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

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/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 #include "ordered-data.h"
  47
  48 struct btrfs_iget_args {
  49         u64 ino;
  50         struct btrfs_root *root;
  51 };
  52
  53 static struct inode_operations btrfs_dir_inode_operations;
  54 static struct inode_operations btrfs_symlink_inode_operations;
  55 static struct inode_operations btrfs_dir_ro_inode_operations;
  56 static struct inode_operations btrfs_special_inode_operations;
  57 static struct inode_operations btrfs_file_inode_operations;
  58 static struct address_space_operations btrfs_aops;
  59 static struct address_space_operations btrfs_symlink_aops;
  60 static struct file_operations btrfs_dir_file_operations;
  61 static struct extent_io_ops btrfs_extent_io_ops;
  62
  63 static struct kmem_cache *btrfs_inode_cachep;
  64 struct kmem_cache *btrfs_trans_handle_cachep;
  65 struct kmem_cache *btrfs_transaction_cachep;
  66 struct kmem_cache *btrfs_bit_radix_cachep;
  67 struct kmem_cache *btrfs_path_cachep;
  68
  69 #define S_SHIFT 12
  70 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  71         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  72         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  73         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  74         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  75         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  76         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  77         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  78 };
  79
  80 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  81                            int for_del)
  82 {
  83         u64 total;
  84         u64 used;
  85         u64 thresh;
  86         unsigned long flags;
  87         int ret = 0;
  88
  89         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
  90         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  91         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  92         if (for_del)
  93                 thresh = total * 90;
  94         else
  95                 thresh = total * 85;
  96
  97         do_div(thresh, 100);
  98
  99         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
 100                 ret = -ENOSPC;
 101         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 102         return ret;
 103 }
 104
 105 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 106 {
 107         struct btrfs_root *root = BTRFS_I(inode)->root;
 108         struct btrfs_trans_handle *trans;
 109         u64 alloc_hint = 0;
 110         u64 num_bytes;
 111         u64 cur_alloc_size;
 112         u64 blocksize = root->sectorsize;
 113         u64 orig_num_bytes;
 114         struct btrfs_key ins;
 115         struct extent_map *em;
 116         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 117         int ret = 0;
 118
 119         trans = btrfs_join_transaction(root, 1);
 120         BUG_ON(!trans);
 121         btrfs_set_trans_block_group(trans, inode);
 122
 123         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 124         num_bytes = max(blocksize,  num_bytes);
 125         orig_num_bytes = num_bytes;
 126
 127         if (alloc_hint == EXTENT_MAP_INLINE)
 128                 goto out;
 129
 130         BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
 131         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 132         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1);
 133         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 134
 135         while(num_bytes > 0) {
 136                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 137                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 138                                            root->sectorsize, 0, 0,
 139                                            (u64)-1, &ins, 1);
 140                 if (ret) {
 141                         WARN_ON(1);
 142                         goto out;
 143                 }
 144                 em = alloc_extent_map(GFP_NOFS);
 145                 em->start = start;
 146                 em->len = ins.offset;
 147                 em->block_start = ins.objectid;
 148                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 149                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 150                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 151                 while(1) {
 152                         spin_lock(&em_tree->lock);
 153                         ret = add_extent_mapping(em_tree, em);
 154                         spin_unlock(&em_tree->lock);
 155                         if (ret != -EEXIST) {
 156                                 free_extent_map(em);
 157                                 break;
 158                         }
 159                         btrfs_drop_extent_cache(inode, start,
 160                                                 start + ins.offset - 1);
 161                 }
 162                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 163
 164                 cur_alloc_size = ins.offset;
 165                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 166                                                ins.offset);
 167                 BUG_ON(ret);
 168                 if (num_bytes < cur_alloc_size) {
 169                         printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
 170                                cur_alloc_size);
 171                         break;
 172                 }
 173                 num_bytes -= cur_alloc_size;
 174                 alloc_hint = ins.objectid + ins.offset;
 175                 start += cur_alloc_size;
 176         }
 177 out:
 178         btrfs_end_transaction(trans, root);
 179         return ret;
 180 }
 181
 182 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 183 {
 184         u64 extent_start;
 185         u64 extent_end;
 186         u64 bytenr;
 187         u64 cow_end;
 188         u64 loops = 0;
 189         u64 total_fs_bytes;
 190         struct btrfs_root *root = BTRFS_I(inode)->root;
 191         struct btrfs_block_group_cache *block_group;
 192         struct extent_buffer *leaf;
 193         int found_type;
 194         struct btrfs_path *path;
 195         struct btrfs_file_extent_item *item;
 196         int ret;
 197         int err;
 198         struct btrfs_key found_key;
 199
 200         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 201         path = btrfs_alloc_path();
 202         BUG_ON(!path);
 203 again:
 204         ret = btrfs_lookup_file_extent(NULL, root, path,
 205                                        inode->i_ino, start, 0);
 206         if (ret < 0) {
 207                 btrfs_free_path(path);
 208                 return ret;
 209         }
 210
 211         cow_end = end;
 212         if (ret != 0) {
 213                 if (path->slots[0] == 0)
 214                         goto not_found;
 215                 path->slots[0]--;
 216         }
 217
 218         leaf = path->nodes[0];
 219         item = btrfs_item_ptr(leaf, path->slots[0],
 220                               struct btrfs_file_extent_item);
 221
 222         /* are we inside the extent that was found? */
 223         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 224         found_type = btrfs_key_type(&found_key);
 225         if (found_key.objectid != inode->i_ino ||
 226             found_type != BTRFS_EXTENT_DATA_KEY)
 227                 goto not_found;
 228
 229         found_type = btrfs_file_extent_type(leaf, item);
 230         extent_start = found_key.offset;
 231         if (found_type == BTRFS_FILE_EXTENT_REG) {
 232                 u64 extent_num_bytes;
 233
 234                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 235                 extent_end = extent_start + extent_num_bytes;
 236                 err = 0;
 237
 238                 if (loops && start != extent_start)
 239                         goto not_found;
 240
 241                 if (start < extent_start || start >= extent_end)
 242                         goto not_found;
 243
 244                 cow_end = min(end, extent_end - 1);
 245                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 246                 if (bytenr == 0)
 247                         goto not_found;
 248
 249                 if (btrfs_count_snapshots_in_path(root, path, inode->i_ino,
 250                                                   bytenr) != 1) {
 251                         goto not_found;
 252                 }
 253
 254                 /*
 255                  * we may be called by the resizer, make sure we're inside
 256                  * the limits of the FS
 257                  */
 258                 block_group = btrfs_lookup_block_group(root->fs_info,
 259                                                        bytenr);
 260                 if (!block_group || block_group->ro)
 261                         goto not_found;
 262
 263                 start = extent_end;
 264         } else {
 265                 goto not_found;
 266         }
 267 loop:
 268         if (start > end) {
 269                 btrfs_free_path(path);
 270                 return 0;
 271         }
 272         btrfs_release_path(root, path);
 273         loops++;
 274         goto again;
 275
 276 not_found:
 277         cow_file_range(inode, start, end);
 278         start = end + 1;
 279         goto loop;
 280 }
 281
 282 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 283 {
 284         struct btrfs_root *root = BTRFS_I(inode)->root;
 285         int ret;
 286
 287         if (btrfs_test_opt(root, NODATACOW) ||
 288             btrfs_test_flag(inode, NODATACOW))
 289                 ret = run_delalloc_nocow(inode, start, end);
 290         else
 291                 ret = cow_file_range(inode, start, end);
 292
 293         return ret;
 294 }
 295
 296 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 297                        unsigned long old, unsigned long bits)
 298 {
 299         unsigned long flags;
 300         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 301                 struct btrfs_root *root = BTRFS_I(inode)->root;
 302                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 303                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 304                 root->fs_info->delalloc_bytes += end - start + 1;
 305                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 306         }
 307         return 0;
 308 }
 309
 310 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 311                          unsigned long old, unsigned long bits)
 312 {
 313         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 314                 struct btrfs_root *root = BTRFS_I(inode)->root;
 315                 unsigned long flags;
 316
 317                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 318                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 319                         printk("warning: delalloc account %Lu %Lu\n",
 320                                end - start + 1, root->fs_info->delalloc_bytes);
 321                         root->fs_info->delalloc_bytes = 0;
 322                         BTRFS_I(inode)->delalloc_bytes = 0;
 323                 } else {
 324                         root->fs_info->delalloc_bytes -= end - start + 1;
 325                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 326                 }
 327                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 328         }
 329         return 0;
 330 }
 331
 332 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 333                          size_t size, struct bio *bio)
 334 {
 335         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 336         struct btrfs_mapping_tree *map_tree;
 337         u64 logical = bio->bi_sector << 9;
 338         u64 length = 0;
 339         u64 map_length;
 340         int ret;
 341
 342         length = bio->bi_size;
 343         map_tree = &root->fs_info->mapping_tree;
 344         map_length = length;
 345         ret = btrfs_map_block(map_tree, READ, logical,
 346                               &map_length, NULL, 0);
 347
 348         if (map_length < length + size) {
 349                 return 1;
 350         }
 351         return 0;
 352 }
 353
 354 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 355                           int mirror_num)
 356 {
 357         struct btrfs_root *root = BTRFS_I(inode)->root;
 358         int ret = 0;
 359
 360         ret = btrfs_csum_one_bio(root, inode, bio);
 361         BUG_ON(ret);
 362
 363         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
 364 }
 365
 366 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 367                           int mirror_num)
 368 {
 369         struct btrfs_root *root = BTRFS_I(inode)->root;
 370         int ret = 0;
 371
 372         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 373         BUG_ON(ret);
 374
 375         if (!(rw & (1 << BIO_RW))) {
 376                 goto mapit;
 377         }
 378
 379         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 380                                    inode, rw, bio, mirror_num,
 381                                    __btrfs_submit_bio_hook);
 382 mapit:
 383         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
 384 }
 385
 386 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
 387                              struct inode *inode, u64 file_offset,
 388                              struct list_head *list)
 389 {
 390         struct list_head *cur;
 391         struct btrfs_ordered_sum *sum;
 392
 393         btrfs_set_trans_block_group(trans, inode);
 394         list_for_each(cur, list) {
 395                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
 396                 mutex_lock(&BTRFS_I(inode)->csum_mutex);
 397                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
 398                                        inode, sum);
 399                 mutex_unlock(&BTRFS_I(inode)->csum_mutex);
 400         }
 401         return 0;
 402 }
 403
 404 struct btrfs_writepage_fixup {
 405         struct page *page;
 406         struct btrfs_work work;
 407 };
 408
 409 /* see btrfs_writepage_start_hook for details on why this is required */
 410 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
 411 {
 412         struct btrfs_writepage_fixup *fixup;
 413         struct btrfs_ordered_extent *ordered;
 414         struct page *page;
 415         struct inode *inode;
 416         u64 page_start;
 417         u64 page_end;
 418
 419         fixup = container_of(work, struct btrfs_writepage_fixup, work);
 420         page = fixup->page;
 421 again:
 422         lock_page(page);
 423         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
 424                 ClearPageChecked(page);
 425                 goto out_page;
 426         }
 427
 428         inode = page->mapping->host;
 429         page_start = page_offset(page);
 430         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
 431
 432         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 433
 434         /* already ordered? We're done */
 435         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
 436                              EXTENT_ORDERED, 0)) {
 437                 goto out;
 438         }
 439
 440         ordered = btrfs_lookup_ordered_extent(inode, page_start);
 441         if (ordered) {
 442                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
 443                               page_end, GFP_NOFS);
 444                 unlock_page(page);
 445                 btrfs_start_ordered_extent(inode, ordered, 1);
 446                 goto again;
 447         }
 448
 449         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start, page_end,
 450                             GFP_NOFS);
 451         ClearPageChecked(page);
 452 out:
 453         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 454 out_page:
 455         unlock_page(page);
 456         page_cache_release(page);
 457 }
 458
 459 /*
 460  * There are a few paths in the higher layers of the kernel that directly
 461  * set the page dirty bit without asking the filesystem if it is a
 462  * good idea.  This causes problems because we want to make sure COW
 463  * properly happens and the data=ordered rules are followed.
 464  *
 465  * In our case any range that doesn't have the EXTENT_ORDERED bit set
 466  * hasn't been properly setup for IO.  We kick off an async process
 467  * to fix it up.  The async helper will wait for ordered extents, set
 468  * the delalloc bit and make it safe to write the page.
