fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/buffer_head.h>
  20 #include <linux/fs.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/string.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mpage.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/version.h>
  35 #include <linux/xattr.h>
  36 #include "ctree.h"
  37 #include "disk-io.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "ioctl.h"
  41 #include "print-tree.h"
  42
  43 struct btrfs_iget_args {
  44         u64 ino;
  45         struct btrfs_root *root;
  46 };
  47
  48 static struct inode_operations btrfs_dir_inode_operations;
  49 static struct inode_operations btrfs_symlink_inode_operations;
  50 static struct inode_operations btrfs_dir_ro_inode_operations;
  51 static struct inode_operations btrfs_special_inode_operations;
  52 static struct inode_operations btrfs_file_inode_operations;
  53 static struct address_space_operations btrfs_aops;
  54 static struct address_space_operations btrfs_symlink_aops;
  55 static struct file_operations btrfs_dir_file_operations;
  56 static struct extent_map_ops btrfs_extent_map_ops;
  57
  58 static struct kmem_cache *btrfs_inode_cachep;
  59 struct kmem_cache *btrfs_trans_handle_cachep;
  60 struct kmem_cache *btrfs_transaction_cachep;
  61 struct kmem_cache *btrfs_bit_radix_cachep;
  62 struct kmem_cache *btrfs_path_cachep;
  63
  64 #define S_SHIFT 12
  65 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  66         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  67         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  68         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  69         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  70         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  71         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  72         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  73 };
  74
  75 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  76                            int for_del)
  77 {
  78         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  79         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  80         u64 thresh;
  81         int ret = 0;
  82
  83         if (for_del)
  84                 thresh = total * 90;
  85         else
  86                 thresh = total * 85;
  87
  88         do_div(thresh, 100);
  89
  90         spin_lock(&root->fs_info->delalloc_lock);
  91         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  92                 ret = -ENOSPC;
  93         spin_unlock(&root->fs_info->delalloc_lock);
  94         return ret;
  95 }
  96
  97 static int cow_file_range(struct inode *inode, u64 start, u64 end)
  98 {
  99         struct btrfs_root *root = BTRFS_I(inode)->root;
 100         struct btrfs_trans_handle *trans;
 101         u64 alloc_hint = 0;
 102         u64 num_bytes;
 103         u64 cur_alloc_size;
 104         u64 blocksize = root->sectorsize;
 105         struct btrfs_key ins;
 106         int ret;
 107
 108         trans = btrfs_start_transaction(root, 1);
 109         BUG_ON(!trans);
 110         btrfs_set_trans_block_group(trans, inode);
 111
 112         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 113         num_bytes = max(blocksize,  num_bytes);
 114         ret = btrfs_drop_extents(trans, root, inode,
 115                                  start, start + num_bytes, start, &alloc_hint);
 116
 117         if (alloc_hint == EXTENT_MAP_INLINE)
 118                 goto out;
 119
 120         while(num_bytes > 0) {
 121                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 122                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 123                                          root->root_key.objectid,
 124                                          trans->transid,
 125                                          inode->i_ino, start, 0,
 126                                          alloc_hint, (u64)-1, &ins, 1);
 127                 if (ret) {
 128                         WARN_ON(1);
 129                         goto out;
 130                 }
 131                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 132                                                start, ins.objectid, ins.offset,
 133                                                ins.offset);
 134                 num_bytes -= cur_alloc_size;
 135                 alloc_hint = ins.objectid + ins.offset;
 136                 start += cur_alloc_size;
 137         }
 138         btrfs_add_ordered_inode(inode);
 139 out:
 140         btrfs_end_transaction(trans, root);
 141         return ret;
 142 }
 143
 144 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 145 {
 146         u64 extent_start;
 147         u64 extent_end;
 148         u64 bytenr;
 149         u64 cow_end;
 150         u64 loops = 0;
 151         u64 total_fs_bytes;
 152         struct btrfs_root *root = BTRFS_I(inode)->root;
 153         struct extent_buffer *leaf;
 154         int found_type;
 155         struct btrfs_path *path;
 156         struct btrfs_file_extent_item *item;
 157         int ret;
 158         int err;
 159         struct btrfs_key found_key;
 160
 161         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 162         path = btrfs_alloc_path();
 163         BUG_ON(!path);
 164 again:
 165         ret = btrfs_lookup_file_extent(NULL, root, path,
 166                                        inode->i_ino, start, 0);
 167         if (ret < 0) {
 168                 btrfs_free_path(path);
 169                 return ret;
 170         }
 171
 172         cow_end = end;
 173         if (ret != 0) {
 174                 if (path->slots[0] == 0)
 175                         goto not_found;
 176                 path->slots[0]--;
 177         }
 178
 179         leaf = path->nodes[0];
 180         item = btrfs_item_ptr(leaf, path->slots[0],
 181                               struct btrfs_file_extent_item);
 182
 183         /* are we inside the extent that was found? */
 184         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 185         found_type = btrfs_key_type(&found_key);
 186         if (found_key.objectid != inode->i_ino ||
 187             found_type != BTRFS_EXTENT_DATA_KEY) {
 188                 goto not_found;
 189         }
 190
 191         found_type = btrfs_file_extent_type(leaf, item);
 192         extent_start = found_key.offset;
 193         if (found_type == BTRFS_FILE_EXTENT_REG) {
 194                 u64 extent_num_bytes;
 195
 196                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 197                 extent_end = extent_start + extent_num_bytes;
 198                 err = 0;
 199
 200                 if (loops && start != extent_start)
 201                         goto not_found;
 202
 203                 if (start < extent_start || start >= extent_end)
 204                         goto not_found;
 205
 206                 cow_end = min(end, extent_end - 1);
 207                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 208                 if (bytenr == 0)
 209                         goto not_found;
 210
 211                 /*
 212                  * we may be called by the resizer, make sure we're inside
 213                  * the limits of the FS
 214                  */
 215                 if (bytenr + extent_num_bytes > total_fs_bytes)
 216                         goto not_found;
 217
 218                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 219                         goto not_found;
 220                 }
 221
 222                 start = extent_end;
 223         } else {
 224                 goto not_found;
 225         }
 226 loop:
 227         if (start > end) {
 228                 btrfs_free_path(path);
 229                 return 0;
 230         }
 231         btrfs_release_path(root, path);
 232         loops++;
 233         goto again;
 234
 235 not_found:
 236         cow_file_range(inode, start, cow_end);
 237         start = cow_end + 1;
 238         goto loop;
 239 }
 240
 241 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 242 {
 243         struct btrfs_root *root = BTRFS_I(inode)->root;
 244         u64 num_bytes;
 245         int ret;
 246         mutex_lock(&root->fs_info->fs_mutex);
 247         if (btrfs_test_opt(root, NODATACOW) ||
 248             btrfs_test_flag(inode, NODATACOW))
 249                 ret = run_delalloc_nocow(inode, start, end);
 250         else
 251                 ret = cow_file_range(inode, start, end);
 252
 253         spin_lock(&root->fs_info->delalloc_lock);
 254         num_bytes = end + 1 - start;
 255         if (root->fs_info->delalloc_bytes < num_bytes) {
 256                 printk("delalloc accounting error total %llu sub %llu\n",
 257                        root->fs_info->delalloc_bytes, num_bytes);
 258         } else {
 259                 root->fs_info->delalloc_bytes -= num_bytes;
 260         }
 261         spin_unlock(&root->fs_info->delalloc_lock);
 262
 263         mutex_unlock(&root->fs_info->fs_mutex);
 264         return ret;
 265 }
 266
 267 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 268 {
 269         struct inode *inode = page->mapping->host;
 270         struct btrfs_root *root = BTRFS_I(inode)->root;
 271         struct btrfs_trans_handle *trans;
 272         char *kaddr;
 273         int ret = 0;
 274         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 275         size_t offset = start - page_start;
 276         if (btrfs_test_opt(root, NODATASUM) ||
 277             btrfs_test_flag(inode, NODATASUM))
 278                 return 0;
 279         mutex_lock(&root->fs_info->fs_mutex);
 280         trans = btrfs_start_transaction(root, 1);
 281         btrfs_set_trans_block_group(trans, inode);
 282         kaddr = kmap(page);
 283         btrfs_csum_file_block(trans, root, inode, inode->i_ino,
 284                               start, kaddr + offset, end - start + 1);
 285         kunmap(page);
 286         ret = btrfs_end_transaction(trans, root);
 287         BUG_ON(ret);
 288         mutex_unlock(&root->fs_info->fs_mutex);
 289         return ret;
 290 }
 291
 292 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 293 {
 294         int ret = 0;
 295         struct inode *inode = page->mapping->host;
 296         struct btrfs_root *root = BTRFS_I(inode)->root;
 297         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 298         struct btrfs_csum_item *item;
 299         struct btrfs_path *path = NULL;
 300         u32 csum;
 301         if (btrfs_test_opt(root, NODATASUM) ||
 302             btrfs_test_flag(inode, NODATASUM))
 303                 return 0;
 304         mutex_lock(&root->fs_info->fs_mutex);
 305         path = btrfs_alloc_path();
 306         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 307         if (IS_ERR(item)) {
 308                 ret = PTR_ERR(item);
 309                 /* a csum that isn't present is a preallocated region. */
 310                 if (ret == -ENOENT || ret == -EFBIG)
 311                         ret = 0;
 312                 csum = 0;
 313                 goto out;
 314         }
 315         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 316                            BTRFS_CRC32_SIZE);
 317         set_state_private(em_tree, start, csum);
 318 out:
 319         if (path)
 320                 btrfs_free_path(path);
 321         mutex_unlock(&root->fs_info->fs_mutex);
 322         return ret;
 323 }
 324
 325 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end)
 326 {
 327         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 328         struct inode *inode = page->mapping->host;
 329         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 330         char *kaddr;
 331         u64 private;
 332         int ret;
 333         struct btrfs_root *root = BTRFS_I(inode)->root;
 334         u32 csum = ~(u32)0;
 335         unsigned long flags;
 336         if (btrfs_test_opt(root, NODATASUM) ||
 337             btrfs_test_flag(inode, NODATASUM))
 338                 return 0;
 339         ret = get_state_private(em_tree, start, &private);
 340         local_irq_save(flags);
 341         kaddr = kmap_atomic(page, KM_IRQ0);
 342         if (ret) {
 343                 goto zeroit;
 344         }
 345         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 346         btrfs_csum_final(csum, (char *)&csum);
 347         if (csum != private) {
 348                 goto zeroit;
 349         }
 350         kunmap_atomic(kaddr, KM_IRQ0);
 351         local_irq_restore(flags);
 352         return 0;
 353
 354 zeroit:
 355         printk("btrfs csum failed ino %lu off %llu\n",
 356                page->mapping->host->i_ino, (unsigned long long)start);
 357         memset(kaddr + offset, 1, end - start + 1);
 358         flush_dcache_page(page);
