SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/bio.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/fs.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/highmem.h>
  24 #include <linux/time.h>
  25 #include <linux/init.h>
  26 #include <linux/string.h>
  27 #include <linux/smp_lock.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/mpage.h>
  30 #include <linux/swap.h>
  31 #include <linux/writeback.h>
  32 #include <linux/statfs.h>
  33 #include <linux/compat.h>
  34 #include <linux/bit_spinlock.h>
  35 #include <linux/version.h>
  36 #include <linux/xattr.h>
  37 #include "ctree.h"
  38 #include "disk-io.h"
  39 #include "transaction.h"
  40 #include "btrfs_inode.h"
  41 #include "ioctl.h"
  42 #include "print-tree.h"
  43 #include "volumes.h"
  44
  45 struct btrfs_iget_args {
  46         u64 ino;
  47         struct btrfs_root *root;
  48 };
  49
  50 static struct inode_operations btrfs_dir_inode_operations;
  51 static struct inode_operations btrfs_symlink_inode_operations;
  52 static struct inode_operations btrfs_dir_ro_inode_operations;
  53 static struct inode_operations btrfs_special_inode_operations;
  54 static struct inode_operations btrfs_file_inode_operations;
  55 static struct address_space_operations btrfs_aops;
  56 static struct address_space_operations btrfs_symlink_aops;
  57 static struct file_operations btrfs_dir_file_operations;
  58 static struct extent_io_ops btrfs_extent_io_ops;
  59
  60 static struct kmem_cache *btrfs_inode_cachep;
  61 struct kmem_cache *btrfs_trans_handle_cachep;
  62 struct kmem_cache *btrfs_transaction_cachep;
  63 struct kmem_cache *btrfs_bit_radix_cachep;
  64 struct kmem_cache *btrfs_path_cachep;
  65
  66 #define S_SHIFT 12
  67 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  68         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  69         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  70         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  71         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  72         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  73         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  74         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  75 };
  76
  77 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  78                            int for_del)
  79 {
  80         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  81         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  82         u64 thresh;
  83         int ret = 0;
  84
  85         if (for_del)
  86                 thresh = total * 90;
  87         else
  88                 thresh = total * 85;
  89
  90         do_div(thresh, 100);
  91
  92         spin_lock(&root->fs_info->delalloc_lock);
  93         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  94                 ret = -ENOSPC;
  95         spin_unlock(&root->fs_info->delalloc_lock);
  96         return ret;
  97 }
  98
  99 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 100 {
 101         struct btrfs_root *root = BTRFS_I(inode)->root;
 102         struct btrfs_trans_handle *trans;
 103         u64 alloc_hint = 0;
 104         u64 num_bytes;
 105         u64 cur_alloc_size;
 106         u64 blocksize = root->sectorsize;
 107         u64 orig_start = start;
 108         u64 orig_num_bytes;
 109         struct btrfs_key ins;
 110         int ret;
 111
 112         trans = btrfs_start_transaction(root, 1);
 113         BUG_ON(!trans);
 114         btrfs_set_trans_block_group(trans, inode);
 115
 116         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 117         num_bytes = max(blocksize,  num_bytes);
 118         ret = btrfs_drop_extents(trans, root, inode,
 119                                  start, start + num_bytes, start, &alloc_hint);
 120         orig_num_bytes = num_bytes;
 121
 122         if (alloc_hint == EXTENT_MAP_INLINE)
 123                 goto out;
 124
 125         while(num_bytes > 0) {
 126                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 127                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 128                                          root->root_key.objectid,
 129                                          trans->transid,
 130                                          inode->i_ino, start, 0,
 131                                          alloc_hint, (u64)-1, &ins, 1);
 132                 if (ret) {
 133                         WARN_ON(1);
 134                         goto out;
 135                 }
 136                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 137                                                start, ins.objectid, ins.offset,
 138                                                ins.offset);
 139                 inode->i_blocks += ins.offset >> 9;
 140                 btrfs_check_file(root, inode);
 141                 num_bytes -= cur_alloc_size;
 142                 alloc_hint = ins.objectid + ins.offset;
 143                 start += cur_alloc_size;
 144         }
 145         btrfs_drop_extent_cache(inode, orig_start,
 146                                 orig_start + orig_num_bytes - 1);
 147         btrfs_add_ordered_inode(inode);
 148         btrfs_update_inode(trans, root, inode);
 149 out:
 150         btrfs_end_transaction(trans, root);
 151         return ret;
 152 }
 153
 154 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 155 {
 156         u64 extent_start;
 157         u64 extent_end;
 158         u64 bytenr;
 159         u64 cow_end;
 160         u64 loops = 0;
 161         u64 total_fs_bytes;
 162         struct btrfs_root *root = BTRFS_I(inode)->root;
 163         struct extent_buffer *leaf;
 164         int found_type;
 165         struct btrfs_path *path;
 166         struct btrfs_file_extent_item *item;
 167         int ret;
 168         int err;
 169         struct btrfs_key found_key;
 170
 171         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 172         path = btrfs_alloc_path();
 173         BUG_ON(!path);
 174 again:
 175         ret = btrfs_lookup_file_extent(NULL, root, path,
 176                                        inode->i_ino, start, 0);
 177         if (ret < 0) {
 178                 btrfs_free_path(path);
 179                 return ret;
 180         }
 181
 182         cow_end = end;
 183         if (ret != 0) {
 184                 if (path->slots[0] == 0)
 185                         goto not_found;
 186                 path->slots[0]--;
 187         }
 188
 189         leaf = path->nodes[0];
 190         item = btrfs_item_ptr(leaf, path->slots[0],
 191                               struct btrfs_file_extent_item);
 192
 193         /* are we inside the extent that was found? */
 194         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 195         found_type = btrfs_key_type(&found_key);
 196         if (found_key.objectid != inode->i_ino ||
 197             found_type != BTRFS_EXTENT_DATA_KEY) {
 198                 goto not_found;
 199         }
 200
 201         found_type = btrfs_file_extent_type(leaf, item);
 202         extent_start = found_key.offset;
 203         if (found_type == BTRFS_FILE_EXTENT_REG) {
 204                 u64 extent_num_bytes;
 205
 206                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 207                 extent_end = extent_start + extent_num_bytes;
 208                 err = 0;
 209
 210                 if (loops && start != extent_start)
 211                         goto not_found;
 212
 213                 if (start < extent_start || start >= extent_end)
 214                         goto not_found;
 215
 216                 cow_end = min(end, extent_end - 1);
 217                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 218                 if (bytenr == 0)
 219                         goto not_found;
 220
 221                 /*
 222                  * we may be called by the resizer, make sure we're inside
 223                  * the limits of the FS
 224                  */
 225                 if (bytenr + extent_num_bytes > total_fs_bytes)
 226                         goto not_found;
 227
 228                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 229                         goto not_found;
 230                 }
 231
 232                 start = extent_end;
 233         } else {
 234                 goto not_found;
 235         }
 236 loop:
 237         if (start > end) {
 238                 btrfs_free_path(path);
 239                 return 0;
 240         }
 241         btrfs_release_path(root, path);
 242         loops++;
 243         goto again;
 244
 245 not_found:
 246         cow_file_range(inode, start, cow_end);
 247         start = cow_end + 1;
 248         goto loop;
 249 }
 250
 251 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 252 {
 253         struct btrfs_root *root = BTRFS_I(inode)->root;
 254         int ret;
 255         mutex_lock(&root->fs_info->fs_mutex);
 256         if (btrfs_test_opt(root, NODATACOW) ||
 257             btrfs_test_flag(inode, NODATACOW))
 258                 ret = run_delalloc_nocow(inode, start, end);
 259         else
 260                 ret = cow_file_range(inode, start, end);
 261
 262         mutex_unlock(&root->fs_info->fs_mutex);
 263         return ret;
 264 }
 265
 266 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 267                        unsigned long old, unsigned long bits)
 268 {
 269         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 270                 struct btrfs_root *root = BTRFS_I(inode)->root;
 271                 spin_lock(&root->fs_info->delalloc_lock);
 272                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 273                 root->fs_info->delalloc_bytes += end - start + 1;
 274                 spin_unlock(&root->fs_info->delalloc_lock);
 275         }
 276         return 0;
 277 }
 278
 279 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 280                          unsigned long old, unsigned long bits)
 281 {
 282         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 283                 struct btrfs_root *root = BTRFS_I(inode)->root;
 284                 spin_lock(&root->fs_info->delalloc_lock);
 285                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 286                         printk("warning: delalloc account %Lu %Lu\n",
 287                                end - start + 1, root->fs_info->delalloc_bytes);
 288                         root->fs_info->delalloc_bytes = 0;
 289                         BTRFS_I(inode)->delalloc_bytes = 0;
 290                 } else {
 291                         root->fs_info->delalloc_bytes -= end - start + 1;
 292                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 293                 }
 294                 spin_unlock(&root->fs_info->delalloc_lock);
 295         }
 296         return 0;
 297 }
 298
 299 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio)
 300 {
 301         struct btrfs_root *root = BTRFS_I(inode)->root;
 302         struct btrfs_trans_handle *trans;
 303         int ret = 0;
 304
 305         if (rw != WRITE) {
 306                 goto mapit;
 307         }
 308
 309         if (btrfs_test_opt(root, NODATASUM) ||
 310             btrfs_test_flag(inode, NODATASUM)) {
 311                 goto mapit;
 312         }
 313
 314         mutex_lock(&root->fs_info->fs_mutex);
 315         trans = btrfs_start_transaction(root, 1);
 316         btrfs_set_trans_block_group(trans, inode);
 317         btrfs_csum_file_blocks(trans, root, inode, bio);
 318         ret = btrfs_end_transaction(trans, root);
 319         BUG_ON(ret);
 320         mutex_unlock(&root->fs_info->fs_mutex);
 321 mapit:
 322         return btrfs_map_bio(root, rw, bio);
 323 }
 324
 325 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 326 {
 327         int ret = 0;
 328         struct inode *inode = page->mapping->host;
 329         struct btrfs_root *root = BTRFS_I(inode)->root;
 330         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 331         struct btrfs_csum_item *item;
 332         struct btrfs_path *path = NULL;
 333         u32 csum;
 334         if (btrfs_test_opt(root, NODATASUM) ||
 335             btrfs_test_flag(inode, NODATASUM))
 336                 return 0;
 337         mutex_lock(&root->fs_info->fs_mutex);
 338         path = btrfs_alloc_path();
 339         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 340         if (IS_ERR(item)) {
 341                 ret = PTR_ERR(item);
 342                 /* a csum that isn't present is a preallocated region. */
 343                 if (ret == -ENOENT || ret == -EFBIG)
 344                         ret = 0;
 345                 csum = 0;
 346                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino, start);
 347                 goto out;
 348         }
 349         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 350                            BTRFS_CRC32_SIZE);
 351         set_state_private(io_tree, start, csum);
 352 out:
 353         if (path)
 354                 btrfs_free_path(path);
 355         mutex_unlock(&root->fs_info->fs_mutex);
 356         return ret;
 357 }
 358
 359 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 360                                struct extent_state *state)
 361 {
 362         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 363         struct inode *inode = page->mapping->host;
 364         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 365         char *kaddr;
 366         u64 private = ~(u32)0;
 367         int ret;
 368         struct btrfs_root *root = BTRFS_I(inode)->root;
 369         u32 csum = ~(u32)0;
 370         unsigned long flags;
 371
 372         if (btrfs_test_opt(root, NODATASUM) ||
