SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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                 drop_nlink(dentry->d_inode);
 536                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 537                 dir->i_sb->s_dirt = 1;
 538         }
 539         return ret;
 540 }
 541
 542 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 543 {
 544         struct btrfs_root *root;
 545         struct btrfs_trans_handle *trans;
 546         int ret;
 547         unsigned long nr;
 548
 549         root = BTRFS_I(dir)->root;
 550         mutex_lock(&root->fs_info->fs_mutex);
 551         trans = btrfs_start_transaction(root, 1);
 552
 553         btrfs_set_trans_block_group(trans, dir);
 554         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 555         nr = trans->blocks_used;
 556
 557         btrfs_end_transaction(trans, root);
 558         mutex_unlock(&root->fs_info->fs_mutex);
 559         btrfs_btree_balance_dirty(root, nr);
 560
 561         return ret;
 562 }
 563
 564 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 565 {
 566         struct inode *inode = dentry->d_inode;
 567         int err;
 568         int ret;
 569         struct btrfs_root *root = BTRFS_I(dir)->root;
 570         struct btrfs_trans_handle *trans;
 571         unsigned long nr;
 572
 573         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 574                 return -ENOTEMPTY;
 575
 576         mutex_lock(&root->fs_info->fs_mutex);
 577         trans = btrfs_start_transaction(root, 1);
 578         btrfs_set_trans_block_group(trans, dir);
 579
 580         /* now the directory is empty */
 581         err = btrfs_unlink_trans(trans, root, dir, dentry);
 582         if (!err) {
 583                 inode->i_size = 0;
 584         }
 585
 586         nr = trans->blocks_used;
 587         ret = btrfs_end_transaction(trans, root);
 588         mutex_unlock(&root->fs_info->fs_mutex);
 589         btrfs_btree_balance_dirty(root, nr);
 590
 591         if (ret && !err)
 592                 err = ret;
 593         return err;
 594 }
 595
 596 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 597                             struct btrfs_root *root,
 598                             struct inode *inode)
 599 {
 600         struct btrfs_path *path;
 601         int ret;
 602
 603         clear_inode(inode);
 604
 605         path = btrfs_alloc_path();
 606         BUG_ON(!path);
 607         ret = btrfs_lookup_inode(trans, root, path,
 608                                  &BTRFS_I(inode)->location, -1);
 609         if (ret > 0)
 610                 ret = -ENOENT;
 611         if (!ret)
 612                 ret = btrfs_del_item(trans, root, path);
 613         btrfs_free_path(path);
 614         return ret;
 615 }
 616
 617 /*
 618  * this can truncate away extent items, csum items and directory items.
 619  * It starts at a high offset and removes keys until it can't find
 620  * any higher than i_size.
 621  *
 622  * csum items that cross the new i_size are truncated to the new size
 623  * as well.
 624  */
 625 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 626                                    struct btrfs_root *root,
 627                                    struct inode *inode)
 628 {
 629         int ret;
 630         struct btrfs_path *path;
 631         struct btrfs_key key;
 632         struct btrfs_key found_key;
 633         u32 found_type;
 634         struct extent_buffer *leaf;
 635         struct btrfs_file_extent_item *fi;
 636         u64 extent_start = 0;
 637         u64 extent_num_bytes = 0;
 638         u64 item_end = 0;
 639         u64 root_gen = 0;
 640         u64 root_owner = 0;
 641         int found_extent;
 642         int del_item;
 643         int extent_type = -1;
 644
 645         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 646         path = btrfs_alloc_path();
 647         path->reada = -1;
 648         BUG_ON(!path);
 649
 650         /* FIXME, add redo link to tree so we don't leak on crash */
 651         key.objectid = inode->i_ino;
 652         key.offset = (u64)-1;
 653         key.type = (u8)-1;
 654
 655         while(1) {
 656                 btrfs_init_path(path);
 657                 fi = NULL;
 658                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 659                 if (ret < 0) {
 660                         goto error;
 661                 }
 662                 if (ret > 0) {
 663                         BUG_ON(path->slots[0] == 0);
 664                         path->slots[0]--;
 665                 }
 666                 leaf = path->nodes[0];
 667                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 668                 found_type = btrfs_key_type(&found_key);
 669
 670                 if (found_key.objectid != inode->i_ino)
 671                         break;
 672
 673                 if (found_type != BTRFS_CSUM_ITEM_KEY &&
 674                     found_type != BTRFS_DIR_ITEM_KEY &&
 675                     found_type != BTRFS_DIR_INDEX_KEY &&
 676                     found_type != BTRFS_EXTENT_DATA_KEY)
 677                         break;
 678
 679                 item_end = found_key.offset;
 680                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 681                         fi = btrfs_item_ptr(leaf, path->slots[0],
 682                                             struct btrfs_file_extent_item);
 683                         extent_type = btrfs_file_extent_type(leaf, fi);
 684                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 685                                 item_end +=
 686                                     btrfs_file_extent_num_bytes(leaf, fi);
 687                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 688                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 689                                                                 path->slots[0]);
 690                                 item_end += btrfs_file_extent_inline_len(leaf,
 691                                                                          item);
 692                         }
 693                         item_end--;
 694                 }
 695                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 696                         ret = btrfs_csum_truncate(trans, root, path,
 697                                                   inode->i_size);
 698                         BUG_ON(ret);
 699                 }
 700                 if (item_end < inode->i_size) {
 701                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 702                                 found_type = BTRFS_INODE_ITEM_KEY;
 703                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 704                                 found_type = BTRFS_CSUM_ITEM_KEY;
 705                         } else if (found_type) {
 706                                 found_type--;
 707                         } else {
 708                                 break;
 709                         }
 710                         btrfs_set_key_type(&key, found_type);
 711                         btrfs_release_path(root, path);
 712                         continue;
 713                 }
 714                 if (found_key.offset >= inode->i_size)
 715                         del_item = 1;
 716                 else
 717                         del_item = 0;
 718                 found_extent = 0;
 719
 720                 /* FIXME, shrink the extent if the ref count is only 1 */
 721                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 722                         goto delete;
 723
 724                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 725                         u64 num_dec;
 726                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 727                         if (!del_item) {
 728                                 u64 orig_num_bytes =
 729                                         btrfs_file_extent_num_bytes(leaf, fi);
 730                                 extent_num_bytes = inode->i_size -
 731                                         found_key.offset + root->sectorsize - 1;
 732                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 733                                                          extent_num_bytes);
 734                                 num_dec = (orig_num_bytes -
 735                                            extent_num_bytes) >> 9;
 736                                 if (extent_start != 0) {
 737                                         inode->i_blocks -= num_dec;
 738                                 }
 739                                 btrfs_mark_buffer_dirty(leaf);
 740                         } else {
 741                                 extent_num_bytes =
 742                                         btrfs_file_extent_disk_num_bytes(leaf,
 743                                                                          fi);
 744                                 /* FIXME blocksize != 4096 */
 745                                 num_dec = btrfs_file_extent_num_bytes(leaf,
 746                                                                        fi) >> 9;
 747                                 if (extent_start != 0) {
 748                                         found_extent = 1;
 749                                         inode->i_blocks -= num_dec;
 750                                 }
 751                                 root_gen = btrfs_header_generation(leaf);
 752                                 root_owner = btrfs_header_owner(leaf);
 753                         }
 754                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE &&
 755                            !del_item) {
 756                         u32 newsize = inode->i_size - found_key.offset;
 757                         newsize = btrfs_file_extent_calc_inline_size(newsize);
 758                         ret = btrfs_truncate_item(trans, root, path,
 759                                                   newsize, 1);
 760                         BUG_ON(ret);
 761                 }
 762 delete:
 763                 if (del_item) {
 764                         ret = btrfs_del_item(trans, root, path);
 765                         if (ret)
 766                                 goto error;
 767                 } else {
 768                         break;
 769                 }
 770                 btrfs_release_path(root, path);
 771                 if (found_extent) {
 772                         ret = btrfs_free_extent(trans, root, extent_start,
 773                                                 extent_num_bytes,
 774                                                 root_owner,
 775                                                 root_gen, inode->i_ino,
 776                                                 found_key.offset, 0);
 777                         BUG_ON(ret);
 778                 }
 779         }
 780         ret = 0;
 781 error:
 782         btrfs_release_path(root, path);
 783         btrfs_free_path(path);
 784         inode->i_sb->s_dirt = 1;
 785         return ret;
 786 }
 787
 788 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 789                               size_t zero_start)
 790 {
 791         char *kaddr;
 792         int ret = 0;
 793         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 794         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 795         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 796
 797         set_page_extent_mapped(page);
 798
 799         lock_extent(em_tree, page_start, page_end, GFP_NOFS);
 800         set_extent_delalloc(&BTRFS_I(inode)->extent_tree, page_start,
 801                             page_end, GFP_NOFS);
 802         if (zero_start != PAGE_CACHE_SIZE) {
 803                 kaddr = kmap(page);
 804                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 805                 flush_dcache_page(page);
 806                 kunmap(page);
 807         }
 808         set_page_dirty(page);
 809         unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
 810
 811         return ret;
 812 }
 813
 814 /*
 815  * taken from block_truncate_page, but does cow as it zeros out
 816  * any bytes left in the last page in the file.
