SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 <linux/xattr.h>
  36 #include "ctree.h"
  37 #include "disk-io.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "ioctl.h"
  41 #include "print-tree.h"
  42
  43 struct btrfs_iget_args {
  44         u64 ino;
  45         struct btrfs_root *root;
  46 };
  47
  48 static struct inode_operations btrfs_dir_inode_operations;
  49 static struct inode_operations btrfs_symlink_inode_operations;
  50 static struct inode_operations btrfs_dir_ro_inode_operations;
  51 static struct inode_operations btrfs_special_inode_operations;
  52 static struct inode_operations btrfs_file_inode_operations;
  53 static struct address_space_operations btrfs_aops;
  54 static struct address_space_operations btrfs_symlink_aops;
  55 static struct file_operations btrfs_dir_file_operations;
  56 static struct extent_map_ops btrfs_extent_map_ops;
  57
  58 static struct kmem_cache *btrfs_inode_cachep;
  59 struct kmem_cache *btrfs_trans_handle_cachep;
  60 struct kmem_cache *btrfs_transaction_cachep;
  61 struct kmem_cache *btrfs_bit_radix_cachep;
  62 struct kmem_cache *btrfs_path_cachep;
  63
  64 #define S_SHIFT 12
  65 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  66         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  67         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  68         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  69         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  70         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  71         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  72         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  73 };
  74
  75 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  76                            int for_del)
  77 {
  78         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  79         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  80         u64 thresh;
  81         int ret = 0;
  82
  83         return 0;
  84
  85         if (for_del)
  86                 thresh = total * 90;
  87         else
  88                 thresh = total * 85;
  89
  90         do_div(thresh, 100);
  91
  92         spin_lock(&root->fs_info->delalloc_lock);
  93         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  94                 ret = -ENOSPC;
  95         spin_unlock(&root->fs_info->delalloc_lock);
  96         return ret;
  97 }
  98
  99 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 100 {
 101         struct btrfs_root *root = BTRFS_I(inode)->root;
 102         struct btrfs_trans_handle *trans;
 103         u64 alloc_hint = 0;
 104         u64 num_bytes;
 105         u64 cur_alloc_size;
 106         u64 blocksize = root->sectorsize;
 107         struct btrfs_key ins;
 108         int ret;
 109
 110         trans = btrfs_start_transaction(root, 1);
 111         BUG_ON(!trans);
 112         btrfs_set_trans_block_group(trans, inode);
 113
 114         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 115         num_bytes = max(blocksize,  num_bytes);
 116         ret = btrfs_drop_extents(trans, root, inode,
 117                                  start, start + num_bytes, start, &alloc_hint);
 118
 119         if (alloc_hint == EXTENT_MAP_INLINE)
 120                 goto out;
 121
 122         while(num_bytes > 0) {
 123                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 124                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 125                                          root->root_key.objectid,
 126                                          trans->transid,
 127                                          inode->i_ino, start, 0,
 128                                          alloc_hint, (u64)-1, &ins, 1);
 129                 if (ret) {
 130                         WARN_ON(1);
 131                         goto out;
 132                 }
 133                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 134                                                start, ins.objectid, ins.offset,
 135                                                ins.offset);
 136                 num_bytes -= cur_alloc_size;
 137                 alloc_hint = ins.objectid + ins.offset;
 138                 start += cur_alloc_size;
 139         }
 140         btrfs_add_ordered_inode(inode);
 141 out:
 142         btrfs_end_transaction(trans, root);
 143         return ret;
 144 }
 145
 146 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 147 {
 148         u64 extent_start;
 149         u64 extent_end;
 150         u64 bytenr;
 151         u64 cow_end;
 152         u64 loops = 0;
 153         u64 total_fs_bytes;
 154         struct btrfs_root *root = BTRFS_I(inode)->root;
 155         struct extent_buffer *leaf;
 156         int found_type;
 157         struct btrfs_path *path;
 158         struct btrfs_file_extent_item *item;
 159         int ret;
 160         int err;
 161         struct btrfs_key found_key;
 162
 163         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 164         path = btrfs_alloc_path();
 165         BUG_ON(!path);
 166 again:
 167         ret = btrfs_lookup_file_extent(NULL, root, path,
 168                                        inode->i_ino, start, 0);
 169         if (ret < 0) {
 170                 btrfs_free_path(path);
 171                 return ret;
 172         }
 173
 174         cow_end = end;
 175         if (ret != 0) {
 176                 if (path->slots[0] == 0)
 177                         goto not_found;
 178                 path->slots[0]--;
 179         }
 180
 181         leaf = path->nodes[0];
 182         item = btrfs_item_ptr(leaf, path->slots[0],
 183                               struct btrfs_file_extent_item);
 184
 185         /* are we inside the extent that was found? */
 186         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 187         found_type = btrfs_key_type(&found_key);
 188         if (found_key.objectid != inode->i_ino ||
 189             found_type != BTRFS_EXTENT_DATA_KEY) {
 190                 goto not_found;
 191         }
 192
 193         found_type = btrfs_file_extent_type(leaf, item);
 194         extent_start = found_key.offset;
 195         if (found_type == BTRFS_FILE_EXTENT_REG) {
 196                 u64 extent_num_bytes;
 197
 198                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 199                 extent_end = extent_start + extent_num_bytes;
 200                 err = 0;
 201
 202                 if (loops && start != extent_start)
 203                         goto not_found;
 204
 205                 if (start < extent_start || start >= extent_end)
 206                         goto not_found;
 207
 208                 cow_end = min(end, extent_end - 1);
 209                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 210                 if (bytenr == 0)
 211                         goto not_found;
 212
 213                 /*
 214                  * we may be called by the resizer, make sure we're inside
 215                  * the limits of the FS
 216                  */
 217                 if (bytenr + extent_num_bytes > total_fs_bytes)
 218                         goto not_found;
 219
 220                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 221                         goto not_found;
 222                 }
 223
 224                 start = extent_end;
 225         } else {
 226                 goto not_found;
 227         }
 228 loop:
 229         if (start > end) {
 230                 btrfs_free_path(path);
 231                 return 0;
 232         }
 233         btrfs_release_path(root, path);
 234         loops++;
 235         goto again;
 236
 237 not_found:
 238         cow_file_range(inode, start, cow_end);
 239         start = cow_end + 1;
 240         goto loop;
 241 }
 242
 243 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 244 {
 245         struct btrfs_root *root = BTRFS_I(inode)->root;
 246         u64 num_bytes;
 247         int ret;
 248         mutex_lock(&root->fs_info->fs_mutex);
 249         if (btrfs_test_opt(root, NODATACOW) ||
 250             btrfs_test_flag(inode, NODATACOW))
 251                 ret = run_delalloc_nocow(inode, start, end);
 252         else
 253                 ret = cow_file_range(inode, start, end);
 254
 255         spin_lock(&root->fs_info->delalloc_lock);
 256         num_bytes = end + 1 - start;
 257         if (root->fs_info->delalloc_bytes < num_bytes) {
 258                 printk("delalloc accounting error total %llu sub %llu\n",
 259                        root->fs_info->delalloc_bytes, num_bytes);
 260         } else {
 261                 root->fs_info->delalloc_bytes -= num_bytes;
 262         }
 263         spin_unlock(&root->fs_info->delalloc_lock);
 264
 265         mutex_unlock(&root->fs_info->fs_mutex);
 266         return ret;
 267 }
 268
 269 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 270 {
 271         struct inode *inode = page->mapping->host;
 272         struct btrfs_root *root = BTRFS_I(inode)->root;
 273         struct btrfs_trans_handle *trans;
 274         char *kaddr;
 275         int ret = 0;
 276         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 277         size_t offset = start - page_start;
 278         if (btrfs_test_opt(root, NODATASUM) ||
 279             btrfs_test_flag(inode, NODATASUM))
 280                 return 0;
 281         mutex_lock(&root->fs_info->fs_mutex);
 282         trans = btrfs_start_transaction(root, 1);
 283         btrfs_set_trans_block_group(trans, inode);
 284         kaddr = kmap(page);
 285         btrfs_csum_file_block(trans, root, inode, inode->i_ino,
 286                               start, kaddr + offset, end - start + 1);
 287         kunmap(page);
 288         ret = btrfs_end_transaction(trans, root);
 289         BUG_ON(ret);
 290         mutex_unlock(&root->fs_info->fs_mutex);
 291         return ret;
 292 }
 293
 294 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 295 {
 296         int ret = 0;
 297         struct inode *inode = page->mapping->host;
 298         struct btrfs_root *root = BTRFS_I(inode)->root;
 299         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 300         struct btrfs_csum_item *item;
 301         struct btrfs_path *path = NULL;
 302         u32 csum;
 303         if (btrfs_test_opt(root, NODATASUM) ||
 304             btrfs_test_flag(inode, NODATASUM))
 305                 return 0;
 306         mutex_lock(&root->fs_info->fs_mutex);
 307         path = btrfs_alloc_path();
 308         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 309         if (IS_ERR(item)) {
 310                 ret = PTR_ERR(item);
 311                 /* a csum that isn't present is a preallocated region. */
 312                 if (ret == -ENOENT || ret == -EFBIG)
 313                         ret = 0;
 314                 csum = 0;
 315                 goto out;
 316         }
 317         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 318                            BTRFS_CRC32_SIZE);
 319         set_state_private(em_tree, start, csum);
 320 out:
 321         if (path)
 322                 btrfs_free_path(path);
 323         mutex_unlock(&root->fs_info->fs_mutex);
 324         return ret;
 325 }
 326
 327 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end)
 328 {
 329         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 330         struct inode *inode = page->mapping->host;
 331         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 332         char *kaddr;
 333         u64 private;
 334         int ret;
 335         struct btrfs_root *root = BTRFS_I(inode)->root;
 336         u32 csum = ~(u32)0;
 337         unsigned long flags;
 338         if (btrfs_test_opt(root, NODATASUM) ||
 339             btrfs_test_flag(inode, NODATASUM))
 340                 return 0;
 341         ret = get_state_private(em_tree, start, &private);
 342         local_irq_save(flags);
 343         kaddr = kmap_atomic(page, KM_IRQ0);
 344         if (ret) {
 345                 goto zeroit;
 346         }
 347         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 348         btrfs_csum_final(csum, (char *)&csum);
 349         if (csum != private) {
 350                 goto zeroit;
 351         }
 352         kunmap_atomic(kaddr, KM_IRQ0);
 353         local_irq_restore(flags);
 354         return 0;
 355
 356 zeroit:
 357         printk("btrfs csum failed ino %lu off %llu\n",
 358                page->mapping->host->i_ino, (unsigned long long)start);
 359         memset(kaddr + offset, 1, end - start + 1);
 360         flush_dcache_page(page);
 361         kunmap_atomic(kaddr, KM_IRQ0);
 362         local_irq_restore(flags);
 363         return 0;
 364 }
 365
 366 void btrfs_read_locked_inode(struct inode *inode)
 367 {
 368         struct btrfs_path *path;
 369         struct extent_buffer *leaf;
 370         struct btrfs_inode_item *inode_item;
 371         struct btrfs_inode_timespec *tspec;
