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/bio.h>
  20 #include <linux/buffer_head.h>
  21 #include <linux/fs.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/highmem.h>
  24 #include <linux/time.h>
  25 #include <linux/init.h>
  26 #include <linux/string.h>
  27 #include <linux/smp_lock.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/mpage.h>
  30 #include <linux/swap.h>
  31 #include <linux/writeback.h>
  32 #include <linux/statfs.h>
  33 #include <linux/compat.h>
  34 #include <linux/bit_spinlock.h>
  35 #include <linux/version.h>
  36 #include <linux/xattr.h>
  37 #include "ctree.h"
  38 #include "disk-io.h"
  39 #include "transaction.h"
  40 #include "btrfs_inode.h"
  41 #include "ioctl.h"
  42 #include "print-tree.h"
  43 #include "volumes.h"
  44
  45 struct btrfs_iget_args {
  46         u64 ino;
  47         struct btrfs_root *root;
  48 };
  49
  50 static struct inode_operations btrfs_dir_inode_operations;
  51 static struct inode_operations btrfs_symlink_inode_operations;
  52 static struct inode_operations btrfs_dir_ro_inode_operations;
  53 static struct inode_operations btrfs_special_inode_operations;
  54 static struct inode_operations btrfs_file_inode_operations;
  55 static struct address_space_operations btrfs_aops;
  56 static struct address_space_operations btrfs_symlink_aops;
  57 static struct file_operations btrfs_dir_file_operations;
  58 static struct extent_io_ops btrfs_extent_io_ops;
  59
  60 static struct kmem_cache *btrfs_inode_cachep;
  61 struct kmem_cache *btrfs_trans_handle_cachep;
  62 struct kmem_cache *btrfs_transaction_cachep;
  63 struct kmem_cache *btrfs_bit_radix_cachep;
  64 struct kmem_cache *btrfs_path_cachep;
  65
  66 #define S_SHIFT 12
  67 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  68         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  69         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  70         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  71         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  72         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  73         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  74         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  75 };
  76
  77 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  78                            int for_del)
  79 {
  80         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  81         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  82         u64 thresh;
  83         int ret = 0;
  84
  85         if (for_del)
  86                 thresh = total * 90;
  87         else
  88                 thresh = total * 85;
  89
  90         do_div(thresh, 100);
  91
  92         spin_lock(&root->fs_info->delalloc_lock);
  93         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  94                 ret = -ENOSPC;
  95         spin_unlock(&root->fs_info->delalloc_lock);
  96         return ret;
  97 }
  98
  99 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 100 {
 101         struct btrfs_root *root = BTRFS_I(inode)->root;
 102         struct btrfs_trans_handle *trans;
 103         u64 alloc_hint = 0;
 104         u64 num_bytes;
 105         u64 cur_alloc_size;
 106         u64 blocksize = root->sectorsize;
 107         u64 orig_start = start;
 108         u64 orig_num_bytes;
 109         struct btrfs_key ins;
 110         int ret;
 111
 112         trans = btrfs_start_transaction(root, 1);
 113         BUG_ON(!trans);
 114         btrfs_set_trans_block_group(trans, inode);
 115
 116         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 117         num_bytes = max(blocksize,  num_bytes);
 118         ret = btrfs_drop_extents(trans, root, inode,
 119                                  start, start + num_bytes, start, &alloc_hint);
 120         orig_num_bytes = num_bytes;
 121
 122         if (alloc_hint == EXTENT_MAP_INLINE)
 123                 goto out;
 124
 125         while(num_bytes > 0) {
 126                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 127                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 128                                          root->root_key.objectid,
 129                                          trans->transid,
 130                                          inode->i_ino, start, 0,
 131                                          alloc_hint, (u64)-1, &ins, 1);
 132                 if (ret) {
 133                         WARN_ON(1);
 134                         goto out;
 135                 }
 136                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 137                                                start, ins.objectid, ins.offset,
 138                                                ins.offset);
 139                 inode->i_blocks += ins.offset >> 9;
 140                 btrfs_check_file(root, inode);
 141                 num_bytes -= cur_alloc_size;
 142                 alloc_hint = ins.objectid + ins.offset;
 143                 start += cur_alloc_size;
 144         }
 145         btrfs_drop_extent_cache(inode, orig_start,
 146                                 orig_start + orig_num_bytes - 1);
 147         btrfs_add_ordered_inode(inode);
 148         btrfs_update_inode(trans, root, inode);
 149 out:
 150         btrfs_end_transaction(trans, root);
 151         return ret;
 152 }
 153
 154 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 155 {
 156         u64 extent_start;
 157         u64 extent_end;
 158         u64 bytenr;
 159         u64 cow_end;
 160         u64 loops = 0;
 161         u64 total_fs_bytes;
 162         struct btrfs_root *root = BTRFS_I(inode)->root;
 163         struct extent_buffer *leaf;
 164         int found_type;
 165         struct btrfs_path *path;
 166         struct btrfs_file_extent_item *item;
 167         int ret;
 168         int err;
 169         struct btrfs_key found_key;
 170
 171         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 172         path = btrfs_alloc_path();
 173         BUG_ON(!path);
 174 again:
 175         ret = btrfs_lookup_file_extent(NULL, root, path,
 176                                        inode->i_ino, start, 0);
 177         if (ret < 0) {
 178                 btrfs_free_path(path);
 179                 return ret;
 180         }
 181
 182         cow_end = end;
 183         if (ret != 0) {
 184                 if (path->slots[0] == 0)
 185                         goto not_found;
 186                 path->slots[0]--;
 187         }
 188
 189         leaf = path->nodes[0];
 190         item = btrfs_item_ptr(leaf, path->slots[0],
 191                               struct btrfs_file_extent_item);
 192
 193         /* are we inside the extent that was found? */
 194         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 195         found_type = btrfs_key_type(&found_key);
 196         if (found_key.objectid != inode->i_ino ||
 197             found_type != BTRFS_EXTENT_DATA_KEY) {
 198                 goto not_found;
 199         }
 200
 201         found_type = btrfs_file_extent_type(leaf, item);
 202         extent_start = found_key.offset;
 203         if (found_type == BTRFS_FILE_EXTENT_REG) {
 204                 u64 extent_num_bytes;
 205
 206                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 207                 extent_end = extent_start + extent_num_bytes;
 208                 err = 0;
 209
 210                 if (loops && start != extent_start)
 211                         goto not_found;
 212
 213                 if (start < extent_start || start >= extent_end)
 214                         goto not_found;
 215
 216                 cow_end = min(end, extent_end - 1);
 217                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 218                 if (bytenr == 0)
 219                         goto not_found;
 220
 221                 /*
 222                  * we may be called by the resizer, make sure we're inside
 223                  * the limits of the FS
 224                  */
 225                 if (bytenr + extent_num_bytes > total_fs_bytes)
 226                         goto not_found;
 227
 228                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 229                         goto not_found;
 230                 }
 231
 232                 start = extent_end;
 233         } else {
 234                 goto not_found;
 235         }
 236 loop:
 237         if (start > end) {
 238                 btrfs_free_path(path);
 239                 return 0;
 240         }
 241         btrfs_release_path(root, path);
 242         loops++;
 243         goto again;
 244
 245 not_found:
 246         cow_file_range(inode, start, cow_end);
 247         start = cow_end + 1;
 248         goto loop;
 249 }
 250
 251 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 252 {
 253         struct btrfs_root *root = BTRFS_I(inode)->root;
 254         int ret;
 255         mutex_lock(&root->fs_info->fs_mutex);
 256         if (btrfs_test_opt(root, NODATACOW) ||
 257             btrfs_test_flag(inode, NODATACOW))
 258                 ret = run_delalloc_nocow(inode, start, end);
 259         else
 260                 ret = cow_file_range(inode, start, end);
 261
 262         mutex_unlock(&root->fs_info->fs_mutex);
 263         return ret;
 264 }
 265
 266 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 267                        unsigned long old, unsigned long bits)
 268 {
 269         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 270                 struct btrfs_root *root = BTRFS_I(inode)->root;
 271                 spin_lock(&root->fs_info->delalloc_lock);
 272                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 273                 root->fs_info->delalloc_bytes += end - start + 1;
 274                 spin_unlock(&root->fs_info->delalloc_lock);
 275         }
 276         return 0;
 277 }
 278
 279 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 280                          unsigned long old, unsigned long bits)
 281 {
 282         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 283                 struct btrfs_root *root = BTRFS_I(inode)->root;
 284                 spin_lock(&root->fs_info->delalloc_lock);
 285                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 286                         printk("warning: delalloc account %Lu %Lu\n",
 287                                end - start + 1, root->fs_info->delalloc_bytes);
 288                         root->fs_info->delalloc_bytes = 0;
 289                         BTRFS_I(inode)->delalloc_bytes = 0;
 290                 } else {
 291                         root->fs_info->delalloc_bytes -= end - start + 1;
 292                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 293                 }
 294                 spin_unlock(&root->fs_info->delalloc_lock);
 295         }
 296         return 0;
 297 }
 298
 299 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 300                          size_t size, struct bio *bio)
 301 {
 302         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 303         struct btrfs_mapping_tree *map_tree;
 304         u64 logical = bio->bi_sector << 9;
 305         u64 length = 0;
 306         u64 map_length;
 307         struct bio_vec *bvec;
 308         int i;
 309         int ret;
 310
 311         bio_for_each_segment(bvec, bio, i) {
 312                 length += bvec->bv_len;
 313         }
 314         map_tree = &root->fs_info->mapping_tree;
 315         map_length = length;
 316         ret = btrfs_map_block(map_tree, READ, logical,
 317                               &map_length, NULL, 0);
 318
 319         if (map_length < length + size) {
 320                 return 1;
 321         }
 322         return 0;
 323 }
 324
 325 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 326                           int mirror_num)
 327 {
 328         struct btrfs_root *root = BTRFS_I(inode)->root;
 329         struct btrfs_trans_handle *trans;
 330         int ret = 0;
 331
 332         if (!(rw & (1 << BIO_RW))) {
 333                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 334                 BUG_ON(ret);
 335                 goto mapit;
 336         }
 337
 338         if (btrfs_test_opt(root, NODATASUM) ||
 339             btrfs_test_flag(inode, NODATASUM)) {
 340                 goto mapit;
 341         }
 342
 343         mutex_lock(&root->fs_info->fs_mutex);
 344         trans = btrfs_start_transaction(root, 1);
 345         btrfs_set_trans_block_group(trans, inode);
 346         btrfs_csum_file_blocks(trans, root, inode, bio);
 347         ret = btrfs_end_transaction(trans, root);
 348         BUG_ON(ret);
 349         mutex_unlock(&root->fs_info->fs_mutex);
 350 mapit:
 351         return btrfs_map_bio(root, rw, bio, mirror_num);
 352 }
 353
 354 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 355 {
 356         int ret = 0;
 357         struct inode *inode = page->mapping->host;
 358         struct btrfs_root *root = BTRFS_I(inode)->root;
 359         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 360         struct btrfs_csum_item *item;
 361         struct btrfs_path *path = NULL;
 362         u32 csum;
 363         if (btrfs_test_opt(root, NODATASUM) ||
 364             btrfs_test_flag(inode, NODATASUM))
 365                 return 0;
 366         mutex_lock(&root->fs_info->fs_mutex);
 367         path = btrfs_alloc_path();
 368         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 369         if (IS_ERR(item)) {
 370                 ret = PTR_ERR(item);
 371                 /* a csum that isn't present is a preallocated region. */
 372                 if (ret == -ENOENT || ret == -EFBIG)
 373                         ret = 0;
 374                 csum = 0;
 375                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino, start);
 376                 goto out;
 377         }
 378         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 379                            BTRFS_CRC32_SIZE);
 380         set_state_private(io_tree, start, csum);
 381 out:
 382         if (path)
 383                 btrfs_free_path(path);
 384         mutex_unlock(&root->fs_info->fs_mutex);
 385         return ret;
 386 }
 387
 388 struct io_failure_record {
 389         struct page *page;
 390         u64 start;
 391         u64 len;
 392         u64 logical;
 393         int last_mirror;
 394 };
 395
 396 int btrfs_readpage_io_failed_hook(struct bio *failed_bio,
 397                                   struct page *page, u64 start, u64 end,
 398                                   struct extent_state *state)
 399 {
 400         struct io_failure_record *failrec = NULL;
 401         u64 private;
 402         struct extent_map *em;
 403         struct inode *inode = page->mapping->host;
 404         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 405         struct bio *bio;
 406         int num_copies;
 407         int ret;
 408         u64 logical;
 409
 410         ret = get_state_private(failure_tree, start, &private);
 411         if (ret) {
 412                 size_t pg_offset = start - page_offset(page);
 413                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 414                 if (!failrec)
 415                         return -ENOMEM;
 416                 failrec->start = start;
 417                 failrec->len = end - start + 1;
 418                 failrec->last_mirror = 0;
 419
 420                 em = btrfs_get_extent(inode, NULL, pg_offset, start,
 421                                       failrec->len, 0);
 422
 423                 if (!em || IS_ERR(em)) {
 424                         kfree(failrec);
 425                         return -EIO;
 426                 }
 427                 logical = start - em->start;
 428                 logical = em->block_start + logical;
 429                 failrec->logical = logical;
 430                 free_extent_map(em);
 431                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 432                                 EXTENT_DIRTY, GFP_NOFS);
 433                 set_state_private(failure_tree, start,
 434                                  (u64)(unsigned long)failrec);
 435         } else {
 436                 failrec = (struct io_failure_record *)(unsigned long)private;
 437         }
 438         num_copies = btrfs_num_copies(
 439                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 440                               failrec->logical, failrec->len);
 441         failrec->last_mirror++;
 442         if (!state) {
 443                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 444                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 445                                                     failrec->start,
