fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/buffer_head.h>
  20 #include <linux/fs.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/string.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mpage.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/version.h>
  35 #include <linux/xattr.h>
  36 #include "ctree.h"
  37 #include "disk-io.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "ioctl.h"
  41 #include "print-tree.h"
  42
  43 struct btrfs_iget_args {
  44         u64 ino;
  45         struct btrfs_root *root;
  46 };
  47
  48 static struct inode_operations btrfs_dir_inode_operations;
  49 static struct inode_operations btrfs_symlink_inode_operations;
  50 static struct inode_operations btrfs_dir_ro_inode_operations;
  51 static struct inode_operations btrfs_special_inode_operations;
  52 static struct inode_operations btrfs_file_inode_operations;
  53 static struct address_space_operations btrfs_aops;
  54 static struct address_space_operations btrfs_symlink_aops;
  55 static struct file_operations btrfs_dir_file_operations;
  56 static struct extent_io_ops btrfs_extent_io_ops;
  57
  58 static struct kmem_cache *btrfs_inode_cachep;
  59 struct kmem_cache *btrfs_trans_handle_cachep;
  60 struct kmem_cache *btrfs_transaction_cachep;
  61 struct kmem_cache *btrfs_bit_radix_cachep;
  62 struct kmem_cache *btrfs_path_cachep;
  63
  64 #define S_SHIFT 12
  65 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  66         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  67         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  68         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  69         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  70         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  71         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  72         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  73 };
  74
  75 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  76                            int for_del)
  77 {
  78         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  79         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  80         u64 thresh;
  81         int ret = 0;
  82
  83         return 0;
  84
  85         if (for_del)
  86                 thresh = total * 90;
  87         else
  88                 thresh = total * 85;
  89
  90         do_div(thresh, 100);
  91
  92         spin_lock(&root->fs_info->delalloc_lock);
  93         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  94                 ret = -ENOSPC;
  95         spin_unlock(&root->fs_info->delalloc_lock);
  96         return ret;
  97 }
  98
  99 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 100 {
 101         struct btrfs_root *root = BTRFS_I(inode)->root;
 102         struct btrfs_trans_handle *trans;
 103         u64 alloc_hint = 0;
 104         u64 num_bytes;
 105         u64 cur_alloc_size;
 106         u64 blocksize = root->sectorsize;
 107         u64 orig_start = start;
 108         u64 orig_num_bytes;
 109         struct btrfs_key ins;
 110         int ret;
 111
 112         trans = btrfs_start_transaction(root, 1);
 113         BUG_ON(!trans);
 114         btrfs_set_trans_block_group(trans, inode);
 115
 116         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 117         num_bytes = max(blocksize,  num_bytes);
 118         ret = btrfs_drop_extents(trans, root, inode,
 119                                  start, start + num_bytes, start, &alloc_hint);
 120         orig_num_bytes = num_bytes;
 121
 122         if (alloc_hint == EXTENT_MAP_INLINE)
 123                 goto out;
 124
 125         while(num_bytes > 0) {
 126                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 127                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 128                                          root->root_key.objectid,
 129                                          trans->transid,
 130                                          inode->i_ino, start, 0,
 131                                          alloc_hint, (u64)-1, &ins, 1);
 132                 if (ret) {
 133                         WARN_ON(1);
 134                         goto out;
 135                 }
 136                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 137                                                start, ins.objectid, ins.offset,
 138                                                ins.offset);
 139                 btrfs_check_file(root, inode);
 140                 num_bytes -= cur_alloc_size;
 141                 alloc_hint = ins.objectid + ins.offset;
 142                 start += cur_alloc_size;
 143         }
 144         btrfs_drop_extent_cache(inode, orig_start,
 145                                 orig_start + orig_num_bytes - 1);
 146         btrfs_add_ordered_inode(inode);
 147 out:
 148         btrfs_end_transaction(trans, root);
 149         return ret;
 150 }
 151
 152 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 153 {
 154         u64 extent_start;
 155         u64 extent_end;
 156         u64 bytenr;
 157         u64 cow_end;
 158         u64 loops = 0;
 159         u64 total_fs_bytes;
 160         struct btrfs_root *root = BTRFS_I(inode)->root;
 161         struct extent_buffer *leaf;
 162         int found_type;
 163         struct btrfs_path *path;
 164         struct btrfs_file_extent_item *item;
 165         int ret;
 166         int err;
 167         struct btrfs_key found_key;
 168
 169         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 170         path = btrfs_alloc_path();
 171         BUG_ON(!path);
 172 again:
 173         ret = btrfs_lookup_file_extent(NULL, root, path,
 174                                        inode->i_ino, start, 0);
 175         if (ret < 0) {
 176                 btrfs_free_path(path);
 177                 return ret;
 178         }
 179
 180         cow_end = end;
 181         if (ret != 0) {
 182                 if (path->slots[0] == 0)
 183                         goto not_found;
 184                 path->slots[0]--;
 185         }
 186
 187         leaf = path->nodes[0];
 188         item = btrfs_item_ptr(leaf, path->slots[0],
 189                               struct btrfs_file_extent_item);
 190
 191         /* are we inside the extent that was found? */
 192         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 193         found_type = btrfs_key_type(&found_key);
 194         if (found_key.objectid != inode->i_ino ||
 195             found_type != BTRFS_EXTENT_DATA_KEY) {
 196                 goto not_found;
 197         }
 198
 199         found_type = btrfs_file_extent_type(leaf, item);
 200         extent_start = found_key.offset;
 201         if (found_type == BTRFS_FILE_EXTENT_REG) {
 202                 u64 extent_num_bytes;
 203
 204                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 205                 extent_end = extent_start + extent_num_bytes;
 206                 err = 0;
 207
 208                 if (loops && start != extent_start)
 209                         goto not_found;
 210
 211                 if (start < extent_start || start >= extent_end)
 212                         goto not_found;
 213
 214                 cow_end = min(end, extent_end - 1);
 215                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 216                 if (bytenr == 0)
 217                         goto not_found;
 218
 219                 /*
 220                  * we may be called by the resizer, make sure we're inside
 221                  * the limits of the FS
 222                  */
 223                 if (bytenr + extent_num_bytes > total_fs_bytes)
 224                         goto not_found;
 225
 226                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 227                         goto not_found;
 228                 }
 229
 230                 start = extent_end;
 231         } else {
 232                 goto not_found;
 233         }
 234 loop:
 235         if (start > end) {
 236                 btrfs_free_path(path);
 237                 return 0;
 238         }
 239         btrfs_release_path(root, path);
 240         loops++;
 241         goto again;
 242
 243 not_found:
 244         cow_file_range(inode, start, cow_end);
 245         start = cow_end + 1;
 246         goto loop;
 247 }
 248
 249 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 250 {
 251         struct btrfs_root *root = BTRFS_I(inode)->root;
 252         u64 num_bytes;
 253         int ret;
 254         mutex_lock(&root->fs_info->fs_mutex);
 255         if (btrfs_test_opt(root, NODATACOW) ||
 256             btrfs_test_flag(inode, NODATACOW))
 257                 ret = run_delalloc_nocow(inode, start, end);
 258         else
 259                 ret = cow_file_range(inode, start, end);
 260
 261         spin_lock(&root->fs_info->delalloc_lock);
 262         num_bytes = end + 1 - start;
 263         if (root->fs_info->delalloc_bytes < num_bytes) {
 264                 printk("delalloc accounting error total %llu sub %llu\n",
 265                        root->fs_info->delalloc_bytes, num_bytes);
 266         } else {
 267                 root->fs_info->delalloc_bytes -= num_bytes;
 268         }
 269         spin_unlock(&root->fs_info->delalloc_lock);
 270
 271         mutex_unlock(&root->fs_info->fs_mutex);
 272         return ret;
 273 }
 274
 275 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 276 {
 277         struct inode *inode = page->mapping->host;
 278         struct btrfs_root *root = BTRFS_I(inode)->root;
 279         struct btrfs_trans_handle *trans;
 280         char *kaddr;
 281         int ret = 0;
 282         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 283         size_t offset = start - page_start;
 284         if (btrfs_test_opt(root, NODATASUM) ||
 285             btrfs_test_flag(inode, NODATASUM))
 286                 return 0;
 287         mutex_lock(&root->fs_info->fs_mutex);
 288         trans = btrfs_start_transaction(root, 1);
 289         btrfs_set_trans_block_group(trans, inode);
 290         kaddr = kmap(page);
 291         btrfs_csum_file_block(trans, root, inode, inode->i_ino,
 292                               start, kaddr + offset, end - start + 1);
 293         kunmap(page);
 294         ret = btrfs_end_transaction(trans, root);
 295         BUG_ON(ret);
 296         mutex_unlock(&root->fs_info->fs_mutex);
 297         return ret;
 298 }
 299
 300 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 301 {
 302         int ret = 0;
 303         struct inode *inode = page->mapping->host;
 304         struct btrfs_root *root = BTRFS_I(inode)->root;
 305         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 306         struct btrfs_csum_item *item;
 307         struct btrfs_path *path = NULL;
 308         u32 csum;
 309         if (btrfs_test_opt(root, NODATASUM) ||
 310             btrfs_test_flag(inode, NODATASUM))
 311                 return 0;
 312         mutex_lock(&root->fs_info->fs_mutex);
 313         path = btrfs_alloc_path();
 314         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 315         if (IS_ERR(item)) {
 316                 ret = PTR_ERR(item);
 317                 /* a csum that isn't present is a preallocated region. */
 318                 if (ret == -ENOENT || ret == -EFBIG)
 319                         ret = 0;
 320                 csum = 0;
 321                 goto out;
 322         }
 323         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 324                            BTRFS_CRC32_SIZE);
 325         set_state_private(io_tree, start, csum);
 326 out:
 327         if (path)
 328                 btrfs_free_path(path);
 329         mutex_unlock(&root->fs_info->fs_mutex);
 330         return ret;
 331 }
 332
 333 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end)
 334 {
 335         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 336         struct inode *inode = page->mapping->host;
 337         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 338         char *kaddr;
 339         u64 private;
 340         int ret;
 341         struct btrfs_root *root = BTRFS_I(inode)->root;
 342         u32 csum = ~(u32)0;
 343         unsigned long flags;
 344
 345         if (btrfs_test_opt(root, NODATASUM) ||
 346             btrfs_test_flag(inode, NODATASUM))
 347                 return 0;
 348
 349         ret = get_state_private(io_tree, start, &private);
 350         local_irq_save(flags);
 351         kaddr = kmap_atomic(page, KM_IRQ0);
 352         if (ret) {
 353                 goto zeroit;
 354         }
 355         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 356         btrfs_csum_final(csum, (char *)&csum);
 357         if (csum != private) {
 358                 goto zeroit;
 359         }
 360         kunmap_atomic(kaddr, KM_IRQ0);
 361         local_irq_restore(flags);
 362         return 0;
 363
 364 zeroit:
 365         printk("btrfs csum failed ino %lu off %llu\n",
 366                page->mapping->host->i_ino, (unsigned long long)start);
 367         memset(kaddr + offset, 1, end - start + 1);
 368         flush_dcache_page(page);
 369         kunmap_atomic(kaddr, KM_IRQ0);
 370         local_irq_restore(flags);
