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         struct btrfs_trans_handle *trans;
 997         struct btrfs_root *root = BTRFS_I(inode)->root;
 998
 999         if (!BTRFS_I(inode)->ordered_trans) {
1000                 generic_drop_inode(inode);
1001                 return;
1002         }
1003         /* nasty, but it prevents a deadlock with data=ordered by preventing
1004          * a commit until after this inode is done
1005          */
1006         trans = btrfs_start_transaction(root, 1);
1007         generic_drop_inode(inode);
1008         /* note, the inode is now untouchable */
1009         btrfs_end_transaction(trans, root);
1010 }
1011
1012 void btrfs_delete_inode(struct inode *inode)
1013 {
1014         struct btrfs_trans_handle *trans;
1015         struct btrfs_root *root = BTRFS_I(inode)->root;
1016         unsigned long nr;
1017         int ret;
1018
1019         truncate_inode_pages(&inode->i_data, 0);
1020         if (is_bad_inode(inode)) {
1021                 goto no_delete;
1022         }
1023
1024         inode->i_size = 0;
1025         mutex_lock(&root->fs_info->fs_mutex);
1026         trans = btrfs_start_transaction(root, 1);
1027
1028         btrfs_set_trans_block_group(trans, inode);
1029         ret = btrfs_truncate_in_trans(trans, root, inode);
1030         if (ret)
1031                 goto no_delete_lock;
1032         ret = btrfs_delete_xattrs(trans, root, inode);
1033         if (ret)
1034                 goto no_delete_lock;
1035         ret = btrfs_free_inode(trans, root, inode);
1036         if (ret)
1037                 goto no_delete_lock;
1038         nr = trans->blocks_used;
1039
1040         btrfs_end_transaction(trans, root);
1041         mutex_unlock(&root->fs_info->fs_mutex);
1042         btrfs_btree_balance_dirty(root, nr);
1043         btrfs_throttle(root);
1044         return;
1045
1046 no_delete_lock:
1047         nr = trans->blocks_used;
1048         btrfs_end_transaction(trans, root);
1049         mutex_unlock(&root->fs_info->fs_mutex);
1050         btrfs_btree_balance_dirty(root, nr);
1051         btrfs_throttle(root);
1052 no_delete:
1053         clear_inode(inode);
1054 }
1055
1056 /*
1057  * this returns the key found in the dir entry in the location pointer.
1058  * If no dir entries were found, location->objectid is 0.
1059  */
1060 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1061                                struct btrfs_key *location)
1062 {
1063         const char *name = dentry->d_name.name;
1064         int namelen = dentry->d_name.len;
1065         struct btrfs_dir_item *di;
1066         struct btrfs_path *path;
1067         struct btrfs_root *root = BTRFS_I(dir)->root;
1068         int ret = 0;
1069
1070         if (namelen == 1 && strcmp(name, ".") == 0) {
1071                 location->objectid = dir->i_ino;
1072                 location->type = BTRFS_INODE_ITEM_KEY;
1073                 location->offset = 0;
1074                 return 0;
1075         }
1076         path = btrfs_alloc_path();
1077         BUG_ON(!path);
1078
1079         if (namelen == 2 && strcmp(name, "..") == 0) {
1080                 struct btrfs_key key;
1081                 struct extent_buffer *leaf;
1082                 u32 nritems;
1083                 int slot;
1084
1085                 key.objectid = dir->i_ino;
1086                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1087                 key.offset = 0;
1088                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1089                 BUG_ON(ret == 0);
1090                 ret = 0;
1091
1092                 leaf = path->nodes[0];
1093                 slot = path->slots[0];
1094                 nritems = btrfs_header_nritems(leaf);
1095                 if (slot >= nritems)
1096                         goto out_err;
1097
1098                 btrfs_item_key_to_cpu(leaf, &key, slot);
1099                 if (key.objectid != dir->i_ino ||
1100                     key.type != BTRFS_INODE_REF_KEY) {
1101                         goto out_err;
1102                 }
1103                 location->objectid = key.offset;
1104                 location->type = BTRFS_INODE_ITEM_KEY;
1105                 location->offset = 0;
1106                 goto out;
1107         }
1108
1109         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1110                                     namelen, 0);
1111         if (IS_ERR(di))
1112                 ret = PTR_ERR(di);
1113         if (!di || IS_ERR(di)) {
1114                 goto out_err;
1115         }
1116         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1117 out:
1118         btrfs_free_path(path);
1119         return ret;
1120 out_err:
1121         location->objectid = 0;
1122         goto out;
1123 }
1124
1125 /*
1126  * when we hit a tree root in a directory, the btrfs part of the inode
1127  * needs to be changed to reflect the root directory of the tree root.  This
1128  * is kind of like crossing a mount point.
1129  */
1130 static int fixup_tree_root_location(struct btrfs_root *root,
1131                              struct btrfs_key *location,
1132                              struct btrfs_root **sub_root,
1133                              struct dentry *dentry)
1134 {
1135         struct btrfs_path *path;
1136         struct btrfs_root_item *ri;
1137
1138         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1139                 return 0;
1140         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1141                 return 0;
1142
1143         path = btrfs_alloc_path();
1144         BUG_ON(!path);
1145         mutex_lock(&root->fs_info->fs_mutex);
1146
1147         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1148                                         dentry->d_name.name,
1149                                         dentry->d_name.len);
1150         if (IS_ERR(*sub_root))
1151                 return PTR_ERR(*sub_root);
1152
1153         ri = &(*sub_root)->root_item;
1154         location->objectid = btrfs_root_dirid(ri);
1155         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1156         location->offset = 0;
1157
1158         btrfs_free_path(path);
1159         mutex_unlock(&root->fs_info->fs_mutex);
1160         return 0;
1161 }
1162
1163 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1164 {
1165         struct btrfs_iget_args *args = p;
1166         inode->i_ino = args->ino;
1167         BTRFS_I(inode)->root = args->root;
1168         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1169                              inode->i_mapping, GFP_NOFS);
1170         return 0;
1171 }
1172
1173 static int btrfs_find_actor(struct inode *inode, void *opaque)
1174 {
1175         struct btrfs_iget_args *args = opaque;
1176         return (args->ino == inode->i_ino &&
1177                 args->root == BTRFS_I(inode)->root);
1178 }
1179
1180 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1181                             u64 root_objectid)
1182 {
1183         struct btrfs_iget_args args;
1184         args.ino = objectid;
1185         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1186
1187         if (!args.root)
1188                 return NULL;
1189
1190         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1191 }
1192
1193 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1194                                 struct btrfs_root *root)
1195 {
1196         struct inode *inode;
1197         struct btrfs_iget_args args;
1198         args.ino = objectid;
1199         args.root = root;
1200
1201         inode = iget5_locked(s, objectid, btrfs_find_actor,
1202                              btrfs_init_locked_inode,
1203                              (void *)&args);
1204         return inode;
1205 }
1206
1207 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1208                                    struct nameidata *nd)
1209 {
1210         struct inode * inode;
1211         struct btrfs_inode *bi = BTRFS_I(dir);
1212         struct btrfs_root *root = bi->root;
1213         struct btrfs_root *sub_root = root;
1214         struct btrfs_key location;
1215         int ret;
1216
1217         if (dentry->d_name.len > BTRFS_NAME_LEN)
1218                 return ERR_PTR(-ENAMETOOLONG);
1219
1220         mutex_lock(&root->fs_info->fs_mutex);
1221         ret = btrfs_inode_by_name(dir, dentry, &location);
1222         mutex_unlock(&root->fs_info->fs_mutex);
1223
1224         if (ret < 0)
1225                 return ERR_PTR(ret);
1226
1227         inode = NULL;
1228         if (location.objectid) {
