git.oblomov.eu Git - 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 <linux/posix_acl.h>
  40 #include "ctree.h"
  41 #include "disk-io.h"
  42 #include "transaction.h"
  43 #include "btrfs_inode.h"
  44 #include "ioctl.h"
  45 #include "print-tree.h"
  46 #include "volumes.h"
  47 #include "ordered-data.h"
  48
  49 struct btrfs_iget_args {
  50         u64 ino;
  51         struct btrfs_root *root;
  52 };
  53
  54 static struct inode_operations btrfs_dir_inode_operations;
  55 static struct inode_operations btrfs_symlink_inode_operations;
  56 static struct inode_operations btrfs_dir_ro_inode_operations;
  57 static struct inode_operations btrfs_special_inode_operations;
  58 static struct inode_operations btrfs_file_inode_operations;
  59 static struct address_space_operations btrfs_aops;
  60 static struct address_space_operations btrfs_symlink_aops;
  61 static struct file_operations btrfs_dir_file_operations;
  62 static struct extent_io_ops btrfs_extent_io_ops;
  63
  64 static struct kmem_cache *btrfs_inode_cachep;
  65 struct kmem_cache *btrfs_trans_handle_cachep;
  66 struct kmem_cache *btrfs_transaction_cachep;
  67 struct kmem_cache *btrfs_bit_radix_cachep;
  68 struct kmem_cache *btrfs_path_cachep;
  69
  70 #define S_SHIFT 12
  71 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  72         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  73         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  74         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  75         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  76         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  77         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  78         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  79 };
  80
  81 static void btrfs_truncate(struct inode *inode);
  82
  83 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  84                            int for_del)
  85 {
  86         u64 total;
  87         u64 used;
  88         u64 thresh;
  89         unsigned long flags;
  90         int ret = 0;
  91
  92         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
  93         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  94         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  95         if (for_del)
  96                 thresh = total * 90;
  97         else
  98                 thresh = total * 85;
  99
 100         do_div(thresh, 100);
 101
 102         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
 103                 ret = -ENOSPC;
 104         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 105         return ret;
 106 }
 107
 108 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 109 {
 110         struct btrfs_root *root = BTRFS_I(inode)->root;
 111         struct btrfs_trans_handle *trans;
 112         u64 alloc_hint = 0;
 113         u64 num_bytes;
 114         u64 cur_alloc_size;
 115         u64 blocksize = root->sectorsize;
 116         u64 orig_num_bytes;
 117         struct btrfs_key ins;
 118         struct extent_map *em;
 119         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 120         int ret = 0;
 121
 122         trans = btrfs_join_transaction(root, 1);
 123         BUG_ON(!trans);
 124         btrfs_set_trans_block_group(trans, inode);
 125
 126         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 127         num_bytes = max(blocksize,  num_bytes);
 128         orig_num_bytes = num_bytes;
 129
 130         if (alloc_hint == EXTENT_MAP_INLINE)
 131                 goto out;
 132
 133         BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
 134         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 135         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1);
 136         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 137
 138         while(num_bytes > 0) {
 139                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 140                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 141                                            root->sectorsize, 0, 0,
 142                                            (u64)-1, &ins, 1);
 143                 if (ret) {
 144                         WARN_ON(1);
 145                         goto out;
 146                 }
 147                 em = alloc_extent_map(GFP_NOFS);
 148                 em->start = start;
 149                 em->len = ins.offset;
 150                 em->block_start = ins.objectid;
 151                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 152                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 153                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 154                 while(1) {
 155                         spin_lock(&em_tree->lock);
 156                         ret = add_extent_mapping(em_tree, em);
 157                         spin_unlock(&em_tree->lock);
 158                         if (ret != -EEXIST) {
 159                                 free_extent_map(em);
 160                                 break;
 161                         }
 162                         btrfs_drop_extent_cache(inode, start,
 163                                                 start + ins.offset - 1);
 164                 }
 165                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 166
 167                 cur_alloc_size = ins.offset;
 168                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 169                                                ins.offset);
 170                 BUG_ON(ret);
 171                 if (num_bytes < cur_alloc_size) {
 172                         printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
 173                                cur_alloc_size);
 174                         break;
 175                 }
 176                 num_bytes -= cur_alloc_size;
 177                 alloc_hint = ins.objectid + ins.offset;
 178                 start += cur_alloc_size;
 179         }
 180 out:
 181         btrfs_end_transaction(trans, root);
 182         return ret;
 183 }
 184
 185 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 186 {
 187         u64 extent_start;
 188         u64 extent_end;
 189         u64 bytenr;
 190         u64 cow_end;
 191         u64 loops = 0;
 192         u64 total_fs_bytes;
 193         struct btrfs_root *root = BTRFS_I(inode)->root;
 194         struct btrfs_block_group_cache *block_group;
 195         struct extent_buffer *leaf;
 196         int found_type;
 197         struct btrfs_path *path;
 198         struct btrfs_file_extent_item *item;
 199         int ret;
 200         int err;
 201         struct btrfs_key found_key;
 202
 203         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 204         path = btrfs_alloc_path();
 205         BUG_ON(!path);
 206 again:
 207         ret = btrfs_lookup_file_extent(NULL, root, path,
 208                                        inode->i_ino, start, 0);
 209         if (ret < 0) {
 210                 btrfs_free_path(path);
 211                 return ret;
 212         }
 213
 214         cow_end = end;
 215         if (ret != 0) {
 216                 if (path->slots[0] == 0)
 217                         goto not_found;
 218                 path->slots[0]--;
 219         }
 220
 221         leaf = path->nodes[0];
 222         item = btrfs_item_ptr(leaf, path->slots[0],
 223                               struct btrfs_file_extent_item);
 224
 225         /* are we inside the extent that was found? */
 226         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 227         found_type = btrfs_key_type(&found_key);
 228         if (found_key.objectid != inode->i_ino ||
 229             found_type != BTRFS_EXTENT_DATA_KEY)
 230                 goto not_found;
 231
 232         found_type = btrfs_file_extent_type(leaf, item);
 233         extent_start = found_key.offset;
 234         if (found_type == BTRFS_FILE_EXTENT_REG) {
 235                 u64 extent_num_bytes;
 236
 237                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 238                 extent_end = extent_start + extent_num_bytes;
 239                 err = 0;
 240
 241                 if (loops && start != extent_start)
 242                         goto not_found;
 243
 244                 if (start < extent_start || start >= extent_end)
 245                         goto not_found;
 246
 247                 cow_end = min(end, extent_end - 1);
 248                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 249                 if (bytenr == 0)
 250                         goto not_found;
 251
 252                 if (btrfs_cross_ref_exists(root, &found_key, bytenr))
 253                         goto not_found;
 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         btrfs_release_path(root, path);
 278         cow_file_range(inode, start, end);
 279         start = end + 1;
 280         goto loop;
 281 }
 282
 283 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 284 {
 285         struct btrfs_root *root = BTRFS_I(inode)->root;
 286         int ret;
 287
 288         if (btrfs_test_opt(root, NODATACOW) ||
 289             btrfs_test_flag(inode, NODATACOW))
 290                 ret = run_delalloc_nocow(inode, start, end);
 291         else
 292                 ret = cow_file_range(inode, start, end);
 293
 294         return ret;
 295 }
 296
 297 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 298                        unsigned long old, unsigned long bits)
 299 {
 300         unsigned long flags;
 301         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 302                 struct btrfs_root *root = BTRFS_I(inode)->root;
 303                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 304                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 305                 root->fs_info->delalloc_bytes += end - start + 1;
 306                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 307         }
 308         return 0;
 309 }
 310
 311 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 312                          unsigned long old, unsigned long bits)
 313 {
 314         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 315                 struct btrfs_root *root = BTRFS_I(inode)->root;
 316                 unsigned long flags;
 317
 318                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 319                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 320                         printk("warning: delalloc account %Lu %Lu\n",
 321                                end - start + 1, root->fs_info->delalloc_bytes);
 322                         root->fs_info->delalloc_bytes = 0;
 323                         BTRFS_I(inode)->delalloc_bytes = 0;
 324                 } else {
 325                         root->fs_info->delalloc_bytes -= end - start + 1;
 326                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 327                 }
 328                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 329         }
 330         return 0;
 331 }
 332
 333 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 334                          size_t size, struct bio *bio)
 335 {
 336         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 337         struct btrfs_mapping_tree *map_tree;
 338         u64 logical = bio->bi_sector << 9;
 339         u64 length = 0;
 340         u64 map_length;
 341         int ret;
 342
 343         length = bio->bi_size;
 344         map_tree = &root->fs_info->mapping_tree;
 345         map_length = length;
 346         ret = btrfs_map_block(map_tree, READ, logical,
 347                               &map_length, NULL, 0);
 348
 349         if (map_length < length + size) {
 350                 return 1;
 351         }
 352         return 0;
 353 }
 354
 355 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 356                           int mirror_num)
 357 {
 358         struct btrfs_root *root = BTRFS_I(inode)->root;
 359         int ret = 0;
 360
 361         ret = btrfs_csum_one_bio(root, inode, bio);
 362         BUG_ON(ret);
 363
 364         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
 365 }
 366
 367 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 368                           int mirror_num)
 369 {
 370         struct btrfs_root *root = BTRFS_I(inode)->root;
 371         int ret = 0;
 372
 373         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 374         BUG_ON(ret);
 375
 376         if (!(rw & (1 << BIO_RW))) {
 377                 if (!btrfs_test_opt(root, NODATASUM) &&
 378                     !btrfs_test_flag(inode, NODATASUM)) {
 379                         btrfs_lookup_bio_sums(root, inode, bio);
 380                 }
 381                 goto mapit;
 382         }
 383
 384         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 385                                    inode, rw, bio, mirror_num,
 386                                    __btrfs_submit_bio_hook);
 387 mapit:
 388         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
 389 }
 390
 391 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
 392                              struct inode *inode, u64 file_offset,
 393                              struct list_head *list)
 394 {
 395         struct list_head *cur;
 396         struct btrfs_ordered_sum *sum;
 397
 398         btrfs_set_trans_block_group(trans, inode);
 399         list_for_each(cur, list) {
 400                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
 401                 mutex_lock(&BTRFS_I(inode)->csum_mutex);
 402                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
 403                                        inode, sum);
 404                 mutex_unlock(&BTRFS_I(inode)->csum_mutex);
 405         }
 406         return 0;
 407 }
 408
 409 struct btrfs_writepage_fixup {
 410         struct page *page;
 411         struct btrfs_work work;
 412 };
 413
 414 /* see btrfs_writepage_start_hook for details on why this is required */
 415 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
 416 {
 417         struct btrfs_writepage_fixup *fixup;
 418         struct btrfs_ordered_extent *ordered;
 419         struct page *page;
 420         struct inode *inode;
 421         u64 page_start;
 422         u64 page_end;
 423
 424         fixup = container_of(work, struct btrfs_writepage_fixup, work);
 425         page = fixup->page;
 426 again:
 427         lock_page(page);
 428         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
 429                 ClearPageChecked(page);
 430                 goto out_page;
 431         }
 432
 433         inode = page->mapping->host;
 434         page_start = page_offset(page);
 435         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
 436
 437         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 438
 439         /* already ordered? We're done */
 440         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
 441                              EXTENT_ORDERED, 0)) {
 442                 goto out;
 443         }
 444
 445         ordered = btrfs_lookup_ordered_extent(inode, page_start);
 446         if (ordered) {
 447                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
 448                               page_end, GFP_NOFS);
 449                 unlock_page(page);
 450                 btrfs_start_ordered_extent(inode, ordered, 1);
 451                 goto again;
 452         }
 453
 454         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start, page_end,
 455                             GFP_NOFS);
 456         ClearPageChecked(page);
 457 out:
 458         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 459 out_page:
 460         unlock_page(page);
 461         page_cache_release(page);
 462 }
 463
 464 /*
 465  * There are a few paths in the higher layers of the kernel that directly
 466  * set the page dirty bit without asking the filesystem if it is a
 467  * good idea.  This causes problems because we want to make sure COW
 468  * properly happens and the data=ordered rules are followed.
 469  *
 470  * In our case any range that doesn't have the EXTENT_ORDERED bit set
 471  * hasn't been properly setup for IO.  We kick off an async process
 472  * to fix it up.  The async helper will wait for ordered extents, set
 473  * the delalloc bit and make it safe to write the page.
