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                                      u64 objectid,
1504                                      struct btrfs_block_group_cache *group,
1505                                      int mode)
1506 {
1507         struct inode *inode;
1508         struct btrfs_inode_item *inode_item;
1509         struct btrfs_key *location;
1510         struct btrfs_path *path;
1511         int ret;
1512         int owner;
1513
1514         path = btrfs_alloc_path();
1515         BUG_ON(!path);
1516
1517         inode = new_inode(root->fs_info->sb);
1518         if (!inode)
1519                 return ERR_PTR(-ENOMEM);
1520
1521         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1522         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1523                              inode->i_mapping, GFP_NOFS);
1524         BTRFS_I(inode)->root = root;
1525
1526         if (mode & S_IFDIR)
1527                 owner = 0;
1528         else
1529                 owner = 1;
1530         group = btrfs_find_block_group(root, group, 0, 0, owner);
1531         BTRFS_I(inode)->block_group = group;
1532         BTRFS_I(inode)->flags = 0;
1533         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1534         if (ret)
1535                 goto fail;
1536
1537         inode->i_uid = current->fsuid;
1538         inode->i_gid = current->fsgid;
1539         inode->i_mode = mode;
1540         inode->i_ino = objectid;
1541         inode->i_blocks = 0;
1542         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1543         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1544                                   struct btrfs_inode_item);
1545         fill_inode_item(path->nodes[0], inode_item, inode);
1546         btrfs_mark_buffer_dirty(path->nodes[0]);
1547         btrfs_free_path(path);
1548
1549         location = &BTRFS_I(inode)->location;
1550         location->objectid = objectid;
1551         location->offset = 0;
1552         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1553
1554         insert_inode_hash(inode);
1555         return inode;
1556 fail:
1557         btrfs_free_path(path);
1558         return ERR_PTR(ret);
1559 }
1560
1561 static inline u8 btrfs_inode_type(struct inode *inode)
1562 {
1563         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1564 }
1565
1566 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1567                             struct dentry *dentry, struct inode *inode)
1568 {
1569         int ret;
1570         struct btrfs_key key;
1571         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1572         struct inode *parent_inode;
1573
1574         key.objectid = inode->i_ino;
1575         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1576         key.offset = 0;
1577
1578         ret = btrfs_insert_dir_item(trans, root,
1579                                     dentry->d_name.name, dentry->d_name.len,
1580                                     dentry->d_parent->d_inode->i_ino,
1581                                     &key, btrfs_inode_type(inode));
1582         if (ret == 0) {
1583                 ret = btrfs_insert_inode_ref(trans, root,
1584                                      dentry->d_name.name,
1585                                      dentry->d_name.len,
1586                                      inode->i_ino,
1587                                      dentry->d_parent->d_inode->i_ino);
1588                 parent_inode = dentry->d_parent->d_inode;
1589                 parent_inode->i_size += dentry->d_name.len * 2;
1590                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1591                 ret = btrfs_update_inode(trans, root,
1592                                          dentry->d_parent->d_inode);
1593         }
1594         return ret;
1595 }
1596
1597 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1598                             struct dentry *dentry, struct inode *inode)
1599 {
1600         int err = btrfs_add_link(trans, dentry, inode);
1601         if (!err) {
1602                 d_instantiate(dentry, inode);
1603                 return 0;
1604         }
1605         if (err > 0)
1606                 err = -EEXIST;
1607         return err;
1608 }
1609
1610 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1611                         int mode, dev_t rdev)
1612 {
1613         struct btrfs_trans_handle *trans;
1614         struct btrfs_root *root = BTRFS_I(dir)->root;
1615         struct inode *inode = NULL;
1616         int err;
1617         int drop_inode = 0;
1618         u64 objectid;
1619         unsigned long nr = 0;
1620
1621         if (!new_valid_dev(rdev))
1622                 return -EINVAL;
1623
1624         mutex_lock(&root->fs_info->fs_mutex);
1625         err = btrfs_check_free_space(root, 1, 0);
1626         if (err)
1627                 goto fail;
1628
1629         trans = btrfs_start_transaction(root, 1);
1630         btrfs_set_trans_block_group(trans, dir);
1631
1632         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1633         if (err) {
1634                 err = -ENOSPC;
1635                 goto out_unlock;
1636         }
1637
1638         inode = btrfs_new_inode(trans, root, objectid,
1639                                 BTRFS_I(dir)->block_group, mode);
1640         err = PTR_ERR(inode);
1641         if (IS_ERR(inode))
1642                 goto out_unlock;
1643
1644         btrfs_set_trans_block_group(trans, inode);
1645         err = btrfs_add_nondir(trans, dentry, inode);
1646         if (err)
1647                 drop_inode = 1;
1648         else {
1649                 inode->i_op = &btrfs_special_inode_operations;
1650                 init_special_inode(inode, inode->i_mode, rdev);
1651                 btrfs_update_inode(trans, root, inode);
1652         }
1653         dir->i_sb->s_dirt = 1;
1654         btrfs_update_inode_block_group(trans, inode);
1655         btrfs_update_inode_block_group(trans, dir);
1656 out_unlock:
1657         nr = trans->blocks_used;
1658         btrfs_end_transaction(trans, root);
1659 fail:
1660         mutex_unlock(&root->fs_info->fs_mutex);
1661
1662         if (drop_inode) {
1663                 inode_dec_link_count(inode);
1664                 iput(inode);
1665         }
1666         btrfs_btree_balance_dirty(root, nr);
1667         btrfs_throttle(root);
1668         return err;
1669 }
1670
1671 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1672                         int mode, struct nameidata *nd)
1673 {
1674         struct btrfs_trans_handle *trans;
1675         struct btrfs_root *root = BTRFS_I(dir)->root;
1676         struct inode *inode = NULL;
1677         int err;
1678         int drop_inode = 0;
1679         unsigned long nr = 0;
1680         u64 objectid;
1681
1682         mutex_lock(&root->fs_info->fs_mutex);
1683         err = btrfs_check_free_space(root, 1, 0);
1684         if (err)
1685                 goto fail;
1686         trans = btrfs_start_transaction(root, 1);
1687         btrfs_set_trans_block_group(trans, dir);
1688
1689         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1690         if (err) {
1691                 err = -ENOSPC;
1692                 goto out_unlock;
1693         }
1694
1695         inode = btrfs_new_inode(trans, root, objectid,
