fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/buffer_head.h>
  20 #include <linux/fs.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/string.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mpage.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/version.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "ioctl.h"
  40 #include "print-tree.h"
  41
  42 struct btrfs_iget_args {
  43         u64 ino;
  44         struct btrfs_root *root;
  45 };
  46
  47 static struct inode_operations btrfs_dir_inode_operations;
  48 static struct inode_operations btrfs_symlink_inode_operations;
  49 static struct inode_operations btrfs_dir_ro_inode_operations;
  50 static struct inode_operations btrfs_special_inode_operations;
  51 static struct inode_operations btrfs_file_inode_operations;
  52 static struct address_space_operations btrfs_aops;
  53 static struct address_space_operations btrfs_symlink_aops;
  54 static struct file_operations btrfs_dir_file_operations;
  55
  56 static struct kmem_cache *btrfs_inode_cachep;
  57 struct kmem_cache *btrfs_trans_handle_cachep;
  58 struct kmem_cache *btrfs_transaction_cachep;
  59 struct kmem_cache *btrfs_bit_radix_cachep;
  60 struct kmem_cache *btrfs_path_cachep;
  61
  62 #define S_SHIFT 12
  63 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  64         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  65         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  66         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  67         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  68         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  69         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  70         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  71 };
  72
  73 void btrfs_read_locked_inode(struct inode *inode)
  74 {
  75         struct btrfs_path *path;
  76         struct btrfs_inode_item *inode_item;
  77         struct btrfs_root *root = BTRFS_I(inode)->root;
  78         struct btrfs_key location;
  79         u64 alloc_group_block;
  80         u32 rdev;
  81         int ret;
  82
  83         path = btrfs_alloc_path();
  84         BUG_ON(!path);
  85         mutex_lock(&root->fs_info->fs_mutex);
  86
  87         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
  88         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
  89         if (ret) {
  90                 btrfs_free_path(path);
  91                 goto make_bad;
  92         }
  93         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
  94                                   path->slots[0],
  95                                   struct btrfs_inode_item);
  96
  97         inode->i_mode = btrfs_inode_mode(inode_item);
  98         inode->i_nlink = btrfs_inode_nlink(inode_item);
  99         inode->i_uid = btrfs_inode_uid(inode_item);
 100         inode->i_gid = btrfs_inode_gid(inode_item);
 101         inode->i_size = btrfs_inode_size(inode_item);
 102         inode->i_atime.tv_sec = btrfs_timespec_sec(&inode_item->atime);
 103         inode->i_atime.tv_nsec = btrfs_timespec_nsec(&inode_item->atime);
 104         inode->i_mtime.tv_sec = btrfs_timespec_sec(&inode_item->mtime);
 105         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(&inode_item->mtime);
 106         inode->i_ctime.tv_sec = btrfs_timespec_sec(&inode_item->ctime);
 107         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(&inode_item->ctime);
 108         inode->i_blocks = btrfs_inode_nblocks(inode_item);
 109         inode->i_generation = btrfs_inode_generation(inode_item);
 110         inode->i_rdev = 0;
 111         rdev = btrfs_inode_rdev(inode_item);
 112         alloc_group_block = btrfs_inode_block_group(inode_item);
 113         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 114                                                        alloc_group_block);
 115
 116         btrfs_free_path(path);
 117         inode_item = NULL;
 118
 119         mutex_unlock(&root->fs_info->fs_mutex);
 120
 121         switch (inode->i_mode & S_IFMT) {
 122         case S_IFREG:
 123                 inode->i_mapping->a_ops = &btrfs_aops;
 124                 inode->i_fop = &btrfs_file_operations;
 125                 inode->i_op = &btrfs_file_inode_operations;
 126                 break;
 127         case S_IFDIR:
 128                 inode->i_fop = &btrfs_dir_file_operations;
 129                 if (root == root->fs_info->tree_root)
 130                         inode->i_op = &btrfs_dir_ro_inode_operations;
 131                 else
 132                         inode->i_op = &btrfs_dir_inode_operations;
 133                 break;
 134         case S_IFLNK:
 135                 inode->i_op = &btrfs_symlink_inode_operations;
 136                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 137                 break;
 138         default:
 139                 init_special_inode(inode, inode->i_mode, rdev);
 140                 break;
 141         }
 142         return;
 143
 144 make_bad:
 145         btrfs_release_path(root, path);
 146         btrfs_free_path(path);
 147         mutex_unlock(&root->fs_info->fs_mutex);
 148         make_bad_inode(inode);
 149 }
 150
 151 static void fill_inode_item(struct btrfs_inode_item *item,
 152                             struct inode *inode)
 153 {
 154         btrfs_set_inode_uid(item, inode->i_uid);
 155         btrfs_set_inode_gid(item, inode->i_gid);
 156         btrfs_set_inode_size(item, inode->i_size);
 157         btrfs_set_inode_mode(item, inode->i_mode);
 158         btrfs_set_inode_nlink(item, inode->i_nlink);
 159         btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
 160         btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
 161         btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
 162         btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
 163         btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
 164         btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
 165         btrfs_set_inode_nblocks(item, inode->i_blocks);
 166         btrfs_set_inode_generation(item, inode->i_generation);
 167         btrfs_set_inode_rdev(item, inode->i_rdev);
 168         btrfs_set_inode_block_group(item,
 169                                     BTRFS_I(inode)->block_group->key.objectid);
 170 }
 171
 172 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
 173                               struct btrfs_root *root,
 174                               struct inode *inode)
 175 {
 176         struct btrfs_inode_item *inode_item;
 177         struct btrfs_path *path;
 178         int ret;
 179
 180         path = btrfs_alloc_path();
 181         BUG_ON(!path);
 182         ret = btrfs_lookup_inode(trans, root, path,
 183                                  &BTRFS_I(inode)->location, 1);
 184         if (ret) {
 185                 if (ret > 0)
 186                         ret = -ENOENT;
 187                 goto failed;
 188         }
 189
 190         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 191                                   path->slots[0],
 192                                   struct btrfs_inode_item);
 193
 194         fill_inode_item(inode_item, inode);
 195         btrfs_mark_buffer_dirty(path->nodes[0]);
 196         ret = 0;
 197 failed:
 198         btrfs_release_path(root, path);
 199         btrfs_free_path(path);
 200         return ret;
 201 }
 202
 203
 204 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 205                               struct btrfs_root *root,
 206                               struct inode *dir,
 207                               struct dentry *dentry)
 208 {
 209         struct btrfs_path *path;
 210         const char *name = dentry->d_name.name;
 211         int name_len = dentry->d_name.len;
 212         int ret = 0;
 213         u64 objectid;
 214         struct btrfs_dir_item *di;
 215
 216         path = btrfs_alloc_path();
 217         if (!path) {
 218                 ret = -ENOMEM;
 219                 goto err;
 220         }
 221
 222         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 223                                     name, name_len, -1);
 224         if (IS_ERR(di)) {
 225                 ret = PTR_ERR(di);
 226                 goto err;
 227         }
 228         if (!di) {
 229                 ret = -ENOENT;
 230                 goto err;
 231         }
 232         objectid = btrfs_disk_key_objectid(&di->location);
 233         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 234         if (ret)
 235                 goto err;
 236         btrfs_release_path(root, path);
 237
 238         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 239                                          objectid, name, name_len, -1);
 240         if (IS_ERR(di)) {
 241                 ret = PTR_ERR(di);
 242                 goto err;
 243         }
 244         if (!di) {
 245                 ret = -ENOENT;
 246                 goto err;
 247         }
 248         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 249
 250         dentry->d_inode->i_ctime = dir->i_ctime;
 251 err:
 252         btrfs_free_path(path);
 253         if (!ret) {
 254                 dir->i_size -= name_len * 2;
 255                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 256                 btrfs_update_inode(trans, root, dir);
 257                 drop_nlink(dentry->d_inode);
 258                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 259                 dir->i_sb->s_dirt = 1;
 260         }
 261         return ret;
 262 }
 263
 264 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 265 {
 266         struct btrfs_root *root;
 267         struct btrfs_trans_handle *trans;
 268         int ret;
 269
 270         root = BTRFS_I(dir)->root;
 271         mutex_lock(&root->fs_info->fs_mutex);
 272         trans = btrfs_start_transaction(root, 1);
 273         btrfs_set_trans_block_group(trans, dir);
 274         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 275         btrfs_end_transaction(trans, root);
 276         mutex_unlock(&root->fs_info->fs_mutex);
 277         btrfs_btree_balance_dirty(root);
 278         return ret;
 279 }
 280
 281 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 282 {
 283         struct inode *inode = dentry->d_inode;
 284         int err;
 285         int ret;
 286         struct btrfs_root *root = BTRFS_I(dir)->root;
 287         struct btrfs_path *path;
 288         struct btrfs_key key;
 289         struct btrfs_trans_handle *trans;
 290         struct btrfs_key found_key;
 291         int found_type;
 292         struct btrfs_leaf *leaf;
 293         char *goodnames = "..";
 294
 295         path = btrfs_alloc_path();
 296         BUG_ON(!path);
 297         mutex_lock(&root->fs_info->fs_mutex);
 298         trans = btrfs_start_transaction(root, 1);
 299         btrfs_set_trans_block_group(trans, dir);
 300         key.objectid = inode->i_ino;
 301         key.offset = (u64)-1;
 302         key.flags = (u32)-1;
 303         while(1) {
 304                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 305                 if (ret < 0) {
 306                         err = ret;
 307                         goto out;
 308                 }
 309                 BUG_ON(ret == 0);
 310                 if (path->slots[0] == 0) {
 311                         err = -ENOENT;
 312                         goto out;
 313                 }
 314                 path->slots[0]--;
 315                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 316                 btrfs_disk_key_to_cpu(&found_key,
 317                                       &leaf->items[path->slots[0]].key);
 318                 found_type = btrfs_key_type(&found_key);
 319                 if (found_key.objectid != inode->i_ino) {
 320                         err = -ENOENT;
 321                         goto out;
 322                 }
 323                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
 324                      found_type != BTRFS_DIR_INDEX_KEY) ||
 325                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
 326                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
 327                         err = -ENOTEMPTY;
 328                         goto out;
 329                 }
 330                 ret = btrfs_del_item(trans, root, path);
 331                 BUG_ON(ret);
 332
 333                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
 334                         break;
 335                 btrfs_release_path(root, path);
 336         }
 337         ret = 0;
 338         btrfs_release_path(root, path);
 339
 340         /* now the directory is empty */
 341         err = btrfs_unlink_trans(trans, root, dir, dentry);
 342         if (!err) {
 343                 inode->i_size = 0;
 344         }
 345 out:
 346         btrfs_release_path(root, path);
 347         btrfs_free_path(path);
 348         mutex_unlock(&root->fs_info->fs_mutex);
 349         ret = btrfs_end_transaction(trans, root);
 350         btrfs_btree_balance_dirty(root);
 351         if (ret && !err)
 352                 err = ret;
 353         return err;
 354 }
 355
 356 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 357                             struct btrfs_root *root,
 358                             struct inode *inode)
 359 {
 360         struct btrfs_path *path;
 361         int ret;
 362
 363         clear_inode(inode);
 364
 365         path = btrfs_alloc_path();
 366         BUG_ON(!path);
 367         ret = btrfs_lookup_inode(trans, root, path,
 368                                  &BTRFS_I(inode)->location, -1);
 369         if (ret > 0)
 370                 ret = -ENOENT;
 371         if (!ret)
 372                 ret = btrfs_del_item(trans, root, path);
 373         btrfs_free_path(path);
 374         return ret;
 375 }
 376
 377 /*
 378  * truncates go from a high offset to a low offset.  So, walk
 379  * from hi to lo in the node and issue readas.  Stop when you find
 380  * keys from a different objectid
 381  */
 382 static void reada_truncate(struct btrfs_root *root, struct btrfs_path *path,
 383                            u64 objectid)
 384 {
 385         struct btrfs_node *node;
 386         int i;
 387         int nritems;
 388         u64 item_objectid;
 389         u64 blocknr;
 390         int slot;
 391         int ret;
 392
 393         if (!path->nodes[1])
 394                 return;
 395         node = btrfs_buffer_node(path->nodes[1]);
 396         slot = path->slots[1];
 397         if (slot == 0)
 398                 return;
 399         nritems = btrfs_header_nritems(&node->header);
 400         for (i = slot - 1; i >= 0; i--) {
 401                 item_objectid = btrfs_disk_key_objectid(&node->ptrs[i].key);
 402                 if (item_objectid != objectid)
 403                         break;
 404                 blocknr = btrfs_node_blockptr(node, i);
 405                 ret = readahead_tree_block(root, blocknr);
 406                 if (ret)
 407                         break;
 408         }
 409 }
 410
 411 /*
 412  * this can truncate away extent items, csum items and directory items.
