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, 0,
 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         u64 alloc_hint = 0;
 532         struct list_head list;
 533         struct btrfs_key ins;
 534         int ret;
 535
 536         ret = btrfs_dec_test_ordered_pending(inode, start, end - start + 1);
 537         if (!ret)
 538                 return 0;
 539
 540         trans = btrfs_join_transaction(root, 1);
 541
 542         ordered_extent = btrfs_lookup_ordered_extent(inode, start);
 543         BUG_ON(!ordered_extent);
 544         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags))
 545                 goto nocow;
 546
 547         lock_extent(io_tree, ordered_extent->file_offset,
 548                     ordered_extent->file_offset + ordered_extent->len - 1,
 549                     GFP_NOFS);
 550
 551         INIT_LIST_HEAD(&list);
 552
 553         ins.objectid = ordered_extent->start;
 554         ins.offset = ordered_extent->len;
 555         ins.type = BTRFS_EXTENT_ITEM_KEY;
 556
 557         ret = btrfs_alloc_reserved_extent(trans, root, root->root_key.objectid,
 558                                           trans->transid, inode->i_ino,
 559                                           ordered_extent->file_offset, &ins);
 560         BUG_ON(ret);
 561
 562         mutex_lock(&BTRFS_I(inode)->extent_mutex);
 563
 564         ret = btrfs_drop_extents(trans, root, inode,
 565                                  ordered_extent->file_offset,
 566                                  ordered_extent->file_offset +
 567                                  ordered_extent->len,
 568                                  ordered_extent->file_offset, &alloc_hint);
 569         BUG_ON(ret);
 570         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 571                                        ordered_extent->file_offset,
 572                                        ordered_extent->start,
 573                                        ordered_extent->len,
 574                                        ordered_extent->len, 0);
 575         BUG_ON(ret);
 576
 577         btrfs_drop_extent_cache(inode, ordered_extent->file_offset,
 578                                 ordered_extent->file_offset +
 579                                 ordered_extent->len - 1);
 580         mutex_unlock(&BTRFS_I(inode)->extent_mutex);
 581
 582         inode->i_blocks += ordered_extent->len >> 9;
 583         unlock_extent(io_tree, ordered_extent->file_offset,
 584                     ordered_extent->file_offset + ordered_extent->len - 1,
 585                     GFP_NOFS);
 586 nocow:
 587         add_pending_csums(trans, inode, ordered_extent->file_offset,
 588                           &ordered_extent->list);
 589
 590         btrfs_ordered_update_i_size(inode, ordered_extent);
 591         btrfs_update_inode(trans, root, inode);
 592         btrfs_remove_ordered_extent(inode, ordered_extent);
 593
 594         /* once for us */
 595         btrfs_put_ordered_extent(ordered_extent);
 596         /* once for the tree */
 597         btrfs_put_ordered_extent(ordered_extent);
 598
 599         btrfs_end_transaction(trans, root);
 600         return 0;
 601 }
 602
 603 int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
 604                                 struct extent_state *state, int uptodate)
 605 {
 606         return btrfs_finish_ordered_io(page->mapping->host, start, end);
 607 }
 608
 609 struct io_failure_record {
 610         struct page *page;
 611         u64 start;
 612         u64 len;
 613         u64 logical;
 614         int last_mirror;
 615 };
 616
 617 int btrfs_io_failed_hook(struct bio *failed_bio,
 618                          struct page *page, u64 start, u64 end,
 619                          struct extent_state *state)
 620 {
 621         struct io_failure_record *failrec = NULL;
 622         u64 private;
 623         struct extent_map *em;
 624         struct inode *inode = page->mapping->host;
 625         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 626         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 627         struct bio *bio;
 628         int num_copies;
 629         int ret;
 630         int rw;
 631         u64 logical;
 632
 633         ret = get_state_private(failure_tree, start, &private);
 634         if (ret) {
 635                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 636                 if (!failrec)
 637                         return -ENOMEM;
 638                 failrec->start = start;
 639                 failrec->len = end - start + 1;
 640                 failrec->last_mirror = 0;
 641
 642                 spin_lock(&em_tree->lock);
 643                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 644                 if (em->start > start || em->start + em->len < start) {
 645                         free_extent_map(em);
 646                         em = NULL;
 647                 }
 648                 spin_unlock(&em_tree->lock);
 649
 650                 if (!em || IS_ERR(em)) {
 651                         kfree(failrec);
 652                         return -EIO;
 653                 }
 654                 logical = start - em->start;
 655                 logical = em->block_start + logical;
 656                 failrec->logical = logical;
 657                 free_extent_map(em);
 658                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 659                                 EXTENT_DIRTY, GFP_NOFS);
 660                 set_state_private(failure_tree, start,
 661                                  (u64)(unsigned long)failrec);
 662         } else {
 663                 failrec = (struct io_failure_record *)(unsigned long)private;
 664         }
 665         num_copies = btrfs_num_copies(
 666                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 667                               failrec->logical, failrec->len);
 668         failrec->last_mirror++;
 669         if (!state) {
 670                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 671                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 672                                                     failrec->start,
 673                                                     EXTENT_LOCKED);
 674                 if (state && state->start != failrec->start)
 675                         state = NULL;
 676                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 677         }
 678         if (!state || failrec->last_mirror > num_copies) {
 679                 set_state_private(failure_tree, failrec->start, 0);
 680                 clear_extent_bits(failure_tree, failrec->start,
 681                                   failrec->start + failrec->len - 1,
 682                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 683                 kfree(failrec);
 684                 return -EIO;
 685         }
 686         bio = bio_alloc(GFP_NOFS, 1);
 687         bio->bi_private = state;
 688         bio->bi_end_io = failed_bio->bi_end_io;
 689         bio->bi_sector = failrec->logical >> 9;
 690         bio->bi_bdev = failed_bio->bi_bdev;
 691         bio->bi_size = 0;
 692         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 693         if (failed_bio->bi_rw & (1 << BIO_RW))
 694                 rw = WRITE;
 695         else
 696                 rw = READ;
 697
 698         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
 699                                                       failrec->last_mirror);
 700         return 0;
 701 }
 702
 703 int btrfs_clean_io_failures(struct inode *inode, u64 start)
 704 {
 705         u64 private;
 706         u64 private_failure;
 707         struct io_failure_record *failure;
 708         int ret;
 709
 710         private = 0;
 711         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 712                              (u64)-1, 1, EXTENT_DIRTY)) {
 713                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 714                                         start, &private_failure);
 715                 if (ret == 0) {
 716                         failure = (struct io_failure_record *)(unsigned long)
 717                                    private_failure;
 718                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 719                                           failure->start, 0);
 720                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 721                                           failure->start,
 722                                           failure->start + failure->len - 1,
 723                                           EXTENT_DIRTY | EXTENT_LOCKED,
 724                                           GFP_NOFS);
 725                         kfree(failure);
 726                 }
 727         }
 728         return 0;
 729 }
 730
 731 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 732                                struct extent_state *state)
 733 {
 734         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 735         struct inode *inode = page->mapping->host;
 736         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 737         char *kaddr;
 738         u64 private = ~(u32)0;
 739         int ret;
 740         struct btrfs_root *root = BTRFS_I(inode)->root;
 741         u32 csum = ~(u32)0;
 742         unsigned long flags;
 743
 744         if (btrfs_test_opt(root, NODATASUM) ||
 745             btrfs_test_flag(inode, NODATASUM))
 746                 return 0;
 747         if (state && state->start == start) {
 748                 private = state->private;
 749                 ret = 0;
 750         } else {
 751                 ret = get_state_private(io_tree, start, &private);
 752         }
 753         local_irq_save(flags);
 754         kaddr = kmap_atomic(page, KM_IRQ0);
 755         if (ret) {
 756                 goto zeroit;
 757         }
 758         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 759         btrfs_csum_final(csum, (char *)&csum);
 760         if (csum != private) {
 761                 goto zeroit;
 762         }
 763         kunmap_atomic(kaddr, KM_IRQ0);
 764         local_irq_restore(flags);
 765
 766         /* if the io failure tree for this inode is non-empty,
 767          * check to see if we've recovered from a failed IO
 768          */
 769         btrfs_clean_io_failures(inode, start);
 770         return 0;
 771
 772 zeroit:
 773         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 774                page->mapping->host->i_ino, (unsigned long long)start, csum,
 775                private);
 776         memset(kaddr + offset, 1, end - start + 1);
 777         flush_dcache_page(page);
 778         kunmap_atomic(kaddr, KM_IRQ0);
 779         local_irq_restore(flags);
 780         if (private == 0)
 781                 return 0;
 782         return -EIO;
 783 }
 784
 785 /*
 786  * This creates an orphan entry for the given inode in case something goes
 787  * wrong in the middle of an unlink/truncate.
 788  */
 789 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
 790 {
 791         struct btrfs_root *root = BTRFS_I(inode)->root;
 792         int ret = 0;
 793
 794         spin_lock(&root->list_lock);
 795
 796         /* already on the orphan list, we're good */
 797         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
 798                 spin_unlock(&root->list_lock);
 799                 return 0;
 800         }
 801
 802         list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 803
 804         spin_unlock(&root->list_lock);
 805
 806         /*
 807          * insert an orphan item to track this unlinked/truncated file
 808          */
 809         ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
 810
 811         return ret;
 812 }
 813
 814 /*
 815  * We have done the truncate/delete so we can go ahead and remove the orphan
 816  * item for this particular inode.
 817  */
 818 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
 819 {
 820         struct btrfs_root *root = BTRFS_I(inode)->root;
 821         int ret = 0;
 822
 823         spin_lock(&root->list_lock);
 824
 825         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
 826                 spin_unlock(&root->list_lock);
 827                 return 0;
 828         }
 829
 830         list_del_init(&BTRFS_I(inode)->i_orphan);
 831         if (!trans) {
 832                 spin_unlock(&root->list_lock);
 833                 return 0;
 834         }
 835
 836         spin_unlock(&root->list_lock);
 837
 838         ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
 839
 840         return ret;
 841 }
 842
 843 /*
 844  * this cleans up any orphans that may be left on the list from the last use
 845  * of this root.
 846  */
 847 void btrfs_orphan_cleanup(struct btrfs_root *root)
 848 {
 849         struct btrfs_path *path;
 850         struct extent_buffer *leaf;
 851         struct btrfs_item *item;
 852         struct btrfs_key key, found_key;
 853         struct btrfs_trans_handle *trans;
 854         struct inode *inode;
 855         int ret = 0, nr_unlink = 0, nr_truncate = 0;
 856
 857         /* don't do orphan cleanup if the fs is readonly. */
 858         if (root->inode->i_sb->s_flags & MS_RDONLY)
 859                 return;
 860
 861         path = btrfs_alloc_path();
 862         if (!path)
 863                 return;
 864         path->reada = -1;
 865
 866         key.objectid = BTRFS_ORPHAN_OBJECTID;
 867         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
 868         key.offset = (u64)-1;
 869
 870         trans = btrfs_start_transaction(root, 1);
 871         btrfs_set_trans_block_group(trans, root->inode);
 872
 873         while (1) {
 874                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 875                 if (ret < 0) {
 876                         printk(KERN_ERR "Error searching slot for orphan: %d"
 877                                "\n", ret);
 878                         break;
 879                 }
 880
 881                 /*
 882                  * if ret == 0 means we found what we were searching for, which
 883                  * is weird, but possible, so only screw with path if we didnt
 884                  * find the key and see if we have stuff that matches
 885                  */
 886                 if (ret > 0) {
 887                         if (path->slots[0] == 0)
 888                                 break;
 889                         path->slots[0]--;
 890                 }
 891
 892                 /* pull out the item */
 893                 leaf = path->nodes[0];
 894                 item = btrfs_item_nr(leaf, path->slots[0]);
 895                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 896
 897                 /* make sure the item matches what we want */
 898                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
 899                         break;
 900                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
 901                         break;
 902
 903                 /* release the path since we're done with it */
 904                 btrfs_release_path(root, path);
 905
 906                 /*
 907                  * this is where we are basically btrfs_lookup, without the
 908                  * crossing root thing.  we store the inode number in the
 909                  * offset of the orphan item.