 373             btrfs_test_flag(inode, NODATASUM))
 374                 return 0;
 375         if (state && state->start == start) {
 376                 private = state->private;
 377                 ret = 0;
 378         } else {
 379                 ret = get_state_private(io_tree, start, &private);
 380         }
 381         local_irq_save(flags);
 382         kaddr = kmap_atomic(page, KM_IRQ0);
 383         if (ret) {
 384                 goto zeroit;
 385         }
 386         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 387         btrfs_csum_final(csum, (char *)&csum);
 388         if (csum != private) {
 389                 goto zeroit;
 390         }
 391         kunmap_atomic(kaddr, KM_IRQ0);
 392         local_irq_restore(flags);
 393         return 0;
 394
 395 zeroit:
 396         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 397                page->mapping->host->i_ino, (unsigned long long)start, csum,
 398                private);
 399         memset(kaddr + offset, 1, end - start + 1);
 400         flush_dcache_page(page);
 401         kunmap_atomic(kaddr, KM_IRQ0);
 402         local_irq_restore(flags);
 403         return 0;
 404 }
 405
 406 void btrfs_read_locked_inode(struct inode *inode)
 407 {
 408         struct btrfs_path *path;
 409         struct extent_buffer *leaf;
 410         struct btrfs_inode_item *inode_item;
 411         struct btrfs_timespec *tspec;
 412         struct btrfs_root *root = BTRFS_I(inode)->root;
 413         struct btrfs_key location;
 414         u64 alloc_group_block;
 415         u32 rdev;
 416         int ret;
 417
 418         path = btrfs_alloc_path();
 419         BUG_ON(!path);
 420         mutex_lock(&root->fs_info->fs_mutex);
 421         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 422
 423         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 424         if (ret)
 425                 goto make_bad;
 426
 427         leaf = path->nodes[0];
 428         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 429                                     struct btrfs_inode_item);
 430
 431         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 432         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 433         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 434         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 435         inode->i_size = btrfs_inode_size(leaf, inode_item);
 436
 437         tspec = btrfs_inode_atime(inode_item);
 438         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 439         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 440
 441         tspec = btrfs_inode_mtime(inode_item);
 442         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 443         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 444
 445         tspec = btrfs_inode_ctime(inode_item);
 446         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 447         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 448
 449         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 450         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 451         inode->i_rdev = 0;
 452         rdev = btrfs_inode_rdev(leaf, inode_item);
 453
 454         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 455         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 456                                                        alloc_group_block);
 457         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 458         if (!BTRFS_I(inode)->block_group) {
 459                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 460                                                  NULL, 0,
 461                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
 462         }
 463         btrfs_free_path(path);
 464         inode_item = NULL;
 465
 466         mutex_unlock(&root->fs_info->fs_mutex);
 467
 468         switch (inode->i_mode & S_IFMT) {
 469         case S_IFREG:
 470                 inode->i_mapping->a_ops = &btrfs_aops;
 471                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 472                 inode->i_fop = &btrfs_file_operations;
 473                 inode->i_op = &btrfs_file_inode_operations;
 474                 break;
 475         case S_IFDIR:
 476                 inode->i_fop = &btrfs_dir_file_operations;
 477                 if (root == root->fs_info->tree_root)
 478                         inode->i_op = &btrfs_dir_ro_inode_operations;
 479                 else
 480                         inode->i_op = &btrfs_dir_inode_operations;
 481                 break;
 482         case S_IFLNK:
 483                 inode->i_op = &btrfs_symlink_inode_operations;
 484                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 485                 break;
 486         default:
 487                 init_special_inode(inode, inode->i_mode, rdev);
 488                 break;
 489         }
 490         return;
 491
 492 make_bad:
 493         btrfs_release_path(root, path);
 494         btrfs_free_path(path);
 495         mutex_unlock(&root->fs_info->fs_mutex);
 496         make_bad_inode(inode);
 497 }
 498
 499 static void fill_inode_item(struct extent_buffer *leaf,
 500                             struct btrfs_inode_item *item,
 501                             struct inode *inode)
 502 {
 503         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 504         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 505         btrfs_set_inode_size(leaf, item, inode->i_size);
 506         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 507         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 508
 509         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 510                                inode->i_atime.tv_sec);
 511         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 512                                 inode->i_atime.tv_nsec);
 513
 514         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 515                                inode->i_mtime.tv_sec);
 516         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 517                                 inode->i_mtime.tv_nsec);
 518
 519         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 520                                inode->i_ctime.tv_sec);
 521         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 522                                 inode->i_ctime.tv_nsec);
 523
 524         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 525         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 526         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 527         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 528         btrfs_set_inode_block_group(leaf, item,
 529                                     BTRFS_I(inode)->block_group->key.objectid);
 530 }
 531
 532 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 533                               struct btrfs_root *root,
 534                               struct inode *inode)
 535 {
 536         struct btrfs_inode_item *inode_item;
 537         struct btrfs_path *path;
 538         struct extent_buffer *leaf;
 539         int ret;
 540
 541         path = btrfs_alloc_path();
 542         BUG_ON(!path);
 543         ret = btrfs_lookup_inode(trans, root, path,
 544                                  &BTRFS_I(inode)->location, 1);
 545         if (ret) {
 546                 if (ret > 0)
 547                         ret = -ENOENT;
 548                 goto failed;
 549         }
 550
 551         leaf = path->nodes[0];
 552         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 553                                   struct btrfs_inode_item);
 554
 555         fill_inode_item(leaf, inode_item, inode);
 556         btrfs_mark_buffer_dirty(leaf);
 557         btrfs_set_inode_last_trans(trans, inode);
 558         ret = 0;
 559 failed:
 560         btrfs_release_path(root, path);
 561         btrfs_free_path(path);
 562         return ret;
 563 }
 564
 565
 566 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 567                               struct btrfs_root *root,
 568                               struct inode *dir,
 569                               struct dentry *dentry)
 570 {
 571         struct btrfs_path *path;
 572         const char *name = dentry->d_name.name;
 573         int name_len = dentry->d_name.len;
 574         int ret = 0;
 575         struct extent_buffer *leaf;
 576         struct btrfs_dir_item *di;
 577         struct btrfs_key key;
 578
 579         path = btrfs_alloc_path();
 580         if (!path) {
 581                 ret = -ENOMEM;
 582                 goto err;
 583         }
 584
 585         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 586                                     name, name_len, -1);
 587         if (IS_ERR(di)) {
 588                 ret = PTR_ERR(di);
 589                 goto err;
 590         }
 591         if (!di) {
 592                 ret = -ENOENT;
 593                 goto err;
 594         }
 595         leaf = path->nodes[0];
 596         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 597         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 598         if (ret)
 599                 goto err;
 600         btrfs_release_path(root, path);
 601
 602         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 603                                          key.objectid, name, name_len, -1);
 604         if (IS_ERR(di)) {
 605                 ret = PTR_ERR(di);
 606                 goto err;
 607         }
 608         if (!di) {
 609                 ret = -ENOENT;
 610                 goto err;
 611         }
 612         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 613
 614         dentry->d_inode->i_ctime = dir->i_ctime;
 615         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 616                                   dentry->d_inode->i_ino,
 617                                   dentry->d_parent->d_inode->i_ino);
 618         if (ret) {
 619                 printk("failed to delete reference to %.*s, "
 620                        "inode %lu parent %lu\n", name_len, name,
 621                        dentry->d_inode->i_ino,
 622                        dentry->d_parent->d_inode->i_ino);
 623         }
 624 err:
 625         btrfs_free_path(path);
 626         if (!ret) {
 627                 dir->i_size -= name_len * 2;
 628                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 629                 btrfs_update_inode(trans, root, dir);
 630 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 631                 dentry->d_inode->i_nlink--;
 632 #else
 633                 drop_nlink(dentry->d_inode);
 634 #endif
 635                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 636                 dir->i_sb->s_dirt = 1;
 637         }
 638         return ret;
 639 }
 640
 641 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 642 {
 643         struct btrfs_root *root;
 644         struct btrfs_trans_handle *trans;
 645         struct inode *inode = dentry->d_inode;
 646         int ret;
 647         unsigned long nr = 0;
 648
 649         root = BTRFS_I(dir)->root;
 650         mutex_lock(&root->fs_info->fs_mutex);
 651
 652         ret = btrfs_check_free_space(root, 1, 1);
 653         if (ret)
 654                 goto fail;
 655
 656         trans = btrfs_start_transaction(root, 1);
 657
 658         btrfs_set_trans_block_group(trans, dir);
 659         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 660         nr = trans->blocks_used;
 661
 662         if (inode->i_nlink == 0) {
 663                 int found;
 664                 /* if the inode isn't linked anywhere,
 665                  * we don't need to worry about
 666                  * data=ordered
 667                  */
 668                 found = btrfs_del_ordered_inode(inode);
 669                 if (found == 1) {
 670                         atomic_dec(&inode->i_count);
 671                 }
 672         }
 673
 674         btrfs_end_transaction(trans, root);
 675 fail:
 676         mutex_unlock(&root->fs_info->fs_mutex);
 677         btrfs_btree_balance_dirty(root, nr);
 678         btrfs_throttle(root);
 679         return ret;
 680 }
 681
 682 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 683 {
 684         struct inode *inode = dentry->d_inode;
 685         int err = 0;
 686         int ret;
 687         struct btrfs_root *root = BTRFS_I(dir)->root;
 688         struct btrfs_trans_handle *trans;
 689         unsigned long nr = 0;
 690
 691         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 692                 return -ENOTEMPTY;
 693
 694         mutex_lock(&root->fs_info->fs_mutex);
 695         ret = btrfs_check_free_space(root, 1, 1);
 696         if (ret)
 697                 goto fail;
 698
 699         trans = btrfs_start_transaction(root, 1);
 700         btrfs_set_trans_block_group(trans, dir);
 701
 702         /* now the directory is empty */
 703         err = btrfs_unlink_trans(trans, root, dir, dentry);
 704         if (!err) {
 705                 inode->i_size = 0;
 706         }
 707
 708         nr = trans->blocks_used;
 709         ret = btrfs_end_transaction(trans, root);
 710 fail:
 711         mutex_unlock(&root->fs_info->fs_mutex);
 712         btrfs_btree_balance_dirty(root, nr);
 713         btrfs_throttle(root);
 714
 715         if (ret && !err)
 716                 err = ret;
 717         return err;
 718 }
 719
 720 /*
 721  * this can truncate away extent items, csum items and directory items.
 722  * It starts at a high offset and removes keys until it can't find
 723  * any higher than i_size.
 724  *
 725  * csum items that cross the new i_size are truncated to the new size
 726  * as well.
