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;
 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 (!di || IS_ERR(di)) {
 857                 location->objectid = 0;
 858                 ret = 0;
 859                 goto out;
 860         }
 861         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
 862 out:
 863         btrfs_release_path(root, path);
 864         btrfs_free_path(path);
 865         return ret;
 866 }
 867
 868 /*
 869  * when we hit a tree root in a directory, the btrfs part of the inode
 870  * needs to be changed to reflect the root directory of the tree root.  This
 871  * is kind of like crossing a mount point.
 872  */
 873 static int fixup_tree_root_location(struct btrfs_root *root,
 874                              struct btrfs_key *location,
 875                              struct btrfs_root **sub_root,
 876                              struct dentry *dentry)
 877 {
 878         struct btrfs_path *path;
 879         struct btrfs_root_item *ri;
 880
 881         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 882                 return 0;
 883         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 884                 return 0;
 885
 886         path = btrfs_alloc_path();
 887         BUG_ON(!path);
 888         mutex_lock(&root->fs_info->fs_mutex);
 889
 890         *sub_root = btrfs_read_fs_root(root->fs_info, location,
 891                                         dentry->d_name.name,
 892                                         dentry->d_name.len);
 893         if (IS_ERR(*sub_root))
 894                 return PTR_ERR(*sub_root);
 895
 896         ri = &(*sub_root)->root_item;
 897         location->objectid = btrfs_root_dirid(ri);
 898         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 899         location->offset = 0;
 900
 901         btrfs_free_path(path);
 902         mutex_unlock(&root->fs_info->fs_mutex);
 903         return 0;
 904 }
 905
 906 static int btrfs_init_locked_inode(struct inode *inode, void *p)
 907 {
 908         struct btrfs_iget_args *args = p;
 909         inode->i_ino = args->ino;
 910         BTRFS_I(inode)->root = args->root;
 911         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
 912                              inode->i_mapping, GFP_NOFS);
 913         return 0;
 914 }
 915
 916 static int btrfs_find_actor(struct inode *inode, void *opaque)
 917 {
 918         struct btrfs_iget_args *args = opaque;
 919         return (args->ino == inode->i_ino &&
 920                 args->root == BTRFS_I(inode)->root);
 921 }
 922
 923 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 924                                 struct btrfs_root *root)
 925 {
 926         struct inode *inode;
 927         struct btrfs_iget_args args;
 928         args.ino = objectid;
 929         args.root = root;
 930
 931         inode = iget5_locked(s, objectid, btrfs_find_actor,
 932                              btrfs_init_locked_inode,
 933                              (void *)&args);
 934         return inode;
 935 }
 936
 937 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 938                                    struct nameidata *nd)
 939 {
 940         struct inode * inode;
 941         struct btrfs_inode *bi = BTRFS_I(dir);
 942         struct btrfs_root *root = bi->root;
 943         struct btrfs_root *sub_root = root;
 944         struct btrfs_key location;
 945         int ret;
 946
 947         if (dentry->d_name.len > BTRFS_NAME_LEN)
 948                 return ERR_PTR(-ENAMETOOLONG);
 949
 950         mutex_lock(&root->fs_info->fs_mutex);
 951         ret = btrfs_inode_by_name(dir, dentry, &location);
 952         mutex_unlock(&root->fs_info->fs_mutex);
 953
 954         if (ret < 0)
 955                 return ERR_PTR(ret);
 956
 957         inode = NULL;
 958         if (location.objectid) {
 959                 ret = fixup_tree_root_location(root, &location, &sub_root,
 960                                                 dentry);
 961                 if (ret < 0)
 962                         return ERR_PTR(ret);
 963                 if (ret > 0)
 964                         return ERR_PTR(-ENOENT);
 965                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 966                                           sub_root);
 967                 if (!inode)
 968                         return ERR_PTR(-EACCES);
 969                 if (inode->i_state & I_NEW) {
 970                         /* the inode and parent dir are two different roots */
 971                         if (sub_root != root) {
 972                                 igrab(inode);
 973                                 sub_root->inode = inode;
 974                         }
 975                         BTRFS_I(inode)->root = sub_root;
 976                         memcpy(&BTRFS_I(inode)->location, &location,
 977                                sizeof(location));
 978                         btrfs_read_locked_inode(inode);
 979                         unlock_new_inode(inode);
 980                 }
 981         }
 982         return d_splice_alias(inode, dentry);
 983 }
 984
 985 static unsigned char btrfs_filetype_table[] = {
 986         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
 987 };
 988
 989 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 990 {
 991         struct inode *inode = filp->f_path.dentry->d_inode;
 992         struct btrfs_root *root = BTRFS_I(inode)->root;
 993         struct btrfs_item *item;
 994         struct btrfs_dir_item *di;
 995         struct btrfs_key key;
 996         struct btrfs_key found_key;
 997         struct btrfs_path *path;
 998         int ret;
 999         u32 nritems;
1000         struct extent_buffer *leaf;
1001         int slot;
1002         int advance;
1003         unsigned char d_type;
1004         int over = 0;
1005         u32 di_cur;
1006         u32 di_total;
1007         u32 di_len;
1008         int key_type = BTRFS_DIR_INDEX_KEY;
1009         char tmp_name[32];
1010         char *name_ptr;
1011         int name_len;
1012
1013         /* FIXME, use a real flag for deciding about the key type */
1014         if (root->fs_info->tree_root == root)
1015                 key_type = BTRFS_DIR_ITEM_KEY;
1016
1017         mutex_lock(&root->fs_info->fs_mutex);
1018         key.objectid = inode->i_ino;
1019         btrfs_set_key_type(&key, key_type);
1020         key.offset = filp->f_pos;
1021
1022         path = btrfs_alloc_path();
1023         path->reada = 2;
1024         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1025         if (ret < 0)
1026                 goto err;
1027         advance = 0;
1028         while(1) {
1029                 leaf = path->nodes[0];
1030                 nritems = btrfs_header_nritems(leaf);
1031                 slot = path->slots[0];
1032                 if (advance || slot >= nritems) {
1033                         if (slot >= nritems -1) {
1034                                 ret = btrfs_next_leaf(root, path);
1035                                 if (ret)
1036                                         break;
1037                                 leaf = path->nodes[0];
1038                                 nritems = btrfs_header_nritems(leaf);
1039                                 slot = path->slots[0];
1040                         } else {
1041                                 slot++;
1042                                 path->slots[0]++;
1043                         }
1044                 }
1045                 advance = 1;
1046                 item = btrfs_item_nr(leaf, slot);
1047                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1048
1049                 if (found_key.objectid != key.objectid)
1050                         break;
1051                 if (btrfs_key_type(&found_key) != key_type)
1052                         break;
1053                 if (found_key.offset < filp->f_pos)
1054                         continue;
1055
1056                 filp->f_pos = found_key.offset;
1057                 advance = 1;
1058                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1059                 di_cur = 0;
1060                 di_total = btrfs_item_size(leaf, item);
1061                 while(di_cur < di_total) {
1062                         struct btrfs_key location;
1063
1064                         name_len = btrfs_dir_name_len(leaf, di);
1065                         if (name_len < 32) {
1066                                 name_ptr = tmp_name;
1067                         } else {
1068                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1069                                 BUG_ON(!name_ptr);
1070                         }
1071                         read_extent_buffer(leaf, name_ptr,
1072                                            (unsigned long)(di + 1), name_len);
1073
1074                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1075                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1076
1077                         over = filldir(dirent, name_ptr, name_len,
1078                                        found_key.offset,
1079                                        location.objectid,
1080                                        d_type);
1081
1082                         if (name_ptr != tmp_name)
1083                                 kfree(name_ptr);
1084
1085                         if (over)
1086                                 goto nopos;
1087                         di_len = btrfs_dir_name_len(leaf, di) + sizeof(*di);
1088                         di_cur += di_len;
1089                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1090                 }
1091         }
1092         filp->f_pos++;
1093 nopos:
1094         ret = 0;
1095 err:
1096         btrfs_release_path(root, path);
1097         btrfs_free_path(path);
1098         mutex_unlock(&root->fs_info->fs_mutex);
1099         return ret;
1100 }
1101
1102 int btrfs_write_inode(struct inode *inode, int wait)
1103 {
1104         struct btrfs_root *root = BTRFS_I(inode)->root;
1105         struct btrfs_trans_handle *trans;
1106         int ret = 0;
1107
1108         if (wait) {
1109                 mutex_lock(&root->fs_info->fs_mutex);
1110                 trans = btrfs_start_transaction(root, 1);
1111                 btrfs_set_trans_block_group(trans, inode);
1112                 ret = btrfs_commit_transaction(trans, root);
1113                 mutex_unlock(&root->fs_info->fs_mutex);
1114         }
1115         return ret;
1116 }
1117
1118 /*
1119  * This is somewhat expensive, updating the tree every time the
1120  * inode changes.  But, it is most likely to find the inode in cache.
