git.oblomov.eu Git - linux-2.6/blob - fs/btrfs/inode.c

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