 474  */
 475 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
 476 {
 477         struct inode *inode = page->mapping->host;
 478         struct btrfs_writepage_fixup *fixup;
 479         struct btrfs_root *root = BTRFS_I(inode)->root;
 480         int ret;
 481
 482         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
 483                              EXTENT_ORDERED, 0);
 484         if (ret)
 485                 return 0;
 486
 487         if (PageChecked(page))
 488                 return -EAGAIN;
 489
 490         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
 491         if (!fixup)
 492                 return -EAGAIN;
 493
 494         SetPageChecked(page);
 495         page_cache_get(page);
 496         fixup->work.func = btrfs_writepage_fixup_worker;
 497         fixup->page = page;
 498         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
 499         return -EAGAIN;
 500 }
 501
 502 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 503 {
 504         struct btrfs_root *root = BTRFS_I(inode)->root;
 505         struct btrfs_trans_handle *trans;
 506         struct btrfs_ordered_extent *ordered_extent;
 507         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 508         u64 alloc_hint = 0;
 509         struct list_head list;
 510         struct btrfs_key ins;
 511         int ret;
 512
 513         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
 514         if (!ret)
 515                 return 0;
 516
 517         trans = btrfs_join_transaction(root, 1);
 518
 519         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 520         BUG_ON(!ordered_extent);
 521
 522         lock_extent(io_tree, ordered_extent->file_offset,
 523                     ordered_extent->file_offset + ordered_extent->len - 1,
 524                     GFP_NOFS);
 525
 526         INIT_LIST_HEAD(&list);
 527
 528         ins.objectid = ordered_extent->start;
 529         ins.offset = ordered_extent->len;
 530         ins.type = BTRFS_EXTENT_ITEM_KEY;
 531
 532         ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
 533                                           trans->transid, inode->i_ino,
 534                                           ordered_extent->file_offset, &ins);
 535         BUG_ON(ret);
 536
 537         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 538
 539         ret = btrfs_drop_extents(trans, root, inode,
 540                                  ordered_extent->file_offset,
 541                                  ordered_extent->file_offset +
 542                                  ordered_extent->len,
 543                                  ordered_extent->file_offset, &alloc_hint);
 544         BUG_ON(ret);
 545         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 546                                        ordered_extent->file_offset,
 547                                        ordered_extent->start,
 548                                        ordered_extent->len,
 549                                        ordered_extent->len, 0);
 550         BUG_ON(ret);
 551
 552         btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 553                                 ordered_extent->file_offset +
 554                                 ordered_extent->len - 1);
 555         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 556
 557         inode->i_blocks += ordered_extent->len >> 9;
 558         unlock_extent(io_tree, ordered_extent->file_offset,
 559                     ordered_extent->file_offset + ordered_extent->len - 1,
 560                     GFP_NOFS);
 561         add_pending_csums(trans, inode, ordered_extent->file_offset,
 562                           &ordered_extent->list);
 563
 564         btrfs_ordered_update_i_size(inode, ordered_extent);
 565         btrfs_remove_ordered_extent(inode, ordered_extent);
 566
 567         /* once for us */
 568         btrfs_put_ordered_extent(ordered_extent);
 569         /* once for the tree */
 570         btrfs_put_ordered_extent(ordered_extent);
 571
 572         btrfs_update_inode(trans, root, inode);
 573         btrfs_end_transaction(trans, root);
 574         return 0;
 575 }
 576
 577 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 578                                 struct extent_state *state, int uptodate)
 579 {
 580         return btrfs_finish_ordered_io(page->mapping->host, start, end);
 581 }
 582
 583 struct io_failure_record {
 584         struct page *page;
 585         u64 start;
 586         u64 len;
 587         u64 logical;
 588         int last_mirror;
 589 };
 590
 591 int btrfs_io_failed_hook(struct bio *failed_bio,
 592                          struct page *page, u64 start, u64 end,
 593                          struct extent_state *state)
 594 {
 595         struct io_failure_record *failrec = NULL;
 596         u64 private;
 597         struct extent_map *em;
 598         struct inode *inode = page->mapping->host;
 599         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 600         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 601         struct bio *bio;
 602         int num_copies;
 603         int ret;
 604         int rw;
 605         u64 logical;
 606
 607         ret = get_state_private(failure_tree, start, &private);
 608         if (ret) {
 609                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 610                 if (!failrec)
 611                         return -ENOMEM;
 612                 failrec->start = start;
 613                 failrec->len = end - start + 1;
 614                 failrec->last_mirror = 0;
 615
 616                 spin_lock(&em_tree->lock);
 617                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 618                 if (em->start > start || em->start + em->len < start) {
 619                         free_extent_map(em);
 620                         em = NULL;
 621                 }
 622                 spin_unlock(&em_tree->lock);
 623
 624                 if (!em || IS_ERR(em)) {
 625                         kfree(failrec);
 626                         return -EIO;
 627                 }
 628                 logical = start - em->start;
 629                 logical = em->block_start + logical;
 630                 failrec->logical = logical;
 631                 free_extent_map(em);
 632                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 633                                 EXTENT_DIRTY, GFP_NOFS);
 634                 set_state_private(failure_tree, start,
 635                                  (u64)(unsigned long)failrec);
 636         } else {
 637                 failrec = (struct io_failure_record *)(unsigned long)private;
 638         }
 639         num_copies = btrfs_num_copies(
 640                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 641                               failrec->logical, failrec->len);
 642         failrec->last_mirror++;
 643         if (!state) {
 644                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 645                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 646                                                     failrec->start,
 647                                                     EXTENT_LOCKED);
 648                 if (state && state->start != failrec->start)
 649                         state = NULL;
 650                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 651         }
 652         if (!state || failrec->last_mirror > num_copies) {
 653                 set_state_private(failure_tree, failrec->start, 0);
 654                 clear_extent_bits(failure_tree, failrec->start,
 655                                   failrec->start + failrec->len - 1,
 656                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 657                 kfree(failrec);
 658                 return -EIO;
 659         }
 660         bio = bio_alloc(GFP_NOFS, 1);
 661         bio->bi_private = state;
 662         bio->bi_end_io = failed_bio->bi_end_io;
 663         bio->bi_sector = failrec->logical >> 9;
 664         bio->bi_bdev = failed_bio->bi_bdev;
 665         bio->bi_size = 0;
 666         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 667         if (failed_bio->bi_rw & (1 << BIO_RW))
 668                 rw = WRITE;
 669         else
 670                 rw = READ;
 671
 672         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 673                                                       failrec->last_mirror);
 674         return 0;
 675 }
 676
 677 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 678 {
 679         u64 private;
 680         u64 private_failure;
 681         struct io_failure_record *failure;
 682         int ret;
 683
 684         private = 0;
 685         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 686                              (u64)-1, 1, EXTENT_DIRTY)) {
 687                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 688                                         start, &private_failure);
 689                 if (ret == 0) {
 690                         failure = (struct io_failure_record *)(unsigned long)
 691                                    private_failure;
 692                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 693                                           failure->start, 0);
 694                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 695                                           failure->start,
 696                                           failure->start + failure->len - 1,
 697                                           EXTENT_DIRTY | EXTENT_LOCKED,
 698                                           GFP_NOFS);
 699                         kfree(failure);
 700                 }
 701         }
 702         return 0;
 703 }
 704
 705 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 706                                struct extent_state *state)
 707 {
 708         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 709         struct inode *inode = page->mapping->host;
 710         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 711         char *kaddr;
 712         u64 private = ~(u32)0;
 713         int ret;
 714         struct btrfs_root *root = BTRFS_I(inode)->root;
 715         u32 csum = ~(u32)0;
 716         unsigned long flags;
 717
 718         if (btrfs_test_opt(root, NODATASUM) ||
 719             btrfs_test_flag(inode, NODATASUM))
 720                 return 0;
 721         if (state && state->start == start) {
 722                 private = state->private;
 723                 ret = 0;
 724         } else {
 725                 ret = get_state_private(io_tree, start, &private);
 726         }
 727         local_irq_save(flags);
 728         kaddr = kmap_atomic(page, KM_IRQ0);
 729         if (ret) {
 730                 goto zeroit;
 731         }
 732         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 733         btrfs_csum_final(csum, (char *)&csum);
 734         if (csum != private) {
 735                 goto zeroit;
 736         }
 737         kunmap_atomic(kaddr, KM_IRQ0);
 738         local_irq_restore(flags);
 739
 740         /* if the io failure tree for this inode is non-empty,
 741          * check to see if we've recovered from a failed IO
 742          */
 743         btrfs_clean_io_failures(inode, start);
 744         return 0;
 745
 746 zeroit:
 747         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 748                page->mapping->host->i_ino, (unsigned long long)start, csum,
 749                private);
 750         memset(kaddr + offset, 1, end - start + 1);
 751         flush_dcache_page(page);
 752         kunmap_atomic(kaddr, KM_IRQ0);
 753         local_irq_restore(flags);
 754         if (private == 0)
 755                 return 0;
 756         return -EIO;
 757 }
 758
 759 /*
 760  * This creates an orphan entry for the given inode in case something goes
 761  * wrong in the middle of an unlink/truncate.
 762  */
 763 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 764 {
 765         struct btrfs_root *root = BTRFS_I(inode)->root;
 766         int ret = 0;
 767
 768         spin_lock(&root->list_lock);
 769
 770         /* already on the orphan list, we're good */
 771         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 772                 spin_unlock(&root->list_lock);
 773                 return 0;
 774         }
 775
 776         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 777
 778         spin_unlock(&root->list_lock);
 779
 780         /*
 781          * insert an orphan item to track this unlinked/truncated file
 782          */
 783         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
 784
 785         return ret;
 786 }
 787
 788 /*
 789  * We have done the truncate/delete so we can go ahead and remove the orphan
 790  * item for this particular inode.
 791  */
 792 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
 793 {
 794         struct btrfs_root *root = BTRFS_I(inode)->root;
 795         int ret = 0;
 796
 797         spin_lock(&root->list_lock);
 798
 799         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
 800                 spin_unlock(&root->list_lock);
 801                 return 0;
 802         }
 803
 804         list_del_init(&BTRFS_I(inode)->i_orphan);
 805         if (!trans) {
 806                 spin_unlock(&root->list_lock);
 807                 return 0;
 808         }
 809
 810         spin_unlock(&root->list_lock);
 811
 812         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
 813
 814         return ret;
 815 }
 816
 817 /*
 818  * this cleans up any orphans that may be left on the list from the last use
 819  * of this root.
 820  */
 821 void btrfs_orphan_cleanup(struct btrfs_root *root)
 822 {
 823         struct btrfs_path *path;
 824         struct extent_buffer *leaf;
 825         struct btrfs_item *item;
 826         struct btrfs_key key, found_key;
 827         struct btrfs_trans_handle *trans;
 828         struct inode *inode;
 829         int ret = 0, nr_unlink = 0, nr_truncate = 0;
 830
 831         /* don't do orphan cleanup if the fs is readonly. */
 832         if (root->inode->i_sb->s_flags & MS_RDONLY)
 833                 return;
 834
 835         path = btrfs_alloc_path();
 836         if (!path)
 837                 return;
 838         path->reada = -1;
 839
 840         key.objectid = BTRFS_ORPHAN_OBJECTID;
 841         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
 842         key.offset = (u64)-1;
 843
 844         trans = btrfs_start_transaction(root, 1);
 845         btrfs_set_trans_block_group(trans, root->inode);
 846
 847         while (1) {
 848                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 849                 if (ret < 0) {
 850                         printk(KERN_ERR "Error searching slot for orphan: %d"
 851                                "\n", ret);
 852                         break;
 853                 }
 854
 855                 /*
 856                  * if ret == 0 means we found what we were searching for, which
 857                  * is weird, but possible, so only screw with path if we didnt
 858                  * find the key and see if we have stuff that matches
 859                  */
 860                 if (ret > 0) {
 861                         if (path->slots[0] == 0)
 862                                 break;
 863                         path->slots[0]--;
 864                 }
 865
 866                 /* pull out the item */
 867                 leaf = path->nodes[0];
 868                 item = btrfs_item_nr(leaf, path->slots[0]);
 869                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 870
 871                 /* make sure the item matches what we want */
 872                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
 873                         break;
 874                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
 875                         break;
 876
 877                 /* release the path since we're done with it */
 878                 btrfs_release_path(root, path);
 879
 880                 /*
 881                  * this is where we are basically btrfs_lookup, without the
 882                  * crossing root thing.  we store the inode number in the
 883                  * offset of the orphan item.
