fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/version.h>
  38 #include <linux/xattr.h>
  39 #include <linux/posix_acl.h>
  40 #include "ctree.h"
  41 #include "disk-io.h"
  42 #include "transaction.h"
  43 #include "btrfs_inode.h"
  44 #include "ioctl.h"
  45 #include "print-tree.h"
  46 #include "volumes.h"
  47 #include "ordered-data.h"
  48
  49 struct btrfs_iget_args {
  50         u64 ino;
  51         struct btrfs_root *root;
  52 };
  53
  54 static struct inode_operations btrfs_dir_inode_operations;
  55 static struct inode_operations btrfs_symlink_inode_operations;
  56 static struct inode_operations btrfs_dir_ro_inode_operations;
  57 static struct inode_operations btrfs_special_inode_operations;
  58 static struct inode_operations btrfs_file_inode_operations;
  59 static struct address_space_operations btrfs_aops;
  60 static struct address_space_operations btrfs_symlink_aops;
  61 static struct file_operations btrfs_dir_file_operations;
  62 static struct extent_io_ops btrfs_extent_io_ops;
  63
  64 static struct kmem_cache *btrfs_inode_cachep;
  65 struct kmem_cache *btrfs_trans_handle_cachep;
  66 struct kmem_cache *btrfs_transaction_cachep;
  67 struct kmem_cache *btrfs_bit_radix_cachep;
  68 struct kmem_cache *btrfs_path_cachep;
  69
  70 #define S_SHIFT 12
  71 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  72         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  73         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  74         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  75         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  76         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  77         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  78         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  79 };
  80
  81 static void btrfs_truncate(struct inode *inode);
  82
  83 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  84                            int for_del)
  85 {
  86         u64 total;
  87         u64 used;
  88         u64 thresh;
  89         unsigned long flags;
  90         int ret = 0;
  91
  92         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
  93         total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  94         used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  95         if (for_del)
  96                 thresh = total * 90;
  97         else
  98                 thresh = total * 85;
  99
 100         do_div(thresh, 100);
 101
 102         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
 103                 ret = -ENOSPC;
 104         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 105         return ret;
 106 }
 107
 108 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 109 {
 110         struct btrfs_root *root = BTRFS_I(inode)->root;
 111         struct btrfs_trans_handle *trans;
 112         u64 alloc_hint = 0;
 113         u64 num_bytes;
 114         u64 cur_alloc_size;
 115         u64 blocksize = root->sectorsize;
 116         u64 orig_num_bytes;
 117         struct btrfs_key ins;
 118         struct extent_map *em;
 119         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 120         int ret = 0;
 121
 122         trans = btrfs_join_transaction(root, 1);
 123         BUG_ON(!trans);
 124         btrfs_set_trans_block_group(trans, inode);
 125
 126         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 127         num_bytes = max(blocksize,  num_bytes);
 128         orig_num_bytes = num_bytes;
 129
 130         if (alloc_hint == EXTENT_MAP_INLINE)
 131                 goto out;
 132
 133         BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
 134         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 135         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1);
 136         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 137
 138         while(num_bytes > 0) {
 139                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 140                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 141                                            root->sectorsize, 0, 0,
 142                                            (u64)-1, &ins, 1);
 143                 if (ret) {
 144                         WARN_ON(1);
 145                         goto out;
 146                 }
 147                 em = alloc_extent_map(GFP_NOFS);
 148                 em->start = start;
 149                 em->len = ins.offset;
 150                 em->block_start = ins.objectid;
 151                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 152                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
 153                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 154                 while(1) {
 155                         spin_lock(&em_tree->lock);
 156                         ret = add_extent_mapping(em_tree, em);
 157                         spin_unlock(&em_tree->lock);
 158                         if (ret != -EEXIST) {
 159                                 free_extent_map(em);
 160                                 break;
 161                         }
 162                         btrfs_drop_extent_cache(inode, start,
 163                                                 start + ins.offset - 1);
 164                 }
 165                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 166
 167                 cur_alloc_size = ins.offset;
 168                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 169                                                ins.offset);
 170                 BUG_ON(ret);
 171                 if (num_bytes < cur_alloc_size) {
 172                         printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
 173                                cur_alloc_size);
 174                         break;
 175                 }
 176                 num_bytes -= cur_alloc_size;
 177                 alloc_hint = ins.objectid + ins.offset;
 178                 start += cur_alloc_size;
 179         }
 180 out:
 181         btrfs_end_transaction(trans, root);
 182         return ret;
 183 }
 184
 185 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 186 {
 187         u64 extent_start;
 188         u64 extent_end;
 189         u64 bytenr;
 190         u64 cow_end;
 191         u64 loops = 0;
 192         u64 total_fs_bytes;
 193         struct btrfs_root *root = BTRFS_I(inode)->root;
 194         struct btrfs_block_group_cache *block_group;
 195         struct extent_buffer *leaf;
 196         int found_type;
 197         struct btrfs_path *path;
 198         struct btrfs_file_extent_item *item;
 199         int ret;
 200         int err;
 201         struct btrfs_key found_key;
 202
 203         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 204         path = btrfs_alloc_path();
 205         BUG_ON(!path);
 206 again:
 207         ret = btrfs_lookup_file_extent(NULL, root, path,
 208                                        inode->i_ino, start, 0);
 209         if (ret < 0) {
 210                 btrfs_free_path(path);
 211                 return ret;
 212         }
 213
 214         cow_end = end;
 215         if (ret != 0) {
 216                 if (path->slots[0] == 0)
 217                         goto not_found;
 218                 path->slots[0]--;
 219         }
 220
 221         leaf = path->nodes[0];
 222         item = btrfs_item_ptr(leaf, path->slots[0],
 223                               struct btrfs_file_extent_item);
 224
 225         /* are we inside the extent that was found? */
 226         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 227         found_type = btrfs_key_type(&found_key);
 228         if (found_key.objectid != inode->i_ino ||
 229             found_type != BTRFS_EXTENT_DATA_KEY)
 230                 goto not_found;
 231
 232         found_type = btrfs_file_extent_type(leaf, item);
 233         extent_start = found_key.offset;
 234         if (found_type == BTRFS_FILE_EXTENT_REG) {
 235                 u64 extent_num_bytes;
 236
 237                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 238                 extent_end = extent_start + extent_num_bytes;
 239                 err = 0;
 240
 241                 if (loops && start != extent_start)
 242                         goto not_found;
 243
 244                 if (start < extent_start || start >= extent_end)
 245                         goto not_found;
 246
 247                 cow_end = min(end, extent_end - 1);
 248                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 249                 if (bytenr == 0)
 250                         goto not_found;
 251
 252                 if (btrfs_cross_ref_exists(root, &found_key, bytenr))
 253                         goto not_found;
 254                 /*
 255                  * we may be called by the resizer, make sure we're inside
 256                  * the limits of the FS
 257                  */
 258                 block_group = btrfs_lookup_block_group(root->fs_info,
 259                                                        bytenr);
 260                 if (!block_group || block_group->ro)
 261                         goto not_found;
 262
 263                 start = extent_end;
 264         } else {
 265                 goto not_found;
 266         }
 267 loop:
 268         if (start > end) {
 269                 btrfs_free_path(path);
 270                 return 0;
 271         }
 272         btrfs_release_path(root, path);
 273         loops++;
 274         goto again;
 275
 276 not_found:
 277         btrfs_release_path(root, path);
 278         cow_file_range(inode, start, end);
 279         start = end + 1;
 280         goto loop;
 281 }
 282
 283 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 284 {
 285         struct btrfs_root *root = BTRFS_I(inode)->root;
 286         int ret;
 287
 288         if (btrfs_test_opt(root, NODATACOW) ||
 289             btrfs_test_flag(inode, NODATACOW))
 290                 ret = run_delalloc_nocow(inode, start, end);
 291         else
 292                 ret = cow_file_range(inode, start, end);
 293
 294         return ret;
 295 }
 296
 297 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 298                        unsigned long old, unsigned long bits)
 299 {
 300         unsigned long flags;
 301         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 302                 struct btrfs_root *root = BTRFS_I(inode)->root;
 303                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 304                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 305                 root->fs_info->delalloc_bytes += end - start + 1;
 306                 if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 307                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
 308                                       &root->fs_info->delalloc_inodes);
 309                 }
 310                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 311         }
 312         return 0;
 313 }
 314
 315 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 316                          unsigned long old, unsigned long bits)
 317 {
 318         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 319                 struct btrfs_root *root = BTRFS_I(inode)->root;
 320                 unsigned long flags;
 321
 322                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 323                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 324                         printk("warning: delalloc account %Lu %Lu\n",
 325                                end - start + 1, root->fs_info->delalloc_bytes);
 326                         root->fs_info->delalloc_bytes = 0;
 327                         BTRFS_I(inode)->delalloc_bytes = 0;
 328                 } else {
 329                         root->fs_info->delalloc_bytes -= end - start + 1;
 330                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 331                 }
 332                 if (BTRFS_I(inode)->delalloc_bytes == 0 &&
 333                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
 334                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
 335                 }
 336                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 337         }
 338         return 0;
 339 }
 340
 341 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 342                          size_t size, struct bio *bio)
 343 {
 344         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 345         struct btrfs_mapping_tree *map_tree;
 346         u64 logical = bio->bi_sector << 9;
 347         u64 length = 0;
 348         u64 map_length;
 349         int ret;
 350
 351         length = bio->bi_size;
 352         map_tree = &root->fs_info->mapping_tree;
 353         map_length = length;
 354         ret = btrfs_map_block(map_tree, READ, logical,
 355                               &map_length, NULL, 0);
 356
 357         if (map_length < length + size) {
 358                 return 1;
 359         }
 360         return 0;
 361 }
 362
 363 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 364                           int mirror_num)
 365 {
 366         struct btrfs_root *root = BTRFS_I(inode)->root;
 367         int ret = 0;
 368
 369         ret = btrfs_csum_one_bio(root, inode, bio);
 370         BUG_ON(ret);
 371
 372         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
 373 }
 374
 375 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 376                           int mirror_num)
 377 {
 378         struct btrfs_root *root = BTRFS_I(inode)->root;
 379         int ret = 0;
 380
 381         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 382         BUG_ON(ret);
 383
 384         if (!(rw & (1 << BIO_RW))) {
 385                 goto mapit;
 386         }
 387
 388         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 389                                    inode, rw, bio, mirror_num,
 390                                    __btrfs_submit_bio_hook);
 391 mapit:
 392         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
 393 }
 394
 395 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
 396                              struct inode *inode, u64 file_offset,
 397                              struct list_head *list)
 398 {
 399         struct list_head *cur;
 400         struct btrfs_ordered_sum *sum;
 401
 402         btrfs_set_trans_block_group(trans, inode);
 403         list_for_each(cur, list) {
 404                 sum = list_entry(cur, struct btrfs_ordered_sum, list);
 405                 mutex_lock(&BTRFS_I(inode)->csum_mutex);
 406                 btrfs_csum_file_blocks(trans, BTRFS_I(inode)->root,
 407                                        inode, sum);
 408                 mutex_unlock(&BTRFS_I(inode)->csum_mutex);
 409         }
 410         return 0;
 411 }
 412
 413 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end)
 414 {
 415         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
 416                                    GFP_NOFS);
 417 }
 418
 419 struct btrfs_writepage_fixup {
 420         struct page *page;
 421         struct btrfs_work work;
 422 };
 423
 424 /* see btrfs_writepage_start_hook for details on why this is required */
 425 void btrfs_writepage_fixup_worker(struct btrfs_work *work)
 426 {
 427         struct btrfs_writepage_fixup *fixup;
 428         struct btrfs_ordered_extent *ordered;
 429         struct page *page;
 430         struct inode *inode;
 431         u64 page_start;
 432         u64 page_end;
 433
 434         fixup = container_of(work, struct btrfs_writepage_fixup, work);
 435         page = fixup->page;
 436 again:
 437         lock_page(page);
 438         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
 439                 ClearPageChecked(page);
 440                 goto out_page;
 441         }
 442
 443         inode = page->mapping->host;
 444         page_start = page_offset(page);
 445         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
 446
 447         lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 448
 449         /* already ordered? We're done */
 450         if (test_range_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
 451                              EXTENT_ORDERED, 0)) {
 452                 goto out;
 453         }
 454
 455         ordered = btrfs_lookup_ordered_extent(inode, page_start);
 456         if (ordered) {
 457                 unlock_extent(&BTRFS_I(inode)->io_tree, page_start,
 458                               page_end, GFP_NOFS);
 459                 unlock_page(page);
 460                 btrfs_start_ordered_extent(inode, ordered, 1);
 461                 goto again;
 462         }
 463
 464         btrfs_set_extent_delalloc(inode, page_start, page_end);
 465         ClearPageChecked(page);
 466 out:
 467         unlock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end, GFP_NOFS);
 468 out_page:
 469         unlock_page(page);
 470         page_cache_release(page);
 471 }
 472
 473 /*
 474  * There are a few paths in the higher layers of the kernel that directly
 475  * set the page dirty bit without asking the filesystem if it is a
 476  * good idea.  This causes problems because we want to make sure COW
 477  * properly happens and the data=ordered rules are followed.
 478  *
 479  * In our case any range that doesn't have the EXTENT_ORDERED bit set
 480  * hasn't been properly setup for IO.  We kick off an async process
 481  * to fix it up.  The async helper will wait for ordered extents, set
 482  * the delalloc bit and make it safe to write the page.