 469  */
 470 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
 471 {
 472         struct inode *inode = page->mapping->host;
 473         struct btrfs_writepage_fixup *fixup;
 474         struct btrfs_root *root = BTRFS_I(inode)->root;
 475         int ret;
 476
 477         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
 478                              EXTENT_ORDERED, 0);
 479         if (ret)
 480                 return 0;
 481
 482         if (PageChecked(page))
 483                 return -EAGAIN;
 484
 485         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
 486         if (!fixup)
 487                 return -EAGAIN;
 488
 489         SetPageChecked(page);
 490         page_cache_get(page);
 491         fixup->work.func = btrfs_writepage_fixup_worker;
 492         fixup->page = page;
 493         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
 494         return -EAGAIN;
 495 }
 496
 497 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 498 {
 499         struct btrfs_root *root = BTRFS_I(inode)->root;
 500         struct btrfs_trans_handle *trans;
 501         struct btrfs_ordered_extent *ordered_extent;
 502         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 503         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 504         struct extent_map *em;
 505         struct extent_map *em_orig;
 506         u64 alloc_hint = 0;
 507         u64 clear_start;
 508         u64 clear_end;
 509         struct list_head list;
 510         struct btrfs_key ins;
 511         struct rb_node *rb;
 512         int ret;
 513
 514         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
 515         if (!ret)
 516                 return 0;
 517
 518         trans = btrfs_join_transaction(root, 1);
 519
 520         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 521         BUG_ON(!ordered_extent);
 522
 523         lock_extent(io_tree, ordered_extent->file_offset,
 524                     ordered_extent->file_offset + ordered_extent->len - 1,
 525                     GFP_NOFS);
 526
 527         INIT_LIST_HEAD(&list);
 528
 529         ins.objectid = ordered_extent->start;
 530         ins.offset = ordered_extent->len;
 531         ins.type = BTRFS_EXTENT_ITEM_KEY;
 532
 533         ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
 534                                           trans->transid, inode->i_ino,
 535                                           ordered_extent->file_offset, &ins);
 536         BUG_ON(ret);
 537
 538         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 539
 540         spin_lock(&em_tree->lock);
 541         clear_start = ordered_extent->file_offset;
 542         clear_end = ordered_extent->file_offset + ordered_extent->len;
 543         em = lookup_extent_mapping(em_tree, clear_start,
 544                                    ordered_extent->len);
 545         em_orig = em;
 546         while(em && clear_start < extent_map_end(em) && clear_end > em->start) {
 547                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 548                 rb = rb_next(&em->rb_node);
 549                 if (!rb)
 550                         break;
 551                 em = rb_entry(rb, struct extent_map, rb_node);
 552         }
 553         free_extent_map(em_orig);
 554         spin_unlock(&em_tree->lock);
 555
 556         ret = btrfs_drop_extents(trans, root, inode,
 557                                  ordered_extent->file_offset,
 558                                  ordered_extent->file_offset +
 559                                  ordered_extent->len,
 560                                  ordered_extent->file_offset, &alloc_hint);
 561         BUG_ON(ret);
 562         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 563                                        ordered_extent->file_offset,
 564                                        ordered_extent->start,
 565                                        ordered_extent->len,
 566                                        ordered_extent->len, 0);
 567         BUG_ON(ret);
 568
 569         btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 570                                 ordered_extent->file_offset +
 571                                 ordered_extent->len - 1);
 572         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 573
 574         inode->i_blocks += ordered_extent->len >> 9;
 575         unlock_extent(io_tree, ordered_extent->file_offset,
 576                     ordered_extent->file_offset + ordered_extent->len - 1,
 577                     GFP_NOFS);
 578         add_pending_csums(trans, inode, ordered_extent->file_offset,
 579                           &ordered_extent->list);
 580
 581         btrfs_ordered_update_i_size(inode, ordered_extent);
 582         btrfs_remove_ordered_extent(inode, ordered_extent);
 583
 584         /* once for us */
 585         btrfs_put_ordered_extent(ordered_extent);
 586         /* once for the tree */
 587         btrfs_put_ordered_extent(ordered_extent);
 588
 589         btrfs_update_inode(trans, root, inode);
 590         btrfs_end_transaction(trans, root);
 591         return 0;
 592 }
 593
 594 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 595                                 struct extent_state *state, int uptodate)
 596 {
 597         return btrfs_finish_ordered_io(page->mapping->host, start, end);
 598 }
 599
 600 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 601 {
 602         int ret = 0;
 603         struct inode *inode = page->mapping->host;
 604         struct btrfs_root *root = BTRFS_I(inode)->root;
 605         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 606         struct btrfs_csum_item *item;
 607         struct btrfs_path *path = NULL;
 608         u32 csum;
 609
 610         if (btrfs_test_opt(root, NODATASUM) ||
 611             btrfs_test_flag(inode, NODATASUM))
 612                 return 0;
 613
 614         /*
 615          * It is possible there is an ordered extent that has
 616          * not yet finished for this range in the file.  If so,
 617          * that extent will have a csum cached, and it will insert
 618          * the sum after all the blocks in the extent are fully
 619          * on disk.  So, look for an ordered extent and use the
 620          * sum if found.  We have to do this before looking in the
 621          * btree because csum items are pre-inserted based on
 622          * the file size.  btrfs_lookup_csum might find an item
 623          * that still hasn't been fully filled.
 624          */
 625         ret = btrfs_find_ordered_sum(inode, start, &csum);
 626         if (ret == 0)
 627                 goto found;
 628
 629         ret = 0;
 630         path = btrfs_alloc_path();
 631         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 632         if (IS_ERR(item)) {
 633                 ret = PTR_ERR(item);
 634                 /* a csum that isn't present is a preallocated region. */
 635                 if (ret == -ENOENT || ret == -EFBIG)
 636                         ret = 0;
 637                 csum = 0;
 638                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino,
 639                        start);
 640                 goto out;
 641         }
 642         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 643                            BTRFS_CRC32_SIZE);
 644 found:
 645         set_state_private(io_tree, start, csum);
 646 out:
 647         if (path)
 648                 btrfs_free_path(path);
 649         return ret;
 650 }
 651
 652 struct io_failure_record {
 653         struct page *page;
 654         u64 start;
 655         u64 len;
 656         u64 logical;
 657         int last_mirror;
 658 };
 659
 660 int btrfs_io_failed_hook(struct bio *failed_bio,
 661                          struct page *page, u64 start, u64 end,
 662                          struct extent_state *state)
 663 {
 664         struct io_failure_record *failrec = NULL;
 665         u64 private;
 666         struct extent_map *em;
 667         struct inode *inode = page->mapping->host;
 668         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 669         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 670         struct bio *bio;
 671         int num_copies;
 672         int ret;
 673         int rw;
 674         u64 logical;
 675
 676         ret = get_state_private(failure_tree, start, &private);
 677         if (ret) {
 678                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 679                 if (!failrec)
 680                         return -ENOMEM;
 681                 failrec->start = start;
 682                 failrec->len = end - start + 1;
 683                 failrec->last_mirror = 0;
 684
 685                 spin_lock(&em_tree->lock);
 686                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 687                 if (em->start > start || em->start + em->len < start) {
 688                         free_extent_map(em);
 689                         em = NULL;
 690                 }
 691                 spin_unlock(&em_tree->lock);
 692
 693                 if (!em || IS_ERR(em)) {
 694                         kfree(failrec);
 695                         return -EIO;
 696                 }
 697                 logical = start - em->start;
 698                 logical = em->block_start + logical;
 699                 failrec->logical = logical;
 700                 free_extent_map(em);
 701                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 702                                 EXTENT_DIRTY, GFP_NOFS);
 703                 set_state_private(failure_tree, start,
 704                                  (u64)(unsigned long)failrec);
 705         } else {
 706                 failrec = (struct io_failure_record *)(unsigned long)private;
 707         }
 708         num_copies = btrfs_num_copies(
 709                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 710                               failrec->logical, failrec->len);
 711         failrec->last_mirror++;
 712         if (!state) {
 713                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 714                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 715                                                     failrec->start,
 716                                                     EXTENT_LOCKED);
 717                 if (state && state->start != failrec->start)
 718                         state = NULL;
 719                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 720         }
 721         if (!state || failrec->last_mirror > num_copies) {
 722                 set_state_private(failure_tree, failrec->start, 0);
 723                 clear_extent_bits(failure_tree, failrec->start,
 724                                   failrec->start + failrec->len - 1,
 725                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 726                 kfree(failrec);
 727                 return -EIO;
 728         }
 729         bio = bio_alloc(GFP_NOFS, 1);
 730         bio->bi_private = state;
 731         bio->bi_end_io = failed_bio->bi_end_io;
 732         bio->bi_sector = failrec->logical >> 9;
 733         bio->bi_bdev = failed_bio->bi_bdev;
 734         bio->bi_size = 0;
 735         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 736         if (failed_bio->bi_rw & (1 << BIO_RW))
 737                 rw = WRITE;
 738         else
 739                 rw = READ;
 740
 741         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 742                                                       failrec->last_mirror);
 743         return 0;
 744 }
 745
 746 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 747 {
 748         u64 private;
 749         u64 private_failure;
 750         struct io_failure_record *failure;
 751         int ret;
 752
 753         private = 0;
 754         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 755                              (u64)-1, 1, EXTENT_DIRTY)) {
 756                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 757                                         start, &private_failure);
 758                 if (ret == 0) {
 759                         failure = (struct io_failure_record *)(unsigned long)
 760                                    private_failure;
 761                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 762                                           failure->start, 0);
 763                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 764                                           failure->start,
 765                                           failure->start + failure->len - 1,
 766                                           EXTENT_DIRTY | EXTENT_LOCKED,
 767                                           GFP_NOFS);
 768                         kfree(failure);
 769                 }
 770         }
 771         return 0;
 772 }
 773
 774 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 775                                struct extent_state *state)
 776 {
 777         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 778         struct inode *inode = page->mapping->host;
 779         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 780         char *kaddr;
 781         u64 private = ~(u32)0;
 782         int ret;
 783         struct btrfs_root *root = BTRFS_I(inode)->root;
 784         u32 csum = ~(u32)0;
 785         unsigned long flags;
 786
 787         if (btrfs_test_opt(root, NODATASUM) ||
 788             btrfs_test_flag(inode, NODATASUM))
 789                 return 0;
 790         if (state && state->start == start) {
 791                 private = state->private;
 792                 ret = 0;
 793         } else {
 794                 ret = get_state_private(io_tree, start, &private);
 795         }
 796         local_irq_save(flags);
 797         kaddr = kmap_atomic(page, KM_IRQ0);
 798         if (ret) {
 799                 goto zeroit;
 800         }
 801         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 802         btrfs_csum_final(csum, (char *)&csum);
 803         if (csum != private) {
 804                 goto zeroit;
 805         }
 806         kunmap_atomic(kaddr, KM_IRQ0);
 807         local_irq_restore(flags);
 808
 809         /* if the io failure tree for this inode is non-empty,
 810          * check to see if we've recovered from a failed IO
 811          */
 812         btrfs_clean_io_failures(inode, start);
 813         return 0;
 814
 815 zeroit:
 816         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 817                page->mapping->host->i_ino, (unsigned long long)start, csum,
 818                private);
 819         memset(kaddr + offset, 1, end - start + 1);
 820         flush_dcache_page(page);
 821         kunmap_atomic(kaddr, KM_IRQ0);
 822         local_irq_restore(flags);
 823         if (private == 0)
 824                 return 0;
 825         return -EIO;
 826 }
 827
 828 void btrfs_read_locked_inode(struct inode *inode)
 829 {
 830         struct btrfs_path *path;
 831         struct extent_buffer *leaf;
 832         struct btrfs_inode_item *inode_item;
 833         struct btrfs_timespec *tspec;
 834         struct btrfs_root *root = BTRFS_I(inode)->root;
 835         struct btrfs_key location;
 836         u64 alloc_group_block;
 837         u32 rdev;
 838         int ret;
 839
 840         path = btrfs_alloc_path();
 841         BUG_ON(!path);
 842         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 843
 844         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 845         if (ret)
 846                 goto make_bad;
 847
 848         leaf = path->nodes[0];
 849         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 850                                     struct btrfs_inode_item);
 851
 852         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 853         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 854         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 855         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 856         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
 857
 858         tspec = btrfs_inode_atime(inode_item);
 859         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 860         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 861
 862         tspec = btrfs_inode_mtime(inode_item);
 863         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 864         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 865
 866         tspec = btrfs_inode_ctime(inode_item);
 867         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 868         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 869
 870         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 871         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 872         inode->i_rdev = 0;
 873         rdev = btrfs_inode_rdev(leaf, inode_item);
 874
 875         BTRFS_I(inode)->index_cnt = (u64)-1;
 876
 877         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 878         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 879                                                        alloc_group_block);
 880         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 881         if (!BTRFS_I(inode)->block_group) {
 882                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 883                                                  NULL, 0,
 884                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
 885         }
 886         btrfs_free_path(path);
 887         inode_item = NULL;
 888
 889         switch (inode->i_mode & S_IFMT) {
 890         case S_IFREG:
 891                 inode->i_mapping->a_ops = &btrfs_aops;
 892                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
 893                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 894                 inode->i_fop = &btrfs_file_operations;
 895                 inode->i_op = &btrfs_file_inode_operations;
 896                 break;
 897         case S_IFDIR:
 898                 inode->i_fop = &btrfs_dir_file_operations;
 899                 if (root == root->fs_info->tree_root)
 900                         inode->i_op = &btrfs_dir_ro_inode_operations;
 901                 else
 902                         inode->i_op = &btrfs_dir_inode_operations;
 903                 break;
 904         case S_IFLNK:
 905                 inode->i_op = &btrfs_symlink_inode_operations;
 906                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 907                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
 908                 break;
 909         default:
 910                 init_special_inode(inode, inode->i_mode, rdev);
 911                 break;
 912         }
 913         return;
 914
 915 make_bad:
 916         btrfs_free_path(path);
 917         make_bad_inode(inode);
 918 }
 919
 920 static void fill_inode_item(struct extent_buffer *leaf,
 921                             struct btrfs_inode_item *item,
 922                             struct inode *inode)
 923 {
 924         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 925         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 926         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
 927         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 928         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 929
 930         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 931                                inode->i_atime.tv_sec);
 932         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 933                                 inode->i_atime.tv_nsec);
 934
 935         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 936                                inode->i_mtime.tv_sec);
 937         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 938                                 inode->i_mtime.tv_nsec);
 939
 940         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 941                                inode->i_ctime.tv_sec);
 942         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 943                                 inode->i_ctime.tv_nsec);
 944
 945         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 946         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 947         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 948         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 949         btrfs_set_inode_block_group(leaf, item,
 950                                     BTRFS_I(inode)->block_group->key.objectid);
 951 }
 952
 953 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
 954                               struct btrfs_root *root,
 955                               struct inode *inode)
 956 {
 957         struct btrfs_inode_item *inode_item;
 958         struct btrfs_path *path;
 959         struct extent_buffer *leaf;
 960         int ret;
 961
 962         path = btrfs_alloc_path();
 963         BUG_ON(!path);
 964         ret = btrfs_lookup_inode(trans, root, path,
 965                                  &BTRFS_I(inode)->location, 1);
 966         if (ret) {
 967                 if (ret > 0)
 968                         ret = -ENOENT;
 969                 goto failed;
 970         }
 971
 972         leaf = path->nodes[0];
 973         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 974                                   struct btrfs_inode_item);
 975
 976         fill_inode_item(leaf, inode_item, inode);
 977         btrfs_mark_buffer_dirty(leaf);
 978         btrfs_set_inode_last_trans(trans, inode);
 979         ret = 0;
 980 failed:
 981         btrfs_free_path(path);
 982         return ret;
 983 }
 984
 985
 986 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 987                               struct btrfs_root *root,
 988                               struct inode *dir,
 989                               struct dentry *dentry)
 990 {
 991         struct btrfs_path *path;
 992         const char *name = dentry->d_name.name;
 993         int name_len = dentry->d_name.len;
 994         int ret = 0;
 995         struct extent_buffer *leaf;
 996         struct btrfs_dir_item *di;
 997         struct btrfs_key key;
 998         u64 index;
 999
1000         path = btrfs_alloc_path();
1001         if (!path) {
1002                 ret = -ENOMEM;
1003                 goto err;
1004         }
1005
1006         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1007                                     name, name_len, -1);
1008         if (IS_ERR(di)) {
1009                 ret = PTR_ERR(di);
1010                 goto err;
1011         }
1012         if (!di) {
1013                 ret = -ENOENT;
1014                 goto err;
1015         }
1016         leaf = path->nodes[0];
1017         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1018         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1019         if (ret)
1020                 goto err;
1021         btrfs_release_path(root, path);
1022
1023         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1024                                   dentry->d_inode->i_ino,
1025                                   dentry->d_parent->d_inode->i_ino, &index);
1026         if (ret) {
1027                 printk("failed to delete reference to %.*s, "
1028                        "inode %lu parent %lu\n", name_len, name,
1029                        dentry->d_inode->i_ino,
1030                        dentry->d_parent->d_inode->i_ino);
1031                 goto err;
1032         }
1033
1034         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1035                                          index, name, name_len, -1);
1036         if (IS_ERR(di)) {
1037                 ret = PTR_ERR(di);
1038                 goto err;
1039         }
1040         if (!di) {
1041                 ret = -ENOENT;
1042                 goto err;
1043         }
1044         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1045         btrfs_release_path(root, path);
1046
1047         dentry->d_inode->i_ctime = dir->i_ctime;
1048 err:
1049         btrfs_free_path(path);
1050         if (!ret) {
1051                 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1052                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1053                 btrfs_update_inode(trans, root, dir);
1054 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1055                 dentry->d_inode->i_nlink--;
1056 #else
1057                 drop_nlink(dentry->d_inode);
1058 #endif
1059                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
1060                 dir->i_sb->s_dirt = 1;
1061         }
1062         return ret;
1063 }
1064
1065 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1066 {
1067         struct btrfs_root *root;
1068         struct btrfs_trans_handle *trans;
1069         int ret;
1070         unsigned long nr = 0;
1071
1072         root = BTRFS_I(dir)->root;
1073
1074         ret = btrfs_check_free_space(root, 1, 1);
1075         if (ret)
1076                 goto fail;
1077
1078         trans = btrfs_start_transaction(root, 1);
1079
1080         btrfs_set_trans_block_group(trans, dir);
1081         ret = btrfs_unlink_trans(trans, root, dir, dentry);
1082         nr = trans->blocks_used;
1083
1084         btrfs_end_transaction_throttle(trans, root);
1085 fail:
1086         btrfs_btree_balance_dirty(root, nr);
1087         return ret;
1088 }
1089
1090 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1091 {
1092         struct inode *inode = dentry->d_inode;
1093         int err = 0;
1094         int ret;
1095         struct btrfs_root *root = BTRFS_I(dir)->root;
1096         struct btrfs_trans_handle *trans;
1097         unsigned long nr = 0;
1098
1099         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1100                 return -ENOTEMPTY;
1101         }
1102
1103         ret = btrfs_check_free_space(root, 1, 1);
1104         if (ret)
1105                 goto fail;
1106
1107         trans = btrfs_start_transaction(root, 1);
1108         btrfs_set_trans_block_group(trans, dir);
1109
1110         /* now the directory is empty */
1111         err = btrfs_unlink_trans(trans, root, dir, dentry);
1112         if (!err) {
1113                 btrfs_i_size_write(inode, 0);
1114         }
1115
1116         nr = trans->blocks_used;
1117         ret = btrfs_end_transaction_throttle(trans, root);
1118 fail:
1119         btrfs_btree_balance_dirty(root, nr);
1120
1121         if (ret && !err)
1122                 err = ret;
1123         return err;
1124 }
1125
1126 /*
1127  * this can truncate away extent items, csum items and directory items.
1128  * It starts at a high offset and removes keys until it can't find
1129  * any higher than i_size.
1130  *
1131  * csum items that cross the new i_size are truncated to the new size
1132  * as well.
1133  */
1134 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1135                                    struct btrfs_root *root,
1136                                    struct inode *inode,
1137                                    u32 min_type)
1138 {
1139         int ret;
1140         struct btrfs_path *path;
1141         struct btrfs_key key;
1142         struct btrfs_key found_key;
1143         u32 found_type;
1144         struct extent_buffer *leaf;
1145         struct btrfs_file_extent_item *fi;
1146         u64 extent_start = 0;
1147         u64 extent_num_bytes = 0;
1148         u64 item_end = 0;
1149         u64 root_gen = 0;
1150         u64 root_owner = 0;
1151         int found_extent;
1152         int del_item;
1153         int pending_del_nr = 0;
1154         int pending_del_slot = 0;
1155         int extent_type = -1;
1156         u64 mask = root->sectorsize - 1;
1157
1158         btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
1159         path = btrfs_alloc_path();
1160         path->reada = -1;
1161         BUG_ON(!path);
1162
1163         /* FIXME, add redo link to tree so we don't leak on crash */
1164         key.objectid = inode->i_ino;
1165         key.offset = (u64)-1;
1166         key.type = (u8)-1;
1167
1168         btrfs_init_path(path);
1169 search_again:
1170         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1171         if (ret < 0) {
1172                 goto error;
1173         }
1174         if (ret > 0) {
1175                 BUG_ON(path->slots[0] == 0);
1176                 path->slots[0]--;
1177         }
1178
1179         while(1) {
1180                 fi = NULL;
1181                 leaf = path->nodes[0];
1182                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1183                 found_type = btrfs_key_type(&found_key);
1184
1185                 if (found_key.objectid != inode->i_ino)
1186                         break;
1187
1188                 if (found_type < min_type)
1189                         break;
1190
1191                 item_end = found_key.offset;
1192                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1193                         fi = btrfs_item_ptr(leaf, path->slots[0],
1194                                             struct btrfs_file_extent_item);
1195                         extent_type = btrfs_file_extent_type(leaf, fi);
1196                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1197                                 item_end +=
1198                                     btrfs_file_extent_num_bytes(leaf, fi);
1199                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1200                                 struct btrfs_item *item = btrfs_item_nr(leaf,
1201                                                                 path->slots[0]);
1202                                 item_end += btrfs_file_extent_inline_len(leaf,
1203                                                                          item);
1204                         }
1205                         item_end--;
1206                 }
1207                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1208                         ret = btrfs_csum_truncate(trans, root, path,
1209                                                   inode->i_size);
1210                         BUG_ON(ret);
1211                 }
1212                 if (item_end < inode->i_size) {
1213                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1214                                 found_type = BTRFS_INODE_ITEM_KEY;
1215                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1216                                 found_type = BTRFS_CSUM_ITEM_KEY;
1217                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1218                                 found_type = BTRFS_XATTR_ITEM_KEY;
1219                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1220                                 found_type = BTRFS_INODE_REF_KEY;
1221                         } else if (found_type) {
1222                                 found_type--;
1223                         } else {
1224                                 break;
1225                         }
1226                         btrfs_set_key_type(&key, found_type);
1227                         goto next;
1228                 }
1229                 if (found_key.offset >= inode->i_size)
1230                         del_item = 1;
1231                 else
1232                         del_item = 0;
1233                 found_extent = 0;
1234
1235                 /* FIXME, shrink the extent if the ref count is only 1 */
1236                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1237                         goto delete;
1238
1239                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1240                         u64 num_dec;
1241                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1242                         if (!del_item) {
1243                                 u64 orig_num_bytes =
1244                                         btrfs_file_extent_num_bytes(leaf, fi);
1245                                 extent_num_bytes = inode->i_size -
1246                                         found_key.offset + root->sectorsize - 1;
1247                                 extent_num_bytes = extent_num_bytes &
1248                                         ~((u64)root->sectorsize - 1);
1249                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1250                                                          extent_num_bytes);
1251                                 num_dec = (orig_num_bytes -
1252                                            extent_num_bytes);
1253                                 if (extent_start != 0)
1254                                         dec_i_blocks(inode, num_dec);
1255                                 btrfs_mark_buffer_dirty(leaf);
1256                         } else {
1257                                 extent_num_bytes =
1258                                         btrfs_file_extent_disk_num_bytes(leaf,
1259                                                                          fi);
1260                                 /* FIXME blocksize != 4096 */
1261                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1262                                 if (extent_start != 0) {
1263                                         found_extent = 1;
1264                                         dec_i_blocks(inode, num_dec);
1265                                 }
1266                                 root_gen = btrfs_header_generation(leaf);
1267                                 root_owner = btrfs_header_owner(leaf);
1268                         }
1269                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1270                         if (!del_item) {
1271                                 u32 newsize = inode->i_size - found_key.offset;
1272                                 dec_i_blocks(inode, item_end + 1 -
1273                                             found_key.offset - newsize);
1274                                 newsize =
1275                                     btrfs_file_extent_calc_inline_size(newsize);
1276                                 ret = btrfs_truncate_item(trans, root, path,
1277                                                           newsize, 1);
1278                                 BUG_ON(ret);
1279                         } else {
1280                                 dec_i_blocks(inode, item_end + 1 -
1281                                              found_key.offset);
1282                         }
1283                 }
1284 delete:
1285                 if (del_item) {
1286                         if (!pending_del_nr) {
1287                                 /* no pending yet, add ourselves */
1288                                 pending_del_slot = path->slots[0];
1289                                 pending_del_nr = 1;
1290                         } else if (pending_del_nr &&
1291                                    path->slots[0] + 1 == pending_del_slot) {
1292                                 /* hop on the pending chunk */
1293                                 pending_del_nr++;
1294                                 pending_del_slot = path->slots[0];
1295                         } else {
1296                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1297                         }
1298                 } else {
1299                         break;
1300                 }
1301                 if (found_extent) {
1302                         ret = btrfs_free_extent(trans, root, extent_start,
1303                                                 extent_num_bytes,
1304                                                 root_owner,
1305                                                 root_gen, inode->i_ino,
1306                                                 found_key.offset, 0);
1307                         BUG_ON(ret);
1308                 }
1309 next:
1310                 if (path->slots[0] == 0) {
1311                         if (pending_del_nr)
1312                                 goto del_pending;
1313                         btrfs_release_path(root, path);
1314                         goto search_again;
1315                 }
1316
1317                 path->slots[0]--;
1318                 if (pending_del_nr &&
1319                     path->slots[0] + 1 != pending_del_slot) {
1320                         struct btrfs_key debug;
1321 del_pending:
1322                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1323                                               pending_del_slot);
1324                         ret = btrfs_del_items(trans, root, path,
1325                                               pending_del_slot,
1326                                               pending_del_nr);
1327                         BUG_ON(ret);
1328                         pending_del_nr = 0;
1329                         btrfs_release_path(root, path);
1330                         goto search_again;
1331                 }
1332         }
1333         ret = 0;
1334 error:
1335         if (pending_del_nr) {
1336                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1337                                       pending_del_nr);
1338         }
1339         btrfs_free_path(path);
1340         inode->i_sb->s_dirt = 1;
1341         return ret;
1342 }
1343
1344 /*
1345  * taken from block_truncate_page, but does cow as it zeros out
1346  * any bytes left in the last page in the file.
