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 = btrfs_super_total_bytes(&root->fs_info->super_copy);
  83         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  84         u64 thresh;
  85         unsigned long flags;
  86         int ret = 0;
  87
  88         if (for_del)
  89                 thresh = total * 90;
  90         else
  91                 thresh = total * 85;
  92
  93         do_div(thresh, 100);
  94
  95         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
  96         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  97                 ret = -ENOSPC;
  98         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
  99         return ret;
 100 }
 101
 102 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 103 {
 104         struct btrfs_root *root = BTRFS_I(inode)->root;
 105         struct btrfs_trans_handle *trans;
 106         u64 alloc_hint = 0;
 107         u64 num_bytes;
 108         u64 cur_alloc_size;
 109         u64 blocksize = root->sectorsize;
 110         u64 orig_start = start;
 111         u64 orig_num_bytes;
 112         struct btrfs_key ins;
 113         int ret;
 114
 115         trans = btrfs_start_transaction(root, 1);
 116         BUG_ON(!trans);
 117         btrfs_set_trans_block_group(trans, inode);
 118
 119         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 120         num_bytes = max(blocksize,  num_bytes);
 121         ret = btrfs_drop_extents(trans, root, inode,
 122                                  start, start + num_bytes, start, &alloc_hint);
 123         orig_num_bytes = num_bytes;
 124
 125         if (alloc_hint == EXTENT_MAP_INLINE)
 126                 goto out;
 127
 128         BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
 129
 130         while(num_bytes > 0) {
 131                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 132                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 133                                          root->sectorsize,
 134                                          root->root_key.objectid,
 135                                          trans->transid,
 136                                          inode->i_ino, start, 0,
 137                                          alloc_hint, (u64)-1, &ins, 1);
 138                 if (ret) {
 139                         WARN_ON(1);
 140                         goto out;
 141                 }
 142                 cur_alloc_size = ins.offset;
 143                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 144                                                start, ins.objectid, ins.offset,
 145                                                ins.offset, 0);
 146                 inode->i_blocks += ins.offset >> 9;
 147                 btrfs_check_file(root, inode);
 148                 if (num_bytes < cur_alloc_size) {
 149                         printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
 150                                cur_alloc_size);
 151                         break;
 152                 }
 153                 num_bytes -= cur_alloc_size;
 154                 alloc_hint = ins.objectid + ins.offset;
 155                 start += cur_alloc_size;
 156         }
 157         btrfs_drop_extent_cache(inode, orig_start,
 158                                 orig_start + orig_num_bytes - 1);
 159         btrfs_add_ordered_inode(inode);
 160         btrfs_update_inode(trans, root, inode);
 161 out:
 162         btrfs_end_transaction(trans, root);
 163         return ret;
 164 }
 165
 166 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 167 {
 168         u64 extent_start;
 169         u64 extent_end;
 170         u64 bytenr;
 171         u64 cow_end;
 172         u64 loops = 0;
 173         u64 total_fs_bytes;
 174         struct btrfs_root *root = BTRFS_I(inode)->root;
 175         struct btrfs_block_group_cache *block_group;
 176         struct extent_buffer *leaf;
 177         int found_type;
 178         struct btrfs_path *path;
 179         struct btrfs_file_extent_item *item;
 180         int ret;
 181         int err;
 182         struct btrfs_key found_key;
 183
 184         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 185         path = btrfs_alloc_path();
 186         BUG_ON(!path);
 187 again:
 188         ret = btrfs_lookup_file_extent(NULL, root, path,
 189                                        inode->i_ino, start, 0);
 190         if (ret < 0) {
 191                 btrfs_free_path(path);
 192                 return ret;
 193         }
 194
 195         cow_end = end;
 196         if (ret != 0) {
 197                 if (path->slots[0] == 0)
 198                         goto not_found;
 199                 path->slots[0]--;
 200         }
 201
 202         leaf = path->nodes[0];
 203         item = btrfs_item_ptr(leaf, path->slots[0],
 204                               struct btrfs_file_extent_item);
 205
 206         /* are we inside the extent that was found? */
 207         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 208         found_type = btrfs_key_type(&found_key);
 209         if (found_key.objectid != inode->i_ino ||
 210             found_type != BTRFS_EXTENT_DATA_KEY)
 211                 goto not_found;
 212
 213         found_type = btrfs_file_extent_type(leaf, item);
 214         extent_start = found_key.offset;
 215         if (found_type == BTRFS_FILE_EXTENT_REG) {
 216                 u64 extent_num_bytes;
 217
 218                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 219                 extent_end = extent_start + extent_num_bytes;
 220                 err = 0;
 221
 222                 if (loops && start != extent_start)
 223                         goto not_found;
 224
 225                 if (start < extent_start || start >= extent_end)
 226                         goto not_found;
 227
 228                 cow_end = min(end, extent_end - 1);
 229                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 230                 if (bytenr == 0)
 231                         goto not_found;
 232
 233                 if (btrfs_count_snapshots_in_path(root, path, inode->i_ino,
 234                                                   bytenr) != 1) {
 235                         goto not_found;
 236                 }
 237
 238                 /*
 239                  * we may be called by the resizer, make sure we're inside
 240                  * the limits of the FS
 241                  */
 242                 block_group = btrfs_lookup_block_group(root->fs_info,
 243                                                        bytenr);
 244                 if (!block_group || block_group->ro)
 245                         goto not_found;
 246
 247                 start = extent_end;
 248         } else {
 249                 goto not_found;
 250         }
 251 loop:
 252         if (start > end) {
 253                 btrfs_free_path(path);
 254                 return 0;
 255         }
 256         btrfs_release_path(root, path);
 257         loops++;
 258         goto again;
 259
 260 not_found:
 261         cow_file_range(inode, start, end);
 262         start = end + 1;
 263         goto loop;
 264 }
 265
 266 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 267 {
 268         struct btrfs_root *root = BTRFS_I(inode)->root;
 269         int ret;
 270         mutex_lock(&root->fs_info->fs_mutex);
 271         if (btrfs_test_opt(root, NODATACOW) ||
 272             btrfs_test_flag(inode, NODATACOW))
 273                 ret = run_delalloc_nocow(inode, start, end);
 274         else
 275                 ret = cow_file_range(inode, start, end);
 276
 277         mutex_unlock(&root->fs_info->fs_mutex);
 278         return ret;
 279 }
 280
 281 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 282                        unsigned long old, unsigned long bits)
 283 {
 284         unsigned long flags;
 285         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 286                 struct btrfs_root *root = BTRFS_I(inode)->root;
 287                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 288                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 289                 root->fs_info->delalloc_bytes += end - start + 1;
 290                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 291         }
 292         return 0;
 293 }
 294
 295 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 296                          unsigned long old, unsigned long bits)
 297 {
 298         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 299                 struct btrfs_root *root = BTRFS_I(inode)->root;
 300                 unsigned long flags;
 301
 302                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 303                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 304                         printk("warning: delalloc account %Lu %Lu\n",
 305                                end - start + 1, root->fs_info->delalloc_bytes);
 306                         root->fs_info->delalloc_bytes = 0;
 307                         BTRFS_I(inode)->delalloc_bytes = 0;
 308                 } else {
 309                         root->fs_info->delalloc_bytes -= end - start + 1;
 310                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 311                 }
 312                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 313         }
 314         return 0;
 315 }
 316
 317 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 318                          size_t size, struct bio *bio)
 319 {
 320         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 321         struct btrfs_mapping_tree *map_tree;
 322         u64 logical = bio->bi_sector << 9;
 323         u64 length = 0;
 324         u64 map_length;
 325         int ret;
 326
 327         length = bio->bi_size;
 328         map_tree = &root->fs_info->mapping_tree;
 329         map_length = length;
 330         ret = btrfs_map_block(map_tree, READ, logical,
 331                               &map_length, NULL, 0);
 332
 333         if (map_length < length + size) {
 334                 return 1;
 335         }
 336         return 0;
 337 }
 338
 339 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 340                           int mirror_num)
 341 {
 342         struct btrfs_root *root = BTRFS_I(inode)->root;
 343         struct btrfs_trans_handle *trans;
 344         int ret = 0;
 345         char *sums = NULL;
 346
 347         ret = btrfs_csum_one_bio(root, bio, &sums);
 348         BUG_ON(ret);
 349
 350         mutex_lock(&root->fs_info->fs_mutex);
 351         trans = btrfs_start_transaction(root, 1);
 352
 353         btrfs_set_trans_block_group(trans, inode);
 354         btrfs_csum_file_blocks(trans, root, inode, bio, sums);
 355
 356         ret = btrfs_end_transaction(trans, root);
 357         BUG_ON(ret);
 358         mutex_unlock(&root->fs_info->fs_mutex);
 359
 360         kfree(sums);
 361
 362         return btrfs_map_bio(root, rw, bio, mirror_num);
 363 }
 364
 365 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 366                           int mirror_num)
 367 {
 368         struct btrfs_root *root = BTRFS_I(inode)->root;
 369         int ret = 0;
 370
 371         if (!(rw & (1 << BIO_RW))) {
 372                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 373                 BUG_ON(ret);
 374                 goto mapit;
 375         }
 376
 377         if (btrfs_test_opt(root, NODATASUM) ||
 378             btrfs_test_flag(inode, NODATASUM)) {
 379                 goto mapit;
 380         }
 381
 382         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 383                                    inode, rw, bio, mirror_num,
 384                                    __btrfs_submit_bio_hook);
 385 mapit:
 386         return btrfs_map_bio(root, rw, bio, mirror_num);
 387 }
 388
 389 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 390 {
 391         int ret = 0;
 392         struct inode *inode = page->mapping->host;
 393         struct btrfs_root *root = BTRFS_I(inode)->root;
 394         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 395         struct btrfs_csum_item *item;
 396         struct btrfs_path *path = NULL;
 397         u32 csum;
 398
 399         if (btrfs_test_opt(root, NODATASUM) ||
 400             btrfs_test_flag(inode, NODATASUM))
 401                 return 0;
 402
 403         mutex_lock(&root->fs_info->fs_mutex);
 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         mutex_unlock(&root->fs_info->fs_mutex);
 422         return ret;
 423 }
 424
 425 struct io_failure_record {
 426         struct page *page;
 427         u64 start;
 428         u64 len;
 429         u64 logical;
 430         int last_mirror;
 431 };
 432
 433 int btrfs_io_failed_hook(struct bio *failed_bio,
 434                          struct page *page, u64 start, u64 end,
 435                          struct extent_state *state)
 436 {
 437         struct io_failure_record *failrec = NULL;
 438         u64 private;
 439         struct extent_map *em;
 440         struct inode *inode = page->mapping->host;
 441         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 442         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 443         struct bio *bio;
 444         int num_copies;
 445         int ret;
 446         int rw;
 447         u64 logical;
 448
 449         ret = get_state_private(failure_tree, start, &private);
 450         if (ret) {
 451                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 452                 if (!failrec)
 453                         return -ENOMEM;
 454                 failrec->start = start;
 455                 failrec->len = end - start + 1;
 456                 failrec->last_mirror = 0;
 457
 458                 spin_lock(&em_tree->lock);
 459                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 460                 if (em->start > start || em->start + em->len < start) {
 461                         free_extent_map(em);
 462                         em = NULL;
 463                 }
 464                 spin_unlock(&em_tree->lock);
 465
 466                 if (!em || IS_ERR(em)) {
 467                         kfree(failrec);
 468                         return -EIO;
 469                 }
 470                 logical = start - em->start;
 471                 logical = em->block_start + logical;
 472                 failrec->logical = logical;
 473                 free_extent_map(em);
 474                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 475                                 EXTENT_DIRTY, GFP_NOFS);
 476                 set_state_private(failure_tree, start,
 477                                  (u64)(unsigned long)failrec);
 478         } else {
 479                 failrec = (struct io_failure_record *)(unsigned long)private;
 480         }
 481         num_copies = btrfs_num_copies(
 482                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 483                               failrec->logical, failrec->len);
 484         failrec->last_mirror++;
 485         if (!state) {
 486                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 487                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 488                                                     failrec->start,
 489                                                     EXTENT_LOCKED);
 490                 if (state && state->start != failrec->start)
 491                         state = NULL;
 492                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 493         }
 494         if (!state || failrec->last_mirror > num_copies) {
 495                 set_state_private(failure_tree, failrec->start, 0);
 496                 clear_extent_bits(failure_tree, failrec->start,
 497                                   failrec->start + failrec->len - 1,
 498                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 499                 kfree(failrec);
 500                 return -EIO;
 501         }
 502         bio = bio_alloc(GFP_NOFS, 1);
 503         bio->bi_private = state;
 504         bio->bi_end_io = failed_bio->bi_end_io;
 505         bio->bi_sector = failrec->logical >> 9;
 506         bio->bi_bdev = failed_bio->bi_bdev;
 507         bio->bi_size = 0;
 508         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 509         if (failed_bio->bi_rw & (1 << BIO_RW))
 510                 rw = WRITE;
 511         else
 512                 rw = READ;
 513
 514         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 515                                                       failrec->last_mirror);
 516         return 0;
 517 }
 518
 519 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 520 {
 521         u64 private;
 522         u64 private_failure;
 523         struct io_failure_record *failure;
 524         int ret;
 525
 526         private = 0;
 527         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 528                              (u64)-1, 1, EXTENT_DIRTY)) {
 529                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 530                                         start, &private_failure);
 531                 if (ret == 0) {
 532                         failure = (struct io_failure_record *)(unsigned long)
 533                                    private_failure;
 534                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 535                                           failure->start, 0);
 536                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 537                                           failure->start,
 538                                           failure->start + failure->len - 1,
 539                                           EXTENT_DIRTY | EXTENT_LOCKED,
 540                                           GFP_NOFS);
 541                         kfree(failure);
 542                 }
 543         }
 544         return 0;
 545 }
 546
 547 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 548                                struct extent_state *state)
 549 {
 550         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 551         struct inode *inode = page->mapping->host;
 552         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 553         char *kaddr;
 554         u64 private = ~(u32)0;
 555         int ret;
 556         struct btrfs_root *root = BTRFS_I(inode)->root;
 557         u32 csum = ~(u32)0;
 558         unsigned long flags;
 559
 560         if (btrfs_test_opt(root, NODATASUM) ||
 561             btrfs_test_flag(inode, NODATASUM))
 562                 return 0;
 563         if (state && state->start == start) {
 564                 private = state->private;
 565                 ret = 0;
 566         } else {
 567                 ret = get_state_private(io_tree, start, &private);
 568         }
 569         local_irq_save(flags);
 570         kaddr = kmap_atomic(page, KM_IRQ0);
 571         if (ret) {
 572                 goto zeroit;
 573         }
 574         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 575         btrfs_csum_final(csum, (char *)&csum);
 576         if (csum != private) {
 577                 goto zeroit;
 578         }
 579         kunmap_atomic(kaddr, KM_IRQ0);
 580         local_irq_restore(flags);
 581
 582         /* if the io failure tree for this inode is non-empty,
 583          * check to see if we've recovered from a failed IO
 584          */
 585         btrfs_clean_io_failures(inode, start);
 586         return 0;
 587
 588 zeroit:
 589         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 590                page->mapping->host->i_ino, (unsigned long long)start, csum,
 591                private);
 592         memset(kaddr + offset, 1, end - start + 1);
 593         flush_dcache_page(page);
 594         kunmap_atomic(kaddr, KM_IRQ0);
 595         local_irq_restore(flags);
 596         if (private == 0)
 597                 return 0;
 598         return -EIO;
 599 }
 600
 601 void btrfs_read_locked_inode(struct inode *inode)
 602 {
 603         struct btrfs_path *path;
 604         struct extent_buffer *leaf;
 605         struct btrfs_inode_item *inode_item;
 606         struct btrfs_timespec *tspec;
 607         struct btrfs_root *root = BTRFS_I(inode)->root;
 608         struct btrfs_key location;
 609         u64 alloc_group_block;
 610         u32 rdev;
 611         int ret;
 612
 613         path = btrfs_alloc_path();
 614         BUG_ON(!path);
 615         mutex_lock(&root->fs_info->fs_mutex);
 616         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 617
 618         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 619         if (ret)
 620                 goto make_bad;
 621
 622         leaf = path->nodes[0];
 623         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 624                                     struct btrfs_inode_item);
 625
 626         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 627         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 628         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 629         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 630         inode->i_size = btrfs_inode_size(leaf, inode_item);
 631
 632         tspec = btrfs_inode_atime(inode_item);
 633         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 634         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 635
 636         tspec = btrfs_inode_mtime(inode_item);
 637         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 638         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 639
 640         tspec = btrfs_inode_ctime(inode_item);
 641         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 642         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 643
 644         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 645         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 646         inode->i_rdev = 0;
 647         rdev = btrfs_inode_rdev(leaf, inode_item);
 648
 649         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 650         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 651                                                        alloc_group_block);