1121  * FIXME, needs more benchmarking...there are no reasons other than performance
1122  * to keep or drop this code.
1123  */
1124 void btrfs_dirty_inode(struct inode *inode)
1125 {
1126         struct btrfs_root *root = BTRFS_I(inode)->root;
1127         struct btrfs_trans_handle *trans;
1128
1129         mutex_lock(&root->fs_info->fs_mutex);
1130         trans = btrfs_start_transaction(root, 1);
1131         btrfs_set_trans_block_group(trans, inode);
1132         btrfs_update_inode(trans, root, inode);
1133         btrfs_end_transaction(trans, root);
1134         mutex_unlock(&root->fs_info->fs_mutex);
1135 }
1136
1137 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1138                                      struct btrfs_root *root,
1139                                      u64 objectid,
1140                                      struct btrfs_block_group_cache *group,
1141                                      int mode)
1142 {
1143         struct inode *inode;
1144         struct btrfs_inode_item *inode_item;
1145         struct btrfs_key *location;
1146         struct btrfs_path *path;
1147         int ret;
1148         int owner;
1149
1150         path = btrfs_alloc_path();
1151         BUG_ON(!path);
1152
1153         inode = new_inode(root->fs_info->sb);
1154         if (!inode)
1155                 return ERR_PTR(-ENOMEM);
1156
1157         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1158                              inode->i_mapping, GFP_NOFS);
1159         BTRFS_I(inode)->root = root;
1160
1161         if (mode & S_IFDIR)
1162                 owner = 0;
1163         else
1164                 owner = 1;
1165         group = btrfs_find_block_group(root, group, 0, 0, owner);
1166         BTRFS_I(inode)->block_group = group;
1167
1168         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1169         if (ret)
1170                 goto fail;
1171
1172         inode->i_uid = current->fsuid;
1173         inode->i_gid = current->fsgid;
1174         inode->i_mode = mode;
1175         inode->i_ino = objectid;
1176         inode->i_blocks = 0;
1177         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1178         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1179                                   struct btrfs_inode_item);
1180         fill_inode_item(path->nodes[0], inode_item, inode);
1181         btrfs_mark_buffer_dirty(path->nodes[0]);
1182         btrfs_free_path(path);
1183
1184         location = &BTRFS_I(inode)->location;
1185         location->objectid = objectid;
1186         location->offset = 0;
1187         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1188
1189         insert_inode_hash(inode);
1190         return inode;
1191 fail:
1192         btrfs_free_path(path);
1193         return ERR_PTR(ret);
1194 }
1195
1196 static inline u8 btrfs_inode_type(struct inode *inode)
1197 {
1198         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1199 }
1200
1201 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1202                             struct dentry *dentry, struct inode *inode)
1203 {
1204         int ret;
1205         struct btrfs_key key;
1206         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1207         struct inode *parent_inode;
1208
1209         key.objectid = inode->i_ino;
1210         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1211         key.offset = 0;
1212
1213         ret = btrfs_insert_dir_item(trans, root,
1214                                     dentry->d_name.name, dentry->d_name.len,
1215                                     dentry->d_parent->d_inode->i_ino,
1216                                     &key, btrfs_inode_type(inode));
1217         if (ret == 0) {
1218                 parent_inode = dentry->d_parent->d_inode;
1219                 parent_inode->i_size += dentry->d_name.len * 2;
1220                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1221                 ret = btrfs_update_inode(trans, root,
1222                                          dentry->d_parent->d_inode);
1223         }
1224         return ret;
1225 }
1226
1227 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1228                             struct dentry *dentry, struct inode *inode)
1229 {
1230         int err = btrfs_add_link(trans, dentry, inode);
1231         if (!err) {
1232                 d_instantiate(dentry, inode);
1233                 return 0;
1234         }
1235         if (err > 0)
1236                 err = -EEXIST;
1237         return err;
1238 }
1239
1240 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1241                         int mode, dev_t rdev)
1242 {
1243         struct btrfs_trans_handle *trans;
1244         struct btrfs_root *root = BTRFS_I(dir)->root;
1245         struct inode *inode;
1246         int err;
1247         int drop_inode = 0;
1248         u64 objectid;
1249         unsigned long nr;
1250
1251         if (!new_valid_dev(rdev))
1252                 return -EINVAL;
1253
1254         mutex_lock(&root->fs_info->fs_mutex);
1255         trans = btrfs_start_transaction(root, 1);
1256         btrfs_set_trans_block_group(trans, dir);
1257
1258         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1259         if (err) {
1260                 err = -ENOSPC;
1261                 goto out_unlock;
1262         }
1263
1264         inode = btrfs_new_inode(trans, root, objectid,
1265                                 BTRFS_I(dir)->block_group, mode);
1266         err = PTR_ERR(inode);
1267         if (IS_ERR(inode))
1268                 goto out_unlock;
1269
1270         btrfs_set_trans_block_group(trans, inode);
1271         err = btrfs_add_nondir(trans, dentry, inode);
1272         if (err)
1273                 drop_inode = 1;
1274         else {
1275                 inode->i_op = &btrfs_special_inode_operations;
1276                 init_special_inode(inode, inode->i_mode, rdev);
1277                 btrfs_update_inode(trans, root, inode);
1278         }
1279         dir->i_sb->s_dirt = 1;
1280         btrfs_update_inode_block_group(trans, inode);
1281         btrfs_update_inode_block_group(trans, dir);
1282 out_unlock:
1283         nr = trans->blocks_used;
1284         btrfs_end_transaction(trans, root);
1285         mutex_unlock(&root->fs_info->fs_mutex);
1286
1287         if (drop_inode) {
1288                 inode_dec_link_count(inode);
1289                 iput(inode);
1290         }
1291         btrfs_btree_balance_dirty(root, nr);
1292         return err;
1293 }
1294
1295 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1296                         int mode, struct nameidata *nd)
1297 {
1298         struct btrfs_trans_handle *trans;
1299         struct btrfs_root *root = BTRFS_I(dir)->root;
1300         struct inode *inode;
1301         int err;
1302         int drop_inode = 0;
1303         unsigned long nr;
1304         u64 objectid;
1305
1306         mutex_lock(&root->fs_info->fs_mutex);
1307         trans = btrfs_start_transaction(root, 1);
1308         btrfs_set_trans_block_group(trans, dir);
1309
1310         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1311         if (err) {
1312                 err = -ENOSPC;
1313                 goto out_unlock;
1314         }
1315
1316         inode = btrfs_new_inode(trans, root, objectid,
1317                                 BTRFS_I(dir)->block_group, mode);
1318         err = PTR_ERR(inode);
1319         if (IS_ERR(inode))
1320                 goto out_unlock;
1321
1322         btrfs_set_trans_block_group(trans, inode);
1323         err = btrfs_add_nondir(trans, dentry, inode);
1324         if (err)
1325                 drop_inode = 1;
1326         else {
1327                 inode->i_mapping->a_ops = &btrfs_aops;
1328                 inode->i_fop = &btrfs_file_operations;
1329                 inode->i_op = &btrfs_file_inode_operations;
