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                 int from_this_trans = 0;
 149
 150                 if (root->fs_info->running_transaction &&
 151                     btrfs_header_generation(buf) ==
 152                     root->fs_info->running_transaction->transid)
 153                         from_this_trans = 1;
 154
 155                 /* FIXME, this is not good */
 156                 if (from_this_trans == 0 &&
 157                     memcmp_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE)) {
 158                         u32 val;
 159                         u32 found = 0;
 160                         memcpy(&found, result, BTRFS_CRC32_SIZE);
 161
 162                         read_extent_buffer(buf, &val, 0, BTRFS_CRC32_SIZE);
 163                         printk("btrfs: %s checksum verify failed on %llu "
 164                                "wanted %X found %X from_this_trans %d\n",
 165                                root->fs_info->sb->s_id,
 166                                buf->start, val, found, from_this_trans);
 167                         return 1;
 168                 }
 169         } else {
 170                 write_extent_buffer(buf, result, 0, BTRFS_CRC32_SIZE);
 171         }
 172         return 0;
 173 }
 174
 175
 176 int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 177 {
 178         struct extent_map_tree *tree;
 179         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 180         u64 found_start;
 181         int found_level;
 182         unsigned long len;
 183         struct extent_buffer *eb;
 184         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 185
 186         if (page->private == EXTENT_PAGE_PRIVATE)
 187                 goto out;
 188         if (!page->private)
 189                 goto out;
 190         len = page->private >> 2;
 191         if (len == 0) {
 192                 WARN_ON(1);
 193         }
 194         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 195         read_extent_buffer_pages(tree, eb, start + PAGE_CACHE_SIZE, 1);
 196         found_start = btrfs_header_bytenr(eb);
 197         if (found_start != start) {
 198                 printk("warning: eb start incorrect %Lu buffer %Lu len %lu\n",
 199                        start, found_start, len);
 200         }
 201         found_level = btrfs_header_level(eb);
 202         csum_tree_block(root, eb, 0);
 203         free_extent_buffer(eb);
 204 out:
 205         return 0;
 206 }
 207
 208 static int btree_writepage_io_hook(struct page *page, u64 start, u64 end)
 209 {
 210         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 211
 212         csum_dirty_buffer(root, page);
 213         return 0;
 214 }
 215
 216 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 217 {
 218         struct extent_map_tree *tree;
 219         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 220         return extent_write_full_page(tree, page, btree_get_extent, wbc);
 221 }
 222
 223 static int btree_writepages(struct address_space *mapping,
 224                             struct writeback_control *wbc)
 225 {
 226         struct extent_map_tree *tree;
 227         tree = &BTRFS_I(mapping->host)->extent_tree;
 228         if (wbc->sync_mode == WB_SYNC_NONE) {
 229                 u64 num_dirty;
 230                 u64 start = 0;
 231                 unsigned long thresh = 96 * 1024 * 1024;
 232
 233                 if (wbc->for_kupdate)
 234                         return 0;
 235
 236                 if (current_is_pdflush()) {
 237                         thresh = 96 * 1024 * 1024;
 238                 } else {
 239                         thresh = 8 * 1024 * 1024;
 240                 }
 241                 num_dirty = count_range_bits(tree, &start, (u64)-1,
 242                                              thresh, EXTENT_DIRTY);
 243                 if (num_dirty < thresh) {
 244                         return 0;
 245                 }
 246         }
 247         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 248 }
 249
 250 int btree_readpage(struct file *file, struct page *page)
 251 {
 252         struct extent_map_tree *tree;
 253         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 254         return extent_read_full_page(tree, page, btree_get_extent);
 255 }
 256
 257 static int btree_releasepage(struct page *page, gfp_t unused_gfp_flags)
 258 {
 259         struct extent_map_tree *tree;
 260         int ret;
 261
 262         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 263         ret = try_release_extent_mapping(tree, page);
 264         if (ret == 1) {
