fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/buffer_head.h>
  20 #include <linux/fs.h>
  21 #include <linux/pagemap.h>
  22 #include <linux/highmem.h>
  23 #include <linux/time.h>
  24 #include <linux/init.h>
  25 #include <linux/string.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/mpage.h>
  29 #include <linux/swap.h>
  30 #include <linux/writeback.h>
  31 #include <linux/statfs.h>
  32 #include <linux/compat.h>
  33 #include <linux/bit_spinlock.h>
  34 #include <linux/version.h>
  35 #include <linux/xattr.h>
  36 #include "ctree.h"
  37 #include "disk-io.h"
  38 #include "transaction.h"
  39 #include "btrfs_inode.h"
  40 #include "ioctl.h"
  41 #include "print-tree.h"
  42
  43 struct btrfs_iget_args {
  44         u64 ino;
  45         struct btrfs_root *root;
  46 };
  47
  48 static struct inode_operations btrfs_dir_inode_operations;
  49 static struct inode_operations btrfs_symlink_inode_operations;
  50 static struct inode_operations btrfs_dir_ro_inode_operations;
  51 static struct inode_operations btrfs_special_inode_operations;
  52 static struct inode_operations btrfs_file_inode_operations;
  53 static struct address_space_operations btrfs_aops;
  54 static struct address_space_operations btrfs_symlink_aops;
  55 static struct file_operations btrfs_dir_file_operations;
  56 static struct extent_io_ops btrfs_extent_io_ops;
  57
  58 static struct kmem_cache *btrfs_inode_cachep;
  59 struct kmem_cache *btrfs_trans_handle_cachep;
  60 struct kmem_cache *btrfs_transaction_cachep;
  61 struct kmem_cache *btrfs_bit_radix_cachep;
  62 struct kmem_cache *btrfs_path_cachep;
  63
  64 #define S_SHIFT 12
  65 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  66         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  67         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  68         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  69         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  70         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  71         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  72         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  73 };
  74
  75 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  76                            int for_del)
  77 {
  78         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  79         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  80         u64 thresh;
  81         int ret = 0;
  82
  83         return 0;
  84
  85         if (for_del)
  86                 thresh = total * 90;
  87         else
  88                 thresh = total * 85;
  89
  90         do_div(thresh, 100);
  91
  92         spin_lock(&root->fs_info->delalloc_lock);
  93         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  94                 ret = -ENOSPC;
  95         spin_unlock(&root->fs_info->delalloc_lock);
  96         return ret;
  97 }
  98
  99 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 100 {
 101         struct btrfs_root *root = BTRFS_I(inode)->root;
 102         struct btrfs_trans_handle *trans;
 103         u64 alloc_hint = 0;
 104         u64 num_bytes;
 105         u64 cur_alloc_size;
 106         u64 blocksize = root->sectorsize;
 107         u64 orig_start = start;
 108         u64 orig_num_bytes;
 109         struct btrfs_key ins;
 110         int ret;
 111
 112         trans = btrfs_start_transaction(root, 1);
 113         BUG_ON(!trans);
 114         btrfs_set_trans_block_group(trans, inode);
 115
 116         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 117         num_bytes = max(blocksize,  num_bytes);
 118         ret = btrfs_drop_extents(trans, root, inode,
 119                                  start, start + num_bytes, start, &alloc_hint);
 120         orig_num_bytes = num_bytes;
 121
 122         if (alloc_hint == EXTENT_MAP_INLINE)
 123                 goto out;
 124
 125         while(num_bytes > 0) {
 126                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 127                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 128                                          root->root_key.objectid,
 129                                          trans->transid,
 130                                          inode->i_ino, start, 0,
 131                                          alloc_hint, (u64)-1, &ins, 1);
 132                 if (ret) {
 133                         WARN_ON(1);
 134                         goto out;
 135                 }
 136                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 137                                                start, ins.objectid, ins.offset,
 138                                                ins.offset);
 139                 btrfs_check_file(root, inode);
 140                 num_bytes -= cur_alloc_size;
 141                 alloc_hint = ins.objectid + ins.offset;
 142                 start += cur_alloc_size;
 143         }
 144         btrfs_drop_extent_cache(inode, orig_start,
 145                                 orig_start + orig_num_bytes - 1);
 146         btrfs_add_ordered_inode(inode);
 147 out:
 148         btrfs_end_transaction(trans, root);
 149         return ret;
 150 }
 151
 152 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 153 {
 154         u64 extent_start;
 155         u64 extent_end;
 156         u64 bytenr;
 157         u64 cow_end;
 158         u64 loops = 0;
 159         u64 total_fs_bytes;
 160         struct btrfs_root *root = BTRFS_I(inode)->root;
 161         struct extent_buffer *leaf;
 162         int found_type;
 163         struct btrfs_path *path;
 164         struct btrfs_file_extent_item *item;
 165         int ret;
 166         int err;
 167         struct btrfs_key found_key;
 168
 169         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 170         path = btrfs_alloc_path();
 171         BUG_ON(!path);
 172 again:
 173         ret = btrfs_lookup_file_extent(NULL, root, path,
 174                                        inode->i_ino, start, 0);
 175         if (ret < 0) {
 176                 btrfs_free_path(path);
 177                 return ret;
 178         }
 179
 180         cow_end = end;
 181         if (ret != 0) {
 182                 if (path->slots[0] == 0)
 183                         goto not_found;
 184                 path->slots[0]--;
 185         }
 186
 187         leaf = path->nodes[0];
 188         item = btrfs_item_ptr(leaf, path->slots[0],
 189                               struct btrfs_file_extent_item);
 190
 191         /* are we inside the extent that was found? */
 192         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 193         found_type = btrfs_key_type(&found_key);
 194         if (found_key.objectid != inode->i_ino ||
 195             found_type != BTRFS_EXTENT_DATA_KEY) {
 196                 goto not_found;
 197         }
 198
 199         found_type = btrfs_file_extent_type(leaf, item);
 200         extent_start = found_key.offset;
 201         if (found_type == BTRFS_FILE_EXTENT_REG) {
 202                 u64 extent_num_bytes;
 203
 204                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 205                 extent_end = extent_start + extent_num_bytes;
 206                 err = 0;
 207
 208                 if (loops && start != extent_start)
 209                         goto not_found;
 210
 211                 if (start < extent_start || start >= extent_end)
 212                         goto not_found;
 213
 214                 cow_end = min(end, extent_end - 1);
 215                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 216                 if (bytenr == 0)
 217                         goto not_found;
 218
 219                 /*
 220                  * we may be called by the resizer, make sure we're inside
 221                  * the limits of the FS
 222                  */
 223                 if (bytenr + extent_num_bytes > total_fs_bytes)
 224                         goto not_found;
 225
 226                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 227                         goto not_found;
 228                 }
 229
 230                 start = extent_end;
 231         } else {
 232                 goto not_found;
 233         }
 234 loop:
 235         if (start > end) {
 236                 btrfs_free_path(path);
 237                 return 0;
 238         }
 239         btrfs_release_path(root, path);
 240         loops++;
 241         goto again;
 242
 243 not_found:
 244         cow_file_range(inode, start, cow_end);
 245         start = cow_end + 1;
 246         goto loop;
 247 }
 248
 249 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 250 {
 251         struct btrfs_root *root = BTRFS_I(inode)->root;
 252         u64 num_bytes;
 253         int ret;
 254         mutex_lock(&root->fs_info->fs_mutex);
 255         if (btrfs_test_opt(root, NODATACOW) ||
 256             btrfs_test_flag(inode, NODATACOW))
 257                 ret = run_delalloc_nocow(inode, start, end);
 258         else
 259                 ret = cow_file_range(inode, start, end);
 260
 261         spin_lock(&root->fs_info->delalloc_lock);
 262         num_bytes = end + 1 - start;
 263         if (root->fs_info->delalloc_bytes < num_bytes) {
 264                 printk("delalloc accounting error total %llu sub %llu\n",
 265                        root->fs_info->delalloc_bytes, num_bytes);
 266         } else {
 267                 root->fs_info->delalloc_bytes -= num_bytes;
 268         }
 269         spin_unlock(&root->fs_info->delalloc_lock);
 270
 271         mutex_unlock(&root->fs_info->fs_mutex);
 272         return ret;
 273 }
 274
 275 int btrfs_writepage_io_hook(struct page *page, u64 start, u64 end)
 276 {
 277         struct inode *inode = page->mapping->host;
 278         struct btrfs_root *root = BTRFS_I(inode)->root;
 279         struct btrfs_trans_handle *trans;
 280         char *kaddr;
 281         int ret = 0;
 282         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 283         size_t offset = start - page_start;
 284         if (btrfs_test_opt(root, NODATASUM) ||
 285             btrfs_test_flag(inode, NODATASUM))
 286                 return 0;
 287         mutex_lock(&root->fs_info->fs_mutex);
 288         trans = btrfs_start_transaction(root, 1);
 289         btrfs_set_trans_block_group(trans, inode);
 290         kaddr = kmap(page);
 291         btrfs_csum_file_block(trans, root, inode, inode->i_ino,
 292                               start, kaddr + offset, end - start + 1);
 293         kunmap(page);
 294         ret = btrfs_end_transaction(trans, root);
 295         BUG_ON(ret);
 296         mutex_unlock(&root->fs_info->fs_mutex);
 297         return ret;
 298 }
 299
 300 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 301 {
 302         int ret = 0;
 303         struct inode *inode = page->mapping->host;
 304         struct btrfs_root *root = BTRFS_I(inode)->root;
 305         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 306         struct btrfs_csum_item *item;
 307         struct btrfs_path *path = NULL;
 308         u32 csum;
 309         if (btrfs_test_opt(root, NODATASUM) ||
 310             btrfs_test_flag(inode, NODATASUM))
 311                 return 0;
 312         mutex_lock(&root->fs_info->fs_mutex);
 313         path = btrfs_alloc_path();
 314         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 315         if (IS_ERR(item)) {
 316                 ret = PTR_ERR(item);
 317                 /* a csum that isn't present is a preallocated region. */
 318                 if (ret == -ENOENT || ret == -EFBIG)
 319                         ret = 0;
 320                 csum = 0;
 321                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino, start);
 322                 goto out;
 323         }
 324         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 325                            BTRFS_CRC32_SIZE);
 326         set_state_private(io_tree, start, csum);
 327 out:
 328         if (path)
 329                 btrfs_free_path(path);
 330         mutex_unlock(&root->fs_info->fs_mutex);
 331         return ret;
 332 }
 333
 334 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 335                                struct extent_state *state)
 336 {
 337         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 338         struct inode *inode = page->mapping->host;
 339         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 340         char *kaddr;
 341         u64 private = ~(u32)0;
 342         int ret;
 343         struct btrfs_root *root = BTRFS_I(inode)->root;
 344         u32 csum = ~(u32)0;
 345         unsigned long flags;
 346
 347         if (btrfs_test_opt(root, NODATASUM) ||
 348             btrfs_test_flag(inode, NODATASUM))
 349                 return 0;
 350
 351         if (state->start == start) {
 352                 private = state->private;
 353                 ret = 0;
 354         } else {
 355                 ret = get_state_private(io_tree, start, &private);
 356         }
 357         local_irq_save(flags);
 358         kaddr = kmap_atomic(page, KM_IRQ0);
 359         if (ret) {
 360                 goto zeroit;
 361         }