1330                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1331                                      inode->i_mapping, GFP_NOFS);
1332                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1333         }
1334         dir->i_sb->s_dirt = 1;
1335         btrfs_update_inode_block_group(trans, inode);
1336         btrfs_update_inode_block_group(trans, dir);
1337 out_unlock:
1338         nr = trans->blocks_used;
1339         btrfs_end_transaction(trans, root);
1340         mutex_unlock(&root->fs_info->fs_mutex);
1341
1342         if (drop_inode) {
1343                 inode_dec_link_count(inode);
1344                 iput(inode);
1345         }
1346         btrfs_btree_balance_dirty(root, nr);
1347         return err;
1348 }
1349
1350 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1351                       struct dentry *dentry)
1352 {
1353         struct btrfs_trans_handle *trans;
1354         struct btrfs_root *root = BTRFS_I(dir)->root;
1355         struct inode *inode = old_dentry->d_inode;
1356         unsigned long nr;
1357         int err;
1358         int drop_inode = 0;
1359
1360         if (inode->i_nlink == 0)
1361                 return -ENOENT;
1362
1363         inc_nlink(inode);
1364         mutex_lock(&root->fs_info->fs_mutex);
1365         trans = btrfs_start_transaction(root, 1);
1366
1367         btrfs_set_trans_block_group(trans, dir);
1368         atomic_inc(&inode->i_count);
1369         err = btrfs_add_nondir(trans, dentry, inode);
1370
1371         if (err)
1372                 drop_inode = 1;
1373
1374         dir->i_sb->s_dirt = 1;
1375         btrfs_update_inode_block_group(trans, dir);
1376         err = btrfs_update_inode(trans, root, inode);
1377
1378         if (err)
1379                 drop_inode = 1;
1380
1381         nr = trans->blocks_used;
1382         btrfs_end_transaction(trans, root);
1383         mutex_unlock(&root->fs_info->fs_mutex);
1384
1385         if (drop_inode) {
1386                 inode_dec_link_count(inode);
1387                 iput(inode);
1388         }
1389         btrfs_btree_balance_dirty(root, nr);
1390         return err;
1391 }
1392
1393 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
1394                                 struct btrfs_root *root,
1395                                 u64 objectid, u64 dirid)
1396 {
1397         int ret;
1398         char buf[2];
1399         struct btrfs_key key;
1400
1401         buf[0] = '.';
1402         buf[1] = '.';
1403
1404         key.objectid = objectid;
1405         key.offset = 0;
1406         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1407
1408         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
1409                                     &key, BTRFS_FT_DIR);
1410         if (ret)
1411                 goto error;
1412
1413         key.objectid = dirid;
1414         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
1415                                     &key, BTRFS_FT_DIR);
1416         if (ret)
1417                 goto error;
1418 error:
1419         return ret;
1420 }
1421
1422 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1423 {
1424         struct inode *inode;
1425         struct btrfs_trans_handle *trans;
1426         struct btrfs_root *root = BTRFS_I(dir)->root;
1427         int err = 0;
1428         int drop_on_err = 0;
1429         u64 objectid;
1430         unsigned long nr = 1;
1431
1432         mutex_lock(&root->fs_info->fs_mutex);
1433         trans = btrfs_start_transaction(root, 1);
1434         btrfs_set_trans_block_group(trans, dir);
1435
1436         if (IS_ERR(trans)) {
1437                 err = PTR_ERR(trans);
1438                 goto out_unlock;
1439         }
1440
1441         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1442         if (err) {
1443                 err = -ENOSPC;
1444                 goto out_unlock;
1445         }
1446
1447         inode = btrfs_new_inode(trans, root, objectid,
1448                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1449         if (IS_ERR(inode)) {
1450                 err = PTR_ERR(inode);
1451                 goto out_fail;
1452         }
1453
1454         drop_on_err = 1;
1455         inode->i_op = &btrfs_dir_inode_operations;
1456         inode->i_fop = &btrfs_dir_file_operations;
1457         btrfs_set_trans_block_group(trans, inode);
1458
1459         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
1460         if (err)
1461                 goto out_fail;
1462
1463         inode->i_size = 6;
1464         err = btrfs_update_inode(trans, root, inode);
1465         if (err)
1466                 goto out_fail;
1467
1468         err = btrfs_add_link(trans, dentry, inode);
1469         if (err)
1470                 goto out_fail;
1471
1472         d_instantiate(dentry, inode);
1473         drop_on_err = 0;
1474         dir->i_sb->s_dirt = 1;
1475         btrfs_update_inode_block_group(trans, inode);
1476         btrfs_update_inode_block_group(trans, dir);
1477
1478 out_fail:
1479         nr = trans->blocks_used;
1480         btrfs_end_transaction(trans, root);
1481
1482 out_unlock:
1483         mutex_unlock(&root->fs_info->fs_mutex);
1484         if (drop_on_err)
1485                 iput(inode);
1486         btrfs_btree_balance_dirty(root, nr);
1487         return err;
1488 }
1489
1490 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1491                                     size_t page_offset, u64 start, u64 end,
1492                                     int create)
1493 {
1494         int ret;
1495         int err = 0;
1496         u64 bytenr;
1497         u64 extent_start = 0;
1498         u64 extent_end = 0;
1499         u64 objectid = inode->i_ino;
1500         u32 found_type;
1501         int failed_insert = 0;
1502         struct btrfs_path *path;
1503         struct btrfs_root *root = BTRFS_I(inode)->root;
1504         struct btrfs_file_extent_item *item;
1505         struct extent_buffer *leaf;
1506         struct btrfs_key found_key;
1507         struct extent_map *em = NULL;
1508         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1509         struct btrfs_trans_handle *trans = NULL;
1510
1511         path = btrfs_alloc_path();
1512         BUG_ON(!path);
1513         mutex_lock(&root->fs_info->fs_mutex);
1514
1515 again:
1516         em = lookup_extent_mapping(em_tree, start, end);
1517         if (em) {
1518                 goto out;
1519         }
1520         if (!em) {
1521                 em = alloc_extent_map(GFP_NOFS);
1522                 if (!em) {
1523                         err = -ENOMEM;
1524                         goto out;
1525                 }
1526                 em->start = EXTENT_MAP_HOLE;
1527                 em->end = EXTENT_MAP_HOLE;
1528         }
1529         em->bdev = inode->i_sb->s_bdev;
1530         ret = btrfs_lookup_file_extent(NULL, root, path,
1531                                        objectid, start, 0);
1532         if (ret < 0) {
1533                 err = ret;
1534                 goto out;
1535         }
1536
1537         if (ret != 0) {
1538                 if (path->slots[0] == 0)
1539                         goto not_found;
1540                 path->slots[0]--;
1541         }
1542
1543         leaf = path->nodes[0];
1544         item = btrfs_item_ptr(leaf, path->slots[0],
1545                               struct btrfs_file_extent_item);
1546         /* are we inside the extent that was found? */
1547         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1548         found_type = btrfs_key_type(&found_key);
1549         if (found_key.objectid != objectid ||
1550             found_type != BTRFS_EXTENT_DATA_KEY) {
1551                 goto not_found;
1552         }
1553
1554         found_type = btrfs_file_extent_type(leaf, item);
1555         extent_start = found_key.offset;