 910                  */
 911                 inode = btrfs_iget_locked(root->inode->i_sb,
 912                                           found_key.offset, root);
 913                 if (!inode)
 914                         break;
 915
 916                 if (inode->i_state & I_NEW) {
 917                         BTRFS_I(inode)->root = root;
 918
 919                         /* have to set the location manually */
 920                         BTRFS_I(inode)->location.objectid = inode->i_ino;
 921                         BTRFS_I(inode)->location.type = BTRFS_INODE_ITEM_KEY;
 922                         BTRFS_I(inode)->location.offset = 0;
 923
 924                         btrfs_read_locked_inode(inode);
 925                         unlock_new_inode(inode);
 926                 }
 927
 928                 /*
 929                  * add this inode to the orphan list so btrfs_orphan_del does
 930                  * the proper thing when we hit it
 931                  */
 932                 spin_lock(&root->list_lock);
 933                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
 934                 spin_unlock(&root->list_lock);
 935
 936                 /*
 937                  * if this is a bad inode, means we actually succeeded in
 938                  * removing the inode, but not the orphan record, which means
 939                  * we need to manually delete the orphan since iput will just
 940                  * do a destroy_inode
 941                  */
 942                 if (is_bad_inode(inode)) {
 943                         btrfs_orphan_del(trans, inode);
 944                         iput(inode);
 945                         continue;
 946                 }
 947
 948                 /* if we have links, this was a truncate, lets do that */
 949                 if (inode->i_nlink) {
 950                         nr_truncate++;
 951                         btrfs_truncate(inode);
 952                 } else {
 953                         nr_unlink++;
 954                 }
 955
 956                 /* this will do delete_inode and everything for us */
 957                 iput(inode);
 958         }
 959
 960         if (nr_unlink)
 961                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
 962         if (nr_truncate)
 963                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
 964
 965         btrfs_free_path(path);
 966         btrfs_end_transaction(trans, root);
 967 }
 968
 969 void btrfs_read_locked_inode(struct inode *inode)
 970 {
 971         struct btrfs_path *path;
 972         struct extent_buffer *leaf;
 973         struct btrfs_inode_item *inode_item;
 974         struct btrfs_timespec *tspec;
 975         struct btrfs_root *root = BTRFS_I(inode)->root;
 976         struct btrfs_key location;
 977         u64 alloc_group_block;
 978         u32 rdev;
 979         int ret;
 980
 981         path = btrfs_alloc_path();
 982         BUG_ON(!path);
 983         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 984
 985         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 986         if (ret)
 987                 goto make_bad;
 988
 989         leaf = path->nodes[0];
 990         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 991                                     struct btrfs_inode_item);
 992
 993         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 994         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 995         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 996         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 997         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
 998
 999         tspec = btrfs_inode_atime(inode_item);
1000         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1001         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1002
1003         tspec = btrfs_inode_mtime(inode_item);
1004         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1005         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1006
1007         tspec = btrfs_inode_ctime(inode_item);
1008         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
1009         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
1010
1011         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
1012         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
1013         inode->i_generation = BTRFS_I(inode)->generation;
1014         inode->i_rdev = 0;
1015         rdev = btrfs_inode_rdev(leaf, inode_item);
1016
1017         BTRFS_I(inode)->index_cnt = (u64)-1;
1018
1019         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
1020         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
1021                                                        alloc_group_block);
1022         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
1023         if (!BTRFS_I(inode)->block_group) {
1024                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
1025                                                  NULL, 0,
1026                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
1027         }
1028         btrfs_free_path(path);
1029         inode_item = NULL;
1030
1031         switch (inode->i_mode & S_IFMT) {
1032         case S_IFREG:
1033                 inode->i_mapping->a_ops = &btrfs_aops;
1034                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1035                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1036                 inode->i_fop = &btrfs_file_operations;
1037                 inode->i_op = &btrfs_file_inode_operations;
1038                 break;
1039         case S_IFDIR:
1040                 inode->i_fop = &btrfs_dir_file_operations;
1041                 if (root == root->fs_info->tree_root)
1042                         inode->i_op = &btrfs_dir_ro_inode_operations;
1043                 else
1044                         inode->i_op = &btrfs_dir_inode_operations;
1045                 break;
1046         case S_IFLNK:
1047                 inode->i_op = &btrfs_symlink_inode_operations;
1048                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
1049                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1050                 break;
1051         default:
1052                 init_special_inode(inode, inode->i_mode, rdev);
1053                 break;
1054         }
1055         return;
1056
1057 make_bad:
1058         btrfs_free_path(path);
1059         make_bad_inode(inode);
1060 }
1061
1062 static void fill_inode_item(struct btrfs_trans_handle *trans,
1063                             struct extent_buffer *leaf,
1064                             struct btrfs_inode_item *item,
1065                             struct inode *inode)
1066 {
1067         btrfs_set_inode_uid(leaf, item, inode->i_uid);
1068         btrfs_set_inode_gid(leaf, item, inode->i_gid);
1069         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
1070         btrfs_set_inode_mode(leaf, item, inode->i_mode);
1071         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
1072
1073         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
1074                                inode->i_atime.tv_sec);
1075         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
1076                                 inode->i_atime.tv_nsec);
1077
1078         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
1079                                inode->i_mtime.tv_sec);
1080         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
1081                                 inode->i_mtime.tv_nsec);
1082
1083         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
1084                                inode->i_ctime.tv_sec);
1085         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
1086                                 inode->i_ctime.tv_nsec);
1087
1088         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
1089         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
1090         btrfs_set_inode_transid(leaf, item, trans->transid);
1091         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
1092         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
1093         btrfs_set_inode_block_group(leaf, item,
1094                                     BTRFS_I(inode)->block_group->key.objectid);
1095 }
1096
1097 int noinline btrfs_update_inode(struct btrfs_trans_handle *trans,
1098                               struct btrfs_root *root,
1099                               struct inode *inode)
1100 {
1101         struct btrfs_inode_item *inode_item;
1102         struct btrfs_path *path;
1103         struct extent_buffer *leaf;
1104         int ret;
1105
1106         path = btrfs_alloc_path();
1107         BUG_ON(!path);
1108         ret = btrfs_lookup_inode(trans, root, path,
1109                                  &BTRFS_I(inode)->location, 1);
1110         if (ret) {
1111                 if (ret > 0)
1112                         ret = -ENOENT;
1113                 goto failed;
1114         }
1115
1116         leaf = path->nodes[0];
1117         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1118                                   struct btrfs_inode_item);
1119
1120         fill_inode_item(trans, leaf, inode_item, inode);
1121         btrfs_mark_buffer_dirty(leaf);
1122         btrfs_set_inode_last_trans(trans, inode);
1123         ret = 0;
1124 failed:
1125         btrfs_free_path(path);
1126         return ret;
1127 }
1128
1129
1130 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
1131                        struct btrfs_root *root,
1132                        struct inode *dir, struct inode *inode,
1133                        const char *name, int name_len)
1134 {
1135         struct btrfs_path *path;
1136         int ret = 0;
1137         struct extent_buffer *leaf;
1138         struct btrfs_dir_item *di;
1139         struct btrfs_key key;
1140         u64 index;
1141
1142         path = btrfs_alloc_path();
1143         if (!path) {
1144                 ret = -ENOMEM;
1145                 goto err;
1146         }
1147
1148         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
1149                                     name, name_len, -1);
1150         if (IS_ERR(di)) {
1151                 ret = PTR_ERR(di);
1152                 goto err;
1153         }
1154         if (!di) {
1155                 ret = -ENOENT;
1156                 goto err;
1157         }
1158         leaf = path->nodes[0];
1159         btrfs_dir_item_key_to_cpu(leaf, di, &key);
1160         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1161         if (ret)
1162                 goto err;
1163         btrfs_release_path(root, path);
1164
1165         ret = btrfs_del_inode_ref(trans, root, name, name_len,
1166                                   inode->i_ino,
1167                                   dir->i_ino, &index);
1168         if (ret) {
1169                 printk("failed to delete reference to %.*s, "
1170                        "inode %lu parent %lu\n", name_len, name,
1171                        inode->i_ino, dir->i_ino);
1172                 goto err;
1173         }
1174
1175         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
1176                                          index, name, name_len, -1);
1177         if (IS_ERR(di)) {
1178                 ret = PTR_ERR(di);
1179                 goto err;
1180         }
1181         if (!di) {
1182                 ret = -ENOENT;
1183                 goto err;
1184         }
1185         ret = btrfs_delete_one_dir_name(trans, root, path, di);
1186         btrfs_release_path(root, path);
1187
1188         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
1189                                          inode, dir->i_ino);
1190         BUG_ON(ret);
1191
1192         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
1193                                            dir, index);
1194         BUG_ON(ret);
1195 err:
1196         btrfs_free_path(path);
1197         if (ret)
1198                 goto out;
1199
1200         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
1201         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1202         btrfs_update_inode(trans, root, dir);
1203         btrfs_drop_nlink(inode);
1204         ret = btrfs_update_inode(trans, root, inode);
1205         dir->i_sb->s_dirt = 1;
1206 out:
1207         return ret;
1208 }
1209
1210 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
1211 {
1212         struct btrfs_root *root;
1213         struct btrfs_trans_handle *trans;
1214         struct inode *inode = dentry->d_inode;
1215         int ret;
1216         unsigned long nr = 0;
1217
1218         root = BTRFS_I(dir)->root;
1219
1220         ret = btrfs_check_free_space(root, 1, 1);
1221         if (ret)
1222                 goto fail;
1223
1224         trans = btrfs_start_transaction(root, 1);
1225
1226         btrfs_set_trans_block_group(trans, dir);
1227         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1228                                  dentry->d_name.name, dentry->d_name.len);
1229
1230         if (inode->i_nlink == 0)
1231                 ret = btrfs_orphan_add(trans, inode);
1232
1233         nr = trans->blocks_used;
1234
1235         btrfs_end_transaction_throttle(trans, root);
1236 fail:
1237         btrfs_btree_balance_dirty(root, nr);
1238         return ret;
1239 }
1240
1241 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
1242 {
1243         struct inode *inode = dentry->d_inode;
1244         int err = 0;
1245         int ret;
1246         struct btrfs_root *root = BTRFS_I(dir)->root;
1247         struct btrfs_trans_handle *trans;
1248         unsigned long nr = 0;
1249
1250         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
1251                 return -ENOTEMPTY;
1252         }
1253
1254         ret = btrfs_check_free_space(root, 1, 1);
1255         if (ret)
1256                 goto fail;
1257
1258         trans = btrfs_start_transaction(root, 1);
1259         btrfs_set_trans_block_group(trans, dir);
1260
1261         err = btrfs_orphan_add(trans, inode);
1262         if (err)
1263                 goto fail_trans;
1264
1265         /* now the directory is empty */
1266         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
1267                                  dentry->d_name.name, dentry->d_name.len);
1268         if (!err) {
1269                 btrfs_i_size_write(inode, 0);
1270         }
1271
1272 fail_trans:
1273         nr = trans->blocks_used;
1274         ret = btrfs_end_transaction_throttle(trans, root);
1275 fail:
1276         btrfs_btree_balance_dirty(root, nr);
1277
1278         if (ret && !err)
1279                 err = ret;
1280         return err;
1281 }
1282
1283 /*
1284  * this can truncate away extent items, csum items and directory items.
1285  * It starts at a high offset and removes keys until it can't find
1286  * any higher than i_size.
1287  *
1288  * csum items that cross the new i_size are truncated to the new size
1289  * as well.
1290  *
1291  * min_type is the minimum key type to truncate down to.  If set to 0, this
1292  * will kill all the items on this inode, including the INODE_ITEM_KEY.
