_blocksize, tree_root, fs_info,
 612                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 613         BUG_ON(ret);
 614
 615         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 616         memset(&fs_info->kobj, 0, sizeof(fs_info->kobj));
 617         kobj_set_kset_s(fs_info, btrfs_subsys);
 618         kobject_set_name(&fs_info->kobj, "%s", sb->s_id);
 619         kobject_register(&fs_info->kobj);
 620         mutex_unlock(&fs_info->fs_mutex);
 621         return tree_root;
 622 }
 623
 624 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 625                       *root)
 626 {
 627         int ret;
 628         struct buffer_head *bh = root->fs_info->sb_buffer;
 629
 630         btrfs_set_super_root(root->fs_info->disk_super,
 631                              bh_blocknr(root->fs_info->tree_root->node));
 632         lock_buffer(bh);
 633         WARN_ON(atomic_read(&bh->b_count) < 1);
 634         clear_buffer_dirty(bh);
 635         csum_tree_block(root, bh, 0);
 636         bh->b_end_io = end_buffer_write_sync;
 637         get_bh(bh);
 638         if (root->fs_info->do_barriers)
 639                 ret = submit_bh(WRITE_BARRIER, bh);
 640         else
 641                 ret = submit_bh(WRITE, bh);
 642         if (ret == -EOPNOTSUPP) {
 643                 set_buffer_uptodate(bh);
 644                 root->fs_info->do_barriers = 0;
 645                 ret = submit_bh(WRITE, bh);
 646         }
 647         wait_on_buffer(bh);
 648         if (!buffer_uptodate(bh)) {
 649                 WARN_ON(1);
 650                 return -EIO;
 651         }
 652         return 0;
 653 }
 654
 655 static int free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 656 {
 657         radix_tree_delete(&fs_info->fs_roots_radix,
 658                           (unsigned long)root->root_key.objectid);
 659         if (root->inode)
 660                 iput(root->inode);
 661         if (root->node)
 662                 brelse(root->node);
 663         if (root->commit_root)
 664                 brelse(root->commit_root);
 665         kfree(root);
 666         return 0;
 667 }
 668
 669 int del_fs_roots(struct btrfs_fs_info *fs_info)
 670 {
 671         int ret;
 672         struct btrfs_root *gang[8];
 673         int i;
 674
 675         while(1) {
 676                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 677                                              (void **)gang, 0,
 678                                              ARRAY_SIZE(gang));
 679                 if (!ret)
 680                         break;
 681                 for (i = 0; i < ret; i++)
 682                         free_fs_root(fs_info, gang[i]);
 683         }
 684         return 0;
 685 }
 686
 687 static int free_dev_radix(struct btrfs_fs_info *fs_info)
 688 {
 689         struct dev_lookup *lookup[8];
 690         struct block_device *super_bdev = fs_info->sb->s_bdev;
 691         int ret;
 692         int i;
 693         while(1) {
 694                 ret = radix_tree_gang_lookup(&fs_info->dev_radix,
 695                                              (void **)lookup, 0,
 696                                              ARRAY_SIZE(lookup));
 697                 if (!ret)
 698                         break;
 699                 for (i = 0; i < ret; i++) {
 700                         if (lookup[i]->bdev != super_bdev)
 701                                 close_bdev_excl(lookup[i]->bdev);
 702                         radix_tree_delete(&fs_info->dev_radix,
 703                                           lookup[i]->block_start +
 704                                           lookup[i]->num_blocks - 1);
 705                         kfree(lookup[i]);
 706                 }
 707         }
 708         return 0;
 709 }
 710
 711 int close_ctree(struct btrfs_root *root)
 712 {
 713         int ret;
 714         struct btrfs_trans_handle *trans;
 715         struct btrfs_fs_info *fs_info = root->fs_info;
 716
 717         mutex_lock(&fs_info->fs_mutex);
 718         trans = btrfs_start_transaction(root, 1);
 719         btrfs_commit_transaction(trans, root);
 720         /* run commit again to  drop the original snapshot */
 721         trans = btrfs_start_transaction(root, 1);
 722         btrfs_commit_transaction(trans, root);
 723         ret = btrfs_write_and_wait_transaction(NULL, root);
 724         BUG_ON(ret);
 725         write_ctree_super(NULL, root);
 726         mutex_unlock(&fs_info->fs_mutex);
 727
 728         if (fs_info->extent_root->node)
 729                 btrfs_block_release(fs_info->extent_root,
 730                                     fs_info->extent_root->node);
 731         if (fs_info->dev_root->node)
 732                 btrfs_block_release(fs_info->dev_root,
 733                                     fs_info->dev_root->node);
 734         if (fs_info->tree_root->node)
 735                 btrfs_block_release(fs_info->tree_root,
 736                                     fs_info->tree_root->node);
 737         btrfs_block_release(root, fs_info->sb_buffer);
 738         crypto_free_hash(fs_info->hash_tfm);
 739         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
 740         iput(fs_info->btree_inode);
 741
 742         free_dev_radix(fs_info);
 743         del_fs_roots(fs_info);
 744         kfree(fs_info->extent_root);
 745         kfree(fs_info->tree_root);
 746         kobject_unregister(&fs_info->kobj);
 747         return 0;
 748 }
 749
 750 void btrfs_block_release(struct btrfs_root *root, struct buffer_head *buf)
 751 {
 752         brelse(buf);
 753 }
 754