SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/inode.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/buffer_head.h>
  20 #include <linux/fs.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/string.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mpage.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/version.h>
  35 #include <linux/xattr.h>
  36 #include "ctree.h"
  37 #include "disk-io.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "ioctl.h"
  41 #include "print-tree.h"
  42
  43 struct btrfs_iget_args {
  44         u64 ino;
  45         struct btrfs_root *root;
  46 };
  47
  48 static struct inode_operations btrfs_dir_inode_operations;
  49 static struct inode_operations btrfs_symlink_inode_operations;
  50 static struct inode_operations btrfs_dir_ro_inode_operations;
  51 static struct inode_operations btrfs_special_inode_operations;
  52 static struct inode_operations btrfs_file_inode_operations;
  53 static struct address_space_operations btrfs_aops;
  54 static struct address_space_operations btrfs_symlink_aops;
  55 static struct file_operations btrfs_dir_file_operations;
  56 static struct extent_map_ops btrfs_extent_map_ops;
  57
  58 static struct kmem_cache *btrfs_inode_cachep;
  59 struct kmem_cache *btrfs_trans_handle_cachep;
  60 struct kmem_cache *btrfs_transaction_cachep;
  61 struct kmem_cache *btrfs_bit_radix_cachep;
  62 struct kmem_cache *btrfs_path_cachep;
  63
  64 #define S_SHIFT 12
  65 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  66         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  67         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  68         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  69         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  70         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  71         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  72         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  73 };
  74
  75 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  76                            int for_del)
  77 {
  78         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  79         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  80         u64 thresh;
  81         int ret = 0;
  82
  83         if (for_del)
  84                 thresh = total * 90;
  85         else
  86                 thresh = total * 85;
  87
  88         do_div(thresh, 100);
  89
  90         spin_lock(&root->fs_info->delalloc_lock);
  91         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  92                 ret = -ENOSPC;
  93         spin_unlock(&root->fs_info->delalloc_lock);
  94         return ret;
  95 }
  96
  97 static int cow_file_range(struct inode *inode, u64 start, u64 end)
  98 {
  99         struct btrfs_root *root = BTRFS_I(inode)->root;
 100         struct btrfs_trans_handle *trans;
 101         u64 alloc_hint = 0;
 102         u64 num_bytes;
 103         u64 cur_alloc_size;
 104         u64 blocksize = root->sectorsize;
 105         struct btrfs_key ins;
 106         int ret;
 107
 108         trans = btrfs_start_transaction(root, 1);
 109         BUG_ON(!trans);
 110         btrfs_set_trans_block_group(trans, inode);
 111
 112         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 113         num_bytes = max(blocksize,  num_bytes);
 114         ret = btrfs_drop_extents(trans, root, inode,
 115                                  start, start + num_bytes, start, &alloc_hint);
 116
 117         if (alloc_hint == EXTENT_MAP_INLINE)
 118                 goto out;
 119
 120         while(num_bytes > 0) {
 121                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 122                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 123                                          root->root_key.objectid,
 124                                          trans->transid,
 125                                          inode->i_ino, start, 0,
 126                                          alloc_hint, (u64)-1, &ins, 1);
 127                 if (ret) {
 128                         WARN_ON(1);
 129                         goto out;
 130                 }
 131                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 132                                                start, ins.objectid, ins.offset,
 133                                                ins.offset);
 134                 num_bytes -= cur_alloc_size;
 135                 alloc_hint = ins.objectid + ins.offset;
 136                 start += cur_alloc_size;
 137         }
 138         btrfs_add_ordered_inode(inode);
 139 out:
 140         btrfs_end_transaction(trans, root);
 141         return ret;
 142 }
 143
 144 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 145 {
 146         u64 extent_start;
 147         u64 extent_end;
 148         u64 bytenr;
 149         u64 cow_end;
 150         u64 loops = 0;
 151         u64 total_fs_bytes;
 152         struct btrfs_root *root = BTRFS_I(inode)->root;
 153         struct extent_buffer *leaf;
 154         int found_type;
 155         struct btrfs_path *path;
 156         struct btrfs_file_extent_item *item;
 157         int ret;
 158         int err;
 159         struct btrfs_key found_key;
 160
 161         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 162         path = btrfs_alloc_path();
 163         BUG_ON(!path);
 164 again:
 165         ret = btrfs_lookup_file_extent(NULL, root, path,
 166                                        inode->i_ino, start, 0);
 167         if (ret < 0) {
 168                 btrfs_free_path(path);
 169                 return ret;
 170         }
 171
 172         cow_end = end;
 173         if (ret != 0) {
 174                 if (path->slots[0] == 0)
 175                         goto not_found;
 176                 path->slots[0]--;
 177         }
 178
 179         leaf = path->nodes[0];
 180         item = btrfs_item_ptr(leaf, path->slots[0],
 181                               struct btrfs_file_extent_item);
 182
 183         /* are we inside the extent that was found? */
 184         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 185         found_type = btrfs_key_type(&found_key);
 186         if (found_key.objectid != inode->i_ino ||
 187             found_type != BTRFS_EXTENT_DATA_KEY) {
 188                 goto not_found;
 189         }
 190
 191         found_type = btrfs_file_extent_type(leaf, item);
 192         extent_start = found_key.offset;
 193         if (found_type == BTRFS_FILE_EXTENT_REG) {
 194                 u64 extent_num_bytes;
 195
 196                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 197                 extent_end = extent_start + extent_num_bytes;
 198                 err = 0;
 199
 200                 if (loops && start != extent_start)
 201                         goto not_found;
 202
 203                 if (start < extent_start || start >= extent_end)
 204                         goto not_found;
 205
 206                 cow_end = min(end, extent_end - 1);
 207                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 208                 if (bytenr == 0)
 209                         goto not_found;
 210
 211                 /*
 212                  * we may be called by the resizer, make sure we're inside
 213                  * the limits of the FS
 214                  */
 215                 if (bytenr + extent_num_bytes > total_fs_bytes)
 216                         goto not_found;
 217
 218                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 219                         goto not_found;
 220                 }
 221
 222                 start = extent_end;
 223         } else {
 224                 goto not_found;
 225         }
 226 loop:
 227         if (start > end) {
 228                 btrfs_free_path(path);
 229                 return 0;
 230         }
 231         btrfs_release_path(root, path);
 232         loops++;
 233         goto again;
 234
 235 not_found:
 236         cow_file_range(inode, start, cow_end);
 237         start = cow_end + 1;
 238         goto loop;
 239 }
 240
 241 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 242 {
 243         struct btrfs_root *root = BTRFS_I(inode)->root;
 244         u64 num_bytes;
 245         int ret;
 246         mutex_lock(&root->fs_info->fs_mutex);
 247         if (btrfs_test_opt(root, NODATACOW) ||
 248             btrfs_test_flag(inode, NODATACOW))
 249                 ret = run_delalloc_nocow(inode, start, end);
 250         else
 251                 ret = cow_file_range(inode, start, end);
 252
 253         spin_lock(&root->fs_info->delalloc_lock);
 254         num_bytes = end + 1 - start;
 255         if (root->fs_info->delalloc_bytes < num_bytes) {
 256                 printk("delalloc accounting error total %llu sub %llu\n",
 257                        root->fs_info->delalloc_bytes, num_bytes);
 258         } else {
 259                 root->fs_info->delalloc_bytes -= num_bytes;
 260         }
 261         spin_unlock(&root->fs_info->delalloc_lock);
 262
 263         mutex_unlock(&root->fs_info->fs_mutex);
 264         return ret;
 265 }
 266
 267 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 268 {
 269         struct inode *inode = page->mapping->host;
 270         struct btrfs_root *root = BTRFS_I(inode)->root;
 271         struct btrfs_trans_handle *trans;
 272         char *kaddr;
 273         int ret = 0;
 274         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 275         size_t offset = start - page_start;
 276         if (btrfs_test_opt(root, NODATASUM) ||
 277             btrfs_test_flag(inode, NODATASUM))
 278                 return 0;
 279         mutex_lock(&root->fs_info->fs_mutex);
 280         trans = btrfs_start_transaction(root, 1);
 281         btrfs_set_trans_block_group(trans, inode);
 282         kaddr = kmap(page);
 283         btrfs_csum_file_block(trans, root, inode, inode->i_ino,
 284                               start, kaddr + offset, end - start + 1);
 285         kunmap(page);
 286         ret = btrfs_end_transaction(trans, root);
 287         BUG_ON(ret);
 288         mutex_unlock(&root->fs_info->fs_mutex);
 289         return ret;
 290 }
 291
 292 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 293 {
 294         int ret = 0;
 295         struct inode *inode = page->mapping->host;
 296         struct btrfs_root *root = BTRFS_I(inode)->root;
 297         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 298         struct btrfs_csum_item *item;
 299         struct btrfs_path *path = NULL;
 300         u32 csum;
 301         if (btrfs_test_opt(root, NODATASUM) ||
 302             btrfs_test_flag(inode, NODATASUM))
 303                 return 0;
 304         mutex_lock(&root->fs_info->fs_mutex);
 305         path = btrfs_alloc_path();
 306         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 307         if (IS_ERR(item)) {
 308                 ret = PTR_ERR(item);
 309                 /* a csum that isn't present is a preallocated region. */
 310                 if (ret == -ENOENT || ret == -EFBIG)
 311                         ret = 0;
 312                 csum = 0;
 313                 goto out;
 314         }
 315         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 316                            BTRFS_CRC32_SIZE);
 317         set_state_private(em_tree, start, csum);
 318 out:
 319         if (path)
 320                 btrfs_free_path(path);
 321         mutex_unlock(&root->fs_info->fs_mutex);
 322         return ret;
 323 }
 324
 325 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end)
 326 {
 327         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 328         struct inode *inode = page->mapping->host;
 329         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 330         char *kaddr;
 331         u64 private;
 332         int ret;
 333         struct btrfs_root *root = BTRFS_I(inode)->root;
 334         u32 csum = ~(u32)0;
 335         unsigned long flags;
 336         if (btrfs_test_opt(root, NODATASUM) ||
 337             btrfs_test_flag(inode, NODATASUM))
 338                 return 0;
 339         ret = get_state_private(em_tree, start, &private);
 340         local_irq_save(flags);
 341         kaddr = kmap_atomic(page, KM_IRQ0);
 342         if (ret) {
 343                 goto zeroit;
 344         }
 345         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 346         btrfs_csum_final(csum, (char *)&csum);
 347         if (csum != private) {
 348                 goto zeroit;
 349         }
 350         kunmap_atomic(kaddr, KM_IRQ0);
 351         local_irq_restore(flags);
 352         return 0;
 353
 354 zeroit:
 355         printk("btrfs csum failed ino %lu off %llu\n",
 356                page->mapping->host->i_ino, (unsigned long long)start);
 357         memset(kaddr + offset, 1, end - start + 1);