 265                 ClearPagePrivate(page);
 266                 set_page_private(page, 0);
 267                 page_cache_release(page);
 268         }
 269         return ret;
 270 }
 271
 272 static void btree_invalidatepage(struct page *page, unsigned long offset)
 273 {
 274         struct extent_map_tree *tree;
 275         tree = &BTRFS_I(page->mapping->host)->extent_tree;
 276         extent_invalidatepage(tree, page, offset);
 277         btree_releasepage(page, GFP_NOFS);
 278 }
 279
 280 #if 0
 281 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 282 {
 283         struct buffer_head *bh;
 284         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 285         struct buffer_head *head;
 286         if (!page_has_buffers(page)) {
 287                 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
 288                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
 289         }
 290         head = page_buffers(page);
 291         bh = head;
 292         do {
 293                 if (buffer_dirty(bh))
 294                         csum_tree_block(root, bh, 0);
 295                 bh = bh->b_this_page;
 296         } while (bh != head);
 297         return block_write_full_page(page, btree_get_block, wbc);
 298 }
 299 #endif
 300
 301 static struct address_space_operations btree_aops = {
 302         .readpage       = btree_readpage,
 303         .writepage      = btree_writepage,
 304         .writepages     = btree_writepages,
 305         .releasepage    = btree_releasepage,
 306         .invalidatepage = btree_invalidatepage,
 307         .sync_page      = block_sync_page,
 308 };
 309
 310 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize)
 311 {
 312         struct extent_buffer *buf = NULL;
 313         struct inode *btree_inode = root->fs_info->btree_inode;
 314         int ret = 0;
 315
 316         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 317         if (!buf)
 318                 return 0;
 319         read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
 320                                  buf, 0, 0);
 321         free_extent_buffer(buf);
 322         return ret;
 323 }
 324
 325 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 326                                       u32 blocksize)
 327 {
 328         struct extent_buffer *buf = NULL;
 329         struct inode *btree_inode = root->fs_info->btree_inode;
 330         struct extent_map_tree *extent_tree;
 331         u64 end;
 332         int ret;
 333
 334         extent_tree = &BTRFS_I(btree_inode)->extent_tree;
 335
 336         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 337         if (!buf)
 338                 return NULL;
 339         read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
 340                                  buf, 0, 1);
 341
 342         if (buf->flags & EXTENT_CSUM)
 343                 return buf;
 344
 345         end = buf->start + PAGE_CACHE_SIZE - 1;
 346         if (test_range_bit(extent_tree, buf->start, end, EXTENT_CSUM, 1)) {
 347                 buf->flags |= EXTENT_CSUM;
 348                 return buf;
 349         }
 350
 351         lock_extent(extent_tree, buf->start, end, GFP_NOFS);
 352
 353         if (test_range_bit(extent_tree, buf->start, end, EXTENT_CSUM, 1)) {
 354                 buf->flags |= EXTENT_CSUM;
 355                 goto out_unlock;
 356         }
 357
 358         ret = csum_tree_block(root, buf, 1);
 359         set_extent_bits(extent_tree, buf->start, end, EXTENT_CSUM, GFP_NOFS);
 360         buf->flags |= EXTENT_CSUM;
 361
 362 out_unlock:
 363         unlock_extent(extent_tree, buf->start, end, GFP_NOFS);
 364         return buf;
 365 }
 366
 367 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 368                      struct extent_buffer *buf)
 369 {
 370         struct inode *btree_inode = root->fs_info->btree_inode;
 371         clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
 372         return 0;
 373 }
 374
 375 int wait_on_tree_block_writeback(struct btrfs_root *root,
 376                                  struct extent_buffer *buf)
 377 {
 378         struct inode *btree_inode = root->fs_info->btree_inode;
 379         wait_on_extent_buffer_writeback(&BTRFS_I(btree_inode)->extent_tree,
 380                                         buf);
 381         return 0;
 382 }
 383
 384 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 385                         u32 stripesize, struct btrfs_root *root,
 386                         struct btrfs_fs_info *fs_info,
 387                         u64 objectid)
 388 {
 389         root->node = NULL;
 390         root->inode = NULL;