 413  * It starts at a high offset and removes keys until it can't find
 414  * any higher than i_size.
 415  *
 416  * csum items that cross the new i_size are truncated to the new size
 417  * as well.
 418  */
 419 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 420                                    struct btrfs_root *root,
 421                                    struct inode *inode)
 422 {
 423         int ret;
 424         struct btrfs_path *path;
 425         struct btrfs_key key;
 426         struct btrfs_disk_key *found_key;
 427         u32 found_type;
 428         struct btrfs_leaf *leaf;
 429         struct btrfs_file_extent_item *fi;
 430         u64 extent_start = 0;
 431         u64 extent_num_blocks = 0;
 432         u64 item_end = 0;
 433         int found_extent;
 434         int del_item;
 435
 436         path = btrfs_alloc_path();
 437         BUG_ON(!path);
 438         /* FIXME, add redo link to tree so we don't leak on crash */
 439         key.objectid = inode->i_ino;
 440         key.offset = (u64)-1;
 441         key.flags = (u32)-1;
 442         while(1) {
 443                 btrfs_init_path(path);
 444                 fi = NULL;
 445                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 446                 if (ret < 0) {
 447                         goto error;
 448                 }
 449                 if (ret > 0) {
 450                         BUG_ON(path->slots[0] == 0);
 451                         path->slots[0]--;
 452                 }
 453                 reada_truncate(root, path, inode->i_ino);
 454                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 455                 found_key = &leaf->items[path->slots[0]].key;
 456                 found_type = btrfs_disk_key_type(found_key);
 457
 458                 if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
 459                         break;
 460                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 461                     found_type != BTRFS_DIR_ITEM_KEY &&
 462                     found_type != BTRFS_DIR_INDEX_KEY &&
 463                     found_type != BTRFS_EXTENT_DATA_KEY)
 464                         break;
 465
 466                 item_end = btrfs_disk_key_offset(found_key);
 467                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 468                         fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 469                                             path->slots[0],
 470                                             struct btrfs_file_extent_item);
 471                         if (btrfs_file_extent_type(fi) !=
 472                             BTRFS_FILE_EXTENT_INLINE) {
 473                                 item_end += btrfs_file_extent_num_blocks(fi) <<
 474                                                 inode->i_blkbits;
 475                         }
 476                 }
 477                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 478                         ret = btrfs_csum_truncate(trans, root, path,
 479                                                   inode->i_size);
 480                         BUG_ON(ret);
 481                 }
 482                 if (item_end < inode->i_size) {
 483                         if (found_type) {
 484                                 btrfs_set_key_type(&key, found_type - 1);
 485                                 continue;
 486                         }
 487                         break;
 488                 }
 489                 if (btrfs_disk_key_offset(found_key) >= inode->i_size)
 490                         del_item = 1;
 491                 else
 492                         del_item = 0;
 493                 found_extent = 0;
 494
 495                 /* FIXME, shrink the extent if the ref count is only 1 */
 496                 if (found_type == BTRFS_EXTENT_DATA_KEY &&
 497                            btrfs_file_extent_type(fi) !=
 498                            BTRFS_FILE_EXTENT_INLINE) {
 499                         u64 num_dec;
 500                         if (!del_item) {
 501                                 u64 orig_num_blocks =
 502                                         btrfs_file_extent_num_blocks(fi);
 503                                 extent_num_blocks = inode->i_size -
 504                                         btrfs_disk_key_offset(found_key) +
 505                                         root->blocksize - 1;
 506                                 extent_num_blocks >>= inode->i_blkbits;
 507                                 btrfs_set_file_extent_num_blocks(fi,
 508                                                          extent_num_blocks);
 509                                 inode->i_blocks -= (orig_num_blocks -
 510                                         extent_num_blocks) << 3;
 511                                 btrfs_mark_buffer_dirty(path->nodes[0]);
 512                         } else {
 513                                 extent_start =
 514                                         btrfs_file_extent_disk_blocknr(fi);
 515                                 extent_num_blocks =
 516                                         btrfs_file_extent_disk_num_blocks(fi);
 517                                 /* FIXME blocksize != 4096 */
 518                                 num_dec = btrfs_file_extent_num_blocks(fi) << 3;
 519                                 if (extent_start != 0) {
 520                                         found_extent = 1;
 521                                         inode->i_blocks -= num_dec;
 522                                 }
 523                         }
 524                 }
 525                 if (del_item) {
 526                         ret = btrfs_del_item(trans, root, path);
 527                         if (ret)
 528                                 goto error;
 529                 } else {
 530                         break;
 531                 }
 532                 btrfs_release_path(root, path);
 533                 if (found_extent) {
 534                         ret = btrfs_free_extent(trans, root, extent_start,
 535                                                 extent_num_blocks, 0);
 536                         BUG_ON(ret);
 537                 }
 538         }
 539         ret = 0;
 540 error:
 541         btrfs_release_path(root, path);
 542         btrfs_free_path(path);
 543         inode->i_sb->s_dirt = 1;
 544         return ret;
 545 }
 546
 547 /*
 548  * taken from block_truncate_page, but does cow as it zeros out
 549  * any bytes left in the last page in the file.
 550  */
 551 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 552 {
 553         struct inode *inode = mapping->host;
 554         unsigned blocksize = 1 << inode->i_blkbits;
 555         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 556         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 557         struct page *page;
 558         char *kaddr;
 559         int ret = 0;
 560         struct btrfs_root *root = BTRFS_I(inode)->root;
 561         u64 alloc_hint = 0;
 562         struct btrfs_key ins;
 563         struct btrfs_trans_handle *trans;
 564
 565         if ((offset & (blocksize - 1)) == 0)
 566                 goto out;
 567
 568         ret = -ENOMEM;
 569         page = grab_cache_page(mapping, index);
 570         if (!page)
 571                 goto out;
 572
 573         if (!PageUptodate(page)) {
 574                 ret = btrfs_readpage(NULL, page);
 575                 lock_page(page);
 576                 if (!PageUptodate(page)) {
 577                         ret = -EIO;
 578                         goto out;
 579                 }
 580         }
 581         mutex_lock(&root->fs_info->fs_mutex);
 582         trans = btrfs_start_transaction(root, 1);
 583         btrfs_set_trans_block_group(trans, inode);
 584
 585         ret = btrfs_drop_extents(trans, root, inode,
 586                                  page->index << PAGE_CACHE_SHIFT,
 587                                  (page->index + 1) << PAGE_CACHE_SHIFT,
 588                                  &alloc_hint);
 589         if (ret)
 590                 goto out;
 591         ret = btrfs_alloc_extent(trans, root, inode->i_ino, 1,
 592                                  alloc_hint, (u64)-1, &ins, 1);
 593         if (ret)
 594                 goto out;
 595         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 596                                        page->index << PAGE_CACHE_SHIFT,
 597                                        ins.objectid, 1, 1);
 598         if (ret)
 599                 goto out;
 600         SetPageChecked(page);
 601         kaddr = kmap(page);
 602         memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
 603         flush_dcache_page(page);
 604         ret = btrfs_csum_file_block(trans, root, inode->i_ino,
 605                               page->index << PAGE_CACHE_SHIFT,
 606                               kaddr, PAGE_CACHE_SIZE);
 607         kunmap(page);
 608         btrfs_end_transaction(trans, root);
 609         mutex_unlock(&root->fs_info->fs_mutex);
 610
 611         set_page_dirty(page);
 612         unlock_page(page);
 613         page_cache_release(page);
 614 out:
 615         return ret;
 616 }
 617
 618 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 619 {
 620         struct inode *inode = dentry->d_inode;
 621         int err;
 622
 623         err = inode_change_ok(inode, attr);
 624         if (err)
 625                 return err;
 626
 627         if (S_ISREG(inode->i_mode) &&
 628             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 629                 struct btrfs_trans_handle *trans;
 630                 struct btrfs_root *root = BTRFS_I(inode)->root;
 631                 u64 mask = root->blocksize - 1;
 632                 u64 pos = (inode->i_size + mask) & ~mask;
 633                 u64 hole_size;
 634
 635                 if (attr->ia_size <= pos)
 636                         goto out;
 637
 638                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 639
 640                 hole_size = (attr->ia_size - pos + mask) & ~mask;
 641                 hole_size >>= inode->i_blkbits;
 642
 643                 mutex_lock(&root->fs_info->fs_mutex);
 644                 trans = btrfs_start_transaction(root, 1);
 645                 btrfs_set_trans_block_group(trans, inode);
 646                 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
 647                                                pos, 0, 0, hole_size);
 648                 btrfs_end_transaction(trans, root);
 649                 mutex_unlock(&root->fs_info->fs_mutex);
 650                 if (err)
 651                         return err;
 652         }
 653 out:
 654         err = inode_setattr(inode, attr);
 655
 656         return err;
 657 }
 658 void btrfs_delete_inode(struct inode *inode)
 659 {
 660         struct btrfs_trans_handle *trans;
 661         struct btrfs_root *root = BTRFS_I(inode)->root;
 662         int ret;
 663
 664         truncate_inode_pages(&inode->i_data, 0);
 665         if (is_bad_inode(inode)) {
 666                 goto no_delete;
 667         }
 668         inode->i_size = 0;
 669         mutex_lock(&root->fs_info->fs_mutex);
 670         trans = btrfs_start_transaction(root, 1);
 671         btrfs_set_trans_block_group(trans, inode);
 672         ret = btrfs_truncate_in_trans(trans, root, inode);
 673         if (ret)
 674                 goto no_delete_lock;
 675         ret = btrfs_free_inode(trans, root, inode);
 676         if (ret)
 677                 goto no_delete_lock;
 678         btrfs_end_transaction(trans, root);
 679         mutex_unlock(&root->fs_info->fs_mutex);
 680         btrfs_btree_balance_dirty(root);
 681         return;
 682
 683 no_delete_lock:
 684         btrfs_end_transaction(trans, root);
 685         mutex_unlock(&root->fs_info->fs_mutex);
 686         btrfs_btree_balance_dirty(root);
 687 no_delete:
 688         clear_inode(inode);
 689 }
 690
 691 /*
 692  * this returns the key found in the dir entry in the location pointer.