1293  */
1294 noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
1295                                         struct btrfs_root *root,
1296                                         struct inode *inode,
1297                                         u64 new_size, u32 min_type)
1298 {
1299         int ret;
1300         struct btrfs_path *path;
1301         struct btrfs_key key;
1302         struct btrfs_key found_key;
1303         u32 found_type;
1304         struct extent_buffer *leaf;
1305         struct btrfs_file_extent_item *fi;
1306         u64 extent_start = 0;
1307         u64 extent_num_bytes = 0;
1308         u64 item_end = 0;
1309         u64 root_gen = 0;
1310         u64 root_owner = 0;
1311         int found_extent;
1312         int del_item;
1313         int pending_del_nr = 0;
1314         int pending_del_slot = 0;
1315         int extent_type = -1;
1316         u64 mask = root->sectorsize - 1;
1317
1318         if (root->ref_cows)
1319                 btrfs_drop_extent_cache(inode,
1320                                         new_size & (~mask), (u64)-1);
1321         path = btrfs_alloc_path();
1322         path->reada = -1;
1323         BUG_ON(!path);
1324
1325         /* FIXME, add redo link to tree so we don't leak on crash */
1326         key.objectid = inode->i_ino;
1327         key.offset = (u64)-1;
1328         key.type = (u8)-1;
1329
1330         btrfs_init_path(path);
1331 search_again:
1332         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1333         if (ret < 0) {
1334                 goto error;
1335         }
1336         if (ret > 0) {
1337                 /* there are no items in the tree for us to truncate, we're
1338                  * done
1339                  */
1340                 if (path->slots[0] == 0) {
1341                         ret = 0;
1342                         goto error;
1343                 }
1344                 path->slots[0]--;
1345         }
1346
1347         while(1) {
1348                 fi = NULL;
1349                 leaf = path->nodes[0];
1350                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1351                 found_type = btrfs_key_type(&found_key);
1352
1353                 if (found_key.objectid != inode->i_ino)
1354                         break;
1355
1356                 if (found_type < min_type)
1357                         break;
1358
1359                 item_end = found_key.offset;
1360                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
1361                         fi = btrfs_item_ptr(leaf, path->slots[0],
1362                                             struct btrfs_file_extent_item);
1363                         extent_type = btrfs_file_extent_type(leaf, fi);
1364                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1365                                 item_end +=
1366                                     btrfs_file_extent_num_bytes(leaf, fi);
1367                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1368                                 struct btrfs_item *item = btrfs_item_nr(leaf,
1369                                                                 path->slots[0]);
1370                                 item_end += btrfs_file_extent_inline_len(leaf,
1371                                                                          item);
1372                         }
1373                         item_end--;
1374                 }
1375                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
1376                         ret = btrfs_csum_truncate(trans, root, path,
1377                                                   new_size);
1378                         BUG_ON(ret);
1379                 }
1380                 if (item_end < new_size) {
1381                         if (found_type == BTRFS_DIR_ITEM_KEY) {
1382                                 found_type = BTRFS_INODE_ITEM_KEY;
1383                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
1384                                 found_type = BTRFS_CSUM_ITEM_KEY;
1385                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
1386                                 found_type = BTRFS_XATTR_ITEM_KEY;
1387                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
1388                                 found_type = BTRFS_INODE_REF_KEY;
1389                         } else if (found_type) {
1390                                 found_type--;
1391                         } else {
1392                                 break;
1393                         }
1394                         btrfs_set_key_type(&key, found_type);
1395                         goto next;
1396                 }
1397                 if (found_key.offset >= new_size)
1398                         del_item = 1;
1399                 else
1400                         del_item = 0;
1401                 found_extent = 0;
1402
1403                 /* FIXME, shrink the extent if the ref count is only 1 */
1404                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1405                         goto delete;
1406
1407                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1408                         u64 num_dec;
1409                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1410                         if (!del_item) {
1411                                 u64 orig_num_bytes =
1412                                         btrfs_file_extent_num_bytes(leaf, fi);
1413                                 extent_num_bytes = new_size -
1414                                         found_key.offset + root->sectorsize - 1;
1415                                 extent_num_bytes = extent_num_bytes &
1416                                         ~((u64)root->sectorsize - 1);
1417                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1418                                                          extent_num_bytes);
1419                                 num_dec = (orig_num_bytes -
1420                                            extent_num_bytes);
1421                                 if (root->ref_cows && extent_start != 0)
1422                                         dec_i_blocks(inode, num_dec);
1423                                 btrfs_mark_buffer_dirty(leaf);
1424                         } else {
1425                                 extent_num_bytes =
1426                                         btrfs_file_extent_disk_num_bytes(leaf,
1427                                                                          fi);
1428                                 /* FIXME blocksize != 4096 */
1429                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1430                                 if (extent_start != 0) {
1431                                         found_extent = 1;
1432                                         if (root->ref_cows)
1433                                                 dec_i_blocks(inode, num_dec);
1434                                 }
1435                                 if (root->ref_cows) {
1436                                         root_gen =
1437                                                 btrfs_header_generation(leaf);
1438                                 }
1439                                 root_owner = btrfs_header_owner(leaf);
1440                         }
1441                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1442                         if (!del_item) {
1443                                 u32 size = new_size - found_key.offset;
1444
1445                                 if (root->ref_cows) {
1446                                         dec_i_blocks(inode, item_end + 1 -
1447                                                     found_key.offset - size);
1448                                 }
1449                                 size =
1450                                     btrfs_file_extent_calc_inline_size(size);
1451                                 ret = btrfs_truncate_item(trans, root, path,
1452                                                           size, 1);
1453                                 BUG_ON(ret);
1454                         } else if (root->ref_cows) {
1455                                 dec_i_blocks(inode, item_end + 1 -
1456                                              found_key.offset);
1457                         }
1458                 }
1459 delete:
1460                 if (del_item) {
1461                         if (!pending_del_nr) {
1462                                 /* no pending yet, add ourselves */
1463                                 pending_del_slot = path->slots[0];
1464                                 pending_del_nr = 1;
1465                         } else if (pending_del_nr &&
1466                                    path->slots[0] + 1 == pending_del_slot) {
1467                                 /* hop on the pending chunk */
1468                                 pending_del_nr++;
1469                                 pending_del_slot = path->slots[0];
1470                         } else {
1471                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1472                         }
1473                 } else {
1474                         break;
1475                 }
1476                 if (found_extent) {
1477                         ret = btrfs_free_extent(trans, root, extent_start,
1478                                                 extent_num_bytes,
1479                                                 root_owner,
1480                                                 root_gen, inode->i_ino,
1481                                                 found_key.offset, 0);
1482                         BUG_ON(ret);
1483                 }
1484 next:
1485                 if (path->slots[0] == 0) {
1486                         if (pending_del_nr)
1487                                 goto del_pending;
1488                         btrfs_release_path(root, path);
1489                         goto search_again;
1490                 }
1491
1492                 path->slots[0]--;
1493                 if (pending_del_nr &&
1494                     path->slots[0] + 1 != pending_del_slot) {
1495                         struct btrfs_key debug;
1496 del_pending:
1497                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1498                                               pending_del_slot);
1499                         ret = btrfs_del_items(trans, root, path,
1500                                               pending_del_slot,
1501                                               pending_del_nr);
1502                         BUG_ON(ret);
1503                         pending_del_nr = 0;
1504                         btrfs_release_path(root, path);
1505                         goto search_again;
1506                 }
1507         }
1508         ret = 0;
1509 error:
1510         if (pending_del_nr) {
1511                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1512                                       pending_del_nr);
1513         }
1514         btrfs_free_path(path);
1515         inode->i_sb->s_dirt = 1;
1516         return ret;
1517 }
1518
1519 /*
1520  * taken from block_truncate_page, but does cow as it zeros out
1521  * any bytes left in the last page in the file.
1522  */
1523 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1524 {
1525         struct inode *inode = mapping->host;
1526         struct btrfs_root *root = BTRFS_I(inode)->root;
1527         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1528         struct btrfs_ordered_extent *ordered;
1529         char *kaddr;
1530         u32 blocksize = root->sectorsize;
1531         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1532         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1533         struct page *page;
1534         int ret = 0;
1535         u64 page_start;
1536         u64 page_end;
1537
1538         if ((offset & (blocksize - 1)) == 0)
1539                 goto out;
1540
1541         ret = -ENOMEM;
1542 again:
1543         page = grab_cache_page(mapping, index);
1544         if (!page)
1545                 goto out;
1546
1547         page_start = page_offset(page);
1548         page_end = page_start + PAGE_CACHE_SIZE - 1;
1549
1550         if (!PageUptodate(page)) {
1551                 ret = btrfs_readpage(NULL, page);
1552                 lock_page(page);
1553                 if (page->mapping != mapping) {
1554                         unlock_page(page);
1555                         page_cache_release(page);
1556                         goto again;
1557                 }
1558                 if (!PageUptodate(page)) {
1559                         ret = -EIO;
1560                         goto out_unlock;
1561                 }
1562         }
1563         wait_on_page_writeback(page);
1564
1565         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1566         set_page_extent_mapped(page);
1567
1568         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1569         if (ordered) {
1570                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1571                 unlock_page(page);
1572                 page_cache_release(page);
1573                 btrfs_start_ordered_extent(inode, ordered, 1);
1574                 btrfs_put_ordered_extent(ordered);
1575                 goto again;
1576         }
1577
1578         btrfs_set_extent_delalloc(inode, page_start, page_end);
1579         ret = 0;
1580         if (offset != PAGE_CACHE_SIZE) {
1581                 kaddr = kmap(page);
1582                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1583                 flush_dcache_page(page);
1584                 kunmap(page);
1585         }
1586         ClearPageChecked(page);
1587         set_page_dirty(page);
1588         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1589
1590 out_unlock:
1591         unlock_page(page);
1592         page_cache_release(page);
1593 out:
1594         return ret;
1595 }
1596
1597 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1598 {
1599         struct inode *inode = dentry->d_inode;
1600         int err;
1601
1602         err = inode_change_ok(inode, attr);
1603         if (err)
1604                 return err;
1605
1606         if (S_ISREG(inode->i_mode) &&
1607             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1608                 struct btrfs_trans_handle *trans;
1609                 struct btrfs_root *root = BTRFS_I(inode)->root;
1610                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1611
1612                 u64 mask = root->sectorsize - 1;
1613                 u64 hole_start = (inode->i_size + mask) & ~mask;
1614                 u64 block_end = (attr->ia_size + mask) & ~mask;
1615                 u64 hole_size;
1616                 u64 alloc_hint = 0;
1617
1618                 if (attr->ia_size <= hole_start)
1619                         goto out;
1620
1621                 err = btrfs_check_free_space(root, 1, 0);
1622                 if (err)
1623                         goto fail;
1624
1625                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1626
1627                 hole_size = block_end - hole_start;
1628                 while(1) {
1629                         struct btrfs_ordered_extent *ordered;
1630                         btrfs_wait_ordered_range(inode, hole_start, hole_size);
1631
1632                         lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1633                         ordered = btrfs_lookup_ordered_extent(inode, hole_start);
1634                         if (ordered) {
1635                                 unlock_extent(io_tree, hole_start,
1636                                               block_end - 1, GFP_NOFS);
1637                                 btrfs_put_ordered_extent(ordered);
1638                         } else {
1639                                 break;
1640                         }
1641                 }
1642
1643                 trans = btrfs_start_transaction(root, 1);
1644                 btrfs_set_trans_block_group(trans, inode);
1645                 mutex_lock(&BTRFS_I(inode)->extent_mutex);
1646                 err = btrfs_drop_extents(trans, root, inode,
1647                                          hole_start, block_end, hole_start,
1648                                          &alloc_hint);
1649
1650                 if (alloc_hint != EXTENT_MAP_INLINE) {
1651                         err = btrfs_insert_file_extent(trans, root,
1652                                                        inode->i_ino,
1653                                                        hole_start, 0, 0,
1654                                                        hole_size, 0);
1655                         btrfs_drop_extent_cache(inode, hole_start,
1656                                                 (u64)-1);
1657                         btrfs_check_file(root, inode);
1658                 }
1659                 mutex_unlock(&BTRFS_I(inode)->extent_mutex);
1660                 btrfs_end_transaction(trans, root);
1661                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1662                 if (err)
1663                         return err;
1664         }
1665 out:
1666         err = inode_setattr(inode, attr);
1667
1668         if (!err && ((attr->ia_valid & ATTR_MODE)))
1669                 err = btrfs_acl_chmod(inode);
1670 fail:
1671         return err;
1672 }
1673
1674 void btrfs_delete_inode(struct inode *inode)
1675 {
1676         struct btrfs_trans_handle *trans;
1677         struct btrfs_root *root = BTRFS_I(inode)->root;
1678         unsigned long nr;
1679         int ret;
1680
1681         truncate_inode_pages(&inode->i_data, 0);
1682         if (is_bad_inode(inode)) {
1683                 btrfs_orphan_del(NULL, inode);
1684                 goto no_delete;
1685         }
1686         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1687
1688         btrfs_i_size_write(inode, 0);
1689         trans = btrfs_start_transaction(root, 1);
1690
1691         btrfs_set_trans_block_group(trans, inode);
1692         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
1693         if (ret) {
1694                 btrfs_orphan_del(NULL, inode);
1695                 goto no_delete_lock;
1696         }
1697
1698         btrfs_orphan_del(trans, inode);
1699
1700         nr = trans->blocks_used;
1701         clear_inode(inode);
1702
1703         btrfs_end_transaction(trans, root);
1704         btrfs_btree_balance_dirty(root, nr);
1705         return;
1706
1707 no_delete_lock:
1708         nr = trans->blocks_used;
1709         btrfs_end_transaction(trans, root);
1710         btrfs_btree_balance_dirty(root, nr);
1711 no_delete:
1712         clear_inode(inode);
1713 }
1714
1715 /*
1716  * this returns the key found in the dir entry in the location pointer.
1717  * If no dir entries were found, location->objectid is 0.
1718  */
1719 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1720                                struct btrfs_key *location)
1721 {
1722         const char *name = dentry->d_name.name;
1723         int namelen = dentry->d_name.len;
1724         struct btrfs_dir_item *di;
1725         struct btrfs_path *path;
1726         struct btrfs_root *root = BTRFS_I(dir)->root;
1727         int ret = 0;
1728
1729         path = btrfs_alloc_path();
1730         BUG_ON(!path);
1731
1732         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1733                                     namelen, 0);
1734         if (IS_ERR(di))
1735                 ret = PTR_ERR(di);
1736         if (!di || IS_ERR(di)) {
1737                 goto out_err;
1738         }
1739         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1740 out:
1741         btrfs_free_path(path);
1742         return ret;
1743 out_err:
1744         location->objectid = 0;
1745         goto out;
1746 }
1747
1748 /*
1749  * when we hit a tree root in a directory, the btrfs part of the inode
1750  * needs to be changed to reflect the root directory of the tree root.  This
1751  * is kind of like crossing a mount point.