 391         root->commit_root = NULL;
 392         root->sectorsize = sectorsize;
 393         root->nodesize = nodesize;
 394         root->leafsize = leafsize;
 395         root->stripesize = stripesize;
 396         root->ref_cows = 0;
 397         root->fs_info = fs_info;
 398         root->objectid = objectid;
 399         root->last_trans = 0;
 400         root->highest_inode = 0;
 401         root->last_inode_alloc = 0;
 402         root->name = NULL;
 403         root->in_sysfs = 0;
 404         memset(&root->root_key, 0, sizeof(root->root_key));
 405         memset(&root->root_item, 0, sizeof(root->root_item));
 406         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 407         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 408         init_completion(&root->kobj_unregister);
 409         root->defrag_running = 0;
 410         root->defrag_level = 0;
 411         root->root_key.objectid = objectid;
 412         return 0;
 413 }
 414
 415 static int find_and_setup_root(struct btrfs_root *tree_root,
 416                                struct btrfs_fs_info *fs_info,
 417                                u64 objectid,
 418                                struct btrfs_root *root)
 419 {
 420         int ret;
 421         u32 blocksize;
 422
 423         __setup_root(tree_root->nodesize, tree_root->leafsize,
 424                      tree_root->sectorsize, tree_root->stripesize,
 425                      root, fs_info, objectid);
 426         ret = btrfs_find_last_root(tree_root, objectid,
 427                                    &root->root_item, &root->root_key);
 428         BUG_ON(ret);
 429
 430         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 431         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 432                                      blocksize);
 433         BUG_ON(!root->node);
 434         return 0;
 435 }
 436
 437 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 438                                                struct btrfs_key *location)
 439 {
 440         struct btrfs_root *root;
 441         struct btrfs_root *tree_root = fs_info->tree_root;
 442         struct btrfs_path *path;
 443         struct extent_buffer *l;
 444         u64 highest_inode;
 445         u32 blocksize;
 446         int ret = 0;
 447
 448         root = kzalloc(sizeof(*root), GFP_NOFS);
 449         if (!root)
 450                 return ERR_PTR(-ENOMEM);
 451         if (location->offset == (u64)-1) {
 452                 ret = find_and_setup_root(tree_root, fs_info,
 453                                           location->objectid, root);
 454                 if (ret) {
 455                         kfree(root);
 456                         return ERR_PTR(ret);
 457                 }
 458                 goto insert;
 459         }
 460
 461         __setup_root(tree_root->nodesize, tree_root->leafsize,
 462                      tree_root->sectorsize, tree_root->stripesize,
 463                      root, fs_info, location->objectid);
 464
 465         path = btrfs_alloc_path();
 466         BUG_ON(!path);
 467         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 468         if (ret != 0) {
 469                 if (ret > 0)
 470                         ret = -ENOENT;
 471                 goto out;
 472         }
 473         l = path->nodes[0];
 474         read_extent_buffer(l, &root->root_item,
 475                btrfs_item_ptr_offset(l, path->slots[0]),
 476                sizeof(root->root_item));
 477         memcpy(&root->root_key, location, sizeof(*location));
 478         ret = 0;
 479 out:
 480         btrfs_release_path(root, path);
 481         btrfs_free_path(path);
 482         if (ret) {
 483                 kfree(root);
 484                 return ERR_PTR(ret);
 485         }
 486         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 487         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 488                                      blocksize);
 489         BUG_ON(!root->node);
 490 insert:
 491         root->ref_cows = 1;
 492         ret = btrfs_find_highest_inode(root, &highest_inode);
 493         if (ret == 0) {
 494                 root->highest_inode = highest_inode;
 495                 root->last_inode_alloc = highest_inode;
 496         }
 497         return root;
 498 }
 499
 500 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
 501                                         u64 root_objectid)
 502 {
 503         struct btrfs_root *root;
 504
 505         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
 506                 return fs_info->tree_root;