 652         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 653         if (!BTRFS_I(inode)->block_group) {
 654                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 655                                                  NULL, 0,
 656                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
 657         }
 658         btrfs_free_path(path);
 659         inode_item = NULL;
 660
 661         mutex_unlock(&root->fs_info->fs_mutex);
 662
 663         switch (inode->i_mode & S_IFMT) {
 664         case S_IFREG:
 665                 inode->i_mapping->a_ops = &btrfs_aops;
 666                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
 667                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 668                 inode->i_fop = &btrfs_file_operations;
 669                 inode->i_op = &btrfs_file_inode_operations;
 670                 break;
 671         case S_IFDIR:
 672                 inode->i_fop = &btrfs_dir_file_operations;
 673                 if (root == root->fs_info->tree_root)
 674                         inode->i_op = &btrfs_dir_ro_inode_operations;
 675                 else
 676                         inode->i_op = &btrfs_dir_inode_operations;
 677                 break;
 678         case S_IFLNK:
 679                 inode->i_op = &btrfs_symlink_inode_operations;
 680                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 681                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
 682                 break;
 683         default:
 684                 init_special_inode(inode, inode->i_mode, rdev);
 685                 break;
 686         }
 687         return;
 688
 689 make_bad:
 690         btrfs_release_path(root, path);
 691         btrfs_free_path(path);
 692         mutex_unlock(&root->fs_info->fs_mutex);
 693         make_bad_inode(inode);
 694 }
 695
 696 static void fill_inode_item(struct extent_buffer *leaf,
 697                             struct btrfs_inode_item *item,
 698                             struct inode *inode)
 699 {
 700         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 701         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 702         btrfs_set_inode_size(leaf, item, inode->i_size);
 703         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 704         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 705
 706         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 707                                inode->i_atime.tv_sec);
 708         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 709                                 inode->i_atime.tv_nsec);
 710
 711         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 712                                inode->i_mtime.tv_sec);
 713         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 714                                 inode->i_mtime.tv_nsec);
 715
 716         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 717                                inode->i_ctime.tv_sec);
 718         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 719                                 inode->i_ctime.tv_nsec);
 720
 721         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 722         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 723         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 724         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 725         btrfs_set_inode_block_group(leaf, item,
 726                                     BTRFS_I(inode)->block_group->key.objectid);
 727 }
 728
 729 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 730                               struct btrfs_root *root,
 731                               struct inode *inode)
 732 {
 733         struct btrfs_inode_item *inode_item;
 734         struct btrfs_path *path;
 735         struct extent_buffer *leaf;
 736         int ret;
 737
 738         path = btrfs_alloc_path();
 739         BUG_ON(!path);
 740         ret = btrfs_lookup_inode(trans, root, path,
 741                                  &BTRFS_I(inode)->location, 1);
 742         if (ret) {
 743                 if (ret > 0)
 744                         ret = -ENOENT;
 745                 goto failed;
 746         }
 747
 748         leaf = path->nodes[0];
 749         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 750                                   struct btrfs_inode_item);
 751
 752         fill_inode_item(leaf, inode_item, inode);
 753         btrfs_mark_buffer_dirty(leaf);
 754         btrfs_set_inode_last_trans(trans, inode);
 755         ret = 0;
 756 failed:
 757         btrfs_release_path(root, path);
 758         btrfs_free_path(path);
 759         return ret;
 760 }
 761
 762
 763 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 764                               struct btrfs_root *root,
 765                               struct inode *dir,
 766                               struct dentry *dentry)
 767 {
 768         struct btrfs_path *path;
 769         const char *name = dentry->d_name.name;
 770         int name_len = dentry->d_name.len;
 771         int ret = 0;
 772         struct extent_buffer *leaf;
 773         struct btrfs_dir_item *di;
 774         struct btrfs_key key;
 775
 776         path = btrfs_alloc_path();
 777         if (!path) {
 778                 ret = -ENOMEM;
 779                 goto err;
 780         }
 781
 782         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 783                                     name, name_len, -1);
 784         if (IS_ERR(di)) {
 785                 ret = PTR_ERR(di);
 786                 goto err;
 787         }
 788         if (!di) {
 789                 ret = -ENOENT;
 790                 goto err;
 791         }
 792         leaf = path->nodes[0];
 793         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 794         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 795         if (ret)
 796                 goto err;
 797         btrfs_release_path(root, path);
 798
 799         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 800                                          key.objectid, name, name_len, -1);
 801         if (IS_ERR(di)) {
 802                 ret = PTR_ERR(di);
 803                 goto err;
 804         }
 805         if (!di) {
 806                 ret = -ENOENT;
 807                 goto err;
 808         }
 809         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 810
 811         dentry->d_inode->i_ctime = dir->i_ctime;
 812         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 813                                   dentry->d_inode->i_ino,
 814                                   dentry->d_parent->d_inode->i_ino);
 815         if (ret) {
 816                 printk("failed to delete reference to %.*s, "
 817                        "inode %lu parent %lu\n", name_len, name,
 818                        dentry->d_inode->i_ino,
 819                        dentry->d_parent->d_inode->i_ino);
 820         }
 821 err:
 822         btrfs_free_path(path);
 823         if (!ret) {
 824                 dir->i_size -= name_len * 2;
 825                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 826                 btrfs_update_inode(trans, root, dir);
 827 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 828                 dentry->d_inode->i_nlink--;
 829 #else
 830                 drop_nlink(dentry->d_inode);
 831 #endif
 832                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 833                 dir->i_sb->s_dirt = 1;
 834         }
 835         return ret;
 836 }
 837
 838 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 839 {
 840         struct btrfs_root *root;
 841         struct btrfs_trans_handle *trans;
 842         struct inode *inode = dentry->d_inode;
 843         int ret;
 844         unsigned long nr = 0;
 845
 846         root = BTRFS_I(dir)->root;
 847         mutex_lock(&root->fs_info->fs_mutex);
 848
 849         ret = btrfs_check_free_space(root, 1, 1);
 850         if (ret)
 851                 goto fail;
 852
 853         trans = btrfs_start_transaction(root, 1);
 854
 855         btrfs_set_trans_block_group(trans, dir);
 856         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 857         nr = trans->blocks_used;
 858
 859         if (inode->i_nlink == 0) {
 860                 int found;
 861                 /* if the inode isn't linked anywhere,
 862                  * we don't need to worry about
 863                  * data=ordered
 864                  */
 865                 found = btrfs_del_ordered_inode(inode);
 866                 if (found == 1) {
 867                         atomic_dec(&inode->i_count);
 868                 }
 869         }
 870
 871         btrfs_end_transaction(trans, root);
 872 fail:
 873         mutex_unlock(&root->fs_info->fs_mutex);
 874         btrfs_btree_balance_dirty(root, nr);
 875         btrfs_throttle(root);
 876         return ret;
 877 }
 878
 879 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 880 {
 881         struct inode *inode = dentry->d_inode;
 882         int err = 0;
 883         int ret;
 884         struct btrfs_root *root = BTRFS_I(dir)->root;
 885         struct btrfs_trans_handle *trans;
 886         unsigned long nr = 0;
 887
 888         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 889                 return -ENOTEMPTY;
 890
 891         mutex_lock(&root->fs_info->fs_mutex);
 892         ret = btrfs_check_free_space(root, 1, 1);
 893         if (ret)
 894                 goto fail;
 895
 896         trans = btrfs_start_transaction(root, 1);
 897         btrfs_set_trans_block_group(trans, dir);
 898
 899         /* now the directory is empty */
 900         err = btrfs_unlink_trans(trans, root, dir, dentry);
 901         if (!err) {
 902                 inode->i_size = 0;
 903         }
 904
 905         nr = trans->blocks_used;
 906         ret = btrfs_end_transaction(trans, root);
 907 fail:
 908         mutex_unlock(&root->fs_info->fs_mutex);
 909         btrfs_btree_balance_dirty(root, nr);
 910         btrfs_throttle(root);
 911
 912         if (ret && !err)
 913                 err = ret;
 914         return err;
 915 }
 916
 917 /*
 918  * this can truncate away extent items, csum items and directory items.
 919  * It starts at a high offset and removes keys until it can't find
 920  * any higher than i_size.
 921  *
 922  * csum items that cross the new i_size are truncated to the new size
 923  * as well.
 924  */
 925 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 926                                    struct btrfs_root *root,
 927                                    struct inode *inode,
 928                                    u32 min_type)
 929 {
 930         int ret;
 931         struct btrfs_path *path;
 932         struct btrfs_key key;
 933         struct btrfs_key found_key;
 934         u32 found_type;
 935         struct extent_buffer *leaf;
 936         struct btrfs_file_extent_item *fi;
 937         u64 extent_start = 0;
 938         u64 extent_num_bytes = 0;
 939         u64 item_end = 0;
 940         u64 root_gen = 0;
 941         u64 root_owner = 0;
 942         int found_extent;
 943         int del_item;
 944         int pending_del_nr = 0;
 945         int pending_del_slot = 0;
 946         int extent_type = -1;
 947         u64 mask = root->sectorsize - 1;
 948
 949         btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
 950         path = btrfs_alloc_path();
 951         path->reada = -1;
 952         BUG_ON(!path);
 953
 954         /* FIXME, add redo link to tree so we don't leak on crash */
 955         key.objectid = inode->i_ino;
 956         key.offset = (u64)-1;
 957         key.type = (u8)-1;
 958
 959         btrfs_init_path(path);
 960 search_again:
 961         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 962         if (ret < 0) {
 963                 goto error;
 964         }
 965         if (ret > 0) {
 966                 BUG_ON(path->slots[0] == 0);
 967                 path->slots[0]--;
 968         }
 969
 970         while(1) {
 971                 fi = NULL;
 972                 leaf = path->nodes[0];
 973                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 974                 found_type = btrfs_key_type(&found_key);
 975
 976                 if (found_key.objectid != inode->i_ino)
 977                         break;
 978
 979                 if (found_type < min_type)
 980                         break;
 981
 982                 item_end = found_key.offset;
 983                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 984                         fi = btrfs_item_ptr(leaf, path->slots[0],
 985                                             struct btrfs_file_extent_item);
 986                         extent_type = btrfs_file_extent_type(leaf, fi);
 987                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 988                                 item_end +=
 989                                     btrfs_file_extent_num_bytes(leaf, fi);
 990                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 991                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 992                                                                 path->slots[0]);
 993                                 item_end += btrfs_file_extent_inline_len(leaf,
 994                                                                          item);
 995                         }
 996                         item_end--;
 997                 }
 998                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 999                         ret = btrfs_csum_truncate(trans, root, path,
1000                                                   inode->i_size);
1001                         BUG_ON(ret);
1002                 }
1003                 if (item_end < inode->i_size) {
1004                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1005                                 found_type = BTRFS_INODE_ITEM_KEY;
1006                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1007                                 found_type = BTRFS_CSUM_ITEM_KEY;
1008                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1009                                 found_type = BTRFS_XATTR_ITEM_KEY;
1010                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1011                                 found_type = BTRFS_INODE_REF_KEY;
1012                         } else if (found_type) {
1013                                 found_type--;
1014                         } else {
1015                                 break;
1016                         }
1017                         btrfs_set_key_type(&key, found_type);
1018                         goto next;
1019                 }
1020                 if (found_key.offset >= inode->i_size)
1021                         del_item = 1;
1022                 else
1023                         del_item = 0;
1024                 found_extent = 0;
1025
1026                 /* FIXME, shrink the extent if the ref count is only 1 */
1027                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1028                         goto delete;
1029
1030                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1031                         u64 num_dec;
1032                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1033                         if (!del_item) {
1034                                 u64 orig_num_bytes =
1035                                         btrfs_file_extent_num_bytes(leaf, fi);
1036                                 extent_num_bytes = inode->i_size -
1037                                         found_key.offset + root->sectorsize - 1;
1038                                 extent_num_bytes = extent_num_bytes &
1039                                         ~((u64)root->sectorsize - 1);
1040                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1041                                                          extent_num_bytes);
1042                                 num_dec = (orig_num_bytes -
1043                                            extent_num_bytes);
1044                                 if (extent_start != 0)
1045                                         dec_i_blocks(inode, num_dec);
1046                                 btrfs_mark_buffer_dirty(leaf);
1047                         } else {
1048                                 extent_num_bytes =
1049                                         btrfs_file_extent_disk_num_bytes(leaf,
1050                                                                          fi);
1051                                 /* FIXME blocksize != 4096 */
1052                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1053                                 if (extent_start != 0) {
1054                                         found_extent = 1;
1055                                         dec_i_blocks(inode, num_dec);
1056                                 }
1057                                 root_gen = btrfs_header_generation(leaf);
1058                                 root_owner = btrfs_header_owner(leaf);
1059                         }
1060                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1061                         if (!del_item) {
1062                                 u32 newsize = inode->i_size - found_key.offset;
1063                                 dec_i_blocks(inode, item_end + 1 -
1064                                             found_key.offset - newsize);
1065                                 newsize =
1066                                     btrfs_file_extent_calc_inline_size(newsize);
1067                                 ret = btrfs_truncate_item(trans, root, path,
1068                                                           newsize, 1);
1069                                 BUG_ON(ret);
1070                         } else {
1071                                 dec_i_blocks(inode, item_end + 1 -
1072                                              found_key.offset);
1073                         }
1074                 }
1075 delete:
1076                 if (del_item) {
1077                         if (!pending_del_nr) {
1078                                 /* no pending yet, add ourselves */
1079                                 pending_del_slot = path->slots[0];
1080                                 pending_del_nr = 1;
1081                         } else if (pending_del_nr &&
1082                                    path->slots[0] + 1 == pending_del_slot) {
1083                                 /* hop on the pending chunk */
1084                                 pending_del_nr++;
1085                                 pending_del_slot = path->slots[0];
1086                         } else {
1087                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1088                         }
1089                 } else {
1090                         break;
1091                 }
1092                 if (found_extent) {
1093                         ret = btrfs_free_extent(trans, root, extent_start,
1094                                                 extent_num_bytes,
1095                                                 root_owner,
1096                                                 root_gen, inode->i_ino,
1097                                                 found_key.offset, 0);
1098                         BUG_ON(ret);
1099                 }
1100 next:
1101                 if (path->slots[0] == 0) {
1102                         if (pending_del_nr)
1103                                 goto del_pending;
1104                         btrfs_release_path(root, path);
1105                         goto search_again;
1106                 }
1107
1108                 path->slots[0]--;
1109                 if (pending_del_nr &&
1110                     path->slots[0] + 1 != pending_del_slot) {
1111                         struct btrfs_key debug;
1112 del_pending:
1113                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1114                                               pending_del_slot);
1115                         ret = btrfs_del_items(trans, root, path,
1116                                               pending_del_slot,
1117                                               pending_del_nr);
1118                         BUG_ON(ret);
1119                         pending_del_nr = 0;
1120                         btrfs_release_path(root, path);
1121                         goto search_again;
1122                 }
1123         }
1124         ret = 0;
1125 error:
1126         if (pending_del_nr) {
1127                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1128                                       pending_del_nr);
1129         }
1130         btrfs_release_path(root, path);
1131         btrfs_free_path(path);
1132         inode->i_sb->s_dirt = 1;
1133         return ret;
1134 }
1135
1136 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
1137                               size_t zero_start)
1138 {
1139         char *kaddr;
1140         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1141         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1142         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1143         int ret = 0;
1144
1145         WARN_ON(!PageLocked(page));
1146         set_page_extent_mapped(page);
1147
1148         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1149         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
1150                             page_end, GFP_NOFS);
1151
1152         if (zero_start != PAGE_CACHE_SIZE) {
1153                 kaddr = kmap(page);
1154                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
1155                 flush_dcache_page(page);
1156                 kunmap(page);
1157         }
1158         set_page_dirty(page);
1159         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1160
1161         return ret;
1162 }
1163
1164 /*
1165  * taken from block_truncate_page, but does cow as it zeros out
1166  * any bytes left in the last page in the file.
