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

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/version.h>
  38 #include <linux/xattr.h>
  39 #include <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, 0);
 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 loops = 0;
 191         u64 total_fs_bytes;
 192         struct btrfs_root *root = BTRFS_I(inode)->root;
 193         struct btrfs_block_group_cache *block_group;
 194         struct btrfs_trans_handle *trans;
 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 = 0;
 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         trans = btrfs_join_transaction(root, 1);
 207         BUG_ON(!trans);
 208 again:
 209         ret = btrfs_lookup_file_extent(NULL, root, path,
 210                                        inode->i_ino, start, 0);
 211         if (ret < 0) {
 212                 err = ret;
 213                 goto out;
 214         }
 215
 216         if (ret != 0) {
 217                 if (path->slots[0] == 0)
 218                         goto not_found;
 219                 path->slots[0]--;
 220         }
 221
 222         leaf = path->nodes[0];
 223         item = btrfs_item_ptr(leaf, path->slots[0],
 224                               struct btrfs_file_extent_item);
 225
 226         /* are we inside the extent that was found? */
 227         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 228         found_type = btrfs_key_type(&found_key);
 229         if (found_key.objectid != inode->i_ino ||
 230             found_type != BTRFS_EXTENT_DATA_KEY)
 231                 goto not_found;
 232
 233         found_type = btrfs_file_extent_type(leaf, item);
 234         extent_start = found_key.offset;
 235         if (found_type == BTRFS_FILE_EXTENT_REG) {
 236                 u64 extent_num_bytes;
 237
 238                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 239                 extent_end = extent_start + extent_num_bytes;
 240                 err = 0;
 241
 242                 if (loops && start != extent_start)
 243                         goto not_found;
 244
 245                 if (start < extent_start || start >= extent_end)
 246                         goto not_found;
 247
 248                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 249                 if (bytenr == 0)
 250                         goto not_found;
 251
 252                 if (btrfs_cross_ref_exists(trans, 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                 bytenr += btrfs_file_extent_offset(leaf, item);
 264                 extent_num_bytes = min(end + 1, extent_end) - start;
 265                 ret = btrfs_add_ordered_extent(inode, start, bytenr,
 266                                                 extent_num_bytes, 1);
 267                 if (ret) {
 268                         err = ret;
 269                         goto out;
 270                 }
 271
 272                 btrfs_release_path(root, path);
 273                 start = extent_end;
 274                 if (start <= end) {
 275                         loops++;
 276                         goto again;
 277                 }
 278         } else {
 279 not_found:
 280                 btrfs_end_transaction(trans, root);
 281                 btrfs_free_path(path);
 282                 return cow_file_range(inode, start, end);
 283         }
 284 out:
 285         WARN_ON(err);
 286         btrfs_end_transaction(trans, root);
 287         btrfs_free_path(path);
 288         return err;
 289 }
 290
 291 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 292 {
 293         struct btrfs_root *root = BTRFS_I(inode)->root;
 294         int ret;
 295
 296         if (btrfs_test_opt(root, NODATACOW) ||
 297             btrfs_test_flag(inode, NODATACOW))
 298                 ret = run_delalloc_nocow(inode, start, end);
 299         else
 300                 ret = cow_file_range(inode, start, end);
 301
 302         return ret;
 303 }
 304
 305 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 306                        unsigned long old, unsigned long bits)
 307 {
 308         unsigned long flags;
 309         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 310                 struct btrfs_root *root = BTRFS_I(inode)->root;
 311                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 312                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 313                 root->fs_info->delalloc_bytes += end - start + 1;
 314                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 315                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
 316                                       &root->fs_info->delalloc_inodes);
 317                 }
 318                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 319         }
 320         return 0;
 321 }
 322
 323 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 324                          unsigned long old, unsigned long bits)
 325 {
 326         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 327                 struct btrfs_root *root = BTRFS_I(inode)->root;
 328                 unsigned long flags;
 329
 330                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 331                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 332                         printk("warning: delalloc account %Lu %Lu\n",
 333                                end - start + 1, root->fs_info->delalloc_bytes);
 334                         root->fs_info->delalloc_bytes = 0;
 335                         BTRFS_I(inode)->delalloc_bytes = 0;
 336                 } else {
 337                         root->fs_info->delalloc_bytes -= end - start + 1;
 338                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 339                 }
 340                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
 341                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 342                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
 343                 }
 344                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 345         }
 346         return 0;
 347 }
 348
 349 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 350                          size_t size, struct bio *bio)
 351 {
 352         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 353         struct btrfs_mapping_tree *map_tree;
 354         u64 logical = bio->bi_sector << 9;
 355         u64 length = 0;
 356         u64 map_length;
 357         int ret;
 358
 359         length = bio->bi_size;
 360         map_tree = &root->fs_info->mapping_tree;
 361         map_length = length;
 362         ret = btrfs_map_block(map_tree, READ, logical,
 363                               &map_length, NULL, 0);
 364
 365         if (map_length < length + size) {
 366                 return 1;
 367         }
 368         return 0;
 369 }
 370
 371 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 372                           int mirror_num)
 373 {
 374         struct btrfs_root *root = BTRFS_I(inode)->root;
 375         int ret = 0;
 376
 377         ret = btrfs_csum_one_bio(root, inode, bio);
 378         BUG_ON(ret);
 379
 380         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
 381 }
 382
 383 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 384                           int mirror_num)
 385 {
 386         struct btrfs_root *root = BTRFS_I(inode)->root;
 387         int ret = 0;
 388
 389         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 390         BUG_ON(ret);
 391
 392         if (btrfs_test_opt(root, NODATASUM) ||
 393             btrfs_test_flag(inode, NODATASUM)) {
 394                 goto mapit;
 395         }
 396
 397         if (!(rw & (1 << BIO_RW))) {
 398                 btrfs_lookup_bio_sums(root, inode, bio);
 399                 goto mapit;
 400         }
 401         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 402                                    inode, rw, bio, mirror_num,
 403                                    __btrfs_submit_bio_hook);
 404 mapit:
 405         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
 406 }
 407
 408 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
 409                              struct inode *inode, u64 file_offset,
 410                              struct list_head *list)
 411 {
 412         struct list_head *cur;
 413         struct btrfs_ordered_sum *sum;
 414
 415         btrfs_set_trans_block_group(trans, inode);
 416         list_for_each(cur, list) {
 417                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
 418                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
 419                                        inode, sum);
 420         }
 421         return 0;
 422 }
 423
 424 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
 425 {
 426         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
 427                                    GFP_NOFS);
 428 }
 429
 430 struct btrfs_writepage_fixup {
 431         struct page *page;
 432         struct btrfs_work work;
 433 };
 434
 435 /* see btrfs_writepage_start_hook for details on why this is required */
 436 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
 437 {
 438         struct btrfs_writepage_fixup *fixup;
 439         struct btrfs_ordered_extent *ordered;
 440         struct page *page;
 441         struct inode *inode;
 442         u64 page_start;
 443         u64 page_end;
 444
 445         fixup = container_of(work, struct btrfs_writepage_fixup, work);
 446         page = fixup->page;
 447 again:
 448         lock_page(page);
 449         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
 450                 ClearPageChecked(page);
 451                 goto out_page;
 452         }
 453
 454         inode = page->mapping->host;
 455         page_start = page_offset(page);
 456         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
 457
 458         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 459
 460         /* already ordered? We're done */
 461         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
 462                              EXTENT_ORDERED, 0)) {
 463                 goto out;
 464         }
 465
 466         ordered = btrfs_lookup_ordered_extent(inode, page_start);
 467         if (ordered) {
 468                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
 469                               page_end, GFP_NOFS);
 470                 unlock_page(page);
 471                 btrfs_start_ordered_extent(inode, ordered, 1);
 472                 goto again;
 473         }
 474
 475         btrfs_set_extent_delalloc(inode, page_start, page_end);
 476         ClearPageChecked(page);
 477 out:
 478         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 479 out_page:
 480         unlock_page(page);
 481         page_cache_release(page);
 482 }
 483
 484 /*
 485  * There are a few paths in the higher layers of the kernel that directly
 486  * set the page dirty bit without asking the filesystem if it is a
 487  * good idea.  This causes problems because we want to make sure COW
 488  * properly happens and the data=ordered rules are followed.
 489  *
 490  * In our case any range that doesn't have the EXTENT_ORDERED bit set
 491  * hasn't been properly setup for IO.  We kick off an async process
 492  * to fix it up.  The async helper will wait for ordered extents, set
 493  * the delalloc bit and make it safe to write the page.
 494  */
 495 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
 496 {
 497         struct inode *inode = page->mapping->host;
 498         struct btrfs_writepage_fixup *fixup;
 499         struct btrfs_root *root = BTRFS_I(inode)->root;
 500         int ret;
 501
 502         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
 503                              EXTENT_ORDERED, 0);
 504         if (ret)
 505                 return 0;
 506
 507         if (PageChecked(page))
 508                 return -EAGAIN;
 509
 510         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
 511         if (!fixup)
 512                 return -EAGAIN;
 513
 514         SetPageChecked(page);
 515         page_cache_get(page);
 516         fixup->work.func = btrfs_writepage_fixup_worker;
 517         fixup->page = page;
 518         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
 519         return -EAGAIN;
 520 }
 521
 522 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 523 {
 524         struct btrfs_root *root = BTRFS_I(inode)->root;
 525         struct btrfs_trans_handle *trans;
 526         struct btrfs_ordered_extent *ordered_extent;
 527         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 528         u64 alloc_hint = 0;
 529         struct list_head list;
 530         struct btrfs_key ins;
 531         int ret;
 532
 533         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
 534         if (!ret)
 535                 return 0;
 536
 537         trans = btrfs_join_transaction(root, 1);
 538
 539         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 540         BUG_ON(!ordered_extent);
 541         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
 542                 goto nocow;
 543
 544         lock_extent(io_tree, ordered_extent->file_offset,
 545                     ordered_extent->file_offset + ordered_extent->len - 1,
 546                     GFP_NOFS);
 547
 548         INIT_LIST_HEAD(&list);
 549
 550         ins.objectid = ordered_extent->start;
 551         ins.offset = ordered_extent->len;
 552         ins.type = BTRFS_EXTENT_ITEM_KEY;
 553
 554         ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
 555                                           trans->transid, inode->i_ino,
 556                                           ordered_extent->file_offset, &ins);
 557         BUG_ON(ret);
 558
 559         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 560
 561         ret = btrfs_drop_extents(trans, root, inode,
 562                                  ordered_extent->file_offset,
 563                                  ordered_extent->file_offset +
 564                                  ordered_extent->len,
 565                                  ordered_extent->file_offset, &alloc_hint);
 566         BUG_ON(ret);
 567         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 568                                        ordered_extent->file_offset,
 569                                        ordered_extent->start,
 570                                        ordered_extent->len,
 571                                        ordered_extent->len, 0);
 572         BUG_ON(ret);
 573
 574         btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 575                                 ordered_extent->file_offset +
 576                                 ordered_extent->len - 1);
 577         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 578
 579         inode->i_blocks += ordered_extent->len >> 9;
 580         unlock_extent(io_tree, ordered_extent->file_offset,
 581                     ordered_extent->file_offset + ordered_extent->len - 1,
 582                     GFP_NOFS);
 583 nocow:
 584         add_pending_csums(trans, inode, ordered_extent->file_offset,
 585                           &ordered_extent->list);
 586
 587         btrfs_ordered_update_i_size(inode, ordered_extent);
 588         btrfs_remove_ordered_extent(inode, ordered_extent);
 589
 590         /* once for us */
 591         btrfs_put_ordered_extent(ordered_extent);
 592         /* once for the tree */
 593         btrfs_put_ordered_extent(ordered_extent);
 594
 595         btrfs_update_inode(trans, root, inode);
 596         btrfs_end_transaction(trans, root);
 597         return 0;
 598 }
 599
 600 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 601                                 struct extent_state *state, int uptodate)
 602 {
 603         return btrfs_finish_ordered_io(page->mapping->host, start, end);
 604 }
 605
 606 struct io_failure_record {
 607         struct page *page;
 608         u64 start;
 609         u64 len;
 610         u64 logical;
 611         int last_mirror;
 612 };
 613
 614 int btrfs_io_failed_hook(struct bio *failed_bio,
 615                          struct page *page, u64 start, u64 end,
 616                          struct extent_state *state)
 617 {
 618         struct io_failure_record *failrec = NULL;
 619         u64 private;
 620         struct extent_map *em;
 621         struct inode *inode = page->mapping->host;
 622         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 623         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 624         struct bio *bio;
 625         int num_copies;
 626         int ret;
 627         int rw;
 628         u64 logical;
 629
 630         ret = get_state_private(failure_tree, start, &private);
 631         if (ret) {
 632                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 633                 if (!failrec)
 634                         return -ENOMEM;
 635                 failrec->start = start;
 636                 failrec->len = end - start + 1;
 637                 failrec->last_mirror = 0;
 638
 639                 spin_lock(&em_tree->lock);
 640                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 641                 if (em->start > start || em->start + em->len < start) {
 642                         free_extent_map(em);
 643                         em = NULL;
 644                 }
 645                 spin_unlock(&em_tree->lock);
 646
 647                 if (!em || IS_ERR(em)) {
 648                         kfree(failrec);
 649                         return -EIO;
 650                 }
 651                 logical = start - em->start;
 652                 logical = em->block_start + logical;
 653                 failrec->logical = logical;
 654                 free_extent_map(em);
 655                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 656                                 EXTENT_DIRTY, GFP_NOFS);
 657                 set_state_private(failure_tree, start,
 658                                  (u64)(unsigned long)failrec);
 659         } else {
 660                 failrec = (struct io_failure_record *)(unsigned long)private;
 661         }
 662         num_copies = btrfs_num_copies(
 663                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 664                               failrec->logical, failrec->len);
 665         failrec->last_mirror++;
 666         if (!state) {
 667                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 668                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 669                                                     failrec->start,
 670                                                     EXTENT_LOCKED);
 671                 if (state && state->start != failrec->start)
 672                         state = NULL;
 673                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 674         }
 675         if (!state || failrec->last_mirror > num_copies) {
 676                 set_state_private(failure_tree, failrec->start, 0);
 677                 clear_extent_bits(failure_tree, failrec->start,
 678                                   failrec->start + failrec->len - 1,
 679                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 680                 kfree(failrec);
 681                 return -EIO;
 682         }
 683         bio = bio_alloc(GFP_NOFS, 1);
 684         bio->bi_private = state;
 685         bio->bi_end_io = failed_bio->bi_end_io;
 686         bio->bi_sector = failrec->logical >> 9;
 687         bio->bi_bdev = failed_bio->bi_bdev;
 688         bio->bi_size = 0;
 689         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 690         if (failed_bio->bi_rw & (1 << BIO_RW))
 691                 rw = WRITE;
 692         else
 693                 rw = READ;
 694
 695         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 696                                                       failrec->last_mirror);
 697         return 0;
 698 }
 699
 700 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 701 {
 702         u64 private;
 703         u64 private_failure;
 704         struct io_failure_record *failure;
 705         int ret;
 706
 707         private = 0;
 708         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 709                              (u64)-1, 1, EXTENT_DIRTY)) {
 710                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 711                                         start, &private_failure);
 712                 if (ret == 0) {
 713                         failure = (struct io_failure_record *)(unsigned long)
 714                                    private_failure;
 715                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 716                                           failure->start, 0);
 717                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 718                                           failure->start,
 719                                           failure->start + failure->len - 1,
 720                                           EXTENT_DIRTY | EXTENT_LOCKED,
 721                                           GFP_NOFS);
 722                         kfree(failure);
 723                 }
 724         }
 725         return 0;
 726 }
 727
 728 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 729                                struct extent_state *state)
 730 {
 731         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 732         struct inode *inode = page->mapping->host;
 733         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 734         char *kaddr;
 735         u64 private = ~(u32)0;
 736         int ret;
 737         struct btrfs_root *root = BTRFS_I(inode)->root;
 738         u32 csum = ~(u32)0;
 739         unsigned long flags;
 740
 741         if (btrfs_test_opt(root, NODATASUM) ||
 742             btrfs_test_flag(inode, NODATASUM))
 743                 return 0;
 744         if (state && state->start == start) {
 745                 private = state->private;
 746                 ret = 0;
 747         } else {
 748                 ret = get_state_private(io_tree, start, &private);
 749         }
 750         local_irq_save(flags);
 751         kaddr = kmap_atomic(page, KM_IRQ0);
 752         if (ret) {
 753                 goto zeroit;
 754         }
 755         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 756         btrfs_csum_final(csum, (char *)&csum);
 757         if (csum != private) {
 758                 goto zeroit;
 759         }
 760         kunmap_atomic(kaddr, KM_IRQ0);
 761         local_irq_restore(flags);
 762
 763         /* if the io failure tree for this inode is non-empty,
 764          * check to see if we've recovered from a failed IO
 765          */
 766         btrfs_clean_io_failures(inode, start);
 767         return 0;
 768
 769 zeroit:
 770         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 771                page->mapping->host->i_ino, (unsigned long long)start, csum,
 772                private);
 773         memset(kaddr + offset, 1, end - start + 1);
 774         flush_dcache_page(page);
 775         kunmap_atomic(kaddr, KM_IRQ0);
 776         local_irq_restore(flags);
 777         if (private == 0)
 778                 return 0;
 779         return -EIO;
 780 }
 781
 782 /*
 783  * This creates an orphan entry for the given inode in case something goes
 784  * wrong in the middle of an unlink/truncate.