1347  */
1348 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1349 {
1350         struct inode *inode = mapping->host;
1351         struct btrfs_root *root = BTRFS_I(inode)->root;
1352         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1353         struct btrfs_ordered_extent *ordered;
1354         char *kaddr;
1355         u32 blocksize = root->sectorsize;
1356         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1357         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1358         struct page *page;
1359         int ret = 0;
1360         u64 page_start;
1361         u64 page_end;
1362
1363         if ((offset & (blocksize - 1)) == 0)
1364                 goto out;
1365
1366         ret = -ENOMEM;
1367 again:
1368         page = grab_cache_page(mapping, index);
1369         if (!page)
1370                 goto out;
1371
1372         page_start = page_offset(page);
1373         page_end = page_start + PAGE_CACHE_SIZE - 1;
1374
1375         if (!PageUptodate(page)) {
1376                 ret = btrfs_readpage(NULL, page);
1377                 lock_page(page);
1378                 if (page->mapping != mapping) {
1379                         unlock_page(page);
1380                         page_cache_release(page);
1381                         goto again;
1382                 }
1383                 if (!PageUptodate(page)) {
1384                         ret = -EIO;
1385                         goto out_unlock;
1386                 }
1387         }
1388         wait_on_page_writeback(page);
1389
1390         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1391         set_page_extent_mapped(page);
1392
1393         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1394         if (ordered) {
1395                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1396                 unlock_page(page);
1397                 page_cache_release(page);
1398                 btrfs_start_ordered_extent(inode, ordered, 1);
1399                 btrfs_put_ordered_extent(ordered);
1400                 goto again;
1401         }
1402
1403         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
1404                             page_end, GFP_NOFS);
1405         ret = 0;
1406         if (offset != PAGE_CACHE_SIZE) {
1407                 kaddr = kmap(page);
1408                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1409                 flush_dcache_page(page);
1410                 kunmap(page);
1411         }
1412         ClearPageChecked(page);
1413         set_page_dirty(page);
1414         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1415
1416 out_unlock:
1417         unlock_page(page);
1418         page_cache_release(page);
1419 out:
1420         return ret;
1421 }
1422
1423 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1424 {
1425         struct inode *inode = dentry->d_inode;
1426         int err;
1427
1428         err = inode_change_ok(inode, attr);
1429         if (err)
1430                 return err;
1431
1432         if (S_ISREG(inode->i_mode) &&
1433             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1434                 struct btrfs_trans_handle *trans;
1435                 struct btrfs_root *root = BTRFS_I(inode)->root;
1436                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1437
1438                 u64 mask = root->sectorsize - 1;
1439                 u64 hole_start = (inode->i_size + mask) & ~mask;
1440                 u64 block_end = (attr->ia_size + mask) & ~mask;
1441                 u64 hole_size;
1442                 u64 alloc_hint = 0;
1443
1444                 if (attr->ia_size <= hole_start)
1445                         goto out;
1446
1447                 err = btrfs_check_free_space(root, 1, 0);
1448                 if (err)
1449                         goto fail;
1450
1451                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1452
1453                 hole_size = block_end - hole_start;
1454                 btrfs_wait_ordered_range(inode, hole_start, hole_size);
1455                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1456
1457                 trans = btrfs_start_transaction(root, 1);
1458                 btrfs_set_trans_block_group(trans, inode);
1459                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1460                 err = btrfs_drop_extents(trans, root, inode,
1461                                          hole_start, block_end, hole_start,
1462                                          &alloc_hint);
1463
1464                 if (alloc_hint != EXTENT_MAP_INLINE) {
1465                         err = btrfs_insert_file_extent(trans, root,
1466                                                        inode->i_ino,
1467                                                        hole_start, 0, 0,
1468                                                        hole_size, 0);
1469                         btrfs_drop_extent_cache(inode, hole_start,
1470                                                 (u64)-1);
1471                         btrfs_check_file(root, inode);
1472                 }
1473                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1474                 btrfs_end_transaction(trans, root);
1475                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1476                 if (err)
1477                         return err;
1478         }
1479 out:
1480         err = inode_setattr(inode, attr);
1481 fail:
1482         return err;
1483 }
1484
1485 void btrfs_delete_inode(struct inode *inode)
1486 {
1487         struct btrfs_trans_handle *trans;
1488         struct btrfs_root *root = BTRFS_I(inode)->root;
1489         unsigned long nr;
1490         int ret;
1491
1492         truncate_inode_pages(&inode->i_data, 0);
1493         if (is_bad_inode(inode)) {
1494                 goto no_delete;
1495         }
1496         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1497
1498         btrfs_i_size_write(inode, 0);
1499         trans = btrfs_start_transaction(root, 1);
1500
1501         btrfs_set_trans_block_group(trans, inode);
1502         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1503         if (ret)
1504                 goto no_delete_lock;
1505
1506         nr = trans->blocks_used;
1507         clear_inode(inode);
1508
1509         btrfs_end_transaction(trans, root);
1510         btrfs_btree_balance_dirty(root, nr);
1511         return;
1512
1513 no_delete_lock:
1514         nr = trans->blocks_used;
1515         btrfs_end_transaction(trans, root);
1516         btrfs_btree_balance_dirty(root, nr);
1517 no_delete:
1518         clear_inode(inode);
1519 }
1520
1521 /*
1522  * this returns the key found in the dir entry in the location pointer.
1523  * If no dir entries were found, location->objectid is 0.
1524  */
1525 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1526                                struct btrfs_key *location)
1527 {
1528         const char *name = dentry->d_name.name;
1529         int namelen = dentry->d_name.len;
1530         struct btrfs_dir_item *di;
1531         struct btrfs_path *path;
1532         struct btrfs_root *root = BTRFS_I(dir)->root;
1533         int ret = 0;
1534
1535         if (namelen == 1 && strcmp(name, ".") == 0) {
1536                 location->objectid = dir->i_ino;
1537                 location->type = BTRFS_INODE_ITEM_KEY;
1538                 location->offset = 0;
1539                 return 0;
1540         }
1541         path = btrfs_alloc_path();
1542         BUG_ON(!path);
1543
1544         if (namelen == 2 && strcmp(name, "..") == 0) {
1545                 struct btrfs_key key;
1546                 struct extent_buffer *leaf;
1547                 u32 nritems;
1548                 int slot;
1549
1550                 key.objectid = dir->i_ino;
1551                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1552                 key.offset = 0;
1553                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1554                 BUG_ON(ret == 0);
1555                 ret = 0;
1556
1557                 leaf = path->nodes[0];
1558                 slot = path->slots[0];
1559                 nritems = btrfs_header_nritems(leaf);
1560                 if (slot >= nritems)
1561                         goto out_err;
1562
1563                 btrfs_item_key_to_cpu(leaf, &key, slot);
1564                 if (key.objectid != dir->i_ino ||
1565                     key.type != BTRFS_INODE_REF_KEY) {
1566                         goto out_err;
1567                 }
1568                 location->objectid = key.offset;
1569                 location->type = BTRFS_INODE_ITEM_KEY;
1570                 location->offset = 0;
1571                 goto out;
1572         }
1573
1574         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1575                                     namelen, 0);
1576         if (IS_ERR(di))
1577                 ret = PTR_ERR(di);
1578         if (!di || IS_ERR(di)) {
1579                 goto out_err;
1580         }
1581         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1582 out:
1583         btrfs_free_path(path);
1584         return ret;
1585 out_err:
1586         location->objectid = 0;
1587         goto out;
1588 }
1589
1590 /*
1591  * when we hit a tree root in a directory, the btrfs part of the inode
1592  * needs to be changed to reflect the root directory of the tree root.  This
1593  * is kind of like crossing a mount point.
1594  */
1595 static int fixup_tree_root_location(struct btrfs_root *root,
1596                              struct btrfs_key *location,
1597                              struct btrfs_root **sub_root,
1598                              struct dentry *dentry)
1599 {
1600         struct btrfs_path *path;
1601         struct btrfs_root_item *ri;
1602
1603         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1604                 return 0;
1605         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1606                 return 0;
1607
1608         path = btrfs_alloc_path();
1609         BUG_ON(!path);
1610
1611         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1612                                         dentry->d_name.name,
1613                                         dentry->d_name.len);
1614         if (IS_ERR(*sub_root))
1615                 return PTR_ERR(*sub_root);
1616
1617         ri = &(*sub_root)->root_item;
1618         location->objectid = btrfs_root_dirid(ri);
1619         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1620         location->offset = 0;
1621
1622         btrfs_free_path(path);
1623         return 0;
1624 }
1625
1626 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1627 {
1628         struct btrfs_iget_args *args = p;
1629         inode->i_ino = args->ino;
1630         BTRFS_I(inode)->root = args->root;
1631         BTRFS_I(inode)->delalloc_bytes = 0;
1632         BTRFS_I(inode)->disk_i_size = 0;
1633         BTRFS_I(inode)->index_cnt = (u64)-1;
1634         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1635         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1636                              inode->i_mapping, GFP_NOFS);
1637         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1638                              inode->i_mapping, GFP_NOFS);
1639         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1640         mutex_init(&BTRFS_I(inode)->csum_mutex);
1641         mutex_init(&BTRFS_I(inode)->extent_mutex);
1642         return 0;
1643 }
1644
1645 static int btrfs_find_actor(struct inode *inode, void *opaque)
1646 {
1647         struct btrfs_iget_args *args = opaque;
1648         return (args->ino == inode->i_ino &&
1649                 args->root == BTRFS_I(inode)->root);
1650 }
1651
1652 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1653                             u64 root_objectid)
1654 {
1655         struct btrfs_iget_args args;
1656         args.ino = objectid;
1657         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1658
1659         if (!args.root)
1660                 return NULL;
1661
1662         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1663 }
1664
1665 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1666                                 struct btrfs_root *root)
1667 {
1668         struct inode *inode;
1669         struct btrfs_iget_args args;
1670         args.ino = objectid;
1671         args.root = root;
1672
1673         inode = iget5_locked(s, objectid, btrfs_find_actor,
1674                              btrfs_init_locked_inode,
1675                              (void *)&args);
1676         return inode;
1677 }
1678
1679 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1680                                    struct nameidata *nd)
1681 {
1682         struct inode * inode;
1683         struct btrfs_inode *bi = BTRFS_I(dir);
1684         struct btrfs_root *root = bi->root;
1685         struct btrfs_root *sub_root = root;
1686         struct btrfs_key location;
1687         int ret;
1688
1689         if (dentry->d_name.len > BTRFS_NAME_LEN)
1690                 return ERR_PTR(-ENAMETOOLONG);
1691
1692         ret = btrfs_inode_by_name(dir, dentry, &location);
1693
1694         if (ret < 0)
1695                 return ERR_PTR(ret);
1696
1697         inode = NULL;
1698         if (location.objectid) {
1699                 ret = fixup_tree_root_location(root, &location, &sub_root,
1700                                                 dentry);
1701                 if (ret < 0)
1702                         return ERR_PTR(ret);
1703                 if (ret > 0)
1704                         return ERR_PTR(-ENOENT);
1705                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1706                                           sub_root);
1707                 if (!inode)
1708                         return ERR_PTR(-EACCES);
1709                 if (inode->i_state & I_NEW) {
1710                         /* the inode and parent dir are two different roots */
1711                         if (sub_root != root) {
1712                                 igrab(inode);
1713                                 sub_root->inode = inode;
1714                         }
1715                         BTRFS_I(inode)->root = sub_root;
1716                         memcpy(&BTRFS_I(inode)->location, &location,
1717                                sizeof(location));
1718                         btrfs_read_locked_inode(inode);
1719                         unlock_new_inode(inode);
1720                 }
1721         }
1722         return d_splice_alias(inode, dentry);
1723 }
1724
1725 static unsigned char btrfs_filetype_table[] = {
1726         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1727 };
1728
1729 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1730 {
1731         struct inode *inode = filp->f_dentry->d_inode;
1732         struct btrfs_root *root = BTRFS_I(inode)->root;
1733         struct btrfs_item *item;
1734         struct btrfs_dir_item *di;
1735         struct btrfs_key key;
1736         struct btrfs_key found_key;
1737         struct btrfs_path *path;
1738         int ret;
1739         u32 nritems;
1740         struct extent_buffer *leaf;
1741         int slot;
1742         int advance;
1743         unsigned char d_type;
1744         int over = 0;
1745         u32 di_cur;
1746         u32 di_total;
1747         u32 di_len;
1748         int key_type = BTRFS_DIR_INDEX_KEY;
1749         char tmp_name[32];
1750         char *name_ptr;
1751         int name_len;
1752
1753         /* FIXME, use a real flag for deciding about the key type */
1754         if (root->fs_info->tree_root == root)
1755                 key_type = BTRFS_DIR_ITEM_KEY;
1756
1757         /* special case for "." */
1758         if (filp->f_pos == 0) {
1759                 over = filldir(dirent, ".", 1,
1760                                1, inode->i_ino,
1761                                DT_DIR);
1762                 if (over)
1763                         return 0;
1764                 filp->f_pos = 1;
1765         }
1766
1767         key.objectid = inode->i_ino;
1768         path = btrfs_alloc_path();
1769         path->reada = 2;
1770
1771         /* special case for .., just use the back ref */
1772         if (filp->f_pos == 1) {
1773                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1774                 key.offset = 0;
1775                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1776                 BUG_ON(ret == 0);
1777                 leaf = path->nodes[0];
1778                 slot = path->slots[0];
1779                 nritems = btrfs_header_nritems(leaf);
1780                 if (slot >= nritems) {
1781                         btrfs_release_path(root, path);
1782                         goto read_dir_items;
1783                 }
1784                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1785                 btrfs_release_path(root, path);
1786                 if (found_key.objectid != key.objectid ||
1787                     found_key.type != BTRFS_INODE_REF_KEY)
1788                         goto read_dir_items;
1789                 over = filldir(dirent, "..", 2,
1790                                2, found_key.offset, DT_DIR);
1791                 if (over)
1792                         goto nopos;
1793                 filp->f_pos = 2;
1794         }
1795
1796 read_dir_items:
1797         btrfs_set_key_type(&key, key_type);
1798         key.offset = filp->f_pos;
1799
1800         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1801         if (ret < 0)
1802                 goto err;
1803         advance = 0;
1804         while(1) {
1805                 leaf = path->nodes[0];
1806                 nritems = btrfs_header_nritems(leaf);
1807                 slot = path->slots[0];
1808                 if (advance || slot >= nritems) {
1809                         if (slot >= nritems -1) {
1810                                 ret = btrfs_next_leaf(root, path);
1811                                 if (ret)
1812                                         break;
1813                                 leaf = path->nodes[0];
1814                                 nritems = btrfs_header_nritems(leaf);
1815                                 slot = path->slots[0];
1816                         } else {
1817                                 slot++;
1818                                 path->slots[0]++;
1819                         }
1820                 }
1821                 advance = 1;
1822                 item = btrfs_item_nr(leaf, slot);
1823                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1824
1825                 if (found_key.objectid != key.objectid)
1826                         break;
1827                 if (btrfs_key_type(&found_key) != key_type)
1828                         break;
1829                 if (found_key.offset < filp->f_pos)
1830                         continue;
1831
1832                 filp->f_pos = found_key.offset;
1833                 advance = 1;
1834                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1835                 di_cur = 0;
1836                 di_total = btrfs_item_size(leaf, item);
1837                 while(di_cur < di_total) {
1838                         struct btrfs_key location;
1839
1840                         name_len = btrfs_dir_name_len(leaf, di);
1841                         if (name_len < 32) {
1842                                 name_ptr = tmp_name;
1843                         } else {
1844                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1845                                 BUG_ON(!name_ptr);
1846                         }
1847                         read_extent_buffer(leaf, name_ptr,
1848                                            (unsigned long)(di + 1), name_len);
1849
1850                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1851                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1852                         over = filldir(dirent, name_ptr, name_len,
1853                                        found_key.offset,
1854                                        location.objectid,
1855                                        d_type);
1856
1857                         if (name_ptr != tmp_name)
1858                                 kfree(name_ptr);
1859
1860                         if (over)
1861                                 goto nopos;
1862                         di_len = btrfs_dir_name_len(leaf, di) +
1863                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1864                         di_cur += di_len;
1865                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1866                 }
1867         }
1868         if (key_type == BTRFS_DIR_INDEX_KEY)
1869                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1870         else
1871                 filp->f_pos++;
1872 nopos:
1873         ret = 0;
1874 err:
1875         btrfs_free_path(path);
1876         return ret;
1877 }
1878
1879 int btrfs_write_inode(struct inode *inode, int wait)
1880 {
1881         struct btrfs_root *root = BTRFS_I(inode)->root;
1882         struct btrfs_trans_handle *trans;
1883         int ret = 0;
1884
1885         if (wait) {
1886                 trans = btrfs_join_transaction(root, 1);
1887                 btrfs_set_trans_block_group(trans, inode);
1888                 ret = btrfs_commit_transaction(trans, root);
1889         }
1890         return ret;
1891 }
1892
1893 /*
1894  * This is somewhat expensive, updating the tree every time the
1895  * inode changes.  But, it is most likely to find the inode in cache.