 884                  */
 885                 inode = btrfs_iget_locked(root->inode->i_sb,
 886                                           found_key.offset, root);
 887                 if (!inode)
 888                         break;
 889
 890                 if (inode->i_state & I_NEW) {
 891                         BTRFS_I(inode)->root = root;
 892
 893                         /* have to set the location manually */
 894                         BTRFS_I(inode)->location.objectid = inode->i_ino;
 895                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
 896                         BTRFS_I(inode)->location.offset = 0;
 897
 898                         btrfs_read_locked_inode(inode);
 899                         unlock_new_inode(inode);
 900                 }
 901
 902                 /*
 903                  * add this inode to the orphan list so btrfs_orphan_del does
 904                  * the proper thing when we hit it
 905                  */
 906                 spin_lock(&root->list_lock);
 907                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 908                 spin_unlock(&root->list_lock);
 909
 910                 /*
 911                  * if this is a bad inode, means we actually succeeded in
 912                  * removing the inode, but not the orphan record, which means
 913                  * we need to manually delete the orphan since iput will just
 914                  * do a destroy_inode
 915                  */
 916                 if (is_bad_inode(inode)) {
 917                         btrfs_orphan_del(trans, inode);
 918                         iput(inode);
 919                         continue;
 920                 }
 921
 922                 /* if we have links, this was a truncate, lets do that */
 923                 if (inode->i_nlink) {
 924                         nr_truncate++;
 925                         btrfs_truncate(inode);
 926                 } else {
 927                         nr_unlink++;
 928                 }
 929
 930                 /* this will do delete_inode and everything for us */
 931                 iput(inode);
 932         }
 933
 934         if (nr_unlink)
 935                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
 936         if (nr_truncate)
 937                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
 938
 939         btrfs_free_path(path);
 940         btrfs_end_transaction(trans, root);
 941 }
 942
 943 void btrfs_read_locked_inode(struct inode *inode)
 944 {
 945         struct btrfs_path *path;
 946         struct extent_buffer *leaf;
 947         struct btrfs_inode_item *inode_item;
 948         struct btrfs_timespec *tspec;
 949         struct btrfs_root *root = BTRFS_I(inode)->root;
 950         struct btrfs_key location;
 951         u64 alloc_group_block;
 952         u32 rdev;
 953         int ret;
 954
 955         path = btrfs_alloc_path();
 956         BUG_ON(!path);
 957         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 958
 959         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 960         if (ret)
 961                 goto make_bad;
 962
 963         leaf = path->nodes[0];
 964         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 965                                     struct btrfs_inode_item);
 966
 967         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 968         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 969         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 970         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 971         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
 972
 973         tspec = btrfs_inode_atime(inode_item);
 974         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 975         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 976
 977         tspec = btrfs_inode_mtime(inode_item);
 978         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 979         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 980
 981         tspec = btrfs_inode_ctime(inode_item);
 982         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 983         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 984
 985         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 986         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 987         inode->i_rdev = 0;
 988         rdev = btrfs_inode_rdev(leaf, inode_item);
 989
 990         BTRFS_I(inode)->index_cnt = (u64)-1;
 991
 992         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 993         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 994                                                        alloc_group_block);
 995         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 996         if (!BTRFS_I(inode)->block_group) {
 997                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 998                                                  NULL, 0,
 999                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1000         }
1001         btrfs_free_path(path);
1002         inode_item = NULL;
1003
1004         switch (inode->i_mode & S_IFMT) {
1005         case S_IFREG:
1006                 inode->i_mapping->a_ops = &btrfs_aops;
1007                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1008                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1009                 inode->i_fop = &btrfs_file_operations;
1010                 inode->i_op = &btrfs_file_inode_operations;
1011                 break;
1012         case S_IFDIR:
1013                 inode->i_fop = &btrfs_dir_file_operations;
1014                 if (root == root->fs_info->tree_root)
1015                         inode->i_op = &btrfs_dir_ro_inode_operations;
1016                 else
1017                         inode->i_op = &btrfs_dir_inode_operations;
1018                 break;
1019         case S_IFLNK:
1020                 inode->i_op = &btrfs_symlink_inode_operations;
1021                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1022                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1023                 break;
1024         default:
1025                 init_special_inode(inode, inode->i_mode, rdev);
1026                 break;
1027         }
1028         return;
1029
1030 make_bad:
1031         btrfs_free_path(path);
1032         make_bad_inode(inode);
1033 }
1034
1035 static void fill_inode_item(struct extent_buffer *leaf,
1036                             struct btrfs_inode_item *item,
1037                             struct inode *inode)
1038 {
1039         btrfs_set_inode_uid(leaf, item, inode->i_uid);
1040         btrfs_set_inode_gid(leaf, item, inode->i_gid);
1041         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1042         btrfs_set_inode_mode(leaf, item, inode->i_mode);
1043         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1044
1045         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1046                                inode->i_atime.tv_sec);
1047         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1048                                 inode->i_atime.tv_nsec);
1049
1050         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1051                                inode->i_mtime.tv_sec);
1052         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1053                                 inode->i_mtime.tv_nsec);
1054
1055         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1056                                inode->i_ctime.tv_sec);
1057         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1058                                 inode->i_ctime.tv_nsec);
1059
1060         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1061         btrfs_set_inode_generation(leaf, item, inode->i_generation);
1062         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1063         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1064         btrfs_set_inode_block_group(leaf, item,
1065                                     BTRFS_I(inode)->block_group->key.objectid);
1066 }
1067
1068 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1069                               struct btrfs_root *root,
1070                               struct inode *inode)
1071 {
1072         struct btrfs_inode_item *inode_item;
1073         struct btrfs_path *path;
1074         struct extent_buffer *leaf;
1075         int ret;
1076
1077         path = btrfs_alloc_path();
1078         BUG_ON(!path);
1079         ret = btrfs_lookup_inode(trans, root, path,
1080                                  &BTRFS_I(inode)->location, 1);
1081         if (ret) {
1082                 if (ret > 0)
1083                         ret = -ENOENT;
1084                 goto failed;
1085         }
1086
1087         leaf = path->nodes[0];
1088         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1089                                   struct btrfs_inode_item);
1090
1091         fill_inode_item(leaf, inode_item, inode);
1092         btrfs_mark_buffer_dirty(leaf);
1093         btrfs_set_inode_last_trans(trans, inode);
1094         ret = 0;
1095 failed:
1096         btrfs_free_path(path);
1097         return ret;
1098 }
1099
1100
1101 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
1102                               struct btrfs_root *root,
1103                               struct inode *dir,
1104                               struct dentry *dentry)
1105 {
1106         struct btrfs_path *path;
1107         const char *name = dentry->d_name.name;
1108         int name_len = dentry->d_name.len;
1109         int ret = 0;
1110         struct extent_buffer *leaf;
1111         struct btrfs_dir_item *di;
1112         struct btrfs_key key;
1113         u64 index;
1114
1115         path = btrfs_alloc_path();
1116         if (!path) {
1117                 ret = -ENOMEM;
1118                 goto err;
1119         }
1120
1121         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1122                                     name, name_len, -1);
1123         if (IS_ERR(di)) {
1124                 ret = PTR_ERR(di);
1125                 goto err;
1126         }
1127         if (!di) {
1128                 ret = -ENOENT;
1129                 goto err;
1130         }
1131         leaf = path->nodes[0];
1132         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1133         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1134         if (ret)
1135                 goto err;
1136         btrfs_release_path(root, path);
1137
1138         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1139                                   dentry->d_inode->i_ino,
1140                                   dentry->d_parent->d_inode->i_ino, &index);
1141         if (ret) {
1142                 printk("failed to delete reference to %.*s, "
1143                        "inode %lu parent %lu\n", name_len, name,
1144                        dentry->d_inode->i_ino,
1145                        dentry->d_parent->d_inode->i_ino);
1146                 goto err;
1147         }
1148
1149         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1150                                          index, name, name_len, -1);
1151         if (IS_ERR(di)) {
1152                 ret = PTR_ERR(di);
1153                 goto err;
1154         }
1155         if (!di) {
1156                 ret = -ENOENT;
1157                 goto err;
1158         }
1159         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1160         btrfs_release_path(root, path);
1161
1162         dentry->d_inode->i_ctime = dir->i_ctime;
1163 err:
1164         btrfs_free_path(path);
1165         if (!ret) {
1166                 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1167                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1168                 btrfs_update_inode(trans, root, dir);
1169 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1170                 dentry->d_inode->i_nlink--;
1171 #else
1172                 drop_nlink(dentry->d_inode);
1173 #endif
1174                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
1175                 dir->i_sb->s_dirt = 1;
1176         }
1177         return ret;
1178 }
1179
1180 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1181 {
1182         struct btrfs_root *root;
1183         struct btrfs_trans_handle *trans;
1184         struct inode *inode = dentry->d_inode;
1185         int ret;
1186         unsigned long nr = 0;
1187
1188         root = BTRFS_I(dir)->root;
1189
1190         ret = btrfs_check_free_space(root, 1, 1);
1191         if (ret)
1192                 goto fail;
1193
1194         trans = btrfs_start_transaction(root, 1);
1195
1196         btrfs_set_trans_block_group(trans, dir);
1197         ret = btrfs_unlink_trans(trans, root, dir, dentry);
1198
1199         if (inode->i_nlink == 0)
1200                 ret = btrfs_orphan_add(trans, inode);
1201
1202         nr = trans->blocks_used;
1203
1204         btrfs_end_transaction_throttle(trans, root);
1205 fail:
1206         btrfs_btree_balance_dirty(root, nr);
1207         return ret;
1208 }
1209
1210 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1211 {
1212         struct inode *inode = dentry->d_inode;
1213         int err = 0;
1214         int ret;
1215         struct btrfs_root *root = BTRFS_I(dir)->root;
1216         struct btrfs_trans_handle *trans;
1217         unsigned long nr = 0;
1218
1219         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1220                 return -ENOTEMPTY;
1221         }
1222
1223         ret = btrfs_check_free_space(root, 1, 1);
1224         if (ret)
1225                 goto fail;
1226
1227         trans = btrfs_start_transaction(root, 1);
1228         btrfs_set_trans_block_group(trans, dir);
1229
1230         err = btrfs_orphan_add(trans, inode);
1231         if (err)
1232                 goto fail_trans;
1233
1234         /* now the directory is empty */
1235         err = btrfs_unlink_trans(trans, root, dir, dentry);
1236         if (!err) {
1237                 btrfs_i_size_write(inode, 0);
1238         }
1239
1240 fail_trans:
1241         nr = trans->blocks_used;
1242         ret = btrfs_end_transaction_throttle(trans, root);
1243 fail:
1244         btrfs_btree_balance_dirty(root, nr);
1245
1246         if (ret && !err)
1247                 err = ret;
1248         return err;
1249 }
1250
1251 /*
1252  * this can truncate away extent items, csum items and directory items.
1253  * It starts at a high offset and removes keys until it can't find
1254  * any higher than i_size.
1255  *
1256  * csum items that cross the new i_size are truncated to the new size
1257  * as well.
1258  *
1259  * min_type is the minimum key type to truncate down to.  If set to 0, this
1260  * will kill all the items on this inode, including the INODE_ITEM_KEY.
1261  */
1262 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1263                                    struct btrfs_root *root,
1264                                    struct inode *inode,
1265                                    u32 min_type)
1266 {
1267         int ret;
1268         struct btrfs_path *path;
1269         struct btrfs_key key;
1270         struct btrfs_key found_key;
1271         u32 found_type;
1272         struct extent_buffer *leaf;
1273         struct btrfs_file_extent_item *fi;
1274         u64 extent_start = 0;
1275         u64 extent_num_bytes = 0;
1276         u64 item_end = 0;
1277         u64 root_gen = 0;
1278         u64 root_owner = 0;
1279         int found_extent;
1280         int del_item;
1281         int pending_del_nr = 0;
1282         int pending_del_slot = 0;
1283         int extent_type = -1;
1284         u64 mask = root->sectorsize - 1;
1285
1286         btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
1287         path = btrfs_alloc_path();
1288         path->reada = -1;
1289         BUG_ON(!path);
1290
1291         /* FIXME, add redo link to tree so we don't leak on crash */
1292         key.objectid = inode->i_ino;
1293         key.offset = (u64)-1;
1294         key.type = (u8)-1;
1295
1296         btrfs_init_path(path);
1297 search_again:
1298         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1299         if (ret < 0) {
1300                 goto error;
1301         }
1302         if (ret > 0) {
1303                 BUG_ON(path->slots[0] == 0);
1304                 path->slots[0]--;
1305         }
1306
1307         while(1) {
1308                 fi = NULL;
1309                 leaf = path->nodes[0];
1310                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1311                 found_type = btrfs_key_type(&found_key);
1312
1313                 if (found_key.objectid != inode->i_ino)
1314                         break;
1315
1316                 if (found_type < min_type)
1317                         break;
1318
1319                 item_end = found_key.offset;
1320                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1321                         fi = btrfs_item_ptr(leaf, path->slots[0],
1322                                             struct btrfs_file_extent_item);
1323                         extent_type = btrfs_file_extent_type(leaf, fi);
1324                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1325                                 item_end +=
1326                                     btrfs_file_extent_num_bytes(leaf, fi);
1327                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1328                                 struct btrfs_item *item = btrfs_item_nr(leaf,
1329                                                                 path->slots[0]);
1330                                 item_end += btrfs_file_extent_inline_len(leaf,
1331                                                                          item);
1332                         }
1333                         item_end--;
1334                 }
1335                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1336                         ret = btrfs_csum_truncate(trans, root, path,
1337                                                   inode->i_size);
1338                         BUG_ON(ret);
1339                 }
1340                 if (item_end < inode->i_size) {
1341                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1342                                 found_type = BTRFS_INODE_ITEM_KEY;
1343                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1344                                 found_type = BTRFS_CSUM_ITEM_KEY;
1345                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1346                                 found_type = BTRFS_XATTR_ITEM_KEY;
1347                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1348                                 found_type = BTRFS_INODE_REF_KEY;
1349                         } else if (found_type) {
1350                                 found_type--;
1351                         } else {
1352                                 break;
1353                         }
1354                         btrfs_set_key_type(&key, found_type);
1355                         goto next;
1356                 }
1357                 if (found_key.offset >= inode->i_size)
1358                         del_item = 1;
1359                 else
1360                         del_item = 0;
1361                 found_extent = 0;
1362
1363                 /* FIXME, shrink the extent if the ref count is only 1 */
1364                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1365                         goto delete;
1366
1367                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1368                         u64 num_dec;
1369                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1370                         if (!del_item) {
1371                                 u64 orig_num_bytes =
1372                                         btrfs_file_extent_num_bytes(leaf, fi);
1373                                 extent_num_bytes = inode->i_size -
1374                                         found_key.offset + root->sectorsize - 1;
1375                                 extent_num_bytes = extent_num_bytes &
1376                                         ~((u64)root->sectorsize - 1);
1377                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1378                                                          extent_num_bytes);
1379                                 num_dec = (orig_num_bytes -
1380                                            extent_num_bytes);
1381                                 if (extent_start != 0)
1382                                         dec_i_blocks(inode, num_dec);
1383                                 btrfs_mark_buffer_dirty(leaf);
1384                         } else {
1385                                 extent_num_bytes =
1386                                         btrfs_file_extent_disk_num_bytes(leaf,
1387                                                                          fi);
1388                                 /* FIXME blocksize != 4096 */
1389                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1390                                 if (extent_start != 0) {
1391                                         found_extent = 1;
1392                                         dec_i_blocks(inode, num_dec);
1393                                 }
1394                                 root_gen = btrfs_header_generation(leaf);
1395                                 root_owner = btrfs_header_owner(leaf);
1396                         }
1397                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1398                         if (!del_item) {
1399                                 u32 newsize = inode->i_size - found_key.offset;
1400                                 dec_i_blocks(inode, item_end + 1 -
1401                                             found_key.offset - newsize);
1402                                 newsize =
1403                                     btrfs_file_extent_calc_inline_size(newsize);
1404                                 ret = btrfs_truncate_item(trans, root, path,
1405                                                           newsize, 1);
1406                                 BUG_ON(ret);
1407                         } else {
1408                                 dec_i_blocks(inode, item_end + 1 -
1409                                              found_key.offset);
1410                         }
1411                 }
1412 delete:
1413                 if (del_item) {
1414                         if (!pending_del_nr) {
1415                                 /* no pending yet, add ourselves */
1416                                 pending_del_slot = path->slots[0];
1417                                 pending_del_nr = 1;
1418                         } else if (pending_del_nr &&
1419                                    path->slots[0] + 1 == pending_del_slot) {
1420                                 /* hop on the pending chunk */
1421                                 pending_del_nr++;
1422                                 pending_del_slot = path->slots[0];
1423                         } else {
1424                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1425                         }
1426                 } else {
1427                         break;
1428                 }
1429                 if (found_extent) {
1430                         ret = btrfs_free_extent(trans, root, extent_start,
1431                                                 extent_num_bytes,
1432                                                 root_owner,
1433                                                 root_gen, inode->i_ino,
1434                                                 found_key.offset, 0);
1435                         BUG_ON(ret);
1436                 }
1437 next:
1438                 if (path->slots[0] == 0) {
1439                         if (pending_del_nr)
1440                                 goto del_pending;
1441                         btrfs_release_path(root, path);
1442                         goto search_again;
1443                 }
1444
1445                 path->slots[0]--;
1446                 if (pending_del_nr &&
1447                     path->slots[0] + 1 != pending_del_slot) {
1448                         struct btrfs_key debug;
1449 del_pending:
1450                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1451                                               pending_del_slot);
1452                         ret = btrfs_del_items(trans, root, path,
1453                                               pending_del_slot,
1454                                               pending_del_nr);
1455                         BUG_ON(ret);
1456                         pending_del_nr = 0;
1457                         btrfs_release_path(root, path);
1458                         goto search_again;
1459                 }
1460         }
1461         ret = 0;
1462 error:
1463         if (pending_del_nr) {
1464                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1465                                       pending_del_nr);
1466         }
1467         btrfs_free_path(path);
1468         inode->i_sb->s_dirt = 1;
1469         return ret;
1470 }
1471
1472 /*
1473  * taken from block_truncate_page, but does cow as it zeros out
1474  * any bytes left in the last page in the file.