 362         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 363         btrfs_csum_final(csum, (char *)&csum);
 364         if (csum != private) {
 365                 goto zeroit;
 366         }
 367         kunmap_atomic(kaddr, KM_IRQ0);
 368         local_irq_restore(flags);
 369         return 0;
 370
 371 zeroit:
 372         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 373                page->mapping->host->i_ino, (unsigned long long)start, csum,
 374                private);
 375         memset(kaddr + offset, 1, end - start + 1);
 376         flush_dcache_page(page);
 377         kunmap_atomic(kaddr, KM_IRQ0);
 378         local_irq_restore(flags);
 379         return 0;
 380 }
 381
 382 void btrfs_read_locked_inode(struct inode *inode)
 383 {
 384         struct btrfs_path *path;
 385         struct extent_buffer *leaf;
 386         struct btrfs_inode_item *inode_item;
 387         struct btrfs_inode_timespec *tspec;
 388         struct btrfs_root *root = BTRFS_I(inode)->root;
 389         struct btrfs_key location;
 390         u64 alloc_group_block;
 391         u32 rdev;
 392         int ret;
 393
 394         path = btrfs_alloc_path();
 395         BUG_ON(!path);
 396         mutex_lock(&root->fs_info->fs_mutex);
 397         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 398
 399         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 400         if (ret)
 401                 goto make_bad;
 402
 403         leaf = path->nodes[0];
 404         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 405                                     struct btrfs_inode_item);
 406
 407         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 408         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 409         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 410         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 411         inode->i_size = btrfs_inode_size(leaf, inode_item);
 412
 413         tspec = btrfs_inode_atime(inode_item);
 414         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 415         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 416
 417         tspec = btrfs_inode_mtime(inode_item);
 418         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 419         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 420
 421         tspec = btrfs_inode_ctime(inode_item);
 422         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 423         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 424
 425         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 426         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 427         inode->i_rdev = 0;
 428         rdev = btrfs_inode_rdev(leaf, inode_item);
 429
 430         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 431         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 432                                                        alloc_group_block);
 433         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 434         if (!BTRFS_I(inode)->block_group) {
 435                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 436                                                          NULL, 0, 0, 0);
 437         }
 438         btrfs_free_path(path);
 439         inode_item = NULL;
 440
 441         mutex_unlock(&root->fs_info->fs_mutex);
 442
 443         switch (inode->i_mode & S_IFMT) {
 444         case S_IFREG:
 445                 inode->i_mapping->a_ops = &btrfs_aops;
 446                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 447                 inode->i_fop = &btrfs_file_operations;
 448                 inode->i_op = &btrfs_file_inode_operations;
 449                 break;
 450         case S_IFDIR:
 451                 inode->i_fop = &btrfs_dir_file_operations;
 452                 if (root == root->fs_info->tree_root)
 453                         inode->i_op = &btrfs_dir_ro_inode_operations;
 454                 else
 455                         inode->i_op = &btrfs_dir_inode_operations;
 456                 break;
 457         case S_IFLNK:
 458                 inode->i_op = &btrfs_symlink_inode_operations;
 459                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 460                 break;
 461         default:
 462                 init_special_inode(inode, inode->i_mode, rdev);
 463                 break;
 464         }
 465         return;
 466
 467 make_bad:
 468         btrfs_release_path(root, path);
 469         btrfs_free_path(path);
 470         mutex_unlock(&root->fs_info->fs_mutex);
 471         make_bad_inode(inode);
 472 }
 473
 474 static void fill_inode_item(struct extent_buffer *leaf,
 475                             struct btrfs_inode_item *item,
 476                             struct inode *inode)
 477 {
 478         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 479         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 480         btrfs_set_inode_size(leaf, item, inode->i_size);
 481         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 482         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 483
 484         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 485                                inode->i_atime.tv_sec);
 486         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 487                                 inode->i_atime.tv_nsec);
 488
 489         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 490                                inode->i_mtime.tv_sec);
 491         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 492                                 inode->i_mtime.tv_nsec);
 493
 494         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 495                                inode->i_ctime.tv_sec);
 496         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 497                                 inode->i_ctime.tv_nsec);
 498
 499         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 500         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 501         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 502         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 503         btrfs_set_inode_block_group(leaf, item,
 504                                     BTRFS_I(inode)->block_group->key.objectid);
 505 }
 506
 507 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 508                               struct btrfs_root *root,
 509                               struct inode *inode)
 510 {
 511         struct btrfs_inode_item *inode_item;
 512         struct btrfs_path *path;
 513         struct extent_buffer *leaf;
 514         int ret;
 515
 516         path = btrfs_alloc_path();
 517         BUG_ON(!path);
 518         ret = btrfs_lookup_inode(trans, root, path,
 519                                  &BTRFS_I(inode)->location, 1);
 520         if (ret) {
 521                 if (ret > 0)
 522                         ret = -ENOENT;
 523                 goto failed;
 524         }
 525
 526         leaf = path->nodes[0];
 527         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 528                                   struct btrfs_inode_item);
 529
 530         fill_inode_item(leaf, inode_item, inode);
 531         btrfs_mark_buffer_dirty(leaf);
 532         btrfs_set_inode_last_trans(trans, inode);
 533         ret = 0;
 534 failed:
 535         btrfs_release_path(root, path);
 536         btrfs_free_path(path);
 537         return ret;
 538 }
 539
 540
 541 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 542                               struct btrfs_root *root,
 543                               struct inode *dir,
 544                               struct dentry *dentry)
 545 {
 546         struct btrfs_path *path;
 547         const char *name = dentry->d_name.name;
 548         int name_len = dentry->d_name.len;
 549         int ret = 0;
 550         struct extent_buffer *leaf;
 551         struct btrfs_dir_item *di;
 552         struct btrfs_key key;
 553
 554         path = btrfs_alloc_path();
 555         if (!path) {
 556                 ret = -ENOMEM;
 557                 goto err;
 558         }
 559
 560         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 561                                     name, name_len, -1);
 562         if (IS_ERR(di)) {
 563                 ret = PTR_ERR(di);
 564                 goto err;
 565         }
 566         if (!di) {
 567                 ret = -ENOENT;
 568                 goto err;
 569         }
 570         leaf = path->nodes[0];
 571         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 572         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 573         if (ret)
 574                 goto err;
 575         btrfs_release_path(root, path);
 576
 577         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 578                                          key.objectid, name, name_len, -1);
 579         if (IS_ERR(di)) {
 580                 ret = PTR_ERR(di);
 581                 goto err;
 582         }
 583         if (!di) {
 584                 ret = -ENOENT;
 585                 goto err;
 586         }
 587         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 588
 589         dentry->d_inode->i_ctime = dir->i_ctime;
 590         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 591                                   dentry->d_inode->i_ino,
 592                                   dentry->d_parent->d_inode->i_ino);
 593         if (ret) {
 594                 printk("failed to delete reference to %.*s, "
 595                        "inode %lu parent %lu\n", name_len, name,
 596                        dentry->d_inode->i_ino,
 597                        dentry->d_parent->d_inode->i_ino);
 598         }
 599 err:
 600         btrfs_free_path(path);
 601         if (!ret) {
 602                 dir->i_size -= name_len * 2;
 603                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 604                 btrfs_update_inode(trans, root, dir);
 605 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 606                 dentry->d_inode->i_nlink--;
 607 #else
 608                 drop_nlink(dentry->d_inode);
 609 #endif
 610                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 611                 dir->i_sb->s_dirt = 1;
 612         }
 613         return ret;
 614 }
 615
 616 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 617 {
 618         struct btrfs_root *root;
 619         struct btrfs_trans_handle *trans;
 620         struct inode *inode = dentry->d_inode;
 621         int ret;
 622         unsigned long nr = 0;
 623
 624         root = BTRFS_I(dir)->root;
 625         mutex_lock(&root->fs_info->fs_mutex);
 626
 627         ret = btrfs_check_free_space(root, 1, 1);
 628         if (ret)
 629                 goto fail;
 630
 631         trans = btrfs_start_transaction(root, 1);
 632
 633         btrfs_set_trans_block_group(trans, dir);
 634         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 635         nr = trans->blocks_used;
 636
 637         if (inode->i_nlink == 0) {
 638                 int found;
 639                 /* if the inode isn't linked anywhere,
 640                  * we don't need to worry about
 641                  * data=ordered
 642                  */
 643                 found = btrfs_del_ordered_inode(inode);
 644                 if (found == 1) {
 645                         atomic_dec(&inode->i_count);
 646                 }
 647         }
 648
 649         btrfs_end_transaction(trans, root);
 650 fail:
 651         mutex_unlock(&root->fs_info->fs_mutex);
 652         btrfs_btree_balance_dirty(root, nr);
 653         btrfs_throttle(root);
 654         return ret;
 655 }
 656
 657 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 658 {
 659         struct inode *inode = dentry->d_inode;
 660         int err = 0;
 661         int ret;
 662         struct btrfs_root *root = BTRFS_I(dir)->root;
 663         struct btrfs_trans_handle *trans;
 664         unsigned long nr = 0;
 665
 666         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 667                 return -ENOTEMPTY;
 668
 669         mutex_lock(&root->fs_info->fs_mutex);
 670         ret = btrfs_check_free_space(root, 1, 1);
 671         if (ret)
 672                 goto fail;
 673
 674         trans = btrfs_start_transaction(root, 1);
 675         btrfs_set_trans_block_group(trans, dir);
 676
 677         /* now the directory is empty */
 678         err = btrfs_unlink_trans(trans, root, dir, dentry);
 679         if (!err) {
 680                 inode->i_size = 0;
 681         }
 682
 683         nr = trans->blocks_used;
 684         ret = btrfs_end_transaction(trans, root);
 685 fail:
 686         mutex_unlock(&root->fs_info->fs_mutex);
 687         btrfs_btree_balance_dirty(root, nr);
 688         btrfs_throttle(root);
 689
 690         if (ret && !err)
 691                 err = ret;
 692         return err;
 693 }
 694
 695 /*
 696  * this can truncate away extent items, csum items and directory items.
 697  * It starts at a high offset and removes keys until it can't find
 698  * any higher than i_size.
 699  *
 700  * csum items that cross the new i_size are truncated to the new size
 701  * as well.
