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