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_bytes;
  81         int ret;
  82         u64 blocksize = root->sectorsize;
  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_bytes = (end - start + blocksize) & ~(blocksize - 1);
  89         ret = btrfs_drop_extents(trans, root, inode,
  90                                  start, start + num_bytes, start, &alloc_hint);
  91
  92         ret = btrfs_alloc_extent(trans, root, inode->i_ino, num_bytes, 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);
 190         flush_dcache_page(page);
 191         kunmap_atomic(kaddr, KM_IRQ0);
 192         return 0;
 193 }
 194
 195 void btrfs_read_locked_inode(struct inode *inode)
 196 {
 197         struct btrfs_path *path;
 198         struct extent_buffer *leaf;
 199         struct btrfs_inode_item *inode_item;
 200         struct btrfs_inode_timespec *tspec;
 201         struct btrfs_root *root = BTRFS_I(inode)->root;
 202         struct btrfs_key location;
 203         u64 alloc_group_block;
 204         u32 rdev;
 205         int ret;
 206
 207         path = btrfs_alloc_path();
 208         BUG_ON(!path);
 209         mutex_lock(&root->fs_info->fs_mutex);
 210
 211         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 212         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 213         if (ret)
 214                 goto make_bad;
 215
 216         leaf = path->nodes[0];
 217         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 218                                     struct btrfs_inode_item);
 219
 220         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 221         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 222         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 223         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 224         inode->i_size = btrfs_inode_size(leaf, inode_item);
 225
 226         tspec = btrfs_inode_atime(inode_item);
 227         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 228         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 229
 230         tspec = btrfs_inode_mtime(inode_item);
 231         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 232         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 233
 234         tspec = btrfs_inode_ctime(inode_item);
 235         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 236         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 237
 238         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 239         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 240         inode->i_rdev = 0;
 241         rdev = btrfs_inode_rdev(leaf, inode_item);
 242
 243         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 244         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 245                                                        alloc_group_block);
 246
 247         btrfs_free_path(path);
 248         inode_item = NULL;
 249
 250         mutex_unlock(&root->fs_info->fs_mutex);
 251
 252         switch (inode->i_mode & S_IFMT) {
 253         case S_IFREG:
 254                 inode->i_mapping->a_ops = &btrfs_aops;
 255                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
 256                 inode->i_fop = &btrfs_file_operations;
 257                 inode->i_op = &btrfs_file_inode_operations;
 258                 break;
 259         case S_IFDIR:
 260                 inode->i_fop = &btrfs_dir_file_operations;
 261                 if (root == root->fs_info->tree_root)
 262                         inode->i_op = &btrfs_dir_ro_inode_operations;
 263                 else
 264                         inode->i_op = &btrfs_dir_inode_operations;
 265                 break;
 266         case S_IFLNK:
 267                 inode->i_op = &btrfs_symlink_inode_operations;
 268                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 269                 break;
 270         default:
 271                 init_special_inode(inode, inode->i_mode, rdev);
 272                 break;
 273         }
 274         return;
 275
 276 make_bad:
 277         btrfs_release_path(root, path);
 278         btrfs_free_path(path);
 279         mutex_unlock(&root->fs_info->fs_mutex);
 280         make_bad_inode(inode);
 281 }
 282
 283 static void fill_inode_item(struct extent_buffer *leaf,
 284                             struct btrfs_inode_item *item,
 285                             struct inode *inode)
 286 {
 287         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 288         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 289         btrfs_set_inode_size(leaf, item, inode->i_size);
 290         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 291         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 292
 293         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 294                                inode->i_atime.tv_sec);
 295         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 296                                 inode->i_atime.tv_nsec);
 297
 298         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 299                                inode->i_mtime.tv_sec);
 300         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 301                                 inode->i_mtime.tv_nsec);
 302
 303         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 304                                inode->i_ctime.tv_sec);
 305         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 306                                 inode->i_ctime.tv_nsec);
 307
 308         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 309         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 310         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 311         btrfs_set_inode_block_group(leaf, item,
 312                                     BTRFS_I(inode)->block_group->key.objectid);
 313 }
 314
 315 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 316                               struct btrfs_root *root,
 317                               struct inode *inode)
 318 {
 319         struct btrfs_inode_item *inode_item;
 320         struct btrfs_path *path;
 321         struct extent_buffer *leaf;
 322         int ret;
 323
 324         path = btrfs_alloc_path();
 325         BUG_ON(!path);
 326         ret = btrfs_lookup_inode(trans, root, path,
 327                                  &BTRFS_I(inode)->location, 1);
 328         if (ret) {
 329                 if (ret > 0)
 330                         ret = -ENOENT;
 331                 goto failed;
 332         }
 333
 334         leaf = path->nodes[0];
 335         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 336                                   struct btrfs_inode_item);
 337
 338         fill_inode_item(leaf, inode_item, inode);
 339         btrfs_mark_buffer_dirty(leaf);
 340         btrfs_set_inode_last_trans(trans, inode);
 341         ret = 0;
 342 failed:
 343         btrfs_release_path(root, path);
 344         btrfs_free_path(path);
 345         return ret;
 346 }
 347
 348
 349 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 350                               struct btrfs_root *root,
 351                               struct inode *dir,
 352                               struct dentry *dentry)
 353 {
 354         struct btrfs_path *path;
 355         const char *name = dentry->d_name.name;
 356         int name_len = dentry->d_name.len;
 357         int ret = 0;
 358         struct extent_buffer *leaf;
 359         struct btrfs_dir_item *di;
 360         struct btrfs_key key;
 361
 362         path = btrfs_alloc_path();
 363         if (!path) {
 364                 ret = -ENOMEM;
 365                 goto err;
 366         }
 367
 368         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 369                                     name, name_len, -1);
 370         if (IS_ERR(di)) {
 371                 ret = PTR_ERR(di);
 372                 goto err;
 373         }
 374         if (!di) {
 375                 ret = -ENOENT;
 376                 goto err;
 377         }
 378         leaf = path->nodes[0];
 379         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 380         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 381         if (ret)
 382                 goto err;
 383         btrfs_release_path(root, path);
 384
 385         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 386                                          key.objectid, name, name_len, -1);
 387         if (IS_ERR(di)) {
 388                 ret = PTR_ERR(di);
 389                 goto err;
 390         }
 391         if (!di) {
 392                 ret = -ENOENT;
 393                 goto err;
 394         }
 395         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 396
 397         dentry->d_inode->i_ctime = dir->i_ctime;
 398 err:
 399         btrfs_free_path(path);
 400         if (!ret) {
 401                 dir->i_size -= name_len * 2;
 402                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 403                 btrfs_update_inode(trans, root, dir);
 404                 drop_nlink(dentry->d_inode);
 405                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 406                 dir->i_sb->s_dirt = 1;
 407         }
 408         return ret;
 409 }
 410
 411 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 412 {
 413         struct btrfs_root *root;
 414         struct btrfs_trans_handle *trans;
 415         int ret;
 416         unsigned long nr;
 417
 418         root = BTRFS_I(dir)->root;
 419         mutex_lock(&root->fs_info->fs_mutex);
 420         trans = btrfs_start_transaction(root, 1);
 421
 422         btrfs_set_trans_block_group(trans, dir);
 423         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 424         nr = trans->blocks_used;
 425
 426         btrfs_end_transaction(trans, root);
 427         mutex_unlock(&root->fs_info->fs_mutex);
 428         btrfs_btree_balance_dirty(root, nr);
 429
 430         return ret;
 431 }
 432
 433 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 434 {
 435         struct inode *inode = dentry->d_inode;
 436         int err;
 437         int ret;
 438         struct btrfs_root *root = BTRFS_I(dir)->root;
 439         struct btrfs_path *path;
 440         struct btrfs_key key;
 441         struct btrfs_trans_handle *trans;
 442         struct btrfs_key found_key;
 443         int found_type;
 444         struct extent_buffer *leaf;
 445         char *goodnames = "..";
 446         unsigned long nr;
 447
 448         path = btrfs_alloc_path();
 449         BUG_ON(!path);
 450         mutex_lock(&root->fs_info->fs_mutex);
 451         trans = btrfs_start_transaction(root, 1);
 452
 453         btrfs_set_trans_block_group(trans, dir);
 454         key.objectid = inode->i_ino;
 455         key.offset = (u64)-1;
 456         key.type = (u8)-1;
 457         while(1) {
 458                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 459                 if (ret < 0) {
 460                         err = ret;
 461                         goto out;
 462                 }
 463                 BUG_ON(ret == 0);
 464                 if (path->slots[0] == 0) {
 465                         err = -ENOENT;
 466                         goto out;
 467                 }
 468                 path->slots[0]--;
 469                 leaf = path->nodes[0];
 470                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 471                 found_type = btrfs_key_type(&found_key);
 472                 if (found_key.objectid != inode->i_ino) {
 473                         err = -ENOENT;
 474                         goto out;
 475                 }
 476                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
 477                      found_type != BTRFS_DIR_INDEX_KEY) ||
 478                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
 479                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
 480                         err = -ENOTEMPTY;
 481                         goto out;
 482                 }
 483                 ret = btrfs_del_item(trans, root, path);
 484                 BUG_ON(ret);
 485
 486                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
 487                         break;
 488                 btrfs_release_path(root, path);
 489         }
 490         ret = 0;
 491         btrfs_release_path(root, path);
 492
 493         /* now the directory is empty */
 494         err = btrfs_unlink_trans(trans, root, dir, dentry);
 495         if (!err) {
 496                 inode->i_size = 0;
 497         }
 498 out:
 499         btrfs_release_path(root, path);
 500         btrfs_free_path(path);
 501         mutex_unlock(&root->fs_info->fs_mutex);
 502         nr = trans->blocks_used;
 503         ret = btrfs_end_transaction(trans, root);
 504         btrfs_btree_balance_dirty(root, nr);
 505         if (ret && !err)
 506                 err = ret;
 507         return err;
 508 }
 509
 510 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 511                             struct btrfs_root *root,
 512                             struct inode *inode)
 513 {
 514         struct btrfs_path *path;
 515         int ret;
 516
 517         clear_inode(inode);
 518
 519         path = btrfs_alloc_path();
 520         BUG_ON(!path);
 521         ret = btrfs_lookup_inode(trans, root, path,
 522                                  &BTRFS_I(inode)->location, -1);
 523         if (ret > 0)
 524                 ret = -ENOENT;
 525         if (!ret)
 526                 ret = btrfs_del_item(trans, root, path);
 527         btrfs_free_path(path);
 528         return ret;
 529 }
 530
 531 /*
 532  * this can truncate away extent items, csum items and directory items.