 371         return 0;
 372 }
 373
 374 void btrfs_read_locked_inode(struct inode *inode)
 375 {
 376         struct btrfs_path *path;
 377         struct extent_buffer *leaf;
 378         struct btrfs_inode_item *inode_item;
 379         struct btrfs_inode_timespec *tspec;
 380         struct btrfs_root *root = BTRFS_I(inode)->root;
 381         struct btrfs_key location;
 382         u64 alloc_group_block;
 383         u32 rdev;
 384         int ret;
 385
 386         path = btrfs_alloc_path();
 387         BUG_ON(!path);
 388         mutex_lock(&root->fs_info->fs_mutex);
 389         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 390
 391         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 392         if (ret)
 393                 goto make_bad;
 394
 395         leaf = path->nodes[0];
 396         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 397                                     struct btrfs_inode_item);
 398
 399         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 400         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 401         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 402         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 403         inode->i_size = btrfs_inode_size(leaf, inode_item);
 404
 405         tspec = btrfs_inode_atime(inode_item);
 406         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 407         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 408
 409         tspec = btrfs_inode_mtime(inode_item);
 410         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 411         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 412
 413         tspec = btrfs_inode_ctime(inode_item);
 414         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 415         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 416
 417         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 418         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 419         inode->i_rdev = 0;
 420         rdev = btrfs_inode_rdev(leaf, inode_item);
 421
 422         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 423         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 424                                                        alloc_group_block);
 425         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 426         if (!BTRFS_I(inode)->block_group) {
 427                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 428                                                          NULL, 0, 0, 0);
 429         }
 430         btrfs_free_path(path);
 431         inode_item = NULL;
 432
 433         mutex_unlock(&root->fs_info->fs_mutex);
 434
 435         switch (inode->i_mode & S_IFMT) {
 436         case S_IFREG:
 437                 inode->i_mapping->a_ops = &btrfs_aops;
 438                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 439                 inode->i_fop = &btrfs_file_operations;
 440                 inode->i_op = &btrfs_file_inode_operations;
 441                 break;
 442         case S_IFDIR:
 443                 inode->i_fop = &btrfs_dir_file_operations;
 444                 if (root == root->fs_info->tree_root)
 445                         inode->i_op = &btrfs_dir_ro_inode_operations;
 446                 else
 447                         inode->i_op = &btrfs_dir_inode_operations;
 448                 break;
 449         case S_IFLNK:
 450                 inode->i_op = &btrfs_symlink_inode_operations;
 451                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 452                 break;
 453         default:
 454                 init_special_inode(inode, inode->i_mode, rdev);
 455                 break;
 456         }
 457         return;
 458
 459 make_bad:
 460         btrfs_release_path(root, path);
 461         btrfs_free_path(path);
 462         mutex_unlock(&root->fs_info->fs_mutex);
 463         make_bad_inode(inode);
 464 }
 465
 466 static void fill_inode_item(struct extent_buffer *leaf,
 467                             struct btrfs_inode_item *item,
 468                             struct inode *inode)
 469 {
 470         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 471         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 472         btrfs_set_inode_size(leaf, item, inode->i_size);
 473         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 474         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 475
 476         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 477                                inode->i_atime.tv_sec);
 478         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 479                                 inode->i_atime.tv_nsec);
 480
 481         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 482                                inode->i_mtime.tv_sec);
 483         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 484                                 inode->i_mtime.tv_nsec);
 485
 486         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 487                                inode->i_ctime.tv_sec);
 488         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 489                                 inode->i_ctime.tv_nsec);
 490
 491         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 492         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 493         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 494         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 495         btrfs_set_inode_block_group(leaf, item,
 496                                     BTRFS_I(inode)->block_group->key.objectid);
 497 }
 498
 499 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 500                               struct btrfs_root *root,
 501                               struct inode *inode)
 502 {
 503         struct btrfs_inode_item *inode_item;
 504         struct btrfs_path *path;
 505         struct extent_buffer *leaf;
 506         int ret;
 507
 508         path = btrfs_alloc_path();
 509         BUG_ON(!path);
 510         ret = btrfs_lookup_inode(trans, root, path,
 511                                  &BTRFS_I(inode)->location, 1);
 512         if (ret) {
 513                 if (ret > 0)
 514                         ret = -ENOENT;
 515                 goto failed;
 516         }
 517
 518         leaf = path->nodes[0];
 519         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 520                                   struct btrfs_inode_item);
 521
 522         fill_inode_item(leaf, inode_item, inode);
 523         btrfs_mark_buffer_dirty(leaf);
 524         btrfs_set_inode_last_trans(trans, inode);
 525         ret = 0;
 526 failed:
 527         btrfs_release_path(root, path);
 528         btrfs_free_path(path);
 529         return ret;
 530 }
 531
 532
 533 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 534                               struct btrfs_root *root,
 535                               struct inode *dir,
 536                               struct dentry *dentry)
 537 {
 538         struct btrfs_path *path;
 539         const char *name = dentry->d_name.name;
 540         int name_len = dentry->d_name.len;
 541         int ret = 0;
 542         struct extent_buffer *leaf;
 543         struct btrfs_dir_item *di;
 544         struct btrfs_key key;
 545
 546         path = btrfs_alloc_path();
 547         if (!path) {
 548                 ret = -ENOMEM;
 549                 goto err;
 550         }
 551
 552         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 553                                     name, name_len, -1);
 554         if (IS_ERR(di)) {
 555                 ret = PTR_ERR(di);
 556                 goto err;
 557         }
 558         if (!di) {
 559                 ret = -ENOENT;
 560                 goto err;
 561         }
 562         leaf = path->nodes[0];
 563         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 564         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 565         if (ret)
 566                 goto err;
 567         btrfs_release_path(root, path);
 568
 569         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 570                                          key.objectid, name, name_len, -1);
 571         if (IS_ERR(di)) {
 572                 ret = PTR_ERR(di);
 573                 goto err;
 574         }
 575         if (!di) {
 576                 ret = -ENOENT;
 577                 goto err;
 578         }
 579         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 580
 581         dentry->d_inode->i_ctime = dir->i_ctime;
 582         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 583                                   dentry->d_inode->i_ino,
 584                                   dentry->d_parent->d_inode->i_ino);
 585         if (ret) {
 586                 printk("failed to delete reference to %.*s, "
 587                        "inode %lu parent %lu\n", name_len, name,
 588                        dentry->d_inode->i_ino,
 589                        dentry->d_parent->d_inode->i_ino);
 590         }
 591 err:
 592         btrfs_free_path(path);
 593         if (!ret) {
 594                 dir->i_size -= name_len * 2;
 595                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 596                 btrfs_update_inode(trans, root, dir);
 597 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 598                 dentry->d_inode->i_nlink--;
 599 #else
 600                 drop_nlink(dentry->d_inode);
 601 #endif
 602                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 603                 dir->i_sb->s_dirt = 1;
 604         }
 605         return ret;
 606 }
 607
 608 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 609 {
 610         struct btrfs_root *root;
 611         struct btrfs_trans_handle *trans;
 612         struct inode *inode = dentry->d_inode;
 613         int ret;
 614         unsigned long nr = 0;
 615
 616         root = BTRFS_I(dir)->root;
 617         mutex_lock(&root->fs_info->fs_mutex);
 618
 619         ret = btrfs_check_free_space(root, 1, 1);
 620         if (ret)
 621                 goto fail;
 622
 623         trans = btrfs_start_transaction(root, 1);
 624
 625         btrfs_set_trans_block_group(trans, dir);
 626         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 627         nr = trans->blocks_used;
 628
 629         if (inode->i_nlink == 0) {
 630                 int found;
 631                 /* if the inode isn't linked anywhere,
 632                  * we don't need to worry about
 633                  * data=ordered
 634                  */
 635                 found = btrfs_del_ordered_inode(inode);
 636                 if (found == 1) {
 637                         atomic_dec(&inode->i_count);
 638                 }
 639         }
 640
 641         btrfs_end_transaction(trans, root);
 642 fail:
 643         mutex_unlock(&root->fs_info->fs_mutex);
 644         btrfs_btree_balance_dirty(root, nr);
 645         btrfs_throttle(root);
 646         return ret;
 647 }
 648
 649 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 650 {
 651         struct inode *inode = dentry->d_inode;
 652         int err = 0;
 653         int ret;
 654         struct btrfs_root *root = BTRFS_I(dir)->root;
 655         struct btrfs_trans_handle *trans;
 656         unsigned long nr = 0;
 657
 658         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 659                 return -ENOTEMPTY;
 660
 661         mutex_lock(&root->fs_info->fs_mutex);
 662         ret = btrfs_check_free_space(root, 1, 1);
 663         if (ret)
 664                 goto fail;
 665
 666         trans = btrfs_start_transaction(root, 1);
 667         btrfs_set_trans_block_group(trans, dir);
 668
 669         /* now the directory is empty */
 670         err = btrfs_unlink_trans(trans, root, dir, dentry);
 671         if (!err) {
 672                 inode->i_size = 0;
 673         }
 674
 675         nr = trans->blocks_used;
 676         ret = btrfs_end_transaction(trans, root);
 677 fail:
 678         mutex_unlock(&root->fs_info->fs_mutex);
 679         btrfs_btree_balance_dirty(root, nr);
 680         btrfs_throttle(root);
 681
 682         if (ret && !err)
 683                 err = ret;
 684         return err;
 685 }
 686
 687 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 688                             struct btrfs_root *root,
 689                             struct inode *inode)
 690 {
 691         struct btrfs_path *path;
 692         int ret;
 693
 694         clear_inode(inode);
 695
 696         path = btrfs_alloc_path();
 697         BUG_ON(!path);
 698         ret = btrfs_lookup_inode(trans, root, path,
 699                                  &BTRFS_I(inode)->location, -1);
 700         if (ret > 0)
 701                 ret = -ENOENT;
 702         if (!ret)
 703                 ret = btrfs_del_item(trans, root, path);
 704         btrfs_free_path(path);
 705         return ret;
 706 }
 707
 708 /*
 709  * this can truncate away extent items, csum items and directory items.
 710  * It starts at a high offset and removes keys until it can't find
 711  * any higher than i_size.
 712  *
 713  * csum items that cross the new i_size are truncated to the new size
 714  * as well.