 702  */
 703 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 704                                    struct btrfs_root *root,
 705                                    struct inode *inode,
 706                                    u32 min_type)
 707 {
 708         int ret;
 709         struct btrfs_path *path;
 710         struct btrfs_key key;
 711         struct btrfs_key found_key;
 712         u32 found_type;
 713         struct extent_buffer *leaf;
 714         struct btrfs_file_extent_item *fi;
 715         u64 extent_start = 0;
 716         u64 extent_num_bytes = 0;
 717         u64 item_end = 0;
 718         u64 root_gen = 0;
 719         u64 root_owner = 0;
 720         int found_extent;
 721         int del_item;
 722         int pending_del_nr = 0;
 723         int pending_del_slot = 0;
 724         int extent_type = -1;
 725
 726         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 727         path = btrfs_alloc_path();
 728         path->reada = -1;
 729         BUG_ON(!path);
 730
 731         /* FIXME, add redo link to tree so we don't leak on crash */
 732         key.objectid = inode->i_ino;
 733         key.offset = (u64)-1;
 734         key.type = (u8)-1;
 735
 736         btrfs_init_path(path);
 737 search_again:
 738         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 739         if (ret < 0) {
 740                 goto error;
 741         }
 742         if (ret > 0) {
 743                 BUG_ON(path->slots[0] == 0);
 744                 path->slots[0]--;
 745         }
 746
 747         while(1) {
 748                 fi = NULL;
 749                 leaf = path->nodes[0];
 750                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 751                 found_type = btrfs_key_type(&found_key);
 752
 753                 if (found_key.objectid != inode->i_ino)
 754                         break;
 755
 756                 if (found_type < min_type)
 757                         break;
 758
 759                 item_end = found_key.offset;
 760                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 761                         fi = btrfs_item_ptr(leaf, path->slots[0],
 762                                             struct btrfs_file_extent_item);
 763                         extent_type = btrfs_file_extent_type(leaf, fi);
 764                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 765                                 item_end +=
 766                                     btrfs_file_extent_num_bytes(leaf, fi);
 767                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 768                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 769                                                                 path->slots[0]);
 770                                 item_end += btrfs_file_extent_inline_len(leaf,
 771                                                                          item);
 772                         }
 773                         item_end--;
 774                 }
 775                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 776                         ret = btrfs_csum_truncate(trans, root, path,
 777                                                   inode->i_size);
 778                         BUG_ON(ret);
 779                 }
 780                 if (item_end < inode->i_size) {
 781                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 782                                 found_type = BTRFS_INODE_ITEM_KEY;
 783                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 784                                 found_type = BTRFS_CSUM_ITEM_KEY;
 785                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
 786                                 found_type = BTRFS_XATTR_ITEM_KEY;
 787                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
 788                                 found_type = BTRFS_INODE_REF_KEY;
 789                         } else if (found_type) {
 790                                 found_type--;
 791                         } else {
 792                                 break;
 793                         }
 794                         btrfs_set_key_type(&key, found_type);
 795                         goto next;
 796                 }
 797                 if (found_key.offset >= inode->i_size)
 798                         del_item = 1;
 799                 else
 800                         del_item = 0;
 801                 found_extent = 0;
 802
 803                 /* FIXME, shrink the extent if the ref count is only 1 */
 804                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 805                         goto delete;
 806
 807                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 808                         u64 num_dec;
 809                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 810                         if (!del_item) {
 811                                 u64 orig_num_bytes =
 812                                         btrfs_file_extent_num_bytes(leaf, fi);
 813                                 extent_num_bytes = inode->i_size -
 814                                         found_key.offset + root->sectorsize - 1;
 815                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 816                                                          extent_num_bytes);
 817                                 num_dec = (orig_num_bytes -
 818                                            extent_num_bytes) >> 9;
 819                                 if (extent_start != 0) {
 820                                         inode->i_blocks -= num_dec;
 821                                 }
 822                                 btrfs_mark_buffer_dirty(leaf);
 823                         } else {
 824                                 extent_num_bytes =
 825                                         btrfs_file_extent_disk_num_bytes(leaf,
 826                                                                          fi);
 827                                 /* FIXME blocksize != 4096 */
 828                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 829                                                                        fi) >> 9;
 830                                 if (extent_start != 0) {
 831                                         found_extent = 1;
 832                                         inode->i_blocks -= num_dec;
 833                                 }
 834                                 root_gen = btrfs_header_generation(leaf);
 835                                 root_owner = btrfs_header_owner(leaf);
 836                         }
 837                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 838                            !del_item) {
 839                         u32 newsize = inode->i_size - found_key.offset;
 840                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 841                         ret = btrfs_truncate_item(trans, root, path,
 842                                                   newsize, 1);
 843                         BUG_ON(ret);
 844                 }
 845 delete:
 846                 if (del_item) {
 847                         if (!pending_del_nr) {
 848                                 /* no pending yet, add ourselves */
 849                                 pending_del_slot = path->slots[0];
 850                                 pending_del_nr = 1;
 851                         } else if (pending_del_nr &&
 852                                    path->slots[0] + 1 == pending_del_slot) {
 853                                 /* hop on the pending chunk */
 854                                 pending_del_nr++;
 855                                 pending_del_slot = path->slots[0];
 856                         } else {
 857                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
 858                         }
 859                 } else {
 860                         break;
 861                 }
 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 next:
 871                 if (path->slots[0] == 0) {
 872                         if (pending_del_nr)
 873                                 goto del_pending;
 874                         btrfs_release_path(root, path);
 875                         goto search_again;
 876                 }
 877
 878                 path->slots[0]--;
 879                 if (pending_del_nr &&
 880                     path->slots[0] + 1 != pending_del_slot) {
 881                         struct btrfs_key debug;
 882 del_pending:
 883                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
 884                                               pending_del_slot);
 885                         ret = btrfs_del_items(trans, root, path,
 886                                               pending_del_slot,
 887                                               pending_del_nr);
 888                         BUG_ON(ret);
 889                         pending_del_nr = 0;
 890                         btrfs_release_path(root, path);
 891                         goto search_again;
 892                 }
 893         }
 894         ret = 0;
 895 error:
 896         if (pending_del_nr) {
 897                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
 898                                       pending_del_nr);
 899         }
 900         btrfs_release_path(root, path);
 901         btrfs_free_path(path);
 902         inode->i_sb->s_dirt = 1;
 903         return ret;
 904 }
 905
 906 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 907                               size_t zero_start)
 908 {
 909         char *kaddr;
 910         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 911         struct btrfs_root *root = BTRFS_I(inode)->root;
 912         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 913         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 914         u64 existing_delalloc;
 915         u64 delalloc_start;
 916         int ret = 0;
 917
 918         WARN_ON(!PageLocked(page));
 919         set_page_extent_mapped(page);
 920
 921         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 922         delalloc_start = page_start;
 923         existing_delalloc = count_range_bits(&BTRFS_I(inode)->io_tree,
 924                                              &delalloc_start, page_end,
 925                                              PAGE_CACHE_SIZE, EXTENT_DELALLOC);
 926         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
 927                             page_end, GFP_NOFS);
 928
 929         spin_lock(&root->fs_info->delalloc_lock);
 930         root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE - existing_delalloc;
 931         spin_unlock(&root->fs_info->delalloc_lock);
 932
 933         if (zero_start != PAGE_CACHE_SIZE) {
 934                 kaddr = kmap(page);
 935                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 936                 flush_dcache_page(page);
 937                 kunmap(page);
 938         }
 939         set_page_dirty(page);
 940         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 941
 942         return ret;
 943 }
 944
 945 /*
 946  * taken from block_truncate_page, but does cow as it zeros out
 947  * any bytes left in the last page in the file.
 948  */
 949 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 950 {
 951         struct inode *inode = mapping->host;
 952         struct btrfs_root *root = BTRFS_I(inode)->root;
 953         u32 blocksize = root->sectorsize;
 954         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 955         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 956         struct page *page;
 957         int ret = 0;
 958         u64 page_start;
 959
 960         if ((offset & (blocksize - 1)) == 0)
 961                 goto out;
 962
 963         ret = -ENOMEM;
 964         page = grab_cache_page(mapping, index);
 965         if (!page)
 966                 goto out;
 967         if (!PageUptodate(page)) {
 968                 ret = btrfs_readpage(NULL, page);
 969                 lock_page(page);
 970                 if (!PageUptodate(page)) {
 971                         ret = -EIO;
 972                         goto out;
 973                 }
 974         }
 975         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 976
 977         ret = btrfs_cow_one_page(inode, page, offset);
 978
 979         unlock_page(page);
 980         page_cache_release(page);
 981 out:
 982         return ret;
 983 }
 984
 985 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 986 {
 987         struct inode *inode = dentry->d_inode;
 988         int err;
 989
 990         err = inode_change_ok(inode, attr);
 991         if (err)
 992                 return err;
 993
 994         if (S_ISREG(inode->i_mode) &&
 995             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 996                 struct btrfs_trans_handle *trans;
 997                 struct btrfs_root *root = BTRFS_I(inode)->root;
 998                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 999
1000                 u64 mask = root->sectorsize - 1;
1001                 u64 pos = (inode->i_size + mask) & ~mask;
1002                 u64 block_end = attr->ia_size | mask;
1003                 u64 hole_start;
1004                 u64 hole_size;
1005                 u64 alloc_hint = 0;
1006
1007                 if (attr->ia_size <= pos)
1008                         goto out;
1009
1010                 if (pos != inode->i_size)
1011                         hole_start = pos + root->sectorsize;
1012                 else
1013                         hole_start = pos;
1014
1015                 mutex_lock(&root->fs_info->fs_mutex);
1016                 err = btrfs_check_free_space(root, 1, 0);
1017                 mutex_unlock(&root->fs_info->fs_mutex);
1018                 if (err)
1019                         goto fail;
1020
1021                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1022
1023                 lock_extent(io_tree, pos, block_end, GFP_NOFS);
1024                 hole_size = block_end - hole_start;
1025
1026                 mutex_lock(&root->fs_info->fs_mutex);
1027                 trans = btrfs_start_transaction(root, 1);
1028                 btrfs_set_trans_block_group(trans, inode);
1029                 err = btrfs_drop_extents(trans, root, inode,
1030                                          pos, block_end, pos,
1031                                          &alloc_hint);
1032
1033                 if (alloc_hint != EXTENT_MAP_INLINE) {
1034                         err = btrfs_insert_file_extent(trans, root,
1035                                                        inode->i_ino,
1036                                                        hole_start, 0, 0,
1037                                                        hole_size);
1038                         btrfs_drop_extent_cache(inode, hole_start,
1039                                                 hole_size - 1);
1040                         btrfs_check_file(root, inode);
1041                 }
1042                 btrfs_end_transaction(trans, root);
1043                 mutex_unlock(&root->fs_info->fs_mutex);
1044                 unlock_extent(io_tree, pos, block_end, GFP_NOFS);
1045                 if (err)
1046                         return err;
1047         }
1048 out:
1049         err = inode_setattr(inode, attr);
1050 fail:
1051         return err;
1052 }
1053
1054 void btrfs_put_inode(struct inode *inode)
1055 {
1056         int ret;
1057
1058         if (!BTRFS_I(inode)->ordered_trans) {
1059                 return;
1060         }
1061
1062         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1063             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1064                 return;
1065
1066         ret = btrfs_del_ordered_inode(inode);
1067         if (ret == 1) {
1068                 atomic_dec(&inode->i_count);
1069         }
1070 }
1071
1072 void btrfs_delete_inode(struct inode *inode)
1073 {
1074         struct btrfs_trans_handle *trans;
1075         struct btrfs_root *root = BTRFS_I(inode)->root;
1076         unsigned long nr;
1077         int ret;
1078
1079         truncate_inode_pages(&inode->i_data, 0);
1080         if (is_bad_inode(inode)) {
1081                 goto no_delete;
1082         }
1083
1084         inode->i_size = 0;
1085         mutex_lock(&root->fs_info->fs_mutex);
1086         trans = btrfs_start_transaction(root, 1);
1087
1088         btrfs_set_trans_block_group(trans, inode);
1089         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1090         if (ret)
1091                 goto no_delete_lock;
1092
1093         nr = trans->blocks_used;
1094         clear_inode(inode);
1095
1096         btrfs_end_transaction(trans, root);
1097         mutex_unlock(&root->fs_info->fs_mutex);
1098         btrfs_btree_balance_dirty(root, nr);
1099         btrfs_throttle(root);
1100         return;
1101
1102 no_delete_lock:
1103         nr = trans->blocks_used;
1104         btrfs_end_transaction(trans, root);
1105         mutex_unlock(&root->fs_info->fs_mutex);
1106         btrfs_btree_balance_dirty(root, nr);
1107         btrfs_throttle(root);
1108 no_delete:
1109         clear_inode(inode);
1110 }
1111
1112 /*
1113  * this returns the key found in the dir entry in the location pointer.
1114  * If no dir entries were found, location->objectid is 0.
1115  */
1116 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1117                                struct btrfs_key *location)
1118 {
1119         const char *name = dentry->d_name.name;
1120         int namelen = dentry->d_name.len;
1121         struct btrfs_dir_item *di;
1122         struct btrfs_path *path;
1123         struct btrfs_root *root = BTRFS_I(dir)->root;
1124         int ret = 0;
1125
1126         if (namelen == 1 && strcmp(name, ".") == 0) {
1127                 location->objectid = dir->i_ino;
1128                 location->type = BTRFS_INODE_ITEM_KEY;
1129                 location->offset = 0;
1130                 return 0;
1131         }
1132         path = btrfs_alloc_path();
1133         BUG_ON(!path);
1134
1135         if (namelen == 2 && strcmp(name, "..") == 0) {
1136                 struct btrfs_key key;
1137                 struct extent_buffer *leaf;
1138                 u32 nritems;
1139                 int slot;
1140
1141                 key.objectid = dir->i_ino;
1142                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1143                 key.offset = 0;
1144                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1145                 BUG_ON(ret == 0);
1146                 ret = 0;
1147
1148                 leaf = path->nodes[0];
1149                 slot = path->slots[0];
1150                 nritems = btrfs_header_nritems(leaf);
1151                 if (slot >= nritems)
1152                         goto out_err;
1153
1154                 btrfs_item_key_to_cpu(leaf, &key, slot);
1155                 if (key.objectid != dir->i_ino ||
1156                     key.type != BTRFS_INODE_REF_KEY) {
1157                         goto out_err;
1158                 }
1159                 location->objectid = key.offset;
1160                 location->type = BTRFS_INODE_ITEM_KEY;
1161                 location->offset = 0;
1162                 goto out;
1163         }
1164
1165         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1166                                     namelen, 0);
1167         if (IS_ERR(di))
1168                 ret = PTR_ERR(di);
1169         if (!di || IS_ERR(di)) {
1170                 goto out_err;
1171         }
1172         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1173 out:
1174         btrfs_free_path(path);
1175         return ret;
1176 out_err:
1177         location->objectid = 0;
1178         goto out;
1179 }
1180
1181 /*
1182  * when we hit a tree root in a directory, the btrfs part of the inode
1183  * needs to be changed to reflect the root directory of the tree root.  This
1184  * is kind of like crossing a mount point.