 358         flush_dcache_page(page);
 359         kunmap_atomic(kaddr, KM_IRQ0);
 360         local_irq_restore(flags);
 361         return 0;
 362 }
 363
 364 void btrfs_read_locked_inode(struct inode *inode)
 365 {
 366         struct btrfs_path *path;
 367         struct extent_buffer *leaf;
 368         struct btrfs_inode_item *inode_item;
 369         struct btrfs_inode_timespec *tspec;
 370         struct btrfs_root *root = BTRFS_I(inode)->root;
 371         struct btrfs_key location;
 372         u64 alloc_group_block;
 373         u32 rdev;
 374         int ret;
 375
 376         path = btrfs_alloc_path();
 377         BUG_ON(!path);
 378         mutex_lock(&root->fs_info->fs_mutex);
 379         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 380
 381         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 382         if (ret)
 383                 goto make_bad;
 384
 385         leaf = path->nodes[0];
 386         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 387                                     struct btrfs_inode_item);
 388
 389         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 390         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 391         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 392         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 393         inode->i_size = btrfs_inode_size(leaf, inode_item);
 394
 395         tspec = btrfs_inode_atime(inode_item);
 396         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 397         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 398
 399         tspec = btrfs_inode_mtime(inode_item);
 400         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 401         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 402
 403         tspec = btrfs_inode_ctime(inode_item);
 404         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 405         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 406
 407         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 408         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 409         inode->i_rdev = 0;
 410         rdev = btrfs_inode_rdev(leaf, inode_item);
 411
 412         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 413         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 414                                                        alloc_group_block);
 415         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 416         if (!BTRFS_I(inode)->block_group) {
 417                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 418                                                          NULL, 0, 0, 0);
 419         }
 420         btrfs_free_path(path);
 421         inode_item = NULL;
 422
 423         mutex_unlock(&root->fs_info->fs_mutex);
 424
 425         switch (inode->i_mode & S_IFMT) {
 426         case S_IFREG:
 427                 inode->i_mapping->a_ops = &btrfs_aops;
 428                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
 429                 inode->i_fop = &btrfs_file_operations;
 430                 inode->i_op = &btrfs_file_inode_operations;
 431                 break;
 432         case S_IFDIR:
 433                 inode->i_fop = &btrfs_dir_file_operations;
 434                 if (root == root->fs_info->tree_root)
 435                         inode->i_op = &btrfs_dir_ro_inode_operations;
 436                 else
 437                         inode->i_op = &btrfs_dir_inode_operations;
 438                 break;
 439         case S_IFLNK:
 440                 inode->i_op = &btrfs_symlink_inode_operations;
 441                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 442                 break;
 443         default:
 444                 init_special_inode(inode, inode->i_mode, rdev);
 445                 break;
 446         }
 447         return;
 448
 449 make_bad:
 450         btrfs_release_path(root, path);
 451         btrfs_free_path(path);
 452         mutex_unlock(&root->fs_info->fs_mutex);
 453         make_bad_inode(inode);
 454 }
 455
 456 static void fill_inode_item(struct extent_buffer *leaf,
 457                             struct btrfs_inode_item *item,
 458                             struct inode *inode)
 459 {
 460         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 461         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 462         btrfs_set_inode_size(leaf, item, inode->i_size);
 463         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 464         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 465
 466         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 467                                inode->i_atime.tv_sec);
 468         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 469                                 inode->i_atime.tv_nsec);
 470
 471         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 472                                inode->i_mtime.tv_sec);
 473         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 474                                 inode->i_mtime.tv_nsec);
 475
 476         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 477                                inode->i_ctime.tv_sec);
 478         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 479                                 inode->i_ctime.tv_nsec);
 480
 481         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 482         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 483         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 484         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 485         btrfs_set_inode_block_group(leaf, item,
 486                                     BTRFS_I(inode)->block_group->key.objectid);
 487 }
 488
 489 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 490                               struct btrfs_root *root,
 491                               struct inode *inode)
 492 {
 493         struct btrfs_inode_item *inode_item;
 494         struct btrfs_path *path;
 495         struct extent_buffer *leaf;
 496         int ret;
 497
 498         path = btrfs_alloc_path();
 499         BUG_ON(!path);
 500         ret = btrfs_lookup_inode(trans, root, path,
 501                                  &BTRFS_I(inode)->location, 1);
 502         if (ret) {
 503                 if (ret > 0)
 504                         ret = -ENOENT;
 505                 goto failed;
 506         }
 507
 508         leaf = path->nodes[0];
 509         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 510                                   struct btrfs_inode_item);
 511
 512         fill_inode_item(leaf, inode_item, inode);
 513         btrfs_mark_buffer_dirty(leaf);
 514         btrfs_set_inode_last_trans(trans, inode);
 515         ret = 0;
 516 failed:
 517         btrfs_release_path(root, path);
 518         btrfs_free_path(path);
 519         return ret;
 520 }
 521
 522
 523 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 524                               struct btrfs_root *root,
 525                               struct inode *dir,
 526                               struct dentry *dentry)
 527 {
 528         struct btrfs_path *path;
 529         const char *name = dentry->d_name.name;
 530         int name_len = dentry->d_name.len;
 531         int ret = 0;
 532         struct extent_buffer *leaf;
 533         struct btrfs_dir_item *di;
 534         struct btrfs_key key;
 535
 536         path = btrfs_alloc_path();
 537         if (!path) {
 538                 ret = -ENOMEM;
 539                 goto err;
 540         }
 541
 542         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 543                                     name, name_len, -1);
 544         if (IS_ERR(di)) {
 545                 ret = PTR_ERR(di);
 546                 goto err;
 547         }
 548         if (!di) {
 549                 ret = -ENOENT;
 550                 goto err;
 551         }
 552         leaf = path->nodes[0];
 553         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 554         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 555         if (ret)
 556                 goto err;
 557         btrfs_release_path(root, path);
 558
 559         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 560                                          key.objectid, name, name_len, -1);
 561         if (IS_ERR(di)) {
 562                 ret = PTR_ERR(di);
 563                 goto err;
 564         }
 565         if (!di) {
 566                 ret = -ENOENT;
 567                 goto err;
 568         }
 569         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 570
 571         dentry->d_inode->i_ctime = dir->i_ctime;
 572         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 573                                   dentry->d_inode->i_ino,
 574                                   dentry->d_parent->d_inode->i_ino);
 575         if (ret) {
 576                 printk("failed to delete reference to %.*s, "
 577                        "inode %lu parent %lu\n", name_len, name,
 578                        dentry->d_inode->i_ino,
 579                        dentry->d_parent->d_inode->i_ino);
 580         }
 581 err:
 582         btrfs_free_path(path);
 583         if (!ret) {
 584                 dir->i_size -= name_len * 2;
 585                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 586                 btrfs_update_inode(trans, root, dir);
 587 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 588                 dentry->d_inode->i_nlink--;
 589 #else
 590                 drop_nlink(dentry->d_inode);
 591 #endif
 592                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 593                 dir->i_sb->s_dirt = 1;
 594         }
 595         return ret;
 596 }
 597
 598 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 599 {
 600         struct btrfs_root *root;
 601         struct btrfs_trans_handle *trans;
 602         int ret;
 603         unsigned long nr = 0;
 604
 605         root = BTRFS_I(dir)->root;
 606         mutex_lock(&root->fs_info->fs_mutex);
 607
 608         ret = btrfs_check_free_space(root, 1, 1);
 609         if (ret)
 610                 goto fail;
 611
 612         trans = btrfs_start_transaction(root, 1);
 613
 614         btrfs_set_trans_block_group(trans, dir);
 615         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 616         nr = trans->blocks_used;
 617
 618         btrfs_end_transaction(trans, root);
 619 fail:
 620         mutex_unlock(&root->fs_info->fs_mutex);
 621         btrfs_btree_balance_dirty(root, nr);
 622         btrfs_throttle(root);
 623         return ret;
 624 }
 625
 626 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 627 {
 628         struct inode *inode = dentry->d_inode;
 629         int err = 0;
 630         int ret;
 631         struct btrfs_root *root = BTRFS_I(dir)->root;
 632         struct btrfs_trans_handle *trans;
 633         unsigned long nr = 0;
 634
 635         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 636                 return -ENOTEMPTY;
 637
 638         mutex_lock(&root->fs_info->fs_mutex);
 639         ret = btrfs_check_free_space(root, 1, 1);
 640         if (ret)
 641                 goto fail;
 642
 643         trans = btrfs_start_transaction(root, 1);
 644         btrfs_set_trans_block_group(trans, dir);
 645
 646         /* now the directory is empty */
 647         err = btrfs_unlink_trans(trans, root, dir, dentry);
 648         if (!err) {
 649                 inode->i_size = 0;
 650         }
 651
 652         nr = trans->blocks_used;
 653         ret = btrfs_end_transaction(trans, root);
 654 fail:
 655         mutex_unlock(&root->fs_info->fs_mutex);
 656         btrfs_btree_balance_dirty(root, nr);
 657         btrfs_throttle(root);
 658
 659         if (ret && !err)
 660                 err = ret;
 661         return err;
 662 }
 663
 664 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 665                             struct btrfs_root *root,
 666                             struct inode *inode)
 667 {
 668         struct btrfs_path *path;
 669         int ret;
 670
 671         clear_inode(inode);
 672
 673         path = btrfs_alloc_path();
 674         BUG_ON(!path);
 675         ret = btrfs_lookup_inode(trans, root, path,
 676                                  &BTRFS_I(inode)->location, -1);
 677         if (ret > 0)
 678                 ret = -ENOENT;
 679         if (!ret)
 680                 ret = btrfs_del_item(trans, root, path);
 681         btrfs_free_path(path);
 682         return ret;
 683 }
 684
 685 /*
 686  * this can truncate away extent items, csum items and directory items.
 687  * It starts at a high offset and removes keys until it can't find
 688  * any higher than i_size.
 689  *
 690  * csum items that cross the new i_size are truncated to the new size
 691  * as well.