 727  */
 728 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 729                                    struct btrfs_root *root,
 730                                    struct inode *inode,
 731                                    u32 min_type)
 732 {
 733         int ret;
 734         struct btrfs_path *path;
 735         struct btrfs_key key;
 736         struct btrfs_key found_key;
 737         u32 found_type;
 738         struct extent_buffer *leaf;
 739         struct btrfs_file_extent_item *fi;
 740         u64 extent_start = 0;
 741         u64 extent_num_bytes = 0;
 742         u64 item_end = 0;
 743         u64 root_gen = 0;
 744         u64 root_owner = 0;
 745         int found_extent;
 746         int del_item;
 747         int pending_del_nr = 0;
 748         int pending_del_slot = 0;
 749         int extent_type = -1;
 750
 751         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 752         path = btrfs_alloc_path();
 753         path->reada = -1;
 754         BUG_ON(!path);
 755
 756         /* FIXME, add redo link to tree so we don't leak on crash */
 757         key.objectid = inode->i_ino;
 758         key.offset = (u64)-1;
 759         key.type = (u8)-1;
 760
 761         btrfs_init_path(path);
 762 search_again:
 763         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 764         if (ret < 0) {
 765                 goto error;
 766         }
 767         if (ret > 0) {
 768                 BUG_ON(path->slots[0] == 0);
 769                 path->slots[0]--;
 770         }
 771
 772         while(1) {
 773                 fi = NULL;
 774                 leaf = path->nodes[0];
 775                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 776                 found_type = btrfs_key_type(&found_key);
 777
 778                 if (found_key.objectid != inode->i_ino)
 779                         break;
 780
 781                 if (found_type < min_type)
 782                         break;
 783
 784                 item_end = found_key.offset;
 785                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 786                         fi = btrfs_item_ptr(leaf, path->slots[0],
 787                                             struct btrfs_file_extent_item);
 788                         extent_type = btrfs_file_extent_type(leaf, fi);
 789                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 790                                 item_end +=
 791                                     btrfs_file_extent_num_bytes(leaf, fi);
 792                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 793                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 794                                                                 path->slots[0]);
 795                                 item_end += btrfs_file_extent_inline_len(leaf,
 796                                                                          item);
 797                         }
 798                         item_end--;
 799                 }
 800                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 801                         ret = btrfs_csum_truncate(trans, root, path,
 802                                                   inode->i_size);
 803                         BUG_ON(ret);
 804                 }
 805                 if (item_end < inode->i_size) {
 806                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 807                                 found_type = BTRFS_INODE_ITEM_KEY;
 808                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 809                                 found_type = BTRFS_CSUM_ITEM_KEY;
 810                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
 811                                 found_type = BTRFS_XATTR_ITEM_KEY;
 812                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
 813                                 found_type = BTRFS_INODE_REF_KEY;
 814                         } else if (found_type) {
 815                                 found_type--;
 816                         } else {
 817                                 break;
 818                         }
 819                         btrfs_set_key_type(&key, found_type);
 820                         goto next;
 821                 }
 822                 if (found_key.offset >= inode->i_size)
 823                         del_item = 1;
 824                 else
 825                         del_item = 0;
 826                 found_extent = 0;
 827
 828                 /* FIXME, shrink the extent if the ref count is only 1 */
 829                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 830                         goto delete;
 831
 832                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 833                         u64 num_dec;
 834                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 835                         if (!del_item) {
 836                                 u64 orig_num_bytes =
 837                                         btrfs_file_extent_num_bytes(leaf, fi);
 838                                 extent_num_bytes = inode->i_size -
 839                                         found_key.offset + root->sectorsize - 1;
 840                                 extent_num_bytes = extent_num_bytes &
 841                                         ~((u64)root->sectorsize - 1);
 842                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 843                                                          extent_num_bytes);
 844                                 num_dec = (orig_num_bytes -
 845                                            extent_num_bytes);
 846                                 if (extent_start != 0)
 847                                         dec_i_blocks(inode, num_dec);
 848                                 btrfs_mark_buffer_dirty(leaf);
 849                         } else {
 850                                 extent_num_bytes =
 851                                         btrfs_file_extent_disk_num_bytes(leaf,
 852                                                                          fi);
 853                                 /* FIXME blocksize != 4096 */
 854                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
 855                                 if (extent_start != 0) {
 856                                         found_extent = 1;
 857                                         dec_i_blocks(inode, num_dec);
 858                                 }
 859                                 root_gen = btrfs_header_generation(leaf);
 860                                 root_owner = btrfs_header_owner(leaf);
 861                         }
 862                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 863                         if (!del_item) {
 864                                 u32 newsize = inode->i_size - found_key.offset;
 865                                 dec_i_blocks(inode, item_end + 1 -
 866                                             found_key.offset - newsize);
 867                                 newsize =
 868                                     btrfs_file_extent_calc_inline_size(newsize);
 869                                 ret = btrfs_truncate_item(trans, root, path,
 870                                                           newsize, 1);
 871                                 BUG_ON(ret);
 872                         } else {
 873                                 dec_i_blocks(inode, item_end + 1 -
 874                                              found_key.offset);
 875                         }
 876                 }
 877 delete:
 878                 if (del_item) {
 879                         if (!pending_del_nr) {
 880                                 /* no pending yet, add ourselves */
 881                                 pending_del_slot = path->slots[0];
 882                                 pending_del_nr = 1;
 883                         } else if (pending_del_nr &&
 884                                    path->slots[0] + 1 == pending_del_slot) {
 885                                 /* hop on the pending chunk */
 886                                 pending_del_nr++;
 887                                 pending_del_slot = path->slots[0];
 888                         } else {
 889                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
 890                         }
 891                 } else {
 892                         break;
 893                 }
 894                 if (found_extent) {
 895                         ret = btrfs_free_extent(trans, root, extent_start,
 896                                                 extent_num_bytes,
 897                                                 root_owner,
 898                                                 root_gen, inode->i_ino,
 899                                                 found_key.offset, 0);
 900                         BUG_ON(ret);
 901                 }
 902 next:
 903                 if (path->slots[0] == 0) {
 904                         if (pending_del_nr)
 905                                 goto del_pending;
 906                         btrfs_release_path(root, path);
 907                         goto search_again;
 908                 }
 909
 910                 path->slots[0]--;
 911                 if (pending_del_nr &&
 912                     path->slots[0] + 1 != pending_del_slot) {
 913                         struct btrfs_key debug;
 914 del_pending:
 915                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
 916                                               pending_del_slot);
 917                         ret = btrfs_del_items(trans, root, path,
 918                                               pending_del_slot,
 919                                               pending_del_nr);
 920                         BUG_ON(ret);
 921                         pending_del_nr = 0;
 922                         btrfs_release_path(root, path);
 923                         goto search_again;
 924                 }
 925         }
 926         ret = 0;
 927 error:
 928         if (pending_del_nr) {
 929                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
 930                                       pending_del_nr);
 931         }
 932         btrfs_release_path(root, path);
 933         btrfs_free_path(path);
 934         inode->i_sb->s_dirt = 1;
 935         return ret;
 936 }
 937
 938 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
 939                               size_t zero_start)
 940 {
 941         char *kaddr;
 942         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 943         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 944         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
 945         int ret = 0;
 946
 947         WARN_ON(!PageLocked(page));
 948         set_page_extent_mapped(page);
 949
 950         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
 951         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
 952                             page_end, GFP_NOFS);
 953
 954         if (zero_start != PAGE_CACHE_SIZE) {
 955                 kaddr = kmap(page);
 956                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
 957                 flush_dcache_page(page);
 958                 kunmap(page);
 959         }
 960         set_page_dirty(page);
 961         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
 962
 963         return ret;
 964 }
 965
 966 /*
 967  * taken from block_truncate_page, but does cow as it zeros out
 968  * any bytes left in the last page in the file.
 969  */
 970 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
 971 {
 972         struct inode *inode = mapping->host;
 973         struct btrfs_root *root = BTRFS_I(inode)->root;
 974         u32 blocksize = root->sectorsize;
 975         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 976         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 977         struct page *page;
 978         int ret = 0;
 979         u64 page_start;
 980
 981         if ((offset & (blocksize - 1)) == 0)
 982                 goto out;
 983
 984         ret = -ENOMEM;
 985         page = grab_cache_page(mapping, index);
 986         if (!page)
 987                 goto out;
 988         if (!PageUptodate(page)) {
 989                 ret = btrfs_readpage(NULL, page);
 990                 lock_page(page);
 991                 if (!PageUptodate(page)) {
 992                         ret = -EIO;
 993                         goto out;
 994                 }
 995         }
 996         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
 997
 998         ret = btrfs_cow_one_page(inode, page, offset);
 999
1000         unlock_page(page);
1001         page_cache_release(page);
1002 out:
1003         return ret;
1004 }
1005
1006 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1007 {
1008         struct inode *inode = dentry->d_inode;
1009         int err;
1010
1011         err = inode_change_ok(inode, attr);
1012         if (err)
1013                 return err;
1014
1015         if (S_ISREG(inode->i_mode) &&
1016             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1017                 struct btrfs_trans_handle *trans;
1018                 struct btrfs_root *root = BTRFS_I(inode)->root;
1019                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1020
1021                 u64 mask = root->sectorsize - 1;
1022                 u64 hole_start = (inode->i_size + mask) & ~mask;
1023                 u64 block_end = (attr->ia_size + mask) & ~mask;
1024                 u64 hole_size;
1025                 u64 alloc_hint = 0;
1026
1027                 if (attr->ia_size <= hole_start)
1028                         goto out;
1029
1030                 mutex_lock(&root->fs_info->fs_mutex);
1031                 err = btrfs_check_free_space(root, 1, 0);
1032                 mutex_unlock(&root->fs_info->fs_mutex);
1033                 if (err)
1034                         goto fail;
1035
1036                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1037
1038                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1039                 hole_size = block_end - hole_start;
1040
1041                 mutex_lock(&root->fs_info->fs_mutex);
1042                 trans = btrfs_start_transaction(root, 1);
1043                 btrfs_set_trans_block_group(trans, inode);
1044                 err = btrfs_drop_extents(trans, root, inode,
1045                                          hole_start, block_end, hole_start,
1046                                          &alloc_hint);
1047
1048                 if (alloc_hint != EXTENT_MAP_INLINE) {
1049                         err = btrfs_insert_file_extent(trans, root,
1050                                                        inode->i_ino,
1051                                                        hole_start, 0, 0,
1052                                                        hole_size);
1053                         btrfs_drop_extent_cache(inode, hole_start,
1054                                                 hole_size - 1);
1055                         btrfs_check_file(root, inode);
1056                 }
1057                 btrfs_end_transaction(trans, root);
1058                 mutex_unlock(&root->fs_info->fs_mutex);
1059                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1060                 if (err)
1061                         return err;
1062         }
1063 out:
1064         err = inode_setattr(inode, attr);
1065 fail:
1066         return err;
1067 }
1068
1069 void btrfs_put_inode(struct inode *inode)
1070 {
1071         int ret;
1072
1073         if (!BTRFS_I(inode)->ordered_trans) {
1074                 return;
1075         }
1076
1077         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1078             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1079                 return;
1080
1081         ret = btrfs_del_ordered_inode(inode);
1082         if (ret == 1) {
1083                 atomic_dec(&inode->i_count);
1084         }
1085 }
1086
1087 void btrfs_delete_inode(struct inode *inode)
1088 {
1089         struct btrfs_trans_handle *trans;
1090         struct btrfs_root *root = BTRFS_I(inode)->root;
1091         unsigned long nr;
1092         int ret;
1093
1094         truncate_inode_pages(&inode->i_data, 0);
1095         if (is_bad_inode(inode)) {
1096                 goto no_delete;
1097         }
1098
1099         inode->i_size = 0;
1100         mutex_lock(&root->fs_info->fs_mutex);
1101         trans = btrfs_start_transaction(root, 1);
1102
1103         btrfs_set_trans_block_group(trans, inode);
1104         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1105         if (ret)
1106                 goto no_delete_lock;
1107
1108         nr = trans->blocks_used;
1109         clear_inode(inode);
1110
1111         btrfs_end_transaction(trans, root);
1112         mutex_unlock(&root->fs_info->fs_mutex);
1113         btrfs_btree_balance_dirty(root, nr);
1114         btrfs_throttle(root);
1115         return;
1116
1117 no_delete_lock:
1118         nr = trans->blocks_used;
1119         btrfs_end_transaction(trans, root);
1120         mutex_unlock(&root->fs_info->fs_mutex);
1121         btrfs_btree_balance_dirty(root, nr);
1122         btrfs_throttle(root);
1123 no_delete:
1124         clear_inode(inode);
1125 }
1126
1127 /*
1128  * this returns the key found in the dir entry in the location pointer.
1129  * If no dir entries were found, location->objectid is 0.
1130  */
1131 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1132                                struct btrfs_key *location)
1133 {
1134         const char *name = dentry->d_name.name;
1135         int namelen = dentry->d_name.len;
1136         struct btrfs_dir_item *di;
1137         struct btrfs_path *path;
1138         struct btrfs_root *root = BTRFS_I(dir)->root;
1139         int ret = 0;
1140
1141         if (namelen == 1 && strcmp(name, ".") == 0) {
1142                 location->objectid = dir->i_ino;
1143                 location->type = BTRFS_INODE_ITEM_KEY;
1144                 location->offset = 0;
1145                 return 0;
1146         }
1147         path = btrfs_alloc_path();
1148         BUG_ON(!path);
1149
1150         if (namelen == 2 && strcmp(name, "..") == 0) {
1151                 struct btrfs_key key;
1152                 struct extent_buffer *leaf;
1153                 u32 nritems;
1154                 int slot;
1155
1156                 key.objectid = dir->i_ino;
1157                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1158                 key.offset = 0;
1159                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1160                 BUG_ON(ret == 0);
1161                 ret = 0;
1162
1163                 leaf = path->nodes[0];
1164                 slot = path->slots[0];
1165                 nritems = btrfs_header_nritems(leaf);
1166                 if (slot >= nritems)
1167                         goto out_err;
1168
1169                 btrfs_item_key_to_cpu(leaf, &key, slot);
1170                 if (key.objectid != dir->i_ino ||
1171                     key.type != BTRFS_INODE_REF_KEY) {
1172                         goto out_err;
1173                 }
1174                 location->objectid = key.offset;
1175                 location->type = BTRFS_INODE_ITEM_KEY;
1176                 location->offset = 0;
1177                 goto out;
1178         }
1179
1180         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1181                                     namelen, 0);
1182         if (IS_ERR(di))
1183                 ret = PTR_ERR(di);
1184         if (!di || IS_ERR(di)) {
1185                 goto out_err;
1186         }
1187         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1188 out:
1189         btrfs_free_path(path);
1190         return ret;
1191 out_err:
1192         location->objectid = 0;
1193         goto out;
1194 }
1195
1196 /*
1197  * when we hit a tree root in a directory, the btrfs part of the inode
1198  * needs to be changed to reflect the root directory of the tree root.  This
1199  * is kind of like crossing a mount point.