1167  */
1168 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1169 {
1170         struct inode *inode = mapping->host;
1171         struct btrfs_root *root = BTRFS_I(inode)->root;
1172         u32 blocksize = root->sectorsize;
1173         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1174         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1175         struct page *page;
1176         int ret = 0;
1177         u64 page_start;
1178
1179         if ((offset & (blocksize - 1)) == 0)
1180                 goto out;
1181
1182         ret = -ENOMEM;
1183         page = grab_cache_page(mapping, index);
1184         if (!page)
1185                 goto out;
1186         if (!PageUptodate(page)) {
1187                 ret = btrfs_readpage(NULL, page);
1188                 lock_page(page);
1189                 if (!PageUptodate(page)) {
1190                         ret = -EIO;
1191                         goto out;
1192                 }
1193         }
1194         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1195
1196         ret = btrfs_cow_one_page(inode, page, offset);
1197
1198         unlock_page(page);
1199         page_cache_release(page);
1200 out:
1201         return ret;
1202 }
1203
1204 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1205 {
1206         struct inode *inode = dentry->d_inode;
1207         int err;
1208
1209         err = inode_change_ok(inode, attr);
1210         if (err)
1211                 return err;
1212
1213         if (S_ISREG(inode->i_mode) &&
1214             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1215                 struct btrfs_trans_handle *trans;
1216                 struct btrfs_root *root = BTRFS_I(inode)->root;
1217                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1218
1219                 u64 mask = root->sectorsize - 1;
1220                 u64 hole_start = (inode->i_size + mask) & ~mask;
1221                 u64 block_end = (attr->ia_size + mask) & ~mask;
1222                 u64 hole_size;
1223                 u64 alloc_hint = 0;
1224
1225                 if (attr->ia_size <= hole_start)
1226                         goto out;
1227
1228                 mutex_lock(&root->fs_info->fs_mutex);
1229                 err = btrfs_check_free_space(root, 1, 0);
1230                 mutex_unlock(&root->fs_info->fs_mutex);
1231                 if (err)
1232                         goto fail;
1233
1234                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1235
1236                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1237                 hole_size = block_end - hole_start;
1238
1239                 mutex_lock(&root->fs_info->fs_mutex);
1240                 trans = btrfs_start_transaction(root, 1);
1241                 btrfs_set_trans_block_group(trans, inode);
1242                 err = btrfs_drop_extents(trans, root, inode,
1243                                          hole_start, block_end, hole_start,
1244                                          &alloc_hint);
1245
1246                 if (alloc_hint != EXTENT_MAP_INLINE) {
1247                         err = btrfs_insert_file_extent(trans, root,
1248                                                        inode->i_ino,
1249                                                        hole_start, 0, 0,
1250                                                        hole_size, 0);
1251                         btrfs_drop_extent_cache(inode, hole_start,
1252                                                 (u64)-1);
1253                         btrfs_check_file(root, inode);
1254                 }
1255                 btrfs_end_transaction(trans, root);
1256                 mutex_unlock(&root->fs_info->fs_mutex);
1257                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1258                 if (err)
1259                         return err;
1260         }
1261 out:
1262         err = inode_setattr(inode, attr);
1263 fail:
1264         return err;
1265 }
1266
1267 void btrfs_put_inode(struct inode *inode)
1268 {
1269         int ret;
1270
1271         if (!BTRFS_I(inode)->ordered_trans) {
1272                 return;
1273         }
1274
1275         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1276             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1277                 return;
1278
1279         ret = btrfs_del_ordered_inode(inode);
1280         if (ret == 1) {
1281                 atomic_dec(&inode->i_count);
1282         }
1283 }
1284
1285 void btrfs_delete_inode(struct inode *inode)
1286 {
1287         struct btrfs_trans_handle *trans;
1288         struct btrfs_root *root = BTRFS_I(inode)->root;
1289         unsigned long nr;
1290         int ret;
1291
1292         truncate_inode_pages(&inode->i_data, 0);
1293         if (is_bad_inode(inode)) {
1294                 goto no_delete;
1295         }
1296
1297         inode->i_size = 0;
1298         mutex_lock(&root->fs_info->fs_mutex);
1299         trans = btrfs_start_transaction(root, 1);
1300
1301         btrfs_set_trans_block_group(trans, inode);
1302         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1303         if (ret)
1304                 goto no_delete_lock;
1305
1306         nr = trans->blocks_used;
1307         clear_inode(inode);
1308
1309         btrfs_end_transaction(trans, root);
1310         mutex_unlock(&root->fs_info->fs_mutex);
1311         btrfs_btree_balance_dirty(root, nr);
1312         btrfs_throttle(root);
1313         return;
1314
1315 no_delete_lock:
1316         nr = trans->blocks_used;
1317         btrfs_end_transaction(trans, root);
1318         mutex_unlock(&root->fs_info->fs_mutex);
1319         btrfs_btree_balance_dirty(root, nr);
1320         btrfs_throttle(root);
1321 no_delete:
1322         clear_inode(inode);
1323 }
1324
1325 /*
1326  * this returns the key found in the dir entry in the location pointer.
1327  * If no dir entries were found, location->objectid is 0.
1328  */
1329 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1330                                struct btrfs_key *location)
1331 {
1332         const char *name = dentry->d_name.name;
1333         int namelen = dentry->d_name.len;
1334         struct btrfs_dir_item *di;
1335         struct btrfs_path *path;
1336         struct btrfs_root *root = BTRFS_I(dir)->root;
1337         int ret = 0;
1338
1339         if (namelen == 1 && strcmp(name, ".") == 0) {
1340                 location->objectid = dir->i_ino;
1341                 location->type = BTRFS_INODE_ITEM_KEY;
1342                 location->offset = 0;
1343                 return 0;
1344         }
1345         path = btrfs_alloc_path();
1346         BUG_ON(!path);
1347
1348         if (namelen == 2 && strcmp(name, "..") == 0) {
1349                 struct btrfs_key key;
1350                 struct extent_buffer *leaf;
1351                 u32 nritems;
1352                 int slot;
1353
1354                 key.objectid = dir->i_ino;
1355                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1356                 key.offset = 0;
1357                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1358                 BUG_ON(ret == 0);
1359                 ret = 0;
1360
1361                 leaf = path->nodes[0];
1362                 slot = path->slots[0];
1363                 nritems = btrfs_header_nritems(leaf);
1364                 if (slot >= nritems)
1365                         goto out_err;
1366
1367                 btrfs_item_key_to_cpu(leaf, &key, slot);
1368                 if (key.objectid != dir->i_ino ||
1369                     key.type != BTRFS_INODE_REF_KEY) {
1370                         goto out_err;
1371                 }
1372                 location->objectid = key.offset;
1373                 location->type = BTRFS_INODE_ITEM_KEY;
1374                 location->offset = 0;
1375                 goto out;
1376         }
1377
1378         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1379                                     namelen, 0);
1380         if (IS_ERR(di))
1381                 ret = PTR_ERR(di);
1382         if (!di || IS_ERR(di)) {
1383                 goto out_err;
1384         }
1385         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1386 out:
1387         btrfs_free_path(path);
1388         return ret;
1389 out_err:
1390         location->objectid = 0;
1391         goto out;
1392 }
1393
1394 /*
1395  * when we hit a tree root in a directory, the btrfs part of the inode
1396  * needs to be changed to reflect the root directory of the tree root.  This
1397  * is kind of like crossing a mount point.
