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 static struct extent_map_ops btrfs_extent_map_ops;
  56
  57 static struct kmem_cache *btrfs_inode_cachep;
  58 struct kmem_cache *btrfs_trans_handle_cachep;
  59 struct kmem_cache *btrfs_transaction_cachep;
  60 struct kmem_cache *btrfs_bit_radix_cachep;
  61 struct kmem_cache *btrfs_path_cachep;
  62
  63 #define S_SHIFT 12
  64 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  65         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  66         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  67         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  68         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  69         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  70         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  71         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  72 };
  73
  74 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
  75 {
  76         struct btrfs_root *root = BTRFS_I(inode)->root;
  77         struct btrfs_trans_handle *trans;
  78         struct btrfs_key ins;
  79         u64 alloc_hint = 0;
  80         u64 num_blocks;
  81         int ret;
  82         u64 blocksize = 1 << inode->i_blkbits;
  83
  84         mutex_lock(&root->fs_info->fs_mutex);
  85         trans = btrfs_start_transaction(root, 1);
  86         btrfs_set_trans_block_group(trans, inode);
  87         BUG_ON(!trans);
  88         num_blocks = (end - start + blocksize) & ~(blocksize - 1);
  89         ret = btrfs_drop_extents(trans, root, inode,
  90                                  start, start + num_blocks, &alloc_hint);
  91         num_blocks = num_blocks >> inode->i_blkbits;
  92         ret = btrfs_alloc_extent(trans, root, inode->i_ino, num_blocks, 0,
  93                                  alloc_hint, (u64)-1, &ins, 1);
  94         if (ret) {
  95                 WARN_ON(1);
  96                 goto out;
  97         }
  98         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
  99                                        start, ins.objectid, ins.offset,
 100                                        ins.offset);
 101 out:
 102         btrfs_end_transaction(trans, root);
 103         mutex_unlock(&root->fs_info->fs_mutex);
 104         return ret;
 105 }
 106
 107 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 108 {
 109         struct inode *inode = page->mapping->host;
 110         struct btrfs_root *root = BTRFS_I(inode)->root;
 111         struct btrfs_trans_handle *trans;
 112         char *kaddr;
 113         int ret;
 114         u64 page_start = page->index << PAGE_CACHE_SHIFT;
 115         size_t offset = start - page_start;
 116
 117         mutex_lock(&root->fs_info->fs_mutex);
 118         trans = btrfs_start_transaction(root, 1);
 119         btrfs_set_trans_block_group(trans, inode);
 120         kaddr = kmap(page);
 121         btrfs_csum_file_block(trans, root, inode->i_ino,
 122                               start, kaddr + offset, end - start + 1);
 123         kunmap(page);
 124         ret = btrfs_end_transaction(trans, root);
 125         BUG_ON(ret);
 126         mutex_unlock(&root->fs_info->fs_mutex);
 127         return ret;
 128 }
 129
 130 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 131 {
 132         int ret = 0;
 133         struct inode *inode = page->mapping->host;
 134         struct btrfs_root *root = BTRFS_I(inode)->root;
 135         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 136         struct btrfs_csum_item *item;
 137         struct btrfs_path *path = NULL;
 138         u64 private;
 139
 140         mutex_lock(&root->fs_info->fs_mutex);
 141         path = btrfs_alloc_path();
 142         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 143         if (IS_ERR(item)) {
 144                 ret = PTR_ERR(item);
 145                 /* a csum that isn't present is a preallocated region. */
 146                 if (ret == -ENOENT || ret == -EFBIG)
 147                         ret = 0;
 148                 private = 0;
 149                 goto out;
 150         }
 151         memcpy((char *)&private, &item->csum, BTRFS_CRC32_SIZE);
 152         set_state_private(em_tree, start, private);
 153 out:
 154         if (path)
 155                 btrfs_free_path(path);
 156         mutex_unlock(&root->fs_info->fs_mutex);
 157         return ret;
 158 }
 159
 160 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end)
 161 {
 162         size_t offset = start - (page->index << PAGE_CACHE_SHIFT);
 163         struct inode *inode = page->mapping->host;
 164         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 165         char *kaddr;
 166         u64 private;
 167         int ret;
 168
 169         ret = get_state_private(em_tree, start, &private);
 170         kaddr = kmap_atomic(page, KM_IRQ0);
 171         if (ret) {
 172                 goto zeroit;
 173         }
 174         /*
 175         struct btrfs_root *root = BTRFS_I(inode)->root;
 176         char csum[BTRFS_CRC32_SIZE];
 177         ret = btrfs_csum_data(root, kaddr + offset, end - start + 1, csum);
 178         BUG_ON(ret);
 179         if (memcmp(csum, &private, BTRFS_CRC32_SIZE)) {
 180                 goto zeroit;
 181         }
 182         */
 183         kunmap_atomic(kaddr, KM_IRQ0);
 184         return 0;
 185
 186 zeroit:
 187         printk("btrfs csum failed ino %lu off %llu\n",
 188                page->mapping->host->i_ino, (unsigned long long)start);
 189         memset(kaddr + offset, 1, end - start + 1); flush_dcache_page(page);
 190         kunmap_atomic(kaddr, KM_IRQ0);
 191         return 0;
 192 }
 193
 194 void btrfs_read_locked_inode(struct inode *inode)
 195 {
 196         struct btrfs_path *path;
 197         struct extent_buffer *leaf;
 198         struct btrfs_inode_item *inode_item;
 199         struct btrfs_inode_timespec *tspec;
 200         struct btrfs_root *root = BTRFS_I(inode)->root;
 201         struct btrfs_key location;
 202         u64 alloc_group_block;
 203         u32 rdev;
 204         int ret;
 205
 206         path = btrfs_alloc_path();
 207         BUG_ON(!path);
 208         mutex_lock(&root->fs_info->fs_mutex);
 209
 210         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 211         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 212         if (ret)
 213                 goto make_bad;
 214
 215         leaf = path->nodes[0];
 216         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 217                                     struct btrfs_inode_item);
 218
 219         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 220         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 221         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 222         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 223         inode->i_size = btrfs_inode_size(leaf, inode_item);
 224
 225         tspec = btrfs_inode_atime(inode_item);
 226         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 227         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 228
 229         tspec = btrfs_inode_mtime(inode_item);
 230         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 231         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 232
 233         tspec = btrfs_inode_ctime(inode_item);
 234         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 235         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 236
 237         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 238         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 239         inode->i_rdev = 0;
 240         rdev = btrfs_inode_rdev(leaf, inode_item);
 241
 242         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 243         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 244                                                        alloc_group_block);
 245
 246         btrfs_free_path(path);
 247         inode_item = NULL;
 248
 249         mutex_unlock(&root->fs_info->fs_mutex);
 250
 251         switch (inode->i_mode & S_IFMT) {
 252         case S_IFREG:
 253                 inode->i_mapping->a_ops = &btrfs_aops;
 254                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
 255                 inode->i_fop = &btrfs_file_operations;
 256                 inode->i_op = &btrfs_file_inode_operations;
 257                 break;
 258         case S_IFDIR:
 259                 inode->i_fop = &btrfs_dir_file_operations;
 260                 if (root == root->fs_info->tree_root)
 261                         inode->i_op = &btrfs_dir_ro_inode_operations;
 262                 else
 263                         inode->i_op = &btrfs_dir_inode_operations;
 264                 break;
 265         case S_IFLNK:
 266                 inode->i_op = &btrfs_symlink_inode_operations;
 267                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 268                 break;
 269         default:
 270                 init_special_inode(inode, inode->i_mode, rdev);
 271                 break;
 272         }
 273         return;
 274
 275 make_bad:
 276         btrfs_release_path(root, path);
 277         btrfs_free_path(path);
 278         mutex_unlock(&root->fs_info->fs_mutex);
 279         make_bad_inode(inode);
 280 }
 281
 282 static void fill_inode_item(struct extent_buffer *leaf,
 283                             struct btrfs_inode_item *item,
 284                             struct inode *inode)
 285 {
 286         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 287         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 288         btrfs_set_inode_size(leaf, item, inode->i_size);
 289         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 290         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 291
 292         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 293                                inode->i_atime.tv_sec);
 294         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 295                                 inode->i_atime.tv_nsec);
 296
 297         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 298                                inode->i_mtime.tv_sec);
 299         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 300                                 inode->i_mtime.tv_nsec);
 301
 302         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 303                                inode->i_ctime.tv_sec);
 304         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 305                                 inode->i_ctime.tv_nsec);
 306
 307         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 308         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 309         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 310         btrfs_set_inode_block_group(leaf, item,
 311                                     BTRFS_I(inode)->block_group->key.objectid);
 312 }
 313
 314 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 315                               struct btrfs_root *root,
 316                               struct inode *inode)
 317 {
 318         struct btrfs_inode_item *inode_item;
 319         struct btrfs_path *path;
 320         struct extent_buffer *leaf;
 321         int ret;
 322
 323         path = btrfs_alloc_path();
 324         BUG_ON(!path);
 325         ret = btrfs_lookup_inode(trans, root, path,
 326                                  &BTRFS_I(inode)->location, 1);
 327         if (ret) {
 328                 if (ret > 0)
 329                         ret = -ENOENT;
 330                 goto failed;
 331         }
 332
 333         leaf = path->nodes[0];
 334         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 335                                   struct btrfs_inode_item);
 336
 337         fill_inode_item(leaf, inode_item, inode);
 338         btrfs_mark_buffer_dirty(leaf);
 339         btrfs_set_inode_last_trans(trans, inode);
 340         ret = 0;
 341 failed:
 342         btrfs_release_path(root, path);
 343         btrfs_free_path(path);
 344         return ret;
 345 }
 346
 347
 348 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 349                               struct btrfs_root *root,
 350                               struct inode *dir,
 351                               struct dentry *dentry)
 352 {
 353         struct btrfs_path *path;
 354         const char *name = dentry->d_name.name;
 355         int name_len = dentry->d_name.len;
 356         int ret = 0;
 357         struct extent_buffer *leaf;
 358         struct btrfs_dir_item *di;
 359         struct btrfs_key key;
 360
 361         path = btrfs_alloc_path();
 362         if (!path) {
 363                 ret = -ENOMEM;
 364                 goto err;
 365         }
 366
 367         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 368                                     name, name_len, -1);
 369         if (IS_ERR(di)) {
 370                 ret = PTR_ERR(di);
 371                 goto err;
 372         }
 373         if (!di) {
 374                 ret = -ENOENT;
 375                 goto err;
 376         }
 377         leaf = path->nodes[0];
 378         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 379         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 380         if (ret)
 381                 goto err;
 382         btrfs_release_path(root, path);
 383
 384         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 385                                          key.objectid, name, name_len, -1);