1229                 ret = fixup_tree_root_location(root, &location, &sub_root,
1230                                                 dentry);
1231                 if (ret < 0)
1232                         return ERR_PTR(ret);
1233                 if (ret > 0)
1234                         return ERR_PTR(-ENOENT);
1235                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1236                                           sub_root);
1237                 if (!inode)
1238                         return ERR_PTR(-EACCES);
1239                 if (inode->i_state & I_NEW) {
1240                         /* the inode and parent dir are two different roots */
1241                         if (sub_root != root) {
1242                                 igrab(inode);
1243                                 sub_root->inode = inode;
1244                         }
1245                         BTRFS_I(inode)->root = sub_root;
1246                         memcpy(&BTRFS_I(inode)->location, &location,
1247                                sizeof(location));
1248                         btrfs_read_locked_inode(inode);
1249                         unlock_new_inode(inode);
1250                 }
1251         }
1252         return d_splice_alias(inode, dentry);
1253 }
1254
1255 static unsigned char btrfs_filetype_table[] = {
1256         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1257 };
1258
1259 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1260 {
1261         struct inode *inode = filp->f_dentry->d_inode;
1262         struct btrfs_root *root = BTRFS_I(inode)->root;
1263         struct btrfs_item *item;
1264         struct btrfs_dir_item *di;
1265         struct btrfs_key key;
1266         struct btrfs_key found_key;
1267         struct btrfs_path *path;
1268         int ret;
1269         u32 nritems;
1270         struct extent_buffer *leaf;
1271         int slot;
1272         int advance;
1273         unsigned char d_type;
1274         int over = 0;
1275         u32 di_cur;
1276         u32 di_total;
1277         u32 di_len;
1278         int key_type = BTRFS_DIR_INDEX_KEY;
1279         char tmp_name[32];
1280         char *name_ptr;
1281         int name_len;
1282
1283         /* FIXME, use a real flag for deciding about the key type */
1284         if (root->fs_info->tree_root == root)
1285                 key_type = BTRFS_DIR_ITEM_KEY;
1286
1287         /* special case for "." */
1288         if (filp->f_pos == 0) {
1289                 over = filldir(dirent, ".", 1,
1290                                1, inode->i_ino,
1291                                DT_DIR);
1292                 if (over)
1293                         return 0;
1294                 filp->f_pos = 1;
1295         }
1296
1297         mutex_lock(&root->fs_info->fs_mutex);
1298         key.objectid = inode->i_ino;
1299         path = btrfs_alloc_path();
1300         path->reada = 2;
1301
1302         /* special case for .., just use the back ref */
1303         if (filp->f_pos == 1) {
1304                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1305                 key.offset = 0;
1306                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1307                 BUG_ON(ret == 0);
1308                 leaf = path->nodes[0];
1309                 slot = path->slots[0];
1310                 nritems = btrfs_header_nritems(leaf);
1311                 if (slot >= nritems) {
1312                         btrfs_release_path(root, path);
1313                         goto read_dir_items;
1314                 }
1315                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1316                 btrfs_release_path(root, path);
1317                 if (found_key.objectid != key.objectid ||
1318                     found_key.type != BTRFS_INODE_REF_KEY)
1319                         goto read_dir_items;
1320                 over = filldir(dirent, "..", 2,
1321                                2, found_key.offset, DT_DIR);
1322                 if (over)
1323                         goto nopos;
1324                 filp->f_pos = 2;
1325         }
1326
1327 read_dir_items:
1328         btrfs_set_key_type(&key, key_type);
1329         key.offset = filp->f_pos;
1330
1331         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1332         if (ret < 0)
1333                 goto err;
1334         advance = 0;
1335         while(1) {
1336                 leaf = path->nodes[0];
1337                 nritems = btrfs_header_nritems(leaf);
1338                 slot = path->slots[0];
1339                 if (advance || slot >= nritems) {
1340                         if (slot >= nritems -1) {
1341                                 ret = btrfs_next_leaf(root, path);
1342                                 if (ret)
1343                                         break;
1344                                 leaf = path->nodes[0];
1345                                 nritems = btrfs_header_nritems(leaf);
1346                                 slot = path->slots[0];
1347                         } else {
1348                                 slot++;
1349                                 path->slots[0]++;
1350                         }
1351                 }
1352                 advance = 1;
1353                 item = btrfs_item_nr(leaf, slot);
1354                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1355
1356                 if (found_key.objectid != key.objectid)
1357                         break;
1358                 if (btrfs_key_type(&found_key) != key_type)
1359                         break;
1360                 if (found_key.offset < filp->f_pos)
1361                         continue;
1362
1363                 filp->f_pos = found_key.offset;
1364                 advance = 1;
1365                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1366                 di_cur = 0;
1367                 di_total = btrfs_item_size(leaf, item);
1368                 while(di_cur < di_total) {
1369                         struct btrfs_key location;
1370
1371                         name_len = btrfs_dir_name_len(leaf, di);
1372                         if (name_len < 32) {
1373                                 name_ptr = tmp_name;
1374                         } else {
1375                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1376                                 BUG_ON(!name_ptr);
1377                         }
1378                         read_extent_buffer(leaf, name_ptr,
1379                                            (unsigned long)(di + 1), name_len);
1380
1381                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1382                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1383                         over = filldir(dirent, name_ptr, name_len,
1384                                        found_key.offset,
1385                                        location.objectid,
1386                                        d_type);
1387
1388                         if (name_ptr != tmp_name)
1389                                 kfree(name_ptr);
1390
1391                         if (over)
1392                                 goto nopos;
1393                         di_len = btrfs_dir_name_len(leaf, di) +
1394                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1395                         di_cur += di_len;
1396                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1397                 }
1398         }
1399         filp->f_pos++;
1400 nopos:
1401         ret = 0;
1402 err:
1403         btrfs_release_path(root, path);
1404         btrfs_free_path(path);
1405         mutex_unlock(&root->fs_info->fs_mutex);
1406         return ret;
1407 }
1408
1409 int btrfs_write_inode(struct inode *inode, int wait)
1410 {
1411         struct btrfs_root *root = BTRFS_I(inode)->root;
1412         struct btrfs_trans_handle *trans;
1413         int ret = 0;
1414
1415         if (wait) {
1416                 mutex_lock(&root->fs_info->fs_mutex);
1417                 trans = btrfs_start_transaction(root, 1);
1418                 btrfs_set_trans_block_group(trans, inode);
1419                 ret = btrfs_commit_transaction(trans, root);
1420                 mutex_unlock(&root->fs_info->fs_mutex);
1421         }
1422         return ret;
1423 }
1424
1425 /*
1426  * This is somewhat expensive, updating the tree every time the
1427  * inode changes.  But, it is most likely to find the inode in cache.
1428  * FIXME, needs more benchmarking...there are no reasons other than performance
1429  * to keep or drop this code.