 359         kunmap_atomic(kaddr, KM_IRQ0);
 360         local_irq_restore(flags);
 361         return 0;
 362 }
 363
 364 void btrfs_read_locked_inode(struct inode *inode)
 365 {
 366         struct btrfs_path *path;
 367         struct extent_buffer *leaf;
 368         struct btrfs_inode_item *inode_item;
 369         struct btrfs_inode_timespec *tspec;
 370         struct btrfs_root *root = BTRFS_I(inode)->root;
 371         struct btrfs_key location;
 372         u64 alloc_group_block;
 373         u32 rdev;
 374         int ret;
 375
 376         path = btrfs_alloc_path();
 377         BUG_ON(!path);
 378         mutex_lock(&root->fs_info->fs_mutex);
 379         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 380
 381         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 382         if (ret)
 383                 goto make_bad;
 384
 385         leaf = path->nodes[0];
 386         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 387                                     struct btrfs_inode_item);
 388
 389         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 390         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 391         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 392         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 393         inode->i_size = btrfs_inode_size(leaf, inode_item);
 394
 395         tspec = btrfs_inode_atime(inode_item);
 396         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 397         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 398
 399         tspec = btrfs_inode_mtime(inode_item);
 400         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 401         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 402
 403         tspec = btrfs_inode_ctime(inode_item);
 404         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 405         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 406
 407         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 408         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 409         inode->i_rdev = 0;
 410         rdev = btrfs_inode_rdev(leaf, inode_item);
 411
 412         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 413         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 414                                                        alloc_group_block);
 415         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 416         if (!BTRFS_I(inode)->block_group) {
 417                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 418                                                          NULL, 0, 0, 0);
 419         }
 420         btrfs_free_path(path);
 421         inode_item = NULL;
 422
 423         mutex_unlock(&root->fs_info->fs_mutex);
 424
 425         switch (inode->i_mode & S_IFMT) {
 426         case S_IFREG:
 427                 inode->i_mapping->a_ops = &btrfs_aops;
 428                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
 429                 inode->i_fop = &btrfs_file_operations;
 430                 inode->i_op = &btrfs_file_inode_operations;
 431                 break;
 432         case S_IFDIR:
 433                 inode->i_fop = &btrfs_dir_file_operations;
 434                 if (root == root->fs_info->tree_root)
 435                         inode->i_op = &btrfs_dir_ro_inode_operations;
 436                 else
 437                         inode->i_op = &btrfs_dir_inode_operations;
 438                 break;
 439         case S_IFLNK:
 440                 inode->i_op = &btrfs_symlink_inode_operations;
 441                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 442                 break;
 443         default:
 444                 init_special_inode(inode, inode->i_mode, rdev);
 445                 break;
 446         }
 447         return;
 448
 449 make_bad:
 450         btrfs_release_path(root, path);
 451         btrfs_free_path(path);
 452         mutex_unlock(&root->fs_info->fs_mutex);
 453         make_bad_inode(inode);
 454 }
 455
 456 static void fill_inode_item(struct extent_buffer *leaf,
 457                             struct btrfs_inode_item *item,
 458                             struct inode *inode)
 459 {
 460         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 461         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 462         btrfs_set_inode_size(leaf, item, inode->i_size);
 463         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 464         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 465
 466         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 467                                inode->i_atime.tv_sec);
 468         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 469                                 inode->i_atime.tv_nsec);
 470
 471         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 472                                inode->i_mtime.tv_sec);
 473         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 474                                 inode->i_mtime.tv_nsec);
 475
 476         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 477                                inode->i_ctime.tv_sec);
 478         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 479                                 inode->i_ctime.tv_nsec);
 480
 481         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 482         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 483         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 484         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 485         btrfs_set_inode_block_group(leaf, item,
 486                                     BTRFS_I(inode)->block_group->key.objectid);
 487 }
 488
 489 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 490                               struct btrfs_root *root,
 491                               struct inode *inode)
 492 {
 493         struct btrfs_inode_item *inode_item;
 494         struct btrfs_path *path;
 495         struct extent_buffer *leaf;
 496         int ret;
 497
 498         path = btrfs_alloc_path();
 499         BUG_ON(!path);
 500         ret = btrfs_lookup_inode(trans, root, path,
 501                                  &BTRFS_I(inode)->location, 1);
 502         if (ret) {
 503                 if (ret > 0)
 504                         ret = -ENOENT;
 505                 goto failed;
 506         }
 507
 508         leaf = path->nodes[0];
 509         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 510                                   struct btrfs_inode_item);
 511
 512         fill_inode_item(leaf, inode_item, inode);
 513         btrfs_mark_buffer_dirty(leaf);
 514         btrfs_set_inode_last_trans(trans, inode);
 515         ret = 0;
 516 failed:
 517         btrfs_release_path(root, path);
 518         btrfs_free_path(path);
 519         return ret;
 520 }
 521
 522
 523 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 524                               struct btrfs_root *root,
 525                               struct inode *dir,
 526                               struct dentry *dentry)
 527 {
 528         struct btrfs_path *path;
 529         const char *name = dentry->d_name.name;
 530         int name_len = dentry->d_name.len;
 531         int ret = 0;
 532         struct extent_buffer *leaf;
 533         struct btrfs_dir_item *di;
 534         struct btrfs_key key;
 535
 536         path = btrfs_alloc_path();
 537         if (!path) {
 538                 ret = -ENOMEM;
 539                 goto err;
 540         }
 541
 542         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 543                                     name, name_len, -1);
 544         if (IS_ERR(di)) {
 545                 ret = PTR_ERR(di);
 546                 goto err;
 547         }
 548         if (!di) {
 549                 ret = -ENOENT;
 550                 goto err;
 551         }
 552         leaf = path->nodes[0];
 553         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 554         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 555         if (ret)
 556                 goto err;
 557         btrfs_release_path(root, path);
 558
 559         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 560                                          key.objectid, name, name_len, -1);
 561         if (IS_ERR(di)) {
 562                 ret = PTR_ERR(di);
 563                 goto err;
 564         }
 565         if (!di) {
 566                 ret = -ENOENT;
 567                 goto err;
 568         }
 569         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 570
 571         dentry->d_inode->i_ctime = dir->i_ctime;
 572         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 573                                   dentry->d_inode->i_ino,
 574                                   dentry->d_parent->d_inode->i_ino);
 575         if (ret) {
 576                 printk("failed to delete reference to %.*s, "
 577                        "inode %lu parent %lu\n", name_len, name,
 578                        dentry->d_inode->i_ino,
 579                        dentry->d_parent->d_inode->i_ino);
 580         }
 581 err:
 582         btrfs_free_path(path);
 583         if (!ret) {
 584                 dir->i_size -= name_len * 2;
 585                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 586                 btrfs_update_inode(trans, root, dir);
 587 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 588                 dentry->d_inode->i_nlink--;
 589 #else
 590                 drop_nlink(dentry->d_inode);
 591 #endif
 592                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 593                 dir->i_sb->s_dirt = 1;
 594         }
 595         return ret;
 596 }
 597
 598 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 599 {
 600         struct btrfs_root *root;
 601         struct btrfs_trans_handle *trans;
 602         int ret;
 603         unsigned long nr = 0;
 604
 605         root = BTRFS_I(dir)->root;
 606         mutex_lock(&root->fs_info->fs_mutex);
 607
 608         ret = btrfs_check_free_space(root, 1, 1);
 609         if (ret)
 610                 goto fail;
 611
 612         trans = btrfs_start_transaction(root, 1);
 613
 614         btrfs_set_trans_block_group(trans, dir);
 615         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 616         nr = trans->blocks_used;
 617
 618         btrfs_end_transaction(trans, root);
 619 fail:
 620         mutex_unlock(&root->fs_info->fs_mutex);
 621         btrfs_btree_balance_dirty(root, nr);
 622         btrfs_throttle(root);
 623         return ret;
 624 }
 625
 626 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 627 {
 628         struct inode *inode = dentry->d_inode;
 629         int err = 0;
 630         int ret;
 631         struct btrfs_root *root = BTRFS_I(dir)->root;
 632         struct btrfs_trans_handle *trans;
 633         unsigned long nr = 0;
 634
 635         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 636                 return -ENOTEMPTY;
 637
 638         mutex_lock(&root->fs_info->fs_mutex);
 639         ret = btrfs_check_free_space(root, 1, 1);
 640         if (ret)
 641                 goto fail;
 642
 643         trans = btrfs_start_transaction(root, 1);
 644         btrfs_set_trans_block_group(trans, dir);
 645
 646         /* now the directory is empty */
 647         err = btrfs_unlink_trans(trans, root, dir, dentry);
 648         if (!err) {
 649                 inode->i_size = 0;
 650         }
 651
 652         nr = trans->blocks_used;
 653         ret = btrfs_end_transaction(trans, root);
 654 fail:
 655         mutex_unlock(&root->fs_info->fs_mutex);
 656         btrfs_btree_balance_dirty(root, nr);
 657         btrfs_throttle(root);
 658
 659         if (ret && !err)
 660                 err = ret;
 661         return err;
 662 }
 663
 664 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 665                             struct btrfs_root *root,
 666                             struct inode *inode)
 667 {
 668         struct btrfs_path *path;
 669         int ret;
 670
 671         clear_inode(inode);
 672
 673         path = btrfs_alloc_path();
 674         BUG_ON(!path);
 675         ret = btrfs_lookup_inode(trans, root, path,
 676                                  &BTRFS_I(inode)->location, -1);
 677         if (ret > 0)
 678                 ret = -ENOENT;
 679         if (!ret)
 680                 ret = btrfs_del_item(trans, root, path);
 681         btrfs_free_path(path);
 682         return ret;
 683 }
 684
 685 /*
 686  * this can truncate away extent items, csum items and directory items.
 687  * It starts at a high offset and removes keys until it can't find
 688  * any higher than i_size.
 689  *
 690  * csum items that cross the new i_size are truncated to the new size
 691  * as well.