 693  * If no dir entries were found, location->objectid is 0.
 694  */
 695 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 696                                struct btrfs_key *location)
 697 {
 698         const char *name = dentry->d_name.name;
 699         int namelen = dentry->d_name.len;
 700         struct btrfs_dir_item *di;
 701         struct btrfs_path *path;
 702         struct btrfs_root *root = BTRFS_I(dir)->root;
 703         int ret;
 704
 705         path = btrfs_alloc_path();
 706         BUG_ON(!path);
 707         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
 708                                     namelen, 0);
 709         if (!di || IS_ERR(di)) {
 710                 location->objectid = 0;
 711                 ret = 0;
 712                 goto out;
 713         }
 714         btrfs_disk_key_to_cpu(location, &di->location);
 715 out:
 716         btrfs_release_path(root, path);
 717         btrfs_free_path(path);
 718         return ret;
 719 }
 720
 721 /*
 722  * when we hit a tree root in a directory, the btrfs part of the inode
 723  * needs to be changed to reflect the root directory of the tree root.  This
 724  * is kind of like crossing a mount point.
 725  */
 726 static int fixup_tree_root_location(struct btrfs_root *root,
 727                              struct btrfs_key *location,
 728                              struct btrfs_root **sub_root)
 729 {
 730         struct btrfs_path *path;
 731         struct btrfs_root_item *ri;
 732
 733         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 734                 return 0;
 735         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 736                 return 0;
 737
 738         path = btrfs_alloc_path();
 739         BUG_ON(!path);
 740         mutex_lock(&root->fs_info->fs_mutex);
 741
 742         *sub_root = btrfs_read_fs_root(root->fs_info, location);
 743         if (IS_ERR(*sub_root))
 744                 return PTR_ERR(*sub_root);
 745
 746         ri = &(*sub_root)->root_item;
 747         location->objectid = btrfs_root_dirid(ri);
 748         location->flags = 0;
 749         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 750         location->offset = 0;
 751
 752         btrfs_free_path(path);
 753         mutex_unlock(&root->fs_info->fs_mutex);
 754         return 0;
 755 }
 756
 757 static int btrfs_init_locked_inode(struct inode *inode, void *p)
 758 {
 759         struct btrfs_iget_args *args = p;
 760         inode->i_ino = args->ino;
 761         BTRFS_I(inode)->root = args->root;
 762         return 0;
 763 }
 764
 765 static int btrfs_find_actor(struct inode *inode, void *opaque)
 766 {
 767         struct btrfs_iget_args *args = opaque;
 768         return (args->ino == inode->i_ino &&
 769                 args->root == BTRFS_I(inode)->root);
 770 }
 771
 772 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 773                                 struct btrfs_root *root)
 774 {
 775         struct inode *inode;
 776         struct btrfs_iget_args args;
 777         args.ino = objectid;
 778         args.root = root;
 779
 780         inode = iget5_locked(s, objectid, btrfs_find_actor,
 781                              btrfs_init_locked_inode,
 782                              (void *)&args);
 783         return inode;
 784 }
 785
 786 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 787                                    struct nameidata *nd)
 788 {
 789         struct inode * inode;
 790         struct btrfs_inode *bi = BTRFS_I(dir);
 791         struct btrfs_root *root = bi->root;
 792         struct btrfs_root *sub_root = root;
 793         struct btrfs_key location;
 794         int ret;
 795
 796         if (dentry->d_name.len > BTRFS_NAME_LEN)
 797                 return ERR_PTR(-ENAMETOOLONG);
 798         mutex_lock(&root->fs_info->fs_mutex);
 799         ret = btrfs_inode_by_name(dir, dentry, &location);
 800         mutex_unlock(&root->fs_info->fs_mutex);
 801         if (ret < 0)
 802                 return ERR_PTR(ret);
 803         inode = NULL;
 804         if (location.objectid) {
 805                 ret = fixup_tree_root_location(root, &location, &sub_root);
 806                 if (ret < 0)
 807                         return ERR_PTR(ret);
 808                 if (ret > 0)
 809                         return ERR_PTR(-ENOENT);
 810                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 811                                           sub_root);
 812                 if (!inode)
 813                         return ERR_PTR(-EACCES);
 814                 if (inode->i_state & I_NEW) {
 815                         /* the inode and parent dir are two different roots */
 816                         if (sub_root != root) {
 817                                 igrab(inode);
 818                                 sub_root->inode = inode;
 819                         }
 820                         BTRFS_I(inode)->root = sub_root;
 821                         memcpy(&BTRFS_I(inode)->location, &location,
 822                                sizeof(location));
 823                         btrfs_read_locked_inode(inode);
 824                         unlock_new_inode(inode);
 825                 }
 826         }
 827         return d_splice_alias(inode, dentry);
 828 }
 829
 830 /*
 831  * readahead one full node of leaves as long as their keys include
 832  * the objectid supplied
 833  */
 834 static void reada_leaves(struct btrfs_root *root, struct btrfs_path *path,
 835                          u64 objectid)
 836 {
 837         struct btrfs_node *node;
 838         int i;
 839         u32 nritems;
 840         u64 item_objectid;
 841         u64 blocknr;
 842         int slot;
 843         int ret;
 844
 845         if (!path->nodes[1])
 846                 return;
 847         node = btrfs_buffer_node(path->nodes[1]);
 848         slot = path->slots[1];
 849         nritems = btrfs_header_nritems(&node->header);
 850         for (i = slot + 1; i < nritems; i++) {
 851                 item_objectid = btrfs_disk_key_objectid(&node->ptrs[i].key);
 852                 if (item_objectid != objectid)
 853                         break;
 854                 blocknr = btrfs_node_blockptr(node, i);
 855                 ret = readahead_tree_block(root, blocknr);
 856                 if (ret)
 857                         break;
 858         }
 859 }
 860 static unsigned char btrfs_filetype_table[] = {
 861         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
 862 };
 863
 864 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 865 {
 866         struct inode *inode = filp->f_path.dentry->d_inode;
 867         struct btrfs_root *root = BTRFS_I(inode)->root;
 868         struct btrfs_item *item;
 869         struct btrfs_dir_item *di;
 870         struct btrfs_key key;
 871         struct btrfs_path *path;
 872         int ret;
 873         u32 nritems;
 874         struct btrfs_leaf *leaf;
 875         int slot;
 876         int advance;
 877         unsigned char d_type;
 878         int over = 0;
 879         u32 di_cur;
 880         u32 di_total;
 881         u32 di_len;
 882         int key_type = BTRFS_DIR_INDEX_KEY;
 883
 884         /* FIXME, use a real flag for deciding about the key type */
 885         if (root->fs_info->tree_root == root)
 886                 key_type = BTRFS_DIR_ITEM_KEY;
 887         mutex_lock(&root->fs_info->fs_mutex);
 888         key.objectid = inode->i_ino;
 889         key.flags = 0;
 890         btrfs_set_key_type(&key, key_type);
 891         key.offset = filp->f_pos;
 892         path = btrfs_alloc_path();
 893         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 894         if (ret < 0)
 895                 goto err;
 896         advance = 0;
 897         reada_leaves(root, path, inode->i_ino);
 898         while(1) {
 899                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 900                 nritems = btrfs_header_nritems(&leaf->header);
 901                 slot = path->slots[0];
 902                 if (advance || slot >= nritems) {
 903                         if (slot >= nritems -1) {
 904                                 reada_leaves(root, path, inode->i_ino);
 905                                 ret = btrfs_next_leaf(root, path);
 906                                 if (ret)
 907                                         break;
 908                                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 909                                 nritems = btrfs_header_nritems(&leaf->header);
 910                                 slot = path->slots[0];
 911                         } else {
 912                                 slot++;
 913                                 path->slots[0]++;
 914                         }
 915                 }
 916                 advance = 1;
 917                 item = leaf->items + slot;
 918                 if (btrfs_disk_key_objectid(&item->key) != key.objectid)
 919                         break;
 920                 if (btrfs_disk_key_type(&item->key) != key_type)
 921                         break;
 922                 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
 923                         continue;
 924                 filp->f_pos = btrfs_disk_key_offset(&item->key);
 925                 advance = 1;
 926                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 927                 di_cur = 0;
 928                 di_total = btrfs_item_size(leaf->items + slot);
 929                 while(di_cur < di_total) {
 930                         d_type = btrfs_filetype_table[btrfs_dir_type(di)];
 931                         over = filldir(dirent, (const char *)(di + 1),
 932                                        btrfs_dir_name_len(di),
 933                                        btrfs_disk_key_offset(&item->key),
 934                                        btrfs_disk_key_objectid(&di->location),
 935                                        d_type);
 936                         if (over)
 937                                 goto nopos;
 938                         di_len = btrfs_dir_name_len(di) + sizeof(*di);
 939                         di_cur += di_len;
 940                         di = (struct btrfs_dir_item *)((char *)di + di_len);
 941                 }
 942         }
 943         filp->f_pos++;
 944 nopos:
 945         ret = 0;
 946 err:
 947         btrfs_release_path(root, path);
 948         btrfs_free_path(path);
 949         mutex_unlock(&root->fs_info->fs_mutex);
 950         return ret;
 951 }
 952
 953 int btrfs_write_inode(struct inode *inode, int wait)
 954 {
 955         struct btrfs_root *root = BTRFS_I(inode)->root;
 956         struct btrfs_trans_handle *trans;
 957         int ret = 0;
 958
 959         if (wait) {
 960                 mutex_lock(&root->fs_info->fs_mutex);
 961                 trans = btrfs_start_transaction(root, 1);
 962                 btrfs_set_trans_block_group(trans, inode);
 963                 ret = btrfs_commit_transaction(trans, root);
 964                 mutex_unlock(&root->fs_info->fs_mutex);
 965         }
 966         return ret;
 967 }
 968
 969 /*
 970  * This is somewhat expensive, updating the tree every time the
 971  * inode changes.  But, it is most likely to find the inode in cache.
 972  * FIXME, needs more benchmarking...there are no reasons other than performance
 973  * to keep or drop this code.
