SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/buffer_head.h>
  20 #include <linux/fs.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/string.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mpage.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/version.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "ioctl.h"
  40 #include "print-tree.h"
  41
  42 struct btrfs_iget_args {
  43         u64 ino;
  44         struct btrfs_root *root;
  45 };
  46
  47 static struct inode_operations btrfs_dir_inode_operations;
  48 static struct inode_operations btrfs_symlink_inode_operations;
  49 static struct inode_operations btrfs_dir_ro_inode_operations;
  50 static struct inode_operations btrfs_special_inode_operations;
  51 static struct inode_operations btrfs_file_inode_operations;
  52 static struct address_space_operations btrfs_aops;
  53 static struct address_space_operations btrfs_symlink_aops;
  54 static struct file_operations btrfs_dir_file_operations;
  55
  56 static struct kmem_cache *btrfs_inode_cachep;
  57 struct kmem_cache *btrfs_trans_handle_cachep;
  58 struct kmem_cache *btrfs_transaction_cachep;
  59 struct kmem_cache *btrfs_bit_radix_cachep;
  60 struct kmem_cache *btrfs_path_cachep;
  61
  62 #define S_SHIFT 12
  63 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  64         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  65         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  66         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  67         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  68         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  69         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  70         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  71 };
  72
  73 void btrfs_read_locked_inode(struct inode *inode)
  74 {
  75         struct btrfs_path *path;
  76         struct btrfs_inode_item *inode_item;
  77         struct btrfs_root *root = BTRFS_I(inode)->root;
  78         struct btrfs_key location;
  79         u64 alloc_group_block;
  80         u32 rdev;
  81         int ret;
  82
  83         path = btrfs_alloc_path();
  84         BUG_ON(!path);
  85         mutex_lock(&root->fs_info->fs_mutex);
  86
  87         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
  88         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
  89         if (ret) {
  90                 btrfs_free_path(path);
  91                 goto make_bad;
  92         }
  93         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
  94                                   path->slots[0],
  95                                   struct btrfs_inode_item);
  96
  97         inode->i_mode = btrfs_inode_mode(inode_item);
  98         inode->i_nlink = btrfs_inode_nlink(inode_item);
  99         inode->i_uid = btrfs_inode_uid(inode_item);
 100         inode->i_gid = btrfs_inode_gid(inode_item);
 101         inode->i_size = btrfs_inode_size(inode_item);
 102         inode->i_atime.tv_sec = btrfs_timespec_sec(&inode_item->atime);
 103         inode->i_atime.tv_nsec = btrfs_timespec_nsec(&inode_item->atime);
 104         inode->i_mtime.tv_sec = btrfs_timespec_sec(&inode_item->mtime);
 105         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(&inode_item->mtime);
 106         inode->i_ctime.tv_sec = btrfs_timespec_sec(&inode_item->ctime);
 107         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(&inode_item->ctime);
 108         inode->i_blocks = btrfs_inode_nblocks(inode_item);
 109         inode->i_generation = btrfs_inode_generation(inode_item);
 110         inode->i_rdev = 0;
 111         rdev = btrfs_inode_rdev(inode_item);
 112         alloc_group_block = btrfs_inode_block_group(inode_item);
 113         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 114                                                        alloc_group_block);
 115
 116         btrfs_free_path(path);
 117         inode_item = NULL;
 118
 119         mutex_unlock(&root->fs_info->fs_mutex);
 120
 121         switch (inode->i_mode & S_IFMT) {
 122         case S_IFREG:
 123                 inode->i_mapping->a_ops = &btrfs_aops;
 124                 inode->i_fop = &btrfs_file_operations;
 125                 inode->i_op = &btrfs_file_inode_operations;
 126                 break;
 127         case S_IFDIR:
 128                 inode->i_fop = &btrfs_dir_file_operations;
 129                 if (root == root->fs_info->tree_root)
 130                         inode->i_op = &btrfs_dir_ro_inode_operations;
 131                 else
 132                         inode->i_op = &btrfs_dir_inode_operations;
 133                 break;
 134         case S_IFLNK:
 135                 inode->i_op = &btrfs_symlink_inode_operations;
 136                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 137                 break;
 138         default:
 139                 init_special_inode(inode, inode->i_mode, rdev);
 140                 break;
 141         }
 142         return;
 143
 144 make_bad:
 145         btrfs_release_path(root, path);
 146         btrfs_free_path(path);
 147         mutex_unlock(&root->fs_info->fs_mutex);
 148         make_bad_inode(inode);
 149 }
 150
 151 static void fill_inode_item(struct btrfs_inode_item *item,
 152                             struct inode *inode)
 153 {
 154         btrfs_set_inode_uid(item, inode->i_uid);
 155         btrfs_set_inode_gid(item, inode->i_gid);
 156         btrfs_set_inode_size(item, inode->i_size);
 157         btrfs_set_inode_mode(item, inode->i_mode);
 158         btrfs_set_inode_nlink(item, inode->i_nlink);
 159         btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
 160         btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
 161         btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
 162         btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
 163         btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
 164         btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
 165         btrfs_set_inode_nblocks(item, inode->i_blocks);
 166         btrfs_set_inode_generation(item, inode->i_generation);
 167         btrfs_set_inode_rdev(item, inode->i_rdev);
 168         btrfs_set_inode_block_group(item,
 169                                     BTRFS_I(inode)->block_group->key.objectid);
 170 }
 171
 172 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
 173                               struct btrfs_root *root,
 174                               struct inode *inode)
 175 {
 176         struct btrfs_inode_item *inode_item;
 177         struct btrfs_path *path;
 178         int ret;
 179
 180         path = btrfs_alloc_path();
 181         BUG_ON(!path);
 182         ret = btrfs_lookup_inode(trans, root, path,
 183                                  &BTRFS_I(inode)->location, 1);
 184         if (ret) {
 185                 if (ret > 0)
 186                         ret = -ENOENT;
 187                 goto failed;
 188         }
 189
 190         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 191                                   path->slots[0],
 192                                   struct btrfs_inode_item);
 193
 194         fill_inode_item(inode_item, inode);
 195         btrfs_mark_buffer_dirty(path->nodes[0]);
 196         btrfs_set_inode_last_trans(trans, inode);
 197         ret = 0;
 198 failed:
 199         btrfs_release_path(root, path);
 200         btrfs_free_path(path);
 201         return ret;
 202 }
 203
 204
 205 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 206                               struct btrfs_root *root,
 207                               struct inode *dir,
 208                               struct dentry *dentry)
 209 {
 210         struct btrfs_path *path;
 211         const char *name = dentry->d_name.name;
 212         int name_len = dentry->d_name.len;
 213         int ret = 0;
 214         u64 objectid;
 215         struct btrfs_dir_item *di;
 216
 217         path = btrfs_alloc_path();
 218         if (!path) {
 219                 ret = -ENOMEM;
 220                 goto err;
 221         }
 222
 223         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 224                                     name, name_len, -1);
 225         if (IS_ERR(di)) {
 226                 ret = PTR_ERR(di);
 227                 goto err;
 228         }
 229         if (!di) {
 230                 ret = -ENOENT;
 231                 goto err;
 232         }
 233         objectid = btrfs_disk_key_objectid(&di->location);
 234         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 235         if (ret)
 236                 goto err;
 237         btrfs_release_path(root, path);
 238
 239         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 240                                          objectid, name, name_len, -1);
 241         if (IS_ERR(di)) {
 242                 ret = PTR_ERR(di);
 243                 goto err;
 244         }
 245         if (!di) {
 246                 ret = -ENOENT;
 247                 goto err;
 248         }
 249         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 250
 251         dentry->d_inode->i_ctime = dir->i_ctime;
 252 err:
 253         btrfs_free_path(path);
 254         if (!ret) {
 255                 dir->i_size -= name_len * 2;
 256                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 257                 btrfs_update_inode(trans, root, dir);
 258                 drop_nlink(dentry->d_inode);
 259                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 260                 dir->i_sb->s_dirt = 1;
 261         }
 262         return ret;
 263 }
 264
 265 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 266 {
 267         struct btrfs_root *root;
 268         struct btrfs_trans_handle *trans;
 269         int ret;
 270
 271         root = BTRFS_I(dir)->root;
 272         mutex_lock(&root->fs_info->fs_mutex);
 273         trans = btrfs_start_transaction(root, 1);
 274         btrfs_set_trans_block_group(trans, dir);
 275         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 276         btrfs_end_transaction(trans, root);
 277         mutex_unlock(&root->fs_info->fs_mutex);
 278         btrfs_btree_balance_dirty(root);
 279         return ret;
 280 }
 281
 282 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 283 {
 284         struct inode *inode = dentry->d_inode;
 285         int err;
 286         int ret;
 287         struct btrfs_root *root = BTRFS_I(dir)->root;
 288         struct btrfs_path *path;
 289         struct btrfs_key key;
 290         struct btrfs_trans_handle *trans;
 291         struct btrfs_key found_key;
 292         int found_type;
 293         struct btrfs_leaf *leaf;
 294         char *goodnames = "..";
 295
 296         path = btrfs_alloc_path();
 297         BUG_ON(!path);
 298         mutex_lock(&root->fs_info->fs_mutex);
 299         trans = btrfs_start_transaction(root, 1);
 300         btrfs_set_trans_block_group(trans, dir);
 301         key.objectid = inode->i_ino;
 302         key.offset = (u64)-1;
 303         key.flags = (u32)-1;
 304         while(1) {
 305                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 306                 if (ret < 0) {
 307                         err = ret;
 308                         goto out;
 309                 }
 310                 BUG_ON(ret == 0);
 311                 if (path->slots[0] == 0) {
 312                         err = -ENOENT;
 313                         goto out;
 314                 }
 315                 path->slots[0]--;
 316                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 317                 btrfs_disk_key_to_cpu(&found_key,
 318                                       &leaf->items[path->slots[0]].key);
 319                 found_type = btrfs_key_type(&found_key);
 320                 if (found_key.objectid != inode->i_ino) {
 321                         err = -ENOENT;
 322                         goto out;
 323                 }
 324                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
 325                      found_type != BTRFS_DIR_INDEX_KEY) ||
 326                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
 327                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
 328                         err = -ENOTEMPTY;
 329                         goto out;
 330                 }
 331                 ret = btrfs_del_item(trans, root, path);
 332                 BUG_ON(ret);
 333
 334                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
 335                         break;
 336                 btrfs_release_path(root, path);
 337         }
 338         ret = 0;
 339         btrfs_release_path(root, path);
 340
 341         /* now the directory is empty */
 342         err = btrfs_unlink_trans(trans, root, dir, dentry);
 343         if (!err) {
 344                 inode->i_size = 0;
 345         }
 346 out:
 347         btrfs_release_path(root, path);
 348         btrfs_free_path(path);
 349         mutex_unlock(&root->fs_info->fs_mutex);
 350         ret = btrfs_end_transaction(trans, root);
 351         btrfs_btree_balance_dirty(root);
 352         if (ret && !err)
 353                 err = ret;
 354         return err;
 355 }
 356
 357 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 358                             struct btrfs_root *root,
 359                             struct inode *inode)
 360 {
 361         struct btrfs_path *path;
 362         int ret;
 363
 364         clear_inode(inode);
 365
 366         path = btrfs_alloc_path();
 367         BUG_ON(!path);
 368         ret = btrfs_lookup_inode(trans, root, path,
 369                                  &BTRFS_I(inode)->location, -1);
 370         if (ret > 0)
 371                 ret = -ENOENT;
 372         if (!ret)
 373                 ret = btrfs_del_item(trans, root, path);
 374         btrfs_free_path(path);
 375         return ret;
 376 }
 377
 378 /*
 379  * this can truncate away extent items, csum items and directory items.
