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         return 0;
  84
  85         if (for_del)
  86                 thresh = total * 90;
  87         else
  88                 thresh = total * 85;
  89
  90         do_div(thresh, 100);
  91
  92         spin_lock(&root->fs_info->delalloc_lock);
  93         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  94                 ret = -ENOSPC;
  95         spin_unlock(&root->fs_info->delalloc_lock);
  96         return ret;
  97 }
  98
  99 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 100 {
 101         struct btrfs_root *root = BTRFS_I(inode)->root;
 102         struct btrfs_trans_handle *trans;
 103         u64 alloc_hint = 0;
 104         u64 num_bytes;
 105         u64 cur_alloc_size;
 106         u64 blocksize = root->sectorsize;
 107         u64 orig_start = start;
 108         u64 orig_num_bytes;
 109         struct btrfs_key ins;
 110         int ret;
 111
 112         trans = btrfs_start_transaction(root, 1);
 113         BUG_ON(!trans);
 114         btrfs_set_trans_block_group(trans, inode);
 115
 116         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 117         num_bytes = max(blocksize,  num_bytes);
 118         ret = btrfs_drop_extents(trans, root, inode,
 119                                  start, start + num_bytes, start, &alloc_hint);
 120         orig_num_bytes = num_bytes;
 121
 122         if (alloc_hint == EXTENT_MAP_INLINE)
 123                 goto out;
 124
 125         while(num_bytes > 0) {
 126                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 127                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 128                                          root->root_key.objectid,
 129                                          trans->transid,
 130                                          inode->i_ino, start, 0,
 131                                          alloc_hint, (u64)-1, &ins, 1);
 132                 if (ret) {
 133                         WARN_ON(1);
 134                         goto out;
 135                 }
 136                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 137                                                start, ins.objectid, ins.offset,
 138                                                ins.offset);
 139                 btrfs_check_file(root, inode);
 140                 num_bytes -= cur_alloc_size;
 141                 alloc_hint = ins.objectid + ins.offset;
 142                 start += cur_alloc_size;
 143         }
 144         btrfs_drop_extent_cache(inode, orig_start,
 145                                 orig_start + orig_num_bytes - 1);
 146         btrfs_add_ordered_inode(inode);
 147 out:
 148         btrfs_end_transaction(trans, root);
 149         return ret;
 150 }
 151
 152 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 153 {
 154         u64 extent_start;
 155         u64 extent_end;
 156         u64 bytenr;
 157         u64 cow_end;
 158         u64 loops = 0;
 159         u64 total_fs_bytes;
 160         struct btrfs_root *root = BTRFS_I(inode)->root;
 161         struct extent_buffer *leaf;
 162         int found_type;
 163         struct btrfs_path *path;
 164         struct btrfs_file_extent_item *item;
 165         int ret;
 166         int err;
 167         struct btrfs_key found_key;
 168
 169         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 170         path = btrfs_alloc_path();
 171         BUG_ON(!path);
 172 again:
 173         ret = btrfs_lookup_file_extent(NULL, root, path,
 174                                        inode->i_ino, start, 0);
 175         if (ret < 0) {
 176                 btrfs_free_path(path);
 177                 return ret;
 178         }
 179
 180         cow_end = end;
 181         if (ret != 0) {
 182                 if (path->slots[0] == 0)
 183                         goto not_found;
 184                 path->slots[0]--;
 185         }
 186
 187         leaf = path->nodes[0];
 188         item = btrfs_item_ptr(leaf, path->slots[0],
 189                               struct btrfs_file_extent_item);
 190
 191         /* are we inside the extent that was found? */
 192         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 193         found_type = btrfs_key_type(&found_key);
 194         if (found_key.objectid != inode->i_ino ||
 195             found_type != BTRFS_EXTENT_DATA_KEY) {
 196                 goto not_found;
 197         }
 198
 199         found_type = btrfs_file_extent_type(leaf, item);
 200         extent_start = found_key.offset;
 201         if (found_type == BTRFS_FILE_EXTENT_REG) {
 202                 u64 extent_num_bytes;
 203
 204                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 205                 extent_end = extent_start + extent_num_bytes;
 206                 err = 0;
 207
 208                 if (loops && start != extent_start)
 209                         goto not_found;
 210
 211                 if (start < extent_start || start >= extent_end)
 212                         goto not_found;
 213
 214                 cow_end = min(end, extent_end - 1);
 215                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 216                 if (bytenr == 0)
 217                         goto not_found;
 218
 219                 /*
 220                  * we may be called by the resizer, make sure we're inside
 221                  * the limits of the FS
 222                  */
 223                 if (bytenr + extent_num_bytes > total_fs_bytes)
 224                         goto not_found;
 225
 226                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 227                         goto not_found;
 228                 }
 229
 230                 start = extent_end;
 231         } else {
 232                 goto not_found;
 233         }
 234 loop:
 235         if (start > end) {
 236                 btrfs_free_path(path);
 237                 return 0;
 238         }
 239         btrfs_release_path(root, path);
 240         loops++;
 241         goto again;
 242
 243 not_found:
 244         cow_file_range(inode, start, cow_end);
 245         start = cow_end + 1;
 246         goto loop;
 247 }
 248
 249 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 250 {
 251         struct btrfs_root *root = BTRFS_I(inode)->root;
 252         u64 num_bytes;
 253         int ret;
 254         mutex_lock(&root->fs_info->fs_mutex);
 255         if (btrfs_test_opt(root, NODATACOW) ||
 256             btrfs_test_flag(inode, NODATACOW))
 257                 ret = run_delalloc_nocow(inode, start, end);
 258         else
 259                 ret = cow_file_range(inode, start, end);
 260
 261         spin_lock(&root->fs_info->delalloc_lock);
 262         num_bytes = end + 1 - start;
 263         if (root->fs_info->delalloc_bytes < num_bytes) {
 264                 printk("delalloc accounting error total %llu sub %llu\n",
 265                        root->fs_info->delalloc_bytes, num_bytes);
 266         } else {
 267                 root->fs_info->delalloc_bytes -= num_bytes;
 268         }
 269         spin_unlock(&root->fs_info->delalloc_lock);
 270
 271         mutex_unlock(&root->fs_info->fs_mutex);
 272         return ret;
 273 }
 274
 275 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 276 {
 277         struct inode *inode = page->mapping->host;
 278         struct btrfs_root *root = BTRFS_I(inode)->root;
 279         struct btrfs_trans_handle *trans;
 280         char *kaddr;
 281         int ret = 0;
 282         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 283         size_t offset = start - page_start;
 284         if (btrfs_test_opt(root, NODATASUM) ||
 285             btrfs_test_flag(inode, NODATASUM))
 286                 return 0;
 287         mutex_lock(&root->fs_info->fs_mutex);
 288         trans = btrfs_start_transaction(root, 1);
 289         btrfs_set_trans_block_group(trans, inode);
 290         kaddr = kmap(page);
 291         btrfs_csum_file_block(trans, root, inode, inode->i_ino,
 292                               start, kaddr + offset, end - start + 1);
 293         kunmap(page);
 294         ret = btrfs_end_transaction(trans, root);
 295         BUG_ON(ret);
 296         mutex_unlock(&root->fs_info->fs_mutex);
 297         return ret;
 298 }
 299
 300 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 301 {
 302         int ret = 0;
 303         struct inode *inode = page->mapping->host;
 304         struct btrfs_root *root = BTRFS_I(inode)->root;
 305         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 306         struct btrfs_csum_item *item;
 307         struct btrfs_path *path = NULL;
 308         u32 csum;
 309         if (btrfs_test_opt(root, NODATASUM) ||
 310             btrfs_test_flag(inode, NODATASUM))
 311                 return 0;
 312         mutex_lock(&root->fs_info->fs_mutex);
 313         path = btrfs_alloc_path();
 314         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 315         if (IS_ERR(item)) {
 316                 ret = PTR_ERR(item);
 317                 /* a csum that isn't present is a preallocated region. */
 318                 if (ret == -ENOENT || ret == -EFBIG)
 319                         ret = 0;
 320                 csum = 0;
 321                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino, start);
 322                 goto out;
 323         }
 324         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 325                            BTRFS_CRC32_SIZE);
 326         set_state_private(io_tree, start, csum);
 327 out:
 328         if (path)
 329                 btrfs_free_path(path);
 330         mutex_unlock(&root->fs_info->fs_mutex);
 331         return ret;
 332 }
 333
 334 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 335                                struct extent_state *state)
 336 {
 337         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 338         struct inode *inode = page->mapping->host;
 339         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 340         char *kaddr;
 341         u64 private = ~(u32)0;
 342         int ret;
 343         struct btrfs_root *root = BTRFS_I(inode)->root;
 344         u32 csum = ~(u32)0;
 345         unsigned long flags;
 346
 347         if (btrfs_test_opt(root, NODATASUM) ||
 348             btrfs_test_flag(inode, NODATASUM))
 349                 return 0;
 350
 351         if (state->start == start) {
 352                 private = state->private;
 353                 ret = 0;
 354         } else {
 355                 ret = get_state_private(io_tree, start, &private);
 356         }
 357         local_irq_save(flags);
 358         kaddr = kmap_atomic(page, KM_IRQ0);
 359         if (ret) {
 360                 goto zeroit;
 361         }
 362         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 363         btrfs_csum_final(csum, (char *)&csum);
 364         if (csum != private) {
 365                 goto zeroit;
 366         }
 367         kunmap_atomic(kaddr, KM_IRQ0);
 368         local_irq_restore(flags);
 369         return 0;
 370
 371 zeroit:
 372         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 373                page->mapping->host->i_ino, (unsigned long long)start, csum,
 374                private);
 375         memset(kaddr + offset, 1, end - start + 1);
 376         flush_dcache_page(page);
 377         kunmap_atomic(kaddr, KM_IRQ0);
 378         local_irq_restore(flags);
 379         return 0;
 380 }
 381
 382 void btrfs_read_locked_inode(struct inode *inode)
 383 {
 384         struct btrfs_path *path;
 385         struct extent_buffer *leaf;
 386         struct btrfs_inode_item *inode_item;
 387         struct btrfs_inode_timespec *tspec;
 388         struct btrfs_root *root = BTRFS_I(inode)->root;
 389         struct btrfs_key location;
 390         u64 alloc_group_block;
 391         u32 rdev;
 392         int ret;
 393
 394         path = btrfs_alloc_path();
 395         BUG_ON(!path);
 396         mutex_lock(&root->fs_info->fs_mutex);
 397         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 398
 399         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 400         if (ret)
 401                 goto make_bad;
 402
 403         leaf = path->nodes[0];
 404         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 405                                     struct btrfs_inode_item);
 406
 407         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 408         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 409         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 410         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 411         inode->i_size = btrfs_inode_size(leaf, inode_item);
 412
 413         tspec = btrfs_inode_atime(inode_item);
 414         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 415         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 416
 417         tspec = btrfs_inode_mtime(inode_item);
 418         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 419         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 420
 421         tspec = btrfs_inode_ctime(inode_item);
 422         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 423         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 424
 425         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 426         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 427         inode->i_rdev = 0;
 428         rdev = btrfs_inode_rdev(leaf, inode_item);
 429
 430         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 431         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 432                                                        alloc_group_block);
 433         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 434         if (!BTRFS_I(inode)->block_group) {
 435                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 436                                                          NULL, 0, 0, 0);
 437         }
 438         btrfs_free_path(path);
 439         inode_item = NULL;
 440
 441         mutex_unlock(&root->fs_info->fs_mutex);
 442
 443         switch (inode->i_mode & S_IFMT) {
 444         case S_IFREG:
 445                 inode->i_mapping->a_ops = &btrfs_aops;
 446                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 447                 inode->i_fop = &btrfs_file_operations;
 448                 inode->i_op = &btrfs_file_inode_operations;
 449                 break;
 450         case S_IFDIR:
 451                 inode->i_fop = &btrfs_dir_file_operations;
 452                 if (root == root->fs_info->tree_root)
 453                         inode->i_op = &btrfs_dir_ro_inode_operations;
 454                 else
 455                         inode->i_op = &btrfs_dir_inode_operations;
 456                 break;
 457         case S_IFLNK:
 458                 inode->i_op = &btrfs_symlink_inode_operations;
 459                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 460                 break;
 461         default:
 462                 init_special_inode(inode, inode->i_mode, rdev);
 463                 break;
 464         }
 465         return;
 466
 467 make_bad:
 468         btrfs_release_path(root, path);
 469         btrfs_free_path(path);
 470         mutex_unlock(&root->fs_info->fs_mutex);
 471         make_bad_inode(inode);
 472 }
 473
 474 static void fill_inode_item(struct extent_buffer *leaf,
 475                             struct btrfs_inode_item *item,
 476                             struct inode *inode)
 477 {
 478         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 479         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 480         btrfs_set_inode_size(leaf, item, inode->i_size);
 481         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 482         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 483
 484         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 485                                inode->i_atime.tv_sec);
 486         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 487                                 inode->i_atime.tv_nsec);
 488
 489         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 490                                inode->i_mtime.tv_sec);
 491         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 492                                 inode->i_mtime.tv_nsec);
 493
 494         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 495                                inode->i_ctime.tv_sec);
 496         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 497                                 inode->i_ctime.tv_nsec);
 498
 499         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 500         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 501         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 502         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 503         btrfs_set_inode_block_group(leaf, item,
 504                                     BTRFS_I(inode)->block_group->key.objectid);
 505 }
 506
 507 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 508                               struct btrfs_root *root,
 509                               struct inode *inode)
 510 {