1200  */
1201 static int fixup_tree_root_location(struct btrfs_root *root,
1202                              struct btrfs_key *location,
1203                              struct btrfs_root **sub_root,
1204                              struct dentry *dentry)
1205 {
1206         struct btrfs_path *path;
1207         struct btrfs_root_item *ri;
1208
1209         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1210                 return 0;
1211         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1212                 return 0;
1213
1214         path = btrfs_alloc_path();
1215         BUG_ON(!path);
1216         mutex_lock(&root->fs_info->fs_mutex);
1217
1218         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1219                                         dentry->d_name.name,
1220                                         dentry->d_name.len);
1221         if (IS_ERR(*sub_root))
1222                 return PTR_ERR(*sub_root);
1223
1224         ri = &(*sub_root)->root_item;
1225         location->objectid = btrfs_root_dirid(ri);
1226         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1227         location->offset = 0;
1228
1229         btrfs_free_path(path);
1230         mutex_unlock(&root->fs_info->fs_mutex);
1231         return 0;
1232 }
1233
1234 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1235 {
1236         struct btrfs_iget_args *args = p;
1237         inode->i_ino = args->ino;
1238         BTRFS_I(inode)->root = args->root;
1239         BTRFS_I(inode)->delalloc_bytes = 0;
1240         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1241         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1242                              inode->i_mapping, GFP_NOFS);
1243         return 0;
1244 }
1245
1246 static int btrfs_find_actor(struct inode *inode, void *opaque)
1247 {
1248         struct btrfs_iget_args *args = opaque;
1249         return (args->ino == inode->i_ino &&
1250                 args->root == BTRFS_I(inode)->root);
1251 }
1252
1253 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1254                             u64 root_objectid)
1255 {
1256         struct btrfs_iget_args args;
1257         args.ino = objectid;
1258         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1259
1260         if (!args.root)
1261                 return NULL;
1262
1263         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1264 }
1265
1266 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1267                                 struct btrfs_root *root)
1268 {
1269         struct inode *inode;
1270         struct btrfs_iget_args args;
1271         args.ino = objectid;
1272         args.root = root;
1273
1274         inode = iget5_locked(s, objectid, btrfs_find_actor,
1275                              btrfs_init_locked_inode,
1276                              (void *)&args);
1277         return inode;
1278 }
1279
1280 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1281                                    struct nameidata *nd)
1282 {
1283         struct inode * inode;
1284         struct btrfs_inode *bi = BTRFS_I(dir);
1285         struct btrfs_root *root = bi->root;
1286         struct btrfs_root *sub_root = root;
1287         struct btrfs_key location;
1288         int ret;
1289
1290         if (dentry->d_name.len > BTRFS_NAME_LEN)
1291                 return ERR_PTR(-ENAMETOOLONG);
1292
1293         mutex_lock(&root->fs_info->fs_mutex);
1294         ret = btrfs_inode_by_name(dir, dentry, &location);
1295         mutex_unlock(&root->fs_info->fs_mutex);
1296
1297         if (ret < 0)
1298                 return ERR_PTR(ret);
1299
1300         inode = NULL;
1301         if (location.objectid) {
1302                 ret = fixup_tree_root_location(root, &location, &sub_root,
1303                                                 dentry);
1304                 if (ret < 0)
1305                         return ERR_PTR(ret);
1306                 if (ret > 0)
1307                         return ERR_PTR(-ENOENT);
1308                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1309                                           sub_root);
1310                 if (!inode)
1311                         return ERR_PTR(-EACCES);
1312                 if (inode->i_state & I_NEW) {
1313                         /* the inode and parent dir are two different roots */
1314                         if (sub_root != root) {
1315                                 igrab(inode);
1316                                 sub_root->inode = inode;
1317                         }
1318                         BTRFS_I(inode)->root = sub_root;
1319                         memcpy(&BTRFS_I(inode)->location, &location,
1320                                sizeof(location));
1321                         btrfs_read_locked_inode(inode);
1322                         unlock_new_inode(inode);
1323                 }
1324         }
1325         return d_splice_alias(inode, dentry);
1326 }
1327
1328 static unsigned char btrfs_filetype_table[] = {
1329         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1330 };
1331
1332 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1333 {
1334         struct inode *inode = filp->f_dentry->d_inode;
1335         struct btrfs_root *root = BTRFS_I(inode)->root;
1336         struct btrfs_item *item;
1337         struct btrfs_dir_item *di;
1338         struct btrfs_key key;
1339         struct btrfs_key found_key;
1340         struct btrfs_path *path;
1341         int ret;
1342         u32 nritems;
1343         struct extent_buffer *leaf;
1344         int slot;
1345         int advance;
1346         unsigned char d_type;
1347         int over = 0;
1348         u32 di_cur;
1349         u32 di_total;
1350         u32 di_len;
1351         int key_type = BTRFS_DIR_INDEX_KEY;
1352         char tmp_name[32];
1353         char *name_ptr;
1354         int name_len;
1355
1356         /* FIXME, use a real flag for deciding about the key type */
1357         if (root->fs_info->tree_root == root)
1358                 key_type = BTRFS_DIR_ITEM_KEY;
1359
1360         /* special case for "." */
1361         if (filp->f_pos == 0) {
1362                 over = filldir(dirent, ".", 1,
1363                                1, inode->i_ino,
1364                                DT_DIR);
1365                 if (over)
1366                         return 0;
1367                 filp->f_pos = 1;
1368         }
1369
1370         mutex_lock(&root->fs_info->fs_mutex);
1371         key.objectid = inode->i_ino;
1372         path = btrfs_alloc_path();
1373         path->reada = 2;
1374
1375         /* special case for .., just use the back ref */
1376         if (filp->f_pos == 1) {
1377                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1378                 key.offset = 0;
1379                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1380                 BUG_ON(ret == 0);
1381                 leaf = path->nodes[0];
1382                 slot = path->slots[0];
1383                 nritems = btrfs_header_nritems(leaf);
1384                 if (slot >= nritems) {
1385                         btrfs_release_path(root, path);
1386                         goto read_dir_items;
1387                 }
1388                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1389                 btrfs_release_path(root, path);
1390                 if (found_key.objectid != key.objectid ||
1391                     found_key.type != BTRFS_INODE_REF_KEY)
1392                         goto read_dir_items;
1393                 over = filldir(dirent, "..", 2,
1394                                2, found_key.offset, DT_DIR);
1395                 if (over)
1396                         goto nopos;
1397                 filp->f_pos = 2;
1398         }
1399
1400 read_dir_items:
1401         btrfs_set_key_type(&key, key_type);
1402         key.offset = filp->f_pos;
1403
1404         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1405         if (ret < 0)
1406                 goto err;
1407         advance = 0;
1408         while(1) {
1409                 leaf = path->nodes[0];
1410                 nritems = btrfs_header_nritems(leaf);
1411                 slot = path->slots[0];
1412                 if (advance || slot >= nritems) {
1413                         if (slot >= nritems -1) {
1414                                 ret = btrfs_next_leaf(root, path);
1415                                 if (ret)
1416                                         break;
1417                                 leaf = path->nodes[0];
1418                                 nritems = btrfs_header_nritems(leaf);
1419                                 slot = path->slots[0];
1420                         } else {
1421                                 slot++;
1422                                 path->slots[0]++;
1423                         }
1424                 }
1425                 advance = 1;
1426                 item = btrfs_item_nr(leaf, slot);
1427                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1428
1429                 if (found_key.objectid != key.objectid)
1430                         break;
1431                 if (btrfs_key_type(&found_key) != key_type)
1432                         break;
1433                 if (found_key.offset < filp->f_pos)
1434                         continue;
1435
1436                 filp->f_pos = found_key.offset;
1437                 advance = 1;
1438                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1439                 di_cur = 0;
1440                 di_total = btrfs_item_size(leaf, item);
1441                 while(di_cur < di_total) {
1442                         struct btrfs_key location;
1443
1444                         name_len = btrfs_dir_name_len(leaf, di);
1445                         if (name_len < 32) {
1446                                 name_ptr = tmp_name;
1447                         } else {
1448                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1449                                 BUG_ON(!name_ptr);
1450                         }
1451                         read_extent_buffer(leaf, name_ptr,
1452                                            (unsigned long)(di + 1), name_len);
1453
1454                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1455                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1456                         over = filldir(dirent, name_ptr, name_len,
1457                                        found_key.offset,
1458                                        location.objectid,
1459                                        d_type);
1460
1461                         if (name_ptr != tmp_name)
1462                                 kfree(name_ptr);
1463
1464                         if (over)
1465                                 goto nopos;
1466                         di_len = btrfs_dir_name_len(leaf, di) +
1467                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1468                         di_cur += di_len;
1469                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1470                 }
1471         }
1472         if (key_type == BTRFS_DIR_INDEX_KEY)
1473                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1474         else
1475                 filp->f_pos++;
1476 nopos:
1477         ret = 0;
1478 err:
1479         btrfs_release_path(root, path);
1480         btrfs_free_path(path);
1481         mutex_unlock(&root->fs_info->fs_mutex);
1482         return ret;
1483 }
1484
1485 int btrfs_write_inode(struct inode *inode, int wait)
1486 {
1487         struct btrfs_root *root = BTRFS_I(inode)->root;
1488         struct btrfs_trans_handle *trans;
1489         int ret = 0;
1490
1491         if (wait) {
1492                 mutex_lock(&root->fs_info->fs_mutex);
1493                 trans = btrfs_start_transaction(root, 1);
1494                 btrfs_set_trans_block_group(trans, inode);
1495                 ret = btrfs_commit_transaction(trans, root);
1496                 mutex_unlock(&root->fs_info->fs_mutex);
1497         }
1498         return ret;
1499 }
1500
1501 /*
1502  * This is somewhat expensive, updating the tree every time the
1503  * inode changes.  But, it is most likely to find the inode in cache.
1504  * FIXME, needs more benchmarking...there are no reasons other than performance
1505  * to keep or drop this code.