1398  */
1399 static int fixup_tree_root_location(struct btrfs_root *root,
1400                              struct btrfs_key *location,
1401                              struct btrfs_root **sub_root,
1402                              struct dentry *dentry)
1403 {
1404         struct btrfs_path *path;
1405         struct btrfs_root_item *ri;
1406
1407         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1408                 return 0;
1409         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1410                 return 0;
1411
1412         path = btrfs_alloc_path();
1413         BUG_ON(!path);
1414         mutex_lock(&root->fs_info->fs_mutex);
1415
1416         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1417                                         dentry->d_name.name,
1418                                         dentry->d_name.len);
1419         if (IS_ERR(*sub_root))
1420                 return PTR_ERR(*sub_root);
1421
1422         ri = &(*sub_root)->root_item;
1423         location->objectid = btrfs_root_dirid(ri);
1424         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1425         location->offset = 0;
1426
1427         btrfs_free_path(path);
1428         mutex_unlock(&root->fs_info->fs_mutex);
1429         return 0;
1430 }
1431
1432 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1433 {
1434         struct btrfs_iget_args *args = p;
1435         inode->i_ino = args->ino;
1436         BTRFS_I(inode)->root = args->root;
1437         BTRFS_I(inode)->delalloc_bytes = 0;
1438         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1439         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1440                              inode->i_mapping, GFP_NOFS);
1441         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1442                              inode->i_mapping, GFP_NOFS);
1443         atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1444         return 0;
1445 }
1446
1447 static int btrfs_find_actor(struct inode *inode, void *opaque)
1448 {
1449         struct btrfs_iget_args *args = opaque;
1450         return (args->ino == inode->i_ino &&
1451                 args->root == BTRFS_I(inode)->root);
1452 }
1453
1454 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1455                             u64 root_objectid)
1456 {
1457         struct btrfs_iget_args args;
1458         args.ino = objectid;
1459         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1460
1461         if (!args.root)
1462                 return NULL;
1463
1464         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1465 }
1466
1467 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1468                                 struct btrfs_root *root)
1469 {
1470         struct inode *inode;
1471         struct btrfs_iget_args args;
1472         args.ino = objectid;
1473         args.root = root;
1474
1475         inode = iget5_locked(s, objectid, btrfs_find_actor,
1476                              btrfs_init_locked_inode,
1477                              (void *)&args);
1478         return inode;
1479 }
1480
1481 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1482                                    struct nameidata *nd)
1483 {
1484         struct inode * inode;
1485         struct btrfs_inode *bi = BTRFS_I(dir);
1486         struct btrfs_root *root = bi->root;
1487         struct btrfs_root *sub_root = root;
1488         struct btrfs_key location;
1489         int ret;
1490
1491         if (dentry->d_name.len > BTRFS_NAME_LEN)
1492                 return ERR_PTR(-ENAMETOOLONG);
1493
1494         mutex_lock(&root->fs_info->fs_mutex);
1495         ret = btrfs_inode_by_name(dir, dentry, &location);
1496         mutex_unlock(&root->fs_info->fs_mutex);
1497
1498         if (ret < 0)
1499                 return ERR_PTR(ret);
1500
1501         inode = NULL;
1502         if (location.objectid) {
1503                 ret = fixup_tree_root_location(root, &location, &sub_root,
1504                                                 dentry);
1505                 if (ret < 0)
1506                         return ERR_PTR(ret);
1507                 if (ret > 0)
1508                         return ERR_PTR(-ENOENT);
1509                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1510                                           sub_root);
1511                 if (!inode)
1512                         return ERR_PTR(-EACCES);
1513                 if (inode->i_state & I_NEW) {
1514                         /* the inode and parent dir are two different roots */
1515                         if (sub_root != root) {
1516                                 igrab(inode);
1517                                 sub_root->inode = inode;
1518                         }
1519                         BTRFS_I(inode)->root = sub_root;
1520                         memcpy(&BTRFS_I(inode)->location, &location,
1521                                sizeof(location));
1522                         btrfs_read_locked_inode(inode);
1523                         unlock_new_inode(inode);
1524                 }
1525         }
1526         return d_splice_alias(inode, dentry);
1527 }
1528
1529 static unsigned char btrfs_filetype_table[] = {
1530         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1531 };
1532
1533 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1534 {
1535         struct inode *inode = filp->f_dentry->d_inode;
1536         struct btrfs_root *root = BTRFS_I(inode)->root;
1537         struct btrfs_item *item;
1538         struct btrfs_dir_item *di;
1539         struct btrfs_key key;
1540         struct btrfs_key found_key;
1541         struct btrfs_path *path;
1542         int ret;
1543         u32 nritems;
1544         struct extent_buffer *leaf;
1545         int slot;
1546         int advance;
1547         unsigned char d_type;
1548         int over = 0;
1549         u32 di_cur;
1550         u32 di_total;
1551         u32 di_len;
1552         int key_type = BTRFS_DIR_INDEX_KEY;
1553         char tmp_name[32];
1554         char *name_ptr;
1555         int name_len;
1556
1557         /* FIXME, use a real flag for deciding about the key type */
1558         if (root->fs_info->tree_root == root)
1559                 key_type = BTRFS_DIR_ITEM_KEY;
1560
1561         /* special case for "." */
1562         if (filp->f_pos == 0) {
1563                 over = filldir(dirent, ".", 1,
1564                                1, inode->i_ino,
1565                                DT_DIR);
1566                 if (over)
1567                         return 0;
1568                 filp->f_pos = 1;
1569         }
1570
1571         mutex_lock(&root->fs_info->fs_mutex);
1572         key.objectid = inode->i_ino;
1573         path = btrfs_alloc_path();
1574         path->reada = 2;
1575
1576         /* special case for .., just use the back ref */
1577         if (filp->f_pos == 1) {
1578                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1579                 key.offset = 0;
1580                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1581                 BUG_ON(ret == 0);
1582                 leaf = path->nodes[0];
1583                 slot = path->slots[0];
1584                 nritems = btrfs_header_nritems(leaf);
1585                 if (slot >= nritems) {
1586                         btrfs_release_path(root, path);
1587                         goto read_dir_items;
1588                 }
1589                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1590                 btrfs_release_path(root, path);
1591                 if (found_key.objectid != key.objectid ||
1592                     found_key.type != BTRFS_INODE_REF_KEY)
1593                         goto read_dir_items;
1594                 over = filldir(dirent, "..", 2,
1595                                2, found_key.offset, DT_DIR);
1596                 if (over)
1597                         goto nopos;
1598                 filp->f_pos = 2;
1599         }
1600
1601 read_dir_items:
1602         btrfs_set_key_type(&key, key_type);
1603         key.offset = filp->f_pos;
1604
1605         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1606         if (ret < 0)
1607                 goto err;
1608         advance = 0;
1609         while(1) {
1610                 leaf = path->nodes[0];
1611                 nritems = btrfs_header_nritems(leaf);
1612                 slot = path->slots[0];
1613                 if (advance || slot >= nritems) {
1614                         if (slot >= nritems -1) {
1615                                 ret = btrfs_next_leaf(root, path);
1616                                 if (ret)
1617                                         break;
1618                                 leaf = path->nodes[0];
1619                                 nritems = btrfs_header_nritems(leaf);
1620                                 slot = path->slots[0];
1621                         } else {
1622                                 slot++;
1623                                 path->slots[0]++;
1624                         }
1625                 }
1626                 advance = 1;
1627                 item = btrfs_item_nr(leaf, slot);
1628                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1629
1630                 if (found_key.objectid != key.objectid)
1631                         break;
1632                 if (btrfs_key_type(&found_key) != key_type)
1633                         break;
1634                 if (found_key.offset < filp->f_pos)
1635                         continue;
1636
1637                 filp->f_pos = found_key.offset;
1638                 advance = 1;
1639                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1640                 di_cur = 0;
1641                 di_total = btrfs_item_size(leaf, item);
1642                 while(di_cur < di_total) {
1643                         struct btrfs_key location;
1644
1645                         name_len = btrfs_dir_name_len(leaf, di);
1646                         if (name_len < 32) {
1647                                 name_ptr = tmp_name;
1648                         } else {
1649                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1650                                 BUG_ON(!name_ptr);
1651                         }
1652                         read_extent_buffer(leaf, name_ptr,
1653                                            (unsigned long)(di + 1), name_len);
1654
1655                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1656                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1657                         over = filldir(dirent, name_ptr, name_len,
1658                                        found_key.offset,
1659                                        location.objectid,
1660                                        d_type);
1661
1662                         if (name_ptr != tmp_name)
1663                                 kfree(name_ptr);
1664
1665                         if (over)
1666                                 goto nopos;
1667                         di_len = btrfs_dir_name_len(leaf, di) +
1668                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1669                         di_cur += di_len;
1670                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1671                 }
1672         }
1673         if (key_type == BTRFS_DIR_INDEX_KEY)
1674                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1675         else
1676                 filp->f_pos++;
1677 nopos:
1678         ret = 0;
1679 err:
1680         btrfs_release_path(root, path);
1681         btrfs_free_path(path);
1682         mutex_unlock(&root->fs_info->fs_mutex);
1683         return ret;
1684 }
1685
1686 int btrfs_write_inode(struct inode *inode, int wait)
1687 {
1688         struct btrfs_root *root = BTRFS_I(inode)->root;
1689         struct btrfs_trans_handle *trans;
1690         int ret = 0;
1691
1692         if (wait) {
1693                 mutex_lock(&root->fs_info->fs_mutex);
1694                 trans = btrfs_start_transaction(root, 1);
1695                 btrfs_set_trans_block_group(trans, inode);
1696                 ret = btrfs_commit_transaction(trans, root);
1697                 mutex_unlock(&root->fs_info->fs_mutex);
1698         }
1699         return ret;
1700 }
1701
1702 /*
1703  * This is somewhat expensive, updating the tree every time the
1704  * inode changes.  But, it is most likely to find the inode in cache.
1705  * FIXME, needs more benchmarking...there are no reasons other than performance
1706  * to keep or drop this code.
1707  */
1708 void btrfs_dirty_inode(struct inode *inode)
1709 {
1710         struct btrfs_root *root = BTRFS_I(inode)->root;
1711         struct btrfs_trans_handle *trans;
1712
1713         mutex_lock(&root->fs_info->fs_mutex);
1714         trans = btrfs_start_transaction(root, 1);
1715         btrfs_set_trans_block_group(trans, inode);
1716         btrfs_update_inode(trans, root, inode);
1717         btrfs_end_transaction(trans, root);
1718         mutex_unlock(&root->fs_info->fs_mutex);
1719 }
1720
1721 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1722                                      struct btrfs_root *root,
1723                                      const char *name, int name_len,
1724                                      u64 ref_objectid,
1725                                      u64 objectid,
1726                                      struct btrfs_block_group_cache *group,
1727                                      int mode)
1728 {
1729         struct inode *inode;
1730         struct btrfs_inode_item *inode_item;
1731         struct btrfs_block_group_cache *new_inode_group;
1732         struct btrfs_key *location;
1733         struct btrfs_path *path;
1734         struct btrfs_inode_ref *ref;
1735         struct btrfs_key key[2];
1736         u32 sizes[2];
1737         unsigned long ptr;
1738         int ret;
1739         int owner;
1740
1741         path = btrfs_alloc_path();
1742         BUG_ON(!path);
1743
1744         inode = new_inode(root->fs_info->sb);
1745         if (!inode)
1746                 return ERR_PTR(-ENOMEM);
1747
1748         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1749         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1750                              inode->i_mapping, GFP_NOFS);
1751         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1752                              inode->i_mapping, GFP_NOFS);
1753         atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1754         BTRFS_I(inode)->delalloc_bytes = 0;
1755         BTRFS_I(inode)->root = root;
1756
1757         if (mode & S_IFDIR)
1758                 owner = 0;
1759         else
1760                 owner = 1;
1761         new_inode_group = btrfs_find_block_group(root, group, 0,
1762                                        BTRFS_BLOCK_GROUP_METADATA, owner);
1763         if (!new_inode_group) {
1764                 printk("find_block group failed\n");
1765                 new_inode_group = group;
1766         }
1767         BTRFS_I(inode)->block_group = new_inode_group;
1768         BTRFS_I(inode)->flags = 0;
1769
1770         key[0].objectid = objectid;
1771         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1772         key[0].offset = 0;
1773
1774         key[1].objectid = objectid;
1775         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1776         key[1].offset = ref_objectid;
1777
1778         sizes[0] = sizeof(struct btrfs_inode_item);
1779         sizes[1] = name_len + sizeof(*ref);
1780
1781         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1782         if (ret != 0)
1783                 goto fail;
1784
1785         if (objectid > root->highest_inode)
1786                 root->highest_inode = objectid;
1787
1788         inode->i_uid = current->fsuid;
1789         inode->i_gid = current->fsgid;
1790         inode->i_mode = mode;
1791         inode->i_ino = objectid;
1792         inode->i_blocks = 0;
1793         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1794         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1795                                   struct btrfs_inode_item);
1796         fill_inode_item(path->nodes[0], inode_item, inode);
1797
1798         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1799                              struct btrfs_inode_ref);
1800         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1801         ptr = (unsigned long)(ref + 1);
1802         write_extent_buffer(path->nodes[0], name, ptr, name_len);
1803
1804         btrfs_mark_buffer_dirty(path->nodes[0]);
1805         btrfs_free_path(path);
1806
1807         location = &BTRFS_I(inode)->location;
1808         location->objectid = objectid;
1809         location->offset = 0;
1810         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1811
1812         insert_inode_hash(inode);
1813         return inode;
1814 fail:
1815         btrfs_free_path(path);
1816         return ERR_PTR(ret);
1817 }
1818
1819 static inline u8 btrfs_inode_type(struct inode *inode)
1820 {
1821         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1822 }
1823
1824 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1825                             struct dentry *dentry, struct inode *inode,
1826                             int add_backref)
1827 {
1828         int ret;
1829         struct btrfs_key key;
1830         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1831         struct inode *parent_inode;
1832
1833         key.objectid = inode->i_ino;
1834         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1835         key.offset = 0;