 817  */
 818 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 819 {
 820         struct inode *inode = mapping->host;
 821         struct btrfs_root *root = BTRFS_I(inode)->root;
 822         u32 blocksize = root->sectorsize;
 823         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 824         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 825         struct page *page;
 826         int ret = 0;
 827         u64 page_start;
 828
 829         if ((offset & (blocksize - 1)) == 0)
 830                 goto out;
 831
 832         down_read(&root->snap_sem);
 833         ret = -ENOMEM;
 834         page = grab_cache_page(mapping, index);
 835         if (!page)
 836                 goto out;
 837         if (!PageUptodate(page)) {
 838                 ret = btrfs_readpage(NULL, page);
 839                 lock_page(page);
 840                 if (!PageUptodate(page)) {
 841                         ret = -EIO;
 842                         goto out;
 843                 }
 844         }
 845         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 846
 847         ret = btrfs_cow_one_page(inode, page, offset);
 848
 849         unlock_page(page);
 850         page_cache_release(page);
 851         up_read(&BTRFS_I(inode)->root->snap_sem);
 852 out:
 853         return ret;
 854 }
 855
 856 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 857 {
 858         struct inode *inode = dentry->d_inode;
 859         int err;
 860
 861         err = inode_change_ok(inode, attr);
 862         if (err)
 863                 return err;
 864
 865         if (S_ISREG(inode->i_mode) &&
 866             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
 867                 struct btrfs_trans_handle *trans;
 868                 struct btrfs_root *root = BTRFS_I(inode)->root;
 869                 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 870
 871                 u64 mask = root->sectorsize - 1;
 872                 u64 pos = (inode->i_size + mask) & ~mask;
 873                 u64 block_end = attr->ia_size | mask;
 874                 u64 hole_size;
 875                 u64 alloc_hint = 0;
 876
 877                 if (attr->ia_size <= pos)
 878                         goto out;
 879
 880                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
 881
 882                 lock_extent(em_tree, pos, block_end, GFP_NOFS);
 883                 hole_size = (attr->ia_size - pos + mask) & ~mask;
 884
 885                 mutex_lock(&root->fs_info->fs_mutex);
 886                 trans = btrfs_start_transaction(root, 1);
 887                 btrfs_set_trans_block_group(trans, inode);
 888                 err = btrfs_drop_extents(trans, root, inode,
 889                                          pos, pos + hole_size, pos,
 890                                          &alloc_hint);
 891
 892                 if (alloc_hint != EXTENT_MAP_INLINE) {
 893                         err = btrfs_insert_file_extent(trans, root,
 894                                                        inode->i_ino,
 895                                                        pos, 0, 0, hole_size);
 896                 }
 897                 btrfs_end_transaction(trans, root);
 898                 mutex_unlock(&root->fs_info->fs_mutex);
 899                 unlock_extent(em_tree, pos, block_end, GFP_NOFS);
 900                 if (err)
 901                         return err;
 902         }
 903 out:
 904         err = inode_setattr(inode, attr);
 905
 906         return err;
 907 }
 908 void btrfs_delete_inode(struct inode *inode)
 909 {
 910         struct btrfs_trans_handle *trans;
 911         struct btrfs_root *root = BTRFS_I(inode)->root;
 912         unsigned long nr;
 913         int ret;
 914
 915         truncate_inode_pages(&inode->i_data, 0);
 916         if (is_bad_inode(inode)) {
 917                 goto no_delete;
 918         }
 919
 920         inode->i_size = 0;
 921         mutex_lock(&root->fs_info->fs_mutex);
 922         trans = btrfs_start_transaction(root, 1);
 923
 924         btrfs_set_trans_block_group(trans, inode);
 925         ret = btrfs_truncate_in_trans(trans, root, inode);
 926         if (ret)
 927                 goto no_delete_lock;
 928         ret = btrfs_delete_xattrs(trans, root, inode);
 929         if (ret)
 930                 goto no_delete_lock;
 931         ret = btrfs_free_inode(trans, root, inode);
 932         if (ret)
 933                 goto no_delete_lock;
 934         nr = trans->blocks_used;
 935
 936         btrfs_end_transaction(trans, root);
 937         mutex_unlock(&root->fs_info->fs_mutex);
 938         btrfs_btree_balance_dirty(root, nr);
 939         return;
 940
 941 no_delete_lock:
 942         nr = trans->blocks_used;
 943         btrfs_end_transaction(trans, root);
 944         mutex_unlock(&root->fs_info->fs_mutex);
 945         btrfs_btree_balance_dirty(root, nr);
 946 no_delete:
 947         clear_inode(inode);
 948 }
 949
 950 /*
 951  * this returns the key found in the dir entry in the location pointer.
 952  * If no dir entries were found, location->objectid is 0.
 953  */
 954 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 955                                struct btrfs_key *location)
 956 {
 957         const char *name = dentry->d_name.name;
 958         int namelen = dentry->d_name.len;
 959         struct btrfs_dir_item *di;
 960         struct btrfs_path *path;
 961         struct btrfs_root *root = BTRFS_I(dir)->root;
 962         int ret = 0;
 963
 964         if (namelen == 1 && strcmp(name, ".") == 0) {
 965                 location->objectid = dir->i_ino;
 966                 location->type = BTRFS_INODE_ITEM_KEY;
 967                 location->offset = 0;
 968                 return 0;
 969         }
 970         path = btrfs_alloc_path();
 971         BUG_ON(!path);
 972
 973         if (namelen == 2 && strcmp(name, "..") == 0) {
 974                 struct btrfs_key key;
 975                 struct extent_buffer *leaf;
 976                 u32 nritems;
 977                 int slot;
 978
 979                 key.objectid = dir->i_ino;
 980                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
 981                 key.offset = 0;
 982                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 983                 BUG_ON(ret == 0);
 984                 ret = 0;
 985
 986                 leaf = path->nodes[0];
 987                 slot = path->slots[0];
 988                 nritems = btrfs_header_nritems(leaf);
 989                 if (slot >= nritems)
 990                         goto out_err;
 991
 992                 btrfs_item_key_to_cpu(leaf, &key, slot);
 993                 if (key.objectid != dir->i_ino ||
 994                     key.type != BTRFS_INODE_REF_KEY) {
 995                         goto out_err;
 996                 }
 997                 location->objectid = key.offset;
 998                 location->type = BTRFS_INODE_ITEM_KEY;
 999                 location->offset = 0;
1000                 goto out;
1001         }
1002
1003         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1004                                     namelen, 0);
1005         if (IS_ERR(di))
1006                 ret = PTR_ERR(di);
1007         if (!di || IS_ERR(di)) {
1008                 goto out_err;
1009         }
1010         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1011 out:
1012         btrfs_free_path(path);
1013         return ret;
1014 out_err:
1015         location->objectid = 0;
1016         goto out;
1017 }
1018
1019 /*
1020  * when we hit a tree root in a directory, the btrfs part of the inode
1021  * needs to be changed to reflect the root directory of the tree root.  This
1022  * is kind of like crossing a mount point.