 386         if (IS_ERR(di)) {
 387                 ret = PTR_ERR(di);
 388                 goto err;
 389         }
 390         if (!di) {
 391                 ret = -ENOENT;
 392                 goto err;
 393         }
 394         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 395
 396         dentry->d_inode->i_ctime = dir->i_ctime;
 397 err:
 398         btrfs_free_path(path);
 399         if (!ret) {
 400                 dir->i_size -= name_len * 2;
 401                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 402                 btrfs_update_inode(trans, root, dir);
 403                 drop_nlink(dentry->d_inode);
 404                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 405                 dir->i_sb->s_dirt = 1;
 406         }
 407         return ret;
 408 }
 409
 410 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 411 {
 412         struct btrfs_root *root;
 413         struct btrfs_trans_handle *trans;
 414         int ret;
 415         unsigned long nr;
 416
 417         root = BTRFS_I(dir)->root;
 418         mutex_lock(&root->fs_info->fs_mutex);
 419         trans = btrfs_start_transaction(root, 1);
 420
 421         btrfs_set_trans_block_group(trans, dir);
 422         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 423         nr = trans->blocks_used;
 424
 425         btrfs_end_transaction(trans, root);
 426         mutex_unlock(&root->fs_info->fs_mutex);
 427         btrfs_btree_balance_dirty(root, nr);
 428
 429         return ret;
 430 }
 431
 432 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 433 {
 434         struct inode *inode = dentry->d_inode;
 435         int err;
 436         int ret;
 437         struct btrfs_root *root = BTRFS_I(dir)->root;
 438         struct btrfs_path *path;
 439         struct btrfs_key key;
 440         struct btrfs_trans_handle *trans;
 441         struct btrfs_key found_key;
 442         int found_type;
 443         struct extent_buffer *leaf;
 444         char *goodnames = "..";
 445         unsigned long nr;
 446
 447         path = btrfs_alloc_path();
 448         BUG_ON(!path);
 449         mutex_lock(&root->fs_info->fs_mutex);
 450         trans = btrfs_start_transaction(root, 1);
 451
 452         btrfs_set_trans_block_group(trans, dir);
 453         key.objectid = inode->i_ino;
 454         key.offset = (u64)-1;
 455         key.type = (u8)-1;
 456         while(1) {
 457                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 458                 if (ret < 0) {
 459                         err = ret;
 460                         goto out;
 461                 }
 462                 BUG_ON(ret == 0);
 463                 if (path->slots[0] == 0) {
 464                         err = -ENOENT;
 465                         goto out;
 466                 }
 467                 path->slots[0]--;
 468                 leaf = path->nodes[0];
 469                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 470                 found_type = btrfs_key_type(&found_key);
 471                 if (found_key.objectid != inode->i_ino) {
 472                         err = -ENOENT;
 473                         goto out;
 474                 }
 475                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
 476                      found_type != BTRFS_DIR_INDEX_KEY) ||
 477                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
 478                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
 479                         err = -ENOTEMPTY;
 480                         goto out;
 481                 }
 482                 ret = btrfs_del_item(trans, root, path);
 483                 BUG_ON(ret);
 484
 485                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
 486                         break;
 487                 btrfs_release_path(root, path);
 488         }
 489         ret = 0;
 490         btrfs_release_path(root, path);
 491
 492         /* now the directory is empty */
 493         err = btrfs_unlink_trans(trans, root, dir, dentry);
 494         if (!err) {
 495                 inode->i_size = 0;
 496         }
 497 out:
 498         btrfs_release_path(root, path);
 499         btrfs_free_path(path);
 500         mutex_unlock(&root->fs_info->fs_mutex);
 501         nr = trans->blocks_used;
 502         ret = btrfs_end_transaction(trans, root);
 503         btrfs_btree_balance_dirty(root, nr);
 504         if (ret && !err)
 505                 err = ret;
 506         return err;
 507 }
 508
 509 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 510                             struct btrfs_root *root,
 511                             struct inode *inode)
 512 {
 513         struct btrfs_path *path;
 514         int ret;
 515
 516         clear_inode(inode);
 517
 518         path = btrfs_alloc_path();
 519         BUG_ON(!path);
 520         ret = btrfs_lookup_inode(trans, root, path,
 521                                  &BTRFS_I(inode)->location, -1);
 522         if (ret > 0)
 523                 ret = -ENOENT;
 524         if (!ret)
 525                 ret = btrfs_del_item(trans, root, path);
 526         btrfs_free_path(path);
 527         return ret;
 528 }
 529
 530 /*
 531  * this can truncate away extent items, csum items and directory items.
 532  * It starts at a high offset and removes keys until it can't find
 533  * any higher than i_size.
 534  *
 535  * csum items that cross the new i_size are truncated to the new size
 536  * as well.
 537  */
 538 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 539                                    struct btrfs_root *root,
 540                                    struct inode *inode)
 541 {
 542         int ret;
 543         struct btrfs_path *path;
 544         struct btrfs_key key;
 545         struct btrfs_key found_key;
 546         u32 found_type;
 547         struct extent_buffer *leaf;
 548         struct btrfs_file_extent_item *fi;
 549         u64 extent_start = 0;
 550         u64 extent_num_blocks = 0;
 551         u64 item_end = 0;
 552         int found_extent;
 553         int del_item;
 554
 555         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 556         path = btrfs_alloc_path();
 557         path->reada = -1;
 558         BUG_ON(!path);
 559
 560         /* FIXME, add redo link to tree so we don't leak on crash */
 561         key.objectid = inode->i_ino;
 562         key.offset = (u64)-1;
 563         key.type = (u8)-1;
 564
 565         while(1) {
 566                 btrfs_init_path(path);
 567                 fi = NULL;
 568                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 569                 if (ret < 0) {
 570                         goto error;
 571                 }
 572                 if (ret > 0) {
 573                         BUG_ON(path->slots[0] == 0);
 574                         path->slots[0]--;
 575                 }
 576                 leaf = path->nodes[0];
 577                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 578                 found_type = btrfs_key_type(&found_key);
 579
 580                 if (found_key.objectid != inode->i_ino)
 581                         break;
 582
 583                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 584                     found_type != BTRFS_DIR_ITEM_KEY &&
 585                     found_type != BTRFS_DIR_INDEX_KEY &&
 586                     found_type != BTRFS_EXTENT_DATA_KEY)
 587                         break;
 588
 589                 item_end = found_key.offset;
 590                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 591                         fi = btrfs_item_ptr(leaf, path->slots[0],
 592                                             struct btrfs_file_extent_item);
 593                         if (btrfs_file_extent_type(leaf, fi) !=
 594                             BTRFS_FILE_EXTENT_INLINE) {
 595                                 item_end +=
 596                                     btrfs_file_extent_num_blocks(leaf, fi) <<
 597                                     inode->i_blkbits;
 598                         }
 599                 }
 600                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 601                         ret = btrfs_csum_truncate(trans, root, path,
 602                                                   inode->i_size);
 603                         BUG_ON(ret);
 604                 }
 605                 if (item_end < inode->i_size) {
 606                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 607                                 found_type = BTRFS_INODE_ITEM_KEY;
 608                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 609                                 found_type = BTRFS_CSUM_ITEM_KEY;
 610                         } else if (found_type) {
 611                                 found_type--;
 612                         } else {
 613                                 break;
 614                         }
 615                         btrfs_set_key_type(&key, found_type);
 616                         continue;
 617                 }
 618                 if (found_key.offset >= inode->i_size)
 619                         del_item = 1;
 620                 else
 621                         del_item = 0;
 622                 found_extent = 0;
 623
 624                 /* FIXME, shrink the extent if the ref count is only 1 */
 625                 if (found_type == BTRFS_EXTENT_DATA_KEY &&
 626                            btrfs_file_extent_type(leaf, fi) !=
 627                            BTRFS_FILE_EXTENT_INLINE) {
 628                         u64 num_dec;
 629                         extent_start = btrfs_file_extent_disk_blocknr(leaf, fi);
 630                         if (!del_item) {
 631                                 u64 orig_num_blocks =
 632                                         btrfs_file_extent_num_blocks(leaf, fi);
 633                                 extent_num_blocks = inode->i_size -
 634                                         found_key.offset + root->sectorsize - 1;
 635                                 extent_num_blocks >>= inode->i_blkbits;
 636                                 btrfs_set_file_extent_num_blocks(leaf, fi,
 637                                                          extent_num_blocks);
 638                                 num_dec = (orig_num_blocks -
 639                                            extent_num_blocks) << 3;
 640                                 if (extent_start != 0) {
 641                                         inode->i_blocks -= num_dec;
 642                                 }
 643                                 btrfs_mark_buffer_dirty(leaf);
 644                         } else {
 645                                 extent_num_blocks =
 646                                         btrfs_file_extent_disk_num_blocks(leaf,
 647                                                                           fi);
 648                                 /* FIXME blocksize != 4096 */
 649                                 num_dec = btrfs_file_extent_num_blocks(leaf,
 650                                                                        fi) << 3;
 651                                 if (extent_start != 0) {
 652                                         found_extent = 1;
 653                                         inode->i_blocks -= num_dec;
 654                                 }
 655                         }
 656                 }
 657                 if (del_item) {
 658                         ret = btrfs_del_item(trans, root, path);
 659                         if (ret)
 660                                 goto error;
 661                 } else {
 662                         break;
 663                 }
 664                 btrfs_release_path(root, path);
 665                 if (found_extent) {
 666                         ret = btrfs_free_extent(trans, root, extent_start,
 667                                                 extent_num_blocks, 0);
 668                         BUG_ON(ret);
 669                 }
 670         }
 671         ret = 0;
 672 error:
 673         btrfs_release_path(root, path);
 674         btrfs_free_path(path);
 675         inode->i_sb->s_dirt = 1;
 676         return ret;
 677 }
 678
 679 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 680                               size_t zero_start)
 681 {
 682         char *kaddr;
 683         int ret = 0;
 684         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 685         u64 page_start = page->index << PAGE_CACHE_SHIFT;
 686         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 687
 688         set_page_extent_mapped(page);
 689
 690         lock_extent(em_tree, page_start, page_end, GFP_NOFS);
 691         set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
 692                             page_end, GFP_NOFS);
 693         if (zero_start != PAGE_CACHE_SIZE) {
 694                 kaddr = kmap(page);
 695                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 696                 flush_dcache_page(page);
 697                 kunmap(page);
 698         }
 699         set_page_dirty(page);
 700         unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
 701
 702         return ret;
 703 }
 704
 705 /*
 706  * taken from block_truncate_page, but does cow as it zeros out
 707  * any bytes left in the last page in the file.