1475  */
1476 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1477 {
1478         struct inode *inode = mapping->host;
1479         struct btrfs_root *root = BTRFS_I(inode)->root;
1480         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1481         struct btrfs_ordered_extent *ordered;
1482         char *kaddr;
1483         u32 blocksize = root->sectorsize;
1484         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1485         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1486         struct page *page;
1487         int ret = 0;
1488         u64 page_start;
1489         u64 page_end;
1490
1491         if ((offset & (blocksize - 1)) == 0)
1492                 goto out;
1493
1494         ret = -ENOMEM;
1495 again:
1496         page = grab_cache_page(mapping, index);
1497         if (!page)
1498                 goto out;
1499
1500         page_start = page_offset(page);
1501         page_end = page_start + PAGE_CACHE_SIZE - 1;
1502
1503         if (!PageUptodate(page)) {
1504                 ret = btrfs_readpage(NULL, page);
1505                 lock_page(page);
1506                 if (page->mapping != mapping) {
1507                         unlock_page(page);
1508                         page_cache_release(page);
1509                         goto again;
1510                 }
1511                 if (!PageUptodate(page)) {
1512                         ret = -EIO;
1513                         goto out_unlock;
1514                 }
1515         }
1516         wait_on_page_writeback(page);
1517
1518         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1519         set_page_extent_mapped(page);
1520
1521         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1522         if (ordered) {
1523                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1524                 unlock_page(page);
1525                 page_cache_release(page);
1526                 btrfs_start_ordered_extent(inode, ordered, 1);
1527                 btrfs_put_ordered_extent(ordered);
1528                 goto again;
1529         }
1530
1531         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
1532                             page_end, GFP_NOFS);
1533         ret = 0;
1534         if (offset != PAGE_CACHE_SIZE) {
1535                 kaddr = kmap(page);
1536                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1537                 flush_dcache_page(page);
1538                 kunmap(page);
1539         }
1540         ClearPageChecked(page);
1541         set_page_dirty(page);
1542         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1543
1544 out_unlock:
1545         unlock_page(page);
1546         page_cache_release(page);
1547 out:
1548         return ret;
1549 }
1550
1551 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1552 {
1553         struct inode *inode = dentry->d_inode;
1554         int err;
1555
1556         err = inode_change_ok(inode, attr);
1557         if (err)
1558                 return err;
1559
1560         if (S_ISREG(inode->i_mode) &&
1561             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1562                 struct btrfs_trans_handle *trans;
1563                 struct btrfs_root *root = BTRFS_I(inode)->root;
1564                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1565
1566                 u64 mask = root->sectorsize - 1;
1567                 u64 hole_start = (inode->i_size + mask) & ~mask;
1568                 u64 block_end = (attr->ia_size + mask) & ~mask;
1569                 u64 hole_size;
1570                 u64 alloc_hint = 0;
1571
1572                 if (attr->ia_size <= hole_start)
1573                         goto out;
1574
1575                 err = btrfs_check_free_space(root, 1, 0);
1576                 if (err)
1577                         goto fail;
1578
1579                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1580
1581                 hole_size = block_end - hole_start;
1582                 btrfs_wait_ordered_range(inode, hole_start, hole_size);
1583                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1584
1585                 trans = btrfs_start_transaction(root, 1);
1586                 btrfs_set_trans_block_group(trans, inode);
1587                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1588                 err = btrfs_drop_extents(trans, root, inode,
1589                                          hole_start, block_end, hole_start,
1590                                          &alloc_hint);
1591
1592                 if (alloc_hint != EXTENT_MAP_INLINE) {
1593                         err = btrfs_insert_file_extent(trans, root,
1594                                                        inode->i_ino,
1595                                                        hole_start, 0, 0,
1596                                                        hole_size, 0);
1597                         btrfs_drop_extent_cache(inode, hole_start,
1598                                                 (u64)-1);
1599                         btrfs_check_file(root, inode);
1600                 }
1601                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1602                 btrfs_end_transaction(trans, root);
1603                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1604                 if (err)
1605                         return err;
1606         }
1607 out:
1608         err = inode_setattr(inode, attr);
1609
1610         if (!err && ((attr->ia_valid & ATTR_MODE)))
1611                 err = btrfs_acl_chmod(inode);
1612 fail:
1613         return err;
1614 }
1615
1616 void btrfs_delete_inode(struct inode *inode)
1617 {
1618         struct btrfs_trans_handle *trans;
1619         struct btrfs_root *root = BTRFS_I(inode)->root;
1620         unsigned long nr;
1621         int ret;
1622
1623         truncate_inode_pages(&inode->i_data, 0);
1624         if (is_bad_inode(inode)) {
1625                 btrfs_orphan_del(NULL, inode);
1626                 goto no_delete;
1627         }
1628         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1629
1630         btrfs_i_size_write(inode, 0);
1631         trans = btrfs_start_transaction(root, 1);
1632
1633         btrfs_set_trans_block_group(trans, inode);
1634         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1635         if (ret) {
1636                 btrfs_orphan_del(NULL, inode);
1637                 goto no_delete_lock;
1638         }
1639
1640         btrfs_orphan_del(trans, inode);
1641
1642         nr = trans->blocks_used;
1643         clear_inode(inode);
1644
1645         btrfs_end_transaction(trans, root);
1646         btrfs_btree_balance_dirty(root, nr);
1647         return;
1648
1649 no_delete_lock:
1650         nr = trans->blocks_used;
1651         btrfs_end_transaction(trans, root);
1652         btrfs_btree_balance_dirty(root, nr);
1653 no_delete:
1654         clear_inode(inode);
1655 }
1656
1657 /*
1658  * this returns the key found in the dir entry in the location pointer.
1659  * If no dir entries were found, location->objectid is 0.
1660  */
1661 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1662                                struct btrfs_key *location)
1663 {
1664         const char *name = dentry->d_name.name;
1665         int namelen = dentry->d_name.len;
1666         struct btrfs_dir_item *di;
1667         struct btrfs_path *path;
1668         struct btrfs_root *root = BTRFS_I(dir)->root;
1669         int ret = 0;
1670
1671         if (namelen == 1 && strcmp(name, ".") == 0) {
1672                 location->objectid = dir->i_ino;
1673                 location->type = BTRFS_INODE_ITEM_KEY;
1674                 location->offset = 0;
1675                 return 0;
1676         }
1677         path = btrfs_alloc_path();
1678         BUG_ON(!path);
1679
1680         if (namelen == 2 && strcmp(name, "..") == 0) {
1681                 struct btrfs_key key;
1682                 struct extent_buffer *leaf;
1683                 int slot;
1684
1685                 key.objectid = dir->i_ino;
1686                 key.offset = (u64)-1;
1687                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1688                 if (ret < 0 || path->slots[0] == 0)
1689                         goto out_err;
1690                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1691                 BUG_ON(ret == 0);
1692                 ret = 0;
1693                 leaf = path->nodes[0];
1694                 slot = path->slots[0] - 1;
1695
1696                 btrfs_item_key_to_cpu(leaf, &key, slot);
1697                 if (key.objectid != dir->i_ino ||
1698                     key.type != BTRFS_INODE_REF_KEY) {
1699                         goto out_err;
1700                 }
1701                 location->objectid = key.offset;
1702                 location->type = BTRFS_INODE_ITEM_KEY;
1703                 location->offset = 0;
1704                 goto out;
1705         }
1706
1707         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1708                                     namelen, 0);
1709         if (IS_ERR(di))
1710                 ret = PTR_ERR(di);
1711         if (!di || IS_ERR(di)) {
1712                 goto out_err;
1713         }
1714         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1715 out:
1716         btrfs_free_path(path);
1717         return ret;
1718 out_err:
1719         location->objectid = 0;
1720         goto out;
1721 }
1722
1723 /*
1724  * when we hit a tree root in a directory, the btrfs part of the inode
1725  * needs to be changed to reflect the root directory of the tree root.  This
1726  * is kind of like crossing a mount point.
1727  */
1728 static int fixup_tree_root_location(struct btrfs_root *root,
1729                              struct btrfs_key *location,
1730                              struct btrfs_root **sub_root,
1731                              struct dentry *dentry)
1732 {
1733         struct btrfs_root_item *ri;
1734
1735         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1736                 return 0;
1737         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1738                 return 0;
1739
1740         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1741                                         dentry->d_name.name,
1742                                         dentry->d_name.len);
1743         if (IS_ERR(*sub_root))
1744                 return PTR_ERR(*sub_root);
1745
1746         ri = &(*sub_root)->root_item;
1747         location->objectid = btrfs_root_dirid(ri);
1748         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1749         location->offset = 0;
1750
1751         return 0;
1752 }
1753
1754 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1755 {
1756         struct btrfs_iget_args *args = p;
1757         inode->i_ino = args->ino;
1758         BTRFS_I(inode)->root = args->root;
1759         BTRFS_I(inode)->delalloc_bytes = 0;
1760         BTRFS_I(inode)->disk_i_size = 0;
1761         BTRFS_I(inode)->index_cnt = (u64)-1;
1762         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1763         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1764                              inode->i_mapping, GFP_NOFS);
1765         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1766                              inode->i_mapping, GFP_NOFS);
1767         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1768         mutex_init(&BTRFS_I(inode)->csum_mutex);
1769         mutex_init(&BTRFS_I(inode)->extent_mutex);
1770         return 0;
1771 }
1772
1773 static int btrfs_find_actor(struct inode *inode, void *opaque)
1774 {
1775         struct btrfs_iget_args *args = opaque;
1776         return (args->ino == inode->i_ino &&
1777                 args->root == BTRFS_I(inode)->root);
1778 }
1779
1780 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1781                             u64 root_objectid)
1782 {
1783         struct btrfs_iget_args args;
1784         args.ino = objectid;
1785         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1786
1787         if (!args.root)
1788                 return NULL;
1789
1790         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1791 }
1792
1793 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1794                                 struct btrfs_root *root)
1795 {
1796         struct inode *inode;
1797         struct btrfs_iget_args args;
1798         args.ino = objectid;
1799         args.root = root;
1800
1801         inode = iget5_locked(s, objectid, btrfs_find_actor,
1802                              btrfs_init_locked_inode,
1803                              (void *)&args);
1804         return inode;
1805 }
1806
1807 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1808                                    struct nameidata *nd)
1809 {
1810         struct inode * inode;
1811         struct btrfs_inode *bi = BTRFS_I(dir);
1812         struct btrfs_root *root = bi->root;
1813         struct btrfs_root *sub_root = root;
1814         struct btrfs_key location;
1815         int ret, do_orphan = 0;
1816
1817         if (dentry->d_name.len > BTRFS_NAME_LEN)
1818                 return ERR_PTR(-ENAMETOOLONG);
1819
1820         ret = btrfs_inode_by_name(dir, dentry, &location);
1821
1822         if (ret < 0)
1823                 return ERR_PTR(ret);
1824
1825         inode = NULL;
1826         if (location.objectid) {
1827                 ret = fixup_tree_root_location(root, &location, &sub_root,
1828                                                 dentry);
1829                 if (ret < 0)
1830                         return ERR_PTR(ret);
1831                 if (ret > 0)
1832                         return ERR_PTR(-ENOENT);
1833
1834                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1835                                           sub_root);
1836                 if (!inode)
1837                         return ERR_PTR(-EACCES);
1838                 if (inode->i_state & I_NEW) {
1839                         /* the inode and parent dir are two different roots */
1840                         if (sub_root != root) {
1841                                 igrab(inode);
1842                                 sub_root->inode = inode;
1843                                 do_orphan = 1;
1844                         }
1845                         BTRFS_I(inode)->root = sub_root;
1846                         memcpy(&BTRFS_I(inode)->location, &location,
1847                                sizeof(location));
1848                         btrfs_read_locked_inode(inode);
1849                         unlock_new_inode(inode);
1850                 }
1851         }
1852
1853         if (unlikely(do_orphan))
1854                 btrfs_orphan_cleanup(sub_root);
1855
1856         return d_splice_alias(inode, dentry);
1857 }
1858
1859 static unsigned char btrfs_filetype_table[] = {
1860         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1861 };
1862
1863 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1864 {
1865         struct inode *inode = filp->f_dentry->d_inode;
1866         struct btrfs_root *root = BTRFS_I(inode)->root;
1867         struct btrfs_item *item;
1868         struct btrfs_dir_item *di;
1869         struct btrfs_key key;
1870         struct btrfs_key found_key;
1871         struct btrfs_path *path;
1872         int ret;
1873         u32 nritems;
1874         struct extent_buffer *leaf;
1875         int slot;
1876         int advance;
1877         unsigned char d_type;
1878         int over = 0;
1879         u32 di_cur;
1880         u32 di_total;
1881         u32 di_len;
1882         int key_type = BTRFS_DIR_INDEX_KEY;
1883         char tmp_name[32];
1884         char *name_ptr;
1885         int name_len;
1886
1887         /* FIXME, use a real flag for deciding about the key type */
1888         if (root->fs_info->tree_root == root)
1889                 key_type = BTRFS_DIR_ITEM_KEY;
1890
1891         /* special case for "." */
1892         if (filp->f_pos == 0) {