 785  */
 786 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 787 {
 788         struct btrfs_root *root = BTRFS_I(inode)->root;
 789         int ret = 0;
 790
 791         spin_lock(&root->list_lock);
 792
 793         /* already on the orphan list, we're good */
 794         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 795                 spin_unlock(&root->list_lock);
 796                 return 0;
 797         }
 798
 799         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 800
 801         spin_unlock(&root->list_lock);
 802
 803         /*
 804          * insert an orphan item to track this unlinked/truncated file
 805          */
 806         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
 807
 808         return ret;
 809 }
 810
 811 /*
 812  * We have done the truncate/delete so we can go ahead and remove the orphan
 813  * item for this particular inode.
 814  */
 815 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
 816 {
 817         struct btrfs_root *root = BTRFS_I(inode)->root;
 818         int ret = 0;
 819
 820         spin_lock(&root->list_lock);
 821
 822         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
 823                 spin_unlock(&root->list_lock);
 824                 return 0;
 825         }
 826
 827         list_del_init(&BTRFS_I(inode)->i_orphan);
 828         if (!trans) {
 829                 spin_unlock(&root->list_lock);
 830                 return 0;
 831         }
 832
 833         spin_unlock(&root->list_lock);
 834
 835         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
 836
 837         return ret;
 838 }
 839
 840 /*
 841  * this cleans up any orphans that may be left on the list from the last use
 842  * of this root.
 843  */
 844 void btrfs_orphan_cleanup(struct btrfs_root *root)
 845 {
 846         struct btrfs_path *path;
 847         struct extent_buffer *leaf;
 848         struct btrfs_item *item;
 849         struct btrfs_key key, found_key;
 850         struct btrfs_trans_handle *trans;
 851         struct inode *inode;
 852         int ret = 0, nr_unlink = 0, nr_truncate = 0;
 853
 854         /* don't do orphan cleanup if the fs is readonly. */
 855         if (root->inode->i_sb->s_flags & MS_RDONLY)
 856                 return;
 857
 858         path = btrfs_alloc_path();
 859         if (!path)
 860                 return;
 861         path->reada = -1;
 862
 863         key.objectid = BTRFS_ORPHAN_OBJECTID;
 864         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
 865         key.offset = (u64)-1;
 866
 867         trans = btrfs_start_transaction(root, 1);
 868         btrfs_set_trans_block_group(trans, root->inode);
 869
 870         while (1) {
 871                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 872                 if (ret < 0) {
 873                         printk(KERN_ERR "Error searching slot for orphan: %d"
 874                                "\n", ret);
 875                         break;
 876                 }
 877
 878                 /*
 879                  * if ret == 0 means we found what we were searching for, which
 880                  * is weird, but possible, so only screw with path if we didnt
 881                  * find the key and see if we have stuff that matches
 882                  */
 883                 if (ret > 0) {
 884                         if (path->slots[0] == 0)
 885                                 break;
 886                         path->slots[0]--;
 887                 }
 888
 889                 /* pull out the item */
 890                 leaf = path->nodes[0];
 891                 item = btrfs_item_nr(leaf, path->slots[0]);
 892                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 893
 894                 /* make sure the item matches what we want */
 895                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
 896                         break;
 897                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
 898                         break;
 899
 900                 /* release the path since we're done with it */
 901                 btrfs_release_path(root, path);
 902
 903                 /*
 904                  * this is where we are basically btrfs_lookup, without the
 905                  * crossing root thing.  we store the inode number in the
 906                  * offset of the orphan item.
 907                  */
 908                 inode = btrfs_iget_locked(root->inode->i_sb,
 909                                           found_key.offset, root);
 910                 if (!inode)
 911                         break;
 912
 913                 if (inode->i_state & I_NEW) {
 914                         BTRFS_I(inode)->root = root;
 915
 916                         /* have to set the location manually */
 917                         BTRFS_I(inode)->location.objectid = inode->i_ino;
 918                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
 919                         BTRFS_I(inode)->location.offset = 0;
 920
 921                         btrfs_read_locked_inode(inode);
 922                         unlock_new_inode(inode);
 923                 }
 924
 925                 /*
 926                  * add this inode to the orphan list so btrfs_orphan_del does
 927                  * the proper thing when we hit it
 928                  */
 929                 spin_lock(&root->list_lock);
 930                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 931                 spin_unlock(&root->list_lock);
 932
 933                 /*
 934                  * if this is a bad inode, means we actually succeeded in
 935                  * removing the inode, but not the orphan record, which means
 936                  * we need to manually delete the orphan since iput will just
 937                  * do a destroy_inode
 938                  */
 939                 if (is_bad_inode(inode)) {
 940                         btrfs_orphan_del(trans, inode);
 941                         iput(inode);
 942                         continue;
 943                 }
 944
 945                 /* if we have links, this was a truncate, lets do that */
 946                 if (inode->i_nlink) {
 947                         nr_truncate++;
 948                         btrfs_truncate(inode);
 949                 } else {
 950                         nr_unlink++;
 951                 }
 952
 953                 /* this will do delete_inode and everything for us */
 954                 iput(inode);
 955         }
 956
 957         if (nr_unlink)
 958                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
 959         if (nr_truncate)
 960                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
 961
 962         btrfs_free_path(path);
 963         btrfs_end_transaction(trans, root);
 964 }
 965
 966 void btrfs_read_locked_inode(struct inode *inode)
 967 {
 968         struct btrfs_path *path;
 969         struct extent_buffer *leaf;
 970         struct btrfs_inode_item *inode_item;
 971         struct btrfs_timespec *tspec;
 972         struct btrfs_root *root = BTRFS_I(inode)->root;
 973         struct btrfs_key location;
 974         u64 alloc_group_block;
 975         u32 rdev;
 976         int ret;
 977
 978         path = btrfs_alloc_path();
 979         BUG_ON(!path);
 980         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 981
 982         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 983         if (ret)
 984                 goto make_bad;
 985
 986         leaf = path->nodes[0];
 987         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 988                                     struct btrfs_inode_item);
 989
 990         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 991         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 992         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 993         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 994         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
 995
 996         tspec = btrfs_inode_atime(inode_item);
 997         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 998         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 999
1000         tspec = btrfs_inode_mtime(inode_item);
1001         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1002         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1003
1004         tspec = btrfs_inode_ctime(inode_item);
1005         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1006         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1007
1008         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1009         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
1010         inode->i_rdev = 0;
1011         rdev = btrfs_inode_rdev(leaf, inode_item);
1012
1013         BTRFS_I(inode)->index_cnt = (u64)-1;
1014
1015         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1016         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1017                                                        alloc_group_block);
1018         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1019         if (!BTRFS_I(inode)->block_group) {
1020                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1021                                                  NULL, 0,
1022                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1023         }
1024         btrfs_free_path(path);
1025         inode_item = NULL;
1026
1027         switch (inode->i_mode & S_IFMT) {
1028         case S_IFREG:
1029                 inode->i_mapping->a_ops = &btrfs_aops;
1030                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1031                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1032                 inode->i_fop = &btrfs_file_operations;
1033                 inode->i_op = &btrfs_file_inode_operations;
1034                 break;
1035         case S_IFDIR:
1036                 inode->i_fop = &btrfs_dir_file_operations;
1037                 if (root == root->fs_info->tree_root)
1038                         inode->i_op = &btrfs_dir_ro_inode_operations;
1039                 else
1040                         inode->i_op = &btrfs_dir_inode_operations;
1041                 break;
1042         case S_IFLNK:
1043                 inode->i_op = &btrfs_symlink_inode_operations;
1044                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1045                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1046                 break;
1047         default:
1048                 init_special_inode(inode, inode->i_mode, rdev);
1049                 break;
1050         }
1051         return;
1052
1053 make_bad:
1054         btrfs_free_path(path);
1055         make_bad_inode(inode);
1056 }
1057
1058 static void fill_inode_item(struct extent_buffer *leaf,
1059                             struct btrfs_inode_item *item,
1060                             struct inode *inode)
1061 {
1062         btrfs_set_inode_uid(leaf, item, inode->i_uid);
1063         btrfs_set_inode_gid(leaf, item, inode->i_gid);
1064         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1065         btrfs_set_inode_mode(leaf, item, inode->i_mode);
1066         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1067
1068         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1069                                inode->i_atime.tv_sec);
1070         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1071                                 inode->i_atime.tv_nsec);
1072
1073         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1074                                inode->i_mtime.tv_sec);
1075         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1076                                 inode->i_mtime.tv_nsec);
1077
1078         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1079                                inode->i_ctime.tv_sec);
1080         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1081                                 inode->i_ctime.tv_nsec);
1082
1083         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1084         btrfs_set_inode_generation(leaf, item, inode->i_generation);
1085         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1086         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1087         btrfs_set_inode_block_group(leaf, item,
1088                                     BTRFS_I(inode)->block_group->key.objectid);
1089 }
1090
1091 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1092                               struct btrfs_root *root,
1093                               struct inode *inode)
1094 {
1095         struct btrfs_inode_item *inode_item;
1096         struct btrfs_path *path;
1097         struct extent_buffer *leaf;
1098         int ret;
1099
1100         path = btrfs_alloc_path();
1101         BUG_ON(!path);
1102         ret = btrfs_lookup_inode(trans, root, path,
1103                                  &BTRFS_I(inode)->location, 1);
1104         if (ret) {
1105                 if (ret > 0)
1106                         ret = -ENOENT;
1107                 goto failed;
1108         }
1109
1110         leaf = path->nodes[0];
1111         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1112                                   struct btrfs_inode_item);
1113
1114         fill_inode_item(leaf, inode_item, inode);
1115         btrfs_mark_buffer_dirty(leaf);
1116         btrfs_set_inode_last_trans(trans, inode);
1117         ret = 0;
1118 failed:
1119         btrfs_free_path(path);
1120         return ret;
1121 }
1122
1123
1124 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
1125                               struct btrfs_root *root,
1126                               struct inode *dir,
1127                               struct dentry *dentry)
1128 {
1129         struct btrfs_path *path;
1130         const char *name = dentry->d_name.name;
1131         int name_len = dentry->d_name.len;
1132         int ret = 0;
1133         struct extent_buffer *leaf;
1134         struct btrfs_dir_item *di;
1135         struct btrfs_key key;
1136         u64 index;
1137
1138         path = btrfs_alloc_path();
1139         if (!path) {
1140                 ret = -ENOMEM;
1141                 goto err;
1142         }
1143
1144         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1145                                     name, name_len, -1);
1146         if (IS_ERR(di)) {
1147                 ret = PTR_ERR(di);
1148                 goto err;
1149         }
1150         if (!di) {
1151                 ret = -ENOENT;
1152                 goto err;
1153         }
1154         leaf = path->nodes[0];
1155         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1156         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1157         if (ret)
1158                 goto err;
1159         btrfs_release_path(root, path);
1160
1161         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1162                                   dentry->d_inode->i_ino,
1163                                   dentry->d_parent->d_inode->i_ino, &index);
1164         if (ret) {
1165                 printk("failed to delete reference to %.*s, "
1166                        "inode %lu parent %lu\n", name_len, name,
1167                        dentry->d_inode->i_ino,
1168                        dentry->d_parent->d_inode->i_ino);
1169                 goto err;
1170         }
1171
1172         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1173                                          index, name, name_len, -1);
1174         if (IS_ERR(di)) {
1175                 ret = PTR_ERR(di);
1176                 goto err;
1177         }
1178         if (!di) {
1179                 ret = -ENOENT;
1180                 goto err;
1181         }
1182         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1183         btrfs_release_path(root, path);
1184
1185         dentry->d_inode->i_ctime = dir->i_ctime;
1186 err:
1187         btrfs_free_path(path);
1188         if (!ret) {
1189                 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1190                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1191                 btrfs_update_inode(trans, root, dir);
1192 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1193                 dentry->d_inode->i_nlink--;
1194 #else
1195                 drop_nlink(dentry->d_inode);
1196 #endif
1197                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
1198                 dir->i_sb->s_dirt = 1;
1199         }
1200         return ret;
1201 }
1202
1203 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1204 {
1205         struct btrfs_root *root;
1206         struct btrfs_trans_handle *trans;
1207         struct inode *inode = dentry->d_inode;
1208         int ret;
1209         unsigned long nr = 0;
1210
1211         root = BTRFS_I(dir)->root;
1212
1213         ret = btrfs_check_free_space(root, 1, 1);
1214         if (ret)
1215                 goto fail;
1216
1217         trans = btrfs_start_transaction(root, 1);
1218
1219         btrfs_set_trans_block_group(trans, dir);
1220         ret = btrfs_unlink_trans(trans, root, dir, dentry);
1221
1222         if (inode->i_nlink == 0)
1223                 ret = btrfs_orphan_add(trans, inode);
1224
1225         nr = trans->blocks_used;
1226
1227         btrfs_end_transaction_throttle(trans, root);
1228 fail:
1229         btrfs_btree_balance_dirty(root, nr);
1230         return ret;
1231 }
1232
1233 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1234 {
1235         struct inode *inode = dentry->d_inode;
1236         int err = 0;
1237         int ret;
1238         struct btrfs_root *root = BTRFS_I(dir)->root;
1239         struct btrfs_trans_handle *trans;
1240         unsigned long nr = 0;
1241
1242         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1243                 return -ENOTEMPTY;
1244         }
1245
1246         ret = btrfs_check_free_space(root, 1, 1);
1247         if (ret)
1248                 goto fail;
1249
1250         trans = btrfs_start_transaction(root, 1);
1251         btrfs_set_trans_block_group(trans, dir);
1252
1253         err = btrfs_orphan_add(trans, inode);
1254         if (err)
1255                 goto fail_trans;
1256
1257         /* now the directory is empty */
1258         err = btrfs_unlink_trans(trans, root, dir, dentry);
1259         if (!err) {
1260                 btrfs_i_size_write(inode, 0);
1261         }
1262
1263 fail_trans:
1264         nr = trans->blocks_used;
1265         ret = btrfs_end_transaction_throttle(trans, root);
1266 fail:
1267         btrfs_btree_balance_dirty(root, nr);
1268
1269         if (ret && !err)
1270                 err = ret;
1271         return err;
1272 }
1273
1274 /*
1275  * this can truncate away extent items, csum items and directory items.