 483  */
 484 int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
 485 {
 486         struct inode *inode = page->mapping->host;
 487         struct btrfs_writepage_fixup *fixup;
 488         struct btrfs_root *root = BTRFS_I(inode)->root;
 489         int ret;
 490
 491         ret = test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
 492                              EXTENT_ORDERED, 0);
 493         if (ret)
 494                 return 0;
 495
 496         if (PageChecked(page))
 497                 return -EAGAIN;
 498
 499         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
 500         if (!fixup)
 501                 return -EAGAIN;
 502
 503         SetPageChecked(page);
 504         page_cache_get(page);
 505         fixup->work.func = btrfs_writepage_fixup_worker;
 506         fixup->page = page;
 507         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
 508         return -EAGAIN;
 509 }
 510
 511 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
 512 {
 513         struct btrfs_root *root = BTRFS_I(inode)->root;
 514         struct btrfs_trans_handle *trans;
 515         struct btrfs_ordered_extent *ordered_extent;
 516         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 517         u64 alloc_hint = 0;
 518         struct list_head list;
 519         struct btrfs_key ins;
 520         int ret;
 521
 522         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
 523         if (!ret)
 524                 return 0;
 525
 526         trans = btrfs_join_transaction(root, 1);
 527
 528         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 529         BUG_ON(!ordered_extent);
 530
 531         lock_extent(io_tree, ordered_extent->file_offset,
 532                     ordered_extent->file_offset + ordered_extent->len - 1,
 533                     GFP_NOFS);
 534
 535         INIT_LIST_HEAD(&list);
 536
 537         ins.objectid = ordered_extent->start;
 538         ins.offset = ordered_extent->len;
 539         ins.type = BTRFS_EXTENT_ITEM_KEY;
 540
 541         ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
 542                                           trans->transid, inode->i_ino,
 543                                           ordered_extent->file_offset, &ins);
 544         BUG_ON(ret);
 545
 546         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 547
 548         ret = btrfs_drop_extents(trans, root, inode,
 549                                  ordered_extent->file_offset,
 550                                  ordered_extent->file_offset +
 551                                  ordered_extent->len,
 552                                  ordered_extent->file_offset, &alloc_hint);
 553         BUG_ON(ret);
 554         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 555                                        ordered_extent->file_offset,
 556                                        ordered_extent->start,
 557                                        ordered_extent->len,
 558                                        ordered_extent->len, 0);
 559         BUG_ON(ret);
 560
 561         btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 562                                 ordered_extent->file_offset +
 563                                 ordered_extent->len - 1);
 564         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 565
 566         inode->i_blocks += ordered_extent->len >> 9;
 567         unlock_extent(io_tree, ordered_extent->file_offset,
 568                     ordered_extent->file_offset + ordered_extent->len - 1,
 569                     GFP_NOFS);
 570         add_pending_csums(trans, inode, ordered_extent->file_offset,
 571                           &ordered_extent->list);
 572
 573         btrfs_ordered_update_i_size(inode, ordered_extent);
 574         btrfs_remove_ordered_extent(inode, ordered_extent);
 575
 576         /* once for us */
 577         btrfs_put_ordered_extent(ordered_extent);
 578         /* once for the tree */
 579         btrfs_put_ordered_extent(ordered_extent);
 580
 581         btrfs_update_inode(trans, root, inode);
 582         btrfs_end_transaction(trans, root);
 583         return 0;
 584 }
 585
 586 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 587                                 struct extent_state *state, int uptodate)
 588 {
 589         return btrfs_finish_ordered_io(page->mapping->host, start, end);
 590 }
 591
 592 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 593 {
 594         int ret = 0;
 595         struct inode *inode = page->mapping->host;
 596         struct btrfs_root *root = BTRFS_I(inode)->root;
 597         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 598         struct btrfs_csum_item *item;
 599         struct btrfs_path *path = NULL;
 600         u32 csum;
 601
 602         if (btrfs_test_opt(root, NODATASUM) ||
 603             btrfs_test_flag(inode, NODATASUM))
 604                 return 0;
 605
 606         /*
 607          * It is possible there is an ordered extent that has
 608          * not yet finished for this range in the file.  If so,
 609          * that extent will have a csum cached, and it will insert
 610          * the sum after all the blocks in the extent are fully
 611          * on disk.  So, look for an ordered extent and use the
 612          * sum if found.  We have to do this before looking in the
 613          * btree because csum items are pre-inserted based on
 614          * the file size.  btrfs_lookup_csum might find an item
 615          * that still hasn't been fully filled.
 616          */
 617         ret = btrfs_find_ordered_sum(inode, start, &csum);
 618         if (ret == 0)
 619                 goto found;
 620
 621         ret = 0;
 622         path = btrfs_alloc_path();
 623         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 624         if (IS_ERR(item)) {
 625                 ret = PTR_ERR(item);
 626                 /* a csum that isn't present is a preallocated region. */
 627                 if (ret == -ENOENT || ret == -EFBIG)
 628                         ret = 0;
 629                 csum = 0;
 630                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino,
 631                        start);
 632                 goto out;
 633         }
 634         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 635                            BTRFS_CRC32_SIZE);
 636 found:
 637         set_state_private(io_tree, start, csum);
 638 out:
 639         if (path)
 640                 btrfs_free_path(path);
 641         return ret;
 642 }
 643
 644 struct io_failure_record {
 645         struct page *page;
 646         u64 start;
 647         u64 len;
 648         u64 logical;
 649         int last_mirror;
 650 };
 651
 652 int btrfs_io_failed_hook(struct bio *failed_bio,
 653                          struct page *page, u64 start, u64 end,
 654                          struct extent_state *state)
 655 {
 656         struct io_failure_record *failrec = NULL;
 657         u64 private;
 658         struct extent_map *em;
 659         struct inode *inode = page->mapping->host;
 660         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 661         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 662         struct bio *bio;
 663         int num_copies;
 664         int ret;
 665         int rw;
 666         u64 logical;
 667
 668         ret = get_state_private(failure_tree, start, &private);
 669         if (ret) {
 670                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 671                 if (!failrec)
 672                         return -ENOMEM;
 673                 failrec->start = start;
 674                 failrec->len = end - start + 1;
 675                 failrec->last_mirror = 0;
 676
 677                 spin_lock(&em_tree->lock);
 678                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 679                 if (em->start > start || em->start + em->len < start) {
 680                         free_extent_map(em);
 681                         em = NULL;
 682                 }
 683                 spin_unlock(&em_tree->lock);
 684
 685                 if (!em || IS_ERR(em)) {
 686                         kfree(failrec);
 687                         return -EIO;
 688                 }
 689                 logical = start - em->start;
 690                 logical = em->block_start + logical;
 691                 failrec->logical = logical;
 692                 free_extent_map(em);
 693                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 694                                 EXTENT_DIRTY, GFP_NOFS);
 695                 set_state_private(failure_tree, start,
 696                                  (u64)(unsigned long)failrec);
 697         } else {
 698                 failrec = (struct io_failure_record *)(unsigned long)private;
 699         }
 700         num_copies = btrfs_num_copies(
 701                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 702                               failrec->logical, failrec->len);
 703         failrec->last_mirror++;
 704         if (!state) {
 705                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 706                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 707                                                     failrec->start,
 708                                                     EXTENT_LOCKED);
 709                 if (state && state->start != failrec->start)
 710                         state = NULL;
 711                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 712         }
 713         if (!state || failrec->last_mirror > num_copies) {
 714                 set_state_private(failure_tree, failrec->start, 0);
 715                 clear_extent_bits(failure_tree, failrec->start,
 716                                   failrec->start + failrec->len - 1,
 717                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 718                 kfree(failrec);
 719                 return -EIO;
 720         }
 721         bio = bio_alloc(GFP_NOFS, 1);
 722         bio->bi_private = state;
 723         bio->bi_end_io = failed_bio->bi_end_io;
 724         bio->bi_sector = failrec->logical >> 9;
 725         bio->bi_bdev = failed_bio->bi_bdev;
 726         bio->bi_size = 0;
 727         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 728         if (failed_bio->bi_rw & (1 << BIO_RW))
 729                 rw = WRITE;
 730         else
 731                 rw = READ;
 732
 733         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 734                                                       failrec->last_mirror);
 735         return 0;
 736 }
 737
 738 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 739 {
 740         u64 private;
 741         u64 private_failure;
 742         struct io_failure_record *failure;
 743         int ret;
 744
 745         private = 0;
 746         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 747                              (u64)-1, 1, EXTENT_DIRTY)) {
 748                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 749                                         start, &private_failure);
 750                 if (ret == 0) {
 751                         failure = (struct io_failure_record *)(unsigned long)
 752                                    private_failure;
 753                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 754                                           failure->start, 0);
 755                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 756                                           failure->start,
 757                                           failure->start + failure->len - 1,
 758                                           EXTENT_DIRTY | EXTENT_LOCKED,
 759                                           GFP_NOFS);
 760                         kfree(failure);
 761                 }
 762         }
 763         return 0;
 764 }
 765
 766 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 767                                struct extent_state *state)
 768 {
 769         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 770         struct inode *inode = page->mapping->host;
 771         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 772         char *kaddr;
 773         u64 private = ~(u32)0;
 774         int ret;
 775         struct btrfs_root *root = BTRFS_I(inode)->root;
 776         u32 csum = ~(u32)0;
 777         unsigned long flags;
 778
 779         if (btrfs_test_opt(root, NODATASUM) ||
 780             btrfs_test_flag(inode, NODATASUM))
 781                 return 0;
 782         if (state && state->start == start) {
 783                 private = state->private;
 784                 ret = 0;
 785         } else {
 786                 ret = get_state_private(io_tree, start, &private);
 787         }
 788         local_irq_save(flags);
 789         kaddr = kmap_atomic(page, KM_IRQ0);
 790         if (ret) {
 791                 goto zeroit;
 792         }
 793         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 794         btrfs_csum_final(csum, (char *)&csum);
 795         if (csum != private) {
 796                 goto zeroit;
 797         }
 798         kunmap_atomic(kaddr, KM_IRQ0);
 799         local_irq_restore(flags);
 800
 801         /* if the io failure tree for this inode is non-empty,
 802          * check to see if we've recovered from a failed IO
 803          */
 804         btrfs_clean_io_failures(inode, start);
 805         return 0;
 806
 807 zeroit:
 808         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 809                page->mapping->host->i_ino, (unsigned long long)start, csum,
 810                private);
 811         memset(kaddr + offset, 1, end - start + 1);
 812         flush_dcache_page(page);
 813         kunmap_atomic(kaddr, KM_IRQ0);
 814         local_irq_restore(flags);
 815         if (private == 0)
 816                 return 0;
 817         return -EIO;
 818 }
 819
 820 /*
 821  * This creates an orphan entry for the given inode in case something goes
 822  * wrong in the middle of an unlink/truncate.
 823  */
 824 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 825 {
 826         struct btrfs_root *root = BTRFS_I(inode)->root;
 827         int ret = 0;
 828
 829         spin_lock(&root->list_lock);
 830
 831         /* already on the orphan list, we're good */
 832         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 833                 spin_unlock(&root->list_lock);
 834                 return 0;
 835         }
 836
 837         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 838
 839         spin_unlock(&root->list_lock);
 840
 841         /*
 842          * insert an orphan item to track this unlinked/truncated file
 843          */
 844         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
 845
 846         return ret;
 847 }
 848
 849 /*
 850  * We have done the truncate/delete so we can go ahead and remove the orphan
 851  * item for this particular inode.
 852  */
 853 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
 854 {
 855         struct btrfs_root *root = BTRFS_I(inode)->root;
 856         int ret = 0;
 857
 858         spin_lock(&root->list_lock);
 859
 860         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
 861                 spin_unlock(&root->list_lock);
 862                 return 0;
 863         }
 864
 865         list_del_init(&BTRFS_I(inode)->i_orphan);
 866         if (!trans) {
 867                 spin_unlock(&root->list_lock);
 868                 return 0;
 869         }
 870
 871         spin_unlock(&root->list_lock);
 872
 873         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
 874
 875         return ret;
 876 }
 877
 878 /*
 879  * this cleans up any orphans that may be left on the list from the last use
 880  * of this root.
 881  */
 882 void btrfs_orphan_cleanup(struct btrfs_root *root)
 883 {
 884         struct btrfs_path *path;
 885         struct extent_buffer *leaf;
 886         struct btrfs_item *item;
 887         struct btrfs_key key, found_key;
 888         struct btrfs_trans_handle *trans;
 889         struct inode *inode;
 890         int ret = 0, nr_unlink = 0, nr_truncate = 0;
 891
 892         /* don't do orphan cleanup if the fs is readonly. */
 893         if (root->inode->i_sb->s_flags & MS_RDONLY)
 894                 return;
 895
 896         path = btrfs_alloc_path();
 897         if (!path)
 898                 return;
 899         path->reada = -1;
 900
 901         key.objectid = BTRFS_ORPHAN_OBJECTID;
 902         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
 903         key.offset = (u64)-1;
 904
 905         trans = btrfs_start_transaction(root, 1);
 906         btrfs_set_trans_block_group(trans, root->inode);
 907
 908         while (1) {
 909                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 910                 if (ret < 0) {
 911                         printk(KERN_ERR "Error searching slot for orphan: %d"
 912                                "\n", ret);
 913                         break;
 914                 }
 915
 916                 /*
 917                  * if ret == 0 means we found what we were searching for, which
 918                  * is weird, but possible, so only screw with path if we didnt
 919                  * find the key and see if we have stuff that matches
 920                  */
 921                 if (ret > 0) {
 922                         if (path->slots[0] == 0)
 923                                 break;
 924                         path->slots[0]--;
 925                 }
 926
 927                 /* pull out the item */
 928                 leaf = path->nodes[0];
 929                 item = btrfs_item_nr(leaf, path->slots[0]);
 930                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 931
 932                 /* make sure the item matches what we want */
 933                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
 934                         break;
 935                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
 936                         break;
 937
 938                 /* release the path since we're done with it */
 939                 btrfs_release_path(root, path);
 940
 941                 /*
 942                  * this is where we are basically btrfs_lookup, without the
 943                  * crossing root thing.  we store the inode number in the
 944                  * offset of the orphan item.
