fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/buffer_head.h>
  20 #include <linux/fs.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/string.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mpage.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/version.h>
  35 #include "ctree.h"
  36 #include "disk-io.h"
  37 #include "transaction.h"
  38 #include "btrfs_inode.h"
  39 #include "ioctl.h"
  40 #include "print-tree.h"
  41
  42 struct btrfs_iget_args {
  43         u64 ino;
  44         struct btrfs_root *root;
  45 };
  46
  47 static struct inode_operations btrfs_dir_inode_operations;
  48 static struct inode_operations btrfs_symlink_inode_operations;
  49 static struct inode_operations btrfs_dir_ro_inode_operations;
  50 static struct inode_operations btrfs_special_inode_operations;
  51 static struct inode_operations btrfs_file_inode_operations;
  52 static struct address_space_operations btrfs_aops;
  53 static struct address_space_operations btrfs_symlink_aops;
  54 static struct file_operations btrfs_dir_file_operations;
  55 static struct extent_map_ops btrfs_extent_map_ops;
  56
  57 static struct kmem_cache *btrfs_inode_cachep;
  58 struct kmem_cache *btrfs_trans_handle_cachep;
  59 struct kmem_cache *btrfs_transaction_cachep;
  60 struct kmem_cache *btrfs_bit_radix_cachep;
  61 struct kmem_cache *btrfs_path_cachep;
  62
  63 #define S_SHIFT 12
  64 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  65         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  66         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  67         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  68         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  69         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  70         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  71         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  72 };
  73
  74 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
  75 {
  76         struct btrfs_root *root = BTRFS_I(inode)->root;
  77         struct btrfs_trans_handle *trans;
  78         struct btrfs_key ins;
  79         u64 alloc_hint = 0;
  80         u64 num_bytes;
  81         int ret;
  82         u64 blocksize = root->sectorsize;
  83
  84         mutex_lock(&root->fs_info->fs_mutex);
  85         trans = btrfs_start_transaction(root, 1);
  86         btrfs_set_trans_block_group(trans, inode);
  87         BUG_ON(!trans);
  88         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
  89         ret = btrfs_drop_extents(trans, root, inode,
  90                                  start, start + num_bytes, start, &alloc_hint);
  91
  92         ret = btrfs_alloc_extent(trans, root, inode->i_ino, num_bytes, 0,
  93                                  alloc_hint, (u64)-1, &ins, 1);
  94         if (ret) {
  95                 WARN_ON(1);
  96                 goto out;
  97         }
  98         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
  99                                        start, ins.objectid, ins.offset,
 100                                        ins.offset);
 101 out:
 102         btrfs_end_transaction(trans, root);
 103         mutex_unlock(&root->fs_info->fs_mutex);
 104         return ret;
 105 }
 106
 107 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 108 {
 109         struct inode *inode = page->mapping->host;
 110         struct btrfs_root *root = BTRFS_I(inode)->root;
 111         struct btrfs_trans_handle *trans;
 112         char *kaddr;
 113         int ret;
 114         u64 page_start = page->index << PAGE_CACHE_SHIFT;
 115         size_t offset = start - page_start;
 116
 117         mutex_lock(&root->fs_info->fs_mutex);
 118         trans = btrfs_start_transaction(root, 1);
 119         btrfs_set_trans_block_group(trans, inode);
 120         kaddr = kmap(page);
 121         btrfs_csum_file_block(trans, root, inode, 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         u32 csum;
 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                 csum = 0;
 149                 goto out;
 150         }
 151         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 152                            BTRFS_CRC32_SIZE);
 153         set_state_private(em_tree, start, csum);
 154 out:
 155         if (path)
 156                 btrfs_free_path(path);
 157         mutex_unlock(&root->fs_info->fs_mutex);
 158         return ret;
 159 }
 160
 161 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end)
 162 {
 163         size_t offset = start - (page->index << PAGE_CACHE_SHIFT);
 164         struct inode *inode = page->mapping->host;
 165         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 166         char *kaddr;
 167         u64 private;
 168         int ret;
 169         struct btrfs_root *root = BTRFS_I(inode)->root;
 170         u32 csum = ~(u32)0;
 171         unsigned long flags;
 172
 173         ret = get_state_private(em_tree, start, &private);
 174         local_irq_save(flags);
 175         kaddr = kmap_atomic(page, KM_IRQ0);
 176         if (ret) {
 177                 goto zeroit;
 178         }
 179         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 180         btrfs_csum_final(csum, (char *)&csum);
 181         if (csum != private) {
 182                 goto zeroit;
 183         }
 184         kunmap_atomic(kaddr, KM_IRQ0);
 185         local_irq_restore(flags);
 186         return 0;
 187
 188 zeroit:
 189         printk("btrfs csum failed ino %lu off %llu\n",
 190                page->mapping->host->i_ino, (unsigned long long)start);
 191         memset(kaddr + offset, 1, end - start + 1);
 192         flush_dcache_page(page);
 193         kunmap_atomic(kaddr, KM_IRQ0);
 194         local_irq_restore(flags);
 195         return 0;
 196 }
 197
 198 void btrfs_read_locked_inode(struct inode *inode)
 199 {
 200         struct btrfs_path *path;
 201         struct extent_buffer *leaf;
 202         struct btrfs_inode_item *inode_item;
 203         struct btrfs_inode_timespec *tspec;
 204         struct btrfs_root *root = BTRFS_I(inode)->root;
 205         struct btrfs_key location;
 206         u64 alloc_group_block;
 207         u32 rdev;
 208         int ret;
 209
 210         path = btrfs_alloc_path();
 211         BUG_ON(!path);
 212         mutex_lock(&root->fs_info->fs_mutex);
 213
 214         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 215         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 216         if (ret)
 217                 goto make_bad;
 218
 219         leaf = path->nodes[0];
 220         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 221                                     struct btrfs_inode_item);
 222
 223         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 224         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 225         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 226         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 227         inode->i_size = btrfs_inode_size(leaf, inode_item);
 228
 229         tspec = btrfs_inode_atime(inode_item);
 230         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 231         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 232
 233         tspec = btrfs_inode_mtime(inode_item);
 234         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 235         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 236
 237         tspec = btrfs_inode_ctime(inode_item);
 238         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 239         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 240
 241         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 242         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 243         inode->i_rdev = 0;
 244         rdev = btrfs_inode_rdev(leaf, inode_item);
 245
 246         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 247         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 248                                                        alloc_group_block);
 249
 250         btrfs_free_path(path);
 251         inode_item = NULL;
 252
 253         mutex_unlock(&root->fs_info->fs_mutex);
 254
 255         switch (inode->i_mode & S_IFMT) {
 256         case S_IFREG:
 257                 inode->i_mapping->a_ops = &btrfs_aops;
 258                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
 259                 inode->i_fop = &btrfs_file_operations;
 260                 inode->i_op = &btrfs_file_inode_operations;
 261                 break;
 262         case S_IFDIR:
 263                 inode->i_fop = &btrfs_dir_file_operations;
 264                 if (root == root->fs_info->tree_root)
 265                         inode->i_op = &btrfs_dir_ro_inode_operations;
 266                 else
 267                         inode->i_op = &btrfs_dir_inode_operations;
 268                 break;
 269         case S_IFLNK:
 270                 inode->i_op = &btrfs_symlink_inode_operations;
 271                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 272                 break;
 273         default:
 274                 init_special_inode(inode, inode->i_mode, rdev);
 275                 break;
 276         }
 277         return;
 278
 279 make_bad:
 280         btrfs_release_path(root, path);
 281         btrfs_free_path(path);
 282         mutex_unlock(&root->fs_info->fs_mutex);
 283         make_bad_inode(inode);
 284 }
 285
 286 static void fill_inode_item(struct extent_buffer *leaf,
 287                             struct btrfs_inode_item *item,
 288                             struct inode *inode)
 289 {
 290         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 291         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 292         btrfs_set_inode_size(leaf, item, inode->i_size);
 293         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 294         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 295
 296         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 297                                inode->i_atime.tv_sec);
 298         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 299                                 inode->i_atime.tv_nsec);
 300
 301         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 302                                inode->i_mtime.tv_sec);
 303         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 304                                 inode->i_mtime.tv_nsec);
 305
 306         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 307                                inode->i_ctime.tv_sec);
 308         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 309                                 inode->i_ctime.tv_nsec);
 310
 311         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 312         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 313         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 314         btrfs_set_inode_block_group(leaf, item,
 315                                     BTRFS_I(inode)->block_group->key.objectid);
 316 }
 317
 318 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 319                               struct btrfs_root *root,
 320                               struct inode *inode)
 321 {
 322         struct btrfs_inode_item *inode_item;
 323         struct btrfs_path *path;
 324         struct extent_buffer *leaf;
 325         int ret;
 326
 327         path = btrfs_alloc_path();
 328         BUG_ON(!path);
 329         ret = btrfs_lookup_inode(trans, root, path,
 330                                  &BTRFS_I(inode)->location, 1);
 331         if (ret) {
 332                 if (ret > 0)
 333                         ret = -ENOENT;
 334                 goto failed;
 335         }
 336
 337         leaf = path->nodes[0];
 338         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 339                                   struct btrfs_inode_item);
 340
 341         fill_inode_item(leaf, inode_item, inode);
 342         btrfs_mark_buffer_dirty(leaf);
 343         btrfs_set_inode_last_trans(trans, inode);
 344         ret = 0;
 345 failed:
 346         btrfs_release_path(root, path);
 347         btrfs_free_path(path);
 348         return ret;
 349 }
 350
 351
 352 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 353                               struct btrfs_root *root,
 354                               struct inode *dir,
 355                               struct dentry *dentry)
 356 {
 357         struct btrfs_path *path;
 358         const char *name = dentry->d_name.name;
 359         int name_len = dentry->d_name.len;
 360         int ret = 0;
 361         struct extent_buffer *leaf;
 362         struct btrfs_dir_item *di;
 363         struct btrfs_key key;
 364
 365         path = btrfs_alloc_path();
 366         if (!path) {
 367                 ret = -ENOMEM;
 368                 goto err;
 369         }
 370
 371         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 372                                     name, name_len, -1);
 373         if (IS_ERR(di)) {
 374                 ret = PTR_ERR(di);
 375                 goto err;
 376         }
 377         if (!di) {
 378                 ret = -ENOENT;
 379                 goto err;
 380         }
 381         leaf = path->nodes[0];
 382         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 383         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 384         if (ret)
 385                 goto err;