 708  */
 709 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 710 {
 711         struct inode *inode = mapping->host;
 712         unsigned blocksize = 1 << inode->i_blkbits;
 713         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 714         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 715         struct page *page;
 716         int ret = 0;
 717         u64 page_start;
 718
 719         if ((offset & (blocksize - 1)) == 0)
 720                 goto out;
 721
 722         down_read(&BTRFS_I(inode)->root->snap_sem);
 723         ret = -ENOMEM;
 724         page = grab_cache_page(mapping, index);
 725         if (!page)
 726                 goto out;
 727         if (!PageUptodate(page)) {
 728                 ret = btrfs_readpage(NULL, page);
 729                 lock_page(page);
 730                 if (!PageUptodate(page)) {
 731                         ret = -EIO;
 732                         goto out;
 733                 }
 734         }
 735         page_start = page->index << PAGE_CACHE_SHIFT;
 736
 737         ret = btrfs_cow_one_page(inode, page, offset);
 738
 739         unlock_page(page);
 740         page_cache_release(page);
 741         up_read(&BTRFS_I(inode)->root->snap_sem);
 742 out:
 743         return ret;
 744 }
 745
 746 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 747 {
 748         struct inode *inode = dentry->d_inode;
 749         int err;
 750
 751         err = inode_change_ok(inode, attr);
 752         if (err)
 753                 return err;
 754
 755         if (S_ISREG(inode->i_mode) &&
 756             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 757                 struct btrfs_trans_handle *trans;
 758                 struct btrfs_root *root = BTRFS_I(inode)->root;
 759                 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 760
 761                 u64 mask = root->sectorsize - 1;
 762                 u64 pos = (inode->i_size + mask) & ~mask;
 763                 u64 block_end = attr->ia_size | mask;
 764                 u64 hole_size;
 765                 u64 alloc_hint;
 766
 767                 if (attr->ia_size <= pos)
 768                         goto out;
 769
 770                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 771
 772                 lock_extent(em_tree, pos, block_end, GFP_NOFS);
 773                 hole_size = (attr->ia_size - pos + mask) & ~mask;
 774
 775                 mutex_lock(&root->fs_info->fs_mutex);
 776                 trans = btrfs_start_transaction(root, 1);
 777                 btrfs_set_trans_block_group(trans, inode);
 778                 err = btrfs_drop_extents(trans, root, inode,
 779                                          pos, pos + hole_size, &alloc_hint);
 780
 781                 hole_size >>= inode->i_blkbits;
 782
 783                 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
 784                                                pos, 0, 0, hole_size);
 785                 btrfs_end_transaction(trans, root);
 786                 mutex_unlock(&root->fs_info->fs_mutex);
 787                 unlock_extent(em_tree, pos, block_end, GFP_NOFS);
 788                 if (err)
 789                         return err;
 790         }
 791 out:
 792         err = inode_setattr(inode, attr);
 793
 794         return err;
 795 }
 796 void btrfs_delete_inode(struct inode *inode)
 797 {
 798         struct btrfs_trans_handle *trans;
 799         struct btrfs_root *root = BTRFS_I(inode)->root;
 800         unsigned long nr;
 801         int ret;
 802
 803         truncate_inode_pages(&inode->i_data, 0);
 804         if (is_bad_inode(inode)) {
 805                 goto no_delete;
 806         }
 807
 808         inode->i_size = 0;
 809         mutex_lock(&root->fs_info->fs_mutex);
 810         trans = btrfs_start_transaction(root, 1);
 811
 812         btrfs_set_trans_block_group(trans, inode);
 813         ret = btrfs_truncate_in_trans(trans, root, inode);
 814         if (ret)
 815                 goto no_delete_lock;
 816         ret = btrfs_free_inode(trans, root, inode);
 817         if (ret)
 818                 goto no_delete_lock;
 819         nr = trans->blocks_used;
 820
 821         btrfs_end_transaction(trans, root);
 822         mutex_unlock(&root->fs_info->fs_mutex);
 823         btrfs_btree_balance_dirty(root, nr);
 824         return;
 825
 826 no_delete_lock:
 827         nr = trans->blocks_used;
 828         btrfs_end_transaction(trans, root);
 829         mutex_unlock(&root->fs_info->fs_mutex);
 830         btrfs_btree_balance_dirty(root, nr);
 831 no_delete:
 832         clear_inode(inode);
 833 }
 834
 835 /*
 836  * this returns the key found in the dir entry in the location pointer.
 837  * If no dir entries were found, location->objectid is 0.
 838  */
 839 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 840                                struct btrfs_key *location)
 841 {
 842         const char *name = dentry->d_name.name;
 843         int namelen = dentry->d_name.len;
 844         struct btrfs_dir_item *di;
 845         struct btrfs_path *path;
 846         struct btrfs_root *root = BTRFS_I(dir)->root;
 847         int ret;
 848
 849         path = btrfs_alloc_path();
 850         BUG_ON(!path);
 851         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
 852                                     namelen, 0);
 853         if (!di || IS_ERR(di)) {
 854                 location->objectid = 0;
 855                 ret = 0;
 856                 goto out;
 857         }
 858         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
 859 out:
 860         btrfs_release_path(root, path);
 861         btrfs_free_path(path);
 862         return ret;
 863 }
 864
 865 /*
 866  * when we hit a tree root in a directory, the btrfs part of the inode
 867  * needs to be changed to reflect the root directory of the tree root.  This
 868  * is kind of like crossing a mount point.
 869  */
 870 static int fixup_tree_root_location(struct btrfs_root *root,
 871                              struct btrfs_key *location,
 872                              struct btrfs_root **sub_root,
 873                              struct dentry *dentry)
 874 {
 875         struct btrfs_path *path;
 876         struct btrfs_root_item *ri;
 877
 878         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 879                 return 0;
 880         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 881                 return 0;
 882
 883         path = btrfs_alloc_path();
 884         BUG_ON(!path);
 885         mutex_lock(&root->fs_info->fs_mutex);
 886
 887         *sub_root = btrfs_read_fs_root(root->fs_info, location,
 888                                         dentry->d_name.name,
 889                                         dentry->d_name.len);
 890         if (IS_ERR(*sub_root))
 891                 return PTR_ERR(*sub_root);
 892
 893         ri = &(*sub_root)->root_item;
 894         location->objectid = btrfs_root_dirid(ri);
 895         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 896         location->offset = 0;
 897
 898         btrfs_free_path(path);
 899         mutex_unlock(&root->fs_info->fs_mutex);
 900         return 0;
 901 }
 902
 903 static int btrfs_init_locked_inode(struct inode *inode, void *p)
 904 {
 905         struct btrfs_iget_args *args = p;
 906         inode->i_ino = args->ino;
 907         BTRFS_I(inode)->root = args->root;
 908         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
 909                              inode->i_mapping, GFP_NOFS);
 910         return 0;
 911 }
 912
 913 static int btrfs_find_actor(struct inode *inode, void *opaque)
 914 {
 915         struct btrfs_iget_args *args = opaque;
 916         return (args->ino == inode->i_ino &&
 917                 args->root == BTRFS_I(inode)->root);
 918 }
 919
 920 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 921                                 struct btrfs_root *root)
 922 {
 923         struct inode *inode;
 924         struct btrfs_iget_args args;
 925         args.ino = objectid;
 926         args.root = root;
 927
 928         inode = iget5_locked(s, objectid, btrfs_find_actor,
 929                              btrfs_init_locked_inode,
 930                              (void *)&args);
 931         return inode;
 932 }
 933
 934 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 935                                    struct nameidata *nd)
 936 {
 937         struct inode * inode;
 938         struct btrfs_inode *bi = BTRFS_I(dir);
 939         struct btrfs_root *root = bi->root;
 940         struct btrfs_root *sub_root = root;
 941         struct btrfs_key location;
 942         int ret;
 943
 944         if (dentry->d_name.len > BTRFS_NAME_LEN)
 945                 return ERR_PTR(-ENAMETOOLONG);
 946
 947         mutex_lock(&root->fs_info->fs_mutex);
 948         ret = btrfs_inode_by_name(dir, dentry, &location);
 949         mutex_unlock(&root->fs_info->fs_mutex);
 950
 951         if (ret < 0)
 952                 return ERR_PTR(ret);
 953
 954         inode = NULL;
 955         if (location.objectid) {
 956                 ret = fixup_tree_root_location(root, &location, &sub_root,
 957                                                 dentry);
 958                 if (ret < 0)
 959                         return ERR_PTR(ret);
 960                 if (ret > 0)
 961                         return ERR_PTR(-ENOENT);
 962                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 963                                           sub_root);
 964                 if (!inode)
 965                         return ERR_PTR(-EACCES);
 966                 if (inode->i_state & I_NEW) {
 967                         /* the inode and parent dir are two different roots */
 968                         if (sub_root != root) {
 969                                 igrab(inode);
 970                                 sub_root->inode = inode;
 971                         }
 972                         BTRFS_I(inode)->root = sub_root;
 973                         memcpy(&BTRFS_I(inode)->location, &location,
 974                                sizeof(location));
 975                         btrfs_read_locked_inode(inode);
 976                         unlock_new_inode(inode);
 977                 }
 978         }
 979         return d_splice_alias(inode, dentry);
 980 }
 981
 982 static unsigned char btrfs_filetype_table[] = {
 983         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
 984 };
 985
 986 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 987 {
 988         struct inode *inode = filp->f_path.dentry->d_inode;
 989         struct btrfs_root *root = BTRFS_I(inode)->root;
 990         struct btrfs_item *item;
 991         struct btrfs_dir_item *di;
 992         struct btrfs_key key;
 993         struct btrfs_key found_key;
 994         struct btrfs_path *path;
 995         int ret;
 996         u32 nritems;
 997         struct extent_buffer *leaf;
 998         int slot;
 999         int advance;
1000         unsigned char d_type;
1001         int over = 0;
1002         u32 di_cur;
1003         u32 di_total;
1004         u32 di_len;
1005         int key_type = BTRFS_DIR_INDEX_KEY;
1006         char tmp_name[32];
1007         char *name_ptr;
1008         int name_len;
1009
1010         /* FIXME, use a real flag for deciding about the key type */
1011         if (root->fs_info->tree_root == root)
1012                 key_type = BTRFS_DIR_ITEM_KEY;
1013