 372         struct btrfs_root *root = BTRFS_I(inode)->root;
 373         struct btrfs_key location;
 374         u64 alloc_group_block;
 375         u32 rdev;
 376         int ret;
 377
 378         path = btrfs_alloc_path();
 379         BUG_ON(!path);
 380         mutex_lock(&root->fs_info->fs_mutex);
 381         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 382
 383         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 384         if (ret)
 385                 goto make_bad;
 386
 387         leaf = path->nodes[0];
 388         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 389                                     struct btrfs_inode_item);
 390
 391         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 392         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 393         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 394         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 395         inode->i_size = btrfs_inode_size(leaf, inode_item);
 396
 397         tspec = btrfs_inode_atime(inode_item);
 398         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 399         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 400
 401         tspec = btrfs_inode_mtime(inode_item);
 402         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 403         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 404
 405         tspec = btrfs_inode_ctime(inode_item);
 406         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 407         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 408
 409         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 410         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 411         inode->i_rdev = 0;
 412         rdev = btrfs_inode_rdev(leaf, inode_item);
 413
 414         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 415         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 416                                                        alloc_group_block);
 417         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 418         if (!BTRFS_I(inode)->block_group) {
 419                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 420                                                          NULL, 0, 0, 0);
 421         }
 422         btrfs_free_path(path);
 423         inode_item = NULL;
 424
 425         mutex_unlock(&root->fs_info->fs_mutex);
 426
 427         switch (inode->i_mode & S_IFMT) {
 428         case S_IFREG:
 429                 inode->i_mapping->a_ops = &btrfs_aops;
 430                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
 431                 inode->i_fop = &btrfs_file_operations;
 432                 inode->i_op = &btrfs_file_inode_operations;
 433                 break;
 434         case S_IFDIR:
 435                 inode->i_fop = &btrfs_dir_file_operations;
 436                 if (root == root->fs_info->tree_root)
 437                         inode->i_op = &btrfs_dir_ro_inode_operations;
 438                 else
 439                         inode->i_op = &btrfs_dir_inode_operations;
 440                 break;
 441         case S_IFLNK:
 442                 inode->i_op = &btrfs_symlink_inode_operations;
 443                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 444                 break;
 445         default:
 446                 init_special_inode(inode, inode->i_mode, rdev);
 447                 break;
 448         }
 449         return;
 450
 451 make_bad:
 452         btrfs_release_path(root, path);
 453         btrfs_free_path(path);
 454         mutex_unlock(&root->fs_info->fs_mutex);
 455         make_bad_inode(inode);
 456 }
 457
 458 static void fill_inode_item(struct extent_buffer *leaf,
 459                             struct btrfs_inode_item *item,
 460                             struct inode *inode)
 461 {
 462         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 463         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 464         btrfs_set_inode_size(leaf, item, inode->i_size);
 465         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 466         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 467
 468         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 469                                inode->i_atime.tv_sec);
 470         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 471                                 inode->i_atime.tv_nsec);
 472
 473         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 474                                inode->i_mtime.tv_sec);
 475         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 476                                 inode->i_mtime.tv_nsec);
 477
 478         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 479                                inode->i_ctime.tv_sec);
 480         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 481                                 inode->i_ctime.tv_nsec);
 482
 483         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 484         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 485         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 486         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 487         btrfs_set_inode_block_group(leaf, item,
 488                                     BTRFS_I(inode)->block_group->key.objectid);
 489 }
 490
 491 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 492                               struct btrfs_root *root,
 493                               struct inode *inode)
 494 {
 495         struct btrfs_inode_item *inode_item;
 496         struct btrfs_path *path;
 497         struct extent_buffer *leaf;
 498         int ret;
 499
 500         path = btrfs_alloc_path();
 501         BUG_ON(!path);
 502         ret = btrfs_lookup_inode(trans, root, path,
 503                                  &BTRFS_I(inode)->location, 1);
 504         if (ret) {
 505                 if (ret > 0)
 506                         ret = -ENOENT;
 507                 goto failed;
 508         }
 509
 510         leaf = path->nodes[0];
 511         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 512                                   struct btrfs_inode_item);
 513
 514         fill_inode_item(leaf, inode_item, inode);
 515         btrfs_mark_buffer_dirty(leaf);
 516         btrfs_set_inode_last_trans(trans, inode);
 517         ret = 0;
 518 failed:
 519         btrfs_release_path(root, path);
 520         btrfs_free_path(path);
 521         return ret;
 522 }
 523
 524
 525 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 526                               struct btrfs_root *root,
 527                               struct inode *dir,
 528                               struct dentry *dentry)
 529 {
 530         struct btrfs_path *path;
 531         const char *name = dentry->d_name.name;
 532         int name_len = dentry->d_name.len;
 533         int ret = 0;
 534         struct extent_buffer *leaf;
 535         struct btrfs_dir_item *di;
 536         struct btrfs_key key;
 537
 538         path = btrfs_alloc_path();
 539         if (!path) {
 540                 ret = -ENOMEM;
 541                 goto err;
 542         }
 543
 544         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 545                                     name, name_len, -1);
 546         if (IS_ERR(di)) {
 547                 ret = PTR_ERR(di);
 548                 goto err;
 549         }
 550         if (!di) {
 551                 ret = -ENOENT;
 552                 goto err;
 553         }
 554         leaf = path->nodes[0];
 555         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 556         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 557         if (ret)
 558                 goto err;
 559         btrfs_release_path(root, path);
 560
 561         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 562                                          key.objectid, name, name_len, -1);
 563         if (IS_ERR(di)) {
 564                 ret = PTR_ERR(di);
 565                 goto err;
 566         }
 567         if (!di) {
 568                 ret = -ENOENT;
 569                 goto err;
 570         }
 571         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 572
 573         dentry->d_inode->i_ctime = dir->i_ctime;
 574         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 575                                   dentry->d_inode->i_ino,
 576                                   dentry->d_parent->d_inode->i_ino);
 577         if (ret) {
 578                 printk("failed to delete reference to %.*s, "
 579                        "inode %lu parent %lu\n", name_len, name,
 580                        dentry->d_inode->i_ino,
 581                        dentry->d_parent->d_inode->i_ino);
 582         }
 583 err:
 584         btrfs_free_path(path);
 585         if (!ret) {
 586                 dir->i_size -= name_len * 2;
 587                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 588                 btrfs_update_inode(trans, root, dir);
 589 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 590                 dentry->d_inode->i_nlink--;
 591 #else
 592                 drop_nlink(dentry->d_inode);
 593 #endif
 594                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 595                 dir->i_sb->s_dirt = 1;
 596         }
 597         return ret;
 598 }
 599
 600 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 601 {
 602         struct btrfs_root *root;
 603         struct btrfs_trans_handle *trans;
 604         struct inode *inode = dentry->d_inode;
 605         int ret;
 606         unsigned long nr = 0;
 607
 608         root = BTRFS_I(dir)->root;
 609         mutex_lock(&root->fs_info->fs_mutex);
 610
 611         ret = btrfs_check_free_space(root, 1, 1);
 612         if (ret)
 613                 goto fail;
 614
 615         trans = btrfs_start_transaction(root, 1);
 616
 617         btrfs_set_trans_block_group(trans, dir);
 618         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 619         nr = trans->blocks_used;
 620
 621         if (inode->i_nlink == 0) {
 622                 int found;
 623                 /* if the inode isn't linked anywhere,
 624                  * we don't need to worry about
 625                  * data=ordered
 626                  */
 627                 found = btrfs_del_ordered_inode(inode);
 628                 if (found == 1) {
 629                         atomic_dec(&inode->i_count);
 630                 }
 631         }
 632
 633         btrfs_end_transaction(trans, root);
 634 fail:
 635         mutex_unlock(&root->fs_info->fs_mutex);
 636         btrfs_btree_balance_dirty(root, nr);
 637         btrfs_throttle(root);
 638         return ret;
 639 }
 640
 641 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 642 {
 643         struct inode *inode = dentry->d_inode;
 644         int err = 0;
 645         int ret;
 646         struct btrfs_root *root = BTRFS_I(dir)->root;
 647         struct btrfs_trans_handle *trans;
 648         unsigned long nr = 0;
 649
 650         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 651                 return -ENOTEMPTY;
 652
 653         mutex_lock(&root->fs_info->fs_mutex);
 654         ret = btrfs_check_free_space(root, 1, 1);
 655         if (ret)
 656                 goto fail;
 657
 658         trans = btrfs_start_transaction(root, 1);
 659         btrfs_set_trans_block_group(trans, dir);
 660
 661         /* now the directory is empty */
 662         err = btrfs_unlink_trans(trans, root, dir, dentry);
 663         if (!err) {
 664                 inode->i_size = 0;
 665         }
 666
 667         nr = trans->blocks_used;
 668         ret = btrfs_end_transaction(trans, root);
 669 fail:
 670         mutex_unlock(&root->fs_info->fs_mutex);
 671         btrfs_btree_balance_dirty(root, nr);
 672         btrfs_throttle(root);
 673
 674         if (ret && !err)
 675                 err = ret;
 676         return err;
 677 }
 678
 679 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 680                             struct btrfs_root *root,
 681                             struct inode *inode)
 682 {
 683         struct btrfs_path *path;
 684         int ret;
 685
 686         clear_inode(inode);
 687
 688         path = btrfs_alloc_path();
 689         BUG_ON(!path);
 690         ret = btrfs_lookup_inode(trans, root, path,
 691                                  &BTRFS_I(inode)->location, -1);
 692         if (ret > 0)
 693                 ret = -ENOENT;
 694         if (!ret)
 695                 ret = btrfs_del_item(trans, root, path);
 696         btrfs_free_path(path);
 697         return ret;
 698 }
 699
 700 /*
 701  * this can truncate away extent items, csum items and directory items.
 702  * It starts at a high offset and removes keys until it can't find
 703  * any higher than i_size.
 704  *
 705  * csum items that cross the new i_size are truncated to the new size
 706  * as well.