1696                                 BTRFS_I(dir)->block_group, mode);
1697         err = PTR_ERR(inode);
1698         if (IS_ERR(inode))
1699                 goto out_unlock;
1700
1701         btrfs_set_trans_block_group(trans, inode);
1702         err = btrfs_add_nondir(trans, dentry, inode);
1703         if (err)
1704                 drop_inode = 1;
1705         else {
1706                 inode->i_mapping->a_ops = &btrfs_aops;
1707                 inode->i_fop = &btrfs_file_operations;
1708                 inode->i_op = &btrfs_file_inode_operations;
1709                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1710                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1711                                      inode->i_mapping, GFP_NOFS);
1712                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1713         }
1714         dir->i_sb->s_dirt = 1;
1715         btrfs_update_inode_block_group(trans, inode);
1716         btrfs_update_inode_block_group(trans, dir);
1717 out_unlock:
1718         nr = trans->blocks_used;
1719         btrfs_end_transaction(trans, root);
1720 fail:
1721         mutex_unlock(&root->fs_info->fs_mutex);
1722
1723         if (drop_inode) {
1724                 inode_dec_link_count(inode);
1725                 iput(inode);
1726         }
1727         btrfs_btree_balance_dirty(root, nr);
1728         btrfs_throttle(root);
1729         return err;
1730 }
1731
1732 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1733                       struct dentry *dentry)
1734 {
1735         struct btrfs_trans_handle *trans;
1736         struct btrfs_root *root = BTRFS_I(dir)->root;
1737         struct inode *inode = old_dentry->d_inode;
1738         unsigned long nr = 0;
1739         int err;
1740         int drop_inode = 0;
1741
1742         if (inode->i_nlink == 0)
1743                 return -ENOENT;
1744
1745 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1746         inode->i_nlink++;
1747 #else
1748         inc_nlink(inode);
1749 #endif
1750         mutex_lock(&root->fs_info->fs_mutex);
1751         err = btrfs_check_free_space(root, 1, 0);
1752         if (err)
1753                 goto fail;
1754         trans = btrfs_start_transaction(root, 1);
1755
1756         btrfs_set_trans_block_group(trans, dir);
1757         atomic_inc(&inode->i_count);
1758         err = btrfs_add_nondir(trans, dentry, inode);
1759
1760         if (err)
1761                 drop_inode = 1;
1762
1763         dir->i_sb->s_dirt = 1;
1764         btrfs_update_inode_block_group(trans, dir);
1765         err = btrfs_update_inode(trans, root, inode);
1766
1767         if (err)
1768                 drop_inode = 1;
1769
1770         nr = trans->blocks_used;
1771         btrfs_end_transaction(trans, root);
1772 fail:
1773         mutex_unlock(&root->fs_info->fs_mutex);
1774
1775         if (drop_inode) {
1776                 inode_dec_link_count(inode);
1777                 iput(inode);
1778         }
1779         btrfs_btree_balance_dirty(root, nr);
1780         btrfs_throttle(root);
1781         return err;
1782 }
1783
1784 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1785 {
1786         struct inode *inode;
1787         struct btrfs_trans_handle *trans;
1788         struct btrfs_root *root = BTRFS_I(dir)->root;
1789         int err = 0;
1790         int drop_on_err = 0;
1791         u64 objectid;
1792         unsigned long nr = 1;
1793
1794         mutex_lock(&root->fs_info->fs_mutex);
1795         err = btrfs_check_free_space(root, 1, 0);
1796         if (err)
1797                 goto out_unlock;
1798
1799         trans = btrfs_start_transaction(root, 1);
1800         btrfs_set_trans_block_group(trans, dir);
1801
1802         if (IS_ERR(trans)) {
1803                 err = PTR_ERR(trans);
1804                 goto out_unlock;
1805         }
1806
1807         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1808         if (err) {
1809                 err = -ENOSPC;
1810                 goto out_unlock;
1811         }
1812
1813         inode = btrfs_new_inode(trans, root, objectid,
1814                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1815         if (IS_ERR(inode)) {
1816                 err = PTR_ERR(inode);
1817                 goto out_fail;
1818         }
1819
1820         drop_on_err = 1;
1821         inode->i_op = &btrfs_dir_inode_operations;
1822         inode->i_fop = &btrfs_dir_file_operations;
1823         btrfs_set_trans_block_group(trans, inode);
1824
1825         inode->i_size = 0;
1826         err = btrfs_update_inode(trans, root, inode);
1827         if (err)
1828                 goto out_fail;
1829
1830         err = btrfs_add_link(trans, dentry, inode);
1831         if (err)
1832                 goto out_fail;
1833
1834         d_instantiate(dentry, inode);
1835         drop_on_err = 0;
1836         dir->i_sb->s_dirt = 1;
1837         btrfs_update_inode_block_group(trans, inode);
1838         btrfs_update_inode_block_group(trans, dir);
1839
1840 out_fail:
1841         nr = trans->blocks_used;
1842         btrfs_end_transaction(trans, root);
1843
1844 out_unlock:
1845         mutex_unlock(&root->fs_info->fs_mutex);
1846         if (drop_on_err)
1847                 iput(inode);
1848         btrfs_btree_balance_dirty(root, nr);
1849         btrfs_throttle(root);
1850         return err;
1851 }
1852
1853 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1854                                     size_t pg_offset, u64 start, u64 len,
1855                                     int create)
1856 {
1857         int ret;
1858         int err = 0;
1859         u64 bytenr;
1860         u64 extent_start = 0;
1861         u64 extent_end = 0;
1862         u64 objectid = inode->i_ino;
1863         u32 found_type;
1864         struct btrfs_path *path;
1865         struct btrfs_root *root = BTRFS_I(inode)->root;
1866         struct btrfs_file_extent_item *item;
1867         struct extent_buffer *leaf;
1868         struct btrfs_key found_key;
1869         struct extent_map *em = NULL;
1870         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1871         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1872         struct btrfs_trans_handle *trans = NULL;
1873
1874         path = btrfs_alloc_path();
1875         BUG_ON(!path);
1876         mutex_lock(&root->fs_info->fs_mutex);
1877
1878 again:
1879         spin_lock(&em_tree->lock);
1880         em = lookup_extent_mapping(em_tree, start, len);
1881         spin_unlock(&em_tree->lock);
1882
1883         if (em) {
1884                 if (em->start > start) {
1885                         printk("get_extent lookup [%Lu %Lu] em [%Lu %Lu]\n",
1886                                start, len, em->start, em->len);
1887                         WARN_ON(1);
1888                 }
1889                 if (em->block_start == EXTENT_MAP_INLINE && page)
1890                         free_extent_map(em);
1891                 else
1892                         goto out;
1893         }
1894         em = alloc_extent_map(GFP_NOFS);
1895         if (!em) {
1896                 err = -ENOMEM;
1897                 goto out;
1898         }
1899
1900         em->start = EXTENT_MAP_HOLE;
1901         em->len = (u64)-1;
1902         em->bdev = inode->i_sb->s_bdev;