 380  * It starts at a high offset and removes keys until it can't find
 381  * any higher than i_size.
 382  *
 383  * csum items that cross the new i_size are truncated to the new size
 384  * as well.
 385  */
 386 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 387                                    struct btrfs_root *root,
 388                                    struct inode *inode)
 389 {
 390         int ret;
 391         struct btrfs_path *path;
 392         struct btrfs_key key;
 393         struct btrfs_disk_key *found_key;
 394         u32 found_type;
 395         struct btrfs_leaf *leaf;
 396         struct btrfs_file_extent_item *fi;
 397         u64 extent_start = 0;
 398         u64 extent_num_blocks = 0;
 399         u64 item_end = 0;
 400         int found_extent;
 401         int del_item;
 402
 403         path = btrfs_alloc_path();
 404         path->reada = -1;
 405         BUG_ON(!path);
 406         /* FIXME, add redo link to tree so we don't leak on crash */
 407         key.objectid = inode->i_ino;
 408         key.offset = (u64)-1;
 409         key.flags = (u32)-1;
 410         while(1) {
 411                 btrfs_init_path(path);
 412                 fi = NULL;
 413                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 414                 if (ret < 0) {
 415                         goto error;
 416                 }
 417                 if (ret > 0) {
 418                         BUG_ON(path->slots[0] == 0);
 419                         path->slots[0]--;
 420                 }
 421                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 422                 found_key = &leaf->items[path->slots[0]].key;
 423                 found_type = btrfs_disk_key_type(found_key);
 424
 425                 if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
 426                         break;
 427                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 428                     found_type != BTRFS_DIR_ITEM_KEY &&
 429                     found_type != BTRFS_DIR_INDEX_KEY &&
 430                     found_type != BTRFS_EXTENT_DATA_KEY)
 431                         break;
 432
 433                 item_end = btrfs_disk_key_offset(found_key);
 434                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 435                         fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 436                                             path->slots[0],
 437                                             struct btrfs_file_extent_item);
 438                         if (btrfs_file_extent_type(fi) !=
 439                             BTRFS_FILE_EXTENT_INLINE) {
 440                                 item_end += btrfs_file_extent_num_blocks(fi) <<
 441                                                 inode->i_blkbits;
 442                         }
 443                 }
 444                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 445                         ret = btrfs_csum_truncate(trans, root, path,
 446                                                   inode->i_size);
 447                         BUG_ON(ret);
 448                 }
 449                 if (item_end < inode->i_size) {
 450                         if (found_type) {
 451                                 btrfs_set_key_type(&key, found_type - 1);
 452                                 continue;
 453                         }
 454                         break;
 455                 }
 456                 if (btrfs_disk_key_offset(found_key) >= inode->i_size)
 457                         del_item = 1;
 458                 else
 459                         del_item = 0;
 460                 found_extent = 0;
 461
 462                 /* FIXME, shrink the extent if the ref count is only 1 */
 463                 if (found_type == BTRFS_EXTENT_DATA_KEY &&
 464                            btrfs_file_extent_type(fi) !=
 465                            BTRFS_FILE_EXTENT_INLINE) {
 466                         u64 num_dec;
 467                         if (!del_item) {
 468                                 u64 orig_num_blocks =
 469                                         btrfs_file_extent_num_blocks(fi);
 470                                 extent_num_blocks = inode->i_size -
 471                                         btrfs_disk_key_offset(found_key) +
 472                                         root->blocksize - 1;
 473                                 extent_num_blocks >>= inode->i_blkbits;
 474                                 btrfs_set_file_extent_num_blocks(fi,
 475                                                          extent_num_blocks);
 476                                 inode->i_blocks -= (orig_num_blocks -
 477                                         extent_num_blocks) << 3;
 478                                 btrfs_mark_buffer_dirty(path->nodes[0]);
 479                         } else {
 480                                 extent_start =
 481                                         btrfs_file_extent_disk_blocknr(fi);
 482                                 extent_num_blocks =
 483                                         btrfs_file_extent_disk_num_blocks(fi);
 484                                 /* FIXME blocksize != 4096 */
 485                                 num_dec = btrfs_file_extent_num_blocks(fi) << 3;
 486                                 if (extent_start != 0) {
 487                                         found_extent = 1;
 488                                         inode->i_blocks -= num_dec;
 489                                 }
 490                         }
 491                 }
 492                 if (del_item) {
 493                         ret = btrfs_del_item(trans, root, path);
 494                         if (ret)
 495                                 goto error;
 496                 } else {
 497                         break;
 498                 }
 499                 btrfs_release_path(root, path);
 500                 if (found_extent) {
 501                         ret = btrfs_free_extent(trans, root, extent_start,
 502                                                 extent_num_blocks, 0);
 503                         BUG_ON(ret);
 504                 }
 505         }
 506         ret = 0;
 507 error:
 508         btrfs_release_path(root, path);
 509         btrfs_free_path(path);
 510         inode->i_sb->s_dirt = 1;
 511         return ret;
 512 }
 513
 514 /*
 515  * taken from block_truncate_page, but does cow as it zeros out
 516  * any bytes left in the last page in the file.
 517  */
 518 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 519 {
 520         struct inode *inode = mapping->host;
 521         unsigned blocksize = 1 << inode->i_blkbits;
 522         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 523         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 524         struct page *page;
 525         char *kaddr;
 526         int ret = 0;
 527         struct btrfs_root *root = BTRFS_I(inode)->root;
 528         u64 alloc_hint = 0;
 529         struct btrfs_key ins;
 530         struct btrfs_trans_handle *trans;
 531
 532         if ((offset & (blocksize - 1)) == 0)
 533                 goto out;
 534
 535         ret = -ENOMEM;
 536         page = grab_cache_page(mapping, index);
 537         if (!page)
 538                 goto out;
 539
 540         if (!PageUptodate(page)) {
 541                 ret = btrfs_readpage(NULL, page);
 542                 lock_page(page);
 543                 if (!PageUptodate(page)) {
 544                         ret = -EIO;
 545                         goto out;
 546                 }
 547         }
 548         mutex_lock(&root->fs_info->fs_mutex);
 549         trans = btrfs_start_transaction(root, 1);
 550         btrfs_set_trans_block_group(trans, inode);
 551
 552         ret = btrfs_drop_extents(trans, root, inode,
 553                                  page->index << PAGE_CACHE_SHIFT,
 554                                  (page->index + 1) << PAGE_CACHE_SHIFT,
 555                                  &alloc_hint);
 556         if (ret)
 557                 goto out;
 558         ret = btrfs_alloc_extent(trans, root, inode->i_ino, 1, 0,
 559                                  alloc_hint, (u64)-1, &ins, 1);
 560         if (ret)
 561                 goto out;
 562         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 563                                        page->index << PAGE_CACHE_SHIFT,
 564                                        ins.objectid, 1, 1);
 565         if (ret)
 566                 goto out;
 567         SetPageChecked(page);
 568         kaddr = kmap(page);
 569         memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
 570         flush_dcache_page(page);
 571         ret = btrfs_csum_file_block(trans, root, inode->i_ino,
 572                               page->index << PAGE_CACHE_SHIFT,
 573                               kaddr, PAGE_CACHE_SIZE);
 574         kunmap(page);
 575         btrfs_end_transaction(trans, root);
 576         mutex_unlock(&root->fs_info->fs_mutex);
 577
 578         set_page_dirty(page);
 579         unlock_page(page);
 580         page_cache_release(page);
 581 out:
 582         return ret;
 583 }
 584
 585 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 586 {
 587         struct inode *inode = dentry->d_inode;
 588         int err;
 589
 590         err = inode_change_ok(inode, attr);
 591         if (err)
 592                 return err;
 593
 594         if (S_ISREG(inode->i_mode) &&
 595             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 596                 struct btrfs_trans_handle *trans;
 597                 struct btrfs_root *root = BTRFS_I(inode)->root;
 598                 u64 mask = root->blocksize - 1;
 599                 u64 pos = (inode->i_size + mask) & ~mask;
 600                 u64 hole_size;
 601
 602                 if (attr->ia_size <= pos)
 603                         goto out;
 604
 605                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 606
 607                 hole_size = (attr->ia_size - pos + mask) & ~mask;
 608                 hole_size >>= inode->i_blkbits;
 609
 610                 mutex_lock(&root->fs_info->fs_mutex);
 611                 trans = btrfs_start_transaction(root, 1);
 612                 btrfs_set_trans_block_group(trans, inode);
 613                 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
 614                                                pos, 0, 0, hole_size);
 615                 btrfs_end_transaction(trans, root);
 616                 mutex_unlock(&root->fs_info->fs_mutex);
 617                 if (err)
 618                         return err;
 619         }
 620 out:
 621         err = inode_setattr(inode, attr);
 622
 623         return err;
 624 }
 625 void btrfs_delete_inode(struct inode *inode)
 626 {
 627         struct btrfs_trans_handle *trans;
 628         struct btrfs_root *root = BTRFS_I(inode)->root;
 629         int ret;
 630
 631         truncate_inode_pages(&inode->i_data, 0);
 632         if (is_bad_inode(inode)) {
 633                 goto no_delete;
 634         }
 635         inode->i_size = 0;
 636         mutex_lock(&root->fs_info->fs_mutex);
 637         trans = btrfs_start_transaction(root, 1);
 638         btrfs_set_trans_block_group(trans, inode);
 639         ret = btrfs_truncate_in_trans(trans, root, inode);
 640         if (ret)
 641                 goto no_delete_lock;
 642         ret = btrfs_free_inode(trans, root, inode);
 643         if (ret)
 644                 goto no_delete_lock;
 645         btrfs_end_transaction(trans, root);
 646         mutex_unlock(&root->fs_info->fs_mutex);
 647         btrfs_btree_balance_dirty(root);
 648         return;
 649
 650 no_delete_lock:
 651         btrfs_end_transaction(trans, root);
 652         mutex_unlock(&root->fs_info->fs_mutex);
 653         btrfs_btree_balance_dirty(root);
 654 no_delete:
 655         clear_inode(inode);
 656 }
 657
 658 /*
 659  * this returns the key found in the dir entry in the location pointer.
 660  * If no dir entries were found, location->objectid is 0.
 661  */
 662 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 663                                struct btrfs_key *location)
 664 {
 665         const char *name = dentry->d_name.name;
 666         int namelen = dentry->d_name.len;
 667         struct btrfs_dir_item *di;
 668         struct btrfs_path *path;
 669         struct btrfs_root *root = BTRFS_I(dir)->root;
 670         int ret;
 671
 672         path = btrfs_alloc_path();
 673         BUG_ON(!path);
 674         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
 675                                     namelen, 0);
 676         if (!di || IS_ERR(di)) {
 677                 location->objectid = 0;
 678                 ret = 0;
 679                 goto out;
 680         }
 681         btrfs_disk_key_to_cpu(location, &di->location);
 682 out:
 683         btrfs_release_path(root, path);
 684         btrfs_free_path(path);
 685         return ret;
 686 }
 687
 688 /*
 689  * when we hit a tree root in a directory, the btrfs part of the inode
 690  * needs to be changed to reflect the root directory of the tree root.  This
 691  * is kind of like crossing a mount point.