 386         btrfs_release_path(root, path);
 387
 388         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 389                                          key.objectid, name, name_len, -1);
 390         if (IS_ERR(di)) {
 391                 ret = PTR_ERR(di);
 392                 goto err;
 393         }
 394         if (!di) {
 395                 ret = -ENOENT;
 396                 goto err;
 397         }
 398         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 399
 400         dentry->d_inode->i_ctime = dir->i_ctime;
 401 err:
 402         btrfs_free_path(path);
 403         if (!ret) {
 404                 dir->i_size -= name_len * 2;
 405                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 406                 btrfs_update_inode(trans, root, dir);
 407                 drop_nlink(dentry->d_inode);
 408                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 409                 dir->i_sb->s_dirt = 1;
 410         }
 411         return ret;
 412 }
 413
 414 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 415 {
 416         struct btrfs_root *root;
 417         struct btrfs_trans_handle *trans;
 418         int ret;
 419         unsigned long nr;
 420
 421         root = BTRFS_I(dir)->root;
 422         mutex_lock(&root->fs_info->fs_mutex);
 423         trans = btrfs_start_transaction(root, 1);
 424
 425         btrfs_set_trans_block_group(trans, dir);
 426         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 427         nr = trans->blocks_used;
 428
 429         btrfs_end_transaction(trans, root);
 430         mutex_unlock(&root->fs_info->fs_mutex);
 431         btrfs_btree_balance_dirty(root, nr);
 432
 433         return ret;
 434 }
 435
 436 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 437 {
 438         struct inode *inode = dentry->d_inode;
 439         int err;
 440         int ret;
 441         struct btrfs_root *root = BTRFS_I(dir)->root;
 442         struct btrfs_path *path;
 443         struct btrfs_key key;
 444         struct btrfs_trans_handle *trans;
 445         struct btrfs_key found_key;
 446         int found_type;
 447         struct extent_buffer *leaf;
 448         char *goodnames = "..";
 449         unsigned long nr;
 450
 451         path = btrfs_alloc_path();
 452         BUG_ON(!path);
 453         mutex_lock(&root->fs_info->fs_mutex);
 454         trans = btrfs_start_transaction(root, 1);
 455
 456         btrfs_set_trans_block_group(trans, dir);
 457         key.objectid = inode->i_ino;
 458         key.offset = (u64)-1;
 459         key.type = (u8)-1;
 460         while(1) {
 461                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 462                 if (ret < 0) {
 463                         err = ret;
 464                         goto out;
 465                 }
 466                 BUG_ON(ret == 0);
 467                 if (path->slots[0] == 0) {
 468                         err = -ENOENT;
 469                         goto out;
 470                 }
 471                 path->slots[0]--;
 472                 leaf = path->nodes[0];
 473                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 474                 found_type = btrfs_key_type(&found_key);
 475                 if (found_key.objectid != inode->i_ino) {
 476                         err = -ENOENT;
 477                         goto out;
 478                 }
 479                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
 480                      found_type != BTRFS_DIR_INDEX_KEY) ||
 481                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
 482                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
 483                         err = -ENOTEMPTY;
 484                         goto out;
 485                 }
 486                 ret = btrfs_del_item(trans, root, path);
 487                 BUG_ON(ret);
 488
 489                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
 490                         break;
 491                 btrfs_release_path(root, path);
 492         }
 493         ret = 0;
 494         btrfs_release_path(root, path);
 495
 496         /* now the directory is empty */
 497         err = btrfs_unlink_trans(trans, root, dir, dentry);
 498         if (!err) {
 499                 inode->i_size = 0;
 500         }
 501 out:
 502         btrfs_release_path(root, path);
 503         btrfs_free_path(path);
 504         mutex_unlock(&root->fs_info->fs_mutex);
 505         nr = trans->blocks_used;
 506         ret = btrfs_end_transaction(trans, root);
 507         btrfs_btree_balance_dirty(root, nr);
 508         if (ret && !err)
 509                 err = ret;
 510         return err;
 511 }
 512
 513 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 514                             struct btrfs_root *root,
 515                             struct inode *inode)
 516 {
 517         struct btrfs_path *path;
 518         int ret;
 519
 520         clear_inode(inode);
 521
 522         path = btrfs_alloc_path();
 523         BUG_ON(!path);
 524         ret = btrfs_lookup_inode(trans, root, path,
 525                                  &BTRFS_I(inode)->location, -1);
 526         if (ret > 0)
 527                 ret = -ENOENT;
 528         if (!ret)
 529                 ret = btrfs_del_item(trans, root, path);
 530         btrfs_free_path(path);
 531         return ret;
 532 }
 533
 534 /*
 535  * this can truncate away extent items, csum items and directory items.
 536  * It starts at a high offset and removes keys until it can't find
 537  * any higher than i_size.
 538  *
 539  * csum items that cross the new i_size are truncated to the new size
 540  * as well.
 541  */
 542 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 543                                    struct btrfs_root *root,
 544                                    struct inode *inode)
 545 {
 546         int ret;
 547         struct btrfs_path *path;
 548         struct btrfs_key key;
 549         struct btrfs_key found_key;
 550         u32 found_type;
 551         struct extent_buffer *leaf;
 552         struct btrfs_file_extent_item *fi;
 553         u64 extent_start = 0;
 554         u64 extent_num_bytes = 0;
 555         u64 item_end = 0;
 556         int found_extent;
 557         int del_item;
 558
 559         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 560         path = btrfs_alloc_path();
 561         path->reada = -1;
 562         BUG_ON(!path);
 563
 564         /* FIXME, add redo link to tree so we don't leak on crash */
 565         key.objectid = inode->i_ino;
 566         key.offset = (u64)-1;
 567         key.type = (u8)-1;
 568
 569         while(1) {
 570                 btrfs_init_path(path);
 571                 fi = NULL;
 572                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 573                 if (ret < 0) {
 574                         goto error;
 575                 }
 576                 if (ret > 0) {
 577                         BUG_ON(path->slots[0] == 0);
 578                         path->slots[0]--;
 579                 }
 580                 leaf = path->nodes[0];
 581                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 582                 found_type = btrfs_key_type(&found_key);
 583
 584                 if (found_key.objectid != inode->i_ino)
 585                         break;
 586
 587                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 588                     found_type != BTRFS_DIR_ITEM_KEY &&
 589                     found_type != BTRFS_DIR_INDEX_KEY &&
 590                     found_type != BTRFS_EXTENT_DATA_KEY)
 591                         break;
 592
 593                 item_end = found_key.offset;
 594                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 595                         fi = btrfs_item_ptr(leaf, path->slots[0],
 596                                             struct btrfs_file_extent_item);
 597                         if (btrfs_file_extent_type(leaf, fi) !=
 598                             BTRFS_FILE_EXTENT_INLINE) {
 599                                 item_end +=
 600                                     btrfs_file_extent_num_bytes(leaf, fi);
 601                         }
 602                 }
 603                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 604                         ret = btrfs_csum_truncate(trans, root, path,
 605                                                   inode->i_size);
 606                         BUG_ON(ret);
 607                 }
 608                 if (item_end < inode->i_size) {
 609                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 610                                 found_type = BTRFS_INODE_ITEM_KEY;
 611                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 612                                 found_type = BTRFS_CSUM_ITEM_KEY;
 613                         } else if (found_type) {
 614                                 found_type--;
 615                         } else {
 616                                 break;
 617                         }
 618                         btrfs_set_key_type(&key, found_type);
 619                         btrfs_release_path(root, path);
 620                         continue;
 621                 }
 622                 if (found_key.offset >= inode->i_size)
 623                         del_item = 1;
 624                 else
 625                         del_item = 0;
 626                 found_extent = 0;
 627
 628                 /* FIXME, shrink the extent if the ref count is only 1 */
 629                 if (found_type == BTRFS_EXTENT_DATA_KEY &&
 630                            btrfs_file_extent_type(leaf, fi) !=
 631                            BTRFS_FILE_EXTENT_INLINE) {
 632                         u64 num_dec;
 633                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 634                         if (!del_item) {
 635                                 u64 orig_num_bytes =
 636                                         btrfs_file_extent_num_bytes(leaf, fi);
 637                                 extent_num_bytes = inode->i_size -
 638                                         found_key.offset + root->sectorsize - 1;
 639                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 640                                                          extent_num_bytes);
 641                                 num_dec = (orig_num_bytes -
 642                                            extent_num_bytes) >> 9;
 643                                 if (extent_start != 0) {
 644                                         inode->i_blocks -= num_dec;
 645                                 }
 646                                 btrfs_mark_buffer_dirty(leaf);
 647                         } else {
 648                                 extent_num_bytes =
 649                                         btrfs_file_extent_disk_num_bytes(leaf,
 650                                                                          fi);
 651                                 /* FIXME blocksize != 4096 */
 652                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 653                                                                        fi) >> 9;
 654                                 if (extent_start != 0) {
 655                                         found_extent = 1;
 656                                         inode->i_blocks -= num_dec;
 657                                 }
 658                         }
 659                 }
 660                 if (del_item) {
 661                         ret = btrfs_del_item(trans, root, path);
 662                         if (ret)
 663                                 goto error;
 664                 } else {
 665                         break;
 666                 }
 667                 btrfs_release_path(root, path);
 668                 if (found_extent) {
 669                         ret = btrfs_free_extent(trans, root, extent_start,
 670                                                 extent_num_bytes, 0);
 671                         BUG_ON(ret);
 672                 }
 673         }
 674         ret = 0;
 675 error:
 676         btrfs_release_path(root, path);
 677         btrfs_free_path(path);
 678         inode->i_sb->s_dirt = 1;
 679         return ret;
 680 }
 681
 682 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 683                               size_t zero_start)
 684 {
 685         char *kaddr;
 686         int ret = 0;
 687         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 688         u64 page_start = page->index << PAGE_CACHE_SHIFT;
 689         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 690
 691         set_page_extent_mapped(page);
 692
 693         lock_extent(em_tree, page_start, page_end, GFP_NOFS);
 694         set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
 695                             page_end, GFP_NOFS);
 696         if (zero_start != PAGE_CACHE_SIZE) {
 697                 kaddr = kmap(page);
 698                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 699                 flush_dcache_page(page);
 700                 kunmap(page);
 701         }
 702         set_page_dirty(page);
 703         unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
 704
 705         return ret;
 706 }
 707
 708 /*
 709  * taken from block_truncate_page, but does cow as it zeros out
 710  * any bytes left in the last page in the file.