1185  */
1186 static int fixup_tree_root_location(struct btrfs_root *root,
1187                              struct btrfs_key *location,
1188                              struct btrfs_root **sub_root,
1189                              struct dentry *dentry)
1190 {
1191         struct btrfs_path *path;
1192         struct btrfs_root_item *ri;
1193
1194         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1195                 return 0;
1196         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1197                 return 0;
1198
1199         path = btrfs_alloc_path();
1200         BUG_ON(!path);
1201         mutex_lock(&root->fs_info->fs_mutex);
1202
1203         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1204                                         dentry->d_name.name,
1205                                         dentry->d_name.len);
1206         if (IS_ERR(*sub_root))
1207                 return PTR_ERR(*sub_root);
1208
1209         ri = &(*sub_root)->root_item;
1210         location->objectid = btrfs_root_dirid(ri);
1211         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1212         location->offset = 0;
1213
1214         btrfs_free_path(path);
1215         mutex_unlock(&root->fs_info->fs_mutex);
1216         return 0;
1217 }
1218
1219 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1220 {
1221         struct btrfs_iget_args *args = p;
1222         inode->i_ino = args->ino;
1223         BTRFS_I(inode)->root = args->root;
1224         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1225         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1226                              inode->i_mapping, GFP_NOFS);
1227         return 0;
1228 }
1229
1230 static int btrfs_find_actor(struct inode *inode, void *opaque)
1231 {
1232         struct btrfs_iget_args *args = opaque;
1233         return (args->ino == inode->i_ino &&
1234                 args->root == BTRFS_I(inode)->root);
1235 }
1236
1237 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1238                             u64 root_objectid)
1239 {
1240         struct btrfs_iget_args args;
1241         args.ino = objectid;
1242         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1243
1244         if (!args.root)
1245                 return NULL;
1246
1247         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1248 }
1249
1250 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1251                                 struct btrfs_root *root)
1252 {
1253         struct inode *inode;
1254         struct btrfs_iget_args args;
1255         args.ino = objectid;
1256         args.root = root;
1257
1258         inode = iget5_locked(s, objectid, btrfs_find_actor,
1259                              btrfs_init_locked_inode,
1260                              (void *)&args);
1261         return inode;
1262 }
1263
1264 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1265                                    struct nameidata *nd)
1266 {
1267         struct inode * inode;
1268         struct btrfs_inode *bi = BTRFS_I(dir);
1269         struct btrfs_root *root = bi->root;
1270         struct btrfs_root *sub_root = root;
1271         struct btrfs_key location;
1272         int ret;
1273
1274         if (dentry->d_name.len > BTRFS_NAME_LEN)
1275                 return ERR_PTR(-ENAMETOOLONG);
1276
1277         mutex_lock(&root->fs_info->fs_mutex);
1278         ret = btrfs_inode_by_name(dir, dentry, &location);
1279         mutex_unlock(&root->fs_info->fs_mutex);
1280
1281         if (ret < 0)
1282                 return ERR_PTR(ret);
1283
1284         inode = NULL;
1285         if (location.objectid) {
1286                 ret = fixup_tree_root_location(root, &location, &sub_root,
1287                                                 dentry);
1288                 if (ret < 0)
1289                         return ERR_PTR(ret);
1290                 if (ret > 0)
1291                         return ERR_PTR(-ENOENT);
1292                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1293                                           sub_root);
1294                 if (!inode)
1295                         return ERR_PTR(-EACCES);
1296                 if (inode->i_state & I_NEW) {
1297                         /* the inode and parent dir are two different roots */
1298                         if (sub_root != root) {
1299                                 igrab(inode);
1300                                 sub_root->inode = inode;
1301                         }
1302                         BTRFS_I(inode)->root = sub_root;
1303                         memcpy(&BTRFS_I(inode)->location, &location,
1304                                sizeof(location));
1305                         btrfs_read_locked_inode(inode);
1306                         unlock_new_inode(inode);
1307                 }
1308         }
1309         return d_splice_alias(inode, dentry);
1310 }
1311
1312 static unsigned char btrfs_filetype_table[] = {
1313         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1314 };
1315
1316 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1317 {
1318         struct inode *inode = filp->f_dentry->d_inode;
1319         struct btrfs_root *root = BTRFS_I(inode)->root;
1320         struct btrfs_item *item;
1321         struct btrfs_dir_item *di;
1322         struct btrfs_key key;
1323         struct btrfs_key found_key;
1324         struct btrfs_path *path;
1325         int ret;
1326         u32 nritems;
1327         struct extent_buffer *leaf;
1328         int slot;
1329         int advance;
1330         unsigned char d_type;
1331         int over = 0;
1332         u32 di_cur;
1333         u32 di_total;
1334         u32 di_len;
1335         int key_type = BTRFS_DIR_INDEX_KEY;
1336         char tmp_name[32];
1337         char *name_ptr;
1338         int name_len;
1339
1340         /* FIXME, use a real flag for deciding about the key type */
1341         if (root->fs_info->tree_root == root)
1342                 key_type = BTRFS_DIR_ITEM_KEY;
1343
1344         /* special case for "." */
1345         if (filp->f_pos == 0) {
1346                 over = filldir(dirent, ".", 1,
1347                                1, inode->i_ino,
1348                                DT_DIR);
1349                 if (over)
1350                         return 0;
1351                 filp->f_pos = 1;
1352         }
1353
1354         mutex_lock(&root->fs_info->fs_mutex);
1355         key.objectid = inode->i_ino;
1356         path = btrfs_alloc_path();
1357         path->reada = 2;
1358
1359         /* special case for .., just use the back ref */
1360         if (filp->f_pos == 1) {
1361                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1362                 key.offset = 0;
1363                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1364                 BUG_ON(ret == 0);
1365                 leaf = path->nodes[0];
1366                 slot = path->slots[0];
1367                 nritems = btrfs_header_nritems(leaf);
1368                 if (slot >= nritems) {
1369                         btrfs_release_path(root, path);
1370                         goto read_dir_items;
1371                 }
1372                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1373                 btrfs_release_path(root, path);
1374                 if (found_key.objectid != key.objectid ||
1375                     found_key.type != BTRFS_INODE_REF_KEY)
1376                         goto read_dir_items;
1377                 over = filldir(dirent, "..", 2,
1378                                2, found_key.offset, DT_DIR);
1379                 if (over)
1380                         goto nopos;
1381                 filp->f_pos = 2;
1382         }
1383
1384 read_dir_items:
1385         btrfs_set_key_type(&key, key_type);
1386         key.offset = filp->f_pos;
1387
1388         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1389         if (ret < 0)
1390                 goto err;
1391         advance = 0;
1392         while(1) {
1393                 leaf = path->nodes[0];
1394                 nritems = btrfs_header_nritems(leaf);
1395                 slot = path->slots[0];
1396                 if (advance || slot >= nritems) {
1397                         if (slot >= nritems -1) {
1398                                 ret = btrfs_next_leaf(root, path);
1399                                 if (ret)
1400                                         break;
1401                                 leaf = path->nodes[0];
1402                                 nritems = btrfs_header_nritems(leaf);
1403                                 slot = path->slots[0];
1404                         } else {
1405                                 slot++;
1406                                 path->slots[0]++;
1407                         }
1408                 }
1409                 advance = 1;
1410                 item = btrfs_item_nr(leaf, slot);
1411                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1412
1413                 if (found_key.objectid != key.objectid)
1414                         break;
1415                 if (btrfs_key_type(&found_key) != key_type)
1416                         break;
1417                 if (found_key.offset < filp->f_pos)
1418                         continue;
1419
1420                 filp->f_pos = found_key.offset;
1421                 advance = 1;
1422                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1423                 di_cur = 0;
1424                 di_total = btrfs_item_size(leaf, item);
1425                 while(di_cur < di_total) {
1426                         struct btrfs_key location;
1427
1428                         name_len = btrfs_dir_name_len(leaf, di);
1429                         if (name_len < 32) {
1430                                 name_ptr = tmp_name;
1431                         } else {
1432                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1433                                 BUG_ON(!name_ptr);
1434                         }
1435                         read_extent_buffer(leaf, name_ptr,
1436                                            (unsigned long)(di + 1), name_len);
1437
1438                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1439                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1440                         over = filldir(dirent, name_ptr, name_len,
1441                                        found_key.offset,
1442                                        location.objectid,
1443                                        d_type);
1444
1445                         if (name_ptr != tmp_name)
1446                                 kfree(name_ptr);
1447
1448                         if (over)
1449                                 goto nopos;
1450                         di_len = btrfs_dir_name_len(leaf, di) +
1451                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1452                         di_cur += di_len;
1453                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1454                 }
1455         }
1456         filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1457 nopos:
1458         ret = 0;
1459 err:
1460         btrfs_release_path(root, path);
1461         btrfs_free_path(path);
1462         mutex_unlock(&root->fs_info->fs_mutex);
1463         return ret;
1464 }
1465
1466 int btrfs_write_inode(struct inode *inode, int wait)
1467 {
1468         struct btrfs_root *root = BTRFS_I(inode)->root;
1469         struct btrfs_trans_handle *trans;
1470         int ret = 0;
1471
1472         if (wait) {
1473                 mutex_lock(&root->fs_info->fs_mutex);
1474                 trans = btrfs_start_transaction(root, 1);
1475                 btrfs_set_trans_block_group(trans, inode);
1476                 ret = btrfs_commit_transaction(trans, root);
1477                 mutex_unlock(&root->fs_info->fs_mutex);
1478         }
1479         return ret;
1480 }
1481
1482 /*
1483  * This is somewhat expensive, updating the tree every time the
1484  * inode changes.  But, it is most likely to find the inode in cache.
1485  * FIXME, needs more benchmarking...there are no reasons other than performance
1486  * to keep or drop this code.
