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                                 extent_num_bytes = extent_num_bytes &
 827                                         ~((u64)root->sectorsize - 1);
 828                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 829                                                          extent_num_bytes);
 830                                 num_dec = (orig_num_bytes -
 831                                            extent_num_bytes) >> 9;
 832                                 if (extent_start != 0) {
 833                                         inode->i_blocks -= num_dec;
 834                                 }
 835                                 btrfs_mark_buffer_dirty(leaf);
 836                         } else {
 837                                 extent_num_bytes =
 838                                         btrfs_file_extent_disk_num_bytes(leaf,
 839                                                                          fi);
 840                                 /* FIXME blocksize != 4096 */
 841                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 842                                                                        fi) >> 9;
 843                                 if (extent_start != 0) {
 844                                         found_extent = 1;
 845                                         inode->i_blocks -= num_dec;
 846                                 }
 847                                 root_gen = btrfs_header_generation(leaf);
 848                                 root_owner = btrfs_header_owner(leaf);
 849                         }
 850                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 851                            !del_item) {
 852                         u32 newsize = inode->i_size - found_key.offset;
 853                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 854                         ret = btrfs_truncate_item(trans, root, path,
 855                                                   newsize, 1);
 856                         BUG_ON(ret);
 857                 }
 858 delete:
 859                 if (del_item) {
 860                         if (!pending_del_nr) {
 861                                 /* no pending yet, add ourselves */
 862                                 pending_del_slot = path->slots[0];
 863                                 pending_del_nr = 1;
 864                         } else if (pending_del_nr &&
 865                                    path->slots[0] + 1 == pending_del_slot) {
 866                                 /* hop on the pending chunk */
 867                                 pending_del_nr++;
 868                                 pending_del_slot = path->slots[0];
 869                         } else {
 870                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
 871                         }
 872                 } else {
 873                         break;
 874                 }
 875                 if (found_extent) {
 876                         ret = btrfs_free_extent(trans, root, extent_start,
 877                                                 extent_num_bytes,
 878                                                 root_owner,
 879                                                 root_gen, inode->i_ino,
 880                                                 found_key.offset, 0);
 881                         BUG_ON(ret);
 882                 }
 883 next:
 884                 if (path->slots[0] == 0) {
 885                         if (pending_del_nr)
 886                                 goto del_pending;
 887                         btrfs_release_path(root, path);
 888                         goto search_again;
 889                 }
 890
 891                 path->slots[0]--;
 892                 if (pending_del_nr &&
 893                     path->slots[0] + 1 != pending_del_slot) {
 894                         struct btrfs_key debug;
 895 del_pending:
 896                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
 897                                               pending_del_slot);
 898                         ret = btrfs_del_items(trans, root, path,
 899                                               pending_del_slot,
 900                                               pending_del_nr);
 901                         BUG_ON(ret);
 902                         pending_del_nr = 0;
 903                         btrfs_release_path(root, path);
 904                         goto search_again;
 905                 }
 906         }
 907         ret = 0;
 908 error:
 909         if (pending_del_nr) {
 910                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
 911                                       pending_del_nr);
 912         }
 913         btrfs_release_path(root, path);
 914         btrfs_free_path(path);
 915         inode->i_sb->s_dirt = 1;
 916         return ret;
 917 }
 918
 919 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 920                               size_t zero_start)
 921 {
 922         char *kaddr;
 923         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 924         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 925         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 926         int ret = 0;
 927
 928         WARN_ON(!PageLocked(page));
 929         set_page_extent_mapped(page);
 930
 931         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 932         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
 933                             page_end, GFP_NOFS);
 934
 935         if (zero_start != PAGE_CACHE_SIZE) {
 936                 kaddr = kmap(page);
 937                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 938                 flush_dcache_page(page);
 939                 kunmap(page);
 940         }
 941         set_page_dirty(page);
 942         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 943
 944         return ret;
 945 }
 946
 947 /*
 948  * taken from block_truncate_page, but does cow as it zeros out
 949  * any bytes left in the last page in the file.
 950  */
 951 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 952 {
 953         struct inode *inode = mapping->host;
 954         struct btrfs_root *root = BTRFS_I(inode)->root;
 955         u32 blocksize = root->sectorsize;
 956         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 957         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 958         struct page *page;
 959         int ret = 0;
 960         u64 page_start;
 961
 962         if ((offset & (blocksize - 1)) == 0)
 963                 goto out;
 964
 965         ret = -ENOMEM;
 966         page = grab_cache_page(mapping, index);
 967         if (!page)
 968                 goto out;
 969         if (!PageUptodate(page)) {
 970                 ret = btrfs_readpage(NULL, page);
 971                 lock_page(page);
 972                 if (!PageUptodate(page)) {
 973                         ret = -EIO;
 974                         goto out;
 975                 }
 976         }
 977         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 978
 979         ret = btrfs_cow_one_page(inode, page, offset);
 980
 981         unlock_page(page);
 982         page_cache_release(page);
 983 out:
 984         return ret;
 985 }
 986
 987 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 988 {
 989         struct inode *inode = dentry->d_inode;
 990         int err;
 991
 992         err = inode_change_ok(inode, attr);
 993         if (err)
 994                 return err;
 995
 996         if (S_ISREG(inode->i_mode) &&
 997             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 998                 struct btrfs_trans_handle *trans;
 999                 struct btrfs_root *root = BTRFS_I(inode)->root;
1000                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1001
1002                 u64 mask = root->sectorsize - 1;
1003                 u64 hole_start = (inode->i_size + mask) & ~mask;
1004                 u64 block_end = (attr->ia_size + mask) & ~mask;
1005                 u64 hole_size;
1006                 u64 alloc_hint = 0;
1007
1008                 if (attr->ia_size <= hole_start)
1009                         goto out;
1010
1011                 mutex_lock(&root->fs_info->fs_mutex);
1012                 err = btrfs_check_free_space(root, 1, 0);
1013                 mutex_unlock(&root->fs_info->fs_mutex);
1014                 if (err)
1015                         goto fail;
1016
1017                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1018
1019                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1020                 hole_size = block_end - hole_start;
1021
1022                 mutex_lock(&root->fs_info->fs_mutex);
1023                 trans = btrfs_start_transaction(root, 1);
1024                 btrfs_set_trans_block_group(trans, inode);
1025                 err = btrfs_drop_extents(trans, root, inode,
1026                                          hole_start, block_end, hole_start,
1027                                          &alloc_hint);
1028
1029                 if (alloc_hint != EXTENT_MAP_INLINE) {
1030                         err = btrfs_insert_file_extent(trans, root,
1031                                                        inode->i_ino,
1032                                                        hole_start, 0, 0,
1033                                                        hole_size);
1034                         btrfs_drop_extent_cache(inode, hole_start,
1035                                                 hole_size - 1);
1036                         btrfs_check_file(root, inode);
1037                 }
1038                 btrfs_end_transaction(trans, root);
1039                 mutex_unlock(&root->fs_info->fs_mutex);
1040                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1041                 if (err)
1042                         return err;
1043         }
1044 out:
1045         err = inode_setattr(inode, attr);
1046 fail:
1047         return err;
1048 }
1049
1050 void btrfs_put_inode(struct inode *inode)
1051 {
1052         int ret;
1053
1054         if (!BTRFS_I(inode)->ordered_trans) {
1055                 return;
1056         }
1057
1058         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1059             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1060                 return;
1061
1062         ret = btrfs_del_ordered_inode(inode);
1063         if (ret == 1) {
1064                 atomic_dec(&inode->i_count);
1065         }
1066 }
1067
1068 void btrfs_delete_inode(struct inode *inode)
1069 {
1070         struct btrfs_trans_handle *trans;
1071         struct btrfs_root *root = BTRFS_I(inode)->root;
1072         unsigned long nr;
1073         int ret;
1074
1075         truncate_inode_pages(&inode->i_data, 0);
1076         if (is_bad_inode(inode)) {
1077                 goto no_delete;
1078         }
1079
1080         inode->i_size = 0;
1081         mutex_lock(&root->fs_info->fs_mutex);
1082         trans = btrfs_start_transaction(root, 1);
1083
1084         btrfs_set_trans_block_group(trans, inode);
1085         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1086         if (ret)
1087                 goto no_delete_lock;
1088
1089         nr = trans->blocks_used;
1090         clear_inode(inode);
1091
1092         btrfs_end_transaction(trans, root);
1093         mutex_unlock(&root->fs_info->fs_mutex);
1094         btrfs_btree_balance_dirty(root, nr);
1095         btrfs_throttle(root);
1096         return;
1097
1098 no_delete_lock:
1099         nr = trans->blocks_used;
1100         btrfs_end_transaction(trans, root);
1101         mutex_unlock(&root->fs_info->fs_mutex);
1102         btrfs_btree_balance_dirty(root, nr);
1103         btrfs_throttle(root);
1104 no_delete:
1105         clear_inode(inode);
1106 }
1107
1108 /*
1109  * this returns the key found in the dir entry in the location pointer.
1110  * If no dir entries were found, location->objectid is 0.