1893                 over = filldir(dirent, ".", 1,
1894                                1, inode->i_ino,
1895                                DT_DIR);
1896                 if (over)
1897                         return 0;
1898                 filp->f_pos = 1;
1899         }
1900
1901         key.objectid = inode->i_ino;
1902         path = btrfs_alloc_path();
1903         path->reada = 2;
1904
1905         /* special case for .., just use the back ref */
1906         if (filp->f_pos == 1) {
1907                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1908                 key.offset = (u64)-1;
1909                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1910                 if (ret < 0 || path->slots[0] == 0) {
1911                         btrfs_release_path(root, path);
1912                         goto read_dir_items;
1913                 }
1914                 BUG_ON(ret == 0);
1915                 leaf = path->nodes[0];
1916                 slot = path->slots[0] - 1;
1917                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1918                 btrfs_release_path(root, path);
1919                 if (found_key.objectid != key.objectid ||
1920                     found_key.type != BTRFS_INODE_REF_KEY)
1921                         goto read_dir_items;
1922                 over = filldir(dirent, "..", 2,
1923                                2, found_key.offset, DT_DIR);
1924                 if (over)
1925                         goto nopos;
1926                 filp->f_pos = 2;
1927         }
1928
1929 read_dir_items:
1930         btrfs_set_key_type(&key, key_type);
1931         key.offset = filp->f_pos;
1932
1933         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1934         if (ret < 0)
1935                 goto err;
1936         advance = 0;
1937         while(1) {
1938                 leaf = path->nodes[0];
1939                 nritems = btrfs_header_nritems(leaf);
1940                 slot = path->slots[0];
1941                 if (advance || slot >= nritems) {
1942                         if (slot >= nritems -1) {
1943                                 ret = btrfs_next_leaf(root, path);
1944                                 if (ret)
1945                                         break;
1946                                 leaf = path->nodes[0];
1947                                 nritems = btrfs_header_nritems(leaf);
1948                                 slot = path->slots[0];
1949                         } else {
1950                                 slot++;
1951                                 path->slots[0]++;
1952                         }
1953                 }
1954                 advance = 1;
1955                 item = btrfs_item_nr(leaf, slot);
1956                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1957
1958                 if (found_key.objectid != key.objectid)
1959                         break;
1960                 if (btrfs_key_type(&found_key) != key_type)
1961                         break;
1962                 if (found_key.offset < filp->f_pos)
1963                         continue;
1964
1965                 filp->f_pos = found_key.offset;
1966                 advance = 1;
1967                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1968                 di_cur = 0;
1969                 di_total = btrfs_item_size(leaf, item);
1970                 while(di_cur < di_total) {
1971                         struct btrfs_key location;
1972
1973                         name_len = btrfs_dir_name_len(leaf, di);
1974                         if (name_len < 32) {
1975                                 name_ptr = tmp_name;
1976                         } else {
1977                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1978                                 BUG_ON(!name_ptr);
1979                         }
1980                         read_extent_buffer(leaf, name_ptr,
1981                                            (unsigned long)(di + 1), name_len);
1982
1983                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1984                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1985                         over = filldir(dirent, name_ptr, name_len,
1986                                        found_key.offset,
1987                                        location.objectid,
1988                                        d_type);
1989
1990                         if (name_ptr != tmp_name)
1991                                 kfree(name_ptr);
1992
1993                         if (over)
1994                                 goto nopos;
1995                         di_len = btrfs_dir_name_len(leaf, di) +
1996                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1997                         di_cur += di_len;
1998                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1999                 }
2000         }
2001         if (key_type == BTRFS_DIR_INDEX_KEY)
2002                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2003         else
2004                 filp->f_pos++;
2005 nopos:
2006         ret = 0;
2007 err:
2008         btrfs_free_path(path);
2009         return ret;
2010 }
2011
2012 int btrfs_write_inode(struct inode *inode, int wait)
2013 {
2014         struct btrfs_root *root = BTRFS_I(inode)->root;
2015         struct btrfs_trans_handle *trans;
2016         int ret = 0;
2017
2018         if (wait) {
2019                 trans = btrfs_join_transaction(root, 1);
2020                 btrfs_set_trans_block_group(trans, inode);
2021                 ret = btrfs_commit_transaction(trans, root);
2022         }
2023         return ret;
2024 }
2025
2026 /*
2027  * This is somewhat expensive, updating the tree every time the
2028  * inode changes.  But, it is most likely to find the inode in cache.
2029  * FIXME, needs more benchmarking...there are no reasons other than performance
2030  * to keep or drop this code.
2031  */
2032 void btrfs_dirty_inode(struct inode *inode)
2033 {
2034         struct btrfs_root *root = BTRFS_I(inode)->root;
2035         struct btrfs_trans_handle *trans;
2036
2037         trans = btrfs_join_transaction(root, 1);
2038         btrfs_set_trans_block_group(trans, inode);
2039         btrfs_update_inode(trans, root, inode);
2040         btrfs_end_transaction(trans, root);
2041 }
2042
2043 static int btrfs_set_inode_index_count(struct inode *inode)
2044 {
2045         struct btrfs_root *root = BTRFS_I(inode)->root;
2046         struct btrfs_key key, found_key;
2047         struct btrfs_path *path;
2048         struct extent_buffer *leaf;
2049         int ret;
2050
2051         key.objectid = inode->i_ino;
2052         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2053         key.offset = (u64)-1;
2054
2055         path = btrfs_alloc_path();
2056         if (!path)
2057                 return -ENOMEM;
2058
2059         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2060         if (ret < 0)
2061                 goto out;
2062         /* FIXME: we should be able to handle this */
2063         if (ret == 0)
2064                 goto out;
2065         ret = 0;
2066
2067         /*
2068          * MAGIC NUMBER EXPLANATION:
2069          * since we search a directory based on f_pos we have to start at 2
2070          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2071          * else has to start at 2
2072          */
2073         if (path->slots[0] == 0) {
2074                 BTRFS_I(inode)->index_cnt = 2;
2075                 goto out;
2076         }
2077
2078         path->slots[0]--;
2079
2080         leaf = path->nodes[0];
2081         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2082
2083         if (found_key.objectid != inode->i_ino ||
2084             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2085                 BTRFS_I(inode)->index_cnt = 2;
2086                 goto out;
2087         }
2088
2089         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2090 out:
2091         btrfs_free_path(path);
2092         return ret;
2093 }
2094
2095 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode)
2096 {
2097         int ret = 0;
2098
2099         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2100                 ret = btrfs_set_inode_index_count(dir);
2101                 if (ret)
2102                         return ret;
2103         }
2104
2105         BTRFS_I(inode)->index = BTRFS_I(dir)->index_cnt;
2106         BTRFS_I(dir)->index_cnt++;
2107
2108         return ret;
2109 }
2110
2111 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2112                                      struct btrfs_root *root,
2113                                      struct inode *dir,
2114                                      const char *name, int name_len,
2115                                      u64 ref_objectid,
2116                                      u64 objectid,
2117                                      struct btrfs_block_group_cache *group,
2118                                      int mode)
2119 {
2120         struct inode *inode;
2121         struct btrfs_inode_item *inode_item;
2122         struct btrfs_block_group_cache *new_inode_group;
2123         struct btrfs_key *location;
2124         struct btrfs_path *path;
2125         struct btrfs_inode_ref *ref;
2126         struct btrfs_key key[2];
2127         u32 sizes[2];
2128         unsigned long ptr;
2129         int ret;
2130         int owner;
2131
2132         path = btrfs_alloc_path();
2133         BUG_ON(!path);
2134
2135         inode = new_inode(root->fs_info->sb);
2136         if (!inode)
2137                 return ERR_PTR(-ENOMEM);
2138
2139         if (dir) {
2140                 ret = btrfs_set_inode_index(dir, inode);
2141                 if (ret)
2142                         return ERR_PTR(ret);
2143         } else {
2144                 BTRFS_I(inode)->index = 0;
2145         }
2146         /*
2147          * index_cnt is ignored for everything but a dir,
2148          * btrfs_get_inode_index_count has an explanation for the magic
2149          * number
2150          */
2151         BTRFS_I(inode)->index_cnt = 2;
2152
2153         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2154         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2155                              inode->i_mapping, GFP_NOFS);
2156         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2157                              inode->i_mapping, GFP_NOFS);
2158         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2159         mutex_init(&BTRFS_I(inode)->csum_mutex);
2160         mutex_init(&BTRFS_I(inode)->extent_mutex);
2161         BTRFS_I(inode)->delalloc_bytes = 0;
2162         BTRFS_I(inode)->disk_i_size = 0;
2163         BTRFS_I(inode)->root = root;
2164
2165         if (mode & S_IFDIR)
2166                 owner = 0;
2167         else
2168                 owner = 1;
2169         new_inode_group = btrfs_find_block_group(root, group, 0,
2170                                        BTRFS_BLOCK_GROUP_METADATA, owner);
2171         if (!new_inode_group) {
2172                 printk("find_block group failed\n");
2173                 new_inode_group = group;
2174         }
2175         BTRFS_I(inode)->block_group = new_inode_group;
2176         BTRFS_I(inode)->flags = 0;
2177
2178         key[0].objectid = objectid;
2179         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2180         key[0].offset = 0;
2181
2182         key[1].objectid = objectid;
2183         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2184         key[1].offset = ref_objectid;
2185
2186         sizes[0] = sizeof(struct btrfs_inode_item);
2187         sizes[1] = name_len + sizeof(*ref);
2188
2189         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2190         if (ret != 0)
2191                 goto fail;
2192
2193         if (objectid > root->highest_inode)
2194                 root->highest_inode = objectid;
2195
2196         inode->i_uid = current->fsuid;
2197         inode->i_gid = current->fsgid;
2198         inode->i_mode = mode;
2199         inode->i_ino = objectid;
2200         inode->i_blocks = 0;
2201         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2202         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2203                                   struct btrfs_inode_item);
2204         fill_inode_item(path->nodes[0], inode_item, inode);
2205
2206         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2207                              struct btrfs_inode_ref);
2208         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2209         btrfs_set_inode_ref_index(path->nodes[0], ref, BTRFS_I(inode)->index);
2210         ptr = (unsigned long)(ref + 1);
2211         write_extent_buffer(path->nodes[0], name, ptr, name_len);
2212
2213         btrfs_mark_buffer_dirty(path->nodes[0]);
2214         btrfs_free_path(path);
2215
2216         location = &BTRFS_I(inode)->location;
2217         location->objectid = objectid;
2218         location->offset = 0;
2219         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2220
2221         insert_inode_hash(inode);
2222         return inode;
2223 fail:
2224         if (dir)
2225                 BTRFS_I(dir)->index_cnt--;
2226         btrfs_free_path(path);
2227         return ERR_PTR(ret);
2228 }
2229
2230 static inline u8 btrfs_inode_type(struct inode *inode)
2231 {
2232         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2233 }
2234
2235 static int btrfs_add_link(struct btrfs_trans_handle *trans,
2236                             struct dentry *dentry, struct inode *inode,
2237                             int add_backref)
2238 {
2239         int ret;
2240         struct btrfs_key key;
2241         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
2242         struct inode *parent_inode = dentry->d_parent->d_inode;
2243
2244         key.objectid = inode->i_ino;
2245         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2246         key.offset = 0;
2247
2248         ret = btrfs_insert_dir_item(trans, root,
2249                                     dentry->d_name.name, dentry->d_name.len,
2250                                     dentry->d_parent->d_inode->i_ino,
2251                                     &key, btrfs_inode_type(inode),
2252                                     BTRFS_I(inode)->index);
2253         if (ret == 0) {
2254                 if (add_backref) {
2255                         ret = btrfs_insert_inode_ref(trans, root,
2256                                              dentry->d_name.name,
2257                                              dentry->d_name.len,
2258                                              inode->i_ino,
2259                                              parent_inode->i_ino,
2260                                              BTRFS_I(inode)->index);
2261                 }
2262                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2263                                    dentry->d_name.len * 2);
2264                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2265                 ret = btrfs_update_inode(trans, root,
2266                                          dentry->d_parent->d_inode);
2267         }
2268         return ret;
2269 }
2270
2271 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2272                             struct dentry *dentry, struct inode *inode,
2273                             int backref)
2274 {
2275         int err = btrfs_add_link(trans, dentry, inode, backref);
2276         if (!err) {
2277                 d_instantiate(dentry, inode);
2278                 return 0;
2279         }
2280         if (err > 0)
2281                 err = -EEXIST;
2282         return err;
2283 }
2284
2285 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2286                         int mode, dev_t rdev)
2287 {
2288         struct btrfs_trans_handle *trans;
2289         struct btrfs_root *root = BTRFS_I(dir)->root;
2290         struct inode *inode = NULL;
2291         int err;
2292         int drop_inode = 0;