 715  */
 716 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 717                                    struct btrfs_root *root,
 718                                    struct inode *inode)
 719 {
 720         int ret;
 721         struct btrfs_path *path;
 722         struct btrfs_key key;
 723         struct btrfs_key found_key;
 724         u32 found_type;
 725         struct extent_buffer *leaf;
 726         struct btrfs_file_extent_item *fi;
 727         u64 extent_start = 0;
 728         u64 extent_num_bytes = 0;
 729         u64 item_end = 0;
 730         u64 root_gen = 0;
 731         u64 root_owner = 0;
 732         int found_extent;
 733         int del_item;
 734         int extent_type = -1;
 735
 736         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 737         path = btrfs_alloc_path();
 738         path->reada = -1;
 739         BUG_ON(!path);
 740
 741         /* FIXME, add redo link to tree so we don't leak on crash */
 742         key.objectid = inode->i_ino;
 743         key.offset = (u64)-1;
 744         key.type = (u8)-1;
 745
 746         while(1) {
 747                 btrfs_init_path(path);
 748                 fi = NULL;
 749                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 750                 if (ret < 0) {
 751                         goto error;
 752                 }
 753                 if (ret > 0) {
 754                         BUG_ON(path->slots[0] == 0);
 755                         path->slots[0]--;
 756                 }
 757                 leaf = path->nodes[0];
 758                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 759                 found_type = btrfs_key_type(&found_key);
 760
 761                 if (found_key.objectid != inode->i_ino)
 762                         break;
 763
 764                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 765                     found_type != BTRFS_DIR_ITEM_KEY &&
 766                     found_type != BTRFS_DIR_INDEX_KEY &&
 767                     found_type != BTRFS_EXTENT_DATA_KEY)
 768                         break;
 769
 770                 item_end = found_key.offset;
 771                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 772                         fi = btrfs_item_ptr(leaf, path->slots[0],
 773                                             struct btrfs_file_extent_item);
 774                         extent_type = btrfs_file_extent_type(leaf, fi);
 775                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 776                                 item_end +=
 777                                     btrfs_file_extent_num_bytes(leaf, fi);
 778                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 779                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 780                                                                 path->slots[0]);
 781                                 item_end += btrfs_file_extent_inline_len(leaf,
 782                                                                          item);
 783                         }
 784                         item_end--;
 785                 }
 786                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 787                         ret = btrfs_csum_truncate(trans, root, path,
 788                                                   inode->i_size);
 789                         BUG_ON(ret);
 790                 }
 791                 if (item_end < inode->i_size) {
 792                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 793                                 found_type = BTRFS_INODE_ITEM_KEY;
 794                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 795                                 found_type = BTRFS_CSUM_ITEM_KEY;
 796                         } else if (found_type) {
 797                                 found_type--;
 798                         } else {
 799                                 break;
 800                         }
 801                         btrfs_set_key_type(&key, found_type);
 802                         btrfs_release_path(root, path);
 803                         continue;
 804                 }
 805                 if (found_key.offset >= inode->i_size)
 806                         del_item = 1;
 807                 else
 808                         del_item = 0;
 809                 found_extent = 0;
 810
 811                 /* FIXME, shrink the extent if the ref count is only 1 */
 812                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 813                         goto delete;
 814
 815                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 816                         u64 num_dec;
 817                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 818                         if (!del_item) {
 819                                 u64 orig_num_bytes =
 820                                         btrfs_file_extent_num_bytes(leaf, fi);
 821                                 extent_num_bytes = inode->i_size -
 822                                         found_key.offset + root->sectorsize - 1;
 823                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 824                                                          extent_num_bytes);
 825                                 num_dec = (orig_num_bytes -
 826                                            extent_num_bytes) >> 9;
 827                                 if (extent_start != 0) {
 828                                         inode->i_blocks -= num_dec;
 829                                 }
 830                                 btrfs_mark_buffer_dirty(leaf);
 831                         } else {
 832                                 extent_num_bytes =
 833                                         btrfs_file_extent_disk_num_bytes(leaf,
 834                                                                          fi);
 835                                 /* FIXME blocksize != 4096 */
 836                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 837                                                                        fi) >> 9;
 838                                 if (extent_start != 0) {
 839                                         found_extent = 1;
 840                                         inode->i_blocks -= num_dec;
 841                                 }
 842                                 root_gen = btrfs_header_generation(leaf);
 843                                 root_owner = btrfs_header_owner(leaf);
 844                         }
 845                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 846                            !del_item) {
 847                         u32 newsize = inode->i_size - found_key.offset;
 848                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 849                         ret = btrfs_truncate_item(trans, root, path,
 850                                                   newsize, 1);
 851                         BUG_ON(ret);
 852                 }
 853 delete:
 854                 if (del_item) {
 855                         ret = btrfs_del_item(trans, root, path);
 856                         if (ret)
 857                                 goto error;
 858                 } else {
 859                         break;
 860                 }
 861                 btrfs_release_path(root, path);
 862                 if (found_extent) {
 863                         ret = btrfs_free_extent(trans, root, extent_start,
 864                                                 extent_num_bytes,
 865                                                 root_owner,
 866                                                 root_gen, inode->i_ino,
 867                                                 found_key.offset, 0);
 868                         BUG_ON(ret);
 869                 }
 870         }
 871         ret = 0;
 872 error:
 873         btrfs_release_path(root, path);
 874         btrfs_free_path(path);
 875         inode->i_sb->s_dirt = 1;
 876         return ret;
 877 }
 878
 879 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 880                               size_t zero_start)
 881 {
 882         char *kaddr;
 883         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 884         struct btrfs_root *root = BTRFS_I(inode)->root;
 885         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 886         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 887         u64 existing_delalloc;
 888         u64 delalloc_start;
 889         int ret = 0;
 890
 891         WARN_ON(!PageLocked(page));
 892         set_page_extent_mapped(page);
 893
 894         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 895         delalloc_start = page_start;
 896         existing_delalloc = count_range_bits(&BTRFS_I(inode)->io_tree,
 897                                              &delalloc_start, page_end,
 898                                              PAGE_CACHE_SIZE, EXTENT_DELALLOC);
 899         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
 900                             page_end, GFP_NOFS);
 901
 902         spin_lock(&root->fs_info->delalloc_lock);
 903         root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE - existing_delalloc;
 904         spin_unlock(&root->fs_info->delalloc_lock);
 905
 906         if (zero_start != PAGE_CACHE_SIZE) {
 907                 kaddr = kmap(page);
 908                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 909                 flush_dcache_page(page);
 910                 kunmap(page);
 911         }
 912         set_page_dirty(page);
 913         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 914
 915         return ret;
 916 }
 917
 918 /*
 919  * taken from block_truncate_page, but does cow as it zeros out
 920  * any bytes left in the last page in the file.
 921  */
 922 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 923 {
 924         struct inode *inode = mapping->host;
 925         struct btrfs_root *root = BTRFS_I(inode)->root;
 926         u32 blocksize = root->sectorsize;
 927         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 928         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 929         struct page *page;
 930         int ret = 0;
 931         u64 page_start;
 932
 933         if ((offset & (blocksize - 1)) == 0)
 934                 goto out;
 935
 936         ret = -ENOMEM;
 937         page = grab_cache_page(mapping, index);
 938         if (!page)
 939                 goto out;
 940         if (!PageUptodate(page)) {
 941                 ret = btrfs_readpage(NULL, page);
 942                 lock_page(page);
 943                 if (!PageUptodate(page)) {
 944                         ret = -EIO;
 945                         goto out;
 946                 }
 947         }
 948         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 949
 950         ret = btrfs_cow_one_page(inode, page, offset);
 951
 952         unlock_page(page);
 953         page_cache_release(page);
 954 out:
 955         return ret;
 956 }
 957
 958 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 959 {
 960         struct inode *inode = dentry->d_inode;
 961         int err;
 962
 963         err = inode_change_ok(inode, attr);
 964         if (err)
 965                 return err;
 966
 967         if (S_ISREG(inode->i_mode) &&
 968             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 969                 struct btrfs_trans_handle *trans;
 970                 struct btrfs_root *root = BTRFS_I(inode)->root;
 971                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 972
 973                 u64 mask = root->sectorsize - 1;
 974                 u64 pos = (inode->i_size + mask) & ~mask;
 975                 u64 block_end = attr->ia_size | mask;
 976                 u64 hole_start;
 977                 u64 hole_size;
 978                 u64 alloc_hint = 0;
 979
 980                 if (attr->ia_size <= pos)
 981                         goto out;
 982
 983                 if (pos != inode->i_size)
 984                         hole_start = pos + root->sectorsize;
 985                 else
 986                         hole_start = pos;
 987
 988                 mutex_lock(&root->fs_info->fs_mutex);
 989                 err = btrfs_check_free_space(root, 1, 0);
 990                 mutex_unlock(&root->fs_info->fs_mutex);
 991                 if (err)
 992                         goto fail;
 993
 994                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 995
 996                 lock_extent(io_tree, pos, block_end, GFP_NOFS);
 997                 hole_size = block_end - hole_start;
 998
 999                 mutex_lock(&root->fs_info->fs_mutex);
1000                 trans = btrfs_start_transaction(root, 1);
1001                 btrfs_set_trans_block_group(trans, inode);
1002                 err = btrfs_drop_extents(trans, root, inode,
1003                                          pos, block_end, pos,
1004                                          &alloc_hint);
1005
1006                 if (alloc_hint != EXTENT_MAP_INLINE) {
1007                         err = btrfs_insert_file_extent(trans, root,
1008                                                        inode->i_ino,
1009                                                        hole_start, 0, 0,
1010                                                        hole_size);
1011                         btrfs_drop_extent_cache(inode, hole_start,
1012                                                 hole_size - 1);
1013                         btrfs_check_file(root, inode);
1014                 }
1015                 btrfs_end_transaction(trans, root);
1016                 mutex_unlock(&root->fs_info->fs_mutex);
1017                 unlock_extent(io_tree, pos, block_end, GFP_NOFS);
1018                 if (err)
1019                         return err;
1020         }
1021 out:
1022         err = inode_setattr(inode, attr);
1023 fail:
1024         return err;
1025 }
1026
1027 void btrfs_put_inode(struct inode *inode)
1028 {
1029         int ret;
1030
1031         if (!BTRFS_I(inode)->ordered_trans) {
1032                 return;
1033         }
1034
1035         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1036             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1037                 return;
1038
1039         ret = btrfs_del_ordered_inode(inode);
1040         if (ret == 1) {
1041                 atomic_dec(&inode->i_count);
1042         }
1043 }
1044
1045 void btrfs_delete_inode(struct inode *inode)
1046 {
1047         struct btrfs_trans_handle *trans;
1048         struct btrfs_root *root = BTRFS_I(inode)->root;
1049         unsigned long nr;
1050         int ret;
1051
1052         truncate_inode_pages(&inode->i_data, 0);
1053         if (is_bad_inode(inode)) {
1054                 goto no_delete;
1055         }
1056
1057         inode->i_size = 0;
1058         mutex_lock(&root->fs_info->fs_mutex);
1059         trans = btrfs_start_transaction(root, 1);
1060
1061         btrfs_set_trans_block_group(trans, inode);
1062         ret = btrfs_truncate_in_trans(trans, root, inode);
1063         if (ret)
1064                 goto no_delete_lock;
1065         ret = btrfs_delete_xattrs(trans, root, inode);
1066         if (ret)
1067                 goto no_delete_lock;
1068         ret = btrfs_free_inode(trans, root, inode);
1069         if (ret)
1070                 goto no_delete_lock;
1071         nr = trans->blocks_used;
1072
1073         btrfs_end_transaction(trans, root);
1074         mutex_unlock(&root->fs_info->fs_mutex);
1075         btrfs_btree_balance_dirty(root, nr);
1076         btrfs_throttle(root);
1077         return;
1078
1079 no_delete_lock:
1080         nr = trans->blocks_used;
1081         btrfs_end_transaction(trans, root);
1082         mutex_unlock(&root->fs_info->fs_mutex);
1083         btrfs_btree_balance_dirty(root, nr);
1084         btrfs_throttle(root);
1085 no_delete:
1086         clear_inode(inode);
1087 }
1088
1089 /*
1090  * this returns the key found in the dir entry in the location pointer.
1091  * If no dir entries were found, location->objectid is 0.
1092  */
1093 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1094                                struct btrfs_key *location)
1095 {
1096         const char *name = dentry->d_name.name;
1097         int namelen = dentry->d_name.len;
1098         struct btrfs_dir_item *di;
1099         struct btrfs_path *path;
1100         struct btrfs_root *root = BTRFS_I(dir)->root;
1101         int ret = 0;
1102
1103         if (namelen == 1 && strcmp(name, ".") == 0) {
1104                 location->objectid = dir->i_ino;
1105                 location->type = BTRFS_INODE_ITEM_KEY;
1106                 location->offset = 0;
1107                 return 0;
1108         }
1109         path = btrfs_alloc_path();
1110         BUG_ON(!path);
1111
1112         if (namelen == 2 && strcmp(name, "..") == 0) {
1113                 struct btrfs_key key;
1114                 struct extent_buffer *leaf;
1115                 u32 nritems;
1116                 int slot;
1117
1118                 key.objectid = dir->i_ino;
1119                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1120                 key.offset = 0;
1121                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1122                 BUG_ON(ret == 0);
1123                 ret = 0;
1124
1125                 leaf = path->nodes[0];
1126                 slot = path->slots[0];
1127                 nritems = btrfs_header_nritems(leaf);
1128                 if (slot >= nritems)
1129                         goto out_err;
1130
1131                 btrfs_item_key_to_cpu(leaf, &key, slot);
1132                 if (key.objectid != dir->i_ino ||
1133                     key.type != BTRFS_INODE_REF_KEY) {
1134                         goto out_err;
1135                 }
1136                 location->objectid = key.offset;
1137                 location->type = BTRFS_INODE_ITEM_KEY;
1138                 location->offset = 0;
1139                 goto out;
1140         }
1141
1142         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1143                                     namelen, 0);
1144         if (IS_ERR(di))
1145                 ret = PTR_ERR(di);
1146         if (!di || IS_ERR(di)) {
1147                 goto out_err;
1148         }
1149         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1150 out:
1151         btrfs_free_path(path);
1152         return ret;
1153 out_err:
1154         location->objectid = 0;
1155         goto out;
1156 }
1157
1158 /*
1159  * when we hit a tree root in a directory, the btrfs part of the inode
1160  * needs to be changed to reflect the root directory of the tree root.  This
1161  * is kind of like crossing a mount point.