 446                                                     EXTENT_LOCKED);
 447                 if (state && state->start != failrec->start)
 448                         state = NULL;
 449                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 450         }
 451         if (!state || failrec->last_mirror > num_copies) {
 452                 set_state_private(failure_tree, failrec->start, 0);
 453                 clear_extent_bits(failure_tree, failrec->start,
 454                                   failrec->start + failrec->len - 1,
 455                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 456                 kfree(failrec);
 457                 return -EIO;
 458         }
 459         bio = bio_alloc(GFP_NOFS, 1);
 460         bio->bi_private = state;
 461         bio->bi_end_io = failed_bio->bi_end_io;
 462         bio->bi_sector = failrec->logical >> 9;
 463         bio->bi_bdev = failed_bio->bi_bdev;
 464         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 465         btrfs_submit_bio_hook(inode, READ, bio, failrec->last_mirror);
 466         return 0;
 467 }
 468
 469 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 470                                struct extent_state *state)
 471 {
 472         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 473         struct inode *inode = page->mapping->host;
 474         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 475         char *kaddr;
 476         u64 private = ~(u32)0;
 477         int ret;
 478         struct btrfs_root *root = BTRFS_I(inode)->root;
 479         u32 csum = ~(u32)0;
 480         unsigned long flags;
 481
 482         if (btrfs_test_opt(root, NODATASUM) ||
 483             btrfs_test_flag(inode, NODATASUM))
 484                 return 0;
 485         if (state && state->start == start) {
 486                 private = state->private;
 487                 ret = 0;
 488         } else {
 489                 ret = get_state_private(io_tree, start, &private);
 490         }
 491         local_irq_save(flags);
 492         kaddr = kmap_atomic(page, KM_IRQ0);
 493         if (ret) {
 494                 goto zeroit;
 495         }
 496         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 497         btrfs_csum_final(csum, (char *)&csum);
 498         if (csum != private) {
 499                 goto zeroit;
 500         }
 501         kunmap_atomic(kaddr, KM_IRQ0);
 502         local_irq_restore(flags);
 503
 504         /* if the io failure tree for this inode is non-empty,
 505          * check to see if we've recovered from a failed IO
 506          */
 507         private = 0;
 508         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 509                              (u64)-1, 1, EXTENT_DIRTY)) {
 510                 u64 private_failure;
 511                 struct io_failure_record *failure;
 512                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 513                                         start, &private_failure);
 514                 if (ret == 0) {
 515                         failure = (struct io_failure_record *)(unsigned long)
 516                                    private_failure;
 517                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 518                                           failure->start, 0);
 519                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 520                                           failure->start,
 521                                           failure->start + failure->len - 1,
 522                                           EXTENT_DIRTY | EXTENT_LOCKED,
 523                                           GFP_NOFS);
 524                         kfree(failure);
 525                 }
 526         }
 527         return 0;
 528
 529 zeroit:
 530         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 531                page->mapping->host->i_ino, (unsigned long long)start, csum,
 532                private);
 533         memset(kaddr + offset, 1, end - start + 1);
 534         flush_dcache_page(page);
 535         kunmap_atomic(kaddr, KM_IRQ0);
 536         local_irq_restore(flags);
 537         return -EIO;
 538 }
 539
 540 void btrfs_read_locked_inode(struct inode *inode)
 541 {
 542         struct btrfs_path *path;
 543         struct extent_buffer *leaf;
 544         struct btrfs_inode_item *inode_item;
 545         struct btrfs_timespec *tspec;
 546         struct btrfs_root *root = BTRFS_I(inode)->root;
 547         struct btrfs_key location;
 548         u64 alloc_group_block;
 549         u32 rdev;
 550         int ret;
 551
 552         path = btrfs_alloc_path();
 553         BUG_ON(!path);
 554         mutex_lock(&root->fs_info->fs_mutex);
 555         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 556
 557         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 558         if (ret)
 559                 goto make_bad;
 560
 561         leaf = path->nodes[0];
 562         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 563                                     struct btrfs_inode_item);
 564
 565         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 566         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 567         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 568         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 569         inode->i_size = btrfs_inode_size(leaf, inode_item);
 570
 571         tspec = btrfs_inode_atime(inode_item);
 572         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 573         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 574
 575         tspec = btrfs_inode_mtime(inode_item);
 576         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 577         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 578
 579         tspec = btrfs_inode_ctime(inode_item);
 580         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 581         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 582
 583         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 584         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 585         inode->i_rdev = 0;
 586         rdev = btrfs_inode_rdev(leaf, inode_item);
 587
 588         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 589         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 590                                                        alloc_group_block);
 591         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 592         if (!BTRFS_I(inode)->block_group) {
 593                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 594                                                  NULL, 0,
 595                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
 596         }
 597         btrfs_free_path(path);
 598         inode_item = NULL;
 599
 600         mutex_unlock(&root->fs_info->fs_mutex);
 601
 602         switch (inode->i_mode & S_IFMT) {
 603         case S_IFREG:
 604                 inode->i_mapping->a_ops = &btrfs_aops;
 605                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
 606                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 607                 inode->i_fop = &btrfs_file_operations;
 608                 inode->i_op = &btrfs_file_inode_operations;
 609                 break;
 610         case S_IFDIR:
 611                 inode->i_fop = &btrfs_dir_file_operations;
 612                 if (root == root->fs_info->tree_root)
 613                         inode->i_op = &btrfs_dir_ro_inode_operations;
 614                 else
 615                         inode->i_op = &btrfs_dir_inode_operations;
 616                 break;
 617         case S_IFLNK:
 618                 inode->i_op = &btrfs_symlink_inode_operations;
 619                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 620                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
 621                 break;
 622         default:
 623                 init_special_inode(inode, inode->i_mode, rdev);
 624                 break;
 625         }
 626         return;
 627
 628 make_bad:
 629         btrfs_release_path(root, path);
 630         btrfs_free_path(path);
 631         mutex_unlock(&root->fs_info->fs_mutex);
 632         make_bad_inode(inode);
 633 }
 634
 635 static void fill_inode_item(struct extent_buffer *leaf,
 636                             struct btrfs_inode_item *item,
 637                             struct inode *inode)
 638 {
 639         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 640         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 641         btrfs_set_inode_size(leaf, item, inode->i_size);
 642         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 643         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 644
 645         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 646                                inode->i_atime.tv_sec);
 647         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 648                                 inode->i_atime.tv_nsec);
 649
 650         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 651                                inode->i_mtime.tv_sec);
 652         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 653                                 inode->i_mtime.tv_nsec);
 654
 655         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 656                                inode->i_ctime.tv_sec);
 657         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 658                                 inode->i_ctime.tv_nsec);
 659
 660         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 661         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 662         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 663         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 664         btrfs_set_inode_block_group(leaf, item,
 665                                     BTRFS_I(inode)->block_group->key.objectid);
 666 }
 667
 668 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 669                               struct btrfs_root *root,
 670                               struct inode *inode)
 671 {
 672         struct btrfs_inode_item *inode_item;
 673         struct btrfs_path *path;
 674         struct extent_buffer *leaf;
 675         int ret;
 676
 677         path = btrfs_alloc_path();
 678         BUG_ON(!path);
 679         ret = btrfs_lookup_inode(trans, root, path,
 680                                  &BTRFS_I(inode)->location, 1);
 681         if (ret) {
 682                 if (ret > 0)
 683                         ret = -ENOENT;
 684                 goto failed;
 685         }
 686
 687         leaf = path->nodes[0];
 688         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 689                                   struct btrfs_inode_item);
 690
 691         fill_inode_item(leaf, inode_item, inode);
 692         btrfs_mark_buffer_dirty(leaf);
 693         btrfs_set_inode_last_trans(trans, inode);
 694         ret = 0;
 695 failed:
 696         btrfs_release_path(root, path);
 697         btrfs_free_path(path);
 698         return ret;
 699 }
 700
 701
 702 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 703                               struct btrfs_root *root,
 704                               struct inode *dir,
 705                               struct dentry *dentry)
 706 {
 707         struct btrfs_path *path;
 708         const char *name = dentry->d_name.name;
 709         int name_len = dentry->d_name.len;
 710         int ret = 0;
 711         struct extent_buffer *leaf;
 712         struct btrfs_dir_item *di;
 713         struct btrfs_key key;
 714
 715         path = btrfs_alloc_path();
 716         if (!path) {
 717                 ret = -ENOMEM;
 718                 goto err;
 719         }
 720
 721         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 722                                     name, name_len, -1);
 723         if (IS_ERR(di)) {
 724                 ret = PTR_ERR(di);
 725                 goto err;
 726         }
 727         if (!di) {
 728                 ret = -ENOENT;
 729                 goto err;
 730         }
 731         leaf = path->nodes[0];
 732         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 733         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 734         if (ret)
 735                 goto err;
 736         btrfs_release_path(root, path);
 737
 738         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 739                                          key.objectid, name, name_len, -1);
 740         if (IS_ERR(di)) {
 741                 ret = PTR_ERR(di);
 742                 goto err;
 743         }
 744         if (!di) {
 745                 ret = -ENOENT;
 746                 goto err;
 747         }
 748         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 749
 750         dentry->d_inode->i_ctime = dir->i_ctime;
 751         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 752                                   dentry->d_inode->i_ino,
 753                                   dentry->d_parent->d_inode->i_ino);
 754         if (ret) {
 755                 printk("failed to delete reference to %.*s, "
 756                        "inode %lu parent %lu\n", name_len, name,
 757                        dentry->d_inode->i_ino,
 758                        dentry->d_parent->d_inode->i_ino);
 759         }
 760 err:
 761         btrfs_free_path(path);
 762         if (!ret) {
 763                 dir->i_size -= name_len * 2;
 764                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 765                 btrfs_update_inode(trans, root, dir);
 766 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 767                 dentry->d_inode->i_nlink--;
 768 #else
 769                 drop_nlink(dentry->d_inode);
 770 #endif
 771                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 772                 dir->i_sb->s_dirt = 1;
 773         }
 774         return ret;
 775 }
 776
 777 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 778 {
 779         struct btrfs_root *root;
 780         struct btrfs_trans_handle *trans;
 781         struct inode *inode = dentry->d_inode;
 782         int ret;
 783         unsigned long nr = 0;
 784
 785         root = BTRFS_I(dir)->root;
 786         mutex_lock(&root->fs_info->fs_mutex);
 787
 788         ret = btrfs_check_free_space(root, 1, 1);
 789         if (ret)
 790                 goto fail;
 791
 792         trans = btrfs_start_transaction(root, 1);
 793
 794         btrfs_set_trans_block_group(trans, dir);
 795         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 796         nr = trans->blocks_used;
 797
 798         if (inode->i_nlink == 0) {
 799                 int found;
 800                 /* if the inode isn't linked anywhere,
 801                  * we don't need to worry about
 802                  * data=ordered
 803                  */
 804                 found = btrfs_del_ordered_inode(inode);
 805                 if (found == 1) {
 806                         atomic_dec(&inode->i_count);
 807                 }
 808         }
 809
 810         btrfs_end_transaction(trans, root);
 811 fail:
 812         mutex_unlock(&root->fs_info->fs_mutex);
 813         btrfs_btree_balance_dirty(root, nr);
 814         btrfs_throttle(root);
 815         return ret;
 816 }
 817
 818 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 819 {
 820         struct inode *inode = dentry->d_inode;
 821         int err = 0;
 822         int ret;
 823         struct btrfs_root *root = BTRFS_I(dir)->root;
 824         struct btrfs_trans_handle *trans;
 825         unsigned long nr = 0;
 826
 827         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 828                 return -ENOTEMPTY;
 829
 830         mutex_lock(&root->fs_info->fs_mutex);
 831         ret = btrfs_check_free_space(root, 1, 1);
 832         if (ret)
 833                 goto fail;
 834
 835         trans = btrfs_start_transaction(root, 1);
 836         btrfs_set_trans_block_group(trans, dir);
 837
 838         /* now the directory is empty */
 839         err = btrfs_unlink_trans(trans, root, dir, dentry);
 840         if (!err) {
 841                 inode->i_size = 0;
 842         }
 843
 844         nr = trans->blocks_used;
 845         ret = btrfs_end_transaction(trans, root);
 846 fail:
 847         mutex_unlock(&root->fs_info->fs_mutex);
 848         btrfs_btree_balance_dirty(root, nr);
 849         btrfs_throttle(root);
 850
 851         if (ret && !err)
 852                 err = ret;
 853         return err;
 854 }
 855
 856 /*
 857  * this can truncate away extent items, csum items and directory items.
 858  * It starts at a high offset and removes keys until it can't find
 859  * any higher than i_size.