1556         if (found_type == BTRFS_FILE_EXTENT_REG) {
1557                 extent_end = extent_start +
1558                        btrfs_file_extent_num_bytes(leaf, item);
1559                 err = 0;
1560                 if (start < extent_start || start >= extent_end) {
1561                         em->start = start;
1562                         if (start < extent_start) {
1563                                 if (end < extent_start)
1564                                         goto not_found;
1565                                 em->end = extent_end - 1;
1566                         } else {
1567                                 em->end = end;
1568                         }
1569                         goto not_found_em;
1570                 }
1571                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1572                 if (bytenr == 0) {
1573                         em->start = extent_start;
1574                         em->end = extent_end - 1;
1575                         em->block_start = EXTENT_MAP_HOLE;
1576                         em->block_end = EXTENT_MAP_HOLE;
1577                         goto insert;
1578                 }
1579                 bytenr += btrfs_file_extent_offset(leaf, item);
1580                 em->block_start = bytenr;
1581                 em->block_end = em->block_start +
1582                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1583                 em->start = extent_start;
1584                 em->end = extent_end - 1;
1585                 goto insert;
1586         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1587                 unsigned long ptr;
1588                 char *map;
1589                 size_t size;
1590                 size_t extent_offset;
1591                 size_t copy_size;
1592
1593                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1594                                                     path->slots[0]));
1595
1596                 extent_end = (extent_start + size) |
1597                         ((u64)root->sectorsize - 1);
1598                 if (start < extent_start || start >= extent_end) {
1599                         em->start = start;
1600                         if (start < extent_start) {
1601                                 if (end < extent_start)
1602                                         goto not_found;
1603                                 em->end = extent_end;
1604                         } else {
1605                                 em->end = end;
1606                         }
1607                         goto not_found_em;
1608                 }
1609
1610                 extent_offset = (page->index << PAGE_CACHE_SHIFT) -
1611                         extent_start;
1612                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1613                 map = kmap(page);
1614                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1615                                 size - extent_offset);
1616
1617                 em->block_start = EXTENT_MAP_INLINE;
1618                 em->block_end = EXTENT_MAP_INLINE;
1619                 em->start = extent_start + extent_offset;
1620                 em->end = (em->start + copy_size -1) |
1621                         ((u64)root->sectorsize -1);
1622
1623                 if (!page) {
1624                         goto insert;
1625                 }
1626
1627                 read_extent_buffer(leaf, map + page_offset, ptr, copy_size);
1628                 /*
1629                 memset(map + page_offset + copy_size, 0,
1630                        PAGE_CACHE_SIZE - copy_size - page_offset);
1631                        */
1632                 flush_dcache_page(page);
1633                 kunmap(page);
1634                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1635                 goto insert;
1636         } else {
1637                 printk("unkknown found_type %d\n", found_type);
1638                 WARN_ON(1);
1639         }
1640 not_found:
1641         em->start = start;
1642         em->end = end;
1643 not_found_em:
1644         em->block_start = EXTENT_MAP_HOLE;
1645         em->block_end = EXTENT_MAP_HOLE;
1646 insert:
1647         btrfs_release_path(root, path);
1648         if (em->start > start || em->end < start) {
1649                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1650                 err = -EIO;
1651                 goto out;
1652         }
1653         ret = add_extent_mapping(em_tree, em);
1654         if (ret == -EEXIST) {
1655                 free_extent_map(em);
1656                 em = NULL;
1657                 failed_insert++;
1658                 if (failed_insert > 5) {
1659                         printk("failing to insert %Lu %Lu\n", start, end);
1660                         err = -EIO;
1661                         goto out;
1662                 }
1663                 goto again;
1664         }
1665         err = 0;
1666 out:
1667         btrfs_free_path(path);
1668         if (trans) {
1669                 ret = btrfs_end_transaction(trans, root);
1670                 if (!err)
1671                         err = ret;
1672         }
1673         mutex_unlock(&root->fs_info->fs_mutex);
1674         if (err) {
1675                 free_extent_map(em);
1676                 WARN_ON(1);
1677                 return ERR_PTR(err);
1678         }
1679         return em;
1680 }
1681
1682 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1683 {
1684         return extent_bmap(mapping, iblock, btrfs_get_extent);
1685 }
1686
1687 static int btrfs_prepare_write(struct file *file, struct page *page,
1688                                unsigned from, unsigned to)
1689 {
1690         return extent_prepare_write(&BTRFS_I(page->mapping->host)->extent_tree,
1691                                     page->mapping->host, page, from, to,
1692                                     btrfs_get_extent);
1693 }
1694
1695 int btrfs_readpage(struct file *file, struct page *page)
1696 {
1697         struct extent_map_tree *tree;
1698         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1699         return extent_read_full_page(tree, page, btrfs_get_extent);
1700 }
1701 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1702 {
1703         struct extent_map_tree *tree;
1704
1705
1706         if (current->flags & PF_MEMALLOC) {
1707                 redirty_page_for_writepage(wbc, page);
1708                 unlock_page(page);
1709                 return 0;
1710         }
1711         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1712         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
1713 }
1714
1715 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1716 {
1717         struct extent_map_tree *tree;
1718         int ret;
1719
1720         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1721         ret = try_release_extent_mapping(tree, page);
1722         if (ret == 1) {
1723                 ClearPagePrivate(page);
1724                 set_page_private(page, 0);
1725                 page_cache_release(page);
1726         }
1727         return ret;
1728 }
1729
1730 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
1731 {
1732         struct extent_map_tree *tree;
1733
1734         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1735         extent_invalidatepage(tree, page, offset);
1736         btrfs_releasepage(page, GFP_NOFS);
1737 }
1738
1739 /*
1740  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
1741  * called from a page fault handler when a page is first dirtied. Hence we must
1742  * be careful to check for EOF conditions here. We set the page up correctly
1743  * for a written page which means we get ENOSPC checking when writing into
1744  * holes and correct delalloc and unwritten extent mapping on filesystems that
1745  * support these features.