 707  */
 708 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 709                                    struct btrfs_root *root,
 710                                    struct inode *inode)
 711 {
 712         int ret;
 713         struct btrfs_path *path;
 714         struct btrfs_key key;
 715         struct btrfs_key found_key;
 716         u32 found_type;
 717         struct extent_buffer *leaf;
 718         struct btrfs_file_extent_item *fi;
 719         u64 extent_start = 0;
 720         u64 extent_num_bytes = 0;
 721         u64 item_end = 0;
 722         u64 root_gen = 0;
 723         u64 root_owner = 0;
 724         int found_extent;
 725         int del_item;
 726         int extent_type = -1;
 727
 728         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 729         path = btrfs_alloc_path();
 730         path->reada = -1;
 731         BUG_ON(!path);
 732
 733         /* FIXME, add redo link to tree so we don't leak on crash */
 734         key.objectid = inode->i_ino;
 735         key.offset = (u64)-1;
 736         key.type = (u8)-1;
 737
 738         while(1) {
 739                 btrfs_init_path(path);
 740                 fi = NULL;
 741                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 742                 if (ret < 0) {
 743                         goto error;
 744                 }
 745                 if (ret > 0) {
 746                         BUG_ON(path->slots[0] == 0);
 747                         path->slots[0]--;
 748                 }
 749                 leaf = path->nodes[0];
 750                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 751                 found_type = btrfs_key_type(&found_key);
 752
 753                 if (found_key.objectid != inode->i_ino)
 754                         break;
 755
 756                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 757                     found_type != BTRFS_DIR_ITEM_KEY &&
 758                     found_type != BTRFS_DIR_INDEX_KEY &&
 759                     found_type != BTRFS_EXTENT_DATA_KEY)
 760                         break;
 761
 762                 item_end = found_key.offset;
 763                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 764                         fi = btrfs_item_ptr(leaf, path->slots[0],
 765                                             struct btrfs_file_extent_item);
 766                         extent_type = btrfs_file_extent_type(leaf, fi);
 767                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 768                                 item_end +=
 769                                     btrfs_file_extent_num_bytes(leaf, fi);
 770                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 771                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 772                                                                 path->slots[0]);
 773                                 item_end += btrfs_file_extent_inline_len(leaf,
 774                                                                          item);
 775                         }
 776                         item_end--;
 777                 }
 778                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 779                         ret = btrfs_csum_truncate(trans, root, path,
 780                                                   inode->i_size);
 781                         BUG_ON(ret);
 782                 }
 783                 if (item_end < inode->i_size) {
 784                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 785                                 found_type = BTRFS_INODE_ITEM_KEY;
 786                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 787                                 found_type = BTRFS_CSUM_ITEM_KEY;
 788                         } else if (found_type) {
 789                                 found_type--;
 790                         } else {
 791                                 break;
 792                         }
 793                         btrfs_set_key_type(&key, found_type);
 794                         btrfs_release_path(root, path);
 795                         continue;
 796                 }
 797                 if (found_key.offset >= inode->i_size)
 798                         del_item = 1;
 799                 else
 800                         del_item = 0;
 801                 found_extent = 0;
 802
 803                 /* FIXME, shrink the extent if the ref count is only 1 */
 804                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 805                         goto delete;
 806
 807                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 808                         u64 num_dec;
 809                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 810                         if (!del_item) {
 811                                 u64 orig_num_bytes =
 812                                         btrfs_file_extent_num_bytes(leaf, fi);
 813                                 extent_num_bytes = inode->i_size -
 814                                         found_key.offset + root->sectorsize - 1;
 815                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 816                                                          extent_num_bytes);
 817                                 num_dec = (orig_num_bytes -
 818                                            extent_num_bytes) >> 9;
 819                                 if (extent_start != 0) {
 820                                         inode->i_blocks -= num_dec;
 821                                 }
 822                                 btrfs_mark_buffer_dirty(leaf);
 823                         } else {
 824                                 extent_num_bytes =
 825                                         btrfs_file_extent_disk_num_bytes(leaf,
 826                                                                          fi);
 827                                 /* FIXME blocksize != 4096 */
 828                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 829                                                                        fi) >> 9;
 830                                 if (extent_start != 0) {
 831                                         found_extent = 1;
 832                                         inode->i_blocks -= num_dec;
 833                                 }
 834                                 root_gen = btrfs_header_generation(leaf);
 835                                 root_owner = btrfs_header_owner(leaf);
 836                         }
 837                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 838                            !del_item) {
 839                         u32 newsize = inode->i_size - found_key.offset;
 840                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 841                         ret = btrfs_truncate_item(trans, root, path,
 842                                                   newsize, 1);
 843                         BUG_ON(ret);
 844                 }
 845 delete:
 846                 if (del_item) {
 847                         ret = btrfs_del_item(trans, root, path);
 848                         if (ret)
 849                                 goto error;
 850                 } else {
 851                         break;
 852                 }
 853                 btrfs_release_path(root, path);
 854                 if (found_extent) {
 855                         ret = btrfs_free_extent(trans, root, extent_start,
 856                                                 extent_num_bytes,
 857                                                 root_owner,
 858                                                 root_gen, inode->i_ino,
 859                                                 found_key.offset, 0);
 860                         BUG_ON(ret);
 861                 }
 862         }
 863         ret = 0;
 864 error:
 865         btrfs_release_path(root, path);
 866         btrfs_free_path(path);
 867         inode->i_sb->s_dirt = 1;
 868         return ret;
 869 }
 870
 871 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 872                               size_t zero_start)
 873 {
 874         char *kaddr;
 875         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 876         struct btrfs_root *root = BTRFS_I(inode)->root;
 877         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 878         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 879         u64 existing_delalloc;
 880         u64 delalloc_start;
 881         int ret = 0;
 882
 883         WARN_ON(!PageLocked(page));
 884         set_page_extent_mapped(page);
 885
 886         lock_extent(em_tree, page_start, page_end, GFP_NOFS);
 887         delalloc_start = page_start;
 888         existing_delalloc = count_range_bits(&BTRFS_I(inode)->extent_tree,
 889                                              &delalloc_start, page_end,
 890                                              PAGE_CACHE_SIZE, EXTENT_DELALLOC);
 891         set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
 892                             page_end, GFP_NOFS);
 893
 894         spin_lock(&root->fs_info->delalloc_lock);
 895         root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE - existing_delalloc;
 896         spin_unlock(&root->fs_info->delalloc_lock);
 897
 898         if (zero_start != PAGE_CACHE_SIZE) {
 899                 kaddr = kmap(page);
 900                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 901                 flush_dcache_page(page);
 902                 kunmap(page);
 903         }
 904         set_page_dirty(page);
 905         unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
 906
 907         return ret;
 908 }
 909
 910 /*
 911  * taken from block_truncate_page, but does cow as it zeros out
 912  * any bytes left in the last page in the file.
 913  */
 914 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 915 {
 916         struct inode *inode = mapping->host;
 917         struct btrfs_root *root = BTRFS_I(inode)->root;
 918         u32 blocksize = root->sectorsize;
 919         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 920         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 921         struct page *page;
 922         int ret = 0;
 923         u64 page_start;
 924
 925         if ((offset & (blocksize - 1)) == 0)
 926                 goto out;
 927
 928         ret = -ENOMEM;
 929         page = grab_cache_page(mapping, index);
 930         if (!page)
 931                 goto out;
 932         if (!PageUptodate(page)) {
 933                 ret = btrfs_readpage(NULL, page);
 934                 lock_page(page);
 935                 if (!PageUptodate(page)) {
 936                         ret = -EIO;
 937                         goto out;
 938                 }
 939         }
 940         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 941
 942         ret = btrfs_cow_one_page(inode, page, offset);
 943
 944         unlock_page(page);
 945         page_cache_release(page);
 946 out:
 947         return ret;
 948 }
 949
 950 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 951 {
 952         struct inode *inode = dentry->d_inode;
 953         int err;
 954
 955         err = inode_change_ok(inode, attr);
 956         if (err)
 957                 return err;
 958
 959         if (S_ISREG(inode->i_mode) &&
 960             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 961                 struct btrfs_trans_handle *trans;
 962                 struct btrfs_root *root = BTRFS_I(inode)->root;
 963                 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 964
 965                 u64 mask = root->sectorsize - 1;
 966                 u64 pos = (inode->i_size + mask) & ~mask;
 967                 u64 block_end = attr->ia_size | mask;
 968                 u64 hole_size;
 969                 u64 alloc_hint = 0;
 970
 971                 if (attr->ia_size <= pos)
 972                         goto out;
 973
 974                 mutex_lock(&root->fs_info->fs_mutex);
 975                 err = btrfs_check_free_space(root, 1, 0);
 976                 mutex_unlock(&root->fs_info->fs_mutex);
 977                 if (err)
 978                         goto fail;
 979
 980                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 981
 982                 lock_extent(em_tree, pos, block_end, GFP_NOFS);
 983                 hole_size = (attr->ia_size - pos + mask) & ~mask;
 984
 985                 mutex_lock(&root->fs_info->fs_mutex);
 986                 trans = btrfs_start_transaction(root, 1);
 987                 btrfs_set_trans_block_group(trans, inode);
 988                 err = btrfs_drop_extents(trans, root, inode,
 989                                          pos, pos + hole_size, pos,
 990                                          &alloc_hint);
 991
 992                 if (alloc_hint != EXTENT_MAP_INLINE) {
 993                         err = btrfs_insert_file_extent(trans, root,
 994                                                        inode->i_ino,
 995                                                        pos, 0, 0, hole_size);
 996                 }
 997                 btrfs_end_transaction(trans, root);
 998                 mutex_unlock(&root->fs_info->fs_mutex);
 999                 unlock_extent(em_tree, pos, block_end, GFP_NOFS);
1000                 if (err)
1001                         return err;
1002         }
1003 out:
1004         err = inode_setattr(inode, attr);
1005 fail:
1006         return err;
1007 }
1008
1009 void btrfs_put_inode(struct inode *inode)
1010 {
1011         int ret;
1012
1013         if (!BTRFS_I(inode)->ordered_trans) {
1014                 return;
1015         }
1016
1017         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1018             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1019                 return;
1020
1021         ret = btrfs_del_ordered_inode(inode);
1022         if (ret == 1) {
1023                 atomic_dec(&inode->i_count);
1024         }
1025 }
1026
1027 void btrfs_delete_inode(struct inode *inode)
1028 {
1029         struct btrfs_trans_handle *trans;
1030         struct btrfs_root *root = BTRFS_I(inode)->root;
1031         unsigned long nr;
1032         int ret;
1033
1034         truncate_inode_pages(&inode->i_data, 0);
1035         if (is_bad_inode(inode)) {
1036                 goto no_delete;
1037         }
1038
1039         inode->i_size = 0;
1040         mutex_lock(&root->fs_info->fs_mutex);
1041         trans = btrfs_start_transaction(root, 1);
1042
1043         btrfs_set_trans_block_group(trans, inode);
1044         ret = btrfs_truncate_in_trans(trans, root, inode);
1045         if (ret)
1046                 goto no_delete_lock;
1047         ret = btrfs_delete_xattrs(trans, root, inode);
1048         if (ret)
1049                 goto no_delete_lock;
1050         ret = btrfs_free_inode(trans, root, inode);
1051         if (ret)
1052                 goto no_delete_lock;
1053         nr = trans->blocks_used;
1054
1055         btrfs_end_transaction(trans, root);
1056         mutex_unlock(&root->fs_info->fs_mutex);
1057         btrfs_btree_balance_dirty(root, nr);
1058         btrfs_throttle(root);
1059         return;
1060
1061 no_delete_lock:
1062         nr = trans->blocks_used;
1063         btrfs_end_transaction(trans, root);
1064         mutex_unlock(&root->fs_info->fs_mutex);
1065         btrfs_btree_balance_dirty(root, nr);
1066         btrfs_throttle(root);
1067 no_delete:
1068         clear_inode(inode);
1069 }
1070
1071 /*
1072  * this returns the key found in the dir entry in the location pointer.
1073  * If no dir entries were found, location->objectid is 0.
1074  */
1075 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1076                                struct btrfs_key *location)
1077 {
1078         const char *name = dentry->d_name.name;
1079         int namelen = dentry->d_name.len;
1080         struct btrfs_dir_item *di;
1081         struct btrfs_path *path;
1082         struct btrfs_root *root = BTRFS_I(dir)->root;
1083         int ret = 0;
1084
1085         if (namelen == 1 && strcmp(name, ".") == 0) {
1086                 location->objectid = dir->i_ino;
1087                 location->type = BTRFS_INODE_ITEM_KEY;
1088                 location->offset = 0;
1089                 return 0;
1090         }
1091         path = btrfs_alloc_path();
1092         BUG_ON(!path);
1093
1094         if (namelen == 2 && strcmp(name, "..") == 0) {
1095                 struct btrfs_key key;
1096                 struct extent_buffer *leaf;
1097                 u32 nritems;
1098                 int slot;
1099
1100                 key.objectid = dir->i_ino;
1101                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1102                 key.offset = 0;
1103                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1104                 BUG_ON(ret == 0);
1105                 ret = 0;
1106
1107                 leaf = path->nodes[0];
1108                 slot = path->slots[0];
1109                 nritems = btrfs_header_nritems(leaf);
1110                 if (slot >= nritems)
1111                         goto out_err;
1112
1113                 btrfs_item_key_to_cpu(leaf, &key, slot);
1114                 if (key.objectid != dir->i_ino ||
1115                     key.type != BTRFS_INODE_REF_KEY) {
1116                         goto out_err;
1117                 }
1118                 location->objectid = key.offset;
1119                 location->type = BTRFS_INODE_ITEM_KEY;
1120                 location->offset = 0;
1121                 goto out;
1122         }
1123
1124         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1125                                     namelen, 0);
1126         if (IS_ERR(di))
1127                 ret = PTR_ERR(di);
1128         if (!di || IS_ERR(di)) {
1129                 goto out_err;
1130         }
1131         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1132 out:
1133         btrfs_free_path(path);
1134         return ret;
1135 out_err:
1136         location->objectid = 0;
1137         goto out;
1138 }
1139
1140 /*
1141  * when we hit a tree root in a directory, the btrfs part of the inode
1142  * needs to be changed to reflect the root directory of the tree root.  This
1143  * is kind of like crossing a mount point.