1276  * It starts at a high offset and removes keys until it can't find
1277  * any higher than i_size.
1278  *
1279  * csum items that cross the new i_size are truncated to the new size
1280  * as well.
1281  *
1282  * min_type is the minimum key type to truncate down to.  If set to 0, this
1283  * will kill all the items on this inode, including the INODE_ITEM_KEY.
1284  */
1285 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1286                                    struct btrfs_root *root,
1287                                    struct inode *inode,
1288                                    u32 min_type)
1289 {
1290         int ret;
1291         struct btrfs_path *path;
1292         struct btrfs_key key;
1293         struct btrfs_key found_key;
1294         u32 found_type;
1295         struct extent_buffer *leaf;
1296         struct btrfs_file_extent_item *fi;
1297         u64 extent_start = 0;
1298         u64 extent_num_bytes = 0;
1299         u64 item_end = 0;
1300         u64 root_gen = 0;
1301         u64 root_owner = 0;
1302         int found_extent;
1303         int del_item;
1304         int pending_del_nr = 0;
1305         int pending_del_slot = 0;
1306         int extent_type = -1;
1307         u64 mask = root->sectorsize - 1;
1308
1309         btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
1310         path = btrfs_alloc_path();
1311         path->reada = -1;
1312         BUG_ON(!path);
1313
1314         /* FIXME, add redo link to tree so we don't leak on crash */
1315         key.objectid = inode->i_ino;
1316         key.offset = (u64)-1;
1317         key.type = (u8)-1;
1318
1319         btrfs_init_path(path);
1320 search_again:
1321         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1322         if (ret < 0) {
1323                 goto error;
1324         }
1325         if (ret > 0) {
1326                 BUG_ON(path->slots[0] == 0);
1327                 path->slots[0]--;
1328         }
1329
1330         while(1) {
1331                 fi = NULL;
1332                 leaf = path->nodes[0];
1333                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1334                 found_type = btrfs_key_type(&found_key);
1335
1336                 if (found_key.objectid != inode->i_ino)
1337                         break;
1338
1339                 if (found_type < min_type)
1340                         break;
1341
1342                 item_end = found_key.offset;
1343                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1344                         fi = btrfs_item_ptr(leaf, path->slots[0],
1345                                             struct btrfs_file_extent_item);
1346                         extent_type = btrfs_file_extent_type(leaf, fi);
1347                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1348                                 item_end +=
1349                                     btrfs_file_extent_num_bytes(leaf, fi);
1350                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1351                                 struct btrfs_item *item = btrfs_item_nr(leaf,
1352                                                                 path->slots[0]);
1353                                 item_end += btrfs_file_extent_inline_len(leaf,
1354                                                                          item);
1355                         }
1356                         item_end--;
1357                 }
1358                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1359                         ret = btrfs_csum_truncate(trans, root, path,
1360                                                   inode->i_size);
1361                         BUG_ON(ret);
1362                 }
1363                 if (item_end < inode->i_size) {
1364                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1365                                 found_type = BTRFS_INODE_ITEM_KEY;
1366                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1367                                 found_type = BTRFS_CSUM_ITEM_KEY;
1368                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1369                                 found_type = BTRFS_XATTR_ITEM_KEY;
1370                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1371                                 found_type = BTRFS_INODE_REF_KEY;
1372                         } else if (found_type) {
1373                                 found_type--;
1374                         } else {
1375                                 break;
1376                         }
1377                         btrfs_set_key_type(&key, found_type);
1378                         goto next;
1379                 }
1380                 if (found_key.offset >= inode->i_size)
1381                         del_item = 1;
1382                 else
1383                         del_item = 0;
1384                 found_extent = 0;
1385
1386                 /* FIXME, shrink the extent if the ref count is only 1 */
1387                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1388                         goto delete;
1389
1390                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1391                         u64 num_dec;
1392                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1393                         if (!del_item) {
1394                                 u64 orig_num_bytes =
1395                                         btrfs_file_extent_num_bytes(leaf, fi);
1396                                 extent_num_bytes = inode->i_size -
1397                                         found_key.offset + root->sectorsize - 1;
1398                                 extent_num_bytes = extent_num_bytes &
1399                                         ~((u64)root->sectorsize - 1);
1400                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1401                                                          extent_num_bytes);
1402                                 num_dec = (orig_num_bytes -
1403                                            extent_num_bytes);
1404                                 if (extent_start != 0)
1405                                         dec_i_blocks(inode, num_dec);
1406                                 btrfs_mark_buffer_dirty(leaf);
1407                         } else {
1408                                 extent_num_bytes =
1409                                         btrfs_file_extent_disk_num_bytes(leaf,
1410                                                                          fi);
1411                                 /* FIXME blocksize != 4096 */
1412                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1413                                 if (extent_start != 0) {
1414                                         found_extent = 1;
1415                                         dec_i_blocks(inode, num_dec);
1416                                 }
1417                                 root_gen = btrfs_header_generation(leaf);
1418                                 root_owner = btrfs_header_owner(leaf);
1419                         }
1420                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1421                         if (!del_item) {
1422                                 u32 newsize = inode->i_size - found_key.offset;
1423                                 dec_i_blocks(inode, item_end + 1 -
1424                                             found_key.offset - newsize);
1425                                 newsize =
1426                                     btrfs_file_extent_calc_inline_size(newsize);
1427                                 ret = btrfs_truncate_item(trans, root, path,
1428                                                           newsize, 1);
1429                                 BUG_ON(ret);
1430                         } else {
1431                                 dec_i_blocks(inode, item_end + 1 -
1432                                              found_key.offset);
1433                         }
1434                 }
1435 delete:
1436                 if (del_item) {
1437                         if (!pending_del_nr) {
1438                                 /* no pending yet, add ourselves */
1439                                 pending_del_slot = path->slots[0];
1440                                 pending_del_nr = 1;
1441                         } else if (pending_del_nr &&
1442                                    path->slots[0] + 1 == pending_del_slot) {
1443                                 /* hop on the pending chunk */
1444                                 pending_del_nr++;
1445                                 pending_del_slot = path->slots[0];
1446                         } else {
1447                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1448                         }
1449                 } else {
1450                         break;
1451                 }
1452                 if (found_extent) {
1453                         ret = btrfs_free_extent(trans, root, extent_start,
1454                                                 extent_num_bytes,
1455                                                 root_owner,
1456                                                 root_gen, inode->i_ino,
1457                                                 found_key.offset, 0);
1458                         BUG_ON(ret);
1459                 }
1460 next:
1461                 if (path->slots[0] == 0) {
1462                         if (pending_del_nr)
1463                                 goto del_pending;
1464                         btrfs_release_path(root, path);
1465                         goto search_again;
1466                 }
1467
1468                 path->slots[0]--;
1469                 if (pending_del_nr &&
1470                     path->slots[0] + 1 != pending_del_slot) {
1471                         struct btrfs_key debug;
1472 del_pending:
1473                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1474                                               pending_del_slot);
1475                         ret = btrfs_del_items(trans, root, path,
1476                                               pending_del_slot,
1477                                               pending_del_nr);
1478                         BUG_ON(ret);
1479                         pending_del_nr = 0;
1480                         btrfs_release_path(root, path);
1481                         goto search_again;
1482                 }
1483         }
1484         ret = 0;
1485 error:
1486         if (pending_del_nr) {
1487                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1488                                       pending_del_nr);
1489         }
1490         btrfs_free_path(path);
1491         inode->i_sb->s_dirt = 1;
1492         return ret;
1493 }
1494
1495 /*
1496  * taken from block_truncate_page, but does cow as it zeros out
1497  * any bytes left in the last page in the file.
