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         init_rwsem(&root->snap_sem);
 410         root->defrag_running = 0;
 411         root->defrag_level = 0;
 412         root->root_key.objectid = objectid;
 413         return 0;
 414 }
 415
 416 static int find_and_setup_root(struct btrfs_root *tree_root,
 417                                struct btrfs_fs_info *fs_info,
 418                                u64 objectid,
 419                                struct btrfs_root *root)
 420 {
 421         int ret;
 422         u32 blocksize;
 423
 424         __setup_root(tree_root->nodesize, tree_root->leafsize,
 425                      tree_root->sectorsize, tree_root->stripesize,
 426                      root, fs_info, objectid);
 427         ret = btrfs_find_last_root(tree_root, objectid,
 428                                    &root->root_item, &root->root_key);
 429         BUG_ON(ret);
 430
 431         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 432         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 433                                      blocksize);
 434         BUG_ON(!root->node);
 435         return 0;
 436 }
 437
 438 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_fs_info *fs_info,
 439                                                struct btrfs_key *location)
 440 {
 441         struct btrfs_root *root;
 442         struct btrfs_root *tree_root = fs_info->tree_root;
 443         struct btrfs_path *path;
 444         struct extent_buffer *l;
 445         u64 highest_inode;
 446         u32 blocksize;
 447         int ret = 0;
 448
 449         root = kzalloc(sizeof(*root), GFP_NOFS);
 450         if (!root)
 451                 return ERR_PTR(-ENOMEM);
 452         if (location->offset == (u64)-1) {
 453                 ret = find_and_setup_root(tree_root, fs_info,
 454                                           location->objectid, root);
 455                 if (ret) {
 456                         kfree(root);
 457                         return ERR_PTR(ret);
 458                 }
 459                 goto insert;
 460         }
 461
 462         __setup_root(tree_root->nodesize, tree_root->leafsize,
 463                      tree_root->sectorsize, tree_root->stripesize,
 464                      root, fs_info, location->objectid);
 465
 466         path = btrfs_alloc_path();
 467         BUG_ON(!path);
 468         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
 469         if (ret != 0) {
 470                 if (ret > 0)
 471                         ret = -ENOENT;
 472                 goto out;
 473         }
 474         l = path->nodes[0];
 475         read_extent_buffer(l, &root->root_item,
 476                btrfs_item_ptr_offset(l, path->slots[0]),
 477                sizeof(root->root_item));
 478         memcpy(&root->root_key, location, sizeof(*location));
 479         ret = 0;
 480 out:
 481         btrfs_release_path(root, path);
 482         btrfs_free_path(path);
 483         if (ret) {
 484                 kfree(root);
 485                 return ERR_PTR(ret);
 486         }
 487         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 488         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 489                                      blocksize);
 490         BUG_ON(!root->node);
 491 insert:
 492         root->ref_cows = 1;
 493         ret = btrfs_find_highest_inode(root, &highest_inode);
 494         if (ret == 0) {
 495                 root->highest_inode = highest_inode;
 496                 root->last_inode_alloc = highest_inode;
 497         }
 498         return root;
 499 }
 500
 501 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
 502                                               struct btrfs_key *location)
 503 {
 504         struct btrfs_root *root;
 505         int ret;
 506
 507         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 508                 return fs_info->tree_root;
 509         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
 510                 return fs_info->extent_root;
 511
 512         root = radix_tree_lookup(&fs_info->fs_roots_radix,
 513                                  (unsigned long)location->objectid);
 514         if (root)
 515                 return root;
 516
 517         root = btrfs_read_fs_root_no_radix(fs_info, location);
 518         if (IS_ERR(root))
 519                 return root;