 692  */
 693 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 694                                    struct btrfs_root *root,
 695                                    struct inode *inode)
 696 {
 697         int ret;
 698         struct btrfs_path *path;
 699         struct btrfs_key key;
 700         struct btrfs_key found_key;
 701         u32 found_type;
 702         struct extent_buffer *leaf;
 703         struct btrfs_file_extent_item *fi;
 704         u64 extent_start = 0;
 705         u64 extent_num_bytes = 0;
 706         u64 item_end = 0;
 707         u64 root_gen = 0;
 708         u64 root_owner = 0;
 709         int found_extent;
 710         int del_item;
 711         int extent_type = -1;
 712
 713         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 714         path = btrfs_alloc_path();
 715         path->reada = -1;
 716         BUG_ON(!path);
 717
 718         /* FIXME, add redo link to tree so we don't leak on crash */
 719         key.objectid = inode->i_ino;
 720         key.offset = (u64)-1;
 721         key.type = (u8)-1;
 722
 723         while(1) {
 724                 btrfs_init_path(path);
 725                 fi = NULL;
 726                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 727                 if (ret < 0) {
 728                         goto error;
 729                 }
 730                 if (ret > 0) {
 731                         BUG_ON(path->slots[0] == 0);
 732                         path->slots[0]--;
 733                 }
 734                 leaf = path->nodes[0];
 735                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 736                 found_type = btrfs_key_type(&found_key);
 737
 738                 if (found_key.objectid != inode->i_ino)
 739                         break;
 740
 741                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 742                     found_type != BTRFS_DIR_ITEM_KEY &&
 743                     found_type != BTRFS_DIR_INDEX_KEY &&
 744                     found_type != BTRFS_EXTENT_DATA_KEY)
 745                         break;
 746
 747                 item_end = found_key.offset;
 748                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 749                         fi = btrfs_item_ptr(leaf, path->slots[0],
 750                                             struct btrfs_file_extent_item);
 751                         extent_type = btrfs_file_extent_type(leaf, fi);
 752                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 753                                 item_end +=
 754                                     btrfs_file_extent_num_bytes(leaf, fi);
 755                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 756                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 757                                                                 path->slots[0]);
 758                                 item_end += btrfs_file_extent_inline_len(leaf,
 759                                                                          item);
 760                         }
 761                         item_end--;
 762                 }
 763                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 764                         ret = btrfs_csum_truncate(trans, root, path,
 765                                                   inode->i_size);
 766                         BUG_ON(ret);
 767                 }
 768                 if (item_end < inode->i_size) {
 769                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 770                                 found_type = BTRFS_INODE_ITEM_KEY;
 771                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 772                                 found_type = BTRFS_CSUM_ITEM_KEY;
 773                         } else if (found_type) {
 774                                 found_type--;
 775                         } else {
 776                                 break;
 777                         }
 778                         btrfs_set_key_type(&key, found_type);
 779                         btrfs_release_path(root, path);
 780                         continue;
 781                 }
 782                 if (found_key.offset >= inode->i_size)
 783                         del_item = 1;
 784                 else
 785                         del_item = 0;
 786                 found_extent = 0;
 787
 788                 /* FIXME, shrink the extent if the ref count is only 1 */
 789                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 790                         goto delete;
 791
 792                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 793                         u64 num_dec;
 794                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 795                         if (!del_item) {
 796                                 u64 orig_num_bytes =
 797                                         btrfs_file_extent_num_bytes(leaf, fi);
 798                                 extent_num_bytes = inode->i_size -
 799                                         found_key.offset + root->sectorsize - 1;
 800                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 801                                                          extent_num_bytes);
 802                                 num_dec = (orig_num_bytes -
 803                                            extent_num_bytes) >> 9;
 804                                 if (extent_start != 0) {
 805                                         inode->i_blocks -= num_dec;
 806                                 }
 807                                 btrfs_mark_buffer_dirty(leaf);
 808                         } else {
 809                                 extent_num_bytes =
 810                                         btrfs_file_extent_disk_num_bytes(leaf,
 811                                                                          fi);
 812                                 /* FIXME blocksize != 4096 */
 813                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 814                                                                        fi) >> 9;
 815                                 if (extent_start != 0) {
 816                                         found_extent = 1;
 817                                         inode->i_blocks -= num_dec;
 818                                 }
 819                                 root_gen = btrfs_header_generation(leaf);
 820                                 root_owner = btrfs_header_owner(leaf);
 821                         }
 822                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 823                            !del_item) {
 824                         u32 newsize = inode->i_size - found_key.offset;
 825                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 826                         ret = btrfs_truncate_item(trans, root, path,
 827                                                   newsize, 1);
 828                         BUG_ON(ret);
 829                 }
 830 delete:
 831                 if (del_item) {
 832                         ret = btrfs_del_item(trans, root, path);
 833                         if (ret)
 834                                 goto error;
 835                 } else {
 836                         break;
 837                 }
 838                 btrfs_release_path(root, path);
 839                 if (found_extent) {
 840                         ret = btrfs_free_extent(trans, root, extent_start,
 841                                                 extent_num_bytes,
 842                                                 root_owner,
 843                                                 root_gen, inode->i_ino,
 844                                                 found_key.offset, 0);
 845                         BUG_ON(ret);
 846                 }
 847         }
 848         ret = 0;
 849 error:
 850         btrfs_release_path(root, path);
 851         btrfs_free_path(path);
 852         inode->i_sb->s_dirt = 1;
 853         return ret;
 854 }
 855
 856 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 857                               size_t zero_start)
 858 {
 859         char *kaddr;
 860         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 861         struct btrfs_root *root = BTRFS_I(inode)->root;
 862         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 863         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 864         u64 existing_delalloc;
 865         u64 delalloc_start;
 866         int ret = 0;
 867
 868         WARN_ON(!PageLocked(page));
 869         set_page_extent_mapped(page);
 870
 871         lock_extent(em_tree, page_start, page_end, GFP_NOFS);
 872         delalloc_start = page_start;
 873         existing_delalloc = count_range_bits(&BTRFS_I(inode)->extent_tree,
 874                                              &delalloc_start, page_end,
 875                                              PAGE_CACHE_SIZE, EXTENT_DELALLOC);
 876         set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
 877                             page_end, GFP_NOFS);
 878
 879         spin_lock(&root->fs_info->delalloc_lock);
 880         root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE - existing_delalloc;
 881         spin_unlock(&root->fs_info->delalloc_lock);
 882
 883         if (zero_start != PAGE_CACHE_SIZE) {
 884                 kaddr = kmap(page);
 885                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 886                 flush_dcache_page(page);
 887                 kunmap(page);
 888         }
 889         set_page_dirty(page);
 890         unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
 891
 892         return ret;
 893 }
 894
 895 /*
 896  * taken from block_truncate_page, but does cow as it zeros out
 897  * any bytes left in the last page in the file.
 898  */
 899 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 900 {
 901         struct inode *inode = mapping->host;
 902         struct btrfs_root *root = BTRFS_I(inode)->root;
 903         u32 blocksize = root->sectorsize;
 904         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 905         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 906         struct page *page;
 907         int ret = 0;
 908         u64 page_start;
 909
 910         if ((offset & (blocksize - 1)) == 0)
 911                 goto out;
 912
 913         ret = -ENOMEM;
 914         page = grab_cache_page(mapping, index);
 915         if (!page)
 916                 goto out;
 917         if (!PageUptodate(page)) {
 918                 ret = btrfs_readpage(NULL, page);
 919                 lock_page(page);
 920                 if (!PageUptodate(page)) {
 921                         ret = -EIO;
 922                         goto out;
 923                 }
 924         }
 925         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 926
 927         ret = btrfs_cow_one_page(inode, page, offset);
 928
 929         unlock_page(page);
 930         page_cache_release(page);
 931 out:
 932         return ret;
 933 }
 934
 935 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 936 {
 937         struct inode *inode = dentry->d_inode;
 938         int err;
 939
 940         err = inode_change_ok(inode, attr);
 941         if (err)
 942                 return err;
 943
 944         if (S_ISREG(inode->i_mode) &&
 945             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 946                 struct btrfs_trans_handle *trans;
 947                 struct btrfs_root *root = BTRFS_I(inode)->root;
 948                 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 949
 950                 u64 mask = root->sectorsize - 1;
 951                 u64 pos = (inode->i_size + mask) & ~mask;
 952                 u64 block_end = attr->ia_size | mask;
 953                 u64 hole_size;
 954                 u64 alloc_hint = 0;
 955
 956                 if (attr->ia_size <= pos)
 957                         goto out;
 958
 959                 mutex_lock(&root->fs_info->fs_mutex);
 960                 err = btrfs_check_free_space(root, 1, 0);
 961                 mutex_unlock(&root->fs_info->fs_mutex);
 962                 if (err)
 963                         goto fail;
 964
 965                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 966
 967                 lock_extent(em_tree, pos, block_end, GFP_NOFS);
 968                 hole_size = (attr->ia_size - pos + mask) & ~mask;
 969
 970                 mutex_lock(&root->fs_info->fs_mutex);
 971                 trans = btrfs_start_transaction(root, 1);
 972                 btrfs_set_trans_block_group(trans, inode);
 973                 err = btrfs_drop_extents(trans, root, inode,
 974                                          pos, pos + hole_size, pos,
 975                                          &alloc_hint);
 976
 977                 if (alloc_hint != EXTENT_MAP_INLINE) {
 978                         err = btrfs_insert_file_extent(trans, root,
 979                                                        inode->i_ino,
 980                                                        pos, 0, 0, hole_size);
 981                 }
 982                 btrfs_end_transaction(trans, root);
 983                 mutex_unlock(&root->fs_info->fs_mutex);
 984                 unlock_extent(em_tree, pos, block_end, GFP_NOFS);
 985                 if (err)
 986                         return err;
 987         }
 988 out:
 989         err = inode_setattr(inode, attr);
 990 fail:
 991         return err;
 992 }
 993 void btrfs_delete_inode(struct inode *inode)
 994 {
 995         struct btrfs_trans_handle *trans;
 996         struct btrfs_root *root = BTRFS_I(inode)->root;
 997         unsigned long nr;
 998         int ret;
 999
1000         truncate_inode_pages(&inode->i_data, 0);
1001         if (is_bad_inode(inode)) {
1002                 goto no_delete;
1003         }
1004
1005         inode->i_size = 0;
1006         mutex_lock(&root->fs_info->fs_mutex);
1007         trans = btrfs_start_transaction(root, 1);
1008
1009         btrfs_set_trans_block_group(trans, inode);
1010         ret = btrfs_truncate_in_trans(trans, root, inode);
1011         if (ret)
1012                 goto no_delete_lock;
1013         ret = btrfs_delete_xattrs(trans, root, inode);
1014         if (ret)
1015                 goto no_delete_lock;
1016         ret = btrfs_free_inode(trans, root, inode);
1017         if (ret)
1018                 goto no_delete_lock;
1019         nr = trans->blocks_used;
1020
1021         btrfs_end_transaction(trans, root);
1022         mutex_unlock(&root->fs_info->fs_mutex);
1023         btrfs_btree_balance_dirty(root, nr);
1024         btrfs_throttle(root);
1025         return;
1026
1027 no_delete_lock:
1028         nr = trans->blocks_used;
1029         btrfs_end_transaction(trans, root);
1030         mutex_unlock(&root->fs_info->fs_mutex);
1031         btrfs_btree_balance_dirty(root, nr);
1032         btrfs_throttle(root);
1033 no_delete:
1034         clear_inode(inode);
1035 }
1036
1037 /*
1038  * this returns the key found in the dir entry in the location pointer.
1039  * If no dir entries were found, location->objectid is 0.
1040  */
1041 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1042                                struct btrfs_key *location)
1043 {
1044         const char *name = dentry->d_name.name;
1045         int namelen = dentry->d_name.len;
1046         struct btrfs_dir_item *di;
1047         struct btrfs_path *path;
1048         struct btrfs_root *root = BTRFS_I(dir)->root;
1049         int ret = 0;
1050
1051         if (namelen == 1 && strcmp(name, ".") == 0) {
1052                 location->objectid = dir->i_ino;
1053                 location->type = BTRFS_INODE_ITEM_KEY;
1054                 location->offset = 0;
1055                 return 0;
1056         }
1057         path = btrfs_alloc_path();
1058         BUG_ON(!path);
1059
1060         if (namelen == 2 && strcmp(name, "..") == 0) {
1061                 struct btrfs_key key;
1062                 struct extent_buffer *leaf;
1063                 u32 nritems;
1064                 int slot;
1065
1066                 key.objectid = dir->i_ino;
1067                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1068                 key.offset = 0;
1069                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1070                 BUG_ON(ret == 0);
1071                 ret = 0;
1072
1073                 leaf = path->nodes[0];
1074                 slot = path->slots[0];
1075                 nritems = btrfs_header_nritems(leaf);
1076                 if (slot >= nritems)
1077                         goto out_err;
1078
1079                 btrfs_item_key_to_cpu(leaf, &key, slot);
1080                 if (key.objectid != dir->i_ino ||
1081                     key.type != BTRFS_INODE_REF_KEY) {
1082                         goto out_err;
1083                 }
1084                 location->objectid = key.offset;
1085                 location->type = BTRFS_INODE_ITEM_KEY;
1086                 location->offset = 0;
1087                 goto out;
1088         }
1089
1090         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1091                                     namelen, 0);
1092         if (IS_ERR(di))
1093                 ret = PTR_ERR(di);
1094         if (!di || IS_ERR(di)) {
1095                 goto out_err;
1096         }
1097         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1098 out:
1099         btrfs_free_path(path);
1100         return ret;
1101 out_err:
1102         location->objectid = 0;
1103         goto out;
1104 }
1105
1106 /*
1107  * when we hit a tree root in a directory, the btrfs part of the inode
1108  * needs to be changed to reflect the root directory of the tree root.  This
1109  * is kind of like crossing a mount point.