 711  */
 712 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 713 {
 714         struct inode *inode = mapping->host;
 715         struct btrfs_root *root = BTRFS_I(inode)->root;
 716         u32 blocksize = root->sectorsize;
 717         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 718         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 719         struct page *page;
 720         int ret = 0;
 721         u64 page_start;
 722
 723         if ((offset & (blocksize - 1)) == 0)
 724                 goto out;
 725
 726         down_read(&root->snap_sem);
 727         ret = -ENOMEM;
 728         page = grab_cache_page(mapping, index);
 729         if (!page)
 730                 goto out;
 731         if (!PageUptodate(page)) {
 732                 ret = btrfs_readpage(NULL, page);
 733                 lock_page(page);
 734                 if (!PageUptodate(page)) {
 735                         ret = -EIO;
 736                         goto out;
 737                 }
 738         }
 739         page_start = page->index << PAGE_CACHE_SHIFT;
 740
 741         ret = btrfs_cow_one_page(inode, page, offset);
 742
 743         unlock_page(page);
 744         page_cache_release(page);
 745         up_read(&BTRFS_I(inode)->root->snap_sem);
 746 out:
 747         return ret;
 748 }
 749
 750 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 751 {
 752         struct inode *inode = dentry->d_inode;
 753         int err;
 754
 755         err = inode_change_ok(inode, attr);
 756         if (err)
 757                 return err;
 758
 759         if (S_ISREG(inode->i_mode) &&
 760             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 761                 struct btrfs_trans_handle *trans;
 762                 struct btrfs_root *root = BTRFS_I(inode)->root;
 763                 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 764
 765                 u64 mask = root->sectorsize - 1;
 766                 u64 pos = (inode->i_size + mask) & ~mask;
 767                 u64 block_end = attr->ia_size | mask;
 768                 u64 hole_size;
 769                 u64 alloc_hint;
 770
 771                 if (attr->ia_size <= pos)
 772                         goto out;
 773
 774                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 775
 776                 lock_extent(em_tree, pos, block_end, GFP_NOFS);
 777                 hole_size = (attr->ia_size - pos + mask) & ~mask;
 778
 779                 mutex_lock(&root->fs_info->fs_mutex);
 780                 trans = btrfs_start_transaction(root, 1);
 781                 btrfs_set_trans_block_group(trans, inode);
 782                 err = btrfs_drop_extents(trans, root, inode,
 783                                          pos, pos + hole_size, pos,
 784                                          &alloc_hint);
 785
 786                 err = btrfs_insert_file_extent(trans, root, inode->i_ino,
 787                                                pos, 0, 0, hole_size);
 788                 btrfs_end_transaction(trans, root);
 789                 mutex_unlock(&root->fs_info->fs_mutex);
 790                 unlock_extent(em_tree, pos, block_end, GFP_NOFS);
 791                 if (err)
 792                         return err;
 793         }
 794 out:
 795         err = inode_setattr(inode, attr);
 796
 797         return err;
 798 }
 799 void btrfs_delete_inode(struct inode *inode)
 800 {
 801         struct btrfs_trans_handle *trans;
 802         struct btrfs_root *root = BTRFS_I(inode)->root;
 803         unsigned long nr;
 804         int ret;
 805
 806         truncate_inode_pages(&inode->i_data, 0);
 807         if (is_bad_inode(inode)) {
 808                 goto no_delete;
 809         }
 810
 811         inode->i_size = 0;
 812         mutex_lock(&root->fs_info->fs_mutex);
 813         trans = btrfs_start_transaction(root, 1);
 814
 815         btrfs_set_trans_block_group(trans, inode);
 816         ret = btrfs_truncate_in_trans(trans, root, inode);
 817         if (ret)
 818                 goto no_delete_lock;
 819         ret = btrfs_free_inode(trans, root, inode);
 820         if (ret)
 821                 goto no_delete_lock;
 822         nr = trans->blocks_used;
 823
 824         btrfs_end_transaction(trans, root);
 825         mutex_unlock(&root->fs_info->fs_mutex);
 826         btrfs_btree_balance_dirty(root, nr);
 827         return;
 828
 829 no_delete_lock:
 830         nr = trans->blocks_used;
 831         btrfs_end_transaction(trans, root);
 832         mutex_unlock(&root->fs_info->fs_mutex);
 833         btrfs_btree_balance_dirty(root, nr);
 834 no_delete:
 835         clear_inode(inode);
 836 }
 837
 838 /*
 839  * this returns the key found in the dir entry in the location pointer.
 840  * If no dir entries were found, location->objectid is 0.
 841  */
 842 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 843                                struct btrfs_key *location)
 844 {
 845         const char *name = dentry->d_name.name;
 846         int namelen = dentry->d_name.len;
 847         struct btrfs_dir_item *di;
 848         struct btrfs_path *path;
 849         struct btrfs_root *root = BTRFS_I(dir)->root;
 850         int ret = 0;
 851
 852         path = btrfs_alloc_path();
 853         BUG_ON(!path);
 854         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
 855                                     namelen, 0);
 856         if (IS_ERR(di))
 857                 ret = PTR_ERR(di);
 858         if (!di || IS_ERR(di)) {
 859                 location->objectid = 0;
 860                 goto out;
 861         }
 862         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
 863 out:
 864         btrfs_release_path(root, path);
 865         btrfs_free_path(path);
 866         return ret;
 867 }
 868
 869 /*
 870  * when we hit a tree root in a directory, the btrfs part of the inode
 871  * needs to be changed to reflect the root directory of the tree root.  This
 872  * is kind of like crossing a mount point.
 873  */
 874 static int fixup_tree_root_location(struct btrfs_root *root,
 875                              struct btrfs_key *location,
 876                              struct btrfs_root **sub_root,
 877                              struct dentry *dentry)
 878 {
 879         struct btrfs_path *path;
 880         struct btrfs_root_item *ri;
 881
 882         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 883                 return 0;
 884         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 885                 return 0;
 886
 887         path = btrfs_alloc_path();
 888         BUG_ON(!path);
 889         mutex_lock(&root->fs_info->fs_mutex);
 890
 891         *sub_root = btrfs_read_fs_root(root->fs_info, location,
 892                                         dentry->d_name.name,
 893                                         dentry->d_name.len);
 894         if (IS_ERR(*sub_root))
 895                 return PTR_ERR(*sub_root);
 896
 897         ri = &(*sub_root)->root_item;
 898         location->objectid = btrfs_root_dirid(ri);
 899         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 900         location->offset = 0;
 901
 902         btrfs_free_path(path);
 903         mutex_unlock(&root->fs_info->fs_mutex);
 904         return 0;
 905 }
 906
 907 static int btrfs_init_locked_inode(struct inode *inode, void *p)
 908 {
 909         struct btrfs_iget_args *args = p;
 910         inode->i_ino = args->ino;
 911         BTRFS_I(inode)->root = args->root;
 912         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
 913                              inode->i_mapping, GFP_NOFS);
 914         return 0;
 915 }
 916
 917 static int btrfs_find_actor(struct inode *inode, void *opaque)
 918 {
 919         struct btrfs_iget_args *args = opaque;
 920         return (args->ino == inode->i_ino &&
 921                 args->root == BTRFS_I(inode)->root);
 922 }
 923
 924 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 925                                 struct btrfs_root *root)
 926 {
 927         struct inode *inode;
 928         struct btrfs_iget_args args;
 929         args.ino = objectid;
 930         args.root = root;
 931
 932         inode = iget5_locked(s, objectid, btrfs_find_actor,
 933                              btrfs_init_locked_inode,
 934                              (void *)&args);
 935         return inode;
 936 }
 937
 938 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 939                                    struct nameidata *nd)
 940 {
 941         struct inode * inode;
 942         struct btrfs_inode *bi = BTRFS_I(dir);
 943         struct btrfs_root *root = bi->root;
 944         struct btrfs_root *sub_root = root;
 945         struct btrfs_key location;
 946         int ret;
 947
 948         if (dentry->d_name.len > BTRFS_NAME_LEN)
 949                 return ERR_PTR(-ENAMETOOLONG);
 950
 951         mutex_lock(&root->fs_info->fs_mutex);
 952         ret = btrfs_inode_by_name(dir, dentry, &location);
 953         mutex_unlock(&root->fs_info->fs_mutex);
 954
 955         if (ret < 0)
 956                 return ERR_PTR(ret);
 957
 958         inode = NULL;
 959         if (location.objectid) {
 960                 ret = fixup_tree_root_location(root, &location, &sub_root,
 961                                                 dentry);
 962                 if (ret < 0)
 963                         return ERR_PTR(ret);
 964                 if (ret > 0)
 965                         return ERR_PTR(-ENOENT);
 966                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 967                                           sub_root);
 968                 if (!inode)
 969                         return ERR_PTR(-EACCES);
 970                 if (inode->i_state & I_NEW) {
 971                         /* the inode and parent dir are two different roots */
 972                         if (sub_root != root) {
 973                                 igrab(inode);
 974                                 sub_root->inode = inode;
 975                         }
 976                         BTRFS_I(inode)->root = sub_root;
 977                         memcpy(&BTRFS_I(inode)->location, &location,
 978                                sizeof(location));
 979                         btrfs_read_locked_inode(inode);
 980                         unlock_new_inode(inode);
 981                 }
 982         }
 983         return d_splice_alias(inode, dentry);
 984 }
 985
 986 static unsigned char btrfs_filetype_table[] = {
 987         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
 988 };
 989
 990 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 991 {
 992         struct inode *inode = filp->f_path.dentry->d_inode;
 993         struct btrfs_root *root = BTRFS_I(inode)->root;
 994         struct btrfs_item *item;
 995         struct btrfs_dir_item *di;
 996         struct btrfs_key key;
 997         struct btrfs_key found_key;
 998         struct btrfs_path *path;
 999         int ret;
1000         u32 nritems;
1001         struct extent_buffer *leaf;
1002         int slot;
1003         int advance;
1004         unsigned char d_type;
1005         int over = 0;
1006         u32 di_cur;
1007         u32 di_total;
1008         u32 di_len;
1009         int key_type = BTRFS_DIR_INDEX_KEY;
1010         char tmp_name[32];
1011         char *name_ptr;
1012         int name_len;
1013
1014         /* FIXME, use a real flag for deciding about the key type */
1015         if (root->fs_info->tree_root == root)
1016                 key_type = BTRFS_DIR_ITEM_KEY;