1498  */
1499 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1500 {
1501         struct inode *inode = mapping->host;
1502         struct btrfs_root *root = BTRFS_I(inode)->root;
1503         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1504         struct btrfs_ordered_extent *ordered;
1505         char *kaddr;
1506         u32 blocksize = root->sectorsize;
1507         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1508         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1509         struct page *page;
1510         int ret = 0;
1511         u64 page_start;
1512         u64 page_end;
1513
1514         if ((offset & (blocksize - 1)) == 0)
1515                 goto out;
1516
1517         ret = -ENOMEM;
1518 again:
1519         page = grab_cache_page(mapping, index);
1520         if (!page)
1521                 goto out;
1522
1523         page_start = page_offset(page);
1524         page_end = page_start + PAGE_CACHE_SIZE - 1;
1525
1526         if (!PageUptodate(page)) {
1527                 ret = btrfs_readpage(NULL, page);
1528                 lock_page(page);
1529                 if (page->mapping != mapping) {
1530                         unlock_page(page);
1531                         page_cache_release(page);
1532                         goto again;
1533                 }
1534                 if (!PageUptodate(page)) {
1535                         ret = -EIO;
1536                         goto out_unlock;
1537                 }
1538         }
1539         wait_on_page_writeback(page);
1540
1541         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1542         set_page_extent_mapped(page);
1543
1544         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1545         if (ordered) {
1546                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1547                 unlock_page(page);
1548                 page_cache_release(page);
1549                 btrfs_start_ordered_extent(inode, ordered, 1);
1550                 btrfs_put_ordered_extent(ordered);
1551                 goto again;
1552         }
1553
1554         btrfs_set_extent_delalloc(inode, page_start, page_end);
1555         ret = 0;
1556         if (offset != PAGE_CACHE_SIZE) {
1557                 kaddr = kmap(page);
1558                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1559                 flush_dcache_page(page);
1560                 kunmap(page);
1561         }
1562         ClearPageChecked(page);
1563         set_page_dirty(page);
1564         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1565
1566 out_unlock:
1567         unlock_page(page);
1568         page_cache_release(page);
1569 out:
1570         return ret;
1571 }
1572
1573 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1574 {
1575         struct inode *inode = dentry->d_inode;
1576         int err;
1577
1578         err = inode_change_ok(inode, attr);
1579         if (err)
1580                 return err;
1581
1582         if (S_ISREG(inode->i_mode) &&
1583             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1584                 struct btrfs_trans_handle *trans;
1585                 struct btrfs_root *root = BTRFS_I(inode)->root;
1586                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1587
1588                 u64 mask = root->sectorsize - 1;
1589                 u64 hole_start = (inode->i_size + mask) & ~mask;
1590                 u64 block_end = (attr->ia_size + mask) & ~mask;
1591                 u64 hole_size;
1592                 u64 alloc_hint = 0;
1593
1594                 if (attr->ia_size <= hole_start)
1595                         goto out;
1596
1597                 err = btrfs_check_free_space(root, 1, 0);
1598                 if (err)
1599                         goto fail;
1600
1601                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1602
1603                 hole_size = block_end - hole_start;
1604                 while(1) {
1605                         struct btrfs_ordered_extent *ordered;
1606                         btrfs_wait_ordered_range(inode, hole_start, hole_size);
1607
1608                         lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1609                         ordered = btrfs_lookup_ordered_extent(inode, hole_start);
1610                         if (ordered) {
1611                                 unlock_extent(io_tree, hole_start,
1612                                               block_end - 1, GFP_NOFS);
1613                                 btrfs_put_ordered_extent(ordered);
1614                         } else {
1615                                 break;
1616                         }
1617                 }
1618
1619                 trans = btrfs_start_transaction(root, 1);
1620                 btrfs_set_trans_block_group(trans, inode);
1621                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1622                 err = btrfs_drop_extents(trans, root, inode,
1623                                          hole_start, block_end, hole_start,
1624                                          &alloc_hint);
1625
1626                 if (alloc_hint != EXTENT_MAP_INLINE) {
1627                         err = btrfs_insert_file_extent(trans, root,
1628                                                        inode->i_ino,
1629                                                        hole_start, 0, 0,
1630                                                        hole_size, 0);
1631                         btrfs_drop_extent_cache(inode, hole_start,
1632                                                 (u64)-1);
1633                         btrfs_check_file(root, inode);
1634                 }
1635                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1636                 btrfs_end_transaction(trans, root);
1637                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1638                 if (err)
1639                         return err;
1640         }
1641 out:
1642         err = inode_setattr(inode, attr);
1643
1644         if (!err && ((attr->ia_valid & ATTR_MODE)))
1645                 err = btrfs_acl_chmod(inode);
1646 fail:
1647         return err;
1648 }
1649
1650 void btrfs_delete_inode(struct inode *inode)
1651 {
1652         struct btrfs_trans_handle *trans;
1653         struct btrfs_root *root = BTRFS_I(inode)->root;
1654         unsigned long nr;
1655         int ret;
1656
1657         truncate_inode_pages(&inode->i_data, 0);
1658         if (is_bad_inode(inode)) {
1659                 btrfs_orphan_del(NULL, inode);
1660                 goto no_delete;
1661         }
1662         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1663
1664         btrfs_i_size_write(inode, 0);
1665         trans = btrfs_start_transaction(root, 1);
1666
1667         btrfs_set_trans_block_group(trans, inode);
1668         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1669         if (ret) {
1670                 btrfs_orphan_del(NULL, inode);
1671                 goto no_delete_lock;
1672         }
1673
1674         btrfs_orphan_del(trans, inode);
1675
1676         nr = trans->blocks_used;
1677         clear_inode(inode);
1678
1679         btrfs_end_transaction(trans, root);
1680         btrfs_btree_balance_dirty(root, nr);
1681         return;
1682
1683 no_delete_lock:
1684         nr = trans->blocks_used;
1685         btrfs_end_transaction(trans, root);
1686         btrfs_btree_balance_dirty(root, nr);
1687 no_delete:
1688         clear_inode(inode);
1689 }
1690
1691 /*
1692  * this returns the key found in the dir entry in the location pointer.
1693  * If no dir entries were found, location->objectid is 0.
1694  */
1695 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1696                                struct btrfs_key *location)
1697 {
1698         const char *name = dentry->d_name.name;
1699         int namelen = dentry->d_name.len;
1700         struct btrfs_dir_item *di;
1701         struct btrfs_path *path;
1702         struct btrfs_root *root = BTRFS_I(dir)->root;
1703         int ret = 0;
1704
1705         path = btrfs_alloc_path();
1706         BUG_ON(!path);
1707
1708         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1709                                     namelen, 0);
1710         if (IS_ERR(di))
1711                 ret = PTR_ERR(di);
1712         if (!di || IS_ERR(di)) {
1713                 goto out_err;
1714         }
1715         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1716 out:
1717         btrfs_free_path(path);
1718         return ret;
1719 out_err:
1720         location->objectid = 0;
1721         goto out;
1722 }
1723
1724 /*
1725  * when we hit a tree root in a directory, the btrfs part of the inode
1726  * needs to be changed to reflect the root directory of the tree root.  This
1727  * is kind of like crossing a mount point.
1728  */
1729 static int fixup_tree_root_location(struct btrfs_root *root,
1730                              struct btrfs_key *location,
1731                              struct btrfs_root **sub_root,
1732                              struct dentry *dentry)
1733 {
1734         struct btrfs_root_item *ri;
1735
1736         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1737                 return 0;
1738         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1739                 return 0;
1740
1741         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1742                                         dentry->d_name.name,
1743                                         dentry->d_name.len);
1744         if (IS_ERR(*sub_root))
1745                 return PTR_ERR(*sub_root);
1746
1747         ri = &(*sub_root)->root_item;
1748         location->objectid = btrfs_root_dirid(ri);
1749         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1750         location->offset = 0;
1751
1752         return 0;
1753 }
1754
1755 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1756 {
1757         struct btrfs_iget_args *args = p;
1758         inode->i_ino = args->ino;
1759         BTRFS_I(inode)->root = args->root;
1760         BTRFS_I(inode)->delalloc_bytes = 0;
1761         inode->i_mapping->writeback_index = 0;
1762         BTRFS_I(inode)->disk_i_size = 0;
1763         BTRFS_I(inode)->index_cnt = (u64)-1;
1764         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1765         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1766                              inode->i_mapping, GFP_NOFS);
1767         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1768                              inode->i_mapping, GFP_NOFS);
1769         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
1770         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1771         mutex_init(&BTRFS_I(inode)->csum_mutex);
1772         mutex_init(&BTRFS_I(inode)->extent_mutex);
1773         return 0;
1774 }
1775
1776 static int btrfs_find_actor(struct inode *inode, void *opaque)
1777 {
1778         struct btrfs_iget_args *args = opaque;
1779         return (args->ino == inode->i_ino &&
1780                 args->root == BTRFS_I(inode)->root);
1781 }
1782
1783 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1784                             u64 root_objectid)
1785 {
1786         struct btrfs_iget_args args;
1787         args.ino = objectid;
1788         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1789
1790         if (!args.root)
1791                 return NULL;
1792
1793         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1794 }
1795
1796 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1797                                 struct btrfs_root *root)
1798 {
1799         struct inode *inode;
1800         struct btrfs_iget_args args;
1801         args.ino = objectid;
1802         args.root = root;
1803
1804         inode = iget5_locked(s, objectid, btrfs_find_actor,
1805                              btrfs_init_locked_inode,
1806                              (void *)&args);
1807         return inode;
1808 }
1809
1810 /* Get an inode object given its location and corresponding root.
1811  * Returns in *is_new if the inode was read from disk
1812  */
1813 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
1814                          struct btrfs_root *root, int *is_new)
1815 {
1816         struct inode *inode;
1817
1818         inode = btrfs_iget_locked(s, location->objectid, root);
1819         if (!inode)
1820                 return ERR_PTR(-EACCES);
1821
1822         if (inode->i_state & I_NEW) {
1823                 BTRFS_I(inode)->root = root;
1824                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
1825                 btrfs_read_locked_inode(inode);
1826                 unlock_new_inode(inode);
1827                 if (is_new)
1828                         *is_new = 1;
1829         } else {
1830                 if (is_new)
1831                         *is_new = 0;
1832         }
1833
1834         return inode;
1835 }
1836
1837 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1838                                    struct nameidata *nd)
1839 {
1840         struct inode * inode;
1841         struct btrfs_inode *bi = BTRFS_I(dir);
1842         struct btrfs_root *root = bi->root;
1843         struct btrfs_root *sub_root = root;
1844         struct btrfs_key location;
1845         int ret, new, do_orphan = 0;
1846
1847         if (dentry->d_name.len > BTRFS_NAME_LEN)
1848                 return ERR_PTR(-ENAMETOOLONG);
1849
1850         ret = btrfs_inode_by_name(dir, dentry, &location);
1851
1852         if (ret < 0)
1853                 return ERR_PTR(ret);
1854
1855         inode = NULL;
1856         if (location.objectid) {
1857                 ret = fixup_tree_root_location(root, &location, &sub_root,
1858                                                 dentry);
1859                 if (ret < 0)
1860                         return ERR_PTR(ret);
1861                 if (ret > 0)
1862                         return ERR_PTR(-ENOENT);
1863                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
1864                 if (IS_ERR(inode))
1865                         return ERR_CAST(inode);
1866
1867                 /* the inode and parent dir are two different roots */
1868                 if (new && root != sub_root) {
1869                         igrab(inode);
1870                         sub_root->inode = inode;
1871                         do_orphan = 1;
1872                 }
1873         }
1874
1875         if (unlikely(do_orphan))
1876                 btrfs_orphan_cleanup(sub_root);
1877
1878         return d_splice_alias(inode, dentry);
1879 }
1880
1881 static unsigned char btrfs_filetype_table[] = {
1882         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1883 };
1884
1885 static int btrfs_real_readdir(struct file *filp, void *dirent,
1886                               filldir_t filldir)
1887 {
1888         struct inode *inode = filp->f_dentry->d_inode;
1889         struct btrfs_root *root = BTRFS_I(inode)->root;
1890         struct btrfs_item *item;
1891         struct btrfs_dir_item *di;
1892         struct btrfs_key key;
1893         struct btrfs_key found_key;
1894         struct btrfs_path *path;
1895         int ret;
1896         u32 nritems;
1897         struct extent_buffer *leaf;
1898         int slot;
1899         int advance;
1900         unsigned char d_type;
1901         int over = 0;
1902         u32 di_cur;
1903         u32 di_total;
1904         u32 di_len;
1905         int key_type = BTRFS_DIR_INDEX_KEY;
1906         char tmp_name[32];
1907         char *name_ptr;
1908         int name_len;
1909
1910         /* FIXME, use a real flag for deciding about the key type */
1911         if (root->fs_info->tree_root == root)
1912                 key_type = BTRFS_DIR_ITEM_KEY;
1913
1914         /* special case for "." */
1915         if (filp->f_pos == 0) {
1916                 over = filldir(dirent, ".", 1,
1917                                1, inode->i_ino,
1918                                DT_DIR);
1919                 if (over)
1920                         return 0;
1921                 filp->f_pos = 1;
1922         }
1923         /* special case for .., just use the back ref */
1924         if (filp->f_pos == 1) {
1925                 u64 pino = parent_ino(filp->f_path.dentry);
1926                 over = filldir(dirent, "..", 2,
1927                                2, pino, DT_DIR);
1928                 if (over)
1929                         return 0;
1930                 filp->f_pos = 2;
1931         }
1932
1933         path = btrfs_alloc_path();
1934         path->reada = 2;
1935
1936         btrfs_set_key_type(&key, key_type);
1937         key.offset = filp->f_pos;
1938         key.objectid = inode->i_ino;
1939
1940         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1941         if (ret < 0)
1942                 goto err;
1943         advance = 0;
1944
1945         while (1) {
1946                 leaf = path->nodes[0];
1947                 nritems = btrfs_header_nritems(leaf);
1948                 slot = path->slots[0];
1949                 if (advance || slot >= nritems) {
1950                         if (slot >= nritems - 1) {
1951                                 ret = btrfs_next_leaf(root, path);
1952                                 if (ret)
1953                                         break;
1954                                 leaf = path->nodes[0];
1955                                 nritems = btrfs_header_nritems(leaf);
1956                                 slot = path->slots[0];
1957                         } else {
1958                                 slot++;
1959                                 path->slots[0]++;
1960                         }
1961                 }
1962                 advance = 1;
1963                 item = btrfs_item_nr(leaf, slot);
1964                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1965
1966                 if (found_key.objectid != key.objectid)
1967                         break;
1968                 if (btrfs_key_type(&found_key) != key_type)
1969                         break;
1970                 if (found_key.offset < filp->f_pos)
1971                         continue;
1972
1973                 filp->f_pos = found_key.offset;
1974
1975                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1976                 di_cur = 0;
1977                 di_total = btrfs_item_size(leaf, item);
1978
1979                 while (di_cur < di_total) {
1980                         struct btrfs_key location;
1981
1982                         name_len = btrfs_dir_name_len(leaf, di);
1983                         if (name_len <= sizeof(tmp_name)) {
1984                                 name_ptr = tmp_name;
1985                         } else {
1986                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1987                                 if (!name_ptr) {
1988                                         ret = -ENOMEM;
1989                                         goto err;
1990                                 }
1991                         }
1992                         read_extent_buffer(leaf, name_ptr,
1993                                            (unsigned long)(di + 1), name_len);
1994
1995                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1996                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1997                         over = filldir(dirent, name_ptr, name_len,
1998                                        found_key.offset, location.objectid,
1999                                        d_type);
2000
2001                         if (name_ptr != tmp_name)
2002                                 kfree(name_ptr);
2003
2004                         if (over)
2005                                 goto nopos;
2006
2007                         di_len = btrfs_dir_name_len(leaf, di) +
2008                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
2009                         di_cur += di_len;
2010                         di = (struct btrfs_dir_item *)((char *)di + di_len);
2011                 }
2012         }
2013
2014         /* Reached end of directory/root. Bump pos past the last item. */
2015         if (key_type == BTRFS_DIR_INDEX_KEY)
2016                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2017         else
2018                 filp->f_pos++;
2019 nopos:
2020         ret = 0;
2021 err:
2022         btrfs_free_path(path);
2023         return ret;
2024 }
2025
2026 /* Kernels earlier than 2.6.28 still have the NFS deadlock where nfsd
2027    will call the file system's ->lookup() method from within its
2028    filldir callback, which in turn was called from the file system's
2029    ->readdir() method. And will deadlock for many file systems. */
2030 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
2031
2032 struct nfshack_dirent {
2033         u64             ino;
2034         loff_t          offset;
2035         int             namlen;
2036         unsigned int    d_type;
2037         char            name[];
2038 };
2039
2040 struct nfshack_readdir {
2041         char            *dirent;
2042         size_t          used;
2043 };
2044
2045
2046
2047 static int btrfs_nfshack_filldir(void *__buf, const char *name, int namlen,
2048                               loff_t offset, u64 ino, unsigned int d_type)
2049 {
2050         struct nfshack_readdir *buf = __buf;
2051         struct nfshack_dirent *de = (void *)(buf->dirent + buf->used);
2052         unsigned int reclen;
2053
2054         reclen = ALIGN(sizeof(struct nfshack_dirent) + namlen, sizeof(u64));
2055         if (buf->used + reclen > PAGE_SIZE)
2056                 return -EINVAL;
2057
2058         de->namlen = namlen;
2059         de->offset = offset;
2060         de->ino = ino;
2061         de->d_type = d_type;
2062         memcpy(de->name, name, namlen);
2063         buf->used += reclen;
2064
2065         return 0;
2066 }
2067
2068 static int btrfs_nfshack_readdir(struct file *file, void *dirent,
2069                                  filldir_t filldir)
2070 {
2071         struct nfshack_readdir buf;
2072         struct nfshack_dirent *de;
2073         int err;
2074         int size;
2075         loff_t offset;
2076
2077         buf.dirent = (void *)__get_free_page(GFP_KERNEL);
2078         if (!buf.dirent)
2079                 return -ENOMEM;
2080
2081         offset = file->f_pos;
2082
2083         while (1) {
2084                 unsigned int reclen;
2085
2086                 buf.used = 0;
2087
2088                 err = btrfs_real_readdir(file, &buf, btrfs_nfshack_filldir);
2089                 if (err)
2090                         break;
2091
2092                 size = buf.used;
2093
2094                 if (!size)
2095                         break;
2096
2097                 de = (struct nfshack_dirent *)buf.dirent;
2098                 while (size > 0) {
2099                         offset = de->offset;
2100
2101                         if (filldir(dirent, de->name, de->namlen, de->offset,
2102                                     de->ino, de->d_type))
2103                                 goto done;
2104                         offset = file->f_pos;
2105
2106                         reclen = ALIGN(sizeof(*de) + de->namlen,
2107                                        sizeof(u64));
2108                         size -= reclen;
2109                         de = (struct nfshack_dirent *)((char *)de + reclen);
2110                 }
2111         }
2112
2113  done:
2114         free_page((unsigned long)buf.dirent);
2115         file->f_pos = offset;
2116
2117         return err;
2118 }
2119 #endif
2120
2121 int btrfs_write_inode(struct inode *inode, int wait)
2122 {
2123         struct btrfs_root *root = BTRFS_I(inode)->root;
2124         struct btrfs_trans_handle *trans;
2125         int ret = 0;
2126
2127         if (root->fs_info->closing > 1)
2128                 return 0;
2129
2130         if (wait) {
2131                 trans = btrfs_join_transaction(root, 1);
2132                 btrfs_set_trans_block_group(trans, inode);
2133                 ret = btrfs_commit_transaction(trans, root);
2134         }
2135         return ret;
2136 }
2137
2138 /*
2139  * This is somewhat expensive, updating the tree every time the
2140  * inode changes.  But, it is most likely to find the inode in cache.
