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