1836
1837         ret = btrfs_insert_dir_item(trans, root,
1838                                     dentry->d_name.name, dentry->d_name.len,
1839                                     dentry->d_parent->d_inode->i_ino,
1840                                     &key, btrfs_inode_type(inode));
1841         if (ret == 0) {
1842                 if (add_backref) {
1843                         ret = btrfs_insert_inode_ref(trans, root,
1844                                              dentry->d_name.name,
1845                                              dentry->d_name.len,
1846                                              inode->i_ino,
1847                                              dentry->d_parent->d_inode->i_ino);
1848                 }
1849                 parent_inode = dentry->d_parent->d_inode;
1850                 parent_inode->i_size += dentry->d_name.len * 2;
1851                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1852                 ret = btrfs_update_inode(trans, root,
1853                                          dentry->d_parent->d_inode);
1854         }
1855         return ret;
1856 }
1857
1858 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1859                             struct dentry *dentry, struct inode *inode,
1860                             int backref)
1861 {
1862         int err = btrfs_add_link(trans, dentry, inode, backref);
1863         if (!err) {
1864                 d_instantiate(dentry, inode);
1865                 return 0;
1866         }
1867         if (err > 0)
1868                 err = -EEXIST;
1869         return err;
1870 }
1871
1872 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1873                         int mode, dev_t rdev)
1874 {
1875         struct btrfs_trans_handle *trans;
1876         struct btrfs_root *root = BTRFS_I(dir)->root;
1877         struct inode *inode = NULL;
1878         int err;
1879         int drop_inode = 0;
1880         u64 objectid;
1881         unsigned long nr = 0;
1882
1883         if (!new_valid_dev(rdev))
1884                 return -EINVAL;
1885
1886         mutex_lock(&root->fs_info->fs_mutex);
1887         err = btrfs_check_free_space(root, 1, 0);
1888         if (err)
1889                 goto fail;
1890
1891         trans = btrfs_start_transaction(root, 1);
1892         btrfs_set_trans_block_group(trans, dir);
1893
1894         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1895         if (err) {
1896                 err = -ENOSPC;
1897                 goto out_unlock;
1898         }
1899
1900         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1901                                 dentry->d_name.len,
1902                                 dentry->d_parent->d_inode->i_ino, objectid,
1903                                 BTRFS_I(dir)->block_group, mode);
1904         err = PTR_ERR(inode);
1905         if (IS_ERR(inode))
1906                 goto out_unlock;
1907
1908         btrfs_set_trans_block_group(trans, inode);
1909         err = btrfs_add_nondir(trans, dentry, inode, 0);
1910         if (err)
1911                 drop_inode = 1;
1912         else {
1913                 inode->i_op = &btrfs_special_inode_operations;
1914                 init_special_inode(inode, inode->i_mode, rdev);
1915                 btrfs_update_inode(trans, root, inode);
1916         }
1917         dir->i_sb->s_dirt = 1;
1918         btrfs_update_inode_block_group(trans, inode);
1919         btrfs_update_inode_block_group(trans, dir);
1920 out_unlock:
1921         nr = trans->blocks_used;
1922         btrfs_end_transaction(trans, root);
1923 fail:
1924         mutex_unlock(&root->fs_info->fs_mutex);
1925
1926         if (drop_inode) {
1927                 inode_dec_link_count(inode);
1928                 iput(inode);
1929         }
1930         btrfs_btree_balance_dirty(root, nr);
1931         btrfs_throttle(root);
1932         return err;
1933 }
1934
1935 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1936                         int mode, struct nameidata *nd)
1937 {
1938         struct btrfs_trans_handle *trans;
1939         struct btrfs_root *root = BTRFS_I(dir)->root;
1940         struct inode *inode = NULL;
1941         int err;
1942         int drop_inode = 0;
1943         unsigned long nr = 0;
1944         u64 objectid;
1945
1946         mutex_lock(&root->fs_info->fs_mutex);
1947         err = btrfs_check_free_space(root, 1, 0);
1948         if (err)
1949                 goto fail;
1950         trans = btrfs_start_transaction(root, 1);
1951         btrfs_set_trans_block_group(trans, dir);
1952
1953         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1954         if (err) {
1955                 err = -ENOSPC;
1956                 goto out_unlock;
1957         }
1958
1959         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1960                                 dentry->d_name.len,
1961                                 dentry->d_parent->d_inode->i_ino,
1962                                 objectid, BTRFS_I(dir)->block_group, mode);
1963         err = PTR_ERR(inode);
1964         if (IS_ERR(inode))
1965                 goto out_unlock;
1966
1967         btrfs_set_trans_block_group(trans, inode);
1968         err = btrfs_add_nondir(trans, dentry, inode, 0);
1969         if (err)
1970                 drop_inode = 1;
1971         else {
1972                 inode->i_mapping->a_ops = &btrfs_aops;
1973                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1974                 inode->i_fop = &btrfs_file_operations;
1975                 inode->i_op = &btrfs_file_inode_operations;
1976                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1977                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1978                                      inode->i_mapping, GFP_NOFS);
1979                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1980                                      inode->i_mapping, GFP_NOFS);
1981                 BTRFS_I(inode)->delalloc_bytes = 0;
1982                 atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1983                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1984         }
1985         dir->i_sb->s_dirt = 1;
1986         btrfs_update_inode_block_group(trans, inode);
1987         btrfs_update_inode_block_group(trans, dir);
1988 out_unlock:
1989         nr = trans->blocks_used;
1990         btrfs_end_transaction(trans, root);
1991 fail:
1992         mutex_unlock(&root->fs_info->fs_mutex);
1993
1994         if (drop_inode) {
1995                 inode_dec_link_count(inode);
1996                 iput(inode);
1997         }
1998         btrfs_btree_balance_dirty(root, nr);
1999         btrfs_throttle(root);
2000         return err;
2001 }
2002
2003 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2004                       struct dentry *dentry)
2005 {
2006         struct btrfs_trans_handle *trans;
2007         struct btrfs_root *root = BTRFS_I(dir)->root;
2008         struct inode *inode = old_dentry->d_inode;
2009         unsigned long nr = 0;
2010         int err;
2011         int drop_inode = 0;
2012
2013         if (inode->i_nlink == 0)
2014                 return -ENOENT;
2015
2016 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2017         inode->i_nlink++;
2018 #else
2019         inc_nlink(inode);
2020 #endif
2021         mutex_lock(&root->fs_info->fs_mutex);
2022         err = btrfs_check_free_space(root, 1, 0);
2023         if (err)
2024                 goto fail;
2025         trans = btrfs_start_transaction(root, 1);
2026
2027         btrfs_set_trans_block_group(trans, dir);
2028         atomic_inc(&inode->i_count);
2029         err = btrfs_add_nondir(trans, dentry, inode, 1);
2030
2031         if (err)
2032                 drop_inode = 1;
2033
2034         dir->i_sb->s_dirt = 1;
2035         btrfs_update_inode_block_group(trans, dir);
2036         err = btrfs_update_inode(trans, root, inode);
2037
2038         if (err)
2039                 drop_inode = 1;
2040
2041         nr = trans->blocks_used;
2042         btrfs_end_transaction(trans, root);
2043 fail:
2044         mutex_unlock(&root->fs_info->fs_mutex);
2045
2046         if (drop_inode) {
2047                 inode_dec_link_count(inode);
2048                 iput(inode);
2049         }
2050         btrfs_btree_balance_dirty(root, nr);
2051         btrfs_throttle(root);
2052         return err;
2053 }
2054
2055 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2056 {
2057         struct inode *inode = NULL;
2058         struct btrfs_trans_handle *trans;
2059         struct btrfs_root *root = BTRFS_I(dir)->root;
2060         int err = 0;
2061         int drop_on_err = 0;
2062         u64 objectid = 0;
2063         unsigned long nr = 1;
2064
2065         mutex_lock(&root->fs_info->fs_mutex);
2066         err = btrfs_check_free_space(root, 1, 0);
2067         if (err)
2068                 goto out_unlock;
2069
2070         trans = btrfs_start_transaction(root, 1);
2071         btrfs_set_trans_block_group(trans, dir);
2072
2073         if (IS_ERR(trans)) {
2074                 err = PTR_ERR(trans);
2075                 goto out_unlock;
2076         }
2077
2078         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2079         if (err) {
2080                 err = -ENOSPC;
2081                 goto out_unlock;
2082         }
2083
2084         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2085                                 dentry->d_name.len,
2086                                 dentry->d_parent->d_inode->i_ino, objectid,
2087                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
2088         if (IS_ERR(inode)) {
2089                 err = PTR_ERR(inode);
2090                 goto out_fail;
2091         }
2092
2093         drop_on_err = 1;
2094         inode->i_op = &btrfs_dir_inode_operations;
2095         inode->i_fop = &btrfs_dir_file_operations;
2096         btrfs_set_trans_block_group(trans, inode);
2097
2098         inode->i_size = 0;
2099         err = btrfs_update_inode(trans, root, inode);
2100         if (err)
2101                 goto out_fail;
2102
2103         err = btrfs_add_link(trans, dentry, inode, 0);
2104         if (err)
2105                 goto out_fail;
2106
2107         d_instantiate(dentry, inode);
2108         drop_on_err = 0;
2109         dir->i_sb->s_dirt = 1;
2110         btrfs_update_inode_block_group(trans, inode);
2111         btrfs_update_inode_block_group(trans, dir);
2112
2113 out_fail:
2114         nr = trans->blocks_used;
2115         btrfs_end_transaction(trans, root);
2116
2117 out_unlock:
2118         mutex_unlock(&root->fs_info->fs_mutex);
2119         if (drop_on_err)
2120                 iput(inode);
2121         btrfs_btree_balance_dirty(root, nr);
2122         btrfs_throttle(root);
2123         return err;
2124 }
2125
2126 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2127                                 struct extent_map *existing,
2128                                 struct extent_map *em)
2129 {
2130         u64 start_diff;
2131         u64 new_end;
2132         int ret = 0;
2133         int real_blocks = existing->block_start < EXTENT_MAP_LAST_BYTE;
2134
2135         if (real_blocks && em->block_start >= EXTENT_MAP_LAST_BYTE)
2136                 goto invalid;
2137
2138         if (!real_blocks && em->block_start != existing->block_start)
2139                 goto invalid;
2140
2141         new_end = max(existing->start + existing->len, em->start + em->len);
2142
2143         if (existing->start >= em->start) {
2144                 if (em->start + em->len < existing->start)
2145                         goto invalid;
2146
2147                 start_diff = existing->start - em->start;
2148                 if (real_blocks && em->block_start + start_diff !=
2149                     existing->block_start)
2150                         goto invalid;
2151
2152                 em->len = new_end - em->start;
2153
2154                 remove_extent_mapping(em_tree, existing);
2155                 /* free for the tree */
2156                 free_extent_map(existing);
2157                 ret = add_extent_mapping(em_tree, em);
2158
2159         } else if (em->start > existing->start) {
2160
2161                 if (existing->start + existing->len < em->start)
2162                         goto invalid;
2163
2164                 start_diff = em->start - existing->start;
2165                 if (real_blocks && existing->block_start + start_diff !=
2166                     em->block_start)
2167                         goto invalid;
2168
2169                 remove_extent_mapping(em_tree, existing);
2170                 em->block_start = existing->block_start;
2171                 em->start = existing->start;
2172                 em->len = new_end - existing->start;
2173                 free_extent_map(existing);
2174
2175                 ret = add_extent_mapping(em_tree, em);
2176         } else {
2177                 goto invalid;
2178         }
2179         return ret;
2180
2181 invalid:
2182         printk("invalid extent map merge [%Lu %Lu %Lu] [%Lu %Lu %Lu]\n",
2183                existing->start, existing->len, existing->block_start,
2184                em->start, em->len, em->block_start);
2185         return -EIO;
2186 }
2187
2188 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2189                                     size_t pg_offset, u64 start, u64 len,
2190                                     int create)
2191 {
2192         int ret;
2193         int err = 0;
2194         u64 bytenr;
2195         u64 extent_start = 0;
2196         u64 extent_end = 0;
2197         u64 objectid = inode->i_ino;
2198         u32 found_type;
2199         struct btrfs_path *path;
2200         struct btrfs_root *root = BTRFS_I(inode)->root;
2201         struct btrfs_file_extent_item *item;
2202         struct extent_buffer *leaf;
2203         struct btrfs_key found_key;
2204         struct extent_map *em = NULL;
2205         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2206         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2207         struct btrfs_trans_handle *trans = NULL;
2208
2209         path = btrfs_alloc_path();
2210         BUG_ON(!path);
2211         mutex_lock(&root->fs_info->fs_mutex);
2212
2213 again:
2214         spin_lock(&em_tree->lock);
2215         em = lookup_extent_mapping(em_tree, start, len);
2216         if (em)
2217                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2218         spin_unlock(&em_tree->lock);
2219
2220         if (em) {
2221                 if (em->start > start || em->start + em->len <= start)
2222                         free_extent_map(em);
2223                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2224                         free_extent_map(em);
2225                 else
2226                         goto out;
2227         }
2228         em = alloc_extent_map(GFP_NOFS);
2229         if (!em) {
2230                 err = -ENOMEM;
2231                 goto out;
2232         }
2233
2234         em->start = EXTENT_MAP_HOLE;
2235         em->len = (u64)-1;
2236         em->bdev = root->fs_info->fs_devices->latest_bdev;
2237         ret = btrfs_lookup_file_extent(trans, root, path,
2238                                        objectid, start, trans != NULL);
2239         if (ret < 0) {
2240                 err = ret;
2241                 goto out;
2242         }
2243
2244         if (ret != 0) {
2245                 if (path->slots[0] == 0)
2246                         goto not_found;
2247                 path->slots[0]--;
2248         }
2249
2250         leaf = path->nodes[0];
2251         item = btrfs_item_ptr(leaf, path->slots[0],
2252                               struct btrfs_file_extent_item);
2253         /* are we inside the extent that was found? */
2254         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2255         found_type = btrfs_key_type(&found_key);
2256         if (found_key.objectid != objectid ||
2257             found_type != BTRFS_EXTENT_DATA_KEY) {
2258                 goto not_found;
2259         }
2260
2261         found_type = btrfs_file_extent_type(leaf, item);
2262         extent_start = found_key.offset;
2263         if (found_type == BTRFS_FILE_EXTENT_REG) {
2264                 extent_end = extent_start +
2265                        btrfs_file_extent_num_bytes(leaf, item);
2266                 err = 0;
2267                 if (start < extent_start || start >= extent_end) {
2268                         em->start = start;
2269                         if (start < extent_start) {
2270                                 if (start + len <= extent_start)
2271                                         goto not_found;
2272                                 em->len = extent_end - extent_start;
2273                         } else {
2274                                 em->len = len;
2275                         }
2276                         goto not_found_em;
2277                 }
2278                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2279                 if (bytenr == 0) {
2280                         em->start = extent_start;
2281                         em->len = extent_end - extent_start;
2282                         em->block_start = EXTENT_MAP_HOLE;
2283                         goto insert;
2284                 }
2285                 bytenr += btrfs_file_extent_offset(leaf, item);
2286                 em->block_start = bytenr;
2287                 em->start = extent_start;
2288                 em->len = extent_end - extent_start;
2289                 goto insert;
2290         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2291                 u64 page_start;