 860  *
 861  * csum items that cross the new i_size are truncated to the new size
 862  * as well.
 863  */
 864 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 865                                    struct btrfs_root *root,
 866                                    struct inode *inode,
 867                                    u32 min_type)
 868 {
 869         int ret;
 870         struct btrfs_path *path;
 871         struct btrfs_key key;
 872         struct btrfs_key found_key;
 873         u32 found_type;
 874         struct extent_buffer *leaf;
 875         struct btrfs_file_extent_item *fi;
 876         u64 extent_start = 0;
 877         u64 extent_num_bytes = 0;
 878         u64 item_end = 0;
 879         u64 root_gen = 0;
 880         u64 root_owner = 0;
 881         int found_extent;
 882         int del_item;
 883         int pending_del_nr = 0;
 884         int pending_del_slot = 0;
 885         int extent_type = -1;
 886
 887         btrfs_drop_extent_cache(inode, inode->i_size, (u64)-1);
 888         path = btrfs_alloc_path();
 889         path->reada = -1;
 890         BUG_ON(!path);
 891
 892         /* FIXME, add redo link to tree so we don't leak on crash */
 893         key.objectid = inode->i_ino;
 894         key.offset = (u64)-1;
 895         key.type = (u8)-1;
 896
 897         btrfs_init_path(path);
 898 search_again:
 899         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 900         if (ret < 0) {
 901                 goto error;
 902         }
 903         if (ret > 0) {
 904                 BUG_ON(path->slots[0] == 0);
 905                 path->slots[0]--;
 906         }
 907
 908         while(1) {
 909                 fi = NULL;
 910                 leaf = path->nodes[0];
 911                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 912                 found_type = btrfs_key_type(&found_key);
 913
 914                 if (found_key.objectid != inode->i_ino)
 915                         break;
 916
 917                 if (found_type < min_type)
 918                         break;
 919
 920                 item_end = found_key.offset;
 921                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 922                         fi = btrfs_item_ptr(leaf, path->slots[0],
 923                                             struct btrfs_file_extent_item);
 924                         extent_type = btrfs_file_extent_type(leaf, fi);
 925                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 926                                 item_end +=
 927                                     btrfs_file_extent_num_bytes(leaf, fi);
 928                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 929                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 930                                                                 path->slots[0]);
 931                                 item_end += btrfs_file_extent_inline_len(leaf,
 932                                                                          item);
 933                         }
 934                         item_end--;
 935                 }
 936                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 937                         ret = btrfs_csum_truncate(trans, root, path,
 938                                                   inode->i_size);
 939                         BUG_ON(ret);
 940                 }
 941                 if (item_end < inode->i_size) {
 942                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 943                                 found_type = BTRFS_INODE_ITEM_KEY;
 944                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 945                                 found_type = BTRFS_CSUM_ITEM_KEY;
 946                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
 947                                 found_type = BTRFS_XATTR_ITEM_KEY;
 948                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
 949                                 found_type = BTRFS_INODE_REF_KEY;
 950                         } else if (found_type) {
 951                                 found_type--;
 952                         } else {
 953                                 break;
 954                         }
 955                         btrfs_set_key_type(&key, found_type);
 956                         goto next;
 957                 }
 958                 if (found_key.offset >= inode->i_size)
 959                         del_item = 1;
 960                 else
 961                         del_item = 0;
 962                 found_extent = 0;
 963
 964                 /* FIXME, shrink the extent if the ref count is only 1 */
 965                 if (found_type != BTRFS_EXTENT_DATA_KEY)
 966                         goto delete;
 967
 968                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 969                         u64 num_dec;
 970                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
 971                         if (!del_item) {
 972                                 u64 orig_num_bytes =
 973                                         btrfs_file_extent_num_bytes(leaf, fi);
 974                                 extent_num_bytes = inode->i_size -
 975                                         found_key.offset + root->sectorsize - 1;
 976                                 extent_num_bytes = extent_num_bytes &
 977                                         ~((u64)root->sectorsize - 1);
 978                                 btrfs_set_file_extent_num_bytes(leaf, fi,
 979                                                          extent_num_bytes);
 980                                 num_dec = (orig_num_bytes -
 981                                            extent_num_bytes);
 982                                 if (extent_start != 0)
 983                                         dec_i_blocks(inode, num_dec);
 984                                 btrfs_mark_buffer_dirty(leaf);
 985                         } else {
 986                                 extent_num_bytes =
 987                                         btrfs_file_extent_disk_num_bytes(leaf,
 988                                                                          fi);
 989                                 /* FIXME blocksize != 4096 */
 990                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
 991                                 if (extent_start != 0) {
 992                                         found_extent = 1;
 993                                         dec_i_blocks(inode, num_dec);
 994                                 }
 995                                 root_gen = btrfs_header_generation(leaf);
 996                                 root_owner = btrfs_header_owner(leaf);
 997                         }
 998                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 999                         if (!del_item) {
1000                                 u32 newsize = inode->i_size - found_key.offset;
1001                                 dec_i_blocks(inode, item_end + 1 -
1002                                             found_key.offset - newsize);
1003                                 newsize =
1004                                     btrfs_file_extent_calc_inline_size(newsize);
1005                                 ret = btrfs_truncate_item(trans, root, path,
1006                                                           newsize, 1);
1007                                 BUG_ON(ret);
1008                         } else {
1009                                 dec_i_blocks(inode, item_end + 1 -
1010                                              found_key.offset);
1011                         }
1012                 }
1013 delete:
1014                 if (del_item) {
1015                         if (!pending_del_nr) {
1016                                 /* no pending yet, add ourselves */
1017                                 pending_del_slot = path->slots[0];
1018                                 pending_del_nr = 1;
1019                         } else if (pending_del_nr &&
1020                                    path->slots[0] + 1 == pending_del_slot) {
1021                                 /* hop on the pending chunk */
1022                                 pending_del_nr++;
1023                                 pending_del_slot = path->slots[0];
1024                         } else {
1025                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1026                         }
1027                 } else {
1028                         break;
1029                 }
1030                 if (found_extent) {
1031                         ret = btrfs_free_extent(trans, root, extent_start,
1032                                                 extent_num_bytes,
1033                                                 root_owner,
1034                                                 root_gen, inode->i_ino,
1035                                                 found_key.offset, 0);
1036                         BUG_ON(ret);
1037                 }
1038 next:
1039                 if (path->slots[0] == 0) {
1040                         if (pending_del_nr)
1041                                 goto del_pending;
1042                         btrfs_release_path(root, path);
1043                         goto search_again;
1044                 }
1045
1046                 path->slots[0]--;
1047                 if (pending_del_nr &&
1048                     path->slots[0] + 1 != pending_del_slot) {
1049                         struct btrfs_key debug;
1050 del_pending:
1051                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1052                                               pending_del_slot);
1053                         ret = btrfs_del_items(trans, root, path,
1054                                               pending_del_slot,
1055                                               pending_del_nr);
1056                         BUG_ON(ret);
1057                         pending_del_nr = 0;
1058                         btrfs_release_path(root, path);
1059                         goto search_again;
1060                 }
1061         }
1062         ret = 0;
1063 error:
1064         if (pending_del_nr) {
1065                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1066                                       pending_del_nr);
1067         }
1068         btrfs_release_path(root, path);
1069         btrfs_free_path(path);
1070         inode->i_sb->s_dirt = 1;
1071         return ret;
1072 }
1073
1074 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
1075                               size_t zero_start)
1076 {
1077         char *kaddr;
1078         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1079         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1080         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1081         int ret = 0;
1082
1083         WARN_ON(!PageLocked(page));
1084         set_page_extent_mapped(page);
1085
1086         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1087         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
1088                             page_end, GFP_NOFS);
1089
1090         if (zero_start != PAGE_CACHE_SIZE) {
1091                 kaddr = kmap(page);
1092                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
1093                 flush_dcache_page(page);
1094                 kunmap(page);
1095         }
1096         set_page_dirty(page);
1097         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1098
1099         return ret;
1100 }
1101
1102 /*
1103  * taken from block_truncate_page, but does cow as it zeros out
1104  * any bytes left in the last page in the file.
1105  */
1106 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1107 {
1108         struct inode *inode = mapping->host;
1109         struct btrfs_root *root = BTRFS_I(inode)->root;
1110         u32 blocksize = root->sectorsize;
1111         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1112         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1113         struct page *page;
1114         int ret = 0;
1115         u64 page_start;
1116
1117         if ((offset & (blocksize - 1)) == 0)
1118                 goto out;
1119
1120         ret = -ENOMEM;
1121         page = grab_cache_page(mapping, index);
1122         if (!page)
1123                 goto out;
1124         if (!PageUptodate(page)) {
1125                 ret = btrfs_readpage(NULL, page);
1126                 lock_page(page);
1127                 if (!PageUptodate(page)) {
1128                         ret = -EIO;
1129                         goto out;
1130                 }
1131         }
1132         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1133
1134         ret = btrfs_cow_one_page(inode, page, offset);
1135
1136         unlock_page(page);
1137         page_cache_release(page);
1138 out:
1139         return ret;
1140 }
1141
1142 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1143 {
1144         struct inode *inode = dentry->d_inode;
1145         int err;
1146
1147         err = inode_change_ok(inode, attr);
1148         if (err)
1149                 return err;
1150
1151         if (S_ISREG(inode->i_mode) &&
1152             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1153                 struct btrfs_trans_handle *trans;
1154                 struct btrfs_root *root = BTRFS_I(inode)->root;
1155                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1156
1157                 u64 mask = root->sectorsize - 1;
1158                 u64 hole_start = (inode->i_size + mask) & ~mask;
1159                 u64 block_end = (attr->ia_size + mask) & ~mask;
1160                 u64 hole_size;
1161                 u64 alloc_hint = 0;
1162
1163                 if (attr->ia_size <= hole_start)
1164                         goto out;
1165
1166                 mutex_lock(&root->fs_info->fs_mutex);
1167                 err = btrfs_check_free_space(root, 1, 0);
1168                 mutex_unlock(&root->fs_info->fs_mutex);
1169                 if (err)
1170                         goto fail;
1171
1172                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1173
1174                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1175                 hole_size = block_end - hole_start;
1176
1177                 mutex_lock(&root->fs_info->fs_mutex);
1178                 trans = btrfs_start_transaction(root, 1);
1179                 btrfs_set_trans_block_group(trans, inode);
1180                 err = btrfs_drop_extents(trans, root, inode,
1181                                          hole_start, block_end, hole_start,
1182                                          &alloc_hint);
1183
1184                 if (alloc_hint != EXTENT_MAP_INLINE) {
1185                         err = btrfs_insert_file_extent(trans, root,
1186                                                        inode->i_ino,
1187                                                        hole_start, 0, 0,
1188                                                        hole_size);
1189                         btrfs_drop_extent_cache(inode, hole_start,
1190                                                 hole_size - 1);
1191                         btrfs_check_file(root, inode);
1192                 }
1193                 btrfs_end_transaction(trans, root);
1194                 mutex_unlock(&root->fs_info->fs_mutex);
1195                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1196                 if (err)
1197                         return err;
1198         }
1199 out:
1200         err = inode_setattr(inode, attr);
1201 fail:
1202         return err;
1203 }
1204
1205 void btrfs_put_inode(struct inode *inode)
1206 {
1207         int ret;
1208
1209         if (!BTRFS_I(inode)->ordered_trans) {
1210                 return;
1211         }
1212
1213         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1214             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1215                 return;
1216
1217         ret = btrfs_del_ordered_inode(inode);
1218         if (ret == 1) {
1219                 atomic_dec(&inode->i_count);
1220         }
1221 }
1222
1223 void btrfs_delete_inode(struct inode *inode)
1224 {
1225         struct btrfs_trans_handle *trans;
1226         struct btrfs_root *root = BTRFS_I(inode)->root;
1227         unsigned long nr;
1228         int ret;
1229
1230         truncate_inode_pages(&inode->i_data, 0);
1231         if (is_bad_inode(inode)) {
1232                 goto no_delete;
1233         }
1234
1235         inode->i_size = 0;
1236         mutex_lock(&root->fs_info->fs_mutex);
1237         trans = btrfs_start_transaction(root, 1);
1238
1239         btrfs_set_trans_block_group(trans, inode);
1240         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1241         if (ret)
1242                 goto no_delete_lock;
1243
1244         nr = trans->blocks_used;
1245         clear_inode(inode);
1246
1247         btrfs_end_transaction(trans, root);
1248         mutex_unlock(&root->fs_info->fs_mutex);
1249         btrfs_btree_balance_dirty(root, nr);
1250         btrfs_throttle(root);
1251         return;
1252
1253 no_delete_lock:
1254         nr = trans->blocks_used;
1255         btrfs_end_transaction(trans, root);
1256         mutex_unlock(&root->fs_info->fs_mutex);
1257         btrfs_btree_balance_dirty(root, nr);
1258         btrfs_throttle(root);
1259 no_delete:
1260         clear_inode(inode);
1261 }
1262
1263 /*
1264  * this returns the key found in the dir entry in the location pointer.
1265  * If no dir entries were found, location->objectid is 0.
