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_io_ops btrfs_extent_io_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         if (for_del)
  84                 thresh = total * 90;
  85         else
  86                 thresh = total * 85;
  87
  88         do_div(thresh, 100);
  89
  90         spin_lock(&root->fs_info->delalloc_lock);
  91         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  92                 ret = -ENOSPC;
  93         spin_unlock(&root->fs_info->delalloc_lock);
  94         return ret;
  95 }
  96
  97 static int cow_file_range(struct inode *inode, u64 start, u64 end)
  98 {
  99         struct btrfs_root *root = BTRFS_I(inode)->root;
 100         struct btrfs_trans_handle *trans;
 101         u64 alloc_hint = 0;
 102         u64 num_bytes;
 103         u64 cur_alloc_size;
 104         u64 blocksize = root->sectorsize;
 105         u64 orig_start = start;
 106         u64 orig_num_bytes;
 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         orig_num_bytes = num_bytes;
 119
 120         if (alloc_hint == EXTENT_MAP_INLINE)
 121                 goto out;
 122
 123         while(num_bytes > 0) {
 124                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 125                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 126                                          root->root_key.objectid,
 127                                          trans->transid,
 128                                          inode->i_ino, start, 0,
 129                                          alloc_hint, (u64)-1, &ins, 1);
 130                 if (ret) {
 131                         WARN_ON(1);
 132                         goto out;
 133                 }
 134                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 135                                                start, ins.objectid, ins.offset,
 136                                                ins.offset);
 137                 btrfs_check_file(root, inode);
 138                 num_bytes -= cur_alloc_size;
 139                 alloc_hint = ins.objectid + ins.offset;
 140                 start += cur_alloc_size;
 141         }
 142         btrfs_drop_extent_cache(inode, orig_start,
 143                                 orig_start + orig_num_bytes - 1);
 144         btrfs_add_ordered_inode(inode);
 145 out:
 146         btrfs_end_transaction(trans, root);
 147         return ret;
 148 }
 149
 150 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 151 {
 152         u64 extent_start;
 153         u64 extent_end;
 154         u64 bytenr;
 155         u64 cow_end;
 156         u64 loops = 0;
 157         u64 total_fs_bytes;
 158         struct btrfs_root *root = BTRFS_I(inode)->root;
 159         struct extent_buffer *leaf;
 160         int found_type;
 161         struct btrfs_path *path;
 162         struct btrfs_file_extent_item *item;
 163         int ret;
 164         int err;
 165         struct btrfs_key found_key;
 166
 167         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 168         path = btrfs_alloc_path();
 169         BUG_ON(!path);
 170 again:
 171         ret = btrfs_lookup_file_extent(NULL, root, path,
 172                                        inode->i_ino, start, 0);
 173         if (ret < 0) {
 174                 btrfs_free_path(path);
 175                 return ret;
 176         }
 177
 178         cow_end = end;
 179         if (ret != 0) {
 180                 if (path->slots[0] == 0)
 181                         goto not_found;
 182                 path->slots[0]--;
 183         }
 184
 185         leaf = path->nodes[0];
 186         item = btrfs_item_ptr(leaf, path->slots[0],
 187                               struct btrfs_file_extent_item);
 188
 189         /* are we inside the extent that was found? */
 190         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 191         found_type = btrfs_key_type(&found_key);
 192         if (found_key.objectid != inode->i_ino ||
 193             found_type != BTRFS_EXTENT_DATA_KEY) {
 194                 goto not_found;
 195         }
 196
 197         found_type = btrfs_file_extent_type(leaf, item);
 198         extent_start = found_key.offset;
 199         if (found_type == BTRFS_FILE_EXTENT_REG) {
 200                 u64 extent_num_bytes;
 201
 202                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 203                 extent_end = extent_start + extent_num_bytes;
 204                 err = 0;
 205
 206                 if (loops && start != extent_start)
 207                         goto not_found;
 208
 209                 if (start < extent_start || start >= extent_end)
 210                         goto not_found;
 211
 212                 cow_end = min(end, extent_end - 1);
 213                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 214                 if (bytenr == 0)
 215                         goto not_found;
 216
 217                 /*
 218                  * we may be called by the resizer, make sure we're inside
 219                  * the limits of the FS
 220                  */
 221                 if (bytenr + extent_num_bytes > total_fs_bytes)
 222                         goto not_found;
 223
 224                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 225                         goto not_found;
 226                 }
 227
 228                 start = extent_end;
 229         } else {
 230                 goto not_found;
 231         }
 232 loop:
 233         if (start > end) {
 234                 btrfs_free_path(path);
 235                 return 0;
 236         }
 237         btrfs_release_path(root, path);
 238         loops++;
 239         goto again;
 240
 241 not_found:
 242         cow_file_range(inode, start, cow_end);
 243         start = cow_end + 1;
 244         goto loop;
 245 }
 246
 247 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 248 {
 249         struct btrfs_root *root = BTRFS_I(inode)->root;
 250         int ret;
 251         mutex_lock(&root->fs_info->fs_mutex);
 252         if (btrfs_test_opt(root, NODATACOW) ||
 253             btrfs_test_flag(inode, NODATACOW))
 254                 ret = run_delalloc_nocow(inode, start, end);
 255         else
 256                 ret = cow_file_range(inode, start, end);
 257
 258         mutex_unlock(&root->fs_info->fs_mutex);
 259         return ret;
 260 }
 261
 262 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 263                        unsigned long bits)
 264 {
 265         if ((bits & EXTENT_DELALLOC)) {
 266                 struct btrfs_root *root = BTRFS_I(inode)->root;
 267                 spin_lock(&root->fs_info->delalloc_lock);
 268                 root->fs_info->delalloc_bytes += end - start + 1;
 269                 spin_unlock(&root->fs_info->delalloc_lock);
 270         }
 271         return 0;
 272 }
 273
 274 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 275                          unsigned long bits)
 276 {
 277         if ((bits & EXTENT_DELALLOC)) {
 278                 struct btrfs_root *root = BTRFS_I(inode)->root;
 279                 spin_lock(&root->fs_info->delalloc_lock);
 280                 root->fs_info->delalloc_bytes -= end - start + 1;
 281                 spin_unlock(&root->fs_info->delalloc_lock);
 282         }
 283         return 0;
 284 }
 285
 286 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 287 {
 288         struct inode *inode = page->mapping->host;
 289         struct btrfs_root *root = BTRFS_I(inode)->root;
 290         struct btrfs_trans_handle *trans;
 291         char *kaddr;
 292         int ret = 0;
 293         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 294         size_t offset = start - page_start;
 295         if (btrfs_test_opt(root, NODATASUM) ||
 296             btrfs_test_flag(inode, NODATASUM))
 297                 return 0;
 298         mutex_lock(&root->fs_info->fs_mutex);
 299         trans = btrfs_start_transaction(root, 1);
 300         btrfs_set_trans_block_group(trans, inode);
 301         kaddr = kmap(page);
 302         btrfs_csum_file_block(trans, root, inode, inode->i_ino,
 303                               start, kaddr + offset, end - start + 1);
 304         kunmap(page);
 305         ret = btrfs_end_transaction(trans, root);
 306         BUG_ON(ret);
 307         mutex_unlock(&root->fs_info->fs_mutex);
 308         return ret;
 309 }
 310
 311 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 312 {
 313         int ret = 0;
 314         struct inode *inode = page->mapping->host;
 315         struct btrfs_root *root = BTRFS_I(inode)->root;
 316         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 317         struct btrfs_csum_item *item;
 318         struct btrfs_path *path = NULL;
 319         u32 csum;
 320         if (btrfs_test_opt(root, NODATASUM) ||
 321             btrfs_test_flag(inode, NODATASUM))
 322                 return 0;
 323         mutex_lock(&root->fs_info->fs_mutex);
 324         path = btrfs_alloc_path();
 325         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 326         if (IS_ERR(item)) {
 327                 ret = PTR_ERR(item);
 328                 /* a csum that isn't present is a preallocated region. */
 329                 if (ret == -ENOENT || ret == -EFBIG)
 330                         ret = 0;
 331                 csum = 0;
 332                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino, start);
 333                 goto out;
 334         }
 335         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 336                            BTRFS_CRC32_SIZE);
 337         set_state_private(io_tree, start, csum);
 338 out:
 339         if (path)
 340                 btrfs_free_path(path);
 341         mutex_unlock(&root->fs_info->fs_mutex);
 342         return ret;
 343 }
 344
 345 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 346                                struct extent_state *state)
 347 {
 348         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 349         struct inode *inode = page->mapping->host;
 350         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 351         char *kaddr;
 352         u64 private = ~(u32)0;
 353         int ret;
 354         struct btrfs_root *root = BTRFS_I(inode)->root;
 355         u32 csum = ~(u32)0;
 356         unsigned long flags;
 357
 358         if (btrfs_test_opt(root, NODATASUM) ||
 359             btrfs_test_flag(inode, NODATASUM))
 360                 return 0;
 361
 362         if (state->start == start) {
 363                 private = state->private;
 364                 ret = 0;
 365         } else {
 366                 ret = get_state_private(io_tree, start, &private);
 367         }
 368         local_irq_save(flags);
 369         kaddr = kmap_atomic(page, KM_IRQ0);
 370         if (ret) {
 371                 goto zeroit;
 372         }
 373         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 374         btrfs_csum_final(csum, (char *)&csum);
 375         if (csum != private) {
 376                 goto zeroit;
 377         }
 378         kunmap_atomic(kaddr, KM_IRQ0);
 379         local_irq_restore(flags);
 380         return 0;
 381
 382 zeroit:
 383         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 384                page->mapping->host->i_ino, (unsigned long long)start, csum,
 385                private);
 386         memset(kaddr + offset, 1, end - start + 1);
 387         flush_dcache_page(page);
 388         kunmap_atomic(kaddr, KM_IRQ0);
 389         local_irq_restore(flags);
 390         return 0;
 391 }
 392
 393 void btrfs_read_locked_inode(struct inode *inode)
 394 {
 395         struct btrfs_path *path;
 396         struct extent_buffer *leaf;
 397         struct btrfs_inode_item *inode_item;
 398         struct btrfs_inode_timespec *tspec;
 399         struct btrfs_root *root = BTRFS_I(inode)->root;
 400         struct btrfs_key location;
 401         u64 alloc_group_block;
 402         u32 rdev;
 403         int ret;
 404
 405         path = btrfs_alloc_path();
 406         BUG_ON(!path);
 407         mutex_lock(&root->fs_info->fs_mutex);
 408         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 409
 410         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 411         if (ret)
 412                 goto make_bad;
 413
 414         leaf = path->nodes[0];
 415         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 416                                     struct btrfs_inode_item);
 417
 418         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 419         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 420         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 421         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 422         inode->i_size = btrfs_inode_size(leaf, inode_item);
 423
 424         tspec = btrfs_inode_atime(inode_item);
 425         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 426         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 427
 428         tspec = btrfs_inode_mtime(inode_item);
 429         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 430         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 431
 432         tspec = btrfs_inode_ctime(inode_item);
 433         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 434         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 435
 436         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 437         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 438         inode->i_rdev = 0;
 439         rdev = btrfs_inode_rdev(leaf, inode_item);
 440
 441         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 442         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 443                                                        alloc_group_block);
 444         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 445         if (!BTRFS_I(inode)->block_group) {
 446                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 447                                                          NULL, 0, 0, 0);
 448         }
 449         btrfs_free_path(path);
 450         inode_item = NULL;
 451
 452         mutex_unlock(&root->fs_info->fs_mutex);
 453
 454         switch (inode->i_mode & S_IFMT) {
 455         case S_IFREG:
 456                 inode->i_mapping->a_ops = &btrfs_aops;
 457                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 458                 inode->i_fop = &btrfs_file_operations;
 459                 inode->i_op = &btrfs_file_inode_operations;
 460                 break;
 461         case S_IFDIR:
 462                 inode->i_fop = &btrfs_dir_file_operations;
 463                 if (root == root->fs_info->tree_root)
 464                         inode->i_op = &btrfs_dir_ro_inode_operations;
 465                 else
 466                         inode->i_op = &btrfs_dir_inode_operations;
 467                 break;
 468         case S_IFLNK:
 469                 inode->i_op = &btrfs_symlink_inode_operations;
 470                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 471                 break;
 472         default:
 473                 init_special_inode(inode, inode->i_mode, rdev);
 474                 break;
 475         }
 476         return;
 477
 478 make_bad:
 479         btrfs_release_path(root, path);
 480         btrfs_free_path(path);
 481         mutex_unlock(&root->fs_info->fs_mutex);
 482         make_bad_inode(inode);
 483 }
 484
 485 static void fill_inode_item(struct extent_buffer *leaf,
 486                             struct btrfs_inode_item *item,
 487                             struct inode *inode)
 488 {
 489         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 490         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 491         btrfs_set_inode_size(leaf, item, inode->i_size);
 492         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 493         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 494
 495         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 496                                inode->i_atime.tv_sec);
 497         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 498                                 inode->i_atime.tv_nsec);
 499
 500         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 501                                inode->i_mtime.tv_sec);
 502         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 503                                 inode->i_mtime.tv_nsec);
 504
 505         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 506                                inode->i_ctime.tv_sec);
 507         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 508                                 inode->i_ctime.tv_nsec);
 509
 510         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 511         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 512         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 513         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 514         btrfs_set_inode_block_group(leaf, item,
 515                                     BTRFS_I(inode)->block_group->key.objectid);
 516 }
 517
 518 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 519                               struct btrfs_root *root,
 520                               struct inode *inode)
 521 {
 522         struct btrfs_inode_item *inode_item;
 523         struct btrfs_path *path;
 524         struct extent_buffer *leaf;
 525         int ret;
 526
 527         path = btrfs_alloc_path();
 528         BUG_ON(!path);