1014         mutex_lock(&root->fs_info->fs_mutex);
1015         key.objectid = inode->i_ino;
1016         btrfs_set_key_type(&key, key_type);
1017         key.offset = filp->f_pos;
1018
1019         path = btrfs_alloc_path();
1020         path->reada = 2;
1021         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1022         if (ret < 0)
1023                 goto err;
1024         advance = 0;
1025         while(1) {
1026                 leaf = path->nodes[0];
1027                 nritems = btrfs_header_nritems(leaf);
1028                 slot = path->slots[0];
1029                 if (advance || slot >= nritems) {
1030                         if (slot >= nritems -1) {
1031                                 ret = btrfs_next_leaf(root, path);
1032                                 if (ret)
1033                                         break;
1034                                 leaf = path->nodes[0];
1035                                 nritems = btrfs_header_nritems(leaf);
1036                                 slot = path->slots[0];
1037                         } else {
1038                                 slot++;
1039                                 path->slots[0]++;
1040                         }
1041                 }
1042                 advance = 1;
1043                 item = btrfs_item_nr(leaf, slot);
1044                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1045
1046                 if (found_key.objectid != key.objectid)
1047                         break;
1048                 if (btrfs_key_type(&found_key) != key_type)
1049                         break;
1050                 if (found_key.offset < filp->f_pos)
1051                         continue;
1052
1053                 filp->f_pos = found_key.offset;
1054                 advance = 1;
1055                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1056                 di_cur = 0;
1057                 di_total = btrfs_item_size(leaf, item);
1058                 while(di_cur < di_total) {
1059                         struct btrfs_key location;
1060
1061                         name_len = btrfs_dir_name_len(leaf, di);
1062                         if (name_len < 32) {
1063                                 name_ptr = tmp_name;
1064                         } else {
1065                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1066                                 BUG_ON(!name_ptr);
1067                         }
1068                         read_extent_buffer(leaf, name_ptr,
1069                                            (unsigned long)(di + 1), name_len);
1070
1071                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1072                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1073
1074                         over = filldir(dirent, name_ptr, name_len,
1075                                        found_key.offset,
1076                                        location.objectid,
1077                                        d_type);
1078
1079                         if (name_ptr != tmp_name)
1080                                 kfree(name_ptr);
1081
1082                         if (over)
1083                                 goto nopos;
1084                         di_len = btrfs_dir_name_len(leaf, di) + sizeof(*di);
1085                         di_cur += di_len;
1086                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1087                 }
1088         }
1089         filp->f_pos++;
1090 nopos:
1091         ret = 0;
1092 err:
1093         btrfs_release_path(root, path);
1094         btrfs_free_path(path);
1095         mutex_unlock(&root->fs_info->fs_mutex);
1096         return ret;
1097 }
1098
1099 int btrfs_write_inode(struct inode *inode, int wait)
1100 {
1101         struct btrfs_root *root = BTRFS_I(inode)->root;
1102         struct btrfs_trans_handle *trans;
1103         int ret = 0;
1104
1105         if (wait) {
1106                 mutex_lock(&root->fs_info->fs_mutex);
1107                 trans = btrfs_start_transaction(root, 1);
1108                 btrfs_set_trans_block_group(trans, inode);
1109                 ret = btrfs_commit_transaction(trans, root);
1110                 mutex_unlock(&root->fs_info->fs_mutex);
1111         }
1112         return ret;
1113 }
1114
1115 /*
1116  * This is somewhat expensive, updating the tree every time the
1117  * inode changes.  But, it is most likely to find the inode in cache.
1118  * FIXME, needs more benchmarking...there are no reasons other than performance
1119  * to keep or drop this code.
1120  */
1121 void btrfs_dirty_inode(struct inode *inode)
1122 {
1123         struct btrfs_root *root = BTRFS_I(inode)->root;
1124         struct btrfs_trans_handle *trans;
1125
1126         mutex_lock(&root->fs_info->fs_mutex);
1127         trans = btrfs_start_transaction(root, 1);
1128         btrfs_set_trans_block_group(trans, inode);
1129         btrfs_update_inode(trans, root, inode);
1130         btrfs_end_transaction(trans, root);
1131         mutex_unlock(&root->fs_info->fs_mutex);
1132 }
1133
1134 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1135                                      struct btrfs_root *root,
1136                                      u64 objectid,
1137                                      struct btrfs_block_group_cache *group,
1138                                      int mode)
1139 {
1140         struct inode *inode;
1141         struct btrfs_inode_item *inode_item;
1142         struct btrfs_key *location;
1143         struct btrfs_path *path;
1144         int ret;
1145         int owner;
1146
1147         path = btrfs_alloc_path();
1148         BUG_ON(!path);
1149
1150         inode = new_inode(root->fs_info->sb);
1151         if (!inode)
1152                 return ERR_PTR(-ENOMEM);
1153
1154         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1155                              inode->i_mapping, GFP_NOFS);
1156         BTRFS_I(inode)->root = root;
1157
1158         if (mode & S_IFDIR)
1159                 owner = 0;
1160         else
1161                 owner = 1;
1162         group = btrfs_find_block_group(root, group, 0, 0, owner);
1163         BTRFS_I(inode)->block_group = group;
1164
1165         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1166         if (ret)
1167                 goto fail;
1168
1169         inode->i_uid = current->fsuid;
1170         inode->i_gid = current->fsgid;
1171         inode->i_mode = mode;
1172         inode->i_ino = objectid;
1173         inode->i_blocks = 0;
1174         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1175         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1176                                   struct btrfs_inode_item);
1177         fill_inode_item(path->nodes[0], inode_item, inode);
1178         btrfs_mark_buffer_dirty(path->nodes[0]);
1179         btrfs_free_path(path);
1180
1181         location = &BTRFS_I(inode)->location;
1182         location->objectid = objectid;
1183         location->offset = 0;
1184         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1185
1186         insert_inode_hash(inode);
1187         return inode;
1188 fail:
1189         btrfs_free_path(path);
1190         return ERR_PTR(ret);
1191 }
1192
1193 static inline u8 btrfs_inode_type(struct inode *inode)
1194 {
1195         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1196 }
1197
1198 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1199                             struct dentry *dentry, struct inode *inode)
1200 {
1201         int ret;
1202         struct btrfs_key key;
1203         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1204         struct inode *parent_inode;
1205
1206         key.objectid = inode->i_ino;
1207         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1208         key.offset = 0;
1209
1210         ret = btrfs_insert_dir_item(trans, root,
1211                                     dentry->d_name.name, dentry->d_name.len,
1212                                     dentry->d_parent->d_inode->i_ino,
1213                                     &key, btrfs_inode_type(inode));
1214         if (ret == 0) {
1215                 parent_inode = dentry->d_parent->d_inode;
1216                 parent_inode->i_size += dentry->d_name.len * 2;
1217                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1218                 ret = btrfs_update_inode(trans, root,
1219                                          dentry->d_parent->d_inode);
1220         }
1221         return ret;
1222 }
1223
1224 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1225                             struct dentry *dentry, struct inode *inode)
1226 {
1227         int err = btrfs_add_link(trans, dentry, inode);
1228         if (!err) {
1229                 d_instantiate(dentry, inode);
1230                 return 0;
1231         }
1232         if (err > 0)
1233                 err = -EEXIST;
1234         return err;
1235 }
1236
1237 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1238                         int mode, dev_t rdev)
1239 {
1240         struct btrfs_trans_handle *trans;
1241         struct btrfs_root *root = BTRFS_I(dir)->root;
1242         struct inode *inode;
1243         int err;
1244         int drop_inode = 0;
1245         u64 objectid;
1246         unsigned long nr;
1247
1248         if (!new_valid_dev(rdev))
1249                 return -EINVAL;
1250
1251         mutex_lock(&root->fs_info->fs_mutex);
1252         trans = btrfs_start_transaction(root, 1);
1253         btrfs_set_trans_block_group(trans, dir);
1254
1255         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1256         if (err) {
1257                 err = -ENOSPC;
1258                 goto out_unlock;
1259         }
1260
1261         inode = btrfs_new_inode(trans, root, objectid,
1262                                 BTRFS_I(dir)->block_group, mode);
1263         err = PTR_ERR(inode);
1264         if (IS_ERR(inode))
1265                 goto out_unlock;
1266
1267         btrfs_set_trans_block_group(trans, inode);
1268         err = btrfs_add_nondir(trans, dentry, inode);
1269         if (err)
1270                 drop_inode = 1;
1271         else {
1272                 inode->i_op = &btrfs_special_inode_operations;
1273                 init_special_inode(inode, inode->i_mode, rdev);
1274                 btrfs_update_inode(trans, root, inode);
1275         }
1276         dir->i_sb->s_dirt = 1;
1277         btrfs_update_inode_block_group(trans, inode);
1278         btrfs_update_inode_block_group(trans, dir);
1279 out_unlock:
1280         nr = trans->blocks_used;
1281         btrfs_end_transaction(trans, root);
1282         mutex_unlock(&root->fs_info->fs_mutex);
1283
1284         if (drop_inode) {
1285                 inode_dec_link_count(inode);
1286                 iput(inode);
1287         }
1288         btrfs_btree_balance_dirty(root, nr);
1289         return err;
1290 }
1291
1292 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1293                         int mode, struct nameidata *nd)
1294 {
1295         struct btrfs_trans_handle *trans;
1296         struct btrfs_root *root = BTRFS_I(dir)->root;
1297         struct inode *inode;
1298         int err;
1299         int drop_inode = 0;
1300         unsigned long nr;
1301         u64 objectid;
1302
1303         mutex_lock(&root->fs_info->fs_mutex);
1304         trans = btrfs_start_transaction(root, 1);
1305         btrfs_set_trans_block_group(trans, dir);
1306