1506  */
1507 void btrfs_dirty_inode(struct inode *inode)
1508 {
1509         struct btrfs_root *root = BTRFS_I(inode)->root;
1510         struct btrfs_trans_handle *trans;
1511
1512         mutex_lock(&root->fs_info->fs_mutex);
1513         trans = btrfs_start_transaction(root, 1);
1514         btrfs_set_trans_block_group(trans, inode);
1515         btrfs_update_inode(trans, root, inode);
1516         btrfs_end_transaction(trans, root);
1517         mutex_unlock(&root->fs_info->fs_mutex);
1518 }
1519
1520 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1521                                      struct btrfs_root *root,
1522                                      const char *name, int name_len,
1523                                      u64 ref_objectid,
1524                                      u64 objectid,
1525                                      struct btrfs_block_group_cache *group,
1526                                      int mode)
1527 {
1528         struct inode *inode;
1529         struct btrfs_inode_item *inode_item;
1530         struct btrfs_key *location;
1531         struct btrfs_path *path;
1532         struct btrfs_inode_ref *ref;
1533         struct btrfs_key key[2];
1534         u32 sizes[2];
1535         unsigned long ptr;
1536         int ret;
1537         int owner;
1538
1539         path = btrfs_alloc_path();
1540         BUG_ON(!path);
1541
1542         inode = new_inode(root->fs_info->sb);
1543         if (!inode)
1544                 return ERR_PTR(-ENOMEM);
1545
1546         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1547         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1548                              inode->i_mapping, GFP_NOFS);
1549         BTRFS_I(inode)->delalloc_bytes = 0;
1550         BTRFS_I(inode)->root = root;
1551
1552         if (mode & S_IFDIR)
1553                 owner = 0;
1554         else
1555                 owner = 1;
1556         group = btrfs_find_block_group(root, group, 0,
1557                                        BTRFS_BLOCK_GROUP_METADATA, owner);
1558         BTRFS_I(inode)->block_group = group;
1559         BTRFS_I(inode)->flags = 0;
1560
1561         key[0].objectid = objectid;
1562         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1563         key[0].offset = 0;
1564
1565         key[1].objectid = objectid;
1566         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1567         key[1].offset = ref_objectid;
1568
1569         sizes[0] = sizeof(struct btrfs_inode_item);
1570         sizes[1] = name_len + sizeof(*ref);
1571
1572         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1573         if (ret != 0)
1574                 goto fail;
1575
1576         if (objectid > root->highest_inode)
1577                 root->highest_inode = objectid;
1578
1579         inode->i_uid = current->fsuid;
1580         inode->i_gid = current->fsgid;
1581         inode->i_mode = mode;
1582         inode->i_ino = objectid;
1583         inode->i_blocks = 0;
1584         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1585         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1586                                   struct btrfs_inode_item);
1587         fill_inode_item(path->nodes[0], inode_item, inode);
1588
1589         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1590                              struct btrfs_inode_ref);
1591         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1592         ptr = (unsigned long)(ref + 1);
1593         write_extent_buffer(path->nodes[0], name, ptr, name_len);
1594
1595         btrfs_mark_buffer_dirty(path->nodes[0]);
1596         btrfs_free_path(path);
1597
1598         location = &BTRFS_I(inode)->location;
1599         location->objectid = objectid;
1600         location->offset = 0;
1601         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1602
1603         insert_inode_hash(inode);
1604         return inode;
1605 fail:
1606         btrfs_free_path(path);
1607         return ERR_PTR(ret);
1608 }
1609
1610 static inline u8 btrfs_inode_type(struct inode *inode)
1611 {
1612         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1613 }
1614
1615 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1616                             struct dentry *dentry, struct inode *inode,
1617                             int add_backref)
1618 {
1619         int ret;
1620         struct btrfs_key key;
1621         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1622         struct inode *parent_inode;
1623
1624         key.objectid = inode->i_ino;
1625         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1626         key.offset = 0;
1627
1628         ret = btrfs_insert_dir_item(trans, root,
1629                                     dentry->d_name.name, dentry->d_name.len,
1630                                     dentry->d_parent->d_inode->i_ino,
1631                                     &key, btrfs_inode_type(inode));
1632         if (ret == 0) {
1633                 if (add_backref) {
1634                         ret = btrfs_insert_inode_ref(trans, root,
1635                                              dentry->d_name.name,
1636                                              dentry->d_name.len,
1637                                              inode->i_ino,
1638                                              dentry->d_parent->d_inode->i_ino);
1639                 }
1640                 parent_inode = dentry->d_parent->d_inode;
1641                 parent_inode->i_size += dentry->d_name.len * 2;
1642                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1643                 ret = btrfs_update_inode(trans, root,
1644                                          dentry->d_parent->d_inode);
1645         }
1646         return ret;
1647 }
1648
1649 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1650                             struct dentry *dentry, struct inode *inode,
1651                             int backref)
1652 {
1653         int err = btrfs_add_link(trans, dentry, inode, backref);
1654         if (!err) {
1655                 d_instantiate(dentry, inode);
1656                 return 0;
1657         }
1658         if (err > 0)
1659                 err = -EEXIST;
1660         return err;
1661 }
1662
1663 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1664                         int mode, dev_t rdev)
1665 {
1666         struct btrfs_trans_handle *trans;
1667         struct btrfs_root *root = BTRFS_I(dir)->root;
1668         struct inode *inode = NULL;
1669         int err;
1670         int drop_inode = 0;
1671         u64 objectid;
1672         unsigned long nr = 0;
1673
1674         if (!new_valid_dev(rdev))
1675                 return -EINVAL;
1676
1677         mutex_lock(&root->fs_info->fs_mutex);
1678         err = btrfs_check_free_space(root, 1, 0);
1679         if (err)
1680                 goto fail;
1681
1682         trans = btrfs_start_transaction(root, 1);
1683         btrfs_set_trans_block_group(trans, dir);
1684
1685         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1686         if (err) {
1687                 err = -ENOSPC;
1688                 goto out_unlock;
1689         }
1690
1691         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1692                                 dentry->d_name.len,
1693                                 dentry->d_parent->d_inode->i_ino, objectid,
1694                                 BTRFS_I(dir)->block_group, mode);
1695         err = PTR_ERR(inode);
1696         if (IS_ERR(inode))
1697                 goto out_unlock;
1698
1699         btrfs_set_trans_block_group(trans, inode);
1700         err = btrfs_add_nondir(trans, dentry, inode, 0);
1701         if (err)
1702                 drop_inode = 1;
1703         else {
1704                 inode->i_op = &btrfs_special_inode_operations;
1705                 init_special_inode(inode, inode->i_mode, rdev);
1706                 btrfs_update_inode(trans, root, inode);
1707         }
1708         dir->i_sb->s_dirt = 1;
1709         btrfs_update_inode_block_group(trans, inode);
1710         btrfs_update_inode_block_group(trans, dir);
1711 out_unlock:
1712         nr = trans->blocks_used;
1713         btrfs_end_transaction(trans, root);
1714 fail:
1715         mutex_unlock(&root->fs_info->fs_mutex);
1716
1717         if (drop_inode) {
1718                 inode_dec_link_count(inode);
1719                 iput(inode);
1720         }
1721         btrfs_btree_balance_dirty(root, nr);
1722         btrfs_throttle(root);
1723         return err;
1724 }
1725
1726 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1727                         int mode, struct nameidata *nd)
1728 {
1729         struct btrfs_trans_handle *trans;
1730         struct btrfs_root *root = BTRFS_I(dir)->root;
1731         struct inode *inode = NULL;
1732         int err;
1733         int drop_inode = 0;
1734         unsigned long nr = 0;
1735         u64 objectid;
1736
1737         mutex_lock(&root->fs_info->fs_mutex);
1738         err = btrfs_check_free_space(root, 1, 0);
1739         if (err)
1740                 goto fail;
1741         trans = btrfs_start_transaction(root, 1);
1742         btrfs_set_trans_block_group(trans, dir);
1743
1744         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1745         if (err) {
1746                 err = -ENOSPC;
1747                 goto out_unlock;
1748         }
1749
1750         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1751                                 dentry->d_name.len,
1752                                 dentry->d_parent->d_inode->i_ino,
1753                                 objectid, BTRFS_I(dir)->block_group, mode);
1754         err = PTR_ERR(inode);
1755         if (IS_ERR(inode))
1756                 goto out_unlock;
1757
1758         btrfs_set_trans_block_group(trans, inode);
1759         err = btrfs_add_nondir(trans, dentry, inode, 0);
1760         if (err)
1761                 drop_inode = 1;
1762         else {
1763                 inode->i_mapping->a_ops = &btrfs_aops;
1764                 inode->i_fop = &btrfs_file_operations;
1765                 inode->i_op = &btrfs_file_inode_operations;
1766                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1767                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1768                                      inode->i_mapping, GFP_NOFS);
1769                 BTRFS_I(inode)->delalloc_bytes = 0;
1770                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1771         }
1772         dir->i_sb->s_dirt = 1;
1773         btrfs_update_inode_block_group(trans, inode);
1774         btrfs_update_inode_block_group(trans, dir);
1775 out_unlock:
1776         nr = trans->blocks_used;
1777         btrfs_end_transaction(trans, root);
1778 fail:
1779         mutex_unlock(&root->fs_info->fs_mutex);
1780
1781         if (drop_inode) {
1782                 inode_dec_link_count(inode);
1783                 iput(inode);
1784         }
1785         btrfs_btree_balance_dirty(root, nr);
1786         btrfs_throttle(root);
1787         return err;
1788 }
1789
1790 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1791                       struct dentry *dentry)
1792 {
1793         struct btrfs_trans_handle *trans;
1794         struct btrfs_root *root = BTRFS_I(dir)->root;
1795         struct inode *inode = old_dentry->d_inode;
1796         unsigned long nr = 0;
1797         int err;
1798         int drop_inode = 0;
1799
1800         if (inode->i_nlink == 0)
1801                 return -ENOENT;
1802
1803 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1804         inode->i_nlink++;
1805 #else
1806         inc_nlink(inode);
1807 #endif
1808         mutex_lock(&root->fs_info->fs_mutex);
1809         err = btrfs_check_free_space(root, 1, 0);
1810         if (err)
1811                 goto fail;
1812         trans = btrfs_start_transaction(root, 1);
1813
1814         btrfs_set_trans_block_group(trans, dir);
1815         atomic_inc(&inode->i_count);
1816         err = btrfs_add_nondir(trans, dentry, inode, 1);
1817
1818         if (err)
1819                 drop_inode = 1;
1820
1821         dir->i_sb->s_dirt = 1;
1822         btrfs_update_inode_block_group(trans, dir);
1823         err = btrfs_update_inode(trans, root, inode);
1824
1825         if (err)
1826                 drop_inode = 1;
1827
1828         nr = trans->blocks_used;
1829         btrfs_end_transaction(trans, root);
1830 fail:
1831         mutex_unlock(&root->fs_info->fs_mutex);
1832
1833         if (drop_inode) {
1834                 inode_dec_link_count(inode);
1835                 iput(inode);
1836         }
1837         btrfs_btree_balance_dirty(root, nr);
1838         btrfs_throttle(root);
1839         return err;
1840 }
1841
1842 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1843 {
1844         struct inode *inode;
1845         struct btrfs_trans_handle *trans;
1846         struct btrfs_root *root = BTRFS_I(dir)->root;
1847         int err = 0;
1848         int drop_on_err = 0;
1849         u64 objectid;
1850         unsigned long nr = 1;
1851
1852         mutex_lock(&root->fs_info->fs_mutex);
1853         err = btrfs_check_free_space(root, 1, 0);
1854         if (err)
1855                 goto out_unlock;
1856
1857         trans = btrfs_start_transaction(root, 1);
1858         btrfs_set_trans_block_group(trans, dir);
1859
1860         if (IS_ERR(trans)) {
1861                 err = PTR_ERR(trans);
1862                 goto out_unlock;
1863         }
1864
1865         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1866         if (err) {
1867                 err = -ENOSPC;
1868                 goto out_unlock;
1869         }
1870
1871         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1872                                 dentry->d_name.len,
1873                                 dentry->d_parent->d_inode->i_ino, objectid,
1874                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
1875         if (IS_ERR(inode)) {
1876                 err = PTR_ERR(inode);
1877                 goto out_fail;
1878         }
1879
1880         drop_on_err = 1;
1881         inode->i_op = &btrfs_dir_inode_operations;
1882         inode->i_fop = &btrfs_dir_file_operations;
1883         btrfs_set_trans_block_group(trans, inode);
1884
1885         inode->i_size = 0;
1886         err = btrfs_update_inode(trans, root, inode);
1887         if (err)
1888                 goto out_fail;
1889
1890         err = btrfs_add_link(trans, dentry, inode, 0);
1891         if (err)
1892                 goto out_fail;
1893
1894         d_instantiate(dentry, inode);
1895         drop_on_err = 0;
1896         dir->i_sb->s_dirt = 1;
1897         btrfs_update_inode_block_group(trans, inode);
1898         btrfs_update_inode_block_group(trans, dir);
1899
1900 out_fail:
1901         nr = trans->blocks_used;
1902         btrfs_end_transaction(trans, root);
1903
1904 out_unlock:
1905         mutex_unlock(&root->fs_info->fs_mutex);
1906         if (drop_on_err)
1907                 iput(inode);
1908         btrfs_btree_balance_dirty(root, nr);
1909         btrfs_throttle(root);