1746  *
1747  * We are not allowed to take the i_mutex here so we have to play games to
1748  * protect against truncate races as the page could now be beyond EOF.  Because
1749  * vmtruncate() writes the inode size before removing pages, once we have the
1750  * page lock we can determine safely if the page is beyond EOF. If it is not
1751  * beyond EOF, then the page is guaranteed safe against truncation until we
1752  * unlock the page.
1753  */
1754 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1755 {
1756         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1757         unsigned long end;
1758         loff_t size;
1759         int ret = -EINVAL;
1760         u64 page_start;
1761
1762         down_read(&BTRFS_I(inode)->root->snap_sem);
1763         lock_page(page);
1764         wait_on_page_writeback(page);
1765         size = i_size_read(inode);
1766         page_start = page->index << PAGE_CACHE_SHIFT;
1767
1768         if ((page->mapping != inode->i_mapping) ||
1769             (page_start > size)) {
1770                 /* page got truncated out from underneath us */
1771                 goto out_unlock;
1772         }
1773
1774         /* page is wholly or partially inside EOF */
1775         if (page_start + PAGE_CACHE_SIZE > size)
1776                 end = size & ~PAGE_CACHE_MASK;
1777         else
1778                 end = PAGE_CACHE_SIZE;
1779
1780         ret = btrfs_cow_one_page(inode, page, end);
1781
1782 out_unlock:
1783         up_read(&BTRFS_I(inode)->root->snap_sem);
1784         unlock_page(page);
1785         return ret;
1786 }
1787
1788 static void btrfs_truncate(struct inode *inode)
1789 {
1790         struct btrfs_root *root = BTRFS_I(inode)->root;
1791         int ret;
1792         struct btrfs_trans_handle *trans;
1793         unsigned long nr;
1794
1795         if (!S_ISREG(inode->i_mode))
1796                 return;
1797         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1798                 return;
1799
1800         btrfs_truncate_page(inode->i_mapping, inode->i_size);
1801
1802         mutex_lock(&root->fs_info->fs_mutex);
1803         trans = btrfs_start_transaction(root, 1);
1804         btrfs_set_trans_block_group(trans, inode);
1805
1806         /* FIXME, add redo link to tree so we don't leak on crash */
1807         ret = btrfs_truncate_in_trans(trans, root, inode);
1808         btrfs_update_inode(trans, root, inode);
1809         nr = trans->blocks_used;
1810
1811         ret = btrfs_end_transaction(trans, root);
1812         BUG_ON(ret);
1813         mutex_unlock(&root->fs_info->fs_mutex);
1814         btrfs_btree_balance_dirty(root, nr);
1815 }
1816
1817 int btrfs_commit_write(struct file *file, struct page *page,
1818                        unsigned from, unsigned to)
1819 {
1820         return extent_commit_write(&BTRFS_I(page->mapping->host)->extent_tree,
1821                                    page->mapping->host, page, from, to);
1822 }
1823
1824 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
1825 {
1826         struct btrfs_trans_handle *trans;
1827         struct btrfs_key key;
1828         struct btrfs_root_item root_item;
1829         struct btrfs_inode_item *inode_item;
1830         struct extent_buffer *leaf;
1831         struct btrfs_root *new_root;
1832         struct inode *inode;
1833         struct inode *dir;
1834         int ret;
1835         int err;
1836         u64 objectid;
1837         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
1838         unsigned long nr = 1;
1839
1840         mutex_lock(&root->fs_info->fs_mutex);
1841         trans = btrfs_start_transaction(root, 1);
1842         BUG_ON(!trans);
1843
1844         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0, 0);
1845         if (IS_ERR(leaf))
1846                 return PTR_ERR(leaf);
1847
1848         btrfs_set_header_nritems(leaf, 0);
1849         btrfs_set_header_level(leaf, 0);
1850         btrfs_set_header_bytenr(leaf, leaf->start);
1851         btrfs_set_header_generation(leaf, trans->transid);
1852         btrfs_set_header_owner(leaf, root->root_key.objectid);
1853         write_extent_buffer(leaf, root->fs_info->fsid,
1854                             (unsigned long)btrfs_header_fsid(leaf),
1855                             BTRFS_FSID_SIZE);
1856         btrfs_mark_buffer_dirty(leaf);
1857
1858         inode_item = &root_item.inode;
1859         memset(inode_item, 0, sizeof(*inode_item));
1860         inode_item->generation = cpu_to_le64(1);
1861         inode_item->size = cpu_to_le64(3);
1862         inode_item->nlink = cpu_to_le32(1);
1863         inode_item->nblocks = cpu_to_le64(1);
1864         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1865
1866         btrfs_set_root_bytenr(&root_item, leaf->start);
1867         btrfs_set_root_level(&root_item, 0);
1868         btrfs_set_root_refs(&root_item, 1);
1869         btrfs_set_root_used(&root_item, 0);
1870
1871         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
1872         root_item.drop_level = 0;
1873
1874         free_extent_buffer(leaf);
1875         leaf = NULL;
1876
1877         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1878                                        0, &objectid);
1879         if (ret)
1880                 goto fail;
1881
1882         btrfs_set_root_dirid(&root_item, new_dirid);
1883
1884         key.objectid = objectid;
1885         key.offset = 1;
1886         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1887         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1888                                 &root_item);
1889         if (ret)
1890                 goto fail;
1891
1892         /*
1893          * insert the directory item
1894          */
1895         key.offset = (u64)-1;
1896         dir = root->fs_info->sb->s_root->d_inode;
1897         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1898                                     name, namelen, dir->i_ino, &key,
1899                                     BTRFS_FT_DIR);
1900         if (ret)
1901                 goto fail;
1902
1903         ret = btrfs_commit_transaction(trans, root);
1904         if (ret)
1905                 goto fail_commit;
1906
1907         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
1908         BUG_ON(!new_root);
1909
1910         trans = btrfs_start_transaction(new_root, 1);
1911         BUG_ON(!trans);
1912
1913         inode = btrfs_new_inode(trans, new_root, new_dirid,
1914                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
1915         if (IS_ERR(inode))
1916                 goto fail;
1917         inode->i_op = &btrfs_dir_inode_operations;
1918         inode->i_fop = &btrfs_dir_file_operations;
1919         new_root->inode = inode;
1920
1921         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
1922         if (ret)
1923                 goto fail;
1924
1925         inode->i_nlink = 1;
1926         inode->i_size = 6;
1927         ret = btrfs_update_inode(trans, new_root, inode);
1928         if (ret)
1929                 goto fail;
1930 fail:
1931         nr = trans->blocks_used;
1932         err = btrfs_commit_transaction(trans, root);
1933         if (err && !ret)
1934                 ret = err;
1935 fail_commit:
1936         mutex_unlock(&root->fs_info->fs_mutex);
1937         btrfs_btree_balance_dirty(root, nr);
1938         return ret;
1939 }
1940
1941 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
1942 {
