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

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