1752  */
1753 static int fixup_tree_root_location(struct btrfs_root *root,
1754                              struct btrfs_key *location,
1755                              struct btrfs_root **sub_root,
1756                              struct dentry *dentry)
1757 {
1758         struct btrfs_root_item *ri;
1759
1760         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1761                 return 0;
1762         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1763                 return 0;
1764
1765         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1766                                         dentry->d_name.name,
1767                                         dentry->d_name.len);
1768         if (IS_ERR(*sub_root))
1769                 return PTR_ERR(*sub_root);
1770
1771         ri = &(*sub_root)->root_item;
1772         location->objectid = btrfs_root_dirid(ri);
1773         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1774         location->offset = 0;
1775
1776         return 0;
1777 }
1778
1779 static noinline void init_btrfs_i(struct inode *inode)
1780 {
1781         struct btrfs_inode *bi = BTRFS_I(inode);
1782
1783         bi->i_acl = NULL;
1784         bi->i_default_acl = NULL;
1785
1786         bi->generation = 0;
1787         bi->last_trans = 0;
1788         bi->logged_trans = 0;
1789         bi->delalloc_bytes = 0;
1790         bi->disk_i_size = 0;
1791         bi->flags = 0;
1792         bi->index_cnt = (u64)-1;
1793         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1794         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1795                              inode->i_mapping, GFP_NOFS);
1796         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1797                              inode->i_mapping, GFP_NOFS);
1798         INIT_LIST_HEAD(&BTRFS_I(inode)->delalloc_inodes);
1799         btrfs_ordered_inode_tree_init(&BTRFS_I(inode)->ordered_tree);
1800         mutex_init(&BTRFS_I(inode)->csum_mutex);
1801         mutex_init(&BTRFS_I(inode)->extent_mutex);
1802         mutex_init(&BTRFS_I(inode)->log_mutex);
1803 }
1804
1805 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1806 {
1807         struct btrfs_iget_args *args = p;
1808         inode->i_ino = args->ino;
1809         init_btrfs_i(inode);
1810         BTRFS_I(inode)->root = args->root;
1811         return 0;
1812 }
1813
1814 static int btrfs_find_actor(struct inode *inode, void *opaque)
1815 {
1816         struct btrfs_iget_args *args = opaque;
1817         return (args->ino == inode->i_ino &&
1818                 args->root == BTRFS_I(inode)->root);
1819 }
1820
1821 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1822                             u64 root_objectid)
1823 {
1824         struct btrfs_iget_args args;
1825         args.ino = objectid;
1826         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1827
1828         if (!args.root)
1829                 return NULL;
1830
1831         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1832 }
1833
1834 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1835                                 struct btrfs_root *root)
1836 {
1837         struct inode *inode;
1838         struct btrfs_iget_args args;
1839         args.ino = objectid;
1840         args.root = root;
1841
1842         inode = iget5_locked(s, objectid, btrfs_find_actor,
1843                              btrfs_init_locked_inode,
1844                              (void *)&args);
1845         return inode;
1846 }
1847
1848 /* Get an inode object given its location and corresponding root.
1849  * Returns in *is_new if the inode was read from disk
1850  */
1851 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
1852                          struct btrfs_root *root, int *is_new)
1853 {
1854         struct inode *inode;
1855
1856         inode = btrfs_iget_locked(s, location->objectid, root);
1857         if (!inode)
1858                 return ERR_PTR(-EACCES);
1859
1860         if (inode->i_state & I_NEW) {
1861                 BTRFS_I(inode)->root = root;
1862                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
1863                 btrfs_read_locked_inode(inode);
1864                 unlock_new_inode(inode);
1865                 if (is_new)
1866                         *is_new = 1;
1867         } else {
1868                 if (is_new)
1869                         *is_new = 0;
1870         }
1871
1872         return inode;
1873 }
1874
1875 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1876                                    struct nameidata *nd)
1877 {
1878         struct inode * inode;
1879         struct btrfs_inode *bi = BTRFS_I(dir);
1880         struct btrfs_root *root = bi->root;
1881         struct btrfs_root *sub_root = root;
1882         struct btrfs_key location;
1883         int ret, new, do_orphan = 0;
1884
1885         if (dentry->d_name.len > BTRFS_NAME_LEN)
1886                 return ERR_PTR(-ENAMETOOLONG);
1887
1888         ret = btrfs_inode_by_name(dir, dentry, &location);
1889
1890         if (ret < 0)
1891                 return ERR_PTR(ret);
1892
1893         inode = NULL;
1894         if (location.objectid) {
1895                 ret = fixup_tree_root_location(root, &location, &sub_root,
1896                                                 dentry);
1897                 if (ret < 0)
1898                         return ERR_PTR(ret);
1899                 if (ret > 0)
1900                         return ERR_PTR(-ENOENT);
1901                 inode = btrfs_iget(dir->i_sb, &location, sub_root, &new);
1902                 if (IS_ERR(inode))
1903                         return ERR_CAST(inode);
1904
1905                 /* the inode and parent dir are two different roots */
1906                 if (new && root != sub_root) {
1907                         igrab(inode);
1908                         sub_root->inode = inode;
1909                         do_orphan = 1;
1910                 }
1911         }
1912
1913         if (unlikely(do_orphan))
1914                 btrfs_orphan_cleanup(sub_root);
1915
1916         return d_splice_alias(inode, dentry);
1917 }
1918
1919 static unsigned char btrfs_filetype_table[] = {
1920         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1921 };
1922
1923 static int btrfs_real_readdir(struct file *filp, void *dirent,
1924                               filldir_t filldir)
1925 {
1926         struct inode *inode = filp->f_dentry->d_inode;
1927         struct btrfs_root *root = BTRFS_I(inode)->root;
1928         struct btrfs_item *item;
1929         struct btrfs_dir_item *di;
1930         struct btrfs_key key;
1931         struct btrfs_key found_key;
1932         struct btrfs_path *path;
1933         int ret;
1934         u32 nritems;
1935         struct extent_buffer *leaf;
1936         int slot;
1937         int advance;
1938         unsigned char d_type;
1939         int over = 0;
1940         u32 di_cur;
1941         u32 di_total;
1942         u32 di_len;
1943         int key_type = BTRFS_DIR_INDEX_KEY;
1944         char tmp_name[32];
1945         char *name_ptr;
1946         int name_len;
1947
1948         /* FIXME, use a real flag for deciding about the key type */
1949         if (root->fs_info->tree_root == root)
1950                 key_type = BTRFS_DIR_ITEM_KEY;
1951
1952         /* special case for "." */
1953         if (filp->f_pos == 0) {
1954                 over = filldir(dirent, ".", 1,
1955                                1, inode->i_ino,
1956                                DT_DIR);
1957                 if (over)
1958                         return 0;
1959                 filp->f_pos = 1;
1960         }
1961         /* special case for .., just use the back ref */
1962         if (filp->f_pos == 1) {
1963                 u64 pino = parent_ino(filp->f_path.dentry);
1964                 over = filldir(dirent, "..", 2,
1965                                2, pino, DT_DIR);
1966                 if (over)
1967                         return 0;
1968                 filp->f_pos = 2;
1969         }
1970
1971         path = btrfs_alloc_path();
1972         path->reada = 2;
1973
1974         btrfs_set_key_type(&key, key_type);
1975         key.offset = filp->f_pos;
1976         key.objectid = inode->i_ino;
1977
1978         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1979         if (ret < 0)
1980                 goto err;
1981         advance = 0;
1982
1983         while (1) {
1984                 leaf = path->nodes[0];
1985                 nritems = btrfs_header_nritems(leaf);
1986                 slot = path->slots[0];
1987                 if (advance || slot >= nritems) {
1988                         if (slot >= nritems - 1) {
1989                                 ret = btrfs_next_leaf(root, path);
1990                                 if (ret)
1991                                         break;
1992                                 leaf = path->nodes[0];
1993                                 nritems = btrfs_header_nritems(leaf);
1994                                 slot = path->slots[0];
1995                         } else {
1996                                 slot++;
1997                                 path->slots[0]++;
1998                         }
1999                 }
2000                 advance = 1;
2001                 item = btrfs_item_nr(leaf, slot);
2002                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2003
2004                 if (found_key.objectid != key.objectid)
2005                         break;
2006                 if (btrfs_key_type(&found_key) != key_type)
2007                         break;
2008                 if (found_key.offset < filp->f_pos)
2009                         continue;
2010
2011                 filp->f_pos = found_key.offset;
2012
2013                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
2014                 di_cur = 0;
2015                 di_total = btrfs_item_size(leaf, item);
2016
2017                 while (di_cur < di_total) {
2018                         struct btrfs_key location;
2019
2020                         name_len = btrfs_dir_name_len(leaf, di);
2021                         if (name_len <= sizeof(tmp_name)) {
2022                                 name_ptr = tmp_name;
2023                         } else {
2024                                 name_ptr = kmalloc(name_len, GFP_NOFS);
2025                                 if (!name_ptr) {
2026                                         ret = -ENOMEM;
2027                                         goto err;
2028                                 }
2029                         }
2030                         read_extent_buffer(leaf, name_ptr,
2031                                            (unsigned long)(di + 1), name_len);
2032
2033                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
2034                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
2035                         over = filldir(dirent, name_ptr, name_len,
2036                                        found_key.offset, location.objectid,
2037                                        d_type);
2038
2039                         if (name_ptr != tmp_name)
2040                                 kfree(name_ptr);
2041
2042                         if (over)
2043                                 goto nopos;
2044
2045                         di_len = btrfs_dir_name_len(leaf, di) +
2046                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
2047                         di_cur += di_len;
2048                         di = (struct btrfs_dir_item *)((char *)di + di_len);
2049                 }
2050         }
2051
2052         /* Reached end of directory/root. Bump pos past the last item. */
2053         if (key_type == BTRFS_DIR_INDEX_KEY)
2054                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
2055         else
2056                 filp->f_pos++;
2057 nopos:
2058         ret = 0;
2059 err:
2060         btrfs_free_path(path);
2061         return ret;
2062 }
2063
2064 /* Kernels earlier than 2.6.28 still have the NFS deadlock where nfsd
2065    will call the file system's ->lookup() method from within its
2066    filldir callback, which in turn was called from the file system's
2067    ->readdir() method. And will deadlock for many file systems. */
2068 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
2069
2070 struct nfshack_dirent {
2071         u64             ino;
2072         loff_t          offset;
2073         int             namlen;
2074         unsigned int    d_type;
2075         char            name[];
2076 };
2077
2078 struct nfshack_readdir {
2079         char            *dirent;
2080         size_t          used;
2081         int             full;
2082 };
2083
2084
2085
2086 static int btrfs_nfshack_filldir(void *__buf, const char *name, int namlen,
2087                               loff_t offset, u64 ino, unsigned int d_type)
2088 {
2089         struct nfshack_readdir *buf = __buf;
2090         struct nfshack_dirent *de = (void *)(buf->dirent + buf->used);
2091         unsigned int reclen;
2092
2093         reclen = ALIGN(sizeof(struct nfshack_dirent) + namlen, sizeof(u64));
2094         if (buf->used + reclen > PAGE_SIZE) {
2095                 buf->full = 1;
2096                 return -EINVAL;
2097         }
2098
2099         de->namlen = namlen;
2100         de->offset = offset;
2101         de->ino = ino;
2102         de->d_type = d_type;
2103         memcpy(de->name, name, namlen);
2104         buf->used += reclen;
2105
2106         return 0;
2107 }
2108
2109 static int btrfs_nfshack_readdir(struct file *file, void *dirent,
2110                                  filldir_t filldir)
2111 {
2112         struct nfshack_readdir buf;
2113         struct nfshack_dirent *de;
2114         int err;
2115         int size;
2116         loff_t offset;
2117
2118         buf.dirent = (void *)__get_free_page(GFP_KERNEL);
2119         if (!buf.dirent)
2120                 return -ENOMEM;
2121
2122         offset = file->f_pos;
2123
2124         do {
2125                 unsigned int reclen;
2126
2127                 buf.used = 0;
2128                 buf.full = 0;
2129                 err = btrfs_real_readdir(file, &buf, btrfs_nfshack_filldir);
2130                 if (err)
2131                         break;
2132
2133                 size = buf.used;
2134
2135                 if (!size)
2136                         break;
2137
2138                 de = (struct nfshack_dirent *)buf.dirent;
2139                 while (size > 0) {
2140                         offset = de->offset;
2141
2142                         if (filldir(dirent, de->name, de->namlen, de->offset,
2143                                     de->ino, de->d_type))
2144                                 goto done;
2145                         offset = file->f_pos;
2146
2147                         reclen = ALIGN(sizeof(*de) + de->namlen,
2148                                        sizeof(u64));
2149                         size -= reclen;
2150                         de = (struct nfshack_dirent *)((char *)de + reclen);
2151                 }
2152         } while (buf.full);
2153
2154  done:
2155         free_page((unsigned long)buf.dirent);
2156         file->f_pos = offset;
2157
2158         return err;
2159 }
2160 #endif
2161
2162 int btrfs_write_inode(struct inode *inode, int wait)
2163 {
2164         struct btrfs_root *root = BTRFS_I(inode)->root;
2165         struct btrfs_trans_handle *trans;
2166         int ret = 0;
2167
2168         if (root->fs_info->closing > 1)
2169                 return 0;
2170
2171         if (wait) {
2172                 trans = btrfs_join_transaction(root, 1);
2173                 btrfs_set_trans_block_group(trans, inode);
2174                 ret = btrfs_commit_transaction(trans, root);
2175         }
2176         return ret;
2177 }
2178
2179 /*
2180  * This is somewhat expensive, updating the tree every time the
2181  * inode changes.  But, it is most likely to find the inode in cache.