 974  */
 975 void btrfs_dirty_inode(struct inode *inode)
 976 {
 977         struct btrfs_root *root = BTRFS_I(inode)->root;
 978         struct btrfs_trans_handle *trans;
 979
 980         mutex_lock(&root->fs_info->fs_mutex);
 981         trans = btrfs_start_transaction(root, 1);
 982         btrfs_set_trans_block_group(trans, inode);
 983         btrfs_update_inode(trans, root, inode);
 984         btrfs_end_transaction(trans, root);
 985         mutex_unlock(&root->fs_info->fs_mutex);
 986 }
 987
 988 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 989                                      struct btrfs_root *root,
 990                                      u64 objectid,
 991                                      struct btrfs_block_group_cache *group,
 992                                      int mode)
 993 {
 994         struct inode *inode;
 995         struct btrfs_inode_item inode_item;
 996         struct btrfs_key *location;
 997         int ret;
 998         int owner;
 999
1000         inode = new_inode(root->fs_info->sb);
1001         if (!inode)
1002                 return ERR_PTR(-ENOMEM);
1003
1004         BTRFS_I(inode)->root = root;
1005         if (mode & S_IFDIR)
1006                 owner = 0;
1007         else
1008                 owner = 1;
1009         group = btrfs_find_block_group(root, group, 0, 0, owner);
1010         BTRFS_I(inode)->block_group = group;
1011
1012         inode->i_uid = current->fsuid;
1013         inode->i_gid = current->fsgid;
1014         inode->i_mode = mode;
1015         inode->i_ino = objectid;
1016         inode->i_blocks = 0;
1017         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1018         fill_inode_item(&inode_item, inode);
1019         location = &BTRFS_I(inode)->location;
1020         location->objectid = objectid;
1021         location->flags = 0;
1022         location->offset = 0;
1023         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1024
1025         ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
1026         if (ret)
1027                 return ERR_PTR(ret);
1028         insert_inode_hash(inode);
1029         return inode;
1030 }
1031
1032 static inline u8 btrfs_inode_type(struct inode *inode)
1033 {
1034         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1035 }
1036
1037 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1038                             struct dentry *dentry, struct inode *inode)
1039 {
1040         int ret;
1041         struct btrfs_key key;
1042         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1043         struct inode *parent_inode;
1044         key.objectid = inode->i_ino;
1045         key.flags = 0;
1046         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1047         key.offset = 0;
1048
1049         ret = btrfs_insert_dir_item(trans, root,
1050                                     dentry->d_name.name, dentry->d_name.len,
1051                                     dentry->d_parent->d_inode->i_ino,
1052                                     &key, btrfs_inode_type(inode));
1053         if (ret == 0) {
1054                 parent_inode = dentry->d_parent->d_inode;
1055                 parent_inode->i_size += dentry->d_name.len * 2;
1056                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1057                 ret = btrfs_update_inode(trans, root,
1058                                          dentry->d_parent->d_inode);
1059         }
1060         return ret;
1061 }
1062
1063 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1064                             struct dentry *dentry, struct inode *inode)
1065 {
1066         int err = btrfs_add_link(trans, dentry, inode);
1067         if (!err) {
1068                 d_instantiate(dentry, inode);
1069                 return 0;
1070         }
1071         if (err > 0)
1072                 err = -EEXIST;
1073         return err;
1074 }
1075
1076 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1077                         int mode, dev_t rdev)
1078 {
1079         struct btrfs_trans_handle *trans;
1080         struct btrfs_root *root = BTRFS_I(dir)->root;
1081         struct inode *inode;
1082         int err;
1083         int drop_inode = 0;
1084         u64 objectid;
1085
1086         if (!new_valid_dev(rdev))
1087                 return -EINVAL;
1088
1089         mutex_lock(&root->fs_info->fs_mutex);
1090         trans = btrfs_start_transaction(root, 1);
1091         btrfs_set_trans_block_group(trans, dir);
1092
1093         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1094         if (err) {
1095                 err = -ENOSPC;
1096                 goto out_unlock;
1097         }
1098
1099         inode = btrfs_new_inode(trans, root, objectid,
1100                                 BTRFS_I(dir)->block_group, mode);
1101         err = PTR_ERR(inode);
1102         if (IS_ERR(inode))
1103                 goto out_unlock;
1104
1105         btrfs_set_trans_block_group(trans, inode);
1106         err = btrfs_add_nondir(trans, dentry, inode);
1107         if (err)
1108                 drop_inode = 1;
1109         else {
1110                 inode->i_op = &btrfs_special_inode_operations;
1111                 init_special_inode(inode, inode->i_mode, rdev);
1112         }
1113         dir->i_sb->s_dirt = 1;
1114         btrfs_update_inode_block_group(trans, inode);
1115         btrfs_update_inode_block_group(trans, dir);
1116 out_unlock:
1117         btrfs_end_transaction(trans, root);
1118         mutex_unlock(&root->fs_info->fs_mutex);
1119
1120         if (drop_inode) {
1121                 inode_dec_link_count(inode);
1122                 iput(inode);
1123         }
1124         btrfs_btree_balance_dirty(root);
1125         return err;
1126 }
1127
1128 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1129                         int mode, struct nameidata *nd)
1130 {
1131         struct btrfs_trans_handle *trans;
1132         struct btrfs_root *root = BTRFS_I(dir)->root;
1133         struct inode *inode;
1134         int err;
1135         int drop_inode = 0;
1136         u64 objectid;
1137
1138         mutex_lock(&root->fs_info->fs_mutex);
1139         trans = btrfs_start_transaction(root, 1);
1140         btrfs_set_trans_block_group(trans, dir);
1141
1142         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1143         if (err) {
1144                 err = -ENOSPC;
1145                 goto out_unlock;
1146         }
1147
1148         inode = btrfs_new_inode(trans, root, objectid,
1149                                 BTRFS_I(dir)->block_group, mode);
1150         err = PTR_ERR(inode);
1151         if (IS_ERR(inode))
1152                 goto out_unlock;
1153
1154         btrfs_set_trans_block_group(trans, inode);
1155         err = btrfs_add_nondir(trans, dentry, inode);
1156         if (err)
1157                 drop_inode = 1;
1158         else {
1159                 inode->i_mapping->a_ops = &btrfs_aops;
1160                 inode->i_fop = &btrfs_file_operations;
1161                 inode->i_op = &btrfs_file_inode_operations;
1162         }
1163         dir->i_sb->s_dirt = 1;
1164         btrfs_update_inode_block_group(trans, inode);
1165         btrfs_update_inode_block_group(trans, dir);
1166 out_unlock:
1167         btrfs_end_transaction(trans, root);
1168         mutex_unlock(&root->fs_info->fs_mutex);
1169
1170         if (drop_inode) {
1171                 inode_dec_link_count(inode);
1172                 iput(inode);
1173         }
1174         btrfs_btree_balance_dirty(root);
1175         return err;
1176 }
1177
1178 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1179                       struct dentry *dentry)
1180 {
1181         struct btrfs_trans_handle *trans;
1182         struct btrfs_root *root = BTRFS_I(dir)->root;
1183         struct inode *inode = old_dentry->d_inode;
1184         int err;
1185         int drop_inode = 0;
1186
1187         if (inode->i_nlink == 0)
1188                 return -ENOENT;
1189
1190         inc_nlink(inode);
1191         mutex_lock(&root->fs_info->fs_mutex);
1192         trans = btrfs_start_transaction(root, 1);
1193         btrfs_set_trans_block_group(trans, dir);
1194         atomic_inc(&inode->i_count);
1195         err = btrfs_add_nondir(trans, dentry, inode);
1196         if (err)
1197                 drop_inode = 1;
1198         dir->i_sb->s_dirt = 1;
1199         btrfs_update_inode_block_group(trans, dir);
1200         err = btrfs_update_inode(trans, root, inode);
1201         if (err)
1202                 drop_inode = 1;
1203
1204         btrfs_end_transaction(trans, root);
1205         mutex_unlock(&root->fs_info->fs_mutex);
1206
1207         if (drop_inode) {
1208                 inode_dec_link_count(inode);
1209                 iput(inode);
1210         }
1211         btrfs_btree_balance_dirty(root);
1212         return err;
1213 }
1214
1215 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
1216                                 struct btrfs_root *root,
1217                                 u64 objectid, u64 dirid)
1218 {
1219         int ret;
1220         char buf[2];
1221         struct btrfs_key key;
1222
1223         buf[0] = '.';
1224         buf[1] = '.';
1225
1226         key.objectid = objectid;
1227         key.offset = 0;
1228         key.flags = 0;
1229         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1230
1231         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
1232                                     &key, BTRFS_FT_DIR);
1233         if (ret)
1234                 goto error;
1235         key.objectid = dirid;
1236         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
1237                                     &key, BTRFS_FT_DIR);
1238         if (ret)
1239                 goto error;
1240 error:
1241         return ret;
1242 }
1243
1244 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1245 {
1246         struct inode *inode;
1247         struct btrfs_trans_handle *trans;
1248         struct btrfs_root *root = BTRFS_I(dir)->root;
1249         int err = 0;
1250         int drop_on_err = 0;
1251         u64 objectid;
1252
1253         mutex_lock(&root->fs_info->fs_mutex);
1254         trans = btrfs_start_transaction(root, 1);
1255         btrfs_set_trans_block_group(trans, dir);
1256         if (IS_ERR(trans)) {
1257                 err = PTR_ERR(trans);
1258                 goto out_unlock;
1259         }
1260
1261         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1262         if (err) {
1263                 err = -ENOSPC;
1264                 goto out_unlock;
1265         }
1266
1267         inode = btrfs_new_inode(trans, root, objectid,
1268                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1269         if (IS_ERR(inode)) {
1270                 err = PTR_ERR(inode);
1271                 goto out_fail;
1272         }
1273         drop_on_err = 1;
1274         inode->i_op = &btrfs_dir_inode_operations;
1275         inode->i_fop = &btrfs_dir_file_operations;
1276         btrfs_set_trans_block_group(trans, inode);
1277
1278         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
1279         if (err)
1280                 goto out_fail;
1281
1282         inode->i_size = 6;
1283         err = btrfs_update_inode(trans, root, inode);
1284         if (err)
1285                 goto out_fail;
1286         err = btrfs_add_link(trans, dentry, inode);
1287         if (err)
1288                 goto out_fail;
1289         d_instantiate(dentry, inode);
1290         drop_on_err = 0;
1291         dir->i_sb->s_dirt = 1;
1292         btrfs_update_inode_block_group(trans, inode);
1293         btrfs_update_inode_block_group(trans, dir);
1294
1295 out_fail:
1296         btrfs_end_transaction(trans, root);
1297 out_unlock:
1298         mutex_unlock(&root->fs_info->fs_mutex);
1299         if (drop_on_err)
1300                 iput(inode);
1301         btrfs_btree_balance_dirty(root);
1302         return err;
1303 }
1304
1305 /*
1306  * FIBMAP and others want to pass in a fake buffer head.  They need to
1307  * use BTRFS_GET_BLOCK_NO_DIRECT to make sure we don't try to memcpy
1308  * any packed file data into the fake bh
1309  */
1310 #define BTRFS_GET_BLOCK_NO_CREATE 0
1311 #define BTRFS_GET_BLOCK_CREATE 1
1312 #define BTRFS_GET_BLOCK_NO_DIRECT 2
1313
1314 /*
1315  * FIXME create==1 doe not work.