1111  */
1112 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1113                                struct btrfs_key *location)
1114 {
1115         const char *name = dentry->d_name.name;
1116         int namelen = dentry->d_name.len;
1117         struct btrfs_dir_item *di;
1118         struct btrfs_path *path;
1119         struct btrfs_root *root = BTRFS_I(dir)->root;
1120         int ret = 0;
1121
1122         if (namelen == 1 && strcmp(name, ".") == 0) {
1123                 location->objectid = dir->i_ino;
1124                 location->type = BTRFS_INODE_ITEM_KEY;
1125                 location->offset = 0;
1126                 return 0;
1127         }
1128         path = btrfs_alloc_path();
1129         BUG_ON(!path);
1130
1131         if (namelen == 2 && strcmp(name, "..") == 0) {
1132                 struct btrfs_key key;
1133                 struct extent_buffer *leaf;
1134                 u32 nritems;
1135                 int slot;
1136
1137                 key.objectid = dir->i_ino;
1138                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1139                 key.offset = 0;
1140                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1141                 BUG_ON(ret == 0);
1142                 ret = 0;
1143
1144                 leaf = path->nodes[0];
1145                 slot = path->slots[0];
1146                 nritems = btrfs_header_nritems(leaf);
1147                 if (slot >= nritems)
1148                         goto out_err;
1149
1150                 btrfs_item_key_to_cpu(leaf, &key, slot);
1151                 if (key.objectid != dir->i_ino ||
1152                     key.type != BTRFS_INODE_REF_KEY) {
1153                         goto out_err;
1154                 }
1155                 location->objectid = key.offset;
1156                 location->type = BTRFS_INODE_ITEM_KEY;
1157                 location->offset = 0;
1158                 goto out;
1159         }
1160
1161         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1162                                     namelen, 0);
1163         if (IS_ERR(di))
1164                 ret = PTR_ERR(di);
1165         if (!di || IS_ERR(di)) {
1166                 goto out_err;
1167         }
1168         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1169 out:
1170         btrfs_free_path(path);
1171         return ret;
1172 out_err:
1173         location->objectid = 0;
1174         goto out;
1175 }
1176
1177 /*
1178  * when we hit a tree root in a directory, the btrfs part of the inode
1179  * needs to be changed to reflect the root directory of the tree root.  This
1180  * is kind of like crossing a mount point.
1181  */
1182 static int fixup_tree_root_location(struct btrfs_root *root,
1183                              struct btrfs_key *location,
1184                              struct btrfs_root **sub_root,
1185                              struct dentry *dentry)
1186 {
1187         struct btrfs_path *path;
1188         struct btrfs_root_item *ri;
1189
1190         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1191                 return 0;
1192         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1193                 return 0;
1194
1195         path = btrfs_alloc_path();
1196         BUG_ON(!path);
1197         mutex_lock(&root->fs_info->fs_mutex);
1198
1199         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1200                                         dentry->d_name.name,
1201                                         dentry->d_name.len);
1202         if (IS_ERR(*sub_root))
1203                 return PTR_ERR(*sub_root);
1204
1205         ri = &(*sub_root)->root_item;
1206         location->objectid = btrfs_root_dirid(ri);
1207         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1208         location->offset = 0;
1209
1210         btrfs_free_path(path);
1211         mutex_unlock(&root->fs_info->fs_mutex);
1212         return 0;
1213 }
1214
1215 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1216 {
1217         struct btrfs_iget_args *args = p;
1218         inode->i_ino = args->ino;
1219         BTRFS_I(inode)->root = args->root;
1220         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1221         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1222                              inode->i_mapping, GFP_NOFS);
1223         return 0;
1224 }
1225
1226 static int btrfs_find_actor(struct inode *inode, void *opaque)
1227 {
1228         struct btrfs_iget_args *args = opaque;
1229         return (args->ino == inode->i_ino &&
1230                 args->root == BTRFS_I(inode)->root);
1231 }
1232
1233 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1234                             u64 root_objectid)
1235 {
1236         struct btrfs_iget_args args;
1237         args.ino = objectid;
1238         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1239
1240         if (!args.root)
1241                 return NULL;
1242
1243         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1244 }
1245
1246 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1247                                 struct btrfs_root *root)
1248 {
1249         struct inode *inode;
1250         struct btrfs_iget_args args;
1251         args.ino = objectid;
1252         args.root = root;
1253
1254         inode = iget5_locked(s, objectid, btrfs_find_actor,
1255                              btrfs_init_locked_inode,
1256                              (void *)&args);
1257         return inode;
1258 }
1259
1260 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1261                                    struct nameidata *nd)
1262 {
1263         struct inode * inode;
1264         struct btrfs_inode *bi = BTRFS_I(dir);
1265         struct btrfs_root *root = bi->root;
1266         struct btrfs_root *sub_root = root;
1267         struct btrfs_key location;
1268         int ret;
1269
1270         if (dentry->d_name.len > BTRFS_NAME_LEN)
1271                 return ERR_PTR(-ENAMETOOLONG);
1272
1273         mutex_lock(&root->fs_info->fs_mutex);
1274         ret = btrfs_inode_by_name(dir, dentry, &location);
1275         mutex_unlock(&root->fs_info->fs_mutex);
1276
1277         if (ret < 0)
1278                 return ERR_PTR(ret);
1279
1280         inode = NULL;
1281         if (location.objectid) {
1282                 ret = fixup_tree_root_location(root, &location, &sub_root,
1283                                                 dentry);
1284                 if (ret < 0)
1285                         return ERR_PTR(ret);
1286                 if (ret > 0)
1287                         return ERR_PTR(-ENOENT);
1288                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1289                                           sub_root);
1290                 if (!inode)
1291                         return ERR_PTR(-EACCES);
1292                 if (inode->i_state & I_NEW) {
1293                         /* the inode and parent dir are two different roots */
1294                         if (sub_root != root) {
1295                                 igrab(inode);
1296                                 sub_root->inode = inode;
1297                         }
1298                         BTRFS_I(inode)->root = sub_root;
1299                         memcpy(&BTRFS_I(inode)->location, &location,
1300                                sizeof(location));
1301                         btrfs_read_locked_inode(inode);
1302                         unlock_new_inode(inode);
1303                 }
1304         }
1305         return d_splice_alias(inode, dentry);
1306 }
1307
1308 static unsigned char btrfs_filetype_table[] = {
1309         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1310 };
1311
1312 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1313 {
1314         struct inode *inode = filp->f_dentry->d_inode;
1315         struct btrfs_root *root = BTRFS_I(inode)->root;
1316         struct btrfs_item *item;
1317         struct btrfs_dir_item *di;
1318         struct btrfs_key key;
1319         struct btrfs_key found_key;
1320         struct btrfs_path *path;
1321         int ret;
1322         u32 nritems;
1323         struct extent_buffer *leaf;
1324         int slot;
1325         int advance;
1326         unsigned char d_type;
1327         int over = 0;
1328         u32 di_cur;
1329         u32 di_total;
1330         u32 di_len;
1331         int key_type = BTRFS_DIR_INDEX_KEY;
1332         char tmp_name[32];
1333         char *name_ptr;
1334         int name_len;
1335
1336         /* FIXME, use a real flag for deciding about the key type */
1337         if (root->fs_info->tree_root == root)
1338                 key_type = BTRFS_DIR_ITEM_KEY;
1339
1340         /* special case for "." */
1341         if (filp->f_pos == 0) {
1342                 over = filldir(dirent, ".", 1,
1343                                1, inode->i_ino,
1344                                DT_DIR);
1345                 if (over)
1346                         return 0;
1347                 filp->f_pos = 1;
1348         }
1349
1350         mutex_lock(&root->fs_info->fs_mutex);
1351         key.objectid = inode->i_ino;
1352         path = btrfs_alloc_path();
1353         path->reada = 2;
1354
1355         /* special case for .., just use the back ref */
1356         if (filp->f_pos == 1) {
1357                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1358                 key.offset = 0;
1359                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1360                 BUG_ON(ret == 0);
1361                 leaf = path->nodes[0];
1362                 slot = path->slots[0];
1363                 nritems = btrfs_header_nritems(leaf);
1364                 if (slot >= nritems) {
1365                         btrfs_release_path(root, path);
1366                         goto read_dir_items;
1367                 }
1368                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1369                 btrfs_release_path(root, path);
1370                 if (found_key.objectid != key.objectid ||
1371                     found_key.type != BTRFS_INODE_REF_KEY)
1372                         goto read_dir_items;
1373                 over = filldir(dirent, "..", 2,
1374                                2, found_key.offset, DT_DIR);
1375                 if (over)
1376                         goto nopos;
1377                 filp->f_pos = 2;
1378         }
1379
1380 read_dir_items:
1381         btrfs_set_key_type(&key, key_type);
1382         key.offset = filp->f_pos;
1383
1384         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1385         if (ret < 0)
1386                 goto err;
1387         advance = 0;
1388         while(1) {
1389                 leaf = path->nodes[0];
1390                 nritems = btrfs_header_nritems(leaf);
1391                 slot = path->slots[0];
1392                 if (advance || slot >= nritems) {
1393                         if (slot >= nritems -1) {
1394                                 ret = btrfs_next_leaf(root, path);
1395                                 if (ret)
1396                                         break;
1397                                 leaf = path->nodes[0];
1398                                 nritems = btrfs_header_nritems(leaf);
1399                                 slot = path->slots[0];
1400                         } else {
1401                                 slot++;
1402                                 path->slots[0]++;
1403                         }
1404                 }
1405                 advance = 1;
1406                 item = btrfs_item_nr(leaf, slot);
1407                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1408
1409                 if (found_key.objectid != key.objectid)
1410                         break;
1411                 if (btrfs_key_type(&found_key) != key_type)
1412                         break;
1413                 if (found_key.offset < filp->f_pos)
1414                         continue;
1415
1416                 filp->f_pos = found_key.offset;
1417                 advance = 1;
1418                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1419                 di_cur = 0;
1420                 di_total = btrfs_item_size(leaf, item);
1421                 while(di_cur < di_total) {
1422                         struct btrfs_key location;
1423
1424                         name_len = btrfs_dir_name_len(leaf, di);
1425                         if (name_len < 32) {
1426                                 name_ptr = tmp_name;
1427                         } else {
1428                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1429                                 BUG_ON(!name_ptr);
1430                         }
1431                         read_extent_buffer(leaf, name_ptr,
1432                                            (unsigned long)(di + 1), name_len);
1433
1434                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1435                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1436                         over = filldir(dirent, name_ptr, name_len,
1437                                        found_key.offset,
1438                                        location.objectid,
1439                                        d_type);
1440
1441                         if (name_ptr != tmp_name)
1442                                 kfree(name_ptr);
1443
1444                         if (over)
1445                                 goto nopos;
1446                         di_len = btrfs_dir_name_len(leaf, di) +
1447                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1448                         di_cur += di_len;
1449                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1450                 }
1451         }
1452         filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1453 nopos:
1454         ret = 0;
1455 err:
1456         btrfs_release_path(root, path);
1457         btrfs_free_path(path);
1458         mutex_unlock(&root->fs_info->fs_mutex);
1459         return ret;
1460 }
1461
1462 int btrfs_write_inode(struct inode *inode, int wait)
1463 {
1464         struct btrfs_root *root = BTRFS_I(inode)->root;
1465         struct btrfs_trans_handle *trans;
1466         int ret = 0;
1467
1468         if (wait) {
1469                 mutex_lock(&root->fs_info->fs_mutex);
1470                 trans = btrfs_start_transaction(root, 1);
1471                 btrfs_set_trans_block_group(trans, inode);
1472                 ret = btrfs_commit_transaction(trans, root);
1473                 mutex_unlock(&root->fs_info->fs_mutex);
1474         }
1475         return ret;
1476 }
1477
1478 /*
1479  * This is somewhat expensive, updating the tree every time the
1480  * inode changes.  But, it is most likely to find the inode in cache.