1487  */
1488 void btrfs_dirty_inode(struct inode *inode)
1489 {
1490         struct btrfs_root *root = BTRFS_I(inode)->root;
1491         struct btrfs_trans_handle *trans;
1492
1493         mutex_lock(&root->fs_info->fs_mutex);
1494         trans = btrfs_start_transaction(root, 1);
1495         btrfs_set_trans_block_group(trans, inode);
1496         btrfs_update_inode(trans, root, inode);
1497         btrfs_end_transaction(trans, root);
1498         mutex_unlock(&root->fs_info->fs_mutex);
1499 }
1500
1501 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1502                                      struct btrfs_root *root,
1503                                      const char *name, int name_len,
1504                                      u64 ref_objectid,
1505                                      u64 objectid,
1506                                      struct btrfs_block_group_cache *group,
1507                                      int mode)
1508 {
1509         struct inode *inode;
1510         struct btrfs_inode_item *inode_item;
1511         struct btrfs_key *location;
1512         struct btrfs_path *path;
1513         struct btrfs_inode_ref *ref;
1514         struct btrfs_key key[2];
1515         u32 sizes[2];
1516         unsigned long ptr;
1517         int ret;
1518         int owner;
1519
1520         path = btrfs_alloc_path();
1521         BUG_ON(!path);
1522
1523         inode = new_inode(root->fs_info->sb);
1524         if (!inode)
1525                 return ERR_PTR(-ENOMEM);
1526
1527         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1528         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1529                              inode->i_mapping, GFP_NOFS);
1530         BTRFS_I(inode)->root = root;
1531
1532         if (mode & S_IFDIR)
1533                 owner = 0;
1534         else
1535                 owner = 1;
1536         group = btrfs_find_block_group(root, group, 0, 0, owner);
1537         BTRFS_I(inode)->block_group = group;
1538         BTRFS_I(inode)->flags = 0;
1539
1540         key[0].objectid = objectid;
1541         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1542         key[0].offset = 0;
1543
1544         key[1].objectid = objectid;
1545         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1546         key[1].offset = ref_objectid;
1547
1548         sizes[0] = sizeof(struct btrfs_inode_item);
1549         sizes[1] = name_len + sizeof(*ref);
1550
1551         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1552         if (ret != 0)
1553                 goto fail;
1554
1555         if (objectid > root->highest_inode)
1556                 root->highest_inode = objectid;
1557
1558         inode->i_uid = current->fsuid;
1559         inode->i_gid = current->fsgid;
1560         inode->i_mode = mode;
1561         inode->i_ino = objectid;
1562         inode->i_blocks = 0;
1563         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1564         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1565                                   struct btrfs_inode_item);
1566         fill_inode_item(path->nodes[0], inode_item, inode);
1567
1568         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1569                              struct btrfs_inode_ref);
1570         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1571         ptr = (unsigned long)(ref + 1);
1572         write_extent_buffer(path->nodes[0], name, ptr, name_len);
1573
1574         btrfs_mark_buffer_dirty(path->nodes[0]);
1575         btrfs_free_path(path);
1576
1577         location = &BTRFS_I(inode)->location;
1578         location->objectid = objectid;
1579         location->offset = 0;
1580         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1581
1582         insert_inode_hash(inode);
1583         return inode;
1584 fail:
1585         btrfs_free_path(path);
1586         return ERR_PTR(ret);
1587 }
1588
1589 static inline u8 btrfs_inode_type(struct inode *inode)
1590 {
1591         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1592 }
1593
1594 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1595                             struct dentry *dentry, struct inode *inode,
1596                             int add_backref)
1597 {
1598         int ret;
1599         struct btrfs_key key;
1600         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1601         struct inode *parent_inode;
1602
1603         key.objectid = inode->i_ino;
1604         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1605         key.offset = 0;
1606
1607         ret = btrfs_insert_dir_item(trans, root,
1608                                     dentry->d_name.name, dentry->d_name.len,
1609                                     dentry->d_parent->d_inode->i_ino,
1610                                     &key, btrfs_inode_type(inode));
1611         if (ret == 0) {
1612                 if (add_backref) {
1613                         ret = btrfs_insert_inode_ref(trans, root,
1614                                              dentry->d_name.name,
1615                                              dentry->d_name.len,
1616                                              inode->i_ino,
1617                                              dentry->d_parent->d_inode->i_ino);
1618                 }
1619                 parent_inode = dentry->d_parent->d_inode;
1620                 parent_inode->i_size += dentry->d_name.len * 2;
1621                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1622                 ret = btrfs_update_inode(trans, root,
1623                                          dentry->d_parent->d_inode);
1624         }
1625         return ret;
1626 }
1627
1628 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1629                             struct dentry *dentry, struct inode *inode,
1630                             int backref)
1631 {
1632         int err = btrfs_add_link(trans, dentry, inode, backref);
1633         if (!err) {
1634                 d_instantiate(dentry, inode);
1635                 return 0;
1636         }
1637         if (err > 0)
1638                 err = -EEXIST;
1639         return err;
1640 }
1641
1642 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1643                         int mode, dev_t rdev)
1644 {
1645         struct btrfs_trans_handle *trans;
1646         struct btrfs_root *root = BTRFS_I(dir)->root;
1647         struct inode *inode = NULL;
1648         int err;
1649         int drop_inode = 0;
1650         u64 objectid;
1651         unsigned long nr = 0;
1652
1653         if (!new_valid_dev(rdev))
1654                 return -EINVAL;
1655
1656         mutex_lock(&root->fs_info->fs_mutex);
1657         err = btrfs_check_free_space(root, 1, 0);
1658         if (err)
1659                 goto fail;
1660
1661         trans = btrfs_start_transaction(root, 1);
1662         btrfs_set_trans_block_group(trans, dir);
1663
1664         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1665         if (err) {
1666                 err = -ENOSPC;
1667                 goto out_unlock;
1668         }
1669
1670         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1671                                 dentry->d_name.len,
1672                                 dentry->d_parent->d_inode->i_ino, 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, 0);
1680         if (err)
1681                 drop_inode = 1;
1682         else {
1683                 inode->i_op = &btrfs_special_inode_operations;
1684                 init_special_inode(inode, inode->i_mode, rdev);
1685                 btrfs_update_inode(trans, root, inode);
1686         }
1687         dir->i_sb->s_dirt = 1;
1688         btrfs_update_inode_block_group(trans, inode);
1689         btrfs_update_inode_block_group(trans, dir);
1690 out_unlock:
1691         nr = trans->blocks_used;
1692         btrfs_end_transaction(trans, root);
1693 fail:
1694         mutex_unlock(&root->fs_info->fs_mutex);
1695
1696         if (drop_inode) {
1697                 inode_dec_link_count(inode);
1698                 iput(inode);
1699         }
1700         btrfs_btree_balance_dirty(root, nr);
1701         btrfs_throttle(root);
1702         return err;
1703 }
1704
1705 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1706                         int mode, struct nameidata *nd)
1707 {
1708         struct btrfs_trans_handle *trans;
1709         struct btrfs_root *root = BTRFS_I(dir)->root;
1710         struct inode *inode = NULL;
1711         int err;
1712         int drop_inode = 0;
1713         unsigned long nr = 0;
1714         u64 objectid;
1715
1716         mutex_lock(&root->fs_info->fs_mutex);
1717         err = btrfs_check_free_space(root, 1, 0);
1718         if (err)
1719                 goto fail;
1720         trans = btrfs_start_transaction(root, 1);
1721         btrfs_set_trans_block_group(trans, dir);
1722
1723         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1724         if (err) {
1725                 err = -ENOSPC;
1726                 goto out_unlock;
1727         }
1728
1729         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1730                                 dentry->d_name.len,
1731                                 dentry->d_parent->d_inode->i_ino,
1732                                 objectid, BTRFS_I(dir)->block_group, mode);
1733         err = PTR_ERR(inode);
1734         if (IS_ERR(inode))
1735                 goto out_unlock;
1736
1737         btrfs_set_trans_block_group(trans, inode);
1738         err = btrfs_add_nondir(trans, dentry, inode, 0);
1739         if (err)
1740                 drop_inode = 1;
1741         else {
1742                 inode->i_mapping->a_ops = &btrfs_aops;
1743                 inode->i_fop = &btrfs_file_operations;
1744                 inode->i_op = &btrfs_file_inode_operations;
1745                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1746                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1747                                      inode->i_mapping, GFP_NOFS);
1748                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1749         }
1750         dir->i_sb->s_dirt = 1;
1751         btrfs_update_inode_block_group(trans, inode);
1752         btrfs_update_inode_block_group(trans, dir);
1753 out_unlock:
1754         nr = trans->blocks_used;
1755         btrfs_end_transaction(trans, root);
1756 fail:
1757         mutex_unlock(&root->fs_info->fs_mutex);
1758
1759         if (drop_inode) {
1760                 inode_dec_link_count(inode);
1761                 iput(inode);
1762         }
1763         btrfs_btree_balance_dirty(root, nr);
1764         btrfs_throttle(root);
1765         return err;
1766 }
1767
1768 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1769                       struct dentry *dentry)
1770 {
1771         struct btrfs_trans_handle *trans;
1772         struct btrfs_root *root = BTRFS_I(dir)->root;
1773         struct inode *inode = old_dentry->d_inode;
1774         unsigned long nr = 0;
1775         int err;
1776         int drop_inode = 0;
1777
1778         if (inode->i_nlink == 0)
1779                 return -ENOENT;
1780
1781 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1782         inode->i_nlink++;
1783 #else
1784         inc_nlink(inode);
1785 #endif
1786         mutex_lock(&root->fs_info->fs_mutex);
1787         err = btrfs_check_free_space(root, 1, 0);
1788         if (err)
1789                 goto fail;
1790         trans = btrfs_start_transaction(root, 1);
1791
1792         btrfs_set_trans_block_group(trans, dir);
1793         atomic_inc(&inode->i_count);
1794         err = btrfs_add_nondir(trans, dentry, inode, 1);
1795
1796         if (err)
1797                 drop_inode = 1;
1798
1799         dir->i_sb->s_dirt = 1;
1800         btrfs_update_inode_block_group(trans, dir);
1801         err = btrfs_update_inode(trans, root, inode);
1802
1803         if (err)
1804                 drop_inode = 1;
1805
1806         nr = trans->blocks_used;
1807         btrfs_end_transaction(trans, root);
1808 fail:
1809         mutex_unlock(&root->fs_info->fs_mutex);
1810
1811         if (drop_inode) {
1812                 inode_dec_link_count(inode);
1813                 iput(inode);
1814         }
1815         btrfs_btree_balance_dirty(root, nr);
1816         btrfs_throttle(root);
1817         return err;
1818 }
1819
1820 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1821 {
1822         struct inode *inode;
1823         struct btrfs_trans_handle *trans;
1824         struct btrfs_root *root = BTRFS_I(dir)->root;
1825         int err = 0;
1826         int drop_on_err = 0;
1827         u64 objectid;
1828         unsigned long nr = 1;
1829
1830         mutex_lock(&root->fs_info->fs_mutex);
1831         err = btrfs_check_free_space(root, 1, 0);
1832         if (err)
1833                 goto out_unlock;
1834
1835         trans = btrfs_start_transaction(root, 1);
1836         btrfs_set_trans_block_group(trans, dir);
1837
1838         if (IS_ERR(trans)) {
1839                 err = PTR_ERR(trans);
1840                 goto out_unlock;
1841         }
1842
1843         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1844         if (err) {
1845                 err = -ENOSPC;
1846                 goto out_unlock;
1847         }
1848
1849         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1850                                 dentry->d_name.len,
1851                                 dentry->d_parent->d_inode->i_ino, objectid,
1852                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1853         if (IS_ERR(inode)) {
1854                 err = PTR_ERR(inode);
1855                 goto out_fail;
1856         }
1857
1858         drop_on_err = 1;
1859         inode->i_op = &btrfs_dir_inode_operations;
1860         inode->i_fop = &btrfs_dir_file_operations;
1861         btrfs_set_trans_block_group(trans, inode);
1862
1863         inode->i_size = 0;
1864         err = btrfs_update_inode(trans, root, inode);
1865         if (err)
1866                 goto out_fail;
1867
1868         err = btrfs_add_link(trans, dentry, inode, 0);
1869         if (err)
1870                 goto out_fail;
1871
1872         d_instantiate(dentry, inode);
1873         drop_on_err = 0;
1874         dir->i_sb->s_dirt = 1;
1875         btrfs_update_inode_block_group(trans, inode);
1876         btrfs_update_inode_block_group(trans, dir);
1877
1878 out_fail:
1879         nr = trans->blocks_used;
1880         btrfs_end_transaction(trans, root);
1881
1882 out_unlock:
1883         mutex_unlock(&root->fs_info->fs_mutex);
1884         if (drop_on_err)
1885                 iput(inode);
1886         btrfs_btree_balance_dirty(root, nr);
1887         btrfs_throttle(root);
1888         return err;
1889 }
1890
1891 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1892                                     size_t pg_offset, u64 start, u64 len,
1893                                     int create)
1894 {
1895         int ret;
1896         int err = 0;
1897         u64 bytenr;
1898         u64 extent_start = 0;
1899         u64 extent_end = 0;
1900         u64 objectid = inode->i_ino;
