fs/btrfs/inode.c

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