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