1430  */
1431 void btrfs_dirty_inode(struct inode *inode)
1432 {
1433         struct btrfs_root *root = BTRFS_I(inode)->root;
1434         struct btrfs_trans_handle *trans;
1435
1436         mutex_lock(&root->fs_info->fs_mutex);
1437         trans = btrfs_start_transaction(root, 1);
1438         btrfs_set_trans_block_group(trans, inode);
1439         btrfs_update_inode(trans, root, inode);
1440         btrfs_end_transaction(trans, root);
1441         mutex_unlock(&root->fs_info->fs_mutex);
1442 }
1443
1444 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1445                                      struct btrfs_root *root,
1446                                      u64 objectid,
1447                                      struct btrfs_block_group_cache *group,
1448                                      int mode)
1449 {
1450         struct inode *inode;
1451         struct btrfs_inode_item *inode_item;
1452         struct btrfs_key *location;
1453         struct btrfs_path *path;
1454         int ret;
1455         int owner;
1456
1457         path = btrfs_alloc_path();
1458         BUG_ON(!path);
1459
1460         inode = new_inode(root->fs_info->sb);
1461         if (!inode)
1462                 return ERR_PTR(-ENOMEM);
1463
1464         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1465                              inode->i_mapping, GFP_NOFS);
1466         BTRFS_I(inode)->root = root;
1467
1468         if (mode & S_IFDIR)
1469                 owner = 0;
1470         else
1471                 owner = 1;
1472         group = btrfs_find_block_group(root, group, 0, 0, owner);
1473         BTRFS_I(inode)->block_group = group;
1474         BTRFS_I(inode)->flags = 0;
1475         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1476         if (ret)
1477                 goto fail;
1478
1479         inode->i_uid = current->fsuid;
1480         inode->i_gid = current->fsgid;
1481         inode->i_mode = mode;
1482         inode->i_ino = objectid;
1483         inode->i_blocks = 0;
1484         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1485         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1486                                   struct btrfs_inode_item);
1487         fill_inode_item(path->nodes[0], inode_item, inode);
1488         btrfs_mark_buffer_dirty(path->nodes[0]);
1489         btrfs_free_path(path);
1490
1491         location = &BTRFS_I(inode)->location;
1492         location->objectid = objectid;
1493         location->offset = 0;
1494         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1495
1496         insert_inode_hash(inode);
1497         return inode;
1498 fail:
1499         btrfs_free_path(path);
1500         return ERR_PTR(ret);
1501 }
1502
1503 static inline u8 btrfs_inode_type(struct inode *inode)
1504 {
1505         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1506 }
1507
1508 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1509                             struct dentry *dentry, struct inode *inode)
1510 {
1511         int ret;
1512         struct btrfs_key key;
1513         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1514         struct inode *parent_inode;
1515
1516         key.objectid = inode->i_ino;
1517         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1518         key.offset = 0;
1519
1520         ret = btrfs_insert_dir_item(trans, root,
1521                                     dentry->d_name.name, dentry->d_name.len,
1522                                     dentry->d_parent->d_inode->i_ino,
1523                                     &key, btrfs_inode_type(inode));
1524         if (ret == 0) {
1525                 ret = btrfs_insert_inode_ref(trans, root,
1526                                      dentry->d_name.name,
1527                                      dentry->d_name.len,
1528                                      inode->i_ino,
1529                                      dentry->d_parent->d_inode->i_ino);
1530                 parent_inode = dentry->d_parent->d_inode;
1531                 parent_inode->i_size += dentry->d_name.len * 2;
1532                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1533                 ret = btrfs_update_inode(trans, root,
1534                                          dentry->d_parent->d_inode);
1535         }
1536         return ret;
1537 }
1538
1539 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1540                             struct dentry *dentry, struct inode *inode)
1541 {
1542         int err = btrfs_add_link(trans, dentry, inode);
1543         if (!err) {
1544                 d_instantiate(dentry, inode);
1545                 return 0;
1546         }
1547         if (err > 0)
1548                 err = -EEXIST;
1549         return err;
1550 }
1551
1552 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1553                         int mode, dev_t rdev)
1554 {
1555         struct btrfs_trans_handle *trans;
1556         struct btrfs_root *root = BTRFS_I(dir)->root;
1557         struct inode *inode = NULL;
1558         int err;
1559         int drop_inode = 0;
1560         u64 objectid;
1561         unsigned long nr = 0;
1562
1563         if (!new_valid_dev(rdev))
1564                 return -EINVAL;
1565
1566         mutex_lock(&root->fs_info->fs_mutex);
1567         err = btrfs_check_free_space(root, 1, 0);
1568         if (err)
1569                 goto fail;
1570
1571         trans = btrfs_start_transaction(root, 1);
1572         btrfs_set_trans_block_group(trans, dir);
1573
1574         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1575         if (err) {
1576                 err = -ENOSPC;
1577                 goto out_unlock;
1578         }
1579
1580         inode = btrfs_new_inode(trans, root, objectid,
1581                                 BTRFS_I(dir)->block_group, mode);
1582         err = PTR_ERR(inode);
1583         if (IS_ERR(inode))
1584                 goto out_unlock;
1585
1586         btrfs_set_trans_block_group(trans, inode);
1587         err = btrfs_add_nondir(trans, dentry, inode);
1588         if (err)
1589                 drop_inode = 1;
1590         else {
1591                 inode->i_op = &btrfs_special_inode_operations;
1592                 init_special_inode(inode, inode->i_mode, rdev);
1593                 btrfs_update_inode(trans, root, inode);
1594         }
1595         dir->i_sb->s_dirt = 1;
1596         btrfs_update_inode_block_group(trans, inode);
1597         btrfs_update_inode_block_group(trans, dir);
1598 out_unlock:
1599         nr = trans->blocks_used;
1600         btrfs_end_transaction(trans, root);
1601 fail:
1602         mutex_unlock(&root->fs_info->fs_mutex);
1603
1604         if (drop_inode) {
1605                 inode_dec_link_count(inode);
1606                 iput(inode);
1607         }
1608         btrfs_btree_balance_dirty(root, nr);
1609         btrfs_throttle(root);
1610         return err;
1611 }
1612
1613 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1614                         int mode, struct nameidata *nd)
1615 {
1616         struct btrfs_trans_handle *trans;
1617         struct btrfs_root *root = BTRFS_I(dir)->root;
1618         struct inode *inode = NULL;
1619         int err;
1620         int drop_inode = 0;
1621         unsigned long nr = 0;
1622         u64 objectid;
1623
1624         mutex_lock(&root->fs_info->fs_mutex);
1625         err = btrfs_check_free_space(root, 1, 0);
1626         if (err)
1627                 goto fail;
1628         trans = btrfs_start_transaction(root, 1);
1629         btrfs_set_trans_block_group(trans, dir);
1630
1631         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1632         if (err) {
1633                 err = -ENOSPC;
1634                 goto out_unlock;
1635         }
1636
1637         inode = btrfs_new_inode(trans, root, objectid,
1638                                 BTRFS_I(dir)->block_group, mode);
1639         err = PTR_ERR(inode);
1640         if (IS_ERR(inode))
1641                 goto out_unlock;
1642
1643         btrfs_set_trans_block_group(trans, inode);
1644         err = btrfs_add_nondir(trans, dentry, inode);
1645         if (err)
1646                 drop_inode = 1;
1647         else {
1648                 inode->i_mapping->a_ops = &btrfs_aops;
1649                 inode->i_fop = &btrfs_file_operations;
1650                 inode->i_op = &btrfs_file_inode_operations;
1651                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1652                                      inode->i_mapping, GFP_NOFS);
1653                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1654         }
1655         dir->i_sb->s_dirt = 1;
1656         btrfs_update_inode_block_group(trans, inode);
1657         btrfs_update_inode_block_group(trans, dir);
1658 out_unlock:
1659         nr = trans->blocks_used;
1660         btrfs_end_transaction(trans, root);
1661 fail:
1662         mutex_unlock(&root->fs_info->fs_mutex);
1663
1664         if (drop_inode) {
1665                 inode_dec_link_count(inode);
1666                 iput(inode);
1667         }
1668         btrfs_btree_balance_dirty(root, nr);
1669         btrfs_throttle(root);
1670         return err;
1671 }
1672
1673 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1674                       struct dentry *dentry)
1675 {
1676         struct btrfs_trans_handle *trans;
1677         struct btrfs_root *root = BTRFS_I(dir)->root;
1678         struct inode *inode = old_dentry->d_inode;
1679         unsigned long nr = 0;
1680         int err;
1681         int drop_inode = 0;
1682
1683         if (inode->i_nlink == 0)
1684                 return -ENOENT;
1685
1686 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1687         inode->i_nlink++;
1688 #else