 507         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
 508                 return fs_info->extent_root;
 509
 510         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 511                                  (unsigned long)root_objectid);
 512         return root;
 513 }
 514
 515 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 516                                               struct btrfs_key *location)
 517 {
 518         struct btrfs_root *root;
 519         int ret;
 520
 521         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 522                 return fs_info->tree_root;
 523         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 524                 return fs_info->extent_root;
 525
 526         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 527                                  (unsigned long)location->objectid);
 528         if (root)
 529                 return root;
 530
 531         root = btrfs_read_fs_root_no_radix(fs_info, location);
 532         if (IS_ERR(root))
 533                 return root;
 534         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 535                                 (unsigned long)root->root_key.objectid,
 536                                 root);
 537         if (ret) {
 538                 free_extent_buffer(root->node);
 539                 kfree(root);
 540                 return ERR_PTR(ret);
 541         }
 542         ret = btrfs_find_dead_roots(fs_info->tree_root,
 543                                     root->root_key.objectid, root);
 544         BUG_ON(ret);
 545
 546         return root;
 547 }
 548
 549 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 550                                       struct btrfs_key *location,
 551                                       const char *name, int namelen)
 552 {
 553         struct btrfs_root *root;
 554         int ret;
 555
 556         root = btrfs_read_fs_root_no_name(fs_info, location);
 557         if (!root)
 558                 return NULL;
 559
 560         if (root->in_sysfs)
 561                 return root;
 562
 563         ret = btrfs_set_root_name(root, name, namelen);
 564         if (ret) {
 565                 free_extent_buffer(root->node);
 566                 kfree(root);
 567                 return ERR_PTR(ret);
 568         }
 569
 570         ret = btrfs_sysfs_add_root(root);
 571         if (ret) {
 572                 free_extent_buffer(root->node);
 573                 kfree(root->name);
 574                 kfree(root);
 575                 return ERR_PTR(ret);
 576         }
 577         root->in_sysfs = 1;
 578         return root;
 579 }
 580 #if 0
 581 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 582         struct btrfs_hasher *hasher;
 583
 584         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 585         if (!hasher)
 586                 return -ENOMEM;
 587         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 588         if (!hasher->hash_tfm) {
 589                 kfree(hasher);
 590                 return -EINVAL;
 591         }
 592         spin_lock(&info->hash_lock);
 593         list_add(&hasher->list, &info->hashers);
 594         spin_unlock(&info->hash_lock);
 595         return 0;
 596 }
 597 #endif
 598 struct btrfs_root *open_ctree(struct super_block *sb)
 599 {
 600         u32 sectorsize;
 601         u32 nodesize;
 602         u32 leafsize;
 603         u32 blocksize;
 604         u32 stripesize;
 605         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 606                                                  GFP_NOFS);
 607         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 608                                                GFP_NOFS);
 609         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 610                                                 GFP_NOFS);
 611         int ret;
 612         int err = -EIO;
 613         struct btrfs_super_block *disk_super;
 614
 615         if (!extent_root || !tree_root || !fs_info) {
 616                 err = -ENOMEM;
 617                 goto fail;
 618         }
 619         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 620         INIT_LIST_HEAD(&fs_info->trans_list);
 621         INIT_LIST_HEAD(&fs_info->dead_roots);
 622         INIT_LIST_HEAD(&fs_info->hashers);
 623         spin_lock_init(&fs_info->hash_lock);
 624         spin_lock_init(&fs_info->delalloc_lock);
 625
 626         memset(&fs_info->super_kobj, 0, sizeof(fs_info->super_kobj));
 627         init_completion(&fs_info->kobj_unregister);
 628         sb_set_blocksize(sb, 4096);
 629         fs_info->running_transaction = NULL;