1162  */
1163 static int fixup_tree_root_location(struct btrfs_root *root,
1164                              struct btrfs_key *location,
1165                              struct btrfs_root **sub_root,
1166                              struct dentry *dentry)
1167 {
1168         struct btrfs_path *path;
1169         struct btrfs_root_item *ri;
1170
1171         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1172                 return 0;
1173         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1174                 return 0;
1175
1176         path = btrfs_alloc_path();
1177         BUG_ON(!path);
1178         mutex_lock(&root->fs_info->fs_mutex);
1179
1180         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1181                                         dentry->d_name.name,
1182                                         dentry->d_name.len);
1183         if (IS_ERR(*sub_root))
1184                 return PTR_ERR(*sub_root);
1185
1186         ri = &(*sub_root)->root_item;
1187         location->objectid = btrfs_root_dirid(ri);
1188         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1189         location->offset = 0;
1190
1191         btrfs_free_path(path);
1192         mutex_unlock(&root->fs_info->fs_mutex);
1193         return 0;
1194 }
1195
1196 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1197 {
1198         struct btrfs_iget_args *args = p;
1199         inode->i_ino = args->ino;
1200         BTRFS_I(inode)->root = args->root;
1201         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1202         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1203                              inode->i_mapping, GFP_NOFS);
1204         return 0;
1205 }
1206
1207 static int btrfs_find_actor(struct inode *inode, void *opaque)
1208 {
1209         struct btrfs_iget_args *args = opaque;
1210         return (args->ino == inode->i_ino &&
1211                 args->root == BTRFS_I(inode)->root);
1212 }
1213
1214 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1215                             u64 root_objectid)
1216 {
1217         struct btrfs_iget_args args;
1218         args.ino = objectid;
1219         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1220
1221         if (!args.root)
1222                 return NULL;
1223
1224         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1225 }
1226
1227 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1228                                 struct btrfs_root *root)
1229 {
1230         struct inode *inode;
1231         struct btrfs_iget_args args;
1232         args.ino = objectid;
1233         args.root = root;
1234
1235         inode = iget5_locked(s, objectid, btrfs_find_actor,
1236                              btrfs_init_locked_inode,
1237                              (void *)&args);
1238         return inode;
1239 }
1240
1241 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1242                                    struct nameidata *nd)
1243 {
1244         struct inode * inode;
1245         struct btrfs_inode *bi = BTRFS_I(dir);
1246         struct btrfs_root *root = bi->root;
1247         struct btrfs_root *sub_root = root;
1248         struct btrfs_key location;
1249         int ret;
1250
1251         if (dentry->d_name.len > BTRFS_NAME_LEN)
1252                 return ERR_PTR(-ENAMETOOLONG);
1253
1254         mutex_lock(&root->fs_info->fs_mutex);
1255         ret = btrfs_inode_by_name(dir, dentry, &location);
1256         mutex_unlock(&root->fs_info->fs_mutex);
1257
1258         if (ret < 0)
1259                 return ERR_PTR(ret);
1260
1261         inode = NULL;
1262         if (location.objectid) {
1263                 ret = fixup_tree_root_location(root, &location, &sub_root,
1264                                                 dentry);
1265                 if (ret < 0)
1266                         return ERR_PTR(ret);
1267                 if (ret > 0)
1268                         return ERR_PTR(-ENOENT);
1269                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1270                                           sub_root);
1271                 if (!inode)
1272                         return ERR_PTR(-EACCES);
1273                 if (inode->i_state & I_NEW) {
1274                         /* the inode and parent dir are two different roots */
1275                         if (sub_root != root) {
1276                                 igrab(inode);
1277                                 sub_root->inode = inode;
1278                         }
1279                         BTRFS_I(inode)->root = sub_root;
1280                         memcpy(&BTRFS_I(inode)->location, &location,
1281                                sizeof(location));
1282                         btrfs_read_locked_inode(inode);
1283                         unlock_new_inode(inode);
1284                 }
1285         }
1286         return d_splice_alias(inode, dentry);
1287 }
1288
1289 static unsigned char btrfs_filetype_table[] = {
1290         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1291 };
1292
1293 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1294 {
1295         struct inode *inode = filp->f_dentry->d_inode;
1296         struct btrfs_root *root = BTRFS_I(inode)->root;
1297         struct btrfs_item *item;
1298         struct btrfs_dir_item *di;
1299         struct btrfs_key key;
1300         struct btrfs_key found_key;
1301         struct btrfs_path *path;
1302         int ret;
1303         u32 nritems;
1304         struct extent_buffer *leaf;
1305         int slot;
1306         int advance;
1307         unsigned char d_type;
1308         int over = 0;
1309         u32 di_cur;
1310         u32 di_total;
1311         u32 di_len;
1312         int key_type = BTRFS_DIR_INDEX_KEY;
1313         char tmp_name[32];
1314         char *name_ptr;
1315         int name_len;
1316
1317         /* FIXME, use a real flag for deciding about the key type */
1318         if (root->fs_info->tree_root == root)
1319                 key_type = BTRFS_DIR_ITEM_KEY;
1320
1321         /* special case for "." */
1322         if (filp->f_pos == 0) {
1323                 over = filldir(dirent, ".", 1,
1324                                1, inode->i_ino,
1325                                DT_DIR);
1326                 if (over)
1327                         return 0;
1328                 filp->f_pos = 1;
1329         }
1330
1331         mutex_lock(&root->fs_info->fs_mutex);
1332         key.objectid = inode->i_ino;
1333         path = btrfs_alloc_path();
1334         path->reada = 2;
1335
1336         /* special case for .., just use the back ref */
1337         if (filp->f_pos == 1) {
1338                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1339                 key.offset = 0;
1340                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1341                 BUG_ON(ret == 0);
1342                 leaf = path->nodes[0];
1343                 slot = path->slots[0];
1344                 nritems = btrfs_header_nritems(leaf);
1345                 if (slot >= nritems) {
1346                         btrfs_release_path(root, path);
1347                         goto read_dir_items;
1348                 }
1349                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1350                 btrfs_release_path(root, path);
1351                 if (found_key.objectid != key.objectid ||
1352                     found_key.type != BTRFS_INODE_REF_KEY)
1353                         goto read_dir_items;
1354                 over = filldir(dirent, "..", 2,
1355                                2, found_key.offset, DT_DIR);
1356                 if (over)
1357                         goto nopos;
1358                 filp->f_pos = 2;
1359         }
1360
1361 read_dir_items:
1362         btrfs_set_key_type(&key, key_type);
1363         key.offset = filp->f_pos;
1364
1365         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1366         if (ret < 0)
1367                 goto err;
1368         advance = 0;
1369         while(1) {
1370                 leaf = path->nodes[0];
1371                 nritems = btrfs_header_nritems(leaf);
1372                 slot = path->slots[0];
1373                 if (advance || slot >= nritems) {
1374                         if (slot >= nritems -1) {
1375                                 ret = btrfs_next_leaf(root, path);
1376                                 if (ret)
1377                                         break;
1378                                 leaf = path->nodes[0];
1379                                 nritems = btrfs_header_nritems(leaf);
1380                                 slot = path->slots[0];
1381                         } else {
1382                                 slot++;
1383                                 path->slots[0]++;
1384                         }
1385                 }
1386                 advance = 1;
1387                 item = btrfs_item_nr(leaf, slot);
1388                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1389
1390                 if (found_key.objectid != key.objectid)
1391                         break;
1392                 if (btrfs_key_type(&found_key) != key_type)
1393                         break;
1394                 if (found_key.offset < filp->f_pos)
1395                         continue;
1396
1397                 filp->f_pos = found_key.offset;
1398                 advance = 1;
1399                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1400                 di_cur = 0;
1401                 di_total = btrfs_item_size(leaf, item);
1402                 while(di_cur < di_total) {
1403                         struct btrfs_key location;
1404
1405                         name_len = btrfs_dir_name_len(leaf, di);
1406                         if (name_len < 32) {
1407                                 name_ptr = tmp_name;
1408                         } else {
1409                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1410                                 BUG_ON(!name_ptr);
1411                         }
1412                         read_extent_buffer(leaf, name_ptr,
1413                                            (unsigned long)(di + 1), name_len);
1414
1415                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1416                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1417                         over = filldir(dirent, name_ptr, name_len,
1418                                        found_key.offset,
1419                                        location.objectid,
1420                                        d_type);
1421
1422                         if (name_ptr != tmp_name)
1423                                 kfree(name_ptr);
1424
1425                         if (over)
1426                                 goto nopos;
1427                         di_len = btrfs_dir_name_len(leaf, di) +
1428                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1429                         di_cur += di_len;
1430                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1431                 }
1432         }
1433         filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1434 nopos:
1435         ret = 0;
1436 err:
1437         btrfs_release_path(root, path);
1438         btrfs_free_path(path);
1439         mutex_unlock(&root->fs_info->fs_mutex);
1440         return ret;
1441 }
1442
1443 int btrfs_write_inode(struct inode *inode, int wait)
1444 {
1445         struct btrfs_root *root = BTRFS_I(inode)->root;
1446         struct btrfs_trans_handle *trans;
1447         int ret = 0;
1448
1449         if (wait) {
1450                 mutex_lock(&root->fs_info->fs_mutex);
1451                 trans = btrfs_start_transaction(root, 1);
1452                 btrfs_set_trans_block_group(trans, inode);
1453                 ret = btrfs_commit_transaction(trans, root);
1454                 mutex_unlock(&root->fs_info->fs_mutex);
1455         }
1456         return ret;
1457 }
1458
1459 /*
1460  * This is somewhat expensive, updating the tree every time the
1461  * inode changes.  But, it is most likely to find the inode in cache.
1462  * FIXME, needs more benchmarking...there are no reasons other than performance
1463  * to keep or drop this code.