 511         struct btrfs_inode_item *inode_item;
 512         struct btrfs_path *path;
 513         struct extent_buffer *leaf;
 514         int ret;
 515
 516         path = btrfs_alloc_path();
 517         BUG_ON(!path);
 518         ret = btrfs_lookup_inode(trans, root, path,
 519                                  &BTRFS_I(inode)->location, 1);
 520         if (ret) {
 521                 if (ret > 0)
 522                         ret = -ENOENT;
 523                 goto failed;
 524         }
 525
 526         leaf = path->nodes[0];
 527         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 528                                   struct btrfs_inode_item);
 529
 530         fill_inode_item(leaf, inode_item, inode);
 531         btrfs_mark_buffer_dirty(leaf);
 532         btrfs_set_inode_last_trans(trans, inode);
 533         ret = 0;
 534 failed:
 535         btrfs_release_path(root, path);
 536         btrfs_free_path(path);
 537         return ret;
 538 }
 539
 540
 541 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 542                               struct btrfs_root *root,
 543                               struct inode *dir,
 544                               struct dentry *dentry)
 545 {
 546         struct btrfs_path *path;
 547         const char *name = dentry->d_name.name;
 548         int name_len = dentry->d_name.len;
 549         int ret = 0;
 550         struct extent_buffer *leaf;
 551         struct btrfs_dir_item *di;
 552         struct btrfs_key key;
 553
 554         path = btrfs_alloc_path();
 555         if (!path) {
 556                 ret = -ENOMEM;
 557                 goto err;
 558         }
 559
 560         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 561                                     name, name_len, -1);
 562         if (IS_ERR(di)) {
 563                 ret = PTR_ERR(di);
 564                 goto err;
 565         }
 566         if (!di) {
 567                 ret = -ENOENT;
 568                 goto err;
 569         }
 570         leaf = path->nodes[0];
 571         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 572         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 573         if (ret)
 574                 goto err;
 575         btrfs_release_path(root, path);
 576
 577         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 578                                          key.objectid, name, name_len, -1);
 579         if (IS_ERR(di)) {
 580                 ret = PTR_ERR(di);
 581                 goto err;
 582         }
 583         if (!di) {
 584                 ret = -ENOENT;
 585                 goto err;
 586         }
 587         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 588
 589         dentry->d_inode->i_ctime = dir->i_ctime;
 590         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 591                                   dentry->d_inode->i_ino,
 592                                   dentry->d_parent->d_inode->i_ino);
 593         if (ret) {
 594                 printk("failed to delete reference to %.*s, "
 595                        "inode %lu parent %lu\n", name_len, name,
 596                        dentry->d_inode->i_ino,
 597                        dentry->d_parent->d_inode->i_ino);
 598         }
 599 err:
 600         btrfs_free_path(path);
 601         if (!ret) {
 602                 dir->i_size -= name_len * 2;
 603                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 604                 btrfs_update_inode(trans, root, dir);
 605 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 606                 dentry->d_inode->i_nlink--;
 607 #else
 608                 drop_nlink(dentry->d_inode);
 609 #endif
 610                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 611                 dir->i_sb->s_dirt = 1;
 612         }
 613         return ret;
 614 }
 615
 616 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 617 {
 618         struct btrfs_root *root;
 619         struct btrfs_trans_handle *trans;
 620         struct inode *inode = dentry->d_inode;
 621         int ret;
 622         unsigned long nr = 0;
 623
 624         root = BTRFS_I(dir)->root;
 625         mutex_lock(&root->fs_info->fs_mutex);
 626
 627         ret = btrfs_check_free_space(root, 1, 1);
 628         if (ret)
 629                 goto fail;
 630
 631         trans = btrfs_start_transaction(root, 1);
 632
 633         btrfs_set_trans_block_group(trans, dir);
 634         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 635         nr = trans->blocks_used;
 636
 637         if (inode->i_nlink == 0) {
 638                 int found;
 639                 /* if the inode isn't linked anywhere,
 640                  * we don't need to worry about
 641                  * data=ordered
 642                  */
 643                 found = btrfs_del_ordered_inode(inode);
 644                 if (found == 1) {
 645                         atomic_dec(&inode->i_count);
 646                 }
 647         }
 648
 649         btrfs_end_transaction(trans, root);
 650 fail:
 651         mutex_unlock(&root->fs_info->fs_mutex);
 652         btrfs_btree_balance_dirty(root, nr);
 653         btrfs_throttle(root);
 654         return ret;
 655 }
 656
 657 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 658 {
 659         struct inode *inode = dentry->d_inode;
 660         int err = 0;
 661         int ret;
 662         struct btrfs_root *root = BTRFS_I(dir)->root;
 663         struct btrfs_trans_handle *trans;
 664         unsigned long nr = 0;
 665
 666         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 667                 return -ENOTEMPTY;
 668
 669         mutex_lock(&root->fs_info->fs_mutex);
 670         ret = btrfs_check_free_space(root, 1, 1);
 671         if (ret)
 672                 goto fail;
 673
 674         trans = btrfs_start_transaction(root, 1);
 675         btrfs_set_trans_block_group(trans, dir);
 676
 677         /* now the directory is empty */
 678         err = btrfs_unlink_trans(trans, root, dir, dentry);
 679         if (!err) {
 680                 inode->i_size = 0;
 681         }
 682
 683         nr = trans->blocks_used;
 684         ret = btrfs_end_transaction(trans, root);
 685 fail:
 686         mutex_unlock(&root->fs_info->fs_mutex);
 687         btrfs_btree_balance_dirty(root, nr);
 688         btrfs_throttle(root);
 689
 690         if (ret && !err)
 691                 err = ret;
 692         return err;
 693 }
 694
 695 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 696                             struct btrfs_root *root,
 697                             struct inode *inode)
 698 {
 699         struct btrfs_path *path;
 700         int ret;
 701
 702         clear_inode(inode);
 703
 704         path = btrfs_alloc_path();
 705         BUG_ON(!path);
 706         ret = btrfs_lookup_inode(trans, root, path,
 707                                  &BTRFS_I(inode)->location, -1);
 708         if (ret > 0)
 709                 ret = -ENOENT;
 710         if (!ret)
 711                 ret = btrfs_del_item(trans, root, path);
 712         btrfs_free_path(path);
 713         return ret;
 714 }
 715
 716 /*
 717  * this can truncate away extent items, csum items and directory items.
 718  * It starts at a high offset and removes keys until it can't find
 719  * any higher than i_size.
 720  *
 721  * csum items that cross the new i_size are truncated to the new size
 722  * as well.
 723  */
 724 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 725                                    struct btrfs_root *root,
 726                                    struct inode *inode)
 727 {
 728         int ret;
 729         struct btrfs_path *path;
 730         struct btrfs_key key;
 731         struct btrfs_key found_key;
 732         u32 found_type;
 733         struct extent_buffer *leaf;
 734         struct btrfs_file_extent_item *fi;
 735         u64 extent_start = 0;
 736         u64 extent_num_bytes = 0;
 737         u64 item_end = 0;
 738         u64 root_gen = 0;
 739         u64 root_owner = 0;
 740         int found_extent;
 741         int del_item;
 742         int extent_type = -1;
 743
 744         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 745         path = btrfs_alloc_path();
 746         path->reada = -1;
 747         BUG_ON(!path);
 748
 749         /* FIXME, add redo link to tree so we don't leak on crash */
 750         key.objectid = inode->i_ino;
 751         key.offset = (u64)-1;
 752         key.type = (u8)-1;
 753
 754         while(1) {
 755                 btrfs_init_path(path);
 756                 fi = NULL;
 757                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 758                 if (ret < 0) {
 759                         goto error;
 760                 }
 761                 if (ret > 0) {
 762                         BUG_ON(path->slots[0] == 0);
 763                         path->slots[0]--;
 764                 }
 765                 leaf = path->nodes[0];
 766                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 767                 found_type = btrfs_key_type(&found_key);
 768
 769                 if (found_key.objectid != inode->i_ino)
 770                         break;
 771
 772                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 773                     found_type != BTRFS_DIR_ITEM_KEY &&
 774                     found_type != BTRFS_DIR_INDEX_KEY &&
 775                     found_type != BTRFS_EXTENT_DATA_KEY)
 776                         break;
 777
 778                 item_end = found_key.offset;
 779                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 780                         fi = btrfs_item_ptr(leaf, path->slots[0],
 781                                             struct btrfs_file_extent_item);
 782                         extent_type = btrfs_file_extent_type(leaf, fi);
 783                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 784                                 item_end +=
 785                                     btrfs_file_extent_num_bytes(leaf, fi);
 786                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 787                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 788                                                                 path->slots[0]);
 789                                 item_end += btrfs_file_extent_inline_len(leaf,
 790                                                                          item);
 791                         }
 792                         item_end--;
 793                 }
 794                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 795                         ret = btrfs_csum_truncate(trans, root, path,
 796                                                   inode->i_size);
 797                         BUG_ON(ret);
 798                 }
 799                 if (item_end < inode->i_size) {
 800                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 801                                 found_type = BTRFS_INODE_ITEM_KEY;
 802                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 803                                 found_type = BTRFS_CSUM_ITEM_KEY;
 804                         } else if (found_type) {
 805                                 found_type--;
 806                         } else {
 807                                 break;
 808                         }
 809                         btrfs_set_key_type(&key, found_type);
 810                         btrfs_release_path(root, path);
 811                         continue;
 812                 }
 813                 if (found_key.offset >= inode->i_size)
 814                         del_item = 1;
 815                 else
 816                         del_item = 0;
 817                 found_extent = 0;
 818
 819                 /* FIXME, shrink the extent if the ref count is only 1 */
 820                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 821                         goto delete;
 822
 823                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 824                         u64 num_dec;
 825                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 826                         if (!del_item) {
 827                                 u64 orig_num_bytes =
 828                                         btrfs_file_extent_num_bytes(leaf, fi);
 829                                 extent_num_bytes = inode->i_size -
 830                                         found_key.offset + root->sectorsize - 1;
 831                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 832                                                          extent_num_bytes);
 833                                 num_dec = (orig_num_bytes -
 834                                            extent_num_bytes) >> 9;
 835                                 if (extent_start != 0) {
 836                                         inode->i_blocks -= num_dec;
 837                                 }
 838                                 btrfs_mark_buffer_dirty(leaf);
 839                         } else {
 840                                 extent_num_bytes =
 841                                         btrfs_file_extent_disk_num_bytes(leaf,
 842                                                                          fi);
 843                                 /* FIXME blocksize != 4096 */
 844                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 845                                                                        fi) >> 9;
 846                                 if (extent_start != 0) {
 847                                         found_extent = 1;
 848                                         inode->i_blocks -= num_dec;
 849                                 }
 850                                 root_gen = btrfs_header_generation(leaf);
 851                                 root_owner = btrfs_header_owner(leaf);
 852                         }
 853                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 854                            !del_item) {
 855                         u32 newsize = inode->i_size - found_key.offset;
 856                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 857                         ret = btrfs_truncate_item(trans, root, path,
 858                                                   newsize, 1);
 859                         BUG_ON(ret);
 860                 }
 861 delete:
 862                 if (del_item) {
 863                         ret = btrfs_del_item(trans, root, path);
 864                         if (ret)
 865                                 goto error;
 866                 } else {
 867                         break;
 868                 }
 869                 btrfs_release_path(root, path);
 870                 if (found_extent) {
 871                         ret = btrfs_free_extent(trans, root, extent_start,
 872                                                 extent_num_bytes,
 873                                                 root_owner,
 874                                                 root_gen, inode->i_ino,
 875                                                 found_key.offset, 0);
 876                         BUG_ON(ret);
 877                 }
 878         }
 879         ret = 0;
 880 error:
 881         btrfs_release_path(root, path);
 882         btrfs_free_path(path);
 883         inode->i_sb->s_dirt = 1;
 884         return ret;
 885 }
 886
 887 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 888                               size_t zero_start)
 889 {
 890         char *kaddr;
 891         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 892         struct btrfs_root *root = BTRFS_I(inode)->root;
 893         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 894         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 895         u64 existing_delalloc;
 896         u64 delalloc_start;
 897         int ret = 0;
 898
 899         WARN_ON(!PageLocked(page));
 900         set_page_extent_mapped(page);
 901
 902         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 903         delalloc_start = page_start;
 904         existing_delalloc = count_range_bits(&BTRFS_I(inode)->io_tree,
 905                                              &delalloc_start, page_end,
 906                                              PAGE_CACHE_SIZE, EXTENT_DELALLOC);
 907         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
 908                             page_end, GFP_NOFS);
 909
 910         spin_lock(&root->fs_info->delalloc_lock);
 911         root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE - existing_delalloc;
 912         spin_unlock(&root->fs_info->delalloc_lock);
 913
 914         if (zero_start != PAGE_CACHE_SIZE) {
 915                 kaddr = kmap(page);
 916                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 917                 flush_dcache_page(page);
 918                 kunmap(page);
 919         }
 920         set_page_dirty(page);
 921         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 922
 923         return ret;
 924 }
 925
 926 /*
 927  * taken from block_truncate_page, but does cow as it zeros out
 928  * any bytes left in the last page in the file.