 533  * It starts at a high offset and removes keys until it can't find
 534  * any higher than i_size.
 535  *
 536  * csum items that cross the new i_size are truncated to the new size
 537  * as well.
 538  */
 539 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 540                                    struct btrfs_root *root,
 541                                    struct inode *inode)
 542 {
 543         int ret;
 544         struct btrfs_path *path;
 545         struct btrfs_key key;
 546         struct btrfs_key found_key;
 547         u32 found_type;
 548         struct extent_buffer *leaf;
 549         struct btrfs_file_extent_item *fi;
 550         u64 extent_start = 0;
 551         u64 extent_num_bytes = 0;
 552         u64 item_end = 0;
 553         int found_extent;
 554         int del_item;
 555
 556         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 557         path = btrfs_alloc_path();
 558         path->reada = -1;
 559         BUG_ON(!path);
 560
 561         /* FIXME, add redo link to tree so we don't leak on crash */
 562         key.objectid = inode->i_ino;
 563         key.offset = (u64)-1;
 564         key.type = (u8)-1;
 565
 566         while(1) {
 567                 btrfs_init_path(path);
 568                 fi = NULL;
 569                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 570                 if (ret < 0) {
 571                         goto error;
 572                 }
 573                 if (ret > 0) {
 574                         BUG_ON(path->slots[0] == 0);
 575                         path->slots[0]--;
 576                 }
 577                 leaf = path->nodes[0];
 578                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 579                 found_type = btrfs_key_type(&found_key);
 580
 581                 if (found_key.objectid != inode->i_ino)
 582                         break;
 583
 584                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 585                     found_type != BTRFS_DIR_ITEM_KEY &&
 586                     found_type != BTRFS_DIR_INDEX_KEY &&
 587                     found_type != BTRFS_EXTENT_DATA_KEY)
 588                         break;
 589
 590                 item_end = found_key.offset;
 591                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 592                         fi = btrfs_item_ptr(leaf, path->slots[0],
 593                                             struct btrfs_file_extent_item);
 594                         if (btrfs_file_extent_type(leaf, fi) !=
 595                             BTRFS_FILE_EXTENT_INLINE) {
 596                                 item_end +=
 597                                     btrfs_file_extent_num_bytes(leaf, fi);
 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_bytenr(leaf, fi);
 630                         if (!del_item) {
 631                                 u64 orig_num_bytes =
 632                                         btrfs_file_extent_num_bytes(leaf, fi);
 633                                 extent_num_bytes = inode->i_size -
 634                                         found_key.offset + root->sectorsize - 1;
 635                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 636                                                          extent_num_bytes);
 637                                 num_dec = (orig_num_bytes -
 638                                            extent_num_bytes) >> 9;
 639                                 if (extent_start != 0) {
 640                                         inode->i_blocks -= num_dec;
 641                                 }
 642                                 btrfs_mark_buffer_dirty(leaf);
 643                         } else {
 644                                 extent_num_bytes =
 645                                         btrfs_file_extent_disk_num_bytes(leaf,
 646                                                                          fi);
 647                                 /* FIXME blocksize != 4096 */
 648                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 649                                                                        fi) >> 9;
 650                                 if (extent_start != 0) {
 651                                         found_extent = 1;
 652                                         inode->i_blocks -= num_dec;
 653                                 }
 654                         }
 655                 }
 656                 if (del_item) {
 657                         ret = btrfs_del_item(trans, root, path);
 658                         if (ret)
 659                                 goto error;
 660                 } else {
 661                         break;
 662                 }
 663                 btrfs_release_path(root, path);
 664                 if (found_extent) {
 665                         ret = btrfs_free_extent(trans, root, extent_start,
 666                                                 extent_num_bytes, 0);
 667                         BUG_ON(ret);
 668                 }
 669         }
 670         ret = 0;
 671 error:
 672         btrfs_release_path(root, path);
 673         btrfs_free_path(path);
 674         inode->i_sb->s_dirt = 1;
 675         return ret;
 676 }
 677
 678 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 679                               size_t zero_start)
 680 {
 681         char *kaddr;
 682         int ret = 0;
 683         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 684         u64 page_start = page->index << PAGE_CACHE_SHIFT;
 685         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 686
 687         set_page_extent_mapped(page);
 688
 689         lock_extent(em_tree, page_start, page_end, GFP_NOFS);
 690         set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
 691                             page_end, GFP_NOFS);
 692         if (zero_start != PAGE_CACHE_SIZE) {
 693                 kaddr = kmap(page);
 694                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 695                 flush_dcache_page(page);
 696                 kunmap(page);
 697         }
 698         set_page_dirty(page);
 699         unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
 700
 701         return ret;
 702 }
 703
 704 /*
 705  * taken from block_truncate_page, but does cow as it zeros out
 706  * any bytes left in the last page in the file.
 707  */
 708 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 709 {
 710         struct inode *inode = mapping->host;
 711         struct btrfs_root *root = BTRFS_I(inode)->root;
 712         u32 blocksize = root->sectorsize;
 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(&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, pos,
 780                                          &alloc_hint);
 781
 782                 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
 783                                                pos, 0, 0, hole_size);
 784                 btrfs_end_transaction(trans, root);
 785                 mutex_unlock(&root->fs_info->fs_mutex);
 786                 unlock_extent(em_tree, pos, block_end, GFP_NOFS);
 787                 if (err)
 788                         return err;
 789         }
 790 out:
 791         err = inode_setattr(inode, attr);
 792
 793         return err;
 794 }
 795 void btrfs_delete_inode(struct inode *inode)
 796 {
 797         struct btrfs_trans_handle *trans;
 798         struct btrfs_root *root = BTRFS_I(inode)->root;
 799         unsigned long nr;
 800         int ret;
 801
 802         truncate_inode_pages(&inode->i_data, 0);
 803         if (is_bad_inode(inode)) {
 804                 goto no_delete;
 805         }
 806
 807         inode->i_size = 0;
 808         mutex_lock(&root->fs_info->fs_mutex);
 809         trans = btrfs_start_transaction(root, 1);
 810
 811         btrfs_set_trans_block_group(trans, inode);
 812         ret = btrfs_truncate_in_trans(trans, root, inode);
 813         if (ret)
 814                 goto no_delete_lock;
 815         ret = btrfs_free_inode(trans, root, inode);
 816         if (ret)
 817                 goto no_delete_lock;
 818         nr = trans->blocks_used;
 819
 820         btrfs_end_transaction(trans, root);
 821         mutex_unlock(&root->fs_info->fs_mutex);
 822         btrfs_btree_balance_dirty(root, nr);
 823         return;
 824
 825 no_delete_lock:
 826         nr = trans->blocks_used;
 827         btrfs_end_transaction(trans, root);
 828         mutex_unlock(&root->fs_info->fs_mutex);
 829         btrfs_btree_balance_dirty(root, nr);
 830 no_delete:
 831         clear_inode(inode);
 832 }
 833
 834 /*
 835  * this returns the key found in the dir entry in the location pointer.
 836  * If no dir entries were found, location->objectid is 0.
 837  */
 838 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 839                                struct btrfs_key *location)
 840 {
 841         const char *name = dentry->d_name.name;
 842         int namelen = dentry->d_name.len;
 843         struct btrfs_dir_item *di;
 844         struct btrfs_path *path;
 845         struct btrfs_root *root = BTRFS_I(dir)->root;
 846         int ret;
 847
 848         path = btrfs_alloc_path();
 849         BUG_ON(!path);
 850         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
 851                                     namelen, 0);
 852         if (!di || IS_ERR(di)) {
 853                 location->objectid = 0;
 854                 ret = 0;
 855                 goto out;
 856         }
 857         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
 858 out:
 859         btrfs_release_path(root, path);
 860         btrfs_free_path(path);
 861         return ret;
 862 }
 863
 864 /*
 865  * when we hit a tree root in a directory, the btrfs part of the inode
 866  * needs to be changed to reflect the root directory of the tree root.  This
 867  * is kind of like crossing a mount point.