1023  */
1024 static int fixup_tree_root_location(struct btrfs_root *root,
1025                              struct btrfs_key *location,
1026                              struct btrfs_root **sub_root,
1027                              struct dentry *dentry)
1028 {
1029         struct btrfs_path *path;
1030         struct btrfs_root_item *ri;
1031
1032         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1033                 return 0;
1034         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1035                 return 0;
1036
1037         path = btrfs_alloc_path();
1038         BUG_ON(!path);
1039         mutex_lock(&root->fs_info->fs_mutex);
1040
1041         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1042                                         dentry->d_name.name,
1043                                         dentry->d_name.len);
1044         if (IS_ERR(*sub_root))
1045                 return PTR_ERR(*sub_root);
1046
1047         ri = &(*sub_root)->root_item;
1048         location->objectid = btrfs_root_dirid(ri);
1049         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1050         location->offset = 0;
1051
1052         btrfs_free_path(path);
1053         mutex_unlock(&root->fs_info->fs_mutex);
1054         return 0;
1055 }
1056
1057 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1058 {
1059         struct btrfs_iget_args *args = p;
1060         inode->i_ino = args->ino;
1061         BTRFS_I(inode)->root = args->root;
1062         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1063                              inode->i_mapping, GFP_NOFS);
1064         return 0;
1065 }
1066
1067 static int btrfs_find_actor(struct inode *inode, void *opaque)
1068 {
1069         struct btrfs_iget_args *args = opaque;
1070         return (args->ino == inode->i_ino &&
1071                 args->root == BTRFS_I(inode)->root);
1072 }
1073
1074 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1075                                 struct btrfs_root *root)
1076 {
1077         struct inode *inode;
1078         struct btrfs_iget_args args;
1079         args.ino = objectid;
1080         args.root = root;
1081
1082         inode = iget5_locked(s, objectid, btrfs_find_actor,
1083                              btrfs_init_locked_inode,
1084                              (void *)&args);
1085         return inode;
1086 }
1087
1088 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1089                                    struct nameidata *nd)
1090 {
1091         struct inode * inode;
1092         struct btrfs_inode *bi = BTRFS_I(dir);
1093         struct btrfs_root *root = bi->root;
1094         struct btrfs_root *sub_root = root;
1095         struct btrfs_key location;
1096         int ret;
1097
1098         if (dentry->d_name.len > BTRFS_NAME_LEN)
1099                 return ERR_PTR(-ENAMETOOLONG);
1100
1101         mutex_lock(&root->fs_info->fs_mutex);
1102         ret = btrfs_inode_by_name(dir, dentry, &location);
1103         mutex_unlock(&root->fs_info->fs_mutex);
1104
1105         if (ret < 0)
1106                 return ERR_PTR(ret);
1107
1108         inode = NULL;
1109         if (location.objectid) {
1110                 ret = fixup_tree_root_location(root, &location, &sub_root,
1111                                                 dentry);
1112                 if (ret < 0)
1113                         return ERR_PTR(ret);
1114                 if (ret > 0)
1115                         return ERR_PTR(-ENOENT);
1116                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1117                                           sub_root);
1118                 if (!inode)
1119                         return ERR_PTR(-EACCES);
1120                 if (inode->i_state & I_NEW) {
1121                         /* the inode and parent dir are two different roots */
1122                         if (sub_root != root) {
1123                                 igrab(inode);
1124                                 sub_root->inode = inode;
1125                         }
1126                         BTRFS_I(inode)->root = sub_root;
1127                         memcpy(&BTRFS_I(inode)->location, &location,
1128                                sizeof(location));
1129                         btrfs_read_locked_inode(inode);
1130                         unlock_new_inode(inode);
1131                 }
1132         }
1133         return d_splice_alias(inode, dentry);
1134 }
1135
1136 static unsigned char btrfs_filetype_table[] = {
1137         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1138 };
1139
1140 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1141 {
1142         struct inode *inode = filp->f_path.dentry->d_inode;
1143         struct btrfs_root *root = BTRFS_I(inode)->root;
1144         struct btrfs_item *item;
1145         struct btrfs_dir_item *di;
1146         struct btrfs_key key;
1147         struct btrfs_key found_key;
1148         struct btrfs_path *path;
1149         int ret;
1150         u32 nritems;
1151         struct extent_buffer *leaf;
1152         int slot;
1153         int advance;
1154         unsigned char d_type;
1155         int over = 0;
1156         u32 di_cur;
1157         u32 di_total;
1158         u32 di_len;
1159         int key_type = BTRFS_DIR_INDEX_KEY;
1160         char tmp_name[32];
1161         char *name_ptr;
1162         int name_len;
1163
1164         /* FIXME, use a real flag for deciding about the key type */
1165         if (root->fs_info->tree_root == root)
1166                 key_type = BTRFS_DIR_ITEM_KEY;
1167
1168         /* special case for "." */
1169         if (filp->f_pos == 0) {
1170                 over = filldir(dirent, ".", 1,
1171                                1, inode->i_ino,
1172                                DT_DIR);
1173                 if (over)
1174                         return 0;
1175                 filp->f_pos = 1;
1176         }
1177
1178         mutex_lock(&root->fs_info->fs_mutex);
1179         key.objectid = inode->i_ino;
1180         path = btrfs_alloc_path();
1181         path->reada = 2;
1182
1183         /* special case for .., just use the back ref */
1184         if (filp->f_pos == 1) {
1185                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1186                 key.offset = 0;
1187                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1188                 BUG_ON(ret == 0);
1189                 leaf = path->nodes[0];
1190                 slot = path->slots[0];
1191                 nritems = btrfs_header_nritems(leaf);
1192                 if (slot >= nritems) {
1193                         btrfs_release_path(root, path);
1194                         goto read_dir_items;
1195                 }
1196                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1197                 btrfs_release_path(root, path);
1198                 if (found_key.objectid != key.objectid ||
1199                     found_key.type != BTRFS_INODE_REF_KEY)
1200                         goto read_dir_items;
1201                 over = filldir(dirent, "..", 2,
1202                                2, found_key.offset, DT_DIR);
1203                 if (over)
1204                         goto nopos;
1205                 filp->f_pos = 2;
1206         }
1207
1208 read_dir_items:
1209         btrfs_set_key_type(&key, key_type);
1210         key.offset = filp->f_pos;
1211
1212         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1213         if (ret < 0)
1214                 goto err;
1215         advance = 0;
1216         while(1) {
1217                 leaf = path->nodes[0];
1218                 nritems = btrfs_header_nritems(leaf);
1219                 slot = path->slots[0];
1220                 if (advance || slot >= nritems) {
1221                         if (slot >= nritems -1) {
1222                                 ret = btrfs_next_leaf(root, path);
1223                                 if (ret)
1224                                         break;
1225                                 leaf = path->nodes[0];
1226                                 nritems = btrfs_header_nritems(leaf);
1227                                 slot = path->slots[0];
1228                         } else {
1229                                 slot++;
1230                                 path->slots[0]++;
1231                         }
1232                 }
1233                 advance = 1;
1234                 item = btrfs_item_nr(leaf, slot);
1235                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1236
1237                 if (found_key.objectid != key.objectid)
1238                         break;
1239                 if (btrfs_key_type(&found_key) != key_type)
1240                         break;
1241                 if (found_key.offset < filp->f_pos)
1242                         continue;
1243
1244                 filp->f_pos = found_key.offset;
1245                 advance = 1;
1246                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1247                 di_cur = 0;
1248                 di_total = btrfs_item_size(leaf, item);
1249                 while(di_cur < di_total) {
1250                         struct btrfs_key location;
1251
1252                         name_len = btrfs_dir_name_len(leaf, di);
1253                         if (name_len < 32) {
1254                                 name_ptr = tmp_name;
1255                         } else {
1256                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1257                                 BUG_ON(!name_ptr);
1258                         }
1259                         read_extent_buffer(leaf, name_ptr,
1260                                            (unsigned long)(di + 1), name_len);
1261
1262                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1263                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1264
1265                         over = filldir(dirent, name_ptr, name_len,
1266                                        found_key.offset,
1267                                        location.objectid,
1268                                        d_type);
1269
1270                         if (name_ptr != tmp_name)
1271                                 kfree(name_ptr);
1272
1273                         if (over)
1274                                 goto nopos;
1275                         di_len = btrfs_dir_name_len(leaf, di) +
1276                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1277                         di_cur += di_len;
1278                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1279                 }
1280         }
1281         filp->f_pos++;
1282 nopos:
1283         ret = 0;
1284 err:
1285         btrfs_release_path(root, path);
1286         btrfs_free_path(path);
1287         mutex_unlock(&root->fs_info->fs_mutex);
1288         return ret;
1289 }
1290
1291 int btrfs_write_inode(struct inode *inode, int wait)
1292 {
1293         struct btrfs_root *root = BTRFS_I(inode)->root;
1294         struct btrfs_trans_handle *trans;
1295         int ret = 0;
1296
1297         if (wait) {
1298                 mutex_lock(&root->fs_info->fs_mutex);
1299                 trans = btrfs_start_transaction(root, 1);
1300                 btrfs_set_trans_block_group(trans, inode);
1301                 ret = btrfs_commit_transaction(trans, root);
1302                 mutex_unlock(&root->fs_info->fs_mutex);
1303         }
1304         return ret;
1305 }
1306
1307 /*
1308  * This is somewhat expensive, updating the tree every time the
1309  * inode changes.  But, it is most likely to find the inode in cache.
1310  * FIXME, needs more benchmarking...there are no reasons other than performance
1311  * to keep or drop this code.