 692  */
 693 static int fixup_tree_root_location(struct btrfs_root *root,
 694                              struct btrfs_key *location,
 695                              struct btrfs_root **sub_root)
 696 {
 697         struct btrfs_path *path;
 698         struct btrfs_root_item *ri;
 699
 700         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 701                 return 0;
 702         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 703                 return 0;
 704
 705         path = btrfs_alloc_path();
 706         BUG_ON(!path);
 707         mutex_lock(&root->fs_info->fs_mutex);
 708
 709         *sub_root = btrfs_read_fs_root(root->fs_info, location);
 710         if (IS_ERR(*sub_root))
 711                 return PTR_ERR(*sub_root);
 712
 713         ri = &(*sub_root)->root_item;
 714         location->objectid = btrfs_root_dirid(ri);
 715         location->flags = 0;
 716         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 717         location->offset = 0;
 718
 719         btrfs_free_path(path);
 720         mutex_unlock(&root->fs_info->fs_mutex);
 721         return 0;
 722 }
 723
 724 static int btrfs_init_locked_inode(struct inode *inode, void *p)
 725 {
 726         struct btrfs_iget_args *args = p;
 727         inode->i_ino = args->ino;
 728         BTRFS_I(inode)->root = args->root;
 729         return 0;
 730 }
 731
 732 static int btrfs_find_actor(struct inode *inode, void *opaque)
 733 {
 734         struct btrfs_iget_args *args = opaque;
 735         return (args->ino == inode->i_ino &&
 736                 args->root == BTRFS_I(inode)->root);
 737 }
 738
 739 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 740                                 struct btrfs_root *root)
 741 {
 742         struct inode *inode;
 743         struct btrfs_iget_args args;
 744         args.ino = objectid;
 745         args.root = root;
 746
 747         inode = iget5_locked(s, objectid, btrfs_find_actor,
 748                              btrfs_init_locked_inode,
 749                              (void *)&args);
 750         return inode;
 751 }
 752
 753 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 754                                    struct nameidata *nd)
 755 {
 756         struct inode * inode;
 757         struct btrfs_inode *bi = BTRFS_I(dir);
 758         struct btrfs_root *root = bi->root;
 759         struct btrfs_root *sub_root = root;
 760         struct btrfs_key location;
 761         int ret;
 762
 763         if (dentry->d_name.len > BTRFS_NAME_LEN)
 764                 return ERR_PTR(-ENAMETOOLONG);
 765         mutex_lock(&root->fs_info->fs_mutex);
 766         ret = btrfs_inode_by_name(dir, dentry, &location);
 767         mutex_unlock(&root->fs_info->fs_mutex);
 768         if (ret < 0)
 769                 return ERR_PTR(ret);
 770         inode = NULL;
 771         if (location.objectid) {
 772                 ret = fixup_tree_root_location(root, &location, &sub_root);
 773                 if (ret < 0)
 774                         return ERR_PTR(ret);
 775                 if (ret > 0)
 776                         return ERR_PTR(-ENOENT);
 777                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 778                                           sub_root);
 779                 if (!inode)
 780                         return ERR_PTR(-EACCES);
 781                 if (inode->i_state & I_NEW) {
 782                         /* the inode and parent dir are two different roots */
 783                         if (sub_root != root) {
 784                                 igrab(inode);
 785                                 sub_root->inode = inode;
 786                         }
 787                         BTRFS_I(inode)->root = sub_root;
 788                         memcpy(&BTRFS_I(inode)->location, &location,
 789                                sizeof(location));
 790                         btrfs_read_locked_inode(inode);
 791                         unlock_new_inode(inode);
 792                 }
 793         }
 794         return d_splice_alias(inode, dentry);
 795 }
 796
 797 static unsigned char btrfs_filetype_table[] = {
 798         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
 799 };
 800
 801 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 802 {
 803         struct inode *inode = filp->f_path.dentry->d_inode;
 804         struct btrfs_root *root = BTRFS_I(inode)->root;
 805         struct btrfs_item *item;
 806         struct btrfs_dir_item *di;
 807         struct btrfs_key key;
 808         struct btrfs_path *path;
 809         int ret;
 810         u32 nritems;
 811         struct btrfs_leaf *leaf;
 812         int slot;
 813         int advance;
 814         unsigned char d_type;
 815         int over = 0;
 816         u32 di_cur;
 817         u32 di_total;
 818         u32 di_len;
 819         int key_type = BTRFS_DIR_INDEX_KEY;
 820
 821         /* FIXME, use a real flag for deciding about the key type */
 822         if (root->fs_info->tree_root == root)
 823                 key_type = BTRFS_DIR_ITEM_KEY;
 824         mutex_lock(&root->fs_info->fs_mutex);
 825         key.objectid = inode->i_ino;
 826         key.flags = 0;
 827         btrfs_set_key_type(&key, key_type);
 828         key.offset = filp->f_pos;
 829         path = btrfs_alloc_path();
 830         path->reada = 2;
 831         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 832         if (ret < 0)
 833                 goto err;
 834         advance = 0;
 835         while(1) {
 836                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 837                 nritems = btrfs_header_nritems(&leaf->header);
 838                 slot = path->slots[0];
 839                 if (advance || slot >= nritems) {
 840                         if (slot >= nritems -1) {
 841                                 ret = btrfs_next_leaf(root, path);
 842                                 if (ret)
 843                                         break;
 844                                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 845                                 nritems = btrfs_header_nritems(&leaf->header);
 846                                 slot = path->slots[0];
 847                         } else {
 848                                 slot++;
 849                                 path->slots[0]++;
 850                         }
 851                 }
 852                 advance = 1;
 853                 item = leaf->items + slot;
 854                 if (btrfs_disk_key_objectid(&item->key) != key.objectid)
 855                         break;
 856                 if (btrfs_disk_key_type(&item->key) != key_type)
 857                         break;
 858                 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
 859                         continue;
 860                 filp->f_pos = btrfs_disk_key_offset(&item->key);
 861                 advance = 1;
 862                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 863                 di_cur = 0;
 864                 di_total = btrfs_item_size(leaf->items + slot);
 865                 while(di_cur < di_total) {
 866                         d_type = btrfs_filetype_table[btrfs_dir_type(di)];
 867                         over = filldir(dirent, (const char *)(di + 1),
 868                                        btrfs_dir_name_len(di),
 869                                        btrfs_disk_key_offset(&item->key),
 870                                        btrfs_disk_key_objectid(&di->location),
 871                                        d_type);
 872                         if (over)
 873                                 goto nopos;
 874                         di_len = btrfs_dir_name_len(di) + sizeof(*di);
 875                         di_cur += di_len;
 876                         di = (struct btrfs_dir_item *)((char *)di + di_len);
 877                 }
 878         }
 879         filp->f_pos++;
 880 nopos:
 881         ret = 0;
 882 err:
 883         btrfs_release_path(root, path);
 884         btrfs_free_path(path);
 885         mutex_unlock(&root->fs_info->fs_mutex);
 886         return ret;
 887 }
 888
 889 int btrfs_write_inode(struct inode *inode, int wait)
 890 {
 891         struct btrfs_root *root = BTRFS_I(inode)->root;
 892         struct btrfs_trans_handle *trans;
 893         int ret = 0;
 894
 895         if (wait) {
 896                 mutex_lock(&root->fs_info->fs_mutex);
 897                 trans = btrfs_start_transaction(root, 1);
 898                 btrfs_set_trans_block_group(trans, inode);
 899                 ret = btrfs_commit_transaction(trans, root);
 900                 mutex_unlock(&root->fs_info->fs_mutex);
 901         }
 902         return ret;
 903 }
 904
 905 /*
 906  * This is somewhat expensive, updating the tree every time the
 907  * inode changes.  But, it is most likely to find the inode in cache.
 908  * FIXME, needs more benchmarking...there are no reasons other than performance
 909  * to keep or drop this code.