1910         return err;
1911 }
1912
1913 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
1914                                     size_t pg_offset, u64 start, u64 len,
1915                                     int create)
1916 {
1917         int ret;
1918         int err = 0;
1919         u64 bytenr;
1920         u64 extent_start = 0;
1921         u64 extent_end = 0;
1922         u64 objectid = inode->i_ino;
1923         u32 found_type;
1924         struct btrfs_path *path;
1925         struct btrfs_root *root = BTRFS_I(inode)->root;
1926         struct btrfs_file_extent_item *item;
1927         struct extent_buffer *leaf;
1928         struct btrfs_key found_key;
1929         struct extent_map *em = NULL;
1930         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1931         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1932         struct btrfs_trans_handle *trans = NULL;
1933
1934         path = btrfs_alloc_path();
1935         BUG_ON(!path);
1936         mutex_lock(&root->fs_info->fs_mutex);
1937
1938 again:
1939         spin_lock(&em_tree->lock);
1940         em = lookup_extent_mapping(em_tree, start, len);
1941         spin_unlock(&em_tree->lock);
1942
1943         if (em) {
1944                 if (em->start > start) {
1945                         printk("get_extent lookup [%Lu %Lu] em [%Lu %Lu]\n",
1946                                start, len, em->start, em->len);
1947                         WARN_ON(1);
1948                 }
1949                 if (em->block_start == EXTENT_MAP_INLINE && page)
1950                         free_extent_map(em);
1951                 else
1952                         goto out;
1953         }
1954         em = alloc_extent_map(GFP_NOFS);
1955         if (!em) {
1956                 err = -ENOMEM;
1957                 goto out;
1958         }
1959
1960         em->start = EXTENT_MAP_HOLE;
1961         em->len = (u64)-1;
1962         em->bdev = inode->i_sb->s_bdev;
1963         ret = btrfs_lookup_file_extent(trans, root, path,
1964                                        objectid, start, trans != NULL);
1965         if (ret < 0) {
1966                 err = ret;
1967                 goto out;
1968         }
1969
1970         if (ret != 0) {
1971                 if (path->slots[0] == 0)
1972                         goto not_found;
1973                 path->slots[0]--;
1974         }
1975
1976         leaf = path->nodes[0];
1977         item = btrfs_item_ptr(leaf, path->slots[0],
1978                               struct btrfs_file_extent_item);
1979         /* are we inside the extent that was found? */
1980         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1981         found_type = btrfs_key_type(&found_key);
1982         if (found_key.objectid != objectid ||
1983             found_type != BTRFS_EXTENT_DATA_KEY) {
1984                 goto not_found;
1985         }
1986
1987         found_type = btrfs_file_extent_type(leaf, item);
1988         extent_start = found_key.offset;
1989         if (found_type == BTRFS_FILE_EXTENT_REG) {
1990                 extent_end = extent_start +
1991                        btrfs_file_extent_num_bytes(leaf, item);
1992                 err = 0;
1993                 if (start < extent_start || start >= extent_end) {
1994                         em->start = start;
1995                         if (start < extent_start) {
1996                                 if (start + len <= extent_start)
1997                                         goto not_found;
1998                                 em->len = extent_end - extent_start;
1999                         } else {
2000                                 em->len = len;
2001                         }
2002                         goto not_found_em;
2003                 }
2004                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2005                 if (bytenr == 0) {
2006                         em->start = extent_start;
2007                         em->len = extent_end - extent_start;
2008                         em->block_start = EXTENT_MAP_HOLE;
2009                         goto insert;
2010                 }
2011                 bytenr += btrfs_file_extent_offset(leaf, item);
2012                 em->block_start = bytenr;
2013                 em->start = extent_start;
2014                 em->len = extent_end - extent_start;
2015                 goto insert;
2016         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2017                 u64 page_start;
2018                 unsigned long ptr;
2019                 char *map;
2020                 size_t size;
2021                 size_t extent_offset;
2022                 size_t copy_size;
2023
2024                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2025                                                     path->slots[0]));
2026                 extent_end = (extent_start + size + root->sectorsize - 1) &
2027                         ~((u64)root->sectorsize - 1);
2028                 if (start < extent_start || start >= extent_end) {
2029                         em->start = start;
2030                         if (start < extent_start) {
2031                                 if (start + len <= extent_start)
2032                                         goto not_found;
2033                                 em->len = extent_end - extent_start;
2034                         } else {
2035                                 em->len = len;
2036                         }
2037                         goto not_found_em;
2038                 }
2039                 em->block_start = EXTENT_MAP_INLINE;
2040
2041                 if (!page) {
2042                         em->start = extent_start;
2043                         em->len = size;
2044                         goto out;
2045                 }
2046
2047                 page_start = page_offset(page) + pg_offset;
2048                 extent_offset = page_start - extent_start;
2049                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2050                                 size - extent_offset);
2051                 em->start = extent_start + extent_offset;
2052                 em->len = (copy_size + root->sectorsize - 1) &
2053                         ~((u64)root->sectorsize - 1);
2054                 map = kmap(page);
2055                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2056                 if (create == 0 && !PageUptodate(page)) {
2057                         read_extent_buffer(leaf, map + pg_offset, ptr,
2058                                            copy_size);
2059                         flush_dcache_page(page);
2060                 } else if (create && PageUptodate(page)) {
2061                         if (!trans) {
2062                                 kunmap(page);
2063                                 free_extent_map(em);
2064                                 em = NULL;
2065                                 btrfs_release_path(root, path);
2066                                 trans = btrfs_start_transaction(root, 1);
2067                                 goto again;
2068                         }
2069                         write_extent_buffer(leaf, map + pg_offset, ptr,
2070                                             copy_size);
2071                         btrfs_mark_buffer_dirty(leaf);
2072                 }
2073                 kunmap(page);
2074                 set_extent_uptodate(io_tree, em->start,
2075                                     extent_map_end(em) - 1, GFP_NOFS);
2076                 goto insert;
2077         } else {
2078                 printk("unkknown found_type %d\n", found_type);
2079                 WARN_ON(1);
2080         }
2081 not_found:
2082         em->start = start;
2083         em->len = len;
2084 not_found_em:
2085         em->block_start = EXTENT_MAP_HOLE;
2086 insert:
2087         btrfs_release_path(root, path);
2088         if (em->start > start || extent_map_end(em) <= start) {
2089                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2090                 err = -EIO;
2091                 goto out;
2092         }
2093
2094         err = 0;
2095         spin_lock(&em_tree->lock);
2096         ret = add_extent_mapping(em_tree, em);
2097         if (ret == -EEXIST) {
2098                 free_extent_map(em);
2099                 em = lookup_extent_mapping(em_tree, start, len);
2100                 if (!em) {
2101                         err = -EIO;
2102                         printk("failing to insert %Lu %Lu\n", start, len);
2103                 }
2104         }
2105         spin_unlock(&em_tree->lock);
2106 out:
2107         btrfs_free_path(path);
2108         if (trans) {
2109                 ret = btrfs_end_transaction(trans, root);
2110                 if (!err)
2111                         err = ret;
2112         }
2113         mutex_unlock(&root->fs_info->fs_mutex);
2114         if (err) {
2115                 free_extent_map(em);
2116                 WARN_ON(1);
2117                 return ERR_PTR(err);
2118         }
2119         return em;
2120 }
2121
2122 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2123 {
2124         return extent_bmap(mapping, iblock, btrfs_get_extent);
2125 }
2126
2127 int btrfs_readpage(struct file *file, struct page *page)
2128 {
2129         struct extent_io_tree *tree;
2130         tree = &BTRFS_I(page->mapping->host)->io_tree;
2131         return extent_read_full_page(tree, page, btrfs_get_extent);
2132 }
2133
2134 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2135 {
2136         struct extent_io_tree *tree;
2137
2138
2139         if (current->flags & PF_MEMALLOC) {
2140                 redirty_page_for_writepage(wbc, page);
2141                 unlock_page(page);
2142                 return 0;
2143         }
2144         tree = &BTRFS_I(page->mapping->host)->io_tree;
2145         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2146 }
2147
2148 static int btrfs_writepages(struct address_space *mapping,
2149                             struct writeback_control *wbc)
2150 {
2151         struct extent_io_tree *tree;
2152         tree = &BTRFS_I(mapping->host)->io_tree;
2153         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2154 }
2155
2156 static int
2157 btrfs_readpages(struct file *file, struct address_space *mapping,
2158                 struct list_head *pages, unsigned nr_pages)
2159 {
2160         struct extent_io_tree *tree;
2161         tree = &BTRFS_I(mapping->host)->io_tree;
2162         return extent_readpages(tree, mapping, pages, nr_pages,
2163                                 btrfs_get_extent);
2164 }
2165
2166 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2167 {
2168         struct extent_io_tree *tree;
2169         struct extent_map_tree *map;
2170         int ret;
2171
2172         tree = &BTRFS_I(page->mapping->host)->io_tree;
2173         map = &BTRFS_I(page->mapping->host)->extent_tree;
2174         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2175         if (ret == 1) {
2176                 ClearPagePrivate(page);
2177                 set_page_private(page, 0);
2178                 page_cache_release(page);
2179         }
2180         return ret;
2181 }
2182
2183 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2184 {
2185         struct extent_io_tree *tree;
2186
2187         tree = &BTRFS_I(page->mapping->host)->io_tree;
2188         extent_invalidatepage(tree, page, offset);
2189         btrfs_releasepage(page, GFP_NOFS);
2190 }
2191
2192 /*
2193  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2194  * called from a page fault handler when a page is first dirtied. Hence we must
2195  * be careful to check for EOF conditions here. We set the page up correctly
2196  * for a written page which means we get ENOSPC checking when writing into
2197  * holes and correct delalloc and unwritten extent mapping on filesystems that
2198  * support these features.
2199  *
2200  * We are not allowed to take the i_mutex here so we have to play games to
2201  * protect against truncate races as the page could now be beyond EOF.  Because
2202  * vmtruncate() writes the inode size before removing pages, once we have the
2203  * page lock we can determine safely if the page is beyond EOF. If it is not
2204  * beyond EOF, then the page is guaranteed safe against truncation until we
2205  * unlock the page.
2206  */
2207 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2208 {
2209         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2210         struct btrfs_root *root = BTRFS_I(inode)->root;
2211         unsigned long end;
2212         loff_t size;
2213         int ret;
2214         u64 page_start;
2215
2216         mutex_lock(&root->fs_info->fs_mutex);
2217         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2218         mutex_unlock(&root->fs_info->fs_mutex);
2219         if (ret)
2220                 goto out;
2221
2222         ret = -EINVAL;
2223
2224         lock_page(page);
2225         wait_on_page_writeback(page);
2226         size = i_size_read(inode);
2227         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2228
2229         if ((page->mapping != inode->i_mapping) ||
2230             (page_start > size)) {
2231                 /* page got truncated out from underneath us */
2232                 goto out_unlock;
2233         }
2234
2235         /* page is wholly or partially inside EOF */
2236         if (page_start + PAGE_CACHE_SIZE > size)
2237                 end = size & ~PAGE_CACHE_MASK;
2238         else
2239                 end = PAGE_CACHE_SIZE;
2240
2241         ret = btrfs_cow_one_page(inode, page, end);
2242
2243 out_unlock:
2244         unlock_page(page);
2245 out:
2246         return ret;
2247 }
2248
2249 static void btrfs_truncate(struct inode *inode)
2250 {
2251         struct btrfs_root *root = BTRFS_I(inode)->root;
2252         int ret;
2253         struct btrfs_trans_handle *trans;
2254         unsigned long nr;
2255
2256         if (!S_ISREG(inode->i_mode))
2257                 return;
2258         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2259                 return;
2260
2261         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2262
2263         mutex_lock(&root->fs_info->fs_mutex);
2264         trans = btrfs_start_transaction(root, 1);
2265         btrfs_set_trans_block_group(trans, inode);
2266
2267         /* FIXME, add redo link to tree so we don't leak on crash */
2268         ret = btrfs_truncate_in_trans(trans, root, inode,
2269                                       BTRFS_EXTENT_DATA_KEY);
2270         btrfs_update_inode(trans, root, inode);
2271         nr = trans->blocks_used;
2272
2273         ret = btrfs_end_transaction(trans, root);
2274         BUG_ON(ret);
2275         mutex_unlock(&root->fs_info->fs_mutex);
2276         btrfs_btree_balance_dirty(root, nr);
2277         btrfs_throttle(root);
2278 }
2279
2280 static int noinline create_subvol(struct btrfs_root *root, char *name,
2281                                   int namelen)
2282 {
2283         struct btrfs_trans_handle *trans;
2284         struct btrfs_key key;
2285         struct btrfs_root_item root_item;
