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