 692  */
 693 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 694                                    struct btrfs_root *root,
 695                                    struct inode *inode)
 696 {
 697         int ret;
 698         struct btrfs_path *path;
 699         struct btrfs_key key;
 700         struct btrfs_key found_key;
 701         u32 found_type;
 702         struct extent_buffer *leaf;
 703         struct btrfs_file_extent_item *fi;
 704         u64 extent_start = 0;
 705         u64 extent_num_bytes = 0;
 706         u64 item_end = 0;
 707         u64 root_gen = 0;
 708         u64 root_owner = 0;
 709         int found_extent;
 710         int del_item;
 711         int extent_type = -1;
 712
 713         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 714         path = btrfs_alloc_path();
 715         path->reada = -1;
 716         BUG_ON(!path);
 717
 718         /* FIXME, add redo link to tree so we don't leak on crash */
 719         key.objectid = inode->i_ino;
 720         key.offset = (u64)-1;
 721         key.type = (u8)-1;
 722
 723         while(1) {
 724                 btrfs_init_path(path);
 725                 fi = NULL;
 726                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 727                 if (ret < 0) {
 728                         goto error;
 729                 }
 730                 if (ret > 0) {
 731                         BUG_ON(path->slots[0] == 0);
 732                         path->slots[0]--;
 733                 }
 734                 leaf = path->nodes[0];
 735                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 736                 found_type = btrfs_key_type(&found_key);
 737
 738                 if (found_key.objectid != inode->i_ino)
 739                         break;
 740
 741                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 742                     found_type != BTRFS_DIR_ITEM_KEY &&
 743                     found_type != BTRFS_DIR_INDEX_KEY &&
 744                     found_type != BTRFS_EXTENT_DATA_KEY)
 745                         break;
 746
 747                 item_end = found_key.offset;
 748                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 749                         fi = btrfs_item_ptr(leaf, path->slots[0],
 750                                             struct btrfs_file_extent_item);
 751                         extent_type = btrfs_file_extent_type(leaf, fi);
 752                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 753                                 item_end +=
 754                                     btrfs_file_extent_num_bytes(leaf, fi);
 755                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 756                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 757                                                                 path->slots[0]);
 758                                 item_end += btrfs_file_extent_inline_len(leaf,
 759                                                                          item);
 760                         }
 761                         item_end--;
 762                 }
 763                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 764                         ret = btrfs_csum_truncate(trans, root, path,
 765                                                   inode->i_size);
 766                         BUG_ON(ret);
 767                 }
 768                 if (item_end < inode->i_size) {
 769                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 770                                 found_type = BTRFS_INODE_ITEM_KEY;
 771                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 772                                 found_type = BTRFS_CSUM_ITEM_KEY;
 773                         } else if (found_type) {
 774                                 found_type--;
 775                         } else {
 776                                 break;
 777                         }
 778                         btrfs_set_key_type(&key, found_type);
 779                         btrfs_release_path(root, path);
 780                         continue;
 781                 }
 782                 if (found_key.offset >= inode->i_size)
 783                         del_item = 1;
 784                 else
 785                         del_item = 0;
 786                 found_extent = 0;
 787
 788                 /* FIXME, shrink the extent if the ref count is only 1 */
 789                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 790                         goto delete;
 791
 792                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 793                         u64 num_dec;
 794                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 795                         if (!del_item) {
 796                                 u64 orig_num_bytes =
 797                                         btrfs_file_extent_num_bytes(leaf, fi);
 798                                 extent_num_bytes = inode->i_size -
 799                                         found_key.offset + root->sectorsize - 1;
 800                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 801                                                          extent_num_bytes);
 802                                 num_dec = (orig_num_bytes -
 803                                            extent_num_bytes) >> 9;
 804                                 if (extent_start != 0) {
 805                                         inode->i_blocks -= num_dec;
 806                                 }
 807                                 btrfs_mark_buffer_dirty(leaf);
 808                         } else {
 809                                 extent_num_bytes =
 810                                         btrfs_file_extent_disk_num_bytes(leaf,
 811                                                                          fi);
 812                                 /* FIXME blocksize != 4096 */
 813                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 814                                                                        fi) >> 9;
 815                                 if (extent_start != 0) {
 816                                         found_extent = 1;
 817                                         inode->i_blocks -= num_dec;
 818                                 }
 819                                 root_gen = btrfs_header_generation(leaf);
 820                                 root_owner = btrfs_header_owner(leaf);
 821                         }
 822                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 823                            !del_item) {
 824                         u32 newsize = inode->i_size - found_key.offset;
 825                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 826                         ret = btrfs_truncate_item(trans, root, path,
 827                                                   newsize, 1);
 828                         BUG_ON(ret);
 829                 }
 830 delete:
 831                 if (del_item) {
 832                         ret = btrfs_del_item(trans, root, path);
 833                         if (ret)
 834                                 goto error;
 835                 } else {
 836                         break;
 837                 }
 838                 btrfs_release_path(root, path);
 839                 if (found_extent) {
 840                         ret = btrfs_free_extent(trans, root, extent_start,
 841                                                 extent_num_bytes,
 842                                                 root_owner,
 843                                                 root_gen, inode->i_ino,
 844                                                 found_key.offset, 0);
 845                         BUG_ON(ret);
 846                 }
 847         }
 848         ret = 0;
 849 error:
 850         btrfs_release_path(root, path);
 851         btrfs_free_path(path);
 852         inode->i_sb->s_dirt = 1;
 853         return ret;
 854 }
 855
 856 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 857                               size_t zero_start)
 858 {
 859         char *kaddr;
 860         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 861         struct btrfs_root *root = BTRFS_I(inode)->root;
 862         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 863         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 864         u64 existing_delalloc;
 865         u64 delalloc_start;
 866         int ret = 0;
 867
 868         WARN_ON(!PageLocked(page));
 869         set_page_extent_mapped(page);
 870
 871         lock_extent(em_tree, page_start, page_end, GFP_NOFS);
 872         delalloc_start = page_start;
 873         existing_delalloc = count_range_bits(&BTRFS_I(inode)->extent_tree,
 874                                              &delalloc_start, page_end,
 875                                              PAGE_CACHE_SIZE, EXTENT_DELALLOC);
 876         set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
 877                             page_end, GFP_NOFS);
 878
 879         spin_lock(&root->fs_info->delalloc_lock);
 880         root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE - existing_delalloc;
 881         spin_unlock(&root->fs_info->delalloc_lock);
 882
 883         if (zero_start != PAGE_CACHE_SIZE) {
 884                 kaddr = kmap(page);
 885                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 886                 flush_dcache_page(page);
 887                 kunmap(page);
 888         }
 889         set_page_dirty(page);
 890         unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
 891
 892         return ret;
 893 }
 894
 895 /*
 896  * taken from block_truncate_page, but does cow as it zeros out
 897  * any bytes left in the last page in the file.
 898  */
 899 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 900 {
 901         struct inode *inode = mapping->host;
 902         struct btrfs_root *root = BTRFS_I(inode)->root;
 903         u32 blocksize = root->sectorsize;
 904         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 905         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 906         struct page *page;
 907         int ret = 0;
 908         u64 page_start;
 909
 910         if ((offset & (blocksize - 1)) == 0)
 911                 goto out;
 912
 913         ret = -ENOMEM;
 914         page = grab_cache_page(mapping, index);
 915         if (!page)
 916                 goto out;
 917         if (!PageUptodate(page)) {
 918                 ret = btrfs_readpage(NULL, page);
 919                 lock_page(page);
 920                 if (!PageUptodate(page)) {
 921                         ret = -EIO;
 922                         goto out;
 923                 }
 924         }
 925         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 926
 927         ret = btrfs_cow_one_page(inode, page, offset);
 928
 929         unlock_page(page);
 930         page_cache_release(page);
 931 out:
 932         return ret;
 933 }
 934
 935 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 936 {
 937         struct inode *inode = dentry->d_inode;
 938         int err;
 939
 940         err = inode_change_ok(inode, attr);
 941         if (err)
 942                 return err;
 943
 944         if (S_ISREG(inode->i_mode) &&
 945             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 946                 struct btrfs_trans_handle *trans;
 947                 struct btrfs_root *root = BTRFS_I(inode)->root;
 948                 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 949
 950                 u64 mask = root->sectorsize - 1;
 951                 u64 pos = (inode->i_size + mask) & ~mask;
 952                 u64 block_end = attr->ia_size | mask;
 953                 u64 hole_size;
 954                 u64 alloc_hint = 0;
 955
 956                 if (attr->ia_size <= pos)
 957                         goto out;
 958
 959                 mutex_lock(&root->fs_info->fs_mutex);
 960                 err = btrfs_check_free_space(root, 1, 0);
 961                 mutex_unlock(&root->fs_info->fs_mutex);
 962                 if (err)
 963                         goto fail;
 964
 965                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 966
 967                 lock_extent(em_tree, pos, block_end, GFP_NOFS);
 968                 hole_size = (attr->ia_size - pos + mask) & ~mask;
 969
 970                 mutex_lock(&root->fs_info->fs_mutex);
 971                 trans = btrfs_start_transaction(root, 1);
 972                 btrfs_set_trans_block_group(trans, inode);
 973                 err = btrfs_drop_extents(trans, root, inode,
 974                                          pos, pos + hole_size, pos,
 975                                          &alloc_hint);
 976
 977                 if (alloc_hint != EXTENT_MAP_INLINE) {
 978                         err = btrfs_insert_file_extent(trans, root,
 979                                                        inode->i_ino,
 980                                                        pos, 0, 0, hole_size);
 981                 }
 982                 btrfs_end_transaction(trans, root);
 983                 mutex_unlock(&root->fs_info->fs_mutex);
 984                 unlock_extent(em_tree, pos, block_end, GFP_NOFS);
 985                 if (err)
 986                         return err;
 987         }
 988 out:
 989         err = inode_setattr(inode, attr);
 990 fail:
 991         return err;
 992 }
 993
 994 void btrfs_drop_inode(struct inode *inode)
 995 {
 996         if (!BTRFS_I(inode)->ordered_trans || inode->i_nlink) {
 997                 generic_drop_inode(inode);
 998                 return;
 999         }
1000         /* FIXME, make sure this delete actually ends up in the transaction */
1001         btrfs_del_ordered_inode(inode);
1002         generic_drop_inode(inode);
1003 }
1004
1005 void btrfs_delete_inode(struct inode *inode)
1006 {
1007         struct btrfs_trans_handle *trans;
1008         struct btrfs_root *root = BTRFS_I(inode)->root;
1009         unsigned long nr;
1010         int ret;
1011
1012         truncate_inode_pages(&inode->i_data, 0);
1013         if (is_bad_inode(inode)) {
1014                 goto no_delete;
1015         }
1016
1017         inode->i_size = 0;
1018         mutex_lock(&root->fs_info->fs_mutex);
1019         trans = btrfs_start_transaction(root, 1);
1020
1021         btrfs_set_trans_block_group(trans, inode);
1022         ret = btrfs_truncate_in_trans(trans, root, inode);
1023         if (ret)
1024                 goto no_delete_lock;
1025         ret = btrfs_delete_xattrs(trans, root, inode);
1026         if (ret)
1027                 goto no_delete_lock;
1028         ret = btrfs_free_inode(trans, root, inode);
1029         if (ret)
1030                 goto no_delete_lock;
1031         nr = trans->blocks_used;
1032
1033         btrfs_end_transaction(trans, root);
1034         mutex_unlock(&root->fs_info->fs_mutex);
1035         btrfs_btree_balance_dirty(root, nr);
1036         btrfs_throttle(root);
1037         return;
1038
1039 no_delete_lock:
1040         nr = trans->blocks_used;
1041         btrfs_end_transaction(trans, root);
1042         mutex_unlock(&root->fs_info->fs_mutex);
1043         btrfs_btree_balance_dirty(root, nr);
1044         btrfs_throttle(root);
1045 no_delete:
1046         clear_inode(inode);
1047 }
1048
1049 /*
1050  * this returns the key found in the dir entry in the location pointer.
1051  * If no dir entries were found, location->objectid is 0.
1052  */
1053 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1054                                struct btrfs_key *location)
1055 {
1056         const char *name = dentry->d_name.name;
1057         int namelen = dentry->d_name.len;
1058         struct btrfs_dir_item *di;
1059         struct btrfs_path *path;
1060         struct btrfs_root *root = BTRFS_I(dir)->root;
1061         int ret = 0;
1062
1063         if (namelen == 1 && strcmp(name, ".") == 0) {
1064                 location->objectid = dir->i_ino;
1065                 location->type = BTRFS_INODE_ITEM_KEY;
1066                 location->offset = 0;
1067                 return 0;
1068         }
1069         path = btrfs_alloc_path();
1070         BUG_ON(!path);
1071
1072         if (namelen == 2 && strcmp(name, "..") == 0) {
1073                 struct btrfs_key key;
1074                 struct extent_buffer *leaf;
1075                 u32 nritems;
1076                 int slot;
1077
1078                 key.objectid = dir->i_ino;
1079                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1080                 key.offset = 0;
1081                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1082                 BUG_ON(ret == 0);
1083                 ret = 0;
1084
1085                 leaf = path->nodes[0];
1086                 slot = path->slots[0];
1087                 nritems = btrfs_header_nritems(leaf);
1088                 if (slot >= nritems)
1089                         goto out_err;
1090
1091                 btrfs_item_key_to_cpu(leaf, &key, slot);
1092                 if (key.objectid != dir->i_ino ||
1093                     key.type != BTRFS_INODE_REF_KEY) {
1094                         goto out_err;
1095                 }
1096                 location->objectid = key.offset;
1097                 location->type = BTRFS_INODE_ITEM_KEY;
1098                 location->offset = 0;
1099                 goto out;
1100         }
1101
1102         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1103                                     namelen, 0);
1104         if (IS_ERR(di))
1105                 ret = PTR_ERR(di);
1106         if (!di || IS_ERR(di)) {
1107                 goto out_err;
1108         }
1109         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1110 out:
1111         btrfs_free_path(path);
1112         return ret;
1113 out_err:
1114         location->objectid = 0;
1115         goto out;
1116 }
1117
1118 /*
1119  * when we hit a tree root in a directory, the btrfs part of the inode
1120  * needs to be changed to reflect the root directory of the tree root.  This
1121  * is kind of like crossing a mount point.