 520         ret = radix_tree_insert(&fs_info->fs_roots_radix,
 521                                 (unsigned long)root->root_key.objectid,
 522                                 root);
 523         if (ret) {
 524                 free_extent_buffer(root->node);
 525                 kfree(root);
 526                 return ERR_PTR(ret);
 527         }
 528         ret = btrfs_find_dead_roots(fs_info->tree_root,
 529                                     root->root_key.objectid, root);
 530         BUG_ON(ret);
 531
 532         return root;
 533 }
 534
 535 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 536                                       struct btrfs_key *location,
 537                                       const char *name, int namelen)
 538 {
 539         struct btrfs_root *root;
 540         int ret;
 541
 542         root = btrfs_read_fs_root_no_name(fs_info, location);
 543         if (!root)
 544                 return NULL;
 545
 546         if (root->in_sysfs)
 547                 return root;
 548
 549         ret = btrfs_set_root_name(root, name, namelen);
 550         if (ret) {
 551                 free_extent_buffer(root->node);
 552                 kfree(root);
 553                 return ERR_PTR(ret);
 554         }
 555
 556         ret = btrfs_sysfs_add_root(root);
 557         if (ret) {
 558                 free_extent_buffer(root->node);
 559                 kfree(root->name);
 560                 kfree(root);
 561                 return ERR_PTR(ret);
 562         }
 563         root->in_sysfs = 1;
 564         return root;
 565 }
 566 #if 0
 567 static int add_hasher(struct btrfs_fs_info *info, char *type) {
 568         struct btrfs_hasher *hasher;
 569
 570         hasher = kmalloc(sizeof(*hasher), GFP_NOFS);
 571         if (!hasher)
 572                 return -ENOMEM;
 573         hasher->hash_tfm = crypto_alloc_hash(type, 0, CRYPTO_ALG_ASYNC);
 574         if (!hasher->hash_tfm) {
 575                 kfree(hasher);
 576                 return -EINVAL;
 577         }
 578         spin_lock(&info->hash_lock);
 579         list_add(&hasher->list, &info->hashers);
 580         spin_unlock(&info->hash_lock);
 581         return 0;
 582 }
 583 #endif
 584 struct btrfs_root *open_ctree(struct super_block *sb)
 585 {
 586         u32 sectorsize;
 587         u32 nodesize;
 588         u32 leafsize;
 589         u32 blocksize;
 590         u32 stripesize;
 591         struct btrfs_root *extent_root = kmalloc(sizeof(struct btrfs_root),
 592                                                  GFP_NOFS);
 593         struct btrfs_root *tree_root = kmalloc(sizeof(struct btrfs_root),
 594                                                GFP_NOFS);
 595         struct btrfs_fs_info *fs_info = kmalloc(sizeof(*fs_info),
 596                                                 GFP_NOFS);
 597         int ret;
 598         int err = -EIO;
 599         struct btrfs_super_block *disk_super;
 600
 601         if (!extent_root || !tree_root || !fs_info) {
 602                 err = -ENOMEM;
 603                 goto fail;
 604         }
 605         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
 606         INIT_LIST_HEAD(&fs_info->trans_list);
 607         INIT_LIST_HEAD(&fs_info->dead_roots);
 608         INIT_LIST_HEAD(&fs_info->hashers);
 609         spin_lock_init(&fs_info->hash_lock);
 610         spin_lock_init(&fs_info->delalloc_lock);
 611
 612         memset(&fs_info->super_kobj, 0, sizeof(fs_info->super_kobj));
 613         init_completion(&fs_info->kobj_unregister);
 614         sb_set_blocksize(sb, 4096);
 615         fs_info->running_transaction = NULL;
 616         fs_info->last_trans_committed = 0;
 617         fs_info->tree_root = tree_root;
 618         fs_info->extent_root = extent_root;
 619         fs_info->sb = sb;
 620         fs_info->mount_opt = 0;
 621         fs_info->max_extent = (u64)-1;
 622         fs_info->delalloc_bytes = 0;
 623         fs_info->btree_inode = new_inode(sb);
 624         fs_info->btree_inode->i_ino = 1;
 625         fs_info->btree_inode->i_nlink = 1;
 626         fs_info->btree_inode->i_size = sb->s_bdev->bd_inode->i_size;
 627         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
 628         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
 629                              fs_info->btree_inode->i_mapping,
 630                              GFP_NOFS);