1689         inc_nlink(inode);
1690 #endif
1691         mutex_lock(&root->fs_info->fs_mutex);
1692         err = btrfs_check_free_space(root, 1, 0);
1693         if (err)
1694                 goto fail;
1695         trans = btrfs_start_transaction(root, 1);
1696
1697         btrfs_set_trans_block_group(trans, dir);
1698         atomic_inc(&inode->i_count);
1699         err = btrfs_add_nondir(trans, dentry, inode);
1700
1701         if (err)
1702                 drop_inode = 1;
1703
1704         dir->i_sb->s_dirt = 1;
1705         btrfs_update_inode_block_group(trans, dir);
1706         err = btrfs_update_inode(trans, root, inode);
1707
1708         if (err)
1709                 drop_inode = 1;
1710
1711         nr = trans->blocks_used;
1712         btrfs_end_transaction(trans, root);
1713 fail:
1714         mutex_unlock(&root->fs_info->fs_mutex);
1715
1716         if (drop_inode) {
1717                 inode_dec_link_count(inode);
1718                 iput(inode);
1719         }
1720         btrfs_btree_balance_dirty(root, nr);
1721         btrfs_throttle(root);
1722         return err;
1723 }
1724
1725 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1726 {
1727         struct inode *inode;
1728         struct btrfs_trans_handle *trans;
1729         struct btrfs_root *root = BTRFS_I(dir)->root;
1730         int err = 0;
1731         int drop_on_err = 0;
1732         u64 objectid;
1733         unsigned long nr = 1;
1734
1735         mutex_lock(&root->fs_info->fs_mutex);
1736         err = btrfs_check_free_space(root, 1, 0);
1737         if (err)
1738                 goto out_unlock;
1739
1740         trans = btrfs_start_transaction(root, 1);
1741         btrfs_set_trans_block_group(trans, dir);
1742
1743         if (IS_ERR(trans)) {
1744                 err = PTR_ERR(trans);
1745                 goto out_unlock;
1746         }
1747
1748         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1749         if (err) {
1750                 err = -ENOSPC;
1751                 goto out_unlock;
1752         }
1753
1754         inode = btrfs_new_inode(trans, root, objectid,
1755                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1756         if (IS_ERR(inode)) {
1757                 err = PTR_ERR(inode);
1758                 goto out_fail;
1759         }
1760
1761         drop_on_err = 1;
1762         inode->i_op = &btrfs_dir_inode_operations;
1763         inode->i_fop = &btrfs_dir_file_operations;
1764         btrfs_set_trans_block_group(trans, inode);
1765
1766         inode->i_size = 0;
1767         err = btrfs_update_inode(trans, root, inode);
1768         if (err)
1769                 goto out_fail;
1770
1771         err = btrfs_add_link(trans, dentry, inode);
1772         if (err)
1773                 goto out_fail;
1774
1775         d_instantiate(dentry, inode);
1776         drop_on_err = 0;
1777         dir->i_sb->s_dirt = 1;
1778         btrfs_update_inode_block_group(trans, inode);
1779         btrfs_update_inode_block_group(trans, dir);
1780
1781 out_fail:
1782         nr = trans->blocks_used;
1783         btrfs_end_transaction(trans, root);
1784
1785 out_unlock:
1786         mutex_unlock(&root->fs_info->fs_mutex);
1787         if (drop_on_err)
1788                 iput(inode);
1789         btrfs_btree_balance_dirty(root, nr);
1790         btrfs_throttle(root);
1791         return err;
1792 }
1793
1794 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1795                                     size_t page_offset, u64 start, u64 end,
1796                                     int create)
1797 {
1798         int ret;
1799         int err = 0;
1800         u64 bytenr;
1801         u64 extent_start = 0;
1802         u64 extent_end = 0;
1803         u64 objectid = inode->i_ino;
1804         u32 found_type;
1805         int failed_insert = 0;
1806         struct btrfs_path *path;
1807         struct btrfs_root *root = BTRFS_I(inode)->root;
1808         struct btrfs_file_extent_item *item;
1809         struct extent_buffer *leaf;
1810         struct btrfs_key found_key;
1811         struct extent_map *em = NULL;
1812         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1813         struct btrfs_trans_handle *trans = NULL;
1814
1815         path = btrfs_alloc_path();
1816         BUG_ON(!path);
1817         mutex_lock(&root->fs_info->fs_mutex);
1818
1819 again:
1820         em = lookup_extent_mapping(em_tree, start, end);
1821         if (em) {
1822                 if (em->start > start) {
1823                         printk("get_extent start %Lu em start %Lu\n",
1824                                start, em->start);
1825                         WARN_ON(1);
1826                 }
1827                 goto out;
1828         }
1829         if (!em) {
1830                 em = alloc_extent_map(GFP_NOFS);
1831                 if (!em) {
1832                         err = -ENOMEM;
1833                         goto out;
1834                 }
1835                 em->start = EXTENT_MAP_HOLE;
1836                 em->end = EXTENT_MAP_HOLE;
1837         }
1838         em->bdev = inode->i_sb->s_bdev;
1839         ret = btrfs_lookup_file_extent(trans, root, path,
1840                                        objectid, start, trans != NULL);
1841         if (ret < 0) {
1842                 err = ret;
1843                 goto out;
1844         }
1845
1846         if (ret != 0) {
1847                 if (path->slots[0] == 0)
1848                         goto not_found;
1849                 path->slots[0]--;
1850         }
1851
1852         leaf = path->nodes[0];
1853         item = btrfs_item_ptr(leaf, path->slots[0],
1854                               struct btrfs_file_extent_item);
1855         /* are we inside the extent that was found? */
1856         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1857         found_type = btrfs_key_type(&found_key);
1858         if (found_key.objectid != objectid ||
1859             found_type != BTRFS_EXTENT_DATA_KEY) {
1860                 goto not_found;
1861         }
1862
1863         found_type = btrfs_file_extent_type(leaf, item);
1864         extent_start = found_key.offset;
1865         if (found_type == BTRFS_FILE_EXTENT_REG) {
1866                 extent_end = extent_start +
1867                        btrfs_file_extent_num_bytes(leaf, item);
1868                 err = 0;
1869                 if (start < extent_start || start >= extent_end) {
1870                         em->start = start;
1871                         if (start < extent_start) {
1872                                 if (end < extent_start)
1873                                         goto not_found;
1874                                 em->end = extent_end - 1;
1875                         } else {
1876                                 em->end = end;
1877                         }
1878                         goto not_found_em;
1879                 }
1880                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1881                 if (bytenr == 0) {
1882                         em->start = extent_start;
1883                         em->end = extent_end - 1;
1884                         em->block_start = EXTENT_MAP_HOLE;
1885                         em->block_end = EXTENT_MAP_HOLE;
1886                         goto insert;
1887                 }
1888                 bytenr += btrfs_file_extent_offset(leaf, item);
1889                 em->block_start = bytenr;
1890                 em->block_end = em->block_start +
1891                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1892                 em->start = extent_start;
1893                 em->end = extent_end - 1;
1894                 goto insert;
1895         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1896                 unsigned long ptr;
1897                 char *map;
1898                 size_t size;
1899                 size_t extent_offset;
1900                 size_t copy_size;
1901
1902                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1903                                                     path->slots[0]));
1904                 extent_end = (extent_start + size - 1) |
1905                         ((u64)root->sectorsize - 1);
1906                 if (start < extent_start || start >= extent_end) {
1907                         em->start = start;
1908                         if (start < extent_start) {
1909                                 if (end < extent_start)
1910                                         goto not_found;
1911                                 em->end = extent_end;
1912                         } else {
1913                                 em->end = end;
1914                         }
1915                         goto not_found_em;
1916                 }
1917                 em->block_start = EXTENT_MAP_INLINE;
1918                 em->block_end = EXTENT_MAP_INLINE;
1919
1920                 if (!page) {
1921                         em->start = extent_start;
1922                         em->end = extent_start + size - 1;
1923                         goto out;
1924                 }
1925
1926                 extent_offset = ((u64)page->index << PAGE_CACHE_SHIFT) -
1927                         extent_start + page_offset;
1928                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1929                                 size - extent_offset);
1930                 em->start = extent_start + extent_offset;
1931                 em->end = (em->start + copy_size -1) |
1932                         ((u64)root->sectorsize -1);
1933                 map = kmap(page);
1934                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1935                 if (create == 0 && !PageUptodate(page)) {
1936                         read_extent_buffer(leaf, map + page_offset, ptr,