 945                  */
 946                 inode = btrfs_iget_locked(root->inode->i_sb,
 947                                           found_key.offset, root);
 948                 if (!inode)
 949                         break;
 950
 951                 if (inode->i_state & I_NEW) {
 952                         BTRFS_I(inode)->root = root;
 953
 954                         /* have to set the location manually */
 955                         BTRFS_I(inode)->location.objectid = inode->i_ino;
 956                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
 957                         BTRFS_I(inode)->location.offset = 0;
 958
 959                         btrfs_read_locked_inode(inode);
 960                         unlock_new_inode(inode);
 961                 }
 962
 963                 /*
 964                  * add this inode to the orphan list so btrfs_orphan_del does
 965                  * the proper thing when we hit it
 966                  */
 967                 spin_lock(&root->list_lock);
 968                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 969                 spin_unlock(&root->list_lock);
 970
 971                 /*
 972                  * if this is a bad inode, means we actually succeeded in
 973                  * removing the inode, but not the orphan record, which means
 974                  * we need to manually delete the orphan since iput will just
 975                  * do a destroy_inode
 976                  */
 977                 if (is_bad_inode(inode)) {
 978                         btrfs_orphan_del(trans, inode);
 979                         iput(inode);
 980                         continue;
 981                 }
 982
 983                 /* if we have links, this was a truncate, lets do that */
 984                 if (inode->i_nlink) {
 985                         nr_truncate++;
 986                         btrfs_truncate(inode);
 987                 } else {
 988                         nr_unlink++;
 989                 }
 990
 991                 /* this will do delete_inode and everything for us */
 992                 iput(inode);
 993         }
 994
 995         if (nr_unlink)
 996                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
 997         if (nr_truncate)
 998                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
 999
1000         btrfs_free_path(path);
1001         btrfs_end_transaction(trans, root);
1002 }
1003
1004 void btrfs_read_locked_inode(struct inode *inode)
1005 {
1006         struct btrfs_path *path;
1007         struct extent_buffer *leaf;
1008         struct btrfs_inode_item *inode_item;
1009         struct btrfs_timespec *tspec;
1010         struct btrfs_root *root = BTRFS_I(inode)->root;
1011         struct btrfs_key location;
1012         u64 alloc_group_block;
1013         u32 rdev;
1014         int ret;
1015
1016         path = btrfs_alloc_path();
1017         BUG_ON(!path);
1018         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
1019
1020         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
1021         if (ret)
1022                 goto make_bad;
1023
1024         leaf = path->nodes[0];
1025         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1026                                     struct btrfs_inode_item);
1027
1028         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
1029         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
1030         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
1031         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
1032         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
1033
1034         tspec = btrfs_inode_atime(inode_item);
1035         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1036         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1037
1038         tspec = btrfs_inode_mtime(inode_item);
1039         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1040         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1041
1042         tspec = btrfs_inode_ctime(inode_item);
1043         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1044         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1045
1046         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1047         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
1048         inode->i_rdev = 0;
1049         rdev = btrfs_inode_rdev(leaf, inode_item);
1050
1051         BTRFS_I(inode)->index_cnt = (u64)-1;
1052
1053         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1054         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1055                                                        alloc_group_block);
1056         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1057         if (!BTRFS_I(inode)->block_group) {
1058                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1059                                                  NULL, 0,
1060                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1061         }
1062         btrfs_free_path(path);
1063         inode_item = NULL;
1064
1065         switch (inode->i_mode & S_IFMT) {
1066         case S_IFREG:
1067                 inode->i_mapping->a_ops = &btrfs_aops;
1068                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1069                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1070                 inode->i_fop = &btrfs_file_operations;
1071                 inode->i_op = &btrfs_file_inode_operations;
1072                 break;
1073         case S_IFDIR:
1074                 inode->i_fop = &btrfs_dir_file_operations;
1075                 if (root == root->fs_info->tree_root)
1076                         inode->i_op = &btrfs_dir_ro_inode_operations;
1077                 else
1078                         inode->i_op = &btrfs_dir_inode_operations;
1079                 break;
1080         case S_IFLNK:
1081                 inode->i_op = &btrfs_symlink_inode_operations;
1082                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1083                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1084                 break;
1085         default:
1086                 init_special_inode(inode, inode->i_mode, rdev);
1087                 break;
1088         }
1089         return;
1090
1091 make_bad:
1092         btrfs_free_path(path);
1093         make_bad_inode(inode);
1094 }
1095
1096 static void fill_inode_item(struct extent_buffer *leaf,
1097                             struct btrfs_inode_item *item,
1098                             struct inode *inode)
1099 {
1100         btrfs_set_inode_uid(leaf, item, inode->i_uid);
1101         btrfs_set_inode_gid(leaf, item, inode->i_gid);
1102         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1103         btrfs_set_inode_mode(leaf, item, inode->i_mode);
1104         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1105
1106         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1107                                inode->i_atime.tv_sec);
1108         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1109                                 inode->i_atime.tv_nsec);
1110
1111         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1112                                inode->i_mtime.tv_sec);
1113         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1114                                 inode->i_mtime.tv_nsec);
1115
1116         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1117                                inode->i_ctime.tv_sec);
1118         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1119                                 inode->i_ctime.tv_nsec);
1120
1121         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1122         btrfs_set_inode_generation(leaf, item, inode->i_generation);
1123         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1124         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1125         btrfs_set_inode_block_group(leaf, item,
1126                                     BTRFS_I(inode)->block_group->key.objectid);
1127 }
1128
1129 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1130                               struct btrfs_root *root,
1131                               struct inode *inode)
1132 {
1133         struct btrfs_inode_item *inode_item;
1134         struct btrfs_path *path;
1135         struct extent_buffer *leaf;
1136         int ret;
1137
1138         path = btrfs_alloc_path();
1139         BUG_ON(!path);
1140         ret = btrfs_lookup_inode(trans, root, path,
1141                                  &BTRFS_I(inode)->location, 1);
1142         if (ret) {
1143                 if (ret > 0)
1144                         ret = -ENOENT;
1145                 goto failed;
1146         }
1147
1148         leaf = path->nodes[0];
1149         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1150                                   struct btrfs_inode_item);
1151
1152         fill_inode_item(leaf, inode_item, inode);
1153         btrfs_mark_buffer_dirty(leaf);
1154         btrfs_set_inode_last_trans(trans, inode);
1155         ret = 0;
1156 failed:
1157         btrfs_free_path(path);
1158         return ret;
1159 }
1160
1161
1162 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
1163                               struct btrfs_root *root,
1164                               struct inode *dir,
1165                               struct dentry *dentry)
1166 {
1167         struct btrfs_path *path;
1168         const char *name = dentry->d_name.name;
1169         int name_len = dentry->d_name.len;
1170         int ret = 0;
1171         struct extent_buffer *leaf;
1172         struct btrfs_dir_item *di;
1173         struct btrfs_key key;
1174         u64 index;
1175
1176         path = btrfs_alloc_path();
1177         if (!path) {
1178                 ret = -ENOMEM;
1179                 goto err;
1180         }
1181
1182         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1183                                     name, name_len, -1);
1184         if (IS_ERR(di)) {
1185                 ret = PTR_ERR(di);
1186                 goto err;
1187         }
1188         if (!di) {
1189                 ret = -ENOENT;
1190                 goto err;
1191         }
1192         leaf = path->nodes[0];
1193         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1194         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1195         if (ret)
1196                 goto err;
1197         btrfs_release_path(root, path);
1198
1199         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1200                                   dentry->d_inode->i_ino,
1201                                   dentry->d_parent->d_inode->i_ino, &index);
1202         if (ret) {
1203                 printk("failed to delete reference to %.*s, "
1204                        "inode %lu parent %lu\n", name_len, name,
1205                        dentry->d_inode->i_ino,
1206                        dentry->d_parent->d_inode->i_ino);
1207                 goto err;
1208         }
1209
1210         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1211                                          index, name, name_len, -1);
1212         if (IS_ERR(di)) {
1213                 ret = PTR_ERR(di);
1214                 goto err;
1215         }
1216         if (!di) {
1217                 ret = -ENOENT;
1218                 goto err;
1219         }
1220         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1221         btrfs_release_path(root, path);
1222
1223         dentry->d_inode->i_ctime = dir->i_ctime;
1224 err:
1225         btrfs_free_path(path);
1226         if (!ret) {
1227                 btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1228                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1229                 btrfs_update_inode(trans, root, dir);
1230 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1231                 dentry->d_inode->i_nlink--;
1232 #else
1233                 drop_nlink(dentry->d_inode);
1234 #endif
1235                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
1236                 dir->i_sb->s_dirt = 1;
1237         }
1238         return ret;
1239 }
1240
1241 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1242 {
1243         struct btrfs_root *root;
1244         struct btrfs_trans_handle *trans;
1245         struct inode *inode = dentry->d_inode;
1246         int ret;
1247         unsigned long nr = 0;
1248
1249         root = BTRFS_I(dir)->root;
1250
1251         ret = btrfs_check_free_space(root, 1, 1);
1252         if (ret)
1253                 goto fail;
1254
1255         trans = btrfs_start_transaction(root, 1);
1256
1257         btrfs_set_trans_block_group(trans, dir);
1258         ret = btrfs_unlink_trans(trans, root, dir, dentry);
1259
1260         if (inode->i_nlink == 0)
1261                 ret = btrfs_orphan_add(trans, inode);
1262
1263         nr = trans->blocks_used;
1264
1265         btrfs_end_transaction_throttle(trans, root);
1266 fail:
1267         btrfs_btree_balance_dirty(root, nr);
1268         return ret;
1269 }
1270
1271 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1272 {
1273         struct inode *inode = dentry->d_inode;
1274         int err = 0;
1275         int ret;
1276         struct btrfs_root *root = BTRFS_I(dir)->root;
1277         struct btrfs_trans_handle *trans;
1278         unsigned long nr = 0;
1279
1280         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1281                 return -ENOTEMPTY;
1282         }
1283
1284         ret = btrfs_check_free_space(root, 1, 1);
1285         if (ret)
1286                 goto fail;
1287
1288         trans = btrfs_start_transaction(root, 1);
1289         btrfs_set_trans_block_group(trans, dir);
1290
1291         err = btrfs_orphan_add(trans, inode);
1292         if (err)
1293                 goto fail_trans;
1294
1295         /* now the directory is empty */
1296         err = btrfs_unlink_trans(trans, root, dir, dentry);
1297         if (!err) {
1298                 btrfs_i_size_write(inode, 0);
1299         }
1300
1301 fail_trans:
1302         nr = trans->blocks_used;
1303         ret = btrfs_end_transaction_throttle(trans, root);
1304 fail:
1305         btrfs_btree_balance_dirty(root, nr);
1306
1307         if (ret && !err)
1308                 err = ret;
1309         return err;
1310 }
1311
1312 /*
1313  * this can truncate away extent items, csum items and directory items.
1314  * It starts at a high offset and removes keys until it can't find
1315  * any higher than i_size.
1316  *
1317  * csum items that cross the new i_size are truncated to the new size
1318  * as well.
1319  *
1320  * min_type is the minimum key type to truncate down to.  If set to 0, this
1321  * will kill all the items on this inode, including the INODE_ITEM_KEY.
1322  */
1323 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
1324                                    struct btrfs_root *root,
1325                                    struct inode *inode,
1326                                    u32 min_type)
1327 {
1328         int ret;
1329         struct btrfs_path *path;
1330         struct btrfs_key key;
1331         struct btrfs_key found_key;
1332         u32 found_type;
1333         struct extent_buffer *leaf;
1334         struct btrfs_file_extent_item *fi;
1335         u64 extent_start = 0;
1336         u64 extent_num_bytes = 0;
1337         u64 item_end = 0;
1338         u64 root_gen = 0;
1339         u64 root_owner = 0;
1340         int found_extent;
1341         int del_item;
1342         int pending_del_nr = 0;
1343         int pending_del_slot = 0;
1344         int extent_type = -1;
1345         u64 mask = root->sectorsize - 1;
1346
1347         btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
1348         path = btrfs_alloc_path();
1349         path->reada = -1;
1350         BUG_ON(!path);
1351
1352         /* FIXME, add redo link to tree so we don't leak on crash */
1353         key.objectid = inode->i_ino;
1354         key.offset = (u64)-1;
1355         key.type = (u8)-1;
1356
1357         btrfs_init_path(path);
1358 search_again:
1359         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1360         if (ret < 0) {
1361                 goto error;
1362         }
1363         if (ret > 0) {
1364                 BUG_ON(path->slots[0] == 0);
1365                 path->slots[0]--;
1366         }
1367
1368         while(1) {
1369                 fi = NULL;
1370                 leaf = path->nodes[0];
1371                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1372                 found_type = btrfs_key_type(&found_key);
1373
1374                 if (found_key.objectid != inode->i_ino)
1375                         break;
1376
1377                 if (found_type < min_type)
1378                         break;
1379
1380                 item_end = found_key.offset;
1381                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1382                         fi = btrfs_item_ptr(leaf, path->slots[0],
1383                                             struct btrfs_file_extent_item);
1384                         extent_type = btrfs_file_extent_type(leaf, fi);
1385                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1386                                 item_end +=
1387                                     btrfs_file_extent_num_bytes(leaf, fi);
1388                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1389                                 struct btrfs_item *item = btrfs_item_nr(leaf,
1390                                                                 path->slots[0]);
1391                                 item_end += btrfs_file_extent_inline_len(leaf,
1392                                                                          item);
1393                         }
1394                         item_end--;
1395                 }
1396                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1397                         ret = btrfs_csum_truncate(trans, root, path,
1398                                                   inode->i_size);
1399                         BUG_ON(ret);
1400                 }
1401                 if (item_end < inode->i_size) {
1402                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1403                                 found_type = BTRFS_INODE_ITEM_KEY;
1404                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1405                                 found_type = BTRFS_CSUM_ITEM_KEY;
1406                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1407                                 found_type = BTRFS_XATTR_ITEM_KEY;
1408                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1409                                 found_type = BTRFS_INODE_REF_KEY;
1410                         } else if (found_type) {
1411                                 found_type--;
1412                         } else {
1413                                 break;
1414                         }
1415                         btrfs_set_key_type(&key, found_type);
1416                         goto next;
1417                 }
1418                 if (found_key.offset >= inode->i_size)
1419                         del_item = 1;
1420                 else
1421                         del_item = 0;
1422                 found_extent = 0;
1423
1424                 /* FIXME, shrink the extent if the ref count is only 1 */
1425                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1426                         goto delete;
1427
1428                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1429                         u64 num_dec;
1430                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1431                         if (!del_item) {
1432                                 u64 orig_num_bytes =
1433                                         btrfs_file_extent_num_bytes(leaf, fi);
1434                                 extent_num_bytes = inode->i_size -
1435                                         found_key.offset + root->sectorsize - 1;
1436                                 extent_num_bytes = extent_num_bytes &
1437                                         ~((u64)root->sectorsize - 1);
1438                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1439                                                          extent_num_bytes);
1440                                 num_dec = (orig_num_bytes -
1441                                            extent_num_bytes);
1442                                 if (extent_start != 0)
1443                                         dec_i_blocks(inode, num_dec);
1444                                 btrfs_mark_buffer_dirty(leaf);
1445                         } else {
1446                                 extent_num_bytes =
1447                                         btrfs_file_extent_disk_num_bytes(leaf,
1448                                                                          fi);
1449                                 /* FIXME blocksize != 4096 */
1450                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1451                                 if (extent_start != 0) {
1452                                         found_extent = 1;
1453                                         dec_i_blocks(inode, num_dec);
1454                                 }
1455                                 root_gen = btrfs_header_generation(leaf);
1456                                 root_owner = btrfs_header_owner(leaf);
1457                         }
1458                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1459                         if (!del_item) {
1460                                 u32 newsize = inode->i_size - found_key.offset;
1461                                 dec_i_blocks(inode, item_end + 1 -
1462                                             found_key.offset - newsize);
1463                                 newsize =
1464                                     btrfs_file_extent_calc_inline_size(newsize);
1465                                 ret = btrfs_truncate_item(trans, root, path,
1466                                                           newsize, 1);
1467                                 BUG_ON(ret);
1468                         } else {
1469                                 dec_i_blocks(inode, item_end + 1 -
1470                                              found_key.offset);
1471                         }
1472                 }
1473 delete:
1474                 if (del_item) {
1475                         if (!pending_del_nr) {
1476                                 /* no pending yet, add ourselves */
1477                                 pending_del_slot = path->slots[0];
1478                                 pending_del_nr = 1;
1479                         } else if (pending_del_nr &&
1480                                    path->slots[0] + 1 == pending_del_slot) {
1481                                 /* hop on the pending chunk */
1482                                 pending_del_nr++;
1483                                 pending_del_slot = path->slots[0];
1484                         } else {
1485                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1486                         }
1487                 } else {
1488                         break;
1489                 }
1490                 if (found_extent) {
1491                         ret = btrfs_free_extent(trans, root, extent_start,
1492                                                 extent_num_bytes,
1493                                                 root_owner,
1494                                                 root_gen, inode->i_ino,
1495                                                 found_key.offset, 0);
1496                         BUG_ON(ret);
1497                 }
1498 next:
1499                 if (path->slots[0] == 0) {
1500                         if (pending_del_nr)
1501                                 goto del_pending;
1502                         btrfs_release_path(root, path);
1503                         goto search_again;
1504                 }
1505
1506                 path->slots[0]--;
1507                 if (pending_del_nr &&
1508                     path->slots[0] + 1 != pending_del_slot) {
1509                         struct btrfs_key debug;
1510 del_pending:
1511                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1512                                               pending_del_slot);
1513                         ret = btrfs_del_items(trans, root, path,
1514                                               pending_del_slot,
1515                                               pending_del_nr);
1516                         BUG_ON(ret);
1517                         pending_del_nr = 0;
1518                         btrfs_release_path(root, path);
1519                         goto search_again;
1520                 }
1521         }
1522         ret = 0;
1523 error:
1524         if (pending_del_nr) {
1525                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1526                                       pending_del_nr);
1527         }
1528         btrfs_free_path(path);
1529         inode->i_sb->s_dirt = 1;
1530         return ret;
1531 }
1532
1533 /*
1534  * taken from block_truncate_page, but does cow as it zeros out
1535  * any bytes left in the last page in the file.