 630         fs_info->last_trans_committed = 0;
 631         fs_info->tree_root = tree_root;
 632         fs_info->extent_root = extent_root;
 633         fs_info->sb = sb;
 634         fs_info->throttles = 0;
 635         fs_info->mount_opt = 0;
 636         fs_info->max_extent = (u64)-1;
 637         fs_info->delalloc_bytes = 0;
 638         fs_info->btree_inode = new_inode(sb);
 639         fs_info->btree_inode->i_ino = 1;
 640         fs_info->btree_inode->i_nlink = 1;
 641         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 642         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 643         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
 644                              fs_info->btree_inode->i_mapping,
 645                              GFP_NOFS);
 646         BTRFS_I(fs_info->btree_inode)->extent_tree.ops = &btree_extent_map_ops;
 647
 648         extent_map_tree_init(&fs_info->free_space_cache,
 649                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 650         extent_map_tree_init(&fs_info->block_group_cache,
 651                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 652         extent_map_tree_init(&fs_info->pinned_extents,
 653                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 654         extent_map_tree_init(&fs_info->pending_del,
 655                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 656         extent_map_tree_init(&fs_info->extent_ins,
 657                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 658         fs_info->do_barriers = 1;
 659         fs_info->closing = 0;
 660         fs_info->total_pinned = 0;
 661 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 662         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
 663 #else
 664         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
 665 #endif
 666         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 667         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 668                sizeof(struct btrfs_key));
 669         insert_inode_hash(fs_info->btree_inode);
 670         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 671
 672         mutex_init(&fs_info->trans_mutex);
 673         mutex_init(&fs_info->fs_mutex);
 674
 675 #if 0
 676         ret = add_hasher(fs_info, "crc32c");
 677         if (ret) {
 678                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
 679                 err = -ENOMEM;
 680                 goto fail_iput;
 681         }
 682 #endif
 683         __setup_root(512, 512, 512, 512, tree_root,
 684                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 685
 686         fs_info->sb_buffer = read_tree_block(tree_root,
 687                                              BTRFS_SUPER_INFO_OFFSET,
 688                                              512);
 689
 690         if (!fs_info->sb_buffer)
 691                 goto fail_iput;
 692
 693         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
 694                            sizeof(fs_info->super_copy));
 695
 696         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
 697                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
 698                            BTRFS_FSID_SIZE);
 699         disk_super = &fs_info->super_copy;
 700         if (!btrfs_super_root(disk_super))
 701                 goto fail_sb_buffer;
 702
 703         nodesize = btrfs_super_nodesize(disk_super);
 704         leafsize = btrfs_super_leafsize(disk_super);
 705         sectorsize = btrfs_super_sectorsize(disk_super);
 706         stripesize = btrfs_super_stripesize(disk_super);
 707         tree_root->nodesize = nodesize;
 708         tree_root->leafsize = leafsize;
 709         tree_root->sectorsize = sectorsize;
 710         tree_root->stripesize = stripesize;
 711         sb_set_blocksize(sb, sectorsize);
 712
 713         i_size_write(fs_info->btree_inode,
 714                      btrfs_super_total_bytes(disk_super));
 715
 716         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 717                     sizeof(disk_super->magic))) {
 718                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
 719                 goto fail_sb_buffer;
 720         }
 721
 722         blocksize = btrfs_level_size(tree_root,
 723                                      btrfs_super_root_level(disk_super));
 724
 725         tree_root->node = read_tree_block(tree_root,
 726                                           btrfs_super_root(disk_super),
 727                                           blocksize);
 728         if (!tree_root->node)
 729                 goto fail_sb_buffer;