1017
1018         mutex_lock(&root->fs_info->fs_mutex);
1019         key.objectid = inode->i_ino;
1020         btrfs_set_key_type(&key, key_type);
1021         key.offset = filp->f_pos;
1022
1023         path = btrfs_alloc_path();
1024         path->reada = 2;
1025         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1026         if (ret < 0)
1027                 goto err;
1028         advance = 0;
1029         while(1) {
1030                 leaf = path->nodes[0];
1031                 nritems = btrfs_header_nritems(leaf);
1032                 slot = path->slots[0];
1033                 if (advance || slot >= nritems) {
1034                         if (slot >= nritems -1) {
1035                                 ret = btrfs_next_leaf(root, path);
1036                                 if (ret)
1037                                         break;
1038                                 leaf = path->nodes[0];
1039                                 nritems = btrfs_header_nritems(leaf);
1040                                 slot = path->slots[0];
1041                         } else {
1042                                 slot++;
1043                                 path->slots[0]++;
1044                         }
1045                 }
1046                 advance = 1;
1047                 item = btrfs_item_nr(leaf, slot);
1048                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1049
1050                 if (found_key.objectid != key.objectid)
1051                         break;
1052                 if (btrfs_key_type(&found_key) != key_type)
1053                         break;
1054                 if (found_key.offset < filp->f_pos)
1055                         continue;
1056
1057                 filp->f_pos = found_key.offset;
1058                 advance = 1;
1059                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1060                 di_cur = 0;
1061                 di_total = btrfs_item_size(leaf, item);
1062                 while(di_cur < di_total) {
1063                         struct btrfs_key location;
1064
1065                         name_len = btrfs_dir_name_len(leaf, di);
1066                         if (name_len < 32) {
1067                                 name_ptr = tmp_name;
1068                         } else {
1069                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1070                                 BUG_ON(!name_ptr);
1071                         }
1072                         read_extent_buffer(leaf, name_ptr,
1073                                            (unsigned long)(di + 1), name_len);
1074
1075                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1076                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1077
1078                         over = filldir(dirent, name_ptr, name_len,
1079                                        found_key.offset,
1080                                        location.objectid,
1081                                        d_type);
1082
1083                         if (name_ptr != tmp_name)
1084                                 kfree(name_ptr);
1085
1086                         if (over)
1087                                 goto nopos;
1088                         di_len = btrfs_dir_name_len(leaf, di) + sizeof(*di);
1089                         di_cur += di_len;
1090                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1091                 }
1092         }
1093         filp->f_pos++;
1094 nopos:
1095         ret = 0;
1096 err:
1097         btrfs_release_path(root, path);
1098         btrfs_free_path(path);
1099         mutex_unlock(&root->fs_info->fs_mutex);
1100         return ret;
1101 }
1102
1103 int btrfs_write_inode(struct inode *inode, int wait)
1104 {
1105         struct btrfs_root *root = BTRFS_I(inode)->root;
1106         struct btrfs_trans_handle *trans;
1107         int ret = 0;
1108
1109         if (wait) {
1110                 mutex_lock(&root->fs_info->fs_mutex);
1111                 trans = btrfs_start_transaction(root, 1);
1112                 btrfs_set_trans_block_group(trans, inode);
1113                 ret = btrfs_commit_transaction(trans, root);
1114                 mutex_unlock(&root->fs_info->fs_mutex);
1115         }
1116         return ret;
1117 }
1118
1119 /*
1120  * This is somewhat expensive, updating the tree every time the
1121  * inode changes.  But, it is most likely to find the inode in cache.
1122  * FIXME, needs more benchmarking...there are no reasons other than performance
1123  * to keep or drop this code.
1124  */
1125 void btrfs_dirty_inode(struct inode *inode)
1126 {
1127         struct btrfs_root *root = BTRFS_I(inode)->root;
1128         struct btrfs_trans_handle *trans;
1129
1130         mutex_lock(&root->fs_info->fs_mutex);
1131         trans = btrfs_start_transaction(root, 1);
1132         btrfs_set_trans_block_group(trans, inode);
1133         btrfs_update_inode(trans, root, inode);
1134         btrfs_end_transaction(trans, root);
1135         mutex_unlock(&root->fs_info->fs_mutex);
1136 }
1137
1138 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1139                                      struct btrfs_root *root,
1140                                      u64 objectid,
1141                                      struct btrfs_block_group_cache *group,
1142                                      int mode)
1143 {
1144         struct inode *inode;
1145         struct btrfs_inode_item *inode_item;
1146         struct btrfs_key *location;
1147         struct btrfs_path *path;
1148         int ret;
1149         int owner;
1150
1151         path = btrfs_alloc_path();
1152         BUG_ON(!path);
1153
1154         inode = new_inode(root->fs_info->sb);
1155         if (!inode)
1156                 return ERR_PTR(-ENOMEM);
1157
1158         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1159                              inode->i_mapping, GFP_NOFS);
1160         BTRFS_I(inode)->root = root;
1161
1162         if (mode & S_IFDIR)
1163                 owner = 0;
1164         else
1165                 owner = 1;
1166         group = btrfs_find_block_group(root, group, 0, 0, owner);
1167         BTRFS_I(inode)->block_group = group;
1168
1169         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1170         if (ret)
1171                 goto fail;
1172
1173         inode->i_uid = current->fsuid;
1174         inode->i_gid = current->fsgid;
1175         inode->i_mode = mode;
1176         inode->i_ino = objectid;
1177         inode->i_blocks = 0;
1178         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1179         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1180                                   struct btrfs_inode_item);
1181         fill_inode_item(path->nodes[0], inode_item, inode);
1182         btrfs_mark_buffer_dirty(path->nodes[0]);
1183         btrfs_free_path(path);
1184
1185         location = &BTRFS_I(inode)->location;
1186         location->objectid = objectid;
1187         location->offset = 0;
1188         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1189
1190         insert_inode_hash(inode);
1191         return inode;
1192 fail:
1193         btrfs_free_path(path);
1194         return ERR_PTR(ret);
1195 }
1196
1197 static inline u8 btrfs_inode_type(struct inode *inode)
1198 {
1199         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1200 }
1201
1202 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1203                             struct dentry *dentry, struct inode *inode)
1204 {
1205         int ret;
1206         struct btrfs_key key;
1207         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1208         struct inode *parent_inode;
1209
1210         key.objectid = inode->i_ino;
1211         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1212         key.offset = 0;
1213
1214         ret = btrfs_insert_dir_item(trans, root,
1215                                     dentry->d_name.name, dentry->d_name.len,
1216                                     dentry->d_parent->d_inode->i_ino,
1217                                     &key, btrfs_inode_type(inode));
1218         if (ret == 0) {
1219                 parent_inode = dentry->d_parent->d_inode;
1220                 parent_inode->i_size += dentry->d_name.len * 2;
1221                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1222                 ret = btrfs_update_inode(trans, root,
1223                                          dentry->d_parent->d_inode);
1224         }
1225         return ret;
1226 }
1227
1228 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1229                             struct dentry *dentry, struct inode *inode)
1230 {
1231         int err = btrfs_add_link(trans, dentry, inode);
1232         if (!err) {
1233                 d_instantiate(dentry, inode);
1234                 return 0;
1235         }
1236         if (err > 0)
1237                 err = -EEXIST;
1238         return err;
1239 }
1240
1241 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1242                         int mode, dev_t rdev)
1243 {
1244         struct btrfs_trans_handle *trans;
1245         struct btrfs_root *root = BTRFS_I(dir)->root;
1246         struct inode *inode;
1247         int err;
1248         int drop_inode = 0;
1249         u64 objectid;
1250         unsigned long nr;
1251
1252         if (!new_valid_dev(rdev))
1253                 return -EINVAL;
1254
1255         mutex_lock(&root->fs_info->fs_mutex);
1256         trans = btrfs_start_transaction(root, 1);
1257         btrfs_set_trans_block_group(trans, dir);
1258
1259         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1260         if (err) {
1261                 err = -ENOSPC;
1262                 goto out_unlock;
1263         }
1264
1265         inode = btrfs_new_inode(trans, root, objectid,
1266                                 BTRFS_I(dir)->block_group, mode);
1267         err = PTR_ERR(inode);
1268         if (IS_ERR(inode))
1269                 goto out_unlock;
1270
1271         btrfs_set_trans_block_group(trans, inode);
1272         err = btrfs_add_nondir(trans, dentry, inode);
1273         if (err)
1274                 drop_inode = 1;
1275         else {
1276                 inode->i_op = &btrfs_special_inode_operations;
1277                 init_special_inode(inode, inode->i_mode, rdev);
1278                 btrfs_update_inode(trans, root, inode);
1279         }
1280         dir->i_sb->s_dirt = 1;
1281         btrfs_update_inode_block_group(trans, inode);
1282         btrfs_update_inode_block_group(trans, dir);
1283 out_unlock:
1284         nr = trans->blocks_used;
1285         btrfs_end_transaction(trans, root);
1286         mutex_unlock(&root->fs_info->fs_mutex);
1287
1288         if (drop_inode) {
1289                 inode_dec_link_count(inode);
1290                 iput(inode);
1291         }
1292         btrfs_btree_balance_dirty(root, nr);
1293         return err;
1294 }
1295
1296 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1297                         int mode, struct nameidata *nd)
1298 {
1299         struct btrfs_trans_handle *trans;
1300         struct btrfs_root *root = BTRFS_I(dir)->root;
1301         struct inode *inode;
1302         int err;
1303         int drop_inode = 0;
1304         unsigned long nr;
1305         u64 objectid;
1306
1307         mutex_lock(&root->fs_info->fs_mutex);
1308         trans = btrfs_start_transaction(root, 1);
1309         btrfs_set_trans_block_group(trans, dir);
1310
1311         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1312         if (err) {