 529         ret = btrfs_lookup_inode(trans, root, path,
 530                                  &BTRFS_I(inode)->location, 1);
 531         if (ret) {
 532                 if (ret > 0)
 533                         ret = -ENOENT;
 534                 goto failed;
 535         }
 536
 537         leaf = path->nodes[0];
 538         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 539                                   struct btrfs_inode_item);
 540
 541         fill_inode_item(leaf, inode_item, inode);
 542         btrfs_mark_buffer_dirty(leaf);
 543         btrfs_set_inode_last_trans(trans, inode);
 544         ret = 0;
 545 failed:
 546         btrfs_release_path(root, path);
 547         btrfs_free_path(path);
 548         return ret;
 549 }
 550
 551
 552 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 553                               struct btrfs_root *root,
 554                               struct inode *dir,
 555                               struct dentry *dentry)
 556 {
 557         struct btrfs_path *path;
 558         const char *name = dentry->d_name.name;
 559         int name_len = dentry->d_name.len;
 560         int ret = 0;
 561         struct extent_buffer *leaf;
 562         struct btrfs_dir_item *di;
 563         struct btrfs_key key;
 564
 565         path = btrfs_alloc_path();
 566         if (!path) {
 567                 ret = -ENOMEM;
 568                 goto err;
 569         }
 570
 571         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 572                                     name, name_len, -1);
 573         if (IS_ERR(di)) {
 574                 ret = PTR_ERR(di);
 575                 goto err;
 576         }
 577         if (!di) {
 578                 ret = -ENOENT;
 579                 goto err;
 580         }
 581         leaf = path->nodes[0];
 582         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 583         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 584         if (ret)
 585                 goto err;
 586         btrfs_release_path(root, path);
 587
 588         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 589                                          key.objectid, name, name_len, -1);
 590         if (IS_ERR(di)) {
 591                 ret = PTR_ERR(di);
 592                 goto err;
 593         }
 594         if (!di) {
 595                 ret = -ENOENT;
 596                 goto err;
 597         }
 598         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 599
 600         dentry->d_inode->i_ctime = dir->i_ctime;
 601         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 602                                   dentry->d_inode->i_ino,
 603                                   dentry->d_parent->d_inode->i_ino);
 604         if (ret) {
 605                 printk("failed to delete reference to %.*s, "
 606                        "inode %lu parent %lu\n", name_len, name,
 607                        dentry->d_inode->i_ino,
 608                        dentry->d_parent->d_inode->i_ino);
 609         }
 610 err:
 611         btrfs_free_path(path);
 612         if (!ret) {
 613                 dir->i_size -= name_len * 2;
 614                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 615                 btrfs_update_inode(trans, root, dir);
 616 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 617                 dentry->d_inode->i_nlink--;
 618 #else
 619                 drop_nlink(dentry->d_inode);
 620 #endif
 621                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 622                 dir->i_sb->s_dirt = 1;
 623         }
 624         return ret;
 625 }
 626
 627 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 628 {
 629         struct btrfs_root *root;
 630         struct btrfs_trans_handle *trans;
 631         struct inode *inode = dentry->d_inode;
 632         int ret;
 633         unsigned long nr = 0;
 634
 635         root = BTRFS_I(dir)->root;
 636         mutex_lock(&root->fs_info->fs_mutex);
 637
 638         ret = btrfs_check_free_space(root, 1, 1);
 639         if (ret)
 640                 goto fail;
 641
 642         trans = btrfs_start_transaction(root, 1);
 643
 644         btrfs_set_trans_block_group(trans, dir);
 645         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 646         nr = trans->blocks_used;
 647
 648         if (inode->i_nlink == 0) {
 649                 int found;
 650                 /* if the inode isn't linked anywhere,
 651                  * we don't need to worry about
 652                  * data=ordered
 653                  */
 654                 found = btrfs_del_ordered_inode(inode);
 655                 if (found == 1) {
 656                         atomic_dec(&inode->i_count);
 657                 }
 658         }
 659
 660         btrfs_end_transaction(trans, root);
 661 fail:
 662         mutex_unlock(&root->fs_info->fs_mutex);
 663         btrfs_btree_balance_dirty(root, nr);
 664         btrfs_throttle(root);
 665         return ret;
 666 }
 667
 668 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 669 {
 670         struct inode *inode = dentry->d_inode;
 671         int err = 0;
 672         int ret;
 673         struct btrfs_root *root = BTRFS_I(dir)->root;
 674         struct btrfs_trans_handle *trans;
 675         unsigned long nr = 0;
 676
 677         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 678                 return -ENOTEMPTY;
 679
 680         mutex_lock(&root->fs_info->fs_mutex);
 681         ret = btrfs_check_free_space(root, 1, 1);
 682         if (ret)
 683                 goto fail;
 684
 685         trans = btrfs_start_transaction(root, 1);
 686         btrfs_set_trans_block_group(trans, dir);
 687
 688         /* now the directory is empty */
 689         err = btrfs_unlink_trans(trans, root, dir, dentry);
 690         if (!err) {
 691                 inode->i_size = 0;
 692         }
 693
 694         nr = trans->blocks_used;
 695         ret = btrfs_end_transaction(trans, root);
 696 fail:
 697         mutex_unlock(&root->fs_info->fs_mutex);
 698         btrfs_btree_balance_dirty(root, nr);
 699         btrfs_throttle(root);
 700
 701         if (ret && !err)
 702                 err = ret;
 703         return err;
 704 }
 705
 706 /*
 707  * this can truncate away extent items, csum items and directory items.
 708  * It starts at a high offset and removes keys until it can't find
 709  * any higher than i_size.
 710  *
 711  * csum items that cross the new i_size are truncated to the new size
 712  * as well.
 713  */
 714 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 715                                    struct btrfs_root *root,
 716                                    struct inode *inode,
 717                                    u32 min_type)
 718 {
 719         int ret;
 720         struct btrfs_path *path;
 721         struct btrfs_key key;
 722         struct btrfs_key found_key;
 723         u32 found_type;
 724         struct extent_buffer *leaf;
 725         struct btrfs_file_extent_item *fi;
 726         u64 extent_start = 0;
 727         u64 extent_num_bytes = 0;
 728         u64 item_end = 0;
 729         u64 root_gen = 0;
 730         u64 root_owner = 0;
 731         int found_extent;
 732         int del_item;
 733         int pending_del_nr = 0;
 734         int pending_del_slot = 0;
 735         int extent_type = -1;
 736
 737         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 738         path = btrfs_alloc_path();
 739         path->reada = -1;
 740         BUG_ON(!path);
 741
 742         /* FIXME, add redo link to tree so we don't leak on crash */
 743         key.objectid = inode->i_ino;
 744         key.offset = (u64)-1;
 745         key.type = (u8)-1;
 746
 747         btrfs_init_path(path);
 748 search_again:
 749         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 750         if (ret < 0) {
 751                 goto error;
 752         }
 753         if (ret > 0) {
 754                 BUG_ON(path->slots[0] == 0);
 755                 path->slots[0]--;
 756         }
 757
 758         while(1) {
 759                 fi = NULL;
 760                 leaf = path->nodes[0];
 761                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 762                 found_type = btrfs_key_type(&found_key);
 763
 764                 if (found_key.objectid != inode->i_ino)
 765                         break;
 766
 767                 if (found_type < min_type)
 768                         break;
 769
 770                 item_end = found_key.offset;
 771                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 772                         fi = btrfs_item_ptr(leaf, path->slots[0],
 773                                             struct btrfs_file_extent_item);
 774                         extent_type = btrfs_file_extent_type(leaf, fi);
 775                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 776                                 item_end +=
 777                                     btrfs_file_extent_num_bytes(leaf, fi);
 778                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 779                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 780                                                                 path->slots[0]);
 781                                 item_end += btrfs_file_extent_inline_len(leaf,
 782                                                                          item);
 783                         }
 784                         item_end--;
 785                 }
 786                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 787                         ret = btrfs_csum_truncate(trans, root, path,
 788                                                   inode->i_size);
 789                         BUG_ON(ret);
 790                 }
 791                 if (item_end < inode->i_size) {
 792                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 793                                 found_type = BTRFS_INODE_ITEM_KEY;
 794                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 795                                 found_type = BTRFS_CSUM_ITEM_KEY;
 796                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
 797                                 found_type = BTRFS_XATTR_ITEM_KEY;
 798                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
 799                                 found_type = BTRFS_INODE_REF_KEY;
 800                         } else if (found_type) {
 801                                 found_type--;
 802                         } else {
 803                                 break;
 804                         }
 805                         btrfs_set_key_type(&key, found_type);
 806                         goto next;
 807                 }
 808                 if (found_key.offset >= inode->i_size)
 809                         del_item = 1;
 810                 else
 811                         del_item = 0;
 812                 found_extent = 0;
 813
 814                 /* FIXME, shrink the extent if the ref count is only 1 */
 815                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 816                         goto delete;
 817
 818                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 819                         u64 num_dec;
 820                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 821                         if (!del_item) {
 822                                 u64 orig_num_bytes =
 823                                         btrfs_file_extent_num_bytes(leaf, fi);
 824                                 extent_num_bytes = inode->i_size -
 825                                         found_key.offset + root->sectorsize - 1;
 826                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 827                                                          extent_num_bytes);
 828                                 num_dec = (orig_num_bytes -
 829                                            extent_num_bytes) >> 9;
 830                                 if (extent_start != 0) {
 831                                         inode->i_blocks -= num_dec;
 832                                 }
 833                                 btrfs_mark_buffer_dirty(leaf);
 834                         } else {
 835                                 extent_num_bytes =
 836                                         btrfs_file_extent_disk_num_bytes(leaf,
 837                                                                          fi);
 838                                 /* FIXME blocksize != 4096 */
 839                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 840                                                                        fi) >> 9;
 841                                 if (extent_start != 0) {
 842                                         found_extent = 1;
 843                                         inode->i_blocks -= num_dec;
 844                                 }
 845                                 root_gen = btrfs_header_generation(leaf);
 846                                 root_owner = btrfs_header_owner(leaf);
 847                         }
 848                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 849                            !del_item) {
 850                         u32 newsize = inode->i_size - found_key.offset;
 851                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 852                         ret = btrfs_truncate_item(trans, root, path,
 853                                                   newsize, 1);
 854                         BUG_ON(ret);
 855                 }
 856 delete:
 857                 if (del_item) {
 858                         if (!pending_del_nr) {
 859                                 /* no pending yet, add ourselves */
 860                                 pending_del_slot = path->slots[0];
 861                                 pending_del_nr = 1;
 862                         } else if (pending_del_nr &&
 863                                    path->slots[0] + 1 == pending_del_slot) {
 864                                 /* hop on the pending chunk */
 865                                 pending_del_nr++;
 866                                 pending_del_slot = path->slots[0];
 867                         } else {
 868                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
 869                         }
 870                 } else {
 871                         break;
 872                 }
 873                 if (found_extent) {
 874                         ret = btrfs_free_extent(trans, root, extent_start,
 875                                                 extent_num_bytes,
 876                                                 root_owner,
 877                                                 root_gen, inode->i_ino,
 878                                                 found_key.offset, 0);
 879                         BUG_ON(ret);
 880                 }
 881 next:
 882                 if (path->slots[0] == 0) {
 883                         if (pending_del_nr)
 884                                 goto del_pending;
 885                         btrfs_release_path(root, path);
 886                         goto search_again;
 887                 }
 888
 889                 path->slots[0]--;
 890                 if (pending_del_nr &&
 891                     path->slots[0] + 1 != pending_del_slot) {
 892                         struct btrfs_key debug;
 893 del_pending:
 894                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
 895                                               pending_del_slot);
 896                         ret = btrfs_del_items(trans, root, path,
 897                                               pending_del_slot,
 898                                               pending_del_nr);
 899                         BUG_ON(ret);
 900                         pending_del_nr = 0;
 901                         btrfs_release_path(root, path);
 902                         goto search_again;
 903                 }
 904         }
 905         ret = 0;
 906 error:
 907         if (pending_del_nr) {
 908                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
 909                                       pending_del_nr);
 910         }
 911         btrfs_release_path(root, path);
 912         btrfs_free_path(path);
 913         inode->i_sb->s_dirt = 1;
 914         return ret;
 915 }
 916
 917 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 918                               size_t zero_start)
 919 {
 920         char *kaddr;
 921         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 922         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 923         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 924         int ret = 0;
 925
 926         WARN_ON(!PageLocked(page));
 927         set_page_extent_mapped(page);
 928
 929         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 930         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
 931                             page_end, GFP_NOFS);
 932
 933         if (zero_start != PAGE_CACHE_SIZE) {
 934                 kaddr = kmap(page);
 935                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 936                 flush_dcache_page(page);
 937                 kunmap(page);
 938         }
 939         set_page_dirty(page);
 940         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 941
 942         return ret;
 943 }
 944
 945 /*
 946  * taken from block_truncate_page, but does cow as it zeros out
 947  * any bytes left in the last page in the file.