1266  */
1267 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1268                                struct btrfs_key *location)
1269 {
1270         const char *name = dentry->d_name.name;
1271         int namelen = dentry->d_name.len;
1272         struct btrfs_dir_item *di;
1273         struct btrfs_path *path;
1274         struct btrfs_root *root = BTRFS_I(dir)->root;
1275         int ret = 0;
1276
1277         if (namelen == 1 && strcmp(name, ".") == 0) {
1278                 location->objectid = dir->i_ino;
1279                 location->type = BTRFS_INODE_ITEM_KEY;
1280                 location->offset = 0;
1281                 return 0;
1282         }
1283         path = btrfs_alloc_path();
1284         BUG_ON(!path);
1285
1286         if (namelen == 2 && strcmp(name, "..") == 0) {
1287                 struct btrfs_key key;
1288                 struct extent_buffer *leaf;
1289                 u32 nritems;
1290                 int slot;
1291
1292                 key.objectid = dir->i_ino;
1293                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1294                 key.offset = 0;
1295                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1296                 BUG_ON(ret == 0);
1297                 ret = 0;
1298
1299                 leaf = path->nodes[0];
1300                 slot = path->slots[0];
1301                 nritems = btrfs_header_nritems(leaf);
1302                 if (slot >= nritems)
1303                         goto out_err;
1304
1305                 btrfs_item_key_to_cpu(leaf, &key, slot);
1306                 if (key.objectid != dir->i_ino ||
1307                     key.type != BTRFS_INODE_REF_KEY) {
1308                         goto out_err;
1309                 }
1310                 location->objectid = key.offset;
1311                 location->type = BTRFS_INODE_ITEM_KEY;
1312                 location->offset = 0;
1313                 goto out;
1314         }
1315
1316         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1317                                     namelen, 0);
1318         if (IS_ERR(di))
1319                 ret = PTR_ERR(di);
1320         if (!di || IS_ERR(di)) {
1321                 goto out_err;
1322         }
1323         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1324 out:
1325         btrfs_free_path(path);
1326         return ret;
1327 out_err:
1328         location->objectid = 0;
1329         goto out;
1330 }
1331
1332 /*
1333  * when we hit a tree root in a directory, the btrfs part of the inode
1334  * needs to be changed to reflect the root directory of the tree root.  This
1335  * is kind of like crossing a mount point.
1336  */
1337 static int fixup_tree_root_location(struct btrfs_root *root,
1338                              struct btrfs_key *location,
1339                              struct btrfs_root **sub_root,
1340                              struct dentry *dentry)
1341 {
1342         struct btrfs_path *path;
1343         struct btrfs_root_item *ri;
1344
1345         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1346                 return 0;
1347         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1348                 return 0;
1349
1350         path = btrfs_alloc_path();
1351         BUG_ON(!path);
1352         mutex_lock(&root->fs_info->fs_mutex);
1353
1354         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1355                                         dentry->d_name.name,
1356                                         dentry->d_name.len);
1357         if (IS_ERR(*sub_root))
1358                 return PTR_ERR(*sub_root);
1359
1360         ri = &(*sub_root)->root_item;
1361         location->objectid = btrfs_root_dirid(ri);
1362         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1363         location->offset = 0;
1364
1365         btrfs_free_path(path);
1366         mutex_unlock(&root->fs_info->fs_mutex);
1367         return 0;
1368 }
1369
1370 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1371 {
1372         struct btrfs_iget_args *args = p;
1373         inode->i_ino = args->ino;
1374         BTRFS_I(inode)->root = args->root;
1375         BTRFS_I(inode)->delalloc_bytes = 0;
1376         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1377         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1378                              inode->i_mapping, GFP_NOFS);
1379         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1380                              inode->i_mapping, GFP_NOFS);
1381         return 0;
1382 }
1383
1384 static int btrfs_find_actor(struct inode *inode, void *opaque)
1385 {
1386         struct btrfs_iget_args *args = opaque;
1387         return (args->ino == inode->i_ino &&
1388                 args->root == BTRFS_I(inode)->root);
1389 }
1390
1391 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1392                             u64 root_objectid)
1393 {
1394         struct btrfs_iget_args args;
1395         args.ino = objectid;
1396         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1397
1398         if (!args.root)
1399                 return NULL;
1400
1401         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1402 }
1403
1404 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1405                                 struct btrfs_root *root)
1406 {
1407         struct inode *inode;
1408         struct btrfs_iget_args args;
1409         args.ino = objectid;
1410         args.root = root;
1411
1412         inode = iget5_locked(s, objectid, btrfs_find_actor,
1413                              btrfs_init_locked_inode,
1414                              (void *)&args);
1415         return inode;
1416 }
1417
1418 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1419                                    struct nameidata *nd)
1420 {
1421         struct inode * inode;
1422         struct btrfs_inode *bi = BTRFS_I(dir);
1423         struct btrfs_root *root = bi->root;
1424         struct btrfs_root *sub_root = root;
1425         struct btrfs_key location;
1426         int ret;
1427
1428         if (dentry->d_name.len > BTRFS_NAME_LEN)
1429                 return ERR_PTR(-ENAMETOOLONG);
1430
1431         mutex_lock(&root->fs_info->fs_mutex);
1432         ret = btrfs_inode_by_name(dir, dentry, &location);
1433         mutex_unlock(&root->fs_info->fs_mutex);
1434
1435         if (ret < 0)
1436                 return ERR_PTR(ret);
1437
1438         inode = NULL;
1439         if (location.objectid) {
1440                 ret = fixup_tree_root_location(root, &location, &sub_root,
1441                                                 dentry);
1442                 if (ret < 0)
1443                         return ERR_PTR(ret);
1444                 if (ret > 0)
1445                         return ERR_PTR(-ENOENT);
1446                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1447                                           sub_root);
1448                 if (!inode)
1449                         return ERR_PTR(-EACCES);
1450                 if (inode->i_state & I_NEW) {
1451                         /* the inode and parent dir are two different roots */
1452                         if (sub_root != root) {
1453                                 igrab(inode);
1454                                 sub_root->inode = inode;
1455                         }
1456                         BTRFS_I(inode)->root = sub_root;
1457                         memcpy(&BTRFS_I(inode)->location, &location,
1458                                sizeof(location));
1459                         btrfs_read_locked_inode(inode);
1460                         unlock_new_inode(inode);
1461                 }
1462         }
1463         return d_splice_alias(inode, dentry);
1464 }
1465
1466 static unsigned char btrfs_filetype_table[] = {
1467         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1468 };
1469
1470 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1471 {
1472         struct inode *inode = filp->f_dentry->d_inode;
1473         struct btrfs_root *root = BTRFS_I(inode)->root;
1474         struct btrfs_item *item;
1475         struct btrfs_dir_item *di;
1476         struct btrfs_key key;
1477         struct btrfs_key found_key;
1478         struct btrfs_path *path;
1479         int ret;
1480         u32 nritems;
1481         struct extent_buffer *leaf;
1482         int slot;
1483         int advance;
1484         unsigned char d_type;
1485         int over = 0;
1486         u32 di_cur;
1487         u32 di_total;
1488         u32 di_len;
1489         int key_type = BTRFS_DIR_INDEX_KEY;
1490         char tmp_name[32];
1491         char *name_ptr;
1492         int name_len;
1493
1494         /* FIXME, use a real flag for deciding about the key type */
1495         if (root->fs_info->tree_root == root)
1496                 key_type = BTRFS_DIR_ITEM_KEY;
1497
1498         /* special case for "." */
1499         if (filp->f_pos == 0) {
1500                 over = filldir(dirent, ".", 1,
1501                                1, inode->i_ino,
1502                                DT_DIR);
1503                 if (over)
1504                         return 0;
1505                 filp->f_pos = 1;
1506         }
1507
1508         mutex_lock(&root->fs_info->fs_mutex);
1509         key.objectid = inode->i_ino;
1510         path = btrfs_alloc_path();
1511         path->reada = 2;
1512
1513         /* special case for .., just use the back ref */
1514         if (filp->f_pos == 1) {
1515                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1516                 key.offset = 0;
1517                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1518                 BUG_ON(ret == 0);
1519                 leaf = path->nodes[0];
1520                 slot = path->slots[0];
1521                 nritems = btrfs_header_nritems(leaf);
1522                 if (slot >= nritems) {
1523                         btrfs_release_path(root, path);
1524                         goto read_dir_items;
1525                 }
1526                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1527                 btrfs_release_path(root, path);
1528                 if (found_key.objectid != key.objectid ||
1529                     found_key.type != BTRFS_INODE_REF_KEY)
1530                         goto read_dir_items;
1531                 over = filldir(dirent, "..", 2,
1532                                2, found_key.offset, DT_DIR);
1533                 if (over)
1534                         goto nopos;
1535                 filp->f_pos = 2;
1536         }
1537
1538 read_dir_items:
1539         btrfs_set_key_type(&key, key_type);
1540         key.offset = filp->f_pos;
1541
1542         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1543         if (ret < 0)
1544                 goto err;
1545         advance = 0;
1546         while(1) {
1547                 leaf = path->nodes[0];
1548                 nritems = btrfs_header_nritems(leaf);
1549                 slot = path->slots[0];
1550                 if (advance || slot >= nritems) {
1551                         if (slot >= nritems -1) {
1552                                 ret = btrfs_next_leaf(root, path);
1553                                 if (ret)
1554                                         break;
1555                                 leaf = path->nodes[0];
1556                                 nritems = btrfs_header_nritems(leaf);
1557                                 slot = path->slots[0];
1558                         } else {
1559                                 slot++;
1560                                 path->slots[0]++;
1561                         }
1562                 }
1563                 advance = 1;
1564                 item = btrfs_item_nr(leaf, slot);
1565                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1566
1567                 if (found_key.objectid != key.objectid)
1568                         break;
1569                 if (btrfs_key_type(&found_key) != key_type)
1570                         break;
1571                 if (found_key.offset < filp->f_pos)
1572                         continue;
1573
1574                 filp->f_pos = found_key.offset;
1575                 advance = 1;
1576                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1577                 di_cur = 0;
1578                 di_total = btrfs_item_size(leaf, item);
1579                 while(di_cur < di_total) {
1580                         struct btrfs_key location;
1581
1582                         name_len = btrfs_dir_name_len(leaf, di);
1583                         if (name_len < 32) {
1584                                 name_ptr = tmp_name;
1585                         } else {
1586                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1587                                 BUG_ON(!name_ptr);
1588                         }
1589                         read_extent_buffer(leaf, name_ptr,
1590                                            (unsigned long)(di + 1), name_len);
1591
1592                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1593                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1594                         over = filldir(dirent, name_ptr, name_len,
1595                                        found_key.offset,
1596                                        location.objectid,
1597                                        d_type);
1598
1599                         if (name_ptr != tmp_name)
1600                                 kfree(name_ptr);
1601
1602                         if (over)
1603                                 goto nopos;
1604                         di_len = btrfs_dir_name_len(leaf, di) +
1605                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1606                         di_cur += di_len;
1607                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1608                 }
1609         }
1610         if (key_type == BTRFS_DIR_INDEX_KEY)
1611                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1612         else
1613                 filp->f_pos++;
1614 nopos:
1615         ret = 0;
1616 err:
1617         btrfs_release_path(root, path);
1618         btrfs_free_path(path);
1619         mutex_unlock(&root->fs_info->fs_mutex);
1620         return ret;
1621 }
1622
1623 int btrfs_write_inode(struct inode *inode, int wait)
1624 {
1625         struct btrfs_root *root = BTRFS_I(inode)->root;
1626         struct btrfs_trans_handle *trans;
1627         int ret = 0;
1628
1629         if (wait) {
1630                 mutex_lock(&root->fs_info->fs_mutex);
1631                 trans = btrfs_start_transaction(root, 1);
1632                 btrfs_set_trans_block_group(trans, inode);
1633                 ret = btrfs_commit_transaction(trans, root);
1634                 mutex_unlock(&root->fs_info->fs_mutex);
1635         }
1636         return ret;
1637 }
1638
1639 /*
1640  * This is somewhat expensive, updating the tree every time the
1641  * inode changes.  But, it is most likely to find the inode in cache.
1642  * FIXME, needs more benchmarking...there are no reasons other than performance
1643  * to keep or drop this code.
