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