 948  */
 949 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 950 {
 951         struct inode *inode = mapping->host;
 952         struct btrfs_root *root = BTRFS_I(inode)->root;
 953         u32 blocksize = root->sectorsize;
 954         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 955         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 956         struct page *page;
 957         int ret = 0;
 958         u64 page_start;
 959
 960         if ((offset & (blocksize - 1)) == 0)
 961                 goto out;
 962
 963         ret = -ENOMEM;
 964         page = grab_cache_page(mapping, index);
 965         if (!page)
 966                 goto out;
 967         if (!PageUptodate(page)) {
 968                 ret = btrfs_readpage(NULL, page);
 969                 lock_page(page);
 970                 if (!PageUptodate(page)) {
 971                         ret = -EIO;
 972                         goto out;
 973                 }
 974         }
 975         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 976
 977         ret = btrfs_cow_one_page(inode, page, offset);
 978
 979         unlock_page(page);
 980         page_cache_release(page);
 981 out:
 982         return ret;
 983 }
 984
 985 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 986 {
 987         struct inode *inode = dentry->d_inode;
 988         int err;
 989
 990         err = inode_change_ok(inode, attr);
 991         if (err)
 992                 return err;
 993
 994         if (S_ISREG(inode->i_mode) &&
 995             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 996                 struct btrfs_trans_handle *trans;
 997                 struct btrfs_root *root = BTRFS_I(inode)->root;
 998                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 999
1000                 u64 mask = root->sectorsize - 1;
1001                 u64 pos = (inode->i_size + mask) & ~mask;
1002                 u64 block_end = attr->ia_size | mask;
1003                 u64 hole_start;
1004                 u64 hole_size;
1005                 u64 alloc_hint = 0;
1006
1007                 if (attr->ia_size <= pos)
1008                         goto out;
1009
1010                 if (pos != inode->i_size)
1011                         hole_start = pos + root->sectorsize;
1012                 else
1013                         hole_start = pos;
1014
1015                 mutex_lock(&root->fs_info->fs_mutex);
1016                 err = btrfs_check_free_space(root, 1, 0);
1017                 mutex_unlock(&root->fs_info->fs_mutex);
1018                 if (err)
1019                         goto fail;
1020
1021                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1022
1023                 lock_extent(io_tree, pos, block_end, GFP_NOFS);
1024                 hole_size = block_end - hole_start;
1025
1026                 mutex_lock(&root->fs_info->fs_mutex);
1027                 trans = btrfs_start_transaction(root, 1);
1028                 btrfs_set_trans_block_group(trans, inode);
1029                 err = btrfs_drop_extents(trans, root, inode,
1030                                          pos, block_end, pos,
1031                                          &alloc_hint);
1032
1033                 if (alloc_hint != EXTENT_MAP_INLINE) {
1034                         err = btrfs_insert_file_extent(trans, root,
1035                                                        inode->i_ino,
1036                                                        hole_start, 0, 0,
1037                                                        hole_size);
1038                         btrfs_drop_extent_cache(inode, hole_start,
1039                                                 hole_size - 1);
1040                         btrfs_check_file(root, inode);
1041                 }
1042                 btrfs_end_transaction(trans, root);
1043                 mutex_unlock(&root->fs_info->fs_mutex);
1044                 unlock_extent(io_tree, pos, block_end, GFP_NOFS);
1045                 if (err)
1046                         return err;
1047         }
1048 out:
1049         err = inode_setattr(inode, attr);
1050 fail:
1051         return err;
1052 }
1053
1054 void btrfs_put_inode(struct inode *inode)
1055 {
1056         int ret;
1057
1058         if (!BTRFS_I(inode)->ordered_trans) {
1059                 return;
1060         }
1061
1062         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1063             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1064                 return;
1065
1066         ret = btrfs_del_ordered_inode(inode);
1067         if (ret == 1) {
1068                 atomic_dec(&inode->i_count);
1069         }
1070 }
1071
1072 void btrfs_delete_inode(struct inode *inode)
1073 {
1074         struct btrfs_trans_handle *trans;
1075         struct btrfs_root *root = BTRFS_I(inode)->root;
1076         unsigned long nr;
1077         int ret;
1078
1079         truncate_inode_pages(&inode->i_data, 0);
1080         if (is_bad_inode(inode)) {
1081                 goto no_delete;
1082         }
1083
1084         inode->i_size = 0;
1085         mutex_lock(&root->fs_info->fs_mutex);
1086         trans = btrfs_start_transaction(root, 1);
1087
1088         btrfs_set_trans_block_group(trans, inode);
1089         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1090         if (ret)
1091                 goto no_delete_lock;
1092
1093         nr = trans->blocks_used;
1094         clear_inode(inode);
1095
1096         btrfs_end_transaction(trans, root);
1097         mutex_unlock(&root->fs_info->fs_mutex);
1098         btrfs_btree_balance_dirty(root, nr);
1099         btrfs_throttle(root);
1100         return;
1101
1102 no_delete_lock:
1103         nr = trans->blocks_used;
1104         btrfs_end_transaction(trans, root);
1105         mutex_unlock(&root->fs_info->fs_mutex);
1106         btrfs_btree_balance_dirty(root, nr);
1107         btrfs_throttle(root);
1108 no_delete:
1109         clear_inode(inode);
1110 }
1111
1112 /*
1113  * this returns the key found in the dir entry in the location pointer.
1114  * If no dir entries were found, location->objectid is 0.
1115  */
1116 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1117                                struct btrfs_key *location)
1118 {
1119         const char *name = dentry->d_name.name;
1120         int namelen = dentry->d_name.len;
1121         struct btrfs_dir_item *di;
1122         struct btrfs_path *path;
1123         struct btrfs_root *root = BTRFS_I(dir)->root;
1124         int ret = 0;
1125
1126         if (namelen == 1 && strcmp(name, ".") == 0) {
1127                 location->objectid = dir->i_ino;
1128                 location->type = BTRFS_INODE_ITEM_KEY;
1129                 location->offset = 0;
1130                 return 0;
1131         }
1132         path = btrfs_alloc_path();
1133         BUG_ON(!path);
1134
1135         if (namelen == 2 && strcmp(name, "..") == 0) {
1136                 struct btrfs_key key;
1137                 struct extent_buffer *leaf;
1138                 u32 nritems;
1139                 int slot;
1140
1141                 key.objectid = dir->i_ino;
1142                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1143                 key.offset = 0;
1144                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1145                 BUG_ON(ret == 0);
1146                 ret = 0;
1147
1148                 leaf = path->nodes[0];
1149                 slot = path->slots[0];
1150                 nritems = btrfs_header_nritems(leaf);
1151                 if (slot >= nritems)
1152                         goto out_err;
1153
1154                 btrfs_item_key_to_cpu(leaf, &key, slot);
1155                 if (key.objectid != dir->i_ino ||
1156                     key.type != BTRFS_INODE_REF_KEY) {
1157                         goto out_err;
1158                 }
1159                 location->objectid = key.offset;
1160                 location->type = BTRFS_INODE_ITEM_KEY;
1161                 location->offset = 0;
1162                 goto out;
1163         }
1164
1165         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1166                                     namelen, 0);
1167         if (IS_ERR(di))
1168                 ret = PTR_ERR(di);
1169         if (!di || IS_ERR(di)) {
1170                 goto out_err;
1171         }
1172         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1173 out:
1174         btrfs_free_path(path);
1175         return ret;
1176 out_err:
1177         location->objectid = 0;
1178         goto out;
1179 }
1180
1181 /*
1182  * when we hit a tree root in a directory, the btrfs part of the inode
1183  * needs to be changed to reflect the root directory of the tree root.  This
1184  * is kind of like crossing a mount point.