1122  */
1123 static int fixup_tree_root_location(struct btrfs_root *root,
1124                              struct btrfs_key *location,
1125                              struct btrfs_root **sub_root,
1126                              struct dentry *dentry)
1127 {
1128         struct btrfs_path *path;
1129         struct btrfs_root_item *ri;
1130
1131         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1132                 return 0;
1133         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1134                 return 0;
1135
1136         path = btrfs_alloc_path();
1137         BUG_ON(!path);
1138         mutex_lock(&root->fs_info->fs_mutex);
1139
1140         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1141                                         dentry->d_name.name,
1142                                         dentry->d_name.len);
1143         if (IS_ERR(*sub_root))
1144                 return PTR_ERR(*sub_root);
1145
1146         ri = &(*sub_root)->root_item;
1147         location->objectid = btrfs_root_dirid(ri);
1148         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1149         location->offset = 0;
1150
1151         btrfs_free_path(path);
1152         mutex_unlock(&root->fs_info->fs_mutex);
1153         return 0;
1154 }
1155
1156 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1157 {
1158         struct btrfs_iget_args *args = p;
1159         inode->i_ino = args->ino;
1160         BTRFS_I(inode)->root = args->root;
1161         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1162                              inode->i_mapping, GFP_NOFS);
1163         return 0;
1164 }
1165
1166 static int btrfs_find_actor(struct inode *inode, void *opaque)
1167 {
1168         struct btrfs_iget_args *args = opaque;
1169         return (args->ino == inode->i_ino &&
1170                 args->root == BTRFS_I(inode)->root);
1171 }
1172
1173 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1174                             u64 root_objectid)
1175 {
1176         struct btrfs_iget_args args;
1177         args.ino = objectid;
1178         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1179
1180         if (!args.root)
1181                 return NULL;
1182
1183         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1184 }
1185
1186 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1187                                 struct btrfs_root *root)
1188 {
1189         struct inode *inode;
1190         struct btrfs_iget_args args;
1191         args.ino = objectid;
1192         args.root = root;
1193
1194         inode = iget5_locked(s, objectid, btrfs_find_actor,
1195                              btrfs_init_locked_inode,
1196                              (void *)&args);
1197         return inode;
1198 }
1199
1200 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1201                                    struct nameidata *nd)
1202 {
1203         struct inode * inode;
1204         struct btrfs_inode *bi = BTRFS_I(dir);
1205         struct btrfs_root *root = bi->root;
1206         struct btrfs_root *sub_root = root;
1207         struct btrfs_key location;
1208         int ret;
1209
1210         if (dentry->d_name.len > BTRFS_NAME_LEN)
1211                 return ERR_PTR(-ENAMETOOLONG);
1212
1213         mutex_lock(&root->fs_info->fs_mutex);
1214         ret = btrfs_inode_by_name(dir, dentry, &location);
1215         mutex_unlock(&root->fs_info->fs_mutex);
1216
1217         if (ret < 0)
1218                 return ERR_PTR(ret);
1219
1220         inode = NULL;
1221         if (location.objectid) {
1222                 ret = fixup_tree_root_location(root, &location, &sub_root,
1223                                                 dentry);
1224                 if (ret < 0)
1225                         return ERR_PTR(ret);
1226                 if (ret > 0)
1227                         return ERR_PTR(-ENOENT);
1228                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1229                                           sub_root);
1230                 if (!inode)
1231                         return ERR_PTR(-EACCES);
1232                 if (inode->i_state & I_NEW) {
1233                         /* the inode and parent dir are two different roots */
1234                         if (sub_root != root) {
1235                                 igrab(inode);
1236                                 sub_root->inode = inode;
1237                         }
1238                         BTRFS_I(inode)->root = sub_root;
1239                         memcpy(&BTRFS_I(inode)->location, &location,
1240                                sizeof(location));
1241                         btrfs_read_locked_inode(inode);
1242                         unlock_new_inode(inode);
1243                 }
1244         }
1245         return d_splice_alias(inode, dentry);
1246 }
1247
1248 static unsigned char btrfs_filetype_table[] = {
1249         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1250 };
1251
1252 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1253 {
1254         struct inode *inode = filp->f_dentry->d_inode;
1255         struct btrfs_root *root = BTRFS_I(inode)->root;
1256         struct btrfs_item *item;
1257         struct btrfs_dir_item *di;
1258         struct btrfs_key key;
1259         struct btrfs_key found_key;
1260         struct btrfs_path *path;
1261         int ret;
1262         u32 nritems;
1263         struct extent_buffer *leaf;
1264         int slot;
1265         int advance;
1266         unsigned char d_type;
1267         int over = 0;
1268         u32 di_cur;
1269         u32 di_total;
1270         u32 di_len;
1271         int key_type = BTRFS_DIR_INDEX_KEY;
1272         char tmp_name[32];
1273         char *name_ptr;
1274         int name_len;
1275
1276         /* FIXME, use a real flag for deciding about the key type */
1277         if (root->fs_info->tree_root == root)
1278                 key_type = BTRFS_DIR_ITEM_KEY;
1279
1280         /* special case for "." */
1281         if (filp->f_pos == 0) {
1282                 over = filldir(dirent, ".", 1,
1283                                1, inode->i_ino,
1284                                DT_DIR);
1285                 if (over)
1286                         return 0;
1287                 filp->f_pos = 1;
1288         }
1289
1290         mutex_lock(&root->fs_info->fs_mutex);
1291         key.objectid = inode->i_ino;
1292         path = btrfs_alloc_path();
1293         path->reada = 2;
1294
1295         /* special case for .., just use the back ref */
1296         if (filp->f_pos == 1) {
1297                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1298                 key.offset = 0;
1299                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1300                 BUG_ON(ret == 0);
1301                 leaf = path->nodes[0];
1302                 slot = path->slots[0];
1303                 nritems = btrfs_header_nritems(leaf);
1304                 if (slot >= nritems) {
1305                         btrfs_release_path(root, path);
1306                         goto read_dir_items;
1307                 }
1308                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1309                 btrfs_release_path(root, path);
1310                 if (found_key.objectid != key.objectid ||
1311                     found_key.type != BTRFS_INODE_REF_KEY)
1312                         goto read_dir_items;
1313                 over = filldir(dirent, "..", 2,
1314                                2, found_key.offset, DT_DIR);
1315                 if (over)
1316                         goto nopos;
1317                 filp->f_pos = 2;
1318         }
1319
1320 read_dir_items:
1321         btrfs_set_key_type(&key, key_type);
1322         key.offset = filp->f_pos;
1323
1324         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1325         if (ret < 0)
1326                 goto err;
1327         advance = 0;
1328         while(1) {
1329                 leaf = path->nodes[0];
1330                 nritems = btrfs_header_nritems(leaf);
1331                 slot = path->slots[0];
1332                 if (advance || slot >= nritems) {
1333                         if (slot >= nritems -1) {
1334                                 ret = btrfs_next_leaf(root, path);
1335                                 if (ret)
1336                                         break;
1337                                 leaf = path->nodes[0];
1338                                 nritems = btrfs_header_nritems(leaf);
1339                                 slot = path->slots[0];
1340                         } else {
1341                                 slot++;
1342                                 path->slots[0]++;
1343                         }
1344                 }
1345                 advance = 1;
1346                 item = btrfs_item_nr(leaf, slot);
1347                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1348
1349                 if (found_key.objectid != key.objectid)
1350                         break;
1351                 if (btrfs_key_type(&found_key) != key_type)
1352                         break;
1353                 if (found_key.offset < filp->f_pos)
1354                         continue;
1355
1356                 filp->f_pos = found_key.offset;
1357                 advance = 1;
1358                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1359                 di_cur = 0;
1360                 di_total = btrfs_item_size(leaf, item);
1361                 while(di_cur < di_total) {
1362                         struct btrfs_key location;
1363
1364                         name_len = btrfs_dir_name_len(leaf, di);
1365                         if (name_len < 32) {
1366                                 name_ptr = tmp_name;
1367                         } else {
1368                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1369                                 BUG_ON(!name_ptr);
1370                         }
1371                         read_extent_buffer(leaf, name_ptr,
1372                                            (unsigned long)(di + 1), name_len);
1373
1374                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1375                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1376                         over = filldir(dirent, name_ptr, name_len,
1377                                        found_key.offset,
1378                                        location.objectid,
1379                                        d_type);
1380
1381                         if (name_ptr != tmp_name)
1382                                 kfree(name_ptr);
1383
1384                         if (over)
1385                                 goto nopos;
1386                         di_len = btrfs_dir_name_len(leaf, di) +
1387                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1388                         di_cur += di_len;
1389                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1390                 }
1391         }
1392         filp->f_pos++;
1393 nopos:
1394         ret = 0;
1395 err:
1396         btrfs_release_path(root, path);
1397         btrfs_free_path(path);
1398         mutex_unlock(&root->fs_info->fs_mutex);
1399         return ret;
1400 }
1401
1402 int btrfs_write_inode(struct inode *inode, int wait)
1403 {
1404         struct btrfs_root *root = BTRFS_I(inode)->root;
1405         struct btrfs_trans_handle *trans;
1406         int ret = 0;
1407
1408         if (wait) {
1409                 mutex_lock(&root->fs_info->fs_mutex);
1410                 trans = btrfs_start_transaction(root, 1);
1411                 btrfs_set_trans_block_group(trans, inode);
1412                 ret = btrfs_commit_transaction(trans, root);
1413                 mutex_unlock(&root->fs_info->fs_mutex);
1414         }
1415         return ret;
1416 }
1417
1418 /*
1419  * This is somewhat expensive, updating the tree every time the
1420  * inode changes.  But, it is most likely to find the inode in cache.
1421  * FIXME, needs more benchmarking...there are no reasons other than performance
1422  * to keep or drop this code.