1144  */
1145 static int fixup_tree_root_location(struct btrfs_root *root,
1146                              struct btrfs_key *location,
1147                              struct btrfs_root **sub_root,
1148                              struct dentry *dentry)
1149 {
1150         struct btrfs_path *path;
1151         struct btrfs_root_item *ri;
1152
1153         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1154                 return 0;
1155         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1156                 return 0;
1157
1158         path = btrfs_alloc_path();
1159         BUG_ON(!path);
1160         mutex_lock(&root->fs_info->fs_mutex);
1161
1162         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1163                                         dentry->d_name.name,
1164                                         dentry->d_name.len);
1165         if (IS_ERR(*sub_root))
1166                 return PTR_ERR(*sub_root);
1167
1168         ri = &(*sub_root)->root_item;
1169         location->objectid = btrfs_root_dirid(ri);
1170         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1171         location->offset = 0;
1172
1173         btrfs_free_path(path);
1174         mutex_unlock(&root->fs_info->fs_mutex);
1175         return 0;
1176 }
1177
1178 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1179 {
1180         struct btrfs_iget_args *args = p;
1181         inode->i_ino = args->ino;
1182         BTRFS_I(inode)->root = args->root;
1183         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1184                              inode->i_mapping, GFP_NOFS);
1185         return 0;
1186 }
1187
1188 static int btrfs_find_actor(struct inode *inode, void *opaque)
1189 {
1190         struct btrfs_iget_args *args = opaque;
1191         return (args->ino == inode->i_ino &&
1192                 args->root == BTRFS_I(inode)->root);
1193 }
1194
1195 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1196                             u64 root_objectid)
1197 {
1198         struct btrfs_iget_args args;
1199         args.ino = objectid;
1200         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1201
1202         if (!args.root)
1203                 return NULL;
1204
1205         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1206 }
1207
1208 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1209                                 struct btrfs_root *root)
1210 {
1211         struct inode *inode;
1212         struct btrfs_iget_args args;
1213         args.ino = objectid;
1214         args.root = root;
1215
1216         inode = iget5_locked(s, objectid, btrfs_find_actor,
1217                              btrfs_init_locked_inode,
1218                              (void *)&args);
1219         return inode;
1220 }
1221
1222 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1223                                    struct nameidata *nd)
1224 {
1225         struct inode * inode;
1226         struct btrfs_inode *bi = BTRFS_I(dir);
1227         struct btrfs_root *root = bi->root;
1228         struct btrfs_root *sub_root = root;
1229         struct btrfs_key location;
1230         int ret;
1231
1232         if (dentry->d_name.len > BTRFS_NAME_LEN)
1233                 return ERR_PTR(-ENAMETOOLONG);
1234
1235         mutex_lock(&root->fs_info->fs_mutex);
1236         ret = btrfs_inode_by_name(dir, dentry, &location);
1237         mutex_unlock(&root->fs_info->fs_mutex);
1238
1239         if (ret < 0)
1240                 return ERR_PTR(ret);
1241
1242         inode = NULL;
1243         if (location.objectid) {
1244                 ret = fixup_tree_root_location(root, &location, &sub_root,
1245                                                 dentry);
1246                 if (ret < 0)
1247                         return ERR_PTR(ret);
1248                 if (ret > 0)
1249                         return ERR_PTR(-ENOENT);
1250                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1251                                           sub_root);
1252                 if (!inode)
1253                         return ERR_PTR(-EACCES);
1254                 if (inode->i_state & I_NEW) {
1255                         /* the inode and parent dir are two different roots */
1256                         if (sub_root != root) {
1257                                 igrab(inode);
1258                                 sub_root->inode = inode;
1259                         }
1260                         BTRFS_I(inode)->root = sub_root;
1261                         memcpy(&BTRFS_I(inode)->location, &location,
1262                                sizeof(location));
1263                         btrfs_read_locked_inode(inode);
1264                         unlock_new_inode(inode);
1265                 }
1266         }
1267         return d_splice_alias(inode, dentry);
1268 }
1269
1270 static unsigned char btrfs_filetype_table[] = {
1271         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1272 };
1273
1274 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1275 {
1276         struct inode *inode = filp->f_dentry->d_inode;
1277         struct btrfs_root *root = BTRFS_I(inode)->root;
1278         struct btrfs_item *item;
1279         struct btrfs_dir_item *di;
1280         struct btrfs_key key;
1281         struct btrfs_key found_key;
1282         struct btrfs_path *path;
1283         int ret;
1284         u32 nritems;
1285         struct extent_buffer *leaf;
1286         int slot;
1287         int advance;
1288         unsigned char d_type;
1289         int over = 0;
1290         u32 di_cur;
1291         u32 di_total;
1292         u32 di_len;
1293         int key_type = BTRFS_DIR_INDEX_KEY;
1294         char tmp_name[32];
1295         char *name_ptr;
1296         int name_len;
1297
1298         /* FIXME, use a real flag for deciding about the key type */
1299         if (root->fs_info->tree_root == root)
1300                 key_type = BTRFS_DIR_ITEM_KEY;
1301
1302         /* special case for "." */
1303         if (filp->f_pos == 0) {
1304                 over = filldir(dirent, ".", 1,
1305                                1, inode->i_ino,
1306                                DT_DIR);
1307                 if (over)
1308                         return 0;
1309                 filp->f_pos = 1;
1310         }
1311
1312         mutex_lock(&root->fs_info->fs_mutex);
1313         key.objectid = inode->i_ino;
1314         path = btrfs_alloc_path();
1315         path->reada = 2;
1316
1317         /* special case for .., just use the back ref */
1318         if (filp->f_pos == 1) {
1319                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1320                 key.offset = 0;
1321                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1322                 BUG_ON(ret == 0);
1323                 leaf = path->nodes[0];
1324                 slot = path->slots[0];
1325                 nritems = btrfs_header_nritems(leaf);
1326                 if (slot >= nritems) {
1327                         btrfs_release_path(root, path);
1328                         goto read_dir_items;
1329                 }
1330                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1331                 btrfs_release_path(root, path);
1332                 if (found_key.objectid != key.objectid ||
1333                     found_key.type != BTRFS_INODE_REF_KEY)
1334                         goto read_dir_items;
1335                 over = filldir(dirent, "..", 2,
1336                                2, found_key.offset, DT_DIR);
1337                 if (over)
1338                         goto nopos;
1339                 filp->f_pos = 2;
1340         }
1341
1342 read_dir_items:
1343         btrfs_set_key_type(&key, key_type);
1344         key.offset = filp->f_pos;
1345
1346         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1347         if (ret < 0)
1348                 goto err;
1349         advance = 0;
1350         while(1) {
1351                 leaf = path->nodes[0];
1352                 nritems = btrfs_header_nritems(leaf);
1353                 slot = path->slots[0];
1354                 if (advance || slot >= nritems) {
1355                         if (slot >= nritems -1) {
1356                                 ret = btrfs_next_leaf(root, path);
1357                                 if (ret)
1358                                         break;
1359                                 leaf = path->nodes[0];
1360                                 nritems = btrfs_header_nritems(leaf);
1361                                 slot = path->slots[0];
1362                         } else {
1363                                 slot++;
1364                                 path->slots[0]++;
1365                         }
1366                 }
1367                 advance = 1;
1368                 item = btrfs_item_nr(leaf, slot);
1369                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1370
1371                 if (found_key.objectid != key.objectid)
1372                         break;
1373                 if (btrfs_key_type(&found_key) != key_type)
1374                         break;
1375                 if (found_key.offset < filp->f_pos)
1376                         continue;
1377
1378                 filp->f_pos = found_key.offset;
1379                 advance = 1;
1380                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1381                 di_cur = 0;
1382                 di_total = btrfs_item_size(leaf, item);
1383                 while(di_cur < di_total) {
1384                         struct btrfs_key location;
1385
1386                         name_len = btrfs_dir_name_len(leaf, di);
1387                         if (name_len < 32) {
1388                                 name_ptr = tmp_name;
1389                         } else {
1390                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1391                                 BUG_ON(!name_ptr);
1392                         }
1393                         read_extent_buffer(leaf, name_ptr,
1394                                            (unsigned long)(di + 1), name_len);
1395
1396                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1397                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1398                         over = filldir(dirent, name_ptr, name_len,
1399                                        found_key.offset,
1400                                        location.objectid,
1401                                        d_type);
1402
1403                         if (name_ptr != tmp_name)
1404                                 kfree(name_ptr);
1405
1406                         if (over)
1407                                 goto nopos;
1408                         di_len = btrfs_dir_name_len(leaf, di) +
1409                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1410                         di_cur += di_len;
1411                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1412                 }
1413         }
1414         filp->f_pos++;
1415 nopos:
1416         ret = 0;
1417 err:
1418         btrfs_release_path(root, path);
1419         btrfs_free_path(path);
1420         mutex_unlock(&root->fs_info->fs_mutex);
1421         return ret;
1422 }
1423
1424 int btrfs_write_inode(struct inode *inode, int wait)
1425 {
1426         struct btrfs_root *root = BTRFS_I(inode)->root;
1427         struct btrfs_trans_handle *trans;
1428         int ret = 0;
1429
1430         if (wait) {
1431                 mutex_lock(&root->fs_info->fs_mutex);
1432                 trans = btrfs_start_transaction(root, 1);
1433                 btrfs_set_trans_block_group(trans, inode);
1434                 ret = btrfs_commit_transaction(trans, root);
1435                 mutex_unlock(&root->fs_info->fs_mutex);
1436         }
1437         return ret;
1438 }
1439
1440 /*
1441  * This is somewhat expensive, updating the tree every time the
1442  * inode changes.  But, it is most likely to find the inode in cache.
1443  * FIXME, needs more benchmarking...there are no reasons other than performance
1444  * to keep or drop this code.