1110  */
1111 static int fixup_tree_root_location(struct btrfs_root *root,
1112                              struct btrfs_key *location,
1113                              struct btrfs_root **sub_root,
1114                              struct dentry *dentry)
1115 {
1116         struct btrfs_path *path;
1117         struct btrfs_root_item *ri;
1118
1119         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1120                 return 0;
1121         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1122                 return 0;
1123
1124         path = btrfs_alloc_path();
1125         BUG_ON(!path);
1126         mutex_lock(&root->fs_info->fs_mutex);
1127
1128         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1129                                         dentry->d_name.name,
1130                                         dentry->d_name.len);
1131         if (IS_ERR(*sub_root))
1132                 return PTR_ERR(*sub_root);
1133
1134         ri = &(*sub_root)->root_item;
1135         location->objectid = btrfs_root_dirid(ri);
1136         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1137         location->offset = 0;
1138
1139         btrfs_free_path(path);
1140         mutex_unlock(&root->fs_info->fs_mutex);
1141         return 0;
1142 }
1143
1144 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1145 {
1146         struct btrfs_iget_args *args = p;
1147         inode->i_ino = args->ino;
1148         BTRFS_I(inode)->root = args->root;
1149         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1150                              inode->i_mapping, GFP_NOFS);
1151         return 0;
1152 }
1153
1154 static int btrfs_find_actor(struct inode *inode, void *opaque)
1155 {
1156         struct btrfs_iget_args *args = opaque;
1157         return (args->ino == inode->i_ino &&
1158                 args->root == BTRFS_I(inode)->root);
1159 }
1160
1161 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1162                             u64 root_objectid)
1163 {
1164         struct btrfs_iget_args args;
1165         args.ino = objectid;
1166         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1167
1168         if (!args.root)
1169                 return NULL;
1170
1171         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1172 }
1173
1174 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1175                                 struct btrfs_root *root)
1176 {
1177         struct inode *inode;
1178         struct btrfs_iget_args args;
1179         args.ino = objectid;
1180         args.root = root;
1181
1182         inode = iget5_locked(s, objectid, btrfs_find_actor,
1183                              btrfs_init_locked_inode,
1184                              (void *)&args);
1185         return inode;
1186 }
1187
1188 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1189                                    struct nameidata *nd)
1190 {
1191         struct inode * inode;
1192         struct btrfs_inode *bi = BTRFS_I(dir);
1193         struct btrfs_root *root = bi->root;
1194         struct btrfs_root *sub_root = root;
1195         struct btrfs_key location;
1196         int ret;
1197
1198         if (dentry->d_name.len > BTRFS_NAME_LEN)
1199                 return ERR_PTR(-ENAMETOOLONG);
1200
1201         mutex_lock(&root->fs_info->fs_mutex);
1202         ret = btrfs_inode_by_name(dir, dentry, &location);
1203         mutex_unlock(&root->fs_info->fs_mutex);
1204
1205         if (ret < 0)
1206                 return ERR_PTR(ret);
1207
1208         inode = NULL;
1209         if (location.objectid) {
1210                 ret = fixup_tree_root_location(root, &location, &sub_root,
1211                                                 dentry);
1212                 if (ret < 0)
1213                         return ERR_PTR(ret);
1214                 if (ret > 0)
1215                         return ERR_PTR(-ENOENT);
1216                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1217                                           sub_root);
1218                 if (!inode)
1219                         return ERR_PTR(-EACCES);
1220                 if (inode->i_state & I_NEW) {
1221                         /* the inode and parent dir are two different roots */
1222                         if (sub_root != root) {
1223                                 igrab(inode);
1224                                 sub_root->inode = inode;
1225                         }
1226                         BTRFS_I(inode)->root = sub_root;
1227                         memcpy(&BTRFS_I(inode)->location, &location,
1228                                sizeof(location));
1229                         btrfs_read_locked_inode(inode);
1230                         unlock_new_inode(inode);
1231                 }
1232         }
1233         return d_splice_alias(inode, dentry);
1234 }
1235
1236 static unsigned char btrfs_filetype_table[] = {
1237         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1238 };
1239
1240 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1241 {
1242         struct inode *inode = filp->f_dentry->d_inode;
1243         struct btrfs_root *root = BTRFS_I(inode)->root;
1244         struct btrfs_item *item;
1245         struct btrfs_dir_item *di;
1246         struct btrfs_key key;
1247         struct btrfs_key found_key;
1248         struct btrfs_path *path;
1249         int ret;
1250         u32 nritems;
1251         struct extent_buffer *leaf;
1252         int slot;
1253         int advance;
1254         unsigned char d_type;
1255         int over = 0;
1256         u32 di_cur;
1257         u32 di_total;
1258         u32 di_len;
1259         int key_type = BTRFS_DIR_INDEX_KEY;
1260         char tmp_name[32];
1261         char *name_ptr;
1262         int name_len;
1263
1264         /* FIXME, use a real flag for deciding about the key type */
1265         if (root->fs_info->tree_root == root)
1266                 key_type = BTRFS_DIR_ITEM_KEY;
1267
1268         /* special case for "." */
1269         if (filp->f_pos == 0) {
1270                 over = filldir(dirent, ".", 1,
1271                                1, inode->i_ino,
1272                                DT_DIR);
1273                 if (over)
1274                         return 0;
1275                 filp->f_pos = 1;
1276         }
1277
1278         mutex_lock(&root->fs_info->fs_mutex);
1279         key.objectid = inode->i_ino;
1280         path = btrfs_alloc_path();
1281         path->reada = 2;
1282
1283         /* special case for .., just use the back ref */
1284         if (filp->f_pos == 1) {
1285                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1286                 key.offset = 0;
1287                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1288                 BUG_ON(ret == 0);
1289                 leaf = path->nodes[0];
1290                 slot = path->slots[0];
1291                 nritems = btrfs_header_nritems(leaf);
1292                 if (slot >= nritems) {
1293                         btrfs_release_path(root, path);
1294                         goto read_dir_items;
1295                 }
1296                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1297                 btrfs_release_path(root, path);
1298                 if (found_key.objectid != key.objectid ||
1299                     found_key.type != BTRFS_INODE_REF_KEY)
1300                         goto read_dir_items;
1301                 over = filldir(dirent, "..", 2,
1302                                2, found_key.offset, DT_DIR);
1303                 if (over)
1304                         goto nopos;
1305                 filp->f_pos = 2;
1306         }
1307
1308 read_dir_items:
1309         btrfs_set_key_type(&key, key_type);
1310         key.offset = filp->f_pos;
1311
1312         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1313         if (ret < 0)
1314                 goto err;
1315         advance = 0;
1316         while(1) {
1317                 leaf = path->nodes[0];
1318                 nritems = btrfs_header_nritems(leaf);
1319                 slot = path->slots[0];
1320                 if (advance || slot >= nritems) {
1321                         if (slot >= nritems -1) {
1322                                 ret = btrfs_next_leaf(root, path);
1323                                 if (ret)
1324                                         break;
1325                                 leaf = path->nodes[0];
1326                                 nritems = btrfs_header_nritems(leaf);
1327                                 slot = path->slots[0];
1328                         } else {
1329                                 slot++;
1330                                 path->slots[0]++;
1331                         }
1332                 }
1333                 advance = 1;
1334                 item = btrfs_item_nr(leaf, slot);
1335                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1336
1337                 if (found_key.objectid != key.objectid)
1338                         break;
1339                 if (btrfs_key_type(&found_key) != key_type)
1340                         break;
1341                 if (found_key.offset < filp->f_pos)
1342                         continue;
1343
1344                 filp->f_pos = found_key.offset;
1345                 advance = 1;
1346                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1347                 di_cur = 0;
1348                 di_total = btrfs_item_size(leaf, item);
1349                 while(di_cur < di_total) {
1350                         struct btrfs_key location;
1351
1352                         name_len = btrfs_dir_name_len(leaf, di);
1353                         if (name_len < 32) {
1354                                 name_ptr = tmp_name;
1355                         } else {
1356                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1357                                 BUG_ON(!name_ptr);
1358                         }
1359                         read_extent_buffer(leaf, name_ptr,
1360                                            (unsigned long)(di + 1), name_len);
1361
1362                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1363                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1364                         over = filldir(dirent, name_ptr, name_len,
1365                                        found_key.offset,
1366                                        location.objectid,
1367                                        d_type);
1368
1369                         if (name_ptr != tmp_name)
1370                                 kfree(name_ptr);
1371
1372                         if (over)
1373                                 goto nopos;
1374                         di_len = btrfs_dir_name_len(leaf, di) +
1375                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1376                         di_cur += di_len;
1377                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1378                 }
1379         }
1380         filp->f_pos++;
1381 nopos:
1382         ret = 0;
1383 err:
1384         btrfs_release_path(root, path);
1385         btrfs_free_path(path);
1386         mutex_unlock(&root->fs_info->fs_mutex);
1387         return ret;
1388 }
1389
1390 int btrfs_write_inode(struct inode *inode, int wait)
1391 {
1392         struct btrfs_root *root = BTRFS_I(inode)->root;
1393         struct btrfs_trans_handle *trans;
1394         int ret = 0;
1395
1396         if (wait) {
1397                 mutex_lock(&root->fs_info->fs_mutex);
1398                 trans = btrfs_start_transaction(root, 1);
1399                 btrfs_set_trans_block_group(trans, inode);
1400                 ret = btrfs_commit_transaction(trans, root);
1401                 mutex_unlock(&root->fs_info->fs_mutex);
1402         }
1403         return ret;
1404 }
1405
1406 /*
1407  * This is somewhat expensive, updating the tree every time the
1408  * inode changes.  But, it is most likely to find the inode in cache.
1409  * FIXME, needs more benchmarking...there are no reasons other than performance
1410  * to keep or drop this code.