 929  */
 930 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 931 {
 932         struct inode *inode = mapping->host;
 933         struct btrfs_root *root = BTRFS_I(inode)->root;
 934         u32 blocksize = root->sectorsize;
 935         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 936         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 937         struct page *page;
 938         int ret = 0;
 939         u64 page_start;
 940
 941         if ((offset & (blocksize - 1)) == 0)
 942                 goto out;
 943
 944         ret = -ENOMEM;
 945         page = grab_cache_page(mapping, index);
 946         if (!page)
 947                 goto out;
 948         if (!PageUptodate(page)) {
 949                 ret = btrfs_readpage(NULL, page);
 950                 lock_page(page);
 951                 if (!PageUptodate(page)) {
 952                         ret = -EIO;
 953                         goto out;
 954                 }
 955         }
 956         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 957
 958         ret = btrfs_cow_one_page(inode, page, offset);
 959
 960         unlock_page(page);
 961         page_cache_release(page);
 962 out:
 963         return ret;
 964 }
 965
 966 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 967 {
 968         struct inode *inode = dentry->d_inode;
 969         int err;
 970
 971         err = inode_change_ok(inode, attr);
 972         if (err)
 973                 return err;
 974
 975         if (S_ISREG(inode->i_mode) &&
 976             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 977                 struct btrfs_trans_handle *trans;
 978                 struct btrfs_root *root = BTRFS_I(inode)->root;
 979                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 980
 981                 u64 mask = root->sectorsize - 1;
 982                 u64 pos = (inode->i_size + mask) & ~mask;
 983                 u64 block_end = attr->ia_size | mask;
 984                 u64 hole_start;
 985                 u64 hole_size;
 986                 u64 alloc_hint = 0;
 987
 988                 if (attr->ia_size <= pos)
 989                         goto out;
 990
 991                 if (pos != inode->i_size)
 992                         hole_start = pos + root->sectorsize;
 993                 else
 994                         hole_start = pos;
 995
 996                 mutex_lock(&root->fs_info->fs_mutex);
 997                 err = btrfs_check_free_space(root, 1, 0);
 998                 mutex_unlock(&root->fs_info->fs_mutex);
 999                 if (err)
1000                         goto fail;
1001
1002                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1003
1004                 lock_extent(io_tree, pos, block_end, GFP_NOFS);
1005                 hole_size = block_end - hole_start;
1006
1007                 mutex_lock(&root->fs_info->fs_mutex);
1008                 trans = btrfs_start_transaction(root, 1);
1009                 btrfs_set_trans_block_group(trans, inode);
1010                 err = btrfs_drop_extents(trans, root, inode,
1011                                          pos, block_end, pos,
1012                                          &alloc_hint);
1013
1014                 if (alloc_hint != EXTENT_MAP_INLINE) {
1015                         err = btrfs_insert_file_extent(trans, root,
1016                                                        inode->i_ino,
1017                                                        hole_start, 0, 0,
1018                                                        hole_size);
1019                         btrfs_drop_extent_cache(inode, hole_start,
1020                                                 hole_size - 1);
1021                         btrfs_check_file(root, inode);
1022                 }
1023                 btrfs_end_transaction(trans, root);
1024                 mutex_unlock(&root->fs_info->fs_mutex);
1025                 unlock_extent(io_tree, pos, block_end, GFP_NOFS);
1026                 if (err)
1027                         return err;
1028         }
1029 out:
1030         err = inode_setattr(inode, attr);
1031 fail:
1032         return err;
1033 }
1034
1035 void btrfs_put_inode(struct inode *inode)
1036 {
1037         int ret;
1038
1039         if (!BTRFS_I(inode)->ordered_trans) {
1040                 return;
1041         }
1042
1043         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1044             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1045                 return;
1046
1047         ret = btrfs_del_ordered_inode(inode);
1048         if (ret == 1) {
1049                 atomic_dec(&inode->i_count);
1050         }
1051 }
1052
1053 void btrfs_delete_inode(struct inode *inode)
1054 {
1055         struct btrfs_trans_handle *trans;
1056         struct btrfs_root *root = BTRFS_I(inode)->root;
1057         unsigned long nr;
1058         int ret;
1059
1060         truncate_inode_pages(&inode->i_data, 0);
1061         if (is_bad_inode(inode)) {
1062                 goto no_delete;
1063         }
1064
1065         inode->i_size = 0;
1066         mutex_lock(&root->fs_info->fs_mutex);
1067         trans = btrfs_start_transaction(root, 1);
1068
1069         btrfs_set_trans_block_group(trans, inode);
1070         ret = btrfs_truncate_in_trans(trans, root, inode);
1071         if (ret)
1072                 goto no_delete_lock;
1073         ret = btrfs_delete_xattrs(trans, root, inode);
1074         if (ret)
1075                 goto no_delete_lock;
1076         ret = btrfs_free_inode(trans, root, inode);
1077         if (ret)
1078                 goto no_delete_lock;
1079         nr = trans->blocks_used;
1080
1081         btrfs_end_transaction(trans, root);
1082         mutex_unlock(&root->fs_info->fs_mutex);
1083         btrfs_btree_balance_dirty(root, nr);
1084         btrfs_throttle(root);
1085         return;
1086
1087 no_delete_lock:
1088         nr = trans->blocks_used;
1089         btrfs_end_transaction(trans, root);
1090         mutex_unlock(&root->fs_info->fs_mutex);
1091         btrfs_btree_balance_dirty(root, nr);
1092         btrfs_throttle(root);
1093 no_delete:
1094         clear_inode(inode);
1095 }
1096
1097 /*
1098  * this returns the key found in the dir entry in the location pointer.
1099  * If no dir entries were found, location->objectid is 0.
1100  */
1101 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1102                                struct btrfs_key *location)
1103 {
1104         const char *name = dentry->d_name.name;
1105         int namelen = dentry->d_name.len;
1106         struct btrfs_dir_item *di;
1107         struct btrfs_path *path;
1108         struct btrfs_root *root = BTRFS_I(dir)->root;
1109         int ret = 0;
1110
1111         if (namelen == 1 && strcmp(name, ".") == 0) {
1112                 location->objectid = dir->i_ino;
1113                 location->type = BTRFS_INODE_ITEM_KEY;
1114                 location->offset = 0;
1115                 return 0;
1116         }
1117         path = btrfs_alloc_path();
1118         BUG_ON(!path);
1119
1120         if (namelen == 2 && strcmp(name, "..") == 0) {
1121                 struct btrfs_key key;
1122                 struct extent_buffer *leaf;
1123                 u32 nritems;
1124                 int slot;
1125
1126                 key.objectid = dir->i_ino;
1127                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1128                 key.offset = 0;
1129                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1130                 BUG_ON(ret == 0);
1131                 ret = 0;
1132
1133                 leaf = path->nodes[0];
1134                 slot = path->slots[0];
1135                 nritems = btrfs_header_nritems(leaf);
1136                 if (slot >= nritems)
1137                         goto out_err;
1138
1139                 btrfs_item_key_to_cpu(leaf, &key, slot);
1140                 if (key.objectid != dir->i_ino ||
1141                     key.type != BTRFS_INODE_REF_KEY) {
1142                         goto out_err;
1143                 }
1144                 location->objectid = key.offset;
1145                 location->type = BTRFS_INODE_ITEM_KEY;
1146                 location->offset = 0;
1147                 goto out;
1148         }
1149
1150         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1151                                     namelen, 0);
1152         if (IS_ERR(di))
1153                 ret = PTR_ERR(di);
1154         if (!di || IS_ERR(di)) {
1155                 goto out_err;
1156         }
1157         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1158 out:
1159         btrfs_free_path(path);
1160         return ret;
1161 out_err:
1162         location->objectid = 0;
1163         goto out;
1164 }
1165
1166 /*
1167  * when we hit a tree root in a directory, the btrfs part of the inode
1168  * needs to be changed to reflect the root directory of the tree root.  This
1169  * is kind of like crossing a mount point.