1307         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1308         if (err) {
1309                 err = -ENOSPC;
1310                 goto out_unlock;
1311         }
1312
1313         inode = btrfs_new_inode(trans, root, objectid,
1314                                 BTRFS_I(dir)->block_group, mode);
1315         err = PTR_ERR(inode);
1316         if (IS_ERR(inode))
1317                 goto out_unlock;
1318
1319         btrfs_set_trans_block_group(trans, inode);
1320         err = btrfs_add_nondir(trans, dentry, inode);
1321         if (err)
1322                 drop_inode = 1;
1323         else {
1324                 inode->i_mapping->a_ops = &btrfs_aops;
1325                 inode->i_fop = &btrfs_file_operations;
1326                 inode->i_op = &btrfs_file_inode_operations;
1327                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1328                                      inode->i_mapping, GFP_NOFS);
1329                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1330         }
1331         dir->i_sb->s_dirt = 1;
1332         btrfs_update_inode_block_group(trans, inode);
1333         btrfs_update_inode_block_group(trans, dir);
1334 out_unlock:
1335         nr = trans->blocks_used;
1336         btrfs_end_transaction(trans, root);
1337         mutex_unlock(&root->fs_info->fs_mutex);
1338
1339         if (drop_inode) {
1340                 inode_dec_link_count(inode);
1341                 iput(inode);
1342         }
1343         btrfs_btree_balance_dirty(root, nr);
1344         return err;
1345 }
1346
1347 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1348                       struct dentry *dentry)
1349 {
1350         struct btrfs_trans_handle *trans;
1351         struct btrfs_root *root = BTRFS_I(dir)->root;
1352         struct inode *inode = old_dentry->d_inode;
1353         unsigned long nr;
1354         int err;
1355         int drop_inode = 0;
1356
1357         if (inode->i_nlink == 0)
1358                 return -ENOENT;
1359
1360         inc_nlink(inode);
1361         mutex_lock(&root->fs_info->fs_mutex);
1362         trans = btrfs_start_transaction(root, 1);
1363
1364         btrfs_set_trans_block_group(trans, dir);
1365         atomic_inc(&inode->i_count);
1366         err = btrfs_add_nondir(trans, dentry, inode);
1367
1368         if (err)
1369                 drop_inode = 1;
1370
1371         dir->i_sb->s_dirt = 1;
1372         btrfs_update_inode_block_group(trans, dir);
1373         err = btrfs_update_inode(trans, root, inode);
1374
1375         if (err)
1376                 drop_inode = 1;
1377
1378         nr = trans->blocks_used;
1379         btrfs_end_transaction(trans, root);
1380         mutex_unlock(&root->fs_info->fs_mutex);
1381
1382         if (drop_inode) {
1383                 inode_dec_link_count(inode);
1384                 iput(inode);
1385         }
1386         btrfs_btree_balance_dirty(root, nr);
1387         return err;
1388 }
1389
1390 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
1391                                 struct btrfs_root *root,
1392                                 u64 objectid, u64 dirid)
1393 {
1394         int ret;
1395         char buf[2];
1396         struct btrfs_key key;
1397
1398         buf[0] = '.';
1399         buf[1] = '.';
1400
1401         key.objectid = objectid;
1402         key.offset = 0;
1403         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1404
1405         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
1406                                     &key, BTRFS_FT_DIR);
1407         if (ret)
1408                 goto error;
1409
1410         key.objectid = dirid;
1411         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
1412                                     &key, BTRFS_FT_DIR);
1413         if (ret)
1414                 goto error;
1415 error:
1416         return ret;
1417 }
1418
1419 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1420 {
1421         struct inode *inode;
1422         struct btrfs_trans_handle *trans;
1423         struct btrfs_root *root = BTRFS_I(dir)->root;
1424         int err = 0;
1425         int drop_on_err = 0;
1426         u64 objectid;
1427         unsigned long nr = 1;
1428
1429         mutex_lock(&root->fs_info->fs_mutex);
1430         trans = btrfs_start_transaction(root, 1);
1431         btrfs_set_trans_block_group(trans, dir);
1432
1433         if (IS_ERR(trans)) {
1434                 err = PTR_ERR(trans);
1435                 goto out_unlock;
1436         }
1437
1438         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1439         if (err) {
1440                 err = -ENOSPC;
1441                 goto out_unlock;
1442         }
1443
1444         inode = btrfs_new_inode(trans, root, objectid,
1445                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1446         if (IS_ERR(inode)) {
1447                 err = PTR_ERR(inode);
1448                 goto out_fail;
1449         }
1450
1451         drop_on_err = 1;
1452         inode->i_op = &btrfs_dir_inode_operations;
1453         inode->i_fop = &btrfs_dir_file_operations;
1454         btrfs_set_trans_block_group(trans, inode);
1455
1456         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
1457         if (err)
1458                 goto out_fail;
1459
1460         inode->i_size = 6;
1461         err = btrfs_update_inode(trans, root, inode);
1462         if (err)
1463                 goto out_fail;
1464
1465         err = btrfs_add_link(trans, dentry, inode);
1466         if (err)
1467                 goto out_fail;
1468
1469         d_instantiate(dentry, inode);
1470         drop_on_err = 0;
1471         dir->i_sb->s_dirt = 1;
1472         btrfs_update_inode_block_group(trans, inode);
1473         btrfs_update_inode_block_group(trans, dir);
1474
1475 out_fail:
1476         nr = trans->blocks_used;
1477         btrfs_end_transaction(trans, root);
1478
1479 out_unlock:
1480         mutex_unlock(&root->fs_info->fs_mutex);
1481         if (drop_on_err)
1482                 iput(inode);
1483         btrfs_btree_balance_dirty(root, nr);
1484         return err;
1485 }
1486
1487 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1488                                     size_t page_offset, u64 start, u64 end,
1489                                     int create)
1490 {
1491         int ret;
1492         int err = 0;
1493         u64 blocknr;
1494         u64 extent_start = 0;
1495         u64 extent_end = 0;
1496         u64 objectid = inode->i_ino;
1497         u32 found_type;
1498         int failed_insert = 0;
1499         struct btrfs_path *path;
1500         struct btrfs_root *root = BTRFS_I(inode)->root;
1501         struct btrfs_file_extent_item *item;
1502         struct extent_buffer *leaf;
1503         struct btrfs_key found_key;
1504         struct extent_map *em = NULL;
1505         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1506         struct btrfs_trans_handle *trans = NULL;
1507
1508         path = btrfs_alloc_path();
1509         BUG_ON(!path);
1510         mutex_lock(&root->fs_info->fs_mutex);
1511
1512 again:
1513         em = lookup_extent_mapping(em_tree, start, end);
1514         if (em) {
1515                 goto out;
1516         }
1517         if (!em) {
1518                 em = alloc_extent_map(GFP_NOFS);
1519                 if (!em) {
1520                         err = -ENOMEM;
1521                         goto out;
1522                 }
1523                 em->start = EXTENT_MAP_HOLE;
1524                 em->end = EXTENT_MAP_HOLE;
1525         }
1526         em->bdev = inode->i_sb->s_bdev;
1527         ret = btrfs_lookup_file_extent(NULL, root, path,
1528                                        objectid, start, 0);
1529         if (ret < 0) {
1530                 err = ret;
1531                 goto out;
1532         }
1533
1534         if (ret != 0) {
1535                 if (path->slots[0] == 0)
1536                         goto not_found;
1537                 path->slots[0]--;
1538         }
1539
1540         leaf = path->nodes[0];
1541         item = btrfs_item_ptr(leaf, path->slots[0],
1542                               struct btrfs_file_extent_item);
1543
1544         blocknr = btrfs_file_extent_disk_blocknr(leaf, item);
1545         blocknr += btrfs_file_extent_offset(leaf, item);
1546
1547         /* are we inside the extent that was found? */
1548         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1549         found_type = btrfs_key_type(&found_key);
1550         if (found_key.objectid != objectid ||
1551             found_type != BTRFS_EXTENT_DATA_KEY) {
1552                 goto not_found;
1553         }
1554
1555         found_type = btrfs_file_extent_type(leaf, item);
1556         extent_start = found_key.offset;
1557         if (found_type == BTRFS_FILE_EXTENT_REG) {
1558                 extent_end = extent_start +
1559                        (btrfs_file_extent_num_blocks(leaf, item) <<
1560                         inode->i_blkbits);
1561                 err = 0;
1562                 if (start < extent_start || start >= extent_end) {
1563                         em->start = start;
1564                         if (start < extent_start) {
1565                                 if (end < extent_start)
1566                                         goto not_found;
1567                                 em->end = extent_end - 1;
1568                         } else {
1569                                 em->end = end;
1570                         }
1571                         goto not_found_em;
1572                 }
1573                 if (btrfs_file_extent_disk_blocknr(leaf, item) == 0) {
1574                         em->start = extent_start;
1575                         em->end = extent_end - 1;
1576                         em->block_start = EXTENT_MAP_HOLE;
1577                         em->block_end = EXTENT_MAP_HOLE;
1578                         goto insert;
1579                 }
1580                 em->block_start = blocknr << inode->i_blkbits;
1581                 em->block_end = em->block_start +
1582                         (btrfs_file_extent_num_blocks(leaf, item) <<
1583                          inode->i_blkbits) - 1;
1584                 em->start = extent_start;
1585                 em->end = extent_end - 1;
1586                 goto insert;
1587         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1588                 unsigned long ptr;
1589                 char *map;
1590                 u32 size;
1591
1592                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1593                                                     path->slots[0]));
1594
1595                 extent_end = extent_start | ((u64)root->sectorsize - 1);
1596                 if (start < extent_start || start >= extent_end) {
1597                         em->start = start;
1598                         if (start < extent_start) {
1599                                 if (end < extent_start)
1600                                         goto not_found;
1601                                 em->end = extent_end;
1602                         } else {
1603                                 em->end = end;
1604                         }
1605                         goto not_found_em;
1606                 }
1607
1608                 em->block_start = EXTENT_MAP_INLINE;
1609                 em->block_end = EXTENT_MAP_INLINE;
1610                 em->start = extent_start;
1611                 em->end = extent_end;
1612
1613                 if (!page) {
1614                         goto insert;