1411  */
1412 void btrfs_dirty_inode(struct inode *inode)
1413 {
1414         struct btrfs_root *root = BTRFS_I(inode)->root;
1415         struct btrfs_trans_handle *trans;
1416
1417         mutex_lock(&root->fs_info->fs_mutex);
1418         trans = btrfs_start_transaction(root, 1);
1419         btrfs_set_trans_block_group(trans, inode);
1420         btrfs_update_inode(trans, root, inode);
1421         btrfs_end_transaction(trans, root);
1422         mutex_unlock(&root->fs_info->fs_mutex);
1423 }
1424
1425 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1426                                      struct btrfs_root *root,
1427                                      u64 objectid,
1428                                      struct btrfs_block_group_cache *group,
1429                                      int mode)
1430 {
1431         struct inode *inode;
1432         struct btrfs_inode_item *inode_item;
1433         struct btrfs_key *location;
1434         struct btrfs_path *path;
1435         int ret;
1436         int owner;
1437
1438         path = btrfs_alloc_path();
1439         BUG_ON(!path);
1440
1441         inode = new_inode(root->fs_info->sb);
1442         if (!inode)
1443                 return ERR_PTR(-ENOMEM);
1444
1445         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1446                              inode->i_mapping, GFP_NOFS);
1447         BTRFS_I(inode)->root = root;
1448
1449         if (mode & S_IFDIR)
1450                 owner = 0;
1451         else
1452                 owner = 1;
1453         group = btrfs_find_block_group(root, group, 0, 0, owner);
1454         BTRFS_I(inode)->block_group = group;
1455         BTRFS_I(inode)->flags = 0;
1456         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1457         if (ret)
1458                 goto fail;
1459
1460         inode->i_uid = current->fsuid;
1461         inode->i_gid = current->fsgid;
1462         inode->i_mode = mode;
1463         inode->i_ino = objectid;
1464         inode->i_blocks = 0;
1465         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1466         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1467                                   struct btrfs_inode_item);
1468         fill_inode_item(path->nodes[0], inode_item, inode);
1469         btrfs_mark_buffer_dirty(path->nodes[0]);
1470         btrfs_free_path(path);
1471
1472         location = &BTRFS_I(inode)->location;
1473         location->objectid = objectid;
1474         location->offset = 0;
1475         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1476
1477         insert_inode_hash(inode);
1478         return inode;
1479 fail:
1480         btrfs_free_path(path);
1481         return ERR_PTR(ret);
1482 }
1483
1484 static inline u8 btrfs_inode_type(struct inode *inode)
1485 {
1486         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1487 }
1488
1489 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1490                             struct dentry *dentry, struct inode *inode)
1491 {
1492         int ret;
1493         struct btrfs_key key;
1494         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1495         struct inode *parent_inode;
1496
1497         key.objectid = inode->i_ino;
1498         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1499         key.offset = 0;
1500
1501         ret = btrfs_insert_dir_item(trans, root,
1502                                     dentry->d_name.name, dentry->d_name.len,
1503                                     dentry->d_parent->d_inode->i_ino,
1504                                     &key, btrfs_inode_type(inode));
1505         if (ret == 0) {
1506                 ret = btrfs_insert_inode_ref(trans, root,
1507                                      dentry->d_name.name,
1508                                      dentry->d_name.len,
1509                                      inode->i_ino,
1510                                      dentry->d_parent->d_inode->i_ino);
1511                 parent_inode = dentry->d_parent->d_inode;
1512                 parent_inode->i_size += dentry->d_name.len * 2;
1513                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1514                 ret = btrfs_update_inode(trans, root,
1515                                          dentry->d_parent->d_inode);
1516         }
1517         return ret;
1518 }
1519
1520 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1521                             struct dentry *dentry, struct inode *inode)
1522 {
1523         int err = btrfs_add_link(trans, dentry, inode);
1524         if (!err) {
1525                 d_instantiate(dentry, inode);
1526                 return 0;
1527         }
1528         if (err > 0)
1529                 err = -EEXIST;
1530         return err;
1531 }
1532
1533 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1534                         int mode, dev_t rdev)
1535 {
1536         struct btrfs_trans_handle *trans;
1537         struct btrfs_root *root = BTRFS_I(dir)->root;
1538         struct inode *inode = NULL;
1539         int err;
1540         int drop_inode = 0;
1541         u64 objectid;
1542         unsigned long nr = 0;
1543
1544         if (!new_valid_dev(rdev))
1545                 return -EINVAL;
1546
1547         mutex_lock(&root->fs_info->fs_mutex);
1548         err = btrfs_check_free_space(root, 1, 0);
1549         if (err)
1550                 goto fail;
1551
1552         trans = btrfs_start_transaction(root, 1);
1553         btrfs_set_trans_block_group(trans, dir);
1554
1555         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1556         if (err) {
1557                 err = -ENOSPC;
1558                 goto out_unlock;
1559         }
1560
1561         inode = btrfs_new_inode(trans, root, objectid,
1562                                 BTRFS_I(dir)->block_group, mode);
1563         err = PTR_ERR(inode);
1564         if (IS_ERR(inode))
1565                 goto out_unlock;
1566
1567         btrfs_set_trans_block_group(trans, inode);
1568         err = btrfs_add_nondir(trans, dentry, inode);
1569         if (err)
1570                 drop_inode = 1;
1571         else {
1572                 inode->i_op = &btrfs_special_inode_operations;
1573                 init_special_inode(inode, inode->i_mode, rdev);
1574                 btrfs_update_inode(trans, root, inode);
1575         }
1576         dir->i_sb->s_dirt = 1;
1577         btrfs_update_inode_block_group(trans, inode);
1578         btrfs_update_inode_block_group(trans, dir);
1579 out_unlock:
1580         nr = trans->blocks_used;
1581         btrfs_end_transaction(trans, root);
1582 fail:
1583         mutex_unlock(&root->fs_info->fs_mutex);
1584
1585         if (drop_inode) {
1586                 inode_dec_link_count(inode);
1587                 iput(inode);
1588         }
1589         btrfs_btree_balance_dirty(root, nr);
1590         btrfs_throttle(root);
1591         return err;
1592 }
1593
1594 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1595                         int mode, struct nameidata *nd)
1596 {
1597         struct btrfs_trans_handle *trans;
1598         struct btrfs_root *root = BTRFS_I(dir)->root;
1599         struct inode *inode = NULL;
1600         int err;
1601         int drop_inode = 0;
1602         unsigned long nr = 0;
1603         u64 objectid;
1604
1605         mutex_lock(&root->fs_info->fs_mutex);
1606         err = btrfs_check_free_space(root, 1, 0);
1607         if (err)
1608                 goto fail;
1609         trans = btrfs_start_transaction(root, 1);
1610         btrfs_set_trans_block_group(trans, dir);
1611
1612         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1613         if (err) {
1614                 err = -ENOSPC;
1615                 goto out_unlock;
1616         }
1617
1618         inode = btrfs_new_inode(trans, root, objectid,
1619                                 BTRFS_I(dir)->block_group, mode);
1620         err = PTR_ERR(inode);
1621         if (IS_ERR(inode))
1622                 goto out_unlock;
1623
1624         btrfs_set_trans_block_group(trans, inode);
1625         err = btrfs_add_nondir(trans, dentry, inode);
1626         if (err)
1627                 drop_inode = 1;
1628         else {
1629                 inode->i_mapping->a_ops = &btrfs_aops;
1630                 inode->i_fop = &btrfs_file_operations;
1631                 inode->i_op = &btrfs_file_inode_operations;
1632                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1633                                      inode->i_mapping, GFP_NOFS);
1634                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1635         }
1636         dir->i_sb->s_dirt = 1;
1637         btrfs_update_inode_block_group(trans, inode);
1638         btrfs_update_inode_block_group(trans, dir);
1639 out_unlock:
1640         nr = trans->blocks_used;
1641         btrfs_end_transaction(trans, root);
1642 fail:
1643         mutex_unlock(&root->fs_info->fs_mutex);
1644
1645         if (drop_inode) {
1646                 inode_dec_link_count(inode);
1647                 iput(inode);
1648         }
1649         btrfs_btree_balance_dirty(root, nr);
1650         btrfs_throttle(root);
1651         return err;
1652 }
1653
1654 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1655                       struct dentry *dentry)
1656 {
1657         struct btrfs_trans_handle *trans;
1658         struct btrfs_root *root = BTRFS_I(dir)->root;
1659         struct inode *inode = old_dentry->d_inode;
1660         unsigned long nr = 0;
1661         int err;
1662         int drop_inode = 0;
1663
1664         if (inode->i_nlink == 0)
1665                 return -ENOENT;
1666
1667 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1668         inode->i_nlink++;
1669 #else
1670         inc_nlink(inode);
1671 #endif
1672         mutex_lock(&root->fs_info->fs_mutex);
1673         err = btrfs_check_free_space(root, 1, 0);
1674         if (err)
1675                 goto fail;
1676         trans = btrfs_start_transaction(root, 1);
1677
1678         btrfs_set_trans_block_group(trans, dir);
1679         atomic_inc(&inode->i_count);
1680         err = btrfs_add_nondir(trans, dentry, inode);
1681
1682         if (err)
1683                 drop_inode = 1;
1684
1685         dir->i_sb->s_dirt = 1;
1686         btrfs_update_inode_block_group(trans, dir);
1687         err = btrfs_update_inode(trans, root, inode);
1688
1689         if (err)
1690                 drop_inode = 1;
1691
1692         nr = trans->blocks_used;
1693         btrfs_end_transaction(trans, root);
1694 fail:
1695         mutex_unlock(&root->fs_info->fs_mutex);
1696
1697         if (drop_inode) {
1698                 inode_dec_link_count(inode);
1699                 iput(inode);
1700         }
1701         btrfs_btree_balance_dirty(root, nr);
1702         btrfs_throttle(root);
1703         return err;
1704 }
1705
1706 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1707 {
1708         struct inode *inode;
1709         struct btrfs_trans_handle *trans;
1710         struct btrfs_root *root = BTRFS_I(dir)->root;
1711         int err = 0;
1712         int drop_on_err = 0;
1713         u64 objectid;
1714         unsigned long nr = 1;
1715
1716         mutex_lock(&root->fs_info->fs_mutex);
1717         err = btrfs_check_free_space(root, 1, 0);
1718         if (err)
1719                 goto out_unlock;
1720
1721         trans = btrfs_start_transaction(root, 1);
1722         btrfs_set_trans_block_group(trans, dir);
1723
1724         if (IS_ERR(trans)) {
1725                 err = PTR_ERR(trans);
1726                 goto out_unlock;
1727         }
1728
1729         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1730         if (err) {
1731                 err = -ENOSPC;
1732                 goto out_unlock;
1733         }
1734
1735         inode = btrfs_new_inode(trans, root, objectid,
1736                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1737         if (IS_ERR(inode)) {
1738                 err = PTR_ERR(inode);
1739                 goto out_fail;
1740         }
1741
1742         drop_on_err = 1;
1743         inode->i_op = &btrfs_dir_inode_operations;
1744         inode->i_fop = &btrfs_dir_file_operations;
1745         btrfs_set_trans_block_group(trans, inode);
1746
1747         inode->i_size = 0;
1748         err = btrfs_update_inode(trans, root, inode);
1749         if (err)
1750                 goto out_fail;
1751
1752         err = btrfs_add_link(trans, dentry, inode);
1753         if (err)
1754                 goto out_fail;
1755
1756         d_instantiate(dentry, inode);
1757         drop_on_err = 0;
1758         dir->i_sb->s_dirt = 1;
1759         btrfs_update_inode_block_group(trans, inode);
1760         btrfs_update_inode_block_group(trans, dir);
1761
1762 out_fail:
1763         nr = trans->blocks_used;
1764         btrfs_end_transaction(trans, root);
1765
1766 out_unlock:
1767         mutex_unlock(&root->fs_info->fs_mutex);
1768         if (drop_on_err)
1769                 iput(inode);
1770         btrfs_btree_balance_dirty(root, nr);
1771         btrfs_throttle(root);
1772         return err;
1773 }
1774
1775 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1776                                     size_t page_offset, u64 start, u64 end,
1777                                     int create)
1778 {
1779         int ret;
1780         int err = 0;
1781         u64 bytenr;
1782         u64 extent_start = 0;
1783         u64 extent_end = 0;
1784         u64 objectid = inode->i_ino;
1785         u32 found_type;
1786         int failed_insert = 0;
1787         struct btrfs_path *path;
1788         struct btrfs_root *root = BTRFS_I(inode)->root;
1789         struct btrfs_file_extent_item *item;
1790         struct extent_buffer *leaf;
1791         struct btrfs_key found_key;
1792         struct extent_map *em = NULL;
1793         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1794         struct btrfs_trans_handle *trans = NULL;
1795
1796         path = btrfs_alloc_path();
1797         BUG_ON(!path);
1798         mutex_lock(&root->fs_info->fs_mutex);
1799
1800 again:
1801         em = lookup_extent_mapping(em_tree, start, end);
1802         if (em) {
1803                 if (em->start > start) {
1804                         printk("get_extent start %Lu em start %Lu\n",
1805                                start, em->start);
1806                         WARN_ON(1);
1807                 }
1808                 goto out;
1809         }
1810         if (!em) {