1644  */
1645 void btrfs_dirty_inode(struct inode *inode)
1646 {
1647         struct btrfs_root *root = BTRFS_I(inode)->root;
1648         struct btrfs_trans_handle *trans;
1649
1650         mutex_lock(&root->fs_info->fs_mutex);
1651         trans = btrfs_start_transaction(root, 1);
1652         btrfs_set_trans_block_group(trans, inode);
1653         btrfs_update_inode(trans, root, inode);
1654         btrfs_end_transaction(trans, root);
1655         mutex_unlock(&root->fs_info->fs_mutex);
1656 }
1657
1658 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1659                                      struct btrfs_root *root,
1660                                      const char *name, int name_len,
1661                                      u64 ref_objectid,
1662                                      u64 objectid,
1663                                      struct btrfs_block_group_cache *group,
1664                                      int mode)
1665 {
1666         struct inode *inode;
1667         struct btrfs_inode_item *inode_item;
1668         struct btrfs_block_group_cache *new_inode_group;
1669         struct btrfs_key *location;
1670         struct btrfs_path *path;
1671         struct btrfs_inode_ref *ref;
1672         struct btrfs_key key[2];
1673         u32 sizes[2];
1674         unsigned long ptr;
1675         int ret;
1676         int owner;
1677
1678         path = btrfs_alloc_path();
1679         BUG_ON(!path);
1680
1681         inode = new_inode(root->fs_info->sb);
1682         if (!inode)
1683                 return ERR_PTR(-ENOMEM);
1684
1685         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1686         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1687                              inode->i_mapping, GFP_NOFS);
1688         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1689                              inode->i_mapping, GFP_NOFS);
1690         BTRFS_I(inode)->delalloc_bytes = 0;
1691         BTRFS_I(inode)->root = root;
1692
1693         if (mode & S_IFDIR)
1694                 owner = 0;
1695         else
1696                 owner = 1;
1697         new_inode_group = btrfs_find_block_group(root, group, 0,
1698                                        BTRFS_BLOCK_GROUP_METADATA, owner);
1699         if (!new_inode_group) {
1700                 printk("find_block group failed\n");
1701                 new_inode_group = group;
1702         }
1703         BTRFS_I(inode)->block_group = new_inode_group;
1704         BTRFS_I(inode)->flags = 0;
1705
1706         key[0].objectid = objectid;
1707         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1708         key[0].offset = 0;
1709
1710         key[1].objectid = objectid;
1711         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1712         key[1].offset = ref_objectid;
1713
1714         sizes[0] = sizeof(struct btrfs_inode_item);
1715         sizes[1] = name_len + sizeof(*ref);
1716
1717         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1718         if (ret != 0)
1719                 goto fail;
1720
1721         if (objectid > root->highest_inode)
1722                 root->highest_inode = objectid;
1723
1724         inode->i_uid = current->fsuid;
1725         inode->i_gid = current->fsgid;
1726         inode->i_mode = mode;
1727         inode->i_ino = objectid;
1728         inode->i_blocks = 0;
1729         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1730         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1731                                   struct btrfs_inode_item);
1732         fill_inode_item(path->nodes[0], inode_item, inode);
1733
1734         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1735                              struct btrfs_inode_ref);
1736         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1737         ptr = (unsigned long)(ref + 1);
1738         write_extent_buffer(path->nodes[0], name, ptr, name_len);
1739
1740         btrfs_mark_buffer_dirty(path->nodes[0]);
1741         btrfs_free_path(path);
1742
1743         location = &BTRFS_I(inode)->location;
1744         location->objectid = objectid;
1745         location->offset = 0;
1746         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1747
1748         insert_inode_hash(inode);
1749         return inode;
1750 fail:
1751         btrfs_free_path(path);
1752         return ERR_PTR(ret);
1753 }
1754
1755 static inline u8 btrfs_inode_type(struct inode *inode)
1756 {
1757         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1758 }
1759
1760 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1761                             struct dentry *dentry, struct inode *inode,
1762                             int add_backref)
1763 {
1764         int ret;
1765         struct btrfs_key key;
1766         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1767         struct inode *parent_inode;
1768
1769         key.objectid = inode->i_ino;
1770         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1771         key.offset = 0;
1772
1773         ret = btrfs_insert_dir_item(trans, root,
1774                                     dentry->d_name.name, dentry->d_name.len,
1775                                     dentry->d_parent->d_inode->i_ino,
1776                                     &key, btrfs_inode_type(inode));
1777         if (ret == 0) {
1778                 if (add_backref) {
1779                         ret = btrfs_insert_inode_ref(trans, root,
1780                                              dentry->d_name.name,
1781                                              dentry->d_name.len,
1782                                              inode->i_ino,
1783                                              dentry->d_parent->d_inode->i_ino);
1784                 }
1785                 parent_inode = dentry->d_parent->d_inode;
1786                 parent_inode->i_size += dentry->d_name.len * 2;
1787                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1788                 ret = btrfs_update_inode(trans, root,
1789                                          dentry->d_parent->d_inode);
1790         }
1791         return ret;
1792 }
1793
1794 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1795                             struct dentry *dentry, struct inode *inode,
1796                             int backref)
1797 {
1798         int err = btrfs_add_link(trans, dentry, inode, backref);
1799         if (!err) {
1800                 d_instantiate(dentry, inode);
1801                 return 0;
1802         }
1803         if (err > 0)
1804                 err = -EEXIST;
1805         return err;
1806 }
1807
1808 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1809                         int mode, dev_t rdev)
1810 {
1811         struct btrfs_trans_handle *trans;
1812         struct btrfs_root *root = BTRFS_I(dir)->root;
1813         struct inode *inode = NULL;
1814         int err;
1815         int drop_inode = 0;
1816         u64 objectid;
1817         unsigned long nr = 0;
1818
1819         if (!new_valid_dev(rdev))
1820                 return -EINVAL;
1821
1822         mutex_lock(&root->fs_info->fs_mutex);
1823         err = btrfs_check_free_space(root, 1, 0);
1824         if (err)
1825                 goto fail;
1826
1827         trans = btrfs_start_transaction(root, 1);
1828         btrfs_set_trans_block_group(trans, dir);
1829
1830         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1831         if (err) {
1832                 err = -ENOSPC;
1833                 goto out_unlock;
1834         }
1835
1836         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1837                                 dentry->d_name.len,
1838                                 dentry->d_parent->d_inode->i_ino, objectid,
1839                                 BTRFS_I(dir)->block_group, mode);
1840         err = PTR_ERR(inode);
1841         if (IS_ERR(inode))
1842                 goto out_unlock;
1843
1844         btrfs_set_trans_block_group(trans, inode);
1845         err = btrfs_add_nondir(trans, dentry, inode, 0);
1846         if (err)
1847                 drop_inode = 1;
1848         else {
1849                 inode->i_op = &btrfs_special_inode_operations;
1850                 init_special_inode(inode, inode->i_mode, rdev);
1851                 btrfs_update_inode(trans, root, inode);
1852         }
1853         dir->i_sb->s_dirt = 1;
1854         btrfs_update_inode_block_group(trans, inode);
1855         btrfs_update_inode_block_group(trans, dir);
1856 out_unlock:
1857         nr = trans->blocks_used;
1858         btrfs_end_transaction(trans, root);
1859 fail:
1860         mutex_unlock(&root->fs_info->fs_mutex);
1861
1862         if (drop_inode) {
1863                 inode_dec_link_count(inode);
1864                 iput(inode);
1865         }
1866         btrfs_btree_balance_dirty(root, nr);
1867         btrfs_throttle(root);
1868         return err;
1869 }
1870
1871 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1872                         int mode, struct nameidata *nd)
1873 {
1874         struct btrfs_trans_handle *trans;
1875         struct btrfs_root *root = BTRFS_I(dir)->root;
1876         struct inode *inode = NULL;
1877         int err;
1878         int drop_inode = 0;
1879         unsigned long nr = 0;
1880         u64 objectid;
1881
1882         mutex_lock(&root->fs_info->fs_mutex);
1883         err = btrfs_check_free_space(root, 1, 0);
1884         if (err)
1885                 goto fail;
1886         trans = btrfs_start_transaction(root, 1);
1887         btrfs_set_trans_block_group(trans, dir);
1888
1889         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1890         if (err) {
1891                 err = -ENOSPC;
1892                 goto out_unlock;
1893         }
1894
1895         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1896                                 dentry->d_name.len,
1897                                 dentry->d_parent->d_inode->i_ino,
1898                                 objectid, BTRFS_I(dir)->block_group, mode);
1899         err = PTR_ERR(inode);
1900         if (IS_ERR(inode))
1901                 goto out_unlock;
1902
1903         btrfs_set_trans_block_group(trans, inode);
1904         err = btrfs_add_nondir(trans, dentry, inode, 0);
1905         if (err)
1906                 drop_inode = 1;
1907         else {
1908                 inode->i_mapping->a_ops = &btrfs_aops;
1909                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1910                 inode->i_fop = &btrfs_file_operations;
1911                 inode->i_op = &btrfs_file_inode_operations;
1912                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1913                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1914                                      inode->i_mapping, GFP_NOFS);
1915                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1916                                      inode->i_mapping, GFP_NOFS);
1917                 BTRFS_I(inode)->delalloc_bytes = 0;
1918                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1919         }
1920         dir->i_sb->s_dirt = 1;
1921         btrfs_update_inode_block_group(trans, inode);
1922         btrfs_update_inode_block_group(trans, dir);
1923 out_unlock:
1924         nr = trans->blocks_used;
1925         btrfs_end_transaction(trans, root);
1926 fail:
1927         mutex_unlock(&root->fs_info->fs_mutex);
1928
1929         if (drop_inode) {
1930                 inode_dec_link_count(inode);
1931                 iput(inode);
1932         }
1933         btrfs_btree_balance_dirty(root, nr);
1934         btrfs_throttle(root);
1935         return err;
1936 }
1937
1938 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1939                       struct dentry *dentry)
1940 {
1941         struct btrfs_trans_handle *trans;
1942         struct btrfs_root *root = BTRFS_I(dir)->root;
1943         struct inode *inode = old_dentry->d_inode;
1944         unsigned long nr = 0;
1945         int err;
1946         int drop_inode = 0;
1947
1948         if (inode->i_nlink == 0)
1949                 return -ENOENT;
1950
1951 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1952         inode->i_nlink++;
1953 #else
1954         inc_nlink(inode);
1955 #endif
1956         mutex_lock(&root->fs_info->fs_mutex);
1957         err = btrfs_check_free_space(root, 1, 0);
1958         if (err)
1959                 goto fail;
1960         trans = btrfs_start_transaction(root, 1);
1961
1962         btrfs_set_trans_block_group(trans, dir);
1963         atomic_inc(&inode->i_count);
1964         err = btrfs_add_nondir(trans, dentry, inode, 1);
1965
1966         if (err)
1967                 drop_inode = 1;
1968
1969         dir->i_sb->s_dirt = 1;
1970         btrfs_update_inode_block_group(trans, dir);
1971         err = btrfs_update_inode(trans, root, inode);
1972
1973         if (err)
1974                 drop_inode = 1;
1975
1976         nr = trans->blocks_used;
1977         btrfs_end_transaction(trans, root);
1978 fail:
1979         mutex_unlock(&root->fs_info->fs_mutex);
1980
1981         if (drop_inode) {
1982                 inode_dec_link_count(inode);
1983                 iput(inode);
1984         }
1985         btrfs_btree_balance_dirty(root, nr);
1986         btrfs_throttle(root);
1987         return err;
1988 }
1989
1990 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1991 {
1992         struct inode *inode;
1993         struct btrfs_trans_handle *trans;
1994         struct btrfs_root *root = BTRFS_I(dir)->root;
1995         int err = 0;
1996         int drop_on_err = 0;
1997         u64 objectid;
1998         unsigned long nr = 1;
1999
2000         mutex_lock(&root->fs_info->fs_mutex);
2001         err = btrfs_check_free_space(root, 1, 0);
2002         if (err)
2003                 goto out_unlock;
2004
2005         trans = btrfs_start_transaction(root, 1);
2006         btrfs_set_trans_block_group(trans, dir);
2007
2008         if (IS_ERR(trans)) {
2009                 err = PTR_ERR(trans);
2010                 goto out_unlock;
2011         }
2012
2013         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2014         if (err) {
2015                 err = -ENOSPC;
2016                 goto out_unlock;
2017         }
2018
2019         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2020                                 dentry->d_name.len,
2021                                 dentry->d_parent->d_inode->i_ino, objectid,
2022                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
2023         if (IS_ERR(inode)) {
2024                 err = PTR_ERR(inode);
2025                 goto out_fail;
2026         }
2027
2028         drop_on_err = 1;
2029         inode->i_op = &btrfs_dir_inode_operations;
2030         inode->i_fop = &btrfs_dir_file_operations;
2031         btrfs_set_trans_block_group(trans, inode);
2032
2033         inode->i_size = 0;
2034         err = btrfs_update_inode(trans, root, inode);
2035         if (err)
2036                 goto out_fail;
2037
2038         err = btrfs_add_link(trans, dentry, inode, 0);
2039         if (err)
2040                 goto out_fail;