1312  */
1313 void btrfs_dirty_inode(struct inode *inode)
1314 {
1315         struct btrfs_root *root = BTRFS_I(inode)->root;
1316         struct btrfs_trans_handle *trans;
1317
1318         mutex_lock(&root->fs_info->fs_mutex);
1319         trans = btrfs_start_transaction(root, 1);
1320         btrfs_set_trans_block_group(trans, inode);
1321         btrfs_update_inode(trans, root, inode);
1322         btrfs_end_transaction(trans, root);
1323         mutex_unlock(&root->fs_info->fs_mutex);
1324 }
1325
1326 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1327                                      struct btrfs_root *root,
1328                                      u64 objectid,
1329                                      struct btrfs_block_group_cache *group,
1330                                      int mode)
1331 {
1332         struct inode *inode;
1333         struct btrfs_inode_item *inode_item;
1334         struct btrfs_key *location;
1335         struct btrfs_path *path;
1336         int ret;
1337         int owner;
1338
1339         path = btrfs_alloc_path();
1340         BUG_ON(!path);
1341
1342         inode = new_inode(root->fs_info->sb);
1343         if (!inode)
1344                 return ERR_PTR(-ENOMEM);
1345
1346         extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1347                              inode->i_mapping, GFP_NOFS);
1348         BTRFS_I(inode)->root = root;
1349
1350         if (mode & S_IFDIR)
1351                 owner = 0;
1352         else
1353                 owner = 1;
1354         group = btrfs_find_block_group(root, group, 0, 0, owner);
1355         BTRFS_I(inode)->block_group = group;
1356
1357         ret = btrfs_insert_empty_inode(trans, root, path, objectid);
1358         if (ret)
1359                 goto fail;
1360
1361         inode->i_uid = current->fsuid;
1362         inode->i_gid = current->fsgid;
1363         inode->i_mode = mode;
1364         inode->i_ino = objectid;
1365         inode->i_blocks = 0;
1366         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1367         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1368                                   struct btrfs_inode_item);
1369         fill_inode_item(path->nodes[0], inode_item, inode);
1370         btrfs_mark_buffer_dirty(path->nodes[0]);
1371         btrfs_free_path(path);
1372
1373         location = &BTRFS_I(inode)->location;
1374         location->objectid = objectid;
1375         location->offset = 0;
1376         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1377
1378         insert_inode_hash(inode);
1379         return inode;
1380 fail:
1381         btrfs_free_path(path);
1382         return ERR_PTR(ret);
1383 }
1384
1385 static inline u8 btrfs_inode_type(struct inode *inode)
1386 {
1387         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1388 }
1389
1390 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1391                             struct dentry *dentry, struct inode *inode)
1392 {
1393         int ret;
1394         struct btrfs_key key;
1395         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1396         struct inode *parent_inode;
1397
1398         key.objectid = inode->i_ino;
1399         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1400         key.offset = 0;
1401
1402         ret = btrfs_insert_dir_item(trans, root,
1403                                     dentry->d_name.name, dentry->d_name.len,
1404                                     dentry->d_parent->d_inode->i_ino,
1405                                     &key, btrfs_inode_type(inode));
1406         if (ret == 0) {
1407                 ret = btrfs_insert_inode_ref(trans, root,
1408                                      dentry->d_name.name,
1409                                      dentry->d_name.len,
1410                                      inode->i_ino,
1411                                      dentry->d_parent->d_inode->i_ino);
1412                 parent_inode = dentry->d_parent->d_inode;
1413                 parent_inode->i_size += dentry->d_name.len * 2;
1414                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1415                 ret = btrfs_update_inode(trans, root,
1416                                          dentry->d_parent->d_inode);
1417         }
1418         return ret;
1419 }
1420
1421 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1422                             struct dentry *dentry, struct inode *inode)
1423 {
1424         int err = btrfs_add_link(trans, dentry, inode);
1425         if (!err) {
1426                 d_instantiate(dentry, inode);
1427                 return 0;
1428         }
1429         if (err > 0)
1430                 err = -EEXIST;
1431         return err;
1432 }
1433
1434 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1435                         int mode, dev_t rdev)
1436 {
1437         struct btrfs_trans_handle *trans;
1438         struct btrfs_root *root = BTRFS_I(dir)->root;
1439         struct inode *inode;
1440         int err;
1441         int drop_inode = 0;
1442         u64 objectid;
1443         unsigned long nr;
1444
1445         if (!new_valid_dev(rdev))
1446                 return -EINVAL;
1447
1448         mutex_lock(&root->fs_info->fs_mutex);
1449         trans = btrfs_start_transaction(root, 1);
1450         btrfs_set_trans_block_group(trans, dir);
1451
1452         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1453         if (err) {
1454                 err = -ENOSPC;
1455                 goto out_unlock;
1456         }
1457
1458         inode = btrfs_new_inode(trans, root, objectid,
1459                                 BTRFS_I(dir)->block_group, mode);
1460         err = PTR_ERR(inode);
1461         if (IS_ERR(inode))
1462                 goto out_unlock;
1463
1464         btrfs_set_trans_block_group(trans, inode);
1465         err = btrfs_add_nondir(trans, dentry, inode);
1466         if (err)
1467                 drop_inode = 1;
1468         else {
1469                 inode->i_op = &btrfs_special_inode_operations;
1470                 init_special_inode(inode, inode->i_mode, rdev);
1471                 btrfs_update_inode(trans, root, inode);
1472         }
1473         dir->i_sb->s_dirt = 1;
1474         btrfs_update_inode_block_group(trans, inode);
1475         btrfs_update_inode_block_group(trans, dir);
1476 out_unlock:
1477         nr = trans->blocks_used;
1478         btrfs_end_transaction(trans, root);
1479         mutex_unlock(&root->fs_info->fs_mutex);
1480
1481         if (drop_inode) {
1482                 inode_dec_link_count(inode);
1483                 iput(inode);
1484         }
1485         btrfs_btree_balance_dirty(root, nr);
1486         return err;
1487 }
1488
1489 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1490                         int mode, struct nameidata *nd)
1491 {
1492         struct btrfs_trans_handle *trans;
1493         struct btrfs_root *root = BTRFS_I(dir)->root;
1494         struct inode *inode;
1495         int err;
1496         int drop_inode = 0;
1497         unsigned long nr;
1498         u64 objectid;
1499
1500         mutex_lock(&root->fs_info->fs_mutex);
1501         trans = btrfs_start_transaction(root, 1);
1502         btrfs_set_trans_block_group(trans, dir);
1503
1504         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1505         if (err) {
1506                 err = -ENOSPC;
1507                 goto out_unlock;
1508         }
1509
1510         inode = btrfs_new_inode(trans, root, objectid,
1511                                 BTRFS_I(dir)->block_group, mode);
1512         err = PTR_ERR(inode);
1513         if (IS_ERR(inode))
1514                 goto out_unlock;
1515
1516         btrfs_set_trans_block_group(trans, inode);
1517         err = btrfs_add_nondir(trans, dentry, inode);
1518         if (err)
1519                 drop_inode = 1;
1520         else {
1521                 inode->i_mapping->a_ops = &btrfs_aops;
1522                 inode->i_fop = &btrfs_file_operations;
1523                 inode->i_op = &btrfs_file_inode_operations;
1524                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
1525                                      inode->i_mapping, GFP_NOFS);
1526                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
1527         }
1528         dir->i_sb->s_dirt = 1;
1529         btrfs_update_inode_block_group(trans, inode);
1530         btrfs_update_inode_block_group(trans, dir);
1531 out_unlock:
1532         nr = trans->blocks_used;
1533         btrfs_end_transaction(trans, root);
1534         mutex_unlock(&root->fs_info->fs_mutex);
1535
1536         if (drop_inode) {
1537                 inode_dec_link_count(inode);
1538                 iput(inode);
1539         }
1540         btrfs_btree_balance_dirty(root, nr);
1541         return err;
1542 }
1543
1544 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1545                       struct dentry *dentry)
1546 {
1547         struct btrfs_trans_handle *trans;
1548         struct btrfs_root *root = BTRFS_I(dir)->root;
1549         struct inode *inode = old_dentry->d_inode;
1550         unsigned long nr;
1551         int err;
1552         int drop_inode = 0;
1553
1554         if (inode->i_nlink == 0)
1555                 return -ENOENT;
1556
1557         inc_nlink(inode);
1558         mutex_lock(&root->fs_info->fs_mutex);
1559         trans = btrfs_start_transaction(root, 1);
1560
1561         btrfs_set_trans_block_group(trans, dir);
1562         atomic_inc(&inode->i_count);
1563         err = btrfs_add_nondir(trans, dentry, inode);
1564
1565         if (err)
1566                 drop_inode = 1;
1567
1568         dir->i_sb->s_dirt = 1;
1569         btrfs_update_inode_block_group(trans, dir);
1570         err = btrfs_update_inode(trans, root, inode);
1571
1572         if (err)
1573                 drop_inode = 1;
1574
1575         nr = trans->blocks_used;
1576         btrfs_end_transaction(trans, root);
1577         mutex_unlock(&root->fs_info->fs_mutex);
1578
1579         if (drop_inode) {
1580                 inode_dec_link_count(inode);
1581                 iput(inode);
1582         }
1583         btrfs_btree_balance_dirty(root, nr);
1584         return err;
1585 }
1586
1587 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1588 {
1589         struct inode *inode;
1590         struct btrfs_trans_handle *trans;
1591         struct btrfs_root *root = BTRFS_I(dir)->root;
1592         int err = 0;
1593         int drop_on_err = 0;
1594         u64 objectid;
1595         unsigned long nr = 1;
1596
1597         mutex_lock(&root->fs_info->fs_mutex);
1598         trans = btrfs_start_transaction(root, 1);
1599         btrfs_set_trans_block_group(trans, dir);
1600
1601         if (IS_ERR(trans)) {
1602                 err = PTR_ERR(trans);
1603                 goto out_unlock;
1604         }
1605
1606         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1607         if (err) {
1608                 err = -ENOSPC;
1609                 goto out_unlock;
1610         }
1611
1612         inode = btrfs_new_inode(trans, root, objectid,
1613                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1614         if (IS_ERR(inode)) {
1615                 err = PTR_ERR(inode);
1616                 goto out_fail;
1617         }
1618
1619         drop_on_err = 1;
1620         inode->i_op = &btrfs_dir_inode_operations;
1621         inode->i_fop = &btrfs_dir_file_operations;
1622         btrfs_set_trans_block_group(trans, inode);
1623
1624         inode->i_size = 0;
1625         err = btrfs_update_inode(trans, root, inode);
1626         if (err)
1627                 goto out_fail;
1628
1629         err = btrfs_add_link(trans, dentry, inode);
1630         if (err)
1631                 goto out_fail;
1632
1633         d_instantiate(dentry, inode);
1634         drop_on_err = 0;
1635         dir->i_sb->s_dirt = 1;
1636         btrfs_update_inode_block_group(trans, inode);
1637         btrfs_update_inode_block_group(trans, dir);
1638
1639 out_fail:
1640         nr = trans->blocks_used;
1641         btrfs_end_transaction(trans, root);
1642