 868  */
 869 static int fixup_tree_root_location(struct btrfs_root *root,
 870                              struct btrfs_key *location,
 871                              struct btrfs_root **sub_root,
 872                              struct dentry *dentry)
 873 {
 874         struct btrfs_path *path;
 875         struct btrfs_root_item *ri;
 876
 877         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 878                 return 0;
 879         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 880                 return 0;
 881
 882         path = btrfs_alloc_path();
 883         BUG_ON(!path);
 884         mutex_lock(&root->fs_info->fs_mutex);
 885
 886         *sub_root = btrfs_read_fs_root(root->fs_info, location,
 887                                         dentry->d_name.name,
 888                                         dentry->d_name.len);
 889         if (IS_ERR(*sub_root))
 890                 return PTR_ERR(*sub_root);
 891
 892         ri = &(*sub_root)->root_item;
 893         location->objectid = btrfs_root_dirid(ri);
 894         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 895         location->offset = 0;
 896
 897         btrfs_free_path(path);
 898         mutex_unlock(&root->fs_info->fs_mutex);
 899         return 0;
 900 }
 901
 902 static int btrfs_init_locked_inode(struct inode *inode, void *p)
 903 {
 904         struct btrfs_iget_args *args = p;
 905         inode->i_ino = args->ino;
 906         BTRFS_I(inode)->root = args->root;
 907         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
 908                              inode->i_mapping, GFP_NOFS);
 909         return 0;
 910 }
 911
 912 static int btrfs_find_actor(struct inode *inode, void *opaque)
 913 {
 914         struct btrfs_iget_args *args = opaque;
 915         return (args->ino == inode->i_ino &&
 916                 args->root == BTRFS_I(inode)->root);
 917 }
 918
 919 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 920                                 struct btrfs_root *root)
 921 {
 922         struct inode *inode;
 923         struct btrfs_iget_args args;
 924         args.ino = objectid;
 925         args.root = root;
 926
 927         inode = iget5_locked(s, objectid, btrfs_find_actor,
 928                              btrfs_init_locked_inode,
 929                              (void *)&args);
 930         return inode;
 931 }
 932
 933 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 934                                    struct nameidata *nd)
 935 {
 936         struct inode * inode;
 937         struct btrfs_inode *bi = BTRFS_I(dir);
 938         struct btrfs_root *root = bi->root;
 939         struct btrfs_root *sub_root = root;
 940         struct btrfs_key location;
 941         int ret;
 942
 943         if (dentry->d_name.len > BTRFS_NAME_LEN)
 944                 return ERR_PTR(-ENAMETOOLONG);
 945
 946         mutex_lock(&root->fs_info->fs_mutex);
 947         ret = btrfs_inode_by_name(dir, dentry, &location);
 948         mutex_unlock(&root->fs_info->fs_mutex);
 949
 950         if (ret < 0)
 951                 return ERR_PTR(ret);
 952
 953         inode = NULL;
 954         if (location.objectid) {
 955                 ret = fixup_tree_root_location(root, &location, &sub_root,
 956                                                 dentry);
 957                 if (ret < 0)
 958                         return ERR_PTR(ret);
 959                 if (ret > 0)
 960                         return ERR_PTR(-ENOENT);
 961                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 962                                           sub_root);
 963                 if (!inode)
 964                         return ERR_PTR(-EACCES);
 965                 if (inode->i_state & I_NEW) {
 966                         /* the inode and parent dir are two different roots */
 967                         if (sub_root != root) {
 968                                 igrab(inode);
 969                                 sub_root->inode = inode;
 970                         }
 971                         BTRFS_I(inode)->root = sub_root;
 972                         memcpy(&BTRFS_I(inode)->location, &location,
 973                                sizeof(location));
 974                         btrfs_read_locked_inode(inode);
 975                         unlock_new_inode(inode);
 976                 }
 977         }
 978         return d_splice_alias(inode, dentry);
 979 }
 980
 981 static unsigned char btrfs_filetype_table[] = {
 982         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
 983 };
 984
 985 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 986 {
 987         struct inode *inode = filp->f_path.dentry->d_inode;
 988         struct btrfs_root *root = BTRFS_I(inode)->root;
 989         struct btrfs_item *item;
 990         struct btrfs_dir_item *di;
 991         struct btrfs_key key;
 992         struct btrfs_key found_key;
 993         struct btrfs_path *path;
 994         int ret;
 995         u32 nritems;
 996         struct extent_buffer *leaf;
 997         int slot;
 998         int advance;
 999         unsigned char d_type;
1000         int over = 0;
1001         u32 di_cur;
1002         u32 di_total;
1003         u32 di_len;
1004         int key_type = BTRFS_DIR_INDEX_KEY;
1005         char tmp_name[32];
1006         char *name_ptr;
1007         int name_len;
1008
1009         /* FIXME, use a real flag for deciding about the key type */
1010         if (root->fs_info->tree_root == root)
1011                 key_type = BTRFS_DIR_ITEM_KEY;
1012
1013         mutex_lock(&root->fs_info->fs_mutex);
1014         key.objectid = inode->i_ino;
1015         btrfs_set_key_type(&key, key_type);
1016         key.offset = filp->f_pos;
1017
1018         path = btrfs_alloc_path();
1019         path->reada = 2;
1020         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1021         if (ret < 0)
1022                 goto err;
1023         advance = 0;
1024         while(1) {
1025                 leaf = path->nodes[0];
1026                 nritems = btrfs_header_nritems(leaf);
1027                 slot = path->slots[0];
1028                 if (advance || slot >= nritems) {
1029                         if (slot >= nritems -1) {
1030                                 ret = btrfs_next_leaf(root, path);
1031                                 if (ret)
1032                                         break;
1033                                 leaf = path->nodes[0];
1034                                 nritems = btrfs_header_nritems(leaf);
1035                                 slot = path->slots[0];
1036                         } else {
1037                                 slot++;
1038                                 path->slots[0]++;
1039                         }
1040                 }
1041                 advance = 1;
1042                 item = btrfs_item_nr(leaf, slot);
1043                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1044
1045                 if (found_key.objectid != key.objectid)
1046                         break;
1047                 if (btrfs_key_type(&found_key) != key_type)
1048                         break;
1049                 if (found_key.offset < filp->f_pos)
1050                         continue;
1051
1052                 filp->f_pos = found_key.offset;
1053                 advance = 1;
1054                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1055                 di_cur = 0;
1056                 di_total = btrfs_item_size(leaf, item);
1057                 while(di_cur < di_total) {
1058                         struct btrfs_key location;
1059
1060                         name_len = btrfs_dir_name_len(leaf, di);
1061                         if (name_len < 32) {
1062                                 name_ptr = tmp_name;
1063                         } else {
1064                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1065                                 BUG_ON(!name_ptr);
1066                         }
1067                         read_extent_buffer(leaf, name_ptr,
1068                                            (unsigned long)(di + 1), name_len);
1069
1070                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1071                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1072
1073                         over = filldir(dirent, name_ptr, name_len,
1074                                        found_key.offset,
1075                                        location.objectid,
1076                                        d_type);
1077
1078                         if (name_ptr != tmp_name)
1079                                 kfree(name_ptr);
1080
1081                         if (over)
1082                                 goto nopos;
1083                         di_len = btrfs_dir_name_len(leaf, di) + sizeof(*di);
1084                         di_cur += di_len;
1085                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1086                 }
1087         }
1088         filp->f_pos++;
1089 nopos:
1090         ret = 0;
1091 err:
1092         btrfs_release_path(root, path);
1093         btrfs_free_path(path);
1094         mutex_unlock(&root->fs_info->fs_mutex);
1095         return ret;
1096 }
1097
1098 int btrfs_write_inode(struct inode *inode, int wait)
1099 {
1100         struct btrfs_root *root = BTRFS_I(inode)->root;
1101         struct btrfs_trans_handle *trans;
1102         int ret = 0;
1103
1104         if (wait) {
1105                 mutex_lock(&root->fs_info->fs_mutex);
1106                 trans = btrfs_start_transaction(root, 1);
1107                 btrfs_set_trans_block_group(trans, inode);
1108                 ret = btrfs_commit_transaction(trans, root);
1109                 mutex_unlock(&root->fs_info->fs_mutex);
1110         }
1111         return ret;
1112 }
1113
1114 /*
1115  * This is somewhat expensive, updating the tree every time the
1116  * inode changes.  But, it is most likely to find the inode in cache.
1117  * FIXME, needs more benchmarking...there are no reasons other than performance
1118  * to keep or drop this code.