1903         ret = btrfs_lookup_file_extent(trans, root, path,
1904                                        objectid, start, trans != NULL);
1905         if (ret < 0) {
1906                 err = ret;
1907                 goto out;
1908         }
1909
1910         if (ret != 0) {
1911                 if (path->slots[0] == 0)
1912                         goto not_found;
1913                 path->slots[0]--;
1914         }
1915
1916         leaf = path->nodes[0];
1917         item = btrfs_item_ptr(leaf, path->slots[0],
1918                               struct btrfs_file_extent_item);
1919         /* are we inside the extent that was found? */
1920         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1921         found_type = btrfs_key_type(&found_key);
1922         if (found_key.objectid != objectid ||
1923             found_type != BTRFS_EXTENT_DATA_KEY) {
1924                 goto not_found;
1925         }
1926
1927         found_type = btrfs_file_extent_type(leaf, item);
1928         extent_start = found_key.offset;
1929         if (found_type == BTRFS_FILE_EXTENT_REG) {
1930                 extent_end = extent_start +
1931                        btrfs_file_extent_num_bytes(leaf, item);
1932                 err = 0;
1933                 if (start < extent_start || start >= extent_end) {
1934                         em->start = start;
1935                         if (start < extent_start) {
1936                                 if (start + len <= extent_start)
1937                                         goto not_found;
1938                                 em->len = extent_end - extent_start;
1939                         } else {
1940                                 em->len = len;
1941                         }
1942                         goto not_found_em;
1943                 }
1944                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1945                 if (bytenr == 0) {
1946                         em->start = extent_start;
1947                         em->len = extent_end - extent_start;
1948                         em->block_start = EXTENT_MAP_HOLE;
1949                         goto insert;
1950                 }
1951                 bytenr += btrfs_file_extent_offset(leaf, item);
1952                 em->block_start = bytenr;
1953                 em->start = extent_start;
1954                 em->len = extent_end - extent_start;
1955                 goto insert;
1956         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1957                 u64 page_start;
1958                 unsigned long ptr;
1959                 char *map;
1960                 size_t size;
1961                 size_t extent_offset;
1962                 size_t copy_size;
1963
1964                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1965                                                     path->slots[0]));
1966                 extent_end = (extent_start + size + root->sectorsize - 1) &
1967                         ~((u64)root->sectorsize - 1);
1968                 if (start < extent_start || start >= extent_end) {
1969                         em->start = start;
1970                         if (start < extent_start) {
1971                                 if (start + len <= extent_start)
1972                                         goto not_found;
1973                                 em->len = extent_end - extent_start;
1974                         } else {
1975                                 em->len = len;
1976                         }
1977                         goto not_found_em;
1978                 }
1979                 em->block_start = EXTENT_MAP_INLINE;
1980
1981                 if (!page) {
1982                         em->start = extent_start;
1983                         em->len = size;
1984                         goto out;
1985                 }
1986
1987                 page_start = page_offset(page) + pg_offset;
1988                 extent_offset = page_start - extent_start;
1989                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
1990                                 size - extent_offset);
1991                 em->start = extent_start + extent_offset;
1992                 em->len = (copy_size + root->sectorsize - 1) &
1993                         ~((u64)root->sectorsize - 1);
1994                 map = kmap(page);
1995                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1996                 if (create == 0 && !PageUptodate(page)) {
1997                         read_extent_buffer(leaf, map + pg_offset, ptr,
1998                                            copy_size);
1999                         flush_dcache_page(page);
2000                 } else if (create && PageUptodate(page)) {
2001                         if (!trans) {
2002                                 kunmap(page);
2003                                 free_extent_map(em);
2004                                 em = NULL;
2005                                 btrfs_release_path(root, path);
2006                                 trans = btrfs_start_transaction(root, 1);
2007                                 goto again;
2008                         }
2009                         write_extent_buffer(leaf, map + pg_offset, ptr,
2010                                             copy_size);
2011                         btrfs_mark_buffer_dirty(leaf);
2012                 }
2013                 kunmap(page);
2014                 set_extent_uptodate(io_tree, em->start,
2015                                     extent_map_end(em) - 1, GFP_NOFS);
2016                 goto insert;
2017         } else {
2018                 printk("unkknown found_type %d\n", found_type);
2019                 WARN_ON(1);
2020         }
2021 not_found:
2022         em->start = start;
2023         em->len = len;
2024 not_found_em:
2025         em->block_start = EXTENT_MAP_HOLE;
2026 insert:
2027         btrfs_release_path(root, path);
2028         if (em->start > start || extent_map_end(em) <= start) {
2029                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2030                 err = -EIO;
2031                 goto out;
2032         }
2033
2034         err = 0;
2035         spin_lock(&em_tree->lock);
2036         ret = add_extent_mapping(em_tree, em);
2037         if (ret == -EEXIST) {
2038                 free_extent_map(em);
2039                 em = lookup_extent_mapping(em_tree, start, len);
2040                 if (!em) {
2041                         err = -EIO;
2042                         printk("failing to insert %Lu %Lu\n", start, len);
2043                 }
2044         }
2045         spin_unlock(&em_tree->lock);
2046 out:
2047         btrfs_free_path(path);
2048         if (trans) {
2049                 ret = btrfs_end_transaction(trans, root);
2050                 if (!err)
2051                         err = ret;
2052         }
2053         mutex_unlock(&root->fs_info->fs_mutex);
2054         if (err) {
2055                 free_extent_map(em);
2056                 WARN_ON(1);
2057                 return ERR_PTR(err);
2058         }
2059         return em;
2060 }
2061
2062 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2063 {
2064         return extent_bmap(mapping, iblock, btrfs_get_extent);
2065 }
2066