1464  */
1465 void btrfs_dirty_inode(struct inode *inode)
1466 {
1467         struct btrfs_root *root = BTRFS_I(inode)->root;
1468         struct btrfs_trans_handle *trans;
1469
1470         mutex_lock(&root->fs_info->fs_mutex);
1471         trans = btrfs_start_transaction(root, 1);
1472         btrfs_set_trans_block_group(trans, inode);
1473         btrfs_update_inode(trans, root, inode);
1474         btrfs_end_transaction(trans, root);
1475         mutex_unlock(&root->fs_info->fs_mutex);
1476 }
1477
1478 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1479                                      struct btrfs_root *root,
1480                                      u64 objectid,
1481                                      struct btrfs_block_group_cache *group,
1482                                      int mode)
1483 {
1484         struct inode *inode;
1485         struct btrfs_inode_item *inode_item;
1486         struct btrfs_key *location;
1487         struct btrfs_path *path;
1488         int ret;
1489         int owner;
1490
1491         path = btrfs_alloc_path();
1492         BUG_ON(!path);
1493
1494         inode = new_inode(root->fs_info->sb);
1495         if (!inode)
1496                 return ERR_PTR(-ENOMEM);
1497
1498         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1499         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1500                              inode->i_mapping, GFP_NOFS);
1501         BTRFS_I(inode)->root = root;
1502
1503         if (mode & S_IFDIR)
1504                 owner = 0;
1505         else
1506                 owner = 1;
1507         group = btrfs_find_block_group(root, group, 0, 0, owner);
1508         BTRFS_I(inode)->block_group = group;
1509         BTRFS_I(inode)->flags = 0;
1510         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1511         if (ret)
1512                 goto fail;
1513
1514         inode->i_uid = current->fsuid;
1515         inode->i_gid = current->fsgid;
1516         inode->i_mode = mode;
1517         inode->i_ino = objectid;
1518         inode->i_blocks = 0;
1519         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1520         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1521                                   struct btrfs_inode_item);
1522         fill_inode_item(path->nodes[0], inode_item, inode);
1523         btrfs_mark_buffer_dirty(path->nodes[0]);
1524         btrfs_free_path(path);
1525
1526         location = &BTRFS_I(inode)->location;
1527         location->objectid = objectid;
1528         location->offset = 0;
1529         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1530
1531         insert_inode_hash(inode);
1532         return inode;
1533 fail:
1534         btrfs_free_path(path);
1535         return ERR_PTR(ret);
1536 }
1537
1538 static inline u8 btrfs_inode_type(struct inode *inode)
1539 {
1540         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1541 }
1542
1543 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1544                             struct dentry *dentry, struct inode *inode)
1545 {
1546         int ret;
1547         struct btrfs_key key;
1548         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1549         struct inode *parent_inode;
1550
1551         key.objectid = inode->i_ino;
1552         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1553         key.offset = 0;
1554
1555         ret = btrfs_insert_dir_item(trans, root,
1556                                     dentry->d_name.name, dentry->d_name.len,
1557                                     dentry->d_parent->d_inode->i_ino,
1558                                     &key, btrfs_inode_type(inode));
1559         if (ret == 0) {
1560                 ret = btrfs_insert_inode_ref(trans, root,
1561                                      dentry->d_name.name,
1562                                      dentry->d_name.len,
1563                                      inode->i_ino,
1564                                      dentry->d_parent->d_inode->i_ino);
1565                 parent_inode = dentry->d_parent->d_inode;
1566                 parent_inode->i_size += dentry->d_name.len * 2;
1567                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1568                 ret = btrfs_update_inode(trans, root,
1569                                          dentry->d_parent->d_inode);
1570         }
1571         return ret;
1572 }
1573
1574 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1575                             struct dentry *dentry, struct inode *inode)
1576 {
1577         int err = btrfs_add_link(trans, dentry, inode);
1578         if (!err) {
1579                 d_instantiate(dentry, inode);
1580                 return 0;
1581         }
1582         if (err > 0)
1583                 err = -EEXIST;
1584         return err;
1585 }
1586
1587 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1588                         int mode, dev_t rdev)
1589 {
1590         struct btrfs_trans_handle *trans;
1591         struct btrfs_root *root = BTRFS_I(dir)->root;
1592         struct inode *inode = NULL;
1593         int err;
1594         int drop_inode = 0;
1595         u64 objectid;
1596         unsigned long nr = 0;
1597
1598         if (!new_valid_dev(rdev))
1599                 return -EINVAL;
1600
1601         mutex_lock(&root->fs_info->fs_mutex);
1602         err = btrfs_check_free_space(root, 1, 0);
1603         if (err)
1604                 goto fail;
1605
1606         trans = btrfs_start_transaction(root, 1);
1607         btrfs_set_trans_block_group(trans, dir);
1608
1609         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1610         if (err) {
1611                 err = -ENOSPC;
1612                 goto out_unlock;
1613         }
1614
1615         inode = btrfs_new_inode(trans, root, objectid,
1616                                 BTRFS_I(dir)->block_group, mode);
1617         err = PTR_ERR(inode);
1618         if (IS_ERR(inode))
1619                 goto out_unlock;
1620
1621         btrfs_set_trans_block_group(trans, inode);
1622         err = btrfs_add_nondir(trans, dentry, inode);
1623         if (err)
1624                 drop_inode = 1;
1625         else {
1626                 inode->i_op = &btrfs_special_inode_operations;
1627                 init_special_inode(inode, inode->i_mode, rdev);
1628                 btrfs_update_inode(trans, root, inode);
1629         }
1630         dir->i_sb->s_dirt = 1;
1631         btrfs_update_inode_block_group(trans, inode);
1632         btrfs_update_inode_block_group(trans, dir);
1633 out_unlock:
1634         nr = trans->blocks_used;
1635         btrfs_end_transaction(trans, root);
1636 fail:
1637         mutex_unlock(&root->fs_info->fs_mutex);
1638
1639         if (drop_inode) {
1640                 inode_dec_link_count(inode);
1641                 iput(inode);
1642         }
1643         btrfs_btree_balance_dirty(root, nr);
1644         btrfs_throttle(root);
1645         return err;
1646 }
1647
1648 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1649                         int mode, struct nameidata *nd)
1650 {
1651         struct btrfs_trans_handle *trans;
1652         struct btrfs_root *root = BTRFS_I(dir)->root;
1653         struct inode *inode = NULL;
1654         int err;
1655         int drop_inode = 0;
1656         unsigned long nr = 0;
1657         u64 objectid;
1658
1659         mutex_lock(&root->fs_info->fs_mutex);
1660         err = btrfs_check_free_space(root, 1, 0);
1661         if (err)
1662                 goto fail;
1663         trans = btrfs_start_transaction(root, 1);
1664         btrfs_set_trans_block_group(trans, dir);
1665
1666         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1667         if (err) {
1668                 err = -ENOSPC;
1669                 goto out_unlock;
1670         }
1671
1672         inode = btrfs_new_inode(trans, root, objectid,
1673                                 BTRFS_I(dir)->block_group, mode);
1674         err = PTR_ERR(inode);
1675         if (IS_ERR(inode))
1676                 goto out_unlock;
1677
1678         btrfs_set_trans_block_group(trans, inode);
1679         err = btrfs_add_nondir(trans, dentry, inode);
1680         if (err)
1681                 drop_inode = 1;
1682         else {
1683                 inode->i_mapping->a_ops = &btrfs_aops;
1684                 inode->i_fop = &btrfs_file_operations;
1685                 inode->i_op = &btrfs_file_inode_operations;
1686                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1687                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1688                                      inode->i_mapping, GFP_NOFS);
1689                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1690         }
1691         dir->i_sb->s_dirt = 1;
1692         btrfs_update_inode_block_group(trans, inode);
1693         btrfs_update_inode_block_group(trans, dir);
1694 out_unlock:
1695         nr = trans->blocks_used;
1696         btrfs_end_transaction(trans, root);
1697 fail:
1698         mutex_unlock(&root->fs_info->fs_mutex);
1699
1700         if (drop_inode) {
1701                 inode_dec_link_count(inode);
1702                 iput(inode);
1703         }
1704         btrfs_btree_balance_dirty(root, nr);
1705         btrfs_throttle(root);
1706         return err;
1707 }
1708
1709 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1710                       struct dentry *dentry)
1711 {
1712         struct btrfs_trans_handle *trans;
1713         struct btrfs_root *root = BTRFS_I(dir)->root;
1714         struct inode *inode = old_dentry->d_inode;
1715         unsigned long nr = 0;
1716         int err;
1717         int drop_inode = 0;
1718
1719         if (inode->i_nlink == 0)
1720                 return -ENOENT;
1721
1722 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1723         inode->i_nlink++;
1724 #else
1725         inc_nlink(inode);
1726 #endif
1727         mutex_lock(&root->fs_info->fs_mutex);
1728         err = btrfs_check_free_space(root, 1, 0);
1729         if (err)
1730                 goto fail;
1731         trans = btrfs_start_transaction(root, 1);
1732
1733         btrfs_set_trans_block_group(trans, dir);
1734         atomic_inc(&inode->i_count);
1735         err = btrfs_add_nondir(trans, dentry, inode);
1736
1737         if (err)
1738                 drop_inode = 1;
1739
1740         dir->i_sb->s_dirt = 1;
1741         btrfs_update_inode_block_group(trans, dir);
1742         err = btrfs_update_inode(trans, root, inode);
1743
1744         if (err)
1745                 drop_inode = 1;
1746
1747         nr = trans->blocks_used;
1748         btrfs_end_transaction(trans, root);
1749 fail:
1750         mutex_unlock(&root->fs_info->fs_mutex);
1751
1752         if (drop_inode) {
1753                 inode_dec_link_count(inode);
1754                 iput(inode);
1755         }
1756         btrfs_btree_balance_dirty(root, nr);
1757         btrfs_throttle(root);
1758         return err;
1759 }
1760
1761 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1762 {
1763         struct inode *inode;
1764         struct btrfs_trans_handle *trans;
1765         struct btrfs_root *root = BTRFS_I(dir)->root;
1766         int err = 0;
1767         int drop_on_err = 0;
1768         u64 objectid;
1769         unsigned long nr = 1;
1770
1771         mutex_lock(&root->fs_info->fs_mutex);
1772         err = btrfs_check_free_space(root, 1, 0);
1773         if (err)
1774                 goto out_unlock;
1775
1776         trans = btrfs_start_transaction(root, 1);
1777         btrfs_set_trans_block_group(trans, dir);
1778
1779         if (IS_ERR(trans)) {
1780                 err = PTR_ERR(trans);
1781                 goto out_unlock;
1782         }
1783
1784         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1785         if (err) {
1786                 err = -ENOSPC;
1787                 goto out_unlock;
1788         }
1789
1790         inode = btrfs_new_inode(trans, root, objectid,
1791                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1792         if (IS_ERR(inode)) {
1793                 err = PTR_ERR(inode);
1794                 goto out_fail;
1795         }
1796
1797         drop_on_err = 1;
1798         inode->i_op = &btrfs_dir_inode_operations;
1799         inode->i_fop = &btrfs_dir_file_operations;
1800         btrfs_set_trans_block_group(trans, inode);
1801
1802         inode->i_size = 0;
1803         err = btrfs_update_inode(trans, root, inode);
1804         if (err)
1805                 goto out_fail;
1806
1807         err = btrfs_add_link(trans, dentry, inode);
1808         if (err)
1809                 goto out_fail;
1810
1811         d_instantiate(dentry, inode);
1812         drop_on_err = 0;
1813         dir->i_sb->s_dirt = 1;
1814         btrfs_update_inode_block_group(trans, inode);
1815         btrfs_update_inode_block_group(trans, dir);
1816
1817 out_fail:
1818         nr = trans->blocks_used;
1819         btrfs_end_transaction(trans, root);
1820
1821 out_unlock:
1822         mutex_unlock(&root->fs_info->fs_mutex);
1823         if (drop_on_err)
1824                 iput(inode);
1825         btrfs_btree_balance_dirty(root, nr);
1826         btrfs_throttle(root);
1827         return err;
1828 }
1829
1830 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1831                                     size_t page_offset, u64 start, u64 len,
1832                                     int create)
1833 {
1834         int ret;
1835         int err = 0;
1836         u64 bytenr;
1837         u64 extent_start = 0;
1838         u64 extent_end = 0;
1839         u64 objectid = inode->i_ino;
1840         u32 found_type;
1841         struct btrfs_path *path;
1842         struct btrfs_root *root = BTRFS_I(inode)->root;
1843         struct btrfs_file_extent_item *item;
1844         struct extent_buffer *leaf;
1845         struct btrfs_key found_key;
1846         struct extent_map *em = NULL;
1847         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1848         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1849         struct btrfs_trans_handle *trans = NULL;
1850
1851         path = btrfs_alloc_path();
1852         BUG_ON(!path);
1853         mutex_lock(&root->fs_info->fs_mutex);
1854
1855 again:
1856         spin_lock(&em_tree->lock);
1857         em = lookup_extent_mapping(em_tree, start, len);
1858         spin_unlock(&em_tree->lock);
1859
1860         if (em) {
1861                 if (em->start > start) {
1862                         printk("get_extent lookup [%Lu %Lu] em [%Lu %Lu]\n",
1863                                start, len, em->start, em->len);