1536  */
1537 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1538 {
1539         struct inode *inode = mapping->host;
1540         struct btrfs_root *root = BTRFS_I(inode)->root;
1541         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1542         struct btrfs_ordered_extent *ordered;
1543         char *kaddr;
1544         u32 blocksize = root->sectorsize;
1545         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1546         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1547         struct page *page;
1548         int ret = 0;
1549         u64 page_start;
1550         u64 page_end;
1551
1552         if ((offset & (blocksize - 1)) == 0)
1553                 goto out;
1554
1555         ret = -ENOMEM;
1556 again:
1557         page = grab_cache_page(mapping, index);
1558         if (!page)
1559                 goto out;
1560
1561         page_start = page_offset(page);
1562         page_end = page_start + PAGE_CACHE_SIZE - 1;
1563
1564         if (!PageUptodate(page)) {
1565                 ret = btrfs_readpage(NULL, page);
1566                 lock_page(page);
1567                 if (page->mapping != mapping) {
1568                         unlock_page(page);
1569                         page_cache_release(page);
1570                         goto again;
1571                 }
1572                 if (!PageUptodate(page)) {
1573                         ret = -EIO;
1574                         goto out_unlock;
1575                 }
1576         }
1577         wait_on_page_writeback(page);
1578
1579         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1580         set_page_extent_mapped(page);
1581
1582         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1583         if (ordered) {
1584                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1585                 unlock_page(page);
1586                 page_cache_release(page);
1587                 btrfs_start_ordered_extent(inode, ordered, 1);
1588                 btrfs_put_ordered_extent(ordered);
1589                 goto again;
1590         }
1591
1592         btrfs_set_extent_delalloc(inode, page_start, page_end);
1593         ret = 0;
1594         if (offset != PAGE_CACHE_SIZE) {
1595                 kaddr = kmap(page);
1596                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1597                 flush_dcache_page(page);
1598                 kunmap(page);
1599         }
1600         ClearPageChecked(page);
1601         set_page_dirty(page);
1602         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1603
1604 out_unlock:
1605         unlock_page(page);
1606         page_cache_release(page);
1607 out:
1608         return ret;
1609 }
1610
1611 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1612 {
1613         struct inode *inode = dentry->d_inode;
1614         int err;
1615
1616         err = inode_change_ok(inode, attr);
1617         if (err)
1618                 return err;
1619
1620         if (S_ISREG(inode->i_mode) &&
1621             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1622                 struct btrfs_trans_handle *trans;
1623                 struct btrfs_root *root = BTRFS_I(inode)->root;
1624                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1625
1626                 u64 mask = root->sectorsize - 1;
1627                 u64 hole_start = (inode->i_size + mask) & ~mask;
1628                 u64 block_end = (attr->ia_size + mask) & ~mask;
1629                 u64 hole_size;
1630                 u64 alloc_hint = 0;
1631
1632                 if (attr->ia_size <= hole_start)
1633                         goto out;
1634
1635                 err = btrfs_check_free_space(root, 1, 0);
1636                 if (err)
1637                         goto fail;
1638
1639                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1640
1641                 hole_size = block_end - hole_start;
1642                 btrfs_wait_ordered_range(inode, hole_start, hole_size);
1643                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1644
1645                 trans = btrfs_start_transaction(root, 1);
1646                 btrfs_set_trans_block_group(trans, inode);
1647                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1648                 err = btrfs_drop_extents(trans, root, inode,
1649                                          hole_start, block_end, hole_start,
1650                                          &alloc_hint);
1651
1652                 if (alloc_hint != EXTENT_MAP_INLINE) {
1653                         err = btrfs_insert_file_extent(trans, root,
1654                                                        inode->i_ino,
1655                                                        hole_start, 0, 0,
1656                                                        hole_size, 0);
1657                         btrfs_drop_extent_cache(inode, hole_start,
1658                                                 (u64)-1);
1659                         btrfs_check_file(root, inode);
1660                 }
1661                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1662                 btrfs_end_transaction(trans, root);
1663                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1664                 if (err)
1665                         return err;
1666         }
1667 out:
1668         err = inode_setattr(inode, attr);
1669
1670         if (!err && ((attr->ia_valid & ATTR_MODE)))
1671                 err = btrfs_acl_chmod(inode);
1672 fail:
1673         return err;
1674 }
1675
1676 void btrfs_delete_inode(struct inode *inode)
1677 {
1678         struct btrfs_trans_handle *trans;
1679         struct btrfs_root *root = BTRFS_I(inode)->root;
1680         unsigned long nr;
1681         int ret;
1682
1683         truncate_inode_pages(&inode->i_data, 0);
1684         if (is_bad_inode(inode)) {
1685                 btrfs_orphan_del(NULL, inode);
1686                 goto no_delete;
1687         }
1688         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1689
1690         btrfs_i_size_write(inode, 0);
1691         trans = btrfs_start_transaction(root, 1);
1692
1693         btrfs_set_trans_block_group(trans, inode);
1694         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1695         if (ret) {
1696                 btrfs_orphan_del(NULL, inode);
1697                 goto no_delete_lock;
1698         }
1699
1700         btrfs_orphan_del(trans, inode);
1701
1702         nr = trans->blocks_used;
1703         clear_inode(inode);
1704
1705         btrfs_end_transaction(trans, root);
1706         btrfs_btree_balance_dirty(root, nr);
1707         return;
1708
1709 no_delete_lock:
1710         nr = trans->blocks_used;
1711         btrfs_end_transaction(trans, root);
1712         btrfs_btree_balance_dirty(root, nr);
1713 no_delete:
1714         clear_inode(inode);
1715 }
1716
1717 /*
1718  * this returns the key found in the dir entry in the location pointer.
1719  * If no dir entries were found, location->objectid is 0.
1720  */
1721 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1722                                struct btrfs_key *location)
1723 {
1724         const char *name = dentry->d_name.name;
1725         int namelen = dentry->d_name.len;
1726         struct btrfs_dir_item *di;
1727         struct btrfs_path *path;
1728         struct btrfs_root *root = BTRFS_I(dir)->root;
1729         int ret = 0;
1730
1731         if (namelen == 1 && strcmp(name, ".") == 0) {
1732                 location->objectid = dir->i_ino;
1733                 location->type = BTRFS_INODE_ITEM_KEY;
1734                 location->offset = 0;
1735                 return 0;
1736         }
1737         path = btrfs_alloc_path();
1738         BUG_ON(!path);
1739
1740         if (namelen == 2 && strcmp(name, "..") == 0) {
1741                 struct btrfs_key key;
1742                 struct extent_buffer *leaf;
1743                 int slot;
1744
1745                 key.objectid = dir->i_ino;
1746                 key.offset = (u64)-1;
1747                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1748                 if (ret < 0 || path->slots[0] == 0)
1749                         goto out_err;
1750                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1751                 BUG_ON(ret == 0);
1752                 ret = 0;
1753                 leaf = path->nodes[0];
1754                 slot = path->slots[0] - 1;
1755
1756                 btrfs_item_key_to_cpu(leaf, &key, slot);
1757                 if (key.objectid != dir->i_ino ||
1758                     key.type != BTRFS_INODE_REF_KEY) {
1759                         goto out_err;
1760                 }
1761                 location->objectid = key.offset;
1762                 location->type = BTRFS_INODE_ITEM_KEY;
1763                 location->offset = 0;
1764                 goto out;
1765         }
1766
1767         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1768                                     namelen, 0);
1769         if (IS_ERR(di))
1770                 ret = PTR_ERR(di);
1771         if (!di || IS_ERR(di)) {
1772                 goto out_err;
1773         }
1774         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1775 out:
1776         btrfs_free_path(path);
1777         return ret;
1778 out_err:
1779         location->objectid = 0;
1780         goto out;
1781 }
1782
1783 /*
1784  * when we hit a tree root in a directory, the btrfs part of the inode
1785  * needs to be changed to reflect the root directory of the tree root.  This
1786  * is kind of like crossing a mount point.