1937                                            copy_size);
1938                         flush_dcache_page(page);
1939                 } else if (create && PageUptodate(page)) {
1940                         if (!trans) {
1941                                 kunmap(page);
1942                                 free_extent_map(em);
1943                                 em = NULL;
1944                                 btrfs_release_path(root, path);
1945                                 trans = btrfs_start_transaction(root, 1);
1946                                 goto again;
1947                         }
1948                         write_extent_buffer(leaf, map + page_offset, ptr,
1949                                             copy_size);
1950                         btrfs_mark_buffer_dirty(leaf);
1951                 }
1952                 kunmap(page);
1953                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1954                 goto insert;
1955         } else {
1956                 printk("unkknown found_type %d\n", found_type);
1957                 WARN_ON(1);
1958         }
1959 not_found:
1960         em->start = start;
1961         em->end = end;
1962 not_found_em:
1963         em->block_start = EXTENT_MAP_HOLE;
1964         em->block_end = EXTENT_MAP_HOLE;
1965 insert:
1966         btrfs_release_path(root, path);
1967         if (em->start > start || em->end < start) {
1968                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1969                 err = -EIO;
1970                 goto out;
1971         }
1972         ret = add_extent_mapping(em_tree, em);
1973         if (ret == -EEXIST) {
1974                 free_extent_map(em);
1975                 em = NULL;
1976                 if (0 && failed_insert == 1) {
1977                         btrfs_drop_extent_cache(inode, start, end);
1978                 }
1979                 failed_insert++;
1980                 if (failed_insert > 5) {
1981                         printk("failing to insert %Lu %Lu\n", start, end);
1982                         err = -EIO;
1983                         goto out;
1984                 }
1985                 goto again;
1986         }
1987         err = 0;
1988 out:
1989         btrfs_free_path(path);
1990         if (trans) {
1991                 ret = btrfs_end_transaction(trans, root);
1992                 if (!err)
1993                         err = ret;
1994         }
1995         mutex_unlock(&root->fs_info->fs_mutex);
1996         if (err) {
1997                 free_extent_map(em);
1998                 WARN_ON(1);
1999                 return ERR_PTR(err);
2000         }
2001         return em;
2002 }
2003
2004 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2005 {
2006         return extent_bmap(mapping, iblock, btrfs_get_extent);
2007 }
2008
2009 int btrfs_readpage(struct file *file, struct page *page)
2010 {
2011         struct extent_map_tree *tree;
2012         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2013         return extent_read_full_page(tree, page, btrfs_get_extent);
2014 }
2015
2016 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2017 {
2018         struct extent_map_tree *tree;
2019
2020
2021         if (current->flags & PF_MEMALLOC) {
2022                 redirty_page_for_writepage(wbc, page);
2023                 unlock_page(page);
2024                 return 0;
2025         }
2026         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2027         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2028 }
2029
2030 static int btrfs_writepages(struct address_space *mapping,
2031                             struct writeback_control *wbc)
2032 {
2033         struct extent_map_tree *tree;
2034         tree = &BTRFS_I(mapping->host)->extent_tree;
2035         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2036 }
2037
2038 static int
2039 btrfs_readpages(struct file *file, struct address_space *mapping,
2040                 struct list_head *pages, unsigned nr_pages)
2041 {
2042         struct extent_map_tree *tree;
2043         tree = &BTRFS_I(mapping->host)->extent_tree;
2044         return extent_readpages(tree, mapping, pages, nr_pages,
2045                                 btrfs_get_extent);
2046 }
2047
2048 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2049 {
2050         struct extent_map_tree *tree;
2051         int ret;
2052
2053         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2054         ret = try_release_extent_mapping(tree, page);
2055         if (ret == 1) {
2056                 ClearPagePrivate(page);
2057                 set_page_private(page, 0);
2058                 page_cache_release(page);
2059         }
2060         return ret;
2061 }
2062
2063 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2064 {
2065         struct extent_map_tree *tree;
2066
2067         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2068         extent_invalidatepage(tree, page, offset);
2069         btrfs_releasepage(page, GFP_NOFS);
2070 }
2071
2072 /*
2073  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2074  * called from a page fault handler when a page is first dirtied. Hence we must
2075  * be careful to check for EOF conditions here. We set the page up correctly
2076  * for a written page which means we get ENOSPC checking when writing into
2077  * holes and correct delalloc and unwritten extent mapping on filesystems that
2078  * support these features.
2079  *
2080  * We are not allowed to take the i_mutex here so we have to play games to
2081  * protect against truncate races as the page could now be beyond EOF.  Because
2082  * vmtruncate() writes the inode size before removing pages, once we have the
2083  * page lock we can determine safely if the page is beyond EOF. If it is not
2084  * beyond EOF, then the page is guaranteed safe against truncation until we
2085  * unlock the page.
2086  */
2087 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2088 {
2089         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2090         struct btrfs_root *root = BTRFS_I(inode)->root;
2091         unsigned long end;
2092         loff_t size;
2093         int ret;
2094         u64 page_start;
2095
2096         mutex_lock(&root->fs_info->fs_mutex);
2097         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2098         mutex_unlock(&root->fs_info->fs_mutex);
2099         if (ret)
2100                 goto out;
2101
2102         ret = -EINVAL;
2103
2104         lock_page(page);
2105         wait_on_page_writeback(page);
2106         size = i_size_read(inode);
2107         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2108
2109         if ((page->mapping != inode->i_mapping) ||
2110             (page_start > size)) {
2111                 /* page got truncated out from underneath us */
2112                 goto out_unlock;
2113         }
2114
2115         /* page is wholly or partially inside EOF */
2116         if (page_start + PAGE_CACHE_SIZE > size)
2117                 end = size & ~PAGE_CACHE_MASK;
2118         else
2119                 end = PAGE_CACHE_SIZE;
2120
2121         ret = btrfs_cow_one_page(inode, page, end);
2122
2123 out_unlock:
2124         unlock_page(page);
2125 out:
2126         return ret;
2127 }
2128
2129 static void btrfs_truncate(struct inode *inode)
2130 {
2131         struct btrfs_root *root = BTRFS_I(inode)->root;
2132         int ret;
2133         struct btrfs_trans_handle *trans;
2134         unsigned long nr;
2135
2136         if (!S_ISREG(inode->i_mode))
2137                 return;
2138         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2139                 return;
2140
2141         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2142
2143         mutex_lock(&root->fs_info->fs_mutex);
2144         trans = btrfs_start_transaction(root, 1);
2145         btrfs_set_trans_block_group(trans, inode);
2146
2147         /* FIXME, add redo link to tree so we don't leak on crash */
2148         ret = btrfs_truncate_in_trans(trans, root, inode);
2149         btrfs_update_inode(trans, root, inode);
2150         nr = trans->blocks_used;
2151
2152         ret = btrfs_end_transaction(trans, root);
2153         BUG_ON(ret);
2154         mutex_unlock(&root->fs_info->fs_mutex);
2155         btrfs_btree_balance_dirty(root, nr);
2156         btrfs_throttle(root);
2157 }
2158
2159 static int noinline create_subvol(struct btrfs_root *root, char *name,
2160                                   int namelen)
2161 {
2162         struct btrfs_trans_handle *trans;
2163         struct btrfs_key key;
2164         struct btrfs_root_item root_item;
2165         struct btrfs_inode_item *inode_item;
2166         struct extent_buffer *leaf;
2167         struct btrfs_root *new_root = root;
2168         struct inode *inode;
2169         struct inode *dir;
2170         int ret;
2171         int err;
2172         u64 objectid;
2173         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2174         unsigned long nr = 1;
2175
2176         mutex_lock(&root->fs_info->fs_mutex);
2177         ret = btrfs_check_free_space(root, 1, 0);
2178         if (ret)
2179                 goto fail_commit;
2180
2181         trans = btrfs_start_transaction(root, 1);
2182         BUG_ON(!trans);
2183
2184         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2185                                        0, &objectid);
2186         if (ret)
2187                 goto fail;
2188
2189         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2190                                         objectid, trans->transid, 0, 0,
2191                                         0, 0);
2192         if (IS_ERR(leaf))
2193                 return PTR_ERR(leaf);
2194