1119  */
1120 void btrfs_dirty_inode(struct inode *inode)
1121 {
1122         struct btrfs_root *root = BTRFS_I(inode)->root;
1123         struct btrfs_trans_handle *trans;
1124
1125         mutex_lock(&root->fs_info->fs_mutex);
1126         trans = btrfs_start_transaction(root, 1);
1127         btrfs_set_trans_block_group(trans, inode);
1128         btrfs_update_inode(trans, root, inode);
1129         btrfs_end_transaction(trans, root);
1130         mutex_unlock(&root->fs_info->fs_mutex);
1131 }
1132
1133 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1134                                      struct btrfs_root *root,
1135                                      u64 objectid,
1136                                      struct btrfs_block_group_cache *group,
1137                                      int mode)
1138 {
1139         struct inode *inode;
1140         struct btrfs_inode_item *inode_item;
1141         struct btrfs_key *location;
1142         struct btrfs_path *path;
1143         int ret;
1144         int owner;
1145
1146         path = btrfs_alloc_path();
1147         BUG_ON(!path);
1148
1149         inode = new_inode(root->fs_info->sb);
1150         if (!inode)
1151                 return ERR_PTR(-ENOMEM);
1152
1153         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1154                              inode->i_mapping, GFP_NOFS);
1155         BTRFS_I(inode)->root = root;
1156
1157         if (mode & S_IFDIR)
1158                 owner = 0;
1159         else
1160                 owner = 1;
1161         group = btrfs_find_block_group(root, group, 0, 0, owner);
1162         BTRFS_I(inode)->block_group = group;
1163
1164         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1165         if (ret)
1166                 goto fail;
1167
1168         inode->i_uid = current->fsuid;
1169         inode->i_gid = current->fsgid;
1170         inode->i_mode = mode;
1171         inode->i_ino = objectid;
1172         inode->i_blocks = 0;
1173         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1174         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1175                                   struct btrfs_inode_item);
1176         fill_inode_item(path->nodes[0], inode_item, inode);
1177         btrfs_mark_buffer_dirty(path->nodes[0]);
1178         btrfs_free_path(path);
1179
1180         location = &BTRFS_I(inode)->location;
1181         location->objectid = objectid;
1182         location->offset = 0;
1183         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1184
1185         insert_inode_hash(inode);
1186         return inode;
1187 fail:
1188         btrfs_free_path(path);
1189         return ERR_PTR(ret);
1190 }
1191
1192 static inline u8 btrfs_inode_type(struct inode *inode)
1193 {
1194         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1195 }
1196
1197 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1198                             struct dentry *dentry, struct inode *inode)
1199 {
1200         int ret;
1201         struct btrfs_key key;
1202         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1203         struct inode *parent_inode;
1204
1205         key.objectid = inode->i_ino;
1206         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1207         key.offset = 0;
1208
1209         ret = btrfs_insert_dir_item(trans, root,
1210                                     dentry->d_name.name, dentry->d_name.len,
1211                                     dentry->d_parent->d_inode->i_ino,
1212                                     &key, btrfs_inode_type(inode));
1213         if (ret == 0) {
1214                 parent_inode = dentry->d_parent->d_inode;
1215                 parent_inode->i_size += dentry->d_name.len * 2;
1216                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1217                 ret = btrfs_update_inode(trans, root,
1218                                          dentry->d_parent->d_inode);
1219         }
1220         return ret;
1221 }
1222
1223 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1224                             struct dentry *dentry, struct inode *inode)
1225 {
1226         int err = btrfs_add_link(trans, dentry, inode);
1227         if (!err) {
1228                 d_instantiate(dentry, inode);
1229                 return 0;
1230         }
1231         if (err > 0)
1232                 err = -EEXIST;
1233         return err;
1234 }
1235
1236 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1237                         int mode, dev_t rdev)
1238 {
1239         struct btrfs_trans_handle *trans;
1240         struct btrfs_root *root = BTRFS_I(dir)->root;
1241         struct inode *inode;
1242         int err;
1243         int drop_inode = 0;
1244         u64 objectid;
1245         unsigned long nr;
1246
1247         if (!new_valid_dev(rdev))
1248                 return -EINVAL;
1249
1250         mutex_lock(&root->fs_info->fs_mutex);
1251         trans = btrfs_start_transaction(root, 1);
1252         btrfs_set_trans_block_group(trans, dir);
1253
1254         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1255         if (err) {
1256                 err = -ENOSPC;
1257                 goto out_unlock;
1258         }
1259
1260         inode = btrfs_new_inode(trans, root, objectid,
1261                                 BTRFS_I(dir)->block_group, mode);
1262         err = PTR_ERR(inode);
1263         if (IS_ERR(inode))
1264                 goto out_unlock;
1265
1266         btrfs_set_trans_block_group(trans, inode);
1267         err = btrfs_add_nondir(trans, dentry, inode);
1268         if (err)
1269                 drop_inode = 1;
1270         else {
1271                 inode->i_op = &btrfs_special_inode_operations;
1272                 init_special_inode(inode, inode->i_mode, rdev);
1273                 btrfs_update_inode(trans, root, inode);
1274         }
1275         dir->i_sb->s_dirt = 1;
1276         btrfs_update_inode_block_group(trans, inode);
1277         btrfs_update_inode_block_group(trans, dir);
1278 out_unlock:
1279         nr = trans->blocks_used;
1280         btrfs_end_transaction(trans, root);
1281         mutex_unlock(&root->fs_info->fs_mutex);
1282
1283         if (drop_inode) {
1284                 inode_dec_link_count(inode);
1285                 iput(inode);
1286         }
1287         btrfs_btree_balance_dirty(root, nr);
1288         return err;
1289 }
1290
1291 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1292                         int mode, struct nameidata *nd)
1293 {
1294         struct btrfs_trans_handle *trans;
1295         struct btrfs_root *root = BTRFS_I(dir)->root;
1296         struct inode *inode;
1297         int err;
1298         int drop_inode = 0;
1299         unsigned long nr;
1300         u64 objectid;
1301
1302         mutex_lock(&root->fs_info->fs_mutex);
1303         trans = btrfs_start_transaction(root, 1);
1304         btrfs_set_trans_block_group(trans, dir);
1305
1306         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1307         if (err) {
1308                 err = -ENOSPC;
1309                 goto out_unlock;
1310         }
1311
1312         inode = btrfs_new_inode(trans, root, objectid,
1313                                 BTRFS_I(dir)->block_group, mode);
1314         err = PTR_ERR(inode);
1315         if (IS_ERR(inode))
1316                 goto out_unlock;
1317
1318         btrfs_set_trans_block_group(trans, inode);
1319         err = btrfs_add_nondir(trans, dentry, inode);
1320         if (err)
1321                 drop_inode = 1;
1322         else {
1323                 inode->i_mapping->a_ops = &btrfs_aops;
1324                 inode->i_fop = &btrfs_file_operations;
1325                 inode->i_op = &btrfs_file_inode_operations;
1326                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1327                                      inode->i_mapping, GFP_NOFS);
1328                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1329         }
1330         dir->i_sb->s_dirt = 1;
1331         btrfs_update_inode_block_group(trans, inode);
1332         btrfs_update_inode_block_group(trans, dir);
1333 out_unlock:
1334         nr = trans->blocks_used;
1335         btrfs_end_transaction(trans, root);
1336         mutex_unlock(&root->fs_info->fs_mutex);
1337
1338         if (drop_inode) {
1339                 inode_dec_link_count(inode);
1340                 iput(inode);
1341         }
1342         btrfs_btree_balance_dirty(root, nr);
1343         return err;
1344 }
1345
1346 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1347                       struct dentry *dentry)
1348 {
1349         struct btrfs_trans_handle *trans;
1350         struct btrfs_root *root = BTRFS_I(dir)->root;
1351         struct inode *inode = old_dentry->d_inode;
1352         unsigned long nr;
1353         int err;
1354         int drop_inode = 0;
1355
1356         if (inode->i_nlink == 0)
1357                 return -ENOENT;
1358
1359         inc_nlink(inode);
1360         mutex_lock(&root->fs_info->fs_mutex);
1361         trans = btrfs_start_transaction(root, 1);
1362
1363         btrfs_set_trans_block_group(trans, dir);
1364         atomic_inc(&inode->i_count);
1365         err = btrfs_add_nondir(trans, dentry, inode);
1366
1367         if (err)
1368                 drop_inode = 1;
1369
1370         dir->i_sb->s_dirt = 1;
1371         btrfs_update_inode_block_group(trans, dir);
1372         err = btrfs_update_inode(trans, root, inode);
1373
1374         if (err)
1375                 drop_inode = 1;
1376
1377         nr = trans->blocks_used;
1378         btrfs_end_transaction(trans, root);
1379         mutex_unlock(&root->fs_info->fs_mutex);
1380
1381         if (drop_inode) {
1382                 inode_dec_link_count(inode);
1383                 iput(inode);
1384         }
1385         btrfs_btree_balance_dirty(root, nr);
1386         return err;
1387 }
1388
1389 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
1390                                 struct btrfs_root *root,
1391                                 u64 objectid, u64 dirid)
1392 {
1393         int ret;
1394         char buf[2];
1395         struct btrfs_key key;
1396
1397         buf[0] = '.';
1398         buf[1] = '.';
1399
1400         key.objectid = objectid;
1401         key.offset = 0;
1402         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1403
1404         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
1405                                     &key, BTRFS_FT_DIR);
1406         if (ret)
1407                 goto error;
1408
1409         key.objectid = dirid;
1410         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
1411                                     &key, BTRFS_FT_DIR);
1412         if (ret)
1413                 goto error;
1414 error:
1415         return ret;
1416 }
1417
1418 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1419 {
1420         struct inode *inode;
1421         struct btrfs_trans_handle *trans;
1422         struct btrfs_root *root = BTRFS_I(dir)->root;
1423         int err = 0;
1424         int drop_on_err = 0;
1425         u64 objectid;
1426         unsigned long nr = 1;
1427
1428         mutex_lock(&root->fs_info->fs_mutex);
1429         trans = btrfs_start_transaction(root, 1);
1430         btrfs_set_trans_block_group(trans, dir);
1431
1432         if (IS_ERR(trans)) {
1433                 err = PTR_ERR(trans);
1434                 goto out_unlock;
1435         }
1436
1437         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1438         if (err) {
1439                 err = -ENOSPC;
1440                 goto out_unlock;
1441         }
1442
1443         inode = btrfs_new_inode(trans, root, objectid,
1444                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1445         if (IS_ERR(inode)) {