1901         u32 found_type;
1902         struct btrfs_path *path;
1903         struct btrfs_root *root = BTRFS_I(inode)->root;
1904         struct btrfs_file_extent_item *item;
1905         struct extent_buffer *leaf;
1906         struct btrfs_key found_key;
1907         struct extent_map *em = NULL;
1908         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1909         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1910         struct btrfs_trans_handle *trans = NULL;
1911
1912         path = btrfs_alloc_path();
1913         BUG_ON(!path);
1914         mutex_lock(&root->fs_info->fs_mutex);
1915
1916 again:
1917         spin_lock(&em_tree->lock);
1918         em = lookup_extent_mapping(em_tree, start, len);
1919         spin_unlock(&em_tree->lock);
1920
1921         if (em) {
1922                 if (em->start > start) {
1923                         printk("get_extent lookup [%Lu %Lu] em [%Lu %Lu]\n",
1924                                start, len, em->start, em->len);
1925                         WARN_ON(1);
1926                 }
1927                 if (em->block_start == EXTENT_MAP_INLINE && page)
1928                         free_extent_map(em);
1929                 else
1930                         goto out;
1931         }
1932         em = alloc_extent_map(GFP_NOFS);
1933         if (!em) {
1934                 err = -ENOMEM;
1935                 goto out;
1936         }
1937
1938         em->start = EXTENT_MAP_HOLE;
1939         em->len = (u64)-1;
1940         em->bdev = inode->i_sb->s_bdev;
1941         ret = btrfs_lookup_file_extent(trans, root, path,
1942                                        objectid, start, trans != NULL);
1943         if (ret < 0) {
1944                 err = ret;
1945                 goto out;
1946         }
1947
1948         if (ret != 0) {
1949                 if (path->slots[0] == 0)
1950                         goto not_found;
1951                 path->slots[0]--;
1952         }
1953
1954         leaf = path->nodes[0];
1955         item = btrfs_item_ptr(leaf, path->slots[0],
1956                               struct btrfs_file_extent_item);
1957         /* are we inside the extent that was found? */
1958         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1959         found_type = btrfs_key_type(&found_key);
1960         if (found_key.objectid != objectid ||
1961             found_type != BTRFS_EXTENT_DATA_KEY) {
1962                 goto not_found;
1963         }
1964
1965         found_type = btrfs_file_extent_type(leaf, item);
1966         extent_start = found_key.offset;
1967         if (found_type == BTRFS_FILE_EXTENT_REG) {
1968                 extent_end = extent_start +
1969                        btrfs_file_extent_num_bytes(leaf, item);
1970                 err = 0;
1971                 if (start < extent_start || start >= extent_end) {
1972                         em->start = start;
1973                         if (start < extent_start) {
1974                                 if (start + len <= extent_start)
1975                                         goto not_found;
1976                                 em->len = extent_end - extent_start;
1977                         } else {
1978                                 em->len = len;
1979                         }
1980                         goto not_found_em;
1981                 }
1982                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1983                 if (bytenr == 0) {
1984                         em->start = extent_start;
1985                         em->len = extent_end - extent_start;
1986                         em->block_start = EXTENT_MAP_HOLE;
1987                         goto insert;
1988                 }
1989                 bytenr += btrfs_file_extent_offset(leaf, item);
1990                 em->block_start = bytenr;
1991                 em->start = extent_start;
1992                 em->len = extent_end - extent_start;
1993                 goto insert;
1994         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1995                 u64 page_start;
1996                 unsigned long ptr;
1997                 char *map;
1998                 size_t size;
1999                 size_t extent_offset;
2000                 size_t copy_size;
2001
2002                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2003                                                     path->slots[0]));
2004                 extent_end = (extent_start + size + root->sectorsize - 1) &
2005                         ~((u64)root->sectorsize - 1);
2006                 if (start < extent_start || start >= extent_end) {
2007                         em->start = start;
2008                         if (start < extent_start) {
2009                                 if (start + len <= extent_start)
2010                                         goto not_found;
2011                                 em->len = extent_end - extent_start;
2012                         } else {
2013                                 em->len = len;
2014                         }
2015                         goto not_found_em;
2016                 }
2017                 em->block_start = EXTENT_MAP_INLINE;
2018
2019                 if (!page) {
2020                         em->start = extent_start;
2021                         em->len = size;
2022                         goto out;
2023                 }
2024
2025                 page_start = page_offset(page) + pg_offset;
2026                 extent_offset = page_start - extent_start;
2027                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2028                                 size - extent_offset);
2029                 em->start = extent_start + extent_offset;
2030                 em->len = (copy_size + root->sectorsize - 1) &
2031                         ~((u64)root->sectorsize - 1);
2032                 map = kmap(page);
2033                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2034                 if (create == 0 && !PageUptodate(page)) {
2035                         read_extent_buffer(leaf, map + pg_offset, ptr,
2036                                            copy_size);
2037                         flush_dcache_page(page);
2038                 } else if (create && PageUptodate(page)) {
2039                         if (!trans) {
2040                                 kunmap(page);
2041                                 free_extent_map(em);
2042                                 em = NULL;
2043                                 btrfs_release_path(root, path);
2044                                 trans = btrfs_start_transaction(root, 1);
2045                                 goto again;
2046                         }
2047                         write_extent_buffer(leaf, map + pg_offset, ptr,
2048                                             copy_size);
2049                         btrfs_mark_buffer_dirty(leaf);
2050                 }
2051                 kunmap(page);
2052                 set_extent_uptodate(io_tree, em->start,
2053                                     extent_map_end(em) - 1, GFP_NOFS);
2054                 goto insert;
2055         } else {
2056                 printk("unkknown found_type %d\n", found_type);
2057                 WARN_ON(1);
2058         }
2059 not_found:
2060         em->start = start;
2061         em->len = len;
2062 not_found_em:
2063         em->block_start = EXTENT_MAP_HOLE;
2064 insert:
2065         btrfs_release_path(root, path);
2066         if (em->start > start || extent_map_end(em) <= start) {
2067                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2068                 err = -EIO;
2069                 goto out;
2070         }
2071
2072         err = 0;
2073         spin_lock(&em_tree->lock);
2074         ret = add_extent_mapping(em_tree, em);
2075         if (ret == -EEXIST) {
2076                 free_extent_map(em);
2077                 em = lookup_extent_mapping(em_tree, start, len);
2078                 if (!em) {
2079                         err = -EIO;
2080                         printk("failing to insert %Lu %Lu\n", start, len);
2081                 }
2082         }
2083         spin_unlock(&em_tree->lock);
2084 out:
2085         btrfs_free_path(path);
2086         if (trans) {
2087                 ret = btrfs_end_transaction(trans, root);
2088                 if (!err)
2089                         err = ret;
2090         }
2091         mutex_unlock(&root->fs_info->fs_mutex);
2092         if (err) {
2093                 free_extent_map(em);
2094                 WARN_ON(1);
2095                 return ERR_PTR(err);
2096         }
2097         return em;
2098 }
2099
2100 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2101 {
2102         return extent_bmap(mapping, iblock, btrfs_get_extent);
2103 }
2104
2105 int btrfs_readpage(struct file *file, struct page *page)
2106 {
2107         struct extent_io_tree *tree;
2108         tree = &BTRFS_I(page->mapping->host)->io_tree;
2109         return extent_read_full_page(tree, page, btrfs_get_extent);
2110 }
2111
2112 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2113 {
2114         struct extent_io_tree *tree;
2115
2116
2117         if (current->flags & PF_MEMALLOC) {
2118                 redirty_page_for_writepage(wbc, page);
2119                 unlock_page(page);
2120                 return 0;
2121         }
2122         tree = &BTRFS_I(page->mapping->host)->io_tree;
2123         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2124 }
2125
2126 static int btrfs_writepages(struct address_space *mapping,
2127                             struct writeback_control *wbc)
2128 {
2129         struct extent_io_tree *tree;
2130         tree = &BTRFS_I(mapping->host)->io_tree;
2131         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2132 }
2133
2134 static int
2135 btrfs_readpages(struct file *file, struct address_space *mapping,
2136                 struct list_head *pages, unsigned nr_pages)
2137 {
2138         struct extent_io_tree *tree;
2139         tree = &BTRFS_I(mapping->host)->io_tree;
2140         return extent_readpages(tree, mapping, pages, nr_pages,
2141                                 btrfs_get_extent);
2142 }
2143
2144 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2145 {
2146         struct extent_io_tree *tree;
2147         struct extent_map_tree *map;
2148         int ret;
2149
2150         tree = &BTRFS_I(page->mapping->host)->io_tree;
2151         map = &BTRFS_I(page->mapping->host)->extent_tree;
2152         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2153         if (ret == 1) {
2154                 ClearPagePrivate(page);
2155                 set_page_private(page, 0);
2156                 page_cache_release(page);
2157         }
2158         return ret;
2159 }
2160
2161 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2162 {
2163         struct extent_io_tree *tree;
2164
2165         tree = &BTRFS_I(page->mapping->host)->io_tree;
2166         extent_invalidatepage(tree, page, offset);
2167         btrfs_releasepage(page, GFP_NOFS);
2168 }
2169
2170 /*
2171  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2172  * called from a page fault handler when a page is first dirtied. Hence we must
2173  * be careful to check for EOF conditions here. We set the page up correctly
2174  * for a written page which means we get ENOSPC checking when writing into
2175  * holes and correct delalloc and unwritten extent mapping on filesystems that
2176  * support these features.
2177  *
2178  * We are not allowed to take the i_mutex here so we have to play games to
2179  * protect against truncate races as the page could now be beyond EOF.  Because
2180  * vmtruncate() writes the inode size before removing pages, once we have the
2181  * page lock we can determine safely if the page is beyond EOF. If it is not
2182  * beyond EOF, then the page is guaranteed safe against truncation until we
2183  * unlock the page.
2184  */
2185 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2186 {
2187         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2188         struct btrfs_root *root = BTRFS_I(inode)->root;
2189         unsigned long end;
2190         loff_t size;
2191         int ret;
2192         u64 page_start;
2193
2194         mutex_lock(&root->fs_info->fs_mutex);
2195         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2196         mutex_unlock(&root->fs_info->fs_mutex);
2197         if (ret)
2198                 goto out;
2199
2200         ret = -EINVAL;
2201
2202         lock_page(page);
2203         wait_on_page_writeback(page);
2204         size = i_size_read(inode);
2205         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2206
2207         if ((page->mapping != inode->i_mapping) ||
2208             (page_start > size)) {
2209                 /* page got truncated out from underneath us */
2210                 goto out_unlock;
2211         }
2212
2213         /* page is wholly or partially inside EOF */
2214         if (page_start + PAGE_CACHE_SIZE > size)
2215                 end = size & ~PAGE_CACHE_MASK;
2216         else
2217                 end = PAGE_CACHE_SIZE;
2218
2219         ret = btrfs_cow_one_page(inode, page, end);
2220
2221 out_unlock:
2222         unlock_page(page);
2223 out:
2224         return ret;
2225 }
2226
2227 static void btrfs_truncate(struct inode *inode)
2228 {
2229         struct btrfs_root *root = BTRFS_I(inode)->root;
2230         int ret;
2231         struct btrfs_trans_handle *trans;
2232         unsigned long nr;
2233
2234         if (!S_ISREG(inode->i_mode))
2235                 return;
2236         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2237                 return;
2238
2239         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2240
2241         mutex_lock(&root->fs_info->fs_mutex);
2242         trans = btrfs_start_transaction(root, 1);
2243         btrfs_set_trans_block_group(trans, inode);
2244
2245         /* FIXME, add redo link to tree so we don't leak on crash */
2246         ret = btrfs_truncate_in_trans(trans, root, inode,
2247                                       BTRFS_EXTENT_DATA_KEY);
2248         btrfs_update_inode(trans, root, inode);
2249         nr = trans->blocks_used;
2250
2251         ret = btrfs_end_transaction(trans, root);
2252         BUG_ON(ret);
2253         mutex_unlock(&root->fs_info->fs_mutex);
2254         btrfs_btree_balance_dirty(root, nr);
2255         btrfs_throttle(root);
2256 }
2257
2258 static int noinline create_subvol(struct btrfs_root *root, char *name,
2259                                   int namelen)
2260 {
2261         struct btrfs_trans_handle *trans;
2262         struct btrfs_key key;
2263         struct btrfs_root_item root_item;
2264         struct btrfs_inode_item *inode_item;
2265         struct extent_buffer *leaf;
2266         struct btrfs_root *new_root = root;
2267         struct inode *inode;
2268         struct inode *dir;
2269         int ret;
2270         int err;
2271         u64 objectid;
2272         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2273         unsigned long nr = 1;
2274
2275         mutex_lock(&root->fs_info->fs_mutex);