 910  */
 911 void btrfs_dirty_inode(struct inode *inode)
 912 {
 913         struct btrfs_root *root = BTRFS_I(inode)->root;
 914         struct btrfs_trans_handle *trans;
 915
 916         mutex_lock(&root->fs_info->fs_mutex);
 917         trans = btrfs_start_transaction(root, 1);
 918         btrfs_set_trans_block_group(trans, inode);
 919         btrfs_update_inode(trans, root, inode);
 920         btrfs_end_transaction(trans, root);
 921         mutex_unlock(&root->fs_info->fs_mutex);
 922 }
 923
 924 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 925                                      struct btrfs_root *root,
 926                                      u64 objectid,
 927                                      struct btrfs_block_group_cache *group,
 928                                      int mode)
 929 {
 930         struct inode *inode;
 931         struct btrfs_inode_item inode_item;
 932         struct btrfs_key *location;
 933         int ret;
 934         int owner;
 935
 936         inode = new_inode(root->fs_info->sb);
 937         if (!inode)
 938                 return ERR_PTR(-ENOMEM);
 939
 940         BTRFS_I(inode)->root = root;
 941         if (mode & S_IFDIR)
 942                 owner = 0;
 943         else
 944                 owner = 1;
 945         group = btrfs_find_block_group(root, group, 0, 0, owner);
 946         BTRFS_I(inode)->block_group = group;
 947
 948         inode->i_uid = current->fsuid;
 949         inode->i_gid = current->fsgid;
 950         inode->i_mode = mode;
 951         inode->i_ino = objectid;
 952         inode->i_blocks = 0;
 953         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 954         fill_inode_item(&inode_item, inode);
 955         location = &BTRFS_I(inode)->location;
 956         location->objectid = objectid;
 957         location->flags = 0;
 958         location->offset = 0;
 959         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 960
 961         ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
 962         if (ret)
 963                 return ERR_PTR(ret);
 964         insert_inode_hash(inode);
 965         return inode;
 966 }
 967
 968 static inline u8 btrfs_inode_type(struct inode *inode)
 969 {
 970         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
 971 }
 972
 973 static int btrfs_add_link(struct btrfs_trans_handle *trans,
 974                             struct dentry *dentry, struct inode *inode)
 975 {
 976         int ret;
 977         struct btrfs_key key;
 978         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
 979         struct inode *parent_inode;
 980         key.objectid = inode->i_ino;
 981         key.flags = 0;
 982         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 983         key.offset = 0;
 984
 985         ret = btrfs_insert_dir_item(trans, root,
 986                                     dentry->d_name.name, dentry->d_name.len,
 987                                     dentry->d_parent->d_inode->i_ino,
 988                                     &key, btrfs_inode_type(inode));
 989         if (ret == 0) {
 990                 parent_inode = dentry->d_parent->d_inode;
 991                 parent_inode->i_size += dentry->d_name.len * 2;
 992                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
 993                 ret = btrfs_update_inode(trans, root,
 994                                          dentry->d_parent->d_inode);
 995         }
 996         return ret;
 997 }
 998
 999 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1000                             struct dentry *dentry, struct inode *inode)
1001 {
1002         int err = btrfs_add_link(trans, dentry, inode);
1003         if (!err) {
1004                 d_instantiate(dentry, inode);
1005                 return 0;
1006         }
1007         if (err > 0)
1008                 err = -EEXIST;
1009         return err;
1010 }
1011
1012 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1013                         int mode, dev_t rdev)
1014 {
1015         struct btrfs_trans_handle *trans;
1016         struct btrfs_root *root = BTRFS_I(dir)->root;
1017         struct inode *inode;
1018         int err;
1019         int drop_inode = 0;
1020         u64 objectid;
1021
1022         if (!new_valid_dev(rdev))
1023                 return -EINVAL;
1024
1025         mutex_lock(&root->fs_info->fs_mutex);
1026         trans = btrfs_start_transaction(root, 1);
1027         btrfs_set_trans_block_group(trans, dir);
1028
1029         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1030         if (err) {
1031                 err = -ENOSPC;
1032                 goto out_unlock;
1033         }
1034
1035         inode = btrfs_new_inode(trans, root, objectid,
1036                                 BTRFS_I(dir)->block_group, mode);
1037         err = PTR_ERR(inode);
1038         if (IS_ERR(inode))
1039                 goto out_unlock;
1040
1041         btrfs_set_trans_block_group(trans, inode);
1042         err = btrfs_add_nondir(trans, dentry, inode);
1043         if (err)
1044                 drop_inode = 1;
1045         else {
1046                 inode->i_op = &btrfs_special_inode_operations;
1047                 init_special_inode(inode, inode->i_mode, rdev);
1048         }
1049         dir->i_sb->s_dirt = 1;
1050         btrfs_update_inode_block_group(trans, inode);
1051         btrfs_update_inode_block_group(trans, dir);
1052 out_unlock:
1053         btrfs_end_transaction(trans, root);
1054         mutex_unlock(&root->fs_info->fs_mutex);
1055
1056         if (drop_inode) {
1057                 inode_dec_link_count(inode);
1058                 iput(inode);
1059         }
1060         btrfs_btree_balance_dirty(root);
1061         return err;
1062 }
1063
1064 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1065                         int mode, struct nameidata *nd)
1066 {
1067         struct btrfs_trans_handle *trans;
1068         struct btrfs_root *root = BTRFS_I(dir)->root;
1069         struct inode *inode;
1070         int err;
1071         int drop_inode = 0;
1072         u64 objectid;
1073
1074         mutex_lock(&root->fs_info->fs_mutex);
1075         trans = btrfs_start_transaction(root, 1);
1076         btrfs_set_trans_block_group(trans, dir);
1077
1078         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1079         if (err) {
1080                 err = -ENOSPC;
1081                 goto out_unlock;
1082         }
1083
1084         inode = btrfs_new_inode(trans, root, objectid,
1085                                 BTRFS_I(dir)->block_group, mode);
1086         err = PTR_ERR(inode);
1087         if (IS_ERR(inode))
1088                 goto out_unlock;
1089
1090         btrfs_set_trans_block_group(trans, inode);
1091         err = btrfs_add_nondir(trans, dentry, inode);
1092         if (err)
1093                 drop_inode = 1;
1094         else {
1095                 inode->i_mapping->a_ops = &btrfs_aops;
1096                 inode->i_fop = &btrfs_file_operations;
1097                 inode->i_op = &btrfs_file_inode_operations;
1098         }
1099         dir->i_sb->s_dirt = 1;
1100         btrfs_update_inode_block_group(trans, inode);
1101         btrfs_update_inode_block_group(trans, dir);
1102 out_unlock:
1103         btrfs_end_transaction(trans, root);
1104         mutex_unlock(&root->fs_info->fs_mutex);
1105
1106         if (drop_inode) {
1107                 inode_dec_link_count(inode);
1108                 iput(inode);
1109         }
1110         btrfs_btree_balance_dirty(root);
1111         return err;
1112 }
1113
1114 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1115                       struct dentry *dentry)
1116 {
1117         struct btrfs_trans_handle *trans;
1118         struct btrfs_root *root = BTRFS_I(dir)->root;
1119         struct inode *inode = old_dentry->d_inode;
1120         int err;
1121         int drop_inode = 0;
1122
1123         if (inode->i_nlink == 0)
1124                 return -ENOENT;
1125
1126         inc_nlink(inode);
1127         mutex_lock(&root->fs_info->fs_mutex);
1128         trans = btrfs_start_transaction(root, 1);
1129         btrfs_set_trans_block_group(trans, dir);
1130         atomic_inc(&inode->i_count);
1131         err = btrfs_add_nondir(trans, dentry, inode);
1132         if (err)
1133                 drop_inode = 1;
1134         dir->i_sb->s_dirt = 1;
1135         btrfs_update_inode_block_group(trans, dir);
1136         err = btrfs_update_inode(trans, root, inode);
1137         if (err)
1138                 drop_inode = 1;
1139
1140         btrfs_end_transaction(trans, root);
1141         mutex_unlock(&root->fs_info->fs_mutex);
1142
1143         if (drop_inode) {
1144                 inode_dec_link_count(inode);
1145                 iput(inode);
1146         }
1147         btrfs_btree_balance_dirty(root);
1148         return err;
1149 }
1150
1151 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
1152                                 struct btrfs_root *root,
1153                                 u64 objectid, u64 dirid)
1154 {
1155         int ret;
1156         char buf[2];
1157         struct btrfs_key key;
1158
1159         buf[0] = '.';
1160         buf[1] = '.';
1161
1162         key.objectid = objectid;
1163         key.offset = 0;
1164         key.flags = 0;
1165         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1166
1167         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
1168                                     &key, BTRFS_FT_DIR);
1169         if (ret)
1170                 goto error;
1171         key.objectid = dirid;
1172         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
1173                                     &key, BTRFS_FT_DIR);
1174         if (ret)
1175                 goto error;
1176 error:
1177         return ret;
1178 }
1179
1180 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1181 {
1182         struct inode *inode;
1183         struct btrfs_trans_handle *trans;
1184         struct btrfs_root *root = BTRFS_I(dir)->root;
1185         int err = 0;
1186         int drop_on_err = 0;
1187         u64 objectid;
1188
1189         mutex_lock(&root->fs_info->fs_mutex);
1190         trans = btrfs_start_transaction(root, 1);
1191         btrfs_set_trans_block_group(trans, dir);
1192         if (IS_ERR(trans)) {
1193                 err = PTR_ERR(trans);
1194                 goto out_unlock;
1195         }
1196
1197         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1198         if (err) {
1199                 err = -ENOSPC;
1200                 goto out_unlock;
1201         }
1202
1203         inode = btrfs_new_inode(trans, root, objectid,
1204                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1205         if (IS_ERR(inode)) {
1206                 err = PTR_ERR(inode);
1207                 goto out_fail;
1208         }
1209         drop_on_err = 1;
1210         inode->i_op = &btrfs_dir_inode_operations;
1211         inode->i_fop = &btrfs_dir_file_operations;
1212         btrfs_set_trans_block_group(trans, inode);
1213
1214         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
1215         if (err)
1216                 goto out_fail;
1217
1218         inode->i_size = 6;
1219         err = btrfs_update_inode(trans, root, inode);
1220         if (err)
1221                 goto out_fail;
1222         err = btrfs_add_link(trans, dentry, inode);
1223         if (err)
1224                 goto out_fail;
1225         d_instantiate(dentry, inode);
1226         drop_on_err = 0;
1227         dir->i_sb->s_dirt = 1;
1228         btrfs_update_inode_block_group(trans, inode);
1229         btrfs_update_inode_block_group(trans, dir);
1230
1231 out_fail:
1232         btrfs_end_transaction(trans, root);
1233 out_unlock:
1234         mutex_unlock(&root->fs_info->fs_mutex);
1235         if (drop_on_err)
1236                 iput(inode);
1237         btrfs_btree_balance_dirty(root);
1238         return err;
1239 }
1240
1241 /*
1242  * FIBMAP and others want to pass in a fake buffer head.  They need to
1243  * use BTRFS_GET_BLOCK_NO_DIRECT to make sure we don't try to memcpy
1244  * any packed file data into the fake bh
1245  */
1246 #define BTRFS_GET_BLOCK_NO_CREATE 0
1247 #define BTRFS_GET_BLOCK_CREATE 1
1248 #define BTRFS_GET_BLOCK_NO_DIRECT 2
1249
1250 /*
1251  * FIXME create==1 doe not work.