1481  * FIXME, needs more benchmarking...there are no reasons other than performance
1482  * to keep or drop this code.
1483  */
1484 void btrfs_dirty_inode(struct inode *inode)
1485 {
1486         struct btrfs_root *root = BTRFS_I(inode)->root;
1487         struct btrfs_trans_handle *trans;
1488
1489         mutex_lock(&root->fs_info->fs_mutex);
1490         trans = btrfs_start_transaction(root, 1);
1491         btrfs_set_trans_block_group(trans, inode);
1492         btrfs_update_inode(trans, root, inode);
1493         btrfs_end_transaction(trans, root);
1494         mutex_unlock(&root->fs_info->fs_mutex);
1495 }
1496
1497 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1498                                      struct btrfs_root *root,
1499                                      const char *name, int name_len,
1500                                      u64 ref_objectid,
1501                                      u64 objectid,
1502                                      struct btrfs_block_group_cache *group,
1503                                      int mode)
1504 {
1505         struct inode *inode;
1506         struct btrfs_inode_item *inode_item;
1507         struct btrfs_key *location;
1508         struct btrfs_path *path;
1509         struct btrfs_inode_ref *ref;
1510         struct btrfs_key key[2];
1511         u32 sizes[2];
1512         unsigned long ptr;
1513         int ret;
1514         int owner;
1515
1516         path = btrfs_alloc_path();
1517         BUG_ON(!path);
1518
1519         inode = new_inode(root->fs_info->sb);
1520         if (!inode)
1521                 return ERR_PTR(-ENOMEM);
1522
1523         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1524         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1525                              inode->i_mapping, GFP_NOFS);
1526         BTRFS_I(inode)->root = root;
1527
1528         if (mode & S_IFDIR)
1529                 owner = 0;
1530         else
1531                 owner = 1;
1532         group = btrfs_find_block_group(root, group, 0, 0, owner);
1533         BTRFS_I(inode)->block_group = group;
1534         BTRFS_I(inode)->flags = 0;
1535
1536         key[0].objectid = objectid;
1537         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1538         key[0].offset = 0;
1539
1540         key[1].objectid = objectid;
1541         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1542         key[1].offset = ref_objectid;
1543
1544         sizes[0] = sizeof(struct btrfs_inode_item);
1545         sizes[1] = name_len + sizeof(*ref);
1546
1547         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1548         if (ret != 0)
1549                 goto fail;
1550
1551         if (objectid > root->highest_inode)
1552                 root->highest_inode = objectid;
1553
1554         inode->i_uid = current->fsuid;
1555         inode->i_gid = current->fsgid;
1556         inode->i_mode = mode;
1557         inode->i_ino = objectid;
1558         inode->i_blocks = 0;
1559         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1560         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1561                                   struct btrfs_inode_item);
1562         fill_inode_item(path->nodes[0], inode_item, inode);
1563
1564         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1565                              struct btrfs_inode_ref);
1566         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1567         ptr = (unsigned long)(ref + 1);
1568         write_extent_buffer(path->nodes[0], name, ptr, name_len);
1569
1570         btrfs_mark_buffer_dirty(path->nodes[0]);
1571         btrfs_free_path(path);
1572
1573         location = &BTRFS_I(inode)->location;
1574         location->objectid = objectid;
1575         location->offset = 0;
1576         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1577
1578         insert_inode_hash(inode);
1579         return inode;
1580 fail:
1581         btrfs_free_path(path);
1582         return ERR_PTR(ret);
1583 }
1584
1585 static inline u8 btrfs_inode_type(struct inode *inode)
1586 {
1587         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1588 }
1589
1590 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1591                             struct dentry *dentry, struct inode *inode,
1592                             int add_backref)
1593 {
1594         int ret;
1595         struct btrfs_key key;
1596         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1597         struct inode *parent_inode;
1598
1599         key.objectid = inode->i_ino;
1600         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1601         key.offset = 0;
1602
1603         ret = btrfs_insert_dir_item(trans, root,
1604                                     dentry->d_name.name, dentry->d_name.len,
1605                                     dentry->d_parent->d_inode->i_ino,
1606                                     &key, btrfs_inode_type(inode));
1607         if (ret == 0) {
1608                 if (add_backref) {
1609                         ret = btrfs_insert_inode_ref(trans, root,
1610                                              dentry->d_name.name,
1611                                              dentry->d_name.len,
1612                                              inode->i_ino,
1613                                              dentry->d_parent->d_inode->i_ino);
1614                 }
1615                 parent_inode = dentry->d_parent->d_inode;
1616                 parent_inode->i_size += dentry->d_name.len * 2;
1617                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1618                 ret = btrfs_update_inode(trans, root,
1619                                          dentry->d_parent->d_inode);
1620         }
1621         return ret;
1622 }
1623
1624 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1625                             struct dentry *dentry, struct inode *inode,
1626                             int backref)
1627 {
1628         int err = btrfs_add_link(trans, dentry, inode, backref);
1629         if (!err) {
1630                 d_instantiate(dentry, inode);
1631                 return 0;
1632         }
1633         if (err > 0)
1634                 err = -EEXIST;
1635         return err;
1636 }
1637
1638 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1639                         int mode, dev_t rdev)
1640 {
1641         struct btrfs_trans_handle *trans;
1642         struct btrfs_root *root = BTRFS_I(dir)->root;
1643         struct inode *inode = NULL;
1644         int err;
1645         int drop_inode = 0;
1646         u64 objectid;
1647         unsigned long nr = 0;
1648
1649         if (!new_valid_dev(rdev))
1650                 return -EINVAL;
1651
1652         mutex_lock(&root->fs_info->fs_mutex);
1653         err = btrfs_check_free_space(root, 1, 0);
1654         if (err)
1655                 goto fail;
1656
1657         trans = btrfs_start_transaction(root, 1);
1658         btrfs_set_trans_block_group(trans, dir);
1659
1660         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1661         if (err) {
1662                 err = -ENOSPC;
1663                 goto out_unlock;
1664         }
1665
1666         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1667                                 dentry->d_name.len,
1668                                 dentry->d_parent->d_inode->i_ino, objectid,
1669                                 BTRFS_I(dir)->block_group, mode);
1670         err = PTR_ERR(inode);
1671         if (IS_ERR(inode))
1672                 goto out_unlock;
1673
1674         btrfs_set_trans_block_group(trans, inode);
1675         err = btrfs_add_nondir(trans, dentry, inode, 0);
1676         if (err)
1677                 drop_inode = 1;
1678         else {
1679                 inode->i_op = &btrfs_special_inode_operations;
1680                 init_special_inode(inode, inode->i_mode, rdev);
1681                 btrfs_update_inode(trans, root, inode);
1682         }
1683         dir->i_sb->s_dirt = 1;
1684         btrfs_update_inode_block_group(trans, inode);
1685         btrfs_update_inode_block_group(trans, dir);
1686 out_unlock:
1687         nr = trans->blocks_used;
1688         btrfs_end_transaction(trans, root);
1689 fail:
1690         mutex_unlock(&root->fs_info->fs_mutex);
1691
1692         if (drop_inode) {
1693                 inode_dec_link_count(inode);
1694                 iput(inode);
1695         }
1696         btrfs_btree_balance_dirty(root, nr);
1697         btrfs_throttle(root);
1698         return err;
1699 }
1700
1701 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1702                         int mode, struct nameidata *nd)
1703 {
1704         struct btrfs_trans_handle *trans;
1705         struct btrfs_root *root = BTRFS_I(dir)->root;
1706         struct inode *inode = NULL;
1707         int err;
1708         int drop_inode = 0;
1709         unsigned long nr = 0;
1710         u64 objectid;
1711
1712         mutex_lock(&root->fs_info->fs_mutex);
1713         err = btrfs_check_free_space(root, 1, 0);
1714         if (err)
1715                 goto fail;
1716         trans = btrfs_start_transaction(root, 1);