1615                 }
1616
1617                 ptr = btrfs_file_extent_inline_start(item);
1618                 map = kmap(page);
1619                 read_extent_buffer(leaf, map + page_offset, ptr, size);
1620                 memset(map + page_offset + size, 0,
1621                        root->sectorsize - (page_offset + size));
1622                 flush_dcache_page(page);
1623                 kunmap(page);
1624                 set_extent_uptodate(em_tree, extent_start,
1625                                     extent_end, GFP_NOFS);
1626                 goto insert;
1627         } else {
1628                 printk("unkknown found_type %d\n", found_type);
1629                 WARN_ON(1);
1630         }
1631 not_found:
1632         em->start = start;
1633         em->end = end;
1634 not_found_em:
1635         em->block_start = EXTENT_MAP_HOLE;
1636         em->block_end = EXTENT_MAP_HOLE;
1637 insert:
1638         btrfs_release_path(root, path);
1639         if (em->start > start || em->end < start) {
1640                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1641                 err = -EIO;
1642                 goto out;
1643         }
1644         ret = add_extent_mapping(em_tree, em);
1645         if (ret == -EEXIST) {
1646                 free_extent_map(em);
1647                 em = NULL;
1648                 failed_insert++;
1649                 if (failed_insert > 5) {
1650                         printk("failing to insert %Lu %Lu\n", start, end);
1651                         err = -EIO;
1652                         goto out;
1653                 }
1654                 goto again;
1655         }
1656         err = 0;
1657 out:
1658         btrfs_free_path(path);
1659         if (trans) {
1660                 ret = btrfs_end_transaction(trans, root);
1661                 if (!err)
1662                         err = ret;
1663         }
1664         mutex_unlock(&root->fs_info->fs_mutex);
1665         if (err) {
1666                 free_extent_map(em);
1667                 WARN_ON(1);
1668                 return ERR_PTR(err);
1669         }
1670         return em;
1671 }
1672
1673 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1674 {
1675         return extent_bmap(mapping, iblock, btrfs_get_extent);
1676 }
1677
1678 static int btrfs_prepare_write(struct file *file, struct page *page,
1679                                unsigned from, unsigned to)
1680 {
1681         return extent_prepare_write(&BTRFS_I(page->mapping->host)->extent_tree,
1682                                     page->mapping->host, page, from, to,
1683                                     btrfs_get_extent);
1684 }
1685
1686 int btrfs_readpage(struct file *file, struct page *page)
1687 {
1688         struct extent_map_tree *tree;
1689         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1690         return extent_read_full_page(tree, page, btrfs_get_extent);
1691 }
1692 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1693 {
1694         struct extent_map_tree *tree;
1695
1696
1697         if (current->flags & PF_MEMALLOC) {
1698                 redirty_page_for_writepage(wbc, page);
1699                 unlock_page(page);
1700                 return 0;
1701         }
1702         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1703         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
1704 }
1705
1706 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1707 {
1708         struct extent_map_tree *tree;
1709         int ret;
1710
1711         if (page->private != 1) {
1712                 WARN_ON(1);
1713                 return try_to_free_buffers(page);
1714         }
1715         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1716         ret = try_release_extent_mapping(tree, page);
1717         if (ret == 1) {
1718                 ClearPagePrivate(page);
1719                 set_page_private(page, 0);
1720                 page_cache_release(page);
1721         }
1722         return ret;
1723 }
1724
1725 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
1726 {
1727         struct extent_map_tree *tree;
1728
1729         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1730         extent_invalidatepage(tree, page, offset);
1731         btrfs_releasepage(page, GFP_NOFS);
1732 }
1733
1734 /*
1735  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
1736  * called from a page fault handler when a page is first dirtied. Hence we must
1737  * be careful to check for EOF conditions here. We set the page up correctly
1738  * for a written page which means we get ENOSPC checking when writing into
1739  * holes and correct delalloc and unwritten extent mapping on filesystems that
1740  * support these features.
1741  *
1742  * We are not allowed to take the i_mutex here so we have to play games to
1743  * protect against truncate races as the page could now be beyond EOF.  Because
1744  * vmtruncate() writes the inode size before removing pages, once we have the
1745  * page lock we can determine safely if the page is beyond EOF. If it is not
1746  * beyond EOF, then the page is guaranteed safe against truncation until we
1747  * unlock the page.
1748  */
1749 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1750 {
1751         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1752         unsigned long end;
1753         loff_t size;
1754         int ret = -EINVAL;
1755         u64 page_start;
1756
1757         down_read(&BTRFS_I(inode)->root->snap_sem);
1758         lock_page(page);
1759         wait_on_page_writeback(page);
1760         size = i_size_read(inode);
1761         page_start = page->index << PAGE_CACHE_SHIFT;
1762
1763         if ((page->mapping != inode->i_mapping) ||
1764             (page_start > size)) {
1765                 /* page got truncated out from underneath us */
1766                 goto out_unlock;
1767         }
1768
1769         /* page is wholly or partially inside EOF */
1770         if (page_start + PAGE_CACHE_SIZE > size)
1771                 end = size & ~PAGE_CACHE_MASK;
1772         else
1773                 end = PAGE_CACHE_SIZE;
1774
1775         ret = btrfs_cow_one_page(inode, page, end);
1776
1777 out_unlock:
1778         up_read(&BTRFS_I(inode)->root->snap_sem);
1779         unlock_page(page);
1780         return ret;
1781 }
1782
1783 static void btrfs_truncate(struct inode *inode)
1784 {
1785         struct btrfs_root *root = BTRFS_I(inode)->root;
1786         int ret;
1787         struct btrfs_trans_handle *trans;
1788         unsigned long nr;
1789
1790         if (!S_ISREG(inode->i_mode))
1791                 return;
1792         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1793                 return;
1794
1795         btrfs_truncate_page(inode->i_mapping, inode->i_size);
1796
1797         mutex_lock(&root->fs_info->fs_mutex);
1798         trans = btrfs_start_transaction(root, 1);
1799         btrfs_set_trans_block_group(trans, inode);
1800
1801         /* FIXME, add redo link to tree so we don't leak on crash */
1802         ret = btrfs_truncate_in_trans(trans, root, inode);
1803         btrfs_update_inode(trans, root, inode);
1804         nr = trans->blocks_used;
1805
1806         ret = btrfs_end_transaction(trans, root);
1807         BUG_ON(ret);
1808         mutex_unlock(&root->fs_info->fs_mutex);
1809         btrfs_btree_balance_dirty(root, nr);
1810 }
1811
1812 int btrfs_commit_write(struct file *file, struct page *page,
1813                        unsigned from, unsigned to)
1814 {
1815         return extent_commit_write(&BTRFS_I(page->mapping->host)->extent_tree,
1816                                    page->mapping->host, page, from, to);
1817 }
1818
1819 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
1820 {
1821         struct btrfs_trans_handle *trans;
1822         struct btrfs_key key;
1823         struct btrfs_root_item root_item;
1824         struct btrfs_inode_item *inode_item;
1825         struct extent_buffer *leaf;
1826         struct btrfs_root *new_root;
1827         struct inode *inode;
1828         struct inode *dir;
1829         int ret;
1830         int err;
1831         u64 objectid;
1832         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
1833         unsigned long nr = 1;
1834
1835         mutex_lock(&root->fs_info->fs_mutex);
1836         trans = btrfs_start_transaction(root, 1);
1837         BUG_ON(!trans);
1838
1839         leaf = btrfs_alloc_free_block(trans, root, 0, 0);
1840         if (IS_ERR(leaf))
1841                 return PTR_ERR(leaf);
1842
1843         btrfs_set_header_nritems(leaf, 0);
1844         btrfs_set_header_level(leaf, 0);
1845         btrfs_set_header_blocknr(leaf, extent_buffer_blocknr(leaf));
1846         btrfs_set_header_generation(leaf, trans->transid);
1847         btrfs_set_header_owner(leaf, root->root_key.objectid);
1848         write_extent_buffer(leaf, root->fs_info->fsid,
1849                             (unsigned long)btrfs_header_fsid(leaf),
1850                             BTRFS_FSID_SIZE);
1851         btrfs_mark_buffer_dirty(leaf);
1852
1853         inode_item = &root_item.inode;
1854         memset(inode_item, 0, sizeof(*inode_item));
1855         inode_item->generation = cpu_to_le64(1);
1856         inode_item->size = cpu_to_le64(3);
1857         inode_item->nlink = cpu_to_le32(1);
1858         inode_item->nblocks = cpu_to_le64(1);
1859         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1860
1861         btrfs_set_root_blocknr(&root_item, extent_buffer_blocknr(leaf));
1862         btrfs_set_root_refs(&root_item, 1);
1863         btrfs_set_root_used(&root_item, 0);
1864
1865         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
1866         root_item.drop_level = 0;
1867
1868         free_extent_buffer(leaf);
1869         leaf = NULL;
1870
1871         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1872                                        0, &objectid);
1873         if (ret)
1874                 goto fail;
1875
1876         btrfs_set_root_dirid(&root_item, new_dirid);
1877
1878         key.objectid = objectid;
1879         key.offset = 1;
1880         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1881         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1882                                 &root_item);
1883         if (ret)
1884                 goto fail;
1885
1886         /*
1887          * insert the directory item
1888          */
1889         key.offset = (u64)-1;
1890         dir = root->fs_info->sb->s_root->d_inode;
1891         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1892                                     name, namelen, dir->i_ino, &key,
1893                                     BTRFS_FT_DIR);
1894         if (ret)
1895                 goto fail;
1896
1897         ret = btrfs_commit_transaction(trans, root);
1898         if (ret)
1899                 goto fail_commit;
1900
1901         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
1902         BUG_ON(!new_root);
1903
1904         trans = btrfs_start_transaction(new_root, 1);
1905         BUG_ON(!trans);
1906
1907         inode = btrfs_new_inode(trans, new_root, new_dirid,
1908                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
1909         if (IS_ERR(inode))
1910                 goto fail;
1911         inode->i_op = &btrfs_dir_inode_operations;
1912         inode->i_fop = &btrfs_dir_file_operations;
1913         new_root->inode = inode;
1914
1915         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
