git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/inode.c

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