1446                 err = PTR_ERR(inode);
1447                 goto out_fail;
1448         }
1449
1450         drop_on_err = 1;
1451         inode->i_op = &btrfs_dir_inode_operations;
1452         inode->i_fop = &btrfs_dir_file_operations;
1453         btrfs_set_trans_block_group(trans, inode);
1454
1455         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
1456         if (err)
1457                 goto out_fail;
1458
1459         inode->i_size = 6;
1460         err = btrfs_update_inode(trans, root, inode);
1461         if (err)
1462                 goto out_fail;
1463
1464         err = btrfs_add_link(trans, dentry, inode);
1465         if (err)
1466                 goto out_fail;
1467
1468         d_instantiate(dentry, inode);
1469         drop_on_err = 0;
1470         dir->i_sb->s_dirt = 1;
1471         btrfs_update_inode_block_group(trans, inode);
1472         btrfs_update_inode_block_group(trans, dir);
1473
1474 out_fail:
1475         nr = trans->blocks_used;
1476         btrfs_end_transaction(trans, root);
1477
1478 out_unlock:
1479         mutex_unlock(&root->fs_info->fs_mutex);
1480         if (drop_on_err)
1481                 iput(inode);
1482         btrfs_btree_balance_dirty(root, nr);
1483         return err;
1484 }
1485
1486 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1487                                     size_t page_offset, u64 start, u64 end,
1488                                     int create)
1489 {
1490         int ret;
1491         int err = 0;
1492         u64 bytenr;
1493         u64 extent_start = 0;
1494         u64 extent_end = 0;
1495         u64 objectid = inode->i_ino;
1496         u32 found_type;
1497         int failed_insert = 0;
1498         struct btrfs_path *path;
1499         struct btrfs_root *root = BTRFS_I(inode)->root;
1500         struct btrfs_file_extent_item *item;
1501         struct extent_buffer *leaf;
1502         struct btrfs_key found_key;
1503         struct extent_map *em = NULL;
1504         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1505         struct btrfs_trans_handle *trans = NULL;
1506
1507         path = btrfs_alloc_path();
1508         BUG_ON(!path);
1509         mutex_lock(&root->fs_info->fs_mutex);
1510
1511 again:
1512         em = lookup_extent_mapping(em_tree, start, end);
1513         if (em) {
1514                 goto out;
1515         }
1516         if (!em) {
1517                 em = alloc_extent_map(GFP_NOFS);
1518                 if (!em) {
1519                         err = -ENOMEM;
1520                         goto out;
1521                 }
1522                 em->start = EXTENT_MAP_HOLE;
1523                 em->end = EXTENT_MAP_HOLE;
1524         }
1525         em->bdev = inode->i_sb->s_bdev;
1526         ret = btrfs_lookup_file_extent(NULL, root, path,
1527                                        objectid, start, 0);
1528         if (ret < 0) {
1529                 err = ret;
1530                 goto out;
1531         }
1532
1533         if (ret != 0) {
1534                 if (path->slots[0] == 0)
1535                         goto not_found;
1536                 path->slots[0]--;
1537         }
1538
1539         leaf = path->nodes[0];
1540         item = btrfs_item_ptr(leaf, path->slots[0],
1541                               struct btrfs_file_extent_item);
1542         /* are we inside the extent that was found? */
1543         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1544         found_type = btrfs_key_type(&found_key);
1545         if (found_key.objectid != objectid ||
1546             found_type != BTRFS_EXTENT_DATA_KEY) {
1547                 goto not_found;
1548         }
1549
1550         found_type = btrfs_file_extent_type(leaf, item);
1551         extent_start = found_key.offset;
1552         if (found_type == BTRFS_FILE_EXTENT_REG) {
1553                 extent_end = extent_start +
1554                        btrfs_file_extent_num_bytes(leaf, item);
1555                 err = 0;
1556                 if (start < extent_start || start >= extent_end) {
1557                         em->start = start;
1558                         if (start < extent_start) {
1559                                 if (end < extent_start)
1560                                         goto not_found;
1561                                 em->end = extent_end - 1;
1562                         } else {
1563                                 em->end = end;
1564                         }
1565                         goto not_found_em;
1566                 }
1567                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1568                 if (bytenr == 0) {
1569                         em->start = extent_start;
1570                         em->end = extent_end - 1;
1571                         em->block_start = EXTENT_MAP_HOLE;
1572                         em->block_end = EXTENT_MAP_HOLE;
1573                         goto insert;
1574                 }
1575                 bytenr += btrfs_file_extent_offset(leaf, item);
1576                 em->block_start = bytenr;
1577                 em->block_end = em->block_start +
1578                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1579                 em->start = extent_start;
1580                 em->end = extent_end - 1;
1581                 goto insert;
1582         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1583                 unsigned long ptr;
1584                 char *map;
1585                 size_t size;
1586                 size_t extent_offset;
1587                 size_t copy_size;
1588
1589                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1590                                                     path->slots[0]));
1591
1592                 extent_end = (extent_start + size) |
1593                         ((u64)root->sectorsize - 1);
1594                 if (start < extent_start || start >= extent_end) {
1595                         em->start = start;
1596                         if (start < extent_start) {
1597                                 if (end < extent_start)
1598                                         goto not_found;
1599                                 em->end = extent_end;
1600                         } else {
1601                                 em->end = end;
1602                         }
1603                         goto not_found_em;
1604                 }
1605
1606                 extent_offset = (page->index << PAGE_CACHE_SHIFT) -
1607                         extent_start;
1608                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1609                 map = kmap(page);
1610                 copy_size = min(PAGE_CACHE_SIZE - page_offset,
1611                                 size - extent_offset);
1612
1613                 em->block_start = EXTENT_MAP_INLINE;
1614                 em->block_end = EXTENT_MAP_INLINE;
1615                 em->start = extent_start + extent_offset;
1616                 em->end = (em->start + copy_size -1) |
1617                         ((u64)root->sectorsize -1);
1618
1619                 if (!page) {
1620                         goto insert;
1621                 }
1622
1623                 read_extent_buffer(leaf, map + page_offset, ptr, copy_size);
1624                 /*
1625                 memset(map + page_offset + copy_size, 0,
1626                        PAGE_CACHE_SIZE - copy_size - page_offset);
1627                        */
1628                 flush_dcache_page(page);
1629                 kunmap(page);
1630                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1631                 goto insert;
1632         } else {
1633                 printk("unkknown found_type %d\n", found_type);
1634                 WARN_ON(1);
1635         }
1636 not_found:
1637         em->start = start;
1638         em->end = end;
1639 not_found_em:
1640         em->block_start = EXTENT_MAP_HOLE;
1641         em->block_end = EXTENT_MAP_HOLE;
1642 insert:
1643         btrfs_release_path(root, path);
1644         if (em->start > start || em->end < start) {
1645                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1646                 err = -EIO;
1647                 goto out;
1648         }
1649         ret = add_extent_mapping(em_tree, em);
1650         if (ret == -EEXIST) {
1651                 free_extent_map(em);
1652                 em = NULL;
1653                 failed_insert++;
1654                 if (failed_insert > 5) {
1655                         printk("failing to insert %Lu %Lu\n", start, end);
1656                         err = -EIO;
1657                         goto out;
1658                 }
1659                 goto again;
1660         }
1661         err = 0;
1662 out:
1663         btrfs_free_path(path);
1664         if (trans) {
1665                 ret = btrfs_end_transaction(trans, root);
1666                 if (!err)
1667                         err = ret;
1668         }
1669         mutex_unlock(&root->fs_info->fs_mutex);
1670         if (err) {
1671                 free_extent_map(em);
1672                 WARN_ON(1);
1673                 return ERR_PTR(err);
1674         }
1675         return em;
1676 }
1677
1678 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1679 {
1680         return extent_bmap(mapping, iblock, btrfs_get_extent);
1681 }
1682
1683 static int btrfs_prepare_write(struct file *file, struct page *page,
1684                                unsigned from, unsigned to)
1685 {
1686         return extent_prepare_write(&BTRFS_I(page->mapping->host)->extent_tree,
1687                                     page->mapping->host, page, from, to,
1688                                     btrfs_get_extent);
1689 }
1690
1691 int btrfs_readpage(struct file *file, struct page *page)
1692 {
1693         struct extent_map_tree *tree;
1694         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1695         return extent_read_full_page(tree, page, btrfs_get_extent);
1696 }
1697 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1698 {
1699         struct extent_map_tree *tree;
1700
1701
1702         if (current->flags & PF_MEMALLOC) {
1703                 redirty_page_for_writepage(wbc, page);
1704                 unlock_page(page);
1705                 return 0;
1706         }
1707         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1708         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
1709 }
1710
1711 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1712 {
1713         struct extent_map_tree *tree;
1714         int ret;
1715
1716         if (page->private != 1) {
1717                 WARN_ON(1);
1718                 return try_to_free_buffers(page);
1719         }
1720         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1721         ret = try_release_extent_mapping(tree, page);
1722         if (ret == 1) {
1723                 ClearPagePrivate(page);
1724                 set_page_private(page, 0);
1725                 page_cache_release(page);
1726         }
1727         return ret;
1728 }
1729
1730 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
1731 {
1732         struct extent_map_tree *tree;
1733
1734         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1735         extent_invalidatepage(tree, page, offset);
1736         btrfs_releasepage(page, GFP_NOFS);
1737 }
1738
1739 /*
1740  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
1741  * called from a page fault handler when a page is first dirtied. Hence we must
1742  * be careful to check for EOF conditions here. We set the page up correctly
1743  * for a written page which means we get ENOSPC checking when writing into
1744  * holes and correct delalloc and unwritten extent mapping on filesystems that
1745  * support these features.