1896  * FIXME, needs more benchmarking...there are no reasons other than performance
1897  * to keep or drop this code.
1898  */
1899 void btrfs_dirty_inode(struct inode *inode)
1900 {
1901         struct btrfs_root *root = BTRFS_I(inode)->root;
1902         struct btrfs_trans_handle *trans;
1903
1904         trans = btrfs_join_transaction(root, 1);
1905         btrfs_set_trans_block_group(trans, inode);
1906         btrfs_update_inode(trans, root, inode);
1907         btrfs_end_transaction(trans, root);
1908 }
1909
1910 static int btrfs_set_inode_index_count(struct inode *inode)
1911 {
1912         struct btrfs_root *root = BTRFS_I(inode)->root;
1913         struct btrfs_key key, found_key;
1914         struct btrfs_path *path;
1915         struct extent_buffer *leaf;
1916         int ret;
1917
1918         key.objectid = inode->i_ino;
1919         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
1920         key.offset = (u64)-1;
1921
1922         path = btrfs_alloc_path();
1923         if (!path)
1924                 return -ENOMEM;
1925
1926         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1927         if (ret < 0)
1928                 goto out;
1929         /* FIXME: we should be able to handle this */
1930         if (ret == 0)
1931                 goto out;
1932         ret = 0;
1933
1934         /*
1935          * MAGIC NUMBER EXPLANATION:
1936          * since we search a directory based on f_pos we have to start at 2
1937          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
1938          * else has to start at 2
1939          */
1940         if (path->slots[0] == 0) {
1941                 BTRFS_I(inode)->index_cnt = 2;
1942                 goto out;
1943         }
1944
1945         path->slots[0]--;
1946
1947         leaf = path->nodes[0];
1948         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1949
1950         if (found_key.objectid != inode->i_ino ||
1951             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
1952                 BTRFS_I(inode)->index_cnt = 2;
1953                 goto out;
1954         }
1955
1956         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
1957 out:
1958         btrfs_free_path(path);
1959         return ret;
1960 }
1961
1962 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode)
1963 {
1964         int ret = 0;
1965
1966         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
1967                 ret = btrfs_set_inode_index_count(dir);
1968                 if (ret)
1969                         return ret;
1970         }
1971
1972         BTRFS_I(inode)->index = BTRFS_I(dir)->index_cnt;
1973         BTRFS_I(dir)->index_cnt++;
1974
1975         return ret;
1976 }
1977
1978 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1979                                      struct btrfs_root *root,
1980                                      struct inode *dir,
1981                                      const char *name, int name_len,
1982                                      u64 ref_objectid,
1983                                      u64 objectid,
1984                                      struct btrfs_block_group_cache *group,
1985                                      int mode)
1986 {
1987         struct inode *inode;
1988         struct btrfs_inode_item *inode_item;
1989         struct btrfs_block_group_cache *new_inode_group;
1990         struct btrfs_key *location;
1991         struct btrfs_path *path;
1992         struct btrfs_inode_ref *ref;
1993         struct btrfs_key key[2];
1994         u32 sizes[2];
1995         unsigned long ptr;
1996         int ret;
1997         int owner;
1998
1999         path = btrfs_alloc_path();
2000         BUG_ON(!path);
2001
2002         inode = new_inode(root->fs_info->sb);
2003         if (!inode)
2004                 return ERR_PTR(-ENOMEM);
2005
2006         if (dir) {
2007                 ret = btrfs_set_inode_index(dir, inode);
2008                 if (ret)
2009                         return ERR_PTR(ret);
2010         } else {
2011                 BTRFS_I(inode)->index = 0;
2012         }
2013         /*
2014          * index_cnt is ignored for everything but a dir,
2015          * btrfs_get_inode_index_count has an explanation for the magic
2016          * number
2017          */
2018         BTRFS_I(inode)->index_cnt = 2;
2019
2020         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2021         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2022                              inode->i_mapping, GFP_NOFS);
2023         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2024                              inode->i_mapping, GFP_NOFS);
2025         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2026         mutex_init(&BTRFS_I(inode)->csum_mutex);
2027         mutex_init(&BTRFS_I(inode)->extent_mutex);
2028         BTRFS_I(inode)->delalloc_bytes = 0;
2029         BTRFS_I(inode)->disk_i_size = 0;
2030         BTRFS_I(inode)->root = root;
2031
2032         if (mode & S_IFDIR)
2033                 owner = 0;
2034         else
2035                 owner = 1;
2036         new_inode_group = btrfs_find_block_group(root, group, 0,
2037                                        BTRFS_BLOCK_GROUP_METADATA, owner);
2038         if (!new_inode_group) {
2039                 printk("find_block group failed\n");
2040                 new_inode_group = group;
2041         }
2042         BTRFS_I(inode)->block_group = new_inode_group;
2043         BTRFS_I(inode)->flags = 0;
2044
2045         key[0].objectid = objectid;
2046         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2047         key[0].offset = 0;
2048
2049         key[1].objectid = objectid;
2050         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2051         key[1].offset = ref_objectid;
2052
2053         sizes[0] = sizeof(struct btrfs_inode_item);
2054         sizes[1] = name_len + sizeof(*ref);
2055
2056         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2057         if (ret != 0)
2058                 goto fail;
2059
2060         if (objectid > root->highest_inode)
2061                 root->highest_inode = objectid;
2062
2063         inode->i_uid = current->fsuid;
2064         inode->i_gid = current->fsgid;
2065         inode->i_mode = mode;
2066         inode->i_ino = objectid;
2067         inode->i_blocks = 0;
2068         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2069         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2070                                   struct btrfs_inode_item);
2071         fill_inode_item(path->nodes[0], inode_item, inode);
2072
2073         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2074                              struct btrfs_inode_ref);
2075         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2076         btrfs_set_inode_ref_index(path->nodes[0], ref, BTRFS_I(inode)->index);
2077         ptr = (unsigned long)(ref + 1);
2078         write_extent_buffer(path->nodes[0], name, ptr, name_len);
2079
2080         btrfs_mark_buffer_dirty(path->nodes[0]);
2081         btrfs_free_path(path);
2082
2083         location = &BTRFS_I(inode)->location;
2084         location->objectid = objectid;
2085         location->offset = 0;
2086         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2087
2088         insert_inode_hash(inode);
2089         return inode;
2090 fail:
2091         if (dir)
2092                 BTRFS_I(dir)->index_cnt--;
2093         btrfs_free_path(path);
2094         return ERR_PTR(ret);
2095 }
2096
2097 static inline u8 btrfs_inode_type(struct inode *inode)
2098 {
2099         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2100 }
2101
2102 static int btrfs_add_link(struct btrfs_trans_handle *trans,
2103                             struct dentry *dentry, struct inode *inode,
2104                             int add_backref)
2105 {
2106         int ret;
2107         struct btrfs_key key;
2108         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
2109         struct inode *parent_inode = dentry->d_parent->d_inode;
2110
2111         key.objectid = inode->i_ino;
2112         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2113         key.offset = 0;
2114
2115         ret = btrfs_insert_dir_item(trans, root,
2116                                     dentry->d_name.name, dentry->d_name.len,
2117                                     dentry->d_parent->d_inode->i_ino,
2118                                     &key, btrfs_inode_type(inode),
2119                                     BTRFS_I(inode)->index);
2120         if (ret == 0) {
2121                 if (add_backref) {
2122                         ret = btrfs_insert_inode_ref(trans, root,
2123                                              dentry->d_name.name,
2124                                              dentry->d_name.len,
2125                                              inode->i_ino,
2126                                              parent_inode->i_ino,
2127                                              BTRFS_I(inode)->index);
2128                 }
2129                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2130                                    dentry->d_name.len * 2);
2131                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2132                 ret = btrfs_update_inode(trans, root,
2133                                          dentry->d_parent->d_inode);
2134         }
2135         return ret;
2136 }
2137
2138 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2139                             struct dentry *dentry, struct inode *inode,
2140                             int backref)
2141 {
2142         int err = btrfs_add_link(trans, dentry, inode, backref);
2143         if (!err) {
2144                 d_instantiate(dentry, inode);
2145                 return 0;
2146         }
2147         if (err > 0)
2148                 err = -EEXIST;
2149         return err;
2150 }
2151
2152 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2153                         int mode, dev_t rdev)
2154 {
2155         struct btrfs_trans_handle *trans;
2156         struct btrfs_root *root = BTRFS_I(dir)->root;
2157         struct inode *inode = NULL;
2158         int err;
2159         int drop_inode = 0;
2160         u64 objectid;
2161         unsigned long nr = 0;
2162
2163         if (!new_valid_dev(rdev))
2164                 return -EINVAL;
2165
2166         err = btrfs_check_free_space(root, 1, 0);
2167         if (err)
2168                 goto fail;
2169
2170         trans = btrfs_start_transaction(root, 1);
2171         btrfs_set_trans_block_group(trans, dir);
2172
2173         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2174         if (err) {
2175                 err = -ENOSPC;
2176                 goto out_unlock;
2177         }
2178
2179         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2180                                 dentry->d_name.len,
2181                                 dentry->d_parent->d_inode->i_ino, objectid,
2182                                 BTRFS_I(dir)->block_group, mode);
2183         err = PTR_ERR(inode);
2184         if (IS_ERR(inode))
2185                 goto out_unlock;
2186
2187         btrfs_set_trans_block_group(trans, inode);
2188         err = btrfs_add_nondir(trans, dentry, inode, 0);
2189         if (err)
2190                 drop_inode = 1;
2191         else {
2192                 inode->i_op = &btrfs_special_inode_operations;
2193                 init_special_inode(inode, inode->i_mode, rdev);
2194                 btrfs_update_inode(trans, root, inode);
2195         }
2196         dir->i_sb->s_dirt = 1;
2197         btrfs_update_inode_block_group(trans, inode);
2198         btrfs_update_inode_block_group(trans, dir);
2199 out_unlock:
2200         nr = trans->blocks_used;
2201         btrfs_end_transaction_throttle(trans, root);
2202 fail:
2203         if (drop_inode) {
2204                 inode_dec_link_count(inode);
2205                 iput(inode);
2206         }
2207         btrfs_btree_balance_dirty(root, nr);
2208         return err;
2209 }
2210
2211 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2212                         int mode, struct nameidata *nd)
2213 {
2214         struct btrfs_trans_handle *trans;
2215         struct btrfs_root *root = BTRFS_I(dir)->root;
2216         struct inode *inode = NULL;
2217         int err;
2218         int drop_inode = 0;
2219         unsigned long nr = 0;
2220         u64 objectid;
2221
2222         err = btrfs_check_free_space(root, 1, 0);
2223         if (err)
2224                 goto fail;
2225         trans = btrfs_start_transaction(root, 1);
2226         btrfs_set_trans_block_group(trans, dir);
2227
2228         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2229         if (err) {
2230                 err = -ENOSPC;
2231                 goto out_unlock;
2232         }
2233
2234         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2235                                 dentry->d_name.len,
2236                                 dentry->d_parent->d_inode->i_ino,
2237                                 objectid, BTRFS_I(dir)->block_group, mode);
2238         err = PTR_ERR(inode);
2239         if (IS_ERR(inode))
2240                 goto out_unlock;
2241
2242         btrfs_set_trans_block_group(trans, inode);
2243         err = btrfs_add_nondir(trans, dentry, inode, 0);
2244         if (err)
2245                 drop_inode = 1;
2246         else {
2247                 inode->i_mapping->a_ops = &btrfs_aops;
2248                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2249                 inode->i_fop = &btrfs_file_operations;
2250                 inode->i_op = &btrfs_file_inode_operations;
2251                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2252                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2253                                      inode->i_mapping, GFP_NOFS);