1185  */
1186 static int fixup_tree_root_location(struct btrfs_root *root,
1187                              struct btrfs_key *location,
1188                              struct btrfs_root **sub_root,
1189                              struct dentry *dentry)
1190 {
1191         struct btrfs_path *path;
1192         struct btrfs_root_item *ri;
1193
1194         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1195                 return 0;
1196         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1197                 return 0;
1198
1199         path = btrfs_alloc_path();
1200         BUG_ON(!path);
1201         mutex_lock(&root->fs_info->fs_mutex);
1202
1203         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1204                                         dentry->d_name.name,
1205                                         dentry->d_name.len);
1206         if (IS_ERR(*sub_root))
1207                 return PTR_ERR(*sub_root);
1208
1209         ri = &(*sub_root)->root_item;
1210         location->objectid = btrfs_root_dirid(ri);
1211         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1212         location->offset = 0;
1213
1214         btrfs_free_path(path);
1215         mutex_unlock(&root->fs_info->fs_mutex);
1216         return 0;
1217 }
1218
1219 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1220 {
1221         struct btrfs_iget_args *args = p;
1222         inode->i_ino = args->ino;
1223         BTRFS_I(inode)->root = args->root;
1224         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1225         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1226                              inode->i_mapping, GFP_NOFS);
1227         return 0;
1228 }
1229
1230 static int btrfs_find_actor(struct inode *inode, void *opaque)
1231 {
1232         struct btrfs_iget_args *args = opaque;
1233         return (args->ino == inode->i_ino &&
1234                 args->root == BTRFS_I(inode)->root);
1235 }
1236
1237 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1238                             u64 root_objectid)
1239 {
1240         struct btrfs_iget_args args;
1241         args.ino = objectid;
1242         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1243
1244         if (!args.root)
1245                 return NULL;
1246
1247         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1248 }
1249
1250 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1251                                 struct btrfs_root *root)
1252 {
1253         struct inode *inode;
1254         struct btrfs_iget_args args;
1255         args.ino = objectid;
1256         args.root = root;
1257
1258         inode = iget5_locked(s, objectid, btrfs_find_actor,
1259                              btrfs_init_locked_inode,
1260                              (void *)&args);
1261         return inode;
1262 }
1263
1264 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1265                                    struct nameidata *nd)
1266 {
1267         struct inode * inode;
1268         struct btrfs_inode *bi = BTRFS_I(dir);
1269         struct btrfs_root *root = bi->root;
1270         struct btrfs_root *sub_root = root;
1271         struct btrfs_key location;
1272         int ret;
1273
1274         if (dentry->d_name.len > BTRFS_NAME_LEN)
1275                 return ERR_PTR(-ENAMETOOLONG);
1276
1277         mutex_lock(&root->fs_info->fs_mutex);
1278         ret = btrfs_inode_by_name(dir, dentry, &location);
1279         mutex_unlock(&root->fs_info->fs_mutex);
1280
1281         if (ret < 0)
1282                 return ERR_PTR(ret);
1283
1284         inode = NULL;
1285         if (location.objectid) {
1286                 ret = fixup_tree_root_location(root, &location, &sub_root,
1287                                                 dentry);
1288                 if (ret < 0)
1289                         return ERR_PTR(ret);
1290                 if (ret > 0)
1291                         return ERR_PTR(-ENOENT);
1292                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1293                                           sub_root);
1294                 if (!inode)
1295                         return ERR_PTR(-EACCES);
1296                 if (inode->i_state & I_NEW) {
1297                         /* the inode and parent dir are two different roots */
1298                         if (sub_root != root) {
1299                                 igrab(inode);
1300                                 sub_root->inode = inode;
1301                         }
1302                         BTRFS_I(inode)->root = sub_root;
1303                         memcpy(&BTRFS_I(inode)->location, &location,
1304                                sizeof(location));
1305                         btrfs_read_locked_inode(inode);
1306                         unlock_new_inode(inode);
1307                 }
1308         }
1309         return d_splice_alias(inode, dentry);
1310 }
1311
1312 static unsigned char btrfs_filetype_table[] = {
1313         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1314 };
1315
1316 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1317 {
1318         struct inode *inode = filp->f_dentry->d_inode;
1319         struct btrfs_root *root = BTRFS_I(inode)->root;
1320         struct btrfs_item *item;
1321         struct btrfs_dir_item *di;
1322         struct btrfs_key key;
1323         struct btrfs_key found_key;
1324         struct btrfs_path *path;
1325         int ret;
1326         u32 nritems;
1327         struct extent_buffer *leaf;
1328         int slot;
1329         int advance;
1330         unsigned char d_type;
1331         int over = 0;
1332         u32 di_cur;
1333         u32 di_total;
1334         u32 di_len;
1335         int key_type = BTRFS_DIR_INDEX_KEY;
1336         char tmp_name[32];
1337         char *name_ptr;
1338         int name_len;
1339
1340         /* FIXME, use a real flag for deciding about the key type */
1341         if (root->fs_info->tree_root == root)
1342                 key_type = BTRFS_DIR_ITEM_KEY;
1343
1344         /* special case for "." */
1345         if (filp->f_pos == 0) {
1346                 over = filldir(dirent, ".", 1,
1347                                1, inode->i_ino,
1348                                DT_DIR);
1349                 if (over)
1350                         return 0;
1351                 filp->f_pos = 1;
1352         }
1353
1354         mutex_lock(&root->fs_info->fs_mutex);
1355         key.objectid = inode->i_ino;
1356         path = btrfs_alloc_path();
1357         path->reada = 2;
1358
1359         /* special case for .., just use the back ref */
1360         if (filp->f_pos == 1) {
1361                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1362                 key.offset = 0;
1363                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1364                 BUG_ON(ret == 0);
1365                 leaf = path->nodes[0];
1366                 slot = path->slots[0];
1367                 nritems = btrfs_header_nritems(leaf);
1368                 if (slot >= nritems) {
1369                         btrfs_release_path(root, path);
1370                         goto read_dir_items;
1371                 }
1372                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1373                 btrfs_release_path(root, path);
1374                 if (found_key.objectid != key.objectid ||
1375                     found_key.type != BTRFS_INODE_REF_KEY)
1376                         goto read_dir_items;
1377                 over = filldir(dirent, "..", 2,
1378                                2, found_key.offset, DT_DIR);
1379                 if (over)
1380                         goto nopos;
1381                 filp->f_pos = 2;
1382         }
1383
1384 read_dir_items:
1385         btrfs_set_key_type(&key, key_type);
1386         key.offset = filp->f_pos;
1387
1388         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1389         if (ret < 0)
1390                 goto err;
1391         advance = 0;
1392         while(1) {
1393                 leaf = path->nodes[0];
1394                 nritems = btrfs_header_nritems(leaf);
1395                 slot = path->slots[0];
1396                 if (advance || slot >= nritems) {
1397                         if (slot >= nritems -1) {
1398                                 ret = btrfs_next_leaf(root, path);
1399                                 if (ret)
1400                                         break;
1401                                 leaf = path->nodes[0];
1402                                 nritems = btrfs_header_nritems(leaf);
1403                                 slot = path->slots[0];
1404                         } else {
1405                                 slot++;
1406                                 path->slots[0]++;
1407                         }
1408                 }
1409                 advance = 1;
1410                 item = btrfs_item_nr(leaf, slot);
1411                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1412
1413                 if (found_key.objectid != key.objectid)
1414                         break;
1415                 if (btrfs_key_type(&found_key) != key_type)
1416                         break;
1417                 if (found_key.offset < filp->f_pos)
1418                         continue;
1419
1420                 filp->f_pos = found_key.offset;
1421                 advance = 1;
1422                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1423                 di_cur = 0;
1424                 di_total = btrfs_item_size(leaf, item);
1425                 while(di_cur < di_total) {
1426                         struct btrfs_key location;
1427
1428                         name_len = btrfs_dir_name_len(leaf, di);
1429                         if (name_len < 32) {
1430                                 name_ptr = tmp_name;
1431                         } else {
1432                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1433                                 BUG_ON(!name_ptr);
1434                         }
1435                         read_extent_buffer(leaf, name_ptr,
1436                                            (unsigned long)(di + 1), name_len);
1437
1438                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1439                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1440                         over = filldir(dirent, name_ptr, name_len,
1441                                        found_key.offset,
1442                                        location.objectid,
1443                                        d_type);
1444
1445                         if (name_ptr != tmp_name)
1446                                 kfree(name_ptr);
1447
1448                         if (over)
1449                                 goto nopos;
1450                         di_len = btrfs_dir_name_len(leaf, di) +
1451                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1452                         di_cur += di_len;
1453                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1454                 }
1455         }
1456         filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1457 nopos:
1458         ret = 0;
1459 err:
1460         btrfs_release_path(root, path);
1461         btrfs_free_path(path);
1462         mutex_unlock(&root->fs_info->fs_mutex);
1463         return ret;
1464 }
1465
1466 int btrfs_write_inode(struct inode *inode, int wait)
1467 {
1468         struct btrfs_root *root = BTRFS_I(inode)->root;
1469         struct btrfs_trans_handle *trans;
1470         int ret = 0;
1471
1472         if (wait) {
1473                 mutex_lock(&root->fs_info->fs_mutex);
1474                 trans = btrfs_start_transaction(root, 1);
1475                 btrfs_set_trans_block_group(trans, inode);
1476                 ret = btrfs_commit_transaction(trans, root);
1477                 mutex_unlock(&root->fs_info->fs_mutex);
1478         }
1479         return ret;
1480 }
1481
1482 /*
1483  * This is somewhat expensive, updating the tree every time the
1484  * inode changes.  But, it is most likely to find the inode in cache.
1485  * FIXME, needs more benchmarking...there are no reasons other than performance
1486  * to keep or drop this code.
1487  */
1488 void btrfs_dirty_inode(struct inode *inode)
1489 {
1490         struct btrfs_root *root = BTRFS_I(inode)->root;
1491         struct btrfs_trans_handle *trans;
1492
1493         mutex_lock(&root->fs_info->fs_mutex);
1494         trans = btrfs_start_transaction(root, 1);
1495         btrfs_set_trans_block_group(trans, inode);
1496         btrfs_update_inode(trans, root, inode);
1497         btrfs_end_transaction(trans, root);
1498         mutex_unlock(&root->fs_info->fs_mutex);
1499 }
1500
1501 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1502                                      struct btrfs_root *root,
1503                                      const char *name, int name_len,
1504                                      u64 ref_objectid,
1505                                      u64 objectid,
1506                                      struct btrfs_block_group_cache *group,
1507                                      int mode)
1508 {
1509         struct inode *inode;
1510         struct btrfs_inode_item *inode_item;
1511         struct btrfs_key *location;
1512         struct btrfs_path *path;
1513         struct btrfs_inode_ref *ref;
1514         struct btrfs_key key[2];
1515         u32 sizes[2];
1516         unsigned long ptr;
1517         int ret;
1518         int owner;
1519
1520         path = btrfs_alloc_path();
1521         BUG_ON(!path);
1522
1523         inode = new_inode(root->fs_info->sb);
1524         if (!inode)
1525                 return ERR_PTR(-ENOMEM);
1526
1527         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1528         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1529                              inode->i_mapping, GFP_NOFS);
1530         BTRFS_I(inode)->root = root;
1531
1532         if (mode & S_IFDIR)
1533                 owner = 0;
1534         else
1535                 owner = 1;
1536         group = btrfs_find_block_group(root, group, 0, 0, owner);
1537         BTRFS_I(inode)->block_group = group;
1538         BTRFS_I(inode)->flags = 0;
1539
1540         key[0].objectid = objectid;
1541         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1542         key[0].offset = 0;
1543
1544         key[1].objectid = objectid;