2293         u64 objectid;
2294         unsigned long nr = 0;
2295
2296         if (!new_valid_dev(rdev))
2297                 return -EINVAL;
2298
2299         err = btrfs_check_free_space(root, 1, 0);
2300         if (err)
2301                 goto fail;
2302
2303         trans = btrfs_start_transaction(root, 1);
2304         btrfs_set_trans_block_group(trans, dir);
2305
2306         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2307         if (err) {
2308                 err = -ENOSPC;
2309                 goto out_unlock;
2310         }
2311
2312         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2313                                 dentry->d_name.len,
2314                                 dentry->d_parent->d_inode->i_ino, objectid,
2315                                 BTRFS_I(dir)->block_group, mode);
2316         err = PTR_ERR(inode);
2317         if (IS_ERR(inode))
2318                 goto out_unlock;
2319
2320         err = btrfs_init_acl(inode, dir);
2321         if (err) {
2322                 drop_inode = 1;
2323                 goto out_unlock;
2324         }
2325
2326         btrfs_set_trans_block_group(trans, inode);
2327         err = btrfs_add_nondir(trans, dentry, inode, 0);
2328         if (err)
2329                 drop_inode = 1;
2330         else {
2331                 inode->i_op = &btrfs_special_inode_operations;
2332                 init_special_inode(inode, inode->i_mode, rdev);
2333                 btrfs_update_inode(trans, root, inode);
2334         }
2335         dir->i_sb->s_dirt = 1;
2336         btrfs_update_inode_block_group(trans, inode);
2337         btrfs_update_inode_block_group(trans, dir);
2338 out_unlock:
2339         nr = trans->blocks_used;
2340         btrfs_end_transaction_throttle(trans, root);
2341 fail:
2342         if (drop_inode) {
2343                 inode_dec_link_count(inode);
2344                 iput(inode);
2345         }
2346         btrfs_btree_balance_dirty(root, nr);
2347         return err;
2348 }
2349
2350 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2351                         int mode, struct nameidata *nd)
2352 {
2353         struct btrfs_trans_handle *trans;
2354         struct btrfs_root *root = BTRFS_I(dir)->root;
2355         struct inode *inode = NULL;
2356         int err;
2357         int drop_inode = 0;
2358         unsigned long nr = 0;
2359         u64 objectid;
2360
2361         err = btrfs_check_free_space(root, 1, 0);
2362         if (err)
2363                 goto fail;
2364         trans = btrfs_start_transaction(root, 1);
2365         btrfs_set_trans_block_group(trans, dir);
2366
2367         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2368         if (err) {
2369                 err = -ENOSPC;
2370                 goto out_unlock;
2371         }
2372
2373         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2374                                 dentry->d_name.len,
2375                                 dentry->d_parent->d_inode->i_ino,
2376                                 objectid, BTRFS_I(dir)->block_group, mode);
2377         err = PTR_ERR(inode);
2378         if (IS_ERR(inode))
2379                 goto out_unlock;
2380
2381         err = btrfs_init_acl(inode, dir);
2382         if (err) {
2383                 drop_inode = 1;
2384                 goto out_unlock;
2385         }
2386
2387         btrfs_set_trans_block_group(trans, inode);
2388         err = btrfs_add_nondir(trans, dentry, inode, 0);
2389         if (err)
2390                 drop_inode = 1;
2391         else {
2392                 inode->i_mapping->a_ops = &btrfs_aops;
2393                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2394                 inode->i_fop = &btrfs_file_operations;
2395                 inode->i_op = &btrfs_file_inode_operations;
2396                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2397                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2398                                      inode->i_mapping, GFP_NOFS);
2399                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2400                                      inode->i_mapping, GFP_NOFS);
2401                 mutex_init(&BTRFS_I(inode)->csum_mutex);
2402                 mutex_init(&BTRFS_I(inode)->extent_mutex);
2403                 BTRFS_I(inode)->delalloc_bytes = 0;
2404                 BTRFS_I(inode)->disk_i_size = 0;
2405                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2406                 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2407         }
2408         dir->i_sb->s_dirt = 1;
2409         btrfs_update_inode_block_group(trans, inode);
2410         btrfs_update_inode_block_group(trans, dir);
2411 out_unlock:
2412         nr = trans->blocks_used;
2413         btrfs_end_transaction_throttle(trans, root);
2414 fail:
2415         if (drop_inode) {
2416                 inode_dec_link_count(inode);
2417                 iput(inode);
2418         }
2419         btrfs_btree_balance_dirty(root, nr);
2420         return err;
2421 }
2422
2423 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2424                       struct dentry *dentry)
2425 {
2426         struct btrfs_trans_handle *trans;
2427         struct btrfs_root *root = BTRFS_I(dir)->root;
2428         struct inode *inode = old_dentry->d_inode;
2429         unsigned long nr = 0;
2430         int err;
2431         int drop_inode = 0;
2432
2433         if (inode->i_nlink == 0)
2434                 return -ENOENT;
2435
2436 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2437         inode->i_nlink++;
2438 #else
2439         inc_nlink(inode);
2440 #endif
2441         err = btrfs_check_free_space(root, 1, 0);
2442         if (err)
2443                 goto fail;
2444         err = btrfs_set_inode_index(dir, inode);
2445         if (err)
2446                 goto fail;
2447
2448         trans = btrfs_start_transaction(root, 1);
2449
2450         btrfs_set_trans_block_group(trans, dir);
2451         atomic_inc(&inode->i_count);
2452
2453         err = btrfs_add_nondir(trans, dentry, inode, 1);
2454
2455         if (err)
2456                 drop_inode = 1;
2457
2458         dir->i_sb->s_dirt = 1;
2459         btrfs_update_inode_block_group(trans, dir);
2460         err = btrfs_update_inode(trans, root, inode);
2461
2462         if (err)
2463                 drop_inode = 1;
2464
2465         nr = trans->blocks_used;
2466         btrfs_end_transaction_throttle(trans, root);
2467 fail:
2468         if (drop_inode) {
2469                 inode_dec_link_count(inode);
2470                 iput(inode);
2471         }
2472         btrfs_btree_balance_dirty(root, nr);
2473         return err;
2474 }
2475
2476 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2477 {
2478         struct inode *inode = NULL;
2479         struct btrfs_trans_handle *trans;
2480         struct btrfs_root *root = BTRFS_I(dir)->root;
2481         int err = 0;
2482         int drop_on_err = 0;
2483         u64 objectid = 0;
2484         unsigned long nr = 1;
2485
2486         err = btrfs_check_free_space(root, 1, 0);
2487         if (err)
2488                 goto out_unlock;
2489
2490         trans = btrfs_start_transaction(root, 1);
2491         btrfs_set_trans_block_group(trans, dir);
2492
2493         if (IS_ERR(trans)) {
2494                 err = PTR_ERR(trans);
2495                 goto out_unlock;
2496         }
2497
2498         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2499         if (err) {
2500                 err = -ENOSPC;
2501                 goto out_unlock;
2502         }
2503
2504         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2505                                 dentry->d_name.len,
2506                                 dentry->d_parent->d_inode->i_ino, objectid,
2507                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
2508         if (IS_ERR(inode)) {
2509                 err = PTR_ERR(inode);
2510                 goto out_fail;
2511         }
2512
2513         drop_on_err = 1;
2514
2515         err = btrfs_init_acl(inode, dir);
2516         if (err)
2517                 goto out_fail;
2518
2519         inode->i_op = &btrfs_dir_inode_operations;
2520         inode->i_fop = &btrfs_dir_file_operations;
2521         btrfs_set_trans_block_group(trans, inode);
2522
2523         btrfs_i_size_write(inode, 0);
2524         err = btrfs_update_inode(trans, root, inode);
2525         if (err)
2526                 goto out_fail;
2527
2528         err = btrfs_add_link(trans, dentry, inode, 0);
2529         if (err)
2530                 goto out_fail;
2531
2532         d_instantiate(dentry, inode);
2533         drop_on_err = 0;
2534         dir->i_sb->s_dirt = 1;
2535         btrfs_update_inode_block_group(trans, inode);
2536         btrfs_update_inode_block_group(trans, dir);
2537
2538 out_fail:
2539         nr = trans->blocks_used;
2540         btrfs_end_transaction_throttle(trans, root);
2541
2542 out_unlock:
2543         if (drop_on_err)
2544                 iput(inode);
2545         btrfs_btree_balance_dirty(root, nr);
2546         return err;
2547 }
2548
2549 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2550                                 struct extent_map *existing,
2551                                 struct extent_map *em,
2552                                 u64 map_start, u64 map_len)
2553 {
2554         u64 start_diff;
2555
2556         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2557         start_diff = map_start - em->start;
2558         em->start = map_start;
2559         em->len = map_len;
2560         if (em->block_start < EXTENT_MAP_LAST_BYTE)
2561                 em->block_start += start_diff;
2562         return add_extent_mapping(em_tree, em);
2563 }
2564
2565 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2566                                     size_t pg_offset, u64 start, u64 len,
2567                                     int create)
2568 {
2569         int ret;
2570         int err = 0;
2571         u64 bytenr;
2572         u64 extent_start = 0;
2573         u64 extent_end = 0;
2574         u64 objectid = inode->i_ino;
2575         u32 found_type;
2576         struct btrfs_path *path = NULL;
2577         struct btrfs_root *root = BTRFS_I(inode)->root;
2578         struct btrfs_file_extent_item *item;
2579         struct extent_buffer *leaf;
2580         struct btrfs_key found_key;
2581         struct extent_map *em = NULL;
2582         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2583         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2584         struct btrfs_trans_handle *trans = NULL;
2585
2586 again:
2587         spin_lock(&em_tree->lock);
2588         em = lookup_extent_mapping(em_tree, start, len);
2589         if (em)
2590                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2591         spin_unlock(&em_tree->lock);
2592
2593         if (em) {
2594                 if (em->start > start || em->start + em->len <= start)
2595                         free_extent_map(em);
2596                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2597                         free_extent_map(em);
2598                 else
2599                         goto out;
2600         }
2601         em = alloc_extent_map(GFP_NOFS);
2602         if (!em) {
2603                 err = -ENOMEM;
2604                 goto out;
2605         }
2606         em->bdev = root->fs_info->fs_devices->latest_bdev;
2607         em->start = EXTENT_MAP_HOLE;
2608         em->len = (u64)-1;
2609
2610         if (!path) {
2611                 path = btrfs_alloc_path();
2612                 BUG_ON(!path);
2613         }
2614
2615         ret = btrfs_lookup_file_extent(trans, root, path,
2616                                        objectid, start, trans != NULL);
2617         if (ret < 0) {
2618                 err = ret;
2619                 goto out;
2620         }
2621
2622         if (ret != 0) {
2623                 if (path->slots[0] == 0)
2624                         goto not_found;
2625                 path->slots[0]--;
2626         }
2627
2628         leaf = path->nodes[0];
2629         item = btrfs_item_ptr(leaf, path->slots[0],
2630                               struct btrfs_file_extent_item);
2631         /* are we inside the extent that was found? */
2632         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2633         found_type = btrfs_key_type(&found_key);
2634         if (found_key.objectid != objectid ||
2635             found_type != BTRFS_EXTENT_DATA_KEY) {
2636                 goto not_found;
2637         }
2638
2639         found_type = btrfs_file_extent_type(leaf, item);
2640         extent_start = found_key.offset;
2641         if (found_type == BTRFS_FILE_EXTENT_REG) {
2642                 extent_end = extent_start +
2643                        btrfs_file_extent_num_bytes(leaf, item);
2644                 err = 0;
2645                 if (start < extent_start || start >= extent_end) {
2646                         em->start = start;
2647                         if (start < extent_start) {
2648                                 if (start + len <= extent_start)
2649                                         goto not_found;
2650                                 em->len = extent_end - extent_start;
2651                         } else {
2652                                 em->len = len;
2653                         }
2654                         goto not_found_em;
2655                 }
2656                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2657                 if (bytenr == 0) {
2658                         em->start = extent_start;
2659                         em->len = extent_end - extent_start;
2660                         em->block_start = EXTENT_MAP_HOLE;
2661                         goto insert;
2662                 }
2663                 bytenr += btrfs_file_extent_offset(leaf, item);
2664                 em->block_start = bytenr;
2665                 em->start = extent_start;
2666                 em->len = extent_end - extent_start;
2667                 goto insert;
2668         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2669                 u64 page_start;
2670                 unsigned long ptr;
2671                 char *map;
2672                 size_t size;
2673                 size_t extent_offset;
2674                 size_t copy_size;
2675
2676                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2677                                                     path->slots[0]));
2678                 extent_end = (extent_start + size + root->sectorsize - 1) &
2679                         ~((u64)root->sectorsize - 1);
2680                 if (start < extent_start || start >= extent_end) {
2681                         em->start = start;
2682                         if (start < extent_start) {
2683                                 if (start + len <= extent_start)
2684                                         goto not_found;
2685                                 em->len = extent_end - extent_start;
2686                         } else {
2687                                 em->len = len;
2688                         }
2689                         goto not_found_em;
2690                 }
2691                 em->block_start = EXTENT_MAP_INLINE;
2692
2693                 if (!page) {
2694                         em->start = extent_start;
2695                         em->len = size;
2696                         goto out;
2697                 }
2698
2699                 page_start = page_offset(page) + pg_offset;
2700                 extent_offset = page_start - extent_start;
2701                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2702                                 size - extent_offset);
2703                 em->start = extent_start + extent_offset;
2704                 em->len = (copy_size + root->sectorsize - 1) &
2705                         ~((u64)root->sectorsize - 1);
2706                 map = kmap(page);