1313                 err = -ENOSPC;
1314                 goto out_unlock;
1315         }
1316
1317         inode = btrfs_new_inode(trans, root, objectid,
1318                                 BTRFS_I(dir)->block_group, mode);
1319         err = PTR_ERR(inode);
1320         if (IS_ERR(inode))
1321                 goto out_unlock;
1322
1323         btrfs_set_trans_block_group(trans, inode);
1324         err = btrfs_add_nondir(trans, dentry, inode);
1325         if (err)
1326                 drop_inode = 1;
1327         else {
1328                 inode->i_mapping->a_ops = &btrfs_aops;
1329                 inode->i_fop = &btrfs_file_operations;
1330                 inode->i_op = &btrfs_file_inode_operations;
1331                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1332                                      inode->i_mapping, GFP_NOFS);
1333                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1334         }
1335         dir->i_sb->s_dirt = 1;
1336         btrfs_update_inode_block_group(trans, inode);
1337         btrfs_update_inode_block_group(trans, dir);
1338 out_unlock:
1339         nr = trans->blocks_used;
1340         btrfs_end_transaction(trans, root);
1341         mutex_unlock(&root->fs_info->fs_mutex);
1342
1343         if (drop_inode) {
1344                 inode_dec_link_count(inode);
1345                 iput(inode);
1346         }
1347         btrfs_btree_balance_dirty(root, nr);
1348         return err;
1349 }
1350
1351 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1352                       struct dentry *dentry)
1353 {
1354         struct btrfs_trans_handle *trans;
1355         struct btrfs_root *root = BTRFS_I(dir)->root;
1356         struct inode *inode = old_dentry->d_inode;
1357         unsigned long nr;
1358         int err;
1359         int drop_inode = 0;
1360
1361         if (inode->i_nlink == 0)
1362                 return -ENOENT;
1363
1364         inc_nlink(inode);
1365         mutex_lock(&root->fs_info->fs_mutex);
1366         trans = btrfs_start_transaction(root, 1);
1367
1368         btrfs_set_trans_block_group(trans, dir);
1369         atomic_inc(&inode->i_count);
1370         err = btrfs_add_nondir(trans, dentry, inode);
1371
1372         if (err)
1373                 drop_inode = 1;
1374
1375         dir->i_sb->s_dirt = 1;
1376         btrfs_update_inode_block_group(trans, dir);
1377         err = btrfs_update_inode(trans, root, inode);
1378
1379         if (err)
1380                 drop_inode = 1;
1381
1382         nr = trans->blocks_used;
1383         btrfs_end_transaction(trans, root);
1384         mutex_unlock(&root->fs_info->fs_mutex);
1385
1386         if (drop_inode) {
1387                 inode_dec_link_count(inode);
1388                 iput(inode);
1389         }
1390         btrfs_btree_balance_dirty(root, nr);
1391         return err;
1392 }
1393
1394 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
1395                                 struct btrfs_root *root,
1396                                 u64 objectid, u64 dirid)
1397 {
1398         int ret;
1399         char buf[2];
1400         struct btrfs_key key;
1401
1402         buf[0] = '.';
1403         buf[1] = '.';
1404
1405         key.objectid = objectid;
1406         key.offset = 0;
1407         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1408
1409         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
1410                                     &key, BTRFS_FT_DIR);
1411         if (ret)
1412                 goto error;
1413
1414         key.objectid = dirid;
1415         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
1416                                     &key, BTRFS_FT_DIR);
1417         if (ret)
1418                 goto error;
1419 error:
1420         return ret;
1421 }
1422
1423 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1424 {
1425         struct inode *inode;
1426         struct btrfs_trans_handle *trans;
1427         struct btrfs_root *root = BTRFS_I(dir)->root;
1428         int err = 0;
1429         int drop_on_err = 0;
1430         u64 objectid;
1431         unsigned long nr = 1;
1432
1433         mutex_lock(&root->fs_info->fs_mutex);
1434         trans = btrfs_start_transaction(root, 1);
1435         btrfs_set_trans_block_group(trans, dir);
1436
1437         if (IS_ERR(trans)) {
1438                 err = PTR_ERR(trans);
1439                 goto out_unlock;
1440         }
1441
1442         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1443         if (err) {
1444                 err = -ENOSPC;
1445                 goto out_unlock;
1446         }
1447
1448         inode = btrfs_new_inode(trans, root, objectid,
1449                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1450         if (IS_ERR(inode)) {
1451                 err = PTR_ERR(inode);
1452                 goto out_fail;
1453         }
1454
1455         drop_on_err = 1;
1456         inode->i_op = &btrfs_dir_inode_operations;
1457         inode->i_fop = &btrfs_dir_file_operations;
1458         btrfs_set_trans_block_group(trans, inode);
1459
1460         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
1461         if (err)
1462                 goto out_fail;
1463
1464         inode->i_size = 6;
1465         err = btrfs_update_inode(trans, root, inode);
1466         if (err)
1467                 goto out_fail;
1468
1469         err = btrfs_add_link(trans, dentry, inode);
1470         if (err)
1471                 goto out_fail;
1472
1473         d_instantiate(dentry, inode);
1474         drop_on_err = 0;
1475         dir->i_sb->s_dirt = 1;
1476         btrfs_update_inode_block_group(trans, inode);
1477         btrfs_update_inode_block_group(trans, dir);
1478
1479 out_fail:
1480         nr = trans->blocks_used;
1481         btrfs_end_transaction(trans, root);
1482
1483 out_unlock:
1484         mutex_unlock(&root->fs_info->fs_mutex);
1485         if (drop_on_err)
1486                 iput(inode);
1487         btrfs_btree_balance_dirty(root, nr);
1488         return err;
1489 }
1490
1491 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1492                                     size_t page_offset, u64 start, u64 end,
1493                                     int create)
1494 {
1495         int ret;
1496         int err = 0;
1497         u64 bytenr;
1498         u64 extent_start = 0;
1499         u64 extent_end = 0;
1500         u64 objectid = inode->i_ino;
1501         u32 found_type;
1502         int failed_insert = 0;
1503         struct btrfs_path *path;
1504         struct btrfs_root *root = BTRFS_I(inode)->root;
1505         struct btrfs_file_extent_item *item;
1506         struct extent_buffer *leaf;
1507         struct btrfs_key found_key;
1508         struct extent_map *em = NULL;
1509         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1510         struct btrfs_trans_handle *trans = NULL;
1511
1512         path = btrfs_alloc_path();
1513         BUG_ON(!path);
1514         mutex_lock(&root->fs_info->fs_mutex);
1515
1516 again:
1517         em = lookup_extent_mapping(em_tree, start, end);
1518         if (em) {
1519                 goto out;
1520         }
1521         if (!em) {
1522                 em = alloc_extent_map(GFP_NOFS);
1523                 if (!em) {
1524                         err = -ENOMEM;
1525                         goto out;
1526                 }
1527                 em->start = EXTENT_MAP_HOLE;
1528                 em->end = EXTENT_MAP_HOLE;
1529         }
1530         em->bdev = inode->i_sb->s_bdev;
1531         ret = btrfs_lookup_file_extent(NULL, root, path,
1532                                        objectid, start, 0);
1533         if (ret < 0) {
1534                 err = ret;
1535                 goto out;
1536         }
1537
1538         if (ret != 0) {
1539                 if (path->slots[0] == 0)
1540                         goto not_found;
1541                 path->slots[0]--;
1542         }
1543
1544         leaf = path->nodes[0];
1545         item = btrfs_item_ptr(leaf, path->slots[0],
1546                               struct btrfs_file_extent_item);
1547         /* are we inside the extent that was found? */
1548         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1549         found_type = btrfs_key_type(&found_key);
1550         if (found_key.objectid != objectid ||
1551             found_type != BTRFS_EXTENT_DATA_KEY) {
1552                 goto not_found;
1553         }
1554
1555         found_type = btrfs_file_extent_type(leaf, item);
1556         extent_start = found_key.offset;
1557         if (found_type == BTRFS_FILE_EXTENT_REG) {
1558                 extent_end = extent_start +
1559                        btrfs_file_extent_num_bytes(leaf, item);
1560                 err = 0;
1561                 if (start < extent_start || start >= extent_end) {
1562                         em->start = start;
1563                         if (start < extent_start) {
1564                                 if (end < extent_start)
1565                                         goto not_found;
1566                                 em->end = extent_end - 1;
1567                         } else {
1568                                 em->end = end;
1569                         }
1570                         goto not_found_em;
1571                 }
1572                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1573                 if (bytenr == 0) {
1574                         em->start = extent_start;
1575                         em->end = extent_end - 1;
1576                         em->block_start = EXTENT_MAP_HOLE;
1577                         em->block_end = EXTENT_MAP_HOLE;
1578                         goto insert;
1579                 }
1580                 bytenr += btrfs_file_extent_offset(leaf, item);
1581                 em->block_start = bytenr;
1582                 em->block_end = em->block_start +
1583                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1584                 em->start = extent_start;
1585                 em->end = extent_end - 1;
1586                 goto insert;
1587         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1588                 unsigned long ptr;
1589                 char *map;
1590                 size_t size;
1591                 size_t extent_offset;
1592                 size_t copy_size;
1593
1594                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1595                                                     path->slots[0]));
1596
1597                 extent_end = (extent_start + size) |
1598                         ((u64)root->sectorsize - 1);
1599                 if (start < extent_start || start >= extent_end) {
1600                         em->start = start;
1601                         if (start < extent_start) {
1602                                 if (end < extent_start)
1603                                         goto not_found;
1604                                 em->end = extent_end;
1605                         } else {
1606                                 em->end = end;
1607                         }
1608                         goto not_found_em;
1609                 }
1610
1611                 extent_offset = (page->index << PAGE_CACHE_SHIFT) -
1612                         extent_start;
1613                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1614                 map = kmap(page);
1615                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1616                                 size - extent_offset);
1617