2292                 unsigned long ptr;
2293                 char *map;
2294                 size_t size;
2295                 size_t extent_offset;
2296                 size_t copy_size;
2297
2298                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2299                                                     path->slots[0]));
2300                 extent_end = (extent_start + size + root->sectorsize - 1) &
2301                         ~((u64)root->sectorsize - 1);
2302                 if (start < extent_start || start >= extent_end) {
2303                         em->start = start;
2304                         if (start < extent_start) {
2305                                 if (start + len <= extent_start)
2306                                         goto not_found;
2307                                 em->len = extent_end - extent_start;
2308                         } else {
2309                                 em->len = len;
2310                         }
2311                         goto not_found_em;
2312                 }
2313                 em->block_start = EXTENT_MAP_INLINE;
2314
2315                 if (!page) {
2316                         em->start = extent_start;
2317                         em->len = size;
2318                         goto out;
2319                 }
2320
2321                 page_start = page_offset(page) + pg_offset;
2322                 extent_offset = page_start - extent_start;
2323                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2324                                 size - extent_offset);
2325                 em->start = extent_start + extent_offset;
2326                 em->len = (copy_size + root->sectorsize - 1) &
2327                         ~((u64)root->sectorsize - 1);
2328                 map = kmap(page);
2329                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2330                 if (create == 0 && !PageUptodate(page)) {
2331                         read_extent_buffer(leaf, map + pg_offset, ptr,
2332                                            copy_size);
2333                         flush_dcache_page(page);
2334                 } else if (create && PageUptodate(page)) {
2335                         if (!trans) {
2336                                 kunmap(page);
2337                                 free_extent_map(em);
2338                                 em = NULL;
2339                                 btrfs_release_path(root, path);
2340                                 trans = btrfs_start_transaction(root, 1);
2341                                 goto again;
2342                         }
2343                         write_extent_buffer(leaf, map + pg_offset, ptr,
2344                                             copy_size);
2345                         btrfs_mark_buffer_dirty(leaf);
2346                 }
2347                 kunmap(page);
2348                 set_extent_uptodate(io_tree, em->start,
2349                                     extent_map_end(em) - 1, GFP_NOFS);
2350                 goto insert;
2351         } else {
2352                 printk("unkknown found_type %d\n", found_type);
2353                 WARN_ON(1);
2354         }
2355 not_found:
2356         em->start = start;
2357         em->len = len;
2358 not_found_em:
2359         em->block_start = EXTENT_MAP_HOLE;
2360 insert:
2361         btrfs_release_path(root, path);
2362         if (em->start > start || extent_map_end(em) <= start) {
2363                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2364                 err = -EIO;
2365                 goto out;
2366         }
2367
2368         err = 0;
2369         spin_lock(&em_tree->lock);
2370         ret = add_extent_mapping(em_tree, em);
2371         /* it is possible that someone inserted the extent into the tree
2372          * while we had the lock dropped.  It is also possible that
2373          * an overlapping map exists in the tree
2374          */
2375         if (ret == -EEXIST) {
2376                 struct extent_map *existing;
2377                 existing = lookup_extent_mapping(em_tree, start, len);
2378                 if (existing && (existing->start > start ||
2379                     existing->start + existing->len <= start)) {
2380                         free_extent_map(existing);
2381                         existing = NULL;
2382                 }
2383                 if (!existing) {
2384                         existing = lookup_extent_mapping(em_tree, em->start,
2385                                                          em->len);
2386                         if (existing) {
2387                                 err = merge_extent_mapping(em_tree, existing,
2388                                                            em);
2389                                 free_extent_map(existing);
2390                                 if (err) {
2391                                         free_extent_map(em);
2392                                         em = NULL;
2393                                 }
2394                         } else {
2395                                 err = -EIO;
2396                                 printk("failing to insert %Lu %Lu\n",
2397                                        start, len);
2398                                 free_extent_map(em);
2399                                 em = NULL;
2400                         }
2401                 } else {
2402                         free_extent_map(em);
2403                         em = existing;
2404                 }
2405         }
2406         spin_unlock(&em_tree->lock);
2407 out:
2408         btrfs_free_path(path);
2409         if (trans) {
2410                 ret = btrfs_end_transaction(trans, root);
2411                 if (!err)
2412                         err = ret;
2413         }
2414         mutex_unlock(&root->fs_info->fs_mutex);
2415         if (err) {
2416                 free_extent_map(em);
2417                 WARN_ON(1);
2418                 return ERR_PTR(err);
2419         }
2420         return em;
2421 }
2422
2423 #if 0 /* waiting for O_DIRECT reads */
2424 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2425                         struct buffer_head *bh_result, int create)
2426 {
2427         struct extent_map *em;
2428         u64 start = (u64)iblock << inode->i_blkbits;
2429         struct btrfs_multi_bio *multi = NULL;
2430         struct btrfs_root *root = BTRFS_I(inode)->root;
2431         u64 len;
2432         u64 logical;
2433         u64 map_length;
2434         int ret = 0;
2435
2436         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2437
2438         if (!em || IS_ERR(em))
2439                 goto out;
2440
2441         if (em->start > start || em->start + em->len <= start) {
2442             goto out;
2443         }
2444
2445         if (em->block_start == EXTENT_MAP_INLINE) {
2446                 ret = -EINVAL;
2447                 goto out;
2448         }
2449
2450         len = em->start + em->len - start;
2451         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2452
2453         if (em->block_start == EXTENT_MAP_HOLE ||
2454             em->block_start == EXTENT_MAP_DELALLOC) {
2455                 bh_result->b_size = len;
2456                 goto out;
2457         }
2458
2459         logical = start - em->start;
2460         logical = em->block_start + logical;
2461
2462         map_length = len;
2463         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2464                               logical, &map_length, &multi, 0);
2465         BUG_ON(ret);
2466         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2467         bh_result->b_size = min(map_length, len);
2468
2469         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2470         set_buffer_mapped(bh_result);
2471         kfree(multi);
2472 out:
2473         free_extent_map(em);
2474         return ret;
2475 }
2476 #endif
2477
2478 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2479                         const struct iovec *iov, loff_t offset,
2480                         unsigned long nr_segs)
2481 {
2482         return -EINVAL;
2483 #if 0
2484         struct file *file = iocb->ki_filp;
2485         struct inode *inode = file->f_mapping->host;
2486
2487         if (rw == WRITE)
2488                 return -EINVAL;
2489
2490         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2491                                   offset, nr_segs, btrfs_get_block, NULL);
2492 #endif
2493 }
2494
2495 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2496 {
2497         return extent_bmap(mapping, iblock, btrfs_get_extent);
2498 }
2499
2500 int btrfs_readpage(struct file *file, struct page *page)
2501 {
2502         struct extent_io_tree *tree;
2503         tree = &BTRFS_I(page->mapping->host)->io_tree;
2504         return extent_read_full_page(tree, page, btrfs_get_extent);
2505 }
2506
2507 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2508 {
2509         struct extent_io_tree *tree;
2510
2511
2512         if (current->flags & PF_MEMALLOC) {
2513                 redirty_page_for_writepage(wbc, page);
2514                 unlock_page(page);
2515                 return 0;
2516         }
2517         tree = &BTRFS_I(page->mapping->host)->io_tree;
2518         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2519 }
2520
2521 static int btrfs_writepages(struct address_space *mapping,
2522                             struct writeback_control *wbc)
2523 {
2524         struct extent_io_tree *tree;
2525         tree = &BTRFS_I(mapping->host)->io_tree;
2526         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2527 }
2528
2529 static int
2530 btrfs_readpages(struct file *file, struct address_space *mapping,
2531                 struct list_head *pages, unsigned nr_pages)
2532 {
2533         struct extent_io_tree *tree;
2534         tree = &BTRFS_I(mapping->host)->io_tree;
2535         return extent_readpages(tree, mapping, pages, nr_pages,
2536                                 btrfs_get_extent);
2537 }
2538
2539 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2540 {
2541         struct extent_io_tree *tree;
2542         struct extent_map_tree *map;
2543         int ret;
2544
2545         tree = &BTRFS_I(page->mapping->host)->io_tree;
2546         map = &BTRFS_I(page->mapping->host)->extent_tree;
2547         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2548         if (ret == 1) {
2549                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
2550                 ClearPagePrivate(page);
2551                 set_page_private(page, 0);
2552                 page_cache_release(page);
2553         }
2554         return ret;
2555 }
2556
2557 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2558 {
2559         struct extent_io_tree *tree;
2560
2561         tree = &BTRFS_I(page->mapping->host)->io_tree;
2562         extent_invalidatepage(tree, page, offset);
2563         btrfs_releasepage(page, GFP_NOFS);
2564         if (PagePrivate(page)) {
2565                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
2566                 ClearPagePrivate(page);
2567                 set_page_private(page, 0);
2568                 page_cache_release(page);
2569         }
2570 }
2571
2572 /*
2573  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2574  * called from a page fault handler when a page is first dirtied. Hence we must
2575  * be careful to check for EOF conditions here. We set the page up correctly
2576  * for a written page which means we get ENOSPC checking when writing into
2577  * holes and correct delalloc and unwritten extent mapping on filesystems that
2578  * support these features.
2579  *
2580  * We are not allowed to take the i_mutex here so we have to play games to
2581  * protect against truncate races as the page could now be beyond EOF.  Because
2582  * vmtruncate() writes the inode size before removing pages, once we have the
2583  * page lock we can determine safely if the page is beyond EOF. If it is not
2584  * beyond EOF, then the page is guaranteed safe against truncation until we
2585  * unlock the page.
2586  */
2587 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2588 {
2589         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2590         struct btrfs_root *root = BTRFS_I(inode)->root;
2591         unsigned long end;
2592         loff_t size;
2593         int ret;
2594         u64 page_start;
2595
2596         mutex_lock(&root->fs_info->fs_mutex);
2597         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2598         mutex_unlock(&root->fs_info->fs_mutex);
2599         if (ret)
2600                 goto out;
2601
2602         ret = -EINVAL;
2603
2604         lock_page(page);
2605         wait_on_page_writeback(page);
2606         size = i_size_read(inode);
2607         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2608
2609         if ((page->mapping != inode->i_mapping) ||
2610             (page_start > size)) {
2611                 /* page got truncated out from underneath us */
2612                 goto out_unlock;
2613         }
2614
2615         /* page is wholly or partially inside EOF */
2616         if (page_start + PAGE_CACHE_SIZE > size)
2617                 end = size & ~PAGE_CACHE_MASK;
2618         else
2619                 end = PAGE_CACHE_SIZE;
2620
2621         ret = btrfs_cow_one_page(inode, page, end);
2622
2623 out_unlock:
2624         unlock_page(page);
2625 out:
2626         return ret;
2627 }
2628
2629 static void btrfs_truncate(struct inode *inode)
2630 {
2631         struct btrfs_root *root = BTRFS_I(inode)->root;
2632         int ret;
2633         struct btrfs_trans_handle *trans;
2634         unsigned long nr;
2635
2636         if (!S_ISREG(inode->i_mode))
2637                 return;
2638         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2639                 return;
2640
2641         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2642
2643         mutex_lock(&root->fs_info->fs_mutex);
2644         trans = btrfs_start_transaction(root, 1);
2645         btrfs_set_trans_block_group(trans, inode);
2646
2647         /* FIXME, add redo link to tree so we don't leak on crash */
2648         ret = btrfs_truncate_in_trans(trans, root, inode,
2649                                       BTRFS_EXTENT_DATA_KEY);
2650         btrfs_update_inode(trans, root, inode);
2651         nr = trans->blocks_used;
2652
2653         ret = btrfs_end_transaction(trans, root);
2654         BUG_ON(ret);
2655         mutex_unlock(&root->fs_info->fs_mutex);
2656         btrfs_btree_balance_dirty(root, nr);
2657         btrfs_throttle(root);
2658 }
2659
2660 static int noinline create_subvol(struct btrfs_root *root, char *name,
2661                                   int namelen)
2662 {
2663         struct btrfs_trans_handle *trans;
2664         struct btrfs_key key;
2665         struct btrfs_root_item root_item;
2666         struct btrfs_inode_item *inode_item;
2667         struct extent_buffer *leaf;
2668         struct btrfs_root *new_root = root;
2669         struct inode *inode;
2670         struct inode *dir;
2671         int ret;
2672         int err;
2673         u64 objectid;
2674         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2675         unsigned long nr = 1;
2676
2677         mutex_lock(&root->fs_info->fs_mutex);
2678         ret = btrfs_check_free_space(root, 1, 0);
2679         if (ret)
2680                 goto fail_commit;
2681
2682         trans = btrfs_start_transaction(root, 1);
2683         BUG_ON(!trans);
2684
2685         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2686                                        0, &objectid);
2687         if (ret)
2688                 goto fail;
2689
2690         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2691                                         objectid, trans->transid, 0, 0,
2692                                         0, 0);
2693         if (IS_ERR(leaf))
2694                 return PTR_ERR(leaf);
2695
2696         btrfs_set_header_nritems(leaf, 0);
2697         btrfs_set_header_level(leaf, 0);
2698         btrfs_set_header_bytenr(leaf, leaf->start);
2699         btrfs_set_header_generation(leaf, trans->transid);
2700         btrfs_set_header_owner(leaf, objectid);
2701
2702         write_extent_buffer(leaf, root->fs_info->fsid,
2703                             (unsigned long)btrfs_header_fsid(leaf),
2704                             BTRFS_FSID_SIZE);
2705         btrfs_mark_buffer_dirty(leaf);
2706
2707         inode_item = &root_item.inode;
2708         memset(inode_item, 0, sizeof(*inode_item));
2709         inode_item->generation = cpu_to_le64(1);
2710         inode_item->size = cpu_to_le64(3);
2711         inode_item->nlink = cpu_to_le32(1);
2712         inode_item->nblocks = cpu_to_le64(1);
2713         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2714
2715         btrfs_set_root_bytenr(&root_item, leaf->start);
2716         btrfs_set_root_level(&root_item, 0);
2717         btrfs_set_root_refs(&root_item, 1);
2718         btrfs_set_root_used(&root_item, 0);
2719
2720         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2721         root_item.drop_level = 0;
2722
2723         free_extent_buffer(leaf);
2724         leaf = NULL;
2725
2726         btrfs_set_root_dirid(&root_item, new_dirid);
2727
2728         key.objectid = objectid;
2729         key.offset = 1;
2730         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2731         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2732                                 &root_item);
2733         if (ret)
2734                 goto fail;
2735