1252  */
1253 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
1254                                 struct buffer_head *result, int create)
1255 {
1256         int ret;
1257         int err = 0;
1258         u64 blocknr;
1259         u64 extent_start = 0;
1260         u64 extent_end = 0;
1261         u64 objectid = inode->i_ino;
1262         u32 found_type;
1263         u64 alloc_hint = 0;
1264         struct btrfs_path *path;
1265         struct btrfs_root *root = BTRFS_I(inode)->root;
1266         struct btrfs_file_extent_item *item;
1267         struct btrfs_leaf *leaf;
1268         struct btrfs_disk_key *found_key;
1269         struct btrfs_trans_handle *trans = NULL;
1270
1271         path = btrfs_alloc_path();
1272         BUG_ON(!path);
1273         if (create & BTRFS_GET_BLOCK_CREATE) {
1274                 /*
1275                  * danger!, this only works if the page is properly up
1276                  * to date somehow
1277                  */
1278                 trans = btrfs_start_transaction(root, 1);
1279                 if (!trans) {
1280                         err = -ENOMEM;
1281                         goto out;
1282                 }
1283                 ret = btrfs_drop_extents(trans, root, inode,
1284                                          iblock << inode->i_blkbits,
1285                                          (iblock + 1) << inode->i_blkbits,
1286                                          &alloc_hint);
1287                 BUG_ON(ret);
1288         }
1289
1290         ret = btrfs_lookup_file_extent(NULL, root, path,
1291                                        objectid,
1292                                        iblock << inode->i_blkbits, 0);
1293         if (ret < 0) {
1294                 err = ret;
1295                 goto out;
1296         }
1297
1298         if (ret != 0) {
1299                 if (path->slots[0] == 0) {
1300                         btrfs_release_path(root, path);
1301                         goto not_found;
1302                 }
1303                 path->slots[0]--;
1304         }
1305
1306         item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
1307                               struct btrfs_file_extent_item);
1308         leaf = btrfs_buffer_leaf(path->nodes[0]);
1309         blocknr = btrfs_file_extent_disk_blocknr(item);
1310         blocknr += btrfs_file_extent_offset(item);
1311
1312         /* are we inside the extent that was found? */
1313         found_key = &leaf->items[path->slots[0]].key;
1314         found_type = btrfs_disk_key_type(found_key);
1315         if (btrfs_disk_key_objectid(found_key) != objectid ||
1316             found_type != BTRFS_EXTENT_DATA_KEY) {
1317                 extent_end = 0;
1318                 extent_start = 0;
1319                 goto not_found;
1320         }
1321         found_type = btrfs_file_extent_type(item);
1322         extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
1323         if (found_type == BTRFS_FILE_EXTENT_REG) {
1324                 extent_start = extent_start >> inode->i_blkbits;
1325                 extent_end = extent_start + btrfs_file_extent_num_blocks(item);
1326                 err = 0;
1327                 if (btrfs_file_extent_disk_blocknr(item) == 0)
1328                         goto out;
1329                 if (iblock >= extent_start && iblock < extent_end) {
1330                         btrfs_map_bh_to_logical(root, result, blocknr +
1331                                                 iblock - extent_start);
1332                         goto out;
1333                 }
1334         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1335                 char *ptr;
1336                 char *map;
1337                 u32 size;
1338
1339                 if (create & BTRFS_GET_BLOCK_NO_DIRECT) {
1340                         err = -EINVAL;
1341                         goto out;
1342                 }
1343                 size = btrfs_file_extent_inline_len(leaf->items +
1344                                                     path->slots[0]);
1345                 extent_end = (extent_start + size) >> inode->i_blkbits;
1346                 extent_start >>= inode->i_blkbits;
1347                 if (iblock < extent_start || iblock > extent_end) {
1348                         goto not_found;
1349                 }
1350                 ptr = btrfs_file_extent_inline_start(item);
1351                 map = kmap(result->b_page);
1352                 memcpy(map, ptr, size);
1353                 memset(map + size, 0, PAGE_CACHE_SIZE - size);
1354                 flush_dcache_page(result->b_page);
1355                 kunmap(result->b_page);
1356                 set_buffer_uptodate(result);
1357                 SetPageChecked(result->b_page);
1358                 btrfs_map_bh_to_logical(root, result, 0);
1359         }
1360 not_found:
1361         if (create & BTRFS_GET_BLOCK_CREATE) {
1362                 struct btrfs_key ins;
1363                 ret = btrfs_alloc_extent(trans, root, inode->i_ino,
1364                                          1, 0, alloc_hint, (u64)-1,
1365                                          &ins, 1);
1366                 if (ret) {
1367                         err = ret;
1368                         goto out;
1369                 }
1370                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
1371                                                iblock << inode->i_blkbits,
1372                                                ins.objectid, ins.offset,
1373                                                ins.offset);
1374                 if (ret) {
1375                         err = ret;
1376                         goto out;
1377                 }
1378                 btrfs_map_bh_to_logical(root, result, ins.objectid);
1379         }
1380 out:
1381         if (trans) {
1382                 ret = btrfs_end_transaction(trans, root);
1383                 if (!err)
1384                         err = ret;
1385         }
1386         btrfs_free_path(path);
1387         return err;
1388 }
1389
1390 int btrfs_get_block(struct inode *inode, sector_t iblock,
1391                     struct buffer_head *result, int create)
1392 {
1393         int err;
1394         struct btrfs_root *root = BTRFS_I(inode)->root;
1395         mutex_lock(&root->fs_info->fs_mutex);
1396         err = btrfs_get_block_lock(inode, iblock, result, create);
1397         mutex_unlock(&root->fs_info->fs_mutex);
1398         return err;
1399 }
1400
1401 static int btrfs_get_block_csum(struct inode *inode, sector_t iblock,
1402                                 struct buffer_head *result, int create)
1403 {
1404         int ret;
1405         struct btrfs_root *root = BTRFS_I(inode)->root;
1406         struct page *page = result->b_page;
1407         u64 offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(result);
1408         struct btrfs_csum_item *item;
1409         struct btrfs_path *path = NULL;
1410
1411         mutex_lock(&root->fs_info->fs_mutex);
1412         ret = btrfs_get_block_lock(inode, iblock, result, create);
1413         if (ret)
1414                 goto out;
1415
1416         path = btrfs_alloc_path();
1417         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, offset, 0);
1418         if (IS_ERR(item)) {
1419                 ret = PTR_ERR(item);
1420                 /* a csum that isn't present is a preallocated region. */
1421                 if (ret == -ENOENT || ret == -EFBIG)
1422                         ret = 0;
1423                 result->b_private = NULL;
1424                 goto out;
1425         }
1426         memcpy((char *)&result->b_private, &item->csum, BTRFS_CRC32_SIZE);
1427 out:
1428         if (path)
1429                 btrfs_free_path(path);
1430         mutex_unlock(&root->fs_info->fs_mutex);
1431         return ret;
1432 }
1433
1434 static int btrfs_get_block_bmap(struct inode *inode, sector_t iblock,
1435                            struct buffer_head *result, int create)
1436 {
1437         struct btrfs_root *root = BTRFS_I(inode)->root;
1438         mutex_lock(&root->fs_info->fs_mutex);
1439         btrfs_get_block_lock(inode, iblock, result, BTRFS_GET_BLOCK_NO_DIRECT);
1440         mutex_unlock(&root->fs_info->fs_mutex);
1441         return 0;
1442 }
1443
1444 static sector_t btrfs_bmap(struct address_space *as, sector_t block)
1445 {
1446         return generic_block_bmap(as, block, btrfs_get_block_bmap);
1447 }
1448
1449 static int btrfs_prepare_write(struct file *file, struct page *page,
1450                                unsigned from, unsigned to)
1451 {
1452         return block_prepare_write(page, from, to, btrfs_get_block);
1453 }
1454
1455 static void buffer_io_error(struct buffer_head *bh)
1456 {
1457         char b[BDEVNAME_SIZE];
1458
1459         printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n",
1460                         bdevname(bh->b_bdev, b),
1461                         (unsigned long long)bh->b_blocknr);
1462 }
1463
1464 /*
1465  * I/O completion handler for block_read_full_page() - pages
1466  * which come unlocked at the end of I/O.
1467  */
1468 static void btrfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
1469 {
1470         unsigned long flags;
1471         struct buffer_head *first;
1472         struct buffer_head *tmp;
1473         struct page *page;
1474         int page_uptodate = 1;
1475         struct inode *inode;
1476         int ret;
1477
1478         BUG_ON(!buffer_async_read(bh));
1479
1480         page = bh->b_page;
1481         inode = page->mapping->host;
1482         if (uptodate) {
1483                 void *kaddr;
1484                 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1485                 if (bh->b_private) {
1486                         char csum[BTRFS_CRC32_SIZE];
1487                         kaddr = kmap_atomic(page, KM_IRQ0);
1488                         ret = btrfs_csum_data(root, kaddr + bh_offset(bh),
1489                                               bh->b_size, csum);
1490                         BUG_ON(ret);
1491                         if (memcmp(csum, &bh->b_private, BTRFS_CRC32_SIZE)) {
1492                                 u64 offset;
1493                                 offset = (page->index << PAGE_CACHE_SHIFT) +
1494                                         bh_offset(bh);
1495                                 printk("btrfs csum failed ino %lu off %llu\n",
1496                                        page->mapping->host->i_ino,
1497                                        (unsigned long long)offset);
1498                                 memset(kaddr + bh_offset(bh), 1, bh->b_size);
1499                                 flush_dcache_page(page);
1500                         }
1501                         kunmap_atomic(kaddr, KM_IRQ0);
1502                 }
1503                 set_buffer_uptodate(bh);
1504         } else {
1505                 clear_buffer_uptodate(bh);
1506                 if (printk_ratelimit())
1507                         buffer_io_error(bh);
1508                 SetPageError(page);
1509         }
1510
1511         /*
1512          * Be _very_ careful from here on. Bad things can happen if
1513          * two buffer heads end IO at almost the same time and both
1514          * decide that the page is now completely done.
1515          */
1516         first = page_buffers(page);
1517         local_irq_save(flags);
1518         bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
1519         clear_buffer_async_read(bh);
1520         unlock_buffer(bh);
1521         tmp = bh;
1522         do {
1523                 if (!buffer_uptodate(tmp))
1524                         page_uptodate = 0;
1525                 if (buffer_async_read(tmp)) {
1526                         BUG_ON(!buffer_locked(tmp));
1527                         goto still_busy;
1528                 }
1529                 tmp = tmp->b_this_page;
1530         } while (tmp != bh);
1531         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
1532         local_irq_restore(flags);
1533
1534         /*
1535          * If none of the buffers had errors and they are all
1536          * uptodate then we can set the page uptodate.
1537          */
1538         if (page_uptodate && !PageError(page))
1539                 SetPageUptodate(page);
1540         unlock_page(page);
1541         return;
1542
1543 still_busy:
1544         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
1545         local_irq_restore(flags);
1546         return;
1547 }
1548
1549 /*
1550  * Generic "read page" function for block devices that have the normal
1551  * get_block functionality. This is most of the block device filesystems.
1552  * Reads the page asynchronously --- the unlock_buffer() and
1553  * set/clear_buffer_uptodate() functions propagate buffer state into the
1554  * page struct once IO has completed.
1555  */
1556 int btrfs_readpage(struct file *file, struct page *page)
1557 {
1558         struct inode *inode = page->mapping->host;
1559         sector_t iblock, lblock;
1560         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
1561         unsigned int blocksize;
1562         int nr, i;
1563         int fully_mapped = 1;
1564
1565         BUG_ON(!PageLocked(page));
1566         blocksize = 1 << inode->i_blkbits;
1567         if (!page_has_buffers(page))
1568                 create_empty_buffers(page, blocksize, 0);
1569         head = page_buffers(page);
1570
1571         iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1572         lblock = (i_size_read(inode)+blocksize-1) >> inode->i_blkbits;
1573         bh = head;
1574         nr = 0;
1575         i = 0;
1576
1577         do {
1578                 if (buffer_uptodate(bh))
1579                         continue;
1580
1581                 if (!buffer_mapped(bh)) {
1582                         int err = 0;
1583
1584                         fully_mapped = 0;
1585                         if (iblock < lblock) {
1586                                 WARN_ON(bh->b_size != blocksize);
1587                                 err = btrfs_get_block_csum(inode, iblock,
1588                                                            bh, 0);
1589                                 if (err)
1590                                         SetPageError(page);
1591                         }
1592                         if (!buffer_mapped(bh)) {
1593                                 void *kaddr = kmap_atomic(page, KM_USER0);
1594                                 memset(kaddr + i * blocksize, 0, blocksize);
1595                                 flush_dcache_page(page);
1596                                 kunmap_atomic(kaddr, KM_USER0);
1597                                 if (!err)
1598                                         set_buffer_uptodate(bh);
1599                                 continue;
1600                         }
1601                         /*
1602                          * get_block() might have updated the buffer
1603                          * synchronously
1604                          */
1605                         if (buffer_uptodate(bh))
1606                                 continue;
1607                 }
1608                 arr[nr++] = bh;
1609         } while (i++, iblock++, (bh = bh->b_this_page) != head);
1610
1611         if (fully_mapped)
1612                 SetPageMappedToDisk(page);
1613
1614         if (!nr) {
1615                 /*
1616                  * All buffers are uptodate - we can set the page uptodate
1617                  * as well. But not if get_block() returned an error.
1618                  */
1619                 if (!PageError(page))
1620                         SetPageUptodate(page);
1621                 unlock_page(page);
1622                 return 0;
1623         }
1624
1625         /* Stage two: lock the buffers */
1626         for (i = 0; i < nr; i++) {
1627                 bh = arr[i];
1628                 lock_buffer(bh);
1629                 bh->b_end_io = btrfs_end_buffer_async_read;
1630                 set_buffer_async_read(bh);
1631         }
1632
1633         /*
1634          * Stage 3: start the IO.  Check for uptodateness
1635          * inside the buffer lock in case another process reading
1636          * the underlying blockdev brought it uptodate (the sct fix).