2195         btrfs_set_header_nritems(leaf, 0);
2196         btrfs_set_header_level(leaf, 0);
2197         btrfs_set_header_bytenr(leaf, leaf->start);
2198         btrfs_set_header_generation(leaf, trans->transid);
2199         btrfs_set_header_owner(leaf, objectid);
2200
2201         write_extent_buffer(leaf, root->fs_info->fsid,
2202                             (unsigned long)btrfs_header_fsid(leaf),
2203                             BTRFS_FSID_SIZE);
2204         btrfs_mark_buffer_dirty(leaf);
2205
2206         inode_item = &root_item.inode;
2207         memset(inode_item, 0, sizeof(*inode_item));
2208         inode_item->generation = cpu_to_le64(1);
2209         inode_item->size = cpu_to_le64(3);
2210         inode_item->nlink = cpu_to_le32(1);
2211         inode_item->nblocks = cpu_to_le64(1);
2212         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2213
2214         btrfs_set_root_bytenr(&root_item, leaf->start);
2215         btrfs_set_root_level(&root_item, 0);
2216         btrfs_set_root_refs(&root_item, 1);
2217         btrfs_set_root_used(&root_item, 0);
2218
2219         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2220         root_item.drop_level = 0;
2221
2222         free_extent_buffer(leaf);
2223         leaf = NULL;
2224
2225         btrfs_set_root_dirid(&root_item, new_dirid);
2226
2227         key.objectid = objectid;
2228         key.offset = 1;
2229         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2230         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2231                                 &root_item);
2232         if (ret)
2233                 goto fail;
2234
2235         /*
2236          * insert the directory item
2237          */
2238         key.offset = (u64)-1;
2239         dir = root->fs_info->sb->s_root->d_inode;
2240         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2241                                     name, namelen, dir->i_ino, &key,
2242                                     BTRFS_FT_DIR);
2243         if (ret)
2244                 goto fail;
2245
2246         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2247                              name, namelen, objectid,
2248                              root->fs_info->sb->s_root->d_inode->i_ino);
2249         if (ret)
2250                 goto fail;
2251
2252         ret = btrfs_commit_transaction(trans, root);
2253         if (ret)
2254                 goto fail_commit;
2255
2256         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2257         BUG_ON(!new_root);
2258
2259         trans = btrfs_start_transaction(new_root, 1);
2260         BUG_ON(!trans);
2261
2262         inode = btrfs_new_inode(trans, new_root, new_dirid,
2263                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2264         if (IS_ERR(inode))
2265                 goto fail;
2266         inode->i_op = &btrfs_dir_inode_operations;
2267         inode->i_fop = &btrfs_dir_file_operations;
2268         new_root->inode = inode;
2269
2270         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2271                                      new_dirid);
2272         inode->i_nlink = 1;
2273         inode->i_size = 0;
2274         ret = btrfs_update_inode(trans, new_root, inode);
2275         if (ret)
2276                 goto fail;
2277 fail:
2278         nr = trans->blocks_used;
2279         err = btrfs_commit_transaction(trans, new_root);
2280         if (err && !ret)
2281                 ret = err;
2282 fail_commit:
2283         mutex_unlock(&root->fs_info->fs_mutex);
2284         btrfs_btree_balance_dirty(root, nr);
2285         btrfs_throttle(root);
2286         return ret;
2287 }
2288
2289 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2290 {
2291         struct btrfs_pending_snapshot *pending_snapshot;
2292         struct btrfs_trans_handle *trans;
2293         int ret;
2294         int err;
2295         unsigned long nr = 0;
2296
2297         if (!root->ref_cows)
2298                 return -EINVAL;
2299
2300         mutex_lock(&root->fs_info->fs_mutex);
2301         ret = btrfs_check_free_space(root, 1, 0);
2302         if (ret)
2303                 goto fail_unlock;
2304
2305         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2306         if (!pending_snapshot) {
2307                 ret = -ENOMEM;
2308                 goto fail_unlock;
2309         }
2310         pending_snapshot->name = kstrndup(name, namelen, GFP_NOFS);
2311         if (!pending_snapshot->name) {
2312                 ret = -ENOMEM;
2313                 kfree(pending_snapshot);
2314                 goto fail_unlock;
2315         }
2316         trans = btrfs_start_transaction(root, 1);
2317         BUG_ON(!trans);
2318
2319         pending_snapshot->root = root;
2320         list_add(&pending_snapshot->list,
2321                  &trans->transaction->pending_snapshots);
2322         ret = btrfs_update_inode(trans, root, root->inode);
2323         err = btrfs_commit_transaction(trans, root);
2324
2325 fail_unlock:
2326         mutex_unlock(&root->fs_info->fs_mutex);
2327         btrfs_btree_balance_dirty(root, nr);
2328         btrfs_throttle(root);
2329         return ret;
2330 }
2331
2332 unsigned long btrfs_force_ra(struct address_space *mapping,
2333                               struct file_ra_state *ra, struct file *file,
2334                               pgoff_t offset, pgoff_t last_index)
2335 {
2336         pgoff_t req_size;
2337
2338 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2339         req_size = last_index - offset + 1;
2340         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2341         return offset;
2342 #else
2343         req_size = min(last_index - offset + 1, (pgoff_t)128);
2344         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2345         return offset + req_size;
2346 #endif
2347 }
2348
2349 int btrfs_defrag_file(struct file *file) {
2350         struct inode *inode = fdentry(file)->d_inode;
2351         struct btrfs_root *root = BTRFS_I(inode)->root;
2352         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2353         struct page *page;
2354         unsigned long last_index;
2355         unsigned long ra_index = 0;
2356         u64 page_start;
2357         u64 page_end;
2358         u64 delalloc_start;
2359         u64 existing_delalloc;
2360         unsigned long i;
2361         int ret;
2362
2363         mutex_lock(&root->fs_info->fs_mutex);
2364         ret = btrfs_check_free_space(root, inode->i_size, 0);
2365         mutex_unlock(&root->fs_info->fs_mutex);
2366         if (ret)
2367                 return -ENOSPC;
2368
2369         mutex_lock(&inode->i_mutex);
2370         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2371         for (i = 0; i <= last_index; i++) {
2372                 if (i == ra_index) {
2373                         ra_index = btrfs_force_ra(inode->i_mapping,
2374                                                   &file->f_ra,
2375                                                   file, ra_index, last_index);
2376                 }
2377                 page = grab_cache_page(inode->i_mapping, i);
2378                 if (!page)
2379                         goto out_unlock;
2380                 if (!PageUptodate(page)) {
2381                         btrfs_readpage(NULL, page);
2382                         lock_page(page);
2383                         if (!PageUptodate(page)) {
2384                                 unlock_page(page);
2385                                 page_cache_release(page);
2386                                 goto out_unlock;
2387                         }
2388                 }
2389                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2390                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2391
2392                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2393                 delalloc_start = page_start;
2394                 existing_delalloc =
2395                         count_range_bits(&BTRFS_I(inode)->extent_tree,
2396                                          &delalloc_start, page_end,
2397                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2398                 set_extent_delalloc(em_tree, page_start,
2399                                     page_end, GFP_NOFS);
2400
2401                 spin_lock(&root->fs_info->delalloc_lock);
2402                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2403                                                  existing_delalloc;
2404                 spin_unlock(&root->fs_info->delalloc_lock);
2405
2406                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2407                 set_page_dirty(page);
2408                 unlock_page(page);
2409                 page_cache_release(page);
2410                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2411         }
2412
2413 out_unlock:
2414         mutex_unlock(&inode->i_mutex);
2415         return 0;
2416 }
2417
2418 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2419 {
2420         u64 new_size;
2421         u64 old_size;
2422         struct btrfs_ioctl_vol_args *vol_args;
2423         struct btrfs_trans_handle *trans;
2424         char *sizestr;
2425         int ret = 0;
2426         int namelen;
2427         int mod = 0;
2428
2429         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2430
2431         if (!vol_args)
2432                 return -ENOMEM;
2433
2434         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2435                 ret = -EFAULT;
2436                 goto out;
2437         }
2438         namelen = strlen(vol_args->name);
2439         if (namelen > BTRFS_VOL_NAME_MAX) {
2440                 ret = -EINVAL;
2441                 goto out;
2442         }
2443
