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