1170  */
1171 static int fixup_tree_root_location(struct btrfs_root *root,
1172                              struct btrfs_key *location,
1173                              struct btrfs_root **sub_root,
1174                              struct dentry *dentry)
1175 {
1176         struct btrfs_path *path;
1177         struct btrfs_root_item *ri;
1178
1179         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1180                 return 0;
1181         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1182                 return 0;
1183
1184         path = btrfs_alloc_path();
1185         BUG_ON(!path);
1186         mutex_lock(&root->fs_info->fs_mutex);
1187
1188         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1189                                         dentry->d_name.name,
1190                                         dentry->d_name.len);
1191         if (IS_ERR(*sub_root))
1192                 return PTR_ERR(*sub_root);
1193
1194         ri = &(*sub_root)->root_item;
1195         location->objectid = btrfs_root_dirid(ri);
1196         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1197         location->offset = 0;
1198
1199         btrfs_free_path(path);
1200         mutex_unlock(&root->fs_info->fs_mutex);
1201         return 0;
1202 }
1203
1204 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1205 {
1206         struct btrfs_iget_args *args = p;
1207         inode->i_ino = args->ino;
1208         BTRFS_I(inode)->root = args->root;
1209         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1210         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1211                              inode->i_mapping, GFP_NOFS);
1212         return 0;
1213 }
1214
1215 static int btrfs_find_actor(struct inode *inode, void *opaque)
1216 {
1217         struct btrfs_iget_args *args = opaque;
1218         return (args->ino == inode->i_ino &&
1219                 args->root == BTRFS_I(inode)->root);
1220 }
1221
1222 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1223                             u64 root_objectid)
1224 {
1225         struct btrfs_iget_args args;
1226         args.ino = objectid;
1227         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1228
1229         if (!args.root)
1230                 return NULL;
1231
1232         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1233 }
1234
1235 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1236                                 struct btrfs_root *root)
1237 {
1238         struct inode *inode;
1239         struct btrfs_iget_args args;
1240         args.ino = objectid;
1241         args.root = root;
1242
1243         inode = iget5_locked(s, objectid, btrfs_find_actor,
1244                              btrfs_init_locked_inode,
1245                              (void *)&args);
1246         return inode;
1247 }
1248
1249 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1250                                    struct nameidata *nd)
1251 {
1252         struct inode * inode;
1253         struct btrfs_inode *bi = BTRFS_I(dir);
1254         struct btrfs_root *root = bi->root;
1255         struct btrfs_root *sub_root = root;
1256         struct btrfs_key location;
1257         int ret;
1258
1259         if (dentry->d_name.len > BTRFS_NAME_LEN)
1260                 return ERR_PTR(-ENAMETOOLONG);
1261
1262         mutex_lock(&root->fs_info->fs_mutex);
1263         ret = btrfs_inode_by_name(dir, dentry, &location);
1264         mutex_unlock(&root->fs_info->fs_mutex);
1265
1266         if (ret < 0)
1267                 return ERR_PTR(ret);
1268
1269         inode = NULL;
1270         if (location.objectid) {
1271                 ret = fixup_tree_root_location(root, &location, &sub_root,
1272                                                 dentry);
1273                 if (ret < 0)
1274                         return ERR_PTR(ret);
1275                 if (ret > 0)
1276                         return ERR_PTR(-ENOENT);
1277                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1278                                           sub_root);
1279                 if (!inode)
1280                         return ERR_PTR(-EACCES);
1281                 if (inode->i_state & I_NEW) {
1282                         /* the inode and parent dir are two different roots */
1283                         if (sub_root != root) {
1284                                 igrab(inode);
1285                                 sub_root->inode = inode;
1286                         }
1287                         BTRFS_I(inode)->root = sub_root;
1288                         memcpy(&BTRFS_I(inode)->location, &location,
1289                                sizeof(location));
1290                         btrfs_read_locked_inode(inode);
1291                         unlock_new_inode(inode);
1292                 }
1293         }
1294         return d_splice_alias(inode, dentry);
1295 }
1296
1297 static unsigned char btrfs_filetype_table[] = {
1298         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1299 };
1300
1301 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1302 {
1303         struct inode *inode = filp->f_dentry->d_inode;
1304         struct btrfs_root *root = BTRFS_I(inode)->root;
1305         struct btrfs_item *item;
1306         struct btrfs_dir_item *di;
1307         struct btrfs_key key;
1308         struct btrfs_key found_key;
1309         struct btrfs_path *path;
1310         int ret;
1311         u32 nritems;
1312         struct extent_buffer *leaf;
1313         int slot;
1314         int advance;
1315         unsigned char d_type;
1316         int over = 0;
1317         u32 di_cur;
1318         u32 di_total;
1319         u32 di_len;
1320         int key_type = BTRFS_DIR_INDEX_KEY;
1321         char tmp_name[32];
1322         char *name_ptr;
1323         int name_len;
1324
1325         /* FIXME, use a real flag for deciding about the key type */
1326         if (root->fs_info->tree_root == root)
1327                 key_type = BTRFS_DIR_ITEM_KEY;
1328
1329         /* special case for "." */
1330         if (filp->f_pos == 0) {
1331                 over = filldir(dirent, ".", 1,
1332                                1, inode->i_ino,
1333                                DT_DIR);
1334                 if (over)
1335                         return 0;
1336                 filp->f_pos = 1;
1337         }
1338
1339         mutex_lock(&root->fs_info->fs_mutex);
1340         key.objectid = inode->i_ino;
1341         path = btrfs_alloc_path();
1342         path->reada = 2;
1343
1344         /* special case for .., just use the back ref */
1345         if (filp->f_pos == 1) {
1346                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1347                 key.offset = 0;
1348                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1349                 BUG_ON(ret == 0);
1350                 leaf = path->nodes[0];
1351                 slot = path->slots[0];
1352                 nritems = btrfs_header_nritems(leaf);
1353                 if (slot >= nritems) {
1354                         btrfs_release_path(root, path);
1355                         goto read_dir_items;
1356                 }
1357                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1358                 btrfs_release_path(root, path);
1359                 if (found_key.objectid != key.objectid ||
1360                     found_key.type != BTRFS_INODE_REF_KEY)
1361                         goto read_dir_items;
1362                 over = filldir(dirent, "..", 2,
1363                                2, found_key.offset, DT_DIR);
1364                 if (over)
1365                         goto nopos;
1366                 filp->f_pos = 2;
1367         }
1368
1369 read_dir_items:
1370         btrfs_set_key_type(&key, key_type);
1371         key.offset = filp->f_pos;
1372
1373         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1374         if (ret < 0)
1375                 goto err;
1376         advance = 0;
1377         while(1) {
1378                 leaf = path->nodes[0];
1379                 nritems = btrfs_header_nritems(leaf);
1380                 slot = path->slots[0];
1381                 if (advance || slot >= nritems) {
1382                         if (slot >= nritems -1) {
1383                                 ret = btrfs_next_leaf(root, path);
1384                                 if (ret)
1385                                         break;
1386                                 leaf = path->nodes[0];
1387                                 nritems = btrfs_header_nritems(leaf);
1388                                 slot = path->slots[0];
1389                         } else {
1390                                 slot++;
1391                                 path->slots[0]++;
1392                         }
1393                 }
1394                 advance = 1;
1395                 item = btrfs_item_nr(leaf, slot);
1396                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1397
1398                 if (found_key.objectid != key.objectid)
1399                         break;
1400                 if (btrfs_key_type(&found_key) != key_type)
1401                         break;
1402                 if (found_key.offset < filp->f_pos)
1403                         continue;
1404
1405                 filp->f_pos = found_key.offset;
1406                 advance = 1;
1407                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1408                 di_cur = 0;
1409                 di_total = btrfs_item_size(leaf, item);
1410                 while(di_cur < di_total) {
1411                         struct btrfs_key location;
1412
1413                         name_len = btrfs_dir_name_len(leaf, di);
1414                         if (name_len < 32) {
1415                                 name_ptr = tmp_name;
1416                         } else {
1417                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1418                                 BUG_ON(!name_ptr);
1419                         }
1420                         read_extent_buffer(leaf, name_ptr,
1421                                            (unsigned long)(di + 1), name_len);
1422
1423                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1424                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1425                         over = filldir(dirent, name_ptr, name_len,
1426                                        found_key.offset,
1427                                        location.objectid,
1428                                        d_type);
1429
1430                         if (name_ptr != tmp_name)
1431                                 kfree(name_ptr);
1432
1433                         if (over)
1434                                 goto nopos;
1435                         di_len = btrfs_dir_name_len(leaf, di) +
1436                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1437                         di_cur += di_len;
1438                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1439                 }
1440         }
1441         filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1442 nopos:
1443         ret = 0;
1444 err:
1445         btrfs_release_path(root, path);
1446         btrfs_free_path(path);
1447         mutex_unlock(&root->fs_info->fs_mutex);
1448         return ret;
1449 }
1450
1451 int btrfs_write_inode(struct inode *inode, int wait)
1452 {
1453         struct btrfs_root *root = BTRFS_I(inode)->root;
1454         struct btrfs_trans_handle *trans;
1455         int ret = 0;
1456
1457         if (wait) {
1458                 mutex_lock(&root->fs_info->fs_mutex);
1459                 trans = btrfs_start_transaction(root, 1);
1460                 btrfs_set_trans_block_group(trans, inode);
1461                 ret = btrfs_commit_transaction(trans, root);
1462                 mutex_unlock(&root->fs_info->fs_mutex);
1463         }
1464         return ret;
1465 }
1466
1467 /*
1468  * This is somewhat expensive, updating the tree every time the
1469  * inode changes.  But, it is most likely to find the inode in cache.
1470  * FIXME, needs more benchmarking...there are no reasons other than performance
1471  * to keep or drop this code.
