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

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