1423  */
1424 void btrfs_dirty_inode(struct inode *inode)
1425 {
1426         struct btrfs_root *root = BTRFS_I(inode)->root;
1427         struct btrfs_trans_handle *trans;
1428
1429         mutex_lock(&root->fs_info->fs_mutex);
1430         trans = btrfs_start_transaction(root, 1);
1431         btrfs_set_trans_block_group(trans, inode);
1432         btrfs_update_inode(trans, root, inode);
1433         btrfs_end_transaction(trans, root);
1434         mutex_unlock(&root->fs_info->fs_mutex);
1435 }
1436
1437 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1438                                      struct btrfs_root *root,
1439                                      u64 objectid,
1440                                      struct btrfs_block_group_cache *group,
1441                                      int mode)
1442 {
1443         struct inode *inode;
1444         struct btrfs_inode_item *inode_item;
1445         struct btrfs_key *location;
1446         struct btrfs_path *path;
1447         int ret;
1448         int owner;
1449
1450         path = btrfs_alloc_path();
1451         BUG_ON(!path);
1452
1453         inode = new_inode(root->fs_info->sb);
1454         if (!inode)
1455                 return ERR_PTR(-ENOMEM);
1456
1457         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1458                              inode->i_mapping, GFP_NOFS);
1459         BTRFS_I(inode)->root = root;
1460
1461         if (mode & S_IFDIR)
1462                 owner = 0;
1463         else
1464                 owner = 1;
1465         group = btrfs_find_block_group(root, group, 0, 0, owner);
1466         BTRFS_I(inode)->block_group = group;
1467         BTRFS_I(inode)->flags = 0;
1468         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1469         if (ret)
1470                 goto fail;
1471
1472         inode->i_uid = current->fsuid;
1473         inode->i_gid = current->fsgid;
1474         inode->i_mode = mode;
1475         inode->i_ino = objectid;
1476         inode->i_blocks = 0;
1477         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1478         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1479                                   struct btrfs_inode_item);
1480         fill_inode_item(path->nodes[0], inode_item, inode);
1481         btrfs_mark_buffer_dirty(path->nodes[0]);
1482         btrfs_free_path(path);
1483
1484         location = &BTRFS_I(inode)->location;
1485         location->objectid = objectid;
1486         location->offset = 0;
1487         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1488
1489         insert_inode_hash(inode);
1490         return inode;
1491 fail:
1492         btrfs_free_path(path);
1493         return ERR_PTR(ret);
1494 }
1495
1496 static inline u8 btrfs_inode_type(struct inode *inode)
1497 {
1498         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1499 }
1500
1501 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1502                             struct dentry *dentry, struct inode *inode)
1503 {
1504         int ret;
1505         struct btrfs_key key;
1506         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1507         struct inode *parent_inode;
1508
1509         key.objectid = inode->i_ino;
1510         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1511         key.offset = 0;
1512
1513         ret = btrfs_insert_dir_item(trans, root,
1514                                     dentry->d_name.name, dentry->d_name.len,
1515                                     dentry->d_parent->d_inode->i_ino,
1516                                     &key, btrfs_inode_type(inode));
1517         if (ret == 0) {
1518                 ret = btrfs_insert_inode_ref(trans, root,
1519                                      dentry->d_name.name,
1520                                      dentry->d_name.len,
1521                                      inode->i_ino,
1522                                      dentry->d_parent->d_inode->i_ino);
1523                 parent_inode = dentry->d_parent->d_inode;
1524                 parent_inode->i_size += dentry->d_name.len * 2;
1525                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1526                 ret = btrfs_update_inode(trans, root,
1527                                          dentry->d_parent->d_inode);
1528         }
1529         return ret;
1530 }
1531
1532 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1533                             struct dentry *dentry, struct inode *inode)
1534 {
1535         int err = btrfs_add_link(trans, dentry, inode);
1536         if (!err) {
1537                 d_instantiate(dentry, inode);
1538                 return 0;
1539         }
1540         if (err > 0)
1541                 err = -EEXIST;
1542         return err;
1543 }
1544
1545 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1546                         int mode, dev_t rdev)
1547 {
1548         struct btrfs_trans_handle *trans;
1549         struct btrfs_root *root = BTRFS_I(dir)->root;
1550         struct inode *inode = NULL;
1551         int err;
1552         int drop_inode = 0;
1553         u64 objectid;
1554         unsigned long nr = 0;
1555
1556         if (!new_valid_dev(rdev))
1557                 return -EINVAL;
1558
1559         mutex_lock(&root->fs_info->fs_mutex);
1560         err = btrfs_check_free_space(root, 1, 0);
1561         if (err)
1562                 goto fail;
1563
1564         trans = btrfs_start_transaction(root, 1);
1565         btrfs_set_trans_block_group(trans, dir);
1566
1567         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1568         if (err) {
1569                 err = -ENOSPC;
1570                 goto out_unlock;
1571         }
1572
1573         inode = btrfs_new_inode(trans, root, objectid,
1574                                 BTRFS_I(dir)->block_group, mode);
1575         err = PTR_ERR(inode);
1576         if (IS_ERR(inode))
1577                 goto out_unlock;
1578
1579         btrfs_set_trans_block_group(trans, inode);
1580         err = btrfs_add_nondir(trans, dentry, inode);
1581         if (err)
1582                 drop_inode = 1;
1583         else {
1584                 inode->i_op = &btrfs_special_inode_operations;
1585                 init_special_inode(inode, inode->i_mode, rdev);
1586                 btrfs_update_inode(trans, root, inode);
1587         }
1588         dir->i_sb->s_dirt = 1;
1589         btrfs_update_inode_block_group(trans, inode);
1590         btrfs_update_inode_block_group(trans, dir);
1591 out_unlock:
1592         nr = trans->blocks_used;
1593         btrfs_end_transaction(trans, root);
1594 fail:
1595         mutex_unlock(&root->fs_info->fs_mutex);
1596
1597         if (drop_inode) {
1598                 inode_dec_link_count(inode);
1599                 iput(inode);
1600         }
1601         btrfs_btree_balance_dirty(root, nr);
1602         btrfs_throttle(root);
1603         return err;
1604 }
1605
1606 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1607                         int mode, struct nameidata *nd)
1608 {
1609         struct btrfs_trans_handle *trans;
1610         struct btrfs_root *root = BTRFS_I(dir)->root;
1611         struct inode *inode = NULL;
1612         int err;
1613         int drop_inode = 0;
1614         unsigned long nr = 0;
1615         u64 objectid;
1616
1617         mutex_lock(&root->fs_info->fs_mutex);
1618         err = btrfs_check_free_space(root, 1, 0);
1619         if (err)
1620                 goto fail;
1621         trans = btrfs_start_transaction(root, 1);
1622         btrfs_set_trans_block_group(trans, dir);
1623
1624         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1625         if (err) {
1626                 err = -ENOSPC;
1627                 goto out_unlock;
1628         }
1629
1630         inode = btrfs_new_inode(trans, root, objectid,
1631                                 BTRFS_I(dir)->block_group, mode);
1632         err = PTR_ERR(inode);
1633         if (IS_ERR(inode))
1634                 goto out_unlock;
1635
1636         btrfs_set_trans_block_group(trans, inode);
1637         err = btrfs_add_nondir(trans, dentry, inode);
1638         if (err)
1639                 drop_inode = 1;
1640         else {
1641                 inode->i_mapping->a_ops = &btrfs_aops;
1642                 inode->i_fop = &btrfs_file_operations;
1643                 inode->i_op = &btrfs_file_inode_operations;
1644                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1645                                      inode->i_mapping, GFP_NOFS);
1646                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1647         }
1648         dir->i_sb->s_dirt = 1;
1649         btrfs_update_inode_block_group(trans, inode);
1650         btrfs_update_inode_block_group(trans, dir);
1651 out_unlock:
1652         nr = trans->blocks_used;
1653         btrfs_end_transaction(trans, root);
1654 fail:
1655         mutex_unlock(&root->fs_info->fs_mutex);
1656
1657         if (drop_inode) {
1658                 inode_dec_link_count(inode);
1659                 iput(inode);
1660         }
1661         btrfs_btree_balance_dirty(root, nr);
1662         btrfs_throttle(root);
1663         return err;
1664 }
1665
1666 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1667                       struct dentry *dentry)
1668 {
1669         struct btrfs_trans_handle *trans;
1670         struct btrfs_root *root = BTRFS_I(dir)->root;
1671         struct inode *inode = old_dentry->d_inode;
1672         unsigned long nr = 0;
1673         int err;
1674         int drop_inode = 0;
1675
1676         if (inode->i_nlink == 0)
1677                 return -ENOENT;
1678
1679 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1680         inode->i_nlink++;
1681 #else
1682         inc_nlink(inode);
1683 #endif
1684         mutex_lock(&root->fs_info->fs_mutex);
1685         err = btrfs_check_free_space(root, 1, 0);
1686         if (err)
1687                 goto fail;
1688         trans = btrfs_start_transaction(root, 1);
1689
1690         btrfs_set_trans_block_group(trans, dir);
1691         atomic_inc(&inode->i_count);
1692         err = btrfs_add_nondir(trans, dentry, inode);
1693
1694         if (err)
1695                 drop_inode = 1;
1696
1697         dir->i_sb->s_dirt = 1;
1698         btrfs_update_inode_block_group(trans, dir);
1699         err = btrfs_update_inode(trans, root, inode);
1700
1701         if (err)
1702                 drop_inode = 1;
1703
1704         nr = trans->blocks_used;
1705         btrfs_end_transaction(trans, root);
1706 fail:
1707         mutex_unlock(&root->fs_info->fs_mutex);
1708
1709         if (drop_inode) {
1710                 inode_dec_link_count(inode);
1711                 iput(inode);
1712         }
1713         btrfs_btree_balance_dirty(root, nr);
1714         btrfs_throttle(root);
1715         return err;
1716 }
1717
1718 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1719 {
1720         struct inode *inode;
1721         struct btrfs_trans_handle *trans;
1722         struct btrfs_root *root = BTRFS_I(dir)->root;
1723         int err = 0;
1724         int drop_on_err = 0;
1725         u64 objectid;
1726         unsigned long nr = 1;
1727
1728         mutex_lock(&root->fs_info->fs_mutex);
1729         err = btrfs_check_free_space(root, 1, 0);
1730         if (err)
1731                 goto out_unlock;
1732
1733         trans = btrfs_start_transaction(root, 1);
1734         btrfs_set_trans_block_group(trans, dir);
1735
1736         if (IS_ERR(trans)) {
1737                 err = PTR_ERR(trans);
1738                 goto out_unlock;
1739         }
1740
1741         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1742         if (err) {
1743                 err = -ENOSPC;
1744                 goto out_unlock;
1745         }
1746
1747         inode = btrfs_new_inode(trans, root, objectid,
1748                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1749         if (IS_ERR(inode)) {
1750                 err = PTR_ERR(inode);
1751                 goto out_fail;
1752         }
1753
1754         drop_on_err = 1;
1755         inode->i_op = &btrfs_dir_inode_operations;
1756         inode->i_fop = &btrfs_dir_file_operations;
1757         btrfs_set_trans_block_group(trans, inode);
1758
1759         inode->i_size = 0;
1760         err = btrfs_update_inode(trans, root, inode);
1761         if (err)
1762                 goto out_fail;
1763
1764         err = btrfs_add_link(trans, dentry, inode);
1765         if (err)
1766                 goto out_fail;
1767
1768         d_instantiate(dentry, inode);
1769         drop_on_err = 0;
1770         dir->i_sb->s_dirt = 1;
1771         btrfs_update_inode_block_group(trans, inode);
1772         btrfs_update_inode_block_group(trans, dir);
1773
1774 out_fail:
1775         nr = trans->blocks_used;
1776         btrfs_end_transaction(trans, root);
1777
1778 out_unlock:
1779         mutex_unlock(&root->fs_info->fs_mutex);
1780         if (drop_on_err)
1781                 iput(inode);
1782         btrfs_btree_balance_dirty(root, nr);
1783         btrfs_throttle(root);
1784         return err;
1785 }
1786
1787 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1788                                     size_t page_offset, u64 start, u64 end,
1789                                     int create)
1790 {
1791         int ret;
1792         int err = 0;
1793         u64 bytenr;
1794         u64 extent_start = 0;
1795         u64 extent_end = 0;
1796         u64 objectid = inode->i_ino;
1797         u32 found_type;
1798         int failed_insert = 0;
1799         struct btrfs_path *path;
1800         struct btrfs_root *root = BTRFS_I(inode)->root;
1801         struct btrfs_file_extent_item *item;
1802         struct extent_buffer *leaf;
1803         struct btrfs_key found_key;
1804         struct extent_map *em = NULL;
1805         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1806         struct btrfs_trans_handle *trans = NULL;
1807
1808         path = btrfs_alloc_path();
1809         BUG_ON(!path);
1810         mutex_lock(&root->fs_info->fs_mutex);
1811
1812 again:
1813         em = lookup_extent_mapping(em_tree, start, end);
1814         if (em) {
1815                 if (em->start > start) {
1816                         printk("get_extent start %Lu em start %Lu\n",
1817                                start, em->start);
1818                         WARN_ON(1);
1819                 }
1820                 goto out;
1821         }
1822         if (!em) {
1823                 em = alloc_extent_map(GFP_NOFS);
1824                 if (!em) {
1825                         err = -ENOMEM;