1943         struct btrfs_trans_handle *trans;
1944         struct btrfs_key key;
1945         struct btrfs_root_item new_root_item;
1946         struct extent_buffer *tmp;
1947         int ret;
1948         int err;
1949         u64 objectid;
1950         unsigned long nr;
1951
1952         if (!root->ref_cows)
1953                 return -EINVAL;
1954
1955         down_write(&root->snap_sem);
1956         freeze_bdev(root->fs_info->sb->s_bdev);
1957         thaw_bdev(root->fs_info->sb->s_bdev, root->fs_info->sb);
1958
1959         mutex_lock(&root->fs_info->fs_mutex);
1960         trans = btrfs_start_transaction(root, 1);
1961         BUG_ON(!trans);
1962
1963         ret = btrfs_update_inode(trans, root, root->inode);
1964         if (ret)
1965                 goto fail;
1966
1967         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1968                                        0, &objectid);
1969         if (ret)
1970                 goto fail;
1971
1972         memcpy(&new_root_item, &root->root_item,
1973                sizeof(new_root_item));
1974
1975         key.objectid = objectid;
1976         key.offset = 1;
1977         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1978
1979         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
1980         btrfs_set_root_bytenr(&new_root_item, root->node->start);
1981         btrfs_set_root_level(&new_root_item, btrfs_header_level(root->node));
1982
1983         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1984                                 &new_root_item);
1985         if (ret)
1986                 goto fail;
1987
1988         /*
1989          * insert the directory item
1990          */
1991         key.offset = (u64)-1;
1992         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1993                                     name, namelen,
1994                                     root->fs_info->sb->s_root->d_inode->i_ino,
1995                                     &key, BTRFS_FT_DIR);
1996
1997         if (ret)
1998                 goto fail;
1999
2000         ret = btrfs_inc_root_ref(trans, root);
2001         if (ret)
2002                 goto fail;
2003 fail:
2004         nr = trans->blocks_used;
2005         err = btrfs_commit_transaction(trans, root);
2006
2007         if (err && !ret)
2008                 ret = err;
2009
2010         mutex_unlock(&root->fs_info->fs_mutex);
2011         up_write(&root->snap_sem);
2012         btrfs_btree_balance_dirty(root, nr);
2013         return ret;
2014 }
2015
2016 static unsigned long force_ra(struct address_space *mapping,
2017                               struct file_ra_state *ra, struct file *file,
2018                               pgoff_t offset, pgoff_t last_index)
2019 {
2020         pgoff_t req_size;
2021
2022 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2023         req_size = last_index - offset + 1;
2024         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2025         return offset;
2026 #else
2027         req_size = min(last_index - offset + 1, (pgoff_t)128);
2028         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2029         return offset + req_size;
2030 #endif
2031 }
2032
2033 int btrfs_defrag_file(struct file *file) {
2034         struct inode *inode = file->f_path.dentry->d_inode;
2035         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2036         struct page *page;
2037         unsigned long last_index;
2038         unsigned long ra_index = 0;
2039         u64 page_start;
2040         u64 page_end;
2041         unsigned long i;
2042
2043         mutex_lock(&inode->i_mutex);
2044         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2045         for (i = 0; i <= last_index; i++) {
2046                 if (i == ra_index) {
2047                         ra_index = force_ra(inode->i_mapping, &file->f_ra,
2048                                             file, ra_index, last_index);
2049                 }
2050                 page = grab_cache_page(inode->i_mapping, i);
2051                 if (!page)
2052                         goto out_unlock;
2053                 if (!PageUptodate(page)) {
2054                         btrfs_readpage(NULL, page);
2055                         lock_page(page);
2056                         if (!PageUptodate(page)) {
2057                                 unlock_page(page);
2058                                 page_cache_release(page);
2059                                 goto out_unlock;
2060                         }
2061                 }
2062                 page_start = page->index << PAGE_CACHE_SHIFT;
2063                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2064
2065                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2066                 set_extent_delalloc(em_tree, page_start,
2067                                     page_end, GFP_NOFS);
2068                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2069                 set_page_dirty(page);
2070                 unlock_page(page);
2071                 page_cache_release(page);
2072                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2073         }
2074
2075 out_unlock:
2076         mutex_unlock(&inode->i_mutex);
2077         return 0;
2078 }
2079
2080 static int btrfs_ioctl_snap_create(struct btrfs_root *root, void __user *arg)
2081 {
2082         struct btrfs_ioctl_vol_args vol_args;
2083         struct btrfs_dir_item *di;
2084         struct btrfs_path *path;
2085         int namelen;
2086         u64 root_dirid;
2087
2088         if (copy_from_user(&vol_args, arg, sizeof(vol_args)))
2089                 return -EFAULT;
2090
2091         namelen = strlen(vol_args.name);
2092         if (namelen > BTRFS_VOL_NAME_MAX)
2093                 return -EINVAL;
2094         if (strchr(vol_args.name, '/'))
2095                 return -EINVAL;
2096
2097         path = btrfs_alloc_path();
2098         if (!path)
2099                 return -ENOMEM;
2100
2101         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2102         mutex_lock(&root->fs_info->fs_mutex);
2103         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2104                             path, root_dirid,
2105                             vol_args.name, namelen, 0);
2106         mutex_unlock(&root->fs_info->fs_mutex);
2107         btrfs_free_path(path);
2108         if (di && !IS_ERR(di))
2109                 return -EEXIST;
2110         if (IS_ERR(di))
2111                 return PTR_ERR(di);
2112
2113         if (root == root->fs_info->tree_root)
2114                 return create_subvol(root, vol_args.name, namelen);
2115         return create_snapshot(root, vol_args.name, namelen);
2116 }
2117
2118 static int btrfs_ioctl_defrag(struct file *file)
2119 {
2120         struct inode *inode = file->f_path.dentry->d_inode;
2121         struct btrfs_root *root = BTRFS_I(inode)->root;
2122
2123         switch (inode->i_mode & S_IFMT) {
2124         case S_IFDIR:
2125                 mutex_lock(&root->fs_info->fs_mutex);
2126                 btrfs_defrag_root(root, 0);
2127                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2128                 mutex_unlock(&root->fs_info->fs_mutex);
2129                 break;
2130         case S_IFREG:
2131                 btrfs_defrag_file(file);
2132                 break;
2133         }
2134
2135         return 0;
2136 }
2137
2138 long btrfs_ioctl(struct file *file, unsigned int
2139                 cmd, unsigned long arg)