1445  */
1446 void btrfs_dirty_inode(struct inode *inode)
1447 {
1448         struct btrfs_root *root = BTRFS_I(inode)->root;
1449         struct btrfs_trans_handle *trans;
1450
1451         mutex_lock(&root->fs_info->fs_mutex);
1452         trans = btrfs_start_transaction(root, 1);
1453         btrfs_set_trans_block_group(trans, inode);
1454         btrfs_update_inode(trans, root, inode);
1455         btrfs_end_transaction(trans, root);
1456         mutex_unlock(&root->fs_info->fs_mutex);
1457 }
1458
1459 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1460                                      struct btrfs_root *root,
1461                                      u64 objectid,
1462                                      struct btrfs_block_group_cache *group,
1463                                      int mode)
1464 {
1465         struct inode *inode;
1466         struct btrfs_inode_item *inode_item;
1467         struct btrfs_key *location;
1468         struct btrfs_path *path;
1469         int ret;
1470         int owner;
1471
1472         path = btrfs_alloc_path();
1473         BUG_ON(!path);
1474
1475         inode = new_inode(root->fs_info->sb);
1476         if (!inode)
1477                 return ERR_PTR(-ENOMEM);
1478
1479         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1480                              inode->i_mapping, GFP_NOFS);
1481         BTRFS_I(inode)->root = root;
1482
1483         if (mode & S_IFDIR)
1484                 owner = 0;
1485         else
1486                 owner = 1;
1487         group = btrfs_find_block_group(root, group, 0, 0, owner);
1488         BTRFS_I(inode)->block_group = group;
1489         BTRFS_I(inode)->flags = 0;
1490         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1491         if (ret)
1492                 goto fail;
1493
1494         inode->i_uid = current->fsuid;
1495         inode->i_gid = current->fsgid;
1496         inode->i_mode = mode;
1497         inode->i_ino = objectid;
1498         inode->i_blocks = 0;
1499         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1500         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1501                                   struct btrfs_inode_item);
1502         fill_inode_item(path->nodes[0], inode_item, inode);
1503         btrfs_mark_buffer_dirty(path->nodes[0]);
1504         btrfs_free_path(path);
1505
1506         location = &BTRFS_I(inode)->location;
1507         location->objectid = objectid;
1508         location->offset = 0;
1509         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1510
1511         insert_inode_hash(inode);
1512         return inode;
1513 fail:
1514         btrfs_free_path(path);
1515         return ERR_PTR(ret);
1516 }
1517
1518 static inline u8 btrfs_inode_type(struct inode *inode)
1519 {
1520         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1521 }
1522
1523 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1524                             struct dentry *dentry, struct inode *inode)
1525 {
1526         int ret;
1527         struct btrfs_key key;
1528         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1529         struct inode *parent_inode;
1530
1531         key.objectid = inode->i_ino;
1532         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1533         key.offset = 0;
1534
1535         ret = btrfs_insert_dir_item(trans, root,
1536                                     dentry->d_name.name, dentry->d_name.len,
1537                                     dentry->d_parent->d_inode->i_ino,
1538                                     &key, btrfs_inode_type(inode));
1539         if (ret == 0) {
1540                 ret = btrfs_insert_inode_ref(trans, root,
1541                                      dentry->d_name.name,
1542                                      dentry->d_name.len,
1543                                      inode->i_ino,
1544                                      dentry->d_parent->d_inode->i_ino);
1545                 parent_inode = dentry->d_parent->d_inode;
1546                 parent_inode->i_size += dentry->d_name.len * 2;
1547                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1548                 ret = btrfs_update_inode(trans, root,
1549                                          dentry->d_parent->d_inode);
1550         }
1551         return ret;
1552 }
1553
1554 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1555                             struct dentry *dentry, struct inode *inode)
1556 {
1557         int err = btrfs_add_link(trans, dentry, inode);
1558         if (!err) {
1559                 d_instantiate(dentry, inode);
1560                 return 0;
1561         }
1562         if (err > 0)
1563                 err = -EEXIST;
1564         return err;
1565 }
1566
1567 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1568                         int mode, dev_t rdev)
1569 {
1570         struct btrfs_trans_handle *trans;
1571         struct btrfs_root *root = BTRFS_I(dir)->root;
1572         struct inode *inode = NULL;
1573         int err;
1574         int drop_inode = 0;
1575         u64 objectid;
1576         unsigned long nr = 0;
1577
1578         if (!new_valid_dev(rdev))
1579                 return -EINVAL;
1580
1581         mutex_lock(&root->fs_info->fs_mutex);
1582         err = btrfs_check_free_space(root, 1, 0);
1583         if (err)
1584                 goto fail;
1585
1586         trans = btrfs_start_transaction(root, 1);
1587         btrfs_set_trans_block_group(trans, dir);
1588
1589         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1590         if (err) {
1591                 err = -ENOSPC;
1592                 goto out_unlock;
1593         }
1594
1595         inode = btrfs_new_inode(trans, root, objectid,
1596                                 BTRFS_I(dir)->block_group, mode);
1597         err = PTR_ERR(inode);
1598         if (IS_ERR(inode))
1599                 goto out_unlock;
1600
1601         btrfs_set_trans_block_group(trans, inode);
1602         err = btrfs_add_nondir(trans, dentry, inode);
1603         if (err)
1604                 drop_inode = 1;
1605         else {
1606                 inode->i_op = &btrfs_special_inode_operations;
1607                 init_special_inode(inode, inode->i_mode, rdev);
1608                 btrfs_update_inode(trans, root, inode);
1609         }
1610         dir->i_sb->s_dirt = 1;
1611         btrfs_update_inode_block_group(trans, inode);
1612         btrfs_update_inode_block_group(trans, dir);
1613 out_unlock:
1614         nr = trans->blocks_used;
1615         btrfs_end_transaction(trans, root);
1616 fail:
1617         mutex_unlock(&root->fs_info->fs_mutex);
1618
1619         if (drop_inode) {
1620                 inode_dec_link_count(inode);
1621                 iput(inode);
1622         }
1623         btrfs_btree_balance_dirty(root, nr);
1624         btrfs_throttle(root);
1625         return err;
1626 }
1627
1628 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1629                         int mode, struct nameidata *nd)
1630 {
1631         struct btrfs_trans_handle *trans;
1632         struct btrfs_root *root = BTRFS_I(dir)->root;
1633         struct inode *inode = NULL;
1634         int err;
1635         int drop_inode = 0;
1636         unsigned long nr = 0;
1637         u64 objectid;
1638
1639         mutex_lock(&root->fs_info->fs_mutex);
1640         err = btrfs_check_free_space(root, 1, 0);
1641         if (err)
1642                 goto fail;
1643         trans = btrfs_start_transaction(root, 1);
1644         btrfs_set_trans_block_group(trans, dir);
1645
1646         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1647         if (err) {
1648                 err = -ENOSPC;
1649                 goto out_unlock;
1650         }
1651
1652         inode = btrfs_new_inode(trans, root, objectid,
1653                                 BTRFS_I(dir)->block_group, mode);
1654         err = PTR_ERR(inode);
1655         if (IS_ERR(inode))
1656                 goto out_unlock;
1657
1658         btrfs_set_trans_block_group(trans, inode);
1659         err = btrfs_add_nondir(trans, dentry, inode);
1660         if (err)
1661                 drop_inode = 1;
1662         else {
1663                 inode->i_mapping->a_ops = &btrfs_aops;
1664                 inode->i_fop = &btrfs_file_operations;
1665                 inode->i_op = &btrfs_file_inode_operations;
1666                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1667                                      inode->i_mapping, GFP_NOFS);
1668                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1669         }
1670         dir->i_sb->s_dirt = 1;
1671         btrfs_update_inode_block_group(trans, inode);
1672         btrfs_update_inode_block_group(trans, dir);
1673 out_unlock:
1674         nr = trans->blocks_used;
1675         btrfs_end_transaction(trans, root);
1676 fail:
1677         mutex_unlock(&root->fs_info->fs_mutex);
1678
1679         if (drop_inode) {
1680                 inode_dec_link_count(inode);
1681                 iput(inode);
1682         }
1683         btrfs_btree_balance_dirty(root, nr);
1684         btrfs_throttle(root);
1685         return err;
1686 }
1687
1688 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1689                       struct dentry *dentry)
1690 {
1691         struct btrfs_trans_handle *trans;
1692         struct btrfs_root *root = BTRFS_I(dir)->root;
1693         struct inode *inode = old_dentry->d_inode;
1694         unsigned long nr = 0;
1695         int err;
1696         int drop_inode = 0;
1697
1698         if (inode->i_nlink == 0)
1699                 return -ENOENT;
1700
1701 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1702         inode->i_nlink++;
1703 #else
1704         inc_nlink(inode);
1705 #endif
1706         mutex_lock(&root->fs_info->fs_mutex);
1707         err = btrfs_check_free_space(root, 1, 0);
1708         if (err)
1709                 goto fail;
1710         trans = btrfs_start_transaction(root, 1);
1711
1712         btrfs_set_trans_block_group(trans, dir);
1713         atomic_inc(&inode->i_count);
1714         err = btrfs_add_nondir(trans, dentry, inode);
1715
1716         if (err)
1717                 drop_inode = 1;
1718
1719         dir->i_sb->s_dirt = 1;
1720         btrfs_update_inode_block_group(trans, dir);
1721         err = btrfs_update_inode(trans, root, inode);
1722
1723         if (err)
1724                 drop_inode = 1;
1725
1726         nr = trans->blocks_used;
1727         btrfs_end_transaction(trans, root);
1728 fail:
1729         mutex_unlock(&root->fs_info->fs_mutex);
1730
1731         if (drop_inode) {
1732                 inode_dec_link_count(inode);
1733                 iput(inode);
1734         }
1735         btrfs_btree_balance_dirty(root, nr);
1736         btrfs_throttle(root);
1737         return err;
1738 }
1739
1740 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1741 {
1742         struct inode *inode;
1743         struct btrfs_trans_handle *trans;
1744         struct btrfs_root *root = BTRFS_I(dir)->root;
1745         int err = 0;
1746         int drop_on_err = 0;
1747         u64 objectid;
1748         unsigned long nr = 1;
1749
1750         mutex_lock(&root->fs_info->fs_mutex);
1751         err = btrfs_check_free_space(root, 1, 0);
1752         if (err)
1753                 goto out_unlock;
1754
1755         trans = btrfs_start_transaction(root, 1);
1756         btrfs_set_trans_block_group(trans, dir);
1757
1758         if (IS_ERR(trans)) {
1759                 err = PTR_ERR(trans);
1760                 goto out_unlock;
1761         }
1762
1763         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1764         if (err) {
1765                 err = -ENOSPC;
1766                 goto out_unlock;
1767         }
1768
1769         inode = btrfs_new_inode(trans, root, objectid,
1770                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1771         if (IS_ERR(inode)) {
1772                 err = PTR_ERR(inode);
1773                 goto out_fail;
1774         }
1775
1776         drop_on_err = 1;
1777         inode->i_op = &btrfs_dir_inode_operations;
1778         inode->i_fop = &btrfs_dir_file_operations;
1779         btrfs_set_trans_block_group(trans, inode);
1780
1781         inode->i_size = 0;
1782         err = btrfs_update_inode(trans, root, inode);
1783         if (err)
1784                 goto out_fail;
1785
1786         err = btrfs_add_link(trans, dentry, inode);
1787         if (err)
1788                 goto out_fail;
1789
1790         d_instantiate(dentry, inode);
1791         drop_on_err = 0;
1792         dir->i_sb->s_dirt = 1;
1793         btrfs_update_inode_block_group(trans, inode);
1794         btrfs_update_inode_block_group(trans, dir);
1795
1796 out_fail:
1797         nr = trans->blocks_used;
1798         btrfs_end_transaction(trans, root);
1799
1800 out_unlock:
1801         mutex_unlock(&root->fs_info->fs_mutex);
1802         if (drop_on_err)
1803                 iput(inode);
1804         btrfs_btree_balance_dirty(root, nr);
1805         btrfs_throttle(root);
1806         return err;
1807 }
1808
1809 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1810                                     size_t page_offset, u64 start, u64 end,
1811                                     int create)
1812 {
1813         int ret;
1814         int err = 0;
1815         u64 bytenr;
1816         u64 extent_start = 0;
1817         u64 extent_end = 0;
1818         u64 objectid = inode->i_ino;
1819         u32 found_type;
1820         int failed_insert = 0;
1821         struct btrfs_path *path;
1822         struct btrfs_root *root = BTRFS_I(inode)->root;
1823         struct btrfs_file_extent_item *item;
1824         struct extent_buffer *leaf;
1825         struct btrfs_key found_key;
1826         struct extent_map *em = NULL;
1827         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1828         struct btrfs_trans_handle *trans = NULL;
1829
1830         path = btrfs_alloc_path();
1831         BUG_ON(!path);
1832         mutex_lock(&root->fs_info->fs_mutex);
1833
1834 again:
1835         em = lookup_extent_mapping(em_tree, start, end);
1836         if (em) {
1837                 if (em->start > start) {
1838                         printk("get_extent start %Lu em start %Lu\n",
1839                                start, em->start);
1840                         WARN_ON(1);
1841                 }
1842                 goto out;
1843         }
1844         if (!em) {
1845                 em = alloc_extent_map(GFP_NOFS);
1846                 if (!em) {
1847                         err = -ENOMEM;