1316  */
1317 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
1318                                 struct buffer_head *result, int create)
1319 {
1320         int ret;
1321         int err = 0;
1322         u64 blocknr;
1323         u64 extent_start = 0;
1324         u64 extent_end = 0;
1325         u64 objectid = inode->i_ino;
1326         u32 found_type;
1327         u64 alloc_hint = 0;
1328         struct btrfs_path *path;
1329         struct btrfs_root *root = BTRFS_I(inode)->root;
1330         struct btrfs_file_extent_item *item;
1331         struct btrfs_leaf *leaf;
1332         struct btrfs_disk_key *found_key;
1333         struct btrfs_trans_handle *trans = NULL;
1334
1335         path = btrfs_alloc_path();
1336         BUG_ON(!path);
1337         if (create & BTRFS_GET_BLOCK_CREATE) {
1338                 /*
1339                  * danger!, this only works if the page is properly up
1340                  * to date somehow
1341                  */
1342                 trans = btrfs_start_transaction(root, 1);
1343                 if (!trans) {
1344                         err = -ENOMEM;
1345                         goto out;
1346                 }
1347                 ret = btrfs_drop_extents(trans, root, inode,
1348                                          iblock << inode->i_blkbits,
1349                                          (iblock + 1) << inode->i_blkbits,
1350                                          &alloc_hint);
1351                 BUG_ON(ret);
1352         }
1353
1354         ret = btrfs_lookup_file_extent(NULL, root, path,
1355                                        objectid,
1356                                        iblock << inode->i_blkbits, 0);
1357         if (ret < 0) {
1358                 err = ret;
1359                 goto out;
1360         }
1361
1362         if (ret != 0) {
1363                 if (path->slots[0] == 0) {
1364                         btrfs_release_path(root, path);
1365                         goto not_found;
1366                 }
1367                 path->slots[0]--;
1368         }
1369
1370         item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
1371                               struct btrfs_file_extent_item);
1372         leaf = btrfs_buffer_leaf(path->nodes[0]);
1373         blocknr = btrfs_file_extent_disk_blocknr(item);
1374         blocknr += btrfs_file_extent_offset(item);
1375
1376         /* are we inside the extent that was found? */
1377         found_key = &leaf->items[path->slots[0]].key;
1378         found_type = btrfs_disk_key_type(found_key);
1379         if (btrfs_disk_key_objectid(found_key) != objectid ||
1380             found_type != BTRFS_EXTENT_DATA_KEY) {
1381                 extent_end = 0;
1382                 extent_start = 0;
1383                 goto not_found;
1384         }
1385         found_type = btrfs_file_extent_type(item);
1386         extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
1387         if (found_type == BTRFS_FILE_EXTENT_REG) {
1388                 extent_start = extent_start >> inode->i_blkbits;
1389                 extent_end = extent_start + btrfs_file_extent_num_blocks(item);
1390                 err = 0;
1391                 if (btrfs_file_extent_disk_blocknr(item) == 0)
1392                         goto out;
1393                 if (iblock >= extent_start && iblock < extent_end) {
1394                         btrfs_map_bh_to_logical(root, result, blocknr +
1395                                                 iblock - extent_start);
1396                         goto out;
1397                 }
1398         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1399                 char *ptr;
1400                 char *map;
1401                 u32 size;
1402
1403                 if (create & BTRFS_GET_BLOCK_NO_DIRECT) {
1404                         err = -EINVAL;
1405                         goto out;
1406                 }
1407                 size = btrfs_file_extent_inline_len(leaf->items +
1408                                                     path->slots[0]);
1409                 extent_end = (extent_start + size) >> inode->i_blkbits;
1410                 extent_start >>= inode->i_blkbits;
1411                 if (iblock < extent_start || iblock > extent_end) {
1412                         goto not_found;
1413                 }
1414                 ptr = btrfs_file_extent_inline_start(item);
1415                 map = kmap(result->b_page);
1416                 memcpy(map, ptr, size);
1417                 memset(map + size, 0, PAGE_CACHE_SIZE - size);
1418                 flush_dcache_page(result->b_page);
1419                 kunmap(result->b_page);
1420                 set_buffer_uptodate(result);
1421                 SetPageChecked(result->b_page);
1422                 btrfs_map_bh_to_logical(root, result, 0);
1423         }
1424 not_found:
1425         if (create & BTRFS_GET_BLOCK_CREATE) {
1426                 struct btrfs_key ins;
1427                 ret = btrfs_alloc_extent(trans, root, inode->i_ino,
1428                                          1, alloc_hint, (u64)-1,
1429                                          &ins, 1);
1430                 if (ret) {
1431                         err = ret;
1432                         goto out;
1433                 }
1434                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
1435                                                iblock << inode->i_blkbits,
1436                                                ins.objectid, ins.offset,
1437                                                ins.offset);
1438                 if (ret) {
1439                         err = ret;
1440                         goto out;
1441                 }
1442                 btrfs_map_bh_to_logical(root, result, ins.objectid);
1443         }
1444 out:
1445         if (trans) {
1446                 ret = btrfs_end_transaction(trans, root);
1447                 if (!err)
1448                         err = ret;
1449         }
1450         btrfs_free_path(path);
1451         return err;
1452 }
1453
1454 int btrfs_get_block(struct inode *inode, sector_t iblock,
1455                     struct buffer_head *result, int create)
1456 {
1457         int err;
1458         struct btrfs_root *root = BTRFS_I(inode)->root;
1459         mutex_lock(&root->fs_info->fs_mutex);
1460         err = btrfs_get_block_lock(inode, iblock, result, create);
1461         mutex_unlock(&root->fs_info->fs_mutex);
1462         return err;
1463 }
1464
1465 static int btrfs_get_block_csum(struct inode *inode, sector_t iblock,
1466                                 struct buffer_head *result, int create)
1467 {
1468         int ret;
1469         struct btrfs_root *root = BTRFS_I(inode)->root;
1470         struct page *page = result->b_page;
1471         u64 offset = (page->index << PAGE_CACHE_SHIFT) + bh_offset(result);
1472         struct btrfs_csum_item *item;
1473         struct btrfs_path *path = NULL;
1474
1475         mutex_lock(&root->fs_info->fs_mutex);
1476         ret = btrfs_get_block_lock(inode, iblock, result, create);
1477         if (ret)
1478                 goto out;
1479
1480         path = btrfs_alloc_path();
1481         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, offset, 0);
1482         if (IS_ERR(item)) {
1483                 ret = PTR_ERR(item);
1484                 /* a csum that isn't present is a preallocated region. */
1485                 if (ret == -ENOENT || ret == -EFBIG)
1486                         ret = 0;
1487                 result->b_private = NULL;
1488                 goto out;
1489         }
1490         memcpy((char *)&result->b_private, &item->csum, BTRFS_CRC32_SIZE);
1491 out:
1492         if (path)
1493                 btrfs_free_path(path);
1494         mutex_unlock(&root->fs_info->fs_mutex);
1495         return ret;
1496 }
1497
1498 static int btrfs_get_block_bmap(struct inode *inode, sector_t iblock,
1499                            struct buffer_head *result, int create)
1500 {
1501         struct btrfs_root *root = BTRFS_I(inode)->root;
1502         mutex_lock(&root->fs_info->fs_mutex);
1503         btrfs_get_block_lock(inode, iblock, result, BTRFS_GET_BLOCK_NO_DIRECT);
1504         mutex_unlock(&root->fs_info->fs_mutex);
1505         return 0;
1506 }
1507
1508 static sector_t btrfs_bmap(struct address_space *as, sector_t block)
1509 {
1510         return generic_block_bmap(as, block, btrfs_get_block_bmap);
1511 }
1512
1513 static int btrfs_prepare_write(struct file *file, struct page *page,
1514                                unsigned from, unsigned to)
1515 {
1516         return block_prepare_write(page, from, to, btrfs_get_block);
1517 }
1518
1519 static void buffer_io_error(struct buffer_head *bh)
1520 {
1521         char b[BDEVNAME_SIZE];
1522
1523         printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n",
1524                         bdevname(bh->b_bdev, b),
1525                         (unsigned long long)bh->b_blocknr);
1526 }
1527
1528 /*
1529  * I/O completion handler for block_read_full_page() - pages
1530  * which come unlocked at the end of I/O.
1531  */
1532 static void btrfs_end_buffer_async_read(struct buffer_head *bh, int uptodate)
1533 {
1534         unsigned long flags;
1535         struct buffer_head *first;
1536         struct buffer_head *tmp;
1537         struct page *page;
1538         int page_uptodate = 1;
1539         struct inode *inode;
1540         int ret;
1541
1542         BUG_ON(!buffer_async_read(bh));
1543
1544         page = bh->b_page;
1545         inode = page->mapping->host;
1546         if (uptodate) {
1547                 void *kaddr;
1548                 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1549                 if (bh->b_private) {
1550                         char csum[BTRFS_CRC32_SIZE];
1551                         kaddr = kmap_atomic(page, KM_IRQ0);
1552                         ret = btrfs_csum_data(root, kaddr + bh_offset(bh),
1553                                               bh->b_size, csum);
1554                         BUG_ON(ret);
1555                         if (memcmp(csum, &bh->b_private, BTRFS_CRC32_SIZE)) {
1556                                 u64 offset;
1557                                 offset = (page->index << PAGE_CACHE_SHIFT) +
1558                                         bh_offset(bh);
1559                                 printk("btrfs csum failed ino %lu off %llu\n",
1560                                        page->mapping->host->i_ino,
1561                                        (unsigned long long)offset);
1562                                 memset(kaddr + bh_offset(bh), 1, bh->b_size);
1563                                 flush_dcache_page(page);
1564                         }
1565                         kunmap_atomic(kaddr, KM_IRQ0);
1566                 }
1567                 set_buffer_uptodate(bh);
1568         } else {
1569                 clear_buffer_uptodate(bh);
1570                 if (printk_ratelimit())
1571                         buffer_io_error(bh);
1572                 SetPageError(page);
1573         }
1574
1575         /*
1576          * Be _very_ careful from here on. Bad things can happen if
1577          * two buffer heads end IO at almost the same time and both
1578          * decide that the page is now completely done.
1579          */
1580         first = page_buffers(page);
1581         local_irq_save(flags);
1582         bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
1583         clear_buffer_async_read(bh);
1584         unlock_buffer(bh);
1585         tmp = bh;
1586         do {
1587                 if (!buffer_uptodate(tmp))
1588                         page_uptodate = 0;
1589                 if (buffer_async_read(tmp)) {
1590                         BUG_ON(!buffer_locked(tmp));
1591                         goto still_busy;
1592                 }
1593                 tmp = tmp->b_this_page;
1594         } while (tmp != bh);
1595         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
1596         local_irq_restore(flags);
1597
1598         /*
1599          * If none of the buffers had errors and they are all
1600          * uptodate then we can set the page uptodate.
1601          */
1602         if (page_uptodate && !PageError(page))
1603                 SetPageUptodate(page);
1604         unlock_page(page);
1605         return;
1606
1607 still_busy:
1608         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
1609         local_irq_restore(flags);
1610         return;
1611 }
1612
1613 /*
1614  * Generic "read page" function for block devices that have the normal
1615  * get_block functionality. This is most of the block device filesystems.
1616  * Reads the page asynchronously --- the unlock_buffer() and
1617  * set/clear_buffer_uptodate() functions propagate buffer state into the
1618  * page struct once IO has completed.
1619  */
1620 int btrfs_readpage(struct file *file, struct page *page)
1621 {
1622         struct inode *inode = page->mapping->host;
1623         sector_t iblock, lblock;
1624         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
1625         unsigned int blocksize;
1626         int nr, i;
1627         int fully_mapped = 1;
1628
1629         BUG_ON(!PageLocked(page));
1630         blocksize = 1 << inode->i_blkbits;
1631         if (!page_has_buffers(page))
1632                 create_empty_buffers(page, blocksize, 0);
1633         head = page_buffers(page);
1634
1635         iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1636         lblock = (i_size_read(inode)+blocksize-1) >> inode->i_blkbits;
1637         bh = head;
1638         nr = 0;
1639         i = 0;
1640
1641         do {
1642                 if (buffer_uptodate(bh))
1643                         continue;
1644
1645                 if (!buffer_mapped(bh)) {
1646                         int err = 0;
1647
1648                         fully_mapped = 0;
1649                         if (iblock < lblock) {
1650                                 WARN_ON(bh->b_size != blocksize);
1651                                 err = btrfs_get_block_csum(inode, iblock,
1652                                                            bh, 0);
1653                                 if (err)
1654                                         SetPageError(page);
1655                         }
1656                         if (!buffer_mapped(bh)) {
1657                                 void *kaddr = kmap_atomic(page, KM_USER0);
1658                                 memset(kaddr + i * blocksize, 0, blocksize);
1659                                 flush_dcache_page(page);
1660                                 kunmap_atomic(kaddr, KM_USER0);
1661                                 if (!err)
1662                                         set_buffer_uptodate(bh);
1663                                 continue;
1664                         }
1665                         /*
1666                          * get_block() might have updated the buffer
1667                          * synchronously
1668                          */
1669                         if (buffer_uptodate(bh))
1670                                 continue;
1671                 }
1672                 arr[nr++] = bh;
1673         } while (i++, iblock++, (bh = bh->b_this_page) != head);
1674
1675         if (fully_mapped)
1676                 SetPageMappedToDisk(page);
1677
1678         if (!nr) {
1679                 /*
1680                  * All buffers are uptodate - we can set the page uptodate
1681                  * as well. But not if get_block() returned an error.