1811                 em = alloc_extent_map(GFP_NOFS);
1812                 if (!em) {
1813                         err = -ENOMEM;
1814                         goto out;
1815                 }
1816                 em->start = EXTENT_MAP_HOLE;
1817                 em->end = EXTENT_MAP_HOLE;
1818         }
1819         em->bdev = inode->i_sb->s_bdev;
1820         ret = btrfs_lookup_file_extent(trans, root, path,
1821                                        objectid, start, trans != NULL);
1822         if (ret < 0) {
1823                 err = ret;
1824                 goto out;
1825         }
1826
1827         if (ret != 0) {
1828                 if (path->slots[0] == 0)
1829                         goto not_found;
1830                 path->slots[0]--;
1831         }
1832
1833         leaf = path->nodes[0];
1834         item = btrfs_item_ptr(leaf, path->slots[0],
1835                               struct btrfs_file_extent_item);
1836         /* are we inside the extent that was found? */
1837         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1838         found_type = btrfs_key_type(&found_key);
1839         if (found_key.objectid != objectid ||
1840             found_type != BTRFS_EXTENT_DATA_KEY) {
1841                 goto not_found;
1842         }
1843
1844         found_type = btrfs_file_extent_type(leaf, item);
1845         extent_start = found_key.offset;
1846         if (found_type == BTRFS_FILE_EXTENT_REG) {
1847                 extent_end = extent_start +
1848                        btrfs_file_extent_num_bytes(leaf, item);
1849                 err = 0;
1850                 if (start < extent_start || start >= extent_end) {
1851                         em->start = start;
1852                         if (start < extent_start) {
1853                                 if (end < extent_start)
1854                                         goto not_found;
1855                                 em->end = extent_end - 1;
1856                         } else {
1857                                 em->end = end;
1858                         }
1859                         goto not_found_em;
1860                 }
1861                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1862                 if (bytenr == 0) {
1863                         em->start = extent_start;
1864                         em->end = extent_end - 1;
1865                         em->block_start = EXTENT_MAP_HOLE;
1866                         em->block_end = EXTENT_MAP_HOLE;
1867                         goto insert;
1868                 }
1869                 bytenr += btrfs_file_extent_offset(leaf, item);
1870                 em->block_start = bytenr;
1871                 em->block_end = em->block_start +
1872                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1873                 em->start = extent_start;
1874                 em->end = extent_end - 1;
1875                 goto insert;
1876         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1877                 unsigned long ptr;
1878                 char *map;
1879                 size_t size;
1880                 size_t extent_offset;
1881                 size_t copy_size;
1882
1883                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1884                                                     path->slots[0]));
1885                 extent_end = (extent_start + size - 1) |
1886                         ((u64)root->sectorsize - 1);
1887                 if (start < extent_start || start >= extent_end) {
1888                         em->start = start;
1889                         if (start < extent_start) {
1890                                 if (end < extent_start)
1891                                         goto not_found;
1892                                 em->end = extent_end;
1893                         } else {
1894                                 em->end = end;
1895                         }
1896                         goto not_found_em;
1897                 }
1898                 em->block_start = EXTENT_MAP_INLINE;
1899                 em->block_end = EXTENT_MAP_INLINE;
1900
1901                 if (!page) {
1902                         em->start = extent_start;
1903                         em->end = extent_start + size - 1;
1904                         goto out;
1905                 }
1906
1907                 extent_offset = ((u64)page->index << PAGE_CACHE_SHIFT) -
1908                         extent_start + page_offset;
1909                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1910                                 size - extent_offset);
1911                 em->start = extent_start + extent_offset;
1912                 em->end = (em->start + copy_size -1) |
1913                         ((u64)root->sectorsize -1);
1914                 map = kmap(page);
1915                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1916                 if (create == 0 && !PageUptodate(page)) {
1917                         read_extent_buffer(leaf, map + page_offset, ptr,
1918                                            copy_size);
1919                         flush_dcache_page(page);
1920                 } else if (create && PageUptodate(page)) {
1921                         if (!trans) {
1922                                 kunmap(page);
1923                                 free_extent_map(em);
1924                                 em = NULL;
1925                                 btrfs_release_path(root, path);
1926                                 trans = btrfs_start_transaction(root, 1);
1927                                 goto again;
1928                         }
1929                         write_extent_buffer(leaf, map + page_offset, ptr,
1930                                             copy_size);
1931                         btrfs_mark_buffer_dirty(leaf);
1932                 }
1933                 kunmap(page);
1934                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1935                 goto insert;
1936         } else {
1937                 printk("unkknown found_type %d\n", found_type);
1938                 WARN_ON(1);
1939         }
1940 not_found:
1941         em->start = start;
1942         em->end = end;
1943 not_found_em:
1944         em->block_start = EXTENT_MAP_HOLE;
1945         em->block_end = EXTENT_MAP_HOLE;
1946 insert:
1947         btrfs_release_path(root, path);
1948         if (em->start > start || em->end < start) {
1949                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1950                 err = -EIO;
1951                 goto out;
1952         }
1953         ret = add_extent_mapping(em_tree, em);
1954         if (ret == -EEXIST) {
1955                 free_extent_map(em);
1956                 em = NULL;
1957                 if (0 && failed_insert == 1) {
1958                         btrfs_drop_extent_cache(inode, start, end);
1959                 }
1960                 failed_insert++;
1961                 if (failed_insert > 5) {
1962                         printk("failing to insert %Lu %Lu\n", start, end);
1963                         err = -EIO;
1964                         goto out;
1965                 }
1966                 goto again;
1967         }
1968         err = 0;
1969 out:
1970         btrfs_free_path(path);
1971         if (trans) {
1972                 ret = btrfs_end_transaction(trans, root);
1973                 if (!err)
1974                         err = ret;
1975         }
1976         mutex_unlock(&root->fs_info->fs_mutex);
1977         if (err) {
1978                 free_extent_map(em);
1979                 WARN_ON(1);
1980                 return ERR_PTR(err);
1981         }
1982         return em;
1983 }
1984
1985 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1986 {
1987         return extent_bmap(mapping, iblock, btrfs_get_extent);
1988 }
1989
1990 int btrfs_readpage(struct file *file, struct page *page)
1991 {
1992         struct extent_map_tree *tree;
1993         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1994         return extent_read_full_page(tree, page, btrfs_get_extent);
1995 }
1996
1997 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1998 {
1999         struct extent_map_tree *tree;
2000
2001
2002         if (current->flags & PF_MEMALLOC) {
2003                 redirty_page_for_writepage(wbc, page);
2004                 unlock_page(page);
2005                 return 0;
2006         }
2007         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2008         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2009 }
2010
2011 static int btrfs_writepages(struct address_space *mapping,
2012                             struct writeback_control *wbc)
2013 {
2014         struct extent_map_tree *tree;
2015         tree = &BTRFS_I(mapping->host)->extent_tree;
2016         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2017 }
2018
2019 static int
2020 btrfs_readpages(struct file *file, struct address_space *mapping,
2021                 struct list_head *pages, unsigned nr_pages)
2022 {
2023         struct extent_map_tree *tree;
2024         tree = &BTRFS_I(mapping->host)->extent_tree;
2025         return extent_readpages(tree, mapping, pages, nr_pages,
2026                                 btrfs_get_extent);
2027 }
2028
2029 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2030 {
2031         struct extent_map_tree *tree;
2032         int ret;
2033
2034         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2035         ret = try_release_extent_mapping(tree, page);
2036         if (ret == 1) {
2037                 ClearPagePrivate(page);
2038                 set_page_private(page, 0);
2039                 page_cache_release(page);
2040         }
2041         return ret;
2042 }
2043
2044 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2045 {
2046         struct extent_map_tree *tree;
2047
2048         tree = &BTRFS_I(page->mapping->host)->extent_tree;
2049         extent_invalidatepage(tree, page, offset);
2050         btrfs_releasepage(page, GFP_NOFS);
2051 }
2052
2053 /*
2054  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2055  * called from a page fault handler when a page is first dirtied. Hence we must
2056  * be careful to check for EOF conditions here. We set the page up correctly
2057  * for a written page which means we get ENOSPC checking when writing into
2058  * holes and correct delalloc and unwritten extent mapping on filesystems that
2059  * support these features.
2060  *
2061  * We are not allowed to take the i_mutex here so we have to play games to
2062  * protect against truncate races as the page could now be beyond EOF.  Because
2063  * vmtruncate() writes the inode size before removing pages, once we have the
2064  * page lock we can determine safely if the page is beyond EOF. If it is not
2065  * beyond EOF, then the page is guaranteed safe against truncation until we
2066  * unlock the page.
2067  */
2068 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2069 {
2070         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2071         struct btrfs_root *root = BTRFS_I(inode)->root;
2072         unsigned long end;
2073         loff_t size;
2074         int ret;
2075         u64 page_start;
2076
2077         mutex_lock(&root->fs_info->fs_mutex);
2078         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2079         mutex_unlock(&root->fs_info->fs_mutex);
2080         if (ret)
2081                 goto out;
2082
2083         ret = -EINVAL;
2084
2085         lock_page(page);
2086         wait_on_page_writeback(page);
2087         size = i_size_read(inode);
2088         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2089
2090         if ((page->mapping != inode->i_mapping) ||
2091             (page_start > size)) {
2092                 /* page got truncated out from underneath us */
2093                 goto out_unlock;
2094         }
2095
2096         /* page is wholly or partially inside EOF */
2097         if (page_start + PAGE_CACHE_SIZE > size)
2098                 end = size & ~PAGE_CACHE_MASK;
2099         else
2100                 end = PAGE_CACHE_SIZE;
2101
2102         ret = btrfs_cow_one_page(inode, page, end);
2103
2104 out_unlock:
2105         unlock_page(page);
2106 out:
2107         return ret;
2108 }
2109
2110 static void btrfs_truncate(struct inode *inode)
2111 {
2112         struct btrfs_root *root = BTRFS_I(inode)->root;
2113         int ret;
2114         struct btrfs_trans_handle *trans;
2115         unsigned long nr;
2116
2117         if (!S_ISREG(inode->i_mode))
2118                 return;
2119         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2120                 return;
2121
2122         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2123
2124         mutex_lock(&root->fs_info->fs_mutex);
2125         trans = btrfs_start_transaction(root, 1);
2126         btrfs_set_trans_block_group(trans, inode);
2127
2128         /* FIXME, add redo link to tree so we don't leak on crash */
2129         ret = btrfs_truncate_in_trans(trans, root, inode);
2130         btrfs_update_inode(trans, root, inode);
2131         nr = trans->blocks_used;
2132
2133         ret = btrfs_end_transaction(trans, root);
2134         BUG_ON(ret);
2135         mutex_unlock(&root->fs_info->fs_mutex);
2136         btrfs_btree_balance_dirty(root, nr);
2137         btrfs_throttle(root);
2138 }
2139
2140 static int noinline create_subvol(struct btrfs_root *root, char *name,
2141                                   int namelen)
2142 {
2143         struct btrfs_trans_handle *trans;
2144         struct btrfs_key key;
2145         struct btrfs_root_item root_item;
2146         struct btrfs_inode_item *inode_item;
2147         struct extent_buffer *leaf;
2148         struct btrfs_root *new_root = root;
2149         struct inode *inode;
2150         struct inode *dir;
2151         int ret;
2152         int err;
2153         u64 objectid;
2154         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2155         unsigned long nr = 1;
2156
2157         mutex_lock(&root->fs_info->fs_mutex);
2158         ret = btrfs_check_free_space(root, 1, 0);
2159         if (ret)
2160                 goto fail_commit;
2161
2162         trans = btrfs_start_transaction(root, 1);
2163         BUG_ON(!trans);
2164
2165         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2166                                        0, &objectid);
2167         if (ret)
2168                 goto fail;
