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

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