2041
2042         d_instantiate(dentry, inode);
2043         drop_on_err = 0;
2044         dir->i_sb->s_dirt = 1;
2045         btrfs_update_inode_block_group(trans, inode);
2046         btrfs_update_inode_block_group(trans, dir);
2047
2048 out_fail:
2049         nr = trans->blocks_used;
2050         btrfs_end_transaction(trans, root);
2051
2052 out_unlock:
2053         mutex_unlock(&root->fs_info->fs_mutex);
2054         if (drop_on_err)
2055                 iput(inode);
2056         btrfs_btree_balance_dirty(root, nr);
2057         btrfs_throttle(root);
2058         return err;
2059 }
2060
2061 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2062                                     size_t pg_offset, u64 start, u64 len,
2063                                     int create)
2064 {
2065         int ret;
2066         int err = 0;
2067         u64 bytenr;
2068         u64 extent_start = 0;
2069         u64 extent_end = 0;
2070         u64 objectid = inode->i_ino;
2071         u32 found_type;
2072         struct btrfs_path *path;
2073         struct btrfs_root *root = BTRFS_I(inode)->root;
2074         struct btrfs_file_extent_item *item;
2075         struct extent_buffer *leaf;
2076         struct btrfs_key found_key;
2077         struct extent_map *em = NULL;
2078         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2079         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2080         struct btrfs_trans_handle *trans = NULL;
2081
2082         path = btrfs_alloc_path();
2083         BUG_ON(!path);
2084         mutex_lock(&root->fs_info->fs_mutex);
2085
2086 again:
2087         spin_lock(&em_tree->lock);
2088         em = lookup_extent_mapping(em_tree, start, len);
2089         spin_unlock(&em_tree->lock);
2090
2091         if (em) {
2092                 if (em->start > start) {
2093                         printk("get_extent lookup [%Lu %Lu] em [%Lu %Lu]\n",
2094                                start, len, em->start, em->len);
2095                         WARN_ON(1);
2096                 }
2097                 if (em->block_start == EXTENT_MAP_INLINE && page)
2098                         free_extent_map(em);
2099                 else
2100                         goto out;
2101         }
2102         em = alloc_extent_map(GFP_NOFS);
2103         if (!em) {
2104                 err = -ENOMEM;
2105                 goto out;
2106         }
2107
2108         em->start = EXTENT_MAP_HOLE;
2109         em->len = (u64)-1;
2110         em->bdev = inode->i_sb->s_bdev;
2111         ret = btrfs_lookup_file_extent(trans, root, path,
2112                                        objectid, start, trans != NULL);
2113         if (ret < 0) {
2114                 err = ret;
2115                 goto out;
2116         }
2117
2118         if (ret != 0) {
2119                 if (path->slots[0] == 0)
2120                         goto not_found;
2121                 path->slots[0]--;
2122         }
2123
2124         leaf = path->nodes[0];
2125         item = btrfs_item_ptr(leaf, path->slots[0],
2126                               struct btrfs_file_extent_item);
2127         /* are we inside the extent that was found? */
2128         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2129         found_type = btrfs_key_type(&found_key);
2130         if (found_key.objectid != objectid ||
2131             found_type != BTRFS_EXTENT_DATA_KEY) {
2132                 goto not_found;
2133         }
2134
2135         found_type = btrfs_file_extent_type(leaf, item);
2136         extent_start = found_key.offset;
2137         if (found_type == BTRFS_FILE_EXTENT_REG) {
2138                 extent_end = extent_start +
2139                        btrfs_file_extent_num_bytes(leaf, item);
2140                 err = 0;
2141                 if (start < extent_start || start >= extent_end) {
2142                         em->start = start;
2143                         if (start < extent_start) {
2144                                 if (start + len <= extent_start)
2145                                         goto not_found;
2146                                 em->len = extent_end - extent_start;
2147                         } else {
2148                                 em->len = len;
2149                         }
2150                         goto not_found_em;
2151                 }
2152                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2153                 if (bytenr == 0) {
2154                         em->start = extent_start;
2155                         em->len = extent_end - extent_start;
2156                         em->block_start = EXTENT_MAP_HOLE;
2157                         goto insert;
2158                 }
2159                 bytenr += btrfs_file_extent_offset(leaf, item);
2160                 em->block_start = bytenr;
2161                 em->start = extent_start;
2162                 em->len = extent_end - extent_start;
2163                 goto insert;
2164         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2165                 u64 page_start;
2166                 unsigned long ptr;
2167                 char *map;
2168                 size_t size;
2169                 size_t extent_offset;
2170                 size_t copy_size;
2171
2172                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2173                                                     path->slots[0]));
2174                 extent_end = (extent_start + size + root->sectorsize - 1) &
2175                         ~((u64)root->sectorsize - 1);
2176                 if (start < extent_start || start >= extent_end) {
2177                         em->start = start;
2178                         if (start < extent_start) {
2179                                 if (start + len <= extent_start)
2180                                         goto not_found;
2181                                 em->len = extent_end - extent_start;
2182                         } else {
2183                                 em->len = len;
2184                         }
2185                         goto not_found_em;
2186                 }
2187                 em->block_start = EXTENT_MAP_INLINE;
2188
2189                 if (!page) {
2190                         em->start = extent_start;
2191                         em->len = size;
2192                         goto out;
2193                 }
2194
2195                 page_start = page_offset(page) + pg_offset;
2196                 extent_offset = page_start - extent_start;
2197                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2198                                 size - extent_offset);
2199                 em->start = extent_start + extent_offset;
2200                 em->len = (copy_size + root->sectorsize - 1) &
2201                         ~((u64)root->sectorsize - 1);
2202                 map = kmap(page);
2203                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2204                 if (create == 0 && !PageUptodate(page)) {
2205                         read_extent_buffer(leaf, map + pg_offset, ptr,
2206                                            copy_size);
2207                         flush_dcache_page(page);
2208                 } else if (create && PageUptodate(page)) {
2209                         if (!trans) {
2210                                 kunmap(page);
2211                                 free_extent_map(em);
2212                                 em = NULL;
2213                                 btrfs_release_path(root, path);
2214                                 trans = btrfs_start_transaction(root, 1);
2215                                 goto again;
2216                         }
2217                         write_extent_buffer(leaf, map + pg_offset, ptr,
2218                                             copy_size);
2219                         btrfs_mark_buffer_dirty(leaf);
2220                 }
2221                 kunmap(page);
2222                 set_extent_uptodate(io_tree, em->start,
2223                                     extent_map_end(em) - 1, GFP_NOFS);
2224                 goto insert;
2225         } else {
2226                 printk("unkknown found_type %d\n", found_type);
2227                 WARN_ON(1);
2228         }
2229 not_found:
2230         em->start = start;
2231         em->len = len;
2232 not_found_em:
2233         em->block_start = EXTENT_MAP_HOLE;
2234 insert:
2235         btrfs_release_path(root, path);
2236         if (em->start > start || extent_map_end(em) <= start) {
2237                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2238                 err = -EIO;
2239                 goto out;
2240         }
2241
2242         err = 0;
2243         spin_lock(&em_tree->lock);
2244         ret = add_extent_mapping(em_tree, em);
2245         if (ret == -EEXIST) {
2246                 free_extent_map(em);
2247                 em = lookup_extent_mapping(em_tree, start, len);
2248                 if (!em) {
2249                         err = -EIO;
2250                         printk("failing to insert %Lu %Lu\n", start, len);
2251                 }
2252         }
2253         spin_unlock(&em_tree->lock);
2254 out:
2255         btrfs_free_path(path);
2256         if (trans) {
2257                 ret = btrfs_end_transaction(trans, root);
2258                 if (!err)
2259                         err = ret;
2260         }
2261         mutex_unlock(&root->fs_info->fs_mutex);
2262         if (err) {
2263                 free_extent_map(em);
2264                 WARN_ON(1);
2265                 return ERR_PTR(err);
2266         }
2267         return em;
2268 }
2269
2270 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2271                         struct buffer_head *bh_result, int create)
2272 {
2273         struct extent_map *em;
2274         u64 start = (u64)iblock << inode->i_blkbits;
2275         struct btrfs_multi_bio *multi = NULL;
2276         struct btrfs_root *root = BTRFS_I(inode)->root;
2277         u64 len;
2278         u64 logical;
2279         u64 map_length;
2280         int ret = 0;
2281
2282         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2283
2284         if (!em || IS_ERR(em))
2285                 goto out;
2286
2287         if (em->start > start || em->start + em->len <= start)
2288             goto out;
2289
2290         if (em->block_start == EXTENT_MAP_INLINE) {
2291                 ret = -EINVAL;
2292                 goto out;
2293         }
2294
2295         if (em->block_start == EXTENT_MAP_HOLE ||
2296             em->block_start == EXTENT_MAP_DELALLOC) {
2297                 goto out;
2298         }
2299
2300         len = em->start + em->len - start;
2301         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2302
2303         logical = start - em->start;
2304         logical = em->block_start + logical;
2305
2306         map_length = len;
2307         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2308                               logical, &map_length, &multi, 0);
2309         BUG_ON(ret);
2310         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2311         bh_result->b_size = min(map_length, len);
2312         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2313         set_buffer_mapped(bh_result);
2314         kfree(multi);
2315 out:
2316         free_extent_map(em);
2317         return ret;
2318 }
2319
2320 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2321                         const struct iovec *iov, loff_t offset,
2322                         unsigned long nr_segs)
2323 {
2324         struct file *file = iocb->ki_filp;
2325         struct inode *inode = file->f_mapping->host;
2326
2327         if (rw == WRITE)
2328                 return -EINVAL;
2329
2330         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2331                                   offset, nr_segs, btrfs_get_block, NULL);
2332 }
2333
2334 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2335 {
2336         return extent_bmap(mapping, iblock, btrfs_get_extent);
2337 }
2338
2339 int btrfs_readpage(struct file *file, struct page *page)
2340 {
2341         struct extent_io_tree *tree;
2342         tree = &BTRFS_I(page->mapping->host)->io_tree;
2343         return extent_read_full_page(tree, page, btrfs_get_extent);
2344 }
2345
2346 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2347 {
2348         struct extent_io_tree *tree;
2349
2350
2351         if (current->flags & PF_MEMALLOC) {
2352                 redirty_page_for_writepage(wbc, page);
2353                 unlock_page(page);
2354                 return 0;
2355         }
2356         tree = &BTRFS_I(page->mapping->host)->io_tree;
2357         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2358 }
2359
2360 static int btrfs_writepages(struct address_space *mapping,
2361                             struct writeback_control *wbc)
2362 {
2363         struct extent_io_tree *tree;
2364         tree = &BTRFS_I(mapping->host)->io_tree;
2365         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2366 }
2367
2368 static int
2369 btrfs_readpages(struct file *file, struct address_space *mapping,
2370                 struct list_head *pages, unsigned nr_pages)
2371 {
2372         struct extent_io_tree *tree;
2373         tree = &BTRFS_I(mapping->host)->io_tree;
2374         return extent_readpages(tree, mapping, pages, nr_pages,
2375                                 btrfs_get_extent);
2376 }
2377
2378 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2379 {
2380         struct extent_io_tree *tree;
2381         struct extent_map_tree *map;
2382         int ret;
2383
2384         tree = &BTRFS_I(page->mapping->host)->io_tree;
2385         map = &BTRFS_I(page->mapping->host)->extent_tree;
2386         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2387         if (ret == 1) {
2388                 ClearPagePrivate(page);
2389                 set_page_private(page, 0);
2390                 page_cache_release(page);
2391         }
2392         return ret;
2393 }
2394
2395 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2396 {
2397         struct extent_io_tree *tree;
2398
2399         tree = &BTRFS_I(page->mapping->host)->io_tree;
2400         extent_invalidatepage(tree, page, offset);
2401         btrfs_releasepage(page, GFP_NOFS);
2402 }
2403
2404 /*
2405  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2406  * called from a page fault handler when a page is first dirtied. Hence we must
2407  * be careful to check for EOF conditions here. We set the page up correctly
2408  * for a written page which means we get ENOSPC checking when writing into
2409  * holes and correct delalloc and unwritten extent mapping on filesystems that
2410  * support these features.
2411  *
2412  * We are not allowed to take the i_mutex here so we have to play games to
2413  * protect against truncate races as the page could now be beyond EOF.  Because
2414  * vmtruncate() writes the inode size before removing pages, once we have the
2415  * page lock we can determine safely if the page is beyond EOF. If it is not
2416  * beyond EOF, then the page is guaranteed safe against truncation until we
2417  * unlock the page.