1472  */
1473 void btrfs_dirty_inode(struct inode *inode)
1474 {
1475         struct btrfs_root *root = BTRFS_I(inode)->root;
1476         struct btrfs_trans_handle *trans;
1477
1478         mutex_lock(&root->fs_info->fs_mutex);
1479         trans = btrfs_start_transaction(root, 1);
1480         btrfs_set_trans_block_group(trans, inode);
1481         btrfs_update_inode(trans, root, inode);
1482         btrfs_end_transaction(trans, root);
1483         mutex_unlock(&root->fs_info->fs_mutex);
1484 }
1485
1486 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1487                                      struct btrfs_root *root,
1488                                      u64 objectid,
1489                                      struct btrfs_block_group_cache *group,
1490                                      int mode)
1491 {
1492         struct inode *inode;
1493         struct btrfs_inode_item *inode_item;
1494         struct btrfs_key *location;
1495         struct btrfs_path *path;
1496         int ret;
1497         int owner;
1498
1499         path = btrfs_alloc_path();
1500         BUG_ON(!path);
1501
1502         inode = new_inode(root->fs_info->sb);
1503         if (!inode)
1504                 return ERR_PTR(-ENOMEM);
1505
1506         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1507         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1508                              inode->i_mapping, GFP_NOFS);
1509         BTRFS_I(inode)->root = root;
1510
1511         if (mode & S_IFDIR)
1512                 owner = 0;
1513         else
1514                 owner = 1;
1515         group = btrfs_find_block_group(root, group, 0, 0, owner);
1516         BTRFS_I(inode)->block_group = group;
1517         BTRFS_I(inode)->flags = 0;
1518         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1519         if (ret)
1520                 goto fail;
1521
1522         inode->i_uid = current->fsuid;
1523         inode->i_gid = current->fsgid;
1524         inode->i_mode = mode;
1525         inode->i_ino = objectid;
1526         inode->i_blocks = 0;
1527         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1528         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1529                                   struct btrfs_inode_item);
1530         fill_inode_item(path->nodes[0], inode_item, inode);
1531         btrfs_mark_buffer_dirty(path->nodes[0]);
1532         btrfs_free_path(path);
1533
1534         location = &BTRFS_I(inode)->location;
1535         location->objectid = objectid;
1536         location->offset = 0;
1537         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1538
1539         insert_inode_hash(inode);
1540         return inode;
1541 fail:
1542         btrfs_free_path(path);
1543         return ERR_PTR(ret);
1544 }
1545
1546 static inline u8 btrfs_inode_type(struct inode *inode)
1547 {
1548         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1549 }
1550
1551 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1552                             struct dentry *dentry, struct inode *inode)
1553 {
1554         int ret;
1555         struct btrfs_key key;
1556         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1557         struct inode *parent_inode;
1558
1559         key.objectid = inode->i_ino;
1560         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1561         key.offset = 0;
1562
1563         ret = btrfs_insert_dir_item(trans, root,
1564                                     dentry->d_name.name, dentry->d_name.len,
1565                                     dentry->d_parent->d_inode->i_ino,
1566                                     &key, btrfs_inode_type(inode));
1567         if (ret == 0) {
1568                 ret = btrfs_insert_inode_ref(trans, root,
1569                                      dentry->d_name.name,
1570                                      dentry->d_name.len,
1571                                      inode->i_ino,
1572                                      dentry->d_parent->d_inode->i_ino);
1573                 parent_inode = dentry->d_parent->d_inode;
1574                 parent_inode->i_size += dentry->d_name.len * 2;
1575                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1576                 ret = btrfs_update_inode(trans, root,
1577                                          dentry->d_parent->d_inode);
1578         }
1579         return ret;
1580 }
1581
1582 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1583                             struct dentry *dentry, struct inode *inode)
1584 {
1585         int err = btrfs_add_link(trans, dentry, inode);
1586         if (!err) {
1587                 d_instantiate(dentry, inode);
1588                 return 0;
1589         }
1590         if (err > 0)
1591                 err = -EEXIST;
1592         return err;
1593 }
1594
1595 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1596                         int mode, dev_t rdev)
1597 {
1598         struct btrfs_trans_handle *trans;
1599         struct btrfs_root *root = BTRFS_I(dir)->root;
1600         struct inode *inode = NULL;
1601         int err;
1602         int drop_inode = 0;
1603         u64 objectid;
1604         unsigned long nr = 0;
1605
1606         if (!new_valid_dev(rdev))
1607                 return -EINVAL;
1608
1609         mutex_lock(&root->fs_info->fs_mutex);
1610         err = btrfs_check_free_space(root, 1, 0);
1611         if (err)
1612                 goto fail;
1613
1614         trans = btrfs_start_transaction(root, 1);
1615         btrfs_set_trans_block_group(trans, dir);
1616
1617         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1618         if (err) {
1619                 err = -ENOSPC;
1620                 goto out_unlock;
1621         }
1622
1623         inode = btrfs_new_inode(trans, root, objectid,
1624                                 BTRFS_I(dir)->block_group, mode);
1625         err = PTR_ERR(inode);
1626         if (IS_ERR(inode))
1627                 goto out_unlock;
1628
1629         btrfs_set_trans_block_group(trans, inode);
1630         err = btrfs_add_nondir(trans, dentry, inode);
1631         if (err)
1632                 drop_inode = 1;
1633         else {
1634                 inode->i_op = &btrfs_special_inode_operations;
1635                 init_special_inode(inode, inode->i_mode, rdev);
1636                 btrfs_update_inode(trans, root, inode);
1637         }
1638         dir->i_sb->s_dirt = 1;
1639         btrfs_update_inode_block_group(trans, inode);
1640         btrfs_update_inode_block_group(trans, dir);
1641 out_unlock:
1642         nr = trans->blocks_used;
1643         btrfs_end_transaction(trans, root);
1644 fail:
1645         mutex_unlock(&root->fs_info->fs_mutex);
1646
1647         if (drop_inode) {
1648                 inode_dec_link_count(inode);
1649                 iput(inode);
1650         }
1651         btrfs_btree_balance_dirty(root, nr);
1652         btrfs_throttle(root);
1653         return err;
1654 }
1655
1656 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1657                         int mode, struct nameidata *nd)
1658 {
1659         struct btrfs_trans_handle *trans;
1660         struct btrfs_root *root = BTRFS_I(dir)->root;
1661         struct inode *inode = NULL;
1662         int err;
1663         int drop_inode = 0;
1664         unsigned long nr = 0;
1665         u64 objectid;
1666
1667         mutex_lock(&root->fs_info->fs_mutex);
1668         err = btrfs_check_free_space(root, 1, 0);
1669         if (err)
1670                 goto fail;
1671         trans = btrfs_start_transaction(root, 1);
1672         btrfs_set_trans_block_group(trans, dir);
1673
1674         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1675         if (err) {
1676                 err = -ENOSPC;
1677                 goto out_unlock;
1678         }
1679
1680         inode = btrfs_new_inode(trans, root, objectid,
1681                                 BTRFS_I(dir)->block_group, mode);
1682         err = PTR_ERR(inode);
1683         if (IS_ERR(inode))
1684                 goto out_unlock;
1685
1686         btrfs_set_trans_block_group(trans, inode);
1687         err = btrfs_add_nondir(trans, dentry, inode);
1688         if (err)
1689                 drop_inode = 1;
1690         else {
1691                 inode->i_mapping->a_ops = &btrfs_aops;
1692                 inode->i_fop = &btrfs_file_operations;
1693                 inode->i_op = &btrfs_file_inode_operations;
1694                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1695                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1696                                      inode->i_mapping, GFP_NOFS);
1697                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1698         }
1699         dir->i_sb->s_dirt = 1;
1700         btrfs_update_inode_block_group(trans, inode);
1701         btrfs_update_inode_block_group(trans, dir);
1702 out_unlock:
1703         nr = trans->blocks_used;
1704         btrfs_end_transaction(trans, root);
1705 fail:
1706         mutex_unlock(&root->fs_info->fs_mutex);
1707
1708         if (drop_inode) {
1709                 inode_dec_link_count(inode);
1710                 iput(inode);
1711         }
1712         btrfs_btree_balance_dirty(root, nr);
1713         btrfs_throttle(root);
1714         return err;
1715 }
1716
1717 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1718                       struct dentry *dentry)
1719 {
1720         struct btrfs_trans_handle *trans;
1721         struct btrfs_root *root = BTRFS_I(dir)->root;
1722         struct inode *inode = old_dentry->d_inode;
1723         unsigned long nr = 0;
1724         int err;
1725         int drop_inode = 0;
1726
1727         if (inode->i_nlink == 0)
1728                 return -ENOENT;
1729
1730 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1731         inode->i_nlink++;
1732 #else
1733         inc_nlink(inode);
1734 #endif
1735         mutex_lock(&root->fs_info->fs_mutex);
1736         err = btrfs_check_free_space(root, 1, 0);
1737         if (err)
1738                 goto fail;
1739         trans = btrfs_start_transaction(root, 1);
1740
1741         btrfs_set_trans_block_group(trans, dir);
1742         atomic_inc(&inode->i_count);
1743         err = btrfs_add_nondir(trans, dentry, inode);
1744
1745         if (err)
1746                 drop_inode = 1;
1747
1748         dir->i_sb->s_dirt = 1;
1749         btrfs_update_inode_block_group(trans, dir);
1750         err = btrfs_update_inode(trans, root, inode);
1751
1752         if (err)
1753                 drop_inode = 1;
1754
1755         nr = trans->blocks_used;
1756         btrfs_end_transaction(trans, root);
1757 fail:
1758         mutex_unlock(&root->fs_info->fs_mutex);
1759
1760         if (drop_inode) {
1761                 inode_dec_link_count(inode);
1762                 iput(inode);
1763         }
1764         btrfs_btree_balance_dirty(root, nr);
1765         btrfs_throttle(root);
1766         return err;
1767 }
1768
1769 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1770 {
1771         struct inode *inode;
1772         struct btrfs_trans_handle *trans;
1773         struct btrfs_root *root = BTRFS_I(dir)->root;
1774         int err = 0;
1775         int drop_on_err = 0;
1776         u64 objectid;
1777         unsigned long nr = 1;
1778
1779         mutex_lock(&root->fs_info->fs_mutex);
1780         err = btrfs_check_free_space(root, 1, 0);
1781         if (err)
1782                 goto out_unlock;
1783
1784         trans = btrfs_start_transaction(root, 1);
1785         btrfs_set_trans_block_group(trans, dir);
1786
1787         if (IS_ERR(trans)) {
1788                 err = PTR_ERR(trans);
1789                 goto out_unlock;
1790         }
1791
1792         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1793         if (err) {
1794                 err = -ENOSPC;
1795                 goto out_unlock;
1796         }
1797
1798         inode = btrfs_new_inode(trans, root, objectid,
1799                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1800         if (IS_ERR(inode)) {
1801                 err = PTR_ERR(inode);
1802                 goto out_fail;
1803         }
1804
1805         drop_on_err = 1;
1806         inode->i_op = &btrfs_dir_inode_operations;
1807         inode->i_fop = &btrfs_dir_file_operations;
1808         btrfs_set_trans_block_group(trans, inode);
1809
1810         inode->i_size = 0;
1811         err = btrfs_update_inode(trans, root, inode);
1812         if (err)
1813                 goto out_fail;
1814
1815         err = btrfs_add_link(trans, dentry, inode);
1816         if (err)
1817                 goto out_fail;
1818
1819         d_instantiate(dentry, inode);
1820         drop_on_err = 0;
1821         dir->i_sb->s_dirt = 1;
1822         btrfs_update_inode_block_group(trans, inode);
1823         btrfs_update_inode_block_group(trans, dir);
1824
1825 out_fail:
1826         nr = trans->blocks_used;
1827         btrfs_end_transaction(trans, root);
1828
1829 out_unlock:
1830         mutex_unlock(&root->fs_info->fs_mutex);
1831         if (drop_on_err)
1832                 iput(inode);
1833         btrfs_btree_balance_dirty(root, nr);
1834         btrfs_throttle(root);
1835         return err;
1836 }
1837
1838 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1839                                     size_t pg_offset, u64 start, u64 len,
1840                                     int create)
1841 {
1842         int ret;
1843         int err = 0;
1844         u64 bytenr;
1845         u64 extent_start = 0;
1846         u64 extent_end = 0;
1847         u64 objectid = inode->i_ino;
1848         u32 found_type;
1849         struct btrfs_path *path;
1850         struct btrfs_root *root = BTRFS_I(inode)->root;
1851         struct btrfs_file_extent_item *item;
1852         struct extent_buffer *leaf;
1853         struct btrfs_key found_key;
1854         struct extent_map *em = NULL;
1855         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1856         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1857         struct btrfs_trans_handle *trans = NULL;
1858
1859         path = btrfs_alloc_path();
1860         BUG_ON(!path);
1861         mutex_lock(&root->fs_info->fs_mutex);
1862
1863 again:
1864         spin_lock(&em_tree->lock);
1865         em = lookup_extent_mapping(em_tree, start, len);
1866         spin_unlock(&em_tree->lock);
1867
1868         if (em) {
1869                 if (em->start > start) {
1870                         printk("get_extent lookup [%Lu %Lu] em [%Lu %Lu]\n",
1871                                start, len, em->start, em->len);
1872                         WARN_ON(1);
1873                 }