2182  * FIXME, needs more benchmarking...there are no reasons other than performance
2183  * to keep or drop this code.
2184  */
2185 void btrfs_dirty_inode(struct inode *inode)
2186 {
2187         struct btrfs_root *root = BTRFS_I(inode)->root;
2188         struct btrfs_trans_handle *trans;
2189
2190         trans = btrfs_join_transaction(root, 1);
2191         btrfs_set_trans_block_group(trans, inode);
2192         btrfs_update_inode(trans, root, inode);
2193         btrfs_end_transaction(trans, root);
2194 }
2195
2196 static int btrfs_set_inode_index_count(struct inode *inode)
2197 {
2198         struct btrfs_root *root = BTRFS_I(inode)->root;
2199         struct btrfs_key key, found_key;
2200         struct btrfs_path *path;
2201         struct extent_buffer *leaf;
2202         int ret;
2203
2204         key.objectid = inode->i_ino;
2205         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
2206         key.offset = (u64)-1;
2207
2208         path = btrfs_alloc_path();
2209         if (!path)
2210                 return -ENOMEM;
2211
2212         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2213         if (ret < 0)
2214                 goto out;
2215         /* FIXME: we should be able to handle this */
2216         if (ret == 0)
2217                 goto out;
2218         ret = 0;
2219
2220         /*
2221          * MAGIC NUMBER EXPLANATION:
2222          * since we search a directory based on f_pos we have to start at 2
2223          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
2224          * else has to start at 2
2225          */
2226         if (path->slots[0] == 0) {
2227                 BTRFS_I(inode)->index_cnt = 2;
2228                 goto out;
2229         }
2230
2231         path->slots[0]--;
2232
2233         leaf = path->nodes[0];
2234         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2235
2236         if (found_key.objectid != inode->i_ino ||
2237             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
2238                 BTRFS_I(inode)->index_cnt = 2;
2239                 goto out;
2240         }
2241
2242         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
2243 out:
2244         btrfs_free_path(path);
2245         return ret;
2246 }
2247
2248 static int btrfs_set_inode_index(struct inode *dir, struct inode *inode,
2249                                  u64 *index)
2250 {
2251         int ret = 0;
2252
2253         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
2254                 ret = btrfs_set_inode_index_count(dir);
2255                 if (ret)
2256                         return ret;
2257         }
2258
2259         *index = BTRFS_I(dir)->index_cnt;
2260         BTRFS_I(dir)->index_cnt++;
2261
2262         return ret;
2263 }
2264
2265 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
2266                                      struct btrfs_root *root,
2267                                      struct inode *dir,
2268                                      const char *name, int name_len,
2269                                      u64 ref_objectid,
2270                                      u64 objectid,
2271                                      struct btrfs_block_group_cache *group,
2272                                      int mode, u64 *index)
2273 {
2274         struct inode *inode;
2275         struct btrfs_inode_item *inode_item;
2276         struct btrfs_block_group_cache *new_inode_group;
2277         struct btrfs_key *location;
2278         struct btrfs_path *path;
2279         struct btrfs_inode_ref *ref;
2280         struct btrfs_key key[2];
2281         u32 sizes[2];
2282         unsigned long ptr;
2283         int ret;
2284         int owner;
2285
2286         path = btrfs_alloc_path();
2287         BUG_ON(!path);
2288
2289         inode = new_inode(root->fs_info->sb);
2290         if (!inode)
2291                 return ERR_PTR(-ENOMEM);
2292
2293         if (dir) {
2294                 ret = btrfs_set_inode_index(dir, inode, index);
2295                 if (ret)
2296                         return ERR_PTR(ret);
2297         }
2298         /*
2299          * index_cnt is ignored for everything but a dir,
2300          * btrfs_get_inode_index_count has an explanation for the magic
2301          * number
2302          */
2303         init_btrfs_i(inode);
2304         BTRFS_I(inode)->index_cnt = 2;
2305         BTRFS_I(inode)->root = root;
2306         BTRFS_I(inode)->generation = trans->transid;
2307
2308         if (mode & S_IFDIR)
2309                 owner = 0;
2310         else
2311                 owner = 1;
2312         new_inode_group = btrfs_find_block_group(root, group, 0,
2313                                        BTRFS_BLOCK_GROUP_METADATA, owner);
2314         if (!new_inode_group) {
2315                 printk("find_block group failed\n");
2316                 new_inode_group = group;
2317         }
2318         BTRFS_I(inode)->block_group = new_inode_group;
2319
2320         key[0].objectid = objectid;
2321         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
2322         key[0].offset = 0;
2323
2324         key[1].objectid = objectid;
2325         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
2326         key[1].offset = ref_objectid;
2327
2328         sizes[0] = sizeof(struct btrfs_inode_item);
2329         sizes[1] = name_len + sizeof(*ref);
2330
2331         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
2332         if (ret != 0)
2333                 goto fail;
2334
2335         if (objectid > root->highest_inode)
2336                 root->highest_inode = objectid;
2337
2338         inode->i_uid = current->fsuid;
2339         inode->i_gid = current->fsgid;
2340         inode->i_mode = mode;
2341         inode->i_ino = objectid;
2342         inode->i_blocks = 0;
2343         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
2344         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
2345                                   struct btrfs_inode_item);
2346         fill_inode_item(trans, path->nodes[0], inode_item, inode);
2347
2348         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
2349                              struct btrfs_inode_ref);
2350         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
2351         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
2352         ptr = (unsigned long)(ref + 1);
2353         write_extent_buffer(path->nodes[0], name, ptr, name_len);
2354
2355         btrfs_mark_buffer_dirty(path->nodes[0]);
2356         btrfs_free_path(path);
2357
2358         location = &BTRFS_I(inode)->location;
2359         location->objectid = objectid;
2360         location->offset = 0;
2361         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
2362
2363         insert_inode_hash(inode);
2364         return inode;
2365 fail:
2366         if (dir)
2367                 BTRFS_I(dir)->index_cnt--;
2368         btrfs_free_path(path);
2369         return ERR_PTR(ret);
2370 }
2371
2372 static inline u8 btrfs_inode_type(struct inode *inode)
2373 {
2374         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
2375 }
2376
2377 int btrfs_add_link(struct btrfs_trans_handle *trans,
2378                    struct inode *parent_inode, struct inode *inode,
2379                    const char *name, int name_len, int add_backref, u64 index)
2380 {
2381         int ret;
2382         struct btrfs_key key;
2383         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
2384
2385         key.objectid = inode->i_ino;
2386         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
2387         key.offset = 0;
2388
2389         ret = btrfs_insert_dir_item(trans, root, name, name_len,
2390                                     parent_inode->i_ino,
2391                                     &key, btrfs_inode_type(inode),
2392                                     index);
2393         if (ret == 0) {
2394                 if (add_backref) {
2395                         ret = btrfs_insert_inode_ref(trans, root,
2396                                                      name, name_len,
2397                                                      inode->i_ino,
2398                                                      parent_inode->i_ino,
2399                                                      index);
2400                 }
2401                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
2402                                    name_len * 2);
2403                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
2404                 ret = btrfs_update_inode(trans, root, parent_inode);
2405         }
2406         return ret;
2407 }
2408
2409 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
2410                             struct dentry *dentry, struct inode *inode,
2411                             int backref, u64 index)
2412 {
2413         int err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2414                                  inode, dentry->d_name.name,
2415                                  dentry->d_name.len, backref, index);
2416         if (!err) {
2417                 d_instantiate(dentry, inode);
2418                 return 0;
2419         }
2420         if (err > 0)
2421                 err = -EEXIST;
2422         return err;
2423 }
2424
2425 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
2426                         int mode, dev_t rdev)
2427 {
2428         struct btrfs_trans_handle *trans;
2429         struct btrfs_root *root = BTRFS_I(dir)->root;
2430         struct inode *inode = NULL;
2431         int err;
2432         int drop_inode = 0;
2433         u64 objectid;
2434         unsigned long nr = 0;
2435         u64 index = 0;
2436
2437         if (!new_valid_dev(rdev))
2438                 return -EINVAL;
2439
2440         err = btrfs_check_free_space(root, 1, 0);
2441         if (err)
2442                 goto fail;
2443
2444         trans = btrfs_start_transaction(root, 1);
2445         btrfs_set_trans_block_group(trans, dir);
2446
2447         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2448         if (err) {
2449                 err = -ENOSPC;
2450                 goto out_unlock;
2451         }
2452
2453         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2454                                 dentry->d_name.len,
2455                                 dentry->d_parent->d_inode->i_ino, objectid,
2456                                 BTRFS_I(dir)->block_group, mode, &index);
2457         err = PTR_ERR(inode);
2458         if (IS_ERR(inode))
2459                 goto out_unlock;
2460
2461         err = btrfs_init_acl(inode, dir);
2462         if (err) {
2463                 drop_inode = 1;
2464                 goto out_unlock;
2465         }
2466
2467         btrfs_set_trans_block_group(trans, inode);
2468         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2469         if (err)
2470                 drop_inode = 1;
2471         else {
2472                 inode->i_op = &btrfs_special_inode_operations;
2473                 init_special_inode(inode, inode->i_mode, rdev);
2474                 btrfs_update_inode(trans, root, inode);
2475         }
2476         dir->i_sb->s_dirt = 1;
2477         btrfs_update_inode_block_group(trans, inode);
2478         btrfs_update_inode_block_group(trans, dir);
2479 out_unlock:
2480         nr = trans->blocks_used;
2481         btrfs_end_transaction_throttle(trans, root);
2482 fail:
2483         if (drop_inode) {
2484                 inode_dec_link_count(inode);
2485                 iput(inode);
2486         }
2487         btrfs_btree_balance_dirty(root, nr);
2488         return err;
2489 }
2490
2491 static int btrfs_create(struct inode *dir, struct dentry *dentry,
2492                         int mode, struct nameidata *nd)
2493 {
2494         struct btrfs_trans_handle *trans;
2495         struct btrfs_root *root = BTRFS_I(dir)->root;
2496         struct inode *inode = NULL;
2497         int err;
2498         int drop_inode = 0;
2499         unsigned long nr = 0;
2500         u64 objectid;
2501         u64 index = 0;
2502
2503         err = btrfs_check_free_space(root, 1, 0);
2504         if (err)
2505                 goto fail;
2506         trans = btrfs_start_transaction(root, 1);
2507         btrfs_set_trans_block_group(trans, dir);
2508
2509         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2510         if (err) {
2511                 err = -ENOSPC;
2512                 goto out_unlock;
2513         }
2514
2515         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2516                                 dentry->d_name.len,
2517                                 dentry->d_parent->d_inode->i_ino,
2518                                 objectid, BTRFS_I(dir)->block_group, mode,
2519                                 &index);
2520         err = PTR_ERR(inode);
2521         if (IS_ERR(inode))
2522                 goto out_unlock;
2523
2524         err = btrfs_init_acl(inode, dir);
2525         if (err) {
2526                 drop_inode = 1;
2527                 goto out_unlock;
2528         }
2529
2530         btrfs_set_trans_block_group(trans, inode);
2531         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
2532         if (err)
2533                 drop_inode = 1;
2534         else {
2535                 inode->i_mapping->a_ops = &btrfs_aops;
2536                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2537                 inode->i_fop = &btrfs_file_operations;
2538                 inode->i_op = &btrfs_file_inode_operations;
2539                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2540         }
2541         dir->i_sb->s_dirt = 1;
2542         btrfs_update_inode_block_group(trans, inode);
2543         btrfs_update_inode_block_group(trans, dir);
2544 out_unlock:
2545         nr = trans->blocks_used;
2546         btrfs_end_transaction_throttle(trans, root);
2547 fail:
2548         if (drop_inode) {
2549                 inode_dec_link_count(inode);
2550                 iput(inode);
2551         }
2552         btrfs_btree_balance_dirty(root, nr);
2553         return err;
2554 }
2555
2556 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
2557                       struct dentry *dentry)
2558 {
2559         struct btrfs_trans_handle *trans;
2560         struct btrfs_root *root = BTRFS_I(dir)->root;
2561         struct inode *inode = old_dentry->d_inode;
2562         u64 index;
2563         unsigned long nr = 0;
2564         int err;
2565         int drop_inode = 0;
2566
2567         if (inode->i_nlink == 0)
2568                 return -ENOENT;
2569
2570         btrfs_inc_nlink(inode);
2571         err = btrfs_check_free_space(root, 1, 0);
2572         if (err)
2573                 goto fail;
2574         err = btrfs_set_inode_index(dir, inode, &index);
2575         if (err)
2576                 goto fail;
2577
2578         trans = btrfs_start_transaction(root, 1);
2579
2580         btrfs_set_trans_block_group(trans, dir);
2581         atomic_inc(&inode->i_count);
2582
2583         err = btrfs_add_nondir(trans, dentry, inode, 1, index);
2584
2585         if (err)
2586                 drop_inode = 1;
2587
2588         dir->i_sb->s_dirt = 1;
2589         btrfs_update_inode_block_group(trans, dir);
2590         err = btrfs_update_inode(trans, root, inode);
2591
2592         if (err)
2593                 drop_inode = 1;
2594
2595         nr = trans->blocks_used;
2596         btrfs_end_transaction_throttle(trans, root);
2597 fail:
2598         if (drop_inode) {
2599                 inode_dec_link_count(inode);