2067 int btrfs_readpage(struct file *file, struct page *page)
2068 {
2069         struct extent_io_tree *tree;
2070         tree = &BTRFS_I(page->mapping->host)->io_tree;
2071         return extent_read_full_page(tree, page, btrfs_get_extent);
2072 }
2073
2074 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2075 {
2076         struct extent_io_tree *tree;
2077
2078
2079         if (current->flags & PF_MEMALLOC) {
2080                 redirty_page_for_writepage(wbc, page);
2081                 unlock_page(page);
2082                 return 0;
2083         }
2084         tree = &BTRFS_I(page->mapping->host)->io_tree;
2085         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2086 }
2087
2088 static int btrfs_writepages(struct address_space *mapping,
2089                             struct writeback_control *wbc)
2090 {
2091         struct extent_io_tree *tree;
2092         tree = &BTRFS_I(mapping->host)->io_tree;
2093         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2094 }
2095
2096 static int
2097 btrfs_readpages(struct file *file, struct address_space *mapping,
2098                 struct list_head *pages, unsigned nr_pages)
2099 {
2100         struct extent_io_tree *tree;
2101         tree = &BTRFS_I(mapping->host)->io_tree;
2102         return extent_readpages(tree, mapping, pages, nr_pages,
2103                                 btrfs_get_extent);
2104 }
2105
2106 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2107 {
2108         struct extent_io_tree *tree;
2109         struct extent_map_tree *map;
2110         int ret;
2111
2112         tree = &BTRFS_I(page->mapping->host)->io_tree;
2113         map = &BTRFS_I(page->mapping->host)->extent_tree;
2114         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2115         if (ret == 1) {
2116                 ClearPagePrivate(page);
2117                 set_page_private(page, 0);
2118                 page_cache_release(page);
2119         }
2120         return ret;
2121 }
2122
2123 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2124 {
2125         struct extent_io_tree *tree;
2126
2127         tree = &BTRFS_I(page->mapping->host)->io_tree;
2128         extent_invalidatepage(tree, page, offset);
2129         btrfs_releasepage(page, GFP_NOFS);
2130 }
2131
2132 /*
2133  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2134  * called from a page fault handler when a page is first dirtied. Hence we must
2135  * be careful to check for EOF conditions here. We set the page up correctly
2136  * for a written page which means we get ENOSPC checking when writing into
2137  * holes and correct delalloc and unwritten extent mapping on filesystems that
2138  * support these features.
2139  *
2140  * We are not allowed to take the i_mutex here so we have to play games to
2141  * protect against truncate races as the page could now be beyond EOF.  Because
2142  * vmtruncate() writes the inode size before removing pages, once we have the
2143  * page lock we can determine safely if the page is beyond EOF. If it is not
2144  * beyond EOF, then the page is guaranteed safe against truncation until we
2145  * unlock the page.
2146  */
2147 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2148 {
2149         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2150         struct btrfs_root *root = BTRFS_I(inode)->root;
2151         unsigned long end;
2152         loff_t size;
2153         int ret;
2154         u64 page_start;
2155
2156         mutex_lock(&root->fs_info->fs_mutex);
2157         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2158         mutex_unlock(&root->fs_info->fs_mutex);
2159         if (ret)
2160                 goto out;
2161
2162         ret = -EINVAL;
2163
2164         lock_page(page);
2165         wait_on_page_writeback(page);
2166         size = i_size_read(inode);
2167         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2168
2169         if ((page->mapping != inode->i_mapping) ||
2170             (page_start > size)) {
2171                 /* page got truncated out from underneath us */
2172                 goto out_unlock;
2173         }
2174
2175         /* page is wholly or partially inside EOF */
2176         if (page_start + PAGE_CACHE_SIZE > size)
2177                 end = size & ~PAGE_CACHE_MASK;
2178         else
2179                 end = PAGE_CACHE_SIZE;
2180
2181         ret = btrfs_cow_one_page(inode, page, end);
2182
2183 out_unlock:
2184         unlock_page(page);
2185 out:
2186         return ret;
2187 }
2188
2189 static void btrfs_truncate(struct inode *inode)
2190 {
2191         struct btrfs_root *root = BTRFS_I(inode)->root;
2192         int ret;
2193         struct btrfs_trans_handle *trans;
2194         unsigned long nr;
2195
2196         if (!S_ISREG(inode->i_mode))
2197                 return;
2198         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2199                 return;
2200
2201         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2202
2203         mutex_lock(&root->fs_info->fs_mutex);
2204         trans = btrfs_start_transaction(root, 1);
2205         btrfs_set_trans_block_group(trans, inode);
2206
2207         /* FIXME, add redo link to tree so we don't leak on crash */
2208         ret = btrfs_truncate_in_trans(trans, root, inode,
2209                                       BTRFS_EXTENT_DATA_KEY);
2210         btrfs_update_inode(trans, root, inode);
2211         nr = trans->blocks_used;
2212
2213         ret = btrfs_end_transaction(trans, root);
2214         BUG_ON(ret);
2215         mutex_unlock(&root->fs_info->fs_mutex);
2216         btrfs_btree_balance_dirty(root, nr);
2217         btrfs_throttle(root);
2218 }
2219
2220 static int noinline create_subvol(struct btrfs_root *root, char *name,
2221                                   int namelen)
2222 {
2223         struct btrfs_trans_handle *trans;
2224         struct btrfs_key key;
2225         struct btrfs_root_item root_item;
2226         struct btrfs_inode_item *inode_item;
2227         struct extent_buffer *leaf;
2228         struct btrfs_root *new_root = root;
2229         struct inode *inode;
2230         struct inode *dir;
2231         int ret;
2232         int err;
2233         u64 objectid;
2234         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2235         unsigned long nr = 1;
2236
2237         mutex_lock(&root->fs_info->fs_mutex);
2238         ret = btrfs_check_free_space(root, 1, 0);
2239         if (ret)
2240                 goto fail_commit;
2241
2242         trans = btrfs_start_transaction(root, 1);
2243         BUG_ON(!trans);
2244
2245         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2246                                        0, &objectid);
2247         if (ret)
2248                 goto fail;
2249
2250         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2251                                         objectid, trans->transid, 0, 0,
2252                                         0, 0);
2253         if (IS_ERR(leaf))
2254                 return PTR_ERR(leaf);
2255
2256         btrfs_set_header_nritems(leaf, 0);
2257         btrfs_set_header_level(leaf, 0);
2258         btrfs_set_header_bytenr(leaf, leaf->start);
2259         btrfs_set_header_generation(leaf, trans->transid);
2260         btrfs_set_header_owner(leaf, objectid);
2261