1618                 em->block_start = EXTENT_MAP_INLINE;
1619                 em->block_end = EXTENT_MAP_INLINE;
1620                 em->start = extent_start + extent_offset;
1621                 em->end = (em->start + copy_size -1) |
1622                         ((u64)root->sectorsize -1);
1623
1624                 if (!page) {
1625                         goto insert;
1626                 }
1627
1628                 read_extent_buffer(leaf, map + page_offset, ptr, copy_size);
1629                 /*
1630                 memset(map + page_offset + copy_size, 0,
1631                        PAGE_CACHE_SIZE - copy_size - page_offset);
1632                        */
1633                 flush_dcache_page(page);
1634                 kunmap(page);
1635                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1636                 goto insert;
1637         } else {
1638                 printk("unkknown found_type %d\n", found_type);
1639                 WARN_ON(1);
1640         }
1641 not_found:
1642         em->start = start;
1643         em->end = end;
1644 not_found_em:
1645         em->block_start = EXTENT_MAP_HOLE;
1646         em->block_end = EXTENT_MAP_HOLE;
1647 insert:
1648         btrfs_release_path(root, path);
1649         if (em->start > start || em->end < start) {
1650                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1651                 err = -EIO;
1652                 goto out;
1653         }
1654         ret = add_extent_mapping(em_tree, em);
1655         if (ret == -EEXIST) {
1656                 free_extent_map(em);
1657                 em = NULL;
1658                 failed_insert++;
1659                 if (failed_insert > 5) {
1660                         printk("failing to insert %Lu %Lu\n", start, end);
1661                         err = -EIO;
1662                         goto out;
1663                 }
1664                 goto again;
1665         }
1666         err = 0;
1667 out:
1668         btrfs_free_path(path);
1669         if (trans) {
1670                 ret = btrfs_end_transaction(trans, root);
1671                 if (!err)
1672                         err = ret;
1673         }
1674         mutex_unlock(&root->fs_info->fs_mutex);
1675         if (err) {
1676                 free_extent_map(em);
1677                 WARN_ON(1);
1678                 return ERR_PTR(err);
1679         }
1680         return em;
1681 }
1682
1683 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1684 {
1685         return extent_bmap(mapping, iblock, btrfs_get_extent);
1686 }
1687
1688 static int btrfs_prepare_write(struct file *file, struct page *page,
1689                                unsigned from, unsigned to)
1690 {
1691         return extent_prepare_write(&BTRFS_I(page->mapping->host)->extent_tree,
1692                                     page->mapping->host, page, from, to,
1693                                     btrfs_get_extent);
1694 }
1695
1696 int btrfs_readpage(struct file *file, struct page *page)
1697 {
1698         struct extent_map_tree *tree;
1699         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1700         return extent_read_full_page(tree, page, btrfs_get_extent);
1701 }
1702 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1703 {
1704         struct extent_map_tree *tree;
1705
1706
1707         if (current->flags & PF_MEMALLOC) {
1708                 redirty_page_for_writepage(wbc, page);
1709                 unlock_page(page);
1710                 return 0;
1711         }
1712         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1713         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
1714 }
1715
1716 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1717 {
1718         struct extent_map_tree *tree;
1719         int ret;
1720
1721         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1722         ret = try_release_extent_mapping(tree, page);
1723         if (ret == 1) {
1724                 ClearPagePrivate(page);
1725                 set_page_private(page, 0);
1726                 page_cache_release(page);
1727         }
1728         return ret;
1729 }
1730
1731 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
1732 {
1733         struct extent_map_tree *tree;
1734
1735         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1736         extent_invalidatepage(tree, page, offset);
1737         btrfs_releasepage(page, GFP_NOFS);
1738 }
1739
1740 /*
1741  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
1742  * called from a page fault handler when a page is first dirtied. Hence we must
1743  * be careful to check for EOF conditions here. We set the page up correctly
1744  * for a written page which means we get ENOSPC checking when writing into
1745  * holes and correct delalloc and unwritten extent mapping on filesystems that
1746  * support these features.
1747  *
1748  * We are not allowed to take the i_mutex here so we have to play games to
1749  * protect against truncate races as the page could now be beyond EOF.  Because
1750  * vmtruncate() writes the inode size before removing pages, once we have the
1751  * page lock we can determine safely if the page is beyond EOF. If it is not
1752  * beyond EOF, then the page is guaranteed safe against truncation until we
1753  * unlock the page.
1754  */
1755 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1756 {
1757         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1758         unsigned long end;
1759         loff_t size;
1760         int ret = -EINVAL;
1761         u64 page_start;
1762
1763         down_read(&BTRFS_I(inode)->root->snap_sem);
1764         lock_page(page);
1765         wait_on_page_writeback(page);
1766         size = i_size_read(inode);
1767         page_start = page->index << PAGE_CACHE_SHIFT;
1768
1769         if ((page->mapping != inode->i_mapping) ||
1770             (page_start > size)) {
1771                 /* page got truncated out from underneath us */
1772                 goto out_unlock;
1773         }
1774
1775         /* page is wholly or partially inside EOF */
1776         if (page_start + PAGE_CACHE_SIZE > size)
1777                 end = size & ~PAGE_CACHE_MASK;
1778         else
1779                 end = PAGE_CACHE_SIZE;
1780
1781         ret = btrfs_cow_one_page(inode, page, end);
1782
1783 out_unlock:
1784         up_read(&BTRFS_I(inode)->root->snap_sem);
1785         unlock_page(page);
1786         return ret;
1787 }
1788
1789 static void btrfs_truncate(struct inode *inode)
1790 {
1791         struct btrfs_root *root = BTRFS_I(inode)->root;
1792         int ret;
1793         struct btrfs_trans_handle *trans;
1794         unsigned long nr;
1795
1796         if (!S_ISREG(inode->i_mode))
1797                 return;
1798         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1799                 return;
1800
1801         btrfs_truncate_page(inode->i_mapping, inode->i_size);
1802
1803         mutex_lock(&root->fs_info->fs_mutex);
1804         trans = btrfs_start_transaction(root, 1);
1805         btrfs_set_trans_block_group(trans, inode);
1806
1807         /* FIXME, add redo link to tree so we don't leak on crash */
1808         ret = btrfs_truncate_in_trans(trans, root, inode);
1809         btrfs_update_inode(trans, root, inode);
1810         nr = trans->blocks_used;
1811
1812         ret = btrfs_end_transaction(trans, root);
1813         BUG_ON(ret);
1814         mutex_unlock(&root->fs_info->fs_mutex);
1815         btrfs_btree_balance_dirty(root, nr);
1816 }
1817
1818 int btrfs_commit_write(struct file *file, struct page *page,
1819                        unsigned from, unsigned to)
1820 {
1821         return extent_commit_write(&BTRFS_I(page->mapping->host)->extent_tree,
1822                                    page->mapping->host, page, from, to);
1823 }
1824
1825 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
1826 {
1827         struct btrfs_trans_handle *trans;
1828         struct btrfs_key key;
1829         struct btrfs_root_item root_item;
1830         struct btrfs_inode_item *inode_item;
1831         struct extent_buffer *leaf;
1832         struct btrfs_root *new_root;
1833         struct inode *inode;
1834         struct inode *dir;
1835         int ret;
1836         int err;
1837         u64 objectid;
1838         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
1839         unsigned long nr = 1;
1840
1841         mutex_lock(&root->fs_info->fs_mutex);
1842         trans = btrfs_start_transaction(root, 1);
1843         BUG_ON(!trans);
1844
1845         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 0);
1846         if (IS_ERR(leaf))
1847                 return PTR_ERR(leaf);
1848
1849         btrfs_set_header_nritems(leaf, 0);
1850         btrfs_set_header_level(leaf, 0);
1851         btrfs_set_header_bytenr(leaf, leaf->start);
1852         btrfs_set_header_generation(leaf, trans->transid);
1853         btrfs_set_header_owner(leaf, root->root_key.objectid);
1854         write_extent_buffer(leaf, root->fs_info->fsid,
1855                             (unsigned long)btrfs_header_fsid(leaf),
1856                             BTRFS_FSID_SIZE);
1857         btrfs_mark_buffer_dirty(leaf);
1858
1859         inode_item = &root_item.inode;
1860         memset(inode_item, 0, sizeof(*inode_item));
1861         inode_item->generation = cpu_to_le64(1);
1862         inode_item->size = cpu_to_le64(3);
1863         inode_item->nlink = cpu_to_le32(1);
1864         inode_item->nblocks = cpu_to_le64(1);
1865         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1866
1867         btrfs_set_root_bytenr(&root_item, leaf->start);
1868         btrfs_set_root_level(&root_item, 0);
1869         btrfs_set_root_refs(&root_item, 1);
1870         btrfs_set_root_used(&root_item, 0);
1871
1872         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
1873         root_item.drop_level = 0;
1874
1875         free_extent_buffer(leaf);
1876         leaf = NULL;
1877
1878         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1879                                        0, &objectid);
1880         if (ret)
1881                 goto fail;
1882
1883         btrfs_set_root_dirid(&root_item, new_dirid);
1884
1885         key.objectid = objectid;
1886         key.offset = 1;
1887         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1888         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1889                                 &root_item);
1890         if (ret)
1891                 goto fail;
1892
1893         /*
1894          * insert the directory item
1895          */
1896         key.offset = (u64)-1;
1897         dir = root->fs_info->sb->s_root->d_inode;
1898         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1899                                     name, namelen, dir->i_ino, &key,
1900                                     BTRFS_FT_DIR);
1901         if (ret)
1902                 goto fail;
1903
1904         ret = btrfs_commit_transaction(trans, root);
1905         if (ret)
1906                 goto fail_commit;
1907
1908         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
1909         BUG_ON(!new_root);
1910
1911         trans = btrfs_start_transaction(new_root, 1);
1912         BUG_ON(!trans);
1913
1914         inode = btrfs_new_inode(trans, new_root, new_dirid,
1915                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
1916         if (IS_ERR(inode))
1917                 goto fail;
1918         inode->i_op = &btrfs_dir_inode_operations;
1919         inode->i_fop = &btrfs_dir_file_operations;
1920         new_root->inode = inode;