1874                 if (em->block_start == EXTENT_MAP_INLINE && page)
1875                         free_extent_map(em);
1876                 else
1877                         goto out;
1878         }
1879         em = alloc_extent_map(GFP_NOFS);
1880         if (!em) {
1881                 err = -ENOMEM;
1882                 goto out;
1883         }
1884
1885         em->start = EXTENT_MAP_HOLE;
1886         em->len = (u64)-1;
1887         em->bdev = inode->i_sb->s_bdev;
1888         ret = btrfs_lookup_file_extent(trans, root, path,
1889                                        objectid, start, trans != NULL);
1890         if (ret < 0) {
1891                 err = ret;
1892                 goto out;
1893         }
1894
1895         if (ret != 0) {
1896                 if (path->slots[0] == 0)
1897                         goto not_found;
1898                 path->slots[0]--;
1899         }
1900
1901         leaf = path->nodes[0];
1902         item = btrfs_item_ptr(leaf, path->slots[0],
1903                               struct btrfs_file_extent_item);
1904         /* are we inside the extent that was found? */
1905         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1906         found_type = btrfs_key_type(&found_key);
1907         if (found_key.objectid != objectid ||
1908             found_type != BTRFS_EXTENT_DATA_KEY) {
1909                 goto not_found;
1910         }
1911
1912         found_type = btrfs_file_extent_type(leaf, item);
1913         extent_start = found_key.offset;
1914         if (found_type == BTRFS_FILE_EXTENT_REG) {
1915                 extent_end = extent_start +
1916                        btrfs_file_extent_num_bytes(leaf, item);
1917                 err = 0;
1918                 if (start < extent_start || start >= extent_end) {
1919                         em->start = start;
1920                         if (start < extent_start) {
1921                                 if (start + len <= extent_start)
1922                                         goto not_found;
1923                                 em->len = extent_end - extent_start;
1924                         } else {
1925                                 em->len = len;
1926                         }
1927                         goto not_found_em;
1928                 }
1929                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1930                 if (bytenr == 0) {
1931                         em->start = extent_start;
1932                         em->len = extent_end - extent_start;
1933                         em->block_start = EXTENT_MAP_HOLE;
1934                         goto insert;
1935                 }
1936                 bytenr += btrfs_file_extent_offset(leaf, item);
1937                 em->block_start = bytenr;
1938                 em->start = extent_start;
1939                 em->len = extent_end - extent_start;
1940                 goto insert;
1941         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1942                 u64 page_start;
1943                 unsigned long ptr;
1944                 char *map;
1945                 size_t size;
1946                 size_t extent_offset;
1947                 size_t copy_size;
1948
1949                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1950                                                     path->slots[0]));
1951                 extent_end = (extent_start + size + root->sectorsize - 1) &
1952                         ~((u64)root->sectorsize - 1);
1953                 if (start < extent_start || start >= extent_end) {
1954                         em->start = start;
1955                         if (start < extent_start) {
1956                                 if (start + len <= extent_start)
1957                                         goto not_found;
1958                                 em->len = extent_end - extent_start;
1959                         } else {
1960                                 em->len = len;
1961                         }
1962                         goto not_found_em;
1963                 }
1964                 em->block_start = EXTENT_MAP_INLINE;
1965
1966                 if (!page) {
1967                         em->start = extent_start;
1968                         em->len = size;
1969                         goto out;
1970                 }
1971
1972                 page_start = page_offset(page) + pg_offset;
1973                 extent_offset = page_start - extent_start;
1974                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
1975                                 size - extent_offset);
1976                 em->start = extent_start + extent_offset;
1977                 em->len = (copy_size + root->sectorsize - 1) &
1978                         ~((u64)root->sectorsize - 1);
1979                 map = kmap(page);
1980                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1981                 if (create == 0 && !PageUptodate(page)) {
1982                         read_extent_buffer(leaf, map + pg_offset, ptr,
1983                                            copy_size);
1984                         flush_dcache_page(page);
1985                 } else if (create && PageUptodate(page)) {
1986                         if (!trans) {
1987                                 kunmap(page);
1988                                 free_extent_map(em);
1989                                 em = NULL;
1990                                 btrfs_release_path(root, path);
1991                                 trans = btrfs_start_transaction(root, 1);
1992                                 goto again;
1993                         }
1994                         write_extent_buffer(leaf, map + pg_offset, ptr,
1995                                             copy_size);
1996                         btrfs_mark_buffer_dirty(leaf);
1997                 }
1998                 kunmap(page);
1999                 set_extent_uptodate(io_tree, em->start,
2000                                     extent_map_end(em) - 1, GFP_NOFS);
2001                 goto insert;
2002         } else {
2003                 printk("unkknown found_type %d\n", found_type);
2004                 WARN_ON(1);
2005         }
2006 not_found:
2007         em->start = start;
2008         em->len = len;
2009 not_found_em:
2010         em->block_start = EXTENT_MAP_HOLE;
2011 insert:
2012         btrfs_release_path(root, path);
2013         if (em->start > start || extent_map_end(em) <= start) {
2014                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2015                 err = -EIO;
2016                 goto out;
2017         }
2018
2019         err = 0;
2020         spin_lock(&em_tree->lock);
2021         ret = add_extent_mapping(em_tree, em);
2022         if (ret == -EEXIST) {
2023                 free_extent_map(em);
2024                 em = lookup_extent_mapping(em_tree, start, len);
2025                 if (!em) {
2026                         err = -EIO;
2027                         printk("failing to insert %Lu %Lu\n", start, len);
2028                 }
2029         }
2030         spin_unlock(&em_tree->lock);
2031 out:
2032         btrfs_free_path(path);
2033         if (trans) {
2034                 ret = btrfs_end_transaction(trans, root);
2035                 if (!err)
2036                         err = ret;
2037         }
2038         mutex_unlock(&root->fs_info->fs_mutex);
2039         if (err) {
2040                 free_extent_map(em);
2041                 WARN_ON(1);
2042                 return ERR_PTR(err);
2043         }
2044         return em;
2045 }
2046
2047 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2048 {
2049         return extent_bmap(mapping, iblock, btrfs_get_extent);
2050 }
2051
2052 int btrfs_readpage(struct file *file, struct page *page)
2053 {
2054         struct extent_io_tree *tree;
2055         tree = &BTRFS_I(page->mapping->host)->io_tree;
2056         return extent_read_full_page(tree, page, btrfs_get_extent);
2057 }
2058
2059 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2060 {
2061         struct extent_io_tree *tree;
2062
2063
2064         if (current->flags & PF_MEMALLOC) {
2065                 redirty_page_for_writepage(wbc, page);
2066                 unlock_page(page);
2067                 return 0;
2068         }
2069         tree = &BTRFS_I(page->mapping->host)->io_tree;
2070         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2071 }
2072
2073 static int btrfs_writepages(struct address_space *mapping,
2074                             struct writeback_control *wbc)
2075 {
2076         struct extent_io_tree *tree;
2077         tree = &BTRFS_I(mapping->host)->io_tree;
2078         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2079 }
2080
2081 static int
2082 btrfs_readpages(struct file *file, struct address_space *mapping,
2083                 struct list_head *pages, unsigned nr_pages)
2084 {
2085         struct extent_io_tree *tree;
2086         tree = &BTRFS_I(mapping->host)->io_tree;
2087         return extent_readpages(tree, mapping, pages, nr_pages,
2088                                 btrfs_get_extent);
2089 }
2090
2091 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2092 {
2093         struct extent_io_tree *tree;
2094         struct extent_map_tree *map;
2095         int ret;
2096
2097         tree = &BTRFS_I(page->mapping->host)->io_tree;
2098         map = &BTRFS_I(page->mapping->host)->extent_tree;
2099         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2100         if (ret == 1) {
2101                 ClearPagePrivate(page);
2102                 set_page_private(page, 0);
2103                 page_cache_release(page);
2104         }
2105         return ret;
2106 }
2107
2108 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2109 {
2110         struct extent_io_tree *tree;
2111
2112         tree = &BTRFS_I(page->mapping->host)->io_tree;
2113         extent_invalidatepage(tree, page, offset);
2114         btrfs_releasepage(page, GFP_NOFS);
2115 }
2116
2117 /*
2118  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2119  * called from a page fault handler when a page is first dirtied. Hence we must
2120  * be careful to check for EOF conditions here. We set the page up correctly
2121  * for a written page which means we get ENOSPC checking when writing into
2122  * holes and correct delalloc and unwritten extent mapping on filesystems that
2123  * support these features.
2124  *
2125  * We are not allowed to take the i_mutex here so we have to play games to
2126  * protect against truncate races as the page could now be beyond EOF.  Because
2127  * vmtruncate() writes the inode size before removing pages, once we have the
2128  * page lock we can determine safely if the page is beyond EOF. If it is not
2129  * beyond EOF, then the page is guaranteed safe against truncation until we
2130  * unlock the page.
2131  */
2132 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2133 {
2134         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2135         struct btrfs_root *root = BTRFS_I(inode)->root;
2136         unsigned long end;
2137         loff_t size;
2138         int ret;
2139         u64 page_start;
2140
2141         mutex_lock(&root->fs_info->fs_mutex);
2142         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2143         mutex_unlock(&root->fs_info->fs_mutex);
2144         if (ret)
2145                 goto out;
2146
2147         ret = -EINVAL;
2148
2149         lock_page(page);
2150         wait_on_page_writeback(page);
2151         size = i_size_read(inode);
2152         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2153
2154         if ((page->mapping != inode->i_mapping) ||
2155             (page_start > size)) {
2156                 /* page got truncated out from underneath us */
2157                 goto out_unlock;
2158         }
2159
2160         /* page is wholly or partially inside EOF */
2161         if (page_start + PAGE_CACHE_SIZE > size)
2162                 end = size & ~PAGE_CACHE_MASK;
2163         else
2164                 end = PAGE_CACHE_SIZE;
2165
2166         ret = btrfs_cow_one_page(inode, page, end);
2167
2168 out_unlock:
2169         unlock_page(page);
2170 out:
2171         return ret;
2172 }
2173
2174 static void btrfs_truncate(struct inode *inode)
2175 {
2176         struct btrfs_root *root = BTRFS_I(inode)->root;
2177         int ret;
2178         struct btrfs_trans_handle *trans;
2179         unsigned long nr;
2180
2181         if (!S_ISREG(inode->i_mode))
2182                 return;
2183         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2184                 return;
2185
2186         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2187
2188         mutex_lock(&root->fs_info->fs_mutex);
2189         trans = btrfs_start_transaction(root, 1);
2190         btrfs_set_trans_block_group(trans, inode);
2191
2192         /* FIXME, add redo link to tree so we don't leak on crash */
2193         ret = btrfs_truncate_in_trans(trans, root, inode);
2194         btrfs_update_inode(trans, root, inode);
2195         nr = trans->blocks_used;
2196
2197         ret = btrfs_end_transaction(trans, root);
2198         BUG_ON(ret);
2199         mutex_unlock(&root->fs_info->fs_mutex);
2200         btrfs_btree_balance_dirty(root, nr);
2201         btrfs_throttle(root);
2202 }
2203
2204 static int noinline create_subvol(struct btrfs_root *root, char *name,
2205                                   int namelen)
2206 {
2207         struct btrfs_trans_handle *trans;
2208         struct btrfs_key key;
2209         struct btrfs_root_item root_item;
2210         struct btrfs_inode_item *inode_item;
2211         struct extent_buffer *leaf;
2212         struct btrfs_root *new_root = root;
2213         struct inode *inode;
2214         struct inode *dir;
2215         int ret;
2216         int err;
2217         u64 objectid;
2218         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2219         unsigned long nr = 1;
2220
2221         mutex_lock(&root->fs_info->fs_mutex);
2222         ret = btrfs_check_free_space(root, 1, 0);
2223         if (ret)
2224                 goto fail_commit;
2225
2226         trans = btrfs_start_transaction(root, 1);
2227         BUG_ON(!trans);
2228
2229         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2230                                        0, &objectid);
2231         if (ret)
2232                 goto fail;
2233
2234         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2235                                         objectid, trans->transid, 0, 0,
2236                                         0, 0);
2237         if (IS_ERR(leaf))
2238                 return PTR_ERR(leaf);
2239