1643 out_unlock:
1644         mutex_unlock(&root->fs_info->fs_mutex);
1645         if (drop_on_err)
1646                 iput(inode);
1647         btrfs_btree_balance_dirty(root, nr);
1648         return err;
1649 }
1650
1651 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1652                                     size_t page_offset, u64 start, u64 end,
1653                                     int create)
1654 {
1655         int ret;
1656         int err = 0;
1657         u64 bytenr;
1658         u64 extent_start = 0;
1659         u64 extent_end = 0;
1660         u64 objectid = inode->i_ino;
1661         u32 found_type;
1662         int failed_insert = 0;
1663         struct btrfs_path *path;
1664         struct btrfs_root *root = BTRFS_I(inode)->root;
1665         struct btrfs_file_extent_item *item;
1666         struct extent_buffer *leaf;
1667         struct btrfs_key found_key;
1668         struct extent_map *em = NULL;
1669         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1670         struct btrfs_trans_handle *trans = NULL;
1671
1672         path = btrfs_alloc_path();
1673         BUG_ON(!path);
1674         mutex_lock(&root->fs_info->fs_mutex);
1675
1676 again:
1677         em = lookup_extent_mapping(em_tree, start, end);
1678         if (em) {
1679                 goto out;
1680         }
1681         if (!em) {
1682                 em = alloc_extent_map(GFP_NOFS);
1683                 if (!em) {
1684                         err = -ENOMEM;
1685                         goto out;
1686                 }
1687                 em->start = EXTENT_MAP_HOLE;
1688                 em->end = EXTENT_MAP_HOLE;
1689         }
1690         em->bdev = inode->i_sb->s_bdev;
1691         ret = btrfs_lookup_file_extent(trans, root, path,
1692                                        objectid, start, trans != NULL);
1693         if (ret < 0) {
1694                 err = ret;
1695                 goto out;
1696         }
1697
1698         if (ret != 0) {
1699                 if (path->slots[0] == 0)
1700                         goto not_found;
1701                 path->slots[0]--;
1702         }
1703
1704         leaf = path->nodes[0];
1705         item = btrfs_item_ptr(leaf, path->slots[0],
1706                               struct btrfs_file_extent_item);
1707         /* are we inside the extent that was found? */
1708         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1709         found_type = btrfs_key_type(&found_key);
1710         if (found_key.objectid != objectid ||
1711             found_type != BTRFS_EXTENT_DATA_KEY) {
1712                 goto not_found;
1713         }
1714
1715         found_type = btrfs_file_extent_type(leaf, item);
1716         extent_start = found_key.offset;
1717         if (found_type == BTRFS_FILE_EXTENT_REG) {
1718                 extent_end = extent_start +
1719                        btrfs_file_extent_num_bytes(leaf, item);
1720                 err = 0;
1721                 if (start < extent_start || start >= extent_end) {
1722                         em->start = start;
1723                         if (start < extent_start) {
1724                                 if (end < extent_start)
1725                                         goto not_found;
1726                                 em->end = extent_end - 1;
1727                         } else {
1728                                 em->end = end;
1729                         }
1730                         goto not_found_em;
1731                 }
1732                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
1733                 if (bytenr == 0) {
1734                         em->start = extent_start;
1735                         em->end = extent_end - 1;
1736                         em->block_start = EXTENT_MAP_HOLE;
1737                         em->block_end = EXTENT_MAP_HOLE;
1738                         goto insert;
1739                 }
1740                 bytenr += btrfs_file_extent_offset(leaf, item);
1741                 em->block_start = bytenr;
1742                 em->block_end = em->block_start +
1743                         btrfs_file_extent_num_bytes(leaf, item) - 1;
1744                 em->start = extent_start;
1745                 em->end = extent_end - 1;
1746                 goto insert;
1747         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1748                 unsigned long ptr;
1749                 char *map;
1750                 size_t size;
1751                 size_t extent_offset;
1752                 size_t copy_size;
1753
1754                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
1755                                                     path->slots[0]));
1756                 extent_end = (extent_start + size - 1) |
1757                         ((u64)root->sectorsize - 1);
1758                 if (start < extent_start || start >= extent_end) {
1759                         em->start = start;
1760                         if (start < extent_start) {
1761                                 if (end < extent_start)
1762                                         goto not_found;
1763                                 em->end = extent_end;
1764                         } else {
1765                                 em->end = end;
1766                         }
1767                         goto not_found_em;
1768                 }
1769                 em->block_start = EXTENT_MAP_INLINE;
1770                 em->block_end = EXTENT_MAP_INLINE;
1771
1772                 if (!page) {
1773                         em->start = extent_start;
1774                         em->end = extent_start + size - 1;
1775                         goto out;
1776                 }
1777
1778                 extent_offset = ((u64)page->index << PAGE_CACHE_SHIFT) -
1779                         extent_start + page_offset;
1780                 copy_size = min_t(u64, PAGE_CACHE_SIZE - page_offset,
1781                                 size - extent_offset);
1782                 em->start = extent_start + extent_offset;
1783                 em->end = (em->start + copy_size -1) |
1784                         ((u64)root->sectorsize -1);
1785                 map = kmap(page);
1786                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
1787                 if (create == 0 && !PageUptodate(page)) {
1788                         read_extent_buffer(leaf, map + page_offset, ptr,
1789                                            copy_size);
1790                         flush_dcache_page(page);
1791                 } else if (create && PageUptodate(page)) {
1792                         if (!trans) {
1793                                 kunmap(page);
1794                                 free_extent_map(em);
1795                                 em = NULL;
1796                                 btrfs_release_path(root, path);
1797                                 trans = btrfs_start_transaction(root, 1);
1798                                 goto again;
1799                         }
1800                         write_extent_buffer(leaf, map + page_offset, ptr,
1801                                             copy_size);
1802                         btrfs_mark_buffer_dirty(leaf);
1803                 }
1804                 kunmap(page);
1805                 set_extent_uptodate(em_tree, em->start, em->end, GFP_NOFS);
1806                 goto insert;
1807         } else {
1808                 printk("unkknown found_type %d\n", found_type);
1809                 WARN_ON(1);
1810         }
1811 not_found:
1812         em->start = start;
1813         em->end = end;
1814 not_found_em:
1815         em->block_start = EXTENT_MAP_HOLE;
1816         em->block_end = EXTENT_MAP_HOLE;
1817 insert:
1818         btrfs_release_path(root, path);
1819         if (em->start > start || em->end < start) {
1820                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->end, start, end);
1821                 err = -EIO;
1822                 goto out;
1823         }
1824         ret = add_extent_mapping(em_tree, em);
1825         if (ret == -EEXIST) {
1826                 free_extent_map(em);
1827                 em = NULL;
1828                 failed_insert++;
1829                 if (failed_insert > 5) {
1830                         printk("failing to insert %Lu %Lu\n", start, end);
1831                         err = -EIO;
1832                         goto out;
1833                 }
1834                 goto again;
1835         }
1836         err = 0;
1837 out:
1838         btrfs_free_path(path);
1839         if (trans) {
1840                 ret = btrfs_end_transaction(trans, root);
1841                 if (!err)
1842                         err = ret;
1843         }
1844         mutex_unlock(&root->fs_info->fs_mutex);
1845         if (err) {
1846                 free_extent_map(em);
1847                 WARN_ON(1);
1848                 return ERR_PTR(err);
1849         }
1850         return em;
1851 }
1852
1853 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
1854 {
1855         return extent_bmap(mapping, iblock, btrfs_get_extent);
1856 }
1857
1858 static int btrfs_prepare_write(struct file *file, struct page *page,
1859                                unsigned from, unsigned to)
1860 {
1861         return extent_prepare_write(&BTRFS_I(page->mapping->host)->extent_tree,
1862                                     page->mapping->host, page, from, to,
1863                                     btrfs_get_extent);
1864 }
1865
1866 int btrfs_readpage(struct file *file, struct page *page)
1867 {
1868         struct extent_map_tree *tree;
1869         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1870         return extent_read_full_page(tree, page, btrfs_get_extent);
1871 }
1872 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1873 {
1874         struct extent_map_tree *tree;
1875
1876
1877         if (current->flags & PF_MEMALLOC) {
1878                 redirty_page_for_writepage(wbc, page);
1879                 unlock_page(page);
1880                 return 0;
1881         }
1882         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1883         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
1884 }
1885
1886 static int btrfs_writepages(struct address_space *mapping,
1887                             struct writeback_control *wbc)
1888 {
1889         struct extent_map_tree *tree;
1890         tree = &BTRFS_I(mapping->host)->extent_tree;
1891         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
1892 }
1893
1894 static int
1895 btrfs_readpages(struct file *file, struct address_space *mapping,
1896                 struct list_head *pages, unsigned nr_pages)
1897 {
1898         struct extent_map_tree *tree;
1899         tree = &BTRFS_I(mapping->host)->extent_tree;
1900         return extent_readpages(tree, mapping, pages, nr_pages,
1901                                 btrfs_get_extent);
1902 }
1903
1904 static int btrfs_releasepage(struct page *page, gfp_t unused_gfp_flags)
1905 {
1906         struct extent_map_tree *tree;
1907         int ret;
1908
1909         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1910         ret = try_release_extent_mapping(tree, page);
1911         if (ret == 1) {
1912                 ClearPagePrivate(page);
1913                 set_page_private(page, 0);
1914                 page_cache_release(page);
1915         }
1916         return ret;
1917 }
1918
1919 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
1920 {
1921         struct extent_map_tree *tree;
1922
1923         tree = &BTRFS_I(page->mapping->host)->extent_tree;
1924         extent_invalidatepage(tree, page, offset);
1925         btrfs_releasepage(page, GFP_NOFS);
1926 }
1927
1928 /*
1929  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
1930  * called from a page fault handler when a page is first dirtied. Hence we must
1931  * be careful to check for EOF conditions here. We set the page up correctly
1932  * for a written page which means we get ENOSPC checking when writing into
1933  * holes and correct delalloc and unwritten extent mapping on filesystems that
1934  * support these features.