1848                         goto out;
1849                 }
1850                 em->start = EXTENT_MAP_HOLE;
1851                 em->end = EXTENT_MAP_HOLE;
1852         }
1853         em->bdev = inode->i_sb->s_bdev;
1854         ret = btrfs_lookup_file_extent(trans, root, path,
1855                                        objectid, start, trans != NULL);
1856         if (ret < 0) {
1857                 err = ret;
1858                 goto out;
1859         }
1860
1861         if (ret != 0) {
1862                 if (path->slots[0] == 0)
1863                         goto not_found;
1864                 path->slots[0]--;
1865         }
1866
1867         leaf = path->nodes[0];
1868         item = btrfs_item_ptr(leaf, path->slots[0],
1869                               struct btrfs_file_extent_item);
1870         /* are we inside the extent that was found? */
1871         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1872         found_type = btrfs_key_type(&found_key);
1873         if (found_key.objectid != objectid ||
1874             found_type != BTRFS_EXTENT_DATA_KEY) {
1875                 goto not_found;
1876         }
1877
1878         found_type = btrfs_file_extent_type(leaf, item);
1879         extent_start = found_key.offset;
1880         if (found_type == BTRFS_FILE_EXTENT_REG) {
1881                 extent_end = extent_start +
1882                        btrfs_file_extent_num_bytes(leaf, item);
1883                 err = 0;
1884                 if (start < extent_start || start >= extent_end) {
1885                         em->start = start;
1886                         if (start < extent_start) {
1887                                 if (end < extent_start)
1888                                         goto not_found;
1889                                 em->end = extent_end - 1;
1890                         } else {
1891                                 em->end = end;
1892                         }
1893                         goto not_found_em;
1894                 }
1895                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1896                 if (bytenr == 0) {
1897                         em->start = extent_start;
1898                         em->end = extent_end - 1;
1899                         em->block_start = EXTENT_MAP_HOLE;
1900                         em->block_end = EXTENT_MAP_HOLE;
1901                         goto insert;
1902                 }
1903                 bytenr += btrfs_file_extent_offset(leaf, item);
1904                 em->block_start = bytenr;
1905                 em->block_end = em->block_start +
1906                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1907                 em->start = extent_start;
1908                 em->end = extent_end - 1;
1909                 goto insert;
1910         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1911                 unsigned long ptr;
1912                 char *map;
1913                 size_t size;
1914                 size_t extent_offset;
1915                 size_t copy_size;
1916
1917                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1918                                                     path->slots[0]));
1919                 extent_end = (extent_start + size - 1) |
1920                         ((u64)root->sectorsize - 1);
1921                 if (start < extent_start || start >= extent_end) {
1922                         em->start = start;
1923                         if (start < extent_start) {
1924                                 if (end < extent_start)
1925                                         goto not_found;
1926                                 em->end = extent_end;
1927                         } else {
1928                                 em->end = end;
1929                         }
1930                         goto not_found_em;
1931                 }
1932                 em->block_start = EXTENT_MAP_INLINE;
1933                 em->block_end = EXTENT_MAP_INLINE;
1934
1935                 if (!page) {
1936                         em->start = extent_start;
1937                         em->end = extent_start + size - 1;
1938                         goto out;
1939                 }
1940
1941                 extent_offset = ((u64)page->index << PAGE_CACHE_SHIFT) -
1942                         extent_start + page_offset;
1943                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1944                                 size - extent_offset);
1945                 em->start = extent_start + extent_offset;
1946                 em->end = (em->start + copy_size -1) |
1947                         ((u64)root->sectorsize -1);
1948                 map = kmap(page);
1949                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1950                 if (create == 0 && !PageUptodate(page)) {
1951                         read_extent_buffer(leaf, map + page_offset, ptr,
1952                                            copy_size);
1953                         flush_dcache_page(page);
1954                 } else if (create && PageUptodate(page)) {
1955                         if (!trans) {
1956                                 kunmap(page);
1957                                 free_extent_map(em);
1958                                 em = NULL;
1959                                 btrfs_release_path(root, path);
1960                                 trans = btrfs_start_transaction(root, 1);
1961                                 goto again;
1962                         }
1963                         write_extent_buffer(leaf, map + page_offset, ptr,
1964                                             copy_size);
1965                         btrfs_mark_buffer_dirty(leaf);
1966                 }
1967                 kunmap(page);
1968                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1969                 goto insert;
1970         } else {
1971                 printk("unkknown found_type %d\n", found_type);
1972                 WARN_ON(1);
1973         }
1974 not_found:
1975         em->start = start;
1976         em->end = end;
1977 not_found_em:
1978         em->block_start = EXTENT_MAP_HOLE;
1979         em->block_end = EXTENT_MAP_HOLE;
1980 insert:
1981         btrfs_release_path(root, path);
1982         if (em->start > start || em->end < start) {
1983                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1984                 err = -EIO;
1985                 goto out;
1986         }
1987         ret = add_extent_mapping(em_tree, em);
1988         if (ret == -EEXIST) {
1989                 free_extent_map(em);
1990                 em = NULL;
1991                 if (0 && failed_insert == 1) {
1992                         btrfs_drop_extent_cache(inode, start, end);
1993                 }
1994                 failed_insert++;
1995                 if (failed_insert > 5) {
1996                         printk("failing to insert %Lu %Lu\n", start, end);
1997                         err = -EIO;
1998                         goto out;
1999                 }
2000                 goto again;
2001         }
2002         err = 0;
2003 out:
2004         btrfs_free_path(path);
2005         if (trans) {
2006                 ret = btrfs_end_transaction(trans, root);
2007                 if (!err)
2008                         err = ret;
2009         }
2010         mutex_unlock(&root->fs_info->fs_mutex);
2011         if (err) {
2012                 free_extent_map(em);
2013                 WARN_ON(1);
2014                 return ERR_PTR(err);
2015         }
2016         return em;
2017 }
2018
2019 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2020 {
2021         return extent_bmap(mapping, iblock, btrfs_get_extent);
2022 }
2023
2024 int btrfs_readpage(struct file *file, struct page *page)
2025 {
2026         struct extent_map_tree *tree;
2027         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2028         return extent_read_full_page(tree, page, btrfs_get_extent);
2029 }
2030
2031 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2032 {
2033         struct extent_map_tree *tree;
2034
2035
2036         if (current->flags & PF_MEMALLOC) {
2037                 redirty_page_for_writepage(wbc, page);
2038                 unlock_page(page);
2039                 return 0;
2040         }
2041         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2042         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2043 }
2044
2045 static int btrfs_writepages(struct address_space *mapping,
2046                             struct writeback_control *wbc)
2047 {
2048         struct extent_map_tree *tree;
2049         tree = &BTRFS_I(mapping->host)->extent_tree;
2050         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2051 }
2052
2053 static int
2054 btrfs_readpages(struct file *file, struct address_space *mapping,
2055                 struct list_head *pages, unsigned nr_pages)
2056 {
2057         struct extent_map_tree *tree;
2058         tree = &BTRFS_I(mapping->host)->extent_tree;
2059         return extent_readpages(tree, mapping, pages, nr_pages,
2060                                 btrfs_get_extent);
2061 }
2062
2063 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2064 {
2065         struct extent_map_tree *tree;
2066         int ret;
2067
2068         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2069         ret = try_release_extent_mapping(tree, page);
2070         if (ret == 1) {
2071                 ClearPagePrivate(page);
2072                 set_page_private(page, 0);
2073                 page_cache_release(page);
2074         }
2075         return ret;
2076 }
2077
2078 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2079 {
2080         struct extent_map_tree *tree;
2081
2082         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2083         extent_invalidatepage(tree, page, offset);
2084         btrfs_releasepage(page, GFP_NOFS);
2085 }
2086
2087 /*
2088  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2089  * called from a page fault handler when a page is first dirtied. Hence we must
2090  * be careful to check for EOF conditions here. We set the page up correctly
2091  * for a written page which means we get ENOSPC checking when writing into
2092  * holes and correct delalloc and unwritten extent mapping on filesystems that
2093  * support these features.
2094  *
2095  * We are not allowed to take the i_mutex here so we have to play games to
2096  * protect against truncate races as the page could now be beyond EOF.  Because
2097  * vmtruncate() writes the inode size before removing pages, once we have the
2098  * page lock we can determine safely if the page is beyond EOF. If it is not
2099  * beyond EOF, then the page is guaranteed safe against truncation until we
2100  * unlock the page.
2101  */
2102 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2103 {
2104         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2105         struct btrfs_root *root = BTRFS_I(inode)->root;
2106         unsigned long end;
2107         loff_t size;
2108         int ret;
2109         u64 page_start;
2110
2111         mutex_lock(&root->fs_info->fs_mutex);
2112         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2113         mutex_unlock(&root->fs_info->fs_mutex);
2114         if (ret)
2115                 goto out;
2116
2117         ret = -EINVAL;
2118
2119         lock_page(page);
2120         wait_on_page_writeback(page);
2121         size = i_size_read(inode);
2122         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2123
2124         if ((page->mapping != inode->i_mapping) ||
2125             (page_start > size)) {
2126                 /* page got truncated out from underneath us */
2127                 goto out_unlock;
2128         }
2129
2130         /* page is wholly or partially inside EOF */
2131         if (page_start + PAGE_CACHE_SIZE > size)
2132                 end = size & ~PAGE_CACHE_MASK;
2133         else
2134                 end = PAGE_CACHE_SIZE;
2135
2136         ret = btrfs_cow_one_page(inode, page, end);
2137
2138 out_unlock:
2139         unlock_page(page);
2140 out:
2141         return ret;
2142 }
2143
2144 static void btrfs_truncate(struct inode *inode)
2145 {
2146         struct btrfs_root *root = BTRFS_I(inode)->root;
2147         int ret;
2148         struct btrfs_trans_handle *trans;
2149         unsigned long nr;
2150
2151         if (!S_ISREG(inode->i_mode))
2152                 return;
2153         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2154                 return;
2155
2156         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2157
2158         mutex_lock(&root->fs_info->fs_mutex);
2159         trans = btrfs_start_transaction(root, 1);
2160         btrfs_set_trans_block_group(trans, inode);
2161
2162         /* FIXME, add redo link to tree so we don't leak on crash */
2163         ret = btrfs_truncate_in_trans(trans, root, inode);
2164         btrfs_update_inode(trans, root, inode);
2165         nr = trans->blocks_used;
2166
2167         ret = btrfs_end_transaction(trans, root);
2168         BUG_ON(ret);
2169         mutex_unlock(&root->fs_info->fs_mutex);
2170         btrfs_btree_balance_dirty(root, nr);
2171         btrfs_throttle(root);
2172 }
2173
2174 static int noinline create_subvol(struct btrfs_root *root, char *name,
2175                                   int namelen)
2176 {
2177         struct btrfs_trans_handle *trans;
2178         struct btrfs_key key;
2179         struct btrfs_root_item root_item;
2180         struct btrfs_inode_item *inode_item;
2181         struct extent_buffer *leaf;
2182         struct btrfs_root *new_root = root;
2183         struct inode *inode;
2184         struct inode *dir;
2185         int ret;
2186         int err;
2187         u64 objectid;
2188         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2189         unsigned long nr = 1;
2190
2191         mutex_lock(&root->fs_info->fs_mutex);
2192         ret = btrfs_check_free_space(root, 1, 0);
2193         if (ret)
2194                 goto fail_commit;
2195
2196         trans = btrfs_start_transaction(root, 1);
2197         BUG_ON(!trans);
2198
2199         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2200                                        0, &objectid);
2201         if (ret)
2202                 goto fail;
2203
2204         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2205                                         objectid, trans->transid, 0, 0,
2206                                         0, 0);
2207         if (IS_ERR(leaf))
2208                 return PTR_ERR(leaf);
2209