1637          */
1638         for (i = 0; i < nr; i++) {
1639                 bh = arr[i];
1640                 if (buffer_uptodate(bh))
1641                         btrfs_end_buffer_async_read(bh, 1);
1642                 else
1643                         submit_bh(READ, bh);
1644         }
1645         return 0;
1646 }
1647
1648 /*
1649  * Aside from a tiny bit of packed file data handling, this is the
1650  * same as the generic code.
1651  *
1652  * While block_write_full_page is writing back the dirty buffers under
1653  * the page lock, whoever dirtied the buffers may decide to clean them
1654  * again at any time.  We handle that by only looking at the buffer
1655  * state inside lock_buffer().
1656  *
1657  * If block_write_full_page() is called for regular writeback
1658  * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1659  * locked buffer.   This only can happen if someone has written the buffer
1660  * directly, with submit_bh().  At the address_space level PageWriteback
1661  * prevents this contention from occurring.
1662  */
1663 static int __btrfs_write_full_page(struct inode *inode, struct page *page,
1664                                    struct writeback_control *wbc)
1665 {
1666         int err;
1667         sector_t block;
1668         sector_t last_block;
1669         struct buffer_head *bh, *head;
1670         const unsigned blocksize = 1 << inode->i_blkbits;
1671         int nr_underway = 0;
1672         struct btrfs_root *root = BTRFS_I(inode)->root;
1673
1674         BUG_ON(!PageLocked(page));
1675
1676         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
1677
1678         /* no csumming allowed when from PF_MEMALLOC */
1679         if (current->flags & PF_MEMALLOC) {
1680                 redirty_page_for_writepage(wbc, page);
1681                 unlock_page(page);
1682                 return 0;
1683         }
1684
1685         if (!page_has_buffers(page)) {
1686                 create_empty_buffers(page, blocksize,
1687                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
1688         }
1689
1690         /*
1691          * Be very careful.  We have no exclusion from __set_page_dirty_buffers
1692          * here, and the (potentially unmapped) buffers may become dirty at
1693          * any time.  If a buffer becomes dirty here after we've inspected it
1694          * then we just miss that fact, and the page stays dirty.
1695          *
1696          * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1697          * handle that here by just cleaning them.
1698          */
1699
1700         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1701         head = page_buffers(page);
1702         bh = head;
1703
1704         /*
1705          * Get all the dirty buffers mapped to disk addresses and
1706          * handle any aliases from the underlying blockdev's mapping.
1707          */
1708         do {
1709                 if (block > last_block) {
1710                         /*
1711                          * mapped buffers outside i_size will occur, because
1712                          * this page can be outside i_size when there is a
1713                          * truncate in progress.
1714                          */
1715                         /*
1716                          * The buffer was zeroed by block_write_full_page()
1717                          */
1718                         clear_buffer_dirty(bh);
1719                         set_buffer_uptodate(bh);
1720                 } else if (!buffer_mapped(bh) && buffer_dirty(bh)) {
1721                         WARN_ON(bh->b_size != blocksize);
1722                         err = btrfs_get_block(inode, block, bh, 0);
1723                         if (err) {
1724                                 goto recover;
1725                         }
1726                         if (buffer_new(bh)) {
1727                                 /* blockdev mappings never come here */
1728                                 clear_buffer_new(bh);
1729                         }
1730                 }
1731                 bh = bh->b_this_page;
1732                 block++;
1733         } while (bh != head);
1734
1735         do {
1736                 if (!buffer_mapped(bh))
1737                         continue;
1738                 /*
1739                  * If it's a fully non-blocking write attempt and we cannot
1740                  * lock the buffer then redirty the page.  Note that this can
1741                  * potentially cause a busy-wait loop from pdflush and kswapd
1742                  * activity, but those code paths have their own higher-level
1743                  * throttling.
1744                  */
1745                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
1746                         lock_buffer(bh);
1747                 } else if (test_set_buffer_locked(bh)) {
1748                         redirty_page_for_writepage(wbc, page);
1749                         continue;
1750                 }
1751                 if (test_clear_buffer_dirty(bh) && bh->b_blocknr != 0) {
1752                         struct btrfs_trans_handle *trans;
1753                         int ret;
1754                         u64 off = page->index << PAGE_CACHE_SHIFT;
1755                         char *kaddr;
1756
1757                         off += bh_offset(bh);
1758                         mutex_lock(&root->fs_info->fs_mutex);
1759                         trans = btrfs_start_transaction(root, 1);
1760                         btrfs_set_trans_block_group(trans, inode);
1761                         kaddr = kmap(page);
1762                         btrfs_csum_file_block(trans, root, inode->i_ino,
1763                                                     off, kaddr + bh_offset(bh),
1764                                                     bh->b_size);
1765                         kunmap(page);
1766                         ret = btrfs_end_transaction(trans, root);
1767                         BUG_ON(ret);
1768                         mutex_unlock(&root->fs_info->fs_mutex);
1769                         mark_buffer_async_write(bh);
1770                 } else {
1771                         unlock_buffer(bh);
1772                 }
1773         } while ((bh = bh->b_this_page) != head);
1774
1775         /*
1776          * The page and its buffers are protected by PageWriteback(), so we can
1777          * drop the bh refcounts early.
1778          */
1779         BUG_ON(PageWriteback(page));
1780         set_page_writeback(page);
1781
1782         do {
1783                 struct buffer_head *next = bh->b_this_page;
1784                 if (buffer_async_write(bh)) {
1785                         submit_bh(WRITE, bh);
1786                         nr_underway++;
1787                 }
1788                 bh = next;
1789         } while (bh != head);
1790         unlock_page(page);
1791
1792         err = 0;
1793 done:
1794         if (nr_underway == 0) {
1795                 /*
1796                  * The page was marked dirty, but the buffers were
1797                  * clean.  Someone wrote them back by hand with
1798                  * ll_rw_block/submit_bh.  A rare case.
1799                  */
1800                 int uptodate = 1;
1801                 do {
1802                         if (!buffer_uptodate(bh)) {
1803                                 uptodate = 0;
1804                                 break;
1805                         }
1806                         bh = bh->b_this_page;
1807                 } while (bh != head);
1808                 if (uptodate)
1809                         SetPageUptodate(page);
1810                 end_page_writeback(page);
1811         }
1812         return err;
1813
1814 recover:
1815         /*
1816          * ENOSPC, or some other error.  We may already have added some
1817          * blocks to the file, so we need to write these out to avoid
1818          * exposing stale data.
1819          * The page is currently locked and not marked for writeback
1820          */
1821         bh = head;
1822         /* Recovery: lock and submit the mapped buffers */
1823         do {
1824                 if (buffer_mapped(bh) && buffer_dirty(bh)) {
1825                         lock_buffer(bh);
1826                         mark_buffer_async_write(bh);
1827                 } else {
1828                         /*
1829                          * The buffer may have been set dirty during
1830                          * attachment to a dirty page.
1831                          */
1832                         clear_buffer_dirty(bh);
1833                 }
1834         } while ((bh = bh->b_this_page) != head);
1835         SetPageError(page);
1836         BUG_ON(PageWriteback(page));
1837         set_page_writeback(page);
1838         do {
1839                 struct buffer_head *next = bh->b_this_page;
1840                 if (buffer_async_write(bh)) {
1841                         clear_buffer_dirty(bh);
1842                         submit_bh(WRITE, bh);
1843                         nr_underway++;
1844                 }
1845                 bh = next;
1846         } while (bh != head);
1847         unlock_page(page);
1848         goto done;
1849 }
1850
1851 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1852 {
1853         struct inode * const inode = page->mapping->host;
1854         loff_t i_size = i_size_read(inode);
1855         const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1856         unsigned offset;
1857         void *kaddr;
1858
1859         /* Is the page fully inside i_size? */
1860         if (page->index < end_index)
1861                 return __btrfs_write_full_page(inode, page, wbc);
1862
1863         /* Is the page fully outside i_size? (truncate in progress) */
1864         offset = i_size & (PAGE_CACHE_SIZE-1);
1865         if (page->index >= end_index+1 || !offset) {
1866                 /*
1867                  * The page may have dirty, unmapped buffers.  For example,
1868                  * they may have been added in ext3_writepage().  Make them
1869                  * freeable here, so the page does not leak.
1870                  */
1871                 block_invalidatepage(page, 0);
1872                 unlock_page(page);
1873                 return 0; /* don't care */
1874         }
1875
1876         /*
1877          * The page straddles i_size.  It must be zeroed out on each and every
1878          * writepage invokation because it may be mmapped.  "A file is mapped
1879          * in multiples of the page size.  For a file that is not a multiple of
1880          * the  page size, the remaining memory is zeroed when mapped, and
1881          * writes to that region are not written out to the file."
1882          */
1883         kaddr = kmap_atomic(page, KM_USER0);
1884         memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1885         flush_dcache_page(page);
1886         kunmap_atomic(kaddr, KM_USER0);
1887         return __btrfs_write_full_page(inode, page, wbc);
1888 }
1889
1890 /*
1891  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
1892  * called from a page fault handler when a page is first dirtied. Hence we must
1893  * be careful to check for EOF conditions here. We set the page up correctly
1894  * for a written page which means we get ENOSPC checking when writing into
1895  * holes and correct delalloc and unwritten extent mapping on filesystems that
1896  * support these features.
1897  *
1898  * We are not allowed to take the i_mutex here so we have to play games to
1899  * protect against truncate races as the page could now be beyond EOF.  Because
1900  * vmtruncate() writes the inode size before removing pages, once we have the
1901  * page lock we can determine safely if the page is beyond EOF. If it is not
1902  * beyond EOF, then the page is guaranteed safe against truncation until we
1903  * unlock the page.