2254                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2255                                      inode->i_mapping, GFP_NOFS);
2256                 mutex_init(&BTRFS_I(inode)->csum_mutex);
2257                 mutex_init(&BTRFS_I(inode)->extent_mutex);
2258                 BTRFS_I(inode)->delalloc_bytes = 0;
2259                 BTRFS_I(inode)->disk_i_size = 0;
2260                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2261                 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2262         }
2263         dir->i_sb->s_dirt = 1;
2264         btrfs_update_inode_block_group(trans, inode);
2265         btrfs_update_inode_block_group(trans, dir);
2266 out_unlock:
2267         nr = trans->blocks_used;
2268         btrfs_end_transaction_throttle(trans, root);
2269 fail:
2270         if (drop_inode) {
2271                 inode_dec_link_count(inode);
2272                 iput(inode);
2273         }
2274         btrfs_btree_balance_dirty(root, nr);
2275         return err;
2276 }
2277
2278 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2279                       struct dentry *dentry)
2280 {
2281         struct btrfs_trans_handle *trans;
2282         struct btrfs_root *root = BTRFS_I(dir)->root;
2283         struct inode *inode = old_dentry->d_inode;
2284         unsigned long nr = 0;
2285         int err;
2286         int drop_inode = 0;
2287
2288         if (inode->i_nlink == 0)
2289                 return -ENOENT;
2290
2291 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2292         inode->i_nlink++;
2293 #else
2294         inc_nlink(inode);
2295 #endif
2296         err = btrfs_check_free_space(root, 1, 0);
2297         if (err)
2298                 goto fail;
2299         err = btrfs_set_inode_index(dir, inode);
2300         if (err)
2301                 goto fail;
2302
2303         trans = btrfs_start_transaction(root, 1);
2304
2305         btrfs_set_trans_block_group(trans, dir);
2306         atomic_inc(&inode->i_count);
2307
2308         err = btrfs_add_nondir(trans, dentry, inode, 1);
2309
2310         if (err)
2311                 drop_inode = 1;
2312
2313         dir->i_sb->s_dirt = 1;
2314         btrfs_update_inode_block_group(trans, dir);
2315         err = btrfs_update_inode(trans, root, inode);
2316
2317         if (err)
2318                 drop_inode = 1;
2319
2320         nr = trans->blocks_used;
2321         btrfs_end_transaction_throttle(trans, root);
2322 fail:
2323         if (drop_inode) {
2324                 inode_dec_link_count(inode);
2325                 iput(inode);
2326         }
2327         btrfs_btree_balance_dirty(root, nr);
2328         return err;
2329 }
2330
2331 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2332 {
2333         struct inode *inode = NULL;
2334         struct btrfs_trans_handle *trans;
2335         struct btrfs_root *root = BTRFS_I(dir)->root;
2336         int err = 0;
2337         int drop_on_err = 0;
2338         u64 objectid = 0;
2339         unsigned long nr = 1;
2340
2341         err = btrfs_check_free_space(root, 1, 0);
2342         if (err)
2343                 goto out_unlock;
2344
2345         trans = btrfs_start_transaction(root, 1);
2346         btrfs_set_trans_block_group(trans, dir);
2347
2348         if (IS_ERR(trans)) {
2349                 err = PTR_ERR(trans);
2350                 goto out_unlock;
2351         }
2352
2353         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2354         if (err) {
2355                 err = -ENOSPC;
2356                 goto out_unlock;
2357         }
2358
2359         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2360                                 dentry->d_name.len,
2361                                 dentry->d_parent->d_inode->i_ino, objectid,
2362                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
2363         if (IS_ERR(inode)) {
2364                 err = PTR_ERR(inode);
2365                 goto out_fail;
2366         }
2367
2368         drop_on_err = 1;
2369         inode->i_op = &btrfs_dir_inode_operations;
2370         inode->i_fop = &btrfs_dir_file_operations;
2371         btrfs_set_trans_block_group(trans, inode);
2372
2373         btrfs_i_size_write(inode, 0);
2374         err = btrfs_update_inode(trans, root, inode);
2375         if (err)
2376                 goto out_fail;
2377
2378         err = btrfs_add_link(trans, dentry, inode, 0);
2379         if (err)
2380                 goto out_fail;
2381
2382         d_instantiate(dentry, inode);
2383         drop_on_err = 0;
2384         dir->i_sb->s_dirt = 1;
2385         btrfs_update_inode_block_group(trans, inode);
2386         btrfs_update_inode_block_group(trans, dir);
2387
2388 out_fail:
2389         nr = trans->blocks_used;
2390         btrfs_end_transaction_throttle(trans, root);
2391
2392 out_unlock:
2393         if (drop_on_err)
2394                 iput(inode);
2395         btrfs_btree_balance_dirty(root, nr);
2396         return err;
2397 }
2398
2399 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2400                                 struct extent_map *existing,
2401                                 struct extent_map *em,
2402                                 u64 map_start, u64 map_len)
2403 {
2404         u64 start_diff;
2405
2406         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2407         start_diff = map_start - em->start;
2408         em->start = map_start;
2409         em->len = map_len;
2410         if (em->block_start < EXTENT_MAP_LAST_BYTE)
2411                 em->block_start += start_diff;
2412         return add_extent_mapping(em_tree, em);
2413 }
2414
2415 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2416                                     size_t pg_offset, u64 start, u64 len,
2417                                     int create)
2418 {
2419         int ret;
2420         int err = 0;
2421         u64 bytenr;
2422         u64 extent_start = 0;
2423         u64 extent_end = 0;
2424         u64 objectid = inode->i_ino;
2425         u32 found_type;
2426         struct btrfs_path *path = NULL;
2427         struct btrfs_root *root = BTRFS_I(inode)->root;
2428         struct btrfs_file_extent_item *item;
2429         struct extent_buffer *leaf;
2430         struct btrfs_key found_key;
2431         struct extent_map *em = NULL;
2432         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2433         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2434         struct btrfs_trans_handle *trans = NULL;
2435
2436 again:
2437         spin_lock(&em_tree->lock);
2438         em = lookup_extent_mapping(em_tree, start, len);
2439         if (em)
2440                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2441         spin_unlock(&em_tree->lock);
2442
2443         if (em) {
2444                 if (em->start > start || em->start + em->len <= start)
2445                         free_extent_map(em);
2446                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2447                         free_extent_map(em);
2448                 else
2449                         goto out;
2450         }
2451         em = alloc_extent_map(GFP_NOFS);
2452         if (!em) {
2453                 err = -ENOMEM;
2454                 goto out;
2455         }
2456         em->bdev = root->fs_info->fs_devices->latest_bdev;
2457         em->start = EXTENT_MAP_HOLE;
2458         em->len = (u64)-1;
2459
2460         if (!path) {
2461                 path = btrfs_alloc_path();
2462                 BUG_ON(!path);
2463         }
2464
2465         ret = btrfs_lookup_file_extent(trans, root, path,
2466                                        objectid, start, trans != NULL);
2467         if (ret < 0) {
2468                 err = ret;
2469                 goto out;
2470         }
2471
2472         if (ret != 0) {
2473                 if (path->slots[0] == 0)
2474                         goto not_found;
2475                 path->slots[0]--;
2476         }
2477
2478         leaf = path->nodes[0];
2479         item = btrfs_item_ptr(leaf, path->slots[0],
2480                               struct btrfs_file_extent_item);
2481         /* are we inside the extent that was found? */
2482         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2483         found_type = btrfs_key_type(&found_key);
2484         if (found_key.objectid != objectid ||
2485             found_type != BTRFS_EXTENT_DATA_KEY) {
2486                 goto not_found;
2487         }
2488
2489         found_type = btrfs_file_extent_type(leaf, item);
2490         extent_start = found_key.offset;
2491         if (found_type == BTRFS_FILE_EXTENT_REG) {
2492                 extent_end = extent_start +
2493                        btrfs_file_extent_num_bytes(leaf, item);
2494                 err = 0;
2495                 if (start < extent_start || start >= extent_end) {
2496                         em->start = start;
2497                         if (start < extent_start) {
2498                                 if (start + len <= extent_start)
2499                                         goto not_found;
2500                                 em->len = extent_end - extent_start;
2501                         } else {
2502                                 em->len = len;
2503                         }
2504                         goto not_found_em;
2505                 }
2506                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2507                 if (bytenr == 0) {
2508                         em->start = extent_start;
2509                         em->len = extent_end - extent_start;
2510                         em->block_start = EXTENT_MAP_HOLE;
2511                         goto insert;
2512                 }
2513                 bytenr += btrfs_file_extent_offset(leaf, item);
2514                 em->block_start = bytenr;
2515                 em->start = extent_start;
2516                 em->len = extent_end - extent_start;
2517                 goto insert;
2518         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2519                 u64 page_start;
2520                 unsigned long ptr;
2521                 char *map;
2522                 size_t size;
2523                 size_t extent_offset;
2524                 size_t copy_size;
2525
2526                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2527                                                     path->slots[0]));
2528                 extent_end = (extent_start + size + root->sectorsize - 1) &
2529                         ~((u64)root->sectorsize - 1);
2530                 if (start < extent_start || start >= extent_end) {
2531                         em->start = start;
2532                         if (start < extent_start) {
2533                                 if (start + len <= extent_start)
2534                                         goto not_found;
2535                                 em->len = extent_end - extent_start;
2536                         } else {
2537                                 em->len = len;
2538                         }
2539                         goto not_found_em;
2540                 }
2541                 em->block_start = EXTENT_MAP_INLINE;
2542
2543                 if (!page) {
2544                         em->start = extent_start;
2545                         em->len = size;
2546                         goto out;
2547                 }
2548
2549                 page_start = page_offset(page) + pg_offset;
2550                 extent_offset = page_start - extent_start;
2551                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2552                                 size - extent_offset);
2553                 em->start = extent_start + extent_offset;
2554                 em->len = (copy_size + root->sectorsize - 1) &
2555                         ~((u64)root->sectorsize - 1);
2556                 map = kmap(page);
2557                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2558                 if (create == 0 && !PageUptodate(page)) {
2559                         read_extent_buffer(leaf, map + pg_offset, ptr,
2560                                            copy_size);
2561                         flush_dcache_page(page);
2562                 } else if (create && PageUptodate(page)) {
2563                         if (!trans) {
2564                                 kunmap(page);
2565                                 free_extent_map(em);
2566                                 em = NULL;
2567                                 btrfs_release_path(root, path);
2568                                 trans = btrfs_join_transaction(root, 1);
2569                                 goto again;
2570                         }
2571                         write_extent_buffer(leaf, map + pg_offset, ptr,
2572                                             copy_size);
2573                         btrfs_mark_buffer_dirty(leaf);
2574                 }
2575                 kunmap(page);
2576                 set_extent_uptodate(io_tree, em->start,
2577                                     extent_map_end(em) - 1, GFP_NOFS);
2578                 goto insert;
2579         } else {
2580                 printk("unkknown found_type %d\n", found_type);
2581                 WARN_ON(1);
2582         }
2583 not_found:
2584         em->start = start;
2585         em->len = len;
2586 not_found_em:
2587         em->block_start = EXTENT_MAP_HOLE;
2588 insert:
2589         btrfs_release_path(root, path);
2590         if (em->start > start || extent_map_end(em) <= start) {
2591                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2592                 err = -EIO;
2593                 goto out;
2594         }
2595
2596         err = 0;
2597         spin_lock(&em_tree->lock);
2598         ret = add_extent_mapping(em_tree, em);
2599         /* it is possible that someone inserted the extent into the tree