1935  *
1936  * We are not allowed to take the i_mutex here so we have to play games to
1937  * protect against truncate races as the page could now be beyond EOF.  Because
1938  * vmtruncate() writes the inode size before removing pages, once we have the
1939  * page lock we can determine safely if the page is beyond EOF. If it is not
1940  * beyond EOF, then the page is guaranteed safe against truncation until we
1941  * unlock the page.
1942  */
1943 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
1944 {
1945         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1946         unsigned long end;
1947         loff_t size;
1948         int ret = -EINVAL;
1949         u64 page_start;
1950
1951         down_read(&BTRFS_I(inode)->root->snap_sem);
1952         lock_page(page);
1953         wait_on_page_writeback(page);
1954         size = i_size_read(inode);
1955         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1956
1957         if ((page->mapping != inode->i_mapping) ||
1958             (page_start > size)) {
1959                 /* page got truncated out from underneath us */
1960                 goto out_unlock;
1961         }
1962
1963         /* page is wholly or partially inside EOF */
1964         if (page_start + PAGE_CACHE_SIZE > size)
1965                 end = size & ~PAGE_CACHE_MASK;
1966         else
1967                 end = PAGE_CACHE_SIZE;
1968
1969         ret = btrfs_cow_one_page(inode, page, end);
1970
1971 out_unlock:
1972         up_read(&BTRFS_I(inode)->root->snap_sem);
1973         unlock_page(page);
1974         return ret;
1975 }
1976
1977 static void btrfs_truncate(struct inode *inode)
1978 {
1979         struct btrfs_root *root = BTRFS_I(inode)->root;
1980         int ret;
1981         struct btrfs_trans_handle *trans;
1982         unsigned long nr;
1983
1984         if (!S_ISREG(inode->i_mode))
1985                 return;
1986         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1987                 return;
1988
1989         btrfs_truncate_page(inode->i_mapping, inode->i_size);
1990
1991         mutex_lock(&root->fs_info->fs_mutex);
1992         trans = btrfs_start_transaction(root, 1);
1993         btrfs_set_trans_block_group(trans, inode);
1994
1995         /* FIXME, add redo link to tree so we don't leak on crash */
1996         ret = btrfs_truncate_in_trans(trans, root, inode);
1997         btrfs_update_inode(trans, root, inode);
1998         nr = trans->blocks_used;
1999
2000         ret = btrfs_end_transaction(trans, root);
2001         BUG_ON(ret);
2002         mutex_unlock(&root->fs_info->fs_mutex);
2003         btrfs_btree_balance_dirty(root, nr);
2004 }
2005
2006 int btrfs_commit_write(struct file *file, struct page *page,
2007                        unsigned from, unsigned to)
2008 {
2009         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
2010         struct inode *inode = page->mapping->host;
2011
2012         btrfs_cow_one_page(inode, page, PAGE_CACHE_SIZE);
2013
2014         if (pos > inode->i_size) {
2015                 i_size_write(inode, pos);
2016                 mark_inode_dirty(inode);
2017         }
2018         return 0;
2019 }
2020
2021 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
2022 {
2023         struct btrfs_trans_handle *trans;
2024         struct btrfs_key key;
2025         struct btrfs_root_item root_item;
2026         struct btrfs_inode_item *inode_item;
2027         struct extent_buffer *leaf;
2028         struct btrfs_root *new_root;
2029         struct inode *inode;
2030         struct inode *dir;
2031         int ret;
2032         int err;
2033         u64 objectid;
2034         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2035         unsigned long nr = 1;
2036
2037         mutex_lock(&root->fs_info->fs_mutex);
2038         trans = btrfs_start_transaction(root, 1);
2039         BUG_ON(!trans);
2040
2041         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2042                                        0, &objectid);
2043         if (ret)
2044                 goto fail;
2045
2046         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2047                                         objectid, trans->transid, 0, 0,
2048                                         0, 0);
2049         if (IS_ERR(leaf))
2050                 return PTR_ERR(leaf);
2051
2052         btrfs_set_header_nritems(leaf, 0);
2053         btrfs_set_header_level(leaf, 0);
2054         btrfs_set_header_bytenr(leaf, leaf->start);
2055         btrfs_set_header_generation(leaf, trans->transid);
2056         btrfs_set_header_owner(leaf, objectid);
2057
2058         write_extent_buffer(leaf, root->fs_info->fsid,
2059                             (unsigned long)btrfs_header_fsid(leaf),
2060                             BTRFS_FSID_SIZE);
2061         btrfs_mark_buffer_dirty(leaf);
2062
2063         inode_item = &root_item.inode;
2064         memset(inode_item, 0, sizeof(*inode_item));
2065         inode_item->generation = cpu_to_le64(1);
2066         inode_item->size = cpu_to_le64(3);
2067         inode_item->nlink = cpu_to_le32(1);
2068         inode_item->nblocks = cpu_to_le64(1);
2069         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2070
2071         btrfs_set_root_bytenr(&root_item, leaf->start);
2072         btrfs_set_root_level(&root_item, 0);
2073         btrfs_set_root_refs(&root_item, 1);
2074         btrfs_set_root_used(&root_item, 0);
2075
2076         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2077         root_item.drop_level = 0;
2078
2079         free_extent_buffer(leaf);
2080         leaf = NULL;
2081
2082         btrfs_set_root_dirid(&root_item, new_dirid);
2083
2084         key.objectid = objectid;
2085         key.offset = 1;
2086         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2087         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2088                                 &root_item);
2089         if (ret)
2090                 goto fail;
2091
2092         /*
2093          * insert the directory item
2094          */
2095         key.offset = (u64)-1;
2096         dir = root->fs_info->sb->s_root->d_inode;
2097         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2098                                     name, namelen, dir->i_ino, &key,
2099                                     BTRFS_FT_DIR);
2100         if (ret)
2101                 goto fail;
2102
2103         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2104                              name, namelen, objectid,
2105                              root->fs_info->sb->s_root->d_inode->i_ino);
2106         if (ret)
2107                 goto fail;
2108
2109         ret = btrfs_commit_transaction(trans, root);
2110         if (ret)
2111                 goto fail_commit;
2112
2113         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2114         BUG_ON(!new_root);
2115
2116         trans = btrfs_start_transaction(new_root, 1);
2117         BUG_ON(!trans);
2118
2119         inode = btrfs_new_inode(trans, new_root, new_dirid,
2120                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2121         if (IS_ERR(inode))
2122                 goto fail;
2123         inode->i_op = &btrfs_dir_inode_operations;
2124         inode->i_fop = &btrfs_dir_file_operations;
2125         new_root->inode = inode;
2126
2127         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2128                                      new_dirid);
2129         inode->i_nlink = 1;
2130         inode->i_size = 0;
2131         ret = btrfs_update_inode(trans, new_root, inode);
2132         if (ret)
2133                 goto fail;
2134 fail:
2135         nr = trans->blocks_used;
2136         err = btrfs_commit_transaction(trans, root);
2137         if (err && !ret)
2138                 ret = err;
2139 fail_commit:
2140         mutex_unlock(&root->fs_info->fs_mutex);
2141         btrfs_btree_balance_dirty(root, nr);
2142         return ret;
2143 }
2144
2145 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2146 {
2147         struct btrfs_trans_handle *trans;
2148         struct btrfs_key key;
2149         struct btrfs_root_item new_root_item;
2150         struct extent_buffer *tmp;
2151         int ret;
2152         int err;
2153         u64 objectid;
2154         unsigned long nr;
2155
2156         if (!root->ref_cows)
2157                 return -EINVAL;
2158
2159         down_write(&root->snap_sem);
2160         freeze_bdev(root->fs_info->sb->s_bdev);
2161         thaw_bdev(root->fs_info->sb->s_bdev, root->fs_info->sb);
2162
2163         mutex_lock(&root->fs_info->fs_mutex);
2164         trans = btrfs_start_transaction(root, 1);
2165         BUG_ON(!trans);
2166
2167         ret = btrfs_update_inode(trans, root, root->inode);
2168         if (ret)
2169                 goto fail;
2170
2171         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2172                                        0, &objectid);
2173         if (ret)
2174                 goto fail;
2175
2176         memcpy(&new_root_item, &root->root_item,
2177                sizeof(new_root_item));
2178
2179         key.objectid = objectid;
2180         key.offset = 1;
2181         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2182
2183         extent_buffer_get(root->node);
2184         btrfs_cow_block(trans, root, root->node, NULL, 0, &tmp);
2185         free_extent_buffer(tmp);
2186
2187         btrfs_copy_root(trans, root, root->node, &tmp, objectid);
2188
2189         btrfs_set_root_bytenr(&new_root_item, tmp->start);
2190         btrfs_set_root_level(&new_root_item, btrfs_header_level(tmp));
2191         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2192                                 &new_root_item);
2193         free_extent_buffer(tmp);
2194         if (ret)
2195                 goto fail;
2196
2197         /*