1864                         WARN_ON(1);
1865                 }
1866                 goto out;
1867         }
1868         em = alloc_extent_map(GFP_NOFS);
1869         if (!em) {
1870                 err = -ENOMEM;
1871                 goto out;
1872         }
1873
1874         em->start = EXTENT_MAP_HOLE;
1875         em->len = (u64)-1;
1876         em->bdev = inode->i_sb->s_bdev;
1877         ret = btrfs_lookup_file_extent(trans, root, path,
1878                                        objectid, start, trans != NULL);
1879         if (ret < 0) {
1880                 err = ret;
1881                 goto out;
1882         }
1883
1884         if (ret != 0) {
1885                 if (path->slots[0] == 0)
1886                         goto not_found;
1887                 path->slots[0]--;
1888         }
1889
1890         leaf = path->nodes[0];
1891         item = btrfs_item_ptr(leaf, path->slots[0],
1892                               struct btrfs_file_extent_item);
1893         /* are we inside the extent that was found? */
1894         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1895         found_type = btrfs_key_type(&found_key);
1896         if (found_key.objectid != objectid ||
1897             found_type != BTRFS_EXTENT_DATA_KEY) {
1898                 goto not_found;
1899         }
1900
1901         found_type = btrfs_file_extent_type(leaf, item);
1902         extent_start = found_key.offset;
1903         if (found_type == BTRFS_FILE_EXTENT_REG) {
1904                 extent_end = extent_start +
1905                        btrfs_file_extent_num_bytes(leaf, item);
1906                 err = 0;
1907                 if (start < extent_start || start >= extent_end) {
1908                         em->start = start;
1909                         if (start < extent_start) {
1910                                 if (start + len <= extent_start)
1911                                         goto not_found;
1912                                 em->len = extent_end - extent_start;
1913                         } else {
1914                                 em->len = len;
1915                         }
1916                         goto not_found_em;
1917                 }
1918                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1919                 if (bytenr == 0) {
1920                         em->start = extent_start;
1921                         em->len = extent_end - extent_start;
1922                         em->block_start = EXTENT_MAP_HOLE;
1923                         goto insert;
1924                 }
1925                 bytenr += btrfs_file_extent_offset(leaf, item);
1926                 em->block_start = bytenr;
1927                 em->start = extent_start;
1928                 em->len = extent_end - extent_start;
1929                 goto insert;
1930         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1931                 unsigned long ptr;
1932                 char *map;
1933                 size_t size;
1934                 size_t extent_offset;
1935                 size_t copy_size;
1936
1937                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1938                                                     path->slots[0]));
1939                 extent_end = (extent_start + size + root->sectorsize - 1) &
1940                         ~((u64)root->sectorsize - 1);
1941                 if (start < extent_start || start >= extent_end) {
1942                         em->start = start;
1943                         if (start < extent_start) {
1944                                 if (start + len <= extent_start)
1945                                         goto not_found;
1946                                 em->len = extent_end - extent_start;
1947                         } else {
1948                                 em->len = len;
1949                         }
1950                         goto not_found_em;
1951                 }
1952                 em->block_start = EXTENT_MAP_INLINE;
1953
1954                 if (!page) {
1955                         em->start = extent_start;
1956                         em->len = size;
1957                         goto out;
1958                 }
1959
1960                 extent_offset = ((u64)page->index << PAGE_CACHE_SHIFT) -
1961                         extent_start + page_offset;
1962                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1963                                 size - extent_offset);
1964                 em->start = extent_start + extent_offset;
1965                 em->len = copy_size;
1966                 map = kmap(page);
1967                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1968                 if (create == 0 && !PageUptodate(page)) {
1969                         read_extent_buffer(leaf, map + page_offset, ptr,
1970                                            copy_size);
1971                         flush_dcache_page(page);
1972                 } else if (create && PageUptodate(page)) {
1973                         if (!trans) {
1974                                 kunmap(page);
1975                                 free_extent_map(em);
1976                                 em = NULL;
1977                                 btrfs_release_path(root, path);
1978                                 trans = btrfs_start_transaction(root, 1);
1979                                 goto again;
1980                         }
1981                         write_extent_buffer(leaf, map + page_offset, ptr,
1982                                             copy_size);
1983                         btrfs_mark_buffer_dirty(leaf);
1984                 }
1985                 kunmap(page);
1986                 set_extent_uptodate(io_tree, em->start,
1987                                     extent_map_end(em) - 1, GFP_NOFS);
1988                 goto insert;
1989         } else {
1990                 printk("unkknown found_type %d\n", found_type);
1991                 WARN_ON(1);
1992         }
1993 not_found:
1994         em->start = start;
1995         em->len = len;
1996 not_found_em:
1997         em->block_start = EXTENT_MAP_HOLE;
1998 insert:
1999         btrfs_release_path(root, path);
2000         if (em->start > start || extent_map_end(em) <= start) {
2001                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2002                 err = -EIO;
2003                 goto out;
2004         }
2005
2006         err = 0;
2007         spin_lock(&em_tree->lock);
2008         ret = add_extent_mapping(em_tree, em);
2009         if (ret == -EEXIST) {
2010                 free_extent_map(em);
2011                 em = lookup_extent_mapping(em_tree, start, len);
2012                 if (!em) {
2013                         err = -EIO;
2014                         printk("failing to insert %Lu %Lu\n", start, len);
2015                 }
2016         }
2017         spin_unlock(&em_tree->lock);
2018 out:
2019         btrfs_free_path(path);
2020         if (trans) {
2021                 ret = btrfs_end_transaction(trans, root);
2022                 if (!err)
2023                         err = ret;
2024         }
2025         mutex_unlock(&root->fs_info->fs_mutex);
2026         if (err) {
2027                 free_extent_map(em);
2028                 WARN_ON(1);
2029                 return ERR_PTR(err);
2030         }
2031         return em;
2032 }
2033
2034 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2035 {
2036         return extent_bmap(mapping, iblock, btrfs_get_extent);
2037 }
2038
2039 int btrfs_readpage(struct file *file, struct page *page)
2040 {
2041         struct extent_io_tree *tree;
2042         tree = &BTRFS_I(page->mapping->host)->io_tree;
2043         return extent_read_full_page(tree, page, btrfs_get_extent);
2044 }
2045
2046 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2047 {
2048         struct extent_io_tree *tree;
2049
2050
2051         if (current->flags & PF_MEMALLOC) {
2052                 redirty_page_for_writepage(wbc, page);
2053                 unlock_page(page);
2054                 return 0;
2055         }
2056         tree = &BTRFS_I(page->mapping->host)->io_tree;
2057         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2058 }
2059
2060 static int btrfs_writepages(struct address_space *mapping,
2061                             struct writeback_control *wbc)
2062 {
2063         struct extent_io_tree *tree;
2064         tree = &BTRFS_I(mapping->host)->io_tree;
2065         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2066 }
2067
2068 static int
2069 btrfs_readpages(struct file *file, struct address_space *mapping,
2070                 struct list_head *pages, unsigned nr_pages)
2071 {
2072         struct extent_io_tree *tree;
2073         tree = &BTRFS_I(mapping->host)->io_tree;
2074         return extent_readpages(tree, mapping, pages, nr_pages,
2075                                 btrfs_get_extent);
2076 }
2077
2078 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
2079 {
2080         struct extent_io_tree *tree;
2081         struct extent_map_tree *map;
2082         int ret;
2083
2084         tree = &BTRFS_I(page->mapping->host)->io_tree;
2085         map = &BTRFS_I(page->mapping->host)->extent_tree;
2086         ret = try_release_extent_mapping(map, tree, page);
2087         if (ret == 1) {
2088                 ClearPagePrivate(page);
2089                 set_page_private(page, 0);
2090                 page_cache_release(page);
2091         }
2092         return ret;
2093 }
2094
2095 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2096 {
2097         struct extent_io_tree *tree;
2098
2099         tree = &BTRFS_I(page->mapping->host)->io_tree;
2100         extent_invalidatepage(tree, page, offset);
2101         btrfs_releasepage(page, GFP_NOFS);
2102 }
2103
2104 /*
2105  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2106  * called from a page fault handler when a page is first dirtied. Hence we must
2107  * be careful to check for EOF conditions here. We set the page up correctly
2108  * for a written page which means we get ENOSPC checking when writing into
2109  * holes and correct delalloc and unwritten extent mapping on filesystems that
2110  * support these features.
2111  *
2112  * We are not allowed to take the i_mutex here so we have to play games to
2113  * protect against truncate races as the page could now be beyond EOF.  Because
2114  * vmtruncate() writes the inode size before removing pages, once we have the
2115  * page lock we can determine safely if the page is beyond EOF. If it is not
2116  * beyond EOF, then the page is guaranteed safe against truncation until we
2117  * unlock the page.
2118  */
2119 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2120 {
2121         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2122         struct btrfs_root *root = BTRFS_I(inode)->root;
2123         unsigned long end;
2124         loff_t size;
2125         int ret;
2126         u64 page_start;
2127
2128         mutex_lock(&root->fs_info->fs_mutex);
2129         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2130         mutex_unlock(&root->fs_info->fs_mutex);
2131         if (ret)
2132                 goto out;
2133
2134         ret = -EINVAL;
2135
2136         lock_page(page);
2137         wait_on_page_writeback(page);
2138         size = i_size_read(inode);
2139         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2140
2141         if ((page->mapping != inode->i_mapping) ||
2142             (page_start > size)) {
2143                 /* page got truncated out from underneath us */
2144                 goto out_unlock;
2145         }
2146
2147         /* page is wholly or partially inside EOF */
2148         if (page_start + PAGE_CACHE_SIZE > size)
2149                 end = size & ~PAGE_CACHE_MASK;
2150         else
2151                 end = PAGE_CACHE_SIZE;
2152
2153         ret = btrfs_cow_one_page(inode, page, end);
2154
2155 out_unlock:
2156         unlock_page(page);
2157 out:
2158         return ret;
2159 }
2160
2161 static void btrfs_truncate(struct inode *inode)
2162 {
2163         struct btrfs_root *root = BTRFS_I(inode)->root;
2164         int ret;
2165         struct btrfs_trans_handle *trans;
2166         unsigned long nr;
2167
2168         if (!S_ISREG(inode->i_mode))
2169                 return;
2170         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2171                 return;
2172
2173         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2174
2175         mutex_lock(&root->fs_info->fs_mutex);
2176         trans = btrfs_start_transaction(root, 1);
2177         btrfs_set_trans_block_group(trans, inode);
2178
2179         /* FIXME, add redo link to tree so we don't leak on crash */
2180         ret = btrfs_truncate_in_trans(trans, root, inode);
2181         btrfs_update_inode(trans, root, inode);
2182         nr = trans->blocks_used;
2183
2184         ret = btrfs_end_transaction(trans, root);
2185         BUG_ON(ret);
2186         mutex_unlock(&root->fs_info->fs_mutex);
2187         btrfs_btree_balance_dirty(root, nr);
2188         btrfs_throttle(root);
2189 }
2190
2191 static int noinline create_subvol(struct btrfs_root *root, char *name,
2192                                   int namelen)
2193 {
2194         struct btrfs_trans_handle *trans;
2195         struct btrfs_key key;
2196         struct btrfs_root_item root_item;
2197         struct btrfs_inode_item *inode_item;
2198         struct extent_buffer *leaf;
2199         struct btrfs_root *new_root = root;
2200         struct inode *inode;
2201         struct inode *dir;
2202         int ret;
2203         int err;
2204         u64 objectid;
2205         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2206         unsigned long nr = 1;
2207
2208         mutex_lock(&root->fs_info->fs_mutex);
2209         ret = btrfs_check_free_space(root, 1, 0);
2210         if (ret)
2211                 goto fail_commit;
2212
2213         trans = btrfs_start_transaction(root, 1);
2214         BUG_ON(!trans);
2215
2216         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2217                                        0, &objectid);
2218         if (ret)
2219                 goto fail;
2220
2221         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2222                                         objectid, trans->transid, 0, 0,
2223                                         0, 0);
2224         if (IS_ERR(leaf))
2225                 return PTR_ERR(leaf);
2226
2227         btrfs_set_header_nritems(leaf, 0);
2228         btrfs_set_header_level(leaf, 0);
2229         btrfs_set_header_bytenr(leaf, leaf->start);