1717         btrfs_set_trans_block_group(trans, dir);
1718
1719         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1720         if (err) {
1721                 err = -ENOSPC;
1722                 goto out_unlock;
1723         }
1724
1725         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1726                                 dentry->d_name.len,
1727                                 dentry->d_parent->d_inode->i_ino,
1728                                 objectid, BTRFS_I(dir)->block_group, mode);
1729         err = PTR_ERR(inode);
1730         if (IS_ERR(inode))
1731                 goto out_unlock;
1732
1733         btrfs_set_trans_block_group(trans, inode);
1734         err = btrfs_add_nondir(trans, dentry, inode, 0);
1735         if (err)
1736                 drop_inode = 1;
1737         else {
1738                 inode->i_mapping->a_ops = &btrfs_aops;
1739                 inode->i_fop = &btrfs_file_operations;
1740                 inode->i_op = &btrfs_file_inode_operations;
1741                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1742                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1743                                      inode->i_mapping, GFP_NOFS);
1744                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1745         }
1746         dir->i_sb->s_dirt = 1;
1747         btrfs_update_inode_block_group(trans, inode);
1748         btrfs_update_inode_block_group(trans, dir);
1749 out_unlock:
1750         nr = trans->blocks_used;
1751         btrfs_end_transaction(trans, root);
1752 fail:
1753         mutex_unlock(&root->fs_info->fs_mutex);
1754
1755         if (drop_inode) {
1756                 inode_dec_link_count(inode);
1757                 iput(inode);
1758         }
1759         btrfs_btree_balance_dirty(root, nr);
1760         btrfs_throttle(root);
1761         return err;
1762 }
1763
1764 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1765                       struct dentry *dentry)
1766 {
1767         struct btrfs_trans_handle *trans;
1768         struct btrfs_root *root = BTRFS_I(dir)->root;
1769         struct inode *inode = old_dentry->d_inode;
1770         unsigned long nr = 0;
1771         int err;
1772         int drop_inode = 0;
1773
1774         if (inode->i_nlink == 0)
1775                 return -ENOENT;
1776
1777 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1778         inode->i_nlink++;
1779 #else
1780         inc_nlink(inode);
1781 #endif
1782         mutex_lock(&root->fs_info->fs_mutex);
1783         err = btrfs_check_free_space(root, 1, 0);
1784         if (err)
1785                 goto fail;
1786         trans = btrfs_start_transaction(root, 1);
1787
1788         btrfs_set_trans_block_group(trans, dir);
1789         atomic_inc(&inode->i_count);
1790         err = btrfs_add_nondir(trans, dentry, inode, 1);
1791
1792         if (err)
1793                 drop_inode = 1;
1794
1795         dir->i_sb->s_dirt = 1;
1796         btrfs_update_inode_block_group(trans, dir);
1797         err = btrfs_update_inode(trans, root, inode);
1798
1799         if (err)
1800                 drop_inode = 1;
1801
1802         nr = trans->blocks_used;
1803         btrfs_end_transaction(trans, root);
1804 fail:
1805         mutex_unlock(&root->fs_info->fs_mutex);
1806
1807         if (drop_inode) {
1808                 inode_dec_link_count(inode);
1809                 iput(inode);
1810         }
1811         btrfs_btree_balance_dirty(root, nr);
1812         btrfs_throttle(root);
1813         return err;
1814 }
1815
1816 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1817 {
1818         struct inode *inode;
1819         struct btrfs_trans_handle *trans;
1820         struct btrfs_root *root = BTRFS_I(dir)->root;
1821         int err = 0;
1822         int drop_on_err = 0;
1823         u64 objectid;
1824         unsigned long nr = 1;
1825
1826         mutex_lock(&root->fs_info->fs_mutex);
1827         err = btrfs_check_free_space(root, 1, 0);
1828         if (err)
1829                 goto out_unlock;
1830
1831         trans = btrfs_start_transaction(root, 1);
1832         btrfs_set_trans_block_group(trans, dir);
1833
1834         if (IS_ERR(trans)) {
1835                 err = PTR_ERR(trans);
1836                 goto out_unlock;
1837         }
1838
1839         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1840         if (err) {
1841                 err = -ENOSPC;
1842                 goto out_unlock;
1843         }
1844
1845         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1846                                 dentry->d_name.len,
1847                                 dentry->d_parent->d_inode->i_ino, objectid,
1848                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1849         if (IS_ERR(inode)) {
1850                 err = PTR_ERR(inode);
1851                 goto out_fail;
1852         }
1853
1854         drop_on_err = 1;
1855         inode->i_op = &btrfs_dir_inode_operations;
1856         inode->i_fop = &btrfs_dir_file_operations;
1857         btrfs_set_trans_block_group(trans, inode);
1858
1859         inode->i_size = 0;
1860         err = btrfs_update_inode(trans, root, inode);
1861         if (err)
1862                 goto out_fail;
1863
1864         err = btrfs_add_link(trans, dentry, inode, 0);
1865         if (err)
1866                 goto out_fail;
1867
1868         d_instantiate(dentry, inode);
1869         drop_on_err = 0;
1870         dir->i_sb->s_dirt = 1;
1871         btrfs_update_inode_block_group(trans, inode);
1872         btrfs_update_inode_block_group(trans, dir);
1873
1874 out_fail:
1875         nr = trans->blocks_used;
1876         btrfs_end_transaction(trans, root);
1877
1878 out_unlock:
1879         mutex_unlock(&root->fs_info->fs_mutex);
1880         if (drop_on_err)
1881                 iput(inode);
1882         btrfs_btree_balance_dirty(root, nr);
1883         btrfs_throttle(root);
1884         return err;
1885 }
1886
1887 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1888                                     size_t pg_offset, u64 start, u64 len,
1889                                     int create)
1890 {
1891         int ret;
1892         int err = 0;
1893         u64 bytenr;
1894         u64 extent_start = 0;
1895         u64 extent_end = 0;
1896         u64 objectid = inode->i_ino;
1897         u32 found_type;
1898         struct btrfs_path *path;
1899         struct btrfs_root *root = BTRFS_I(inode)->root;
1900         struct btrfs_file_extent_item *item;
1901         struct extent_buffer *leaf;
1902         struct btrfs_key found_key;
1903         struct extent_map *em = NULL;
1904         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1905         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1906         struct btrfs_trans_handle *trans = NULL;
1907
1908         path = btrfs_alloc_path();
1909         BUG_ON(!path);
1910         mutex_lock(&root->fs_info->fs_mutex);
1911
1912 again:
1913         spin_lock(&em_tree->lock);
1914         em = lookup_extent_mapping(em_tree, start, len);
1915         spin_unlock(&em_tree->lock);
1916
1917         if (em) {
1918                 if (em->start > start) {
1919                         printk("get_extent lookup [%Lu %Lu] em [%Lu %Lu]\n",
1920                                start, len, em->start, em->len);
1921                         WARN_ON(1);
1922                 }
1923                 if (em->block_start == EXTENT_MAP_INLINE && page)
1924                         free_extent_map(em);
1925                 else
1926                         goto out;
1927         }
1928         em = alloc_extent_map(GFP_NOFS);
1929         if (!em) {
1930                 err = -ENOMEM;
1931                 goto out;
1932         }
1933
1934         em->start = EXTENT_MAP_HOLE;
1935         em->len = (u64)-1;
1936         em->bdev = inode->i_sb->s_bdev;
1937         ret = btrfs_lookup_file_extent(trans, root, path,
1938                                        objectid, start, trans != NULL);
1939         if (ret < 0) {
1940                 err = ret;
1941                 goto out;
1942         }
1943
1944         if (ret != 0) {
1945                 if (path->slots[0] == 0)
1946                         goto not_found;
1947                 path->slots[0]--;
1948         }
1949
1950         leaf = path->nodes[0];
1951         item = btrfs_item_ptr(leaf, path->slots[0],
1952                               struct btrfs_file_extent_item);
1953         /* are we inside the extent that was found? */
1954         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1955         found_type = btrfs_key_type(&found_key);
1956         if (found_key.objectid != objectid ||
1957             found_type != BTRFS_EXTENT_DATA_KEY) {
1958                 goto not_found;
1959         }
1960
1961         found_type = btrfs_file_extent_type(leaf, item);
1962         extent_start = found_key.offset;
1963         if (found_type == BTRFS_FILE_EXTENT_REG) {
1964                 extent_end = extent_start +
1965                        btrfs_file_extent_num_bytes(leaf, item);
1966                 err = 0;