2140 {
2141         struct btrfs_root *root = BTRFS_I(file->f_path.dentry->d_inode)->root;
2142
2143         switch (cmd) {
2144         case BTRFS_IOC_SNAP_CREATE:
2145                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2146         case BTRFS_IOC_DEFRAG:
2147                 return btrfs_ioctl_defrag(file);
2148         }
2149
2150         return -ENOTTY;
2151 }
2152
2153 /*
2154  * Called inside transaction, so use GFP_NOFS
2155  */
2156 struct inode *btrfs_alloc_inode(struct super_block *sb)
2157 {
2158         struct btrfs_inode *ei;
2159
2160         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2161         if (!ei)
2162                 return NULL;
2163         ei->last_trans = 0;
2164         return &ei->vfs_inode;
2165 }
2166
2167 void btrfs_destroy_inode(struct inode *inode)
2168 {
2169         WARN_ON(!list_empty(&inode->i_dentry));
2170         WARN_ON(inode->i_data.nrpages);
2171
2172         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2173 }
2174
2175 static void init_once(void * foo, struct kmem_cache * cachep,
2176                       unsigned long flags)
2177 {
2178         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2179
2180         inode_init_once(&ei->vfs_inode);
2181 }
2182
2183 void btrfs_destroy_cachep(void)
2184 {
2185         if (btrfs_inode_cachep)
2186                 kmem_cache_destroy(btrfs_inode_cachep);
2187         if (btrfs_trans_handle_cachep)
2188                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2189         if (btrfs_transaction_cachep)
2190                 kmem_cache_destroy(btrfs_transaction_cachep);
2191         if (btrfs_bit_radix_cachep)
2192                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2193         if (btrfs_path_cachep)
2194                 kmem_cache_destroy(btrfs_path_cachep);
2195 }
2196
2197 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2198                                        unsigned long extra_flags,
2199                                        void (*ctor)(void *, struct kmem_cache *,
2200                                                     unsigned long))
2201 {
2202         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2203                                  SLAB_MEM_SPREAD | extra_flags), ctor
2204 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2205                                  ,NULL
2206 #endif
2207                                 );
2208 }
2209
2210 int btrfs_init_cachep(void)
2211 {
2212         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2213                                           sizeof(struct btrfs_inode),
2214                                           0, init_once);
2215         if (!btrfs_inode_cachep)
2216                 goto fail;
2217         btrfs_trans_handle_cachep =
2218                         btrfs_cache_create("btrfs_trans_handle_cache",
2219                                            sizeof(struct btrfs_trans_handle),
2220                                            0, NULL);
2221         if (!btrfs_trans_handle_cachep)
2222                 goto fail;
2223         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2224                                              sizeof(struct btrfs_transaction),
2225                                              0, NULL);
2226         if (!btrfs_transaction_cachep)
2227                 goto fail;
2228         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2229                                          sizeof(struct btrfs_path),
2230                                          0, NULL);
2231         if (!btrfs_path_cachep)
2232                 goto fail;
2233         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2234                                               SLAB_DESTROY_BY_RCU, NULL);
2235         if (!btrfs_bit_radix_cachep)
2236                 goto fail;
2237         return 0;
2238 fail:
2239         btrfs_destroy_cachep();
2240         return -ENOMEM;
2241 }
2242
2243 static int btrfs_getattr(struct vfsmount *mnt,
2244                          struct dentry *dentry, struct kstat *stat)
2245 {
2246         struct inode *inode = dentry->d_inode;
2247         generic_fillattr(inode, stat);
2248         stat->blksize = 256 * 1024;
2249         return 0;
2250 }
2251
2252 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2253                            struct inode * new_dir,struct dentry *new_dentry)
2254 {
2255         struct btrfs_trans_handle *trans;
2256         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2257         struct inode *new_inode = new_dentry->d_inode;
2258         struct inode *old_inode = old_dentry->d_inode;
2259         struct timespec ctime = CURRENT_TIME;
2260         struct btrfs_path *path;
2261         struct btrfs_dir_item *di;
2262         int ret;
2263
2264         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2265             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2266                 return -ENOTEMPTY;
2267         }
2268
2269         mutex_lock(&root->fs_info->fs_mutex);
2270         trans = btrfs_start_transaction(root, 1);
2271
2272         btrfs_set_trans_block_group(trans, new_dir);
2273         path = btrfs_alloc_path();
2274         if (!path) {
2275                 ret = -ENOMEM;
2276                 goto out_fail;
2277         }
2278
2279         old_dentry->d_inode->i_nlink++;
2280         old_dir->i_ctime = old_dir->i_mtime = ctime;
2281         new_dir->i_ctime = new_dir->i_mtime = ctime;
2282         old_inode->i_ctime = ctime;
2283
2284         if (S_ISDIR(old_inode->i_mode) && old_dir != new_dir) {
2285                 struct btrfs_key *location = &BTRFS_I(new_dir)->location;
2286                 struct btrfs_key old_parent_key;
2287                 di = btrfs_lookup_dir_item(trans, root, path, old_inode->i_ino,
2288                                            "..", 2, -1);
2289                 if (IS_ERR(di)) {
2290                         ret = PTR_ERR(di);
2291                         goto out_fail;
2292                 }
2293                 if (!di) {
2294                         ret = -ENOENT;
2295                         goto out_fail;
2296                 }
2297                 btrfs_dir_item_key_to_cpu(path->nodes[0], di, &old_parent_key);
2298                 ret = btrfs_del_item(trans, root, path);
2299                 if (ret) {
2300                         goto out_fail;
2301                 }
2302                 btrfs_release_path(root, path);
2303
2304                 di = btrfs_lookup_dir_index_item(trans, root, path,
2305                                                  old_inode->i_ino,
2306                                                  old_parent_key.objectid,
2307                                                  "..", 2, -1);
2308                 if (IS_ERR(di)) {
2309                         ret = PTR_ERR(di);
2310                         goto out_fail;
2311                 }
2312                 if (!di) {
2313                         ret = -ENOENT;
2314                         goto out_fail;
2315                 }
2316                 ret = btrfs_del_item(trans, root, path);
2317                 if (ret) {
2318                         goto out_fail;
2319                 }
2320                 btrfs_release_path(root, path);
2321
2322                 ret = btrfs_insert_dir_item(trans, root, "..", 2,
2323                                             old_inode->i_ino, location,
