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