2262         write_extent_buffer(leaf, root->fs_info->fsid,
2263                             (unsigned long)btrfs_header_fsid(leaf),
2264                             BTRFS_FSID_SIZE);
2265         btrfs_mark_buffer_dirty(leaf);
2266
2267         inode_item = &root_item.inode;
2268         memset(inode_item, 0, sizeof(*inode_item));
2269         inode_item->generation = cpu_to_le64(1);
2270         inode_item->size = cpu_to_le64(3);
2271         inode_item->nlink = cpu_to_le32(1);
2272         inode_item->nblocks = cpu_to_le64(1);
2273         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2274
2275         btrfs_set_root_bytenr(&root_item, leaf->start);
2276         btrfs_set_root_level(&root_item, 0);
2277         btrfs_set_root_refs(&root_item, 1);
2278         btrfs_set_root_used(&root_item, 0);
2279
2280         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2281         root_item.drop_level = 0;
2282
2283         free_extent_buffer(leaf);
2284         leaf = NULL;
2285
2286         btrfs_set_root_dirid(&root_item, new_dirid);
2287
2288         key.objectid = objectid;
2289         key.offset = 1;
2290         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2291         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2292                                 &root_item);
2293         if (ret)
2294                 goto fail;
2295
2296         /*
2297          * insert the directory item
2298          */
2299         key.offset = (u64)-1;
2300         dir = root->fs_info->sb->s_root->d_inode;
2301         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2302                                     name, namelen, dir->i_ino, &key,
2303                                     BTRFS_FT_DIR);
2304         if (ret)
2305                 goto fail;
2306
2307         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2308                              name, namelen, objectid,
2309                              root->fs_info->sb->s_root->d_inode->i_ino);
2310         if (ret)
2311                 goto fail;
2312
2313         ret = btrfs_commit_transaction(trans, root);
2314         if (ret)
2315                 goto fail_commit;
2316
2317         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2318         BUG_ON(!new_root);
2319
2320         trans = btrfs_start_transaction(new_root, 1);
2321         BUG_ON(!trans);
2322
2323         inode = btrfs_new_inode(trans, new_root, new_dirid,
2324                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2325         if (IS_ERR(inode))
2326                 goto fail;
2327         inode->i_op = &btrfs_dir_inode_operations;
2328         inode->i_fop = &btrfs_dir_file_operations;
2329         new_root->inode = inode;
2330
2331         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2332                                      new_dirid);
2333         inode->i_nlink = 1;
2334         inode->i_size = 0;
2335         ret = btrfs_update_inode(trans, new_root, inode);
2336         if (ret)
2337                 goto fail;
2338 fail:
2339         nr = trans->blocks_used;
2340         err = btrfs_commit_transaction(trans, new_root);
2341         if (err && !ret)
2342                 ret = err;
2343 fail_commit:
2344         mutex_unlock(&root->fs_info->fs_mutex);
2345         btrfs_btree_balance_dirty(root, nr);
2346         btrfs_throttle(root);
2347         return ret;
2348 }
2349
2350 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2351 {
2352         struct btrfs_pending_snapshot *pending_snapshot;
2353         struct btrfs_trans_handle *trans;
2354         int ret;
2355         int err;
2356         unsigned long nr = 0;
2357
2358         if (!root->ref_cows)
2359                 return -EINVAL;
2360
2361         mutex_lock(&root->fs_info->fs_mutex);
2362         ret = btrfs_check_free_space(root, 1, 0);
2363         if (ret)
2364                 goto fail_unlock;
2365
2366         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2367         if (!pending_snapshot) {
2368                 ret = -ENOMEM;
2369                 goto fail_unlock;
2370         }
2371         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2372         if (!pending_snapshot->name) {
2373                 ret = -ENOMEM;
2374                 kfree(pending_snapshot);
2375                 goto fail_unlock;
2376         }
2377         memcpy(pending_snapshot->name, name, namelen);
2378         pending_snapshot->name[namelen] = '\0';
2379         trans = btrfs_start_transaction(root, 1);
2380         BUG_ON(!trans);
2381         pending_snapshot->root = root;
2382         list_add(&pending_snapshot->list,
2383                  &trans->transaction->pending_snapshots);
2384         ret = btrfs_update_inode(trans, root, root->inode);
2385         err = btrfs_commit_transaction(trans, root);
2386
2387 fail_unlock:
2388         mutex_unlock(&root->fs_info->fs_mutex);
2389         btrfs_btree_balance_dirty(root, nr);
2390         btrfs_throttle(root);
2391         return ret;
2392 }
2393
2394 unsigned long btrfs_force_ra(struct address_space *mapping,
2395                               struct file_ra_state *ra, struct file *file,
2396                               pgoff_t offset, pgoff_t last_index)
2397 {
2398         pgoff_t req_size;
2399
2400 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2401         req_size = last_index - offset + 1;
2402         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2403         return offset;
2404 #else
2405         req_size = min(last_index - offset + 1, (pgoff_t)128);
2406         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2407         return offset + req_size;
2408 #endif
2409 }
2410
2411 int btrfs_defrag_file(struct file *file) {
2412         struct inode *inode = fdentry(file)->d_inode;
2413         struct btrfs_root *root = BTRFS_I(inode)->root;
2414         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2415         struct page *page;
2416         unsigned long last_index;
2417         unsigned long ra_index = 0;
2418         u64 page_start;
2419         u64 page_end;
2420         u64 delalloc_start;
2421         u64 existing_delalloc;
2422         unsigned long i;
2423         int ret;
2424
2425         mutex_lock(&root->fs_info->fs_mutex);
2426         ret = btrfs_check_free_space(root, inode->i_size, 0);
2427         mutex_unlock(&root->fs_info->fs_mutex);
2428         if (ret)
2429                 return -ENOSPC;
2430
2431         mutex_lock(&inode->i_mutex);
2432         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2433         for (i = 0; i <= last_index; i++) {
2434                 if (i == ra_index) {
2435                         ra_index = btrfs_force_ra(inode->i_mapping,
2436                                                   &file->f_ra,
2437                                                   file, ra_index, last_index);
2438                 }
2439                 page = grab_cache_page(inode->i_mapping, i);
2440                 if (!page)
2441                         goto out_unlock;
2442                 if (!PageUptodate(page)) {
2443                         btrfs_readpage(NULL, page);
2444                         lock_page(page);
2445                         if (!PageUptodate(page)) {
2446                                 unlock_page(page);