1904  */
1905 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1906 {
1907         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1908         unsigned long end;
1909         loff_t size;
1910         int ret = -EINVAL;
1911
1912         lock_page(page);
1913         wait_on_page_writeback(page);
1914         size = i_size_read(inode);
1915         if ((page->mapping != inode->i_mapping) ||
1916             ((page->index << PAGE_CACHE_SHIFT) > size)) {
1917                 /* page got truncated out from underneath us */
1918                 goto out_unlock;
1919         }
1920
1921         /* page is wholly or partially inside EOF */
1922         if (((page->index + 1) << PAGE_CACHE_SHIFT) > size)
1923                 end = size & ~PAGE_CACHE_MASK;
1924         else
1925                 end = PAGE_CACHE_SIZE;
1926
1927         ret = btrfs_prepare_write(NULL, page, 0, end);
1928         if (!ret)
1929                 ret = btrfs_commit_write(NULL, page, 0, end);
1930
1931 out_unlock:
1932         unlock_page(page);
1933         return ret;
1934 }
1935
1936 static void btrfs_truncate(struct inode *inode)
1937 {
1938         struct btrfs_root *root = BTRFS_I(inode)->root;
1939         int ret;
1940         struct btrfs_trans_handle *trans;
1941
1942         if (!S_ISREG(inode->i_mode))
1943                 return;
1944         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1945                 return;
1946
1947         btrfs_truncate_page(inode->i_mapping, inode->i_size);
1948
1949         mutex_lock(&root->fs_info->fs_mutex);
1950         trans = btrfs_start_transaction(root, 1);
1951         btrfs_set_trans_block_group(trans, inode);
1952
1953         /* FIXME, add redo link to tree so we don't leak on crash */
1954         ret = btrfs_truncate_in_trans(trans, root, inode);
1955         btrfs_update_inode(trans, root, inode);
1956         ret = btrfs_end_transaction(trans, root);
1957         BUG_ON(ret);
1958         mutex_unlock(&root->fs_info->fs_mutex);
1959         btrfs_btree_balance_dirty(root);
1960 }
1961
1962 int btrfs_commit_write(struct file *file, struct page *page,
1963                        unsigned from, unsigned to)
1964 {
1965         struct inode *inode = page->mapping->host;
1966         struct buffer_head *bh;
1967         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1968
1969         SetPageUptodate(page);
1970         bh = page_buffers(page);
1971         set_buffer_uptodate(bh);
1972         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
1973                 set_page_dirty(page);
1974         }
1975         if (pos > inode->i_size) {
1976                 i_size_write(inode, pos);
1977                 mark_inode_dirty(inode);
1978         }
1979         return 0;
1980 }
1981
1982 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
1983 {
1984         struct btrfs_trans_handle *trans;
1985         struct btrfs_key key;
1986         struct btrfs_root_item root_item;
1987         struct btrfs_inode_item *inode_item;
1988         struct buffer_head *subvol;
1989         struct btrfs_leaf *leaf;
1990         struct btrfs_root *new_root;
1991         struct inode *inode;
1992         struct inode *dir;
1993         int ret;
1994         int err;
1995         u64 objectid;
1996         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
1997
1998         mutex_lock(&root->fs_info->fs_mutex);
1999         trans = btrfs_start_transaction(root, 1);
2000         BUG_ON(!trans);
2001
2002         subvol = btrfs_alloc_free_block(trans, root, 0, 0);
2003         if (IS_ERR(subvol))
2004                 return PTR_ERR(subvol);
2005         leaf = btrfs_buffer_leaf(subvol);
2006         btrfs_set_header_nritems(&leaf->header, 0);
2007         btrfs_set_header_level(&leaf->header, 0);
2008         btrfs_set_header_blocknr(&leaf->header, bh_blocknr(subvol));
2009         btrfs_set_header_generation(&leaf->header, trans->transid);
2010         btrfs_set_header_owner(&leaf->header, root->root_key.objectid);
2011         memcpy(leaf->header.fsid, root->fs_info->disk_super->fsid,
2012                sizeof(leaf->header.fsid));
2013         btrfs_mark_buffer_dirty(subvol);
2014
2015         inode_item = &root_item.inode;
2016         memset(inode_item, 0, sizeof(*inode_item));
2017         btrfs_set_inode_generation(inode_item, 1);
2018         btrfs_set_inode_size(inode_item, 3);
2019         btrfs_set_inode_nlink(inode_item, 1);
2020         btrfs_set_inode_nblocks(inode_item, 1);
2021         btrfs_set_inode_mode(inode_item, S_IFDIR | 0755);
2022
2023         btrfs_set_root_blocknr(&root_item, bh_blocknr(subvol));
2024         btrfs_set_root_refs(&root_item, 1);
2025         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2026         root_item.drop_level = 0;
2027         brelse(subvol);
2028         subvol = NULL;
2029
2030         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2031                                        0, &objectid);
2032         if (ret)
2033                 goto fail;
2034
2035         btrfs_set_root_dirid(&root_item, new_dirid);
2036
2037         key.objectid = objectid;
2038         key.offset = 1;
2039         key.flags = 0;
2040         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2041         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2042                                 &root_item);
2043         if (ret)
2044                 goto fail;
2045
2046         /*
2047          * insert the directory item
2048          */
2049         key.offset = (u64)-1;
2050         dir = root->fs_info->sb->s_root->d_inode;
2051         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2052                                     name, namelen, dir->i_ino, &key,
2053                                     BTRFS_FT_DIR);
2054         if (ret)
2055                 goto fail;
2056
2057         ret = btrfs_commit_transaction(trans, root);
2058         if (ret)
2059                 goto fail_commit;
2060
2061         new_root = btrfs_read_fs_root(root->fs_info, &key);
2062         BUG_ON(!new_root);
2063
2064         trans = btrfs_start_transaction(new_root, 1);
2065         BUG_ON(!trans);
2066
2067         inode = btrfs_new_inode(trans, new_root, new_dirid,
2068                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2069         if (IS_ERR(inode))
2070                 goto fail;
2071         inode->i_op = &btrfs_dir_inode_operations;
2072         inode->i_fop = &btrfs_dir_file_operations;
2073         new_root->inode = inode;
2074
2075         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
2076         if (ret)
2077                 goto fail;
2078
2079         inode->i_nlink = 1;
2080         inode->i_size = 6;
2081         ret = btrfs_update_inode(trans, new_root, inode);
2082         if (ret)
2083                 goto fail;
2084 fail:
2085         err = btrfs_commit_transaction(trans, root);
2086         if (err && !ret)
2087                 ret = err;
2088 fail_commit:
2089         mutex_unlock(&root->fs_info->fs_mutex);
2090         btrfs_btree_balance_dirty(root);
2091         return ret;
2092 }
2093
2094 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2095 {
2096         struct btrfs_trans_handle *trans;
2097         struct btrfs_key key;
2098         struct btrfs_root_item new_root_item;
2099         struct buffer_head *tmp;
2100         int ret;
2101         int err;
2102         u64 objectid;
2103
2104         if (!root->ref_cows)
2105                 return -EINVAL;
2106
2107         mutex_lock(&root->fs_info->fs_mutex);
2108         trans = btrfs_start_transaction(root, 1);
2109         BUG_ON(!trans);
2110
2111         ret = btrfs_update_inode(trans, root, root->inode);
2112         if (ret)
2113                 goto fail;
2114
2115         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2116                                        0, &objectid);
2117         if (ret)
2118                 goto fail;
2119
2120         memcpy(&new_root_item, &root->root_item,
2121                sizeof(new_root_item));
2122
2123         key.objectid = objectid;
2124         key.offset = 1;
2125         key.flags = 0;
2126         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2127         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
2128         btrfs_set_root_blocknr(&new_root_item, bh_blocknr(root->node));
2129
2130         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2131                                 &new_root_item);
2132         if (ret)
2133                 goto fail;
2134
2135         /*
2136          * insert the directory item
2137          */
2138         key.offset = (u64)-1;
2139         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2140                                     name, namelen,
2141                                     root->fs_info->sb->s_root->d_inode->i_ino,
2142                                     &key, BTRFS_FT_DIR);
2143
2144         if (ret)
2145                 goto fail;
2146
2147         ret = btrfs_inc_root_ref(trans, root);
2148         if (ret)
2149                 goto fail;
2150
2151 fail:
2152         err = btrfs_commit_transaction(trans, root);
2153         if (err && !ret)
2154                 ret = err;
2155         mutex_unlock(&root->fs_info->fs_mutex);
2156         btrfs_btree_balance_dirty(root);
2157         return ret;
2158 }
2159
2160 int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
2161                 cmd, unsigned long arg)
2162 {
2163         struct btrfs_root *root = BTRFS_I(inode)->root;
2164         struct btrfs_ioctl_vol_args vol_args;
2165         int ret = 0;
2166         struct btrfs_dir_item *di;
2167         int namelen;
2168         struct btrfs_path *path;
2169         u64 root_dirid;
2170
2171         switch (cmd) {
2172         case BTRFS_IOC_SNAP_CREATE:
2173                 if (copy_from_user(&vol_args,
2174                                    (struct btrfs_ioctl_vol_args __user *)arg,
2175                                    sizeof(vol_args)))
2176                         return -EFAULT;
2177                 namelen = strlen(vol_args.name);
2178                 if (namelen > BTRFS_VOL_NAME_MAX)
2179                         return -EINVAL;
2180                 if (strchr(vol_args.name, '/'))
2181                         return -EINVAL;
2182                 path = btrfs_alloc_path();
2183                 if (!path)
2184                         return -ENOMEM;
2185                 root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2186                 mutex_lock(&root->fs_info->fs_mutex);
2187                 di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2188                                     path, root_dirid,
2189                                     vol_args.name, namelen, 0);
2190                 mutex_unlock(&root->fs_info->fs_mutex);
2191                 btrfs_free_path(path);
2192                 if (di && !IS_ERR(di))
2193                         return -EEXIST;
2194                 if (IS_ERR(di))
2195                         return PTR_ERR(di);
2196
2197                 if (root == root->fs_info->tree_root)
2198                         ret = create_subvol(root, vol_args.name, namelen);
2199                 else
2200                         ret = create_snapshot(root, vol_args.name, namelen);
2201                 break;
2202
2203         case BTRFS_IOC_DEFRAG:
2204                 mutex_lock(&root->fs_info->fs_mutex);
2205                 btrfs_defrag_root(root, 0);
2206                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2207                 mutex_unlock(&root->fs_info->fs_mutex);
2208                 ret = 0;
2209                 break;
2210         default:
2211                 return -ENOTTY;
2212         }
2213         return ret;
2214 }
2215
2216 #ifdef CONFIG_COMPAT
2217 long btrfs_compat_ioctl(struct file *file, unsigned int cmd,
2218                                unsigned long arg)
2219 {
2220         struct inode *inode = file->f_path.dentry->d_inode;
2221         int ret;
2222         lock_kernel();
2223         ret = btrfs_ioctl(inode, file, cmd, (unsigned long) compat_ptr(arg));
2224         unlock_kernel();
2225         return ret;
2226
2227 }
2228 #endif
2229
2230 /*
2231  * Called inside transaction, so use GFP_NOFS
2232  */
2233 struct inode *btrfs_alloc_inode(struct super_block *sb)
2234 {
2235         struct btrfs_inode *ei;
2236
2237         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2238         if (!ei)
2239                 return NULL;
2240         ei->last_trans = 0;
2241         return &ei->vfs_inode;
2242 }
2243
2244 void btrfs_destroy_inode(struct inode *inode)
2245 {
2246         WARN_ON(!list_empty(&inode->i_dentry));
2247         WARN_ON(inode->i_data.nrpages);
2248
2249         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2250 }
2251
2252 static void init_once(void * foo, struct kmem_cache * cachep,
2253                       unsigned long flags)
2254 {
2255         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2256
2257         inode_init_once(&ei->vfs_inode);
2258 }
2259
2260 void btrfs_destroy_cachep(void)
