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