2169
2170         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2171                                         objectid, trans->transid, 0, 0,
2172                                         0, 0);
2173         if (IS_ERR(leaf))
2174                 return PTR_ERR(leaf);
2175
2176         btrfs_set_header_nritems(leaf, 0);
2177         btrfs_set_header_level(leaf, 0);
2178         btrfs_set_header_bytenr(leaf, leaf->start);
2179         btrfs_set_header_generation(leaf, trans->transid);
2180         btrfs_set_header_owner(leaf, objectid);
2181
2182         write_extent_buffer(leaf, root->fs_info->fsid,
2183                             (unsigned long)btrfs_header_fsid(leaf),
2184                             BTRFS_FSID_SIZE);
2185         btrfs_mark_buffer_dirty(leaf);
2186
2187         inode_item = &root_item.inode;
2188         memset(inode_item, 0, sizeof(*inode_item));
2189         inode_item->generation = cpu_to_le64(1);
2190         inode_item->size = cpu_to_le64(3);
2191         inode_item->nlink = cpu_to_le32(1);
2192         inode_item->nblocks = cpu_to_le64(1);
2193         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2194
2195         btrfs_set_root_bytenr(&root_item, leaf->start);
2196         btrfs_set_root_level(&root_item, 0);
2197         btrfs_set_root_refs(&root_item, 1);
2198         btrfs_set_root_used(&root_item, 0);
2199
2200         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2201         root_item.drop_level = 0;
2202
2203         free_extent_buffer(leaf);
2204         leaf = NULL;
2205
2206         btrfs_set_root_dirid(&root_item, new_dirid);
2207
2208         key.objectid = objectid;
2209         key.offset = 1;
2210         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2211         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2212                                 &root_item);
2213         if (ret)
2214                 goto fail;
2215
2216         /*
2217          * insert the directory item
2218          */
2219         key.offset = (u64)-1;
2220         dir = root->fs_info->sb->s_root->d_inode;
2221         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2222                                     name, namelen, dir->i_ino, &key,
2223                                     BTRFS_FT_DIR);
2224         if (ret)
2225                 goto fail;
2226
2227         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2228                              name, namelen, objectid,
2229                              root->fs_info->sb->s_root->d_inode->i_ino);
2230         if (ret)
2231                 goto fail;
2232
2233         ret = btrfs_commit_transaction(trans, root);
2234         if (ret)
2235                 goto fail_commit;
2236
2237         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2238         BUG_ON(!new_root);
2239
2240         trans = btrfs_start_transaction(new_root, 1);
2241         BUG_ON(!trans);
2242
2243         inode = btrfs_new_inode(trans, new_root, new_dirid,
2244                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2245         if (IS_ERR(inode))
2246                 goto fail;
2247         inode->i_op = &btrfs_dir_inode_operations;
2248         inode->i_fop = &btrfs_dir_file_operations;
2249         new_root->inode = inode;
2250
2251         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2252                                      new_dirid);
2253         inode->i_nlink = 1;
2254         inode->i_size = 0;
2255         ret = btrfs_update_inode(trans, new_root, inode);
2256         if (ret)
2257                 goto fail;
2258 fail:
2259         nr = trans->blocks_used;
2260         err = btrfs_commit_transaction(trans, new_root);
2261         if (err && !ret)
2262                 ret = err;
2263 fail_commit:
2264         mutex_unlock(&root->fs_info->fs_mutex);
2265         btrfs_btree_balance_dirty(root, nr);
2266         btrfs_throttle(root);
2267         return ret;
2268 }
2269
2270 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2271 {
2272         struct btrfs_pending_snapshot *pending_snapshot;
2273         struct btrfs_trans_handle *trans;
2274         int ret;
2275         int err;
2276         unsigned long nr = 0;
2277
2278         if (!root->ref_cows)
2279                 return -EINVAL;
2280
2281         mutex_lock(&root->fs_info->fs_mutex);
2282         ret = btrfs_check_free_space(root, 1, 0);
2283         if (ret)
2284                 goto fail_unlock;
2285
2286         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2287         if (!pending_snapshot) {
2288                 ret = -ENOMEM;
2289                 goto fail_unlock;
2290         }
2291         pending_snapshot->name = kstrndup(name, namelen, GFP_NOFS);
2292         if (!pending_snapshot->name) {
2293                 ret = -ENOMEM;
2294                 kfree(pending_snapshot);
2295                 goto fail_unlock;
2296         }
2297         trans = btrfs_start_transaction(root, 1);
2298         BUG_ON(!trans);
2299
2300         pending_snapshot->root = root;
2301         list_add(&pending_snapshot->list,
2302                  &trans->transaction->pending_snapshots);
2303         ret = btrfs_update_inode(trans, root, root->inode);
2304         err = btrfs_commit_transaction(trans, root);
2305
2306 fail_unlock:
2307         mutex_unlock(&root->fs_info->fs_mutex);
2308         btrfs_btree_balance_dirty(root, nr);
2309         btrfs_throttle(root);
2310         return ret;
2311 }
2312
2313 unsigned long btrfs_force_ra(struct address_space *mapping,
2314                               struct file_ra_state *ra, struct file *file,
2315                               pgoff_t offset, pgoff_t last_index)
2316 {
2317         pgoff_t req_size;
2318
2319 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2320         req_size = last_index - offset + 1;
2321         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2322         return offset;
2323 #else
2324         req_size = min(last_index - offset + 1, (pgoff_t)128);
2325         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2326         return offset + req_size;
2327 #endif
2328 }
2329
2330 int btrfs_defrag_file(struct file *file) {
2331         struct inode *inode = fdentry(file)->d_inode;
2332         struct btrfs_root *root = BTRFS_I(inode)->root;
2333         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2334         struct page *page;
2335         unsigned long last_index;
2336         unsigned long ra_index = 0;
2337         u64 page_start;
2338         u64 page_end;
2339         u64 delalloc_start;
2340         u64 existing_delalloc;
2341         unsigned long i;
2342         int ret;
2343
2344         mutex_lock(&root->fs_info->fs_mutex);
2345         ret = btrfs_check_free_space(root, inode->i_size, 0);
2346         mutex_unlock(&root->fs_info->fs_mutex);
2347         if (ret)
2348                 return -ENOSPC;
2349
2350         mutex_lock(&inode->i_mutex);
2351         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2352         for (i = 0; i <= last_index; i++) {
2353                 if (i == ra_index) {
2354                         ra_index = btrfs_force_ra(inode->i_mapping,
2355                                                   &file->f_ra,
2356                                                   file, ra_index, last_index);
2357                 }
2358                 page = grab_cache_page(inode->i_mapping, i);
2359                 if (!page)
2360                         goto out_unlock;
2361                 if (!PageUptodate(page)) {
2362                         btrfs_readpage(NULL, page);
2363                         lock_page(page);
2364                         if (!PageUptodate(page)) {
2365                                 unlock_page(page);
2366                                 page_cache_release(page);
2367                                 goto out_unlock;
2368                         }
2369                 }
2370                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2371                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2372
2373                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2374                 delalloc_start = page_start;
2375                 existing_delalloc =
2376                         count_range_bits(&BTRFS_I(inode)->extent_tree,
2377                                          &delalloc_start, page_end,
2378                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2379                 set_extent_delalloc(em_tree, page_start,
2380                                     page_end, GFP_NOFS);
2381
2382                 spin_lock(&root->fs_info->delalloc_lock);
2383                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2384                                                  existing_delalloc;
2385                 spin_unlock(&root->fs_info->delalloc_lock);
2386
2387                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2388                 set_page_dirty(page);
2389                 unlock_page(page);
2390                 page_cache_release(page);
2391                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2392         }
2393
2394 out_unlock:
2395         mutex_unlock(&inode->i_mutex);
2396         return 0;
2397 }
2398
2399 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2400 {
2401         u64 new_size;
2402         u64 old_size;
2403         struct btrfs_ioctl_vol_args *vol_args;
2404         struct btrfs_trans_handle *trans;
2405         char *sizestr;
2406         int ret = 0;
2407         int namelen;
2408         int mod = 0;
2409
2410         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2411
2412         if (!vol_args)
2413                 return -ENOMEM;
2414
2415         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2416                 ret = -EFAULT;
2417                 goto out;
2418         }
2419         namelen = strlen(vol_args->name);
2420         if (namelen > BTRFS_VOL_NAME_MAX) {
2421                 ret = -EINVAL;
2422                 goto out;
2423         }
2424
2425         sizestr = vol_args->name;
2426         if (!strcmp(sizestr, "max"))
2427                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2428         else {
2429                 if (sizestr[0] == '-') {
2430                         mod = -1;
2431                         sizestr++;
2432                 } else if (sizestr[0] == '+') {
2433                         mod = 1;
2434                         sizestr++;
2435                 }
2436                 new_size = btrfs_parse_size(sizestr);
2437                 if (new_size == 0) {
2438                         ret = -EINVAL;
2439                         goto out;
2440                 }
2441         }
2442
2443         mutex_lock(&root->fs_info->fs_mutex);
2444         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2445
2446         if (mod < 0) {
2447                 if (new_size > old_size) {
2448                         ret = -EINVAL;
2449                         goto out_unlock;
2450                 }
2451                 new_size = old_size - new_size;
2452         } else if (mod > 0) {
2453                 new_size = old_size + new_size;
2454         }
2455
2456         if (new_size < 256 * 1024 * 1024) {
2457                 ret = -EINVAL;
2458                 goto out_unlock;
2459         }
2460         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2461                 ret = -EFBIG;
2462                 goto out_unlock;
2463         }
2464
2465         do_div(new_size, root->sectorsize);
2466         new_size *= root->sectorsize;
2467
2468 printk("new size is %Lu\n", new_size);
2469         if (new_size > old_size) {
2470                 trans = btrfs_start_transaction(root, 1);
2471                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2472                 btrfs_commit_transaction(trans, root);
2473         } else {
2474                 ret = btrfs_shrink_extent_tree(root, new_size);
2475         }
2476
2477 out_unlock:
2478         mutex_unlock(&root->fs_info->fs_mutex);
2479 out:
2480         kfree(vol_args);
2481         return ret;
2482 }
2483
2484 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2485                                             void __user *arg)
2486 {
2487         struct btrfs_ioctl_vol_args *vol_args;
2488         struct btrfs_dir_item *di;
2489         struct btrfs_path *path;
2490         u64 root_dirid;
2491         int namelen;
2492         int ret;
2493
2494         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2495
2496         if (!vol_args)
2497                 return -ENOMEM;
2498
2499         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2500                 ret = -EFAULT;
2501                 goto out;
2502         }
2503
2504         namelen = strlen(vol_args->name);
2505         if (namelen > BTRFS_VOL_NAME_MAX) {
2506                 ret = -EINVAL;
2507                 goto out;
2508         }
2509         if (strchr(vol_args->name, '/')) {
2510                 ret = -EINVAL;
2511                 goto out;
2512         }
2513
2514         path = btrfs_alloc_path();
2515         if (!path) {
2516                 ret = -ENOMEM;
2517                 goto out;
2518         }
2519
2520         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2521         mutex_lock(&root->fs_info->fs_mutex);
2522         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2523                             path, root_dirid,
2524                             vol_args->name, namelen, 0);
2525         mutex_unlock(&root->fs_info->fs_mutex);
2526         btrfs_free_path(path);
2527
2528         if (di && !IS_ERR(di)) {
2529                 ret = -EEXIST;
2530                 goto out;
2531         }
2532
2533         if (IS_ERR(di)) {
2534                 ret = PTR_ERR(di);
2535                 goto out;
2536         }
2537
2538         if (root == root->fs_info->tree_root)
2539                 ret = create_subvol(root, vol_args->name, namelen);
2540         else
2541                 ret = create_snapshot(root, vol_args->name, namelen);
2542 out:
2543         kfree(vol_args);
2544         return ret;
2545 }
2546
2547 static int btrfs_ioctl_defrag(struct file *file)
2548 {
2549         struct inode *inode = fdentry(file)->d_inode;
2550         struct btrfs_root *root = BTRFS_I(inode)->root;
2551
2552         switch (inode->i_mode & S_IFMT) {
2553         case S_IFDIR:
2554                 mutex_lock(&root->fs_info->fs_mutex);