 730
 731         mutex_lock(&fs_info->fs_mutex);
 732
 733         ret = find_and_setup_root(tree_root, fs_info,
 734                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 735         if (ret) {
 736                 mutex_unlock(&fs_info->fs_mutex);
 737                 goto fail_tree_root;
 738         }
 739
 740         btrfs_read_block_groups(extent_root);
 741
 742         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 743         mutex_unlock(&fs_info->fs_mutex);
 744         return tree_root;
 745
 746 fail_tree_root:
 747         free_extent_buffer(tree_root->node);
 748 fail_sb_buffer:
 749         free_extent_buffer(fs_info->sb_buffer);
 750 fail_iput:
 751         iput(fs_info->btree_inode);
 752 fail:
 753         kfree(extent_root);
 754         kfree(tree_root);
 755         kfree(fs_info);
 756         return ERR_PTR(err);
 757 }
 758
 759 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 760                       *root)
 761 {
 762         int ret;
 763         struct extent_buffer *super = root->fs_info->sb_buffer;
 764         struct inode *btree_inode = root->fs_info->btree_inode;
 765
 766         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, super);
 767         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
 768                                      super->start, super->len);
 769         return ret;
 770 }
 771
 772 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 773 {
 774         radix_tree_delete(&fs_info->fs_roots_radix,
 775                           (unsigned long)root->root_key.objectid);
 776         btrfs_sysfs_del_root(root);
 777         if (root->inode)
 778                 iput(root->inode);
 779         if (root->node)
 780                 free_extent_buffer(root->node);
 781         if (root->commit_root)
 782                 free_extent_buffer(root->commit_root);
 783         if (root->name)
 784                 kfree(root->name);
 785         kfree(root);
 786         return 0;
 787 }
 788
 789 static int del_fs_roots(struct btrfs_fs_info *fs_info)
 790 {
 791         int ret;
 792         struct btrfs_root *gang[8];
 793         int i;
 794
 795         while(1) {
 796                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 797                                              (void **)gang, 0,
 798                                              ARRAY_SIZE(gang));
 799                 if (!ret)
 800                         break;
 801                 for (i = 0; i < ret; i++)
 802                         btrfs_free_fs_root(fs_info, gang[i]);
 803         }
 804         return 0;
 805 }
 806
 807 int close_ctree(struct btrfs_root *root)
 808 {
 809         int ret;
 810         struct btrfs_trans_handle *trans;
 811         struct btrfs_fs_info *fs_info = root->fs_info;
 812
 813         fs_info->closing = 1;
 814         btrfs_transaction_flush_work(root);
 815         mutex_lock(&fs_info->fs_mutex);
 816         btrfs_defrag_dirty_roots(root->fs_info);
 817         trans = btrfs_start_transaction(root, 1);
 818         ret = btrfs_commit_transaction(trans, root);
 819         /* run commit again to  drop the original snapshot */
 820         trans = btrfs_start_transaction(root, 1);
 821         btrfs_commit_transaction(trans, root);
 822         ret = btrfs_write_and_wait_transaction(NULL, root);
 823         BUG_ON(ret);
 824         write_ctree_super(NULL, root);
 825         mutex_unlock(&fs_info->fs_mutex);
 826
 827         if (fs_info->extent_root->node)
 828                 free_extent_buffer(fs_info->extent_root->node);
 829
 830         if (fs_info->tree_root->node)
 831                 free_extent_buffer(fs_info->tree_root->node);
 832
 833         free_extent_buffer(fs_info->sb_buffer);
 834
 835         btrfs_free_block_groups(root->fs_info);
 836         del_fs_roots(fs_info);
 837
 838         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
 839
 840         extent_map_tree_empty_lru(&fs_info->free_space_cache);
 841         extent_map_tree_empty_lru(&fs_info->block_group_cache);
 842         extent_map_tree_empty_lru(&fs_info->pinned_extents);
 843         extent_map_tree_empty_lru(&fs_info->pending_del);
 844         extent_map_tree_empty_lru(&fs_info->extent_ins);
 845         extent_map_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->extent_tree);
 846
 847         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
 848
 849         iput(fs_info->btree_inode);
 850 #if 0
 851         while(!list_empty(&fs_info->hashers)) {
 852                 struct btrfs_hasher *hasher;
 853                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