2418  */
2419 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2420 {
2421         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2422         struct btrfs_root *root = BTRFS_I(inode)->root;
2423         unsigned long end;
2424         loff_t size;
2425         int ret;
2426         u64 page_start;
2427
2428         mutex_lock(&root->fs_info->fs_mutex);
2429         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2430         mutex_unlock(&root->fs_info->fs_mutex);
2431         if (ret)
2432                 goto out;
2433
2434         ret = -EINVAL;
2435
2436         lock_page(page);
2437         wait_on_page_writeback(page);
2438         size = i_size_read(inode);
2439         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2440
2441         if ((page->mapping != inode->i_mapping) ||
2442             (page_start > size)) {
2443                 /* page got truncated out from underneath us */
2444                 goto out_unlock;
2445         }
2446
2447         /* page is wholly or partially inside EOF */
2448         if (page_start + PAGE_CACHE_SIZE > size)
2449                 end = size & ~PAGE_CACHE_MASK;
2450         else
2451                 end = PAGE_CACHE_SIZE;
2452
2453         ret = btrfs_cow_one_page(inode, page, end);
2454
2455 out_unlock:
2456         unlock_page(page);
2457 out:
2458         return ret;
2459 }
2460
2461 static void btrfs_truncate(struct inode *inode)
2462 {
2463         struct btrfs_root *root = BTRFS_I(inode)->root;
2464         int ret;
2465         struct btrfs_trans_handle *trans;
2466         unsigned long nr;
2467
2468         if (!S_ISREG(inode->i_mode))
2469                 return;
2470         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2471                 return;
2472
2473         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2474
2475         mutex_lock(&root->fs_info->fs_mutex);
2476         trans = btrfs_start_transaction(root, 1);
2477         btrfs_set_trans_block_group(trans, inode);
2478
2479         /* FIXME, add redo link to tree so we don't leak on crash */
2480         ret = btrfs_truncate_in_trans(trans, root, inode,
2481                                       BTRFS_EXTENT_DATA_KEY);
2482         btrfs_update_inode(trans, root, inode);
2483         nr = trans->blocks_used;
2484
2485         ret = btrfs_end_transaction(trans, root);
2486         BUG_ON(ret);
2487         mutex_unlock(&root->fs_info->fs_mutex);
2488         btrfs_btree_balance_dirty(root, nr);
2489         btrfs_throttle(root);
2490 }
2491
2492 static int noinline create_subvol(struct btrfs_root *root, char *name,
2493                                   int namelen)
2494 {
2495         struct btrfs_trans_handle *trans;
2496         struct btrfs_key key;
2497         struct btrfs_root_item root_item;
2498         struct btrfs_inode_item *inode_item;
2499         struct extent_buffer *leaf;
2500         struct btrfs_root *new_root = root;
2501         struct inode *inode;
2502         struct inode *dir;
2503         int ret;
2504         int err;
2505         u64 objectid;
2506         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2507         unsigned long nr = 1;
2508
2509         mutex_lock(&root->fs_info->fs_mutex);
2510         ret = btrfs_check_free_space(root, 1, 0);
2511         if (ret)
2512                 goto fail_commit;
2513
2514         trans = btrfs_start_transaction(root, 1);
2515         BUG_ON(!trans);
2516
2517         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2518                                        0, &objectid);
2519         if (ret)
2520                 goto fail;
2521
2522         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2523                                         objectid, trans->transid, 0, 0,
2524                                         0, 0);
2525         if (IS_ERR(leaf))
2526                 return PTR_ERR(leaf);
2527
2528         btrfs_set_header_nritems(leaf, 0);
2529         btrfs_set_header_level(leaf, 0);
2530         btrfs_set_header_bytenr(leaf, leaf->start);
2531         btrfs_set_header_generation(leaf, trans->transid);
2532         btrfs_set_header_owner(leaf, objectid);
2533
2534         write_extent_buffer(leaf, root->fs_info->fsid,
2535                             (unsigned long)btrfs_header_fsid(leaf),
2536                             BTRFS_FSID_SIZE);
2537         btrfs_mark_buffer_dirty(leaf);
2538
2539         inode_item = &root_item.inode;
2540         memset(inode_item, 0, sizeof(*inode_item));
2541         inode_item->generation = cpu_to_le64(1);
2542         inode_item->size = cpu_to_le64(3);
2543         inode_item->nlink = cpu_to_le32(1);
2544         inode_item->nblocks = cpu_to_le64(1);
2545         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2546
2547         btrfs_set_root_bytenr(&root_item, leaf->start);
2548         btrfs_set_root_level(&root_item, 0);
2549         btrfs_set_root_refs(&root_item, 1);
2550         btrfs_set_root_used(&root_item, 0);
2551
2552         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2553         root_item.drop_level = 0;
2554
2555         free_extent_buffer(leaf);
2556         leaf = NULL;
2557
2558         btrfs_set_root_dirid(&root_item, new_dirid);
2559
2560         key.objectid = objectid;
2561         key.offset = 1;
2562         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2563         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2564                                 &root_item);
2565         if (ret)
2566                 goto fail;
2567
2568         /*
2569          * insert the directory item
2570          */
2571         key.offset = (u64)-1;
2572         dir = root->fs_info->sb->s_root->d_inode;
2573         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2574                                     name, namelen, dir->i_ino, &key,
2575                                     BTRFS_FT_DIR);
2576         if (ret)
2577                 goto fail;
2578
2579         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2580                              name, namelen, objectid,
2581                              root->fs_info->sb->s_root->d_inode->i_ino);
2582         if (ret)
2583                 goto fail;
2584
2585         ret = btrfs_commit_transaction(trans, root);
2586         if (ret)
2587                 goto fail_commit;
2588
2589         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2590         BUG_ON(!new_root);
2591
2592         trans = btrfs_start_transaction(new_root, 1);
2593         BUG_ON(!trans);
2594
2595         inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2596                                 new_dirid,
2597                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2598         if (IS_ERR(inode))
2599                 goto fail;
2600         inode->i_op = &btrfs_dir_inode_operations;
2601         inode->i_fop = &btrfs_dir_file_operations;
2602         new_root->inode = inode;
2603
2604         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2605                                      new_dirid);
2606         inode->i_nlink = 1;
2607         inode->i_size = 0;
2608         ret = btrfs_update_inode(trans, new_root, inode);
2609         if (ret)
2610                 goto fail;
2611 fail:
2612         nr = trans->blocks_used;
2613         err = btrfs_commit_transaction(trans, new_root);
2614         if (err && !ret)
2615                 ret = err;
2616 fail_commit:
2617         mutex_unlock(&root->fs_info->fs_mutex);
2618         btrfs_btree_balance_dirty(root, nr);
2619         btrfs_throttle(root);
2620         return ret;
2621 }
2622
2623 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2624 {
2625         struct btrfs_pending_snapshot *pending_snapshot;
2626         struct btrfs_trans_handle *trans;
2627         int ret;
2628         int err;
2629         unsigned long nr = 0;
2630
2631         if (!root->ref_cows)
2632                 return -EINVAL;
2633
2634         mutex_lock(&root->fs_info->fs_mutex);
2635         ret = btrfs_check_free_space(root, 1, 0);
2636         if (ret)
2637                 goto fail_unlock;
2638
2639         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2640         if (!pending_snapshot) {
2641                 ret = -ENOMEM;
2642                 goto fail_unlock;
2643         }
2644         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2645         if (!pending_snapshot->name) {
2646                 ret = -ENOMEM;
2647                 kfree(pending_snapshot);
2648                 goto fail_unlock;
2649         }
2650         memcpy(pending_snapshot->name, name, namelen);
2651         pending_snapshot->name[namelen] = '\0';
2652         trans = btrfs_start_transaction(root, 1);
2653         BUG_ON(!trans);
2654         pending_snapshot->root = root;
2655         list_add(&pending_snapshot->list,
2656                  &trans->transaction->pending_snapshots);
2657         ret = btrfs_update_inode(trans, root, root->inode);
2658         err = btrfs_commit_transaction(trans, root);
2659
2660 fail_unlock:
2661         mutex_unlock(&root->fs_info->fs_mutex);
2662         btrfs_btree_balance_dirty(root, nr);
2663         btrfs_throttle(root);
2664         return ret;
2665 }
2666
2667 unsigned long btrfs_force_ra(struct address_space *mapping,
2668                               struct file_ra_state *ra, struct file *file,
2669                               pgoff_t offset, pgoff_t last_index)
2670 {
2671         pgoff_t req_size;
2672
2673 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2674         req_size = last_index - offset + 1;
2675         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2676         return offset;
2677 #else
2678         req_size = min(last_index - offset + 1, (pgoff_t)128);
2679         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2680         return offset + req_size;
2681 #endif
2682 }
2683
2684 int btrfs_defrag_file(struct file *file) {
2685         struct inode *inode = fdentry(file)->d_inode;
2686         struct btrfs_root *root = BTRFS_I(inode)->root;
2687         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2688         struct page *page;
2689         unsigned long last_index;
2690         unsigned long ra_index = 0;
2691         u64 page_start;
2692         u64 page_end;
2693         unsigned long i;
2694         int ret;
2695
2696         mutex_lock(&root->fs_info->fs_mutex);
2697         ret = btrfs_check_free_space(root, inode->i_size, 0);
2698         mutex_unlock(&root->fs_info->fs_mutex);
2699         if (ret)
2700                 return -ENOSPC;
2701
2702         mutex_lock(&inode->i_mutex);
2703         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2704         for (i = 0; i <= last_index; i++) {
2705                 if (i == ra_index) {
2706                         ra_index = btrfs_force_ra(inode->i_mapping,
2707                                                   &file->f_ra,
2708                                                   file, ra_index, last_index);
2709                 }
2710                 page = grab_cache_page(inode->i_mapping, i);
2711                 if (!page)
2712                         goto out_unlock;
2713                 if (!PageUptodate(page)) {
2714                         btrfs_readpage(NULL, page);
2715                         lock_page(page);
2716                         if (!PageUptodate(page)) {
2717                                 unlock_page(page);
2718                                 page_cache_release(page);
2719                                 goto out_unlock;
2720                         }
2721                 }
2722                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2723                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2724
2725                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2726                 set_extent_delalloc(io_tree, page_start,
2727                                     page_end, GFP_NOFS);
2728
2729                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2730                 set_page_dirty(page);
2731                 unlock_page(page);
2732                 page_cache_release(page);
2733                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2734         }
2735
2736 out_unlock:
2737         mutex_unlock(&inode->i_mutex);
2738         return 0;
2739 }
2740
2741 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2742 {
2743         u64 new_size;
2744         u64 old_size;
2745         struct btrfs_ioctl_vol_args *vol_args;
2746         struct btrfs_trans_handle *trans;
2747         char *sizestr;
2748         int ret = 0;
2749         int namelen;
2750         int mod = 0;
2751
2752         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2753
2754         if (!vol_args)
2755                 return -ENOMEM;
2756
2757         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2758                 ret = -EFAULT;
2759                 goto out;
2760         }
2761         namelen = strlen(vol_args->name);
2762         if (namelen > BTRFS_VOL_NAME_MAX) {
2763                 ret = -EINVAL;
2764                 goto out;
2765         }
2766
2767         sizestr = vol_args->name;
2768         if (!strcmp(sizestr, "max"))
2769                 new_size = root->fs_info->sb->s_bdev->bd_inode->i_size;
2770         else {
2771                 if (sizestr[0] == '-') {
2772                         mod = -1;
2773                         sizestr++;
2774                 } else if (sizestr[0] == '+') {
2775                         mod = 1;
2776                         sizestr++;
2777                 }
2778                 new_size = btrfs_parse_size(sizestr);
2779                 if (new_size == 0) {
2780                         ret = -EINVAL;
2781                         goto out;
2782                 }
2783         }
2784
2785         mutex_lock(&root->fs_info->fs_mutex);
2786         old_size = btrfs_super_total_bytes(&root->fs_info->super_copy);
2787
2788         if (mod < 0) {
2789                 if (new_size > old_size) {
2790                         ret = -EINVAL;
2791                         goto out_unlock;
2792                 }
2793                 new_size = old_size - new_size;
2794         } else if (mod > 0) {
2795                 new_size = old_size + new_size;
2796         }
2797
2798         if (new_size < 256 * 1024 * 1024) {
2799                 ret = -EINVAL;
2800                 goto out_unlock;
2801         }
2802         if (new_size > root->fs_info->sb->s_bdev->bd_inode->i_size) {
2803                 ret = -EFBIG;
2804                 goto out_unlock;
2805         }
2806
2807         do_div(new_size, root->sectorsize);
2808         new_size *= root->sectorsize;
2809
2810 printk("new size is %Lu\n", new_size);
2811         if (new_size > old_size) {
2812                 trans = btrfs_start_transaction(root, 1);
2813                 ret = btrfs_grow_extent_tree(trans, root, new_size);
2814                 btrfs_commit_transaction(trans, root);
2815         } else {
2816                 ret = btrfs_shrink_extent_tree(root, new_size);
2817         }
2818
2819 out_unlock:
2820         mutex_unlock(&root->fs_info->fs_mutex);
2821 out:
2822         kfree(vol_args);
2823         return ret;
2824 }
2825
2826 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
2827                                             void __user *arg)
2828 {
2829         struct btrfs_ioctl_vol_args *vol_args;
2830         struct btrfs_dir_item *di;
2831         struct btrfs_path *path;
2832         u64 root_dirid;
2833         int namelen;
2834         int ret;
2835
2836         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2837
2838         if (!vol_args)
2839                 return -ENOMEM;
2840
2841         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2842                 ret = -EFAULT;
2843                 goto out;
2844         }
2845
2846         namelen = strlen(vol_args->name);
2847         if (namelen > BTRFS_VOL_NAME_MAX) {
2848                 ret = -EINVAL;
2849                 goto out;
2850         }
2851         if (strchr(vol_args->name, '/')) {
2852                 ret = -EINVAL;
2853                 goto out;
2854         }
2855
2856         path = btrfs_alloc_path();