2555                 btrfs_defrag_root(root, 0);
2556                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2557                 mutex_unlock(&root->fs_info->fs_mutex);
2558                 break;
2559         case S_IFREG:
2560                 btrfs_defrag_file(file);
2561                 break;
2562         }
2563
2564         return 0;
2565 }
2566
2567 long btrfs_ioctl(struct file *file, unsigned int
2568                 cmd, unsigned long arg)
2569 {
2570         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2571
2572         switch (cmd) {
2573         case BTRFS_IOC_SNAP_CREATE:
2574                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2575         case BTRFS_IOC_DEFRAG:
2576                 return btrfs_ioctl_defrag(file);
2577         case BTRFS_IOC_RESIZE:
2578                 return btrfs_ioctl_resize(root, (void __user *)arg);
2579         }
2580
2581         return -ENOTTY;
2582 }
2583
2584 /*
2585  * Called inside transaction, so use GFP_NOFS
2586  */
2587 struct inode *btrfs_alloc_inode(struct super_block *sb)
2588 {
2589         struct btrfs_inode *ei;
2590
2591         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2592         if (!ei)
2593                 return NULL;
2594         ei->last_trans = 0;
2595         ei->ordered_trans = 0;
2596         return &ei->vfs_inode;
2597 }
2598
2599 void btrfs_destroy_inode(struct inode *inode)
2600 {
2601         WARN_ON(!list_empty(&inode->i_dentry));
2602         WARN_ON(inode->i_data.nrpages);
2603
2604         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2605 }
2606
2607 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2608 static void init_once(struct kmem_cache * cachep, void *foo)
2609 #else
2610 static void init_once(void * foo, struct kmem_cache * cachep,
2611                       unsigned long flags)
2612 #endif
2613 {
2614         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2615
2616         inode_init_once(&ei->vfs_inode);
2617 }
2618
2619 void btrfs_destroy_cachep(void)
2620 {
2621         if (btrfs_inode_cachep)
2622                 kmem_cache_destroy(btrfs_inode_cachep);
2623         if (btrfs_trans_handle_cachep)
2624                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2625         if (btrfs_transaction_cachep)
2626                 kmem_cache_destroy(btrfs_transaction_cachep);
2627         if (btrfs_bit_radix_cachep)
2628                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2629         if (btrfs_path_cachep)
2630                 kmem_cache_destroy(btrfs_path_cachep);
2631 }
2632
2633 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2634                                        unsigned long extra_flags,
2635 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2636                                        void (*ctor)(struct kmem_cache *, void *)
2637 #else
2638                                        void (*ctor)(void *, struct kmem_cache *,
2639                                                     unsigned long)
2640 #endif
2641                                      )
2642 {
2643         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2644                                  SLAB_MEM_SPREAD | extra_flags), ctor
2645 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2646                                  ,NULL
2647 #endif
2648                                 );
2649 }
2650
2651 int btrfs_init_cachep(void)
2652 {
2653         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2654                                           sizeof(struct btrfs_inode),
2655                                           0, init_once);
2656         if (!btrfs_inode_cachep)
2657                 goto fail;
2658         btrfs_trans_handle_cachep =
2659                         btrfs_cache_create("btrfs_trans_handle_cache",
2660                                            sizeof(struct btrfs_trans_handle),
2661                                            0, NULL);
2662         if (!btrfs_trans_handle_cachep)
2663                 goto fail;
2664         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2665                                              sizeof(struct btrfs_transaction),
2666                                              0, NULL);
2667         if (!btrfs_transaction_cachep)
2668                 goto fail;
2669         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2670                                          sizeof(struct btrfs_path),
2671                                          0, NULL);
2672         if (!btrfs_path_cachep)
2673                 goto fail;
2674         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2675                                               SLAB_DESTROY_BY_RCU, NULL);
2676         if (!btrfs_bit_radix_cachep)
2677                 goto fail;
2678         return 0;
2679 fail:
2680         btrfs_destroy_cachep();
2681         return -ENOMEM;
2682 }
2683
2684 static int btrfs_getattr(struct vfsmount *mnt,
2685                          struct dentry *dentry, struct kstat *stat)
2686 {
2687         struct inode *inode = dentry->d_inode;
2688         generic_fillattr(inode, stat);
2689         stat->blksize = PAGE_CACHE_SIZE;
2690         return 0;
2691 }
2692
2693 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2694                            struct inode * new_dir,struct dentry *new_dentry)
2695 {
2696         struct btrfs_trans_handle *trans;
2697         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2698         struct inode *new_inode = new_dentry->d_inode;
2699         struct inode *old_inode = old_dentry->d_inode;
2700         struct timespec ctime = CURRENT_TIME;
2701         struct btrfs_path *path;
2702         int ret;
2703
2704         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2705             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2706                 return -ENOTEMPTY;
2707         }
2708
2709         mutex_lock(&root->fs_info->fs_mutex);
2710         ret = btrfs_check_free_space(root, 1, 0);
2711         if (ret)
2712                 goto out_unlock;
2713
2714         trans = btrfs_start_transaction(root, 1);
2715
2716         btrfs_set_trans_block_group(trans, new_dir);
2717         path = btrfs_alloc_path();
2718         if (!path) {
2719                 ret = -ENOMEM;
2720                 goto out_fail;
2721         }
2722
2723         old_dentry->d_inode->i_nlink++;
2724         old_dir->i_ctime = old_dir->i_mtime = ctime;
2725         new_dir->i_ctime = new_dir->i_mtime = ctime;
2726         old_inode->i_ctime = ctime;
2727
2728         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2729         if (ret)
2730                 goto out_fail;
2731
2732         if (new_inode) {
2733                 new_inode->i_ctime = CURRENT_TIME;
2734                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2735                 if (ret)
2736                         goto out_fail;
2737         }
2738         ret = btrfs_add_link(trans, new_dentry, old_inode);
2739         if (ret)
2740                 goto out_fail;
2741
2742 out_fail:
2743         btrfs_free_path(path);
2744         btrfs_end_transaction(trans, root);
2745 out_unlock:
2746         mutex_unlock(&root->fs_info->fs_mutex);
2747         return ret;
2748 }
2749
2750 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2751                          const char *symname)
2752 {
2753         struct btrfs_trans_handle *trans;
2754         struct btrfs_root *root = BTRFS_I(dir)->root;
2755         struct btrfs_path *path;
2756         struct btrfs_key key;
2757         struct inode *inode = NULL;
2758         int err;
2759         int drop_inode = 0;
2760         u64 objectid;
2761         int name_len;
2762         int datasize;
2763         unsigned long ptr;
2764         struct btrfs_file_extent_item *ei;
2765         struct extent_buffer *leaf;
2766         unsigned long nr = 0;
2767
2768         name_len = strlen(symname) + 1;
2769         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2770                 return -ENAMETOOLONG;
2771
2772         mutex_lock(&root->fs_info->fs_mutex);
2773         err = btrfs_check_free_space(root, 1, 0);
2774         if (err)
2775                 goto out_fail;
2776
2777         trans = btrfs_start_transaction(root, 1);
2778         btrfs_set_trans_block_group(trans, dir);
2779
2780         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2781         if (err) {
2782                 err = -ENOSPC;
2783                 goto out_unlock;
2784         }
2785
2786         inode = btrfs_new_inode(trans, root, objectid,
2787                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2788         err = PTR_ERR(inode);
2789         if (IS_ERR(inode))
2790                 goto out_unlock;
2791
2792         btrfs_set_trans_block_group(trans, inode);
2793         err = btrfs_add_nondir(trans, dentry, inode);
2794         if (err)
2795                 drop_inode = 1;
2796         else {
2797                 inode->i_mapping->a_ops = &btrfs_aops;
2798                 inode->i_fop = &btrfs_file_operations;
2799                 inode->i_op = &btrfs_file_inode_operations;
2800                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2801                                      inode->i_mapping, GFP_NOFS);
2802                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2803         }
2804         dir->i_sb->s_dirt = 1;
2805         btrfs_update_inode_block_group(trans, inode);
2806         btrfs_update_inode_block_group(trans, dir);
2807         if (drop_inode)
2808                 goto out_unlock;
2809
2810         path = btrfs_alloc_path();
2811         BUG_ON(!path);
2812         key.objectid = inode->i_ino;
2813         key.offset = 0;
2814         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2815         datasize = btrfs_file_extent_calc_inline_size(name_len);
2816         err = btrfs_insert_empty_item(trans, root, path, &key,
2817                                       datasize);
2818         if (err) {
2819                 drop_inode = 1;
2820                 goto out_unlock;
2821         }
2822         leaf = path->nodes[0];
2823         ei = btrfs_item_ptr(leaf, path->slots[0],
2824                             struct btrfs_file_extent_item);
2825         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2826         btrfs_set_file_extent_type(leaf, ei,
2827                                    BTRFS_FILE_EXTENT_INLINE);
2828         ptr = btrfs_file_extent_inline_start(ei);
2829         write_extent_buffer(leaf, symname, ptr, name_len);
2830         btrfs_mark_buffer_dirty(leaf);
2831         btrfs_free_path(path);
2832
2833         inode->i_op = &btrfs_symlink_inode_operations;
2834         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2835         inode->i_size = name_len - 1;
2836         err = btrfs_update_inode(trans, root, inode);
2837         if (err)
2838                 drop_inode = 1;
2839
2840 out_unlock:
2841         nr = trans->blocks_used;
2842         btrfs_end_transaction(trans, root);
2843 out_fail:
2844         mutex_unlock(&root->fs_info->fs_mutex);
2845         if (drop_inode) {
2846                 inode_dec_link_count(inode);
2847                 iput(inode);
2848         }
2849         btrfs_btree_balance_dirty(root, nr);
2850         btrfs_throttle(root);
2851         return err;
2852 }
2853
2854 static struct inode_operations btrfs_dir_inode_operations = {
2855         .lookup         = btrfs_lookup,
2856         .create         = btrfs_create,
2857         .unlink         = btrfs_unlink,
2858         .link           = btrfs_link,
2859         .mkdir          = btrfs_mkdir,
2860         .rmdir          = btrfs_rmdir,
2861         .rename         = btrfs_rename,
2862         .symlink        = btrfs_symlink,
2863         .setattr        = btrfs_setattr,
2864         .mknod          = btrfs_mknod,
2865         .setxattr       = generic_setxattr,
2866         .getxattr       = generic_getxattr,
2867         .listxattr      = btrfs_listxattr,
2868         .removexattr    = generic_removexattr,
2869 };
2870
2871 static struct inode_operations btrfs_dir_ro_inode_operations = {
2872         .lookup         = btrfs_lookup,
2873 };
2874
2875 static struct file_operations btrfs_dir_file_operations = {
2876         .llseek         = generic_file_llseek,
2877         .read           = generic_read_dir,
2878         .readdir        = btrfs_readdir,
2879         .unlocked_ioctl = btrfs_ioctl,
2880 #ifdef CONFIG_COMPAT
2881         .compat_ioctl   = btrfs_ioctl,
2882 #endif
2883 };
2884
2885 static struct extent_map_ops btrfs_extent_map_ops = {
2886         .fill_delalloc = run_delalloc_range,
2887         .writepage_io_hook = btrfs_writepage_io_hook,
2888         .readpage_io_hook = btrfs_readpage_io_hook,
2889         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2890 };
2891
2892 static struct address_space_operations btrfs_aops = {
2893         .readpage       = btrfs_readpage,
2894         .writepage      = btrfs_writepage,
2895         .writepages     = btrfs_writepages,
2896         .readpages      = btrfs_readpages,
2897         .sync_page      = block_sync_page,
2898         .bmap           = btrfs_bmap,
2899         .invalidatepage = btrfs_invalidatepage,
2900         .releasepage    = btrfs_releasepage,
2901         .set_page_dirty = __set_page_dirty_nobuffers,
2902 };
2903
2904 static struct address_space_operations btrfs_symlink_aops = {
2905         .readpage       = btrfs_readpage,
2906         .writepage      = btrfs_writepage,
2907         .invalidatepage = btrfs_invalidatepage,
2908         .releasepage    = btrfs_releasepage,
2909 };
2910
2911 static struct inode_operations btrfs_file_inode_operations = {
2912         .truncate       = btrfs_truncate,
2913         .getattr        = btrfs_getattr,
2914         .setattr        = btrfs_setattr,
2915         .setxattr       = generic_setxattr,
2916         .getxattr       = generic_getxattr,
2917         .listxattr      = btrfs_listxattr,
2918         .removexattr    = generic_removexattr,
2919 };
2920
2921 static struct inode_operations btrfs_special_inode_operations = {
2922         .getattr        = btrfs_getattr,
2923         .setattr        = btrfs_setattr,
2924 };
2925
2926 static struct inode_operations btrfs_symlink_inode_operations = {
2927         .readlink       = generic_readlink,
2928         .follow_link    = page_follow_link_light,
2929         .put_link       = page_put_link,
2930 };