 631         BTRFS_I(fs_info->btree_inode)->extent_tree.ops = &btree_extent_map_ops;
 632
 633         extent_map_tree_init(&fs_info->free_space_cache,
 634                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 635         extent_map_tree_init(&fs_info->block_group_cache,
 636                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 637         extent_map_tree_init(&fs_info->pinned_extents,
 638                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 639         extent_map_tree_init(&fs_info->pending_del,
 640                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 641         extent_map_tree_init(&fs_info->extent_ins,
 642                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 643         fs_info->do_barriers = 1;
 644         fs_info->closing = 0;
 645         fs_info->total_pinned = 0;
 646 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 647         INIT_WORK(&fs_info->trans_work, btrfs_transaction_cleaner, fs_info);
 648 #else
 649         INIT_DELAYED_WORK(&fs_info->trans_work, btrfs_transaction_cleaner);
 650 #endif
 651         BTRFS_I(fs_info->btree_inode)->root = tree_root;
 652         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
 653                sizeof(struct btrfs_key));
 654         insert_inode_hash(fs_info->btree_inode);
 655         mapping_set_gfp_mask(fs_info->btree_inode->i_mapping, GFP_NOFS);
 656
 657         mutex_init(&fs_info->trans_mutex);
 658         mutex_init(&fs_info->fs_mutex);
 659
 660 #if 0
 661         ret = add_hasher(fs_info, "crc32c");
 662         if (ret) {
 663                 printk("btrfs: failed hash setup, modprobe cryptomgr?\n");
 664                 err = -ENOMEM;
 665                 goto fail_iput;
 666         }
 667 #endif
 668         __setup_root(512, 512, 512, 512, tree_root,
 669                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
 670
 671         fs_info->sb_buffer = read_tree_block(tree_root,
 672                                              BTRFS_SUPER_INFO_OFFSET,
 673                                              512);
 674
 675         if (!fs_info->sb_buffer)
 676                 goto fail_iput;
 677
 678         read_extent_buffer(fs_info->sb_buffer, &fs_info->super_copy, 0,
 679                            sizeof(fs_info->super_copy));
 680
 681         read_extent_buffer(fs_info->sb_buffer, fs_info->fsid,
 682                            (unsigned long)btrfs_super_fsid(fs_info->sb_buffer),
 683                            BTRFS_FSID_SIZE);
 684         disk_super = &fs_info->super_copy;
 685         if (!btrfs_super_root(disk_super))
 686                 goto fail_sb_buffer;
 687
 688         nodesize = btrfs_super_nodesize(disk_super);
 689         leafsize = btrfs_super_leafsize(disk_super);
 690         sectorsize = btrfs_super_sectorsize(disk_super);
 691         stripesize = btrfs_super_stripesize(disk_super);
 692         tree_root->nodesize = nodesize;
 693         tree_root->leafsize = leafsize;
 694         tree_root->sectorsize = sectorsize;
 695         tree_root->stripesize = stripesize;
 696         sb_set_blocksize(sb, sectorsize);
 697
 698         i_size_write(fs_info->btree_inode,
 699                      btrfs_super_total_bytes(disk_super));
 700
 701         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
 702                     sizeof(disk_super->magic))) {
 703                 printk("btrfs: valid FS not found on %s\n", sb->s_id);
 704                 goto fail_sb_buffer;
 705         }
 706
 707         blocksize = btrfs_level_size(tree_root,
 708                                      btrfs_super_root_level(disk_super));
 709
 710         tree_root->node = read_tree_block(tree_root,
 711                                           btrfs_super_root(disk_super),
 712                                           blocksize);
 713         if (!tree_root->node)
 714                 goto fail_sb_buffer;
 715
 716         mutex_lock(&fs_info->fs_mutex);
 717
 718         ret = find_and_setup_root(tree_root, fs_info,
 719                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
 720         if (ret) {
 721                 mutex_unlock(&fs_info->fs_mutex);
 722                 goto fail_tree_root;
 723         }
 724
 725         btrfs_read_block_groups(extent_root);
 726
 727         fs_info->generation = btrfs_super_generation(disk_super) + 1;