2600                 iput(inode);
2601         }
2602         btrfs_btree_balance_dirty(root, nr);
2603         return err;
2604 }
2605
2606 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2607 {
2608         struct inode *inode = NULL;
2609         struct btrfs_trans_handle *trans;
2610         struct btrfs_root *root = BTRFS_I(dir)->root;
2611         int err = 0;
2612         int drop_on_err = 0;
2613         u64 objectid = 0;
2614         u64 index = 0;
2615         unsigned long nr = 1;
2616
2617         err = btrfs_check_free_space(root, 1, 0);
2618         if (err)
2619                 goto out_unlock;
2620
2621         trans = btrfs_start_transaction(root, 1);
2622         btrfs_set_trans_block_group(trans, dir);
2623
2624         if (IS_ERR(trans)) {
2625                 err = PTR_ERR(trans);
2626                 goto out_unlock;
2627         }
2628
2629         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2630         if (err) {
2631                 err = -ENOSPC;
2632                 goto out_unlock;
2633         }
2634
2635         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
2636                                 dentry->d_name.len,
2637                                 dentry->d_parent->d_inode->i_ino, objectid,
2638                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
2639                                 &index);
2640         if (IS_ERR(inode)) {
2641                 err = PTR_ERR(inode);
2642                 goto out_fail;
2643         }
2644
2645         drop_on_err = 1;
2646
2647         err = btrfs_init_acl(inode, dir);
2648         if (err)
2649                 goto out_fail;
2650
2651         inode->i_op = &btrfs_dir_inode_operations;
2652         inode->i_fop = &btrfs_dir_file_operations;
2653         btrfs_set_trans_block_group(trans, inode);
2654
2655         btrfs_i_size_write(inode, 0);
2656         err = btrfs_update_inode(trans, root, inode);
2657         if (err)
2658                 goto out_fail;
2659
2660         err = btrfs_add_link(trans, dentry->d_parent->d_inode,
2661                                  inode, dentry->d_name.name,
2662                                  dentry->d_name.len, 0, index);
2663         if (err)
2664                 goto out_fail;
2665
2666         d_instantiate(dentry, inode);
2667         drop_on_err = 0;
2668         dir->i_sb->s_dirt = 1;
2669         btrfs_update_inode_block_group(trans, inode);
2670         btrfs_update_inode_block_group(trans, dir);
2671
2672 out_fail:
2673         nr = trans->blocks_used;
2674         btrfs_end_transaction_throttle(trans, root);
2675
2676 out_unlock:
2677         if (drop_on_err)
2678                 iput(inode);
2679         btrfs_btree_balance_dirty(root, nr);
2680         return err;
2681 }
2682
2683 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2684                                 struct extent_map *existing,
2685                                 struct extent_map *em,
2686                                 u64 map_start, u64 map_len)
2687 {
2688         u64 start_diff;
2689
2690         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
2691         start_diff = map_start - em->start;
2692         em->start = map_start;
2693         em->len = map_len;
2694         if (em->block_start < EXTENT_MAP_LAST_BYTE)
2695                 em->block_start += start_diff;
2696         return add_extent_mapping(em_tree, em);
2697 }
2698
2699 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2700                                     size_t pg_offset, u64 start, u64 len,
2701                                     int create)
2702 {
2703         int ret;
2704         int err = 0;
2705         u64 bytenr;
2706         u64 extent_start = 0;
2707         u64 extent_end = 0;
2708         u64 objectid = inode->i_ino;
2709         u32 found_type;
2710         struct btrfs_path *path = NULL;
2711         struct btrfs_root *root = BTRFS_I(inode)->root;
2712         struct btrfs_file_extent_item *item;
2713         struct extent_buffer *leaf;
2714         struct btrfs_key found_key;
2715         struct extent_map *em = NULL;
2716         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2717         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2718         struct btrfs_trans_handle *trans = NULL;
2719
2720 again:
2721         spin_lock(&em_tree->lock);
2722         em = lookup_extent_mapping(em_tree, start, len);
2723         if (em)
2724                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2725         spin_unlock(&em_tree->lock);
2726
2727         if (em) {
2728                 if (em->start > start || em->start + em->len <= start)
2729                         free_extent_map(em);
2730                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2731                         free_extent_map(em);
2732                 else
2733                         goto out;
2734         }
2735         em = alloc_extent_map(GFP_NOFS);
2736         if (!em) {
2737                 err = -ENOMEM;
2738                 goto out;
2739         }
2740         em->bdev = root->fs_info->fs_devices->latest_bdev;
2741         em->start = EXTENT_MAP_HOLE;
2742         em->len = (u64)-1;
2743
2744         if (!path) {
2745                 path = btrfs_alloc_path();
2746                 BUG_ON(!path);
2747         }
2748
2749         ret = btrfs_lookup_file_extent(trans, root, path,
2750                                        objectid, start, trans != NULL);
2751         if (ret < 0) {
2752                 err = ret;
2753                 goto out;
2754         }
2755
2756         if (ret != 0) {
2757                 if (path->slots[0] == 0)
2758                         goto not_found;
2759                 path->slots[0]--;
2760         }
2761
2762         leaf = path->nodes[0];
2763         item = btrfs_item_ptr(leaf, path->slots[0],
2764                               struct btrfs_file_extent_item);
2765         /* are we inside the extent that was found? */
2766         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2767         found_type = btrfs_key_type(&found_key);
2768         if (found_key.objectid != objectid ||
2769             found_type != BTRFS_EXTENT_DATA_KEY) {
2770                 goto not_found;
2771         }
2772
2773         found_type = btrfs_file_extent_type(leaf, item);
2774         extent_start = found_key.offset;
2775         if (found_type == BTRFS_FILE_EXTENT_REG) {
2776                 extent_end = extent_start +
2777                        btrfs_file_extent_num_bytes(leaf, item);
2778                 err = 0;
2779                 if (start < extent_start || start >= extent_end) {
2780                         em->start = start;
2781                         if (start < extent_start) {
2782                                 if (start + len <= extent_start)
2783                                         goto not_found;
2784                                 em->len = extent_end - extent_start;
2785                         } else {
2786                                 em->len = len;
2787                         }
2788                         goto not_found_em;
2789                 }
2790                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2791                 if (bytenr == 0) {
2792                         em->start = extent_start;
2793                         em->len = extent_end - extent_start;
2794                         em->block_start = EXTENT_MAP_HOLE;
2795                         goto insert;
2796                 }
2797                 bytenr += btrfs_file_extent_offset(leaf, item);
2798                 em->block_start = bytenr;
2799                 em->start = extent_start;
2800                 em->len = extent_end - extent_start;
2801                 goto insert;
2802         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2803                 u64 page_start;
2804                 unsigned long ptr;
2805                 char *map;
2806                 size_t size;
2807                 size_t extent_offset;
2808                 size_t copy_size;
2809
2810                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2811                                                     path->slots[0]));
2812                 extent_end = (extent_start + size + root->sectorsize - 1) &
2813                         ~((u64)root->sectorsize - 1);
2814                 if (start < extent_start || start >= extent_end) {
2815                         em->start = start;
2816                         if (start < extent_start) {
2817                                 if (start + len <= extent_start)
2818                                         goto not_found;
2819                                 em->len = extent_end - extent_start;
2820                         } else {
2821                                 em->len = len;
2822                         }
2823                         goto not_found_em;
2824                 }
2825                 em->block_start = EXTENT_MAP_INLINE;
2826
2827                 if (!page) {
2828                         em->start = extent_start;
2829                         em->len = size;
2830                         goto out;
2831                 }
2832
2833                 page_start = page_offset(page) + pg_offset;
2834                 extent_offset = page_start - extent_start;
2835                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2836                                 size - extent_offset);
2837                 em->start = extent_start + extent_offset;
2838                 em->len = (copy_size + root->sectorsize - 1) &
2839                         ~((u64)root->sectorsize - 1);
2840                 map = kmap(page);
2841                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2842                 if (create == 0 && !PageUptodate(page)) {
2843                         read_extent_buffer(leaf, map + pg_offset, ptr,
2844                                            copy_size);
2845                         flush_dcache_page(page);
2846                 } else if (create && PageUptodate(page)) {
2847                         if (!trans) {
2848                                 kunmap(page);
2849                                 free_extent_map(em);
2850                                 em = NULL;
2851                                 btrfs_release_path(root, path);
2852                                 trans = btrfs_join_transaction(root, 1);
2853                                 goto again;
2854                         }
2855                         write_extent_buffer(leaf, map + pg_offset, ptr,
2856                                             copy_size);
2857                         btrfs_mark_buffer_dirty(leaf);
2858                 }
2859                 kunmap(page);
2860                 set_extent_uptodate(io_tree, em->start,
2861                                     extent_map_end(em) - 1, GFP_NOFS);
2862                 goto insert;
2863         } else {
2864                 printk("unkknown found_type %d\n", found_type);
2865                 WARN_ON(1);
2866         }
2867 not_found:
2868         em->start = start;
2869         em->len = len;
2870 not_found_em:
2871         em->block_start = EXTENT_MAP_HOLE;
2872 insert:
2873         btrfs_release_path(root, path);
2874         if (em->start > start || extent_map_end(em) <= start) {
2875                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2876                 err = -EIO;
2877                 goto out;
2878         }
2879
2880         err = 0;
2881         spin_lock(&em_tree->lock);
2882         ret = add_extent_mapping(em_tree, em);
2883         /* it is possible that someone inserted the extent into the tree
2884          * while we had the lock dropped.  It is also possible that
2885          * an overlapping map exists in the tree
2886          */
2887         if (ret == -EEXIST) {
2888                 struct extent_map *existing;
2889
2890                 ret = 0;
2891
2892                 existing = lookup_extent_mapping(em_tree, start, len);
2893                 if (existing && (existing->start > start ||
2894                     existing->start + existing->len <= start)) {
2895                         free_extent_map(existing);
2896                         existing = NULL;
2897                 }
2898                 if (!existing) {
2899                         existing = lookup_extent_mapping(em_tree, em->start,
2900                                                          em->len);
2901                         if (existing) {
2902                                 err = merge_extent_mapping(em_tree, existing,
2903                                                            em, start,
2904                                                            root->sectorsize);
2905                                 free_extent_map(existing);
2906                                 if (err) {
2907                                         free_extent_map(em);
2908                                         em = NULL;
2909                                 }
2910                         } else {
2911                                 err = -EIO;
2912                                 printk("failing to insert %Lu %Lu\n",
2913                                        start, len);
2914                                 free_extent_map(em);
2915                                 em = NULL;
2916                         }
2917                 } else {
2918                         free_extent_map(em);
2919                         em = existing;
2920                         err = 0;
2921                 }
2922         }
2923         spin_unlock(&em_tree->lock);
2924 out:
2925         if (path)
2926                 btrfs_free_path(path);
2927         if (trans) {
2928                 ret = btrfs_end_transaction(trans, root);
2929                 if (!err) {
2930                         err = ret;
2931                 }
2932         }
2933         if (err) {
2934                 free_extent_map(em);
2935                 WARN_ON(1);
2936                 return ERR_PTR(err);
2937         }
2938         return em;
2939 }
2940
2941 #if 0 /* waiting for O_DIRECT reads */
2942 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2943                         struct buffer_head *bh_result, int create)
2944 {
2945         struct extent_map *em;
2946         u64 start = (u64)iblock << inode->i_blkbits;
2947         struct btrfs_multi_bio *multi = NULL;
2948         struct btrfs_root *root = BTRFS_I(inode)->root;
2949         u64 len;
2950         u64 logical;
2951         u64 map_length;
2952         int ret = 0;
2953
2954         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2955
2956         if (!em || IS_ERR(em))
2957                 goto out;
2958
2959         if (em->start > start || em->start + em->len <= start) {
2960             goto out;
2961         }
2962
2963         if (em->block_start == EXTENT_MAP_INLINE) {
2964                 ret = -EINVAL;
2965                 goto out;
2966         }
2967
2968         len = em->start + em->len - start;
2969         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2970
2971         if (em->block_start == EXTENT_MAP_HOLE ||
2972             em->block_start == EXTENT_MAP_DELALLOC) {
2973                 bh_result->b_size = len;
2974                 goto out;
2975         }
2976
2977         logical = start - em->start;
2978         logical = em->block_start + logical;
2979
2980         map_length = len;
2981         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2982                               logical, &map_length, &multi, 0);