2707                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2708                 if (create == 0 && !PageUptodate(page)) {
2709                         read_extent_buffer(leaf, map + pg_offset, ptr,
2710                                            copy_size);
2711                         flush_dcache_page(page);
2712                 } else if (create && PageUptodate(page)) {
2713                         if (!trans) {
2714                                 kunmap(page);
2715                                 free_extent_map(em);
2716                                 em = NULL;
2717                                 btrfs_release_path(root, path);
2718                                 trans = btrfs_join_transaction(root, 1);
2719                                 goto again;
2720                         }
2721                         write_extent_buffer(leaf, map + pg_offset, ptr,
2722                                             copy_size);
2723                         btrfs_mark_buffer_dirty(leaf);
2724                 }
2725                 kunmap(page);
2726                 set_extent_uptodate(io_tree, em->start,
2727                                     extent_map_end(em) - 1, GFP_NOFS);
2728                 goto insert;
2729         } else {
2730                 printk("unkknown found_type %d\n", found_type);
2731                 WARN_ON(1);
2732         }
2733 not_found:
2734         em->start = start;
2735         em->len = len;
2736 not_found_em:
2737         em->block_start = EXTENT_MAP_HOLE;
2738 insert:
2739         btrfs_release_path(root, path);
2740         if (em->start > start || extent_map_end(em) <= start) {
2741                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2742                 err = -EIO;
2743                 goto out;
2744         }
2745
2746         err = 0;
2747         spin_lock(&em_tree->lock);
2748         ret = add_extent_mapping(em_tree, em);
2749         /* it is possible that someone inserted the extent into the tree
2750          * while we had the lock dropped.  It is also possible that
2751          * an overlapping map exists in the tree
2752          */
2753         if (ret == -EEXIST) {
2754                 struct extent_map *existing;
2755
2756                 ret = 0;
2757
2758                 existing = lookup_extent_mapping(em_tree, start, len);
2759                 if (existing && (existing->start > start ||
2760                     existing->start + existing->len <= start)) {
2761                         free_extent_map(existing);
2762                         existing = NULL;
2763                 }
2764                 if (!existing) {
2765                         existing = lookup_extent_mapping(em_tree, em->start,
2766                                                          em->len);
2767                         if (existing) {
2768                                 err = merge_extent_mapping(em_tree, existing,
2769                                                            em, start,
2770                                                            root->sectorsize);
2771                                 free_extent_map(existing);
2772                                 if (err) {
2773                                         free_extent_map(em);
2774                                         em = NULL;
2775                                 }
2776                         } else {
2777                                 err = -EIO;
2778                                 printk("failing to insert %Lu %Lu\n",
2779                                        start, len);
2780                                 free_extent_map(em);
2781                                 em = NULL;
2782                         }
2783                 } else {
2784                         free_extent_map(em);
2785                         em = existing;
2786                         err = 0;
2787                 }
2788         }
2789         spin_unlock(&em_tree->lock);
2790 out:
2791         if (path)
2792                 btrfs_free_path(path);
2793         if (trans) {
2794                 ret = btrfs_end_transaction(trans, root);
2795                 if (!err) {
2796                         err = ret;
2797                 }
2798         }
2799         if (err) {
2800                 free_extent_map(em);
2801                 WARN_ON(1);
2802                 return ERR_PTR(err);
2803         }
2804         return em;
2805 }
2806
2807 #if 0 /* waiting for O_DIRECT reads */
2808 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2809                         struct buffer_head *bh_result, int create)
2810 {
2811         struct extent_map *em;
2812         u64 start = (u64)iblock << inode->i_blkbits;
2813         struct btrfs_multi_bio *multi = NULL;
2814         struct btrfs_root *root = BTRFS_I(inode)->root;
2815         u64 len;
2816         u64 logical;
2817         u64 map_length;
2818         int ret = 0;
2819
2820         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2821
2822         if (!em || IS_ERR(em))
2823                 goto out;
2824
2825         if (em->start > start || em->start + em->len <= start) {
2826             goto out;
2827         }
2828
2829         if (em->block_start == EXTENT_MAP_INLINE) {
2830                 ret = -EINVAL;
2831                 goto out;
2832         }
2833
2834         len = em->start + em->len - start;
2835         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2836
2837         if (em->block_start == EXTENT_MAP_HOLE ||
2838             em->block_start == EXTENT_MAP_DELALLOC) {
2839                 bh_result->b_size = len;
2840                 goto out;
2841         }
2842
2843         logical = start - em->start;
2844         logical = em->block_start + logical;
2845
2846         map_length = len;
2847         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2848                               logical, &map_length, &multi, 0);
2849         BUG_ON(ret);
2850         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2851         bh_result->b_size = min(map_length, len);
2852
2853         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2854         set_buffer_mapped(bh_result);
2855         kfree(multi);
2856 out:
2857         free_extent_map(em);
2858         return ret;
2859 }
2860 #endif
2861
2862 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2863                         const struct iovec *iov, loff_t offset,
2864                         unsigned long nr_segs)
2865 {
2866         return -EINVAL;
2867 #if 0
2868         struct file *file = iocb->ki_filp;
2869         struct inode *inode = file->f_mapping->host;
2870
2871         if (rw == WRITE)
2872                 return -EINVAL;
2873
2874         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2875                                   offset, nr_segs, btrfs_get_block, NULL);
2876 #endif
2877 }
2878
2879 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2880 {
2881         return extent_bmap(mapping, iblock, btrfs_get_extent);
2882 }
2883
2884 int btrfs_readpage(struct file *file, struct page *page)
2885 {
2886         struct extent_io_tree *tree;
2887         tree = &BTRFS_I(page->mapping->host)->io_tree;
2888         return extent_read_full_page(tree, page, btrfs_get_extent);
2889 }
2890
2891 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2892 {
2893         struct extent_io_tree *tree;
2894
2895
2896         if (current->flags & PF_MEMALLOC) {
2897                 redirty_page_for_writepage(wbc, page);
2898                 unlock_page(page);
2899                 return 0;
2900         }
2901         tree = &BTRFS_I(page->mapping->host)->io_tree;
2902         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2903 }
2904
2905 int btrfs_writepages(struct address_space *mapping,
2906                      struct writeback_control *wbc)
2907 {
2908         struct extent_io_tree *tree;
2909         tree = &BTRFS_I(mapping->host)->io_tree;
2910         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2911 }
2912
2913 static int
2914 btrfs_readpages(struct file *file, struct address_space *mapping,
2915                 struct list_head *pages, unsigned nr_pages)
2916 {
2917         struct extent_io_tree *tree;
2918         tree = &BTRFS_I(mapping->host)->io_tree;
2919         return extent_readpages(tree, mapping, pages, nr_pages,
2920                                 btrfs_get_extent);
2921 }
2922 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2923 {
2924         struct extent_io_tree *tree;
2925         struct extent_map_tree *map;
2926         int ret;
2927
2928         tree = &BTRFS_I(page->mapping->host)->io_tree;
2929         map = &BTRFS_I(page->mapping->host)->extent_tree;
2930         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2931         if (ret == 1) {
2932                 ClearPagePrivate(page);
2933                 set_page_private(page, 0);
2934                 page_cache_release(page);
2935         }
2936         return ret;
2937 }
2938
2939 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2940 {
2941         return __btrfs_releasepage(page, gfp_flags);
2942 }
2943
2944 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2945 {
2946         struct extent_io_tree *tree;
2947         struct btrfs_ordered_extent *ordered;
2948         u64 page_start = page_offset(page);
2949         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
2950
2951         wait_on_page_writeback(page);
2952         tree = &BTRFS_I(page->mapping->host)->io_tree;
2953         if (offset) {
2954                 btrfs_releasepage(page, GFP_NOFS);
2955                 return;
2956         }
2957
2958         lock_extent(tree, page_start, page_end, GFP_NOFS);
2959         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
2960                                            page_offset(page));
2961         if (ordered) {
2962                 /*
2963                  * IO on this page will never be started, so we need
2964                  * to account for any ordered extents now
2965                  */
2966                 clear_extent_bit(tree, page_start, page_end,
2967                                  EXTENT_DIRTY | EXTENT_DELALLOC |
2968                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
2969                 btrfs_finish_ordered_io(page->mapping->host,
2970                                         page_start, page_end);
2971                 btrfs_put_ordered_extent(ordered);
2972                 lock_extent(tree, page_start, page_end, GFP_NOFS);
2973         }
2974         clear_extent_bit(tree, page_start, page_end,
2975                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
2976                  EXTENT_ORDERED,
2977                  1, 1, GFP_NOFS);
2978         __btrfs_releasepage(page, GFP_NOFS);
2979
2980         ClearPageChecked(page);
2981         if (PagePrivate(page)) {
2982                 ClearPagePrivate(page);
2983                 set_page_private(page, 0);
2984                 page_cache_release(page);
2985         }
2986 }
2987
2988 /*
2989  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2990  * called from a page fault handler when a page is first dirtied. Hence we must
2991  * be careful to check for EOF conditions here. We set the page up correctly
2992  * for a written page which means we get ENOSPC checking when writing into
2993  * holes and correct delalloc and unwritten extent mapping on filesystems that
2994  * support these features.
2995  *
2996  * We are not allowed to take the i_mutex here so we have to play games to
2997  * protect against truncate races as the page could now be beyond EOF.  Because
2998  * vmtruncate() writes the inode size before removing pages, once we have the
2999  * page lock we can determine safely if the page is beyond EOF. If it is not
3000  * beyond EOF, then the page is guaranteed safe against truncation until we
3001  * unlock the page.
3002  */
3003 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3004 {
3005         struct inode *inode = fdentry(vma->vm_file)->d_inode;
3006         struct btrfs_root *root = BTRFS_I(inode)->root;
3007         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3008         struct btrfs_ordered_extent *ordered;
3009         char *kaddr;
3010         unsigned long zero_start;
3011         loff_t size;
3012         int ret;
3013         u64 page_start;
3014         u64 page_end;
3015
3016         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3017         if (ret)
3018                 goto out;
3019
3020         ret = -EINVAL;
3021 again:
3022         lock_page(page);
3023         size = i_size_read(inode);
3024         page_start = page_offset(page);
3025         page_end = page_start + PAGE_CACHE_SIZE - 1;
3026
3027         if ((page->mapping != inode->i_mapping) ||
3028             (page_start >= size)) {
3029                 /* page got truncated out from underneath us */
3030                 goto out_unlock;
3031         }
3032         wait_on_page_writeback(page);
3033
3034         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3035         set_page_extent_mapped(page);
3036
3037         /*
3038          * we can't set the delalloc bits if there are pending ordered
3039          * extents.  Drop our locks and wait for them to finish
3040          */
3041         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3042         if (ordered) {
3043                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3044                 unlock_page(page);
3045                 btrfs_start_ordered_extent(inode, ordered, 1);
3046                 btrfs_put_ordered_extent(ordered);
3047                 goto again;
3048         }
3049
3050         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
3051                             page_end, GFP_NOFS);
3052         ret = 0;
3053
3054         /* page is wholly or partially inside EOF */
3055         if (page_start + PAGE_CACHE_SIZE > size)
3056                 zero_start = size & ~PAGE_CACHE_MASK;
3057         else
3058                 zero_start = PAGE_CACHE_SIZE;
3059
3060         if (zero_start != PAGE_CACHE_SIZE) {
3061                 kaddr = kmap(page);
3062                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3063                 flush_dcache_page(page);
3064                 kunmap(page);
3065         }
3066         ClearPageChecked(page);
3067         set_page_dirty(page);
3068         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3069
3070 out_unlock:
3071         unlock_page(page);
3072 out:
3073         return ret;
3074 }
3075
3076 static void btrfs_truncate(struct inode *inode)
3077 {
3078         struct btrfs_root *root = BTRFS_I(inode)->root;
3079         int ret;
3080         struct btrfs_trans_handle *trans;
3081         unsigned long nr;
3082         u64 mask = root->sectorsize - 1;
3083
3084         if (!S_ISREG(inode->i_mode))
3085                 return;
3086         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3087                 return;
3088
3089         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3090         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3091
3092         trans = btrfs_start_transaction(root, 1);
3093         btrfs_set_trans_block_group(trans, inode);
3094         btrfs_i_size_write(inode, inode->i_size);
3095
3096         ret = btrfs_orphan_add(trans, inode);
3097         if (ret)
3098                 goto out;
3099         /* FIXME, add redo link to tree so we don't leak on crash */
3100         ret = btrfs_truncate_in_trans(trans, root, inode,
3101                                       BTRFS_EXTENT_DATA_KEY);
3102         btrfs_update_inode(trans, root, inode);
3103
3104         ret = btrfs_orphan_del(trans, inode);
3105         BUG_ON(ret);
3106
3107 out:
3108         nr = trans->blocks_used;
3109         ret = btrfs_end_transaction_throttle(trans, root);
3110         BUG_ON(ret);
3111         btrfs_btree_balance_dirty(root, nr);
3112 }
3113
3114 /*
3115  * Invalidate a single dcache entry at the root of the filesystem.
3116  * Needed after creation of snapshot or subvolume.