2141  * FIXME, needs more benchmarking...there are no reasons other than performance
2142  * to keep or drop this code.
2143  */
2144 void btrfs_dirty_inode(struct inode *inode)
2145 {
2146         struct btrfs_root *root = BTRFS_I(inode)->root;
2147         struct btrfs_trans_handle *trans;
2148
2149         trans = btrfs_join_transaction(root, 1);
2150         btrfs_set_trans_block_group(trans, inode);
2151         btrfs_update_inode(trans, root, inode);
2152         btrfs_end_transaction(trans, root);
2153 }
2154
2155 static int btrfs_set_inode_index_count(struct inode *inode)
2156 {
2157         struct btrfs_root *root = BTRFS_I(inode)->root;
2158         struct btrfs_key key, found_key;
2159         struct btrfs_path *path;
2160         struct extent_buffer *leaf;
2161         int ret;
2162
2163         key.objectid = inode->i_ino;
2164         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2165         key.offset = (u64)-1;
2166
2167         path = btrfs_alloc_path();
2168         if (!path)
2169                 return -ENOMEM;
2170
2171         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2172         if (ret < 0)
2173                 goto out;
2174         /* FIXME: we should be able to handle this */
2175         if (ret == 0)
2176                 goto out;
2177         ret = 0;
2178
2179         /*
2180          * MAGIC NUMBER EXPLANATION:
2181          * since we search a directory based on f_pos we have to start at 2
2182          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2183          * else has to start at 2
2184          */
2185         if (path->slots[0] == 0) {
2186                 BTRFS_I(inode)->index_cnt = 2;
2187                 goto out;
2188         }
2189
2190         path->slots[0]--;
2191
2192         leaf = path->nodes[0];
2193         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2194
2195         if (found_key.objectid != inode->i_ino ||
2196             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2197                 BTRFS_I(inode)->index_cnt = 2;
2198                 goto out;
2199         }
2200
2201         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2202 out:
2203         btrfs_free_path(path);
2204         return ret;
2205 }
2206
2207 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2208                                  u64 *index)
2209 {
2210         int ret = 0;
2211
2212         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2213                 ret = btrfs_set_inode_index_count(dir);
2214                 if (ret)
2215                         return ret;
2216         }
2217
2218         *index = BTRFS_I(dir)->index_cnt;
2219         BTRFS_I(dir)->index_cnt++;
2220
2221         return ret;
2222 }
2223
2224 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2225                                      struct btrfs_root *root,
2226                                      struct inode *dir,
2227                                      const char *name, int name_len,
2228                                      u64 ref_objectid,
2229                                      u64 objectid,
2230                                      struct btrfs_block_group_cache *group,
2231                                      int mode, u64 *index)
2232 {
2233         struct inode *inode;
2234         struct btrfs_inode_item *inode_item;
2235         struct btrfs_block_group_cache *new_inode_group;
2236         struct btrfs_key *location;
2237         struct btrfs_path *path;
2238         struct btrfs_inode_ref *ref;
2239         struct btrfs_key key[2];
2240         u32 sizes[2];
2241         unsigned long ptr;
2242         int ret;
2243         int owner;
2244
2245         path = btrfs_alloc_path();
2246         BUG_ON(!path);
2247
2248         inode = new_inode(root->fs_info->sb);
2249         if (!inode)
2250                 return ERR_PTR(-ENOMEM);
2251
2252         if (dir) {
2253                 ret = btrfs_set_inode_index(dir, inode, index);
2254                 if (ret)
2255                         return ERR_PTR(ret);
2256         }
2257         /*
2258          * index_cnt is ignored for everything but a dir,
2259          * btrfs_get_inode_index_count has an explanation for the magic
2260          * number
2261          */
2262         BTRFS_I(inode)->index_cnt = 2;
2263
2264         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2265         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2266                              inode->i_mapping, GFP_NOFS);
2267         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2268                              inode->i_mapping, GFP_NOFS);
2269         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2270         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2271         mutex_init(&BTRFS_I(inode)->csum_mutex);
2272         mutex_init(&BTRFS_I(inode)->extent_mutex);
2273         BTRFS_I(inode)->delalloc_bytes = 0;
2274         inode->i_mapping->writeback_index = 0;
2275         BTRFS_I(inode)->disk_i_size = 0;
2276         BTRFS_I(inode)->root = root;
2277
2278         if (mode & S_IFDIR)
2279                 owner = 0;
2280         else
2281                 owner = 1;
2282         new_inode_group = btrfs_find_block_group(root, group, 0,
2283                                        BTRFS_BLOCK_GROUP_METADATA, owner);
2284         if (!new_inode_group) {
2285                 printk("find_block group failed\n");
2286                 new_inode_group = group;
2287         }
2288         BTRFS_I(inode)->block_group = new_inode_group;
2289         BTRFS_I(inode)->flags = 0;
2290
2291         key[0].objectid = objectid;
2292         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2293         key[0].offset = 0;
2294
2295         key[1].objectid = objectid;
2296         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2297         key[1].offset = ref_objectid;
2298
2299         sizes[0] = sizeof(struct btrfs_inode_item);
2300         sizes[1] = name_len + sizeof(*ref);
2301
2302         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2303         if (ret != 0)
2304                 goto fail;
2305
2306         if (objectid > root->highest_inode)
2307                 root->highest_inode = objectid;
2308
2309         inode->i_uid = current->fsuid;
2310         inode->i_gid = current->fsgid;
2311         inode->i_mode = mode;
2312         inode->i_ino = objectid;
2313         inode->i_blocks = 0;
2314         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2315         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2316                                   struct btrfs_inode_item);
2317         fill_inode_item(path->nodes[0], inode_item, inode);
2318
2319         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2320                              struct btrfs_inode_ref);
2321         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2322         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
2323         ptr = (unsigned long)(ref + 1);
2324         write_extent_buffer(path->nodes[0], name, ptr, name_len);
2325
2326         btrfs_mark_buffer_dirty(path->nodes[0]);
2327         btrfs_free_path(path);
2328
2329         location = &BTRFS_I(inode)->location;
2330         location->objectid = objectid;
2331         location->offset = 0;
2332         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2333
2334         insert_inode_hash(inode);
2335         return inode;
2336 fail:
2337         if (dir)
2338                 BTRFS_I(dir)->index_cnt--;
2339         btrfs_free_path(path);
2340         return ERR_PTR(ret);
2341 }
2342
2343 static inline u8 btrfs_inode_type(struct inode *inode)
2344 {
2345         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2346 }
2347
2348 static int btrfs_add_link(struct btrfs_trans_handle *trans,
2349                             struct dentry *dentry, struct inode *inode,
2350                             int add_backref, u64 index)
2351 {
2352         int ret;
2353         struct btrfs_key key;
2354         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
2355         struct inode *parent_inode = dentry->d_parent->d_inode;
2356
2357         key.objectid = inode->i_ino;
2358         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2359         key.offset = 0;
2360
2361         ret = btrfs_insert_dir_item(trans, root,
2362                                     dentry->d_name.name, dentry->d_name.len,
2363                                     dentry->d_parent->d_inode->i_ino,
2364                                     &key, btrfs_inode_type(inode),
2365                                     index);
2366         if (ret == 0) {
2367                 if (add_backref) {
2368                         ret = btrfs_insert_inode_ref(trans, root,
2369                                              dentry->d_name.name,
2370                                              dentry->d_name.len,
2371                                              inode->i_ino,
2372                                              parent_inode->i_ino,
2373                                              index);
2374                 }
2375                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2376                                    dentry->d_name.len * 2);
2377                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2378                 ret = btrfs_update_inode(trans, root,
2379                                          dentry->d_parent->d_inode);
2380         }
2381         return ret;
2382 }
2383
2384 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2385                             struct dentry *dentry, struct inode *inode,
2386                             int backref, u64 index)
2387 {
2388         int err = btrfs_add_link(trans, dentry, inode, backref, index);
2389         if (!err) {
2390                 d_instantiate(dentry, inode);
2391                 return 0;
2392         }
2393         if (err > 0)
2394                 err = -EEXIST;
2395         return err;
2396 }
2397
2398 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2399                         int mode, dev_t rdev)
2400 {
2401         struct btrfs_trans_handle *trans;
2402         struct btrfs_root *root = BTRFS_I(dir)->root;
2403         struct inode *inode = NULL;
2404         int err;
2405         int drop_inode = 0;
2406         u64 objectid;
2407         unsigned long nr = 0;
2408         u64 index = 0;
2409
2410         if (!new_valid_dev(rdev))
2411                 return -EINVAL;
2412
2413         err = btrfs_check_free_space(root, 1, 0);
2414         if (err)
2415                 goto fail;
2416
2417         trans = btrfs_start_transaction(root, 1);
2418         btrfs_set_trans_block_group(trans, dir);
2419
2420         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2421         if (err) {
2422                 err = -ENOSPC;
2423                 goto out_unlock;
2424         }
2425
2426         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2427                                 dentry->d_name.len,
2428                                 dentry->d_parent->d_inode->i_ino, objectid,
2429                                 BTRFS_I(dir)->block_group, mode, &index);
2430         err = PTR_ERR(inode);
2431         if (IS_ERR(inode))
2432                 goto out_unlock;
2433
2434         err = btrfs_init_acl(inode, dir);
2435         if (err) {
2436                 drop_inode = 1;
2437                 goto out_unlock;
2438         }
2439
2440         btrfs_set_trans_block_group(trans, inode);
2441         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2442         if (err)
2443                 drop_inode = 1;
2444         else {
2445                 inode->i_op = &btrfs_special_inode_operations;
2446                 init_special_inode(inode, inode->i_mode, rdev);
2447                 btrfs_update_inode(trans, root, inode);
2448         }
2449         dir->i_sb->s_dirt = 1;
2450         btrfs_update_inode_block_group(trans, inode);
2451         btrfs_update_inode_block_group(trans, dir);
2452 out_unlock:
2453         nr = trans->blocks_used;
2454         btrfs_end_transaction_throttle(trans, root);
2455 fail:
2456         if (drop_inode) {
2457                 inode_dec_link_count(inode);
2458                 iput(inode);
2459         }
2460         btrfs_btree_balance_dirty(root, nr);
2461         return err;
2462 }
2463
2464 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2465                         int mode, struct nameidata *nd)
2466 {
2467         struct btrfs_trans_handle *trans;
2468         struct btrfs_root *root = BTRFS_I(dir)->root;
2469         struct inode *inode = NULL;
2470         int err;
2471         int drop_inode = 0;
2472         unsigned long nr = 0;
2473         u64 objectid;
2474         u64 index = 0;
2475
2476         err = btrfs_check_free_space(root, 1, 0);
2477         if (err)
2478                 goto fail;
2479         trans = btrfs_start_transaction(root, 1);
2480         btrfs_set_trans_block_group(trans, dir);
2481
2482         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2483         if (err) {
2484                 err = -ENOSPC;
2485                 goto out_unlock;
2486         }
2487
2488         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2489                                 dentry->d_name.len,
2490                                 dentry->d_parent->d_inode->i_ino,
2491                                 objectid, BTRFS_I(dir)->block_group, mode,
2492                                 &index);
2493         err = PTR_ERR(inode);
2494         if (IS_ERR(inode))
2495                 goto out_unlock;
2496
2497         err = btrfs_init_acl(inode, dir);
2498         if (err) {
2499                 drop_inode = 1;
2500                 goto out_unlock;
2501         }
2502
2503         btrfs_set_trans_block_group(trans, inode);
2504         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2505         if (err)
2506                 drop_inode = 1;
2507         else {
2508                 inode->i_mapping->a_ops = &btrfs_aops;
2509                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2510                 inode->i_fop = &btrfs_file_operations;
2511                 inode->i_op = &btrfs_file_inode_operations;