1967                 if (start < extent_start || start >= extent_end) {
1968                         em->start = start;
1969                         if (start < extent_start) {
1970                                 if (start + len <= extent_start)
1971                                         goto not_found;
1972                                 em->len = extent_end - extent_start;
1973                         } else {
1974                                 em->len = len;
1975                         }
1976                         goto not_found_em;
1977                 }
1978                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1979                 if (bytenr == 0) {
1980                         em->start = extent_start;
1981                         em->len = extent_end - extent_start;
1982                         em->block_start = EXTENT_MAP_HOLE;
1983                         goto insert;
1984                 }
1985                 bytenr += btrfs_file_extent_offset(leaf, item);
1986                 em->block_start = bytenr;
1987                 em->start = extent_start;
1988                 em->len = extent_end - extent_start;
1989                 goto insert;
1990         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1991                 u64 page_start;
1992                 unsigned long ptr;
1993                 char *map;
1994                 size_t size;
1995                 size_t extent_offset;
1996                 size_t copy_size;
1997
1998                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1999                                                     path->slots[0]));
2000                 extent_end = (extent_start + size + root->sectorsize - 1) &
2001                         ~((u64)root->sectorsize - 1);
2002                 if (start < extent_start || start >= extent_end) {
2003                         em->start = start;
2004                         if (start < extent_start) {
2005                                 if (start + len <= extent_start)
2006                                         goto not_found;
2007                                 em->len = extent_end - extent_start;
2008                         } else {
2009                                 em->len = len;
2010                         }
2011                         goto not_found_em;
2012                 }
2013                 em->block_start = EXTENT_MAP_INLINE;
2014
2015                 if (!page) {
2016                         em->start = extent_start;
2017                         em->len = size;
2018                         goto out;
2019                 }
2020
2021                 page_start = page_offset(page) + pg_offset;
2022                 extent_offset = page_start - extent_start;
2023                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2024                                 size - extent_offset);
2025                 em->start = extent_start + extent_offset;
2026                 em->len = (copy_size + root->sectorsize - 1) &
2027                         ~((u64)root->sectorsize - 1);
2028                 map = kmap(page);
2029                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2030                 if (create == 0 && !PageUptodate(page)) {
2031                         read_extent_buffer(leaf, map + pg_offset, ptr,
2032                                            copy_size);
2033                         flush_dcache_page(page);
2034                 } else if (create && PageUptodate(page)) {
2035                         if (!trans) {
2036                                 kunmap(page);
2037                                 free_extent_map(em);
2038                                 em = NULL;
2039                                 btrfs_release_path(root, path);
2040                                 trans = btrfs_start_transaction(root, 1);
2041                                 goto again;
2042                         }
2043                         write_extent_buffer(leaf, map + pg_offset, ptr,
2044                                             copy_size);
2045                         btrfs_mark_buffer_dirty(leaf);
2046                 }
2047                 kunmap(page);
2048                 set_extent_uptodate(io_tree, em->start,
2049                                     extent_map_end(em) - 1, GFP_NOFS);
2050                 goto insert;
2051         } else {
2052                 printk("unkknown found_type %d\n", found_type);
2053                 WARN_ON(1);
2054         }
2055 not_found:
2056         em->start = start;
2057         em->len = len;
2058 not_found_em:
2059         em->block_start = EXTENT_MAP_HOLE;
2060 insert:
2061         btrfs_release_path(root, path);
2062         if (em->start > start || extent_map_end(em) <= start) {
2063                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2064                 err = -EIO;
2065                 goto out;
2066         }
2067
2068         err = 0;
2069         spin_lock(&em_tree->lock);
2070         ret = add_extent_mapping(em_tree, em);
2071         if (ret == -EEXIST) {
2072                 free_extent_map(em);
2073                 em = lookup_extent_mapping(em_tree, start, len);
2074                 if (!em) {
2075                         err = -EIO;
2076                         printk("failing to insert %Lu %Lu\n", start, len);
2077                 }
2078         }
2079         spin_unlock(&em_tree->lock);
2080 out:
2081         btrfs_free_path(path);
2082         if (trans) {
2083                 ret = btrfs_end_transaction(trans, root);
2084                 if (!err)
2085                         err = ret;
2086         }
2087         mutex_unlock(&root->fs_info->fs_mutex);
2088         if (err) {
2089                 free_extent_map(em);
2090                 WARN_ON(1);
2091                 return ERR_PTR(err);
2092         }
2093         return em;
2094 }
2095
2096 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2097 {
2098         return extent_bmap(mapping, iblock, btrfs_get_extent);
2099 }
2100
2101 int btrfs_readpage(struct file *file, struct page *page)
2102 {
2103         struct extent_io_tree *tree;
2104         tree = &BTRFS_I(page->mapping->host)->io_tree;
2105         return extent_read_full_page(tree, page, btrfs_get_extent);
2106 }
2107
2108 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2109 {
2110         struct extent_io_tree *tree;
2111
2112
2113         if (current->flags & PF_MEMALLOC) {
2114                 redirty_page_for_writepage(wbc, page);
2115                 unlock_page(page);
2116                 return 0;
2117         }
2118         tree = &BTRFS_I(page->mapping->host)->io_tree;
2119         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2120 }
2121
2122 static int btrfs_writepages(struct address_space *mapping,
2123                             struct writeback_control *wbc)
2124 {
2125         struct extent_io_tree *tree;
2126         tree = &BTRFS_I(mapping->host)->io_tree;
2127         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2128 }
2129
2130 static int
2131 btrfs_readpages(struct file *file, struct address_space *mapping,
2132                 struct list_head *pages, unsigned nr_pages)
2133 {
2134         struct extent_io_tree *tree;
2135         tree = &BTRFS_I(mapping->host)->io_tree;
2136         return extent_readpages(tree, mapping, pages, nr_pages,
2137                                 btrfs_get_extent);
2138 }
2139
2140 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2141 {
2142         struct extent_io_tree *tree;
2143         struct extent_map_tree *map;
2144         int ret;
2145
2146         tree = &BTRFS_I(page->mapping->host)->io_tree;
2147         map = &BTRFS_I(page->mapping->host)->extent_tree;
2148         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2149         if (ret == 1) {
2150                 ClearPagePrivate(page);
2151                 set_page_private(page, 0);
2152                 page_cache_release(page);
2153         }
2154         return ret;
2155 }
2156
2157 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2158 {
2159         struct extent_io_tree *tree;
2160
2161         tree = &BTRFS_I(page->mapping->host)->io_tree;
2162         extent_invalidatepage(tree, page, offset);
2163         btrfs_releasepage(page, GFP_NOFS);
2164 }
2165
2166 /*
2167  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2168  * called from a page fault handler when a page is first dirtied. Hence we must
2169  * be careful to check for EOF conditions here. We set the page up correctly
2170  * for a written page which means we get ENOSPC checking when writing into
2171  * holes and correct delalloc and unwritten extent mapping on filesystems that
2172  * support these features.
2173  *
2174  * We are not allowed to take the i_mutex here so we have to play games to
2175  * protect against truncate races as the page could now be beyond EOF.  Because
2176  * vmtruncate() writes the inode size before removing pages, once we have the
2177  * page lock we can determine safely if the page is beyond EOF. If it is not
2178  * beyond EOF, then the page is guaranteed safe against truncation until we
2179  * unlock the page.