2286         struct btrfs_inode_item *inode_item;
2287         struct extent_buffer *leaf;
2288         struct btrfs_root *new_root = root;
2289         struct inode *inode;
2290         struct inode *dir;
2291         int ret;
2292         int err;
2293         u64 objectid;
2294         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2295         unsigned long nr = 1;
2296
2297         mutex_lock(&root->fs_info->fs_mutex);
2298         ret = btrfs_check_free_space(root, 1, 0);
2299         if (ret)
2300                 goto fail_commit;
2301
2302         trans = btrfs_start_transaction(root, 1);
2303         BUG_ON(!trans);
2304
2305         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2306                                        0, &objectid);
2307         if (ret)
2308                 goto fail;
2309
2310         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2311                                         objectid, trans->transid, 0, 0,
2312                                         0, 0);
2313         if (IS_ERR(leaf))
2314                 return PTR_ERR(leaf);
2315
2316         btrfs_set_header_nritems(leaf, 0);
2317         btrfs_set_header_level(leaf, 0);
2318         btrfs_set_header_bytenr(leaf, leaf->start);
2319         btrfs_set_header_generation(leaf, trans->transid);
2320         btrfs_set_header_owner(leaf, objectid);
2321
2322         write_extent_buffer(leaf, root->fs_info->fsid,
2323                             (unsigned long)btrfs_header_fsid(leaf),
2324                             BTRFS_FSID_SIZE);
2325         btrfs_mark_buffer_dirty(leaf);
2326
2327         inode_item = &root_item.inode;
2328         memset(inode_item, 0, sizeof(*inode_item));
2329         inode_item->generation = cpu_to_le64(1);
2330         inode_item->size = cpu_to_le64(3);
2331         inode_item->nlink = cpu_to_le32(1);
2332         inode_item->nblocks = cpu_to_le64(1);
2333         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2334
2335         btrfs_set_root_bytenr(&root_item, leaf->start);
2336         btrfs_set_root_level(&root_item, 0);
2337         btrfs_set_root_refs(&root_item, 1);
2338         btrfs_set_root_used(&root_item, 0);
2339
2340         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2341         root_item.drop_level = 0;
2342
2343         free_extent_buffer(leaf);
2344         leaf = NULL;
2345
2346         btrfs_set_root_dirid(&root_item, new_dirid);
2347
2348         key.objectid = objectid;
2349         key.offset = 1;
2350         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2351         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2352                                 &root_item);
2353         if (ret)
2354                 goto fail;
2355
2356         /*
2357          * insert the directory item
2358          */
2359         key.offset = (u64)-1;
2360         dir = root->fs_info->sb->s_root->d_inode;
2361         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2362                                     name, namelen, dir->i_ino, &key,
2363                                     BTRFS_FT_DIR);
2364         if (ret)
2365                 goto fail;
2366
2367         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2368                              name, namelen, objectid,
2369                              root->fs_info->sb->s_root->d_inode->i_ino);
2370         if (ret)
2371                 goto fail;
2372
2373         ret = btrfs_commit_transaction(trans, root);
2374         if (ret)
2375                 goto fail_commit;
2376
2377         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2378         BUG_ON(!new_root);
2379
2380         trans = btrfs_start_transaction(new_root, 1);
2381         BUG_ON(!trans);
2382
2383         inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2384                                 new_dirid,
2385                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2386         if (IS_ERR(inode))
2387                 goto fail;
2388         inode->i_op = &btrfs_dir_inode_operations;
2389         inode->i_fop = &btrfs_dir_file_operations;
2390         new_root->inode = inode;
2391
2392         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2393                                      new_dirid);
2394         inode->i_nlink = 1;
2395         inode->i_size = 0;
2396         ret = btrfs_update_inode(trans, new_root, inode);
2397         if (ret)
2398                 goto fail;
2399 fail:
2400         nr = trans->blocks_used;
2401         err = btrfs_commit_transaction(trans, new_root);
2402         if (err && !ret)
2403                 ret = err;
2404 fail_commit:
2405         mutex_unlock(&root->fs_info->fs_mutex);
2406         btrfs_btree_balance_dirty(root, nr);
2407         btrfs_throttle(root);
2408         return ret;
2409 }
2410
2411 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2412 {
2413         struct btrfs_pending_snapshot *pending_snapshot;
2414         struct btrfs_trans_handle *trans;
2415         int ret;
2416         int err;
2417         unsigned long nr = 0;
2418
2419         if (!root->ref_cows)
2420                 return -EINVAL;
2421
2422         mutex_lock(&root->fs_info->fs_mutex);
2423         ret = btrfs_check_free_space(root, 1, 0);
2424         if (ret)
2425                 goto fail_unlock;
2426
2427         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2428         if (!pending_snapshot) {
2429                 ret = -ENOMEM;
2430                 goto fail_unlock;
2431         }
2432         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2433         if (!pending_snapshot->name) {
2434                 ret = -ENOMEM;
2435                 kfree(pending_snapshot);
2436                 goto fail_unlock;
2437         }
2438         memcpy(pending_snapshot->name, name, namelen);
2439         pending_snapshot->name[namelen] = '\0';
2440         trans = btrfs_start_transaction(root, 1);
2441         BUG_ON(!trans);
2442         pending_snapshot->root = root;
2443         list_add(&pending_snapshot->list,
2444                  &trans->transaction->pending_snapshots);
2445         ret = btrfs_update_inode(trans, root, root->inode);
2446         err = btrfs_commit_transaction(trans, root);
2447
2448 fail_unlock:
2449         mutex_unlock(&root->fs_info->fs_mutex);
2450         btrfs_btree_balance_dirty(root, nr);
2451         btrfs_throttle(root);
2452         return ret;
2453 }
2454
2455 unsigned long btrfs_force_ra(struct address_space *mapping,
2456                               struct file_ra_state *ra, struct file *file,
2457                               pgoff_t offset, pgoff_t last_index)
2458 {
2459         pgoff_t req_size;
2460
2461 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2462         req_size = last_index - offset + 1;
2463         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2464         return offset;
2465 #else
2466         req_size = min(last_index - offset + 1, (pgoff_t)128);
2467         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2468         return offset + req_size;
2469 #endif
2470 }
2471
2472 int btrfs_defrag_file(struct file *file) {
2473         struct inode *inode = fdentry(file)->d_inode;
2474         struct btrfs_root *root = BTRFS_I(inode)->root;
2475         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2476         struct page *page;
2477         unsigned long last_index;
2478         unsigned long ra_index = 0;
2479         u64 page_start;
2480         u64 page_end;
2481         unsigned long i;
2482         int ret;
2483
2484         mutex_lock(&root->fs_info->fs_mutex);
2485         ret = btrfs_check_free_space(root, inode->i_size, 0);
2486         mutex_unlock(&root->fs_info->fs_mutex);
2487         if (ret)
2488                 return -ENOSPC;
2489
2490         mutex_lock(&inode->i_mutex);
2491         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2492         for (i = 0; i <= last_index; i++) {
2493                 if (i == ra_index) {
2494                         ra_index = btrfs_force_ra(inode->i_mapping,
2495                                                   &file->f_ra,
2496                                                   file, ra_index, last_index);
2497                 }
2498                 page = grab_cache_page(inode->i_mapping, i);
2499                 if (!page)
2500                         goto out_unlock;
2501                 if (!PageUptodate(page)) {
2502                         btrfs_readpage(NULL, page);
2503                         lock_page(page);
2504                         if (!PageUptodate(page)) {
2505                                 unlock_page(page);
2506                                 page_cache_release(page);
2507                                 goto out_unlock;
2508                         }
2509                 }
2510                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2511                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2512
2513                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2514                 set_extent_delalloc(io_tree, page_start,
2515                                     page_end, GFP_NOFS);
2516
2517                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2518                 set_page_dirty(page);
2519                 unlock_page(page);
2520                 page_cache_release(page);
2521                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2522         }
2523
2524 out_unlock:
2525         mutex_unlock(&inode->i_mutex);
2526         return 0;
2527 }
2528
2529 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2530 {
2531         u64 new_size;
2532         u64 old_size;
2533         struct btrfs_ioctl_vol_args *vol_args;
2534         struct btrfs_trans_handle *trans;
2535         char *sizestr;
2536         int ret = 0;
2537         int namelen;
2538         int mod = 0;
2539
2540         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2541
2542         if (!vol_args)
2543                 return -ENOMEM;
2544
2545         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2546                 ret = -EFAULT;
2547                 goto out;
2548         }
2549         namelen = strlen(vol_args->name);
2550         if (namelen > BTRFS_VOL_NAME_MAX) {
2551                 ret = -EINVAL;
2552                 goto out;
2553         }
2554
2555         sizestr = vol_args->name;
2556         if (!strcmp(sizestr, "max"))
2557                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2558         else {
2559                 if (sizestr[0] == '-') {
2560                         mod = -1;
2561                         sizestr++;
2562                 } else if (sizestr[0] == '+') {
2563                         mod = 1;
2564                         sizestr++;
2565                 }
2566                 new_size = btrfs_parse_size(sizestr);
2567                 if (new_size == 0) {
2568                         ret = -EINVAL;
2569                         goto out;
2570                 }
2571         }
2572
2573         mutex_lock(&root->fs_info->fs_mutex);
2574         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2575
2576         if (mod < 0) {
2577                 if (new_size > old_size) {
2578                         ret = -EINVAL;
2579                         goto out_unlock;
2580                 }
2581                 new_size = old_size - new_size;
2582         } else if (mod > 0) {
2583                 new_size = old_size + new_size;
2584         }
2585
2586         if (new_size < 256 * 1024 * 1024) {
2587                 ret = -EINVAL;
2588                 goto out_unlock;
2589         }
2590         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2591                 ret = -EFBIG;
2592                 goto out_unlock;
2593         }
2594
2595         do_div(new_size, root->sectorsize);
2596         new_size *= root->sectorsize;
2597
2598 printk("new size is %Lu\n", new_size);
2599         if (new_size > old_size) {
2600                 trans = btrfs_start_transaction(root, 1);
2601                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2602                 btrfs_commit_transaction(trans, root);
2603         } else {
2604                 ret = btrfs_shrink_extent_tree(root, new_size);
2605         }
2606
2607 out_unlock:
2608         mutex_unlock(&root->fs_info->fs_mutex);
2609 out:
2610         kfree(vol_args);
2611         return ret;
2612 }
2613
2614 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2615                                             void __user *arg)
2616 {
2617         struct btrfs_ioctl_vol_args *vol_args;
2618         struct btrfs_dir_item *di;
2619         struct btrfs_path *path;
2620         u64 root_dirid;
2621         int namelen;
2622         int ret;
2623
2624         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2625
2626         if (!vol_args)
2627                 return -ENOMEM;
2628
2629         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2630                 ret = -EFAULT;
2631                 goto out;
2632         }
2633
2634         namelen = strlen(vol_args->name);
2635         if (namelen > BTRFS_VOL_NAME_MAX) {
2636                 ret = -EINVAL;
2637                 goto out;
2638         }
2639         if (strchr(vol_args->name, '/')) {
2640                 ret = -EINVAL;
2641                 goto out;
2642         }
2643
2644         path = btrfs_alloc_path();
2645         if (!path) {
2646                 ret = -ENOMEM;
2647                 goto out;
2648         }
2649
2650         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2651         mutex_lock(&root->fs_info->fs_mutex);
2652         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2653                             path, root_dirid,
2654                             vol_args->name, namelen, 0);
2655         mutex_unlock(&root->fs_info->fs_mutex);
2656         btrfs_free_path(path);
2657
2658         if (di && !IS_ERR(di)) {
2659                 ret = -EEXIST;
2660                 goto out;
2661         }
2662
2663         if (IS_ERR(di)) {
2664                 ret = PTR_ERR(di);
2665                 goto out;
2666         }
2667
2668         if (root == root->fs_info->tree_root)
2669                 ret = create_subvol(root, vol_args->name, namelen);
2670         else
2671                 ret = create_snapshot(root, vol_args->name, namelen);
2672 out:
2673         kfree(vol_args);
2674         return ret;
2675 }
2676
2677 static int btrfs_ioctl_defrag(struct file *file)
2678 {
2679         struct inode *inode = fdentry(file)->d_inode;
2680         struct btrfs_root *root = BTRFS_I(inode)->root;
2681
2682         switch (inode->i_mode & S_IFMT) {
2683         case S_IFDIR:
2684                 mutex_lock(&root->fs_info->fs_mutex);
2685                 btrfs_defrag_root(root, 0);
2686                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2687                 mutex_unlock(&root->fs_info->fs_mutex);
2688                 break;
2689         case S_IFREG:
2690                 btrfs_defrag_file(file);
2691                 break;
2692         }
2693
2694         return 0;
2695 }
2696