2276         ret = btrfs_check_free_space(root, 1, 0);
2277         if (ret)
2278                 goto fail_commit;
2279
2280         trans = btrfs_start_transaction(root, 1);
2281         BUG_ON(!trans);
2282
2283         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2284                                        0, &objectid);
2285         if (ret)
2286                 goto fail;
2287
2288         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2289                                         objectid, trans->transid, 0, 0,
2290                                         0, 0);
2291         if (IS_ERR(leaf))
2292                 return PTR_ERR(leaf);
2293
2294         btrfs_set_header_nritems(leaf, 0);
2295         btrfs_set_header_level(leaf, 0);
2296         btrfs_set_header_bytenr(leaf, leaf->start);
2297         btrfs_set_header_generation(leaf, trans->transid);
2298         btrfs_set_header_owner(leaf, objectid);
2299
2300         write_extent_buffer(leaf, root->fs_info->fsid,
2301                             (unsigned long)btrfs_header_fsid(leaf),
2302                             BTRFS_FSID_SIZE);
2303         btrfs_mark_buffer_dirty(leaf);
2304
2305         inode_item = &root_item.inode;
2306         memset(inode_item, 0, sizeof(*inode_item));
2307         inode_item->generation = cpu_to_le64(1);
2308         inode_item->size = cpu_to_le64(3);
2309         inode_item->nlink = cpu_to_le32(1);
2310         inode_item->nblocks = cpu_to_le64(1);
2311         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2312
2313         btrfs_set_root_bytenr(&root_item, leaf->start);
2314         btrfs_set_root_level(&root_item, 0);
2315         btrfs_set_root_refs(&root_item, 1);
2316         btrfs_set_root_used(&root_item, 0);
2317
2318         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2319         root_item.drop_level = 0;
2320
2321         free_extent_buffer(leaf);
2322         leaf = NULL;
2323
2324         btrfs_set_root_dirid(&root_item, new_dirid);
2325
2326         key.objectid = objectid;
2327         key.offset = 1;
2328         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2329         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2330                                 &root_item);
2331         if (ret)
2332                 goto fail;
2333
2334         /*
2335          * insert the directory item
2336          */
2337         key.offset = (u64)-1;
2338         dir = root->fs_info->sb->s_root->d_inode;
2339         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2340                                     name, namelen, dir->i_ino, &key,
2341                                     BTRFS_FT_DIR);
2342         if (ret)
2343                 goto fail;
2344
2345         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2346                              name, namelen, objectid,
2347                              root->fs_info->sb->s_root->d_inode->i_ino);
2348         if (ret)
2349                 goto fail;
2350
2351         ret = btrfs_commit_transaction(trans, root);
2352         if (ret)
2353                 goto fail_commit;
2354
2355         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2356         BUG_ON(!new_root);
2357
2358         trans = btrfs_start_transaction(new_root, 1);
2359         BUG_ON(!trans);
2360
2361         inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2362                                 new_dirid,
2363                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2364         if (IS_ERR(inode))
2365                 goto fail;
2366         inode->i_op = &btrfs_dir_inode_operations;
2367         inode->i_fop = &btrfs_dir_file_operations;
2368         new_root->inode = inode;
2369
2370         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2371                                      new_dirid);
2372         inode->i_nlink = 1;
2373         inode->i_size = 0;
2374         ret = btrfs_update_inode(trans, new_root, inode);
2375         if (ret)
2376                 goto fail;
2377 fail:
2378         nr = trans->blocks_used;
2379         err = btrfs_commit_transaction(trans, new_root);
2380         if (err && !ret)
2381                 ret = err;
2382 fail_commit:
2383         mutex_unlock(&root->fs_info->fs_mutex);
2384         btrfs_btree_balance_dirty(root, nr);
2385         btrfs_throttle(root);
2386         return ret;
2387 }
2388
2389 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2390 {
2391         struct btrfs_pending_snapshot *pending_snapshot;
2392         struct btrfs_trans_handle *trans;
2393         int ret;
2394         int err;
2395         unsigned long nr = 0;
2396
2397         if (!root->ref_cows)
2398                 return -EINVAL;
2399
2400         mutex_lock(&root->fs_info->fs_mutex);
2401         ret = btrfs_check_free_space(root, 1, 0);
2402         if (ret)
2403                 goto fail_unlock;
2404
2405         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2406         if (!pending_snapshot) {
2407                 ret = -ENOMEM;
2408                 goto fail_unlock;
2409         }
2410         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2411         if (!pending_snapshot->name) {
2412                 ret = -ENOMEM;
2413                 kfree(pending_snapshot);
2414                 goto fail_unlock;
2415         }
2416         memcpy(pending_snapshot->name, name, namelen);
2417         pending_snapshot->name[namelen] = '\0';
2418         trans = btrfs_start_transaction(root, 1);
2419         BUG_ON(!trans);
2420         pending_snapshot->root = root;
2421         list_add(&pending_snapshot->list,
2422                  &trans->transaction->pending_snapshots);
2423         ret = btrfs_update_inode(trans, root, root->inode);
2424         err = btrfs_commit_transaction(trans, root);
2425
2426 fail_unlock:
2427         mutex_unlock(&root->fs_info->fs_mutex);
2428         btrfs_btree_balance_dirty(root, nr);
2429         btrfs_throttle(root);
2430         return ret;
2431 }
2432
2433 unsigned long btrfs_force_ra(struct address_space *mapping,
2434                               struct file_ra_state *ra, struct file *file,
2435                               pgoff_t offset, pgoff_t last_index)
2436 {
2437         pgoff_t req_size;
2438
2439 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2440         req_size = last_index - offset + 1;
2441         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2442         return offset;
2443 #else
2444         req_size = min(last_index - offset + 1, (pgoff_t)128);
2445         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2446         return offset + req_size;
2447 #endif
2448 }
2449
2450 int btrfs_defrag_file(struct file *file) {
2451         struct inode *inode = fdentry(file)->d_inode;
2452         struct btrfs_root *root = BTRFS_I(inode)->root;
2453         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2454         struct page *page;
2455         unsigned long last_index;
2456         unsigned long ra_index = 0;
2457         u64 page_start;
2458         u64 page_end;
2459         u64 delalloc_start;
2460         u64 existing_delalloc;
2461         unsigned long i;
2462         int ret;
2463
2464         mutex_lock(&root->fs_info->fs_mutex);
2465         ret = btrfs_check_free_space(root, inode->i_size, 0);
2466         mutex_unlock(&root->fs_info->fs_mutex);
2467         if (ret)
2468                 return -ENOSPC;
2469
2470         mutex_lock(&inode->i_mutex);
2471         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2472         for (i = 0; i <= last_index; i++) {
2473                 if (i == ra_index) {
2474                         ra_index = btrfs_force_ra(inode->i_mapping,
2475                                                   &file->f_ra,
2476                                                   file, ra_index, last_index);
2477                 }
2478                 page = grab_cache_page(inode->i_mapping, i);
2479                 if (!page)
2480                         goto out_unlock;
2481                 if (!PageUptodate(page)) {
2482                         btrfs_readpage(NULL, page);
2483                         lock_page(page);
2484                         if (!PageUptodate(page)) {
2485                                 unlock_page(page);
2486                                 page_cache_release(page);
2487                                 goto out_unlock;
2488                         }
2489                 }
2490                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2491                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2492
2493                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2494                 delalloc_start = page_start;
2495                 existing_delalloc =
2496                         count_range_bits(&BTRFS_I(inode)->io_tree,
2497                                          &delalloc_start, page_end,
2498                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2499                 set_extent_delalloc(io_tree, page_start,
2500                                     page_end, GFP_NOFS);
2501
2502                 spin_lock(&root->fs_info->delalloc_lock);
2503                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2504                                                  existing_delalloc;
2505                 spin_unlock(&root->fs_info->delalloc_lock);
2506
2507                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2508                 set_page_dirty(page);
2509                 unlock_page(page);
2510                 page_cache_release(page);
2511                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2512         }
2513
2514 out_unlock:
2515         mutex_unlock(&inode->i_mutex);
2516         return 0;
2517 }
2518
2519 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2520 {
2521         u64 new_size;
2522         u64 old_size;
2523         struct btrfs_ioctl_vol_args *vol_args;
2524         struct btrfs_trans_handle *trans;
2525         char *sizestr;
2526         int ret = 0;
2527         int namelen;
2528         int mod = 0;
2529
2530         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2531
2532         if (!vol_args)
2533                 return -ENOMEM;
2534
2535         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2536                 ret = -EFAULT;
2537                 goto out;
2538         }
2539         namelen = strlen(vol_args->name);
2540         if (namelen > BTRFS_VOL_NAME_MAX) {
2541                 ret = -EINVAL;
2542                 goto out;
2543         }
2544
2545         sizestr = vol_args->name;
2546         if (!strcmp(sizestr, "max"))
2547                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2548         else {
2549                 if (sizestr[0] == '-') {
2550                         mod = -1;
2551                         sizestr++;
2552                 } else if (sizestr[0] == '+') {
2553                         mod = 1;
2554                         sizestr++;
2555                 }
2556                 new_size = btrfs_parse_size(sizestr);
2557                 if (new_size == 0) {
2558                         ret = -EINVAL;
2559                         goto out;
2560                 }
2561         }
2562
2563         mutex_lock(&root->fs_info->fs_mutex);
2564         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2565
2566         if (mod < 0) {
2567                 if (new_size > old_size) {
2568                         ret = -EINVAL;
2569                         goto out_unlock;
2570                 }
2571                 new_size = old_size - new_size;
2572         } else if (mod > 0) {
2573                 new_size = old_size + new_size;
2574         }
2575
2576         if (new_size < 256 * 1024 * 1024) {
2577                 ret = -EINVAL;
2578                 goto out_unlock;
2579         }
2580         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2581                 ret = -EFBIG;
2582                 goto out_unlock;
2583         }
2584
2585         do_div(new_size, root->sectorsize);
2586         new_size *= root->sectorsize;
2587
2588 printk("new size is %Lu\n", new_size);
2589         if (new_size > old_size) {
2590                 trans = btrfs_start_transaction(root, 1);
2591                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2592                 btrfs_commit_transaction(trans, root);
2593         } else {
2594                 ret = btrfs_shrink_extent_tree(root, new_size);
2595         }
2596
2597 out_unlock:
2598         mutex_unlock(&root->fs_info->fs_mutex);
2599 out:
2600         kfree(vol_args);
2601         return ret;
2602 }
2603
2604 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2605                                             void __user *arg)
2606 {
2607         struct btrfs_ioctl_vol_args *vol_args;
2608         struct btrfs_dir_item *di;
2609         struct btrfs_path *path;
2610         u64 root_dirid;
2611         int namelen;
2612         int ret;
2613
2614         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2615
2616         if (!vol_args)
2617                 return -ENOMEM;
2618
2619         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2620                 ret = -EFAULT;
2621                 goto out;
2622         }
2623
2624         namelen = strlen(vol_args->name);
2625         if (namelen > BTRFS_VOL_NAME_MAX) {
2626                 ret = -EINVAL;
2627                 goto out;
2628         }
2629         if (strchr(vol_args->name, '/')) {
2630                 ret = -EINVAL;
2631                 goto out;
2632         }
2633
2634         path = btrfs_alloc_path();
2635         if (!path) {
2636                 ret = -ENOMEM;
2637                 goto out;
2638         }
2639
2640         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2641         mutex_lock(&root->fs_info->fs_mutex);
2642         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2643                             path, root_dirid,
2644                             vol_args->name, namelen, 0);
2645         mutex_unlock(&root->fs_info->fs_mutex);
2646         btrfs_free_path(path);
2647
2648         if (di && !IS_ERR(di)) {
2649                 ret = -EEXIST;
2650                 goto out;
2651         }
2652
2653         if (IS_ERR(di)) {
2654                 ret = PTR_ERR(di);
2655                 goto out;
2656         }
2657
2658         if (root == root->fs_info->tree_root)
2659                 ret = create_subvol(root, vol_args->name, namelen);
2660         else
2661                 ret = create_snapshot(root, vol_args->name, namelen);
2662 out:
2663         kfree(vol_args);
2664         return ret;
2665 }
2666
2667 static int btrfs_ioctl_defrag(struct file *file)
2668 {
2669         struct inode *inode = fdentry(file)->d_inode;
2670         struct btrfs_root *root = BTRFS_I(inode)->root;
2671
2672         switch (inode->i_mode & S_IFMT) {
2673         case S_IFDIR:
2674                 mutex_lock(&root->fs_info->fs_mutex);
2675                 btrfs_defrag_root(root, 0);
2676                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2677                 mutex_unlock(&root->fs_info->fs_mutex);