2324                                             BTRFS_FT_DIR);
2325                 if (ret)
2326                         goto out_fail;
2327         }
2328
2329
2330         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2331         if (ret)
2332                 goto out_fail;
2333
2334         if (new_inode) {
2335                 new_inode->i_ctime = CURRENT_TIME;
2336                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2337                 if (ret)
2338                         goto out_fail;
2339         }
2340         ret = btrfs_add_link(trans, new_dentry, old_inode);
2341         if (ret)
2342                 goto out_fail;
2343
2344 out_fail:
2345         btrfs_free_path(path);
2346         btrfs_end_transaction(trans, root);
2347         mutex_unlock(&root->fs_info->fs_mutex);
2348         return ret;
2349 }
2350
2351 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2352                          const char *symname)
2353 {
2354         struct btrfs_trans_handle *trans;
2355         struct btrfs_root *root = BTRFS_I(dir)->root;
2356         struct btrfs_path *path;
2357         struct btrfs_key key;
2358         struct inode *inode;
2359         int err;
2360         int drop_inode = 0;
2361         u64 objectid;
2362         int name_len;
2363         int datasize;
2364         unsigned long ptr;
2365         struct btrfs_file_extent_item *ei;
2366         struct extent_buffer *leaf;
2367         unsigned long nr;
2368
2369         name_len = strlen(symname) + 1;
2370         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2371                 return -ENAMETOOLONG;
2372         mutex_lock(&root->fs_info->fs_mutex);
2373         trans = btrfs_start_transaction(root, 1);
2374         btrfs_set_trans_block_group(trans, dir);
2375
2376         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2377         if (err) {
2378                 err = -ENOSPC;
2379                 goto out_unlock;
2380         }
2381
2382         inode = btrfs_new_inode(trans, root, objectid,
2383                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2384         err = PTR_ERR(inode);
2385         if (IS_ERR(inode))
2386                 goto out_unlock;
2387
2388         btrfs_set_trans_block_group(trans, inode);
2389         err = btrfs_add_nondir(trans, dentry, inode);
2390         if (err)
2391                 drop_inode = 1;
2392         else {
2393                 inode->i_mapping->a_ops = &btrfs_aops;
2394                 inode->i_fop = &btrfs_file_operations;
2395                 inode->i_op = &btrfs_file_inode_operations;
2396                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2397                                      inode->i_mapping, GFP_NOFS);
2398                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2399         }
2400         dir->i_sb->s_dirt = 1;
2401         btrfs_update_inode_block_group(trans, inode);
2402         btrfs_update_inode_block_group(trans, dir);
2403         if (drop_inode)
2404                 goto out_unlock;
2405
2406         path = btrfs_alloc_path();
2407         BUG_ON(!path);
2408         key.objectid = inode->i_ino;
2409         key.offset = 0;
2410         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2411         datasize = btrfs_file_extent_calc_inline_size(name_len);
2412         err = btrfs_insert_empty_item(trans, root, path, &key,
2413                                       datasize);
2414         if (err) {
2415                 drop_inode = 1;
2416                 goto out_unlock;
2417         }
2418         leaf = path->nodes[0];
2419         ei = btrfs_item_ptr(leaf, path->slots[0],
2420                             struct btrfs_file_extent_item);
2421         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2422         btrfs_set_file_extent_type(leaf, ei,
2423                                    BTRFS_FILE_EXTENT_INLINE);
2424         ptr = btrfs_file_extent_inline_start(ei);
2425         write_extent_buffer(leaf, symname, ptr, name_len);
2426         btrfs_mark_buffer_dirty(leaf);
2427         btrfs_free_path(path);
2428
2429         inode->i_op = &btrfs_symlink_inode_operations;
2430         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2431         inode->i_size = name_len - 1;
2432         err = btrfs_update_inode(trans, root, inode);
2433         if (err)
2434                 drop_inode = 1;
2435
2436 out_unlock:
2437         nr = trans->blocks_used;
2438         btrfs_end_transaction(trans, root);
2439         mutex_unlock(&root->fs_info->fs_mutex);
2440         if (drop_inode) {
2441                 inode_dec_link_count(inode);
2442                 iput(inode);
2443         }
2444         btrfs_btree_balance_dirty(root, nr);
2445         return err;
2446 }
2447
2448 static struct inode_operations btrfs_dir_inode_operations = {
2449         .lookup         = btrfs_lookup,
2450         .create         = btrfs_create,
2451         .unlink         = btrfs_unlink,
2452         .link           = btrfs_link,
2453         .mkdir          = btrfs_mkdir,
2454         .rmdir          = btrfs_rmdir,
2455         .rename         = btrfs_rename,
2456         .symlink        = btrfs_symlink,
2457         .setattr        = btrfs_setattr,
2458         .mknod          = btrfs_mknod,
2459 };
2460
2461 static struct inode_operations btrfs_dir_ro_inode_operations = {
2462         .lookup         = btrfs_lookup,
2463 };
2464
2465 static struct file_operations btrfs_dir_file_operations = {
2466         .llseek         = generic_file_llseek,
2467         .read           = generic_read_dir,
2468         .readdir        = btrfs_readdir,
2469         .unlocked_ioctl = btrfs_ioctl,
2470 #ifdef CONFIG_COMPAT
2471         .compat_ioctl   = btrfs_ioctl,
2472 #endif
2473 };
2474
2475 static struct extent_map_ops btrfs_extent_map_ops = {
2476         .fill_delalloc = run_delalloc_range,
2477         .writepage_io_hook = btrfs_writepage_io_hook,
2478         .readpage_io_hook = btrfs_readpage_io_hook,
2479         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2480 };
2481
2482 static struct address_space_operations btrfs_aops = {
2483         .readpage       = btrfs_readpage,
2484         .writepage      = btrfs_writepage,
2485         .sync_page      = block_sync_page,
2486         .prepare_write  = btrfs_prepare_write,
2487         .commit_write   = btrfs_commit_write,
2488         .bmap           = btrfs_bmap,
2489         .invalidatepage = btrfs_invalidatepage,
2490         .releasepage    = btrfs_releasepage,
2491         .set_page_dirty = __set_page_dirty_nobuffers,
2492 };
2493
2494 static struct address_space_operations btrfs_symlink_aops = {
2495         .readpage       = btrfs_readpage,
2496         .writepage      = btrfs_writepage,
2497         .invalidatepage = btrfs_invalidatepage,
2498         .releasepage    = btrfs_releasepage,
2499 };
2500
2501 static struct inode_operations btrfs_file_inode_operations = {
2502         .truncate       = btrfs_truncate,
2503         .getattr        = btrfs_getattr,
2504         .setattr        = btrfs_setattr,
2505 };
2506
2507 static struct inode_operations btrfs_special_inode_operations = {
2508         .getattr        = btrfs_getattr,
2509         .setattr        = btrfs_setattr,
2510 };
2511
2512 static struct inode_operations btrfs_symlink_inode_operations = {
2513         .readlink       = generic_readlink,
2514         .follow_link    = page_follow_link_light,
2515         .put_link       = page_put_link,
2516 };