2697 long btrfs_ioctl(struct file *file, unsigned int
2698                 cmd, unsigned long arg)
2699 {
2700         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2701
2702         switch (cmd) {
2703         case BTRFS_IOC_SNAP_CREATE:
2704                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2705         case BTRFS_IOC_DEFRAG:
2706                 return btrfs_ioctl_defrag(file);
2707         case BTRFS_IOC_RESIZE:
2708                 return btrfs_ioctl_resize(root, (void __user *)arg);
2709         }
2710
2711         return -ENOTTY;
2712 }
2713
2714 /*
2715  * Called inside transaction, so use GFP_NOFS
2716  */
2717 struct inode *btrfs_alloc_inode(struct super_block *sb)
2718 {
2719         struct btrfs_inode *ei;
2720
2721         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2722         if (!ei)
2723                 return NULL;
2724         ei->last_trans = 0;
2725         ei->ordered_trans = 0;
2726         return &ei->vfs_inode;
2727 }
2728
2729 void btrfs_destroy_inode(struct inode *inode)
2730 {
2731         WARN_ON(!list_empty(&inode->i_dentry));
2732         WARN_ON(inode->i_data.nrpages);
2733
2734         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2735         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2736 }
2737
2738 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2739 static void init_once(struct kmem_cache * cachep, void *foo)
2740 #else
2741 static void init_once(void * foo, struct kmem_cache * cachep,
2742                       unsigned long flags)
2743 #endif
2744 {
2745         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2746
2747         inode_init_once(&ei->vfs_inode);
2748 }
2749
2750 void btrfs_destroy_cachep(void)
2751 {
2752         if (btrfs_inode_cachep)
2753                 kmem_cache_destroy(btrfs_inode_cachep);
2754         if (btrfs_trans_handle_cachep)
2755                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2756         if (btrfs_transaction_cachep)
2757                 kmem_cache_destroy(btrfs_transaction_cachep);
2758         if (btrfs_bit_radix_cachep)
2759                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2760         if (btrfs_path_cachep)
2761                 kmem_cache_destroy(btrfs_path_cachep);
2762 }
2763
2764 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2765                                        unsigned long extra_flags,
2766 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2767                                        void (*ctor)(struct kmem_cache *, void *)
2768 #else
2769                                        void (*ctor)(void *, struct kmem_cache *,
2770                                                     unsigned long)
2771 #endif
2772                                      )
2773 {
2774         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2775                                  SLAB_MEM_SPREAD | extra_flags), ctor
2776 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2777                                  ,NULL
2778 #endif
2779                                 );
2780 }
2781
2782 int btrfs_init_cachep(void)
2783 {
2784         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2785                                           sizeof(struct btrfs_inode),
2786                                           0, init_once);
2787         if (!btrfs_inode_cachep)
2788                 goto fail;
2789         btrfs_trans_handle_cachep =
2790                         btrfs_cache_create("btrfs_trans_handle_cache",
2791                                            sizeof(struct btrfs_trans_handle),
2792                                            0, NULL);
2793         if (!btrfs_trans_handle_cachep)
2794                 goto fail;
2795         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
2796                                              sizeof(struct btrfs_transaction),
2797                                              0, NULL);
2798         if (!btrfs_transaction_cachep)
2799                 goto fail;
2800         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
2801                                          sizeof(struct btrfs_path),
2802                                          0, NULL);
2803         if (!btrfs_path_cachep)
2804                 goto fail;
2805         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
2806                                               SLAB_DESTROY_BY_RCU, NULL);
2807         if (!btrfs_bit_radix_cachep)
2808                 goto fail;
2809         return 0;
2810 fail:
2811         btrfs_destroy_cachep();
2812         return -ENOMEM;
2813 }
2814
2815 static int btrfs_getattr(struct vfsmount *mnt,
2816                          struct dentry *dentry, struct kstat *stat)
2817 {
2818         struct inode *inode = dentry->d_inode;
2819         generic_fillattr(inode, stat);
2820         stat->blksize = PAGE_CACHE_SIZE;
2821         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
2822         return 0;
2823 }
2824
2825 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
2826                            struct inode * new_dir,struct dentry *new_dentry)
2827 {
2828         struct btrfs_trans_handle *trans;
2829         struct btrfs_root *root = BTRFS_I(old_dir)->root;
2830         struct inode *new_inode = new_dentry->d_inode;
2831         struct inode *old_inode = old_dentry->d_inode;
2832         struct timespec ctime = CURRENT_TIME;
2833         struct btrfs_path *path;
2834         int ret;
2835
2836         if (S_ISDIR(old_inode->i_mode) && new_inode &&
2837             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
2838                 return -ENOTEMPTY;
2839         }
2840
2841         mutex_lock(&root->fs_info->fs_mutex);
2842         ret = btrfs_check_free_space(root, 1, 0);
2843         if (ret)
2844                 goto out_unlock;
2845
2846         trans = btrfs_start_transaction(root, 1);
2847
2848         btrfs_set_trans_block_group(trans, new_dir);
2849         path = btrfs_alloc_path();
2850         if (!path) {
2851                 ret = -ENOMEM;
2852                 goto out_fail;
2853         }
2854
2855         old_dentry->d_inode->i_nlink++;
2856         old_dir->i_ctime = old_dir->i_mtime = ctime;
2857         new_dir->i_ctime = new_dir->i_mtime = ctime;
2858         old_inode->i_ctime = ctime;
2859
2860         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
2861         if (ret)
2862                 goto out_fail;
2863
2864         if (new_inode) {
2865                 new_inode->i_ctime = CURRENT_TIME;
2866                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
2867                 if (ret)
2868                         goto out_fail;
2869         }
2870         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
2871         if (ret)
2872                 goto out_fail;
2873
2874 out_fail:
2875         btrfs_free_path(path);
2876         btrfs_end_transaction(trans, root);
2877 out_unlock:
2878         mutex_unlock(&root->fs_info->fs_mutex);
2879         return ret;
2880 }
2881
2882 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
2883                          const char *symname)
2884 {
2885         struct btrfs_trans_handle *trans;
2886         struct btrfs_root *root = BTRFS_I(dir)->root;
2887         struct btrfs_path *path;
2888         struct btrfs_key key;
2889         struct inode *inode = NULL;
2890         int err;
2891         int drop_inode = 0;
2892         u64 objectid;
2893         int name_len;
2894         int datasize;
2895         unsigned long ptr;
2896         struct btrfs_file_extent_item *ei;
2897         struct extent_buffer *leaf;
2898         unsigned long nr = 0;
2899
2900         name_len = strlen(symname) + 1;
2901         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
2902                 return -ENAMETOOLONG;
2903
2904         mutex_lock(&root->fs_info->fs_mutex);
2905         err = btrfs_check_free_space(root, 1, 0);
2906         if (err)
2907                 goto out_fail;
2908
2909         trans = btrfs_start_transaction(root, 1);
2910         btrfs_set_trans_block_group(trans, dir);
2911
2912         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2913         if (err) {
2914                 err = -ENOSPC;
2915                 goto out_unlock;
2916         }
2917
2918         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2919                                 dentry->d_name.len,
2920                                 dentry->d_parent->d_inode->i_ino, objectid,
2921                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
2922         err = PTR_ERR(inode);
2923         if (IS_ERR(inode))
2924                 goto out_unlock;
2925
2926         btrfs_set_trans_block_group(trans, inode);
2927         err = btrfs_add_nondir(trans, dentry, inode, 0);
2928         if (err)
2929                 drop_inode = 1;
2930         else {
2931                 inode->i_mapping->a_ops = &btrfs_aops;
2932                 inode->i_fop = &btrfs_file_operations;
2933                 inode->i_op = &btrfs_file_inode_operations;
2934                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2935                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2936                                      inode->i_mapping, GFP_NOFS);
2937                 BTRFS_I(inode)->delalloc_bytes = 0;
2938                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2939         }
2940         dir->i_sb->s_dirt = 1;
2941         btrfs_update_inode_block_group(trans, inode);
2942         btrfs_update_inode_block_group(trans, dir);
2943         if (drop_inode)
2944                 goto out_unlock;
2945
2946         path = btrfs_alloc_path();
2947         BUG_ON(!path);
2948         key.objectid = inode->i_ino;
2949         key.offset = 0;
2950         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2951         datasize = btrfs_file_extent_calc_inline_size(name_len);
2952         err = btrfs_insert_empty_item(trans, root, path, &key,
2953                                       datasize);
2954         if (err) {
2955                 drop_inode = 1;
2956                 goto out_unlock;
2957         }
2958         leaf = path->nodes[0];
2959         ei = btrfs_item_ptr(leaf, path->slots[0],
2960                             struct btrfs_file_extent_item);
2961         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
2962         btrfs_set_file_extent_type(leaf, ei,
2963                                    BTRFS_FILE_EXTENT_INLINE);
2964         ptr = btrfs_file_extent_inline_start(ei);
2965         write_extent_buffer(leaf, symname, ptr, name_len);
2966         btrfs_mark_buffer_dirty(leaf);
2967         btrfs_free_path(path);
2968
2969         inode->i_op = &btrfs_symlink_inode_operations;
2970         inode->i_mapping->a_ops = &btrfs_symlink_aops;
2971         inode->i_size = name_len - 1;
2972         err = btrfs_update_inode(trans, root, inode);
2973         if (err)
2974                 drop_inode = 1;
2975
2976 out_unlock:
2977         nr = trans->blocks_used;
2978         btrfs_end_transaction(trans, root);
2979 out_fail:
2980         mutex_unlock(&root->fs_info->fs_mutex);
2981         if (drop_inode) {
2982                 inode_dec_link_count(inode);
2983                 iput(inode);
2984         }
2985         btrfs_btree_balance_dirty(root, nr);
2986         btrfs_throttle(root);
2987         return err;
2988 }
2989 static int btrfs_permission(struct inode *inode, int mask,
2990                             struct nameidata *nd)
2991 {
2992         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
2993                 return -EACCES;
2994         return generic_permission(inode, mask, NULL);
2995 }
2996
2997 static struct inode_operations btrfs_dir_inode_operations = {
2998         .lookup         = btrfs_lookup,
2999         .create         = btrfs_create,
3000         .unlink         = btrfs_unlink,
3001         .link           = btrfs_link,
3002         .mkdir          = btrfs_mkdir,
3003         .rmdir          = btrfs_rmdir,
3004         .rename         = btrfs_rename,
3005         .symlink        = btrfs_symlink,
3006         .setattr        = btrfs_setattr,
3007         .mknod          = btrfs_mknod,
3008         .setxattr       = generic_setxattr,
3009         .getxattr       = generic_getxattr,
3010         .listxattr      = btrfs_listxattr,
3011         .removexattr    = generic_removexattr,
3012         .permission     = btrfs_permission,
3013 };
3014 static struct inode_operations btrfs_dir_ro_inode_operations = {
3015         .lookup         = btrfs_lookup,
3016         .permission     = btrfs_permission,
3017 };
3018 static struct file_operations btrfs_dir_file_operations = {
3019         .llseek         = generic_file_llseek,
3020         .read           = generic_read_dir,
3021         .readdir        = btrfs_readdir,
3022         .unlocked_ioctl = btrfs_ioctl,
3023 #ifdef CONFIG_COMPAT
3024         .compat_ioctl   = btrfs_ioctl,
3025 #endif
3026 };
3027
3028 static struct extent_io_ops btrfs_extent_io_ops = {
3029         .fill_delalloc = run_delalloc_range,
3030         .submit_bio_hook = btrfs_submit_bio_hook,
3031         .readpage_io_hook = btrfs_readpage_io_hook,
3032         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3033         .set_bit_hook = btrfs_set_bit_hook,
3034         .clear_bit_hook = btrfs_clear_bit_hook,
3035 };
3036
3037 static struct address_space_operations btrfs_aops = {
3038         .readpage       = btrfs_readpage,
3039         .writepage      = btrfs_writepage,
3040         .writepages     = btrfs_writepages,
3041         .readpages      = btrfs_readpages,
3042         .sync_page      = block_sync_page,
3043         .bmap           = btrfs_bmap,
3044         .invalidatepage = btrfs_invalidatepage,
3045         .releasepage    = btrfs_releasepage,
3046         .set_page_dirty = __set_page_dirty_nobuffers,
3047 };
3048
3049 static struct address_space_operations btrfs_symlink_aops = {
3050         .readpage       = btrfs_readpage,
3051         .writepage      = btrfs_writepage,
3052         .invalidatepage = btrfs_invalidatepage,
3053         .releasepage    = btrfs_releasepage,
3054 };
3055
3056 static struct inode_operations btrfs_file_inode_operations = {
3057         .truncate       = btrfs_truncate,
3058         .getattr        = btrfs_getattr,
3059         .setattr        = btrfs_setattr,
3060         .setxattr       = generic_setxattr,
3061         .getxattr       = generic_getxattr,
3062         .listxattr      = btrfs_listxattr,
3063         .removexattr    = generic_removexattr,
3064         .permission     = btrfs_permission,
3065 };
3066 static struct inode_operations btrfs_special_inode_operations = {
3067         .getattr        = btrfs_getattr,
3068         .setattr        = btrfs_setattr,
3069         .permission     = btrfs_permission,
3070 };
3071 static struct inode_operations btrfs_symlink_inode_operations = {
3072         .readlink       = generic_readlink,
3073         .follow_link    = page_follow_link_light,
3074         .put_link       = page_put_link,
3075         .permission     = btrfs_permission,
3076 };