1826                         goto out;
1827                 }
1828                 em->start = EXTENT_MAP_HOLE;
1829                 em->end = EXTENT_MAP_HOLE;
1830         }
1831         em->bdev = inode->i_sb->s_bdev;
1832         ret = btrfs_lookup_file_extent(trans, root, path,
1833                                        objectid, start, trans != NULL);
1834         if (ret < 0) {
1835                 err = ret;
1836                 goto out;
1837         }
1838
1839         if (ret != 0) {
1840                 if (path->slots[0] == 0)
1841                         goto not_found;
1842                 path->slots[0]--;
1843         }
1844
1845         leaf = path->nodes[0];
1846         item = btrfs_item_ptr(leaf, path->slots[0],
1847                               struct btrfs_file_extent_item);
1848         /* are we inside the extent that was found? */
1849         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1850         found_type = btrfs_key_type(&found_key);
1851         if (found_key.objectid != objectid ||
1852             found_type != BTRFS_EXTENT_DATA_KEY) {
1853                 goto not_found;
1854         }
1855
1856         found_type = btrfs_file_extent_type(leaf, item);
1857         extent_start = found_key.offset;
1858         if (found_type == BTRFS_FILE_EXTENT_REG) {
1859                 extent_end = extent_start +
1860                        btrfs_file_extent_num_bytes(leaf, item);
1861                 err = 0;
1862                 if (start < extent_start || start >= extent_end) {
1863                         em->start = start;
1864                         if (start < extent_start) {
1865                                 if (end < extent_start)
1866                                         goto not_found;
1867                                 em->end = extent_end - 1;
1868                         } else {
1869                                 em->end = end;
1870                         }
1871                         goto not_found_em;
1872                 }
1873                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1874                 if (bytenr == 0) {
1875                         em->start = extent_start;
1876                         em->end = extent_end - 1;
1877                         em->block_start = EXTENT_MAP_HOLE;
1878                         em->block_end = EXTENT_MAP_HOLE;
1879                         goto insert;
1880                 }
1881                 bytenr += btrfs_file_extent_offset(leaf, item);
1882                 em->block_start = bytenr;
1883                 em->block_end = em->block_start +
1884                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1885                 em->start = extent_start;
1886                 em->end = extent_end - 1;
1887                 goto insert;
1888         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1889                 unsigned long ptr;
1890                 char *map;
1891                 size_t size;
1892                 size_t extent_offset;
1893                 size_t copy_size;
1894
1895                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1896                                                     path->slots[0]));
1897                 extent_end = (extent_start + size - 1) |
1898                         ((u64)root->sectorsize - 1);
1899                 if (start < extent_start || start >= extent_end) {
1900                         em->start = start;
1901                         if (start < extent_start) {
1902                                 if (end < extent_start)
1903                                         goto not_found;
1904                                 em->end = extent_end;
1905                         } else {
1906                                 em->end = end;
1907                         }
1908                         goto not_found_em;
1909                 }
1910                 em->block_start = EXTENT_MAP_INLINE;
1911                 em->block_end = EXTENT_MAP_INLINE;
1912
1913                 if (!page) {
1914                         em->start = extent_start;
1915                         em->end = extent_start + size - 1;
1916                         goto out;
1917                 }
1918
1919                 extent_offset = ((u64)page->index << PAGE_CACHE_SHIFT) -
1920                         extent_start + page_offset;
1921                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1922                                 size - extent_offset);
1923                 em->start = extent_start + extent_offset;
1924                 em->end = (em->start + copy_size -1) |
1925                         ((u64)root->sectorsize -1);
1926                 map = kmap(page);
1927                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1928                 if (create == 0 && !PageUptodate(page)) {
1929                         read_extent_buffer(leaf, map + page_offset, ptr,
1930                                            copy_size);
1931                         flush_dcache_page(page);
1932                 } else if (create && PageUptodate(page)) {
1933                         if (!trans) {
1934                                 kunmap(page);
1935                                 free_extent_map(em);
1936                                 em = NULL;
1937                                 btrfs_release_path(root, path);
1938                                 trans = btrfs_start_transaction(root, 1);
1939                                 goto again;
1940                         }
1941                         write_extent_buffer(leaf, map + page_offset, ptr,
1942                                             copy_size);
1943                         btrfs_mark_buffer_dirty(leaf);
1944                 }
1945                 kunmap(page);
1946                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1947                 goto insert;
1948         } else {
1949                 printk("unkknown found_type %d\n", found_type);
1950                 WARN_ON(1);
1951         }
1952 not_found:
1953         em->start = start;
1954         em->end = end;
1955 not_found_em:
1956         em->block_start = EXTENT_MAP_HOLE;
1957         em->block_end = EXTENT_MAP_HOLE;
1958 insert:
1959         btrfs_release_path(root, path);
1960         if (em->start > start || em->end < start) {
1961                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1962                 err = -EIO;
1963                 goto out;
1964         }
1965         ret = add_extent_mapping(em_tree, em);
1966         if (ret == -EEXIST) {
1967                 free_extent_map(em);
1968                 em = NULL;
1969                 if (0 && failed_insert == 1) {
1970                         btrfs_drop_extent_cache(inode, start, end);
1971                 }
1972                 failed_insert++;
1973                 if (failed_insert > 5) {
1974                         printk("failing to insert %Lu %Lu\n", start, end);
1975                         err = -EIO;
1976                         goto out;
1977                 }
1978                 goto again;
1979         }
1980         err = 0;
1981 out:
1982         btrfs_free_path(path);
1983         if (trans) {
1984                 ret = btrfs_end_transaction(trans, root);
1985                 if (!err)
1986                         err = ret;
1987         }
1988         mutex_unlock(&root->fs_info->fs_mutex);
1989         if (err) {
1990                 free_extent_map(em);
1991                 WARN_ON(1);
1992                 return ERR_PTR(err);
1993         }
1994         return em;
1995 }
1996
1997 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1998 {
1999         return extent_bmap(mapping, iblock, btrfs_get_extent);
2000 }
2001
2002 int btrfs_readpage(struct file *file, struct page *page)
2003 {
2004         struct extent_map_tree *tree;
2005         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2006         return extent_read_full_page(tree, page, btrfs_get_extent);
2007 }
2008
2009 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2010 {
2011         struct extent_map_tree *tree;
2012
2013
2014         if (current->flags & PF_MEMALLOC) {
2015                 redirty_page_for_writepage(wbc, page);
2016                 unlock_page(page);
2017                 return 0;
2018         }
2019         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2020         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2021 }
2022
2023 static int btrfs_writepages(struct address_space *mapping,
2024                             struct writeback_control *wbc)
2025 {
2026         struct extent_map_tree *tree;
2027         tree = &BTRFS_I(mapping->host)->extent_tree;
2028         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2029 }
2030
2031 static int
2032 btrfs_readpages(struct file *file, struct address_space *mapping,
2033                 struct list_head *pages, unsigned nr_pages)
2034 {
2035         struct extent_map_tree *tree;
2036         tree = &BTRFS_I(mapping->host)->extent_tree;
2037         return extent_readpages(tree, mapping, pages, nr_pages,
2038                                 btrfs_get_extent);
2039 }
2040
2041 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2042 {
2043         struct extent_map_tree *tree;
2044         int ret;
2045
2046         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2047         ret = try_release_extent_mapping(tree, page);
2048         if (ret == 1) {
2049                 ClearPagePrivate(page);
2050                 set_page_private(page, 0);
2051                 page_cache_release(page);
2052         }
2053         return ret;
2054 }
2055
2056 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2057 {
2058         struct extent_map_tree *tree;
2059
2060         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2061         extent_invalidatepage(tree, page, offset);
2062         btrfs_releasepage(page, GFP_NOFS);
2063 }
2064
2065 /*
2066  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2067  * called from a page fault handler when a page is first dirtied. Hence we must
2068  * be careful to check for EOF conditions here. We set the page up correctly
2069  * for a written page which means we get ENOSPC checking when writing into
2070  * holes and correct delalloc and unwritten extent mapping on filesystems that
2071  * support these features.
2072  *
2073  * We are not allowed to take the i_mutex here so we have to play games to
2074  * protect against truncate races as the page could now be beyond EOF.  Because
2075  * vmtruncate() writes the inode size before removing pages, once we have the
2076  * page lock we can determine safely if the page is beyond EOF. If it is not
2077  * beyond EOF, then the page is guaranteed safe against truncation until we
2078  * unlock the page.
2079  */
2080 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2081 {
2082         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2083         struct btrfs_root *root = BTRFS_I(inode)->root;
2084         unsigned long end;
2085         loff_t size;
2086         int ret;
2087         u64 page_start;
2088
2089         mutex_lock(&root->fs_info->fs_mutex);
2090         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2091         mutex_unlock(&root->fs_info->fs_mutex);
2092         if (ret)
2093                 goto out;
2094
2095         ret = -EINVAL;
2096
2097         lock_page(page);
2098         wait_on_page_writeback(page);
2099         size = i_size_read(inode);
2100         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2101
2102         if ((page->mapping != inode->i_mapping) ||
2103             (page_start > size)) {
2104                 /* page got truncated out from underneath us */
2105                 goto out_unlock;
2106         }
2107
2108         /* page is wholly or partially inside EOF */
2109         if (page_start + PAGE_CACHE_SIZE > size)
2110                 end = size & ~PAGE_CACHE_MASK;
2111         else
2112                 end = PAGE_CACHE_SIZE;
2113
2114         ret = btrfs_cow_one_page(inode, page, end);
2115
2116 out_unlock:
2117         unlock_page(page);
2118 out:
2119         return ret;
2120 }
2121
2122 static void btrfs_truncate(struct inode *inode)
2123 {
2124         struct btrfs_root *root = BTRFS_I(inode)->root;
2125         int ret;
2126         struct btrfs_trans_handle *trans;
2127         unsigned long nr;
2128
2129         if (!S_ISREG(inode->i_mode))
2130                 return;
2131         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2132                 return;
2133
2134         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2135
2136         mutex_lock(&root->fs_info->fs_mutex);
2137         trans = btrfs_start_transaction(root, 1);
2138         btrfs_set_trans_block_group(trans, inode);
2139
2140         /* FIXME, add redo link to tree so we don't leak on crash */
2141         ret = btrfs_truncate_in_trans(trans, root, inode);
2142         btrfs_update_inode(trans, root, inode);
2143         nr = trans->blocks_used;
2144
2145         ret = btrfs_end_transaction(trans, root);
2146         BUG_ON(ret);
2147         mutex_unlock(&root->fs_info->fs_mutex);
2148         btrfs_btree_balance_dirty(root, nr);
2149         btrfs_throttle(root);
2150 }
2151
2152 static int noinline create_subvol(struct btrfs_root *root, char *name,
2153                                   int namelen)
2154 {
2155         struct btrfs_trans_handle *trans;
2156         struct btrfs_key key;
2157         struct btrfs_root_item root_item;
2158         struct btrfs_inode_item *inode_item;
2159         struct extent_buffer *leaf;
2160         struct btrfs_root *new_root = root;
2161         struct inode *inode;
2162         struct inode *dir;
2163         int ret;
2164         int err;
2165         u64 objectid;
2166         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2167         unsigned long nr = 1;
2168
2169         mutex_lock(&root->fs_info->fs_mutex);
2170         ret = btrfs_check_free_space(root, 1, 0);
2171         if (ret)
2172                 goto fail_commit;
2173
2174         trans = btrfs_start_transaction(root, 1);
2175         BUG_ON(!trans);
2176
2177         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2178                                        0, &objectid);
2179         if (ret)
2180                 goto fail;
2181
2182         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2183                                         objectid, trans->transid, 0, 0,
2184                                         0, 0);
2185         if (IS_ERR(leaf))
2186                 return PTR_ERR(leaf);
2187