2210         btrfs_set_header_nritems(leaf, 0);
2211         btrfs_set_header_level(leaf, 0);
2212         btrfs_set_header_bytenr(leaf, leaf->start);
2213         btrfs_set_header_generation(leaf, trans->transid);
2214         btrfs_set_header_owner(leaf, objectid);
2215
2216         write_extent_buffer(leaf, root->fs_info->fsid,
2217                             (unsigned long)btrfs_header_fsid(leaf),
2218                             BTRFS_FSID_SIZE);
2219         btrfs_mark_buffer_dirty(leaf);
2220
2221         inode_item = &root_item.inode;
2222         memset(inode_item, 0, sizeof(*inode_item));
2223         inode_item->generation = cpu_to_le64(1);
2224         inode_item->size = cpu_to_le64(3);
2225         inode_item->nlink = cpu_to_le32(1);
2226         inode_item->nblocks = cpu_to_le64(1);
2227         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2228
2229         btrfs_set_root_bytenr(&root_item, leaf->start);
2230         btrfs_set_root_level(&root_item, 0);
2231         btrfs_set_root_refs(&root_item, 1);
2232         btrfs_set_root_used(&root_item, 0);
2233
2234         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2235         root_item.drop_level = 0;
2236
2237         free_extent_buffer(leaf);
2238         leaf = NULL;
2239
2240         btrfs_set_root_dirid(&root_item, new_dirid);
2241
2242         key.objectid = objectid;
2243         key.offset = 1;
2244         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2245         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2246                                 &root_item);
2247         if (ret)
2248                 goto fail;
2249
2250         /*
2251          * insert the directory item
2252          */
2253         key.offset = (u64)-1;
2254         dir = root->fs_info->sb->s_root->d_inode;
2255         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2256                                     name, namelen, dir->i_ino, &key,
2257                                     BTRFS_FT_DIR);
2258         if (ret)
2259                 goto fail;
2260
2261         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2262                              name, namelen, objectid,
2263                              root->fs_info->sb->s_root->d_inode->i_ino);
2264         if (ret)
2265                 goto fail;
2266
2267         ret = btrfs_commit_transaction(trans, root);
2268         if (ret)
2269                 goto fail_commit;
2270
2271         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2272         BUG_ON(!new_root);
2273
2274         trans = btrfs_start_transaction(new_root, 1);
2275         BUG_ON(!trans);
2276
2277         inode = btrfs_new_inode(trans, new_root, new_dirid,
2278                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2279         if (IS_ERR(inode))
2280                 goto fail;
2281         inode->i_op = &btrfs_dir_inode_operations;
2282         inode->i_fop = &btrfs_dir_file_operations;
2283         new_root->inode = inode;
2284
2285         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2286                                      new_dirid);
2287         inode->i_nlink = 1;
2288         inode->i_size = 0;
2289         ret = btrfs_update_inode(trans, new_root, inode);
2290         if (ret)
2291                 goto fail;
2292 fail:
2293         nr = trans->blocks_used;
2294         err = btrfs_commit_transaction(trans, new_root);
2295         if (err && !ret)
2296                 ret = err;
2297 fail_commit:
2298         mutex_unlock(&root->fs_info->fs_mutex);
2299         btrfs_btree_balance_dirty(root, nr);
2300         btrfs_throttle(root);
2301         return ret;
2302 }
2303
2304 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2305 {
2306         struct btrfs_pending_snapshot *pending_snapshot;
2307         struct btrfs_trans_handle *trans;
2308         int ret;
2309         int err;
2310         unsigned long nr = 0;
2311
2312         if (!root->ref_cows)
2313                 return -EINVAL;
2314
2315         mutex_lock(&root->fs_info->fs_mutex);
2316         ret = btrfs_check_free_space(root, 1, 0);
2317         if (ret)
2318                 goto fail_unlock;
2319
2320         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2321         if (!pending_snapshot) {
2322                 ret = -ENOMEM;
2323                 goto fail_unlock;
2324         }
2325         pending_snapshot->name = kstrndup(name, namelen, GFP_NOFS);
2326         if (!pending_snapshot->name) {
2327                 ret = -ENOMEM;
2328                 kfree(pending_snapshot);
2329                 goto fail_unlock;
2330         }
2331         trans = btrfs_start_transaction(root, 1);
2332         BUG_ON(!trans);
2333
2334         pending_snapshot->root = root;
2335         list_add(&pending_snapshot->list,
2336                  &trans->transaction->pending_snapshots);
2337         ret = btrfs_update_inode(trans, root, root->inode);
2338         err = btrfs_commit_transaction(trans, root);
2339
2340 fail_unlock:
2341         mutex_unlock(&root->fs_info->fs_mutex);
2342         btrfs_btree_balance_dirty(root, nr);
2343         btrfs_throttle(root);
2344         return ret;
2345 }
2346
2347 unsigned long btrfs_force_ra(struct address_space *mapping,
2348                               struct file_ra_state *ra, struct file *file,
2349                               pgoff_t offset, pgoff_t last_index)
2350 {
2351         pgoff_t req_size;
2352
2353 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2354         req_size = last_index - offset + 1;
2355         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2356         return offset;
2357 #else
2358         req_size = min(last_index - offset + 1, (pgoff_t)128);
2359         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2360         return offset + req_size;
2361 #endif
2362 }
2363
2364 int btrfs_defrag_file(struct file *file) {
2365         struct inode *inode = fdentry(file)->d_inode;
2366         struct btrfs_root *root = BTRFS_I(inode)->root;
2367         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2368         struct page *page;
2369         unsigned long last_index;
2370         unsigned long ra_index = 0;
2371         u64 page_start;
2372         u64 page_end;
2373         u64 delalloc_start;
2374         u64 existing_delalloc;
2375         unsigned long i;
2376         int ret;
2377
2378         mutex_lock(&root->fs_info->fs_mutex);
2379         ret = btrfs_check_free_space(root, inode->i_size, 0);
2380         mutex_unlock(&root->fs_info->fs_mutex);
2381         if (ret)
2382                 return -ENOSPC;
2383
2384         mutex_lock(&inode->i_mutex);
2385         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2386         for (i = 0; i <= last_index; i++) {
2387                 if (i == ra_index) {
2388                         ra_index = btrfs_force_ra(inode->i_mapping,
2389                                                   &file->f_ra,
2390                                                   file, ra_index, last_index);
2391                 }
2392                 page = grab_cache_page(inode->i_mapping, i);
2393                 if (!page)
2394                         goto out_unlock;
2395                 if (!PageUptodate(page)) {
2396                         btrfs_readpage(NULL, page);
2397                         lock_page(page);
2398                         if (!PageUptodate(page)) {
2399                                 unlock_page(page);
2400                                 page_cache_release(page);
2401                                 goto out_unlock;
2402                         }
2403                 }
2404                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2405                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2406
2407                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2408                 delalloc_start = page_start;
2409                 existing_delalloc =
2410                         count_range_bits(&BTRFS_I(inode)->extent_tree,
2411                                          &delalloc_start, page_end,
2412                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2413                 set_extent_delalloc(em_tree, page_start,
2414                                     page_end, GFP_NOFS);
2415
2416                 spin_lock(&root->fs_info->delalloc_lock);
2417                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2418                                                  existing_delalloc;
2419                 spin_unlock(&root->fs_info->delalloc_lock);
2420
2421                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2422                 set_page_dirty(page);
2423                 unlock_page(page);
2424                 page_cache_release(page);
2425                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2426         }
2427
2428 out_unlock:
2429         mutex_unlock(&inode->i_mutex);
2430         return 0;
2431 }
2432
2433 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2434 {
2435         u64 new_size;
2436         u64 old_size;
2437         struct btrfs_ioctl_vol_args *vol_args;
2438         struct btrfs_trans_handle *trans;
2439         char *sizestr;
2440         int ret = 0;
2441         int namelen;
2442         int mod = 0;
2443
2444         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2445
2446         if (!vol_args)
2447                 return -ENOMEM;
2448
2449         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2450                 ret = -EFAULT;
2451                 goto out;
2452         }
2453         namelen = strlen(vol_args->name);
2454         if (namelen > BTRFS_VOL_NAME_MAX) {
2455                 ret = -EINVAL;
2456                 goto out;
2457         }
2458
2459         sizestr = vol_args->name;
2460         if (!strcmp(sizestr, "max"))
2461                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2462         else {
2463                 if (sizestr[0] == '-') {
2464                         mod = -1;
2465                         sizestr++;
2466                 } else if (sizestr[0] == '+') {
2467                         mod = 1;
2468                         sizestr++;
2469                 }
2470                 new_size = btrfs_parse_size(sizestr);
2471                 if (new_size == 0) {
2472                         ret = -EINVAL;
2473                         goto out;
2474                 }
2475         }
2476
2477         mutex_lock(&root->fs_info->fs_mutex);
2478         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2479
2480         if (mod < 0) {
2481                 if (new_size > old_size) {
2482                         ret = -EINVAL;
2483                         goto out_unlock;
2484                 }
2485                 new_size = old_size - new_size;
2486         } else if (mod > 0) {
2487                 new_size = old_size + new_size;
2488         }
2489
2490         if (new_size < 256 * 1024 * 1024) {
2491                 ret = -EINVAL;
2492                 goto out_unlock;
2493         }
2494         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2495                 ret = -EFBIG;
2496                 goto out_unlock;
2497         }
2498
2499         do_div(new_size, root->sectorsize);
2500         new_size *= root->sectorsize;
2501
2502 printk("new size is %Lu\n", new_size);
2503         if (new_size > old_size) {
2504                 trans = btrfs_start_transaction(root, 1);
2505                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2506                 btrfs_commit_transaction(trans, root);
2507         } else {
2508                 ret = btrfs_shrink_extent_tree(root, new_size);
2509         }
2510
2511 out_unlock:
2512         mutex_unlock(&root->fs_info->fs_mutex);
2513 out:
2514         kfree(vol_args);
2515         return ret;
2516 }
2517
2518 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2519                                             void __user *arg)
2520 {
2521         struct btrfs_ioctl_vol_args *vol_args;
2522         struct btrfs_dir_item *di;
2523         struct btrfs_path *path;
2524         u64 root_dirid;
2525         int namelen;
2526         int ret;
2527
2528         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2529
2530         if (!vol_args)
2531                 return -ENOMEM;
2532
2533         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2534                 ret = -EFAULT;
2535                 goto out;
2536         }
2537
2538         namelen = strlen(vol_args->name);
2539         if (namelen > BTRFS_VOL_NAME_MAX) {
2540                 ret = -EINVAL;
2541                 goto out;
2542         }
2543         if (strchr(vol_args->name, '/')) {
2544                 ret = -EINVAL;
2545                 goto out;
2546         }
2547
2548         path = btrfs_alloc_path();
2549         if (!path) {
2550                 ret = -ENOMEM;
2551                 goto out;
2552         }
2553
2554         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2555         mutex_lock(&root->fs_info->fs_mutex);
2556         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2557                             path, root_dirid,
2558                             vol_args->name, namelen, 0);
2559         mutex_unlock(&root->fs_info->fs_mutex);
2560         btrfs_free_path(path);
2561
2562         if (di && !IS_ERR(di)) {
2563                 ret = -EEXIST;
2564                 goto out;
2565         }
2566
2567         if (IS_ERR(di)) {
2568                 ret = PTR_ERR(di);
2569                 goto out;
2570         }
2571
2572         if (root == root->fs_info->tree_root)
2573                 ret = create_subvol(root, vol_args->name, namelen);
2574         else
2575                 ret = create_snapshot(root, vol_args->name, namelen);
2576 out:
2577         kfree(vol_args);
2578         return ret;
2579 }
2580
2581 static int btrfs_ioctl_defrag(struct file *file)
2582 {
2583         struct inode *inode = fdentry(file)->d_inode;
2584         struct btrfs_root *root = BTRFS_I(inode)->root;
2585
2586         switch (inode->i_mode & S_IFMT) {
2587         case S_IFDIR:
2588                 mutex_lock(&root->fs_info->fs_mutex);
2589                 btrfs_defrag_root(root, 0);
2590                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2591                 mutex_unlock(&root->fs_info->fs_mutex);
2592                 break;
2593         case S_IFREG:
2594                 btrfs_defrag_file(file);
2595                 break;
2596         }
2597
2598         return 0;
2599 }
2600
2601 long btrfs_ioctl(struct file *file, unsigned int
2602                 cmd, unsigned long arg)