1787  */
1788 static int fixup_tree_root_location(struct btrfs_root *root,
1789                              struct btrfs_key *location,
1790                              struct btrfs_root **sub_root,
1791                              struct dentry *dentry)
1792 {
1793         struct btrfs_root_item *ri;
1794
1795         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1796                 return 0;
1797         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1798                 return 0;
1799
1800         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1801                                         dentry->d_name.name,
1802                                         dentry->d_name.len);
1803         if (IS_ERR(*sub_root))
1804                 return PTR_ERR(*sub_root);
1805
1806         ri = &(*sub_root)->root_item;
1807         location->objectid = btrfs_root_dirid(ri);
1808         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1809         location->offset = 0;
1810
1811         return 0;
1812 }
1813
1814 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1815 {
1816         struct btrfs_iget_args *args = p;
1817         inode->i_ino = args->ino;
1818         BTRFS_I(inode)->root = args->root;
1819         BTRFS_I(inode)->delalloc_bytes = 0;
1820         BTRFS_I(inode)->disk_i_size = 0;
1821         BTRFS_I(inode)->index_cnt = (u64)-1;
1822         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1823         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1824                              inode->i_mapping, GFP_NOFS);
1825         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1826                              inode->i_mapping, GFP_NOFS);
1827         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
1828         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1829         mutex_init(&BTRFS_I(inode)->csum_mutex);
1830         mutex_init(&BTRFS_I(inode)->extent_mutex);
1831         return 0;
1832 }
1833
1834 static int btrfs_find_actor(struct inode *inode, void *opaque)
1835 {
1836         struct btrfs_iget_args *args = opaque;
1837         return (args->ino == inode->i_ino &&
1838                 args->root == BTRFS_I(inode)->root);
1839 }
1840
1841 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1842                             u64 root_objectid)
1843 {
1844         struct btrfs_iget_args args;
1845         args.ino = objectid;
1846         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1847
1848         if (!args.root)
1849                 return NULL;
1850
1851         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1852 }
1853
1854 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1855                                 struct btrfs_root *root)
1856 {
1857         struct inode *inode;
1858         struct btrfs_iget_args args;
1859         args.ino = objectid;
1860         args.root = root;
1861
1862         inode = iget5_locked(s, objectid, btrfs_find_actor,
1863                              btrfs_init_locked_inode,
1864                              (void *)&args);
1865         return inode;
1866 }
1867
1868 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1869                                    struct nameidata *nd)
1870 {
1871         struct inode * inode;
1872         struct btrfs_inode *bi = BTRFS_I(dir);
1873         struct btrfs_root *root = bi->root;
1874         struct btrfs_root *sub_root = root;
1875         struct btrfs_key location;
1876         int ret, do_orphan = 0;
1877
1878         if (dentry->d_name.len > BTRFS_NAME_LEN)
1879                 return ERR_PTR(-ENAMETOOLONG);
1880
1881         ret = btrfs_inode_by_name(dir, dentry, &location);
1882
1883         if (ret < 0)
1884                 return ERR_PTR(ret);
1885
1886         inode = NULL;
1887         if (location.objectid) {
1888                 ret = fixup_tree_root_location(root, &location, &sub_root,
1889                                                 dentry);
1890                 if (ret < 0)
1891                         return ERR_PTR(ret);
1892                 if (ret > 0)
1893                         return ERR_PTR(-ENOENT);
1894
1895                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1896                                           sub_root);
1897                 if (!inode)
1898                         return ERR_PTR(-EACCES);
1899                 if (inode->i_state & I_NEW) {
1900                         /* the inode and parent dir are two different roots */
1901                         if (sub_root != root) {
1902                                 igrab(inode);
1903                                 sub_root->inode = inode;
1904                                 do_orphan = 1;
1905                         }
1906                         BTRFS_I(inode)->root = sub_root;
1907                         memcpy(&BTRFS_I(inode)->location, &location,
1908                                sizeof(location));
1909                         btrfs_read_locked_inode(inode);
1910                         unlock_new_inode(inode);
1911                 }
1912         }
1913
1914         if (unlikely(do_orphan))
1915                 btrfs_orphan_cleanup(sub_root);
1916
1917         return d_splice_alias(inode, dentry);
1918 }
1919
1920 static unsigned char btrfs_filetype_table[] = {
1921         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1922 };
1923
1924 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1925 {
1926         struct inode *inode = filp->f_dentry->d_inode;
1927         struct btrfs_root *root = BTRFS_I(inode)->root;
1928         struct btrfs_item *item;
1929         struct btrfs_dir_item *di;
1930         struct btrfs_key key;
1931         struct btrfs_key found_key;
1932         struct btrfs_path *path;
1933         int ret;
1934         u32 nritems;
1935         struct extent_buffer *leaf;
1936         int slot;
1937         int advance;
1938         unsigned char d_type;
1939         int over = 0;
1940         u32 di_cur;
1941         u32 di_total;
1942         u32 di_len;
1943         int key_type = BTRFS_DIR_INDEX_KEY;
1944         char tmp_name[32];
1945         char *name_ptr;
1946         int name_len;
1947
1948         /* FIXME, use a real flag for deciding about the key type */
1949         if (root->fs_info->tree_root == root)
1950                 key_type = BTRFS_DIR_ITEM_KEY;
1951
1952         /* special case for "." */
1953         if (filp->f_pos == 0) {
1954                 over = filldir(dirent, ".", 1,
1955                                1, inode->i_ino,
1956                                DT_DIR);
1957                 if (over)
1958                         return 0;
1959                 filp->f_pos = 1;
1960         }
1961
1962         key.objectid = inode->i_ino;
1963         path = btrfs_alloc_path();
1964         path->reada = 2;
1965
1966         /* special case for .., just use the back ref */
1967         if (filp->f_pos == 1) {
1968                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1969                 key.offset = (u64)-1;
1970                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1971                 if (ret < 0 || path->slots[0] == 0) {
1972                         btrfs_release_path(root, path);
1973                         goto read_dir_items;
1974                 }
1975                 BUG_ON(ret == 0);
1976                 leaf = path->nodes[0];
1977                 slot = path->slots[0] - 1;
1978                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1979                 btrfs_release_path(root, path);
1980                 if (found_key.objectid != key.objectid ||
1981                     found_key.type != BTRFS_INODE_REF_KEY)
1982                         goto read_dir_items;
1983                 over = filldir(dirent, "..", 2,
1984                                2, found_key.offset, DT_DIR);
1985                 if (over)
1986                         goto nopos;
1987                 filp->f_pos = 2;
1988         }
1989
1990 read_dir_items:
1991         btrfs_set_key_type(&key, key_type);
1992         key.offset = filp->f_pos;
1993
1994         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1995         if (ret < 0)
1996                 goto err;
1997         advance = 0;
1998         while(1) {
1999                 leaf = path->nodes[0];
2000                 nritems = btrfs_header_nritems(leaf);
2001                 slot = path->slots[0];
2002                 if (advance || slot >= nritems) {
2003                         if (slot >= nritems -1) {
2004                                 ret = btrfs_next_leaf(root, path);
2005                                 if (ret)
2006                                         break;
2007                                 leaf = path->nodes[0];
2008                                 nritems = btrfs_header_nritems(leaf);
2009                                 slot = path->slots[0];
2010                         } else {
2011                                 slot++;
2012                                 path->slots[0]++;
2013                         }
2014                 }
2015                 advance = 1;
2016                 item = btrfs_item_nr(leaf, slot);
2017                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2018
2019                 if (found_key.objectid != key.objectid)
2020                         break;
2021                 if (btrfs_key_type(&found_key) != key_type)
2022                         break;
2023                 if (found_key.offset < filp->f_pos)
2024                         continue;
2025
2026                 filp->f_pos = found_key.offset;
2027                 advance = 1;
2028                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2029                 di_cur = 0;
2030                 di_total = btrfs_item_size(leaf, item);
2031                 while(di_cur < di_total) {
2032                         struct btrfs_key location;
2033
2034                         name_len = btrfs_dir_name_len(leaf, di);
2035                         if (name_len < 32) {
2036                                 name_ptr = tmp_name;
2037                         } else {
2038                                 name_ptr = kmalloc(name_len, GFP_NOFS);
2039                                 BUG_ON(!name_ptr);
2040                         }
2041                         read_extent_buffer(leaf, name_ptr,
2042                                            (unsigned long)(di + 1), name_len);
2043
2044                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2045                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
2046                         over = filldir(dirent, name_ptr, name_len,
2047                                        found_key.offset,
2048                                        location.objectid,
2049                                        d_type);
2050
2051                         if (name_ptr != tmp_name)
2052                                 kfree(name_ptr);
2053
2054                         if (over)
2055                                 goto nopos;
2056                         di_len = btrfs_dir_name_len(leaf, di) +
2057                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
2058                         di_cur += di_len;
2059                         di = (struct btrfs_dir_item *)((char *)di + di_len);
2060                 }
2061         }
2062         if (key_type == BTRFS_DIR_INDEX_KEY)
2063                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2064         else
2065                 filp->f_pos++;
2066 nopos:
2067         ret = 0;
2068 err:
2069         btrfs_free_path(path);
2070         return ret;
2071 }
2072
2073 int btrfs_write_inode(struct inode *inode, int wait)
2074 {
2075         struct btrfs_root *root = BTRFS_I(inode)->root;
2076         struct btrfs_trans_handle *trans;
2077         int ret = 0;
2078
2079         if (wait) {
2080                 trans = btrfs_join_transaction(root, 1);
2081                 btrfs_set_trans_block_group(trans, inode);
2082                 ret = btrfs_commit_transaction(trans, root);
2083         }
2084         return ret;
2085 }
2086
2087 /*
2088  * This is somewhat expensive, updating the tree every time the
2089  * inode changes.  But, it is most likely to find the inode in cache.
2090  * FIXME, needs more benchmarking...there are no reasons other than performance
2091  * to keep or drop this code.
2092  */
2093 void btrfs_dirty_inode(struct inode *inode)
2094 {
2095         struct btrfs_root *root = BTRFS_I(inode)->root;
2096         struct btrfs_trans_handle *trans;
2097
2098         trans = btrfs_join_transaction(root, 1);
2099         btrfs_set_trans_block_group(trans, inode);
2100         btrfs_update_inode(trans, root, inode);
2101         btrfs_end_transaction(trans, root);
2102 }
2103
2104 static int btrfs_set_inode_index_count(struct inode *inode)
2105 {
2106         struct btrfs_root *root = BTRFS_I(inode)->root;
2107         struct btrfs_key key, found_key;
2108         struct btrfs_path *path;
2109         struct extent_buffer *leaf;
2110         int ret;
2111
2112         key.objectid = inode->i_ino;
2113         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2114         key.offset = (u64)-1;
2115
2116         path = btrfs_alloc_path();
2117         if (!path)
2118                 return -ENOMEM;
2119
2120         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2121         if (ret < 0)
2122                 goto out;
2123         /* FIXME: we should be able to handle this */
2124         if (ret == 0)
2125                 goto out;
2126         ret = 0;
2127
2128         /*
2129          * MAGIC NUMBER EXPLANATION:
2130          * since we search a directory based on f_pos we have to start at 2
2131          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2132          * else has to start at 2
2133          */
2134         if (path->slots[0] == 0) {
2135                 BTRFS_I(inode)->index_cnt = 2;
2136                 goto out;
2137         }
2138
2139         path->slots[0]--;
2140
2141         leaf = path->nodes[0];
2142         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2143
2144         if (found_key.objectid != inode->i_ino ||
2145             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2146                 BTRFS_I(inode)->index_cnt = 2;
2147                 goto out;
2148         }
2149
2150         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2151 out:
2152         btrfs_free_path(path);
2153         return ret;
2154 }
2155
2156 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2157                                  u64 *index)
2158 {
2159         int ret = 0;
2160
2161         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2162                 ret = btrfs_set_inode_index_count(dir);
2163                 if (ret)
2164                         return ret;
2165         }
2166
2167         *index = BTRFS_I(dir)->index_cnt;
2168         BTRFS_I(dir)->index_cnt++;
2169
2170         return ret;
2171 }
2172
2173 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2174                                      struct btrfs_root *root,
2175                                      struct inode *dir,
2176                                      const char *name, int name_len,
2177                                      u64 ref_objectid,
2178                                      u64 objectid,
2179                                      struct btrfs_block_group_cache *group,
2180                                      int mode, u64 *index)
2181 {
2182         struct inode *inode;
2183         struct btrfs_inode_item *inode_item;
2184         struct btrfs_block_group_cache *new_inode_group;
2185         struct btrfs_key *location;
2186         struct btrfs_path *path;
2187         struct btrfs_inode_ref *ref;
2188         struct btrfs_key key[2];
2189         u32 sizes[2];
2190         unsigned long ptr;
2191         int ret;
2192         int owner;
2193
2194         path = btrfs_alloc_path();
2195         BUG_ON(!path);
2196
2197         inode = new_inode(root->fs_info->sb);
2198         if (!inode)
2199                 return ERR_PTR(-ENOMEM);
2200
2201         if (dir) {
2202                 ret = btrfs_set_inode_index(dir, inode, index);
2203                 if (ret)
2204                         return ERR_PTR(ret);
2205         }
2206         /*
2207          * index_cnt is ignored for everything but a dir,
2208          * btrfs_get_inode_index_count has an explanation for the magic
2209          * number
2210          */
2211         BTRFS_I(inode)->index_cnt = 2;
2212
2213         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2214         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2215                              inode->i_mapping, GFP_NOFS);
2216         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2217                              inode->i_mapping, GFP_NOFS);
2218         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2219         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2220         mutex_init(&BTRFS_I(inode)->csum_mutex);
2221         mutex_init(&BTRFS_I(inode)->extent_mutex);
2222         BTRFS_I(inode)->delalloc_bytes = 0;
2223         BTRFS_I(inode)->disk_i_size = 0;
2224         BTRFS_I(inode)->root = root;
2225
2226         if (mode & S_IFDIR)
2227                 owner = 0;
2228         else
2229                 owner = 1;
2230         new_inode_group = btrfs_find_block_group(root, group, 0,
2231                                        BTRFS_BLOCK_GROUP_METADATA, owner);
2232         if (!new_inode_group) {
2233                 printk("find_block group failed\n");
2234                 new_inode_group = group;
2235         }
2236         BTRFS_I(inode)->block_group = new_inode_group;
2237         BTRFS_I(inode)->flags = 0;
2238
2239         key[0].objectid = objectid;
2240         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2241         key[0].offset = 0;
2242
2243         key[1].objectid = objectid;
2244         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2245         key[1].offset = ref_objectid;
2246
2247         sizes[0] = sizeof(struct btrfs_inode_item);
2248         sizes[1] = name_len + sizeof(*ref);
2249
2250         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2251         if (ret != 0)