 854                                     hashers);
 855                 list_del(&hasher->hashers);
 856                 crypto_free_hash(&fs_info->hash_tfm);
 857                 kfree(hasher);
 858         }
 859 #endif
 860         kfree(fs_info->extent_root);
 861         kfree(fs_info->tree_root);
 862         return 0;
 863 }
 864
 865 int btrfs_buffer_uptodate(struct extent_buffer *buf)
 866 {
 867         struct inode *btree_inode = buf->first_page->mapping->host;
 868         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree, buf);
 869 }
 870
 871 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
 872 {
 873         struct inode *btree_inode = buf->first_page->mapping->host;
 874         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree,
 875                                           buf);
 876 }
 877
 878 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 879 {
 880         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 881         u64 transid = btrfs_header_generation(buf);
 882         struct inode *btree_inode = root->fs_info->btree_inode;
 883
 884         if (transid != root->fs_info->generation) {
 885                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
 886                         (unsigned long long)buf->start,
 887                         transid, root->fs_info->generation);
 888                 WARN_ON(1);
 889         }
 890         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
 891 }
 892
 893 void btrfs_throttle(struct btrfs_root *root)
 894 {
 895         if (root->fs_info->throttles)
 896                 congestion_wait(WRITE, HZ/10);
 897 }
 898
 899 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
 900 {
 901         balance_dirty_pages_ratelimited_nr(
 902                         root->fs_info->btree_inode->i_mapping, 1);
 903 }
 904
 905 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
 906 {
 907         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 908         struct inode *btree_inode = root->fs_info->btree_inode;
 909         set_extent_bits(&BTRFS_I(btree_inode)->extent_tree, buf->start,
 910                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
 911 }
 912
 913 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
 914 {
 915         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 916         struct inode *btree_inode = root->fs_info->btree_inode;
 917         set_extent_bits(&BTRFS_I(btree_inode)->extent_tree, buf->start,
 918                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
 919                         GFP_NOFS);
 920 }
 921
 922 int btrfs_buffer_defrag(struct extent_buffer *buf)
 923 {
 924         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 925         struct inode *btree_inode = root->fs_info->btree_inode;
 926         return test_range_bit(&BTRFS_I(btree_inode)->extent_tree,
 927                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
 928 }
 929
 930 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
 931 {
 932         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 933         struct inode *btree_inode = root->fs_info->btree_inode;
 934         return test_range_bit(&BTRFS_I(btree_inode)->extent_tree,
 935                      buf->start, buf->start + buf->len - 1,
 936                      EXTENT_DEFRAG_DONE, 0);
 937 }
 938
 939 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
 940 {
 941         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 942         struct inode *btree_inode = root->fs_info->btree_inode;
 943         return clear_extent_bits(&BTRFS_I(btree_inode)->extent_tree,
 944                      buf->start, buf->start + buf->len - 1,
 945                      EXTENT_DEFRAG_DONE, GFP_NOFS);
 946 }
 947
 948 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
 949 {
 950         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 951         struct inode *btree_inode = root->fs_info->btree_inode;
 952         return clear_extent_bits(&BTRFS_I(btree_inode)->extent_tree,
 953                      buf->start, buf->start + buf->len - 1,
 954                      EXTENT_DEFRAG, GFP_NOFS);
 955 }
 956
 957 int btrfs_read_buffer(struct extent_buffer *buf)
 958 {
 959         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 960         struct inode *btree_inode = root->fs_info->btree_inode;
 961         return read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
 962                                         buf, 0, 1);
 963 }
 964
 965 static struct extent_map_ops btree_extent_map_ops = {
 966         .writepage_io_hook = btree_writepage_io_hook,
 967 };