2447                                 page_cache_release(page);
2448                                 goto out_unlock;
2449                         }
2450                 }
2451                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2452                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2453
2454                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2455                 delalloc_start = page_start;
2456                 existing_delalloc =
2457                         count_range_bits(&BTRFS_I(inode)->io_tree,
2458                                          &delalloc_start, page_end,
2459                                          PAGE_CACHE_SIZE, EXTENT_DELALLOC);
2460                 set_extent_delalloc(io_tree, page_start,
2461                                     page_end, GFP_NOFS);
2462
2463                 spin_lock(&root->fs_info->delalloc_lock);
2464                 root->fs_info->delalloc_bytes += PAGE_CACHE_SIZE -
2465                                                  existing_delalloc;
2466                 spin_unlock(&root->fs_info->delalloc_lock);
2467
2468                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2469                 set_page_dirty(page);
2470                 unlock_page(page);
2471                 page_cache_release(page);
2472                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2473         }
2474
2475 out_unlock:
2476         mutex_unlock(&inode->i_mutex);
2477         return 0;
2478 }
2479
2480 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2481 {
2482         u64 new_size;
2483         u64 old_size;
2484         struct btrfs_ioctl_vol_args *vol_args;
2485         struct btrfs_trans_handle *trans;
2486         char *sizestr;
2487         int ret = 0;
2488         int namelen;
2489         int mod = 0;
2490
2491         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2492
2493         if (!vol_args)
2494                 return -ENOMEM;
2495
2496         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2497                 ret = -EFAULT;
2498                 goto out;
2499         }
2500         namelen = strlen(vol_args->name);
2501         if (namelen > BTRFS_VOL_NAME_MAX) {
2502                 ret = -EINVAL;
2503                 goto out;
2504         }
2505
2506         sizestr = vol_args->name;
2507         if (!strcmp(sizestr, "max"))
2508                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2509         else {
2510                 if (sizestr[0] == '-') {
2511                         mod = -1;
2512                         sizestr++;
2513                 } else if (sizestr[0] == '+') {
2514                         mod = 1;
2515                         sizestr++;
2516                 }
2517                 new_size = btrfs_parse_size(sizestr);
2518                 if (new_size == 0) {
2519                         ret = -EINVAL;
2520                         goto out;
2521                 }
2522         }
2523
2524         mutex_lock(&root->fs_info->fs_mutex);
2525         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2526
2527         if (mod < 0) {
2528                 if (new_size > old_size) {
2529                         ret = -EINVAL;
2530                         goto out_unlock;
2531                 }
2532                 new_size = old_size - new_size;
2533         } else if (mod > 0) {
2534                 new_size = old_size + new_size;
2535         }
2536
2537         if (new_size < 256 * 1024 * 1024) {
2538                 ret = -EINVAL;
2539                 goto out_unlock;
2540         }
2541         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2542                 ret = -EFBIG;
2543                 goto out_unlock;
2544         }
2545
2546         do_div(new_size, root->sectorsize);
2547         new_size *= root->sectorsize;
2548
2549 printk("new size is %Lu\n", new_size);
2550         if (new_size > old_size) {
2551                 trans = btrfs_start_transaction(root, 1);
2552                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2553                 btrfs_commit_transaction(trans, root);
2554         } else {
2555                 ret = btrfs_shrink_extent_tree(root, new_size);
2556         }
2557
2558 out_unlock:
2559         mutex_unlock(&root->fs_info->fs_mutex);
2560 out:
2561         kfree(vol_args);
2562         return ret;
2563 }
2564
2565 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2566                                             void __user *arg)
2567 {
2568         struct btrfs_ioctl_vol_args *vol_args;
2569         struct btrfs_dir_item *di;
2570         struct btrfs_path *path;
2571         u64 root_dirid;
2572         int namelen;
2573         int ret;
2574
2575         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2576
2577         if (!vol_args)
2578                 return -ENOMEM;
2579
2580         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2581                 ret = -EFAULT;
2582                 goto out;
2583         }
2584
2585         namelen = strlen(vol_args->name);
2586         if (namelen > BTRFS_VOL_NAME_MAX) {
2587                 ret = -EINVAL;
2588                 goto out;
2589         }
2590         if (strchr(vol_args->name, '/')) {
2591                 ret = -EINVAL;
2592                 goto out;
2593         }
2594
2595         path = btrfs_alloc_path();
2596         if (!path) {
2597                 ret = -ENOMEM;
2598                 goto out;
2599         }
2600
2601         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2602         mutex_lock(&root->fs_info->fs_mutex);
2603         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2604                             path, root_dirid,
2605                             vol_args->name, namelen, 0);
2606         mutex_unlock(&root->fs_info->fs_mutex);
2607         btrfs_free_path(path);
2608
2609         if (di && !IS_ERR(di)) {
2610                 ret = -EEXIST;
2611                 goto out;
2612         }
2613
2614         if (IS_ERR(di)) {
2615                 ret = PTR_ERR(di);
2616                 goto out;
2617         }
2618
2619         if (root == root->fs_info->tree_root)
2620                 ret = create_subvol(root, vol_args->name, namelen);
2621         else
2622                 ret = create_snapshot(root, vol_args->name, namelen);
2623 out:
2624         kfree(vol_args);
2625         return ret;
2626 }
2627
2628 static int btrfs_ioctl_defrag(struct file *file)
2629 {
2630         struct inode *inode = fdentry(file)->d_inode;
2631         struct btrfs_root *root = BTRFS_I(inode)->root;
2632
2633         switch (inode->i_mode & S_IFMT) {
2634         case S_IFDIR:
2635                 mutex_lock(&root->fs_info->fs_mutex);
2636                 btrfs_defrag_root(root, 0);
2637                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2638                 mutex_unlock(&root->fs_info->fs_mutex);
2639                 break;
2640         case S_IFREG:
2641                 btrfs_defrag_file(file);
2642                 break;
2643         }
2644
2645         return 0;
2646 }
2647
2648 long btrfs_ioctl(struct file *file, unsigned int
2649                 cmd, unsigned long arg)
2650 {
2651         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2652
2653         switch (cmd) {