2600          * while we had the lock dropped.  It is also possible that
2601          * an overlapping map exists in the tree
2602          */
2603         if (ret == -EEXIST) {
2604                 struct extent_map *existing;
2605
2606                 ret = 0;
2607
2608                 existing = lookup_extent_mapping(em_tree, start, len);
2609                 if (existing && (existing->start > start ||
2610                     existing->start + existing->len <= start)) {
2611                         free_extent_map(existing);
2612                         existing = NULL;
2613                 }
2614                 if (!existing) {
2615                         existing = lookup_extent_mapping(em_tree, em->start,
2616                                                          em->len);
2617                         if (existing) {
2618                                 err = merge_extent_mapping(em_tree, existing,
2619                                                            em, start,
2620                                                            root->sectorsize);
2621                                 free_extent_map(existing);
2622                                 if (err) {
2623                                         free_extent_map(em);
2624                                         em = NULL;
2625                                 }
2626                         } else {
2627                                 err = -EIO;
2628                                 printk("failing to insert %Lu %Lu\n",
2629                                        start, len);
2630                                 free_extent_map(em);
2631                                 em = NULL;
2632                         }
2633                 } else {
2634                         free_extent_map(em);
2635                         em = existing;
2636                         err = 0;
2637                 }
2638         }
2639         spin_unlock(&em_tree->lock);
2640 out:
2641         if (path)
2642                 btrfs_free_path(path);
2643         if (trans) {
2644                 ret = btrfs_end_transaction(trans, root);
2645                 if (!err) {
2646                         err = ret;
2647                 }
2648         }
2649         if (err) {
2650                 free_extent_map(em);
2651                 WARN_ON(1);
2652                 return ERR_PTR(err);
2653         }
2654         return em;
2655 }
2656
2657 #if 0 /* waiting for O_DIRECT reads */
2658 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2659                         struct buffer_head *bh_result, int create)
2660 {
2661         struct extent_map *em;
2662         u64 start = (u64)iblock << inode->i_blkbits;
2663         struct btrfs_multi_bio *multi = NULL;
2664         struct btrfs_root *root = BTRFS_I(inode)->root;
2665         u64 len;
2666         u64 logical;
2667         u64 map_length;
2668         int ret = 0;
2669
2670         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2671
2672         if (!em || IS_ERR(em))
2673                 goto out;
2674
2675         if (em->start > start || em->start + em->len <= start) {
2676             goto out;
2677         }
2678
2679         if (em->block_start == EXTENT_MAP_INLINE) {
2680                 ret = -EINVAL;
2681                 goto out;
2682         }
2683
2684         len = em->start + em->len - start;
2685         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2686
2687         if (em->block_start == EXTENT_MAP_HOLE ||
2688             em->block_start == EXTENT_MAP_DELALLOC) {
2689                 bh_result->b_size = len;
2690                 goto out;
2691         }
2692
2693         logical = start - em->start;
2694         logical = em->block_start + logical;
2695
2696         map_length = len;
2697         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2698                               logical, &map_length, &multi, 0);
2699         BUG_ON(ret);
2700         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2701         bh_result->b_size = min(map_length, len);
2702
2703         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2704         set_buffer_mapped(bh_result);
2705         kfree(multi);
2706 out:
2707         free_extent_map(em);
2708         return ret;
2709 }
2710 #endif
2711
2712 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2713                         const struct iovec *iov, loff_t offset,
2714                         unsigned long nr_segs)
2715 {
2716         return -EINVAL;
2717 #if 0
2718         struct file *file = iocb->ki_filp;
2719         struct inode *inode = file->f_mapping->host;
2720
2721         if (rw == WRITE)
2722                 return -EINVAL;
2723
2724         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2725                                   offset, nr_segs, btrfs_get_block, NULL);
2726 #endif
2727 }
2728
2729 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2730 {
2731         return extent_bmap(mapping, iblock, btrfs_get_extent);
2732 }
2733
2734 int btrfs_readpage(struct file *file, struct page *page)
2735 {
2736         struct extent_io_tree *tree;
2737         tree = &BTRFS_I(page->mapping->host)->io_tree;
2738         return extent_read_full_page(tree, page, btrfs_get_extent);
2739 }
2740
2741 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2742 {
2743         struct extent_io_tree *tree;
2744
2745
2746         if (current->flags & PF_MEMALLOC) {
2747                 redirty_page_for_writepage(wbc, page);
2748                 unlock_page(page);
2749                 return 0;
2750         }
2751         tree = &BTRFS_I(page->mapping->host)->io_tree;
2752         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2753 }
2754
2755 int btrfs_writepages(struct address_space *mapping,
2756                      struct writeback_control *wbc)
2757 {
2758         struct extent_io_tree *tree;
2759         tree = &BTRFS_I(mapping->host)->io_tree;
2760         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2761 }
2762
2763 static int
2764 btrfs_readpages(struct file *file, struct address_space *mapping,
2765                 struct list_head *pages, unsigned nr_pages)
2766 {
2767         struct extent_io_tree *tree;
2768         tree = &BTRFS_I(mapping->host)->io_tree;
2769         return extent_readpages(tree, mapping, pages, nr_pages,
2770                                 btrfs_get_extent);
2771 }
2772 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2773 {
2774         struct extent_io_tree *tree;
2775         struct extent_map_tree *map;
2776         int ret;
2777
2778         tree = &BTRFS_I(page->mapping->host)->io_tree;
2779         map = &BTRFS_I(page->mapping->host)->extent_tree;
2780         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2781         if (ret == 1) {
2782                 ClearPagePrivate(page);
2783                 set_page_private(page, 0);
2784                 page_cache_release(page);
2785         }
2786         return ret;
2787 }
2788
2789 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2790 {
2791         return __btrfs_releasepage(page, gfp_flags);
2792 }
2793
2794 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2795 {
2796         struct extent_io_tree *tree;
2797         struct btrfs_ordered_extent *ordered;
2798         u64 page_start = page_offset(page);
2799         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2800
2801         wait_on_page_writeback(page);
2802         tree = &BTRFS_I(page->mapping->host)->io_tree;
2803         if (offset) {
2804                 btrfs_releasepage(page, GFP_NOFS);
2805                 return;
2806         }
2807
2808         lock_extent(tree, page_start, page_end, GFP_NOFS);
2809         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
2810                                            page_offset(page));
2811         if (ordered) {
2812                 /*
2813                  * IO on this page will never be started, so we need
2814                  * to account for any ordered extents now
2815                  */
2816                 clear_extent_bit(tree, page_start, page_end,
2817                                  EXTENT_DIRTY | EXTENT_DELALLOC |
2818                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
2819                 btrfs_finish_ordered_io(page->mapping->host,
2820                                         page_start, page_end);
2821                 btrfs_put_ordered_extent(ordered);
2822                 lock_extent(tree, page_start, page_end, GFP_NOFS);
2823         }
2824         clear_extent_bit(tree, page_start, page_end,
2825                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2826                  EXTENT_ORDERED,
2827                  1, 1, GFP_NOFS);
2828         __btrfs_releasepage(page, GFP_NOFS);
2829
2830         ClearPageChecked(page);
2831         if (PagePrivate(page)) {
2832                 ClearPagePrivate(page);
2833                 set_page_private(page, 0);
2834                 page_cache_release(page);
2835         }
2836 }
2837
2838 /*
2839  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2840  * called from a page fault handler when a page is first dirtied. Hence we must
2841  * be careful to check for EOF conditions here. We set the page up correctly
2842  * for a written page which means we get ENOSPC checking when writing into
2843  * holes and correct delalloc and unwritten extent mapping on filesystems that
2844  * support these features.
2845  *
2846  * We are not allowed to take the i_mutex here so we have to play games to
2847  * protect against truncate races as the page could now be beyond EOF.  Because
2848  * vmtruncate() writes the inode size before removing pages, once we have the
2849  * page lock we can determine safely if the page is beyond EOF. If it is not
2850  * beyond EOF, then the page is guaranteed safe against truncation until we
2851  * unlock the page.
2852  */
2853 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2854 {
2855         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2856         struct btrfs_root *root = BTRFS_I(inode)->root;
2857         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2858         struct btrfs_ordered_extent *ordered;
2859         char *kaddr;
2860         unsigned long zero_start;
2861         loff_t size;
2862         int ret;
2863         u64 page_start;
2864         u64 page_end;
2865
2866         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2867         if (ret)
2868                 goto out;
2869
2870         ret = -EINVAL;
2871 again:
2872         lock_page(page);
2873         size = i_size_read(inode);
2874         page_start = page_offset(page);
2875         page_end = page_start + PAGE_CACHE_SIZE - 1;
2876
2877         if ((page->mapping != inode->i_mapping) ||
2878             (page_start >= size)) {
2879                 /* page got truncated out from underneath us */
2880                 goto out_unlock;
2881         }
2882         wait_on_page_writeback(page);
2883
2884         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2885         set_page_extent_mapped(page);
2886
2887         /*
2888          * we can't set the delalloc bits if there are pending ordered
2889          * extents.  Drop our locks and wait for them to finish
2890          */
2891         ordered = btrfs_lookup_ordered_extent(inode, page_start);
2892         if (ordered) {
2893                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2894                 unlock_page(page);
2895                 btrfs_start_ordered_extent(inode, ordered, 1);
2896                 btrfs_put_ordered_extent(ordered);
2897                 goto again;
2898         }
2899
2900         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
2901                             page_end, GFP_NOFS);
2902         ret = 0;
2903
2904         /* page is wholly or partially inside EOF */
2905         if (page_start + PAGE_CACHE_SIZE > size)
2906                 zero_start = size & ~PAGE_CACHE_MASK;
2907         else
2908                 zero_start = PAGE_CACHE_SIZE;
2909
2910         if (zero_start != PAGE_CACHE_SIZE) {
2911                 kaddr = kmap(page);
2912                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
2913                 flush_dcache_page(page);
2914                 kunmap(page);
2915         }
2916         ClearPageChecked(page);
2917         set_page_dirty(page);
2918         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2919
2920 out_unlock:
2921         unlock_page(page);
2922 out:
2923         return ret;
2924 }
2925
2926 static void btrfs_truncate(struct inode *inode)
2927 {
2928         struct btrfs_root *root = BTRFS_I(inode)->root;
2929         int ret;
2930         struct btrfs_trans_handle *trans;
2931         unsigned long nr;
2932         u64 mask = root->sectorsize - 1;
2933
2934         if (!S_ISREG(inode->i_mode))
2935                 return;
2936         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2937                 return;
2938
2939         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2940         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
2941
2942         trans = btrfs_start_transaction(root, 1);
2943         btrfs_set_trans_block_group(trans, inode);
2944         btrfs_i_size_write(inode, inode->i_size);
2945
2946         /* FIXME, add redo link to tree so we don't leak on crash */
2947         ret = btrfs_truncate_in_trans(trans, root, inode,
2948                                       BTRFS_EXTENT_DATA_KEY);
2949         btrfs_update_inode(trans, root, inode);
2950         nr = trans->blocks_used;
2951
2952         ret = btrfs_end_transaction_throttle(trans, root);
2953         BUG_ON(ret);
2954         btrfs_btree_balance_dirty(root, nr);
2955 }
2956
2957 /*
2958  * Invalidate a single dcache entry at the root of the filesystem.
2959  * Needed after creation of snapshot or subvolume.
