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