2188         btrfs_set_header_nritems(leaf, 0);
2189         btrfs_set_header_level(leaf, 0);
2190         btrfs_set_header_bytenr(leaf, leaf->start);
2191         btrfs_set_header_generation(leaf, trans->transid);
2192         btrfs_set_header_owner(leaf, objectid);
2193
2194         write_extent_buffer(leaf, root->fs_info->fsid,
2195                             (unsigned long)btrfs_header_fsid(leaf),
2196                             BTRFS_FSID_SIZE);
2197         btrfs_mark_buffer_dirty(leaf);
2198
2199         inode_item = &root_item.inode;
2200         memset(inode_item, 0, sizeof(*inode_item));
2201         inode_item->generation = cpu_to_le64(1);
2202         inode_item->size = cpu_to_le64(3);
2203         inode_item->nlink = cpu_to_le32(1);
2204         inode_item->nblocks = cpu_to_le64(1);
2205         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2206
2207         btrfs_set_root_bytenr(&root_item, leaf->start);
2208         btrfs_set_root_level(&root_item, 0);
2209         btrfs_set_root_refs(&root_item, 1);
2210         btrfs_set_root_used(&root_item, 0);
2211
2212         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2213         root_item.drop_level = 0;
2214
2215         free_extent_buffer(leaf);
2216         leaf = NULL;
2217
2218         btrfs_set_root_dirid(&root_item, new_dirid);
2219
2220         key.objectid = objectid;
2221         key.offset = 1;
2222         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2223         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2224                                 &root_item);
2225         if (ret)
2226                 goto fail;
2227
2228         /*
2229          * insert the directory item
2230          */
2231         key.offset = (u64)-1;
2232         dir = root->fs_info->sb->s_root->d_inode;
2233         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2234                                     name, namelen, dir->i_ino, &key,
2235                                     BTRFS_FT_DIR);
2236         if (ret)
2237                 goto fail;
2238
2239         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2240                              name, namelen, objectid,
2241                              root->fs_info->sb->s_root->d_inode->i_ino);
2242         if (ret)
2243                 goto fail;
2244
2245         ret = btrfs_commit_transaction(trans, root);
2246         if (ret)
2247                 goto fail_commit;
2248
2249         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2250         BUG_ON(!new_root);
2251
2252         trans = btrfs_start_transaction(new_root, 1);
2253         BUG_ON(!trans);
2254
2255         inode = btrfs_new_inode(trans, new_root, new_dirid,
2256                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2257         if (IS_ERR(inode))
2258                 goto fail;
2259         inode->i_op = &btrfs_dir_inode_operations;
2260         inode->i_fop = &btrfs_dir_file_operations;
2261         new_root->inode = inode;
2262
2263         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2264                                      new_dirid);
2265         inode->i_nlink = 1;
2266         inode->i_size = 0;
2267         ret = btrfs_update_inode(trans, new_root, inode);
2268         if (ret)
2269                 goto fail;
2270 fail:
2271         nr = trans->blocks_used;
2272         err = btrfs_commit_transaction(trans, new_root);
2273         if (err && !ret)
2274                 ret = err;
2275 fail_commit:
2276         mutex_unlock(&root->fs_info->fs_mutex);
2277         btrfs_btree_balance_dirty(root, nr);
2278         btrfs_throttle(root);
2279         return ret;
2280 }
2281
2282 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2283 {
2284         struct btrfs_pending_snapshot *pending_snapshot;
2285         struct btrfs_trans_handle *trans;
2286         int ret;
2287         int err;
2288         unsigned long nr = 0;
2289
2290         if (!root->ref_cows)
2291                 return -EINVAL;
2292
2293         mutex_lock(&root->fs_info->fs_mutex);
2294         ret = btrfs_check_free_space(root, 1, 0);
2295         if (ret)
2296                 goto fail_unlock;
2297
2298         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2299         if (!pending_snapshot) {
2300                 ret = -ENOMEM;
2301                 goto fail_unlock;
2302         }
2303         pending_snapshot->name = kstrndup(name, namelen, GFP_NOFS);
2304         if (!pending_snapshot->name) {
2305                 ret = -ENOMEM;
2306                 kfree(pending_snapshot);
2307                 goto fail_unlock;
2308         }
2309         trans = btrfs_start_transaction(root, 1);
2310         BUG_ON(!trans);
2311
2312         pending_snapshot->root = root;
2313         list_add(&pending_snapshot->list,
2314                  &trans->transaction->pending_snapshots);
2315         ret = btrfs_update_inode(trans, root, root->inode);
2316         err = btrfs_commit_transaction(trans, root);
2317
2318 fail_unlock:
2319         mutex_unlock(&root->fs_info->fs_mutex);
2320         btrfs_btree_balance_dirty(root, nr);
2321         btrfs_throttle(root);
2322         return ret;
2323 }
2324
2325 unsigned long btrfs_force_ra(struct address_space *mapping,
2326                               struct file_ra_state *ra, struct file *file,
2327                               pgoff_t offset, pgoff_t last_index)
2328 {
2329         pgoff_t req_size;
2330
2331 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2332         req_size = last_index - offset + 1;
2333         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2334         return offset;
2335 #else
2336         req_size = min(last_index - offset + 1, (pgoff_t)128);
2337         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2338         return offset + req_size;
2339 #endif
2340 }
2341
2342 int btrfs_defrag_file(struct file *file) {
2343         struct inode *inode = fdentry(file)->d_inode;
2344         struct btrfs_root *root = BTRFS_I(inode)->root;
2345         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2346         struct page *page;
2347         unsigned long last_index;
2348         unsigned long ra_index = 0;
2349         u64 page_start;
2350         u64 page_end;
2351         u64 delalloc_start;
2352         u64 existing_delalloc;
2353         unsigned long i;
2354         int ret;
2355
2356         mutex_lock(&root->fs_info->fs_mutex);
2357         ret = btrfs_check_free_space(root, inode->i_size, 0);
2358         mutex_unlock(&root->fs_info->fs_mutex);
2359         if (ret)
2360                 return -ENOSPC;
2361
2362         mutex_lock(&inode->i_mutex);
2363         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2364         for (i = 0; i <= last_index; i++) {
2365                 if (i == ra_index) {
2366                         ra_index = btrfs_force_ra(inode->i_mapping,
2367                                                   &file->f_ra,
2368                                                   file, ra_index, last_index);
2369                 }
2370                 page = grab_cache_page(inode->i_mapping, i);
2371                 if (!page)
2372                         goto out_unlock;
2373                 if (!PageUptodate(page)) {
2374                         btrfs_readpage(NULL, page);
2375                         lock_page(page);
2376                         if (!PageUptodate(page)) {
2377                                 unlock_page(page);
2378                                 page_cache_release(page);
2379                                 goto out_unlock;
2380                         }
2381                 }
2382                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2383                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2384
2385                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2386                 delalloc_start = page_start;
2387                 existing_delalloc =
2388                         count_range_bits(&BTRFS_I(inode)->extent_tree,
2389                                          &delalloc_start, page_end,
2390                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2391                 set_extent_delalloc(em_tree, page_start,
2392                                     page_end, GFP_NOFS);
2393
2394                 spin_lock(&root->fs_info->delalloc_lock);
2395                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2396                                                  existing_delalloc;
2397                 spin_unlock(&root->fs_info->delalloc_lock);
2398
2399                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2400                 set_page_dirty(page);
2401                 unlock_page(page);
2402                 page_cache_release(page);
2403                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2404         }
2405
2406 out_unlock:
2407         mutex_unlock(&inode->i_mutex);
2408         return 0;
2409 }
2410
2411 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2412 {
2413         u64 new_size;
2414         u64 old_size;
2415         struct btrfs_ioctl_vol_args *vol_args;
2416         struct btrfs_trans_handle *trans;
2417         char *sizestr;
2418         int ret = 0;
2419         int namelen;
2420         int mod = 0;
2421
2422         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2423
2424         if (!vol_args)
2425                 return -ENOMEM;
2426
2427         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2428                 ret = -EFAULT;
2429                 goto out;
2430         }
2431         namelen = strlen(vol_args->name);
2432         if (namelen > BTRFS_VOL_NAME_MAX) {
2433                 ret = -EINVAL;
2434                 goto out;
2435         }
2436
2437         sizestr = vol_args->name;
2438         if (!strcmp(sizestr, "max"))
2439                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2440         else {
2441                 if (sizestr[0] == '-') {
2442                         mod = -1;
2443                         sizestr++;
2444                 } else if (sizestr[0] == '+') {
2445                         mod = 1;
2446                         sizestr++;
2447                 }
2448                 new_size = btrfs_parse_size(sizestr);
2449                 if (new_size == 0) {
2450                         ret = -EINVAL;
2451                         goto out;
2452                 }
2453         }
2454
2455         mutex_lock(&root->fs_info->fs_mutex);
2456         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2457
2458         if (mod < 0) {
2459                 if (new_size > old_size) {
2460                         ret = -EINVAL;
2461                         goto out_unlock;
2462                 }
2463                 new_size = old_size - new_size;
2464         } else if (mod > 0) {
2465                 new_size = old_size + new_size;
2466         }
2467
2468         if (new_size < 256 * 1024 * 1024) {
2469                 ret = -EINVAL;
2470                 goto out_unlock;
2471         }
2472         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2473                 ret = -EFBIG;
2474                 goto out_unlock;
2475         }
2476
2477         do_div(new_size, root->sectorsize);
2478         new_size *= root->sectorsize;
2479
2480 printk("new size is %Lu\n", new_size);
2481         if (new_size > old_size) {
2482                 trans = btrfs_start_transaction(root, 1);
2483                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2484                 btrfs_commit_transaction(trans, root);
2485         } else {
2486                 ret = btrfs_shrink_extent_tree(root, new_size);
2487         }
2488
2489 out_unlock:
2490         mutex_unlock(&root->fs_info->fs_mutex);
2491 out:
2492         kfree(vol_args);
2493         return ret;
2494 }
2495
2496 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2497                                             void __user *arg)
2498 {
2499         struct btrfs_ioctl_vol_args *vol_args;
2500         struct btrfs_dir_item *di;
2501         struct btrfs_path *path;
2502         u64 root_dirid;
2503         int namelen;
2504         int ret;
2505
2506         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2507
2508         if (!vol_args)
2509                 return -ENOMEM;
2510
2511         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2512                 ret = -EFAULT;
2513                 goto out;
2514         }
2515
2516         namelen = strlen(vol_args->name);
2517         if (namelen > BTRFS_VOL_NAME_MAX) {
2518                 ret = -EINVAL;
2519                 goto out;
2520         }
2521         if (strchr(vol_args->name, '/')) {
2522                 ret = -EINVAL;
2523                 goto out;
2524         }
2525
2526         path = btrfs_alloc_path();
2527         if (!path) {
2528                 ret = -ENOMEM;
2529                 goto out;
2530         }
2531
2532         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2533         mutex_lock(&root->fs_info->fs_mutex);
2534         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2535                             path, root_dirid,
2536                             vol_args->name, namelen, 0);
2537         mutex_unlock(&root->fs_info->fs_mutex);
2538         btrfs_free_path(path);
2539
2540         if (di && !IS_ERR(di)) {
2541                 ret = -EEXIST;
2542                 goto out;
2543         }
2544
2545         if (IS_ERR(di)) {
2546                 ret = PTR_ERR(di);
2547                 goto out;
2548         }
2549
2550         if (root == root->fs_info->tree_root)
2551                 ret = create_subvol(root, vol_args->name, namelen);
2552         else
2553                 ret = create_snapshot(root, vol_args->name, namelen);
2554 out:
2555         kfree(vol_args);
2556         return ret;
2557 }
2558
2559 static int btrfs_ioctl_defrag(struct file *file)
2560 {
2561         struct inode *inode = fdentry(file)->d_inode;
2562         struct btrfs_root *root = BTRFS_I(inode)->root;
2563
2564         switch (inode->i_mode & S_IFMT) {
2565         case S_IFDIR:
2566                 mutex_lock(&root->fs_info->fs_mutex);
2567                 btrfs_defrag_root(root, 0);
2568                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2569                 mutex_unlock(&root->fs_info->fs_mutex);
2570                 break;
2571         case S_IFREG:
2572                 btrfs_defrag_file(file);
2573                 break;
2574         }
2575
2576         return 0;
2577 }
2578
2579 long btrfs_ioctl(struct file *file, unsigned int
2580                 cmd, unsigned long arg)