2603 {
2604         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2605
2606         switch (cmd) {
2607         case BTRFS_IOC_SNAP_CREATE:
2608                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2609         case BTRFS_IOC_DEFRAG:
2610                 return btrfs_ioctl_defrag(file);
2611         case BTRFS_IOC_RESIZE:
2612                 return btrfs_ioctl_resize(root, (void __user *)arg);
2613         }
2614
2615         return -ENOTTY;
2616 }
2617
2618 /*
2619  * Called inside transaction, so use GFP_NOFS
2620  */
2621 struct inode *btrfs_alloc_inode(struct super_block *sb)
2622 {
2623         struct btrfs_inode *ei;
2624
2625         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2626         if (!ei)
2627                 return NULL;
2628         ei->last_trans = 0;
2629         ei->ordered_trans = 0;
2630         return &ei->vfs_inode;
2631 }
2632
2633 void btrfs_destroy_inode(struct inode *inode)
2634 {
2635         WARN_ON(!list_empty(&inode->i_dentry));
2636         WARN_ON(inode->i_data.nrpages);
2637
2638         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2639         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2640 }
2641
2642 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2643 static void init_once(struct kmem_cache * cachep, void *foo)
2644 #else
2645 static void init_once(void * foo, struct kmem_cache * cachep,
2646                       unsigned long flags)
2647 #endif
2648 {
2649         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2650
2651         inode_init_once(&ei->vfs_inode);
2652 }
2653
2654 void btrfs_destroy_cachep(void)
2655 {
2656         if (btrfs_inode_cachep)
2657                 kmem_cache_destroy(btrfs_inode_cachep);
2658         if (btrfs_trans_handle_cachep)
2659                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2660         if (btrfs_transaction_cachep)
2661                 kmem_cache_destroy(btrfs_transaction_cachep);
2662         if (btrfs_bit_radix_cachep)
2663                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2664         if (btrfs_path_cachep)
2665                 kmem_cache_destroy(btrfs_path_cachep);
2666 }
2667
2668 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2669                                        unsigned long extra_flags,
2670 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2671                                        void (*ctor)(struct kmem_cache *, void *)
2672 #else
2673                                        void (*ctor)(void *, struct kmem_cache *,
2674                                                     unsigned long)
2675 #endif
2676                                      )
2677 {
2678         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2679                                  SLAB_MEM_SPREAD | extra_flags), ctor
2680 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2681                                  ,NULL
2682 #endif
2683                                 );
2684 }
2685
2686 int btrfs_init_cachep(void)
2687 {
2688         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2689                                           sizeof(struct btrfs_inode),
2690                                           0, init_once);
2691         if (!btrfs_inode_cachep)
2692                 goto fail;
2693         btrfs_trans_handle_cachep =
2694                         btrfs_cache_create("btrfs_trans_handle_cache",
2695                                            sizeof(struct btrfs_trans_handle),
2696                                            0, NULL);
2697         if (!btrfs_trans_handle_cachep)
2698                 goto fail;
2699         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2700                                              sizeof(struct btrfs_transaction),
2701                                              0, NULL);
2702         if (!btrfs_transaction_cachep)
2703                 goto fail;
2704         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2705                                          sizeof(struct btrfs_path),
2706                                          0, NULL);
2707         if (!btrfs_path_cachep)
2708                 goto fail;
2709         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2710                                               SLAB_DESTROY_BY_RCU, NULL);
2711         if (!btrfs_bit_radix_cachep)
2712                 goto fail;
2713         return 0;
2714 fail:
2715         btrfs_destroy_cachep();
2716         return -ENOMEM;
2717 }
2718
2719 static int btrfs_getattr(struct vfsmount *mnt,
2720                          struct dentry *dentry, struct kstat *stat)
2721 {
2722         struct inode *inode = dentry->d_inode;
2723         generic_fillattr(inode, stat);
2724         stat->blksize = PAGE_CACHE_SIZE;
2725         return 0;
2726 }
2727
2728 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2729                            struct inode * new_dir,struct dentry *new_dentry)
2730 {
2731         struct btrfs_trans_handle *trans;
2732         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2733         struct inode *new_inode = new_dentry->d_inode;
2734         struct inode *old_inode = old_dentry->d_inode;
2735         struct timespec ctime = CURRENT_TIME;
2736         struct btrfs_path *path;
2737         int ret;
2738
2739         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2740             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2741                 return -ENOTEMPTY;
2742         }
2743
2744         mutex_lock(&root->fs_info->fs_mutex);
2745         ret = btrfs_check_free_space(root, 1, 0);
2746         if (ret)
2747                 goto out_unlock;
2748
2749         trans = btrfs_start_transaction(root, 1);
2750
2751         btrfs_set_trans_block_group(trans, new_dir);
2752         path = btrfs_alloc_path();
2753         if (!path) {
2754                 ret = -ENOMEM;
2755                 goto out_fail;
2756         }
2757
2758         old_dentry->d_inode->i_nlink++;
2759         old_dir->i_ctime = old_dir->i_mtime = ctime;
2760         new_dir->i_ctime = new_dir->i_mtime = ctime;
2761         old_inode->i_ctime = ctime;
2762
2763         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2764         if (ret)
2765                 goto out_fail;
2766
2767         if (new_inode) {
2768                 new_inode->i_ctime = CURRENT_TIME;
2769                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2770                 if (ret)
2771                         goto out_fail;
2772         }
2773         ret = btrfs_add_link(trans, new_dentry, old_inode);
2774         if (ret)
2775                 goto out_fail;
2776
2777 out_fail:
2778         btrfs_free_path(path);
2779         btrfs_end_transaction(trans, root);
2780 out_unlock:
2781         mutex_unlock(&root->fs_info->fs_mutex);
2782         return ret;
2783 }
2784
2785 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2786                          const char *symname)
2787 {
2788         struct btrfs_trans_handle *trans;
2789         struct btrfs_root *root = BTRFS_I(dir)->root;
2790         struct btrfs_path *path;
2791         struct btrfs_key key;
2792         struct inode *inode = NULL;
2793         int err;
2794         int drop_inode = 0;
2795         u64 objectid;
2796         int name_len;
2797         int datasize;
2798         unsigned long ptr;
2799         struct btrfs_file_extent_item *ei;
2800         struct extent_buffer *leaf;
2801         unsigned long nr = 0;
2802
2803         name_len = strlen(symname) + 1;
2804         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2805                 return -ENAMETOOLONG;
2806
2807         mutex_lock(&root->fs_info->fs_mutex);
2808         err = btrfs_check_free_space(root, 1, 0);
2809         if (err)
2810                 goto out_fail;
2811
2812         trans = btrfs_start_transaction(root, 1);
2813         btrfs_set_trans_block_group(trans, dir);
2814
2815         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2816         if (err) {
2817                 err = -ENOSPC;
2818                 goto out_unlock;
2819         }
2820
2821         inode = btrfs_new_inode(trans, root, objectid,
2822                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2823         err = PTR_ERR(inode);
2824         if (IS_ERR(inode))
2825                 goto out_unlock;
2826
2827         btrfs_set_trans_block_group(trans, inode);
2828         err = btrfs_add_nondir(trans, dentry, inode);
2829         if (err)
2830                 drop_inode = 1;
2831         else {
2832                 inode->i_mapping->a_ops = &btrfs_aops;
2833                 inode->i_fop = &btrfs_file_operations;
2834                 inode->i_op = &btrfs_file_inode_operations;
2835                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2836                                      inode->i_mapping, GFP_NOFS);
2837                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2838         }
2839         dir->i_sb->s_dirt = 1;
2840         btrfs_update_inode_block_group(trans, inode);
2841         btrfs_update_inode_block_group(trans, dir);
2842         if (drop_inode)
2843                 goto out_unlock;
2844
2845         path = btrfs_alloc_path();
2846         BUG_ON(!path);
2847         key.objectid = inode->i_ino;
2848         key.offset = 0;
2849         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2850         datasize = btrfs_file_extent_calc_inline_size(name_len);
2851         err = btrfs_insert_empty_item(trans, root, path, &key,
2852                                       datasize);
2853         if (err) {
2854                 drop_inode = 1;
2855                 goto out_unlock;
2856         }
2857         leaf = path->nodes[0];
2858         ei = btrfs_item_ptr(leaf, path->slots[0],
2859                             struct btrfs_file_extent_item);
2860         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2861         btrfs_set_file_extent_type(leaf, ei,
2862                                    BTRFS_FILE_EXTENT_INLINE);
2863         ptr = btrfs_file_extent_inline_start(ei);
2864         write_extent_buffer(leaf, symname, ptr, name_len);
2865         btrfs_mark_buffer_dirty(leaf);
2866         btrfs_free_path(path);
2867
2868         inode->i_op = &btrfs_symlink_inode_operations;
2869         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2870         inode->i_size = name_len - 1;
2871         err = btrfs_update_inode(trans, root, inode);
2872         if (err)
2873                 drop_inode = 1;
2874
2875 out_unlock:
2876         nr = trans->blocks_used;
2877         btrfs_end_transaction(trans, root);
2878 out_fail:
2879         mutex_unlock(&root->fs_info->fs_mutex);
2880         if (drop_inode) {
2881                 inode_dec_link_count(inode);
2882                 iput(inode);
2883         }
2884         btrfs_btree_balance_dirty(root, nr);
2885         btrfs_throttle(root);
2886         return err;
2887 }
2888 static int btrfs_permission(struct inode *inode, int mask,
2889                             struct nameidata *nd)
2890 {
2891         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2892                 return -EACCES;
2893         return generic_permission(inode, mask, NULL);
2894 }
2895
2896 static struct inode_operations btrfs_dir_inode_operations = {
2897         .lookup         = btrfs_lookup,
2898         .create         = btrfs_create,
2899         .unlink         = btrfs_unlink,
2900         .link           = btrfs_link,
2901         .mkdir          = btrfs_mkdir,
2902         .rmdir          = btrfs_rmdir,
2903         .rename         = btrfs_rename,
2904         .symlink        = btrfs_symlink,
2905         .setattr        = btrfs_setattr,
2906         .mknod          = btrfs_mknod,
2907         .setxattr       = generic_setxattr,
2908         .getxattr       = generic_getxattr,
2909         .listxattr      = btrfs_listxattr,
2910         .removexattr    = generic_removexattr,
2911         .permission     = btrfs_permission,
2912 };
2913 static struct inode_operations btrfs_dir_ro_inode_operations = {
2914         .lookup         = btrfs_lookup,
2915         .permission     = btrfs_permission,
2916 };
2917 static struct file_operations btrfs_dir_file_operations = {
2918         .llseek         = generic_file_llseek,
2919         .read           = generic_read_dir,
2920         .readdir        = btrfs_readdir,
2921         .unlocked_ioctl = btrfs_ioctl,
2922 #ifdef CONFIG_COMPAT
2923         .compat_ioctl   = btrfs_ioctl,
2924 #endif
2925 };
2926
2927 static struct extent_map_ops btrfs_extent_map_ops = {
2928         .fill_delalloc = run_delalloc_range,
2929         .writepage_io_hook = btrfs_writepage_io_hook,
2930         .readpage_io_hook = btrfs_readpage_io_hook,
2931         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2932 };
2933
2934 static struct address_space_operations btrfs_aops = {
2935         .readpage       = btrfs_readpage,
2936         .writepage      = btrfs_writepage,
2937         .writepages     = btrfs_writepages,
2938         .readpages      = btrfs_readpages,
2939         .sync_page      = block_sync_page,
2940         .bmap           = btrfs_bmap,
2941         .invalidatepage = btrfs_invalidatepage,
2942         .releasepage    = btrfs_releasepage,
2943         .set_page_dirty = __set_page_dirty_nobuffers,
2944 };
2945
2946 static struct address_space_operations btrfs_symlink_aops = {
2947         .readpage       = btrfs_readpage,
2948         .writepage      = btrfs_writepage,
2949         .invalidatepage = btrfs_invalidatepage,
2950         .releasepage    = btrfs_releasepage,
2951 };
2952
2953 static struct inode_operations btrfs_file_inode_operations = {
2954         .truncate       = btrfs_truncate,
2955         .getattr        = btrfs_getattr,
2956         .setattr        = btrfs_setattr,
2957         .setxattr       = generic_setxattr,
2958         .getxattr       = generic_getxattr,
2959         .listxattr      = btrfs_listxattr,
2960         .removexattr    = generic_removexattr,
2961         .permission     = btrfs_permission,
2962 };
2963 static struct inode_operations btrfs_special_inode_operations = {
2964         .getattr        = btrfs_getattr,
2965         .setattr        = btrfs_setattr,
2966         .permission     = btrfs_permission,
2967 };
2968 static struct inode_operations btrfs_symlink_inode_operations = {
2969         .readlink       = generic_readlink,
2970         .follow_link    = page_follow_link_light,
2971         .put_link       = page_put_link,
2972         .permission     = btrfs_permission,
2973 };