2252                 goto fail;
2253
2254         if (objectid > root->highest_inode)
2255                 root->highest_inode = objectid;
2256
2257         inode->i_uid = current->fsuid;
2258         inode->i_gid = current->fsgid;
2259         inode->i_mode = mode;
2260         inode->i_ino = objectid;
2261         inode->i_blocks = 0;
2262         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2263         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2264                                   struct btrfs_inode_item);
2265         fill_inode_item(path->nodes[0], inode_item, inode);
2266
2267         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2268                              struct btrfs_inode_ref);
2269         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2270         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
2271         ptr = (unsigned long)(ref + 1);
2272         write_extent_buffer(path->nodes[0], name, ptr, name_len);
2273
2274         btrfs_mark_buffer_dirty(path->nodes[0]);
2275         btrfs_free_path(path);
2276
2277         location = &BTRFS_I(inode)->location;
2278         location->objectid = objectid;
2279         location->offset = 0;
2280         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2281
2282         insert_inode_hash(inode);
2283         return inode;
2284 fail:
2285         if (dir)
2286                 BTRFS_I(dir)->index_cnt--;
2287         btrfs_free_path(path);
2288         return ERR_PTR(ret);
2289 }
2290
2291 static inline u8 btrfs_inode_type(struct inode *inode)
2292 {
2293         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2294 }
2295
2296 static int btrfs_add_link(struct btrfs_trans_handle *trans,
2297                             struct dentry *dentry, struct inode *inode,
2298                             int add_backref, u64 index)
2299 {
2300         int ret;
2301         struct btrfs_key key;
2302         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
2303         struct inode *parent_inode = dentry->d_parent->d_inode;
2304
2305         key.objectid = inode->i_ino;
2306         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2307         key.offset = 0;
2308
2309         ret = btrfs_insert_dir_item(trans, root,
2310                                     dentry->d_name.name, dentry->d_name.len,
2311                                     dentry->d_parent->d_inode->i_ino,
2312                                     &key, btrfs_inode_type(inode),
2313                                     index);
2314         if (ret == 0) {
2315                 if (add_backref) {
2316                         ret = btrfs_insert_inode_ref(trans, root,
2317                                              dentry->d_name.name,
2318                                              dentry->d_name.len,
2319                                              inode->i_ino,
2320                                              parent_inode->i_ino,
2321                                              index);
2322                 }
2323                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2324                                    dentry->d_name.len * 2);
2325                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2326                 ret = btrfs_update_inode(trans, root,
2327                                          dentry->d_parent->d_inode);
2328         }
2329         return ret;
2330 }
2331
2332 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2333                             struct dentry *dentry, struct inode *inode,
2334                             int backref, u64 index)
2335 {
2336         int err = btrfs_add_link(trans, dentry, inode, backref, index);
2337         if (!err) {
2338                 d_instantiate(dentry, inode);
2339                 return 0;
2340         }
2341         if (err > 0)
2342                 err = -EEXIST;
2343         return err;
2344 }
2345
2346 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2347                         int mode, dev_t rdev)
2348 {
2349         struct btrfs_trans_handle *trans;
2350         struct btrfs_root *root = BTRFS_I(dir)->root;
2351         struct inode *inode = NULL;
2352         int err;
2353         int drop_inode = 0;
2354         u64 objectid;
2355         unsigned long nr = 0;
2356         u64 index = 0;
2357
2358         if (!new_valid_dev(rdev))
2359                 return -EINVAL;
2360
2361         err = btrfs_check_free_space(root, 1, 0);
2362         if (err)
2363                 goto fail;
2364
2365         trans = btrfs_start_transaction(root, 1);
2366         btrfs_set_trans_block_group(trans, dir);
2367
2368         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2369         if (err) {
2370                 err = -ENOSPC;
2371                 goto out_unlock;
2372         }
2373
2374         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2375                                 dentry->d_name.len,
2376                                 dentry->d_parent->d_inode->i_ino, objectid,
2377                                 BTRFS_I(dir)->block_group, mode, &index);
2378         err = PTR_ERR(inode);
2379         if (IS_ERR(inode))
2380                 goto out_unlock;
2381
2382         err = btrfs_init_acl(inode, dir);
2383         if (err) {
2384                 drop_inode = 1;
2385                 goto out_unlock;
2386         }
2387
2388         btrfs_set_trans_block_group(trans, inode);
2389         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2390         if (err)
2391                 drop_inode = 1;
2392         else {
2393                 inode->i_op = &btrfs_special_inode_operations;
2394                 init_special_inode(inode, inode->i_mode, rdev);
2395                 btrfs_update_inode(trans, root, inode);
2396         }
2397         dir->i_sb->s_dirt = 1;
2398         btrfs_update_inode_block_group(trans, inode);
2399         btrfs_update_inode_block_group(trans, dir);
2400 out_unlock:
2401         nr = trans->blocks_used;
2402         btrfs_end_transaction_throttle(trans, root);
2403 fail:
2404         if (drop_inode) {
2405                 inode_dec_link_count(inode);
2406                 iput(inode);
2407         }
2408         btrfs_btree_balance_dirty(root, nr);
2409         return err;
2410 }
2411
2412 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2413                         int mode, struct nameidata *nd)
2414 {
2415         struct btrfs_trans_handle *trans;
2416         struct btrfs_root *root = BTRFS_I(dir)->root;
2417         struct inode *inode = NULL;
2418         int err;
2419         int drop_inode = 0;
2420         unsigned long nr = 0;
2421         u64 objectid;
2422         u64 index = 0;
2423
2424         err = btrfs_check_free_space(root, 1, 0);
2425         if (err)
2426                 goto fail;
2427         trans = btrfs_start_transaction(root, 1);
2428         btrfs_set_trans_block_group(trans, dir);
2429
2430         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2431         if (err) {
2432                 err = -ENOSPC;
2433                 goto out_unlock;
2434         }
2435
2436         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2437                                 dentry->d_name.len,
2438                                 dentry->d_parent->d_inode->i_ino,
2439                                 objectid, BTRFS_I(dir)->block_group, mode,
2440                                 &index);
2441         err = PTR_ERR(inode);
2442         if (IS_ERR(inode))
2443                 goto out_unlock;
2444
2445         err = btrfs_init_acl(inode, dir);
2446         if (err) {
2447                 drop_inode = 1;
2448                 goto out_unlock;
2449         }
2450
2451         btrfs_set_trans_block_group(trans, inode);
2452         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2453         if (err)
2454                 drop_inode = 1;
2455         else {
2456                 inode->i_mapping->a_ops = &btrfs_aops;
2457                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2458                 inode->i_fop = &btrfs_file_operations;
2459                 inode->i_op = &btrfs_file_inode_operations;
2460                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
2461                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
2462                                      inode->i_mapping, GFP_NOFS);
2463                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
2464                                      inode->i_mapping, GFP_NOFS);
2465                 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
2466                 mutex_init(&BTRFS_I(inode)->csum_mutex);
2467                 mutex_init(&BTRFS_I(inode)->extent_mutex);
2468                 BTRFS_I(inode)->delalloc_bytes = 0;
2469                 BTRFS_I(inode)->disk_i_size = 0;
2470                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2471                 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
2472         }
2473         dir->i_sb->s_dirt = 1;
2474         btrfs_update_inode_block_group(trans, inode);
2475         btrfs_update_inode_block_group(trans, dir);
2476 out_unlock:
2477         nr = trans->blocks_used;
2478         btrfs_end_transaction_throttle(trans, root);
2479 fail:
2480         if (drop_inode) {
2481                 inode_dec_link_count(inode);
2482                 iput(inode);
2483         }
2484         btrfs_btree_balance_dirty(root, nr);
2485         return err;
2486 }
2487
2488 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2489                       struct dentry *dentry)
2490 {
2491         struct btrfs_trans_handle *trans;
2492         struct btrfs_root *root = BTRFS_I(dir)->root;
2493         struct inode *inode = old_dentry->d_inode;
2494         u64 index;
2495         unsigned long nr = 0;
2496         int err;
2497         int drop_inode = 0;
2498
2499         if (inode->i_nlink == 0)
2500                 return -ENOENT;
2501
2502 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2503         inode->i_nlink++;
2504 #else
2505         inc_nlink(inode);
2506 #endif
2507         err = btrfs_check_free_space(root, 1, 0);
2508         if (err)
2509                 goto fail;
2510         err = btrfs_set_inode_index(dir, inode, &index);
2511         if (err)
2512                 goto fail;
2513
2514         trans = btrfs_start_transaction(root, 1);
2515
2516         btrfs_set_trans_block_group(trans, dir);
2517         atomic_inc(&inode->i_count);
2518
2519         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
2520
2521         if (err)
2522                 drop_inode = 1;
2523
2524         dir->i_sb->s_dirt = 1;
2525         btrfs_update_inode_block_group(trans, dir);
2526         err = btrfs_update_inode(trans, root, inode);
2527
2528         if (err)
2529                 drop_inode = 1;
2530
2531         nr = trans->blocks_used;
2532         btrfs_end_transaction_throttle(trans, root);
2533 fail:
2534         if (drop_inode) {
2535                 inode_dec_link_count(inode);
2536                 iput(inode);
2537         }
2538         btrfs_btree_balance_dirty(root, nr);
2539         return err;
2540 }
2541
2542 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2543 {
2544         struct inode *inode = NULL;
2545         struct btrfs_trans_handle *trans;
2546         struct btrfs_root *root = BTRFS_I(dir)->root;
2547         int err = 0;
2548         int drop_on_err = 0;
2549         u64 objectid = 0;
2550         u64 index = 0;
2551         unsigned long nr = 1;
2552
2553         err = btrfs_check_free_space(root, 1, 0);
2554         if (err)
2555                 goto out_unlock;
2556
2557         trans = btrfs_start_transaction(root, 1);
2558         btrfs_set_trans_block_group(trans, dir);
2559
2560         if (IS_ERR(trans)) {
2561                 err = PTR_ERR(trans);
2562                 goto out_unlock;
2563         }
2564
2565         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2566         if (err) {
2567                 err = -ENOSPC;
2568                 goto out_unlock;
2569         }
2570
2571         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2572                                 dentry->d_name.len,
2573                                 dentry->d_parent->d_inode->i_ino, objectid,
2574                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
2575                                 &index);
2576         if (IS_ERR(inode)) {
2577                 err = PTR_ERR(inode);
2578                 goto out_fail;
2579         }
2580
2581         drop_on_err = 1;
2582
2583         err = btrfs_init_acl(inode, dir);
2584         if (err)
2585                 goto out_fail;
2586
2587         inode->i_op = &btrfs_dir_inode_operations;
2588         inode->i_fop = &btrfs_dir_file_operations;
2589         btrfs_set_trans_block_group(trans, inode);
2590
2591         btrfs_i_size_write(inode, 0);
2592         err = btrfs_update_inode(trans, root, inode);
2593         if (err)
2594                 goto out_fail;
2595
2596         err = btrfs_add_link(trans, dentry, inode, 0, index);
2597         if (err)
2598                 goto out_fail;
2599
2600         d_instantiate(dentry, inode);
2601         drop_on_err = 0;
2602         dir->i_sb->s_dirt = 1;
2603         btrfs_update_inode_block_group(trans, inode);
2604         btrfs_update_inode_block_group(trans, dir);
2605
2606 out_fail:
2607         nr = trans->blocks_used;
2608         btrfs_end_transaction_throttle(trans, root);
2609
2610 out_unlock:
2611         if (drop_on_err)
2612                 iput(inode);
2613         btrfs_btree_balance_dirty(root, nr);
2614         return err;
2615 }
2616
2617 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2618                                 struct extent_map *existing,
2619                                 struct extent_map *em,
2620                                 u64 map_start, u64 map_len)
2621 {
2622         u64 start_diff;
2623
2624         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2625         start_diff = map_start - em->start;
2626         em->start = map_start;
2627         em->len = map_len;
2628         if (em->block_start < EXTENT_MAP_LAST_BYTE)
2629                 em->block_start += start_diff;
2630         return add_extent_mapping(em_tree, em);
2631 }
2632
2633 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2634                                     size_t pg_offset, u64 start, u64 len,
2635                                     int create)
2636 {
2637         int ret;
2638         int err = 0;
2639         u64 bytenr;
2640         u64 extent_start = 0;
2641         u64 extent_end = 0;
2642         u64 objectid = inode->i_ino;
2643         u32 found_type;
2644         struct btrfs_path *path = NULL;
2645         struct btrfs_root *root = BTRFS_I(inode)->root;
2646         struct btrfs_file_extent_item *item;
2647         struct extent_buffer *leaf;
2648         struct btrfs_key found_key;
2649         struct extent_map *em = NULL;
2650         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2651         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2652         struct btrfs_trans_handle *trans = NULL;
2653
2654 again:
2655         spin_lock(&em_tree->lock);
2656         em = lookup_extent_mapping(em_tree, start, len);
2657         if (em)
2658                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2659         spin_unlock(&em_tree->lock);
2660
2661         if (em) {
2662                 if (em->start > start || em->start + em->len <= start)
2663                         free_extent_map(em);
2664                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2665                         free_extent_map(em);
2666                 else
2667                         goto out;
2668         }
2669         em = alloc_extent_map(GFP_NOFS);
2670         if (!em) {
2671                 err = -ENOMEM;
2672                 goto out;
2673         }
2674         em->bdev = root->fs_info->fs_devices->latest_bdev;
2675         em->start = EXTENT_MAP_HOLE;
2676         em->len = (u64)-1;
2677
2678         if (!path) {
2679                 path = btrfs_alloc_path();
2680                 BUG_ON(!path);
2681         }
2682
2683         ret = btrfs_lookup_file_extent(trans, root, path,
2684                                        objectid, start, trans != NULL);
2685         if (ret < 0) {
2686                 err = ret;
2687                 goto out;
2688         }
2689
2690         if (ret != 0) {
2691                 if (path->slots[0] == 0)
2692                         goto not_found;
2693                 path->slots[0]--;
2694         }
2695
2696         leaf = path->nodes[0];
2697         item = btrfs_item_ptr(leaf, path->slots[0],
2698                               struct btrfs_file_extent_item);
2699         /* are we inside the extent that was found? */
2700         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2701         found_type = btrfs_key_type(&found_key);
2702         if (found_key.objectid != objectid ||
2703             found_type != BTRFS_EXTENT_DATA_KEY) {
2704                 goto not_found;
2705         }
2706
2707         found_type = btrfs_file_extent_type(leaf, item);
2708         extent_start = found_key.offset;
2709         if (found_type == BTRFS_FILE_EXTENT_REG) {
2710                 extent_end = extent_start +
2711                        btrfs_file_extent_num_bytes(leaf, item);
2712                 err = 0;
2713                 if (start < extent_start || start >= extent_end) {
2714                         em->start = start;
2715                         if (start < extent_start) {
2716                                 if (start + len <= extent_start)
2717                                         goto not_found;
2718                                 em->len = extent_end - extent_start;
2719                         } else {
2720                                 em->len = len;
2721                         }
2722                         goto not_found_em;
2723                 }
2724                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2725                 if (bytenr == 0) {
2726                         em->start = extent_start;
2727                         em->len = extent_end - extent_start;