3117  */
3118 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3119                                   int namelen)
3120 {
3121         struct dentry *alias, *entry;
3122         struct qstr qstr;
3123
3124         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3125         if (alias) {
3126                 qstr.name = name;
3127                 qstr.len = namelen;
3128                 /* change me if btrfs ever gets a d_hash operation */
3129                 qstr.hash = full_name_hash(qstr.name, qstr.len);
3130                 entry = d_lookup(alias, &qstr);
3131                 dput(alias);
3132                 if (entry) {
3133                         d_invalidate(entry);
3134                         dput(entry);
3135                 }
3136         }
3137 }
3138
3139 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3140                 struct btrfs_trans_handle *trans, u64 new_dirid,
3141                 struct btrfs_block_group_cache *block_group)
3142 {
3143         struct inode *inode;
3144
3145         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3146                                 new_dirid, block_group, S_IFDIR | 0700);
3147         if (IS_ERR(inode))
3148                 return PTR_ERR(inode);
3149         inode->i_op = &btrfs_dir_inode_operations;
3150         inode->i_fop = &btrfs_dir_file_operations;
3151         new_root->inode = inode;
3152
3153         inode->i_nlink = 1;
3154         btrfs_i_size_write(inode, 0);
3155
3156         return btrfs_update_inode(trans, new_root, inode);
3157 }
3158
3159 unsigned long btrfs_force_ra(struct address_space *mapping,
3160                               struct file_ra_state *ra, struct file *file,
3161                               pgoff_t offset, pgoff_t last_index)
3162 {
3163         pgoff_t req_size = last_index - offset + 1;
3164
3165 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3166         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3167         return offset;
3168 #else
3169         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3170         return offset + req_size;
3171 #endif
3172 }
3173
3174 struct inode *btrfs_alloc_inode(struct super_block *sb)
3175 {
3176         struct btrfs_inode *ei;
3177
3178         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3179         if (!ei)
3180                 return NULL;
3181         ei->last_trans = 0;
3182         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3183         ei->i_acl = BTRFS_ACL_NOT_CACHED;
3184         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3185         INIT_LIST_HEAD(&ei->i_orphan);
3186         return &ei->vfs_inode;
3187 }
3188
3189 void btrfs_destroy_inode(struct inode *inode)
3190 {
3191         struct btrfs_ordered_extent *ordered;
3192         WARN_ON(!list_empty(&inode->i_dentry));
3193         WARN_ON(inode->i_data.nrpages);
3194
3195         if (BTRFS_I(inode)->i_acl &&
3196             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3197                 posix_acl_release(BTRFS_I(inode)->i_acl);
3198         if (BTRFS_I(inode)->i_default_acl &&
3199             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3200                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3201
3202         spin_lock(&BTRFS_I(inode)->root->list_lock);
3203         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3204                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3205                        " list\n", inode->i_ino);
3206                 dump_stack();
3207         }
3208         spin_unlock(&BTRFS_I(inode)->root->list_lock);
3209
3210         while(1) {
3211                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3212                 if (!ordered)
3213                         break;
3214                 else {
3215                         printk("found ordered extent %Lu %Lu\n",
3216                                ordered->file_offset, ordered->len);
3217                         btrfs_remove_ordered_extent(inode, ordered);
3218                         btrfs_put_ordered_extent(ordered);
3219                         btrfs_put_ordered_extent(ordered);
3220                 }
3221         }
3222         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3223         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3224 }
3225
3226 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3227 static void init_once(void *foo)
3228 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3229 static void init_once(struct kmem_cache * cachep, void *foo)
3230 #else
3231 static void init_once(void * foo, struct kmem_cache * cachep,
3232                       unsigned long flags)
3233 #endif
3234 {
3235         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3236
3237         inode_init_once(&ei->vfs_inode);
3238 }
3239
3240 void btrfs_destroy_cachep(void)
3241 {
3242         if (btrfs_inode_cachep)
3243                 kmem_cache_destroy(btrfs_inode_cachep);
3244         if (btrfs_trans_handle_cachep)
3245                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3246         if (btrfs_transaction_cachep)
3247                 kmem_cache_destroy(btrfs_transaction_cachep);
3248         if (btrfs_bit_radix_cachep)
3249                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3250         if (btrfs_path_cachep)
3251                 kmem_cache_destroy(btrfs_path_cachep);
3252 }
3253
3254 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3255                                        unsigned long extra_flags,
3256 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3257                                        void (*ctor)(void *)
3258 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3259                                        void (*ctor)(struct kmem_cache *, void *)
3260 #else
3261                                        void (*ctor)(void *, struct kmem_cache *,
3262                                                     unsigned long)
3263 #endif
3264                                      )
3265 {
3266         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3267                                  SLAB_MEM_SPREAD | extra_flags), ctor
3268 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3269                                  ,NULL
3270 #endif
3271                                 );
3272 }
3273
3274 int btrfs_init_cachep(void)
3275 {
3276         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3277                                           sizeof(struct btrfs_inode),
3278                                           0, init_once);
3279         if (!btrfs_inode_cachep)
3280                 goto fail;
3281         btrfs_trans_handle_cachep =
3282                         btrfs_cache_create("btrfs_trans_handle_cache",
3283                                            sizeof(struct btrfs_trans_handle),
3284                                            0, NULL);
3285         if (!btrfs_trans_handle_cachep)
3286                 goto fail;
3287         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3288                                              sizeof(struct btrfs_transaction),
3289                                              0, NULL);
3290         if (!btrfs_transaction_cachep)
3291                 goto fail;
3292         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3293                                          sizeof(struct btrfs_path),
3294                                          0, NULL);
3295         if (!btrfs_path_cachep)
3296                 goto fail;
3297         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3298                                               SLAB_DESTROY_BY_RCU, NULL);
3299         if (!btrfs_bit_radix_cachep)
3300                 goto fail;
3301         return 0;
3302 fail:
3303         btrfs_destroy_cachep();
3304         return -ENOMEM;
3305 }
3306
3307 static int btrfs_getattr(struct vfsmount *mnt,
3308                          struct dentry *dentry, struct kstat *stat)
3309 {
3310         struct inode *inode = dentry->d_inode;
3311         generic_fillattr(inode, stat);
3312         stat->blksize = PAGE_CACHE_SIZE;
3313         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3314         return 0;
3315 }
3316
3317 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3318                            struct inode * new_dir,struct dentry *new_dentry)
3319 {
3320         struct btrfs_trans_handle *trans;
3321         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3322         struct inode *new_inode = new_dentry->d_inode;
3323         struct inode *old_inode = old_dentry->d_inode;
3324         struct timespec ctime = CURRENT_TIME;
3325         int ret;
3326
3327         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3328             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3329                 return -ENOTEMPTY;
3330         }
3331
3332         ret = btrfs_check_free_space(root, 1, 0);
3333         if (ret)
3334                 goto out_unlock;
3335
3336         trans = btrfs_start_transaction(root, 1);
3337
3338         btrfs_set_trans_block_group(trans, new_dir);
3339
3340         old_dentry->d_inode->i_nlink++;
3341         old_dir->i_ctime = old_dir->i_mtime = ctime;
3342         new_dir->i_ctime = new_dir->i_mtime = ctime;
3343         old_inode->i_ctime = ctime;
3344
3345         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3346         if (ret)
3347                 goto out_fail;
3348
3349         if (new_inode) {
3350                 new_inode->i_ctime = CURRENT_TIME;
3351                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3352                 if (ret)
3353                         goto out_fail;
3354                 if (new_inode->i_nlink == 0) {
3355                         ret = btrfs_orphan_add(trans, new_inode);
3356                         if (ret)
3357                                 goto out_fail;
3358                 }
3359         }
3360         ret = btrfs_set_inode_index(new_dir, old_inode);
3361         if (ret)
3362                 goto out_fail;
3363
3364         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
3365         if (ret)
3366                 goto out_fail;
3367
3368 out_fail:
3369         btrfs_end_transaction_throttle(trans, root);
3370 out_unlock:
3371         return ret;
3372 }
3373
3374 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3375                          const char *symname)
3376 {
3377         struct btrfs_trans_handle *trans;
3378         struct btrfs_root *root = BTRFS_I(dir)->root;
3379         struct btrfs_path *path;
3380         struct btrfs_key key;
3381         struct inode *inode = NULL;
3382         int err;
3383         int drop_inode = 0;
3384         u64 objectid;
3385         int name_len;
3386         int datasize;
3387         unsigned long ptr;
3388         struct btrfs_file_extent_item *ei;
3389         struct extent_buffer *leaf;
3390         unsigned long nr = 0;
3391
3392         name_len = strlen(symname) + 1;
3393         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3394                 return -ENAMETOOLONG;
3395
3396         err = btrfs_check_free_space(root, 1, 0);
3397         if (err)
3398                 goto out_fail;
3399
3400         trans = btrfs_start_transaction(root, 1);
3401         btrfs_set_trans_block_group(trans, dir);
3402
3403         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3404         if (err) {
3405                 err = -ENOSPC;
3406                 goto out_unlock;
3407         }
3408
3409         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3410                                 dentry->d_name.len,
3411                                 dentry->d_parent->d_inode->i_ino, objectid,
3412                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
3413         err = PTR_ERR(inode);
3414         if (IS_ERR(inode))
3415                 goto out_unlock;
3416
3417         err = btrfs_init_acl(inode, dir);
3418         if (err) {
3419                 drop_inode = 1;
3420                 goto out_unlock;
3421         }
3422
3423         btrfs_set_trans_block_group(trans, inode);
3424         err = btrfs_add_nondir(trans, dentry, inode, 0);
3425         if (err)
3426                 drop_inode = 1;
3427         else {
3428                 inode->i_mapping->a_ops = &btrfs_aops;
3429                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3430                 inode->i_fop = &btrfs_file_operations;
3431                 inode->i_op = &btrfs_file_inode_operations;
3432                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3433                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3434                                      inode->i_mapping, GFP_NOFS);
3435                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3436                                      inode->i_mapping, GFP_NOFS);
3437                 mutex_init(&BTRFS_I(inode)->csum_mutex);
3438                 mutex_init(&BTRFS_I(inode)->extent_mutex);
3439                 BTRFS_I(inode)->delalloc_bytes = 0;
3440                 BTRFS_I(inode)->disk_i_size = 0;
3441                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3442                 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3443         }
3444         dir->i_sb->s_dirt = 1;
3445         btrfs_update_inode_block_group(trans, inode);
3446         btrfs_update_inode_block_group(trans, dir);
3447         if (drop_inode)
3448                 goto out_unlock;
3449
3450         path = btrfs_alloc_path();
3451         BUG_ON(!path);
3452         key.objectid = inode->i_ino;
3453         key.offset = 0;
3454         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3455         datasize = btrfs_file_extent_calc_inline_size(name_len);
3456         err = btrfs_insert_empty_item(trans, root, path, &key,
3457                                       datasize);
3458         if (err) {
3459                 drop_inode = 1;
3460                 goto out_unlock;
3461         }
3462         leaf = path->nodes[0];
3463         ei = btrfs_item_ptr(leaf, path->slots[0],
3464                             struct btrfs_file_extent_item);
3465         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3466         btrfs_set_file_extent_type(leaf, ei,
3467                                    BTRFS_FILE_EXTENT_INLINE);
3468         ptr = btrfs_file_extent_inline_start(ei);
3469         write_extent_buffer(leaf, symname, ptr, name_len);
3470         btrfs_mark_buffer_dirty(leaf);
3471         btrfs_free_path(path);
3472
3473         inode->i_op = &btrfs_symlink_inode_operations;
3474         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3475         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3476         btrfs_i_size_write(inode, name_len - 1);
3477         err = btrfs_update_inode(trans, root, inode);
3478         if (err)
3479                 drop_inode = 1;
3480
3481 out_unlock:
3482         nr = trans->blocks_used;
3483         btrfs_end_transaction_throttle(trans, root);
3484 out_fail:
3485         if (drop_inode) {
3486                 inode_dec_link_count(inode);
3487                 iput(inode);
3488         }
3489         btrfs_btree_balance_dirty(root, nr);
3490         return err;
3491 }
3492
3493 static int btrfs_set_page_dirty(struct page *page)
3494 {
3495         return __set_page_dirty_nobuffers(page);
3496 }
3497
3498 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3499 static int btrfs_permission(struct inode *inode, int mask)
3500 #else
3501 static int btrfs_permission(struct inode *inode, int mask,
3502                             struct nameidata *nd)
3503 #endif
3504 {
3505         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3506                 return -EACCES;
3507         return generic_permission(inode, mask, btrfs_check_acl);
3508 }
3509
3510 static struct inode_operations btrfs_dir_inode_operations = {
3511         .lookup         = btrfs_lookup,
3512         .create         = btrfs_create,
3513         .unlink         = btrfs_unlink,
3514         .link           = btrfs_link,
3515         .mkdir          = btrfs_mkdir,
3516         .rmdir          = btrfs_rmdir,
3517         .rename         = btrfs_rename,
3518         .symlink        = btrfs_symlink,
3519         .setattr        = btrfs_setattr,
3520         .mknod          = btrfs_mknod,
3521         .setxattr       = generic_setxattr,
3522         .getxattr       = generic_getxattr,
3523         .listxattr      = btrfs_listxattr,
3524         .removexattr    = generic_removexattr,
3525         .permission     = btrfs_permission,
3526 };
3527 static struct inode_operations btrfs_dir_ro_inode_operations = {
3528         .lookup         = btrfs_lookup,
3529         .permission     = btrfs_permission,
3530 };
3531 static struct file_operations btrfs_dir_file_operations = {
3532         .llseek         = generic_file_llseek,
3533         .read           = generic_read_dir,
3534         .readdir        = btrfs_readdir,
3535         .unlocked_ioctl = btrfs_ioctl,
3536 #ifdef CONFIG_COMPAT
3537         .compat_ioctl   = btrfs_ioctl,
3538 #endif
3539         .release        = btrfs_release_file,
3540 };
3541
3542 static struct extent_io_ops btrfs_extent_io_ops = {
3543         .fill_delalloc = run_delalloc_range,
3544         .submit_bio_hook = btrfs_submit_bio_hook,
3545         .merge_bio_hook = btrfs_merge_bio_hook,
3546         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3547         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3548         .writepage_start_hook = btrfs_writepage_start_hook,
3549         .readpage_io_failed_hook = btrfs_io_failed_hook,
3550         .set_bit_hook = btrfs_set_bit_hook,
3551         .clear_bit_hook = btrfs_clear_bit_hook,
3552 };
3553
3554 static struct address_space_operations btrfs_aops = {
3555         .readpage       = btrfs_readpage,
3556         .writepage      = btrfs_writepage,
3557         .writepages     = btrfs_writepages,
3558         .readpages      = btrfs_readpages,
3559         .sync_page      = block_sync_page,
3560         .bmap           = btrfs_bmap,
3561         .direct_IO      = btrfs_direct_IO,
3562         .invalidatepage = btrfs_invalidatepage,
3563         .releasepage    = btrfs_releasepage,
3564         .set_page_dirty = btrfs_set_page_dirty,
3565 };
3566
3567 static struct address_space_operations btrfs_symlink_aops = {
3568         .readpage       = btrfs_readpage,
3569         .writepage      = btrfs_writepage,
3570         .invalidatepage = btrfs_invalidatepage,
3571         .releasepage    = btrfs_releasepage,
3572 };
3573
3574 static struct inode_operations btrfs_file_inode_operations = {
3575         .truncate       = btrfs_truncate,
3576         .getattr        = btrfs_getattr,
3577         .setattr        = btrfs_setattr,
3578         .setxattr       = generic_setxattr,
3579         .getxattr       = generic_getxattr,
3580         .listxattr      = btrfs_listxattr,
3581         .removexattr    = generic_removexattr,
3582         .permission     = btrfs_permission,
3583 };
3584 static struct inode_operations btrfs_special_inode_operations = {
3585         .getattr        = btrfs_getattr,
3586         .setattr        = btrfs_setattr,
3587         .permission     = btrfs_permission,
3588         .setxattr       = generic_setxattr,
3589         .getxattr       = generic_getxattr,
3590         .listxattr      = btrfs_listxattr,
3591         .removexattr    = generic_removexattr,
3592 };
3593 static struct inode_operations btrfs_symlink_inode_operations = {
3594         .readlink       = generic_readlink,
3595         .follow_link    = page_follow_link_light,
3596         .put_link       = page_put_link,
3597         .permission     = btrfs_permission,
3598 };