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