2240         btrfs_set_header_nritems(leaf, 0);
2241         btrfs_set_header_level(leaf, 0);
2242         btrfs_set_header_bytenr(leaf, leaf->start);
2243         btrfs_set_header_generation(leaf, trans->transid);
2244         btrfs_set_header_owner(leaf, objectid);
2245
2246         write_extent_buffer(leaf, root->fs_info->fsid,
2247                             (unsigned long)btrfs_header_fsid(leaf),
2248                             BTRFS_FSID_SIZE);
2249         btrfs_mark_buffer_dirty(leaf);
2250
2251         inode_item = &root_item.inode;
2252         memset(inode_item, 0, sizeof(*inode_item));
2253         inode_item->generation = cpu_to_le64(1);
2254         inode_item->size = cpu_to_le64(3);
2255         inode_item->nlink = cpu_to_le32(1);
2256         inode_item->nblocks = cpu_to_le64(1);
2257         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2258
2259         btrfs_set_root_bytenr(&root_item, leaf->start);
2260         btrfs_set_root_level(&root_item, 0);
2261         btrfs_set_root_refs(&root_item, 1);
2262         btrfs_set_root_used(&root_item, 0);
2263
2264         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2265         root_item.drop_level = 0;
2266
2267         free_extent_buffer(leaf);
2268         leaf = NULL;
2269
2270         btrfs_set_root_dirid(&root_item, new_dirid);
2271
2272         key.objectid = objectid;
2273         key.offset = 1;
2274         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2275         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2276                                 &root_item);
2277         if (ret)
2278                 goto fail;
2279
2280         /*
2281          * insert the directory item
2282          */
2283         key.offset = (u64)-1;
2284         dir = root->fs_info->sb->s_root->d_inode;
2285         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2286                                     name, namelen, dir->i_ino, &key,
2287                                     BTRFS_FT_DIR);
2288         if (ret)
2289                 goto fail;
2290
2291         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2292                              name, namelen, objectid,
2293                              root->fs_info->sb->s_root->d_inode->i_ino);
2294         if (ret)
2295                 goto fail;
2296
2297         ret = btrfs_commit_transaction(trans, root);
2298         if (ret)
2299                 goto fail_commit;
2300
2301         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2302         BUG_ON(!new_root);
2303
2304         trans = btrfs_start_transaction(new_root, 1);
2305         BUG_ON(!trans);
2306
2307         inode = btrfs_new_inode(trans, new_root, new_dirid,
2308                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2309         if (IS_ERR(inode))
2310                 goto fail;
2311         inode->i_op = &btrfs_dir_inode_operations;
2312         inode->i_fop = &btrfs_dir_file_operations;
2313         new_root->inode = inode;
2314
2315         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2316                                      new_dirid);
2317         inode->i_nlink = 1;
2318         inode->i_size = 0;
2319         ret = btrfs_update_inode(trans, new_root, inode);
2320         if (ret)
2321                 goto fail;
2322 fail:
2323         nr = trans->blocks_used;
2324         err = btrfs_commit_transaction(trans, new_root);
2325         if (err && !ret)
2326                 ret = err;
2327 fail_commit:
2328         mutex_unlock(&root->fs_info->fs_mutex);
2329         btrfs_btree_balance_dirty(root, nr);
2330         btrfs_throttle(root);
2331         return ret;
2332 }
2333
2334 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2335 {
2336         struct btrfs_pending_snapshot *pending_snapshot;
2337         struct btrfs_trans_handle *trans;
2338         int ret;
2339         int err;
2340         unsigned long nr = 0;
2341
2342         if (!root->ref_cows)
2343                 return -EINVAL;
2344
2345         mutex_lock(&root->fs_info->fs_mutex);
2346         ret = btrfs_check_free_space(root, 1, 0);
2347         if (ret)
2348                 goto fail_unlock;
2349
2350         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2351         if (!pending_snapshot) {
2352                 ret = -ENOMEM;
2353                 goto fail_unlock;
2354         }
2355         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2356         if (!pending_snapshot->name) {
2357                 ret = -ENOMEM;
2358                 kfree(pending_snapshot);
2359                 goto fail_unlock;
2360         }
2361         memcpy(pending_snapshot->name, name, namelen);
2362         pending_snapshot->name[namelen] = '\0';
2363         trans = btrfs_start_transaction(root, 1);
2364         BUG_ON(!trans);
2365         pending_snapshot->root = root;
2366         list_add(&pending_snapshot->list,
2367                  &trans->transaction->pending_snapshots);
2368         ret = btrfs_update_inode(trans, root, root->inode);
2369         err = btrfs_commit_transaction(trans, root);
2370
2371 fail_unlock:
2372         mutex_unlock(&root->fs_info->fs_mutex);
2373         btrfs_btree_balance_dirty(root, nr);
2374         btrfs_throttle(root);
2375         return ret;
2376 }
2377
2378 unsigned long btrfs_force_ra(struct address_space *mapping,
2379                               struct file_ra_state *ra, struct file *file,
2380                               pgoff_t offset, pgoff_t last_index)
2381 {
2382         pgoff_t req_size;
2383
2384 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2385         req_size = last_index - offset + 1;
2386         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2387         return offset;
2388 #else
2389         req_size = min(last_index - offset + 1, (pgoff_t)128);
2390         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2391         return offset + req_size;
2392 #endif
2393 }
2394
2395 int btrfs_defrag_file(struct file *file) {
2396         struct inode *inode = fdentry(file)->d_inode;
2397         struct btrfs_root *root = BTRFS_I(inode)->root;
2398         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2399         struct page *page;
2400         unsigned long last_index;
2401         unsigned long ra_index = 0;
2402         u64 page_start;
2403         u64 page_end;
2404         u64 delalloc_start;
2405         u64 existing_delalloc;
2406         unsigned long i;
2407         int ret;
2408
2409         mutex_lock(&root->fs_info->fs_mutex);
2410         ret = btrfs_check_free_space(root, inode->i_size, 0);
2411         mutex_unlock(&root->fs_info->fs_mutex);
2412         if (ret)
2413                 return -ENOSPC;
2414
2415         mutex_lock(&inode->i_mutex);
2416         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2417         for (i = 0; i <= last_index; i++) {
2418                 if (i == ra_index) {
2419                         ra_index = btrfs_force_ra(inode->i_mapping,
2420                                                   &file->f_ra,
2421                                                   file, ra_index, last_index);
2422                 }
2423                 page = grab_cache_page(inode->i_mapping, i);
2424                 if (!page)
2425                         goto out_unlock;
2426                 if (!PageUptodate(page)) {
2427                         btrfs_readpage(NULL, page);
2428                         lock_page(page);
2429                         if (!PageUptodate(page)) {
2430                                 unlock_page(page);
2431                                 page_cache_release(page);
2432                                 goto out_unlock;
2433                         }
2434                 }
2435                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2436                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2437
2438                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2439                 delalloc_start = page_start;
2440                 existing_delalloc =
2441                         count_range_bits(&BTRFS_I(inode)->io_tree,
2442                                          &delalloc_start, page_end,
2443                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2444                 set_extent_delalloc(io_tree, page_start,
2445                                     page_end, GFP_NOFS);
2446
2447                 spin_lock(&root->fs_info->delalloc_lock);
2448                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2449                                                  existing_delalloc;
2450                 spin_unlock(&root->fs_info->delalloc_lock);
2451
2452                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2453                 set_page_dirty(page);
2454                 unlock_page(page);
2455                 page_cache_release(page);
2456                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2457         }
2458
2459 out_unlock:
2460         mutex_unlock(&inode->i_mutex);
2461         return 0;
2462 }
2463
2464 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2465 {
2466         u64 new_size;
2467         u64 old_size;
2468         struct btrfs_ioctl_vol_args *vol_args;
2469         struct btrfs_trans_handle *trans;
2470         char *sizestr;
2471         int ret = 0;
2472         int namelen;
2473         int mod = 0;
2474
2475         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2476
2477         if (!vol_args)
2478                 return -ENOMEM;
2479
2480         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2481                 ret = -EFAULT;
2482                 goto out;
2483         }
2484         namelen = strlen(vol_args->name);
2485         if (namelen > BTRFS_VOL_NAME_MAX) {
2486                 ret = -EINVAL;
2487                 goto out;
2488         }
2489
2490         sizestr = vol_args->name;
2491         if (!strcmp(sizestr, "max"))
2492                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2493         else {
2494                 if (sizestr[0] == '-') {
2495                         mod = -1;
2496                         sizestr++;
2497                 } else if (sizestr[0] == '+') {
2498                         mod = 1;
2499                         sizestr++;
2500                 }
2501                 new_size = btrfs_parse_size(sizestr);
2502                 if (new_size == 0) {
2503                         ret = -EINVAL;
2504                         goto out;
2505                 }
2506         }
2507
2508         mutex_lock(&root->fs_info->fs_mutex);
2509         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2510
2511         if (mod < 0) {
2512                 if (new_size > old_size) {
2513                         ret = -EINVAL;
2514                         goto out_unlock;
2515                 }
2516                 new_size = old_size - new_size;
2517         } else if (mod > 0) {
2518                 new_size = old_size + new_size;
2519         }
2520
2521         if (new_size < 256 * 1024 * 1024) {
2522                 ret = -EINVAL;
2523                 goto out_unlock;
2524         }
2525         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2526                 ret = -EFBIG;
2527                 goto out_unlock;
2528         }
2529
2530         do_div(new_size, root->sectorsize);
2531         new_size *= root->sectorsize;
2532
2533 printk("new size is %Lu\n", new_size);
2534         if (new_size > old_size) {
2535                 trans = btrfs_start_transaction(root, 1);
2536                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2537                 btrfs_commit_transaction(trans, root);
2538         } else {
2539                 ret = btrfs_shrink_extent_tree(root, new_size);
2540         }
2541
2542 out_unlock:
2543         mutex_unlock(&root->fs_info->fs_mutex);
2544 out:
2545         kfree(vol_args);
2546         return ret;
2547 }
2548
2549 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2550                                             void __user *arg)
2551 {
2552         struct btrfs_ioctl_vol_args *vol_args;
2553         struct btrfs_dir_item *di;
2554         struct btrfs_path *path;
2555         u64 root_dirid;
2556         int namelen;
2557         int ret;
2558
2559         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2560
2561         if (!vol_args)
2562                 return -ENOMEM;
2563
2564         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2565                 ret = -EFAULT;
2566                 goto out;
2567         }
2568
2569         namelen = strlen(vol_args->name);
2570         if (namelen > BTRFS_VOL_NAME_MAX) {
2571                 ret = -EINVAL;
2572                 goto out;
2573         }
2574         if (strchr(vol_args->name, '/')) {
2575                 ret = -EINVAL;
2576                 goto out;
2577         }
2578
2579         path = btrfs_alloc_path();
2580         if (!path) {
2581                 ret = -ENOMEM;
2582                 goto out;
2583         }
2584
2585         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2586         mutex_lock(&root->fs_info->fs_mutex);
2587         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2588                             path, root_dirid,
2589                             vol_args->name, namelen, 0);
2590         mutex_unlock(&root->fs_info->fs_mutex);
2591         btrfs_free_path(path);
2592
2593         if (di && !IS_ERR(di)) {
2594                 ret = -EEXIST;
2595                 goto out;
2596         }
2597
2598         if (IS_ERR(di)) {
2599                 ret = PTR_ERR(di);
2600                 goto out;
2601         }
2602
2603         if (root == root->fs_info->tree_root)
2604                 ret = create_subvol(root, vol_args->name, namelen);
2605         else
2606                 ret = create_snapshot(root, vol_args->name, namelen);
2607 out:
2608         kfree(vol_args);
2609         return ret;
2610 }
2611
2612 static int btrfs_ioctl_defrag(struct file *file)
2613 {
2614         struct inode *inode = fdentry(file)->d_inode;
2615         struct btrfs_root *root = BTRFS_I(inode)->root;
2616
2617         switch (inode->i_mode & S_IFMT) {
2618         case S_IFDIR:
2619                 mutex_lock(&root->fs_info->fs_mutex);
2620                 btrfs_defrag_root(root, 0);
2621                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2622                 mutex_unlock(&root->fs_info->fs_mutex);
2623                 break;
2624         case S_IFREG:
2625                 btrfs_defrag_file(file);
2626                 break;
2627         }
2628
2629         return 0;
2630 }
2631
2632 long btrfs_ioctl(struct file *file, unsigned int
2633                 cmd, unsigned long arg)
