fs/btrfs/inode.c

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