1682                  */
1683                 if (!PageError(page))
1684                         SetPageUptodate(page);
1685                 unlock_page(page);
1686                 return 0;
1687         }
1688
1689         /* Stage two: lock the buffers */
1690         for (i = 0; i < nr; i++) {
1691                 bh = arr[i];
1692                 lock_buffer(bh);
1693                 bh->b_end_io = btrfs_end_buffer_async_read;
1694                 set_buffer_async_read(bh);
1695         }
1696
1697         /*
1698          * Stage 3: start the IO.  Check for uptodateness
1699          * inside the buffer lock in case another process reading
1700          * the underlying blockdev brought it uptodate (the sct fix).
1701          */
1702         for (i = 0; i < nr; i++) {
1703                 bh = arr[i];
1704                 if (buffer_uptodate(bh))
1705                         btrfs_end_buffer_async_read(bh, 1);
1706                 else
1707                         submit_bh(READ, bh);
1708         }
1709         return 0;
1710 }
1711
1712 /*
1713  * Aside from a tiny bit of packed file data handling, this is the
1714  * same as the generic code.
1715  *
1716  * While block_write_full_page is writing back the dirty buffers under
1717  * the page lock, whoever dirtied the buffers may decide to clean them
1718  * again at any time.  We handle that by only looking at the buffer
1719  * state inside lock_buffer().
1720  *
1721  * If block_write_full_page() is called for regular writeback
1722  * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1723  * locked buffer.   This only can happen if someone has written the buffer
1724  * directly, with submit_bh().  At the address_space level PageWriteback
1725  * prevents this contention from occurring.
1726  */
1727 static int __btrfs_write_full_page(struct inode *inode, struct page *page,
1728                                    struct writeback_control *wbc)
1729 {
1730         int err;
1731         sector_t block;
1732         sector_t last_block;
1733         struct buffer_head *bh, *head;
1734         const unsigned blocksize = 1 << inode->i_blkbits;
1735         int nr_underway = 0;
1736         struct btrfs_root *root = BTRFS_I(inode)->root;
1737
1738         BUG_ON(!PageLocked(page));
1739
1740         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
1741
1742         /* no csumming allowed when from PF_MEMALLOC */
1743         if (current->flags & PF_MEMALLOC) {
1744                 redirty_page_for_writepage(wbc, page);
1745                 unlock_page(page);
1746                 return 0;
1747         }
1748
1749         if (!page_has_buffers(page)) {
1750                 create_empty_buffers(page, blocksize,
1751                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
1752         }
1753
1754         /*
1755          * Be very careful.  We have no exclusion from __set_page_dirty_buffers
1756          * here, and the (potentially unmapped) buffers may become dirty at
1757          * any time.  If a buffer becomes dirty here after we've inspected it
1758          * then we just miss that fact, and the page stays dirty.
1759          *
1760          * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1761          * handle that here by just cleaning them.
1762          */
1763
1764         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1765         head = page_buffers(page);
1766         bh = head;
1767
1768         /*
1769          * Get all the dirty buffers mapped to disk addresses and
1770          * handle any aliases from the underlying blockdev's mapping.
1771          */
1772         do {
1773                 if (block > last_block) {
1774                         /*
1775                          * mapped buffers outside i_size will occur, because
1776                          * this page can be outside i_size when there is a
1777                          * truncate in progress.
1778                          */
1779                         /*
1780                          * The buffer was zeroed by block_write_full_page()
1781                          */
1782                         clear_buffer_dirty(bh);
1783                         set_buffer_uptodate(bh);
1784                 } else if (!buffer_mapped(bh) && buffer_dirty(bh)) {
1785                         WARN_ON(bh->b_size != blocksize);
1786                         err = btrfs_get_block(inode, block, bh, 0);
1787                         if (err) {
1788                                 goto recover;
1789                         }
1790                         if (buffer_new(bh)) {
1791                                 /* blockdev mappings never come here */
1792                                 clear_buffer_new(bh);
1793                         }
1794                 }
1795                 bh = bh->b_this_page;
1796                 block++;
1797         } while (bh != head);
1798
1799         do {
1800                 if (!buffer_mapped(bh))
1801                         continue;
1802                 /*
1803                  * If it's a fully non-blocking write attempt and we cannot
1804                  * lock the buffer then redirty the page.  Note that this can
1805                  * potentially cause a busy-wait loop from pdflush and kswapd
1806                  * activity, but those code paths have their own higher-level
1807                  * throttling.
1808                  */
1809                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
1810                         lock_buffer(bh);
1811                 } else if (test_set_buffer_locked(bh)) {
1812                         redirty_page_for_writepage(wbc, page);
1813                         continue;
1814                 }
1815                 if (test_clear_buffer_dirty(bh) && bh->b_blocknr != 0) {
1816                         struct btrfs_trans_handle *trans;
1817                         int ret;
1818                         u64 off = page->index << PAGE_CACHE_SHIFT;
1819                         char *kaddr;
1820
1821                         off += bh_offset(bh);
1822                         mutex_lock(&root->fs_info->fs_mutex);
1823                         trans = btrfs_start_transaction(root, 1);
1824                         btrfs_set_trans_block_group(trans, inode);
1825                         kaddr = kmap(page);
1826                         btrfs_csum_file_block(trans, root, inode->i_ino,
1827                                                     off, kaddr + bh_offset(bh),
1828                                                     bh->b_size);
1829                         kunmap(page);
1830                         ret = btrfs_end_transaction(trans, root);
1831                         BUG_ON(ret);
1832                         mutex_unlock(&root->fs_info->fs_mutex);
1833                         mark_buffer_async_write(bh);
1834                 } else {
1835                         unlock_buffer(bh);
1836                 }
1837         } while ((bh = bh->b_this_page) != head);
1838
1839         /*
1840          * The page and its buffers are protected by PageWriteback(), so we can
1841          * drop the bh refcounts early.
1842          */
1843         BUG_ON(PageWriteback(page));
1844         set_page_writeback(page);
1845
1846         do {
1847                 struct buffer_head *next = bh->b_this_page;
1848                 if (buffer_async_write(bh)) {
1849                         submit_bh(WRITE, bh);
1850                         nr_underway++;
1851                 }
1852                 bh = next;
1853         } while (bh != head);
1854         unlock_page(page);
1855
1856         err = 0;
1857 done:
1858         if (nr_underway == 0) {
1859                 /*
1860                  * The page was marked dirty, but the buffers were
1861                  * clean.  Someone wrote them back by hand with
1862                  * ll_rw_block/submit_bh.  A rare case.
1863                  */
1864                 int uptodate = 1;
1865                 do {
1866                         if (!buffer_uptodate(bh)) {
1867                                 uptodate = 0;
1868                                 break;
1869                         }
1870                         bh = bh->b_this_page;
1871                 } while (bh != head);
1872                 if (uptodate)
1873                         SetPageUptodate(page);
1874                 end_page_writeback(page);
1875         }
1876         return err;
1877
1878 recover:
1879         /*
1880          * ENOSPC, or some other error.  We may already have added some
1881          * blocks to the file, so we need to write these out to avoid
1882          * exposing stale data.
1883          * The page is currently locked and not marked for writeback
1884          */
1885         bh = head;
1886         /* Recovery: lock and submit the mapped buffers */
1887         do {
1888                 if (buffer_mapped(bh) && buffer_dirty(bh)) {
1889                         lock_buffer(bh);
1890                         mark_buffer_async_write(bh);
1891                 } else {
1892                         /*
1893                          * The buffer may have been set dirty during
1894                          * attachment to a dirty page.
1895                          */
1896                         clear_buffer_dirty(bh);
1897                 }
1898         } while ((bh = bh->b_this_page) != head);
1899         SetPageError(page);
1900         BUG_ON(PageWriteback(page));
1901         set_page_writeback(page);
1902         do {
1903                 struct buffer_head *next = bh->b_this_page;
1904                 if (buffer_async_write(bh)) {
1905                         clear_buffer_dirty(bh);
1906                         submit_bh(WRITE, bh);
1907                         nr_underway++;
1908                 }
1909                 bh = next;
1910         } while (bh != head);
1911         unlock_page(page);
1912         goto done;
1913 }
1914
1915 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1916 {
1917         struct inode * const inode = page->mapping->host;
1918         loff_t i_size = i_size_read(inode);
1919         const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1920         unsigned offset;
1921         void *kaddr;
1922
1923         /* Is the page fully inside i_size? */
1924         if (page->index < end_index)
1925                 return __btrfs_write_full_page(inode, page, wbc);
1926
1927         /* Is the page fully outside i_size? (truncate in progress) */
1928         offset = i_size & (PAGE_CACHE_SIZE-1);
1929         if (page->index >= end_index+1 || !offset) {
1930                 /*
1931                  * The page may have dirty, unmapped buffers.  For example,
1932                  * they may have been added in ext3_writepage().  Make them
1933                  * freeable here, so the page does not leak.
1934                  */
1935                 block_invalidatepage(page, 0);
1936                 unlock_page(page);
1937                 return 0; /* don't care */
1938         }
1939
1940         /*
1941          * The page straddles i_size.  It must be zeroed out on each and every
1942          * writepage invokation because it may be mmapped.  "A file is mapped
1943          * in multiples of the page size.  For a file that is not a multiple of
1944          * the  page size, the remaining memory is zeroed when mapped, and
1945          * writes to that region are not written out to the file."
1946          */
1947         kaddr = kmap_atomic(page, KM_USER0);
1948         memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1949         flush_dcache_page(page);
1950         kunmap_atomic(kaddr, KM_USER0);
1951         return __btrfs_write_full_page(inode, page, wbc);
1952 }
1953
1954 /*
1955  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
1956  * called from a page fault handler when a page is first dirtied. Hence we must
1957  * be careful to check for EOF conditions here. We set the page up correctly
1958  * for a written page which means we get ENOSPC checking when writing into
1959  * holes and correct delalloc and unwritten extent mapping on filesystems that
1960  * support these features.
1961  *
1962  * We are not allowed to take the i_mutex here so we have to play games to
1963  * protect against truncate races as the page could now be beyond EOF.  Because
1964  * vmtruncate() writes the inode size before removing pages, once we have the
1965  * page lock we can determine safely if the page is beyond EOF. If it is not
1966  * beyond EOF, then the page is guaranteed safe against truncation until we
1967  * unlock the page.