2230         btrfs_set_header_generation(leaf, trans->transid);
2231         btrfs_set_header_owner(leaf, objectid);
2232
2233         write_extent_buffer(leaf, root->fs_info->fsid,
2234                             (unsigned long)btrfs_header_fsid(leaf),
2235                             BTRFS_FSID_SIZE);
2236         btrfs_mark_buffer_dirty(leaf);
2237
2238         inode_item = &root_item.inode;
2239         memset(inode_item, 0, sizeof(*inode_item));
2240         inode_item->generation = cpu_to_le64(1);
2241         inode_item->size = cpu_to_le64(3);
2242         inode_item->nlink = cpu_to_le32(1);
2243         inode_item->nblocks = cpu_to_le64(1);
2244         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2245
2246         btrfs_set_root_bytenr(&root_item, leaf->start);
2247         btrfs_set_root_level(&root_item, 0);
2248         btrfs_set_root_refs(&root_item, 1);
2249         btrfs_set_root_used(&root_item, 0);
2250
2251         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2252         root_item.drop_level = 0;
2253
2254         free_extent_buffer(leaf);
2255         leaf = NULL;
2256
2257         btrfs_set_root_dirid(&root_item, new_dirid);
2258
2259         key.objectid = objectid;
2260         key.offset = 1;
2261         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2262         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2263                                 &root_item);
2264         if (ret)
2265                 goto fail;
2266
2267         /*
2268          * insert the directory item
2269          */
2270         key.offset = (u64)-1;
2271         dir = root->fs_info->sb->s_root->d_inode;
2272         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2273                                     name, namelen, dir->i_ino, &key,
2274                                     BTRFS_FT_DIR);
2275         if (ret)
2276                 goto fail;
2277
2278         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2279                              name, namelen, objectid,
2280                              root->fs_info->sb->s_root->d_inode->i_ino);
2281         if (ret)
2282                 goto fail;
2283
2284         ret = btrfs_commit_transaction(trans, root);
2285         if (ret)
2286                 goto fail_commit;
2287
2288         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2289         BUG_ON(!new_root);
2290
2291         trans = btrfs_start_transaction(new_root, 1);
2292         BUG_ON(!trans);
2293
2294         inode = btrfs_new_inode(trans, new_root, new_dirid,
2295                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2296         if (IS_ERR(inode))
2297                 goto fail;
2298         inode->i_op = &btrfs_dir_inode_operations;
2299         inode->i_fop = &btrfs_dir_file_operations;
2300         new_root->inode = inode;
2301
2302         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2303                                      new_dirid);
2304         inode->i_nlink = 1;
2305         inode->i_size = 0;
2306         ret = btrfs_update_inode(trans, new_root, inode);
2307         if (ret)
2308                 goto fail;
2309 fail:
2310         nr = trans->blocks_used;
2311         err = btrfs_commit_transaction(trans, new_root);
2312         if (err && !ret)
2313                 ret = err;
2314 fail_commit:
2315         mutex_unlock(&root->fs_info->fs_mutex);
2316         btrfs_btree_balance_dirty(root, nr);
2317         btrfs_throttle(root);
2318         return ret;
2319 }
2320
2321 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2322 {
2323         struct btrfs_pending_snapshot *pending_snapshot;
2324         struct btrfs_trans_handle *trans;
2325         int ret;
2326         int err;
2327         unsigned long nr = 0;
2328
2329         if (!root->ref_cows)
2330                 return -EINVAL;
2331
2332         mutex_lock(&root->fs_info->fs_mutex);
2333         ret = btrfs_check_free_space(root, 1, 0);
2334         if (ret)
2335                 goto fail_unlock;
2336
2337         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2338         if (!pending_snapshot) {
2339                 ret = -ENOMEM;
2340                 goto fail_unlock;
2341         }
2342         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2343         if (!pending_snapshot->name) {
2344                 ret = -ENOMEM;
2345                 kfree(pending_snapshot);
2346                 goto fail_unlock;
2347         }
2348         memcpy(pending_snapshot->name, name, namelen);
2349         pending_snapshot->name[namelen] = '\0';
2350         trans = btrfs_start_transaction(root, 1);
2351         BUG_ON(!trans);
2352         pending_snapshot->root = root;
2353         list_add(&pending_snapshot->list,
2354                  &trans->transaction->pending_snapshots);
2355         ret = btrfs_update_inode(trans, root, root->inode);
2356         err = btrfs_commit_transaction(trans, root);
2357
2358 fail_unlock:
2359         mutex_unlock(&root->fs_info->fs_mutex);
2360         btrfs_btree_balance_dirty(root, nr);
2361         btrfs_throttle(root);
2362         return ret;
2363 }
2364
2365 unsigned long btrfs_force_ra(struct address_space *mapping,
2366                               struct file_ra_state *ra, struct file *file,
2367                               pgoff_t offset, pgoff_t last_index)
2368 {
2369         pgoff_t req_size;
2370
2371 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2372         req_size = last_index - offset + 1;
2373         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2374         return offset;
2375 #else
2376         req_size = min(last_index - offset + 1, (pgoff_t)128);
2377         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2378         return offset + req_size;
2379 #endif
2380 }
2381
2382 int btrfs_defrag_file(struct file *file) {
2383         struct inode *inode = fdentry(file)->d_inode;
2384         struct btrfs_root *root = BTRFS_I(inode)->root;
2385         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2386         struct page *page;
2387         unsigned long last_index;
2388         unsigned long ra_index = 0;
2389         u64 page_start;
2390         u64 page_end;
2391         u64 delalloc_start;
2392         u64 existing_delalloc;
2393         unsigned long i;
2394         int ret;
2395
2396         mutex_lock(&root->fs_info->fs_mutex);
2397         ret = btrfs_check_free_space(root, inode->i_size, 0);
2398         mutex_unlock(&root->fs_info->fs_mutex);
2399         if (ret)
2400                 return -ENOSPC;
2401
2402         mutex_lock(&inode->i_mutex);
2403         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2404         for (i = 0; i <= last_index; i++) {
2405                 if (i == ra_index) {
2406                         ra_index = btrfs_force_ra(inode->i_mapping,
2407                                                   &file->f_ra,
2408                                                   file, ra_index, last_index);
2409                 }
2410                 page = grab_cache_page(inode->i_mapping, i);
2411                 if (!page)
2412                         goto out_unlock;
2413                 if (!PageUptodate(page)) {
2414                         btrfs_readpage(NULL, page);
2415                         lock_page(page);
2416                         if (!PageUptodate(page)) {
2417                                 unlock_page(page);
2418                                 page_cache_release(page);
2419                                 goto out_unlock;
2420                         }
2421                 }
2422                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2423                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2424
2425                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2426                 delalloc_start = page_start;
2427                 existing_delalloc =
2428                         count_range_bits(&BTRFS_I(inode)->io_tree,
2429                                          &delalloc_start, page_end,
2430                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2431                 set_extent_delalloc(io_tree, page_start,
2432                                     page_end, GFP_NOFS);
2433
2434                 spin_lock(&root->fs_info->delalloc_lock);
2435                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2436                                                  existing_delalloc;
2437                 spin_unlock(&root->fs_info->delalloc_lock);
2438
2439                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2440                 set_page_dirty(page);
2441                 unlock_page(page);
2442                 page_cache_release(page);
2443                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2444         }
2445
2446 out_unlock:
2447         mutex_unlock(&inode->i_mutex);
2448         return 0;
2449 }
2450
2451 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2452 {
2453         u64 new_size;
2454         u64 old_size;
2455         struct btrfs_ioctl_vol_args *vol_args;
2456         struct btrfs_trans_handle *trans;
2457         char *sizestr;
2458         int ret = 0;
2459         int namelen;
2460         int mod = 0;
2461
2462         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2463
2464         if (!vol_args)
2465                 return -ENOMEM;
2466
2467         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2468                 ret = -EFAULT;
2469                 goto out;
2470         }
2471         namelen = strlen(vol_args->name);
2472         if (namelen > BTRFS_VOL_NAME_MAX) {
2473                 ret = -EINVAL;
2474                 goto out;
2475         }
2476
2477         sizestr = vol_args->name;
2478         if (!strcmp(sizestr, "max"))
2479                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2480         else {
2481                 if (sizestr[0] == '-') {
2482                         mod = -1;
2483                         sizestr++;
2484                 } else if (sizestr[0] == '+') {
2485                         mod = 1;
2486                         sizestr++;
2487                 }
2488                 new_size = btrfs_parse_size(sizestr);
2489                 if (new_size == 0) {
2490                         ret = -EINVAL;
2491                         goto out;
2492                 }
2493         }
2494
2495         mutex_lock(&root->fs_info->fs_mutex);
2496         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2497
2498         if (mod < 0) {
2499                 if (new_size > old_size) {
2500                         ret = -EINVAL;
2501                         goto out_unlock;
2502                 }
2503                 new_size = old_size - new_size;
2504         } else if (mod > 0) {
2505                 new_size = old_size + new_size;
2506         }
2507
2508         if (new_size < 256 * 1024 * 1024) {
2509                 ret = -EINVAL;
2510                 goto out_unlock;
2511         }
2512         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2513                 ret = -EFBIG;
2514                 goto out_unlock;
2515         }
2516
2517         do_div(new_size, root->sectorsize);
2518         new_size *= root->sectorsize;
2519
2520 printk("new size is %Lu\n", new_size);
2521         if (new_size > old_size) {
2522                 trans = btrfs_start_transaction(root, 1);
2523                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2524                 btrfs_commit_transaction(trans, root);
2525         } else {
2526                 ret = btrfs_shrink_extent_tree(root, new_size);
2527         }
2528
2529 out_unlock:
2530         mutex_unlock(&root->fs_info->fs_mutex);
2531 out:
2532         kfree(vol_args);
2533         return ret;
2534 }
2535
2536 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2537                                             void __user *arg)
2538 {
2539         struct btrfs_ioctl_vol_args *vol_args;
2540         struct btrfs_dir_item *di;
2541         struct btrfs_path *path;
2542         u64 root_dirid;
2543         int namelen;
2544         int ret;
2545
2546         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2547
2548         if (!vol_args)
2549                 return -ENOMEM;
2550
2551         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2552                 ret = -EFAULT;
2553                 goto out;
2554         }
2555
2556         namelen = strlen(vol_args->name);
2557         if (namelen > BTRFS_VOL_NAME_MAX) {
2558                 ret = -EINVAL;
2559                 goto out;
2560         }
2561         if (strchr(vol_args->name, '/')) {
2562                 ret = -EINVAL;
2563                 goto out;
2564         }
2565
2566         path = btrfs_alloc_path();
2567         if (!path) {
2568                 ret = -ENOMEM;
2569                 goto out;
2570         }
2571
2572         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2573         mutex_lock(&root->fs_info->fs_mutex);
2574         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2575                             path, root_dirid,
2576                             vol_args->name, namelen, 0);
2577         mutex_unlock(&root->fs_info->fs_mutex);
2578         btrfs_free_path(path);
2579
2580         if (di && !IS_ERR(di)) {
2581                 ret = -EEXIST;
2582                 goto out;
2583         }
2584
2585         if (IS_ERR(di)) {
2586                 ret = PTR_ERR(di);
2587                 goto out;
2588         }
2589
2590         if (root == root->fs_info->tree_root)
2591                 ret = create_subvol(root, vol_args->name, namelen);
2592         else
2593                 ret = create_snapshot(root, vol_args->name, namelen);
2594 out:
2595         kfree(vol_args);
2596         return ret;
2597 }
2598
2599 static int btrfs_ioctl_defrag(struct file *file)
2600 {
2601         struct inode *inode = fdentry(file)->d_inode;
2602         struct btrfs_root *root = BTRFS_I(inode)->root;
2603
2604         switch (inode->i_mode & S_IFMT) {
2605         case S_IFDIR:
2606                 mutex_lock(&root->fs_info->fs_mutex);
2607                 btrfs_defrag_root(root, 0);
2608                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2609                 mutex_unlock(&root->fs_info->fs_mutex);
2610                 break;
2611         case S_IFREG:
2612                 btrfs_defrag_file(file);
2613                 break;
2614         }
2615
2616         return 0;
2617 }
2618
2619 long btrfs_ioctl(struct file *file, unsigned int
2620                 cmd, unsigned long arg)
2621 {
2622         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2623