2180  */
2181 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2182 {
2183         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2184         struct btrfs_root *root = BTRFS_I(inode)->root;
2185         unsigned long end;
2186         loff_t size;
2187         int ret;
2188         u64 page_start;
2189
2190         mutex_lock(&root->fs_info->fs_mutex);
2191         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2192         mutex_unlock(&root->fs_info->fs_mutex);
2193         if (ret)
2194                 goto out;
2195
2196         ret = -EINVAL;
2197
2198         lock_page(page);
2199         wait_on_page_writeback(page);
2200         size = i_size_read(inode);
2201         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2202
2203         if ((page->mapping != inode->i_mapping) ||
2204             (page_start > size)) {
2205                 /* page got truncated out from underneath us */
2206                 goto out_unlock;
2207         }
2208
2209         /* page is wholly or partially inside EOF */
2210         if (page_start + PAGE_CACHE_SIZE > size)
2211                 end = size & ~PAGE_CACHE_MASK;
2212         else
2213                 end = PAGE_CACHE_SIZE;
2214
2215         ret = btrfs_cow_one_page(inode, page, end);
2216
2217 out_unlock:
2218         unlock_page(page);
2219 out:
2220         return ret;
2221 }
2222
2223 static void btrfs_truncate(struct inode *inode)
2224 {
2225         struct btrfs_root *root = BTRFS_I(inode)->root;
2226         int ret;
2227         struct btrfs_trans_handle *trans;
2228         unsigned long nr;
2229
2230         if (!S_ISREG(inode->i_mode))
2231                 return;
2232         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2233                 return;
2234
2235         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2236
2237         mutex_lock(&root->fs_info->fs_mutex);
2238         trans = btrfs_start_transaction(root, 1);
2239         btrfs_set_trans_block_group(trans, inode);
2240
2241         /* FIXME, add redo link to tree so we don't leak on crash */
2242         ret = btrfs_truncate_in_trans(trans, root, inode,
2243                                       BTRFS_EXTENT_DATA_KEY);
2244         btrfs_update_inode(trans, root, inode);
2245         nr = trans->blocks_used;
2246
2247         ret = btrfs_end_transaction(trans, root);
2248         BUG_ON(ret);
2249         mutex_unlock(&root->fs_info->fs_mutex);
2250         btrfs_btree_balance_dirty(root, nr);
2251         btrfs_throttle(root);
2252 }
2253
2254 static int noinline create_subvol(struct btrfs_root *root, char *name,
2255                                   int namelen)
2256 {
2257         struct btrfs_trans_handle *trans;
2258         struct btrfs_key key;
2259         struct btrfs_root_item root_item;
2260         struct btrfs_inode_item *inode_item;
2261         struct extent_buffer *leaf;
2262         struct btrfs_root *new_root = root;
2263         struct inode *inode;
2264         struct inode *dir;
2265         int ret;
2266         int err;
2267         u64 objectid;
2268         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2269         unsigned long nr = 1;
2270
2271         mutex_lock(&root->fs_info->fs_mutex);
2272         ret = btrfs_check_free_space(root, 1, 0);
2273         if (ret)
2274                 goto fail_commit;
2275
2276         trans = btrfs_start_transaction(root, 1);
2277         BUG_ON(!trans);
2278
2279         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2280                                        0, &objectid);
2281         if (ret)
2282                 goto fail;
2283
2284         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2285                                         objectid, trans->transid, 0, 0,
2286                                         0, 0);
2287         if (IS_ERR(leaf))
2288                 return PTR_ERR(leaf);
2289
2290         btrfs_set_header_nritems(leaf, 0);
2291         btrfs_set_header_level(leaf, 0);
2292         btrfs_set_header_bytenr(leaf, leaf->start);
2293         btrfs_set_header_generation(leaf, trans->transid);
2294         btrfs_set_header_owner(leaf, objectid);
2295
2296         write_extent_buffer(leaf, root->fs_info->fsid,
2297                             (unsigned long)btrfs_header_fsid(leaf),
2298                             BTRFS_FSID_SIZE);
2299         btrfs_mark_buffer_dirty(leaf);
2300
2301         inode_item = &root_item.inode;
2302         memset(inode_item, 0, sizeof(*inode_item));
2303         inode_item->generation = cpu_to_le64(1);
2304         inode_item->size = cpu_to_le64(3);
2305         inode_item->nlink = cpu_to_le32(1);
2306         inode_item->nblocks = cpu_to_le64(1);
2307         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2308
2309         btrfs_set_root_bytenr(&root_item, leaf->start);
2310         btrfs_set_root_level(&root_item, 0);
2311         btrfs_set_root_refs(&root_item, 1);
2312         btrfs_set_root_used(&root_item, 0);
2313
2314         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2315         root_item.drop_level = 0;
2316
2317         free_extent_buffer(leaf);
2318         leaf = NULL;
2319
2320         btrfs_set_root_dirid(&root_item, new_dirid);
2321
2322         key.objectid = objectid;
2323         key.offset = 1;
2324         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2325         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2326                                 &root_item);
2327         if (ret)
2328                 goto fail;
2329
2330         /*
2331          * insert the directory item
2332          */
2333         key.offset = (u64)-1;
2334         dir = root->fs_info->sb->s_root->d_inode;
2335         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2336                                     name, namelen, dir->i_ino, &key,
2337                                     BTRFS_FT_DIR);
2338         if (ret)
2339                 goto fail;
2340
2341         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2342                              name, namelen, objectid,
2343                              root->fs_info->sb->s_root->d_inode->i_ino);
2344         if (ret)
2345                 goto fail;
2346
2347         ret = btrfs_commit_transaction(trans, root);
2348         if (ret)
2349                 goto fail_commit;
2350
2351         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2352         BUG_ON(!new_root);
2353
2354         trans = btrfs_start_transaction(new_root, 1);
2355         BUG_ON(!trans);
2356
2357         inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2358                                 new_dirid,
2359                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2360         if (IS_ERR(inode))
2361                 goto fail;
2362         inode->i_op = &btrfs_dir_inode_operations;
2363         inode->i_fop = &btrfs_dir_file_operations;
2364         new_root->inode = inode;
2365
2366         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2367                                      new_dirid);
2368         inode->i_nlink = 1;
2369         inode->i_size = 0;
2370         ret = btrfs_update_inode(trans, new_root, inode);
2371         if (ret)
2372                 goto fail;
2373 fail:
2374         nr = trans->blocks_used;
2375         err = btrfs_commit_transaction(trans, new_root);
2376         if (err && !ret)
2377                 ret = err;
2378 fail_commit:
2379         mutex_unlock(&root->fs_info->fs_mutex);
2380         btrfs_btree_balance_dirty(root, nr);
2381         btrfs_throttle(root);
2382         return ret;
2383 }
2384
2385 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2386 {
2387         struct btrfs_pending_snapshot *pending_snapshot;
2388         struct btrfs_trans_handle *trans;
2389         int ret;
2390         int err;
2391         unsigned long nr = 0;
2392
2393         if (!root->ref_cows)
2394                 return -EINVAL;
2395
2396         mutex_lock(&root->fs_info->fs_mutex);
2397         ret = btrfs_check_free_space(root, 1, 0);
2398         if (ret)
2399                 goto fail_unlock;
2400
2401         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2402         if (!pending_snapshot) {
2403                 ret = -ENOMEM;
2404                 goto fail_unlock;
2405         }
2406         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2407         if (!pending_snapshot->name) {
2408                 ret = -ENOMEM;
2409                 kfree(pending_snapshot);
2410                 goto fail_unlock;
2411         }
2412         memcpy(pending_snapshot->name, name, namelen);
2413         pending_snapshot->name[namelen] = '\0';
2414         trans = btrfs_start_transaction(root, 1);
2415         BUG_ON(!trans);
2416         pending_snapshot->root = root;
2417         list_add(&pending_snapshot->list,
2418                  &trans->transaction->pending_snapshots);
2419         ret = btrfs_update_inode(trans, root, root->inode);
2420         err = btrfs_commit_transaction(trans, root);
2421
2422 fail_unlock:
2423         mutex_unlock(&root->fs_info->fs_mutex);
2424         btrfs_btree_balance_dirty(root, nr);
2425         btrfs_throttle(root);
2426         return ret;
2427 }
2428
2429 unsigned long btrfs_force_ra(struct address_space *mapping,
2430                               struct file_ra_state *ra, struct file *file,
2431                               pgoff_t offset, pgoff_t last_index)
2432 {
2433         pgoff_t req_size;
2434
2435 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2436         req_size = last_index - offset + 1;
2437         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2438         return offset;
2439 #else
2440         req_size = min(last_index - offset + 1, (pgoff_t)128);
2441         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2442         return offset + req_size;
2443 #endif
2444 }
2445
2446 int btrfs_defrag_file(struct file *file) {
2447         struct inode *inode = fdentry(file)->d_inode;
2448         struct btrfs_root *root = BTRFS_I(inode)->root;
2449         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2450         struct page *page;
2451         unsigned long last_index;
2452         unsigned long ra_index = 0;
2453         u64 page_start;
2454         u64 page_end;
2455         unsigned long i;
2456         int ret;
2457
2458         mutex_lock(&root->fs_info->fs_mutex);
2459         ret = btrfs_check_free_space(root, inode->i_size, 0);
2460         mutex_unlock(&root->fs_info->fs_mutex);
2461         if (ret)
2462                 return -ENOSPC;
2463
2464         mutex_lock(&inode->i_mutex);
2465         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2466         for (i = 0; i <= last_index; i++) {
2467                 if (i == ra_index) {
2468                         ra_index = btrfs_force_ra(inode->i_mapping,
2469                                                   &file->f_ra,