1916         if (ret)
1917                 goto fail;
1918
1919         inode->i_nlink = 1;
1920         inode->i_size = 6;
1921         ret = btrfs_update_inode(trans, new_root, inode);
1922         if (ret)
1923                 goto fail;
1924 fail:
1925         nr = trans->blocks_used;
1926         err = btrfs_commit_transaction(trans, root);
1927         if (err && !ret)
1928                 ret = err;
1929 fail_commit:
1930         mutex_unlock(&root->fs_info->fs_mutex);
1931         btrfs_btree_balance_dirty(root, nr);
1932         return ret;
1933 }
1934
1935 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
1936 {
1937         struct btrfs_trans_handle *trans;
1938         struct btrfs_key key;
1939         struct btrfs_root_item new_root_item;
1940         struct extent_buffer *tmp;
1941         int ret;
1942         int err;
1943         u64 objectid;
1944         unsigned long nr;
1945
1946         if (!root->ref_cows)
1947                 return -EINVAL;
1948
1949         down_write(&root->snap_sem);
1950         freeze_bdev(root->fs_info->sb->s_bdev);
1951         thaw_bdev(root->fs_info->sb->s_bdev, root->fs_info->sb);
1952
1953         mutex_lock(&root->fs_info->fs_mutex);
1954         trans = btrfs_start_transaction(root, 1);
1955         BUG_ON(!trans);
1956
1957         ret = btrfs_update_inode(trans, root, root->inode);
1958         if (ret)
1959                 goto fail;
1960
1961         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1962                                        0, &objectid);
1963         if (ret)
1964                 goto fail;
1965
1966         memcpy(&new_root_item, &root->root_item,
1967                sizeof(new_root_item));
1968
1969         key.objectid = objectid;
1970         key.offset = 1;
1971         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1972
1973         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
1974         btrfs_set_root_blocknr(&new_root_item,
1975                                extent_buffer_blocknr(root->node));
1976
1977         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1978                                 &new_root_item);
1979         if (ret)
1980                 goto fail;
1981
1982         /*
1983          * insert the directory item
1984          */
1985         key.offset = (u64)-1;
1986         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1987                                     name, namelen,
1988                                     root->fs_info->sb->s_root->d_inode->i_ino,
1989                                     &key, BTRFS_FT_DIR);
1990
1991         if (ret)
1992                 goto fail;
1993
1994         ret = btrfs_inc_root_ref(trans, root);
1995         if (ret)
1996                 goto fail;
1997 fail:
1998         nr = trans->blocks_used;
1999         err = btrfs_commit_transaction(trans, root);
2000
2001         if (err && !ret)
2002                 ret = err;
2003
2004         mutex_unlock(&root->fs_info->fs_mutex);
2005         up_write(&root->snap_sem);
2006         btrfs_btree_balance_dirty(root, nr);
2007         return ret;
2008 }
2009
2010 static unsigned long force_ra(struct address_space *mapping,
2011                               struct file_ra_state *ra, struct file *file,
2012                               pgoff_t offset, pgoff_t last_index)
2013 {
2014         pgoff_t req_size;
2015
2016 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2017         req_size = last_index - offset + 1;
2018         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2019         return offset;
2020 #else
2021         req_size = min(last_index - offset + 1, (pgoff_t)128);
2022         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2023         return offset + req_size;
2024 #endif
2025 }
2026
2027 int btrfs_defrag_file(struct file *file) {
2028         struct inode *inode = file->f_path.dentry->d_inode;
2029         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2030         struct page *page;
2031         unsigned long last_index;
2032         unsigned long ra_index = 0;
2033         u64 page_start;
2034         u64 page_end;
2035         unsigned long i;
2036
2037         mutex_lock(&inode->i_mutex);
2038         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2039         for (i = 0; i <= last_index; i++) {
2040                 if (i == ra_index) {
2041                         ra_index = force_ra(inode->i_mapping, &file->f_ra,
2042                                             file, ra_index, last_index);
2043                 }
2044                 page = grab_cache_page(inode->i_mapping, i);
2045                 if (!page)
2046                         goto out_unlock;
2047                 if (!PageUptodate(page)) {
2048                         btrfs_readpage(NULL, page);
2049                         lock_page(page);
2050                         if (!PageUptodate(page)) {
2051                                 unlock_page(page);
2052                                 page_cache_release(page);
2053                                 goto out_unlock;
2054                         }
2055                 }
2056                 page_start = page->index << PAGE_CACHE_SHIFT;
2057                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2058
2059                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2060                 set_extent_delalloc(em_tree, page_start,
2061                                     page_end, GFP_NOFS);
2062                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2063                 set_page_dirty(page);
2064                 unlock_page(page);
2065                 page_cache_release(page);
2066                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2067         }
2068
2069 out_unlock:
2070         mutex_unlock(&inode->i_mutex);
2071         return 0;
2072 }
2073
2074 static int btrfs_ioctl_snap_create(struct btrfs_root *root, void __user *arg)
2075 {
2076         struct btrfs_ioctl_vol_args vol_args;
2077         struct btrfs_dir_item *di;
2078         struct btrfs_path *path;
2079         int namelen;
2080         u64 root_dirid;
2081
2082         if (copy_from_user(&vol_args, arg, sizeof(vol_args)))
2083                 return -EFAULT;
2084
2085         namelen = strlen(vol_args.name);
2086         if (namelen > BTRFS_VOL_NAME_MAX)
2087                 return -EINVAL;
2088         if (strchr(vol_args.name, '/'))
2089                 return -EINVAL;
2090
2091         path = btrfs_alloc_path();
2092         if (!path)
2093                 return -ENOMEM;
2094
2095         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2096         mutex_lock(&root->fs_info->fs_mutex);
2097         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2098                             path, root_dirid,
2099                             vol_args.name, namelen, 0);
2100         mutex_unlock(&root->fs_info->fs_mutex);
2101         btrfs_free_path(path);
2102         if (di && !IS_ERR(di))
2103                 return -EEXIST;
2104         if (IS_ERR(di))
2105                 return PTR_ERR(di);
2106
2107         if (root == root->fs_info->tree_root)
2108                 return create_subvol(root, vol_args.name, namelen);
2109         return create_snapshot(root, vol_args.name, namelen);
2110 }
2111
2112 static int btrfs_ioctl_defrag(struct file *file)
2113 {
2114         struct inode *inode = file->f_path.dentry->d_inode;
2115         struct btrfs_root *root = BTRFS_I(inode)->root;
2116
2117         switch (inode->i_mode & S_IFMT) {
2118         case S_IFDIR:
2119                 mutex_lock(&root->fs_info->fs_mutex);
2120                 btrfs_defrag_root(root, 0);
2121                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2122                 mutex_unlock(&root->fs_info->fs_mutex);
2123                 break;
2124         case S_IFREG:
2125                 btrfs_defrag_file(file);
2126                 break;
2127         }
2128
2129         return 0;
2130 }
2131
2132 long btrfs_ioctl(struct file *file, unsigned int
2133                 cmd, unsigned long arg)
2134 {
2135         struct btrfs_root *root = BTRFS_I(file->f_path.dentry->d_inode)->root;
2136
2137         switch (cmd) {
2138         case BTRFS_IOC_SNAP_CREATE:
2139                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2140         case BTRFS_IOC_DEFRAG:
2141                 return btrfs_ioctl_defrag(file);
2142         }
2143
2144         return -ENOTTY;
2145 }
2146
2147 /*
2148  * Called inside transaction, so use GFP_NOFS
2149  */
2150 struct inode *btrfs_alloc_inode(struct super_block *sb)
2151 {
2152         struct btrfs_inode *ei;
2153
2154         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2155         if (!ei)
2156                 return NULL;
2157         ei->last_trans = 0;
2158         return &ei->vfs_inode;
2159 }
2160
2161 void btrfs_destroy_inode(struct inode *inode)
2162 {
2163         WARN_ON(!list_empty(&inode->i_dentry));
2164         WARN_ON(inode->i_data.nrpages);
2165
2166         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2167 }
2168
2169 static void init_once(void * foo, struct kmem_cache * cachep,
2170                       unsigned long flags)
2171 {
2172         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2173
2174         inode_init_once(&ei->vfs_inode);
2175 }
2176
2177 void btrfs_destroy_cachep(void)
2178 {
2179         if (btrfs_inode_cachep)
2180                 kmem_cache_destroy(btrfs_inode_cachep);
2181         if (btrfs_trans_handle_cachep)
2182                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2183         if (btrfs_transaction_cachep)
2184                 kmem_cache_destroy(btrfs_transaction_cachep);
2185         if (btrfs_bit_radix_cachep)
2186                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2187         if (btrfs_path_cachep)
2188                 kmem_cache_destroy(btrfs_path_cachep);
2189 }
2190
2191 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2192                                        unsigned long extra_flags,
2193                                        void (*ctor)(void *, struct kmem_cache *,
2194                                                     unsigned long))
2195 {
2196         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2197                                  SLAB_MEM_SPREAD | extra_flags), ctor
2198 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2199                                  ,NULL
2200 #endif
2201                                 );
2202 }
2203
2204 int btrfs_init_cachep(void)
2205 {
2206         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2207                                           sizeof(struct btrfs_inode),
2208                                           0, init_once);
2209         if (!btrfs_inode_cachep)
2210                 goto fail;
2211         btrfs_trans_handle_cachep =
2212                         btrfs_cache_create("btrfs_trans_handle_cache",
2213                                            sizeof(struct btrfs_trans_handle),
2214                                            0, NULL);
2215         if (!btrfs_trans_handle_cachep)
2216                 goto fail;
2217         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2218                                              sizeof(struct btrfs_transaction),
2219                                              0, NULL);
2220         if (!btrfs_transaction_cachep)
2221                 goto fail;