2654         case BTRFS_IOC_SNAP_CREATE:
2655                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2656         case BTRFS_IOC_DEFRAG:
2657                 return btrfs_ioctl_defrag(file);
2658         case BTRFS_IOC_RESIZE:
2659                 return btrfs_ioctl_resize(root, (void __user *)arg);
2660         }
2661
2662         return -ENOTTY;
2663 }
2664
2665 /*
2666  * Called inside transaction, so use GFP_NOFS
2667  */
2668 struct inode *btrfs_alloc_inode(struct super_block *sb)
2669 {
2670         struct btrfs_inode *ei;
2671
2672         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2673         if (!ei)
2674                 return NULL;
2675         ei->last_trans = 0;
2676         ei->ordered_trans = 0;
2677         return &ei->vfs_inode;
2678 }
2679
2680 void btrfs_destroy_inode(struct inode *inode)
2681 {
2682         WARN_ON(!list_empty(&inode->i_dentry));
2683         WARN_ON(inode->i_data.nrpages);
2684
2685         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2686         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2687 }
2688
2689 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2690 static void init_once(struct kmem_cache * cachep, void *foo)
2691 #else
2692 static void init_once(void * foo, struct kmem_cache * cachep,
2693                       unsigned long flags)
2694 #endif
2695 {
2696         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2697
2698         inode_init_once(&ei->vfs_inode);
2699 }
2700
2701 void btrfs_destroy_cachep(void)
2702 {
2703         if (btrfs_inode_cachep)
2704                 kmem_cache_destroy(btrfs_inode_cachep);
2705         if (btrfs_trans_handle_cachep)
2706                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2707         if (btrfs_transaction_cachep)
2708                 kmem_cache_destroy(btrfs_transaction_cachep);
2709         if (btrfs_bit_radix_cachep)
2710                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2711         if (btrfs_path_cachep)
2712                 kmem_cache_destroy(btrfs_path_cachep);
2713 }
2714
2715 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2716                                        unsigned long extra_flags,
2717 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2718                                        void (*ctor)(struct kmem_cache *, void *)
2719 #else
2720                                        void (*ctor)(void *, struct kmem_cache *,
2721                                                     unsigned long)
2722 #endif
2723                                      )
2724 {
2725         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2726                                  SLAB_MEM_SPREAD | extra_flags), ctor
2727 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2728                                  ,NULL
2729 #endif
2730                                 );
2731 }
2732
2733 int btrfs_init_cachep(void)
2734 {
2735         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2736                                           sizeof(struct btrfs_inode),
2737                                           0, init_once);
2738         if (!btrfs_inode_cachep)
2739                 goto fail;
2740         btrfs_trans_handle_cachep =
2741                         btrfs_cache_create("btrfs_trans_handle_cache",
2742                                            sizeof(struct btrfs_trans_handle),
2743                                            0, NULL);
2744         if (!btrfs_trans_handle_cachep)
2745                 goto fail;
2746         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2747                                              sizeof(struct btrfs_transaction),
2748                                              0, NULL);
2749         if (!btrfs_transaction_cachep)
2750                 goto fail;
2751         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2752                                          sizeof(struct btrfs_path),
2753                                          0, NULL);
2754         if (!btrfs_path_cachep)
2755                 goto fail;
2756         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2757                                               SLAB_DESTROY_BY_RCU, NULL);
2758         if (!btrfs_bit_radix_cachep)
2759                 goto fail;
2760         return 0;
2761 fail:
2762         btrfs_destroy_cachep();
2763         return -ENOMEM;
2764 }
2765
2766 static int btrfs_getattr(struct vfsmount *mnt,
2767                          struct dentry *dentry, struct kstat *stat)
2768 {
2769         struct inode *inode = dentry->d_inode;
2770         generic_fillattr(inode, stat);
2771         stat->blksize = PAGE_CACHE_SIZE;
2772         return 0;
2773 }
2774
2775 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2776                            struct inode * new_dir,struct dentry *new_dentry)
2777 {
2778         struct btrfs_trans_handle *trans;
2779         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2780         struct inode *new_inode = new_dentry->d_inode;
2781         struct inode *old_inode = old_dentry->d_inode;
2782         struct timespec ctime = CURRENT_TIME;
2783         struct btrfs_path *path;
2784         int ret;
2785
2786         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2787             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2788                 return -ENOTEMPTY;
2789         }
2790
2791         mutex_lock(&root->fs_info->fs_mutex);
2792         ret = btrfs_check_free_space(root, 1, 0);
2793         if (ret)
2794                 goto out_unlock;
2795
2796         trans = btrfs_start_transaction(root, 1);
2797
2798         btrfs_set_trans_block_group(trans, new_dir);
2799         path = btrfs_alloc_path();
2800         if (!path) {
2801                 ret = -ENOMEM;
2802                 goto out_fail;
2803         }
2804
2805         old_dentry->d_inode->i_nlink++;
2806         old_dir->i_ctime = old_dir->i_mtime = ctime;
2807         new_dir->i_ctime = new_dir->i_mtime = ctime;
2808         old_inode->i_ctime = ctime;
2809
2810         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2811         if (ret)
2812                 goto out_fail;
2813
2814         if (new_inode) {
2815                 new_inode->i_ctime = CURRENT_TIME;
2816                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2817                 if (ret)
2818                         goto out_fail;
2819         }
2820         ret = btrfs_add_link(trans, new_dentry, old_inode);
2821         if (ret)
2822                 goto out_fail;
2823
2824 out_fail:
2825         btrfs_free_path(path);
2826         btrfs_end_transaction(trans, root);
2827 out_unlock:
2828         mutex_unlock(&root->fs_info->fs_mutex);
2829         return ret;
2830 }
2831
2832 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2833                          const char *symname)
2834 {
2835         struct btrfs_trans_handle *trans;
2836         struct btrfs_root *root = BTRFS_I(dir)->root;
2837         struct btrfs_path *path;
2838         struct btrfs_key key;
2839         struct inode *inode = NULL;
2840         int err;
2841         int drop_inode = 0;
2842         u64 objectid;
2843         int name_len;
2844         int datasize;
2845         unsigned long ptr;