1545         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1546         key[1].offset = ref_objectid;
1547
1548         sizes[0] = sizeof(struct btrfs_inode_item);
1549         sizes[1] = name_len + sizeof(*ref);
1550
1551         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1552         if (ret != 0)
1553                 goto fail;
1554
1555         if (objectid > root->highest_inode)
1556                 root->highest_inode = objectid;
1557
1558         inode->i_uid = current->fsuid;
1559         inode->i_gid = current->fsgid;
1560         inode->i_mode = mode;
1561         inode->i_ino = objectid;
1562         inode->i_blocks = 0;
1563         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1564         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1565                                   struct btrfs_inode_item);
1566         fill_inode_item(path->nodes[0], inode_item, inode);
1567
1568         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1569                              struct btrfs_inode_ref);
1570         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1571         ptr = (unsigned long)(ref + 1);
1572         write_extent_buffer(path->nodes[0], name, ptr, name_len);
1573
1574         btrfs_mark_buffer_dirty(path->nodes[0]);
1575         btrfs_free_path(path);
1576
1577         location = &BTRFS_I(inode)->location;
1578         location->objectid = objectid;
1579         location->offset = 0;
1580         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1581
1582         insert_inode_hash(inode);
1583         return inode;
1584 fail:
1585         btrfs_free_path(path);
1586         return ERR_PTR(ret);
1587 }
1588
1589 static inline u8 btrfs_inode_type(struct inode *inode)
1590 {
1591         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1592 }
1593
1594 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1595                             struct dentry *dentry, struct inode *inode,
1596                             int add_backref)
1597 {
1598         int ret;
1599         struct btrfs_key key;
1600         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1601         struct inode *parent_inode;
1602
1603         key.objectid = inode->i_ino;
1604         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1605         key.offset = 0;
1606
1607         ret = btrfs_insert_dir_item(trans, root,
1608                                     dentry->d_name.name, dentry->d_name.len,
1609                                     dentry->d_parent->d_inode->i_ino,
1610                                     &key, btrfs_inode_type(inode));
1611         if (ret == 0) {
1612                 if (add_backref) {
1613                         ret = btrfs_insert_inode_ref(trans, root,
1614                                              dentry->d_name.name,
1615                                              dentry->d_name.len,
1616                                              inode->i_ino,
1617                                              dentry->d_parent->d_inode->i_ino);
1618                 }
1619                 parent_inode = dentry->d_parent->d_inode;
1620                 parent_inode->i_size += dentry->d_name.len * 2;
1621                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1622                 ret = btrfs_update_inode(trans, root,
1623                                          dentry->d_parent->d_inode);
1624         }
1625         return ret;
1626 }
1627
1628 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1629                             struct dentry *dentry, struct inode *inode,
1630                             int backref)
1631 {
1632         int err = btrfs_add_link(trans, dentry, inode, backref);
1633         if (!err) {
1634                 d_instantiate(dentry, inode);
1635                 return 0;
1636         }
1637         if (err > 0)
1638                 err = -EEXIST;
1639         return err;
1640 }
1641
1642 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1643                         int mode, dev_t rdev)
1644 {
1645         struct btrfs_trans_handle *trans;
1646         struct btrfs_root *root = BTRFS_I(dir)->root;
1647         struct inode *inode = NULL;
1648         int err;
1649         int drop_inode = 0;
1650         u64 objectid;
1651         unsigned long nr = 0;
1652
1653         if (!new_valid_dev(rdev))
1654                 return -EINVAL;
1655
1656         mutex_lock(&root->fs_info->fs_mutex);
1657         err = btrfs_check_free_space(root, 1, 0);
1658         if (err)
1659                 goto fail;
1660
1661         trans = btrfs_start_transaction(root, 1);
1662         btrfs_set_trans_block_group(trans, dir);
1663
1664         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1665         if (err) {
1666                 err = -ENOSPC;
1667                 goto out_unlock;
1668         }
1669
1670         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1671                                 dentry->d_name.len,
1672                                 dentry->d_parent->d_inode->i_ino, objectid,
1673                                 BTRFS_I(dir)->block_group, mode);
1674         err = PTR_ERR(inode);
1675         if (IS_ERR(inode))
1676                 goto out_unlock;
1677
1678         btrfs_set_trans_block_group(trans, inode);
1679         err = btrfs_add_nondir(trans, dentry, inode, 0);
1680         if (err)
1681                 drop_inode = 1;
1682         else {
1683                 inode->i_op = &btrfs_special_inode_operations;
1684                 init_special_inode(inode, inode->i_mode, rdev);
1685                 btrfs_update_inode(trans, root, inode);
1686         }
1687         dir->i_sb->s_dirt = 1;
1688         btrfs_update_inode_block_group(trans, inode);
1689         btrfs_update_inode_block_group(trans, dir);
1690 out_unlock:
1691         nr = trans->blocks_used;
1692         btrfs_end_transaction(trans, root);
1693 fail:
1694         mutex_unlock(&root->fs_info->fs_mutex);
1695
1696         if (drop_inode) {
1697                 inode_dec_link_count(inode);
1698                 iput(inode);
1699         }
1700         btrfs_btree_balance_dirty(root, nr);
1701         btrfs_throttle(root);
1702         return err;
1703 }
1704
1705 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1706                         int mode, struct nameidata *nd)
1707 {
1708         struct btrfs_trans_handle *trans;
1709         struct btrfs_root *root = BTRFS_I(dir)->root;
1710         struct inode *inode = NULL;
1711         int err;
1712         int drop_inode = 0;
1713         unsigned long nr = 0;
1714         u64 objectid;
1715
1716         mutex_lock(&root->fs_info->fs_mutex);
1717         err = btrfs_check_free_space(root, 1, 0);
1718         if (err)
1719                 goto fail;
1720         trans = btrfs_start_transaction(root, 1);
1721         btrfs_set_trans_block_group(trans, dir);
1722
1723         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1724         if (err) {
1725                 err = -ENOSPC;
1726                 goto out_unlock;
1727         }
1728
1729         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1730                                 dentry->d_name.len,
1731                                 dentry->d_parent->d_inode->i_ino,
1732                                 objectid, BTRFS_I(dir)->block_group, mode);
1733         err = PTR_ERR(inode);
1734         if (IS_ERR(inode))
1735                 goto out_unlock;
1736
1737         btrfs_set_trans_block_group(trans, inode);
1738         err = btrfs_add_nondir(trans, dentry, inode, 0);
1739         if (err)
1740                 drop_inode = 1;
1741         else {
1742                 inode->i_mapping->a_ops = &btrfs_aops;
1743                 inode->i_fop = &btrfs_file_operations;
1744                 inode->i_op = &btrfs_file_inode_operations;
1745                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1746                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1747                                      inode->i_mapping, GFP_NOFS);
1748                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1749         }
1750         dir->i_sb->s_dirt = 1;
1751         btrfs_update_inode_block_group(trans, inode);
1752         btrfs_update_inode_block_group(trans, dir);
1753 out_unlock:
1754         nr = trans->blocks_used;
1755         btrfs_end_transaction(trans, root);
1756 fail:
1757         mutex_unlock(&root->fs_info->fs_mutex);
1758
1759         if (drop_inode) {
1760                 inode_dec_link_count(inode);
1761                 iput(inode);
1762         }
1763         btrfs_btree_balance_dirty(root, nr);
1764         btrfs_throttle(root);
1765         return err;
1766 }
1767
1768 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1769                       struct dentry *dentry)
1770 {
1771         struct btrfs_trans_handle *trans;
1772         struct btrfs_root *root = BTRFS_I(dir)->root;
1773         struct inode *inode = old_dentry->d_inode;
1774         unsigned long nr = 0;
1775         int err;
1776         int drop_inode = 0;
1777
1778         if (inode->i_nlink == 0)
1779                 return -ENOENT;
1780
1781 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1782         inode->i_nlink++;
1783 #else
1784         inc_nlink(inode);
1785 #endif
1786         mutex_lock(&root->fs_info->fs_mutex);
1787         err = btrfs_check_free_space(root, 1, 0);
1788         if (err)
1789                 goto fail;
1790         trans = btrfs_start_transaction(root, 1);
1791
1792         btrfs_set_trans_block_group(trans, dir);
1793         atomic_inc(&inode->i_count);
1794         err = btrfs_add_nondir(trans, dentry, inode, 1);
1795
1796         if (err)
1797                 drop_inode = 1;
1798
1799         dir->i_sb->s_dirt = 1;
1800         btrfs_update_inode_block_group(trans, dir);
1801         err = btrfs_update_inode(trans, root, inode);
1802
1803         if (err)
1804                 drop_inode = 1;
1805
1806         nr = trans->blocks_used;
1807         btrfs_end_transaction(trans, root);
1808 fail:
1809         mutex_unlock(&root->fs_info->fs_mutex);
1810
1811         if (drop_inode) {
1812                 inode_dec_link_count(inode);
1813                 iput(inode);
1814         }
1815         btrfs_btree_balance_dirty(root, nr);
1816         btrfs_throttle(root);
1817         return err;
1818 }
1819
1820 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1821 {
1822         struct inode *inode;
1823         struct btrfs_trans_handle *trans;
1824         struct btrfs_root *root = BTRFS_I(dir)->root;
1825         int err = 0;
1826         int drop_on_err = 0;
1827         u64 objectid;
1828         unsigned long nr = 1;
1829
1830         mutex_lock(&root->fs_info->fs_mutex);
1831         err = btrfs_check_free_space(root, 1, 0);
1832         if (err)
1833                 goto out_unlock;
1834
1835         trans = btrfs_start_transaction(root, 1);
1836         btrfs_set_trans_block_group(trans, dir);
1837
1838         if (IS_ERR(trans)) {
1839                 err = PTR_ERR(trans);
1840                 goto out_unlock;
1841         }
1842
1843         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1844         if (err) {
1845                 err = -ENOSPC;
1846                 goto out_unlock;
1847         }
1848
1849         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1850                                 dentry->d_name.len,
1851                                 dentry->d_parent->d_inode->i_ino, objectid,
1852                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1853         if (IS_ERR(inode)) {
1854                 err = PTR_ERR(inode);
1855                 goto out_fail;
1856         }
1857
1858         drop_on_err = 1;
1859         inode->i_op = &btrfs_dir_inode_operations;
1860         inode->i_fop = &btrfs_dir_file_operations;
1861         btrfs_set_trans_block_group(trans, inode);
1862
1863         inode->i_size = 0;
1864         err = btrfs_update_inode(trans, root, inode);
1865         if (err)
1866                 goto out_fail;
1867
1868         err = btrfs_add_link(trans, dentry, inode, 0);
1869         if (err)
1870                 goto out_fail;
1871
1872         d_instantiate(dentry, inode);
1873         drop_on_err = 0;
1874         dir->i_sb->s_dirt = 1;
1875         btrfs_update_inode_block_group(trans, inode);
1876         btrfs_update_inode_block_group(trans, dir);
1877
1878 out_fail:
1879         nr = trans->blocks_used;
1880         btrfs_end_transaction(trans, root);
1881
1882 out_unlock:
1883         mutex_unlock(&root->fs_info->fs_mutex);
1884         if (drop_on_err)
1885                 iput(inode);
1886         btrfs_btree_balance_dirty(root, nr);
1887         btrfs_throttle(root);
1888         return err;
1889 }
1890
1891 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1892                                     size_t pg_offset, u64 start, u64 len,
1893                                     int create)
1894 {
1895         int ret;
1896         int err = 0;
1897         u64 bytenr;
1898         u64 extent_start = 0;
1899         u64 extent_end = 0;
1900         u64 objectid = inode->i_ino;
1901         u32 found_type;
1902         struct btrfs_path *path;
1903         struct btrfs_root *root = BTRFS_I(inode)->root;
1904         struct btrfs_file_extent_item *item;
1905         struct extent_buffer *leaf;
1906         struct btrfs_key found_key;
1907         struct extent_map *em = NULL;
1908         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1909         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1910         struct btrfs_trans_handle *trans = NULL;
1911
1912         path = btrfs_alloc_path();
1913         BUG_ON(!path);
1914         mutex_lock(&root->fs_info->fs_mutex);
1915
1916 again:
1917         spin_lock(&em_tree->lock);
1918         em = lookup_extent_mapping(em_tree, start, len);
1919         spin_unlock(&em_tree->lock);
1920
1921         if (em) {
1922                 if (em->start > start) {
1923                         printk("get_extent lookup [%Lu %Lu] em [%Lu %Lu]\n",
1924                                start, len, em->start, em->len);
1925                         WARN_ON(1);
1926                 }
1927                 if (em->block_start == EXTENT_MAP_INLINE && page)
1928                         free_extent_map(em);
1929                 else
1930                         goto out;
1931         }
1932         em = alloc_extent_map(GFP_NOFS);
1933         if (!em) {