2581 {
2582         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2583
2584         switch (cmd) {
2585         case BTRFS_IOC_SNAP_CREATE:
2586                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2587         case BTRFS_IOC_DEFRAG:
2588                 return btrfs_ioctl_defrag(file);
2589         case BTRFS_IOC_RESIZE:
2590                 return btrfs_ioctl_resize(root, (void __user *)arg);
2591         }
2592
2593         return -ENOTTY;
2594 }
2595
2596 /*
2597  * Called inside transaction, so use GFP_NOFS
2598  */
2599 struct inode *btrfs_alloc_inode(struct super_block *sb)
2600 {
2601         struct btrfs_inode *ei;
2602
2603         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2604         if (!ei)
2605                 return NULL;
2606         ei->last_trans = 0;
2607         ei->ordered_trans = 0;
2608         return &ei->vfs_inode;
2609 }
2610
2611 void btrfs_destroy_inode(struct inode *inode)
2612 {
2613         WARN_ON(!list_empty(&inode->i_dentry));
2614         WARN_ON(inode->i_data.nrpages);
2615
2616         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2617         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2618 }
2619
2620 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2621 static void init_once(struct kmem_cache * cachep, void *foo)
2622 #else
2623 static void init_once(void * foo, struct kmem_cache * cachep,
2624                       unsigned long flags)
2625 #endif
2626 {
2627         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2628
2629         inode_init_once(&ei->vfs_inode);
2630 }
2631
2632 void btrfs_destroy_cachep(void)
2633 {
2634         if (btrfs_inode_cachep)
2635                 kmem_cache_destroy(btrfs_inode_cachep);
2636         if (btrfs_trans_handle_cachep)
2637                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2638         if (btrfs_transaction_cachep)
2639                 kmem_cache_destroy(btrfs_transaction_cachep);
2640         if (btrfs_bit_radix_cachep)
2641                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2642         if (btrfs_path_cachep)
2643                 kmem_cache_destroy(btrfs_path_cachep);
2644 }
2645
2646 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2647                                        unsigned long extra_flags,
2648 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2649                                        void (*ctor)(struct kmem_cache *, void *)
2650 #else
2651                                        void (*ctor)(void *, struct kmem_cache *,
2652                                                     unsigned long)
2653 #endif
2654                                      )
2655 {
2656         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2657                                  SLAB_MEM_SPREAD | extra_flags), ctor
2658 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2659                                  ,NULL
2660 #endif
2661                                 );
2662 }
2663
2664 int btrfs_init_cachep(void)
2665 {
2666         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2667                                           sizeof(struct btrfs_inode),
2668                                           0, init_once);
2669         if (!btrfs_inode_cachep)
2670                 goto fail;
2671         btrfs_trans_handle_cachep =
2672                         btrfs_cache_create("btrfs_trans_handle_cache",
2673                                            sizeof(struct btrfs_trans_handle),
2674                                            0, NULL);
2675         if (!btrfs_trans_handle_cachep)
2676                 goto fail;
2677         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2678                                              sizeof(struct btrfs_transaction),
2679                                              0, NULL);
2680         if (!btrfs_transaction_cachep)
2681                 goto fail;
2682         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2683                                          sizeof(struct btrfs_path),
2684                                          0, NULL);
2685         if (!btrfs_path_cachep)
2686                 goto fail;
2687         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2688                                               SLAB_DESTROY_BY_RCU, NULL);
2689         if (!btrfs_bit_radix_cachep)
2690                 goto fail;
2691         return 0;
2692 fail:
2693         btrfs_destroy_cachep();
2694         return -ENOMEM;
2695 }
2696
2697 static int btrfs_getattr(struct vfsmount *mnt,
2698                          struct dentry *dentry, struct kstat *stat)
2699 {
2700         struct inode *inode = dentry->d_inode;
2701         generic_fillattr(inode, stat);
2702         stat->blksize = PAGE_CACHE_SIZE;
2703         return 0;
2704 }
2705
2706 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2707                            struct inode * new_dir,struct dentry *new_dentry)
2708 {
2709         struct btrfs_trans_handle *trans;
2710         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2711         struct inode *new_inode = new_dentry->d_inode;
2712         struct inode *old_inode = old_dentry->d_inode;
2713         struct timespec ctime = CURRENT_TIME;
2714         struct btrfs_path *path;
2715         int ret;
2716
2717         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2718             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2719                 return -ENOTEMPTY;
2720         }
2721
2722         mutex_lock(&root->fs_info->fs_mutex);
2723         ret = btrfs_check_free_space(root, 1, 0);
2724         if (ret)
2725                 goto out_unlock;
2726
2727         trans = btrfs_start_transaction(root, 1);
2728
2729         btrfs_set_trans_block_group(trans, new_dir);
2730         path = btrfs_alloc_path();
2731         if (!path) {
2732                 ret = -ENOMEM;
2733                 goto out_fail;
2734         }
2735
2736         old_dentry->d_inode->i_nlink++;
2737         old_dir->i_ctime = old_dir->i_mtime = ctime;
2738         new_dir->i_ctime = new_dir->i_mtime = ctime;
2739         old_inode->i_ctime = ctime;
2740
2741         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2742         if (ret)
2743                 goto out_fail;
2744
2745         if (new_inode) {
2746                 new_inode->i_ctime = CURRENT_TIME;
2747                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2748                 if (ret)
2749                         goto out_fail;
2750         }
2751         ret = btrfs_add_link(trans, new_dentry, old_inode);
2752         if (ret)
2753                 goto out_fail;
2754
2755 out_fail:
2756         btrfs_free_path(path);
2757         btrfs_end_transaction(trans, root);
2758 out_unlock:
2759         mutex_unlock(&root->fs_info->fs_mutex);
2760         return ret;
2761 }
2762
2763 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2764                          const char *symname)
2765 {
2766         struct btrfs_trans_handle *trans;
2767         struct btrfs_root *root = BTRFS_I(dir)->root;
2768         struct btrfs_path *path;
2769         struct btrfs_key key;
2770         struct inode *inode = NULL;
2771         int err;
2772         int drop_inode = 0;
2773         u64 objectid;
2774         int name_len;
2775         int datasize;
2776         unsigned long ptr;
2777         struct btrfs_file_extent_item *ei;
2778         struct extent_buffer *leaf;
2779         unsigned long nr = 0;
2780
2781         name_len = strlen(symname) + 1;
2782         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2783                 return -ENAMETOOLONG;
2784
2785         mutex_lock(&root->fs_info->fs_mutex);
2786         err = btrfs_check_free_space(root, 1, 0);
2787         if (err)
2788                 goto out_fail;
2789
2790         trans = btrfs_start_transaction(root, 1);
2791         btrfs_set_trans_block_group(trans, dir);
2792
2793         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2794         if (err) {
2795                 err = -ENOSPC;
2796                 goto out_unlock;
2797         }
2798
2799         inode = btrfs_new_inode(trans, root, objectid,
2800                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2801         err = PTR_ERR(inode);
2802         if (IS_ERR(inode))
2803                 goto out_unlock;
2804
2805         btrfs_set_trans_block_group(trans, inode);
2806         err = btrfs_add_nondir(trans, dentry, inode);
2807         if (err)
2808                 drop_inode = 1;
2809         else {
2810                 inode->i_mapping->a_ops = &btrfs_aops;
2811                 inode->i_fop = &btrfs_file_operations;
2812                 inode->i_op = &btrfs_file_inode_operations;
2813                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2814                                      inode->i_mapping, GFP_NOFS);
2815                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2816         }
2817         dir->i_sb->s_dirt = 1;
2818         btrfs_update_inode_block_group(trans, inode);
2819         btrfs_update_inode_block_group(trans, dir);
2820         if (drop_inode)
2821                 goto out_unlock;
2822
2823         path = btrfs_alloc_path();
2824         BUG_ON(!path);
2825         key.objectid = inode->i_ino;
2826         key.offset = 0;
2827         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2828         datasize = btrfs_file_extent_calc_inline_size(name_len);
2829         err = btrfs_insert_empty_item(trans, root, path, &key,
2830                                       datasize);
2831         if (err) {
2832                 drop_inode = 1;
2833                 goto out_unlock;
2834         }
2835         leaf = path->nodes[0];
2836         ei = btrfs_item_ptr(leaf, path->slots[0],
2837                             struct btrfs_file_extent_item);
2838         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2839         btrfs_set_file_extent_type(leaf, ei,
2840                                    BTRFS_FILE_EXTENT_INLINE);
2841         ptr = btrfs_file_extent_inline_start(ei);
2842         write_extent_buffer(leaf, symname, ptr, name_len);
2843         btrfs_mark_buffer_dirty(leaf);
2844         btrfs_free_path(path);
2845
2846         inode->i_op = &btrfs_symlink_inode_operations;
2847         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2848         inode->i_size = name_len - 1;
2849         err = btrfs_update_inode(trans, root, inode);
2850         if (err)
2851                 drop_inode = 1;
2852
2853 out_unlock:
2854         nr = trans->blocks_used;
2855         btrfs_end_transaction(trans, root);
2856 out_fail:
2857         mutex_unlock(&root->fs_info->fs_mutex);
2858         if (drop_inode) {
2859                 inode_dec_link_count(inode);
2860                 iput(inode);
2861         }
2862         btrfs_btree_balance_dirty(root, nr);
2863         btrfs_throttle(root);
2864         return err;
2865 }
2866 static int btrfs_permission(struct inode *inode, int mask,
2867                             struct nameidata *nd)
2868 {
2869         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2870                 return -EACCES;
2871         return generic_permission(inode, mask, NULL);
2872 }
2873
2874 static struct inode_operations btrfs_dir_inode_operations = {
2875         .lookup         = btrfs_lookup,
2876         .create         = btrfs_create,
2877         .unlink         = btrfs_unlink,
2878         .link           = btrfs_link,
2879         .mkdir          = btrfs_mkdir,
2880         .rmdir          = btrfs_rmdir,
2881         .rename         = btrfs_rename,
2882         .symlink        = btrfs_symlink,
2883         .setattr        = btrfs_setattr,
2884         .mknod          = btrfs_mknod,
2885         .setxattr       = generic_setxattr,
2886         .getxattr       = generic_getxattr,
2887         .listxattr      = btrfs_listxattr,
2888         .removexattr    = generic_removexattr,
2889         .permission     = btrfs_permission,
2890 };
2891 static struct inode_operations btrfs_dir_ro_inode_operations = {
2892         .lookup         = btrfs_lookup,
2893         .permission     = btrfs_permission,
2894 };
2895 static struct file_operations btrfs_dir_file_operations = {
2896         .llseek         = generic_file_llseek,
2897         .read           = generic_read_dir,
2898         .readdir        = btrfs_readdir,
2899         .unlocked_ioctl = btrfs_ioctl,
2900 #ifdef CONFIG_COMPAT
2901         .compat_ioctl   = btrfs_ioctl,
2902 #endif
2903 };
2904
2905 static struct extent_map_ops btrfs_extent_map_ops = {
2906         .fill_delalloc = run_delalloc_range,
2907         .writepage_io_hook = btrfs_writepage_io_hook,
2908         .readpage_io_hook = btrfs_readpage_io_hook,
2909         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2910 };
2911
2912 static struct address_space_operations btrfs_aops = {
2913         .readpage       = btrfs_readpage,
2914         .writepage      = btrfs_writepage,
2915         .writepages     = btrfs_writepages,
2916         .readpages      = btrfs_readpages,
2917         .sync_page      = block_sync_page,
2918         .bmap           = btrfs_bmap,
2919         .invalidatepage = btrfs_invalidatepage,
2920         .releasepage    = btrfs_releasepage,
2921         .set_page_dirty = __set_page_dirty_nobuffers,
2922 };
2923
2924 static struct address_space_operations btrfs_symlink_aops = {
2925         .readpage       = btrfs_readpage,
2926         .writepage      = btrfs_writepage,
2927         .invalidatepage = btrfs_invalidatepage,
2928         .releasepage    = btrfs_releasepage,
2929 };
2930
2931 static struct inode_operations btrfs_file_inode_operations = {
2932         .truncate       = btrfs_truncate,
2933         .getattr        = btrfs_getattr,
2934         .setattr        = btrfs_setattr,
2935         .setxattr       = generic_setxattr,
2936         .getxattr       = generic_getxattr,
2937         .listxattr      = btrfs_listxattr,
2938         .removexattr    = generic_removexattr,
2939         .permission     = btrfs_permission,
2940 };
2941 static struct inode_operations btrfs_special_inode_operations = {
2942         .getattr        = btrfs_getattr,
2943         .setattr        = btrfs_setattr,
2944         .permission     = btrfs_permission,
2945 };
2946 static struct inode_operations btrfs_symlink_inode_operations = {
2947         .readlink       = generic_readlink,
2948         .follow_link    = page_follow_link_light,
2949         .put_link       = page_put_link,
2950         .permission     = btrfs_permission,
2951 };