1921
1922         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
1923         if (ret)
1924                 goto fail;
1925
1926         inode->i_nlink = 1;
1927         inode->i_size = 6;
1928         ret = btrfs_update_inode(trans, new_root, inode);
1929         if (ret)
1930                 goto fail;
1931 fail:
1932         nr = trans->blocks_used;
1933         err = btrfs_commit_transaction(trans, root);
1934         if (err && !ret)
1935                 ret = err;
1936 fail_commit:
1937         mutex_unlock(&root->fs_info->fs_mutex);
1938         btrfs_btree_balance_dirty(root, nr);
1939         return ret;
1940 }
1941
1942 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
1943 {
1944         struct btrfs_trans_handle *trans;
1945         struct btrfs_key key;
1946         struct btrfs_root_item new_root_item;
1947         struct extent_buffer *tmp;
1948         int ret;
1949         int err;
1950         u64 objectid;
1951         unsigned long nr;
1952
1953         if (!root->ref_cows)
1954                 return -EINVAL;
1955
1956         down_write(&root->snap_sem);
1957         freeze_bdev(root->fs_info->sb->s_bdev);
1958         thaw_bdev(root->fs_info->sb->s_bdev, root->fs_info->sb);
1959
1960         mutex_lock(&root->fs_info->fs_mutex);
1961         trans = btrfs_start_transaction(root, 1);
1962         BUG_ON(!trans);
1963
1964         ret = btrfs_update_inode(trans, root, root->inode);
1965         if (ret)
1966                 goto fail;
1967
1968         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1969                                        0, &objectid);
1970         if (ret)
1971                 goto fail;
1972
1973         memcpy(&new_root_item, &root->root_item,
1974                sizeof(new_root_item));
1975
1976         key.objectid = objectid;
1977         key.offset = 1;
1978         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1979
1980         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
1981         btrfs_set_root_bytenr(&new_root_item, root->node->start);
1982         btrfs_set_root_level(&new_root_item, btrfs_header_level(root->node));
1983
1984         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1985                                 &new_root_item);
1986         if (ret)
1987                 goto fail;
1988
1989         /*
1990          * insert the directory item
1991          */
1992         key.offset = (u64)-1;
1993         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1994                                     name, namelen,
1995                                     root->fs_info->sb->s_root->d_inode->i_ino,
1996                                     &key, BTRFS_FT_DIR);
1997
1998         if (ret)
1999                 goto fail;
2000
2001         ret = btrfs_inc_root_ref(trans, root);
2002         if (ret)
2003                 goto fail;
2004 fail:
2005         nr = trans->blocks_used;
2006         err = btrfs_commit_transaction(trans, root);
2007
2008         if (err && !ret)
2009                 ret = err;
2010
2011         mutex_unlock(&root->fs_info->fs_mutex);
2012         up_write(&root->snap_sem);
2013         btrfs_btree_balance_dirty(root, nr);
2014         return ret;
2015 }
2016
2017 static unsigned long force_ra(struct address_space *mapping,
2018                               struct file_ra_state *ra, struct file *file,
2019                               pgoff_t offset, pgoff_t last_index)
2020 {
2021         pgoff_t req_size;
2022
2023 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2024         req_size = last_index - offset + 1;
2025         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2026         return offset;
2027 #else
2028         req_size = min(last_index - offset + 1, (pgoff_t)128);
2029         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2030         return offset + req_size;
2031 #endif
2032 }
2033
2034 int btrfs_defrag_file(struct file *file) {
2035         struct inode *inode = file->f_path.dentry->d_inode;
2036         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2037         struct page *page;
2038         unsigned long last_index;
2039         unsigned long ra_index = 0;
2040         u64 page_start;
2041         u64 page_end;
2042         unsigned long i;
2043
2044         mutex_lock(&inode->i_mutex);
2045         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2046         for (i = 0; i <= last_index; i++) {
2047                 if (i == ra_index) {
2048                         ra_index = force_ra(inode->i_mapping, &file->f_ra,
2049                                             file, ra_index, last_index);
2050                 }
2051                 page = grab_cache_page(inode->i_mapping, i);
2052                 if (!page)
2053                         goto out_unlock;
2054                 if (!PageUptodate(page)) {
2055                         btrfs_readpage(NULL, page);
2056                         lock_page(page);
2057                         if (!PageUptodate(page)) {
2058                                 unlock_page(page);
2059                                 page_cache_release(page);
2060                                 goto out_unlock;
2061                         }
2062                 }
2063                 page_start = page->index << PAGE_CACHE_SHIFT;
2064                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2065
2066                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2067                 set_extent_delalloc(em_tree, page_start,
2068                                     page_end, GFP_NOFS);
2069                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2070                 set_page_dirty(page);
2071                 unlock_page(page);
2072                 page_cache_release(page);
2073                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2074         }
2075
2076 out_unlock:
2077         mutex_unlock(&inode->i_mutex);
2078         return 0;
2079 }
2080
2081 static int btrfs_ioctl_snap_create(struct btrfs_root *root, void __user *arg)
2082 {
2083         struct btrfs_ioctl_vol_args vol_args;
2084         struct btrfs_dir_item *di;
2085         struct btrfs_path *path;
2086         int namelen;
2087         u64 root_dirid;
2088
2089         if (copy_from_user(&vol_args, arg, sizeof(vol_args)))
2090                 return -EFAULT;
2091
2092         namelen = strlen(vol_args.name);
2093         if (namelen > BTRFS_VOL_NAME_MAX)
2094                 return -EINVAL;
2095         if (strchr(vol_args.name, '/'))
2096                 return -EINVAL;
2097
2098         path = btrfs_alloc_path();
2099         if (!path)
2100                 return -ENOMEM;
2101
2102         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2103         mutex_lock(&root->fs_info->fs_mutex);
2104         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2105                             path, root_dirid,
2106                             vol_args.name, namelen, 0);
2107         mutex_unlock(&root->fs_info->fs_mutex);
2108         btrfs_free_path(path);
2109         if (di && !IS_ERR(di))
2110                 return -EEXIST;
2111         if (IS_ERR(di))
2112                 return PTR_ERR(di);
2113
2114         if (root == root->fs_info->tree_root)
2115                 return create_subvol(root, vol_args.name, namelen);
2116         return create_snapshot(root, vol_args.name, namelen);
2117 }
2118
2119 static int btrfs_ioctl_defrag(struct file *file)
2120 {
2121         struct inode *inode = file->f_path.dentry->d_inode;
2122         struct btrfs_root *root = BTRFS_I(inode)->root;
2123
2124         switch (inode->i_mode & S_IFMT) {
2125         case S_IFDIR:
2126                 mutex_lock(&root->fs_info->fs_mutex);
2127                 btrfs_defrag_root(root, 0);
2128                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2129                 mutex_unlock(&root->fs_info->fs_mutex);
2130                 break;
2131         case S_IFREG:
2132                 btrfs_defrag_file(file);
2133                 break;
2134         }
2135
2136         return 0;
2137 }
2138
2139 long btrfs_ioctl(struct file *file, unsigned int
2140                 cmd, unsigned long arg)
2141 {
2142         struct btrfs_root *root = BTRFS_I(file->f_path.dentry->d_inode)->root;
2143
2144         switch (cmd) {
2145         case BTRFS_IOC_SNAP_CREATE:
2146                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2147         case BTRFS_IOC_DEFRAG:
2148                 return btrfs_ioctl_defrag(file);
2149         }
2150
2151         return -ENOTTY;
2152 }
2153
2154 /*
2155  * Called inside transaction, so use GFP_NOFS
2156  */
2157 struct inode *btrfs_alloc_inode(struct super_block *sb)
2158 {
2159         struct btrfs_inode *ei;
2160
2161         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2162         if (!ei)
2163                 return NULL;
2164         ei->last_trans = 0;
2165         return &ei->vfs_inode;
2166 }
2167
2168 void btrfs_destroy_inode(struct inode *inode)
2169 {
2170         WARN_ON(!list_empty(&inode->i_dentry));
2171         WARN_ON(inode->i_data.nrpages);
2172
2173         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2174 }
2175
2176 static void init_once(void * foo, struct kmem_cache * cachep,
2177                       unsigned long flags)
2178 {
2179         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2180
2181         inode_init_once(&ei->vfs_inode);
2182 }
2183
2184 void btrfs_destroy_cachep(void)
2185 {
2186         if (btrfs_inode_cachep)
2187                 kmem_cache_destroy(btrfs_inode_cachep);
2188         if (btrfs_trans_handle_cachep)
2189                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2190         if (btrfs_transaction_cachep)
2191                 kmem_cache_destroy(btrfs_transaction_cachep);
2192         if (btrfs_bit_radix_cachep)
2193                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2194         if (btrfs_path_cachep)
2195                 kmem_cache_destroy(btrfs_path_cachep);
2196 }
2197
2198 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2199                                        unsigned long extra_flags,
2200                                        void (*ctor)(void *, struct kmem_cache *,
2201                                                     unsigned long))
2202 {
2203         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2204                                  SLAB_MEM_SPREAD | extra_flags), ctor
2205 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2206                                  ,NULL
2207 #endif
2208                                 );
2209 }
2210
2211 int btrfs_init_cachep(void)
2212 {
2213         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2214                                           sizeof(struct btrfs_inode),
2215                                           0, init_once);
2216         if (!btrfs_inode_cachep)
2217                 goto fail;
2218         btrfs_trans_handle_cachep =
2219                         btrfs_cache_create("btrfs_trans_handle_cache",
2220                                            sizeof(struct btrfs_trans_handle),
2221                                            0, NULL);
2222         if (!btrfs_trans_handle_cachep)
2223                 goto fail;
2224         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2225                                              sizeof(struct btrfs_transaction),
2226                                              0, NULL);
2227         if (!btrfs_transaction_cachep)
2228                 goto fail;