2512                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2513                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2514                                      inode->i_mapping, GFP_NOFS);
2515                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2516                                      inode->i_mapping, GFP_NOFS);
2517                 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2518                 mutex_init(&BTRFS_I(inode)->csum_mutex);
2519                 mutex_init(&BTRFS_I(inode)->extent_mutex);
2520                 BTRFS_I(inode)->delalloc_bytes = 0;
2521                 BTRFS_I(inode)->disk_i_size = 0;
2522                 inode->i_mapping->writeback_index = 0;
2523                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2524                 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2525         }
2526         dir->i_sb->s_dirt = 1;
2527         btrfs_update_inode_block_group(trans, inode);
2528         btrfs_update_inode_block_group(trans, dir);
2529 out_unlock:
2530         nr = trans->blocks_used;
2531         btrfs_end_transaction_throttle(trans, root);
2532 fail:
2533         if (drop_inode) {
2534                 inode_dec_link_count(inode);
2535                 iput(inode);
2536         }
2537         btrfs_btree_balance_dirty(root, nr);
2538         return err;
2539 }
2540
2541 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2542                       struct dentry *dentry)
2543 {
2544         struct btrfs_trans_handle *trans;
2545         struct btrfs_root *root = BTRFS_I(dir)->root;
2546         struct inode *inode = old_dentry->d_inode;
2547         u64 index;
2548         unsigned long nr = 0;
2549         int err;
2550         int drop_inode = 0;
2551
2552         if (inode->i_nlink == 0)
2553                 return -ENOENT;
2554
2555 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2556         inode->i_nlink++;
2557 #else
2558         inc_nlink(inode);
2559 #endif
2560         err = btrfs_check_free_space(root, 1, 0);
2561         if (err)
2562                 goto fail;
2563         err = btrfs_set_inode_index(dir, inode, &index);
2564         if (err)
2565                 goto fail;
2566
2567         trans = btrfs_start_transaction(root, 1);
2568
2569         btrfs_set_trans_block_group(trans, dir);
2570         atomic_inc(&inode->i_count);
2571
2572         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
2573
2574         if (err)
2575                 drop_inode = 1;
2576
2577         dir->i_sb->s_dirt = 1;
2578         btrfs_update_inode_block_group(trans, dir);
2579         err = btrfs_update_inode(trans, root, inode);
2580
2581         if (err)
2582                 drop_inode = 1;
2583
2584         nr = trans->blocks_used;
2585         btrfs_end_transaction_throttle(trans, root);
2586 fail:
2587         if (drop_inode) {
2588                 inode_dec_link_count(inode);
2589                 iput(inode);
2590         }
2591         btrfs_btree_balance_dirty(root, nr);
2592         return err;
2593 }
2594
2595 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2596 {
2597         struct inode *inode = NULL;
2598         struct btrfs_trans_handle *trans;
2599         struct btrfs_root *root = BTRFS_I(dir)->root;
2600         int err = 0;
2601         int drop_on_err = 0;
2602         u64 objectid = 0;
2603         u64 index = 0;
2604         unsigned long nr = 1;
2605
2606         err = btrfs_check_free_space(root, 1, 0);
2607         if (err)
2608                 goto out_unlock;
2609
2610         trans = btrfs_start_transaction(root, 1);
2611         btrfs_set_trans_block_group(trans, dir);
2612
2613         if (IS_ERR(trans)) {
2614                 err = PTR_ERR(trans);
2615                 goto out_unlock;
2616         }
2617
2618         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2619         if (err) {
2620                 err = -ENOSPC;
2621                 goto out_unlock;
2622         }
2623
2624         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2625                                 dentry->d_name.len,
2626                                 dentry->d_parent->d_inode->i_ino, objectid,
2627                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
2628                                 &index);
2629         if (IS_ERR(inode)) {
2630                 err = PTR_ERR(inode);
2631                 goto out_fail;
2632         }
2633
2634         drop_on_err = 1;
2635
2636         err = btrfs_init_acl(inode, dir);
2637         if (err)
2638                 goto out_fail;
2639
2640         inode->i_op = &btrfs_dir_inode_operations;
2641         inode->i_fop = &btrfs_dir_file_operations;
2642         btrfs_set_trans_block_group(trans, inode);
2643
2644         btrfs_i_size_write(inode, 0);
2645         err = btrfs_update_inode(trans, root, inode);
2646         if (err)
2647                 goto out_fail;
2648
2649         err = btrfs_add_link(trans, dentry, inode, 0, index);
2650         if (err)
2651                 goto out_fail;
2652
2653         d_instantiate(dentry, inode);
2654         drop_on_err = 0;
2655         dir->i_sb->s_dirt = 1;
2656         btrfs_update_inode_block_group(trans, inode);
2657         btrfs_update_inode_block_group(trans, dir);
2658
2659 out_fail:
2660         nr = trans->blocks_used;
2661         btrfs_end_transaction_throttle(trans, root);
2662
2663 out_unlock:
2664         if (drop_on_err)
2665                 iput(inode);
2666         btrfs_btree_balance_dirty(root, nr);
2667         return err;
2668 }
2669
2670 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2671                                 struct extent_map *existing,
2672                                 struct extent_map *em,
2673                                 u64 map_start, u64 map_len)
2674 {
2675         u64 start_diff;
2676
2677         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2678         start_diff = map_start - em->start;
2679         em->start = map_start;
2680         em->len = map_len;
2681         if (em->block_start < EXTENT_MAP_LAST_BYTE)
2682                 em->block_start += start_diff;
2683         return add_extent_mapping(em_tree, em);
2684 }
2685
2686 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2687                                     size_t pg_offset, u64 start, u64 len,
2688                                     int create)
2689 {
2690         int ret;
2691         int err = 0;
2692         u64 bytenr;
2693         u64 extent_start = 0;
2694         u64 extent_end = 0;
2695         u64 objectid = inode->i_ino;
2696         u32 found_type;
2697         struct btrfs_path *path = NULL;
2698         struct btrfs_root *root = BTRFS_I(inode)->root;
2699         struct btrfs_file_extent_item *item;
2700         struct extent_buffer *leaf;
2701         struct btrfs_key found_key;
2702         struct extent_map *em = NULL;
2703         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2704         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2705         struct btrfs_trans_handle *trans = NULL;
2706
2707 again:
2708         spin_lock(&em_tree->lock);
2709         em = lookup_extent_mapping(em_tree, start, len);
2710         if (em)
2711                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2712         spin_unlock(&em_tree->lock);
2713
2714         if (em) {
2715                 if (em->start > start || em->start + em->len <= start)
2716                         free_extent_map(em);
2717                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2718                         free_extent_map(em);
2719                 else
2720                         goto out;
2721         }
2722         em = alloc_extent_map(GFP_NOFS);
2723         if (!em) {
2724                 err = -ENOMEM;
2725                 goto out;
2726         }
2727         em->bdev = root->fs_info->fs_devices->latest_bdev;
2728         em->start = EXTENT_MAP_HOLE;
2729         em->len = (u64)-1;
2730
2731         if (!path) {
2732                 path = btrfs_alloc_path();
2733                 BUG_ON(!path);
2734         }
2735
2736         ret = btrfs_lookup_file_extent(trans, root, path,
2737                                        objectid, start, trans != NULL);
2738         if (ret < 0) {
2739                 err = ret;
2740                 goto out;
2741         }
2742
2743         if (ret != 0) {
2744                 if (path->slots[0] == 0)
2745                         goto not_found;
2746                 path->slots[0]--;
2747         }
2748
2749         leaf = path->nodes[0];
2750         item = btrfs_item_ptr(leaf, path->slots[0],
2751                               struct btrfs_file_extent_item);
2752         /* are we inside the extent that was found? */
2753         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2754         found_type = btrfs_key_type(&found_key);
2755         if (found_key.objectid != objectid ||
2756             found_type != BTRFS_EXTENT_DATA_KEY) {
2757                 goto not_found;
2758         }
2759
2760         found_type = btrfs_file_extent_type(leaf, item);
2761         extent_start = found_key.offset;
2762         if (found_type == BTRFS_FILE_EXTENT_REG) {
2763                 extent_end = extent_start +
2764                        btrfs_file_extent_num_bytes(leaf, item);
2765                 err = 0;
2766                 if (start < extent_start || start >= extent_end) {
2767                         em->start = start;
2768                         if (start < extent_start) {
2769                                 if (start + len <= extent_start)
2770                                         goto not_found;
2771                                 em->len = extent_end - extent_start;
2772                         } else {
2773                                 em->len = len;
2774                         }
2775                         goto not_found_em;
2776                 }
2777                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2778                 if (bytenr == 0) {
2779                         em->start = extent_start;
2780                         em->len = extent_end - extent_start;
2781                         em->block_start = EXTENT_MAP_HOLE;
2782                         goto insert;
2783                 }
2784                 bytenr += btrfs_file_extent_offset(leaf, item);
2785                 em->block_start = bytenr;
2786                 em->start = extent_start;
2787                 em->len = extent_end - extent_start;
2788                 goto insert;
2789         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2790                 u64 page_start;
2791                 unsigned long ptr;
2792                 char *map;
2793                 size_t size;
2794                 size_t extent_offset;
2795                 size_t copy_size;
2796
2797                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2798                                                     path->slots[0]));
2799                 extent_end = (extent_start + size + root->sectorsize - 1) &
2800                         ~((u64)root->sectorsize - 1);
2801                 if (start < extent_start || start >= extent_end) {
2802                         em->start = start;
2803                         if (start < extent_start) {
2804                                 if (start + len <= extent_start)
2805                                         goto not_found;
2806                                 em->len = extent_end - extent_start;
2807                         } else {
2808                                 em->len = len;
2809                         }
2810                         goto not_found_em;
2811                 }
2812                 em->block_start = EXTENT_MAP_INLINE;
2813
2814                 if (!page) {
2815                         em->start = extent_start;
2816                         em->len = size;
2817                         goto out;
2818                 }
2819
2820                 page_start = page_offset(page) + pg_offset;
2821                 extent_offset = page_start - extent_start;
2822                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2823                                 size - extent_offset);
2824                 em->start = extent_start + extent_offset;
2825                 em->len = (copy_size + root->sectorsize - 1) &
2826                         ~((u64)root->sectorsize - 1);
2827                 map = kmap(page);
2828                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2829                 if (create == 0 && !PageUptodate(page)) {
2830                         read_extent_buffer(leaf, map + pg_offset, ptr,
2831                                            copy_size);
2832                         flush_dcache_page(page);
2833                 } else if (create && PageUptodate(page)) {
2834                         if (!trans) {
2835                                 kunmap(page);
2836                                 free_extent_map(em);
2837                                 em = NULL;
2838                                 btrfs_release_path(root, path);
2839                                 trans = btrfs_join_transaction(root, 1);
2840                                 goto again;
2841                         }
2842                         write_extent_buffer(leaf, map + pg_offset, ptr,
2843                                             copy_size);
2844                         btrfs_mark_buffer_dirty(leaf);
2845                 }
2846                 kunmap(page);
2847                 set_extent_uptodate(io_tree, em->start,
2848                                     extent_map_end(em) - 1, GFP_NOFS);
2849                 goto insert;
2850         } else {
2851                 printk("unkknown found_type %d\n", found_type);
2852                 WARN_ON(1);
2853         }
2854 not_found:
2855         em->start = start;
2856         em->len = len;
2857 not_found_em:
2858         em->block_start = EXTENT_MAP_HOLE;
2859 insert:
2860         btrfs_release_path(root, path);
2861         if (em->start > start || extent_map_end(em) <= start) {
2862                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2863                 err = -EIO;
2864                 goto out;
