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