2198          * insert the directory item
2199          */
2200         key.offset = (u64)-1;
2201         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2202                                     name, namelen,
2203                                     root->fs_info->sb->s_root->d_inode->i_ino,
2204                                     &key, BTRFS_FT_DIR);
2205
2206         if (ret)
2207                 goto fail;
2208
2209         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2210                              name, namelen, objectid,
2211                              root->fs_info->sb->s_root->d_inode->i_ino);
2212
2213         if (ret)
2214                 goto fail;
2215 fail:
2216         nr = trans->blocks_used;
2217         err = btrfs_commit_transaction(trans, root);
2218
2219         if (err && !ret)
2220                 ret = err;
2221
2222         mutex_unlock(&root->fs_info->fs_mutex);
2223         up_write(&root->snap_sem);
2224         btrfs_btree_balance_dirty(root, nr);
2225         return ret;
2226 }
2227
2228 static unsigned long force_ra(struct address_space *mapping,
2229                               struct file_ra_state *ra, struct file *file,
2230                               pgoff_t offset, pgoff_t last_index)
2231 {
2232         pgoff_t req_size;
2233
2234 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2235         req_size = last_index - offset + 1;
2236         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2237         return offset;
2238 #else
2239         req_size = min(last_index - offset + 1, (pgoff_t)128);
2240         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2241         return offset + req_size;
2242 #endif
2243 }
2244
2245 int btrfs_defrag_file(struct file *file) {
2246         struct inode *inode = file->f_path.dentry->d_inode;
2247         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2248         struct page *page;
2249         unsigned long last_index;
2250         unsigned long ra_index = 0;
2251         u64 page_start;
2252         u64 page_end;
2253         unsigned long i;
2254
2255         mutex_lock(&inode->i_mutex);
2256         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2257         for (i = 0; i <= last_index; i++) {
2258                 if (i == ra_index) {
2259                         ra_index = force_ra(inode->i_mapping, &file->f_ra,
2260                                             file, ra_index, last_index);
2261                 }
2262                 page = grab_cache_page(inode->i_mapping, i);
2263                 if (!page)
2264                         goto out_unlock;
2265                 if (!PageUptodate(page)) {
2266                         btrfs_readpage(NULL, page);
2267                         lock_page(page);
2268                         if (!PageUptodate(page)) {
2269                                 unlock_page(page);
2270                                 page_cache_release(page);
2271                                 goto out_unlock;
2272                         }
2273                 }
2274                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2275                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2276
2277                 lock_extent(em_tree, page_start, page_end, GFP_NOFS);
2278                 set_extent_delalloc(em_tree, page_start,
2279                                     page_end, GFP_NOFS);
2280                 unlock_extent(em_tree, page_start, page_end, GFP_NOFS);
2281                 set_page_dirty(page);
2282                 unlock_page(page);
2283                 page_cache_release(page);
2284                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2285         }
2286
2287 out_unlock:
2288         mutex_unlock(&inode->i_mutex);
2289         return 0;
2290 }
2291
2292 static int btrfs_ioctl_snap_create(struct btrfs_root *root, void __user *arg)
2293 {
2294         struct btrfs_ioctl_vol_args vol_args;
2295         struct btrfs_dir_item *di;
2296         struct btrfs_path *path;
2297         int namelen;
2298         u64 root_dirid;
2299
2300         if (copy_from_user(&vol_args, arg, sizeof(vol_args)))
2301                 return -EFAULT;
2302
2303         namelen = strlen(vol_args.name);
2304         if (namelen > BTRFS_VOL_NAME_MAX)
2305                 return -EINVAL;
2306         if (strchr(vol_args.name, '/'))
2307                 return -EINVAL;
2308
2309         path = btrfs_alloc_path();
2310         if (!path)
2311                 return -ENOMEM;
2312
2313         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2314         mutex_lock(&root->fs_info->fs_mutex);
2315         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2316                             path, root_dirid,
2317                             vol_args.name, namelen, 0);
2318         mutex_unlock(&root->fs_info->fs_mutex);
2319         btrfs_free_path(path);
2320         if (di && !IS_ERR(di))
2321                 return -EEXIST;
2322         if (IS_ERR(di))
2323                 return PTR_ERR(di);
2324
2325         if (root == root->fs_info->tree_root)
2326                 return create_subvol(root, vol_args.name, namelen);
2327         return create_snapshot(root, vol_args.name, namelen);
2328 }
2329
2330 static int btrfs_ioctl_defrag(struct file *file)
2331 {
2332         struct inode *inode = file->f_path.dentry->d_inode;
2333         struct btrfs_root *root = BTRFS_I(inode)->root;
2334
2335         switch (inode->i_mode & S_IFMT) {
2336         case S_IFDIR:
2337                 mutex_lock(&root->fs_info->fs_mutex);
2338                 btrfs_defrag_root(root, 0);
2339                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2340                 mutex_unlock(&root->fs_info->fs_mutex);
2341                 break;
2342         case S_IFREG:
2343                 btrfs_defrag_file(file);
2344                 break;
2345         }
2346
2347         return 0;
2348 }
2349
2350 long btrfs_ioctl(struct file *file, unsigned int
2351                 cmd, unsigned long arg)
2352 {
2353         struct btrfs_root *root = BTRFS_I(file->f_path.dentry->d_inode)->root;
2354
2355         switch (cmd) {
2356         case BTRFS_IOC_SNAP_CREATE:
2357                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2358         case BTRFS_IOC_DEFRAG:
2359                 return btrfs_ioctl_defrag(file);
2360         }
2361
2362         return -ENOTTY;
2363 }
2364
2365 /*
2366  * Called inside transaction, so use GFP_NOFS
2367  */
2368 struct inode *btrfs_alloc_inode(struct super_block *sb)
2369 {
2370         struct btrfs_inode *ei;
2371
2372         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2373         if (!ei)
2374                 return NULL;
2375         ei->last_trans = 0;
2376         return &ei->vfs_inode;
2377 }
2378
2379 void btrfs_destroy_inode(struct inode *inode)
2380 {
2381         WARN_ON(!list_empty(&inode->i_dentry));
2382         WARN_ON(inode->i_data.nrpages);
2383
2384         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2385 }
2386
2387 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2388 static void init_once(struct kmem_cache * cachep, void *foo)
2389 #else
2390 static void init_once(void * foo, struct kmem_cache * cachep,
2391                       unsigned long flags)
2392 #endif
2393 {
2394         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2395
2396         inode_init_once(&ei->vfs_inode);
2397 }
2398
2399 void btrfs_destroy_cachep(void)
2400 {
2401         if (btrfs_inode_cachep)
2402                 kmem_cache_destroy(btrfs_inode_cachep);
2403         if (btrfs_trans_handle_cachep)
2404                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2405         if (btrfs_transaction_cachep)
2406                 kmem_cache_destroy(btrfs_transaction_cachep);
2407         if (btrfs_bit_radix_cachep)
2408                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2409         if (btrfs_path_cachep)
2410                 kmem_cache_destroy(btrfs_path_cachep);
2411 }
2412
2413 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2414                                        unsigned long extra_flags,
2415 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2416                                        void (*ctor)(struct kmem_cache *, void *)
2417 #else
2418                                        void (*ctor)(void *, struct kmem_cache *,
2419                                                     unsigned long)
2420 #endif
2421                                      )
2422 {
2423         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2424                                  SLAB_MEM_SPREAD | extra_flags), ctor
2425 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2426                                  ,NULL
2427 #endif
2428                                 );
2429 }
2430
2431 int btrfs_init_cachep(void)
2432 {
2433         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2434                                           sizeof(struct btrfs_inode),
2435                                           0, init_once);
2436         if (!btrfs_inode_cachep)
2437                 goto fail;
2438         btrfs_trans_handle_cachep =
2439                         btrfs_cache_create("btrfs_trans_handle_cache",
2440                                            sizeof(struct btrfs_trans_handle),
2441                                            0, NULL);
2442         if (!btrfs_trans_handle_cachep)
2443                 goto fail;
2444         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2445                                              sizeof(struct btrfs_transaction),
2446                                              0, NULL);
2447         if (!btrfs_transaction_cachep)
2448                 goto fail;
2449         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2450                                          sizeof(struct btrfs_path),