2470                                                   file, ra_index, last_index);
2471                 }
2472                 page = grab_cache_page(inode->i_mapping, i);
2473                 if (!page)
2474                         goto out_unlock;
2475                 if (!PageUptodate(page)) {
2476                         btrfs_readpage(NULL, page);
2477                         lock_page(page);
2478                         if (!PageUptodate(page)) {
2479                                 unlock_page(page);
2480                                 page_cache_release(page);
2481                                 goto out_unlock;
2482                         }
2483                 }
2484                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2485                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2486
2487                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2488                 set_extent_delalloc(io_tree, page_start,
2489                                     page_end, GFP_NOFS);
2490
2491                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2492                 set_page_dirty(page);
2493                 unlock_page(page);
2494                 page_cache_release(page);
2495                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2496         }
2497
2498 out_unlock:
2499         mutex_unlock(&inode->i_mutex);
2500         return 0;
2501 }
2502
2503 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2504 {
2505         u64 new_size;
2506         u64 old_size;
2507         struct btrfs_ioctl_vol_args *vol_args;
2508         struct btrfs_trans_handle *trans;
2509         char *sizestr;
2510         int ret = 0;
2511         int namelen;
2512         int mod = 0;
2513
2514         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2515
2516         if (!vol_args)
2517                 return -ENOMEM;
2518
2519         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2520                 ret = -EFAULT;
2521                 goto out;
2522         }
2523         namelen = strlen(vol_args->name);
2524         if (namelen > BTRFS_VOL_NAME_MAX) {
2525                 ret = -EINVAL;
2526                 goto out;
2527         }
2528
2529         sizestr = vol_args->name;
2530         if (!strcmp(sizestr, "max"))
2531                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2532         else {
2533                 if (sizestr[0] == '-') {
2534                         mod = -1;
2535                         sizestr++;
2536                 } else if (sizestr[0] == '+') {
2537                         mod = 1;
2538                         sizestr++;
2539                 }
2540                 new_size = btrfs_parse_size(sizestr);
2541                 if (new_size == 0) {
2542                         ret = -EINVAL;
2543                         goto out;
2544                 }
2545         }
2546
2547         mutex_lock(&root->fs_info->fs_mutex);
2548         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2549
2550         if (mod < 0) {
2551                 if (new_size > old_size) {
2552                         ret = -EINVAL;
2553                         goto out_unlock;
2554                 }
2555                 new_size = old_size - new_size;
2556         } else if (mod > 0) {
2557                 new_size = old_size + new_size;
2558         }
2559
2560         if (new_size < 256 * 1024 * 1024) {
2561                 ret = -EINVAL;
2562                 goto out_unlock;
2563         }
2564         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2565                 ret = -EFBIG;
2566                 goto out_unlock;
2567         }
2568
2569         do_div(new_size, root->sectorsize);
2570         new_size *= root->sectorsize;
2571
2572 printk("new size is %Lu\n", new_size);
2573         if (new_size > old_size) {
2574                 trans = btrfs_start_transaction(root, 1);
2575                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2576                 btrfs_commit_transaction(trans, root);
2577         } else {
2578                 ret = btrfs_shrink_extent_tree(root, new_size);
2579         }
2580
2581 out_unlock:
2582         mutex_unlock(&root->fs_info->fs_mutex);
2583 out:
2584         kfree(vol_args);
2585         return ret;
2586 }
2587
2588 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2589                                             void __user *arg)
2590 {
2591         struct btrfs_ioctl_vol_args *vol_args;
2592         struct btrfs_dir_item *di;
2593         struct btrfs_path *path;
2594         u64 root_dirid;
2595         int namelen;
2596         int ret;
2597
2598         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2599
2600         if (!vol_args)
2601                 return -ENOMEM;
2602
2603         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2604                 ret = -EFAULT;
2605                 goto out;
2606         }
2607
2608         namelen = strlen(vol_args->name);
2609         if (namelen > BTRFS_VOL_NAME_MAX) {
2610                 ret = -EINVAL;
2611                 goto out;
2612         }
2613         if (strchr(vol_args->name, '/')) {
2614                 ret = -EINVAL;
2615                 goto out;
2616         }
2617
2618         path = btrfs_alloc_path();
2619         if (!path) {
2620                 ret = -ENOMEM;
2621                 goto out;
2622         }
2623
2624         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2625         mutex_lock(&root->fs_info->fs_mutex);
2626         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2627                             path, root_dirid,
2628                             vol_args->name, namelen, 0);
2629         mutex_unlock(&root->fs_info->fs_mutex);
2630         btrfs_free_path(path);
2631
2632         if (di && !IS_ERR(di)) {
2633                 ret = -EEXIST;
2634                 goto out;
2635         }
2636
2637         if (IS_ERR(di)) {
2638                 ret = PTR_ERR(di);
2639                 goto out;
2640         }
2641
2642         if (root == root->fs_info->tree_root)
2643                 ret = create_subvol(root, vol_args->name, namelen);
2644         else
2645                 ret = create_snapshot(root, vol_args->name, namelen);
2646 out:
2647         kfree(vol_args);
2648         return ret;
2649 }
2650
2651 static int btrfs_ioctl_defrag(struct file *file)
2652 {
2653         struct inode *inode = fdentry(file)->d_inode;
2654         struct btrfs_root *root = BTRFS_I(inode)->root;
2655
2656         switch (inode->i_mode & S_IFMT) {
2657         case S_IFDIR:
2658                 mutex_lock(&root->fs_info->fs_mutex);
2659                 btrfs_defrag_root(root, 0);
2660                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2661                 mutex_unlock(&root->fs_info->fs_mutex);
2662                 break;
2663         case S_IFREG:
2664                 btrfs_defrag_file(file);
2665                 break;
2666         }
2667
2668         return 0;
2669 }
2670
2671 long btrfs_ioctl(struct file *file, unsigned int
2672                 cmd, unsigned long arg)
2673 {
2674         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2675
2676         switch (cmd) {
2677         case BTRFS_IOC_SNAP_CREATE:
2678                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2679         case BTRFS_IOC_DEFRAG:
2680                 return btrfs_ioctl_defrag(file);
2681         case BTRFS_IOC_RESIZE:
2682                 return btrfs_ioctl_resize(root, (void __user *)arg);
2683         }
2684
2685         return -ENOTTY;
2686 }
2687
2688 /*
2689  * Called inside transaction, so use GFP_NOFS
2690  */
2691 struct inode *btrfs_alloc_inode(struct super_block *sb)
2692 {
2693         struct btrfs_inode *ei;
2694
2695         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2696         if (!ei)
2697                 return NULL;
2698         ei->last_trans = 0;
2699         ei->ordered_trans = 0;
2700         return &ei->vfs_inode;
2701 }
2702
2703 void btrfs_destroy_inode(struct inode *inode)
2704 {
2705         WARN_ON(!list_empty(&inode->i_dentry));
2706         WARN_ON(inode->i_data.nrpages);
2707
2708         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2709         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2710 }
2711
2712 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2713 static void init_once(struct kmem_cache * cachep, void *foo)
2714 #else
2715 static void init_once(void * foo, struct kmem_cache * cachep,
2716                       unsigned long flags)
2717 #endif
2718 {
2719         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2720
2721         inode_init_once(&ei->vfs_inode);
2722 }
2723
2724 void btrfs_destroy_cachep(void)
2725 {
2726         if (btrfs_inode_cachep)
2727                 kmem_cache_destroy(btrfs_inode_cachep);
2728         if (btrfs_trans_handle_cachep)
2729                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2730         if (btrfs_transaction_cachep)
2731                 kmem_cache_destroy(btrfs_transaction_cachep);
2732         if (btrfs_bit_radix_cachep)
2733                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2734         if (btrfs_path_cachep)
2735                 kmem_cache_destroy(btrfs_path_cachep);
2736 }
2737
2738 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2739                                        unsigned long extra_flags,
2740 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2741                                        void (*ctor)(struct kmem_cache *, void *)
2742 #else
2743                                        void (*ctor)(void *, struct kmem_cache *,
2744                                                     unsigned long)
2745 #endif
2746                                      )
2747 {
2748         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2749                                  SLAB_MEM_SPREAD | extra_flags), ctor
2750 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2751                                  ,NULL
2752 #endif
2753                                 );
2754 }
2755
2756 int btrfs_init_cachep(void)
2757 {
2758         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2759                                           sizeof(struct btrfs_inode),
2760                                           0, init_once);
2761         if (!btrfs_inode_cachep)
2762                 goto fail;
2763         btrfs_trans_handle_cachep =
2764                         btrfs_cache_create("btrfs_trans_handle_cache",
2765                                            sizeof(struct btrfs_trans_handle),
2766                                            0, NULL);
2767         if (!btrfs_trans_handle_cachep)
2768                 goto fail;
2769         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2770                                              sizeof(struct btrfs_transaction),
2771                                              0, NULL);
2772         if (!btrfs_transaction_cachep)