2624         switch (cmd) {
2625         case BTRFS_IOC_SNAP_CREATE:
2626                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2627         case BTRFS_IOC_DEFRAG:
2628                 return btrfs_ioctl_defrag(file);
2629         case BTRFS_IOC_RESIZE:
2630                 return btrfs_ioctl_resize(root, (void __user *)arg);
2631         }
2632
2633         return -ENOTTY;
2634 }
2635
2636 /*
2637  * Called inside transaction, so use GFP_NOFS
2638  */
2639 struct inode *btrfs_alloc_inode(struct super_block *sb)
2640 {
2641         struct btrfs_inode *ei;
2642
2643         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2644         if (!ei)
2645                 return NULL;
2646         ei->last_trans = 0;
2647         ei->ordered_trans = 0;
2648         return &ei->vfs_inode;
2649 }
2650
2651 void btrfs_destroy_inode(struct inode *inode)
2652 {
2653         WARN_ON(!list_empty(&inode->i_dentry));
2654         WARN_ON(inode->i_data.nrpages);
2655
2656         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2657         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2658 }
2659
2660 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2661 static void init_once(struct kmem_cache * cachep, void *foo)
2662 #else
2663 static void init_once(void * foo, struct kmem_cache * cachep,
2664                       unsigned long flags)
2665 #endif
2666 {
2667         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2668
2669         inode_init_once(&ei->vfs_inode);
2670 }
2671
2672 void btrfs_destroy_cachep(void)
2673 {
2674         if (btrfs_inode_cachep)
2675                 kmem_cache_destroy(btrfs_inode_cachep);
2676         if (btrfs_trans_handle_cachep)
2677                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2678         if (btrfs_transaction_cachep)
2679                 kmem_cache_destroy(btrfs_transaction_cachep);
2680         if (btrfs_bit_radix_cachep)
2681                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2682         if (btrfs_path_cachep)
2683                 kmem_cache_destroy(btrfs_path_cachep);
2684 }
2685
2686 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2687                                        unsigned long extra_flags,
2688 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2689                                        void (*ctor)(struct kmem_cache *, void *)
2690 #else
2691                                        void (*ctor)(void *, struct kmem_cache *,
2692                                                     unsigned long)
2693 #endif
2694                                      )
2695 {
2696         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2697                                  SLAB_MEM_SPREAD | extra_flags), ctor
2698 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2699                                  ,NULL
2700 #endif
2701                                 );
2702 }
2703
2704 int btrfs_init_cachep(void)
2705 {
2706         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2707                                           sizeof(struct btrfs_inode),
2708                                           0, init_once);
2709         if (!btrfs_inode_cachep)
2710                 goto fail;
2711         btrfs_trans_handle_cachep =
2712                         btrfs_cache_create("btrfs_trans_handle_cache",
2713                                            sizeof(struct btrfs_trans_handle),
2714                                            0, NULL);
2715         if (!btrfs_trans_handle_cachep)
2716                 goto fail;
2717         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2718                                              sizeof(struct btrfs_transaction),
2719                                              0, NULL);
2720         if (!btrfs_transaction_cachep)
2721                 goto fail;
2722         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2723                                          sizeof(struct btrfs_path),
2724                                          0, NULL);
2725         if (!btrfs_path_cachep)
2726                 goto fail;
2727         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2728                                               SLAB_DESTROY_BY_RCU, NULL);
2729         if (!btrfs_bit_radix_cachep)
2730                 goto fail;
2731         return 0;
2732 fail:
2733         btrfs_destroy_cachep();
2734         return -ENOMEM;
2735 }
2736
2737 static int btrfs_getattr(struct vfsmount *mnt,
2738                          struct dentry *dentry, struct kstat *stat)
2739 {
2740         struct inode *inode = dentry->d_inode;
2741         generic_fillattr(inode, stat);
2742         stat->blksize = PAGE_CACHE_SIZE;
2743         return 0;
2744 }
2745
2746 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2747                            struct inode * new_dir,struct dentry *new_dentry)
2748 {
2749         struct btrfs_trans_handle *trans;
2750         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2751         struct inode *new_inode = new_dentry->d_inode;
2752         struct inode *old_inode = old_dentry->d_inode;
2753         struct timespec ctime = CURRENT_TIME;
2754         struct btrfs_path *path;
2755         int ret;
2756
2757         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2758             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2759                 return -ENOTEMPTY;
2760         }
2761
2762         mutex_lock(&root->fs_info->fs_mutex);
2763         ret = btrfs_check_free_space(root, 1, 0);
2764         if (ret)
2765                 goto out_unlock;
2766
2767         trans = btrfs_start_transaction(root, 1);
2768
2769         btrfs_set_trans_block_group(trans, new_dir);
2770         path = btrfs_alloc_path();
2771         if (!path) {
2772                 ret = -ENOMEM;
2773                 goto out_fail;
2774         }
2775
2776         old_dentry->d_inode->i_nlink++;
2777         old_dir->i_ctime = old_dir->i_mtime = ctime;
2778         new_dir->i_ctime = new_dir->i_mtime = ctime;
2779         old_inode->i_ctime = ctime;
2780
2781         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2782         if (ret)
2783                 goto out_fail;
2784
2785         if (new_inode) {
2786                 new_inode->i_ctime = CURRENT_TIME;
2787                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2788                 if (ret)
2789                         goto out_fail;
2790         }
2791         ret = btrfs_add_link(trans, new_dentry, old_inode);
2792         if (ret)
2793                 goto out_fail;
2794
2795 out_fail:
2796         btrfs_free_path(path);
2797         btrfs_end_transaction(trans, root);
2798 out_unlock:
2799         mutex_unlock(&root->fs_info->fs_mutex);
2800         return ret;
2801 }
2802
2803 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2804                          const char *symname)
2805 {
2806         struct btrfs_trans_handle *trans;
2807         struct btrfs_root *root = BTRFS_I(dir)->root;
2808         struct btrfs_path *path;
2809         struct btrfs_key key;
2810         struct inode *inode = NULL;
2811         int err;
2812         int drop_inode = 0;
2813         u64 objectid;
2814         int name_len;
2815         int datasize;
2816         unsigned long ptr;
2817         struct btrfs_file_extent_item *ei;
2818         struct extent_buffer *leaf;
2819         unsigned long nr = 0;
2820
2821         name_len = strlen(symname) + 1;
2822         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2823                 return -ENAMETOOLONG;
2824
2825         mutex_lock(&root->fs_info->fs_mutex);
2826         err = btrfs_check_free_space(root, 1, 0);
2827         if (err)
2828                 goto out_fail;
2829
2830         trans = btrfs_start_transaction(root, 1);
2831         btrfs_set_trans_block_group(trans, dir);
2832
2833         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2834         if (err) {
2835                 err = -ENOSPC;
2836                 goto out_unlock;
2837         }
2838
2839         inode = btrfs_new_inode(trans, root, objectid,
2840                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2841         err = PTR_ERR(inode);
2842         if (IS_ERR(inode))
2843                 goto out_unlock;
2844
2845         btrfs_set_trans_block_group(trans, inode);
2846         err = btrfs_add_nondir(trans, dentry, inode);
2847         if (err)
2848                 drop_inode = 1;
2849         else {
2850                 inode->i_mapping->a_ops = &btrfs_aops;
2851                 inode->i_fop = &btrfs_file_operations;
2852                 inode->i_op = &btrfs_file_inode_operations;
2853                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2854                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2855                                      inode->i_mapping, GFP_NOFS);
2856                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2857         }
2858         dir->i_sb->s_dirt = 1;
2859         btrfs_update_inode_block_group(trans, inode);
2860         btrfs_update_inode_block_group(trans, dir);
2861         if (drop_inode)
2862                 goto out_unlock;
2863
2864         path = btrfs_alloc_path();
2865         BUG_ON(!path);
2866         key.objectid = inode->i_ino;
2867         key.offset = 0;
2868         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2869         datasize = btrfs_file_extent_calc_inline_size(name_len);
2870         err = btrfs_insert_empty_item(trans, root, path, &key,
2871                                       datasize);
2872         if (err) {
2873                 drop_inode = 1;
2874                 goto out_unlock;
2875         }
2876         leaf = path->nodes[0];
2877         ei = btrfs_item_ptr(leaf, path->slots[0],
2878                             struct btrfs_file_extent_item);
2879         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2880         btrfs_set_file_extent_type(leaf, ei,
2881                                    BTRFS_FILE_EXTENT_INLINE);
2882         ptr = btrfs_file_extent_inline_start(ei);
2883         write_extent_buffer(leaf, symname, ptr, name_len);
2884         btrfs_mark_buffer_dirty(leaf);
2885         btrfs_free_path(path);
2886
2887         inode->i_op = &btrfs_symlink_inode_operations;
2888         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2889         inode->i_size = name_len - 1;
2890         err = btrfs_update_inode(trans, root, inode);
2891         if (err)
2892                 drop_inode = 1;
2893
2894 out_unlock:
2895         nr = trans->blocks_used;
2896         btrfs_end_transaction(trans, root);
2897 out_fail:
2898         mutex_unlock(&root->fs_info->fs_mutex);
2899         if (drop_inode) {
2900                 inode_dec_link_count(inode);
2901                 iput(inode);
2902         }
2903         btrfs_btree_balance_dirty(root, nr);
2904         btrfs_throttle(root);
2905         return err;
2906 }
2907 static int btrfs_permission(struct inode *inode, int mask,
2908                             struct nameidata *nd)
2909 {
2910         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2911                 return -EACCES;
2912         return generic_permission(inode, mask, NULL);
2913 }
2914
2915 static struct inode_operations btrfs_dir_inode_operations = {
2916         .lookup         = btrfs_lookup,
2917         .create         = btrfs_create,
2918         .unlink         = btrfs_unlink,
2919         .link           = btrfs_link,
2920         .mkdir          = btrfs_mkdir,
2921         .rmdir          = btrfs_rmdir,
2922         .rename         = btrfs_rename,
2923         .symlink        = btrfs_symlink,
2924         .setattr        = btrfs_setattr,
2925         .mknod          = btrfs_mknod,
2926         .setxattr       = generic_setxattr,
2927         .getxattr       = generic_getxattr,
2928         .listxattr      = btrfs_listxattr,
2929         .removexattr    = generic_removexattr,
2930         .permission     = btrfs_permission,
2931 };
2932 static struct inode_operations btrfs_dir_ro_inode_operations = {
2933         .lookup         = btrfs_lookup,
2934         .permission     = btrfs_permission,
2935 };
2936 static struct file_operations btrfs_dir_file_operations = {
2937         .llseek         = generic_file_llseek,
2938         .read           = generic_read_dir,
2939         .readdir        = btrfs_readdir,
2940         .unlocked_ioctl = btrfs_ioctl,
2941 #ifdef CONFIG_COMPAT
2942         .compat_ioctl   = btrfs_ioctl,
2943 #endif
2944 };
2945
2946 static struct extent_io_ops btrfs_extent_io_ops = {
2947         .fill_delalloc = run_delalloc_range,
2948         .writepage_io_hook = btrfs_writepage_io_hook,
2949         .readpage_io_hook = btrfs_readpage_io_hook,
2950         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2951 };
2952
2953 static struct address_space_operations btrfs_aops = {
2954         .readpage       = btrfs_readpage,
2955         .writepage      = btrfs_writepage,
2956         .writepages     = btrfs_writepages,
2957         .readpages      = btrfs_readpages,
2958         .sync_page      = block_sync_page,
2959         .bmap           = btrfs_bmap,
2960         .invalidatepage = btrfs_invalidatepage,
2961         .releasepage    = btrfs_releasepage,
2962         .set_page_dirty = __set_page_dirty_nobuffers,
2963 };
2964
2965 static struct address_space_operations btrfs_symlink_aops = {
2966         .readpage       = btrfs_readpage,
2967         .writepage      = btrfs_writepage,
2968         .invalidatepage = btrfs_invalidatepage,
2969         .releasepage    = btrfs_releasepage,
2970 };
2971
2972 static struct inode_operations btrfs_file_inode_operations = {
2973         .truncate       = btrfs_truncate,
2974         .getattr        = btrfs_getattr,
2975         .setattr        = btrfs_setattr,
2976         .setxattr       = generic_setxattr,
2977         .getxattr       = generic_getxattr,
2978         .listxattr      = btrfs_listxattr,
2979         .removexattr    = generic_removexattr,
2980         .permission     = btrfs_permission,
2981 };
2982 static struct inode_operations btrfs_special_inode_operations = {
2983         .getattr        = btrfs_getattr,
2984         .setattr        = btrfs_setattr,
2985         .permission     = btrfs_permission,
2986 };
2987 static struct inode_operations btrfs_symlink_inode_operations = {
2988         .readlink       = generic_readlink,
2989         .follow_link    = page_follow_link_light,
2990         .put_link       = page_put_link,
2991         .permission     = btrfs_permission,
2992 };