2983         BUG_ON(ret);
2984         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2985         bh_result->b_size = min(map_length, len);
2986
2987         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2988         set_buffer_mapped(bh_result);
2989         kfree(multi);
2990 out:
2991         free_extent_map(em);
2992         return ret;
2993 }
2994 #endif
2995
2996 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2997                         const struct iovec *iov, loff_t offset,
2998                         unsigned long nr_segs)
2999 {
3000         return -EINVAL;
3001 #if 0
3002         struct file *file = iocb->ki_filp;
3003         struct inode *inode = file->f_mapping->host;
3004
3005         if (rw == WRITE)
3006                 return -EINVAL;
3007
3008         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
3009                                   offset, nr_segs, btrfs_get_block, NULL);
3010 #endif
3011 }
3012
3013 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
3014 {
3015         return extent_bmap(mapping, iblock, btrfs_get_extent);
3016 }
3017
3018 int btrfs_readpage(struct file *file, struct page *page)
3019 {
3020         struct extent_io_tree *tree;
3021         tree = &BTRFS_I(page->mapping->host)->io_tree;
3022         return extent_read_full_page(tree, page, btrfs_get_extent);
3023 }
3024
3025 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
3026 {
3027         struct extent_io_tree *tree;
3028
3029
3030         if (current->flags & PF_MEMALLOC) {
3031                 redirty_page_for_writepage(wbc, page);
3032                 unlock_page(page);
3033                 return 0;
3034         }
3035         tree = &BTRFS_I(page->mapping->host)->io_tree;
3036         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
3037 }
3038
3039 int btrfs_writepages(struct address_space *mapping,
3040                      struct writeback_control *wbc)
3041 {
3042         struct extent_io_tree *tree;
3043         tree = &BTRFS_I(mapping->host)->io_tree;
3044         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
3045 }
3046
3047 static int
3048 btrfs_readpages(struct file *file, struct address_space *mapping,
3049                 struct list_head *pages, unsigned nr_pages)
3050 {
3051         struct extent_io_tree *tree;
3052         tree = &BTRFS_I(mapping->host)->io_tree;
3053         return extent_readpages(tree, mapping, pages, nr_pages,
3054                                 btrfs_get_extent);
3055 }
3056 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3057 {
3058         struct extent_io_tree *tree;
3059         struct extent_map_tree *map;
3060         int ret;
3061
3062         tree = &BTRFS_I(page->mapping->host)->io_tree;
3063         map = &BTRFS_I(page->mapping->host)->extent_tree;
3064         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
3065         if (ret == 1) {
3066                 ClearPagePrivate(page);
3067                 set_page_private(page, 0);
3068                 page_cache_release(page);
3069         }
3070         return ret;
3071 }
3072
3073 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
3074 {
3075         return __btrfs_releasepage(page, gfp_flags);
3076 }
3077
3078 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
3079 {
3080         struct extent_io_tree *tree;
3081         struct btrfs_ordered_extent *ordered;
3082         u64 page_start = page_offset(page);
3083         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
3084
3085         wait_on_page_writeback(page);
3086         tree = &BTRFS_I(page->mapping->host)->io_tree;
3087         if (offset) {
3088                 btrfs_releasepage(page, GFP_NOFS);
3089                 return;
3090         }
3091
3092         lock_extent(tree, page_start, page_end, GFP_NOFS);
3093         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
3094                                            page_offset(page));
3095         if (ordered) {
3096                 /*
3097                  * IO on this page will never be started, so we need
3098                  * to account for any ordered extents now
3099                  */
3100                 clear_extent_bit(tree, page_start, page_end,
3101                                  EXTENT_DIRTY | EXTENT_DELALLOC |
3102                                  EXTENT_LOCKED, 1, 0, GFP_NOFS);
3103                 btrfs_finish_ordered_io(page->mapping->host,
3104                                         page_start, page_end);
3105                 btrfs_put_ordered_extent(ordered);
3106                 lock_extent(tree, page_start, page_end, GFP_NOFS);
3107         }
3108         clear_extent_bit(tree, page_start, page_end,
3109                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
3110                  EXTENT_ORDERED,
3111                  1, 1, GFP_NOFS);
3112         __btrfs_releasepage(page, GFP_NOFS);
3113
3114         ClearPageChecked(page);
3115         if (PagePrivate(page)) {
3116                 ClearPagePrivate(page);
3117                 set_page_private(page, 0);
3118                 page_cache_release(page);
3119         }
3120 }
3121
3122 /*
3123  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
3124  * called from a page fault handler when a page is first dirtied. Hence we must
3125  * be careful to check for EOF conditions here. We set the page up correctly
3126  * for a written page which means we get ENOSPC checking when writing into
3127  * holes and correct delalloc and unwritten extent mapping on filesystems that
3128  * support these features.
3129  *
3130  * We are not allowed to take the i_mutex here so we have to play games to
3131  * protect against truncate races as the page could now be beyond EOF.  Because
3132  * vmtruncate() writes the inode size before removing pages, once we have the
3133  * page lock we can determine safely if the page is beyond EOF. If it is not
3134  * beyond EOF, then the page is guaranteed safe against truncation until we
3135  * unlock the page.
3136  */
3137 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
3138 {
3139         struct inode *inode = fdentry(vma->vm_file)->d_inode;
3140         struct btrfs_root *root = BTRFS_I(inode)->root;
3141         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3142         struct btrfs_ordered_extent *ordered;
3143         char *kaddr;
3144         unsigned long zero_start;
3145         loff_t size;
3146         int ret;
3147         u64 page_start;
3148         u64 page_end;
3149
3150         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
3151         if (ret)
3152                 goto out;
3153
3154         ret = -EINVAL;
3155 again:
3156         lock_page(page);
3157         size = i_size_read(inode);
3158         page_start = page_offset(page);
3159         page_end = page_start + PAGE_CACHE_SIZE - 1;
3160
3161         if ((page->mapping != inode->i_mapping) ||
3162             (page_start >= size)) {
3163                 /* page got truncated out from underneath us */
3164                 goto out_unlock;
3165         }
3166         wait_on_page_writeback(page);
3167
3168         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
3169         set_page_extent_mapped(page);
3170
3171         /*
3172          * we can't set the delalloc bits if there are pending ordered
3173          * extents.  Drop our locks and wait for them to finish
3174          */
3175         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3176         if (ordered) {
3177                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3178                 unlock_page(page);
3179                 btrfs_start_ordered_extent(inode, ordered, 1);
3180                 btrfs_put_ordered_extent(ordered);
3181                 goto again;
3182         }
3183
3184         btrfs_set_extent_delalloc(inode, page_start, page_end);
3185         ret = 0;
3186
3187         /* page is wholly or partially inside EOF */
3188         if (page_start + PAGE_CACHE_SIZE > size)
3189                 zero_start = size & ~PAGE_CACHE_MASK;
3190         else
3191                 zero_start = PAGE_CACHE_SIZE;
3192
3193         if (zero_start != PAGE_CACHE_SIZE) {
3194                 kaddr = kmap(page);
3195                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
3196                 flush_dcache_page(page);
3197                 kunmap(page);
3198         }
3199         ClearPageChecked(page);
3200         set_page_dirty(page);
3201         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
3202
3203 out_unlock:
3204         unlock_page(page);
3205 out:
3206         return ret;
3207 }
3208
3209 static void btrfs_truncate(struct inode *inode)
3210 {
3211         struct btrfs_root *root = BTRFS_I(inode)->root;
3212         int ret;
3213         struct btrfs_trans_handle *trans;
3214         unsigned long nr;
3215         u64 mask = root->sectorsize - 1;
3216
3217         if (!S_ISREG(inode->i_mode))
3218                 return;
3219         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3220                 return;
3221
3222         btrfs_truncate_page(inode->i_mapping, inode->i_size);
3223         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
3224
3225         trans = btrfs_start_transaction(root, 1);
3226         btrfs_set_trans_block_group(trans, inode);
3227         btrfs_i_size_write(inode, inode->i_size);
3228
3229         ret = btrfs_orphan_add(trans, inode);
3230         if (ret)
3231                 goto out;
3232         /* FIXME, add redo link to tree so we don't leak on crash */
3233         ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
3234                                       BTRFS_EXTENT_DATA_KEY);
3235         btrfs_update_inode(trans, root, inode);
3236
3237         ret = btrfs_orphan_del(trans, inode);
3238         BUG_ON(ret);
3239
3240 out:
3241         nr = trans->blocks_used;
3242         ret = btrfs_end_transaction_throttle(trans, root);
3243         BUG_ON(ret);
3244         btrfs_btree_balance_dirty(root, nr);
3245 }
3246
3247 /*
3248  * Invalidate a single dcache entry at the root of the filesystem.
3249  * Needed after creation of snapshot or subvolume.
3250  */
3251 void btrfs_invalidate_dcache_root(struct btrfs_root *root, char *name,
3252                                   int namelen)
3253 {
3254         struct dentry *alias, *entry;
3255         struct qstr qstr;
3256
3257         alias = d_find_alias(root->fs_info->sb->s_root->d_inode);
3258         if (alias) {
3259                 qstr.name = name;
3260                 qstr.len = namelen;
3261                 /* change me if btrfs ever gets a d_hash operation */
3262                 qstr.hash = full_name_hash(qstr.name, qstr.len);
3263                 entry = d_lookup(alias, &qstr);
3264                 dput(alias);
3265                 if (entry) {
3266                         d_invalidate(entry);
3267                         dput(entry);
3268                 }
3269         }
3270 }
3271
3272 int btrfs_create_subvol_root(struct btrfs_root *new_root,
3273                 struct btrfs_trans_handle *trans, u64 new_dirid,
3274                 struct btrfs_block_group_cache *block_group)
3275 {
3276         struct inode *inode;
3277         u64 index = 0;
3278
3279         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
3280                                 new_dirid, block_group, S_IFDIR | 0700, &index);
3281         if (IS_ERR(inode))
3282                 return PTR_ERR(inode);
3283         inode->i_op = &btrfs_dir_inode_operations;
3284         inode->i_fop = &btrfs_dir_file_operations;
3285         new_root->inode = inode;
3286
3287         inode->i_nlink = 1;
3288         btrfs_i_size_write(inode, 0);
3289
3290         return btrfs_update_inode(trans, new_root, inode);
3291 }
3292
3293 unsigned long btrfs_force_ra(struct address_space *mapping,
3294                               struct file_ra_state *ra, struct file *file,
3295                               pgoff_t offset, pgoff_t last_index)
3296 {
3297         pgoff_t req_size = last_index - offset + 1;
3298
3299 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3300         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
3301         return offset;
3302 #else
3303         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
3304         return offset + req_size;
3305 #endif
3306 }
3307
3308 struct inode *btrfs_alloc_inode(struct super_block *sb)
3309 {
3310         struct btrfs_inode *ei;
3311
3312         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3313         if (!ei)
3314                 return NULL;
3315         ei->last_trans = 0;
3316         ei->logged_trans = 0;
3317         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
3318         ei->i_acl = BTRFS_ACL_NOT_CACHED;
3319         ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
3320         INIT_LIST_HEAD(&ei->i_orphan);
3321         return &ei->vfs_inode;
3322 }
3323
3324 void btrfs_destroy_inode(struct inode *inode)
3325 {
3326         struct btrfs_ordered_extent *ordered;
3327         WARN_ON(!list_empty(&inode->i_dentry));
3328         WARN_ON(inode->i_data.nrpages);
3329
3330         if (BTRFS_I(inode)->i_acl &&
3331             BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
3332                 posix_acl_release(BTRFS_I(inode)->i_acl);
3333         if (BTRFS_I(inode)->i_default_acl &&
3334             BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
3335                 posix_acl_release(BTRFS_I(inode)->i_default_acl);
3336
3337         spin_lock(&BTRFS_I(inode)->root->list_lock);
3338         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
3339                 printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
3340                        " list\n", inode->i_ino);
3341                 dump_stack();
3342         }
3343         spin_unlock(&BTRFS_I(inode)->root->list_lock);
3344
3345         while(1) {
3346                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
3347                 if (!ordered)
3348                         break;
3349                 else {
3350                         printk("found ordered extent %Lu %Lu\n",
3351                                ordered->file_offset, ordered->len);
3352                         btrfs_remove_ordered_extent(inode, ordered);
3353                         btrfs_put_ordered_extent(ordered);
3354                         btrfs_put_ordered_extent(ordered);
3355                 }
3356         }
3357         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3358         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3359 }
3360
3361 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3362 static void init_once(void *foo)
3363 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3364 static void init_once(struct kmem_cache * cachep, void *foo)
3365 #else
3366 static void init_once(void * foo, struct kmem_cache * cachep,
3367                       unsigned long flags)
3368 #endif
3369 {
3370         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3371
3372         inode_init_once(&ei->vfs_inode);
3373 }
3374
3375 void btrfs_destroy_cachep(void)
3376 {
3377         if (btrfs_inode_cachep)
3378                 kmem_cache_destroy(btrfs_inode_cachep);
3379         if (btrfs_trans_handle_cachep)
3380                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3381         if (btrfs_transaction_cachep)
3382                 kmem_cache_destroy(btrfs_transaction_cachep);
3383         if (btrfs_bit_radix_cachep)
3384                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3385         if (btrfs_path_cachep)