2773                 goto fail;
2774         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2775                                          sizeof(struct btrfs_path),
2776                                          0, NULL);
2777         if (!btrfs_path_cachep)
2778                 goto fail;
2779         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2780                                               SLAB_DESTROY_BY_RCU, NULL);
2781         if (!btrfs_bit_radix_cachep)
2782                 goto fail;
2783         return 0;
2784 fail:
2785         btrfs_destroy_cachep();
2786         return -ENOMEM;
2787 }
2788
2789 static int btrfs_getattr(struct vfsmount *mnt,
2790                          struct dentry *dentry, struct kstat *stat)
2791 {
2792         struct inode *inode = dentry->d_inode;
2793         generic_fillattr(inode, stat);
2794         stat->blksize = PAGE_CACHE_SIZE;
2795         return 0;
2796 }
2797
2798 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2799                            struct inode * new_dir,struct dentry *new_dentry)
2800 {
2801         struct btrfs_trans_handle *trans;
2802         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2803         struct inode *new_inode = new_dentry->d_inode;
2804         struct inode *old_inode = old_dentry->d_inode;
2805         struct timespec ctime = CURRENT_TIME;
2806         struct btrfs_path *path;
2807         int ret;
2808
2809         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2810             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2811                 return -ENOTEMPTY;
2812         }
2813
2814         mutex_lock(&root->fs_info->fs_mutex);
2815         ret = btrfs_check_free_space(root, 1, 0);
2816         if (ret)
2817                 goto out_unlock;
2818
2819         trans = btrfs_start_transaction(root, 1);
2820
2821         btrfs_set_trans_block_group(trans, new_dir);
2822         path = btrfs_alloc_path();
2823         if (!path) {
2824                 ret = -ENOMEM;
2825                 goto out_fail;
2826         }
2827
2828         old_dentry->d_inode->i_nlink++;
2829         old_dir->i_ctime = old_dir->i_mtime = ctime;
2830         new_dir->i_ctime = new_dir->i_mtime = ctime;
2831         old_inode->i_ctime = ctime;
2832
2833         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2834         if (ret)
2835                 goto out_fail;
2836
2837         if (new_inode) {
2838                 new_inode->i_ctime = CURRENT_TIME;
2839                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2840                 if (ret)
2841                         goto out_fail;
2842         }
2843         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
2844         if (ret)
2845                 goto out_fail;
2846
2847 out_fail:
2848         btrfs_free_path(path);
2849         btrfs_end_transaction(trans, root);
2850 out_unlock:
2851         mutex_unlock(&root->fs_info->fs_mutex);
2852         return ret;
2853 }
2854
2855 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2856                          const char *symname)
2857 {
2858         struct btrfs_trans_handle *trans;
2859         struct btrfs_root *root = BTRFS_I(dir)->root;
2860         struct btrfs_path *path;
2861         struct btrfs_key key;
2862         struct inode *inode = NULL;
2863         int err;
2864         int drop_inode = 0;
2865         u64 objectid;
2866         int name_len;
2867         int datasize;
2868         unsigned long ptr;
2869         struct btrfs_file_extent_item *ei;
2870         struct extent_buffer *leaf;
2871         unsigned long nr = 0;
2872
2873         name_len = strlen(symname) + 1;
2874         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2875                 return -ENAMETOOLONG;
2876
2877         mutex_lock(&root->fs_info->fs_mutex);
2878         err = btrfs_check_free_space(root, 1, 0);
2879         if (err)
2880                 goto out_fail;
2881
2882         trans = btrfs_start_transaction(root, 1);
2883         btrfs_set_trans_block_group(trans, dir);
2884
2885         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2886         if (err) {
2887                 err = -ENOSPC;
2888                 goto out_unlock;
2889         }
2890
2891         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2892                                 dentry->d_name.len,
2893                                 dentry->d_parent->d_inode->i_ino, objectid,
2894                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2895         err = PTR_ERR(inode);
2896         if (IS_ERR(inode))
2897                 goto out_unlock;
2898
2899         btrfs_set_trans_block_group(trans, inode);
2900         err = btrfs_add_nondir(trans, dentry, inode, 0);
2901         if (err)
2902                 drop_inode = 1;
2903         else {
2904                 inode->i_mapping->a_ops = &btrfs_aops;
2905                 inode->i_fop = &btrfs_file_operations;
2906                 inode->i_op = &btrfs_file_inode_operations;
2907                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2908                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2909                                      inode->i_mapping, GFP_NOFS);
2910                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2911         }
2912         dir->i_sb->s_dirt = 1;
2913         btrfs_update_inode_block_group(trans, inode);
2914         btrfs_update_inode_block_group(trans, dir);
2915         if (drop_inode)
2916                 goto out_unlock;
2917
2918         path = btrfs_alloc_path();
2919         BUG_ON(!path);
2920         key.objectid = inode->i_ino;
2921         key.offset = 0;
2922         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2923         datasize = btrfs_file_extent_calc_inline_size(name_len);
2924         err = btrfs_insert_empty_item(trans, root, path, &key,
2925                                       datasize);
2926         if (err) {
2927                 drop_inode = 1;
2928                 goto out_unlock;
2929         }
2930         leaf = path->nodes[0];
2931         ei = btrfs_item_ptr(leaf, path->slots[0],
2932                             struct btrfs_file_extent_item);
2933         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2934         btrfs_set_file_extent_type(leaf, ei,
2935                                    BTRFS_FILE_EXTENT_INLINE);
2936         ptr = btrfs_file_extent_inline_start(ei);
2937         write_extent_buffer(leaf, symname, ptr, name_len);
2938         btrfs_mark_buffer_dirty(leaf);
2939         btrfs_free_path(path);
2940
2941         inode->i_op = &btrfs_symlink_inode_operations;
2942         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2943         inode->i_size = name_len - 1;
2944         err = btrfs_update_inode(trans, root, inode);
2945         if (err)
2946                 drop_inode = 1;
2947
2948 out_unlock:
2949         nr = trans->blocks_used;
2950         btrfs_end_transaction(trans, root);
2951 out_fail:
2952         mutex_unlock(&root->fs_info->fs_mutex);
2953         if (drop_inode) {
2954                 inode_dec_link_count(inode);
2955                 iput(inode);
2956         }
2957         btrfs_btree_balance_dirty(root, nr);
2958         btrfs_throttle(root);
2959         return err;
2960 }
2961 static int btrfs_permission(struct inode *inode, int mask,
2962                             struct nameidata *nd)
2963 {
2964         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2965                 return -EACCES;
2966         return generic_permission(inode, mask, NULL);
2967 }
2968
2969 static struct inode_operations btrfs_dir_inode_operations = {
2970         .lookup         = btrfs_lookup,
2971         .create         = btrfs_create,
2972         .unlink         = btrfs_unlink,
2973         .link           = btrfs_link,
2974         .mkdir          = btrfs_mkdir,
2975         .rmdir          = btrfs_rmdir,
2976         .rename         = btrfs_rename,
2977         .symlink        = btrfs_symlink,
2978         .setattr        = btrfs_setattr,
2979         .mknod          = btrfs_mknod,
2980         .setxattr       = generic_setxattr,
2981         .getxattr       = generic_getxattr,
2982         .listxattr      = btrfs_listxattr,
2983         .removexattr    = generic_removexattr,
2984         .permission     = btrfs_permission,
2985 };
2986 static struct inode_operations btrfs_dir_ro_inode_operations = {
2987         .lookup         = btrfs_lookup,
2988         .permission     = btrfs_permission,
2989 };
2990 static struct file_operations btrfs_dir_file_operations = {
2991         .llseek         = generic_file_llseek,
2992         .read           = generic_read_dir,
2993         .readdir        = btrfs_readdir,
2994         .unlocked_ioctl = btrfs_ioctl,
2995 #ifdef CONFIG_COMPAT
2996         .compat_ioctl   = btrfs_ioctl,
2997 #endif
2998 };
2999
3000 static struct extent_io_ops btrfs_extent_io_ops = {
3001         .fill_delalloc = run_delalloc_range,
3002         .writepage_io_hook = btrfs_writepage_io_hook,
3003         .readpage_io_hook = btrfs_readpage_io_hook,
3004         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3005 };
3006
3007 static struct address_space_operations btrfs_aops = {
3008         .readpage       = btrfs_readpage,
3009         .writepage      = btrfs_writepage,
3010         .writepages     = btrfs_writepages,
3011         .readpages      = btrfs_readpages,
3012         .sync_page      = block_sync_page,
3013         .bmap           = btrfs_bmap,
3014         .invalidatepage = btrfs_invalidatepage,
3015         .releasepage    = btrfs_releasepage,
3016         .set_page_dirty = __set_page_dirty_nobuffers,
3017 };
3018
3019 static struct address_space_operations btrfs_symlink_aops = {
3020         .readpage       = btrfs_readpage,
3021         .writepage      = btrfs_writepage,
3022         .invalidatepage = btrfs_invalidatepage,
3023         .releasepage    = btrfs_releasepage,
3024 };
3025
3026 static struct inode_operations btrfs_file_inode_operations = {
3027         .truncate       = btrfs_truncate,
3028         .getattr        = btrfs_getattr,
3029         .setattr        = btrfs_setattr,
3030         .setxattr       = generic_setxattr,
3031         .getxattr       = generic_getxattr,
3032         .listxattr      = btrfs_listxattr,
3033         .removexattr    = generic_removexattr,
3034         .permission     = btrfs_permission,
3035 };
3036 static struct inode_operations btrfs_special_inode_operations = {
3037         .getattr        = btrfs_getattr,
3038         .setattr        = btrfs_setattr,
3039         .permission     = btrfs_permission,
3040 };
3041 static struct inode_operations btrfs_symlink_inode_operations = {
3042         .readlink       = generic_readlink,
3043         .follow_link    = page_follow_link_light,
3044         .put_link       = page_put_link,
3045         .permission     = btrfs_permission,
3046 };