2229         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2230                                          sizeof(struct btrfs_path),
2231                                          0, NULL);
2232         if (!btrfs_path_cachep)
2233                 goto fail;
2234         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2235                                               SLAB_DESTROY_BY_RCU, NULL);
2236         if (!btrfs_bit_radix_cachep)
2237                 goto fail;
2238         return 0;
2239 fail:
2240         btrfs_destroy_cachep();
2241         return -ENOMEM;
2242 }
2243
2244 static int btrfs_getattr(struct vfsmount *mnt,
2245                          struct dentry *dentry, struct kstat *stat)
2246 {
2247         struct inode *inode = dentry->d_inode;
2248         generic_fillattr(inode, stat);
2249         stat->blksize = 256 * 1024;
2250         return 0;
2251 }
2252
2253 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2254                            struct inode * new_dir,struct dentry *new_dentry)
2255 {
2256         struct btrfs_trans_handle *trans;
2257         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2258         struct inode *new_inode = new_dentry->d_inode;
2259         struct inode *old_inode = old_dentry->d_inode;
2260         struct timespec ctime = CURRENT_TIME;
2261         struct btrfs_path *path;
2262         struct btrfs_dir_item *di;
2263         int ret;
2264
2265         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2266             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2267                 return -ENOTEMPTY;
2268         }
2269
2270         mutex_lock(&root->fs_info->fs_mutex);
2271         trans = btrfs_start_transaction(root, 1);
2272
2273         btrfs_set_trans_block_group(trans, new_dir);
2274         path = btrfs_alloc_path();
2275         if (!path) {
2276                 ret = -ENOMEM;
2277                 goto out_fail;
2278         }
2279
2280         old_dentry->d_inode->i_nlink++;
2281         old_dir->i_ctime = old_dir->i_mtime = ctime;
2282         new_dir->i_ctime = new_dir->i_mtime = ctime;
2283         old_inode->i_ctime = ctime;
2284
2285         if (S_ISDIR(old_inode->i_mode) && old_dir != new_dir) {
2286                 struct btrfs_key *location = &BTRFS_I(new_dir)->location;
2287                 struct btrfs_key old_parent_key;
2288                 di = btrfs_lookup_dir_item(trans, root, path, old_inode->i_ino,
2289                                            "..", 2, -1);
2290                 if (IS_ERR(di)) {
2291                         ret = PTR_ERR(di);
2292                         goto out_fail;
2293                 }
2294                 if (!di) {
2295                         ret = -ENOENT;
2296                         goto out_fail;
2297                 }
2298                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &old_parent_key);
2299                 ret = btrfs_del_item(trans, root, path);
2300                 if (ret) {
2301                         goto out_fail;
2302                 }
2303                 btrfs_release_path(root, path);
2304
2305                 di = btrfs_lookup_dir_index_item(trans, root, path,
2306                                                  old_inode->i_ino,
2307                                                  old_parent_key.objectid,
2308                                                  "..", 2, -1);
2309                 if (IS_ERR(di)) {
2310                         ret = PTR_ERR(di);
2311                         goto out_fail;
2312                 }
2313                 if (!di) {
2314                         ret = -ENOENT;
2315                         goto out_fail;
2316                 }
2317                 ret = btrfs_del_item(trans, root, path);
2318                 if (ret) {
2319                         goto out_fail;
2320                 }
2321                 btrfs_release_path(root, path);
2322
2323                 ret = btrfs_insert_dir_item(trans, root, "..", 2,
2324                                             old_inode->i_ino, location,
2325                                             BTRFS_FT_DIR);
2326                 if (ret)
2327                         goto out_fail;
2328         }
2329
2330
2331         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2332         if (ret)
2333                 goto out_fail;
2334
2335         if (new_inode) {
2336                 new_inode->i_ctime = CURRENT_TIME;
2337                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2338                 if (ret)
2339                         goto out_fail;
2340         }
2341         ret = btrfs_add_link(trans, new_dentry, old_inode);
2342         if (ret)
2343                 goto out_fail;
2344
2345 out_fail:
2346         btrfs_free_path(path);
2347         btrfs_end_transaction(trans, root);
2348         mutex_unlock(&root->fs_info->fs_mutex);
2349         return ret;
2350 }
2351
2352 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2353                          const char *symname)
2354 {
2355         struct btrfs_trans_handle *trans;
2356         struct btrfs_root *root = BTRFS_I(dir)->root;
2357         struct btrfs_path *path;
2358         struct btrfs_key key;
2359         struct inode *inode;
2360         int err;
2361         int drop_inode = 0;
2362         u64 objectid;
2363         int name_len;
2364         int datasize;
2365         unsigned long ptr;
2366         struct btrfs_file_extent_item *ei;
2367         struct extent_buffer *leaf;
2368         unsigned long nr;
2369
2370         name_len = strlen(symname) + 1;
2371         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2372                 return -ENAMETOOLONG;
2373         mutex_lock(&root->fs_info->fs_mutex);
2374         trans = btrfs_start_transaction(root, 1);
2375         btrfs_set_trans_block_group(trans, dir);
2376
2377         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2378         if (err) {
2379                 err = -ENOSPC;
2380                 goto out_unlock;
2381         }
2382
2383         inode = btrfs_new_inode(trans, root, objectid,
2384                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2385         err = PTR_ERR(inode);
2386         if (IS_ERR(inode))
2387                 goto out_unlock;
2388
2389         btrfs_set_trans_block_group(trans, inode);
2390         err = btrfs_add_nondir(trans, dentry, inode);
2391         if (err)
2392                 drop_inode = 1;
2393         else {
2394                 inode->i_mapping->a_ops = &btrfs_aops;
2395                 inode->i_fop = &btrfs_file_operations;
2396                 inode->i_op = &btrfs_file_inode_operations;
2397                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2398                                      inode->i_mapping, GFP_NOFS);
2399                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2400         }
2401         dir->i_sb->s_dirt = 1;
2402         btrfs_update_inode_block_group(trans, inode);
2403         btrfs_update_inode_block_group(trans, dir);
2404         if (drop_inode)
2405                 goto out_unlock;
2406
2407         path = btrfs_alloc_path();
2408         BUG_ON(!path);
2409         key.objectid = inode->i_ino;
2410         key.offset = 0;
2411         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2412         datasize = btrfs_file_extent_calc_inline_size(name_len);
2413         err = btrfs_insert_empty_item(trans, root, path, &key,
2414                                       datasize);
2415         if (err) {
2416                 drop_inode = 1;
2417                 goto out_unlock;
2418         }
2419         leaf = path->nodes[0];
2420         ei = btrfs_item_ptr(leaf, path->slots[0],
2421                             struct btrfs_file_extent_item);
2422         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2423         btrfs_set_file_extent_type(leaf, ei,
2424                                    BTRFS_FILE_EXTENT_INLINE);
2425         ptr = btrfs_file_extent_inline_start(ei);
2426         write_extent_buffer(leaf, symname, ptr, name_len);
2427         btrfs_mark_buffer_dirty(leaf);
2428         btrfs_free_path(path);
2429
2430         inode->i_op = &btrfs_symlink_inode_operations;
2431         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2432         inode->i_size = name_len - 1;
2433         err = btrfs_update_inode(trans, root, inode);
2434         if (err)
2435                 drop_inode = 1;
2436
2437 out_unlock:
2438         nr = trans->blocks_used;
2439         btrfs_end_transaction(trans, root);
2440         mutex_unlock(&root->fs_info->fs_mutex);
2441         if (drop_inode) {
2442                 inode_dec_link_count(inode);
2443                 iput(inode);
2444         }
2445         btrfs_btree_balance_dirty(root, nr);
2446         return err;
2447 }
2448
2449 static struct inode_operations btrfs_dir_inode_operations = {
2450         .lookup         = btrfs_lookup,
2451         .create         = btrfs_create,
2452         .unlink         = btrfs_unlink,
2453         .link           = btrfs_link,
2454         .mkdir          = btrfs_mkdir,
2455         .rmdir          = btrfs_rmdir,
2456         .rename         = btrfs_rename,
2457         .symlink        = btrfs_symlink,
2458         .setattr        = btrfs_setattr,
2459         .mknod          = btrfs_mknod,
2460 };
2461
2462 static struct inode_operations btrfs_dir_ro_inode_operations = {
2463         .lookup         = btrfs_lookup,
2464 };
2465
2466 static struct file_operations btrfs_dir_file_operations = {
2467         .llseek         = generic_file_llseek,
2468         .read           = generic_read_dir,
2469         .readdir        = btrfs_readdir,
2470         .unlocked_ioctl = btrfs_ioctl,
2471 #ifdef CONFIG_COMPAT
2472         .compat_ioctl   = btrfs_ioctl,
2473 #endif
2474 };
2475
2476 static struct extent_map_ops btrfs_extent_map_ops = {
2477         .fill_delalloc = run_delalloc_range,
2478         .writepage_io_hook = btrfs_writepage_io_hook,
2479         .readpage_io_hook = btrfs_readpage_io_hook,
2480         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2481 };
2482
2483 static struct address_space_operations btrfs_aops = {
2484         .readpage       = btrfs_readpage,
2485         .writepage      = btrfs_writepage,
2486         .sync_page      = block_sync_page,
2487         .prepare_write  = btrfs_prepare_write,
2488         .commit_write   = btrfs_commit_write,
2489         .bmap           = btrfs_bmap,
2490         .invalidatepage = btrfs_invalidatepage,
2491         .releasepage    = btrfs_releasepage,
2492         .set_page_dirty = __set_page_dirty_nobuffers,
2493 };
2494
2495 static struct address_space_operations btrfs_symlink_aops = {
2496         .readpage       = btrfs_readpage,
2497         .writepage      = btrfs_writepage,
2498         .invalidatepage = btrfs_invalidatepage,
2499         .releasepage    = btrfs_releasepage,
2500 };
2501
2502 static struct inode_operations btrfs_file_inode_operations = {
2503         .truncate       = btrfs_truncate,
2504         .getattr        = btrfs_getattr,
2505         .setattr        = btrfs_setattr,
2506 };
2507
2508 static struct inode_operations btrfs_special_inode_operations = {
2509         .getattr        = btrfs_getattr,
2510         .setattr        = btrfs_setattr,
2511 };
2512
2513 static struct inode_operations btrfs_symlink_inode_operations = {
2514         .readlink       = generic_readlink,
2515         .follow_link    = page_follow_link_light,
2516         .put_link       = page_put_link,
2517 };