1968  */
1969 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1970 {
1971         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1972         unsigned long end;
1973         loff_t size;
1974         int ret = -EINVAL;
1975
1976         lock_page(page);
1977         wait_on_page_writeback(page);
1978         size = i_size_read(inode);
1979         if ((page->mapping != inode->i_mapping) ||
1980             ((page->index << PAGE_CACHE_SHIFT) > size)) {
1981                 /* page got truncated out from underneath us */
1982                 goto out_unlock;
1983         }
1984
1985         /* page is wholly or partially inside EOF */
1986         if (((page->index + 1) << PAGE_CACHE_SHIFT) > size)
1987                 end = size & ~PAGE_CACHE_MASK;
1988         else
1989                 end = PAGE_CACHE_SIZE;
1990
1991         ret = btrfs_prepare_write(NULL, page, 0, end);
1992         if (!ret)
1993                 ret = btrfs_commit_write(NULL, page, 0, end);
1994
1995 out_unlock:
1996         unlock_page(page);
1997         return ret;
1998 }
1999
2000 static void btrfs_truncate(struct inode *inode)
2001 {
2002         struct btrfs_root *root = BTRFS_I(inode)->root;
2003         int ret;
2004         struct btrfs_trans_handle *trans;
2005
2006         if (!S_ISREG(inode->i_mode))
2007                 return;
2008         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2009                 return;
2010
2011         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2012
2013         mutex_lock(&root->fs_info->fs_mutex);
2014         trans = btrfs_start_transaction(root, 1);
2015         btrfs_set_trans_block_group(trans, inode);
2016
2017         /* FIXME, add redo link to tree so we don't leak on crash */
2018         ret = btrfs_truncate_in_trans(trans, root, inode);
2019         btrfs_update_inode(trans, root, inode);
2020         ret = btrfs_end_transaction(trans, root);
2021         BUG_ON(ret);
2022         mutex_unlock(&root->fs_info->fs_mutex);
2023         btrfs_btree_balance_dirty(root);
2024 }
2025
2026 int btrfs_commit_write(struct file *file, struct page *page,
2027                        unsigned from, unsigned to)
2028 {
2029         struct inode *inode = page->mapping->host;
2030         struct buffer_head *bh;
2031         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2032
2033         SetPageUptodate(page);
2034         bh = page_buffers(page);
2035         set_buffer_uptodate(bh);
2036         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
2037                 set_page_dirty(page);
2038         }
2039         if (pos > inode->i_size) {
2040                 i_size_write(inode, pos);
2041                 mark_inode_dirty(inode);
2042         }
2043         return 0;
2044 }
2045
2046 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
2047 {
2048         struct btrfs_trans_handle *trans;
2049         struct btrfs_key key;
2050         struct btrfs_root_item root_item;
2051         struct btrfs_inode_item *inode_item;
2052         struct buffer_head *subvol;
2053         struct btrfs_leaf *leaf;
2054         struct btrfs_root *new_root;
2055         struct inode *inode;
2056         struct inode *dir;
2057         int ret;
2058         int err;
2059         u64 objectid;
2060         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2061
2062         mutex_lock(&root->fs_info->fs_mutex);
2063         trans = btrfs_start_transaction(root, 1);
2064         BUG_ON(!trans);
2065
2066         subvol = btrfs_alloc_free_block(trans, root, 0);
2067         if (IS_ERR(subvol))
2068                 return PTR_ERR(subvol);
2069         leaf = btrfs_buffer_leaf(subvol);
2070         btrfs_set_header_nritems(&leaf->header, 0);
2071         btrfs_set_header_level(&leaf->header, 0);
2072         btrfs_set_header_blocknr(&leaf->header, bh_blocknr(subvol));
2073         btrfs_set_header_generation(&leaf->header, trans->transid);
2074         btrfs_set_header_owner(&leaf->header, root->root_key.objectid);
2075         memcpy(leaf->header.fsid, root->fs_info->disk_super->fsid,
2076                sizeof(leaf->header.fsid));
2077         btrfs_mark_buffer_dirty(subvol);
2078
2079         inode_item = &root_item.inode;
2080         memset(inode_item, 0, sizeof(*inode_item));
2081         btrfs_set_inode_generation(inode_item, 1);
2082         btrfs_set_inode_size(inode_item, 3);
2083         btrfs_set_inode_nlink(inode_item, 1);
2084         btrfs_set_inode_nblocks(inode_item, 1);
2085         btrfs_set_inode_mode(inode_item, S_IFDIR | 0755);
2086
2087         btrfs_set_root_blocknr(&root_item, bh_blocknr(subvol));
2088         btrfs_set_root_refs(&root_item, 1);
2089         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2090         root_item.drop_level = 0;
2091         brelse(subvol);
2092         subvol = NULL;
2093
2094         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2095                                        0, &objectid);
2096         if (ret)
2097                 goto fail;
2098
2099         btrfs_set_root_dirid(&root_item, new_dirid);
2100
2101         key.objectid = objectid;
2102         key.offset = 1;
2103         key.flags = 0;
2104         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2105         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2106                                 &root_item);
2107         if (ret)
2108                 goto fail;
2109
2110         /*
2111          * insert the directory item
2112          */
2113         key.offset = (u64)-1;
2114         dir = root->fs_info->sb->s_root->d_inode;
2115         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2116                                     name, namelen, dir->i_ino, &key,
2117                                     BTRFS_FT_DIR);
2118         if (ret)
2119                 goto fail;
2120
2121         ret = btrfs_commit_transaction(trans, root);
2122         if (ret)
2123                 goto fail_commit;
2124
2125         new_root = btrfs_read_fs_root(root->fs_info, &key);
2126         BUG_ON(!new_root);
2127
2128         trans = btrfs_start_transaction(new_root, 1);
2129         BUG_ON(!trans);
2130
2131         inode = btrfs_new_inode(trans, new_root, new_dirid,
2132                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2133         if (IS_ERR(inode))
2134                 goto fail;
2135         inode->i_op = &btrfs_dir_inode_operations;
2136         inode->i_fop = &btrfs_dir_file_operations;
2137         new_root->inode = inode;
2138
2139         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
2140         if (ret)
2141                 goto fail;
2142
2143         inode->i_nlink = 1;
2144         inode->i_size = 6;
2145         ret = btrfs_update_inode(trans, new_root, inode);
2146         if (ret)
2147                 goto fail;
2148 fail:
2149         err = btrfs_commit_transaction(trans, root);
2150         if (err && !ret)
2151                 ret = err;
2152 fail_commit:
2153         mutex_unlock(&root->fs_info->fs_mutex);
2154         btrfs_btree_balance_dirty(root);
2155         return ret;
2156 }
2157
2158 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2159 {
2160         struct btrfs_trans_handle *trans;
2161         struct btrfs_key key;
2162         struct btrfs_root_item new_root_item;
2163         int ret;
2164         int err;
2165         u64 objectid;
2166
2167         if (!root->ref_cows)
2168                 return -EINVAL;
2169
2170         mutex_lock(&root->fs_info->fs_mutex);
2171         trans = btrfs_start_transaction(root, 1);
2172         BUG_ON(!trans);
2173
2174         ret = btrfs_update_inode(trans, root, root->inode);
2175         if (ret)
2176                 goto fail;
2177
2178         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2179                                        0, &objectid);
2180         if (ret)
2181                 goto fail;
2182
2183         memcpy(&new_root_item, &root->root_item,
2184                sizeof(new_root_item));
2185
2186         key.objectid = objectid;
2187         key.offset = 1;
2188         key.flags = 0;
2189         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2190         btrfs_set_root_blocknr(&new_root_item, bh_blocknr(root->node));
2191
2192         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2193                                 &new_root_item);
2194         if (ret)
2195                 goto fail;
2196
2197         /*
2198          * insert the directory item
2199          */
2200         key.offset = (u64)-1;
2201         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2202                                     name, namelen,
2203                                     root->fs_info->sb->s_root->d_inode->i_ino,
2204                                     &key, BTRFS_FT_DIR);
2205
2206         if (ret)
2207                 goto fail;
2208
2209         ret = btrfs_inc_root_ref(trans, root);
2210         if (ret)
2211                 goto fail;
2212
2213 fail:
2214         err = btrfs_commit_transaction(trans, root);
2215         if (err && !ret)
2216                 ret = err;
2217         mutex_unlock(&root->fs_info->fs_mutex);
2218         btrfs_btree_balance_dirty(root);
2219         return ret;
2220 }
2221
2222 int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
2223                 cmd, unsigned long arg)
2224 {
2225         struct btrfs_root *root = BTRFS_I(inode)->root;
2226         struct btrfs_ioctl_vol_args vol_args;
2227         int ret = 0;
2228         struct btrfs_dir_item *di;
2229         int namelen;
2230         struct btrfs_path *path;
2231         u64 root_dirid;
2232
2233         switch (cmd) {
2234         case BTRFS_IOC_SNAP_CREATE:
2235                 if (copy_from_user(&vol_args,
2236                                    (struct btrfs_ioctl_vol_args __user *)arg,
2237                                    sizeof(vol_args)))
2238                         return -EFAULT;
2239                 namelen = strlen(vol_args.name);
2240                 if (namelen > BTRFS_VOL_NAME_MAX)
2241                         return -EINVAL;
2242                 if (strchr(vol_args.name, '/'))
2243                         return -EINVAL;
2244                 path = btrfs_alloc_path();
2245                 if (!path)
2246                         return -ENOMEM;
2247                 root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2248                 mutex_lock(&root->fs_info->fs_mutex);
2249                 di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2250                                     path, root_dirid,
2251                                     vol_args.name, namelen, 0);
2252                 mutex_unlock(&root->fs_info->fs_mutex);
2253                 btrfs_free_path(path);
2254                 if (di && !IS_ERR(di))
2255                         return -EEXIST;
2256                 if (IS_ERR(di))
2257                         return PTR_ERR(di);
2258
2259                 if (root == root->fs_info->tree_root)
2260                         ret = create_subvol(root, vol_args.name, namelen);
2261                 else
2262                         ret = create_snapshot(root, vol_args.name, namelen);
2263                 break;
2264         default:
2265                 return -ENOTTY;
2266         }
2267         return ret;
2268 }
2269
2270 #ifdef CONFIG_COMPAT
2271 long btrfs_compat_ioctl(struct file *file, unsigned int cmd,
2272                                unsigned long arg)
2273 {
2274         struct inode *inode = file->f_path.dentry->d_inode;
2275         int ret;
2276         lock_kernel();
2277         ret = btrfs_ioctl(inode, file, cmd, (unsigned long) compat_ptr(arg));
2278         unlock_kernel();
2279         return ret;
2280
2281 }
2282 #endif
2283
2284 /*
2285  * Called inside transaction, so use GFP_NOFS
2286  */
2287 struct inode *btrfs_alloc_inode(struct super_block *sb)
2288 {
2289         struct btrfs_inode *ei;
2290
2291         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2292         if (!ei)
2293                 return NULL;
2294         return &ei->vfs_inode;
2295 }
2296
2297 void btrfs_destroy_inode(struct inode *inode)
2298 {
2299         WARN_ON(!list_empty(&inode->i_dentry));
2300         WARN_ON(inode->i_data.nrpages);
2301
2302         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2303 }
2304
2305 static void init_once(void * foo, struct kmem_cache * cachep,
2306                       unsigned long flags)
2307 {
2308         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2309
2310         inode_init_once(&ei->vfs_inode);
2311 }
2312
2313 void btrfs_destroy_cachep(void)
2314 {
2315         if (btrfs_inode_cachep)
2316                 kmem_cache_destroy(btrfs_inode_cachep);
2317         if (btrfs_trans_handle_cachep)
2318                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2319         if (btrfs_transaction_cachep)