3386                 kmem_cache_destroy(btrfs_path_cachep);
3387 }
3388
3389 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3390                                        unsigned long extra_flags,
3391 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3392                                        void (*ctor)(void *)
3393 #elif LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3394                                        void (*ctor)(struct kmem_cache *, void *)
3395 #else
3396                                        void (*ctor)(void *, struct kmem_cache *,
3397                                                     unsigned long)
3398 #endif
3399                                      )
3400 {
3401         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3402                                  SLAB_MEM_SPREAD | extra_flags), ctor
3403 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3404                                  ,NULL
3405 #endif
3406                                 );
3407 }
3408
3409 int btrfs_init_cachep(void)
3410 {
3411         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3412                                           sizeof(struct btrfs_inode),
3413                                           0, init_once);
3414         if (!btrfs_inode_cachep)
3415                 goto fail;
3416         btrfs_trans_handle_cachep =
3417                         btrfs_cache_create("btrfs_trans_handle_cache",
3418                                            sizeof(struct btrfs_trans_handle),
3419                                            0, NULL);
3420         if (!btrfs_trans_handle_cachep)
3421                 goto fail;
3422         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3423                                              sizeof(struct btrfs_transaction),
3424                                              0, NULL);
3425         if (!btrfs_transaction_cachep)
3426                 goto fail;
3427         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3428                                          sizeof(struct btrfs_path),
3429                                          0, NULL);
3430         if (!btrfs_path_cachep)
3431                 goto fail;
3432         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3433                                               SLAB_DESTROY_BY_RCU, NULL);
3434         if (!btrfs_bit_radix_cachep)
3435                 goto fail;
3436         return 0;
3437 fail:
3438         btrfs_destroy_cachep();
3439         return -ENOMEM;
3440 }
3441
3442 static int btrfs_getattr(struct vfsmount *mnt,
3443                          struct dentry *dentry, struct kstat *stat)
3444 {
3445         struct inode *inode = dentry->d_inode;
3446         generic_fillattr(inode, stat);
3447         stat->blksize = PAGE_CACHE_SIZE;
3448         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3449         return 0;
3450 }
3451
3452 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3453                            struct inode * new_dir,struct dentry *new_dentry)
3454 {
3455         struct btrfs_trans_handle *trans;
3456         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3457         struct inode *new_inode = new_dentry->d_inode;
3458         struct inode *old_inode = old_dentry->d_inode;
3459         struct timespec ctime = CURRENT_TIME;
3460         u64 index = 0;
3461         int ret;
3462
3463         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3464             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3465                 return -ENOTEMPTY;
3466         }
3467
3468         ret = btrfs_check_free_space(root, 1, 0);
3469         if (ret)
3470                 goto out_unlock;
3471
3472         trans = btrfs_start_transaction(root, 1);
3473
3474         btrfs_set_trans_block_group(trans, new_dir);
3475
3476         btrfs_inc_nlink(old_dentry->d_inode);
3477         old_dir->i_ctime = old_dir->i_mtime = ctime;
3478         new_dir->i_ctime = new_dir->i_mtime = ctime;
3479         old_inode->i_ctime = ctime;
3480
3481         ret = btrfs_unlink_inode(trans, root, old_dir, old_dentry->d_inode,
3482                                  old_dentry->d_name.name,
3483                                  old_dentry->d_name.len);
3484         if (ret)
3485                 goto out_fail;
3486
3487         if (new_inode) {
3488                 new_inode->i_ctime = CURRENT_TIME;
3489                 ret = btrfs_unlink_inode(trans, root, new_dir,
3490                                          new_dentry->d_inode,
3491                                          new_dentry->d_name.name,
3492                                          new_dentry->d_name.len);
3493                 if (ret)
3494                         goto out_fail;
3495                 if (new_inode->i_nlink == 0) {
3496                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
3497                         if (ret)
3498                                 goto out_fail;
3499                 }
3500
3501         }
3502         ret = btrfs_set_inode_index(new_dir, old_inode, &index);
3503         if (ret)
3504                 goto out_fail;
3505
3506         ret = btrfs_add_link(trans, new_dentry->d_parent->d_inode,
3507                              old_inode, new_dentry->d_name.name,
3508                              new_dentry->d_name.len, 1, index);
3509         if (ret)
3510                 goto out_fail;
3511
3512 out_fail:
3513         btrfs_end_transaction_throttle(trans, root);
3514 out_unlock:
3515         return ret;
3516 }
3517
3518 int btrfs_start_delalloc_inodes(struct btrfs_root *root)
3519 {
3520         struct list_head *head = &root->fs_info->delalloc_inodes;
3521         struct btrfs_inode *binode;
3522         unsigned long flags;
3523
3524         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3525         while(!list_empty(head)) {
3526                 binode = list_entry(head->next, struct btrfs_inode,
3527                                     delalloc_inodes);
3528                 atomic_inc(&binode->vfs_inode.i_count);
3529                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3530                 filemap_write_and_wait(binode->vfs_inode.i_mapping);
3531                 iput(&binode->vfs_inode);
3532                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
3533         }
3534         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
3535         return 0;
3536 }
3537
3538 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3539                          const char *symname)
3540 {
3541         struct btrfs_trans_handle *trans;
3542         struct btrfs_root *root = BTRFS_I(dir)->root;
3543         struct btrfs_path *path;
3544         struct btrfs_key key;
3545         struct inode *inode = NULL;
3546         int err;
3547         int drop_inode = 0;
3548         u64 objectid;
3549         u64 index = 0 ;
3550         int name_len;
3551         int datasize;
3552         unsigned long ptr;
3553         struct btrfs_file_extent_item *ei;
3554         struct extent_buffer *leaf;
3555         unsigned long nr = 0;
3556
3557         name_len = strlen(symname) + 1;
3558         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3559                 return -ENAMETOOLONG;
3560
3561         err = btrfs_check_free_space(root, 1, 0);
3562         if (err)
3563                 goto out_fail;
3564
3565         trans = btrfs_start_transaction(root, 1);
3566         btrfs_set_trans_block_group(trans, dir);
3567
3568         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3569         if (err) {
3570                 err = -ENOSPC;
3571                 goto out_unlock;
3572         }
3573
3574         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
3575                                 dentry->d_name.len,
3576                                 dentry->d_parent->d_inode->i_ino, objectid,
3577                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
3578                                 &index);
3579         err = PTR_ERR(inode);
3580         if (IS_ERR(inode))
3581                 goto out_unlock;
3582
3583         err = btrfs_init_acl(inode, dir);
3584         if (err) {
3585                 drop_inode = 1;
3586                 goto out_unlock;
3587         }
3588
3589         btrfs_set_trans_block_group(trans, inode);
3590         err = btrfs_add_nondir(trans, dentry, inode, 0, index);
3591         if (err)
3592                 drop_inode = 1;
3593         else {
3594                 inode->i_mapping->a_ops = &btrfs_aops;
3595                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3596                 inode->i_fop = &btrfs_file_operations;
3597                 inode->i_op = &btrfs_file_inode_operations;
3598                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3599         }
3600         dir->i_sb->s_dirt = 1;
3601         btrfs_update_inode_block_group(trans, inode);
3602         btrfs_update_inode_block_group(trans, dir);
3603         if (drop_inode)
3604                 goto out_unlock;
3605
3606         path = btrfs_alloc_path();
3607         BUG_ON(!path);
3608         key.objectid = inode->i_ino;
3609         key.offset = 0;
3610         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3611         datasize = btrfs_file_extent_calc_inline_size(name_len);
3612         err = btrfs_insert_empty_item(trans, root, path, &key,
3613                                       datasize);
3614         if (err) {
3615                 drop_inode = 1;
3616                 goto out_unlock;
3617         }
3618         leaf = path->nodes[0];
3619         ei = btrfs_item_ptr(leaf, path->slots[0],
3620                             struct btrfs_file_extent_item);
3621         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3622         btrfs_set_file_extent_type(leaf, ei,
3623                                    BTRFS_FILE_EXTENT_INLINE);
3624         ptr = btrfs_file_extent_inline_start(ei);
3625         write_extent_buffer(leaf, symname, ptr, name_len);
3626         btrfs_mark_buffer_dirty(leaf);
3627         btrfs_free_path(path);
3628
3629         inode->i_op = &btrfs_symlink_inode_operations;
3630         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3631         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3632         btrfs_i_size_write(inode, name_len - 1);
3633         err = btrfs_update_inode(trans, root, inode);
3634         if (err)
3635                 drop_inode = 1;
3636
3637 out_unlock:
3638         nr = trans->blocks_used;
3639         btrfs_end_transaction_throttle(trans, root);
3640 out_fail:
3641         if (drop_inode) {
3642                 inode_dec_link_count(inode);
3643                 iput(inode);
3644         }
3645         btrfs_btree_balance_dirty(root, nr);
3646         return err;
3647 }
3648
3649 static int btrfs_set_page_dirty(struct page *page)
3650 {
3651         return __set_page_dirty_nobuffers(page);
3652 }
3653
3654 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,26)
3655 static int btrfs_permission(struct inode *inode, int mask)
3656 #else
3657 static int btrfs_permission(struct inode *inode, int mask,
3658                             struct nameidata *nd)
3659 #endif
3660 {
3661         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3662                 return -EACCES;
3663         return generic_permission(inode, mask, btrfs_check_acl);
3664 }
3665
3666 static struct inode_operations btrfs_dir_inode_operations = {
3667         .lookup         = btrfs_lookup,
3668         .create         = btrfs_create,
3669         .unlink         = btrfs_unlink,
3670         .link           = btrfs_link,
3671         .mkdir          = btrfs_mkdir,
3672         .rmdir          = btrfs_rmdir,
3673         .rename         = btrfs_rename,
3674         .symlink        = btrfs_symlink,
3675         .setattr        = btrfs_setattr,
3676         .mknod          = btrfs_mknod,
3677         .setxattr       = btrfs_setxattr,
3678         .getxattr       = btrfs_getxattr,
3679         .listxattr      = btrfs_listxattr,
3680         .removexattr    = btrfs_removexattr,
3681         .permission     = btrfs_permission,
3682 };
3683 static struct inode_operations btrfs_dir_ro_inode_operations = {
3684         .lookup         = btrfs_lookup,
3685         .permission     = btrfs_permission,
3686 };
3687 static struct file_operations btrfs_dir_file_operations = {
3688         .llseek         = generic_file_llseek,
3689         .read           = generic_read_dir,
3690 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,28)
3691         .readdir        = btrfs_nfshack_readdir,
3692 #else /* NFSd readdir/lookup deadlock is fixed */
3693         .readdir        = btrfs_real_readdir,
3694 #endif
3695         .unlocked_ioctl = btrfs_ioctl,
3696 #ifdef CONFIG_COMPAT
3697         .compat_ioctl   = btrfs_ioctl,
3698 #endif
3699         .release        = btrfs_release_file,
3700         .fsync          = btrfs_sync_file,
3701 };
3702
3703 static struct extent_io_ops btrfs_extent_io_ops = {
3704         .fill_delalloc = run_delalloc_range,
3705         .submit_bio_hook = btrfs_submit_bio_hook,
3706         .merge_bio_hook = btrfs_merge_bio_hook,
3707         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3708         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
3709         .writepage_start_hook = btrfs_writepage_start_hook,
3710         .readpage_io_failed_hook = btrfs_io_failed_hook,
3711         .set_bit_hook = btrfs_set_bit_hook,
3712         .clear_bit_hook = btrfs_clear_bit_hook,
3713 };
3714
3715 static struct address_space_operations btrfs_aops = {
3716         .readpage       = btrfs_readpage,
3717         .writepage      = btrfs_writepage,
3718         .writepages     = btrfs_writepages,
3719         .readpages      = btrfs_readpages,
3720         .sync_page      = block_sync_page,
3721         .bmap           = btrfs_bmap,
3722         .direct_IO      = btrfs_direct_IO,
3723         .invalidatepage = btrfs_invalidatepage,
3724         .releasepage    = btrfs_releasepage,
3725         .set_page_dirty = btrfs_set_page_dirty,
3726 };
3727
3728 static struct address_space_operations btrfs_symlink_aops = {
3729         .readpage       = btrfs_readpage,
3730         .writepage      = btrfs_writepage,
3731         .invalidatepage = btrfs_invalidatepage,
3732         .releasepage    = btrfs_releasepage,
3733 };
3734
3735 static struct inode_operations btrfs_file_inode_operations = {
3736         .truncate       = btrfs_truncate,
3737         .getattr        = btrfs_getattr,
3738         .setattr        = btrfs_setattr,
3739         .setxattr       = btrfs_setxattr,
3740         .getxattr       = btrfs_getxattr,
3741         .listxattr      = btrfs_listxattr,
3742         .removexattr    = btrfs_removexattr,
3743         .permission     = btrfs_permission,
3744 };
3745 static struct inode_operations btrfs_special_inode_operations = {
3746         .getattr        = btrfs_getattr,
3747         .setattr        = btrfs_setattr,
3748         .permission     = btrfs_permission,
3749         .setxattr       = btrfs_setxattr,
3750         .getxattr       = btrfs_getxattr,
3751         .listxattr      = btrfs_listxattr,
3752         .removexattr    = btrfs_removexattr,
3753 };
3754 static struct inode_operations btrfs_symlink_inode_operations = {
3755         .readlink       = generic_readlink,
3756         .follow_link    = page_follow_link_light,
3757         .put_link       = page_put_link,
3758         .permission     = btrfs_permission,
3759 };