1746  *
1747  * We are not allowed to take the i_mutex here so we have to play games to
1748  * protect against truncate races as the page could now be beyond EOF.  Because
1749  * vmtruncate() writes the inode size before removing pages, once we have the
1750  * page lock we can determine safely if the page is beyond EOF. If it is not
1751  * beyond EOF, then the page is guaranteed safe against truncation until we
1752  * unlock the page.
1753  */
1754 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1755 {
1756         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1757         unsigned long end;
1758         loff_t size;
1759         int ret = -EINVAL;
1760         u64 page_start;
1761
1762         down_read(&BTRFS_I(inode)->root->snap_sem);
1763         lock_page(page);
1764         wait_on_page_writeback(page);
1765         size = i_size_read(inode);
1766         page_start = page->index << PAGE_CACHE_SHIFT;
1767
1768         if ((page->mapping != inode->i_mapping) ||
1769             (page_start > size)) {
1770                 /* page got truncated out from underneath us */
1771                 goto out_unlock;
1772         }
1773
1774         /* page is wholly or partially inside EOF */
1775         if (page_start + PAGE_CACHE_SIZE > size)
1776                 end = size & ~PAGE_CACHE_MASK;
1777         else
1778                 end = PAGE_CACHE_SIZE;
1779
1780         ret = btrfs_cow_one_page(inode, page, end);
1781
1782 out_unlock:
1783         up_read(&BTRFS_I(inode)->root->snap_sem);
1784         unlock_page(page);
1785         return ret;
1786 }
1787
1788 static void btrfs_truncate(struct inode *inode)
1789 {
1790         struct btrfs_root *root = BTRFS_I(inode)->root;
1791         int ret;
1792         struct btrfs_trans_handle *trans;
1793         unsigned long nr;
1794
1795         if (!S_ISREG(inode->i_mode))
1796                 return;
1797         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1798                 return;
1799
1800         btrfs_truncate_page(inode->i_mapping, inode->i_size);
1801
1802         mutex_lock(&root->fs_info->fs_mutex);
1803         trans = btrfs_start_transaction(root, 1);
1804         btrfs_set_trans_block_group(trans, inode);
1805
1806         /* FIXME, add redo link to tree so we don't leak on crash */
1807         ret = btrfs_truncate_in_trans(trans, root, inode);
1808         btrfs_update_inode(trans, root, inode);
1809         nr = trans->blocks_used;
1810
1811         ret = btrfs_end_transaction(trans, root);
1812         BUG_ON(ret);
1813         mutex_unlock(&root->fs_info->fs_mutex);
1814         btrfs_btree_balance_dirty(root, nr);
1815 }
1816
1817 int btrfs_commit_write(struct file *file, struct page *page,
1818                        unsigned from, unsigned to)
1819 {
1820         return extent_commit_write(&BTRFS_I(page->mapping->host)->extent_tree,
1821                                    page->mapping->host, page, from, to);
1822 }
1823
1824 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
1825 {
1826         struct btrfs_trans_handle *trans;
1827         struct btrfs_key key;
1828         struct btrfs_root_item root_item;
1829         struct btrfs_inode_item *inode_item;
1830         struct extent_buffer *leaf;
1831         struct btrfs_root *new_root;
1832         struct inode *inode;
1833         struct inode *dir;
1834         int ret;
1835         int err;
1836         u64 objectid;
1837         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
1838         unsigned long nr = 1;
1839
1840         mutex_lock(&root->fs_info->fs_mutex);
1841         trans = btrfs_start_transaction(root, 1);
1842         BUG_ON(!trans);
1843
1844         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 0);
1845         if (IS_ERR(leaf))
1846                 return PTR_ERR(leaf);
1847
1848         btrfs_set_header_nritems(leaf, 0);
1849         btrfs_set_header_level(leaf, 0);
1850         btrfs_set_header_bytenr(leaf, leaf->start);
1851         btrfs_set_header_generation(leaf, trans->transid);
1852         btrfs_set_header_owner(leaf, root->root_key.objectid);
1853         write_extent_buffer(leaf, root->fs_info->fsid,
1854                             (unsigned long)btrfs_header_fsid(leaf),
1855                             BTRFS_FSID_SIZE);
1856         btrfs_mark_buffer_dirty(leaf);
1857
1858         inode_item = &root_item.inode;
1859         memset(inode_item, 0, sizeof(*inode_item));
1860         inode_item->generation = cpu_to_le64(1);
1861         inode_item->size = cpu_to_le64(3);
1862         inode_item->nlink = cpu_to_le32(1);
1863         inode_item->nblocks = cpu_to_le64(1);
1864         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1865
1866         btrfs_set_root_bytenr(&root_item, leaf->start);
1867         btrfs_set_root_level(&root_item, 0);
1868         btrfs_set_root_refs(&root_item, 1);
1869         btrfs_set_root_used(&root_item, 0);
1870
1871         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
1872         root_item.drop_level = 0;
1873
1874         free_extent_buffer(leaf);
1875         leaf = NULL;
1876
1877         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1878                                        0, &objectid);
1879         if (ret)
1880                 goto fail;
1881
1882         btrfs_set_root_dirid(&root_item, new_dirid);
1883
1884         key.objectid = objectid;
1885         key.offset = 1;
1886         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1887         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1888                                 &root_item);
1889         if (ret)
1890                 goto fail;
1891
1892         /*
1893          * insert the directory item
1894          */
1895         key.offset = (u64)-1;
1896         dir = root->fs_info->sb->s_root->d_inode;
1897         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1898                                     name, namelen, dir->i_ino, &key,
1899                                     BTRFS_FT_DIR);
1900         if (ret)
1901                 goto fail;
1902
1903         ret = btrfs_commit_transaction(trans, root);
1904         if (ret)
1905                 goto fail_commit;
1906
1907         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
1908         BUG_ON(!new_root);
1909
1910         trans = btrfs_start_transaction(new_root, 1);
1911         BUG_ON(!trans);
1912
1913         inode = btrfs_new_inode(trans, new_root, new_dirid,
1914                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
1915         if (IS_ERR(inode))
1916                 goto fail;
1917         inode->i_op = &btrfs_dir_inode_operations;
1918         inode->i_fop = &btrfs_dir_file_operations;
1919         new_root->inode = inode;
1920
1921         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
1922         if (ret)
1923                 goto fail;
1924
1925         inode->i_nlink = 1;
1926         inode->i_size = 6;
1927         ret = btrfs_update_inode(trans, new_root, inode);
1928         if (ret)
1929                 goto fail;
1930 fail:
1931         nr = trans->blocks_used;
1932         err = btrfs_commit_transaction(trans, root);
1933         if (err && !ret)
1934                 ret = err;
1935 fail_commit:
1936         mutex_unlock(&root->fs_info->fs_mutex);
1937         btrfs_btree_balance_dirty(root, nr);
1938         return ret;
1939 }
1940
1941 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
1942 {
1943         struct btrfs_trans_handle *trans;
1944         struct btrfs_key key;
1945         struct btrfs_root_item new_root_item;
1946         struct extent_buffer *tmp;
1947         int ret;
1948         int err;
1949         u64 objectid;
1950         unsigned long nr;
1951
1952         if (!root->ref_cows)
1953                 return -EINVAL;
1954
1955         down_write(&root->snap_sem);
1956         freeze_bdev(root->fs_info->sb->s_bdev);
1957         thaw_bdev(root->fs_info->sb->s_bdev, root->fs_info->sb);
1958
1959         mutex_lock(&root->fs_info->fs_mutex);
1960         trans = btrfs_start_transaction(root, 1);
1961         BUG_ON(!trans);
1962
1963         ret = btrfs_update_inode(trans, root, root->inode);
1964         if (ret)
1965                 goto fail;
1966
1967         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1968                                        0, &objectid);
1969         if (ret)
1970                 goto fail;
1971
1972         memcpy(&new_root_item, &root->root_item,
1973                sizeof(new_root_item));
1974
1975         key.objectid = objectid;
1976         key.offset = 1;
1977         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1978
1979         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
1980         btrfs_set_root_bytenr(&new_root_item, root->node->start);
1981         btrfs_set_root_level(&new_root_item, btrfs_header_level(root->node));
1982
1983         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1984                                 &new_root_item);
1985         if (ret)
1986                 goto fail;
1987
1988         /*
1989          * insert the directory item
1990          */
1991         key.offset = (u64)-1;
1992         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1993                                     name, namelen,
1994                                     root->fs_info->sb->s_root->d_inode->i_ino,
1995                                     &key, BTRFS_FT_DIR);
1996
1997         if (ret)
1998                 goto fail;
1999
2000         ret = btrfs_inc_root_ref(trans, root);
2001         if (ret)
2002                 goto fail;
2003 fail:
2004         nr = trans->blocks_used;
2005         err = btrfs_commit_transaction(trans, root);
2006
2007         if (err && !ret)
2008                 ret = err;
2009
2010         mutex_unlock(&root->fs_info->fs_mutex);
2011         up_write(&root->snap_sem);
2012         btrfs_btree_balance_dirty(root, nr);
2013         return ret;