2451                                          0, NULL);
2452         if (!btrfs_path_cachep)
2453                 goto fail;
2454         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2455                                               SLAB_DESTROY_BY_RCU, NULL);
2456         if (!btrfs_bit_radix_cachep)
2457                 goto fail;
2458         return 0;
2459 fail:
2460         btrfs_destroy_cachep();
2461         return -ENOMEM;
2462 }
2463
2464 static int btrfs_getattr(struct vfsmount *mnt,
2465                          struct dentry *dentry, struct kstat *stat)
2466 {
2467         struct inode *inode = dentry->d_inode;
2468         generic_fillattr(inode, stat);
2469         stat->blksize = 256 * 1024;
2470         return 0;
2471 }
2472
2473 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2474                            struct inode * new_dir,struct dentry *new_dentry)
2475 {
2476         struct btrfs_trans_handle *trans;
2477         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2478         struct inode *new_inode = new_dentry->d_inode;
2479         struct inode *old_inode = old_dentry->d_inode;
2480         struct timespec ctime = CURRENT_TIME;
2481         struct btrfs_path *path;
2482         int ret;
2483
2484         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2485             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2486                 return -ENOTEMPTY;
2487         }
2488
2489         mutex_lock(&root->fs_info->fs_mutex);
2490         trans = btrfs_start_transaction(root, 1);
2491
2492         btrfs_set_trans_block_group(trans, new_dir);
2493         path = btrfs_alloc_path();
2494         if (!path) {
2495                 ret = -ENOMEM;
2496                 goto out_fail;
2497         }
2498
2499         old_dentry->d_inode->i_nlink++;
2500         old_dir->i_ctime = old_dir->i_mtime = ctime;
2501         new_dir->i_ctime = new_dir->i_mtime = ctime;
2502         old_inode->i_ctime = ctime;
2503
2504         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2505         if (ret)
2506                 goto out_fail;
2507
2508         if (new_inode) {
2509                 new_inode->i_ctime = CURRENT_TIME;
2510                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2511                 if (ret)
2512                         goto out_fail;
2513         }
2514         ret = btrfs_add_link(trans, new_dentry, old_inode);
2515         if (ret)
2516                 goto out_fail;
2517
2518 out_fail:
2519         btrfs_free_path(path);
2520         btrfs_end_transaction(trans, root);
2521         mutex_unlock(&root->fs_info->fs_mutex);
2522         return ret;
2523 }
2524
2525 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2526                          const char *symname)
2527 {
2528         struct btrfs_trans_handle *trans;
2529         struct btrfs_root *root = BTRFS_I(dir)->root;
2530         struct btrfs_path *path;
2531         struct btrfs_key key;
2532         struct inode *inode;
2533         int err;
2534         int drop_inode = 0;
2535         u64 objectid;
2536         int name_len;
2537         int datasize;
2538         unsigned long ptr;
2539         struct btrfs_file_extent_item *ei;
2540         struct extent_buffer *leaf;
2541         unsigned long nr;
2542
2543         name_len = strlen(symname) + 1;
2544         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2545                 return -ENAMETOOLONG;
2546         mutex_lock(&root->fs_info->fs_mutex);
2547         trans = btrfs_start_transaction(root, 1);
2548         btrfs_set_trans_block_group(trans, dir);
2549
2550         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2551         if (err) {
2552                 err = -ENOSPC;
2553                 goto out_unlock;
2554         }
2555
2556         inode = btrfs_new_inode(trans, root, objectid,
2557                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2558         err = PTR_ERR(inode);
2559         if (IS_ERR(inode))
2560                 goto out_unlock;
2561
2562         btrfs_set_trans_block_group(trans, inode);
2563         err = btrfs_add_nondir(trans, dentry, inode);
2564         if (err)
2565                 drop_inode = 1;
2566         else {
2567                 inode->i_mapping->a_ops = &btrfs_aops;
2568                 inode->i_fop = &btrfs_file_operations;
2569                 inode->i_op = &btrfs_file_inode_operations;
2570                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree,
2571                                      inode->i_mapping, GFP_NOFS);
2572                 BTRFS_I(inode)->extent_tree.ops = &btrfs_extent_map_ops;
2573         }
2574         dir->i_sb->s_dirt = 1;
2575         btrfs_update_inode_block_group(trans, inode);
2576         btrfs_update_inode_block_group(trans, dir);
2577         if (drop_inode)
2578                 goto out_unlock;
2579
2580         path = btrfs_alloc_path();
2581         BUG_ON(!path);
2582         key.objectid = inode->i_ino;
2583         key.offset = 0;
2584         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2585         datasize = btrfs_file_extent_calc_inline_size(name_len);
2586         err = btrfs_insert_empty_item(trans, root, path, &key,
2587                                       datasize);
2588         if (err) {
2589                 drop_inode = 1;
2590                 goto out_unlock;
2591         }
2592         leaf = path->nodes[0];
2593         ei = btrfs_item_ptr(leaf, path->slots[0],
2594                             struct btrfs_file_extent_item);
2595         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2596         btrfs_set_file_extent_type(leaf, ei,
2597                                    BTRFS_FILE_EXTENT_INLINE);
2598         ptr = btrfs_file_extent_inline_start(ei);
2599         write_extent_buffer(leaf, symname, ptr, name_len);
2600         btrfs_mark_buffer_dirty(leaf);
2601         btrfs_free_path(path);
2602
2603         inode->i_op = &btrfs_symlink_inode_operations;
2604         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2605         inode->i_size = name_len - 1;
2606         err = btrfs_update_inode(trans, root, inode);
2607         if (err)
2608                 drop_inode = 1;
2609
2610 out_unlock:
2611         nr = trans->blocks_used;
2612         btrfs_end_transaction(trans, root);
2613         mutex_unlock(&root->fs_info->fs_mutex);
2614         if (drop_inode) {
2615                 inode_dec_link_count(inode);
2616                 iput(inode);
2617         }
2618         btrfs_btree_balance_dirty(root, nr);
2619         return err;
2620 }
2621
2622 static struct inode_operations btrfs_dir_inode_operations = {
2623         .lookup         = btrfs_lookup,
2624         .create         = btrfs_create,
2625         .unlink         = btrfs_unlink,
2626         .link           = btrfs_link,
2627         .mkdir          = btrfs_mkdir,
2628         .rmdir          = btrfs_rmdir,
2629         .rename         = btrfs_rename,
2630         .symlink        = btrfs_symlink,
2631         .setattr        = btrfs_setattr,
2632         .mknod          = btrfs_mknod,
2633         .setxattr       = generic_setxattr,
2634         .getxattr       = generic_getxattr,
2635         .listxattr      = btrfs_listxattr,
2636         .removexattr    = generic_removexattr,
2637 };
2638
2639 static struct inode_operations btrfs_dir_ro_inode_operations = {
2640         .lookup         = btrfs_lookup,
2641 };
2642
2643 static struct file_operations btrfs_dir_file_operations = {
2644         .llseek         = generic_file_llseek,
2645         .read           = generic_read_dir,
2646         .readdir        = btrfs_readdir,
2647         .unlocked_ioctl = btrfs_ioctl,
2648 #ifdef CONFIG_COMPAT
2649         .compat_ioctl   = btrfs_ioctl,
2650 #endif
2651 };
2652
2653 static struct extent_map_ops btrfs_extent_map_ops = {
2654         .fill_delalloc = run_delalloc_range,
2655         .writepage_io_hook = btrfs_writepage_io_hook,
2656         .readpage_io_hook = btrfs_readpage_io_hook,
2657         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
2658 };
2659
2660 static struct address_space_operations btrfs_aops = {
2661         .readpage       = btrfs_readpage,
2662         .writepage      = btrfs_writepage,
2663         .writepages     = btrfs_writepages,
2664         .readpages      = btrfs_readpages,
2665         .sync_page      = block_sync_page,
2666         .prepare_write  = btrfs_prepare_write,
2667         .commit_write   = btrfs_commit_write,
2668         .bmap           = btrfs_bmap,
2669         .invalidatepage = btrfs_invalidatepage,
2670         .releasepage    = btrfs_releasepage,
2671         .set_page_dirty = __set_page_dirty_nobuffers,
2672 };
2673
2674 static struct address_space_operations btrfs_symlink_aops = {
2675         .readpage       = btrfs_readpage,
2676         .writepage      = btrfs_writepage,
2677         .invalidatepage = btrfs_invalidatepage,
2678         .releasepage    = btrfs_releasepage,
2679 };
2680
2681 static struct inode_operations btrfs_file_inode_operations = {
2682         .truncate       = btrfs_truncate,
2683         .getattr        = btrfs_getattr,
2684         .setattr        = btrfs_setattr,
2685         .setxattr       = generic_setxattr,
2686         .getxattr       = generic_getxattr,
2687         .listxattr      = btrfs_listxattr,
2688         .removexattr    = generic_removexattr,
2689 };
2690
2691 static struct inode_operations btrfs_special_inode_operations = {
2692         .getattr        = btrfs_getattr,
2693         .setattr        = btrfs_setattr,
2694 };
2695
2696 static struct inode_operations btrfs_symlink_inode_operations = {
2697         .readlink       = generic_readlink,
2698         .follow_link    = page_follow_link_light,
2699         .put_link       = page_put_link,
2700 };