1934                 err = -ENOMEM;
1935                 goto out;
1936         }
1937
1938         em->start = EXTENT_MAP_HOLE;
1939         em->len = (u64)-1;
1940         em->bdev = inode->i_sb->s_bdev;
1941         ret = btrfs_lookup_file_extent(trans, root, path,
1942                                        objectid, start, trans != NULL);
1943         if (ret < 0) {
1944                 err = ret;
1945                 goto out;
1946         }
1947
1948         if (ret != 0) {
1949                 if (path->slots[0] == 0)
1950                         goto not_found;
1951                 path->slots[0]--;
1952         }
1953
1954         leaf = path->nodes[0];
1955         item = btrfs_item_ptr(leaf, path->slots[0],
1956                               struct btrfs_file_extent_item);
1957         /* are we inside the extent that was found? */
1958         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1959         found_type = btrfs_key_type(&found_key);
1960         if (found_key.objectid != objectid ||
1961             found_type != BTRFS_EXTENT_DATA_KEY) {
1962                 goto not_found;
1963         }
1964
1965         found_type = btrfs_file_extent_type(leaf, item);
1966         extent_start = found_key.offset;
1967         if (found_type == BTRFS_FILE_EXTENT_REG) {
1968                 extent_end = extent_start +
1969                        btrfs_file_extent_num_bytes(leaf, item);
1970                 err = 0;
1971                 if (start < extent_start || start >= extent_end) {
1972                         em->start = start;
1973                         if (start < extent_start) {
1974                                 if (start + len <= extent_start)
1975                                         goto not_found;
1976                                 em->len = extent_end - extent_start;
1977                         } else {
1978                                 em->len = len;
1979                         }
1980                         goto not_found_em;
1981                 }
1982                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1983                 if (bytenr == 0) {
1984                         em->start = extent_start;
1985                         em->len = extent_end - extent_start;
1986                         em->block_start = EXTENT_MAP_HOLE;
1987                         goto insert;
1988                 }
1989                 bytenr += btrfs_file_extent_offset(leaf, item);
1990                 em->block_start = bytenr;
1991                 em->start = extent_start;
1992                 em->len = extent_end - extent_start;
1993                 goto insert;
1994         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1995                 u64 page_start;
1996                 unsigned long ptr;
1997                 char *map;
1998                 size_t size;
1999                 size_t extent_offset;
2000                 size_t copy_size;
2001
2002                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2003                                                     path->slots[0]));
2004                 extent_end = (extent_start + size + root->sectorsize - 1) &
2005                         ~((u64)root->sectorsize - 1);
2006                 if (start < extent_start || start >= extent_end) {
2007                         em->start = start;
2008                         if (start < extent_start) {
2009                                 if (start + len <= extent_start)
2010                                         goto not_found;
2011                                 em->len = extent_end - extent_start;
2012                         } else {
2013                                 em->len = len;
2014                         }
2015                         goto not_found_em;
2016                 }
2017                 em->block_start = EXTENT_MAP_INLINE;
2018
2019                 if (!page) {
2020                         em->start = extent_start;
2021                         em->len = size;
2022                         goto out;
2023                 }
2024
2025                 page_start = page_offset(page) + pg_offset;
2026                 extent_offset = page_start - extent_start;
2027                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2028                                 size - extent_offset);
2029                 em->start = extent_start + extent_offset;
2030                 em->len = (copy_size + root->sectorsize - 1) &
2031                         ~((u64)root->sectorsize - 1);
2032                 map = kmap(page);
2033                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2034                 if (create == 0 && !PageUptodate(page)) {
2035                         read_extent_buffer(leaf, map + pg_offset, ptr,
2036                                            copy_size);
2037                         flush_dcache_page(page);
2038                 } else if (create && PageUptodate(page)) {
2039                         if (!trans) {
2040                                 kunmap(page);
2041                                 free_extent_map(em);
2042                                 em = NULL;
2043                                 btrfs_release_path(root, path);
2044                                 trans = btrfs_start_transaction(root, 1);
2045                                 goto again;
2046                         }
2047                         write_extent_buffer(leaf, map + pg_offset, ptr,
2048                                             copy_size);
2049                         btrfs_mark_buffer_dirty(leaf);
2050                 }
2051                 kunmap(page);
2052                 set_extent_uptodate(io_tree, em->start,
2053                                     extent_map_end(em) - 1, GFP_NOFS);
2054                 goto insert;
2055         } else {
2056                 printk("unkknown found_type %d\n", found_type);
2057                 WARN_ON(1);
2058         }
2059 not_found:
2060         em->start = start;
2061         em->len = len;
2062 not_found_em:
2063         em->block_start = EXTENT_MAP_HOLE;
2064 insert:
2065         btrfs_release_path(root, path);
2066         if (em->start > start || extent_map_end(em) <= start) {
2067                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2068                 err = -EIO;
2069                 goto out;
2070         }
2071
2072         err = 0;
2073         spin_lock(&em_tree->lock);
2074         ret = add_extent_mapping(em_tree, em);
2075         if (ret == -EEXIST) {
2076                 free_extent_map(em);
2077                 em = lookup_extent_mapping(em_tree, start, len);
2078                 if (!em) {
2079                         err = -EIO;
2080                         printk("failing to insert %Lu %Lu\n", start, len);
2081                 }
2082         }
2083         spin_unlock(&em_tree->lock);
2084 out:
2085         btrfs_free_path(path);
2086         if (trans) {
2087                 ret = btrfs_end_transaction(trans, root);
2088                 if (!err)
2089                         err = ret;
2090         }
2091         mutex_unlock(&root->fs_info->fs_mutex);
2092         if (err) {
2093                 free_extent_map(em);
2094                 WARN_ON(1);
2095                 return ERR_PTR(err);
2096         }
2097         return em;
2098 }
2099
2100 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2101 {
2102         return extent_bmap(mapping, iblock, btrfs_get_extent);
2103 }
2104
2105 int btrfs_readpage(struct file *file, struct page *page)
2106 {
2107         struct extent_io_tree *tree;
2108         tree = &BTRFS_I(page->mapping->host)->io_tree;
2109         return extent_read_full_page(tree, page, btrfs_get_extent);
2110 }
2111
2112 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2113 {
2114         struct extent_io_tree *tree;
2115
2116
2117         if (current->flags & PF_MEMALLOC) {
2118                 redirty_page_for_writepage(wbc, page);
2119                 unlock_page(page);
2120                 return 0;
2121         }
2122         tree = &BTRFS_I(page->mapping->host)->io_tree;
2123         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2124 }
2125
2126 static int btrfs_writepages(struct address_space *mapping,
2127                             struct writeback_control *wbc)
2128 {
2129         struct extent_io_tree *tree;
2130         tree = &BTRFS_I(mapping->host)->io_tree;
2131         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2132 }
2133
2134 static int
2135 btrfs_readpages(struct file *file, struct address_space *mapping,
2136                 struct list_head *pages, unsigned nr_pages)
2137 {
2138         struct extent_io_tree *tree;
2139         tree = &BTRFS_I(mapping->host)->io_tree;
2140         return extent_readpages(tree, mapping, pages, nr_pages,
2141                                 btrfs_get_extent);
2142 }
2143
2144 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2145 {
2146         struct extent_io_tree *tree;
2147         struct extent_map_tree *map;
2148         int ret;
2149
2150         tree = &BTRFS_I(page->mapping->host)->io_tree;
2151         map = &BTRFS_I(page->mapping->host)->extent_tree;
2152         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2153         if (ret == 1) {
2154                 ClearPagePrivate(page);
2155                 set_page_private(page, 0);
2156                 page_cache_release(page);
2157         }
2158         return ret;
2159 }
2160
2161 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2162 {
2163         struct extent_io_tree *tree;
2164
2165         tree = &BTRFS_I(page->mapping->host)->io_tree;
2166         extent_invalidatepage(tree, page, offset);
2167         btrfs_releasepage(page, GFP_NOFS);
2168 }
2169
2170 /*
2171  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2172  * called from a page fault handler when a page is first dirtied. Hence we must
2173  * be careful to check for EOF conditions here. We set the page up correctly
2174  * for a written page which means we get ENOSPC checking when writing into
2175  * holes and correct delalloc and unwritten extent mapping on filesystems that
2176  * support these features.
2177  *
2178  * We are not allowed to take the i_mutex here so we have to play games to
2179  * protect against truncate races as the page could now be beyond EOF.  Because
2180  * vmtruncate() writes the inode size before removing pages, once we have the
2181  * page lock we can determine safely if the page is beyond EOF. If it is not
2182  * beyond EOF, then the page is guaranteed safe against truncation until we
2183  * unlock the page.
2184  */
2185 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2186 {
2187         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2188         struct btrfs_root *root = BTRFS_I(inode)->root;
2189         unsigned long end;
2190         loff_t size;
2191         int ret;
2192         u64 page_start;
2193
2194         mutex_lock(&root->fs_info->fs_mutex);
2195         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2196         mutex_unlock(&root->fs_info->fs_mutex);
2197         if (ret)
2198                 goto out;
2199
2200         ret = -EINVAL;
2201
2202         lock_page(page);
2203         wait_on_page_writeback(page);
2204         size = i_size_read(inode);
2205         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2206
2207         if ((page->mapping != inode->i_mapping) ||
2208             (page_start > size)) {
2209                 /* page got truncated out from underneath us */
2210                 goto out_unlock;
2211         }
2212
2213         /* page is wholly or partially inside EOF */
2214         if (page_start + PAGE_CACHE_SIZE > size)
2215                 end = size & ~PAGE_CACHE_MASK;
2216         else
2217                 end = PAGE_CACHE_SIZE;
2218
2219         ret = btrfs_cow_one_page(inode, page, end);
2220
2221 out_unlock:
2222         unlock_page(page);
2223 out:
2224         return ret;
2225 }
2226
2227 static void btrfs_truncate(struct inode *inode)
2228 {
2229         struct btrfs_root *root = BTRFS_I(inode)->root;
2230         int ret;
2231         struct btrfs_trans_handle *trans;
2232         unsigned long nr;
2233
2234         if (!S_ISREG(inode->i_mode))
2235                 return;
2236         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2237                 return;
2238
2239         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2240
2241         mutex_lock(&root->fs_info->fs_mutex);
2242         trans = btrfs_start_transaction(root, 1);
2243         btrfs_set_trans_block_group(trans, inode);
2244
2245         /* FIXME, add redo link to tree so we don't leak on crash */
2246         ret = btrfs_truncate_in_trans(trans, root, inode,
2247                                       BTRFS_EXTENT_DATA_KEY);
2248         btrfs_update_inode(trans, root, inode);
2249         nr = trans->blocks_used;
2250
2251         ret = btrfs_end_transaction(trans, root);
2252         BUG_ON(ret);
2253         mutex_unlock(&root->fs_info->fs_mutex);
2254         btrfs_btree_balance_dirty(root, nr);
2255         btrfs_throttle(root);
2256 }
2257
2258 static int noinline create_subvol(struct btrfs_root *root, char *name,
2259                                   int namelen)
2260 {
2261         struct btrfs_trans_handle *trans;
2262         struct btrfs_key key;
2263         struct btrfs_root_item root_item;
2264         struct btrfs_inode_item *inode_item;
2265         struct extent_buffer *leaf;
2266         struct btrfs_root *new_root = root;
2267         struct inode *inode;
2268         struct inode *dir;
2269         int ret;
2270         int err;
2271         u64 objectid;
2272         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2273         unsigned long nr = 1;
2274
2275         mutex_lock(&root->fs_info->fs_mutex);
2276         ret = btrfs_check_free_space(root, 1, 0);
2277         if (ret)
2278                 goto fail_commit;
2279
2280         trans = btrfs_start_transaction(root, 1);
2281         BUG_ON(!trans);
2282
2283         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2284                                        0, &objectid);
2285         if (ret)
2286                 goto fail;
2287
2288         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2289                                         objectid, trans->transid, 0, 0,
2290                                         0, 0);
2291         if (IS_ERR(leaf))
2292                 return PTR_ERR(leaf);