2261 {
2262         if (btrfs_inode_cachep)
2263                 kmem_cache_destroy(btrfs_inode_cachep);
2264         if (btrfs_trans_handle_cachep)
2265                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2266         if (btrfs_transaction_cachep)
2267                 kmem_cache_destroy(btrfs_transaction_cachep);
2268         if (btrfs_bit_radix_cachep)
2269                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2270         if (btrfs_path_cachep)
2271                 kmem_cache_destroy(btrfs_path_cachep);
2272 }
2273
2274 static struct kmem_cache *cache_create(const char *name, size_t size,
2275                                        unsigned long extra_flags,
2276                                        void (*ctor)(void *, struct kmem_cache *,
2277                                                     unsigned long))
2278 {
2279         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2280                                  SLAB_MEM_SPREAD | extra_flags), ctor
2281 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2282                                  ,NULL
2283 #endif
2284                                 );
2285 }
2286
2287 int btrfs_init_cachep(void)
2288 {
2289         btrfs_inode_cachep = cache_create("btrfs_inode_cache",
2290                                           sizeof(struct btrfs_inode),
2291                                           0, init_once);
2292         if (!btrfs_inode_cachep)
2293                 goto fail;
2294         btrfs_trans_handle_cachep = cache_create("btrfs_trans_handle_cache",
2295                                              sizeof(struct btrfs_trans_handle),
2296                                              0, NULL);
2297         if (!btrfs_trans_handle_cachep)
2298                 goto fail;
2299         btrfs_transaction_cachep = cache_create("btrfs_transaction_cache",
2300                                              sizeof(struct btrfs_transaction),
2301                                              0, NULL);
2302         if (!btrfs_transaction_cachep)
2303                 goto fail;
2304         btrfs_path_cachep = cache_create("btrfs_path_cache",
2305                                          sizeof(struct btrfs_transaction),
2306                                          0, NULL);
2307         if (!btrfs_path_cachep)
2308                 goto fail;
2309         btrfs_bit_radix_cachep = cache_create("btrfs_radix", 256,
2310                                               SLAB_DESTROY_BY_RCU, NULL);
2311         if (!btrfs_bit_radix_cachep)
2312                 goto fail;
2313         return 0;
2314 fail:
2315         btrfs_destroy_cachep();
2316         return -ENOMEM;
2317 }
2318
2319 static int btrfs_getattr(struct vfsmount *mnt,
2320                          struct dentry *dentry, struct kstat *stat)
2321 {
2322         struct inode *inode = dentry->d_inode;
2323         generic_fillattr(inode, stat);
2324         stat->blksize = 256 * 1024;
2325         return 0;
2326 }
2327
2328 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2329                            struct inode * new_dir,struct dentry *new_dentry)
2330 {
2331         struct btrfs_trans_handle *trans;
2332         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2333         struct inode *new_inode = new_dentry->d_inode;
2334         struct inode *old_inode = old_dentry->d_inode;
2335         struct timespec ctime = CURRENT_TIME;
2336         struct btrfs_path *path;
2337         struct btrfs_dir_item *di;
2338         int ret;
2339
2340         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2341             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2342                 return -ENOTEMPTY;
2343         }
2344         mutex_lock(&root->fs_info->fs_mutex);
2345         trans = btrfs_start_transaction(root, 1);
2346         btrfs_set_trans_block_group(trans, new_dir);
2347         path = btrfs_alloc_path();
2348         if (!path) {
2349                 ret = -ENOMEM;
2350                 goto out_fail;
2351         }
2352
2353         old_dentry->d_inode->i_nlink++;
2354         old_dir->i_ctime = old_dir->i_mtime = ctime;
2355         new_dir->i_ctime = new_dir->i_mtime = ctime;
2356         old_inode->i_ctime = ctime;
2357         if (S_ISDIR(old_inode->i_mode) && old_dir != new_dir) {
2358                 struct btrfs_key *location = &BTRFS_I(new_dir)->location;
2359                 u64 old_parent_oid;
2360                 di = btrfs_lookup_dir_item(trans, root, path, old_inode->i_ino,
2361                                            "..", 2, -1);
2362                 if (IS_ERR(di)) {
2363                         ret = PTR_ERR(di);
2364                         goto out_fail;
2365                 }
2366                 if (!di) {
2367                         ret = -ENOENT;
2368                         goto out_fail;
2369                 }
2370                 old_parent_oid = btrfs_disk_key_objectid(&di->location);
2371                 ret = btrfs_del_item(trans, root, path);
2372                 if (ret) {
2373                         goto out_fail;
2374                 }
2375                 btrfs_release_path(root, path);
2376
2377                 di = btrfs_lookup_dir_index_item(trans, root, path,
2378                                                  old_inode->i_ino,
2379                                                  old_parent_oid,
2380                                                  "..", 2, -1);
2381                 if (IS_ERR(di)) {
2382                         ret = PTR_ERR(di);
2383                         goto out_fail;
2384                 }
2385                 if (!di) {
2386                         ret = -ENOENT;
2387                         goto out_fail;
2388                 }
2389                 ret = btrfs_del_item(trans, root, path);
2390                 if (ret) {
2391                         goto out_fail;
2392                 }
2393                 btrfs_release_path(root, path);
2394
2395                 ret = btrfs_insert_dir_item(trans, root, "..", 2,
2396                                             old_inode->i_ino, location,
2397                                             BTRFS_FT_DIR);
2398                 if (ret)
2399                         goto out_fail;
2400         }
2401
2402
2403         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2404         if (ret)
2405                 goto out_fail;
2406
2407         if (new_inode) {
2408                 new_inode->i_ctime = CURRENT_TIME;
2409                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2410                 if (ret)
2411                         goto out_fail;
2412                 if (S_ISDIR(new_inode->i_mode))
2413                         clear_nlink(new_inode);
2414                 else
2415                         drop_nlink(new_inode);
2416                 ret = btrfs_update_inode(trans, root, new_inode);
2417                 if (ret)
2418                         goto out_fail;
2419         }
2420         ret = btrfs_add_link(trans, new_dentry, old_inode);
2421         if (ret)
2422                 goto out_fail;
2423
2424 out_fail:
2425         btrfs_free_path(path);
2426         btrfs_end_transaction(trans, root);
2427         mutex_unlock(&root->fs_info->fs_mutex);
2428         return ret;
2429 }
2430
2431 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2432                          const char *symname)
2433 {
2434         struct btrfs_trans_handle *trans;
2435         struct btrfs_root *root = BTRFS_I(dir)->root;
2436         struct btrfs_path *path;
2437         struct btrfs_key key;
2438         struct inode *inode;
2439         int err;
2440         int drop_inode = 0;
2441         u64 objectid;
2442         int name_len;
2443         int datasize;
2444         char *ptr;
2445         struct btrfs_file_extent_item *ei;
2446
2447         name_len = strlen(symname) + 1;
2448         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2449                 return -ENAMETOOLONG;
2450         mutex_lock(&root->fs_info->fs_mutex);
2451         trans = btrfs_start_transaction(root, 1);
2452         btrfs_set_trans_block_group(trans, dir);
2453
2454         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2455         if (err) {
2456                 err = -ENOSPC;
2457                 goto out_unlock;
2458         }
2459
2460         inode = btrfs_new_inode(trans, root, objectid,
2461                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2462         err = PTR_ERR(inode);
2463         if (IS_ERR(inode))
2464                 goto out_unlock;
2465
2466         btrfs_set_trans_block_group(trans, inode);
2467         err = btrfs_add_nondir(trans, dentry, inode);
2468         if (err)
2469                 drop_inode = 1;
2470         else {
2471                 inode->i_mapping->a_ops = &btrfs_aops;
2472                 inode->i_fop = &btrfs_file_operations;
2473                 inode->i_op = &btrfs_file_inode_operations;
2474         }
2475         dir->i_sb->s_dirt = 1;
2476         btrfs_update_inode_block_group(trans, inode);
2477         btrfs_update_inode_block_group(trans, dir);
2478         if (drop_inode)
2479                 goto out_unlock;
2480
2481         path = btrfs_alloc_path();
2482         BUG_ON(!path);
2483         key.objectid = inode->i_ino;
2484         key.offset = 0;
2485         key.flags = 0;
2486         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2487         datasize = btrfs_file_extent_calc_inline_size(name_len);
2488         err = btrfs_insert_empty_item(trans, root, path, &key,
2489                                       datasize);
2490         if (err) {
2491                 drop_inode = 1;
2492                 goto out_unlock;
2493         }
2494         ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
2495                path->slots[0], struct btrfs_file_extent_item);
2496         btrfs_set_file_extent_generation(ei, trans->transid);
2497         btrfs_set_file_extent_type(ei,
2498                                    BTRFS_FILE_EXTENT_INLINE);
2499         ptr = btrfs_file_extent_inline_start(ei);
2500         btrfs_memcpy(root, path->nodes[0]->b_data,
2501                      ptr, symname, name_len);
2502         btrfs_mark_buffer_dirty(path->nodes[0]);
2503         btrfs_free_path(path);
2504         inode->i_op = &btrfs_symlink_inode_operations;
2505         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2506         inode->i_size = name_len - 1;
2507         err = btrfs_update_inode(trans, root, inode);
2508         if (err)
2509                 drop_inode = 1;
2510
2511 out_unlock:
2512         btrfs_end_transaction(trans, root);
2513         mutex_unlock(&root->fs_info->fs_mutex);
2514         if (drop_inode) {
2515                 inode_dec_link_count(inode);
2516                 iput(inode);
2517         }
2518         btrfs_btree_balance_dirty(root);
2519         return err;
2520 }
2521
2522 static struct inode_operations btrfs_dir_inode_operations = {
2523         .lookup         = btrfs_lookup,
2524         .create         = btrfs_create,
2525         .unlink         = btrfs_unlink,
2526         .link           = btrfs_link,
2527         .mkdir          = btrfs_mkdir,
2528         .rmdir          = btrfs_rmdir,
2529         .rename         = btrfs_rename,
2530         .symlink        = btrfs_symlink,
2531         .setattr        = btrfs_setattr,
2532         .mknod          = btrfs_mknod,
2533 };
2534
2535 static struct inode_operations btrfs_dir_ro_inode_operations = {
2536         .lookup         = btrfs_lookup,
2537 };
2538
2539 static struct file_operations btrfs_dir_file_operations = {
2540         .llseek         = generic_file_llseek,
2541         .read           = generic_read_dir,
2542         .readdir        = btrfs_readdir,
2543         .ioctl          = btrfs_ioctl,
2544 #ifdef CONFIG_COMPAT
2545         .compat_ioctl   = btrfs_compat_ioctl,
2546 #endif
2547 };
2548
2549 static struct address_space_operations btrfs_aops = {
2550         .readpage       = btrfs_readpage,
2551         .writepage      = btrfs_writepage,
2552         .sync_page      = block_sync_page,
2553         .prepare_write  = btrfs_prepare_write,
2554         .commit_write   = btrfs_commit_write,
2555         .bmap           = btrfs_bmap,
2556 };
2557
2558 static struct address_space_operations btrfs_symlink_aops = {
2559         .readpage       = btrfs_readpage,
2560         .writepage      = btrfs_writepage,
2561 };
2562
2563 static struct inode_operations btrfs_file_inode_operations = {
2564         .truncate       = btrfs_truncate,
2565         .getattr        = btrfs_getattr,
2566         .setattr        = btrfs_setattr,
2567 };
2568
2569 static struct inode_operations btrfs_special_inode_operations = {
2570         .getattr        = btrfs_getattr,
2571         .setattr        = btrfs_setattr,
2572 };
2573
2574 static struct inode_operations btrfs_symlink_inode_operations = {
2575         .readlink       = generic_readlink,
2576         .follow_link    = page_follow_link_light,
2577         .put_link       = page_put_link,
2578 };