 728         mutex_unlock(&fs_info->fs_mutex);
 729         return tree_root;
 730
 731 fail_tree_root:
 732         free_extent_buffer(tree_root->node);
 733 fail_sb_buffer:
 734         free_extent_buffer(fs_info->sb_buffer);
 735 fail_iput:
 736         iput(fs_info->btree_inode);
 737 fail:
 738         kfree(extent_root);
 739         kfree(tree_root);
 740         kfree(fs_info);
 741         return ERR_PTR(err);
 742 }
 743
 744 int write_ctree_super(struct btrfs_trans_handle *trans, struct btrfs_root
 745                       *root)
 746 {
 747         int ret;
 748         struct extent_buffer *super = root->fs_info->sb_buffer;
 749         struct inode *btree_inode = root->fs_info->btree_inode;
 750
 751         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, super);
 752         ret = sync_page_range_nolock(btree_inode, btree_inode->i_mapping,
 753                                      super->start, super->len);
 754         return ret;
 755 }
 756
 757 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
 758 {
 759         radix_tree_delete(&fs_info->fs_roots_radix,
 760                           (unsigned long)root->root_key.objectid);
 761         btrfs_sysfs_del_root(root);
 762         if (root->inode)
 763                 iput(root->inode);
 764         if (root->node)
 765                 free_extent_buffer(root->node);
 766         if (root->commit_root)
 767                 free_extent_buffer(root->commit_root);
 768         if (root->name)
 769                 kfree(root->name);
 770         kfree(root);
 771         return 0;
 772 }
 773
 774 static int del_fs_roots(struct btrfs_fs_info *fs_info)
 775 {
 776         int ret;
 777         struct btrfs_root *gang[8];
 778         int i;
 779
 780         while(1) {
 781                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
 782                                              (void **)gang, 0,
 783                                              ARRAY_SIZE(gang));
 784                 if (!ret)
 785                         break;
 786                 for (i = 0; i < ret; i++)
 787                         btrfs_free_fs_root(fs_info, gang[i]);
 788         }
 789         return 0;
 790 }
 791
 792 int close_ctree(struct btrfs_root *root)
 793 {
 794         int ret;
 795         struct btrfs_trans_handle *trans;
 796         struct btrfs_fs_info *fs_info = root->fs_info;
 797
 798         fs_info->closing = 1;
 799         btrfs_transaction_flush_work(root);
 800         mutex_lock(&fs_info->fs_mutex);
 801         btrfs_defrag_dirty_roots(root->fs_info);
 802         trans = btrfs_start_transaction(root, 1);
 803         ret = btrfs_commit_transaction(trans, root);
 804         /* run commit again to  drop the original snapshot */
 805         trans = btrfs_start_transaction(root, 1);
 806         btrfs_commit_transaction(trans, root);
 807         ret = btrfs_write_and_wait_transaction(NULL, root);
 808         BUG_ON(ret);
 809         write_ctree_super(NULL, root);
 810         mutex_unlock(&fs_info->fs_mutex);
 811
 812         if (fs_info->extent_root->node)
 813                 free_extent_buffer(fs_info->extent_root->node);
 814
 815         if (fs_info->tree_root->node)
 816                 free_extent_buffer(fs_info->tree_root->node);
 817
 818         free_extent_buffer(fs_info->sb_buffer);
 819
 820         btrfs_free_block_groups(root->fs_info);
 821         del_fs_roots(fs_info);
 822
 823         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
 824
 825         extent_map_tree_empty_lru(&fs_info->free_space_cache);
 826         extent_map_tree_empty_lru(&fs_info->block_group_cache);
 827         extent_map_tree_empty_lru(&fs_info->pinned_extents);
 828         extent_map_tree_empty_lru(&fs_info->pending_del);
 829         extent_map_tree_empty_lru(&fs_info->extent_ins);
 830         extent_map_tree_empty_lru(&BTRFS_I(fs_info->btree_inode)->extent_tree);
 831
 832         truncate_inode_pages(fs_info->btree_inode->i_mapping, 0);
 833
 834         iput(fs_info->btree_inode);
 835 #if 0
 836         while(!list_empty(&fs_info->hashers)) {
 837                 struct btrfs_hasher *hasher;
 838                 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
 839                                     hashers);
 840                 list_del(&hasher->hashers);
 841                 crypto_free_hash(&fs_info->hash_tfm);