2222         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2223                                          sizeof(struct btrfs_path),
2224                                          0, NULL);
2225         if (!btrfs_path_cachep)
2226                 goto fail;
2227         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2228                                               SLAB_DESTROY_BY_RCU, NULL);
2229         if (!btrfs_bit_radix_cachep)
2230                 goto fail;
2231         return 0;
2232 fail:
2233         btrfs_destroy_cachep();
2234         return -ENOMEM;
2235 }
2236
2237 static int btrfs_getattr(struct vfsmount *mnt,
2238                          struct dentry *dentry, struct kstat *stat)
2239 {
2240         struct inode *inode = dentry->d_inode;
2241         generic_fillattr(inode, stat);
2242         stat->blksize = 256 * 1024;
2243         return 0;
2244 }
2245
2246 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2247                            struct inode * new_dir,struct dentry *new_dentry)
2248 {
2249         struct btrfs_trans_handle *trans;
2250         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2251         struct inode *new_inode = new_dentry->d_inode;
2252         struct inode *old_inode = old_dentry->d_inode;
2253         struct timespec ctime = CURRENT_TIME;
2254         struct btrfs_path *path;
2255         struct btrfs_dir_item *di;
2256         int ret;
2257
2258         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2259             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2260                 return -ENOTEMPTY;
2261         }
2262
2263         mutex_lock(&root->fs_info->fs_mutex);
2264         trans = btrfs_start_transaction(root, 1);
2265
2266         btrfs_set_trans_block_group(trans, new_dir);
2267         path = btrfs_alloc_path();
2268         if (!path) {
2269                 ret = -ENOMEM;
2270                 goto out_fail;
2271         }
2272
2273         old_dentry->d_inode->i_nlink++;
2274         old_dir->i_ctime = old_dir->i_mtime = ctime;
2275         new_dir->i_ctime = new_dir->i_mtime = ctime;
2276         old_inode->i_ctime = ctime;
2277
2278         if (S_ISDIR(old_inode->i_mode) && old_dir != new_dir) {
2279                 struct btrfs_key *location = &BTRFS_I(new_dir)->location;
2280                 struct btrfs_key old_parent_key;
2281                 di = btrfs_lookup_dir_item(trans, root, path, old_inode->i_ino,
2282                                            "..", 2, -1);
2283                 if (IS_ERR(di)) {
2284                         ret = PTR_ERR(di);
2285                         goto out_fail;
2286                 }
2287                 if (!di) {
2288                         ret = -ENOENT;
2289                         goto out_fail;
2290                 }
2291                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &old_parent_key);
2292                 ret = btrfs_del_item(trans, root, path);
2293                 if (ret) {
2294                         goto out_fail;
2295                 }
2296                 btrfs_release_path(root, path);
2297
2298                 di = btrfs_lookup_dir_index_item(trans, root, path,
2299                                                  old_inode->i_ino,
2300                                                  old_parent_key.objectid,
2301                                                  "..", 2, -1);
2302                 if (IS_ERR(di)) {
2303                         ret = PTR_ERR(di);
2304                         goto out_fail;
2305                 }
2306                 if (!di) {
2307                         ret = -ENOENT;
2308                         goto out_fail;
2309                 }
2310                 ret = btrfs_del_item(trans, root, path);
2311                 if (ret) {
2312                         goto out_fail;
2313                 }
2314                 btrfs_release_path(root, path);
2315
2316                 ret = btrfs_insert_dir_item(trans, root, "..", 2,
2317                                             old_inode->i_ino, location,
2318                                             BTRFS_FT_DIR);
2319                 if (ret)
2320                         goto out_fail;
2321         }
2322
2323
2324         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2325         if (ret)
2326                 goto out_fail;
2327
2328         if (new_inode) {
2329                 new_inode->i_ctime = CURRENT_TIME;
2330                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2331                 if (ret)
2332                         goto out_fail;
2333         }
2334         ret = btrfs_add_link(trans, new_dentry, old_inode);
2335         if (ret)
2336                 goto out_fail;
2337
2338 out_fail:
2339         btrfs_free_path(path);
2340         btrfs_end_transaction(trans, root);
2341         mutex_unlock(&root->fs_info->fs_mutex);
2342         return ret;
2343 }
2344
2345 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2346                          const char *symname)
2347 {
2348         struct btrfs_trans_handle *trans;
2349         struct btrfs_root *root = BTRFS_I(dir)->root;
2350         struct btrfs_path *path;
2351         struct btrfs_key key;
2352         struct inode *inode;
2353         int err;
2354         int drop_inode = 0;
2355         u64 objectid;
2356         int name_len;
2357         int datasize;
2358         unsigned long ptr;
2359         struct btrfs_file_extent_item *ei;
2360         struct extent_buffer *leaf;
2361         unsigned long nr;
2362
2363         name_len = strlen(symname) + 1;
2364         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2365                 return -ENAMETOOLONG;
2366         mutex_lock(&root->fs_info->fs_mutex);
2367         trans = btrfs_start_transaction(root, 1);
2368         btrfs_set_trans_block_group(trans, dir);
2369
2370         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2371         if (err) {
2372                 err = -ENOSPC;
2373                 goto out_unlock;
2374         }
2375
2376         inode = btrfs_new_inode(trans, root, objectid,
2377                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2378         err = PTR_ERR(inode);
2379         if (IS_ERR(inode))
2380                 goto out_unlock;
2381
2382         btrfs_set_trans_block_group(trans, inode);
2383         err = btrfs_add_nondir(trans, dentry, inode);
2384         if (err)
2385                 drop_inode = 1;
2386         else {
2387                 inode->i_mapping->a_ops = &btrfs_aops;
2388                 inode->i_fop = &btrfs_file_operations;
2389                 inode->i_op = &btrfs_file_inode_operations;
2390                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2391                                      inode->i_mapping, GFP_NOFS);
2392                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2393         }
2394         dir->i_sb->s_dirt = 1;
2395         btrfs_update_inode_block_group(trans, inode);
2396         btrfs_update_inode_block_group(trans, dir);
2397         if (drop_inode)
2398                 goto out_unlock;
2399
2400         path = btrfs_alloc_path();
2401         BUG_ON(!path);
2402         key.objectid = inode->i_ino;
2403         key.offset = 0;
2404         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2405         datasize = btrfs_file_extent_calc_inline_size(name_len);
2406         err = btrfs_insert_empty_item(trans, root, path, &key,
2407                                       datasize);
2408         if (err) {
2409                 drop_inode = 1;
2410                 goto out_unlock;
2411         }
2412         leaf = path->nodes[0];
2413         ei = btrfs_item_ptr(leaf, path->slots[0],
2414                             struct btrfs_file_extent_item);
2415         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2416         btrfs_set_file_extent_type(leaf, ei,
2417                                    BTRFS_FILE_EXTENT_INLINE);
2418         ptr = btrfs_file_extent_inline_start(ei);
2419         write_extent_buffer(leaf, symname, ptr, name_len);
2420         btrfs_mark_buffer_dirty(leaf);
2421         btrfs_free_path(path);
2422
2423         inode->i_op = &btrfs_symlink_inode_operations;
2424         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2425         inode->i_size = name_len - 1;
2426         err = btrfs_update_inode(trans, root, inode);
2427         if (err)
2428                 drop_inode = 1;
2429
2430 out_unlock:
2431         nr = trans->blocks_used;
2432         btrfs_end_transaction(trans, root);
2433         mutex_unlock(&root->fs_info->fs_mutex);
2434         if (drop_inode) {
2435                 inode_dec_link_count(inode);
2436                 iput(inode);
2437         }
2438         btrfs_btree_balance_dirty(root, nr);
2439         return err;
2440 }
2441
2442 static struct inode_operations btrfs_dir_inode_operations = {
2443         .lookup         = btrfs_lookup,
2444         .create         = btrfs_create,
2445         .unlink         = btrfs_unlink,
2446         .link           = btrfs_link,
2447         .mkdir          = btrfs_mkdir,
2448         .rmdir          = btrfs_rmdir,
2449         .rename         = btrfs_rename,
2450         .symlink        = btrfs_symlink,
2451         .setattr        = btrfs_setattr,
2452         .mknod          = btrfs_mknod,
2453 };
2454
2455 static struct inode_operations btrfs_dir_ro_inode_operations = {
2456         .lookup         = btrfs_lookup,
2457 };
2458
2459 static struct file_operations btrfs_dir_file_operations = {
2460         .llseek         = generic_file_llseek,
2461         .read           = generic_read_dir,
2462         .readdir        = btrfs_readdir,
2463         .unlocked_ioctl = btrfs_ioctl,
2464 #ifdef CONFIG_COMPAT
2465         .compat_ioctl   = btrfs_ioctl,
2466 #endif
2467 };
2468
2469 static struct extent_map_ops btrfs_extent_map_ops = {
2470         .fill_delalloc = run_delalloc_range,
2471         .writepage_io_hook = btrfs_writepage_io_hook,
2472         .readpage_io_hook = btrfs_readpage_io_hook,
2473         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2474 };
2475
2476 static struct address_space_operations btrfs_aops = {
2477         .readpage       = btrfs_readpage,
2478         .writepage      = btrfs_writepage,
2479         .sync_page      = block_sync_page,
2480         .prepare_write  = btrfs_prepare_write,
2481         .commit_write   = btrfs_commit_write,
2482         .bmap           = btrfs_bmap,
2483         .invalidatepage = btrfs_invalidatepage,
2484         .releasepage    = btrfs_releasepage,
2485         .set_page_dirty = __set_page_dirty_nobuffers,
2486 };
2487
2488 static struct address_space_operations btrfs_symlink_aops = {
2489         .readpage       = btrfs_readpage,
2490         .writepage      = btrfs_writepage,
2491         .invalidatepage = btrfs_invalidatepage,
2492         .releasepage    = btrfs_releasepage,
2493 };
2494
2495 static struct inode_operations btrfs_file_inode_operations = {
2496         .truncate       = btrfs_truncate,
2497         .getattr        = btrfs_getattr,
2498         .setattr        = btrfs_setattr,
2499 };
2500
2501 static struct inode_operations btrfs_special_inode_operations = {
2502         .getattr        = btrfs_getattr,
2503         .setattr        = btrfs_setattr,
2504 };
2505
2506 static struct inode_operations btrfs_symlink_inode_operations = {
2507         .readlink       = generic_readlink,
2508         .follow_link    = page_follow_link_light,
2509         .put_link       = page_put_link,
2510 };