2728                         em->block_start = EXTENT_MAP_HOLE;
2729                         goto insert;
2730                 }
2731                 bytenr += btrfs_file_extent_offset(leaf, item);
2732                 em->block_start = bytenr;
2733                 em->start = extent_start;
2734                 em->len = extent_end - extent_start;
2735                 goto insert;
2736         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2737                 u64 page_start;
2738                 unsigned long ptr;
2739                 char *map;
2740                 size_t size;
2741                 size_t extent_offset;
2742                 size_t copy_size;
2743
2744                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2745                                                     path->slots[0]));
2746                 extent_end = (extent_start + size + root->sectorsize - 1) &
2747                         ~((u64)root->sectorsize - 1);
2748                 if (start < extent_start || start >= extent_end) {
2749                         em->start = start;
2750                         if (start < extent_start) {
2751                                 if (start + len <= extent_start)
2752                                         goto not_found;
2753                                 em->len = extent_end - extent_start;
2754                         } else {
2755                                 em->len = len;
2756                         }
2757                         goto not_found_em;
2758                 }
2759                 em->block_start = EXTENT_MAP_INLINE;
2760
2761                 if (!page) {
2762                         em->start = extent_start;
2763                         em->len = size;
2764                         goto out;
2765                 }
2766
2767                 page_start = page_offset(page) + pg_offset;
2768                 extent_offset = page_start - extent_start;
2769                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2770                                 size - extent_offset);
2771                 em->start = extent_start + extent_offset;
2772                 em->len = (copy_size + root->sectorsize - 1) &
2773                         ~((u64)root->sectorsize - 1);
2774                 map = kmap(page);
2775                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2776                 if (create == 0 && !PageUptodate(page)) {
2777                         read_extent_buffer(leaf, map + pg_offset, ptr,
2778                                            copy_size);
2779                         flush_dcache_page(page);
2780                 } else if (create && PageUptodate(page)) {
2781                         if (!trans) {
2782                                 kunmap(page);
2783                                 free_extent_map(em);
2784                                 em = NULL;
2785                                 btrfs_release_path(root, path);
2786                                 trans = btrfs_join_transaction(root, 1);
2787                                 goto again;
2788                         }
2789                         write_extent_buffer(leaf, map + pg_offset, ptr,
2790                                             copy_size);
2791                         btrfs_mark_buffer_dirty(leaf);
2792                 }
2793                 kunmap(page);
2794                 set_extent_uptodate(io_tree, em->start,
2795                                     extent_map_end(em) - 1, GFP_NOFS);
2796                 goto insert;
2797         } else {
2798                 printk("unkknown found_type %d\n", found_type);
2799                 WARN_ON(1);
2800         }
2801 not_found:
2802         em->start = start;
2803         em->len = len;
2804 not_found_em:
2805         em->block_start = EXTENT_MAP_HOLE;
2806 insert:
2807         btrfs_release_path(root, path);
2808         if (em->start > start || extent_map_end(em) <= start) {
2809                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2810                 err = -EIO;
2811                 goto out;
2812         }
2813
2814         err = 0;
2815         spin_lock(&em_tree->lock);
2816         ret = add_extent_mapping(em_tree, em);
2817         /* it is possible that someone inserted the extent into the tree
2818          * while we had the lock dropped.  It is also possible that
2819          * an overlapping map exists in the tree
2820          */
2821         if (ret == -EEXIST) {
2822                 struct extent_map *existing;
2823
2824                 ret = 0;
2825
2826                 existing = lookup_extent_mapping(em_tree, start, len);
2827                 if (existing && (existing->start > start ||
2828                     existing->start + existing->len <= start)) {
2829                         free_extent_map(existing);
2830                         existing = NULL;
2831                 }
2832                 if (!existing) {
2833                         existing = lookup_extent_mapping(em_tree, em->start,
2834                                                          em->len);
2835                         if (existing) {
2836                                 err = merge_extent_mapping(em_tree, existing,
2837                                                            em, start,
2838                                                            root->sectorsize);
2839                                 free_extent_map(existing);
2840                                 if (err) {
2841                                         free_extent_map(em);
2842                                         em = NULL;
2843                                 }
2844                         } else {
2845                                 err = -EIO;
2846                                 printk("failing to insert %Lu %Lu\n",
2847                                        start, len);
2848                                 free_extent_map(em);
2849                                 em = NULL;
2850                         }
2851                 } else {
2852                         free_extent_map(em);
2853                         em = existing;
2854                         err = 0;
2855                 }
2856         }
2857         spin_unlock(&em_tree->lock);
2858 out:
2859         if (path)
2860                 btrfs_free_path(path);
2861         if (trans) {
2862                 ret = btrfs_end_transaction(trans, root);
2863                 if (!err) {
2864                         err = ret;
2865                 }
2866         }
2867         if (err) {
2868                 free_extent_map(em);
2869                 WARN_ON(1);
2870                 return ERR_PTR(err);
2871         }
2872         return em;
2873 }
2874
2875 #if 0 /* waiting for O_DIRECT reads */
2876 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2877                         struct buffer_head *bh_result, int create)
2878 {
2879         struct extent_map *em;
2880         u64 start = (u64)iblock << inode->i_blkbits;
2881         struct btrfs_multi_bio *multi = NULL;
2882         struct btrfs_root *root = BTRFS_I(inode)->root;
2883         u64 len;
2884         u64 logical;
2885         u64 map_length;
2886         int ret = 0;
2887
2888         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2889
2890         if (!em || IS_ERR(em))
2891                 goto out;
2892
2893         if (em->start > start || em->start + em->len <= start) {
2894             goto out;
2895         }
2896
2897         if (em->block_start == EXTENT_MAP_INLINE) {
2898                 ret = -EINVAL;
2899                 goto out;
2900         }
2901
2902         len = em->start + em->len - start;
2903         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2904
2905         if (em->block_start == EXTENT_MAP_HOLE ||
2906             em->block_start == EXTENT_MAP_DELALLOC) {
2907                 bh_result->b_size = len;
2908                 goto out;
2909         }
2910
2911         logical = start - em->start;
2912         logical = em->block_start + logical;
2913
2914         map_length = len;
2915         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2916                               logical, &map_length, &multi, 0);
2917         BUG_ON(ret);
2918         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2919         bh_result->b_size = min(map_length, len);
2920
2921         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2922         set_buffer_mapped(bh_result);
2923         kfree(multi);
2924 out:
2925         free_extent_map(em);
2926         return ret;
2927 }
2928 #endif
2929
2930 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2931                         const struct iovec *iov, loff_t offset,
2932                         unsigned long nr_segs)
2933 {
2934         return -EINVAL;
2935 #if 0
2936         struct file *file = iocb->ki_filp;
2937         struct inode *inode = file->f_mapping->host;
2938
2939         if (rw == WRITE)
2940                 return -EINVAL;
2941
2942         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2943                                   offset, nr_segs, btrfs_get_block, NULL);
2944 #endif
2945 }
2946
2947 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2948 {
2949         return extent_bmap(mapping, iblock, btrfs_get_extent);
2950 }
2951
2952 int btrfs_readpage(struct file *file, struct page *page)
2953 {
2954         struct extent_io_tree *tree;
2955         tree = &BTRFS_I(page->mapping->host)->io_tree;
2956         return extent_read_full_page(tree, page, btrfs_get_extent);
2957 }
2958
2959 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2960 {
2961         struct extent_io_tree *tree;
2962
2963
2964         if (current->flags & PF_MEMALLOC) {
2965                 redirty_page_for_writepage(wbc, page);
2966                 unlock_page(page);
2967                 return 0;
2968         }
2969         tree = &BTRFS_I(page->mapping->host)->io_tree;
2970         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2971 }
2972
2973 int btrfs_writepages(struct address_space *mapping,
2974                      struct writeback_control *wbc)
2975 {
2976         struct extent_io_tree *tree;
2977         tree = &BTRFS_I(mapping->host)->io_tree;
2978         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2979 }
2980
2981 static int
2982 btrfs_readpages(struct file *file, struct address_space *mapping,
2983                 struct list_head *pages, unsigned nr_pages)
2984 {
2985         struct extent_io_tree *tree;
2986         tree = &BTRFS_I(mapping->host)->io_tree;
2987         return extent_readpages(tree, mapping, pages, nr_pages,
2988                                 btrfs_get_extent);
2989 }
2990 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2991 {
2992         struct extent_io_tree *tree;
2993         struct extent_map_tree *map;
2994         int ret;
2995
2996         tree = &BTRFS_I(page->mapping->host)->io_tree;
2997         map = &BTRFS_I(page->mapping->host)->extent_tree;
2998         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2999         if (ret == 1) {
3000                 ClearPagePrivate(page);
3001                 set_page_private(page, 0);
3002                 page_cache_release(page);
3003         }
3004         return ret;
3005 }
3006
3007 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3008 {
3009         return __btrfs_releasepage(page, gfp_flags);
3010 }
3011
3012 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3013 {
3014         struct extent_io_tree *tree;
3015         struct btrfs_ordered_extent *ordered;
3016         u64 page_start = page_offset(page);
3017         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3018
3019         wait_on_page_writeback(page);
3020         tree = &BTRFS_I(page->mapping->host)->io_tree;
3021         if (offset) {
3022                 btrfs_releasepage(page, GFP_NOFS);
3023                 return;
3024         }
3025
3026         lock_extent(tree, page_start, page_end, GFP_NOFS);
3027         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3028                                            page_offset(page));
3029         if (ordered) {
3030                 /*
3031                  * IO on this page will never be started, so we need
3032                  * to account for any ordered extents now
3033                  */
3034                 clear_extent_bit(tree, page_start, page_end,
3035                                  EXTENT_DIRTY | EXTENT_DELALLOC |
3036                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
3037                 btrfs_finish_ordered_io(page->mapping->host,
3038                                         page_start, page_end);
3039                 btrfs_put_ordered_extent(ordered);
3040                 lock_extent(tree, page_start, page_end, GFP_NOFS);
3041         }
3042         clear_extent_bit(tree, page_start, page_end,
3043                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3044                  EXTENT_ORDERED,
3045                  1, 1, GFP_NOFS);
3046         __btrfs_releasepage(page, GFP_NOFS);
3047
3048         ClearPageChecked(page);
3049         if (PagePrivate(page)) {
3050                 ClearPagePrivate(page);
3051                 set_page_private(page, 0);
3052                 page_cache_release(page);
3053         }
3054 }
3055
3056 /*
3057  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3058  * called from a page fault handler when a page is first dirtied. Hence we must
3059  * be careful to check for EOF conditions here. We set the page up correctly
3060  * for a written page which means we get ENOSPC checking when writing into
3061  * holes and correct delalloc and unwritten extent mapping on filesystems that
3062  * support these features.
3063  *
3064  * We are not allowed to take the i_mutex here so we have to play games to
3065  * protect against truncate races as the page could now be beyond EOF.  Because
3066  * vmtruncate() writes the inode size before removing pages, once we have the
3067  * page lock we can determine safely if the page is beyond EOF. If it is not
3068  * beyond EOF, then the page is guaranteed safe against truncation until we
3069  * unlock the page.
3070  */
3071 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3072 {
3073         struct inode *inode = fdentry(vma->vm_file)->d_inode;
3074         struct btrfs_root *root = BTRFS_I(inode)->root;
3075         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3076         struct btrfs_ordered_extent *ordered;
3077         char *kaddr;
3078         unsigned long zero_start;
3079         loff_t size;
3080         int ret;
3081         u64 page_start;
3082         u64 page_end;
3083
3084         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3085         if (ret)
3086                 goto out;
3087
3088         ret = -EINVAL;
3089 again:
3090         lock_page(page);
3091         size = i_size_read(inode);
3092         page_start = page_offset(page);
3093         page_end = page_start + PAGE_CACHE_SIZE - 1;
3094
3095         if ((page->mapping != inode->i_mapping) ||
3096             (page_start >= size)) {
3097                 /* page got truncated out from underneath us */
3098                 goto out_unlock;
3099         }
3100         wait_on_page_writeback(page);
3101
3102         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3103         set_page_extent_mapped(page);
3104
3105         /*
3106          * we can't set the delalloc bits if there are pending ordered
3107          * extents.  Drop our locks and wait for them to finish
3108          */
3109         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3110         if (ordered) {
3111                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3112                 unlock_page(page);
3113                 btrfs_start_ordered_extent(inode, ordered, 1);
3114                 btrfs_put_ordered_extent(ordered);
3115                 goto again;
3116         }
3117
3118         btrfs_set_extent_delalloc(inode, page_start, page_end);
3119         ret = 0;
3120
3121         /* page is wholly or partially inside EOF */
3122         if (page_start + PAGE_CACHE_SIZE > size)
3123                 zero_start = size & ~PAGE_CACHE_MASK;
3124         else
3125                 zero_start = PAGE_CACHE_SIZE;
3126
3127         if (zero_start != PAGE_CACHE_SIZE) {
3128                 kaddr = kmap(page);
3129                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3130                 flush_dcache_page(page);
3131                 kunmap(page);
3132         }
3133         ClearPageChecked(page);
3134         set_page_dirty(page);
3135         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3136
3137 out_unlock:
3138         unlock_page(page);
3139 out:
3140         return ret;
3141 }
3142
3143 static void btrfs_truncate(struct inode *inode)
3144 {
3145         struct btrfs_root *root = BTRFS_I(inode)->root;
3146         int ret;
3147         struct btrfs_trans_handle *trans;
3148         unsigned long nr;
3149         u64 mask = root->sectorsize - 1;
3150
3151         if (!S_ISREG(inode->i_mode))
3152                 return;
3153         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3154                 return;
3155
3156         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3157         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3158
3159         trans = btrfs_start_transaction(root, 1);
3160         btrfs_set_trans_block_group(trans, inode);
3161         btrfs_i_size_write(inode, inode->i_size);
3162
3163         ret = btrfs_orphan_add(trans, inode);
3164         if (ret)
3165                 goto out;
3166         /* FIXME, add redo link to tree so we don't leak on crash */
3167         ret = btrfs_truncate_in_trans(trans, root, inode,
3168                                       BTRFS_EXTENT_DATA_KEY);
3169         btrfs_update_inode(trans, root, inode);
3170
3171         ret = btrfs_orphan_del(trans, inode);
3172         BUG_ON(ret);
3173
3174 out:
3175         nr = trans->blocks_used;
3176         ret = btrfs_end_transaction_throttle(trans, root);
3177         BUG_ON(ret);
3178         btrfs_btree_balance_dirty(root, nr);
3179 }
3180
3181 /*
3182  * Invalidate a single dcache entry at the root of the filesystem.
3183  * Needed after creation of snapshot or subvolume.