 842                 kfree(hasher);
 843         }
 844 #endif
 845         kfree(fs_info->extent_root);
 846         kfree(fs_info->tree_root);
 847         return 0;
 848 }
 849
 850 int btrfs_buffer_uptodate(struct extent_buffer *buf)
 851 {
 852         struct inode *btree_inode = buf->first_page->mapping->host;
 853         return extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree, buf);
 854 }
 855
 856 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
 857 {
 858         struct inode *btree_inode = buf->first_page->mapping->host;
 859         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->extent_tree,
 860                                           buf);
 861 }
 862
 863 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
 864 {
 865         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 866         u64 transid = btrfs_header_generation(buf);
 867         struct inode *btree_inode = root->fs_info->btree_inode;
 868
 869         if (transid != root->fs_info->generation) {
 870                 printk(KERN_CRIT "transid mismatch buffer %llu, found %Lu running %Lu\n",
 871                         (unsigned long long)buf->start,
 872                         transid, root->fs_info->generation);
 873                 WARN_ON(1);
 874         }
 875         set_extent_buffer_dirty(&BTRFS_I(btree_inode)->extent_tree, buf);
 876 }
 877
 878 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
 879 {
 880         balance_dirty_pages_ratelimited_nr(
 881                         root->fs_info->btree_inode->i_mapping, 1);
 882 }
 883
 884 void btrfs_set_buffer_defrag(struct extent_buffer *buf)
 885 {
 886         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 887         struct inode *btree_inode = root->fs_info->btree_inode;
 888         set_extent_bits(&BTRFS_I(btree_inode)->extent_tree, buf->start,
 889                         buf->start + buf->len - 1, EXTENT_DEFRAG, GFP_NOFS);
 890 }
 891
 892 void btrfs_set_buffer_defrag_done(struct extent_buffer *buf)
 893 {
 894         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 895         struct inode *btree_inode = root->fs_info->btree_inode;
 896         set_extent_bits(&BTRFS_I(btree_inode)->extent_tree, buf->start,
 897                         buf->start + buf->len - 1, EXTENT_DEFRAG_DONE,
 898                         GFP_NOFS);
 899 }
 900
 901 int btrfs_buffer_defrag(struct extent_buffer *buf)
 902 {
 903         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 904         struct inode *btree_inode = root->fs_info->btree_inode;
 905         return test_range_bit(&BTRFS_I(btree_inode)->extent_tree,
 906                      buf->start, buf->start + buf->len - 1, EXTENT_DEFRAG, 0);
 907 }
 908
 909 int btrfs_buffer_defrag_done(struct extent_buffer *buf)
 910 {
 911         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 912         struct inode *btree_inode = root->fs_info->btree_inode;
 913         return test_range_bit(&BTRFS_I(btree_inode)->extent_tree,
 914                      buf->start, buf->start + buf->len - 1,
 915                      EXTENT_DEFRAG_DONE, 0);
 916 }
 917
 918 int btrfs_clear_buffer_defrag_done(struct extent_buffer *buf)
 919 {
 920         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 921         struct inode *btree_inode = root->fs_info->btree_inode;
 922         return clear_extent_bits(&BTRFS_I(btree_inode)->extent_tree,
 923                      buf->start, buf->start + buf->len - 1,
 924                      EXTENT_DEFRAG_DONE, GFP_NOFS);
 925 }
 926
 927 int btrfs_clear_buffer_defrag(struct extent_buffer *buf)
 928 {
 929         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 930         struct inode *btree_inode = root->fs_info->btree_inode;
 931         return clear_extent_bits(&BTRFS_I(btree_inode)->extent_tree,
 932                      buf->start, buf->start + buf->len - 1,
 933                      EXTENT_DEFRAG, GFP_NOFS);
 934 }
 935
 936 int btrfs_read_buffer(struct extent_buffer *buf)
 937 {
 938         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
 939         struct inode *btree_inode = root->fs_info->btree_inode;
 940         return read_extent_buffer_pages(&BTRFS_I(btree_inode)->extent_tree,
 941                                         buf, 0, 1);
 942 }
 943
 944 static struct extent_map_ops btree_extent_map_ops = {
 945         .writepage_io_hook = btree_writepage_io_hook,
 946 };