2444         sizestr = vol_args->name;
2445         if (!strcmp(sizestr, "max"))
2446                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2447         else {
2448                 if (sizestr[0] == '-') {
2449                         mod = -1;
2450                         sizestr++;
2451                 } else if (sizestr[0] == '+') {
2452                         mod = 1;
2453                         sizestr++;
2454                 }
2455                 new_size = btrfs_parse_size(sizestr);
2456                 if (new_size == 0) {
2457                         ret = -EINVAL;
2458                         goto out;
2459                 }
2460         }
2461
2462         mutex_lock(&root->fs_info->fs_mutex);
2463         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2464
2465         if (mod < 0) {
2466                 if (new_size > old_size) {
2467                         ret = -EINVAL;
2468                         goto out_unlock;
2469                 }
2470                 new_size = old_size - new_size;
2471         } else if (mod > 0) {
2472                 new_size = old_size + new_size;
2473         }
2474
2475         if (new_size < 256 * 1024 * 1024) {
2476                 ret = -EINVAL;
2477                 goto out_unlock;
2478         }
2479         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2480                 ret = -EFBIG;
2481                 goto out_unlock;
2482         }
2483
2484         do_div(new_size, root->sectorsize);
2485         new_size *= root->sectorsize;
2486
2487 printk("new size is %Lu\n", new_size);
2488         if (new_size > old_size) {
2489                 trans = btrfs_start_transaction(root, 1);
2490                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2491                 btrfs_commit_transaction(trans, root);
2492         } else {
2493                 ret = btrfs_shrink_extent_tree(root, new_size);
2494         }
2495
2496 out_unlock:
2497         mutex_unlock(&root->fs_info->fs_mutex);
2498 out:
2499         kfree(vol_args);
2500         return ret;
2501 }
2502
2503 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2504                                             void __user *arg)
2505 {
2506         struct btrfs_ioctl_vol_args *vol_args;
2507         struct btrfs_dir_item *di;
2508         struct btrfs_path *path;
2509         u64 root_dirid;
2510         int namelen;
2511         int ret;
2512
2513         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2514
2515         if (!vol_args)
2516                 return -ENOMEM;
2517
2518         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2519                 ret = -EFAULT;
2520                 goto out;
2521         }
2522
2523         namelen = strlen(vol_args->name);
2524         if (namelen > BTRFS_VOL_NAME_MAX) {
2525                 ret = -EINVAL;
2526                 goto out;
2527         }
2528         if (strchr(vol_args->name, '/')) {
2529                 ret = -EINVAL;
2530                 goto out;
2531         }
2532
2533         path = btrfs_alloc_path();
2534         if (!path) {
2535                 ret = -ENOMEM;
2536                 goto out;
2537         }
2538
2539         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2540         mutex_lock(&root->fs_info->fs_mutex);
2541         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2542                             path, root_dirid,
2543                             vol_args->name, namelen, 0);
2544         mutex_unlock(&root->fs_info->fs_mutex);
2545         btrfs_free_path(path);
2546
2547         if (di && !IS_ERR(di)) {
2548                 ret = -EEXIST;
2549                 goto out;
2550         }
2551
2552         if (IS_ERR(di)) {
2553                 ret = PTR_ERR(di);
2554                 goto out;
2555         }
2556
2557         if (root == root->fs_info->tree_root)
2558                 ret = create_subvol(root, vol_args->name, namelen);
2559         else
2560                 ret = create_snapshot(root, vol_args->name, namelen);
2561 out:
2562         kfree(vol_args);
2563         return ret;
2564 }
2565
2566 static int btrfs_ioctl_defrag(struct file *file)
2567 {
2568         struct inode *inode = fdentry(file)->d_inode;
2569         struct btrfs_root *root = BTRFS_I(inode)->root;
2570
2571         switch (inode->i_mode & S_IFMT) {
2572         case S_IFDIR:
2573                 mutex_lock(&root->fs_info->fs_mutex);
2574                 btrfs_defrag_root(root, 0);
2575                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2576                 mutex_unlock(&root->fs_info->fs_mutex);
2577                 break;
2578         case S_IFREG:
2579                 btrfs_defrag_file(file);
2580                 break;
2581         }
2582
2583         return 0;
2584 }
2585
2586 long btrfs_ioctl(struct file *file, unsigned int
2587                 cmd, unsigned long arg)
2588 {
2589         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2590
2591         switch (cmd) {
2592         case BTRFS_IOC_SNAP_CREATE:
2593                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2594         case BTRFS_IOC_DEFRAG:
2595                 return btrfs_ioctl_defrag(file);
2596         case BTRFS_IOC_RESIZE:
2597                 return btrfs_ioctl_resize(root, (void __user *)arg);
2598         }
2599
2600         return -ENOTTY;
2601 }
2602
2603 /*
2604  * Called inside transaction, so use GFP_NOFS
2605  */
2606 struct inode *btrfs_alloc_inode(struct super_block *sb)
2607 {
2608         struct btrfs_inode *ei;
2609
2610         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2611         if (!ei)
2612                 return NULL;
2613         ei->last_trans = 0;
2614         ei->ordered_trans = 0;
2615         return &ei->vfs_inode;
2616 }
2617
2618 void btrfs_destroy_inode(struct inode *inode)
2619 {
2620         WARN_ON(!list_empty(&inode->i_dentry));
2621         WARN_ON(inode->i_data.nrpages);
2622
2623         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2624         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2625 }
2626
2627 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2628 static void init_once(struct kmem_cache * cachep, void *foo)
2629 #else
2630 static void init_once(void * foo, struct kmem_cache * cachep,
2631                       unsigned long flags)
2632 #endif
2633 {
2634         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2635
2636         inode_init_once(&ei->vfs_inode);
2637 }
2638
2639 void btrfs_destroy_cachep(void)
2640 {
2641         if (btrfs_inode_cachep)
2642                 kmem_cache_destroy(btrfs_inode_cachep);
2643         if (btrfs_trans_handle_cachep)
2644                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2645         if (btrfs_transaction_cachep)
2646                 kmem_cache_destroy(btrfs_transaction_cachep);
2647         if (btrfs_bit_radix_cachep)
2648                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2649         if (btrfs_path_cachep)
2650                 kmem_cache_destroy(btrfs_path_cachep);
2651 }
2652
2653 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2654                                        unsigned long extra_flags,
2655 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2656                                        void (*ctor)(struct kmem_cache *, void *)
2657 #else
2658                                        void (*ctor)(void *, struct kmem_cache *,
2659                                                     unsigned long)
2660 #endif
2661                                      )
2662 {
2663         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2664                                  SLAB_MEM_SPREAD | extra_flags), ctor
2665 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2666                                  ,NULL
2667 #endif
2668                                 );
2669 }
2670
2671 int btrfs_init_cachep(void)
2672 {
2673         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2674                                           sizeof(struct btrfs_inode),
2675                                           0, init_once);
2676         if (!btrfs_inode_cachep)
2677                 goto fail;
2678         btrfs_trans_handle_cachep =
2679                         btrfs_cache_create("btrfs_trans_handle_cache",
2680                                            sizeof(struct btrfs_trans_handle),
2681                                            0, NULL);
2682         if (!btrfs_trans_handle_cachep)
2683                 goto fail;
2684         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2685                                              sizeof(struct btrfs_transaction),
2686                                              0, NULL);
2687         if (!btrfs_transaction_cachep)
2688                 goto fail;
2689         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2690                                          sizeof(struct btrfs_path),
2691                                          0, NULL);
2692         if (!btrfs_path_cachep)
2693                 goto fail;
2694         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2695                                               SLAB_DESTROY_BY_RCU, NULL);
2696         if (!btrfs_bit_radix_cachep)
2697                 goto fail;
2698         return 0;
2699 fail:
2700         btrfs_destroy_cachep();
2701         return -ENOMEM;
2702 }
2703
2704 static int btrfs_getattr(struct vfsmount *mnt,
2705                          struct dentry *dentry, struct kstat *stat)
2706 {
2707         struct inode *inode = dentry->d_inode;
2708         generic_fillattr(inode, stat);
2709         stat->blksize = PAGE_CACHE_SIZE;
2710         return 0;
2711 }
2712