2736         /*
2737          * insert the directory item
2738          */
2739         key.offset = (u64)-1;
2740         dir = root->fs_info->sb->s_root->d_inode;
2741         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2742                                     name, namelen, dir->i_ino, &key,
2743                                     BTRFS_FT_DIR);
2744         if (ret)
2745                 goto fail;
2746
2747         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2748                              name, namelen, objectid,
2749                              root->fs_info->sb->s_root->d_inode->i_ino);
2750         if (ret)
2751                 goto fail;
2752
2753         ret = btrfs_commit_transaction(trans, root);
2754         if (ret)
2755                 goto fail_commit;
2756
2757         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2758         BUG_ON(!new_root);
2759
2760         trans = btrfs_start_transaction(new_root, 1);
2761         BUG_ON(!trans);
2762
2763         inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2764                                 new_dirid,
2765                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2766         if (IS_ERR(inode))
2767                 goto fail;
2768         inode->i_op = &btrfs_dir_inode_operations;
2769         inode->i_fop = &btrfs_dir_file_operations;
2770         new_root->inode = inode;
2771
2772         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2773                                      new_dirid);
2774         inode->i_nlink = 1;
2775         inode->i_size = 0;
2776         ret = btrfs_update_inode(trans, new_root, inode);
2777         if (ret)
2778                 goto fail;
2779 fail:
2780         nr = trans->blocks_used;
2781         err = btrfs_commit_transaction(trans, new_root);
2782         if (err && !ret)
2783                 ret = err;
2784 fail_commit:
2785         mutex_unlock(&root->fs_info->fs_mutex);
2786         btrfs_btree_balance_dirty(root, nr);
2787         btrfs_throttle(root);
2788         return ret;
2789 }
2790
2791 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2792 {
2793         struct btrfs_pending_snapshot *pending_snapshot;
2794         struct btrfs_trans_handle *trans;
2795         int ret;
2796         int err;
2797         unsigned long nr = 0;
2798
2799         if (!root->ref_cows)
2800                 return -EINVAL;
2801
2802         mutex_lock(&root->fs_info->fs_mutex);
2803         ret = btrfs_check_free_space(root, 1, 0);
2804         if (ret)
2805                 goto fail_unlock;
2806
2807         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2808         if (!pending_snapshot) {
2809                 ret = -ENOMEM;
2810                 goto fail_unlock;
2811         }
2812         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2813         if (!pending_snapshot->name) {
2814                 ret = -ENOMEM;
2815                 kfree(pending_snapshot);
2816                 goto fail_unlock;
2817         }
2818         memcpy(pending_snapshot->name, name, namelen);
2819         pending_snapshot->name[namelen] = '\0';
2820         trans = btrfs_start_transaction(root, 1);
2821         BUG_ON(!trans);
2822         pending_snapshot->root = root;
2823         list_add(&pending_snapshot->list,
2824                  &trans->transaction->pending_snapshots);
2825         ret = btrfs_update_inode(trans, root, root->inode);
2826         err = btrfs_commit_transaction(trans, root);
2827
2828 fail_unlock:
2829         mutex_unlock(&root->fs_info->fs_mutex);
2830         btrfs_btree_balance_dirty(root, nr);
2831         btrfs_throttle(root);
2832         return ret;
2833 }
2834
2835 unsigned long btrfs_force_ra(struct address_space *mapping,
2836                               struct file_ra_state *ra, struct file *file,
2837                               pgoff_t offset, pgoff_t last_index)
2838 {
2839         pgoff_t req_size = last_index - offset + 1;
2840
2841 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2842         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2843         return offset;
2844 #else
2845         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2846         return offset + req_size;
2847 #endif
2848 }
2849
2850 int btrfs_defrag_file(struct file *file) {
2851         struct inode *inode = fdentry(file)->d_inode;
2852         struct btrfs_root *root = BTRFS_I(inode)->root;
2853         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2854         struct page *page;
2855         unsigned long last_index;
2856         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
2857         unsigned long total_read = 0;
2858         u64 page_start;
2859         u64 page_end;
2860         unsigned long i;
2861         int ret;
2862
2863         mutex_lock(&root->fs_info->fs_mutex);
2864         ret = btrfs_check_free_space(root, inode->i_size, 0);
2865         mutex_unlock(&root->fs_info->fs_mutex);
2866         if (ret)
2867                 return -ENOSPC;
2868
2869         mutex_lock(&inode->i_mutex);
2870         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2871         for (i = 0; i <= last_index; i++) {
2872                 if (total_read % ra_pages == 0) {
2873                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
2874                                        min(last_index, i + ra_pages - 1));
2875                 }
2876                 total_read++;
2877                 page = grab_cache_page(inode->i_mapping, i);
2878                 if (!page)
2879                         goto out_unlock;
2880                 if (!PageUptodate(page)) {
2881                         btrfs_readpage(NULL, page);
2882                         lock_page(page);
2883                         if (!PageUptodate(page)) {
2884                                 unlock_page(page);
2885                                 page_cache_release(page);
2886                                 goto out_unlock;
2887                         }
2888                 }
2889
2890 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2891                 ClearPageDirty(page);
2892 #else
2893                 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2894 #endif
2895                 wait_on_page_writeback(page);
2896                 set_page_extent_mapped(page);
2897
2898                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2899                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2900
2901                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2902                 set_extent_delalloc(io_tree, page_start,
2903                                     page_end, GFP_NOFS);
2904
2905                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2906                 set_page_dirty(page);
2907                 unlock_page(page);
2908                 page_cache_release(page);
2909                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2910         }
2911
2912 out_unlock:
2913         mutex_unlock(&inode->i_mutex);
2914         return 0;
2915 }
2916
2917 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2918 {
2919         u64 new_size;
2920         u64 old_size;
2921         u64 devid = 1;
2922         struct btrfs_ioctl_vol_args *vol_args;
2923         struct btrfs_trans_handle *trans;
2924         struct btrfs_device *device = NULL;
2925         char *sizestr;
2926         char *devstr = NULL;
2927         int ret = 0;
2928         int namelen;
2929         int mod = 0;
2930
2931         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2932
2933         if (!vol_args)
2934                 return -ENOMEM;
2935
2936         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2937                 ret = -EFAULT;
2938                 goto out;
2939         }
2940         namelen = strlen(vol_args->name);
2941         if (namelen > BTRFS_VOL_NAME_MAX) {
2942                 ret = -EINVAL;
2943                 goto out;
2944         }
2945
2946         mutex_lock(&root->fs_info->fs_mutex);
2947         sizestr = vol_args->name;
2948         devstr = strchr(sizestr, ':');
2949         if (devstr) {
2950                 char *end;
2951                 sizestr = devstr + 1;
2952                 *devstr = '\0';
2953                 devstr = vol_args->name;
2954                 devid = simple_strtoull(devstr, &end, 10);
2955 printk("resizing devid %Lu\n", devid);
2956         }
2957         device = btrfs_find_device(root, devid, NULL);
2958         if (!device) {
2959                 printk("resizer unable to find device %Lu\n", devid);
2960                 ret = -EINVAL;
2961                 goto out_unlock;
2962         }
2963         if (!strcmp(sizestr, "max"))
2964                 new_size = device->bdev->bd_inode->i_size;
2965         else {
2966                 if (sizestr[0] == '-') {
2967                         mod = -1;
2968                         sizestr++;
2969                 } else if (sizestr[0] == '+') {
2970                         mod = 1;
2971                         sizestr++;
2972                 }
2973                 new_size = btrfs_parse_size(sizestr);
2974                 if (new_size == 0) {
2975                         ret = -EINVAL;
2976                         goto out_unlock;
2977                 }
2978         }
2979
2980         old_size = device->total_bytes;
2981
2982         if (mod < 0) {
2983                 if (new_size > old_size) {
2984                         ret = -EINVAL;
2985                         goto out_unlock;
2986                 }
2987                 new_size = old_size - new_size;
2988         } else if (mod > 0) {
2989                 new_size = old_size + new_size;
2990         }
2991
2992         if (new_size < 256 * 1024 * 1024) {
2993                 ret = -EINVAL;
2994                 goto out_unlock;
2995         }
2996         if (new_size > device->bdev->bd_inode->i_size) {
2997                 ret = -EFBIG;
2998                 goto out_unlock;
2999         }
3000
3001         do_div(new_size, root->sectorsize);
3002         new_size *= root->sectorsize;
3003
3004 printk("new size for %s is %llu\n", device->name, (unsigned long long)new_size);
3005
3006         if (new_size > old_size) {
3007                 trans = btrfs_start_transaction(root, 1);
3008                 ret = btrfs_grow_device(trans, device, new_size);
3009                 btrfs_commit_transaction(trans, root);
3010         } else {
3011                 ret = btrfs_shrink_device(device, new_size);
3012         }
3013
3014 out_unlock:
3015         mutex_unlock(&root->fs_info->fs_mutex);
3016 out:
3017         kfree(vol_args);
3018         return ret;
3019 }
3020
3021 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
3022                                             void __user *arg)
3023 {
3024         struct btrfs_ioctl_vol_args *vol_args;
3025         struct btrfs_dir_item *di;
3026         struct btrfs_path *path;
3027         u64 root_dirid;
3028         int namelen;
3029         int ret;
3030
3031         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
3032
3033         if (!vol_args)
3034                 return -ENOMEM;
3035
3036         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
3037                 ret = -EFAULT;
3038                 goto out;
3039         }
3040
3041         namelen = strlen(vol_args->name);
3042         if (namelen > BTRFS_VOL_NAME_MAX) {
3043                 ret = -EINVAL;
3044                 goto out;
3045         }
3046         if (strchr(vol_args->name, '/')) {
3047                 ret = -EINVAL;
3048                 goto out;
3049         }
3050
3051         path = btrfs_alloc_path();
3052         if (!path) {
3053                 ret = -ENOMEM;
3054                 goto out;
3055         }
3056
3057         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
3058         mutex_lock(&root->fs_info->fs_mutex);
3059         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
3060                             path, root_dirid,
3061                             vol_args->name, namelen, 0);
3062         mutex_unlock(&root->fs_info->fs_mutex);
3063         btrfs_free_path(path);
3064
3065         if (di && !IS_ERR(di)) {
3066                 ret = -EEXIST;
3067                 goto out;
3068         }
3069
3070         if (IS_ERR(di)) {
3071                 ret = PTR_ERR(di);
3072                 goto out;
3073         }
3074
3075         if (root == root->fs_info->tree_root)
3076                 ret = create_subvol(root, vol_args->name, namelen);
3077         else
3078                 ret = create_snapshot(root, vol_args->name, namelen);
3079 out:
3080         kfree(vol_args);
3081         return ret;
3082 }
3083
3084 static int btrfs_ioctl_defrag(struct file *file)
3085 {
3086         struct inode *inode = fdentry(file)->d_inode;
3087         struct btrfs_root *root = BTRFS_I(inode)->root;
3088
3089         switch (inode->i_mode & S_IFMT) {
3090         case S_IFDIR:
3091                 mutex_lock(&root->fs_info->fs_mutex);
3092                 btrfs_defrag_root(root, 0);
3093                 btrfs_defrag_root(root->fs_info->extent_root, 0);
3094                 mutex_unlock(&root->fs_info->fs_mutex);
3095                 break;
3096         case S_IFREG:
3097                 btrfs_defrag_file(file);
3098                 break;
3099         }
3100
3101         return 0;
3102 }
3103
3104 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
3105 {
3106         struct btrfs_ioctl_vol_args *vol_args;
3107         int ret;
3108
3109         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
3110
3111         if (!vol_args)
3112                 return -ENOMEM;
3113
3114         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
3115                 ret = -EFAULT;
3116                 goto out;
3117         }
3118         ret = btrfs_init_new_device(root, vol_args->name);
3119
3120 out:
3121         kfree(vol_args);
3122         return ret;
3123 }
3124
3125 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
3126 {
3127         struct btrfs_ioctl_vol_args *vol_args;
3128         int ret;
3129
3130         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
3131
3132         if (!vol_args)
3133                 return -ENOMEM;
3134
3135         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
3136                 ret = -EFAULT;
3137                 goto out;
3138         }
3139         ret = btrfs_rm_device(root, vol_args->name);
3140
3141 out:
3142         kfree(vol_args);
3143         return ret;
3144 }
3145
3146 int dup_item_to_inode(struct btrfs_trans_handle *trans,
3147                        struct btrfs_root *root,
3148                        struct btrfs_path *path,
3149                        struct extent_buffer *leaf,
3150                        int slot,
3151                        struct btrfs_key *key,
3152                        u64 destino)
3153 {
3154         char *dup;
3155         int len = btrfs_item_size_nr(leaf, slot);
3156         struct btrfs_key ckey = *key;
3157         int ret = 0;
3158
3159         dup = kmalloc(len, GFP_NOFS);
3160         if (!dup)
3161                 return -ENOMEM;
3162
3163         read_extent_buffer(leaf, dup, btrfs_item_ptr_offset(leaf, slot), len);
3164         btrfs_release_path(root, path);
3165
3166         ckey.objectid = destino;
3167         ret = btrfs_insert_item(trans, root, &ckey, dup, len);
3168         kfree(dup);
3169         return ret;
3170 }
3171
3172 long btrfs_ioctl_clone(struct file *file, unsigned long src_fd)
3173 {
3174         struct inode *inode = fdentry(file)->d_inode;
3175         struct btrfs_root *root = BTRFS_I(inode)->root;
3176         struct file *src_file;
3177         struct inode *src;
3178         struct btrfs_trans_handle *trans;
3179         int ret;
3180         u64 pos;
3181         struct btrfs_path *path;
3182         struct btrfs_key key;
3183         struct extent_buffer *leaf;
3184         u32 nritems;
3185         int slot;
3186
3187         src_file = fget(src_fd);
3188         if (!src_file)
3189                 return -EBADF;
3190         src = src_file->f_dentry->d_inode;
3191
3192         ret = -EXDEV;
3193         if (src->i_sb != inode->i_sb)
3194                 goto out_fput;
3195
3196         if (inode < src) {
3197                 mutex_lock(&inode->i_mutex);
3198                 mutex_lock(&src->i_mutex);
3199         } else {
3200                 mutex_lock(&src->i_mutex);
3201                 mutex_lock(&inode->i_mutex);
3202         }
3203
3204         ret = -ENOTEMPTY;
3205         if (inode->i_size)
3206                 goto out_unlock;
3207
3208         /* do any pending delalloc/csum calc on src, one way or
3209            another, and lock file content */
3210         while (1) {
3211                 filemap_write_and_wait(src->i_mapping);
3212                 lock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
3213                 if (BTRFS_I(src)->delalloc_bytes == 0)
3214                         break;
3215                 unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
3216         }
3217
3218         mutex_lock(&root->fs_info->fs_mutex);
3219         trans = btrfs_start_transaction(root, 0);
3220         path = btrfs_alloc_path();
3221         if (!path) {
3222                 ret = -ENOMEM;
3223                 goto out;
3224         }
3225         key.offset = 0;
3226         key.type = BTRFS_EXTENT_DATA_KEY;
3227         key.objectid = src->i_ino;
3228         pos = 0;
3229         path->reada = 2;
3230
3231         while (1) {
3232                 /*
3233                  * note the key will change type as we walk through the
3234                  * tree.