2846         struct btrfs_file_extent_item *ei;
2847         struct extent_buffer *leaf;
2848         unsigned long nr = 0;
2849
2850         name_len = strlen(symname) + 1;
2851         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2852                 return -ENAMETOOLONG;
2853
2854         mutex_lock(&root->fs_info->fs_mutex);
2855         err = btrfs_check_free_space(root, 1, 0);
2856         if (err)
2857                 goto out_fail;
2858
2859         trans = btrfs_start_transaction(root, 1);
2860         btrfs_set_trans_block_group(trans, dir);
2861
2862         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2863         if (err) {
2864                 err = -ENOSPC;
2865                 goto out_unlock;
2866         }
2867
2868         inode = btrfs_new_inode(trans, root, objectid,
2869                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2870         err = PTR_ERR(inode);
2871         if (IS_ERR(inode))
2872                 goto out_unlock;
2873
2874         btrfs_set_trans_block_group(trans, inode);
2875         err = btrfs_add_nondir(trans, dentry, inode);
2876         if (err)
2877                 drop_inode = 1;
2878         else {
2879                 inode->i_mapping->a_ops = &btrfs_aops;
2880                 inode->i_fop = &btrfs_file_operations;
2881                 inode->i_op = &btrfs_file_inode_operations;
2882                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2883                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2884                                      inode->i_mapping, GFP_NOFS);
2885                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2886         }
2887         dir->i_sb->s_dirt = 1;
2888         btrfs_update_inode_block_group(trans, inode);
2889         btrfs_update_inode_block_group(trans, dir);
2890         if (drop_inode)
2891                 goto out_unlock;
2892
2893         path = btrfs_alloc_path();
2894         BUG_ON(!path);
2895         key.objectid = inode->i_ino;
2896         key.offset = 0;
2897         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2898         datasize = btrfs_file_extent_calc_inline_size(name_len);
2899         err = btrfs_insert_empty_item(trans, root, path, &key,
2900                                       datasize);
2901         if (err) {
2902                 drop_inode = 1;
2903                 goto out_unlock;
2904         }
2905         leaf = path->nodes[0];
2906         ei = btrfs_item_ptr(leaf, path->slots[0],
2907                             struct btrfs_file_extent_item);
2908         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2909         btrfs_set_file_extent_type(leaf, ei,
2910                                    BTRFS_FILE_EXTENT_INLINE);
2911         ptr = btrfs_file_extent_inline_start(ei);
2912         write_extent_buffer(leaf, symname, ptr, name_len);
2913         btrfs_mark_buffer_dirty(leaf);
2914         btrfs_free_path(path);
2915
2916         inode->i_op = &btrfs_symlink_inode_operations;
2917         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2918         inode->i_size = name_len - 1;
2919         err = btrfs_update_inode(trans, root, inode);
2920         if (err)
2921                 drop_inode = 1;
2922
2923 out_unlock:
2924         nr = trans->blocks_used;
2925         btrfs_end_transaction(trans, root);
2926 out_fail:
2927         mutex_unlock(&root->fs_info->fs_mutex);
2928         if (drop_inode) {
2929                 inode_dec_link_count(inode);
2930                 iput(inode);
2931         }
2932         btrfs_btree_balance_dirty(root, nr);
2933         btrfs_throttle(root);
2934         return err;
2935 }
2936 static int btrfs_permission(struct inode *inode, int mask,
2937                             struct nameidata *nd)
2938 {
2939         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2940                 return -EACCES;
2941         return generic_permission(inode, mask, NULL);
2942 }
2943
2944 static struct inode_operations btrfs_dir_inode_operations = {
2945         .lookup         = btrfs_lookup,
2946         .create         = btrfs_create,
2947         .unlink         = btrfs_unlink,
2948         .link           = btrfs_link,
2949         .mkdir          = btrfs_mkdir,
2950         .rmdir          = btrfs_rmdir,
2951         .rename         = btrfs_rename,
2952         .symlink        = btrfs_symlink,
2953         .setattr        = btrfs_setattr,
2954         .mknod          = btrfs_mknod,
2955         .setxattr       = generic_setxattr,
2956         .getxattr       = generic_getxattr,
2957         .listxattr      = btrfs_listxattr,
2958         .removexattr    = generic_removexattr,
2959         .permission     = btrfs_permission,
2960 };
2961 static struct inode_operations btrfs_dir_ro_inode_operations = {
2962         .lookup         = btrfs_lookup,
2963         .permission     = btrfs_permission,
2964 };
2965 static struct file_operations btrfs_dir_file_operations = {
2966         .llseek         = generic_file_llseek,
2967         .read           = generic_read_dir,
2968         .readdir        = btrfs_readdir,
2969         .unlocked_ioctl = btrfs_ioctl,
2970 #ifdef CONFIG_COMPAT
2971         .compat_ioctl   = btrfs_ioctl,
2972 #endif
2973 };
2974
2975 static struct extent_io_ops btrfs_extent_io_ops = {
2976         .fill_delalloc = run_delalloc_range,
2977         .writepage_io_hook = btrfs_writepage_io_hook,
2978         .readpage_io_hook = btrfs_readpage_io_hook,
2979         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2980 };
2981
2982 static struct address_space_operations btrfs_aops = {
2983         .readpage       = btrfs_readpage,
2984         .writepage      = btrfs_writepage,
2985         .writepages     = btrfs_writepages,
2986         .readpages      = btrfs_readpages,
2987         .sync_page      = block_sync_page,
2988         .bmap           = btrfs_bmap,
2989         .invalidatepage = btrfs_invalidatepage,
2990         .releasepage    = btrfs_releasepage,
2991         .set_page_dirty = __set_page_dirty_nobuffers,
2992 };
2993
2994 static struct address_space_operations btrfs_symlink_aops = {
2995         .readpage       = btrfs_readpage,
2996         .writepage      = btrfs_writepage,
2997         .invalidatepage = btrfs_invalidatepage,
2998         .releasepage    = btrfs_releasepage,
2999 };
3000
3001 static struct inode_operations btrfs_file_inode_operations = {
3002         .truncate       = btrfs_truncate,
3003         .getattr        = btrfs_getattr,
3004         .setattr        = btrfs_setattr,
3005         .setxattr       = generic_setxattr,
3006         .getxattr       = generic_getxattr,
3007         .listxattr      = btrfs_listxattr,
3008         .removexattr    = generic_removexattr,
3009         .permission     = btrfs_permission,
3010 };
3011 static struct inode_operations btrfs_special_inode_operations = {
3012         .getattr        = btrfs_getattr,
3013         .setattr        = btrfs_setattr,
3014         .permission     = btrfs_permission,
3015 };
3016 static struct inode_operations btrfs_symlink_inode_operations = {
3017         .readlink       = generic_readlink,
3018         .follow_link    = page_follow_link_light,
3019         .put_link       = page_put_link,
3020         .permission     = btrfs_permission,
3021 };