2713 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2714                            struct inode * new_dir,struct dentry *new_dentry)
2715 {
2716         struct btrfs_trans_handle *trans;
2717         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2718         struct inode *new_inode = new_dentry->d_inode;
2719         struct inode *old_inode = old_dentry->d_inode;
2720         struct timespec ctime = CURRENT_TIME;
2721         struct btrfs_path *path;
2722         int ret;
2723
2724         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2725             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2726                 return -ENOTEMPTY;
2727         }
2728
2729         mutex_lock(&root->fs_info->fs_mutex);
2730         ret = btrfs_check_free_space(root, 1, 0);
2731         if (ret)
2732                 goto out_unlock;
2733
2734         trans = btrfs_start_transaction(root, 1);
2735
2736         btrfs_set_trans_block_group(trans, new_dir);
2737         path = btrfs_alloc_path();
2738         if (!path) {
2739                 ret = -ENOMEM;
2740                 goto out_fail;
2741         }
2742
2743         old_dentry->d_inode->i_nlink++;
2744         old_dir->i_ctime = old_dir->i_mtime = ctime;
2745         new_dir->i_ctime = new_dir->i_mtime = ctime;
2746         old_inode->i_ctime = ctime;
2747
2748         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2749         if (ret)
2750                 goto out_fail;
2751
2752         if (new_inode) {
2753                 new_inode->i_ctime = CURRENT_TIME;
2754                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2755                 if (ret)
2756                         goto out_fail;
2757         }
2758         ret = btrfs_add_link(trans, new_dentry, old_inode);
2759         if (ret)
2760                 goto out_fail;
2761
2762 out_fail:
2763         btrfs_free_path(path);
2764         btrfs_end_transaction(trans, root);
2765 out_unlock:
2766         mutex_unlock(&root->fs_info->fs_mutex);
2767         return ret;
2768 }
2769
2770 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2771                          const char *symname)
2772 {
2773         struct btrfs_trans_handle *trans;
2774         struct btrfs_root *root = BTRFS_I(dir)->root;
2775         struct btrfs_path *path;
2776         struct btrfs_key key;
2777         struct inode *inode = NULL;
2778         int err;
2779         int drop_inode = 0;
2780         u64 objectid;
2781         int name_len;
2782         int datasize;
2783         unsigned long ptr;
2784         struct btrfs_file_extent_item *ei;
2785         struct extent_buffer *leaf;
2786         unsigned long nr = 0;
2787
2788         name_len = strlen(symname) + 1;
2789         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2790                 return -ENAMETOOLONG;
2791
2792         mutex_lock(&root->fs_info->fs_mutex);
2793         err = btrfs_check_free_space(root, 1, 0);
2794         if (err)
2795                 goto out_fail;
2796
2797         trans = btrfs_start_transaction(root, 1);
2798         btrfs_set_trans_block_group(trans, dir);
2799
2800         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2801         if (err) {
2802                 err = -ENOSPC;
2803                 goto out_unlock;
2804         }
2805
2806         inode = btrfs_new_inode(trans, root, objectid,
2807                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2808         err = PTR_ERR(inode);
2809         if (IS_ERR(inode))
2810                 goto out_unlock;
2811
2812         btrfs_set_trans_block_group(trans, inode);
2813         err = btrfs_add_nondir(trans, dentry, inode);
2814         if (err)
2815                 drop_inode = 1;
2816         else {
2817                 inode->i_mapping->a_ops = &btrfs_aops;
2818                 inode->i_fop = &btrfs_file_operations;
2819                 inode->i_op = &btrfs_file_inode_operations;
2820                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2821                                      inode->i_mapping, GFP_NOFS);
2822                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2823         }
2824         dir->i_sb->s_dirt = 1;
2825         btrfs_update_inode_block_group(trans, inode);
2826         btrfs_update_inode_block_group(trans, dir);
2827         if (drop_inode)
2828                 goto out_unlock;
2829
2830         path = btrfs_alloc_path();
2831         BUG_ON(!path);
2832         key.objectid = inode->i_ino;
2833         key.offset = 0;
2834         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2835         datasize = btrfs_file_extent_calc_inline_size(name_len);
2836         err = btrfs_insert_empty_item(trans, root, path, &key,
2837                                       datasize);
2838         if (err) {
2839                 drop_inode = 1;
2840                 goto out_unlock;
2841         }
2842         leaf = path->nodes[0];
2843         ei = btrfs_item_ptr(leaf, path->slots[0],
2844                             struct btrfs_file_extent_item);
2845         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2846         btrfs_set_file_extent_type(leaf, ei,
2847                                    BTRFS_FILE_EXTENT_INLINE);
2848         ptr = btrfs_file_extent_inline_start(ei);
2849         write_extent_buffer(leaf, symname, ptr, name_len);
2850         btrfs_mark_buffer_dirty(leaf);
2851         btrfs_free_path(path);
2852
2853         inode->i_op = &btrfs_symlink_inode_operations;
2854         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2855         inode->i_size = name_len - 1;
2856         err = btrfs_update_inode(trans, root, inode);
2857         if (err)
2858                 drop_inode = 1;
2859
2860 out_unlock:
2861         nr = trans->blocks_used;
2862         btrfs_end_transaction(trans, root);
2863 out_fail:
2864         mutex_unlock(&root->fs_info->fs_mutex);
2865         if (drop_inode) {
2866                 inode_dec_link_count(inode);
2867                 iput(inode);
2868         }
2869         btrfs_btree_balance_dirty(root, nr);
2870         btrfs_throttle(root);
2871         return err;
2872 }
2873 static int btrfs_permission(struct inode *inode, int mask,
2874                             struct nameidata *nd)
2875 {
2876         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2877                 return -EACCES;
2878         return generic_permission(inode, mask, NULL);
2879 }
2880
2881 static struct inode_operations btrfs_dir_inode_operations = {
2882         .lookup         = btrfs_lookup,
2883         .create         = btrfs_create,
2884         .unlink         = btrfs_unlink,
2885         .link           = btrfs_link,
2886         .mkdir          = btrfs_mkdir,
2887         .rmdir          = btrfs_rmdir,
2888         .rename         = btrfs_rename,
2889         .symlink        = btrfs_symlink,
2890         .setattr        = btrfs_setattr,
2891         .mknod          = btrfs_mknod,
2892         .setxattr       = generic_setxattr,
2893         .getxattr       = generic_getxattr,
2894         .listxattr      = btrfs_listxattr,
2895         .removexattr    = generic_removexattr,
2896         .permission     = btrfs_permission,
2897 };
2898 static struct inode_operations btrfs_dir_ro_inode_operations = {
2899         .lookup         = btrfs_lookup,
2900         .permission     = btrfs_permission,
2901 };
2902 static struct file_operations btrfs_dir_file_operations = {
2903         .llseek         = generic_file_llseek,
2904         .read           = generic_read_dir,
2905         .readdir        = btrfs_readdir,
2906         .unlocked_ioctl = btrfs_ioctl,
2907 #ifdef CONFIG_COMPAT
2908         .compat_ioctl   = btrfs_ioctl,
2909 #endif
2910 };
2911
2912 static struct extent_map_ops btrfs_extent_map_ops = {
2913         .fill_delalloc = run_delalloc_range,
2914         .writepage_io_hook = btrfs_writepage_io_hook,
2915         .readpage_io_hook = btrfs_readpage_io_hook,
2916         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2917 };
2918
2919 static struct address_space_operations btrfs_aops = {
2920         .readpage       = btrfs_readpage,
2921         .writepage      = btrfs_writepage,
2922         .writepages     = btrfs_writepages,
2923         .readpages      = btrfs_readpages,
2924         .sync_page      = block_sync_page,
2925         .bmap           = btrfs_bmap,
2926         .invalidatepage = btrfs_invalidatepage,
2927         .releasepage    = btrfs_releasepage,
2928         .set_page_dirty = __set_page_dirty_nobuffers,
2929 };
2930
2931 static struct address_space_operations btrfs_symlink_aops = {
2932         .readpage       = btrfs_readpage,
2933         .writepage      = btrfs_writepage,
2934         .invalidatepage = btrfs_invalidatepage,
2935         .releasepage    = btrfs_releasepage,
2936 };
2937
2938 static struct inode_operations btrfs_file_inode_operations = {
2939         .truncate       = btrfs_truncate,
2940         .getattr        = btrfs_getattr,
2941         .setattr        = btrfs_setattr,
2942         .setxattr       = generic_setxattr,
2943         .getxattr       = generic_getxattr,
2944         .listxattr      = btrfs_listxattr,
2945         .removexattr    = generic_removexattr,
2946         .permission     = btrfs_permission,
2947 };
2948 static struct inode_operations btrfs_special_inode_operations = {
2949         .getattr        = btrfs_getattr,
2950         .setattr        = btrfs_setattr,
2951         .permission     = btrfs_permission,
2952 };
2953 static struct inode_operations btrfs_symlink_inode_operations = {
2954         .readlink       = generic_readlink,
2955         .follow_link    = page_follow_link_light,
2956         .put_link       = page_put_link,
2957         .permission     = btrfs_permission,
2958 };