2960  */
2961 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
2962                                   int namelen)
2963 {
2964         struct dentry *alias, *entry;
2965         struct qstr qstr;
2966
2967         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
2968         if (alias) {
2969                 qstr.name = name;
2970                 qstr.len = namelen;
2971                 /* change me if btrfs ever gets a d_hash operation */
2972                 qstr.hash = full_name_hash(qstr.name, qstr.len);
2973                 entry = d_lookup(alias, &qstr);
2974                 dput(alias);
2975                 if (entry) {
2976                         d_invalidate(entry);
2977                         dput(entry);
2978                 }
2979         }
2980 }
2981
2982 int btrfs_create_subvol_root(struct btrfs_root *new_root,
2983                 struct btrfs_trans_handle *trans, u64 new_dirid,
2984                 struct btrfs_block_group_cache *block_group)
2985 {
2986         struct inode *inode;
2987
2988         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
2989                                 new_dirid, block_group, S_IFDIR | 0700);
2990         if (IS_ERR(inode))
2991                 return PTR_ERR(inode);
2992         inode->i_op = &btrfs_dir_inode_operations;
2993         inode->i_fop = &btrfs_dir_file_operations;
2994         new_root->inode = inode;
2995
2996         inode->i_nlink = 1;
2997         btrfs_i_size_write(inode, 0);
2998
2999         return btrfs_update_inode(trans, new_root, inode);
3000 }
3001
3002 unsigned long btrfs_force_ra(struct address_space *mapping,
3003                               struct file_ra_state *ra, struct file *file,
3004                               pgoff_t offset, pgoff_t last_index)
3005 {
3006         pgoff_t req_size = last_index - offset + 1;
3007
3008 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3009         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3010         return offset;
3011 #else
3012         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3013         return offset + req_size;
3014 #endif
3015 }
3016
3017 struct inode *btrfs_alloc_inode(struct super_block *sb)
3018 {
3019         struct btrfs_inode *ei;
3020
3021         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3022         if (!ei)
3023                 return NULL;
3024         ei->last_trans = 0;
3025         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3026         return &ei->vfs_inode;
3027 }
3028
3029 void btrfs_destroy_inode(struct inode *inode)
3030 {
3031         struct btrfs_ordered_extent *ordered;
3032         WARN_ON(!list_empty(&inode->i_dentry));
3033         WARN_ON(inode->i_data.nrpages);
3034
3035         while(1) {
3036                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3037                 if (!ordered)
3038                         break;
3039                 else {
3040                         printk("found ordered extent %Lu %Lu\n",
3041                                ordered->file_offset, ordered->len);
3042                         btrfs_remove_ordered_extent(inode, ordered);
3043                         btrfs_put_ordered_extent(ordered);
3044                         btrfs_put_ordered_extent(ordered);
3045                 }
3046         }
3047         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3048         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3049 }
3050
3051 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3052 static void init_once(struct kmem_cache * cachep, void *foo)
3053 #else
3054 static void init_once(void * foo, struct kmem_cache * cachep,
3055                       unsigned long flags)
3056 #endif
3057 {
3058         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3059
3060         inode_init_once(&ei->vfs_inode);
3061 }
3062
3063 void btrfs_destroy_cachep(void)
3064 {
3065         if (btrfs_inode_cachep)
3066                 kmem_cache_destroy(btrfs_inode_cachep);
3067         if (btrfs_trans_handle_cachep)
3068                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3069         if (btrfs_transaction_cachep)
3070                 kmem_cache_destroy(btrfs_transaction_cachep);
3071         if (btrfs_bit_radix_cachep)
3072                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3073         if (btrfs_path_cachep)
3074                 kmem_cache_destroy(btrfs_path_cachep);
3075 }
3076
3077 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3078                                        unsigned long extra_flags,
3079 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3080                                        void (*ctor)(struct kmem_cache *, void *)
3081 #else
3082                                        void (*ctor)(void *, struct kmem_cache *,
3083                                                     unsigned long)
3084 #endif
3085                                      )
3086 {
3087         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3088                                  SLAB_MEM_SPREAD | extra_flags), ctor
3089 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3090                                  ,NULL
3091 #endif
3092                                 );
3093 }
3094
3095 int btrfs_init_cachep(void)
3096 {
3097         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3098                                           sizeof(struct btrfs_inode),
3099                                           0, init_once);
3100         if (!btrfs_inode_cachep)
3101                 goto fail;
3102         btrfs_trans_handle_cachep =
3103                         btrfs_cache_create("btrfs_trans_handle_cache",
3104                                            sizeof(struct btrfs_trans_handle),
3105                                            0, NULL);
3106         if (!btrfs_trans_handle_cachep)
3107                 goto fail;
3108         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3109                                              sizeof(struct btrfs_transaction),
3110                                              0, NULL);
3111         if (!btrfs_transaction_cachep)
3112                 goto fail;
3113         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3114                                          sizeof(struct btrfs_path),
3115                                          0, NULL);
3116         if (!btrfs_path_cachep)
3117                 goto fail;
3118         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3119                                               SLAB_DESTROY_BY_RCU, NULL);
3120         if (!btrfs_bit_radix_cachep)
3121                 goto fail;
3122         return 0;
3123 fail:
3124         btrfs_destroy_cachep();
3125         return -ENOMEM;
3126 }
3127
3128 static int btrfs_getattr(struct vfsmount *mnt,
3129                          struct dentry *dentry, struct kstat *stat)
3130 {
3131         struct inode *inode = dentry->d_inode;
3132         generic_fillattr(inode, stat);
3133         stat->blksize = PAGE_CACHE_SIZE;
3134         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3135         return 0;
3136 }
3137
3138 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3139                            struct inode * new_dir,struct dentry *new_dentry)
3140 {
3141         struct btrfs_trans_handle *trans;
3142         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3143         struct inode *new_inode = new_dentry->d_inode;
3144         struct inode *old_inode = old_dentry->d_inode;
3145         struct timespec ctime = CURRENT_TIME;
3146         int ret;
3147
3148         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3149             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3150                 return -ENOTEMPTY;
3151         }
3152
3153         ret = btrfs_check_free_space(root, 1, 0);
3154         if (ret)
3155                 goto out_unlock;
3156
3157         trans = btrfs_start_transaction(root, 1);
3158
3159         btrfs_set_trans_block_group(trans, new_dir);
3160
3161         old_dentry->d_inode->i_nlink++;
3162         old_dir->i_ctime = old_dir->i_mtime = ctime;
3163         new_dir->i_ctime = new_dir->i_mtime = ctime;
3164         old_inode->i_ctime = ctime;
3165
3166         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3167         if (ret)
3168                 goto out_fail;
3169
3170         if (new_inode) {
3171                 new_inode->i_ctime = CURRENT_TIME;
3172                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3173                 if (ret)
3174                         goto out_fail;
3175         }
3176         ret = btrfs_set_inode_index(new_dir, old_inode);
3177         if (ret)
3178                 goto out_fail;
3179
3180         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
3181         if (ret)
3182                 goto out_fail;
3183
3184 out_fail:
3185         btrfs_end_transaction(trans, root);
3186 out_unlock:
3187         return ret;
3188 }
3189
3190 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3191                          const char *symname)
3192 {
3193         struct btrfs_trans_handle *trans;
3194         struct btrfs_root *root = BTRFS_I(dir)->root;
3195         struct btrfs_path *path;
3196         struct btrfs_key key;
3197         struct inode *inode = NULL;
3198         int err;
3199         int drop_inode = 0;
3200         u64 objectid;
3201         int name_len;
3202         int datasize;
3203         unsigned long ptr;
3204         struct btrfs_file_extent_item *ei;
3205         struct extent_buffer *leaf;
3206         unsigned long nr = 0;
3207
3208         name_len = strlen(symname) + 1;
3209         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3210                 return -ENAMETOOLONG;
3211
3212         err = btrfs_check_free_space(root, 1, 0);
3213         if (err)
3214                 goto out_fail;
3215
3216         trans = btrfs_start_transaction(root, 1);
3217         btrfs_set_trans_block_group(trans, dir);
3218
3219         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3220         if (err) {
3221                 err = -ENOSPC;
3222                 goto out_unlock;
3223         }
3224
3225         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3226                                 dentry->d_name.len,
3227                                 dentry->d_parent->d_inode->i_ino, objectid,
3228                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
3229         err = PTR_ERR(inode);
3230         if (IS_ERR(inode))
3231                 goto out_unlock;
3232
3233         btrfs_set_trans_block_group(trans, inode);
3234         err = btrfs_add_nondir(trans, dentry, inode, 0);
3235         if (err)
3236                 drop_inode = 1;
3237         else {
3238                 inode->i_mapping->a_ops = &btrfs_aops;
3239                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3240                 inode->i_fop = &btrfs_file_operations;
3241                 inode->i_op = &btrfs_file_inode_operations;
3242                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3243                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3244                                      inode->i_mapping, GFP_NOFS);
3245                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3246                                      inode->i_mapping, GFP_NOFS);
3247                 mutex_init(&BTRFS_I(inode)->csum_mutex);
3248                 mutex_init(&BTRFS_I(inode)->extent_mutex);
3249                 BTRFS_I(inode)->delalloc_bytes = 0;
3250                 BTRFS_I(inode)->disk_i_size = 0;
3251                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3252                 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3253         }
3254         dir->i_sb->s_dirt = 1;
3255         btrfs_update_inode_block_group(trans, inode);
3256         btrfs_update_inode_block_group(trans, dir);
3257         if (drop_inode)
3258                 goto out_unlock;
3259
3260         path = btrfs_alloc_path();
3261         BUG_ON(!path);
3262         key.objectid = inode->i_ino;
3263         key.offset = 0;
3264         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3265         datasize = btrfs_file_extent_calc_inline_size(name_len);
3266         err = btrfs_insert_empty_item(trans, root, path, &key,
3267                                       datasize);
3268         if (err) {
3269                 drop_inode = 1;
3270                 goto out_unlock;
3271         }
3272         leaf = path->nodes[0];
3273         ei = btrfs_item_ptr(leaf, path->slots[0],
3274                             struct btrfs_file_extent_item);
3275         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3276         btrfs_set_file_extent_type(leaf, ei,
3277                                    BTRFS_FILE_EXTENT_INLINE);
3278         ptr = btrfs_file_extent_inline_start(ei);
3279         write_extent_buffer(leaf, symname, ptr, name_len);
3280         btrfs_mark_buffer_dirty(leaf);
3281         btrfs_free_path(path);
3282
3283         inode->i_op = &btrfs_symlink_inode_operations;
3284         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3285         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3286         btrfs_i_size_write(inode, name_len - 1);
3287         err = btrfs_update_inode(trans, root, inode);
3288         if (err)
3289                 drop_inode = 1;
3290
3291 out_unlock:
3292         nr = trans->blocks_used;
3293         btrfs_end_transaction_throttle(trans, root);
3294 out_fail:
3295         if (drop_inode) {
3296                 inode_dec_link_count(inode);
3297                 iput(inode);
3298         }
3299         btrfs_btree_balance_dirty(root, nr);
3300         return err;
3301 }
3302
3303 static int btrfs_set_page_dirty(struct page *page)
3304 {
3305         return __set_page_dirty_nobuffers(page);
3306 }
3307
3308 static int btrfs_permission(struct inode *inode, int mask,
3309                             struct nameidata *nd)
3310 {
3311         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3312                 return -EACCES;
3313         return generic_permission(inode, mask, NULL);
3314 }
3315
3316 static struct inode_operations btrfs_dir_inode_operations = {
3317         .lookup         = btrfs_lookup,
3318         .create         = btrfs_create,
3319         .unlink         = btrfs_unlink,
3320         .link           = btrfs_link,
3321         .mkdir          = btrfs_mkdir,
3322         .rmdir          = btrfs_rmdir,
3323         .rename         = btrfs_rename,
3324         .symlink        = btrfs_symlink,
3325         .setattr        = btrfs_setattr,
3326         .mknod          = btrfs_mknod,
3327         .setxattr       = generic_setxattr,
3328         .getxattr       = generic_getxattr,
3329         .listxattr      = btrfs_listxattr,
3330         .removexattr    = generic_removexattr,
3331         .permission     = btrfs_permission,
3332 };
3333 static struct inode_operations btrfs_dir_ro_inode_operations = {
3334         .lookup         = btrfs_lookup,
3335         .permission     = btrfs_permission,
3336 };
3337 static struct file_operations btrfs_dir_file_operations = {
3338         .llseek         = generic_file_llseek,
3339         .read           = generic_read_dir,
3340         .readdir        = btrfs_readdir,
3341         .unlocked_ioctl = btrfs_ioctl,
3342 #ifdef CONFIG_COMPAT
3343         .compat_ioctl   = btrfs_ioctl,
3344 #endif
3345         .release        = btrfs_release_file,
3346 };
3347
3348 static struct extent_io_ops btrfs_extent_io_ops = {
3349         .fill_delalloc = run_delalloc_range,
3350         .submit_bio_hook = btrfs_submit_bio_hook,
3351         .merge_bio_hook = btrfs_merge_bio_hook,
3352         .readpage_io_hook = btrfs_readpage_io_hook,
3353         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3354         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3355         .writepage_start_hook = btrfs_writepage_start_hook,
3356         .readpage_io_failed_hook = btrfs_io_failed_hook,
3357         .set_bit_hook = btrfs_set_bit_hook,
3358         .clear_bit_hook = btrfs_clear_bit_hook,
3359 };
3360
3361 static struct address_space_operations btrfs_aops = {
3362         .readpage       = btrfs_readpage,
3363         .writepage      = btrfs_writepage,
3364         .writepages     = btrfs_writepages,
3365         .readpages      = btrfs_readpages,
3366         .sync_page      = block_sync_page,
3367         .bmap           = btrfs_bmap,
3368         .direct_IO      = btrfs_direct_IO,
3369         .invalidatepage = btrfs_invalidatepage,
3370         .releasepage    = btrfs_releasepage,
3371         .set_page_dirty = btrfs_set_page_dirty,
3372 };
3373
3374 static struct address_space_operations btrfs_symlink_aops = {
3375         .readpage       = btrfs_readpage,
3376         .writepage      = btrfs_writepage,
3377         .invalidatepage = btrfs_invalidatepage,
3378         .releasepage    = btrfs_releasepage,
3379 };
3380
3381 static struct inode_operations btrfs_file_inode_operations = {
3382         .truncate       = btrfs_truncate,
3383         .getattr        = btrfs_getattr,
3384         .setattr        = btrfs_setattr,
3385         .setxattr       = generic_setxattr,
3386         .getxattr       = generic_getxattr,
3387         .listxattr      = btrfs_listxattr,
3388         .removexattr    = generic_removexattr,
3389         .permission     = btrfs_permission,
3390 };
3391 static struct inode_operations btrfs_special_inode_operations = {
3392         .getattr        = btrfs_getattr,
3393         .setattr        = btrfs_setattr,
3394         .permission     = btrfs_permission,
3395 };
3396 static struct inode_operations btrfs_symlink_inode_operations = {
3397         .readlink       = generic_readlink,
3398         .follow_link    = page_follow_link_light,
3399         .put_link       = page_put_link,
3400         .permission     = btrfs_permission,
3401 };