2634 {
2635         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2636
2637         switch (cmd) {
2638         case BTRFS_IOC_SNAP_CREATE:
2639                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2640         case BTRFS_IOC_DEFRAG:
2641                 return btrfs_ioctl_defrag(file);
2642         case BTRFS_IOC_RESIZE:
2643                 return btrfs_ioctl_resize(root, (void __user *)arg);
2644         }
2645
2646         return -ENOTTY;
2647 }
2648
2649 /*
2650  * Called inside transaction, so use GFP_NOFS
2651  */
2652 struct inode *btrfs_alloc_inode(struct super_block *sb)
2653 {
2654         struct btrfs_inode *ei;
2655
2656         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2657         if (!ei)
2658                 return NULL;
2659         ei->last_trans = 0;
2660         ei->ordered_trans = 0;
2661         return &ei->vfs_inode;
2662 }
2663
2664 void btrfs_destroy_inode(struct inode *inode)
2665 {
2666         WARN_ON(!list_empty(&inode->i_dentry));
2667         WARN_ON(inode->i_data.nrpages);
2668
2669         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2670         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2671 }
2672
2673 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2674 static void init_once(struct kmem_cache * cachep, void *foo)
2675 #else
2676 static void init_once(void * foo, struct kmem_cache * cachep,
2677                       unsigned long flags)
2678 #endif
2679 {
2680         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2681
2682         inode_init_once(&ei->vfs_inode);
2683 }
2684
2685 void btrfs_destroy_cachep(void)
2686 {
2687         if (btrfs_inode_cachep)
2688                 kmem_cache_destroy(btrfs_inode_cachep);
2689         if (btrfs_trans_handle_cachep)
2690                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2691         if (btrfs_transaction_cachep)
2692                 kmem_cache_destroy(btrfs_transaction_cachep);
2693         if (btrfs_bit_radix_cachep)
2694                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2695         if (btrfs_path_cachep)
2696                 kmem_cache_destroy(btrfs_path_cachep);
2697 }
2698
2699 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2700                                        unsigned long extra_flags,
2701 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2702                                        void (*ctor)(struct kmem_cache *, void *)
2703 #else
2704                                        void (*ctor)(void *, struct kmem_cache *,
2705                                                     unsigned long)
2706 #endif
2707                                      )
2708 {
2709         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2710                                  SLAB_MEM_SPREAD | extra_flags), ctor
2711 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2712                                  ,NULL
2713 #endif
2714                                 );
2715 }
2716
2717 int btrfs_init_cachep(void)
2718 {
2719         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2720                                           sizeof(struct btrfs_inode),
2721                                           0, init_once);
2722         if (!btrfs_inode_cachep)
2723                 goto fail;
2724         btrfs_trans_handle_cachep =
2725                         btrfs_cache_create("btrfs_trans_handle_cache",
2726                                            sizeof(struct btrfs_trans_handle),
2727                                            0, NULL);
2728         if (!btrfs_trans_handle_cachep)
2729                 goto fail;
2730         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2731                                              sizeof(struct btrfs_transaction),
2732                                              0, NULL);
2733         if (!btrfs_transaction_cachep)
2734                 goto fail;
2735         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2736                                          sizeof(struct btrfs_path),
2737                                          0, NULL);
2738         if (!btrfs_path_cachep)
2739                 goto fail;
2740         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2741                                               SLAB_DESTROY_BY_RCU, NULL);
2742         if (!btrfs_bit_radix_cachep)
2743                 goto fail;
2744         return 0;
2745 fail:
2746         btrfs_destroy_cachep();
2747         return -ENOMEM;
2748 }
2749
2750 static int btrfs_getattr(struct vfsmount *mnt,
2751                          struct dentry *dentry, struct kstat *stat)
2752 {
2753         struct inode *inode = dentry->d_inode;
2754         generic_fillattr(inode, stat);
2755         stat->blksize = PAGE_CACHE_SIZE;
2756         return 0;
2757 }
2758
2759 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2760                            struct inode * new_dir,struct dentry *new_dentry)
2761 {
2762         struct btrfs_trans_handle *trans;
2763         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2764         struct inode *new_inode = new_dentry->d_inode;
2765         struct inode *old_inode = old_dentry->d_inode;
2766         struct timespec ctime = CURRENT_TIME;
2767         struct btrfs_path *path;
2768         int ret;
2769
2770         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2771             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2772                 return -ENOTEMPTY;
2773         }
2774
2775         mutex_lock(&root->fs_info->fs_mutex);
2776         ret = btrfs_check_free_space(root, 1, 0);
2777         if (ret)
2778                 goto out_unlock;
2779
2780         trans = btrfs_start_transaction(root, 1);
2781
2782         btrfs_set_trans_block_group(trans, new_dir);
2783         path = btrfs_alloc_path();
2784         if (!path) {
2785                 ret = -ENOMEM;
2786                 goto out_fail;
2787         }
2788
2789         old_dentry->d_inode->i_nlink++;
2790         old_dir->i_ctime = old_dir->i_mtime = ctime;
2791         new_dir->i_ctime = new_dir->i_mtime = ctime;
2792         old_inode->i_ctime = ctime;
2793
2794         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2795         if (ret)
2796                 goto out_fail;
2797
2798         if (new_inode) {
2799                 new_inode->i_ctime = CURRENT_TIME;
2800                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2801                 if (ret)
2802                         goto out_fail;
2803         }
2804         ret = btrfs_add_link(trans, new_dentry, old_inode);
2805         if (ret)
2806                 goto out_fail;
2807
2808 out_fail:
2809         btrfs_free_path(path);
2810         btrfs_end_transaction(trans, root);
2811 out_unlock:
2812         mutex_unlock(&root->fs_info->fs_mutex);
2813         return ret;
2814 }
2815
2816 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2817                          const char *symname)
2818 {
2819         struct btrfs_trans_handle *trans;
2820         struct btrfs_root *root = BTRFS_I(dir)->root;
2821         struct btrfs_path *path;
2822         struct btrfs_key key;
2823         struct inode *inode = NULL;
2824         int err;
2825         int drop_inode = 0;
2826         u64 objectid;
2827         int name_len;
2828         int datasize;
2829         unsigned long ptr;
2830         struct btrfs_file_extent_item *ei;
2831         struct extent_buffer *leaf;
2832         unsigned long nr = 0;
2833
2834         name_len = strlen(symname) + 1;
2835         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2836                 return -ENAMETOOLONG;
2837
2838         mutex_lock(&root->fs_info->fs_mutex);
2839         err = btrfs_check_free_space(root, 1, 0);
2840         if (err)
2841                 goto out_fail;
2842
2843         trans = btrfs_start_transaction(root, 1);
2844         btrfs_set_trans_block_group(trans, dir);
2845
2846         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2847         if (err) {
2848                 err = -ENOSPC;
2849                 goto out_unlock;
2850         }
2851
2852         inode = btrfs_new_inode(trans, root, objectid,
2853                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2854         err = PTR_ERR(inode);
2855         if (IS_ERR(inode))
2856                 goto out_unlock;
2857
2858         btrfs_set_trans_block_group(trans, inode);
2859         err = btrfs_add_nondir(trans, dentry, inode);
2860         if (err)
2861                 drop_inode = 1;
2862         else {
2863                 inode->i_mapping->a_ops = &btrfs_aops;
2864                 inode->i_fop = &btrfs_file_operations;
2865                 inode->i_op = &btrfs_file_inode_operations;
2866                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2867                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2868                                      inode->i_mapping, GFP_NOFS);
2869                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2870         }
2871         dir->i_sb->s_dirt = 1;
2872         btrfs_update_inode_block_group(trans, inode);
2873         btrfs_update_inode_block_group(trans, dir);
2874         if (drop_inode)
2875                 goto out_unlock;
2876
2877         path = btrfs_alloc_path();
2878         BUG_ON(!path);
2879         key.objectid = inode->i_ino;
2880         key.offset = 0;
2881         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2882         datasize = btrfs_file_extent_calc_inline_size(name_len);
2883         err = btrfs_insert_empty_item(trans, root, path, &key,
2884                                       datasize);
2885         if (err) {
2886                 drop_inode = 1;
2887                 goto out_unlock;
2888         }
2889         leaf = path->nodes[0];
2890         ei = btrfs_item_ptr(leaf, path->slots[0],
2891                             struct btrfs_file_extent_item);
2892         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2893         btrfs_set_file_extent_type(leaf, ei,
2894                                    BTRFS_FILE_EXTENT_INLINE);
2895         ptr = btrfs_file_extent_inline_start(ei);
2896         write_extent_buffer(leaf, symname, ptr, name_len);
2897         btrfs_mark_buffer_dirty(leaf);
2898         btrfs_free_path(path);
2899
2900         inode->i_op = &btrfs_symlink_inode_operations;
2901         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2902         inode->i_size = name_len - 1;
2903         err = btrfs_update_inode(trans, root, inode);
2904         if (err)
2905                 drop_inode = 1;
2906
2907 out_unlock:
2908         nr = trans->blocks_used;
2909         btrfs_end_transaction(trans, root);
2910 out_fail:
2911         mutex_unlock(&root->fs_info->fs_mutex);
2912         if (drop_inode) {
2913                 inode_dec_link_count(inode);
2914                 iput(inode);
2915         }
2916         btrfs_btree_balance_dirty(root, nr);
2917         btrfs_throttle(root);
2918         return err;
2919 }
2920 static int btrfs_permission(struct inode *inode, int mask,
2921                             struct nameidata *nd)
2922 {
2923         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2924                 return -EACCES;
2925         return generic_permission(inode, mask, NULL);
2926 }
2927
2928 static struct inode_operations btrfs_dir_inode_operations = {
2929         .lookup         = btrfs_lookup,
2930         .create         = btrfs_create,
2931         .unlink         = btrfs_unlink,
2932         .link           = btrfs_link,
2933         .mkdir          = btrfs_mkdir,
2934         .rmdir          = btrfs_rmdir,
2935         .rename         = btrfs_rename,
2936         .symlink        = btrfs_symlink,
2937         .setattr        = btrfs_setattr,
2938         .mknod          = btrfs_mknod,
2939         .setxattr       = generic_setxattr,
2940         .getxattr       = generic_getxattr,
2941         .listxattr      = btrfs_listxattr,
2942         .removexattr    = generic_removexattr,
2943         .permission     = btrfs_permission,
2944 };
2945 static struct inode_operations btrfs_dir_ro_inode_operations = {
2946         .lookup         = btrfs_lookup,
2947         .permission     = btrfs_permission,
2948 };
2949 static struct file_operations btrfs_dir_file_operations = {
2950         .llseek         = generic_file_llseek,
2951         .read           = generic_read_dir,
2952         .readdir        = btrfs_readdir,
2953         .unlocked_ioctl = btrfs_ioctl,
2954 #ifdef CONFIG_COMPAT
2955         .compat_ioctl   = btrfs_ioctl,
2956 #endif
2957 };
2958
2959 static struct extent_io_ops btrfs_extent_io_ops = {
2960         .fill_delalloc = run_delalloc_range,
2961         .writepage_io_hook = btrfs_writepage_io_hook,
2962         .readpage_io_hook = btrfs_readpage_io_hook,
2963         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2964 };
2965
2966 static struct address_space_operations btrfs_aops = {
2967         .readpage       = btrfs_readpage,
2968         .writepage      = btrfs_writepage,
2969         .writepages     = btrfs_writepages,
2970         .readpages      = btrfs_readpages,
2971         .sync_page      = block_sync_page,
2972         .bmap           = btrfs_bmap,
2973         .invalidatepage = btrfs_invalidatepage,
2974         .releasepage    = btrfs_releasepage,
2975         .set_page_dirty = __set_page_dirty_nobuffers,
2976 };
2977
2978 static struct address_space_operations btrfs_symlink_aops = {
2979         .readpage       = btrfs_readpage,
2980         .writepage      = btrfs_writepage,
2981         .invalidatepage = btrfs_invalidatepage,
2982         .releasepage    = btrfs_releasepage,
2983 };
2984
2985 static struct inode_operations btrfs_file_inode_operations = {
2986         .truncate       = btrfs_truncate,
2987         .getattr        = btrfs_getattr,
2988         .setattr        = btrfs_setattr,
2989         .setxattr       = generic_setxattr,
2990         .getxattr       = generic_getxattr,
2991         .listxattr      = btrfs_listxattr,
2992         .removexattr    = generic_removexattr,
2993         .permission     = btrfs_permission,
2994 };
2995 static struct inode_operations btrfs_special_inode_operations = {
2996         .getattr        = btrfs_getattr,
2997         .setattr        = btrfs_setattr,
2998         .permission     = btrfs_permission,
2999 };
3000 static struct inode_operations btrfs_symlink_inode_operations = {
3001         .readlink       = generic_readlink,
3002         .follow_link    = page_follow_link_light,
3003         .put_link       = page_put_link,
3004         .permission     = btrfs_permission,
3005 };