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