2865         }
2866
2867         err = 0;
2868         spin_lock(&em_tree->lock);
2869         ret = add_extent_mapping(em_tree, em);
2870         /* it is possible that someone inserted the extent into the tree
2871          * while we had the lock dropped.  It is also possible that
2872          * an overlapping map exists in the tree
2873          */
2874         if (ret == -EEXIST) {
2875                 struct extent_map *existing;
2876
2877                 ret = 0;
2878
2879                 existing = lookup_extent_mapping(em_tree, start, len);
2880                 if (existing && (existing->start > start ||
2881                     existing->start + existing->len <= start)) {
2882                         free_extent_map(existing);
2883                         existing = NULL;
2884                 }
2885                 if (!existing) {
2886                         existing = lookup_extent_mapping(em_tree, em->start,
2887                                                          em->len);
2888                         if (existing) {
2889                                 err = merge_extent_mapping(em_tree, existing,
2890                                                            em, start,
2891                                                            root->sectorsize);
2892                                 free_extent_map(existing);
2893                                 if (err) {
2894                                         free_extent_map(em);
2895                                         em = NULL;
2896                                 }
2897                         } else {
2898                                 err = -EIO;
2899                                 printk("failing to insert %Lu %Lu\n",
2900                                        start, len);
2901                                 free_extent_map(em);
2902                                 em = NULL;
2903                         }
2904                 } else {
2905                         free_extent_map(em);
2906                         em = existing;
2907                         err = 0;
2908                 }
2909         }
2910         spin_unlock(&em_tree->lock);
2911 out:
2912         if (path)
2913                 btrfs_free_path(path);
2914         if (trans) {
2915                 ret = btrfs_end_transaction(trans, root);
2916                 if (!err) {
2917                         err = ret;
2918                 }
2919         }
2920         if (err) {
2921                 free_extent_map(em);
2922                 WARN_ON(1);
2923                 return ERR_PTR(err);
2924         }
2925         return em;
2926 }
2927
2928 #if 0 /* waiting for O_DIRECT reads */
2929 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2930                         struct buffer_head *bh_result, int create)
2931 {
2932         struct extent_map *em;
2933         u64 start = (u64)iblock << inode->i_blkbits;
2934         struct btrfs_multi_bio *multi = NULL;
2935         struct btrfs_root *root = BTRFS_I(inode)->root;
2936         u64 len;
2937         u64 logical;
2938         u64 map_length;
2939         int ret = 0;
2940
2941         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2942
2943         if (!em || IS_ERR(em))
2944                 goto out;
2945
2946         if (em->start > start || em->start + em->len <= start) {
2947             goto out;
2948         }
2949
2950         if (em->block_start == EXTENT_MAP_INLINE) {
2951                 ret = -EINVAL;
2952                 goto out;
2953         }
2954
2955         len = em->start + em->len - start;
2956         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2957
2958         if (em->block_start == EXTENT_MAP_HOLE ||
2959             em->block_start == EXTENT_MAP_DELALLOC) {
2960                 bh_result->b_size = len;
2961                 goto out;
2962         }
2963
2964         logical = start - em->start;
2965         logical = em->block_start + logical;
2966
2967         map_length = len;
2968         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2969                               logical, &map_length, &multi, 0);
2970         BUG_ON(ret);
2971         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2972         bh_result->b_size = min(map_length, len);
2973
2974         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2975         set_buffer_mapped(bh_result);
2976         kfree(multi);
2977 out:
2978         free_extent_map(em);
2979         return ret;
2980 }
2981 #endif
2982
2983 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2984                         const struct iovec *iov, loff_t offset,
2985                         unsigned long nr_segs)
2986 {
2987         return -EINVAL;
2988 #if 0
2989         struct file *file = iocb->ki_filp;
2990         struct inode *inode = file->f_mapping->host;
2991
2992         if (rw == WRITE)
2993                 return -EINVAL;
2994
2995         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2996                                   offset, nr_segs, btrfs_get_block, NULL);
2997 #endif
2998 }
2999
3000 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
3001 {
3002         return extent_bmap(mapping, iblock, btrfs_get_extent);
3003 }
3004
3005 int btrfs_readpage(struct file *file, struct page *page)
3006 {
3007         struct extent_io_tree *tree;
3008         tree = &BTRFS_I(page->mapping->host)->io_tree;
3009         return extent_read_full_page(tree, page, btrfs_get_extent);
3010 }
3011
3012 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
3013 {
3014         struct extent_io_tree *tree;
3015
3016
3017         if (current->flags & PF_MEMALLOC) {
3018                 redirty_page_for_writepage(wbc, page);
3019                 unlock_page(page);
3020                 return 0;
3021         }
3022         tree = &BTRFS_I(page->mapping->host)->io_tree;
3023         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
3024 }
3025
3026 int btrfs_writepages(struct address_space *mapping,
3027                      struct writeback_control *wbc)
3028 {
3029         struct extent_io_tree *tree;
3030         tree = &BTRFS_I(mapping->host)->io_tree;
3031         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3032 }
3033
3034 static int
3035 btrfs_readpages(struct file *file, struct address_space *mapping,
3036                 struct list_head *pages, unsigned nr_pages)
3037 {
3038         struct extent_io_tree *tree;
3039         tree = &BTRFS_I(mapping->host)->io_tree;
3040         return extent_readpages(tree, mapping, pages, nr_pages,
3041                                 btrfs_get_extent);
3042 }
3043 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3044 {
3045         struct extent_io_tree *tree;
3046         struct extent_map_tree *map;
3047         int ret;
3048
3049         tree = &BTRFS_I(page->mapping->host)->io_tree;
3050         map = &BTRFS_I(page->mapping->host)->extent_tree;
3051         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
3052         if (ret == 1) {
3053                 ClearPagePrivate(page);
3054                 set_page_private(page, 0);
3055                 page_cache_release(page);
3056         }
3057         return ret;
3058 }
3059
3060 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3061 {
3062         return __btrfs_releasepage(page, gfp_flags);
3063 }
3064
3065 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3066 {
3067         struct extent_io_tree *tree;
3068         struct btrfs_ordered_extent *ordered;
3069         u64 page_start = page_offset(page);
3070         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3071
3072         wait_on_page_writeback(page);
3073         tree = &BTRFS_I(page->mapping->host)->io_tree;
3074         if (offset) {
3075                 btrfs_releasepage(page, GFP_NOFS);
3076                 return;
3077         }
3078
3079         lock_extent(tree, page_start, page_end, GFP_NOFS);
3080         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3081                                            page_offset(page));
3082         if (ordered) {
3083                 /*
3084                  * IO on this page will never be started, so we need
3085                  * to account for any ordered extents now
3086                  */
3087                 clear_extent_bit(tree, page_start, page_end,
3088                                  EXTENT_DIRTY | EXTENT_DELALLOC |
3089                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
3090                 btrfs_finish_ordered_io(page->mapping->host,
3091                                         page_start, page_end);
3092                 btrfs_put_ordered_extent(ordered);
3093                 lock_extent(tree, page_start, page_end, GFP_NOFS);
3094         }
3095         clear_extent_bit(tree, page_start, page_end,
3096                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3097                  EXTENT_ORDERED,
3098                  1, 1, GFP_NOFS);
3099         __btrfs_releasepage(page, GFP_NOFS);
3100
3101         ClearPageChecked(page);
3102         if (PagePrivate(page)) {
3103                 ClearPagePrivate(page);
3104                 set_page_private(page, 0);
3105                 page_cache_release(page);
3106         }
3107 }
3108
3109 /*
3110  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3111  * called from a page fault handler when a page is first dirtied. Hence we must
3112  * be careful to check for EOF conditions here. We set the page up correctly
3113  * for a written page which means we get ENOSPC checking when writing into
3114  * holes and correct delalloc and unwritten extent mapping on filesystems that
3115  * support these features.
3116  *
3117  * We are not allowed to take the i_mutex here so we have to play games to
3118  * protect against truncate races as the page could now be beyond EOF.  Because
3119  * vmtruncate() writes the inode size before removing pages, once we have the
3120  * page lock we can determine safely if the page is beyond EOF. If it is not
3121  * beyond EOF, then the page is guaranteed safe against truncation until we
3122  * unlock the page.
3123  */
3124 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3125 {
3126         struct inode *inode = fdentry(vma->vm_file)->d_inode;
3127         struct btrfs_root *root = BTRFS_I(inode)->root;
3128         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3129         struct btrfs_ordered_extent *ordered;
3130         char *kaddr;
3131         unsigned long zero_start;
3132         loff_t size;
3133         int ret;
3134         u64 page_start;
3135         u64 page_end;
3136
3137         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3138         if (ret)
3139                 goto out;
3140
3141         ret = -EINVAL;
3142 again:
3143         lock_page(page);
3144         size = i_size_read(inode);
3145         page_start = page_offset(page);
3146         page_end = page_start + PAGE_CACHE_SIZE - 1;
3147
3148         if ((page->mapping != inode->i_mapping) ||
3149             (page_start >= size)) {
3150                 /* page got truncated out from underneath us */
3151                 goto out_unlock;
3152         }
3153         wait_on_page_writeback(page);
3154
3155         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3156         set_page_extent_mapped(page);
3157
3158         /*
3159          * we can't set the delalloc bits if there are pending ordered
3160          * extents.  Drop our locks and wait for them to finish
3161          */
3162         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3163         if (ordered) {
3164                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3165                 unlock_page(page);
3166                 btrfs_start_ordered_extent(inode, ordered, 1);
3167                 btrfs_put_ordered_extent(ordered);
3168                 goto again;
3169         }
3170
3171         btrfs_set_extent_delalloc(inode, page_start, page_end);
3172         ret = 0;
3173
3174         /* page is wholly or partially inside EOF */
3175         if (page_start + PAGE_CACHE_SIZE > size)
3176                 zero_start = size & ~PAGE_CACHE_MASK;
3177         else
3178                 zero_start = PAGE_CACHE_SIZE;
3179
3180         if (zero_start != PAGE_CACHE_SIZE) {
3181                 kaddr = kmap(page);
3182                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3183                 flush_dcache_page(page);
3184                 kunmap(page);
3185         }
3186         ClearPageChecked(page);
3187         set_page_dirty(page);
3188         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3189
3190 out_unlock:
3191         unlock_page(page);
3192 out:
3193         return ret;
3194 }
3195
3196 static void btrfs_truncate(struct inode *inode)
3197 {
3198         struct btrfs_root *root = BTRFS_I(inode)->root;
3199         int ret;
3200         struct btrfs_trans_handle *trans;
3201         unsigned long nr;
3202         u64 mask = root->sectorsize - 1;
3203
3204         if (!S_ISREG(inode->i_mode))
3205                 return;
3206         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3207                 return;
3208
3209         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3210         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3211
3212         trans = btrfs_start_transaction(root, 1);
3213         btrfs_set_trans_block_group(trans, inode);
3214         btrfs_i_size_write(inode, inode->i_size);
3215
3216         ret = btrfs_orphan_add(trans, inode);
3217         if (ret)
3218                 goto out;
3219         /* FIXME, add redo link to tree so we don't leak on crash */
3220         ret = btrfs_truncate_in_trans(trans, root, inode,
3221                                       BTRFS_EXTENT_DATA_KEY);
3222         btrfs_update_inode(trans, root, inode);
3223
3224         ret = btrfs_orphan_del(trans, inode);
3225         BUG_ON(ret);
3226
3227 out:
3228         nr = trans->blocks_used;
3229         ret = btrfs_end_transaction_throttle(trans, root);
3230         BUG_ON(ret);
3231         btrfs_btree_balance_dirty(root, nr);
3232 }
3233
3234 /*
3235  * Invalidate a single dcache entry at the root of the filesystem.
3236  * Needed after creation of snapshot or subvolume.