3235                  */
3236                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
3237                 if (ret < 0)
3238                         goto out;
3239
3240                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3241                         ret = btrfs_next_leaf(root, path);
3242                         if (ret < 0)
3243                                 goto out;
3244                         if (ret > 0)
3245                                 break;
3246                 }
3247                 leaf = path->nodes[0];
3248                 slot = path->slots[0];
3249                 btrfs_item_key_to_cpu(leaf, &key, slot);
3250                 nritems = btrfs_header_nritems(leaf);
3251
3252                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
3253                     key.objectid != src->i_ino)
3254                         break;
3255
3256                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
3257                         struct btrfs_file_extent_item *extent;
3258                         int found_type;
3259                         pos = key.offset;
3260                         extent = btrfs_item_ptr(leaf, slot,
3261                                                 struct btrfs_file_extent_item);
3262                         found_type = btrfs_file_extent_type(leaf, extent);
3263                         if (found_type == BTRFS_FILE_EXTENT_REG) {
3264                                 u64 len = btrfs_file_extent_num_bytes(leaf,
3265                                                                       extent);
3266                                 u64 ds = btrfs_file_extent_disk_bytenr(leaf,
3267                                                                        extent);
3268                                 u64 dl = btrfs_file_extent_disk_num_bytes(leaf,
3269                                                                  extent);
3270                                 u64 off = btrfs_file_extent_offset(leaf,
3271                                                                    extent);
3272                                 btrfs_insert_file_extent(trans, root,
3273                                                          inode->i_ino, pos,
3274                                                          ds, dl, len, off);
3275                                 /* ds == 0 means there's a hole */
3276                                 if (ds != 0) {
3277                                         btrfs_inc_extent_ref(trans, root,
3278                                                      ds, dl,
3279                                                      root->root_key.objectid,
3280                                                      trans->transid,
3281                                                      inode->i_ino, pos);
3282                                 }
3283                                 pos = key.offset + len;
3284                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3285                                 ret = dup_item_to_inode(trans, root, path,
3286                                                         leaf, slot, &key,
3287                                                         inode->i_ino);
3288                                 if (ret)
3289                                         goto out;
3290                                 pos = key.offset + btrfs_item_size_nr(leaf,
3291                                                                       slot);
3292                         }
3293                 } else if (btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
3294                         ret = dup_item_to_inode(trans, root, path, leaf,
3295                                                 slot, &key, inode->i_ino);
3296
3297                         if (ret)
3298                                 goto out;
3299                 }
3300                 key.offset++;
3301                 btrfs_release_path(root, path);
3302         }
3303
3304         ret = 0;
3305 out:
3306         btrfs_free_path(path);
3307
3308         inode->i_blocks = src->i_blocks;
3309         i_size_write(inode, src->i_size);
3310         btrfs_update_inode(trans, root, inode);
3311
3312         unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
3313
3314         btrfs_end_transaction(trans, root);
3315         mutex_unlock(&root->fs_info->fs_mutex);
3316
3317 out_unlock:
3318         mutex_unlock(&src->i_mutex);
3319         mutex_unlock(&inode->i_mutex);
3320 out_fput:
3321         fput(src_file);
3322         return ret;
3323 }
3324
3325 long btrfs_ioctl(struct file *file, unsigned int
3326                 cmd, unsigned long arg)
3327 {
3328         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3329
3330         switch (cmd) {
3331         case BTRFS_IOC_SNAP_CREATE:
3332                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
3333         case BTRFS_IOC_DEFRAG:
3334                 return btrfs_ioctl_defrag(file);
3335         case BTRFS_IOC_RESIZE:
3336                 return btrfs_ioctl_resize(root, (void __user *)arg);
3337         case BTRFS_IOC_ADD_DEV:
3338                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
3339         case BTRFS_IOC_RM_DEV:
3340                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
3341         case BTRFS_IOC_BALANCE:
3342                 return btrfs_balance(root->fs_info->dev_root);
3343         case BTRFS_IOC_CLONE:
3344                 return btrfs_ioctl_clone(file, arg);
3345         }
3346
3347         return -ENOTTY;
3348 }
3349
3350 /*
3351  * Called inside transaction, so use GFP_NOFS
3352  */
3353 struct inode *btrfs_alloc_inode(struct super_block *sb)
3354 {
3355         struct btrfs_inode *ei;
3356
3357         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3358         if (!ei)
3359                 return NULL;
3360         ei->last_trans = 0;
3361         ei->ordered_trans = 0;
3362         return &ei->vfs_inode;
3363 }
3364
3365 void btrfs_destroy_inode(struct inode *inode)
3366 {
3367         WARN_ON(!list_empty(&inode->i_dentry));
3368         WARN_ON(inode->i_data.nrpages);
3369
3370         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3371         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3372 }
3373
3374 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3375 static void init_once(struct kmem_cache * cachep, void *foo)
3376 #else
3377 static void init_once(void * foo, struct kmem_cache * cachep,
3378                       unsigned long flags)
3379 #endif
3380 {
3381         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3382
3383         inode_init_once(&ei->vfs_inode);
3384 }
3385
3386 void btrfs_destroy_cachep(void)
3387 {
3388         if (btrfs_inode_cachep)
3389                 kmem_cache_destroy(btrfs_inode_cachep);
3390         if (btrfs_trans_handle_cachep)
3391                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3392         if (btrfs_transaction_cachep)
3393                 kmem_cache_destroy(btrfs_transaction_cachep);
3394         if (btrfs_bit_radix_cachep)
3395                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3396         if (btrfs_path_cachep)
3397                 kmem_cache_destroy(btrfs_path_cachep);
3398 }
3399
3400 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3401                                        unsigned long extra_flags,
3402 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3403                                        void (*ctor)(struct kmem_cache *, void *)
3404 #else
3405                                        void (*ctor)(void *, struct kmem_cache *,
3406                                                     unsigned long)
3407 #endif
3408                                      )
3409 {
3410         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3411                                  SLAB_MEM_SPREAD | extra_flags), ctor
3412 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3413                                  ,NULL
3414 #endif
3415                                 );
3416 }
3417
3418 int btrfs_init_cachep(void)
3419 {
3420         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3421                                           sizeof(struct btrfs_inode),
3422                                           0, init_once);
3423         if (!btrfs_inode_cachep)
3424                 goto fail;
3425         btrfs_trans_handle_cachep =
3426                         btrfs_cache_create("btrfs_trans_handle_cache",
3427                                            sizeof(struct btrfs_trans_handle),
3428                                            0, NULL);
3429         if (!btrfs_trans_handle_cachep)
3430                 goto fail;
3431         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3432                                              sizeof(struct btrfs_transaction),
3433                                              0, NULL);
3434         if (!btrfs_transaction_cachep)
3435                 goto fail;
3436         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3437                                          sizeof(struct btrfs_path),
3438                                          0, NULL);
3439         if (!btrfs_path_cachep)
3440                 goto fail;
3441         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3442                                               SLAB_DESTROY_BY_RCU, NULL);
3443         if (!btrfs_bit_radix_cachep)
3444                 goto fail;
3445         return 0;
3446 fail:
3447         btrfs_destroy_cachep();
3448         return -ENOMEM;
3449 }
3450
3451 static int btrfs_getattr(struct vfsmount *mnt,
3452                          struct dentry *dentry, struct kstat *stat)
3453 {
3454         struct inode *inode = dentry->d_inode;
3455         generic_fillattr(inode, stat);
3456         stat->blksize = PAGE_CACHE_SIZE;
3457         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3458         return 0;
3459 }
3460
3461 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3462                            struct inode * new_dir,struct dentry *new_dentry)
3463 {
3464         struct btrfs_trans_handle *trans;
3465         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3466         struct inode *new_inode = new_dentry->d_inode;
3467         struct inode *old_inode = old_dentry->d_inode;
3468         struct timespec ctime = CURRENT_TIME;
3469         struct btrfs_path *path;
3470         int ret;
3471
3472         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3473             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3474                 return -ENOTEMPTY;
3475         }
3476
3477         mutex_lock(&root->fs_info->fs_mutex);
3478         ret = btrfs_check_free_space(root, 1, 0);
3479         if (ret)
3480                 goto out_unlock;
3481
3482         trans = btrfs_start_transaction(root, 1);
3483
3484         btrfs_set_trans_block_group(trans, new_dir);
3485         path = btrfs_alloc_path();
3486         if (!path) {
3487                 ret = -ENOMEM;
3488                 goto out_fail;
3489         }
3490
3491         old_dentry->d_inode->i_nlink++;
3492         old_dir->i_ctime = old_dir->i_mtime = ctime;
3493         new_dir->i_ctime = new_dir->i_mtime = ctime;
3494         old_inode->i_ctime = ctime;
3495
3496         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3497         if (ret)
3498                 goto out_fail;
3499
3500         if (new_inode) {
3501                 new_inode->i_ctime = CURRENT_TIME;
3502                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3503                 if (ret)
3504                         goto out_fail;
3505         }
3506         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
3507         if (ret)
3508                 goto out_fail;
3509
3510 out_fail:
3511         btrfs_free_path(path);
3512         btrfs_end_transaction(trans, root);
3513 out_unlock:
3514         mutex_unlock(&root->fs_info->fs_mutex);
3515         return ret;
3516 }
3517
3518 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3519                          const char *symname)
3520 {
3521         struct btrfs_trans_handle *trans;
3522         struct btrfs_root *root = BTRFS_I(dir)->root;
3523         struct btrfs_path *path;
3524         struct btrfs_key key;
3525         struct inode *inode = NULL;
3526         int err;
3527         int drop_inode = 0;
3528         u64 objectid;
3529         int name_len;
3530         int datasize;
3531         unsigned long ptr;
3532         struct btrfs_file_extent_item *ei;
3533         struct extent_buffer *leaf;
3534         unsigned long nr = 0;
3535
3536         name_len = strlen(symname) + 1;
3537         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3538                 return -ENAMETOOLONG;
3539
3540         mutex_lock(&root->fs_info->fs_mutex);
3541         err = btrfs_check_free_space(root, 1, 0);
3542         if (err)
3543                 goto out_fail;
3544
3545         trans = btrfs_start_transaction(root, 1);
3546         btrfs_set_trans_block_group(trans, dir);
3547
3548         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3549         if (err) {
3550                 err = -ENOSPC;
3551                 goto out_unlock;
3552         }
3553
3554         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
3555                                 dentry->d_name.len,
3556                                 dentry->d_parent->d_inode->i_ino, objectid,
3557                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
3558         err = PTR_ERR(inode);
3559         if (IS_ERR(inode))
3560                 goto out_unlock;
3561
3562         btrfs_set_trans_block_group(trans, inode);
3563         err = btrfs_add_nondir(trans, dentry, inode, 0);
3564         if (err)
3565                 drop_inode = 1;
3566         else {
3567                 inode->i_mapping->a_ops = &btrfs_aops;
3568                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3569                 inode->i_fop = &btrfs_file_operations;
3570                 inode->i_op = &btrfs_file_inode_operations;
3571                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3572                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3573                                      inode->i_mapping, GFP_NOFS);
3574                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3575                                      inode->i_mapping, GFP_NOFS);
3576                 BTRFS_I(inode)->delalloc_bytes = 0;
3577                 atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
3578                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3579         }
3580         dir->i_sb->s_dirt = 1;
3581         btrfs_update_inode_block_group(trans, inode);
3582         btrfs_update_inode_block_group(trans, dir);
3583         if (drop_inode)
3584                 goto out_unlock;
3585
3586         path = btrfs_alloc_path();
3587         BUG_ON(!path);
3588         key.objectid = inode->i_ino;
3589         key.offset = 0;
3590         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3591         datasize = btrfs_file_extent_calc_inline_size(name_len);
3592         err = btrfs_insert_empty_item(trans, root, path, &key,
3593                                       datasize);
3594         if (err) {
3595                 drop_inode = 1;
3596                 goto out_unlock;
3597         }
3598         leaf = path->nodes[0];
3599         ei = btrfs_item_ptr(leaf, path->slots[0],
3600                             struct btrfs_file_extent_item);
3601         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3602         btrfs_set_file_extent_type(leaf, ei,
3603                                    BTRFS_FILE_EXTENT_INLINE);
3604         ptr = btrfs_file_extent_inline_start(ei);
3605         write_extent_buffer(leaf, symname, ptr, name_len);
3606         btrfs_mark_buffer_dirty(leaf);
3607         btrfs_free_path(path);
3608
3609         inode->i_op = &btrfs_symlink_inode_operations;
3610         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3611         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3612         inode->i_size = name_len - 1;
3613         err = btrfs_update_inode(trans, root, inode);
3614         if (err)
3615                 drop_inode = 1;
3616
3617 out_unlock:
3618         nr = trans->blocks_used;
3619         btrfs_end_transaction(trans, root);
3620 out_fail:
3621         mutex_unlock(&root->fs_info->fs_mutex);
3622         if (drop_inode) {
3623                 inode_dec_link_count(inode);
3624                 iput(inode);
3625         }
3626         btrfs_btree_balance_dirty(root, nr);
3627         btrfs_throttle(root);
3628         return err;
3629 }
3630
3631 static int btrfs_permission(struct inode *inode, int mask,
3632                             struct nameidata *nd)
3633 {
3634         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3635                 return -EACCES;
3636         return generic_permission(inode, mask, NULL);
3637 }
3638
3639 static struct inode_operations btrfs_dir_inode_operations = {
3640         .lookup         = btrfs_lookup,
3641         .create         = btrfs_create,
3642         .unlink         = btrfs_unlink,
3643         .link           = btrfs_link,
3644         .mkdir          = btrfs_mkdir,
3645         .rmdir          = btrfs_rmdir,
3646         .rename         = btrfs_rename,
3647         .symlink        = btrfs_symlink,
3648         .setattr        = btrfs_setattr,
3649         .mknod          = btrfs_mknod,
3650         .setxattr       = generic_setxattr,
3651         .getxattr       = generic_getxattr,
3652         .listxattr      = btrfs_listxattr,
3653         .removexattr    = generic_removexattr,
3654         .permission     = btrfs_permission,
3655 };
3656 static struct inode_operations btrfs_dir_ro_inode_operations = {
3657         .lookup         = btrfs_lookup,
3658         .permission     = btrfs_permission,
3659 };
3660 static struct file_operations btrfs_dir_file_operations = {
3661         .llseek         = generic_file_llseek,
3662         .read           = generic_read_dir,
3663         .readdir        = btrfs_readdir,
3664         .unlocked_ioctl = btrfs_ioctl,
3665 #ifdef CONFIG_COMPAT
3666         .compat_ioctl   = btrfs_ioctl,
3667 #endif
3668 };
3669
3670 static struct extent_io_ops btrfs_extent_io_ops = {
3671         .fill_delalloc = run_delalloc_range,
3672         .submit_bio_hook = btrfs_submit_bio_hook,
3673         .merge_bio_hook = btrfs_merge_bio_hook,
3674         .readpage_io_hook = btrfs_readpage_io_hook,
3675         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3676         .readpage_io_failed_hook = btrfs_io_failed_hook,
3677         .set_bit_hook = btrfs_set_bit_hook,
3678         .clear_bit_hook = btrfs_clear_bit_hook,
3679 };
3680
3681 static struct address_space_operations btrfs_aops = {
3682         .readpage       = btrfs_readpage,
3683         .writepage      = btrfs_writepage,
3684         .writepages     = btrfs_writepages,
3685         .readpages      = btrfs_readpages,
3686         .sync_page      = block_sync_page,
3687         .bmap           = btrfs_bmap,
3688         .direct_IO      = btrfs_direct_IO,
3689         .invalidatepage = btrfs_invalidatepage,
3690         .releasepage    = btrfs_releasepage,
3691         .set_page_dirty = __set_page_dirty_nobuffers,
3692 };
3693
3694 static struct address_space_operations btrfs_symlink_aops = {
3695         .readpage       = btrfs_readpage,
3696         .writepage      = btrfs_writepage,
3697         .invalidatepage = btrfs_invalidatepage,
3698         .releasepage    = btrfs_releasepage,
3699 };
3700
3701 static struct inode_operations btrfs_file_inode_operations = {
3702         .truncate       = btrfs_truncate,
3703         .getattr        = btrfs_getattr,
3704         .setattr        = btrfs_setattr,
3705         .setxattr       = generic_setxattr,
3706         .getxattr       = generic_getxattr,
3707         .listxattr      = btrfs_listxattr,
3708         .removexattr    = generic_removexattr,
3709         .permission     = btrfs_permission,
3710 };
3711 static struct inode_operations btrfs_special_inode_operations = {
3712         .getattr        = btrfs_getattr,
3713         .setattr        = btrfs_setattr,
3714         .permission     = btrfs_permission,
3715 };
3716 static struct inode_operations btrfs_symlink_inode_operations = {
3717         .readlink       = generic_readlink,
3718         .follow_link    = page_follow_link_light,
3719         .put_link       = page_put_link,
3720         .permission     = btrfs_permission,
3721 };