2678                 break;
2679         case S_IFREG:
2680                 btrfs_defrag_file(file);
2681                 break;
2682         }
2683
2684         return 0;
2685 }
2686
2687 long btrfs_ioctl(struct file *file, unsigned int
2688                 cmd, unsigned long arg)
2689 {
2690         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2691
2692         switch (cmd) {
2693         case BTRFS_IOC_SNAP_CREATE:
2694                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2695         case BTRFS_IOC_DEFRAG:
2696                 return btrfs_ioctl_defrag(file);
2697         case BTRFS_IOC_RESIZE:
2698                 return btrfs_ioctl_resize(root, (void __user *)arg);
2699         }
2700
2701         return -ENOTTY;
2702 }
2703
2704 /*
2705  * Called inside transaction, so use GFP_NOFS
2706  */
2707 struct inode *btrfs_alloc_inode(struct super_block *sb)
2708 {
2709         struct btrfs_inode *ei;
2710
2711         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2712         if (!ei)
2713                 return NULL;
2714         ei->last_trans = 0;
2715         ei->ordered_trans = 0;
2716         return &ei->vfs_inode;
2717 }
2718
2719 void btrfs_destroy_inode(struct inode *inode)
2720 {
2721         WARN_ON(!list_empty(&inode->i_dentry));
2722         WARN_ON(inode->i_data.nrpages);
2723
2724         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2725         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2726 }
2727
2728 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2729 static void init_once(struct kmem_cache * cachep, void *foo)
2730 #else
2731 static void init_once(void * foo, struct kmem_cache * cachep,
2732                       unsigned long flags)
2733 #endif
2734 {
2735         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2736
2737         inode_init_once(&ei->vfs_inode);
2738 }
2739
2740 void btrfs_destroy_cachep(void)
2741 {
2742         if (btrfs_inode_cachep)
2743                 kmem_cache_destroy(btrfs_inode_cachep);
2744         if (btrfs_trans_handle_cachep)
2745                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2746         if (btrfs_transaction_cachep)
2747                 kmem_cache_destroy(btrfs_transaction_cachep);
2748         if (btrfs_bit_radix_cachep)
2749                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2750         if (btrfs_path_cachep)
2751                 kmem_cache_destroy(btrfs_path_cachep);
2752 }
2753
2754 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2755                                        unsigned long extra_flags,
2756 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2757                                        void (*ctor)(struct kmem_cache *, void *)
2758 #else
2759                                        void (*ctor)(void *, struct kmem_cache *,
2760                                                     unsigned long)
2761 #endif
2762                                      )
2763 {
2764         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2765                                  SLAB_MEM_SPREAD | extra_flags), ctor
2766 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2767                                  ,NULL
2768 #endif
2769                                 );
2770 }
2771
2772 int btrfs_init_cachep(void)
2773 {
2774         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2775                                           sizeof(struct btrfs_inode),
2776                                           0, init_once);
2777         if (!btrfs_inode_cachep)
2778                 goto fail;
2779         btrfs_trans_handle_cachep =
2780                         btrfs_cache_create("btrfs_trans_handle_cache",
2781                                            sizeof(struct btrfs_trans_handle),
2782                                            0, NULL);
2783         if (!btrfs_trans_handle_cachep)
2784                 goto fail;
2785         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2786                                              sizeof(struct btrfs_transaction),
2787                                              0, NULL);
2788         if (!btrfs_transaction_cachep)
2789                 goto fail;
2790         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2791                                          sizeof(struct btrfs_path),
2792                                          0, NULL);
2793         if (!btrfs_path_cachep)
2794                 goto fail;
2795         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2796                                               SLAB_DESTROY_BY_RCU, NULL);
2797         if (!btrfs_bit_radix_cachep)
2798                 goto fail;
2799         return 0;
2800 fail:
2801         btrfs_destroy_cachep();
2802         return -ENOMEM;
2803 }
2804
2805 static int btrfs_getattr(struct vfsmount *mnt,
2806                          struct dentry *dentry, struct kstat *stat)
2807 {
2808         struct inode *inode = dentry->d_inode;
2809         generic_fillattr(inode, stat);
2810         stat->blksize = PAGE_CACHE_SIZE;
2811         return 0;
2812 }
2813
2814 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2815                            struct inode * new_dir,struct dentry *new_dentry)
2816 {
2817         struct btrfs_trans_handle *trans;
2818         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2819         struct inode *new_inode = new_dentry->d_inode;
2820         struct inode *old_inode = old_dentry->d_inode;
2821         struct timespec ctime = CURRENT_TIME;
2822         struct btrfs_path *path;
2823         int ret;
2824
2825         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2826             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2827                 return -ENOTEMPTY;
2828         }
2829
2830         mutex_lock(&root->fs_info->fs_mutex);
2831         ret = btrfs_check_free_space(root, 1, 0);
2832         if (ret)
2833                 goto out_unlock;
2834
2835         trans = btrfs_start_transaction(root, 1);
2836
2837         btrfs_set_trans_block_group(trans, new_dir);
2838         path = btrfs_alloc_path();
2839         if (!path) {
2840                 ret = -ENOMEM;
2841                 goto out_fail;
2842         }
2843
2844         old_dentry->d_inode->i_nlink++;
2845         old_dir->i_ctime = old_dir->i_mtime = ctime;
2846         new_dir->i_ctime = new_dir->i_mtime = ctime;
2847         old_inode->i_ctime = ctime;
2848
2849         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2850         if (ret)
2851                 goto out_fail;
2852
2853         if (new_inode) {
2854                 new_inode->i_ctime = CURRENT_TIME;
2855                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2856                 if (ret)
2857                         goto out_fail;
2858         }
2859         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
2860         if (ret)
2861                 goto out_fail;
2862
2863 out_fail:
2864         btrfs_free_path(path);
2865         btrfs_end_transaction(trans, root);
2866 out_unlock:
2867         mutex_unlock(&root->fs_info->fs_mutex);
2868         return ret;
2869 }
2870
2871 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2872                          const char *symname)
2873 {
2874         struct btrfs_trans_handle *trans;
2875         struct btrfs_root *root = BTRFS_I(dir)->root;
2876         struct btrfs_path *path;
2877         struct btrfs_key key;
2878         struct inode *inode = NULL;
2879         int err;
2880         int drop_inode = 0;
2881         u64 objectid;
2882         int name_len;
2883         int datasize;
2884         unsigned long ptr;
2885         struct btrfs_file_extent_item *ei;
2886         struct extent_buffer *leaf;
2887         unsigned long nr = 0;
2888
2889         name_len = strlen(symname) + 1;
2890         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2891                 return -ENAMETOOLONG;
2892
2893         mutex_lock(&root->fs_info->fs_mutex);
2894         err = btrfs_check_free_space(root, 1, 0);
2895         if (err)
2896                 goto out_fail;
2897
2898         trans = btrfs_start_transaction(root, 1);
2899         btrfs_set_trans_block_group(trans, dir);
2900
2901         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2902         if (err) {
2903                 err = -ENOSPC;
2904                 goto out_unlock;
2905         }
2906
2907         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2908                                 dentry->d_name.len,
2909                                 dentry->d_parent->d_inode->i_ino, objectid,
2910                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2911         err = PTR_ERR(inode);
2912         if (IS_ERR(inode))
2913                 goto out_unlock;
2914
2915         btrfs_set_trans_block_group(trans, inode);
2916         err = btrfs_add_nondir(trans, dentry, inode, 0);
2917         if (err)
2918                 drop_inode = 1;
2919         else {
2920                 inode->i_mapping->a_ops = &btrfs_aops;
2921                 inode->i_fop = &btrfs_file_operations;
2922                 inode->i_op = &btrfs_file_inode_operations;
2923                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2924                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2925                                      inode->i_mapping, GFP_NOFS);
2926                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2927         }
2928         dir->i_sb->s_dirt = 1;
2929         btrfs_update_inode_block_group(trans, inode);
2930         btrfs_update_inode_block_group(trans, dir);
2931         if (drop_inode)
2932                 goto out_unlock;
2933
2934         path = btrfs_alloc_path();
2935         BUG_ON(!path);
2936         key.objectid = inode->i_ino;
2937         key.offset = 0;
2938         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2939         datasize = btrfs_file_extent_calc_inline_size(name_len);
2940         err = btrfs_insert_empty_item(trans, root, path, &key,
2941                                       datasize);
2942         if (err) {
2943                 drop_inode = 1;
2944                 goto out_unlock;
2945         }
2946         leaf = path->nodes[0];
2947         ei = btrfs_item_ptr(leaf, path->slots[0],
2948                             struct btrfs_file_extent_item);
2949         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2950         btrfs_set_file_extent_type(leaf, ei,
2951                                    BTRFS_FILE_EXTENT_INLINE);
2952         ptr = btrfs_file_extent_inline_start(ei);
2953         write_extent_buffer(leaf, symname, ptr, name_len);
2954         btrfs_mark_buffer_dirty(leaf);
2955         btrfs_free_path(path);
2956
2957         inode->i_op = &btrfs_symlink_inode_operations;
2958         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2959         inode->i_size = name_len - 1;
2960         err = btrfs_update_inode(trans, root, inode);
2961         if (err)
2962                 drop_inode = 1;
2963
2964 out_unlock:
2965         nr = trans->blocks_used;
2966         btrfs_end_transaction(trans, root);
2967 out_fail:
2968         mutex_unlock(&root->fs_info->fs_mutex);
2969         if (drop_inode) {
2970                 inode_dec_link_count(inode);
2971                 iput(inode);
2972         }
2973         btrfs_btree_balance_dirty(root, nr);
2974         btrfs_throttle(root);
2975         return err;
2976 }
2977 static int btrfs_permission(struct inode *inode, int mask,
2978                             struct nameidata *nd)
2979 {
2980         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2981                 return -EACCES;
2982         return generic_permission(inode, mask, NULL);
2983 }
2984
2985 static struct inode_operations btrfs_dir_inode_operations = {
2986         .lookup         = btrfs_lookup,
2987         .create         = btrfs_create,
2988         .unlink         = btrfs_unlink,
2989         .link           = btrfs_link,
2990         .mkdir          = btrfs_mkdir,
2991         .rmdir          = btrfs_rmdir,
2992         .rename         = btrfs_rename,
2993         .symlink        = btrfs_symlink,
2994         .setattr        = btrfs_setattr,
2995         .mknod          = btrfs_mknod,
2996         .setxattr       = generic_setxattr,
2997         .getxattr       = generic_getxattr,
2998         .listxattr      = btrfs_listxattr,
2999         .removexattr    = generic_removexattr,
3000         .permission     = btrfs_permission,
3001 };
3002 static struct inode_operations btrfs_dir_ro_inode_operations = {
3003         .lookup         = btrfs_lookup,
3004         .permission     = btrfs_permission,
3005 };
3006 static struct file_operations btrfs_dir_file_operations = {
3007         .llseek         = generic_file_llseek,
3008         .read           = generic_read_dir,
3009         .readdir        = btrfs_readdir,
3010         .unlocked_ioctl = btrfs_ioctl,
3011 #ifdef CONFIG_COMPAT
3012         .compat_ioctl   = btrfs_ioctl,
3013 #endif
3014 };
3015
3016 static struct extent_io_ops btrfs_extent_io_ops = {
3017         .fill_delalloc = run_delalloc_range,
3018         .writepage_io_hook = btrfs_writepage_io_hook,
3019         .readpage_io_hook = btrfs_readpage_io_hook,
3020         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3021 };
3022
3023 static struct address_space_operations btrfs_aops = {
3024         .readpage       = btrfs_readpage,
3025         .writepage      = btrfs_writepage,
3026         .writepages     = btrfs_writepages,
3027         .readpages      = btrfs_readpages,
3028         .sync_page      = block_sync_page,
3029         .bmap           = btrfs_bmap,
3030         .invalidatepage = btrfs_invalidatepage,
3031         .releasepage    = btrfs_releasepage,
3032         .set_page_dirty = __set_page_dirty_nobuffers,
3033 };
3034
3035 static struct address_space_operations btrfs_symlink_aops = {
3036         .readpage       = btrfs_readpage,
3037         .writepage      = btrfs_writepage,
3038         .invalidatepage = btrfs_invalidatepage,
3039         .releasepage    = btrfs_releasepage,
3040 };
3041
3042 static struct inode_operations btrfs_file_inode_operations = {
3043         .truncate       = btrfs_truncate,
3044         .getattr        = btrfs_getattr,
3045         .setattr        = btrfs_setattr,
3046         .setxattr       = generic_setxattr,
3047         .getxattr       = generic_getxattr,
3048         .listxattr      = btrfs_listxattr,
3049         .removexattr    = generic_removexattr,
3050         .permission     = btrfs_permission,
3051 };
3052 static struct inode_operations btrfs_special_inode_operations = {
3053         .getattr        = btrfs_getattr,
3054         .setattr        = btrfs_setattr,
3055         .permission     = btrfs_permission,
3056 };
3057 static struct inode_operations btrfs_symlink_inode_operations = {
3058         .readlink       = generic_readlink,
3059         .follow_link    = page_follow_link_light,
3060         .put_link       = page_put_link,
3061         .permission     = btrfs_permission,
3062 };