3237  */
3238 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3239                                   int namelen)
3240 {
3241         struct dentry *alias, *entry;
3242         struct qstr qstr;
3243
3244         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3245         if (alias) {
3246                 qstr.name = name;
3247                 qstr.len = namelen;
3248                 /* change me if btrfs ever gets a d_hash operation */
3249                 qstr.hash = full_name_hash(qstr.name, qstr.len);
3250                 entry = d_lookup(alias, &qstr);
3251                 dput(alias);
3252                 if (entry) {
3253                         d_invalidate(entry);
3254                         dput(entry);
3255                 }
3256         }
3257 }
3258
3259 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3260                 struct btrfs_trans_handle *trans, u64 new_dirid,
3261                 struct btrfs_block_group_cache *block_group)
3262 {
3263         struct inode *inode;
3264         u64 index = 0;
3265
3266         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3267                                 new_dirid, block_group, S_IFDIR | 0700, &index);
3268         if (IS_ERR(inode))
3269                 return PTR_ERR(inode);
3270         inode->i_op = &btrfs_dir_inode_operations;
3271         inode->i_fop = &btrfs_dir_file_operations;
3272         new_root->inode = inode;
3273
3274         inode->i_nlink = 1;
3275         btrfs_i_size_write(inode, 0);
3276
3277         return btrfs_update_inode(trans, new_root, inode);
3278 }
3279
3280 unsigned long btrfs_force_ra(struct address_space *mapping,
3281                               struct file_ra_state *ra, struct file *file,
3282                               pgoff_t offset, pgoff_t last_index)
3283 {
3284         pgoff_t req_size = last_index - offset + 1;
3285
3286 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3287         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3288         return offset;
3289 #else
3290         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3291         return offset + req_size;
3292 #endif
3293 }
3294
3295 struct inode *btrfs_alloc_inode(struct super_block *sb)
3296 {
3297         struct btrfs_inode *ei;
3298
3299         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3300         if (!ei)
3301                 return NULL;
3302         ei->last_trans = 0;
3303         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3304         ei->i_acl = BTRFS_ACL_NOT_CACHED;
3305         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3306         INIT_LIST_HEAD(&ei->i_orphan);
3307         return &ei->vfs_inode;
3308 }
3309
3310 void btrfs_destroy_inode(struct inode *inode)
3311 {
3312         struct btrfs_ordered_extent *ordered;
3313         WARN_ON(!list_empty(&inode->i_dentry));
3314         WARN_ON(inode->i_data.nrpages);
3315
3316         if (BTRFS_I(inode)->i_acl &&
3317             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3318                 posix_acl_release(BTRFS_I(inode)->i_acl);
3319         if (BTRFS_I(inode)->i_default_acl &&
3320             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3321                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3322
3323         spin_lock(&BTRFS_I(inode)->root->list_lock);
3324         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3325                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3326                        " list\n", inode->i_ino);
3327                 dump_stack();
3328         }
3329         spin_unlock(&BTRFS_I(inode)->root->list_lock);
3330
3331         while(1) {
3332                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3333                 if (!ordered)
3334                         break;
3335                 else {
3336                         printk("found ordered extent %Lu %Lu\n",
3337                                ordered->file_offset, ordered->len);
3338                         btrfs_remove_ordered_extent(inode, ordered);
3339                         btrfs_put_ordered_extent(ordered);
3340                         btrfs_put_ordered_extent(ordered);
3341                 }
3342         }
3343         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3344         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3345 }
3346
3347 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3348 static void init_once(void *foo)
3349 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3350 static void init_once(struct kmem_cache * cachep, void *foo)
3351 #else
3352 static void init_once(void * foo, struct kmem_cache * cachep,
3353                       unsigned long flags)
3354 #endif
3355 {
3356         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3357
3358         inode_init_once(&ei->vfs_inode);
3359 }
3360
3361 void btrfs_destroy_cachep(void)
3362 {
3363         if (btrfs_inode_cachep)
3364                 kmem_cache_destroy(btrfs_inode_cachep);
3365         if (btrfs_trans_handle_cachep)
3366                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3367         if (btrfs_transaction_cachep)
3368                 kmem_cache_destroy(btrfs_transaction_cachep);
3369         if (btrfs_bit_radix_cachep)
3370                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3371         if (btrfs_path_cachep)
3372                 kmem_cache_destroy(btrfs_path_cachep);
3373 }
3374
3375 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3376                                        unsigned long extra_flags,
3377 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3378                                        void (*ctor)(void *)
3379 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3380                                        void (*ctor)(struct kmem_cache *, void *)
3381 #else
3382                                        void (*ctor)(void *, struct kmem_cache *,
3383                                                     unsigned long)
3384 #endif
3385                                      )
3386 {
3387         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3388                                  SLAB_MEM_SPREAD | extra_flags), ctor
3389 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3390                                  ,NULL
3391 #endif
3392                                 );
3393 }
3394
3395 int btrfs_init_cachep(void)
3396 {
3397         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3398                                           sizeof(struct btrfs_inode),
3399                                           0, init_once);
3400         if (!btrfs_inode_cachep)
3401                 goto fail;
3402         btrfs_trans_handle_cachep =
3403                         btrfs_cache_create("btrfs_trans_handle_cache",
3404                                            sizeof(struct btrfs_trans_handle),
3405                                            0, NULL);
3406         if (!btrfs_trans_handle_cachep)
3407                 goto fail;
3408         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3409                                              sizeof(struct btrfs_transaction),
3410                                              0, NULL);
3411         if (!btrfs_transaction_cachep)
3412                 goto fail;
3413         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3414                                          sizeof(struct btrfs_path),
3415                                          0, NULL);
3416         if (!btrfs_path_cachep)
3417                 goto fail;
3418         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3419                                               SLAB_DESTROY_BY_RCU, NULL);
3420         if (!btrfs_bit_radix_cachep)
3421                 goto fail;
3422         return 0;
3423 fail:
3424         btrfs_destroy_cachep();
3425         return -ENOMEM;
3426 }
3427
3428 static int btrfs_getattr(struct vfsmount *mnt,
3429                          struct dentry *dentry, struct kstat *stat)
3430 {
3431         struct inode *inode = dentry->d_inode;
3432         generic_fillattr(inode, stat);
3433         stat->blksize = PAGE_CACHE_SIZE;
3434         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3435         return 0;
3436 }
3437
3438 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3439                            struct inode * new_dir,struct dentry *new_dentry)
3440 {
3441         struct btrfs_trans_handle *trans;
3442         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3443         struct inode *new_inode = new_dentry->d_inode;
3444         struct inode *old_inode = old_dentry->d_inode;
3445         struct timespec ctime = CURRENT_TIME;
3446         u64 index = 0;
3447         int ret;
3448
3449         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3450             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3451                 return -ENOTEMPTY;
3452         }
3453
3454         ret = btrfs_check_free_space(root, 1, 0);
3455         if (ret)
3456                 goto out_unlock;
3457
3458         trans = btrfs_start_transaction(root, 1);
3459
3460         btrfs_set_trans_block_group(trans, new_dir);
3461
3462         old_dentry->d_inode->i_nlink++;
3463         old_dir->i_ctime = old_dir->i_mtime = ctime;
3464         new_dir->i_ctime = new_dir->i_mtime = ctime;
3465         old_inode->i_ctime = ctime;
3466
3467         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3468         if (ret)
3469                 goto out_fail;
3470
3471         if (new_inode) {
3472                 new_inode->i_ctime = CURRENT_TIME;
3473                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3474                 if (ret)
3475                         goto out_fail;
3476                 if (new_inode->i_nlink == 0) {
3477                         ret = btrfs_orphan_add(trans, new_inode);
3478                         if (ret)
3479                                 goto out_fail;
3480                 }
3481         }
3482         ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3483         if (ret)
3484                 goto out_fail;
3485
3486         ret = btrfs_add_link(trans, new_dentry, old_inode, 1, index);
3487         if (ret)
3488                 goto out_fail;
3489
3490 out_fail:
3491         btrfs_end_transaction_throttle(trans, root);
3492 out_unlock:
3493         return ret;
3494 }
3495
3496 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
3497 {
3498         struct list_head *head = &root->fs_info->delalloc_inodes;
3499         struct btrfs_inode *binode;
3500         unsigned long flags;
3501
3502         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3503         while(!list_empty(head)) {
3504                 binode = list_entry(head->next, struct btrfs_inode,
3505                                     delalloc_inodes);
3506                 atomic_inc(&binode->vfs_inode.i_count);
3507                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3508                 filemap_write_and_wait(binode->vfs_inode.i_mapping);
3509                 iput(&binode->vfs_inode);
3510                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3511         }
3512         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3513         return 0;
3514 }
3515
3516 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3517                          const char *symname)
3518 {
3519         struct btrfs_trans_handle *trans;
3520         struct btrfs_root *root = BTRFS_I(dir)->root;
3521         struct btrfs_path *path;
3522         struct btrfs_key key;
3523         struct inode *inode = NULL;
3524         int err;
3525         int drop_inode = 0;
3526         u64 objectid;
3527         u64 index = 0 ;
3528         int name_len;
3529         int datasize;
3530         unsigned long ptr;
3531         struct btrfs_file_extent_item *ei;
3532         struct extent_buffer *leaf;
3533         unsigned long nr = 0;
3534
3535         name_len = strlen(symname) + 1;
3536         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3537                 return -ENAMETOOLONG;
3538
3539         err = btrfs_check_free_space(root, 1, 0);
3540         if (err)
3541                 goto out_fail;
3542
3543         trans = btrfs_start_transaction(root, 1);
3544         btrfs_set_trans_block_group(trans, dir);
3545
3546         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3547         if (err) {
3548                 err = -ENOSPC;
3549                 goto out_unlock;
3550         }
3551
3552         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3553                                 dentry->d_name.len,
3554                                 dentry->d_parent->d_inode->i_ino, objectid,
3555                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
3556                                 &index);
3557         err = PTR_ERR(inode);
3558         if (IS_ERR(inode))
3559                 goto out_unlock;
3560
3561         err = btrfs_init_acl(inode, dir);
3562         if (err) {
3563                 drop_inode = 1;
3564                 goto out_unlock;
3565         }
3566
3567         btrfs_set_trans_block_group(trans, inode);
3568         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3569         if (err)
3570                 drop_inode = 1;
3571         else {
3572                 inode->i_mapping->a_ops = &btrfs_aops;
3573                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3574                 inode->i_fop = &btrfs_file_operations;
3575                 inode->i_op = &btrfs_file_inode_operations;
3576                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3577                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3578                                      inode->i_mapping, GFP_NOFS);
3579                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3580                                      inode->i_mapping, GFP_NOFS);
3581                 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3582                 mutex_init(&BTRFS_I(inode)->csum_mutex);
3583                 mutex_init(&BTRFS_I(inode)->extent_mutex);
3584                 BTRFS_I(inode)->delalloc_bytes = 0;
3585                 BTRFS_I(inode)->disk_i_size = 0;
3586                 inode->i_mapping->writeback_index = 0;
3587                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3588                 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3589         }
3590         dir->i_sb->s_dirt = 1;
3591         btrfs_update_inode_block_group(trans, inode);
3592         btrfs_update_inode_block_group(trans, dir);
3593         if (drop_inode)
3594                 goto out_unlock;
3595
3596         path = btrfs_alloc_path();
3597         BUG_ON(!path);
3598         key.objectid = inode->i_ino;
3599         key.offset = 0;
3600         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3601         datasize = btrfs_file_extent_calc_inline_size(name_len);
3602         err = btrfs_insert_empty_item(trans, root, path, &key,
3603                                       datasize);
3604         if (err) {
3605                 drop_inode = 1;
3606                 goto out_unlock;
3607         }
3608         leaf = path->nodes[0];
3609         ei = btrfs_item_ptr(leaf, path->slots[0],
3610                             struct btrfs_file_extent_item);
3611         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3612         btrfs_set_file_extent_type(leaf, ei,
3613                                    BTRFS_FILE_EXTENT_INLINE);
3614         ptr = btrfs_file_extent_inline_start(ei);
3615         write_extent_buffer(leaf, symname, ptr, name_len);
3616         btrfs_mark_buffer_dirty(leaf);
3617         btrfs_free_path(path);
3618
3619         inode->i_op = &btrfs_symlink_inode_operations;
3620         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3621         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3622         btrfs_i_size_write(inode, name_len - 1);
3623         err = btrfs_update_inode(trans, root, inode);
3624         if (err)
3625                 drop_inode = 1;
3626
3627 out_unlock:
3628         nr = trans->blocks_used;
3629         btrfs_end_transaction_throttle(trans, root);
3630 out_fail:
3631         if (drop_inode) {
3632                 inode_dec_link_count(inode);
3633                 iput(inode);
3634         }
3635         btrfs_btree_balance_dirty(root, nr);
3636         return err;
3637 }
3638
3639 static int btrfs_set_page_dirty(struct page *page)
3640 {
3641         return __set_page_dirty_nobuffers(page);
3642 }
3643
3644 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3645 static int btrfs_permission(struct inode *inode, int mask)
3646 #else
3647 static int btrfs_permission(struct inode *inode, int mask,
3648                             struct nameidata *nd)
3649 #endif
3650 {
3651         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3652                 return -EACCES;
3653         return generic_permission(inode, mask, btrfs_check_acl);
3654 }
3655
3656 static struct inode_operations btrfs_dir_inode_operations = {
3657         .lookup         = btrfs_lookup,
3658         .create         = btrfs_create,
3659         .unlink         = btrfs_unlink,
3660         .link           = btrfs_link,
3661         .mkdir          = btrfs_mkdir,
3662         .rmdir          = btrfs_rmdir,
3663         .rename         = btrfs_rename,
3664         .symlink        = btrfs_symlink,
3665         .setattr        = btrfs_setattr,
3666         .mknod          = btrfs_mknod,
3667         .setxattr       = generic_setxattr,
3668         .getxattr       = generic_getxattr,
3669         .listxattr      = btrfs_listxattr,
3670         .removexattr    = generic_removexattr,
3671         .permission     = btrfs_permission,
3672 };
3673 static struct inode_operations btrfs_dir_ro_inode_operations = {
3674         .lookup         = btrfs_lookup,
3675         .permission     = btrfs_permission,
3676 };
3677 static struct file_operations btrfs_dir_file_operations = {
3678         .llseek         = generic_file_llseek,
3679         .read           = generic_read_dir,
3680 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
3681         .readdir        = btrfs_nfshack_readdir,
3682 #else /* NFSd readdir/lookup deadlock is fixed */
3683         .readdir        = btrfs_real_readdir,
3684 #endif
3685         .unlocked_ioctl = btrfs_ioctl,
3686 #ifdef CONFIG_COMPAT
3687         .compat_ioctl   = btrfs_ioctl,
3688 #endif
3689         .release        = btrfs_release_file,
3690 };
3691
3692 static struct extent_io_ops btrfs_extent_io_ops = {
3693         .fill_delalloc = run_delalloc_range,
3694         .submit_bio_hook = btrfs_submit_bio_hook,
3695         .merge_bio_hook = btrfs_merge_bio_hook,
3696         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3697         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3698         .writepage_start_hook = btrfs_writepage_start_hook,
3699         .readpage_io_failed_hook = btrfs_io_failed_hook,
3700         .set_bit_hook = btrfs_set_bit_hook,
3701         .clear_bit_hook = btrfs_clear_bit_hook,
3702 };
3703
3704 static struct address_space_operations btrfs_aops = {
3705         .readpage       = btrfs_readpage,
3706         .writepage      = btrfs_writepage,
3707         .writepages     = btrfs_writepages,
3708         .readpages      = btrfs_readpages,
3709         .sync_page      = block_sync_page,
3710         .bmap           = btrfs_bmap,
3711         .direct_IO      = btrfs_direct_IO,
3712         .invalidatepage = btrfs_invalidatepage,
3713         .releasepage    = btrfs_releasepage,
3714         .set_page_dirty = btrfs_set_page_dirty,
3715 };
3716
3717 static struct address_space_operations btrfs_symlink_aops = {
3718         .readpage       = btrfs_readpage,
3719         .writepage      = btrfs_writepage,
3720         .invalidatepage = btrfs_invalidatepage,
3721         .releasepage    = btrfs_releasepage,
3722 };
3723
3724 static struct inode_operations btrfs_file_inode_operations = {
3725         .truncate       = btrfs_truncate,
3726         .getattr        = btrfs_getattr,
3727         .setattr        = btrfs_setattr,
3728         .setxattr       = generic_setxattr,
3729         .getxattr       = generic_getxattr,
3730         .listxattr      = btrfs_listxattr,
3731         .removexattr    = generic_removexattr,
3732         .permission     = btrfs_permission,
3733 };
3734 static struct inode_operations btrfs_special_inode_operations = {
3735         .getattr        = btrfs_getattr,
3736         .setattr        = btrfs_setattr,
3737         .permission     = btrfs_permission,
3738         .setxattr       = generic_setxattr,
3739         .getxattr       = generic_getxattr,
3740         .listxattr      = btrfs_listxattr,
3741         .removexattr    = generic_removexattr,
3742 };
3743 static struct inode_operations btrfs_symlink_inode_operations = {
3744         .readlink       = generic_readlink,
3745         .follow_link    = page_follow_link_light,
3746         .put_link       = page_put_link,
3747         .permission     = btrfs_permission,
3748 };