3184  */
3185 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3186                                   int namelen)
3187 {
3188         struct dentry *alias, *entry;
3189         struct qstr qstr;
3190
3191         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3192         if (alias) {
3193                 qstr.name = name;
3194                 qstr.len = namelen;
3195                 /* change me if btrfs ever gets a d_hash operation */
3196                 qstr.hash = full_name_hash(qstr.name, qstr.len);
3197                 entry = d_lookup(alias, &qstr);
3198                 dput(alias);
3199                 if (entry) {
3200                         d_invalidate(entry);
3201                         dput(entry);
3202                 }
3203         }
3204 }
3205
3206 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3207                 struct btrfs_trans_handle *trans, u64 new_dirid,
3208                 struct btrfs_block_group_cache *block_group)
3209 {
3210         struct inode *inode;
3211         u64 index = 0;
3212
3213         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3214                                 new_dirid, block_group, S_IFDIR | 0700, &index);
3215         if (IS_ERR(inode))
3216                 return PTR_ERR(inode);
3217         inode->i_op = &btrfs_dir_inode_operations;
3218         inode->i_fop = &btrfs_dir_file_operations;
3219         new_root->inode = inode;
3220
3221         inode->i_nlink = 1;
3222         btrfs_i_size_write(inode, 0);
3223
3224         return btrfs_update_inode(trans, new_root, inode);
3225 }
3226
3227 unsigned long btrfs_force_ra(struct address_space *mapping,
3228                               struct file_ra_state *ra, struct file *file,
3229                               pgoff_t offset, pgoff_t last_index)
3230 {
3231         pgoff_t req_size = last_index - offset + 1;
3232
3233 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3234         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3235         return offset;
3236 #else
3237         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3238         return offset + req_size;
3239 #endif
3240 }
3241
3242 struct inode *btrfs_alloc_inode(struct super_block *sb)
3243 {
3244         struct btrfs_inode *ei;
3245
3246         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3247         if (!ei)
3248                 return NULL;
3249         ei->last_trans = 0;
3250         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3251         ei->i_acl = BTRFS_ACL_NOT_CACHED;
3252         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3253         INIT_LIST_HEAD(&ei->i_orphan);
3254         return &ei->vfs_inode;
3255 }
3256
3257 void btrfs_destroy_inode(struct inode *inode)
3258 {
3259         struct btrfs_ordered_extent *ordered;
3260         WARN_ON(!list_empty(&inode->i_dentry));
3261         WARN_ON(inode->i_data.nrpages);
3262
3263         if (BTRFS_I(inode)->i_acl &&
3264             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3265                 posix_acl_release(BTRFS_I(inode)->i_acl);
3266         if (BTRFS_I(inode)->i_default_acl &&
3267             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3268                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3269
3270         spin_lock(&BTRFS_I(inode)->root->list_lock);
3271         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3272                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3273                        " list\n", inode->i_ino);
3274                 dump_stack();
3275         }
3276         spin_unlock(&BTRFS_I(inode)->root->list_lock);
3277
3278         while(1) {
3279                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3280                 if (!ordered)
3281                         break;
3282                 else {
3283                         printk("found ordered extent %Lu %Lu\n",
3284                                ordered->file_offset, ordered->len);
3285                         btrfs_remove_ordered_extent(inode, ordered);
3286                         btrfs_put_ordered_extent(ordered);
3287                         btrfs_put_ordered_extent(ordered);
3288                 }
3289         }
3290         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3291         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3292 }
3293
3294 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3295 static void init_once(void *foo)
3296 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3297 static void init_once(struct kmem_cache * cachep, void *foo)
3298 #else
3299 static void init_once(void * foo, struct kmem_cache * cachep,
3300                       unsigned long flags)
3301 #endif
3302 {
3303         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3304
3305         inode_init_once(&ei->vfs_inode);
3306 }
3307
3308 void btrfs_destroy_cachep(void)
3309 {
3310         if (btrfs_inode_cachep)
3311                 kmem_cache_destroy(btrfs_inode_cachep);
3312         if (btrfs_trans_handle_cachep)
3313                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3314         if (btrfs_transaction_cachep)
3315                 kmem_cache_destroy(btrfs_transaction_cachep);
3316         if (btrfs_bit_radix_cachep)
3317                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3318         if (btrfs_path_cachep)
3319                 kmem_cache_destroy(btrfs_path_cachep);
3320 }
3321
3322 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3323                                        unsigned long extra_flags,
3324 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3325                                        void (*ctor)(void *)
3326 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3327                                        void (*ctor)(struct kmem_cache *, void *)
3328 #else
3329                                        void (*ctor)(void *, struct kmem_cache *,
3330                                                     unsigned long)
3331 #endif
3332                                      )
3333 {
3334         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3335                                  SLAB_MEM_SPREAD | extra_flags), ctor
3336 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3337                                  ,NULL
3338 #endif
3339                                 );
3340 }
3341
3342 int btrfs_init_cachep(void)
3343 {
3344         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3345                                           sizeof(struct btrfs_inode),
3346                                           0, init_once);
3347         if (!btrfs_inode_cachep)
3348                 goto fail;
3349         btrfs_trans_handle_cachep =
3350                         btrfs_cache_create("btrfs_trans_handle_cache",
3351                                            sizeof(struct btrfs_trans_handle),
3352                                            0, NULL);
3353         if (!btrfs_trans_handle_cachep)
3354                 goto fail;
3355         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3356                                              sizeof(struct btrfs_transaction),
3357                                              0, NULL);
3358         if (!btrfs_transaction_cachep)
3359                 goto fail;
3360         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3361                                          sizeof(struct btrfs_path),
3362                                          0, NULL);
3363         if (!btrfs_path_cachep)
3364                 goto fail;
3365         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3366                                               SLAB_DESTROY_BY_RCU, NULL);
3367         if (!btrfs_bit_radix_cachep)
3368                 goto fail;
3369         return 0;
3370 fail:
3371         btrfs_destroy_cachep();
3372         return -ENOMEM;
3373 }
3374
3375 static int btrfs_getattr(struct vfsmount *mnt,
3376                          struct dentry *dentry, struct kstat *stat)
3377 {
3378         struct inode *inode = dentry->d_inode;
3379         generic_fillattr(inode, stat);
3380         stat->blksize = PAGE_CACHE_SIZE;
3381         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3382         return 0;
3383 }
3384
3385 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3386                            struct inode * new_dir,struct dentry *new_dentry)
3387 {
3388         struct btrfs_trans_handle *trans;
3389         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3390         struct inode *new_inode = new_dentry->d_inode;
3391         struct inode *old_inode = old_dentry->d_inode;
3392         struct timespec ctime = CURRENT_TIME;
3393         u64 index = 0;
3394         int ret;
3395
3396         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3397             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3398                 return -ENOTEMPTY;
3399         }
3400
3401         ret = btrfs_check_free_space(root, 1, 0);
3402         if (ret)
3403                 goto out_unlock;
3404
3405         trans = btrfs_start_transaction(root, 1);
3406
3407         btrfs_set_trans_block_group(trans, new_dir);
3408
3409         old_dentry->d_inode->i_nlink++;
3410         old_dir->i_ctime = old_dir->i_mtime = ctime;
3411         new_dir->i_ctime = new_dir->i_mtime = ctime;
3412         old_inode->i_ctime = ctime;
3413
3414         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3415         if (ret)
3416                 goto out_fail;
3417
3418         if (new_inode) {
3419                 new_inode->i_ctime = CURRENT_TIME;
3420                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3421                 if (ret)
3422                         goto out_fail;
3423                 if (new_inode->i_nlink == 0) {
3424                         ret = btrfs_orphan_add(trans, new_inode);
3425                         if (ret)
3426                                 goto out_fail;
3427                 }
3428         }
3429         ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3430         if (ret)
3431                 goto out_fail;
3432
3433         ret = btrfs_add_link(trans, new_dentry, old_inode, 1, index);
3434         if (ret)
3435                 goto out_fail;
3436
3437 out_fail:
3438         btrfs_end_transaction_throttle(trans, root);
3439 out_unlock:
3440         return ret;
3441 }
3442
3443 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
3444 {
3445         struct list_head *head = &root->fs_info->delalloc_inodes;
3446         struct btrfs_inode *binode;
3447         unsigned long flags;
3448
3449         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3450         while(!list_empty(head)) {
3451                 binode = list_entry(head->next, struct btrfs_inode,
3452                                     delalloc_inodes);
3453                 atomic_inc(&binode->vfs_inode.i_count);
3454                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3455                 filemap_write_and_wait(binode->vfs_inode.i_mapping);
3456                 iput(&binode->vfs_inode);
3457                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3458         }
3459         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3460         return 0;
3461 }
3462
3463 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3464                          const char *symname)
3465 {
3466         struct btrfs_trans_handle *trans;
3467         struct btrfs_root *root = BTRFS_I(dir)->root;
3468         struct btrfs_path *path;
3469         struct btrfs_key key;
3470         struct inode *inode = NULL;
3471         int err;
3472         int drop_inode = 0;
3473         u64 objectid;
3474         u64 index = 0 ;
3475         int name_len;
3476         int datasize;
3477         unsigned long ptr;
3478         struct btrfs_file_extent_item *ei;
3479         struct extent_buffer *leaf;
3480         unsigned long nr = 0;
3481
3482         name_len = strlen(symname) + 1;
3483         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3484                 return -ENAMETOOLONG;
3485
3486         err = btrfs_check_free_space(root, 1, 0);
3487         if (err)
3488                 goto out_fail;
3489
3490         trans = btrfs_start_transaction(root, 1);
3491         btrfs_set_trans_block_group(trans, dir);
3492
3493         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3494         if (err) {
3495                 err = -ENOSPC;
3496                 goto out_unlock;
3497         }
3498
3499         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3500                                 dentry->d_name.len,
3501                                 dentry->d_parent->d_inode->i_ino, objectid,
3502                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
3503                                 &index);
3504         err = PTR_ERR(inode);
3505         if (IS_ERR(inode))
3506                 goto out_unlock;
3507
3508         err = btrfs_init_acl(inode, dir);
3509         if (err) {
3510                 drop_inode = 1;
3511                 goto out_unlock;
3512         }
3513
3514         btrfs_set_trans_block_group(trans, inode);
3515         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3516         if (err)
3517                 drop_inode = 1;
3518         else {
3519                 inode->i_mapping->a_ops = &btrfs_aops;
3520                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3521                 inode->i_fop = &btrfs_file_operations;
3522                 inode->i_op = &btrfs_file_inode_operations;
3523                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3524                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3525                                      inode->i_mapping, GFP_NOFS);
3526                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3527                                      inode->i_mapping, GFP_NOFS);
3528                 INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
3529                 mutex_init(&BTRFS_I(inode)->csum_mutex);
3530                 mutex_init(&BTRFS_I(inode)->extent_mutex);
3531                 BTRFS_I(inode)->delalloc_bytes = 0;
3532                 BTRFS_I(inode)->disk_i_size = 0;
3533                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3534                 btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
3535         }
3536         dir->i_sb->s_dirt = 1;
3537         btrfs_update_inode_block_group(trans, inode);
3538         btrfs_update_inode_block_group(trans, dir);
3539         if (drop_inode)
3540                 goto out_unlock;
3541
3542         path = btrfs_alloc_path();
3543         BUG_ON(!path);
3544         key.objectid = inode->i_ino;
3545         key.offset = 0;
3546         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3547         datasize = btrfs_file_extent_calc_inline_size(name_len);
3548         err = btrfs_insert_empty_item(trans, root, path, &key,
3549                                       datasize);
3550         if (err) {
3551                 drop_inode = 1;
3552                 goto out_unlock;
3553         }
3554         leaf = path->nodes[0];
3555         ei = btrfs_item_ptr(leaf, path->slots[0],
3556                             struct btrfs_file_extent_item);
3557         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3558         btrfs_set_file_extent_type(leaf, ei,
3559                                    BTRFS_FILE_EXTENT_INLINE);
3560         ptr = btrfs_file_extent_inline_start(ei);
3561         write_extent_buffer(leaf, symname, ptr, name_len);
3562         btrfs_mark_buffer_dirty(leaf);
3563         btrfs_free_path(path);
3564
3565         inode->i_op = &btrfs_symlink_inode_operations;
3566         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3567         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3568         btrfs_i_size_write(inode, name_len - 1);
3569         err = btrfs_update_inode(trans, root, inode);
3570         if (err)
3571                 drop_inode = 1;
3572
3573 out_unlock:
3574         nr = trans->blocks_used;
3575         btrfs_end_transaction_throttle(trans, root);
3576 out_fail:
3577         if (drop_inode) {
3578                 inode_dec_link_count(inode);
3579                 iput(inode);
3580         }
3581         btrfs_btree_balance_dirty(root, nr);
3582         return err;
3583 }
3584
3585 static int btrfs_set_page_dirty(struct page *page)
3586 {
3587         return __set_page_dirty_nobuffers(page);
3588 }
3589
3590 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3591 static int btrfs_permission(struct inode *inode, int mask)
3592 #else
3593 static int btrfs_permission(struct inode *inode, int mask,
3594                             struct nameidata *nd)
3595 #endif
3596 {
3597         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3598                 return -EACCES;
3599         return generic_permission(inode, mask, btrfs_check_acl);
3600 }
3601
3602 static struct inode_operations btrfs_dir_inode_operations = {
3603         .lookup         = btrfs_lookup,
3604         .create         = btrfs_create,
3605         .unlink         = btrfs_unlink,
3606         .link           = btrfs_link,
3607         .mkdir          = btrfs_mkdir,
3608         .rmdir          = btrfs_rmdir,
3609         .rename         = btrfs_rename,
3610         .symlink        = btrfs_symlink,
3611         .setattr        = btrfs_setattr,
3612         .mknod          = btrfs_mknod,
3613         .setxattr       = generic_setxattr,
3614         .getxattr       = generic_getxattr,
3615         .listxattr      = btrfs_listxattr,
3616         .removexattr    = generic_removexattr,
3617         .permission     = btrfs_permission,
3618 };
3619 static struct inode_operations btrfs_dir_ro_inode_operations = {
3620         .lookup         = btrfs_lookup,
3621         .permission     = btrfs_permission,
3622 };
3623 static struct file_operations btrfs_dir_file_operations = {
3624         .llseek         = generic_file_llseek,
3625         .read           = generic_read_dir,
3626         .readdir        = btrfs_readdir,
3627         .unlocked_ioctl = btrfs_ioctl,
3628 #ifdef CONFIG_COMPAT
3629         .compat_ioctl   = btrfs_ioctl,
3630 #endif
3631         .release        = btrfs_release_file,
3632 };
3633
3634 static struct extent_io_ops btrfs_extent_io_ops = {
3635         .fill_delalloc = run_delalloc_range,
3636         .submit_bio_hook = btrfs_submit_bio_hook,
3637         .merge_bio_hook = btrfs_merge_bio_hook,
3638         .readpage_io_hook = btrfs_readpage_io_hook,
3639         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3640         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3641         .writepage_start_hook = btrfs_writepage_start_hook,
3642         .readpage_io_failed_hook = btrfs_io_failed_hook,
3643         .set_bit_hook = btrfs_set_bit_hook,
3644         .clear_bit_hook = btrfs_clear_bit_hook,
3645 };
3646
3647 static struct address_space_operations btrfs_aops = {
3648         .readpage       = btrfs_readpage,
3649         .writepage      = btrfs_writepage,
3650         .writepages     = btrfs_writepages,
3651         .readpages      = btrfs_readpages,
3652         .sync_page      = block_sync_page,
3653         .bmap           = btrfs_bmap,
3654         .direct_IO      = btrfs_direct_IO,
3655         .invalidatepage = btrfs_invalidatepage,
3656         .releasepage    = btrfs_releasepage,
3657         .set_page_dirty = btrfs_set_page_dirty,
3658 };
3659
3660 static struct address_space_operations btrfs_symlink_aops = {
3661         .readpage       = btrfs_readpage,
3662         .writepage      = btrfs_writepage,
3663         .invalidatepage = btrfs_invalidatepage,
3664         .releasepage    = btrfs_releasepage,
3665 };
3666
3667 static struct inode_operations btrfs_file_inode_operations = {
3668         .truncate       = btrfs_truncate,
3669         .getattr        = btrfs_getattr,
3670         .setattr        = btrfs_setattr,
3671         .setxattr       = generic_setxattr,
3672         .getxattr       = generic_getxattr,
3673         .listxattr      = btrfs_listxattr,
3674         .removexattr    = generic_removexattr,
3675         .permission     = btrfs_permission,
3676 };
3677 static struct inode_operations btrfs_special_inode_operations = {
3678         .getattr        = btrfs_getattr,
3679         .setattr        = btrfs_setattr,
3680         .permission     = btrfs_permission,
3681         .setxattr       = generic_setxattr,
3682         .getxattr       = generic_getxattr,
3683         .listxattr      = btrfs_listxattr,
3684         .removexattr    = generic_removexattr,
3685 };
3686 static struct inode_operations btrfs_symlink_inode_operations = {
3687         .readlink       = generic_readlink,
3688         .follow_link    = page_follow_link_light,
3689         .put_link       = page_put_link,
3690         .permission     = btrfs_permission,
3691 };