2320                 kmem_cache_destroy(btrfs_transaction_cachep);
2321         if (btrfs_bit_radix_cachep)
2322                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2323         if (btrfs_path_cachep)
2324                 kmem_cache_destroy(btrfs_path_cachep);
2325 }
2326
2327 static struct kmem_cache *cache_create(const char *name, size_t size,
2328                                        unsigned long extra_flags,
2329                                        void (*ctor)(void *, struct kmem_cache *,
2330                                                     unsigned long))
2331 {
2332         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2333                                  SLAB_MEM_SPREAD | extra_flags), ctor
2334 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2335                                  ,NULL
2336 #endif
2337                                 );
2338 }
2339
2340 int btrfs_init_cachep(void)
2341 {
2342         btrfs_inode_cachep = cache_create("btrfs_inode_cache",
2343                                           sizeof(struct btrfs_inode),
2344                                           0, init_once);
2345         if (!btrfs_inode_cachep)
2346                 goto fail;
2347         btrfs_trans_handle_cachep = cache_create("btrfs_trans_handle_cache",
2348                                              sizeof(struct btrfs_trans_handle),
2349                                              0, NULL);
2350         if (!btrfs_trans_handle_cachep)
2351                 goto fail;
2352         btrfs_transaction_cachep = cache_create("btrfs_transaction_cache",
2353                                              sizeof(struct btrfs_transaction),
2354                                              0, NULL);
2355         if (!btrfs_transaction_cachep)
2356                 goto fail;
2357         btrfs_path_cachep = cache_create("btrfs_path_cache",
2358                                          sizeof(struct btrfs_transaction),
2359                                          0, NULL);
2360         if (!btrfs_path_cachep)
2361                 goto fail;
2362         btrfs_bit_radix_cachep = cache_create("btrfs_radix", 256,
2363                                               SLAB_DESTROY_BY_RCU, NULL);
2364         if (!btrfs_bit_radix_cachep)
2365                 goto fail;
2366         return 0;
2367 fail:
2368         btrfs_destroy_cachep();
2369         return -ENOMEM;
2370 }
2371
2372 static int btrfs_getattr(struct vfsmount *mnt,
2373                          struct dentry *dentry, struct kstat *stat)
2374 {
2375         struct inode *inode = dentry->d_inode;
2376         generic_fillattr(inode, stat);
2377         stat->blksize = 256 * 1024;
2378         return 0;
2379 }
2380
2381 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2382                            struct inode * new_dir,struct dentry *new_dentry)
2383 {
2384         struct btrfs_trans_handle *trans;
2385         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2386         struct inode *new_inode = new_dentry->d_inode;
2387         struct inode *old_inode = old_dentry->d_inode;
2388         struct timespec ctime = CURRENT_TIME;
2389         struct btrfs_path *path;
2390         struct btrfs_dir_item *di;
2391         int ret;
2392
2393         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2394             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2395                 return -ENOTEMPTY;
2396         }
2397         mutex_lock(&root->fs_info->fs_mutex);
2398         trans = btrfs_start_transaction(root, 1);
2399         btrfs_set_trans_block_group(trans, new_dir);
2400         path = btrfs_alloc_path();
2401         if (!path) {
2402                 ret = -ENOMEM;
2403                 goto out_fail;
2404         }
2405
2406         old_dentry->d_inode->i_nlink++;
2407         old_dir->i_ctime = old_dir->i_mtime = ctime;
2408         new_dir->i_ctime = new_dir->i_mtime = ctime;
2409         old_inode->i_ctime = ctime;
2410         if (S_ISDIR(old_inode->i_mode) && old_dir != new_dir) {
2411                 struct btrfs_key *location = &BTRFS_I(new_dir)->location;
2412                 u64 old_parent_oid;
2413                 di = btrfs_lookup_dir_item(trans, root, path, old_inode->i_ino,
2414                                            "..", 2, -1);
2415                 if (IS_ERR(di)) {
2416                         ret = PTR_ERR(di);
2417                         goto out_fail;
2418                 }
2419                 if (!di) {
2420                         ret = -ENOENT;
2421                         goto out_fail;
2422                 }
2423                 old_parent_oid = btrfs_disk_key_objectid(&di->location);
2424                 ret = btrfs_del_item(trans, root, path);
2425                 if (ret) {
2426                         goto out_fail;
2427                 }
2428                 btrfs_release_path(root, path);
2429
2430                 di = btrfs_lookup_dir_index_item(trans, root, path,
2431                                                  old_inode->i_ino,
2432                                                  old_parent_oid,
2433                                                  "..", 2, -1);
2434                 if (IS_ERR(di)) {
2435                         ret = PTR_ERR(di);
2436                         goto out_fail;
2437                 }
2438                 if (!di) {
2439                         ret = -ENOENT;
2440                         goto out_fail;
2441                 }
2442                 ret = btrfs_del_item(trans, root, path);
2443                 if (ret) {
2444                         goto out_fail;
2445                 }
2446                 btrfs_release_path(root, path);
2447
2448                 ret = btrfs_insert_dir_item(trans, root, "..", 2,
2449                                             old_inode->i_ino, location,
2450                                             BTRFS_FT_DIR);
2451                 if (ret)
2452                         goto out_fail;
2453         }
2454
2455
2456         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2457         if (ret)
2458                 goto out_fail;
2459
2460         if (new_inode) {
2461                 new_inode->i_ctime = CURRENT_TIME;
2462                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2463                 if (ret)
2464                         goto out_fail;
2465                 if (S_ISDIR(new_inode->i_mode))
2466                         clear_nlink(new_inode);
2467                 else
2468                         drop_nlink(new_inode);
2469                 ret = btrfs_update_inode(trans, root, new_inode);
2470                 if (ret)
2471                         goto out_fail;
2472         }
2473         ret = btrfs_add_link(trans, new_dentry, old_inode);
2474         if (ret)
2475                 goto out_fail;
2476
2477 out_fail:
2478         btrfs_free_path(path);
2479         btrfs_end_transaction(trans, root);
2480         mutex_unlock(&root->fs_info->fs_mutex);
2481         return ret;
2482 }
2483
2484 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2485                          const char *symname)
2486 {
2487         struct btrfs_trans_handle *trans;
2488         struct btrfs_root *root = BTRFS_I(dir)->root;
2489         struct btrfs_path *path;
2490         struct btrfs_key key;
2491         struct inode *inode;
2492         int err;
2493         int drop_inode = 0;
2494         u64 objectid;
2495         int name_len;
2496         int datasize;
2497         char *ptr;
2498         struct btrfs_file_extent_item *ei;
2499
2500         name_len = strlen(symname) + 1;
2501         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2502                 return -ENAMETOOLONG;
2503         mutex_lock(&root->fs_info->fs_mutex);
2504         trans = btrfs_start_transaction(root, 1);
2505         btrfs_set_trans_block_group(trans, dir);
2506
2507         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2508         if (err) {
2509                 err = -ENOSPC;
2510                 goto out_unlock;
2511         }
2512
2513         inode = btrfs_new_inode(trans, root, objectid,
2514                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2515         err = PTR_ERR(inode);
2516         if (IS_ERR(inode))
2517                 goto out_unlock;
2518
2519         btrfs_set_trans_block_group(trans, inode);
2520         err = btrfs_add_nondir(trans, dentry, inode);
2521         if (err)
2522                 drop_inode = 1;
2523         else {
2524                 inode->i_mapping->a_ops = &btrfs_aops;
2525                 inode->i_fop = &btrfs_file_operations;
2526                 inode->i_op = &btrfs_file_inode_operations;
2527         }
2528         dir->i_sb->s_dirt = 1;
2529         btrfs_update_inode_block_group(trans, inode);
2530         btrfs_update_inode_block_group(trans, dir);
2531         if (drop_inode)
2532                 goto out_unlock;
2533
2534         path = btrfs_alloc_path();
2535         BUG_ON(!path);
2536         key.objectid = inode->i_ino;
2537         key.offset = 0;
2538         key.flags = 0;
2539         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2540         datasize = btrfs_file_extent_calc_inline_size(name_len);
2541         err = btrfs_insert_empty_item(trans, root, path, &key,
2542                                       datasize);
2543         if (err) {
2544                 drop_inode = 1;
2545                 goto out_unlock;
2546         }
2547         ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
2548                path->slots[0], struct btrfs_file_extent_item);
2549         btrfs_set_file_extent_generation(ei, trans->transid);
2550         btrfs_set_file_extent_type(ei,
2551                                    BTRFS_FILE_EXTENT_INLINE);
2552         ptr = btrfs_file_extent_inline_start(ei);
2553         btrfs_memcpy(root, path->nodes[0]->b_data,
2554                      ptr, symname, name_len);
2555         btrfs_mark_buffer_dirty(path->nodes[0]);
2556         btrfs_free_path(path);
2557         inode->i_op = &btrfs_symlink_inode_operations;
2558         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2559         inode->i_size = name_len - 1;
2560         err = btrfs_update_inode(trans, root, inode);
2561         if (err)
2562                 drop_inode = 1;
2563
2564 out_unlock:
2565         btrfs_end_transaction(trans, root);
2566         mutex_unlock(&root->fs_info->fs_mutex);
2567         if (drop_inode) {
2568                 inode_dec_link_count(inode);
2569                 iput(inode);
2570         }
2571         btrfs_btree_balance_dirty(root);
2572         return err;
2573 }
2574
2575 static struct inode_operations btrfs_dir_inode_operations = {
2576         .lookup         = btrfs_lookup,
2577         .create         = btrfs_create,
2578         .unlink         = btrfs_unlink,
2579         .link           = btrfs_link,
2580         .mkdir          = btrfs_mkdir,
2581         .rmdir          = btrfs_rmdir,
2582         .rename         = btrfs_rename,
2583         .symlink        = btrfs_symlink,
2584         .setattr        = btrfs_setattr,
2585         .mknod          = btrfs_mknod,
2586 };
2587
2588 static struct inode_operations btrfs_dir_ro_inode_operations = {
2589         .lookup         = btrfs_lookup,
2590 };
2591
2592 static struct file_operations btrfs_dir_file_operations = {
2593         .llseek         = generic_file_llseek,
2594         .read           = generic_read_dir,
2595         .readdir        = btrfs_readdir,
2596         .ioctl          = btrfs_ioctl,
2597 #ifdef CONFIG_COMPAT
2598         .compat_ioctl   = btrfs_compat_ioctl,
2599 #endif
2600 };
2601
2602 static struct address_space_operations btrfs_aops = {
2603         .readpage       = btrfs_readpage,
2604         .writepage      = btrfs_writepage,
2605         .sync_page      = block_sync_page,
2606         .prepare_write  = btrfs_prepare_write,
2607         .commit_write   = btrfs_commit_write,
2608         .bmap           = btrfs_bmap,
2609 };
2610
2611 static struct address_space_operations btrfs_symlink_aops = {
2612         .readpage       = btrfs_readpage,
2613         .writepage      = btrfs_writepage,
2614 };
2615
2616 static struct inode_operations btrfs_file_inode_operations = {
2617         .truncate       = btrfs_truncate,
2618         .getattr        = btrfs_getattr,
2619         .setattr        = btrfs_setattr,
2620 };
2621
2622 static struct inode_operations btrfs_special_inode_operations = {
2623         .getattr        = btrfs_getattr,
2624         .setattr        = btrfs_setattr,
2625 };
2626
2627 static struct inode_operations btrfs_symlink_inode_operations = {
2628         .readlink       = generic_readlink,
2629         .follow_link    = page_follow_link_light,
2630         .put_link       = page_put_link,
2631 };