2014 }
2015
2016 static unsigned long force_ra(struct address_space *mapping,
2017                               struct file_ra_state *ra, struct file *file,
2018                               pgoff_t offset, pgoff_t last_index)
2019 {
2020         pgoff_t req_size;
2021
2022 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2023         req_size = last_index - offset + 1;
2024         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2025         return offset;
2026 #else
2027         req_size = min(last_index - offset + 1, (pgoff_t)128);
2028         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2029         return offset + req_size;
2030 #endif
2031 }
2032
2033 int btrfs_defrag_file(struct file *file) {
2034         struct inode *inode = file->f_path.dentry->d_inode;
2035         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2036         struct page *page;
2037         unsigned long last_index;
2038         unsigned long ra_index = 0;
2039         u64 page_start;
2040         u64 page_end;
2041         unsigned long i;
2042
2043         mutex_lock(&inode->i_mutex);
2044         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2045         for (i = 0; i <= last_index; i++) {
2046                 if (i == ra_index) {
2047                         ra_index = force_ra(inode->i_mapping, &file->f_ra,
2048                                             file, ra_index, last_index);
2049                 }
2050                 page = grab_cache_page(inode->i_mapping, i);
2051                 if (!page)
2052                         goto out_unlock;
2053                 if (!PageUptodate(page)) {
2054                         btrfs_readpage(NULL, page);
2055                         lock_page(page);
2056                         if (!PageUptodate(page)) {
2057                                 unlock_page(page);
2058                                 page_cache_release(page);
2059                                 goto out_unlock;
2060                         }
2061                 }
2062                 page_start = page->index << PAGE_CACHE_SHIFT;
2063                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2064
2065                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2066                 set_extent_delalloc(em_tree, page_start,
2067                                     page_end, GFP_NOFS);
2068                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2069                 set_page_dirty(page);
2070                 unlock_page(page);
2071                 page_cache_release(page);
2072                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2073         }
2074
2075 out_unlock:
2076         mutex_unlock(&inode->i_mutex);
2077         return 0;
2078 }
2079
2080 static int btrfs_ioctl_snap_create(struct btrfs_root *root, void __user *arg)
2081 {
2082         struct btrfs_ioctl_vol_args vol_args;
2083         struct btrfs_dir_item *di;
2084         struct btrfs_path *path;
2085         int namelen;
2086         u64 root_dirid;
2087
2088         if (copy_from_user(&vol_args, arg, sizeof(vol_args)))
2089                 return -EFAULT;
2090
2091         namelen = strlen(vol_args.name);
2092         if (namelen > BTRFS_VOL_NAME_MAX)
2093                 return -EINVAL;
2094         if (strchr(vol_args.name, '/'))
2095                 return -EINVAL;
2096
2097         path = btrfs_alloc_path();
2098         if (!path)
2099                 return -ENOMEM;
2100
2101         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2102         mutex_lock(&root->fs_info->fs_mutex);
2103         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2104                             path, root_dirid,
2105                             vol_args.name, namelen, 0);
2106         mutex_unlock(&root->fs_info->fs_mutex);
2107         btrfs_free_path(path);
2108         if (di && !IS_ERR(di))
2109                 return -EEXIST;
2110         if (IS_ERR(di))
2111                 return PTR_ERR(di);
2112
2113         if (root == root->fs_info->tree_root)
2114                 return create_subvol(root, vol_args.name, namelen);
2115         return create_snapshot(root, vol_args.name, namelen);
2116 }
2117
2118 static int btrfs_ioctl_defrag(struct file *file)
2119 {
2120         struct inode *inode = file->f_path.dentry->d_inode;
2121         struct btrfs_root *root = BTRFS_I(inode)->root;
2122
2123         switch (inode->i_mode & S_IFMT) {
2124         case S_IFDIR:
2125                 mutex_lock(&root->fs_info->fs_mutex);
2126                 btrfs_defrag_root(root, 0);
2127                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2128                 mutex_unlock(&root->fs_info->fs_mutex);
2129                 break;
2130         case S_IFREG:
2131                 btrfs_defrag_file(file);
2132                 break;
2133         }
2134
2135         return 0;
2136 }
2137
2138 long btrfs_ioctl(struct file *file, unsigned int
2139                 cmd, unsigned long arg)
2140 {
2141         struct btrfs_root *root = BTRFS_I(file->f_path.dentry->d_inode)->root;
2142
2143         switch (cmd) {
2144         case BTRFS_IOC_SNAP_CREATE:
2145                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2146         case BTRFS_IOC_DEFRAG:
2147                 return btrfs_ioctl_defrag(file);
2148         }
2149
2150         return -ENOTTY;
2151 }
2152
2153 /*
2154  * Called inside transaction, so use GFP_NOFS
2155  */
2156 struct inode *btrfs_alloc_inode(struct super_block *sb)
2157 {
2158         struct btrfs_inode *ei;
2159
2160         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2161         if (!ei)
2162                 return NULL;
2163         ei->last_trans = 0;
2164         return &ei->vfs_inode;
2165 }
2166
2167 void btrfs_destroy_inode(struct inode *inode)
2168 {
2169         WARN_ON(!list_empty(&inode->i_dentry));
2170         WARN_ON(inode->i_data.nrpages);
2171
2172         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2173 }
2174
2175 static void init_once(void * foo, struct kmem_cache * cachep,
2176                       unsigned long flags)
2177 {
2178         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2179
2180         inode_init_once(&ei->vfs_inode);
2181 }
2182
2183 void btrfs_destroy_cachep(void)
2184 {
2185         if (btrfs_inode_cachep)
2186                 kmem_cache_destroy(btrfs_inode_cachep);
2187         if (btrfs_trans_handle_cachep)
2188                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2189         if (btrfs_transaction_cachep)
2190                 kmem_cache_destroy(btrfs_transaction_cachep);
2191         if (btrfs_bit_radix_cachep)
2192                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2193         if (btrfs_path_cachep)
2194                 kmem_cache_destroy(btrfs_path_cachep);
2195 }
2196
2197 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2198                                        unsigned long extra_flags,
2199                                        void (*ctor)(void *, struct kmem_cache *,
2200                                                     unsigned long))
2201 {
2202         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2203                                  SLAB_MEM_SPREAD | extra_flags), ctor
2204 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2205                                  ,NULL
2206 #endif
2207                                 );
2208 }
2209
2210 int btrfs_init_cachep(void)
2211 {
2212         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2213                                           sizeof(struct btrfs_inode),
2214                                           0, init_once);
2215         if (!btrfs_inode_cachep)
2216                 goto fail;
2217         btrfs_trans_handle_cachep =
2218                         btrfs_cache_create("btrfs_trans_handle_cache",
2219                                            sizeof(struct btrfs_trans_handle),
2220                                            0, NULL);
2221         if (!btrfs_trans_handle_cachep)
2222                 goto fail;
2223         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2224                                              sizeof(struct btrfs_transaction),
2225                                              0, NULL);
2226         if (!btrfs_transaction_cachep)
2227                 goto fail;
2228         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2229                                          sizeof(struct btrfs_path),
2230                                          0, NULL);
2231         if (!btrfs_path_cachep)
2232                 goto fail;
2233         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2234                                               SLAB_DESTROY_BY_RCU, NULL);
2235         if (!btrfs_bit_radix_cachep)
2236                 goto fail;
2237         return 0;
2238 fail:
2239         btrfs_destroy_cachep();
2240         return -ENOMEM;
2241 }
2242
2243 static int btrfs_getattr(struct vfsmount *mnt,
2244                          struct dentry *dentry, struct kstat *stat)
2245 {
2246         struct inode *inode = dentry->d_inode;
2247         generic_fillattr(inode, stat);
2248         stat->blksize = 256 * 1024;
2249         return 0;
2250 }
2251
2252 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2253                            struct inode * new_dir,struct dentry *new_dentry)
2254 {
2255         struct btrfs_trans_handle *trans;
2256         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2257         struct inode *new_inode = new_dentry->d_inode;
2258         struct inode *old_inode = old_dentry->d_inode;
2259         struct timespec ctime = CURRENT_TIME;
2260         struct btrfs_path *path;
2261         struct btrfs_dir_item *di;
2262         int ret;
2263
2264         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2265             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2266                 return -ENOTEMPTY;