2293
2294         btrfs_set_header_nritems(leaf, 0);
2295         btrfs_set_header_level(leaf, 0);
2296         btrfs_set_header_bytenr(leaf, leaf->start);
2297         btrfs_set_header_generation(leaf, trans->transid);
2298         btrfs_set_header_owner(leaf, objectid);
2299
2300         write_extent_buffer(leaf, root->fs_info->fsid,
2301                             (unsigned long)btrfs_header_fsid(leaf),
2302                             BTRFS_FSID_SIZE);
2303         btrfs_mark_buffer_dirty(leaf);
2304
2305         inode_item = &root_item.inode;
2306         memset(inode_item, 0, sizeof(*inode_item));
2307         inode_item->generation = cpu_to_le64(1);
2308         inode_item->size = cpu_to_le64(3);
2309         inode_item->nlink = cpu_to_le32(1);
2310         inode_item->nblocks = cpu_to_le64(1);
2311         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2312
2313         btrfs_set_root_bytenr(&root_item, leaf->start);
2314         btrfs_set_root_level(&root_item, 0);
2315         btrfs_set_root_refs(&root_item, 1);
2316         btrfs_set_root_used(&root_item, 0);
2317
2318         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2319         root_item.drop_level = 0;
2320
2321         free_extent_buffer(leaf);
2322         leaf = NULL;
2323
2324         btrfs_set_root_dirid(&root_item, new_dirid);
2325
2326         key.objectid = objectid;
2327         key.offset = 1;
2328         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2329         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2330                                 &root_item);
2331         if (ret)
2332                 goto fail;
2333
2334         /*
2335          * insert the directory item
2336          */
2337         key.offset = (u64)-1;
2338         dir = root->fs_info->sb->s_root->d_inode;
2339         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2340                                     name, namelen, dir->i_ino, &key,
2341                                     BTRFS_FT_DIR);
2342         if (ret)
2343                 goto fail;
2344
2345         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2346                              name, namelen, objectid,
2347                              root->fs_info->sb->s_root->d_inode->i_ino);
2348         if (ret)
2349                 goto fail;
2350
2351         ret = btrfs_commit_transaction(trans, root);
2352         if (ret)
2353                 goto fail_commit;
2354
2355         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2356         BUG_ON(!new_root);
2357
2358         trans = btrfs_start_transaction(new_root, 1);
2359         BUG_ON(!trans);
2360
2361         inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2362                                 new_dirid,
2363                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2364         if (IS_ERR(inode))
2365                 goto fail;
2366         inode->i_op = &btrfs_dir_inode_operations;
2367         inode->i_fop = &btrfs_dir_file_operations;
2368         new_root->inode = inode;
2369
2370         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2371                                      new_dirid);
2372         inode->i_nlink = 1;
2373         inode->i_size = 0;
2374         ret = btrfs_update_inode(trans, new_root, inode);
2375         if (ret)
2376                 goto fail;
2377 fail:
2378         nr = trans->blocks_used;
2379         err = btrfs_commit_transaction(trans, new_root);
2380         if (err && !ret)
2381                 ret = err;
2382 fail_commit:
2383         mutex_unlock(&root->fs_info->fs_mutex);
2384         btrfs_btree_balance_dirty(root, nr);
2385         btrfs_throttle(root);
2386         return ret;
2387 }
2388
2389 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2390 {
2391         struct btrfs_pending_snapshot *pending_snapshot;
2392         struct btrfs_trans_handle *trans;
2393         int ret;
2394         int err;
2395         unsigned long nr = 0;
2396
2397         if (!root->ref_cows)
2398                 return -EINVAL;
2399
2400         mutex_lock(&root->fs_info->fs_mutex);
2401         ret = btrfs_check_free_space(root, 1, 0);
2402         if (ret)
2403                 goto fail_unlock;
2404
2405         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2406         if (!pending_snapshot) {
2407                 ret = -ENOMEM;
2408                 goto fail_unlock;
2409         }
2410         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2411         if (!pending_snapshot->name) {
2412                 ret = -ENOMEM;
2413                 kfree(pending_snapshot);
2414                 goto fail_unlock;
2415         }
2416         memcpy(pending_snapshot->name, name, namelen);
2417         pending_snapshot->name[namelen] = '\0';
2418         trans = btrfs_start_transaction(root, 1);
2419         BUG_ON(!trans);
2420         pending_snapshot->root = root;
2421         list_add(&pending_snapshot->list,
2422                  &trans->transaction->pending_snapshots);
2423         ret = btrfs_update_inode(trans, root, root->inode);
2424         err = btrfs_commit_transaction(trans, root);
2425
2426 fail_unlock:
2427         mutex_unlock(&root->fs_info->fs_mutex);
2428         btrfs_btree_balance_dirty(root, nr);
2429         btrfs_throttle(root);
2430         return ret;
2431 }
2432
2433 unsigned long btrfs_force_ra(struct address_space *mapping,
2434                               struct file_ra_state *ra, struct file *file,
2435                               pgoff_t offset, pgoff_t last_index)
2436 {
2437         pgoff_t req_size;
2438
2439 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2440         req_size = last_index - offset + 1;
2441         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2442         return offset;
2443 #else
2444         req_size = min(last_index - offset + 1, (pgoff_t)128);
2445         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2446         return offset + req_size;
2447 #endif
2448 }
2449
2450 int btrfs_defrag_file(struct file *file) {
2451         struct inode *inode = fdentry(file)->d_inode;
2452         struct btrfs_root *root = BTRFS_I(inode)->root;
2453         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2454         struct page *page;
2455         unsigned long last_index;
2456         unsigned long ra_index = 0;
2457         u64 page_start;
2458         u64 page_end;
2459         unsigned long i;
2460         int ret;
2461
2462         mutex_lock(&root->fs_info->fs_mutex);
2463         ret = btrfs_check_free_space(root, inode->i_size, 0);
2464         mutex_unlock(&root->fs_info->fs_mutex);
2465         if (ret)
2466                 return -ENOSPC;
2467
2468         mutex_lock(&inode->i_mutex);
2469         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2470         for (i = 0; i <= last_index; i++) {
2471                 if (i == ra_index) {
2472                         ra_index = btrfs_force_ra(inode->i_mapping,
2473                                                   &file->f_ra,
2474                                                   file, ra_index, last_index);
2475                 }
2476                 page = grab_cache_page(inode->i_mapping, i);
2477                 if (!page)
2478                         goto out_unlock;
2479                 if (!PageUptodate(page)) {
2480                         btrfs_readpage(NULL, page);
2481                         lock_page(page);
2482                         if (!PageUptodate(page)) {
2483                                 unlock_page(page);
2484                                 page_cache_release(page);
2485                                 goto out_unlock;
2486                         }
2487                 }
2488                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2489                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2490
2491                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2492                 set_extent_delalloc(io_tree, page_start,
2493                                     page_end, GFP_NOFS);
2494
2495                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2496                 set_page_dirty(page);
2497                 unlock_page(page);
2498                 page_cache_release(page);
2499                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2500         }
2501
2502 out_unlock:
2503         mutex_unlock(&inode->i_mutex);
2504         return 0;
2505 }
2506
2507 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2508 {
2509         u64 new_size;
2510         u64 old_size;
2511         struct btrfs_ioctl_vol_args *vol_args;
2512         struct btrfs_trans_handle *trans;
2513         char *sizestr;
2514         int ret = 0;
2515         int namelen;
2516         int mod = 0;
2517
2518         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2519
2520         if (!vol_args)
2521                 return -ENOMEM;
2522
2523         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2524                 ret = -EFAULT;
2525                 goto out;
2526         }
2527         namelen = strlen(vol_args->name);
2528         if (namelen > BTRFS_VOL_NAME_MAX) {
2529                 ret = -EINVAL;
2530                 goto out;
2531         }
2532
2533         sizestr = vol_args->name;
2534         if (!strcmp(sizestr, "max"))
2535                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2536         else {
2537                 if (sizestr[0] == '-') {
2538                         mod = -1;
2539                         sizestr++;
2540                 } else if (sizestr[0] == '+') {
2541                         mod = 1;
2542                         sizestr++;
2543                 }
2544                 new_size = btrfs_parse_size(sizestr);
2545                 if (new_size == 0) {
2546                         ret = -EINVAL;
2547                         goto out;
2548                 }
2549         }
2550
2551         mutex_lock(&root->fs_info->fs_mutex);
2552         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2553
2554         if (mod < 0) {
2555                 if (new_size > old_size) {
2556                         ret = -EINVAL;
2557                         goto out_unlock;
2558                 }
2559                 new_size = old_size - new_size;
2560         } else if (mod > 0) {
2561                 new_size = old_size + new_size;
2562         }
2563
2564         if (new_size < 256 * 1024 * 1024) {
2565                 ret = -EINVAL;
2566                 goto out_unlock;
2567         }
2568         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2569                 ret = -EFBIG;
2570                 goto out_unlock;
2571         }
2572
2573         do_div(new_size, root->sectorsize);
2574         new_size *= root->sectorsize;
2575
2576 printk("new size is %Lu\n", new_size);
2577         if (new_size > old_size) {
2578                 trans = btrfs_start_transaction(root, 1);
2579                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2580                 btrfs_commit_transaction(trans, root);
2581         } else {
2582                 ret = btrfs_shrink_extent_tree(root, new_size);
2583         }
2584
2585 out_unlock:
2586         mutex_unlock(&root->fs_info->fs_mutex);
2587 out:
2588         kfree(vol_args);
2589         return ret;
2590 }
2591
2592 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2593                                             void __user *arg)
2594 {
2595         struct btrfs_ioctl_vol_args *vol_args;
2596         struct btrfs_dir_item *di;
2597         struct btrfs_path *path;
2598         u64 root_dirid;
2599         int namelen;
2600         int ret;
2601
2602         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2603
2604         if (!vol_args)
2605                 return -ENOMEM;
2606
2607         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2608                 ret = -EFAULT;
2609                 goto out;
2610         }
2611
2612         namelen = strlen(vol_args->name);
2613         if (namelen > BTRFS_VOL_NAME_MAX) {
2614                 ret = -EINVAL;
2615                 goto out;
2616         }
2617         if (strchr(vol_args->name, '/')) {
2618                 ret = -EINVAL;
2619                 goto out;
2620         }
2621
2622         path = btrfs_alloc_path();
2623         if (!path) {
2624                 ret = -ENOMEM;
2625                 goto out;
2626         }
2627
2628         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2629         mutex_lock(&root->fs_info->fs_mutex);
2630         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2631                             path, root_dirid,
2632                             vol_args->name, namelen, 0);
2633         mutex_unlock(&root->fs_info->fs_mutex);
2634         btrfs_free_path(path);
2635
2636         if (di && !IS_ERR(di)) {
2637                 ret = -EEXIST;
2638                 goto out;
2639         }
2640
2641         if (IS_ERR(di)) {
2642                 ret = PTR_ERR(di);
2643                 goto out;
2644         }
2645
2646         if (root == root->fs_info->tree_root)
2647                 ret = create_subvol(root, vol_args->name, namelen);
2648         else
2649                 ret = create_snapshot(root, vol_args->name, namelen);
2650 out:
2651         kfree(vol_args);
2652         return ret;
2653 }
2654
2655 static int btrfs_ioctl_defrag(struct file *file)
2656 {
2657         struct inode *inode = fdentry(file)->d_inode;
2658         struct btrfs_root *root = BTRFS_I(inode)->root;
2659
2660         switch (inode->i_mode & S_IFMT) {
2661         case S_IFDIR:
2662                 mutex_lock(&root->fs_info->fs_mutex);
2663                 btrfs_defrag_root(root, 0);
2664                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2665                 mutex_unlock(&root->fs_info->fs_mutex);
2666                 break;
2667         case S_IFREG:
2668                 btrfs_defrag_file(file);
2669                 break;
2670         }
2671
2672         return 0;
2673 }
2674
2675 long btrfs_ioctl(struct file *file, unsigned int
2676                 cmd, unsigned long arg)
2677 {
2678         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2679
2680         switch (cmd) {
2681         case BTRFS_IOC_SNAP_CREATE:
2682                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2683         case BTRFS_IOC_DEFRAG:
2684                 return btrfs_ioctl_defrag(file);
2685         case BTRFS_IOC_RESIZE:
2686                 return btrfs_ioctl_resize(root, (void __user *)arg);
2687         }
2688
2689         return -ENOTTY;
2690 }
2691
2692 /*
2693  * Called inside transaction, so use GFP_NOFS
2694  */
2695 struct inode *btrfs_alloc_inode(struct super_block *sb)
2696 {
2697         struct btrfs_inode *ei;
2698
2699         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2700         if (!ei)
2701                 return NULL;
2702         ei->last_trans = 0;
2703         ei->ordered_trans = 0;
2704         return &ei->vfs_inode;
2705 }
2706
2707 void btrfs_destroy_inode(struct inode *inode)
2708 {
2709         WARN_ON(!list_empty(&inode->i_dentry));
2710         WARN_ON(inode->i_data.nrpages);
2711
2712         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2713         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2714 }
2715
2716 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2717 static void init_once(struct kmem_cache * cachep, void *foo)
2718 #else
2719 static void init_once(void * foo, struct kmem_cache * cachep,
2720                       unsigned long flags)
2721 #endif
2722 {
2723         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2724
2725         inode_init_once(&ei->vfs_inode);
2726 }
2727
2728 void btrfs_destroy_cachep(void)
2729 {
2730         if (btrfs_inode_cachep)
2731                 kmem_cache_destroy(btrfs_inode_cachep);
2732         if (btrfs_trans_handle_cachep)
2733                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2734         if (btrfs_transaction_cachep)
2735                 kmem_cache_destroy(btrfs_transaction_cachep);
2736         if (btrfs_bit_radix_cachep)
2737                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2738         if (btrfs_path_cachep)
2739                 kmem_cache_destroy(btrfs_path_cachep);
2740 }
2741
2742 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2743                                        unsigned long extra_flags,
2744 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2745                                        void (*ctor)(struct kmem_cache *, void *)
2746 #else
2747                                        void (*ctor)(void *, struct kmem_cache *,
2748                                                     unsigned long)
2749 #endif
2750                                      )
2751 {
2752         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2753                                  SLAB_MEM_SPREAD | extra_flags), ctor
2754 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2755                                  ,NULL
2756 #endif
2757                                 );
2758 }
2759
2760 int btrfs_init_cachep(void)
2761 {
2762         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2763                                           sizeof(struct btrfs_inode),
2764                                           0, init_once);
2765         if (!btrfs_inode_cachep)
2766                 goto fail;
2767         btrfs_trans_handle_cachep =
2768                         btrfs_cache_create("btrfs_trans_handle_cache",
2769                                            sizeof(struct btrfs_trans_handle),
2770                                            0, NULL);
2771         if (!btrfs_trans_handle_cachep)
2772                 goto fail;
2773         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2774                                              sizeof(struct btrfs_transaction),
2775                                              0, NULL);
2776         if (!btrfs_transaction_cachep)
2777                 goto fail;
2778         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2779                                          sizeof(struct btrfs_path),
2780                                          0, NULL);
2781         if (!btrfs_path_cachep)
2782                 goto fail;
2783         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2784                                               SLAB_DESTROY_BY_RCU, NULL);
2785         if (!btrfs_bit_radix_cachep)
2786                 goto fail;
2787         return 0;
2788 fail:
2789         btrfs_destroy_cachep();
2790         return -ENOMEM;
2791 }
2792
2793 static int btrfs_getattr(struct vfsmount *mnt,
2794                          struct dentry *dentry, struct kstat *stat)
2795 {
2796         struct inode *inode = dentry->d_inode;
2797         generic_fillattr(inode, stat);
2798         stat->blksize = PAGE_CACHE_SIZE;
2799         return 0;
2800 }
2801
2802 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2803                            struct inode * new_dir,struct dentry *new_dentry)
2804 {
2805         struct btrfs_trans_handle *trans;
2806         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2807         struct inode *new_inode = new_dentry->d_inode;
2808         struct inode *old_inode = old_dentry->d_inode;
2809         struct timespec ctime = CURRENT_TIME;
2810         struct btrfs_path *path;
2811         int ret;
2812
2813         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2814             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2815                 return -ENOTEMPTY;
2816         }
2817
2818         mutex_lock(&root->fs_info->fs_mutex);
2819         ret = btrfs_check_free_space(root, 1, 0);
2820         if (ret)
2821                 goto out_unlock;
2822
2823         trans = btrfs_start_transaction(root, 1);
2824
2825         btrfs_set_trans_block_group(trans, new_dir);
2826         path = btrfs_alloc_path();
2827         if (!path) {
2828                 ret = -ENOMEM;
2829                 goto out_fail;
2830         }
2831
2832         old_dentry->d_inode->i_nlink++;
2833         old_dir->i_ctime = old_dir->i_mtime = ctime;
2834         new_dir->i_ctime = new_dir->i_mtime = ctime;
2835         old_inode->i_ctime = ctime;
2836
2837         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2838         if (ret)
2839                 goto out_fail;
2840
2841         if (new_inode) {
2842                 new_inode->i_ctime = CURRENT_TIME;
2843                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2844                 if (ret)
2845                         goto out_fail;
2846         }
2847         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
2848         if (ret)
2849                 goto out_fail;
2850
2851 out_fail:
2852         btrfs_free_path(path);
2853         btrfs_end_transaction(trans, root);
2854 out_unlock:
2855         mutex_unlock(&root->fs_info->fs_mutex);
2856         return ret;
2857 }
2858
2859 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2860                          const char *symname)
2861 {
2862         struct btrfs_trans_handle *trans;
2863         struct btrfs_root *root = BTRFS_I(dir)->root;
2864         struct btrfs_path *path;
2865         struct btrfs_key key;
2866         struct inode *inode = NULL;
2867         int err;
2868         int drop_inode = 0;
2869         u64 objectid;
2870         int name_len;
2871         int datasize;
2872         unsigned long ptr;
2873         struct btrfs_file_extent_item *ei;
2874         struct extent_buffer *leaf;
2875         unsigned long nr = 0;
2876
2877         name_len = strlen(symname) + 1;
2878         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2879                 return -ENAMETOOLONG;
2880
2881         mutex_lock(&root->fs_info->fs_mutex);
2882         err = btrfs_check_free_space(root, 1, 0);
2883         if (err)
2884                 goto out_fail;
2885
2886         trans = btrfs_start_transaction(root, 1);
2887         btrfs_set_trans_block_group(trans, dir);
2888
2889         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2890         if (err) {
2891                 err = -ENOSPC;
2892                 goto out_unlock;
2893         }
2894
2895         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2896                                 dentry->d_name.len,
2897                                 dentry->d_parent->d_inode->i_ino, objectid,
2898                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2899         err = PTR_ERR(inode);
2900         if (IS_ERR(inode))
2901                 goto out_unlock;
2902
2903         btrfs_set_trans_block_group(trans, inode);
2904         err = btrfs_add_nondir(trans, dentry, inode, 0);
2905         if (err)
2906                 drop_inode = 1;
2907         else {
2908                 inode->i_mapping->a_ops = &btrfs_aops;
2909                 inode->i_fop = &btrfs_file_operations;
2910                 inode->i_op = &btrfs_file_inode_operations;
2911                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2912                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2913                                      inode->i_mapping, GFP_NOFS);
2914                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2915         }
2916         dir->i_sb->s_dirt = 1;
2917         btrfs_update_inode_block_group(trans, inode);
2918         btrfs_update_inode_block_group(trans, dir);
2919         if (drop_inode)
2920                 goto out_unlock;
2921
2922         path = btrfs_alloc_path();
2923         BUG_ON(!path);
2924         key.objectid = inode->i_ino;
2925         key.offset = 0;
2926         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2927         datasize = btrfs_file_extent_calc_inline_size(name_len);
2928         err = btrfs_insert_empty_item(trans, root, path, &key,
2929                                       datasize);
2930         if (err) {
2931                 drop_inode = 1;
2932                 goto out_unlock;
2933         }
2934         leaf = path->nodes[0];
2935         ei = btrfs_item_ptr(leaf, path->slots[0],
2936                             struct btrfs_file_extent_item);
2937         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2938         btrfs_set_file_extent_type(leaf, ei,
2939                                    BTRFS_FILE_EXTENT_INLINE);
2940         ptr = btrfs_file_extent_inline_start(ei);
2941         write_extent_buffer(leaf, symname, ptr, name_len);
2942         btrfs_mark_buffer_dirty(leaf);
2943         btrfs_free_path(path);
2944
2945         inode->i_op = &btrfs_symlink_inode_operations;
2946         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2947         inode->i_size = name_len - 1;
2948         err = btrfs_update_inode(trans, root, inode);
2949         if (err)
2950                 drop_inode = 1;
2951
2952 out_unlock:
2953         nr = trans->blocks_used;
2954         btrfs_end_transaction(trans, root);
2955 out_fail:
2956         mutex_unlock(&root->fs_info->fs_mutex);
2957         if (drop_inode) {
2958                 inode_dec_link_count(inode);
2959                 iput(inode);
2960         }
2961         btrfs_btree_balance_dirty(root, nr);
2962         btrfs_throttle(root);
2963         return err;
2964 }
2965 static int btrfs_permission(struct inode *inode, int mask,
2966                             struct nameidata *nd)
2967 {
2968         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2969                 return -EACCES;
2970         return generic_permission(inode, mask, NULL);
2971 }
2972
2973 static struct inode_operations btrfs_dir_inode_operations = {
2974         .lookup         = btrfs_lookup,
2975         .create         = btrfs_create,
2976         .unlink         = btrfs_unlink,
2977         .link           = btrfs_link,
2978         .mkdir          = btrfs_mkdir,
2979         .rmdir          = btrfs_rmdir,
2980         .rename         = btrfs_rename,
2981         .symlink        = btrfs_symlink,
2982         .setattr        = btrfs_setattr,
2983         .mknod          = btrfs_mknod,
2984         .setxattr       = generic_setxattr,
2985         .getxattr       = generic_getxattr,
2986         .listxattr      = btrfs_listxattr,
2987         .removexattr    = generic_removexattr,
2988         .permission     = btrfs_permission,
2989 };
2990 static struct inode_operations btrfs_dir_ro_inode_operations = {
2991         .lookup         = btrfs_lookup,
2992         .permission     = btrfs_permission,
2993 };
2994 static struct file_operations btrfs_dir_file_operations = {
2995         .llseek         = generic_file_llseek,
2996         .read           = generic_read_dir,
2997         .readdir        = btrfs_readdir,
2998         .unlocked_ioctl = btrfs_ioctl,
2999 #ifdef CONFIG_COMPAT
3000         .compat_ioctl   = btrfs_ioctl,
3001 #endif
3002 };
3003
3004 static struct extent_io_ops btrfs_extent_io_ops = {
3005         .fill_delalloc = run_delalloc_range,
3006         .writepage_io_hook = btrfs_writepage_io_hook,
3007         .readpage_io_hook = btrfs_readpage_io_hook,
3008         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3009 };
3010
3011 static struct address_space_operations btrfs_aops = {
3012         .readpage       = btrfs_readpage,
3013         .writepage      = btrfs_writepage,
3014         .writepages     = btrfs_writepages,
3015         .readpages      = btrfs_readpages,
3016         .sync_page      = block_sync_page,
3017         .bmap           = btrfs_bmap,
3018         .invalidatepage = btrfs_invalidatepage,
3019         .releasepage    = btrfs_releasepage,
3020         .set_page_dirty = __set_page_dirty_nobuffers,
3021 };
3022
3023 static struct address_space_operations btrfs_symlink_aops = {
3024         .readpage       = btrfs_readpage,
3025         .writepage      = btrfs_writepage,
3026         .invalidatepage = btrfs_invalidatepage,
3027         .releasepage    = btrfs_releasepage,
3028 };
3029
3030 static struct inode_operations btrfs_file_inode_operations = {
3031         .truncate       = btrfs_truncate,
3032         .getattr        = btrfs_getattr,
3033         .setattr        = btrfs_setattr,
3034         .setxattr       = generic_setxattr,
3035         .getxattr       = generic_getxattr,
3036         .listxattr      = btrfs_listxattr,
3037         .removexattr    = generic_removexattr,
3038         .permission     = btrfs_permission,
3039 };
3040 static struct inode_operations btrfs_special_inode_operations = {
3041         .getattr        = btrfs_getattr,
3042         .setattr        = btrfs_setattr,
3043         .permission     = btrfs_permission,
3044 };
3045 static struct inode_operations btrfs_symlink_inode_operations = {
3046         .readlink       = generic_readlink,
3047         .follow_link    = page_follow_link_light,
3048         .put_link       = page_put_link,
3049         .permission     = btrfs_permission,
3050 };