2267         }
2268
2269         mutex_lock(&root->fs_info->fs_mutex);
2270         trans = btrfs_start_transaction(root, 1);
2271
2272         btrfs_set_trans_block_group(trans, new_dir);
2273         path = btrfs_alloc_path();
2274         if (!path) {
2275                 ret = -ENOMEM;
2276                 goto out_fail;
2277         }
2278
2279         old_dentry->d_inode->i_nlink++;
2280         old_dir->i_ctime = old_dir->i_mtime = ctime;
2281         new_dir->i_ctime = new_dir->i_mtime = ctime;
2282         old_inode->i_ctime = ctime;
2283
2284         if (S_ISDIR(old_inode->i_mode) && old_dir != new_dir) {
2285                 struct btrfs_key *location = &BTRFS_I(new_dir)->location;
2286                 struct btrfs_key old_parent_key;
2287                 di = btrfs_lookup_dir_item(trans, root, path, old_inode->i_ino,
2288                                            "..", 2, -1);
2289                 if (IS_ERR(di)) {
2290                         ret = PTR_ERR(di);
2291                         goto out_fail;
2292                 }
2293                 if (!di) {
2294                         ret = -ENOENT;
2295                         goto out_fail;
2296                 }
2297                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &old_parent_key);
2298                 ret = btrfs_del_item(trans, root, path);
2299                 if (ret) {
2300                         goto out_fail;
2301                 }
2302                 btrfs_release_path(root, path);
2303
2304                 di = btrfs_lookup_dir_index_item(trans, root, path,
2305                                                  old_inode->i_ino,
2306                                                  old_parent_key.objectid,
2307                                                  "..", 2, -1);
2308                 if (IS_ERR(di)) {
2309                         ret = PTR_ERR(di);
2310                         goto out_fail;
2311                 }
2312                 if (!di) {
2313                         ret = -ENOENT;
2314                         goto out_fail;
2315                 }
2316                 ret = btrfs_del_item(trans, root, path);
2317                 if (ret) {
2318                         goto out_fail;
2319                 }
2320                 btrfs_release_path(root, path);
2321
2322                 ret = btrfs_insert_dir_item(trans, root, "..", 2,
2323                                             old_inode->i_ino, location,
2324                                             BTRFS_FT_DIR);
2325                 if (ret)
2326                         goto out_fail;
2327         }
2328
2329
2330         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2331         if (ret)
2332                 goto out_fail;
2333
2334         if (new_inode) {
2335                 new_inode->i_ctime = CURRENT_TIME;
2336                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2337                 if (ret)
2338                         goto out_fail;
2339         }
2340         ret = btrfs_add_link(trans, new_dentry, old_inode);
2341         if (ret)
2342                 goto out_fail;
2343
2344 out_fail:
2345         btrfs_free_path(path);
2346         btrfs_end_transaction(trans, root);
2347         mutex_unlock(&root->fs_info->fs_mutex);
2348         return ret;
2349 }
2350
2351 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2352                          const char *symname)
2353 {
2354         struct btrfs_trans_handle *trans;
2355         struct btrfs_root *root = BTRFS_I(dir)->root;
2356         struct btrfs_path *path;
2357         struct btrfs_key key;
2358         struct inode *inode;
2359         int err;
2360         int drop_inode = 0;
2361         u64 objectid;
2362         int name_len;
2363         int datasize;
2364         unsigned long ptr;
2365         struct btrfs_file_extent_item *ei;
2366         struct extent_buffer *leaf;
2367         unsigned long nr;
2368
2369         name_len = strlen(symname) + 1;
2370         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2371                 return -ENAMETOOLONG;
2372         mutex_lock(&root->fs_info->fs_mutex);
2373         trans = btrfs_start_transaction(root, 1);
2374         btrfs_set_trans_block_group(trans, dir);
2375
2376         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2377         if (err) {
2378                 err = -ENOSPC;
2379                 goto out_unlock;
2380         }
2381
2382         inode = btrfs_new_inode(trans, root, objectid,
2383                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2384         err = PTR_ERR(inode);
2385         if (IS_ERR(inode))
2386                 goto out_unlock;
2387
2388         btrfs_set_trans_block_group(trans, inode);
2389         err = btrfs_add_nondir(trans, dentry, inode);
2390         if (err)
2391                 drop_inode = 1;
2392         else {
2393                 inode->i_mapping->a_ops = &btrfs_aops;
2394                 inode->i_fop = &btrfs_file_operations;
2395                 inode->i_op = &btrfs_file_inode_operations;
2396                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2397                                      inode->i_mapping, GFP_NOFS);
2398                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2399         }
2400         dir->i_sb->s_dirt = 1;
2401         btrfs_update_inode_block_group(trans, inode);
2402         btrfs_update_inode_block_group(trans, dir);
2403         if (drop_inode)
2404                 goto out_unlock;
2405
2406         path = btrfs_alloc_path();
2407         BUG_ON(!path);
2408         key.objectid = inode->i_ino;
2409         key.offset = 0;
2410         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2411         datasize = btrfs_file_extent_calc_inline_size(name_len);
2412         err = btrfs_insert_empty_item(trans, root, path, &key,
2413                                       datasize);
2414         if (err) {
2415                 drop_inode = 1;
2416                 goto out_unlock;
2417         }
2418         leaf = path->nodes[0];
2419         ei = btrfs_item_ptr(leaf, path->slots[0],
2420                             struct btrfs_file_extent_item);
2421         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2422         btrfs_set_file_extent_type(leaf, ei,
2423                                    BTRFS_FILE_EXTENT_INLINE);
2424         ptr = btrfs_file_extent_inline_start(ei);
2425         write_extent_buffer(leaf, symname, ptr, name_len);
2426         btrfs_mark_buffer_dirty(leaf);
2427         btrfs_free_path(path);
2428
2429         inode->i_op = &btrfs_symlink_inode_operations;
2430         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2431         inode->i_size = name_len - 1;
2432         err = btrfs_update_inode(trans, root, inode);
2433         if (err)
2434                 drop_inode = 1;
2435
2436 out_unlock:
2437         nr = trans->blocks_used;
2438         btrfs_end_transaction(trans, root);
2439         mutex_unlock(&root->fs_info->fs_mutex);
2440         if (drop_inode) {
2441                 inode_dec_link_count(inode);
2442                 iput(inode);
2443         }
2444         btrfs_btree_balance_dirty(root, nr);
2445         return err;
2446 }
2447
2448 static struct inode_operations btrfs_dir_inode_operations = {
2449         .lookup         = btrfs_lookup,
2450         .create         = btrfs_create,
2451         .unlink         = btrfs_unlink,
2452         .link           = btrfs_link,
2453         .mkdir          = btrfs_mkdir,
2454         .rmdir          = btrfs_rmdir,
2455         .rename         = btrfs_rename,
2456         .symlink        = btrfs_symlink,
2457         .setattr        = btrfs_setattr,
2458         .mknod          = btrfs_mknod,
2459 };
2460
2461 static struct inode_operations btrfs_dir_ro_inode_operations = {
2462         .lookup         = btrfs_lookup,
2463 };
2464
2465 static struct file_operations btrfs_dir_file_operations = {
2466         .llseek         = generic_file_llseek,
2467         .read           = generic_read_dir,
2468         .readdir        = btrfs_readdir,
2469         .unlocked_ioctl = btrfs_ioctl,
2470 #ifdef CONFIG_COMPAT
2471         .compat_ioctl   = btrfs_ioctl,
2472 #endif
2473 };
2474
2475 static struct extent_map_ops btrfs_extent_map_ops = {
2476         .fill_delalloc = run_delalloc_range,
2477         .writepage_io_hook = btrfs_writepage_io_hook,
2478         .readpage_io_hook = btrfs_readpage_io_hook,
2479         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2480 };
2481
2482 static struct address_space_operations btrfs_aops = {
2483         .readpage       = btrfs_readpage,
2484         .writepage      = btrfs_writepage,
2485         .sync_page      = block_sync_page,
2486         .prepare_write  = btrfs_prepare_write,
2487         .commit_write   = btrfs_commit_write,
2488         .bmap           = btrfs_bmap,
2489         .invalidatepage = btrfs_invalidatepage,
2490         .releasepage    = btrfs_releasepage,
2491         .set_page_dirty = __set_page_dirty_nobuffers,
2492 };
2493
2494 static struct address_space_operations btrfs_symlink_aops = {
2495         .readpage       = btrfs_readpage,
2496         .writepage      = btrfs_writepage,
2497         .invalidatepage = btrfs_invalidatepage,
2498         .releasepage    = btrfs_releasepage,
2499 };
2500
2501 static struct inode_operations btrfs_file_inode_operations = {
2502         .truncate       = btrfs_truncate,
2503         .getattr        = btrfs_getattr,
2504         .setattr        = btrfs_setattr,
2505 };
2506
2507 static struct inode_operations btrfs_special_inode_operations = {
2508         .getattr        = btrfs_getattr,
2509         .setattr        = btrfs_setattr,
2510 };
2511
2512 static struct inode_operations btrfs_symlink_inode_operations = {
2513         .readlink       = generic_readlink,
2514         .follow_link    = page_follow_link_light,
2515         .put_link       = page_put_link,
2516 };