2857         if (!path) {
2858                 ret = -ENOMEM;
2859                 goto out;
2860         }
2861
2862         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2863         mutex_lock(&root->fs_info->fs_mutex);
2864         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2865                             path, root_dirid,
2866                             vol_args->name, namelen, 0);
2867         mutex_unlock(&root->fs_info->fs_mutex);
2868         btrfs_free_path(path);
2869
2870         if (di && !IS_ERR(di)) {
2871                 ret = -EEXIST;
2872                 goto out;
2873         }
2874
2875         if (IS_ERR(di)) {
2876                 ret = PTR_ERR(di);
2877                 goto out;
2878         }
2879
2880         if (root == root->fs_info->tree_root)
2881                 ret = create_subvol(root, vol_args->name, namelen);
2882         else
2883                 ret = create_snapshot(root, vol_args->name, namelen);
2884 out:
2885         kfree(vol_args);
2886         return ret;
2887 }
2888
2889 static int btrfs_ioctl_defrag(struct file *file)
2890 {
2891         struct inode *inode = fdentry(file)->d_inode;
2892         struct btrfs_root *root = BTRFS_I(inode)->root;
2893
2894         switch (inode->i_mode & S_IFMT) {
2895         case S_IFDIR:
2896                 mutex_lock(&root->fs_info->fs_mutex);
2897                 btrfs_defrag_root(root, 0);
2898                 btrfs_defrag_root(root->fs_info->extent_root, 0);
2899                 mutex_unlock(&root->fs_info->fs_mutex);
2900                 break;
2901         case S_IFREG:
2902                 btrfs_defrag_file(file);
2903                 break;
2904         }
2905
2906         return 0;
2907 }
2908
2909 long btrfs_ioctl(struct file *file, unsigned int
2910                 cmd, unsigned long arg)
2911 {
2912         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
2913
2914         switch (cmd) {
2915         case BTRFS_IOC_SNAP_CREATE:
2916                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
2917         case BTRFS_IOC_DEFRAG:
2918                 return btrfs_ioctl_defrag(file);
2919         case BTRFS_IOC_RESIZE:
2920                 return btrfs_ioctl_resize(root, (void __user *)arg);
2921         }
2922
2923         return -ENOTTY;
2924 }
2925
2926 /*
2927  * Called inside transaction, so use GFP_NOFS
2928  */
2929 struct inode *btrfs_alloc_inode(struct super_block *sb)
2930 {
2931         struct btrfs_inode *ei;
2932
2933         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2934         if (!ei)
2935                 return NULL;
2936         ei->last_trans = 0;
2937         ei->ordered_trans = 0;
2938         return &ei->vfs_inode;
2939 }
2940
2941 void btrfs_destroy_inode(struct inode *inode)
2942 {
2943         WARN_ON(!list_empty(&inode->i_dentry));
2944         WARN_ON(inode->i_data.nrpages);
2945
2946         btrfs_drop_extent_cache(inode, 0, (u64)-1);
2947         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2948 }
2949
2950 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2951 static void init_once(struct kmem_cache * cachep, void *foo)
2952 #else
2953 static void init_once(void * foo, struct kmem_cache * cachep,
2954                       unsigned long flags)
2955 #endif
2956 {
2957         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2958
2959         inode_init_once(&ei->vfs_inode);
2960 }
2961
2962 void btrfs_destroy_cachep(void)
2963 {
2964         if (btrfs_inode_cachep)
2965                 kmem_cache_destroy(btrfs_inode_cachep);
2966         if (btrfs_trans_handle_cachep)
2967                 kmem_cache_destroy(btrfs_trans_handle_cachep);
2968         if (btrfs_transaction_cachep)
2969                 kmem_cache_destroy(btrfs_transaction_cachep);
2970         if (btrfs_bit_radix_cachep)
2971                 kmem_cache_destroy(btrfs_bit_radix_cachep);
2972         if (btrfs_path_cachep)
2973                 kmem_cache_destroy(btrfs_path_cachep);
2974 }
2975
2976 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
2977                                        unsigned long extra_flags,
2978 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
2979                                        void (*ctor)(struct kmem_cache *, void *)
2980 #else
2981                                        void (*ctor)(void *, struct kmem_cache *,
2982                                                     unsigned long)
2983 #endif
2984                                      )
2985 {
2986         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
2987                                  SLAB_MEM_SPREAD | extra_flags), ctor
2988 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2989                                  ,NULL
2990 #endif
2991                                 );
2992 }
2993
2994 int btrfs_init_cachep(void)
2995 {
2996         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
2997                                           sizeof(struct btrfs_inode),
2998                                           0, init_once);
2999         if (!btrfs_inode_cachep)
3000                 goto fail;
3001         btrfs_trans_handle_cachep =
3002                         btrfs_cache_create("btrfs_trans_handle_cache",
3003                                            sizeof(struct btrfs_trans_handle),
3004                                            0, NULL);
3005         if (!btrfs_trans_handle_cachep)
3006                 goto fail;
3007         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3008                                              sizeof(struct btrfs_transaction),
3009                                              0, NULL);
3010         if (!btrfs_transaction_cachep)
3011                 goto fail;
3012         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3013                                          sizeof(struct btrfs_path),
3014                                          0, NULL);
3015         if (!btrfs_path_cachep)
3016                 goto fail;
3017         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3018                                               SLAB_DESTROY_BY_RCU, NULL);
3019         if (!btrfs_bit_radix_cachep)
3020                 goto fail;
3021         return 0;
3022 fail:
3023         btrfs_destroy_cachep();
3024         return -ENOMEM;
3025 }
3026
3027 static int btrfs_getattr(struct vfsmount *mnt,
3028                          struct dentry *dentry, struct kstat *stat)
3029 {
3030         struct inode *inode = dentry->d_inode;
3031         generic_fillattr(inode, stat);
3032         stat->blksize = PAGE_CACHE_SIZE;
3033         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3034         return 0;
3035 }
3036
3037 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3038                            struct inode * new_dir,struct dentry *new_dentry)
3039 {
3040         struct btrfs_trans_handle *trans;
3041         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3042         struct inode *new_inode = new_dentry->d_inode;
3043         struct inode *old_inode = old_dentry->d_inode;
3044         struct timespec ctime = CURRENT_TIME;
3045         struct btrfs_path *path;
3046         int ret;
3047
3048         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3049             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3050                 return -ENOTEMPTY;
3051         }
3052
3053         mutex_lock(&root->fs_info->fs_mutex);
3054         ret = btrfs_check_free_space(root, 1, 0);
3055         if (ret)
3056                 goto out_unlock;
3057
3058         trans = btrfs_start_transaction(root, 1);
3059
3060         btrfs_set_trans_block_group(trans, new_dir);
3061         path = btrfs_alloc_path();
3062         if (!path) {
3063                 ret = -ENOMEM;
3064                 goto out_fail;
3065         }
3066
3067         old_dentry->d_inode->i_nlink++;
3068         old_dir->i_ctime = old_dir->i_mtime = ctime;
3069         new_dir->i_ctime = new_dir->i_mtime = ctime;
3070         old_inode->i_ctime = ctime;
3071
3072         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3073         if (ret)
3074                 goto out_fail;
3075
3076         if (new_inode) {
3077                 new_inode->i_ctime = CURRENT_TIME;
3078                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3079                 if (ret)
3080                         goto out_fail;
3081         }
3082         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
3083         if (ret)
3084                 goto out_fail;
3085
3086 out_fail:
3087         btrfs_free_path(path);
3088         btrfs_end_transaction(trans, root);
3089 out_unlock:
3090         mutex_unlock(&root->fs_info->fs_mutex);
3091         return ret;
3092 }
3093
3094 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3095                          const char *symname)
3096 {
3097         struct btrfs_trans_handle *trans;
3098         struct btrfs_root *root = BTRFS_I(dir)->root;
3099         struct btrfs_path *path;
3100         struct btrfs_key key;
3101         struct inode *inode = NULL;
3102         int err;
3103         int drop_inode = 0;
3104         u64 objectid;
3105         int name_len;
3106         int datasize;
3107         unsigned long ptr;
3108         struct btrfs_file_extent_item *ei;
3109         struct extent_buffer *leaf;
3110         unsigned long nr = 0;
3111
3112         name_len = strlen(symname) + 1;
3113         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3114                 return -ENAMETOOLONG;
3115
3116         mutex_lock(&root->fs_info->fs_mutex);
3117         err = btrfs_check_free_space(root, 1, 0);
3118         if (err)
3119                 goto out_fail;
3120
3121         trans = btrfs_start_transaction(root, 1);
3122         btrfs_set_trans_block_group(trans, dir);
3123
3124         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3125         if (err) {
3126                 err = -ENOSPC;
3127                 goto out_unlock;
3128         }
3129
3130         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
3131                                 dentry->d_name.len,
3132                                 dentry->d_parent->d_inode->i_ino, objectid,
3133                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
3134         err = PTR_ERR(inode);
3135         if (IS_ERR(inode))
3136                 goto out_unlock;
3137
3138         btrfs_set_trans_block_group(trans, inode);
3139         err = btrfs_add_nondir(trans, dentry, inode, 0);
3140         if (err)
3141                 drop_inode = 1;
3142         else {
3143                 inode->i_mapping->a_ops = &btrfs_aops;
3144                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3145                 inode->i_fop = &btrfs_file_operations;
3146                 inode->i_op = &btrfs_file_inode_operations;
3147                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3148                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3149                                      inode->i_mapping, GFP_NOFS);
3150                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3151                                      inode->i_mapping, GFP_NOFS);
3152                 BTRFS_I(inode)->delalloc_bytes = 0;
3153                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3154         }
3155         dir->i_sb->s_dirt = 1;
3156         btrfs_update_inode_block_group(trans, inode);
3157         btrfs_update_inode_block_group(trans, dir);
3158         if (drop_inode)
3159                 goto out_unlock;
3160
3161         path = btrfs_alloc_path();
3162         BUG_ON(!path);
3163         key.objectid = inode->i_ino;
3164         key.offset = 0;
3165         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3166         datasize = btrfs_file_extent_calc_inline_size(name_len);
3167         err = btrfs_insert_empty_item(trans, root, path, &key,
3168                                       datasize);
3169         if (err) {
3170                 drop_inode = 1;
3171                 goto out_unlock;
3172         }
3173         leaf = path->nodes[0];
3174         ei = btrfs_item_ptr(leaf, path->slots[0],
3175                             struct btrfs_file_extent_item);
3176         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3177         btrfs_set_file_extent_type(leaf, ei,
3178                                    BTRFS_FILE_EXTENT_INLINE);
3179         ptr = btrfs_file_extent_inline_start(ei);
3180         write_extent_buffer(leaf, symname, ptr, name_len);
3181         btrfs_mark_buffer_dirty(leaf);
3182         btrfs_free_path(path);
3183
3184         inode->i_op = &btrfs_symlink_inode_operations;
3185         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3186         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3187         inode->i_size = name_len - 1;
3188         err = btrfs_update_inode(trans, root, inode);
3189         if (err)
3190                 drop_inode = 1;
3191
3192 out_unlock:
3193         nr = trans->blocks_used;
3194         btrfs_end_transaction(trans, root);
3195 out_fail:
3196         mutex_unlock(&root->fs_info->fs_mutex);
3197         if (drop_inode) {
3198                 inode_dec_link_count(inode);
3199                 iput(inode);
3200         }
3201         btrfs_btree_balance_dirty(root, nr);
3202         btrfs_throttle(root);
3203         return err;
3204 }
3205
3206 static int btrfs_permission(struct inode *inode, int mask,
3207                             struct nameidata *nd)
3208 {
3209         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3210                 return -EACCES;
3211         return generic_permission(inode, mask, NULL);
3212 }
3213
3214 static struct inode_operations btrfs_dir_inode_operations = {
3215         .lookup         = btrfs_lookup,
3216         .create         = btrfs_create,
3217         .unlink         = btrfs_unlink,
3218         .link           = btrfs_link,
3219         .mkdir          = btrfs_mkdir,
3220         .rmdir          = btrfs_rmdir,
3221         .rename         = btrfs_rename,
3222         .symlink        = btrfs_symlink,
3223         .setattr        = btrfs_setattr,
3224         .mknod          = btrfs_mknod,
3225         .setxattr       = generic_setxattr,
3226         .getxattr       = generic_getxattr,
3227         .listxattr      = btrfs_listxattr,
3228         .removexattr    = generic_removexattr,
3229         .permission     = btrfs_permission,
3230 };
3231 static struct inode_operations btrfs_dir_ro_inode_operations = {
3232         .lookup         = btrfs_lookup,
3233         .permission     = btrfs_permission,
3234 };
3235 static struct file_operations btrfs_dir_file_operations = {
3236         .llseek         = generic_file_llseek,
3237         .read           = generic_read_dir,
3238         .readdir        = btrfs_readdir,
3239         .unlocked_ioctl = btrfs_ioctl,
3240 #ifdef CONFIG_COMPAT
3241         .compat_ioctl   = btrfs_ioctl,
3242 #endif
3243 };
3244
3245 static struct extent_io_ops btrfs_extent_io_ops = {
3246         .fill_delalloc = run_delalloc_range,
3247         .submit_bio_hook = btrfs_submit_bio_hook,
3248         .merge_bio_hook = btrfs_merge_bio_hook,
3249         .readpage_io_hook = btrfs_readpage_io_hook,
3250         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3251         .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
3252         .set_bit_hook = btrfs_set_bit_hook,
3253         .clear_bit_hook = btrfs_clear_bit_hook,
3254 };
3255
3256 static struct address_space_operations btrfs_aops = {
3257         .readpage       = btrfs_readpage,
3258         .writepage      = btrfs_writepage,
3259         .writepages     = btrfs_writepages,
3260         .readpages      = btrfs_readpages,
3261         .sync_page      = block_sync_page,
3262         .bmap           = btrfs_bmap,
3263         .direct_IO      = btrfs_direct_IO,
3264         .invalidatepage = btrfs_invalidatepage,
3265         .releasepage    = btrfs_releasepage,
3266         .set_page_dirty = __set_page_dirty_nobuffers,
3267 };
3268
3269 static struct address_space_operations btrfs_symlink_aops = {
3270         .readpage       = btrfs_readpage,
3271         .writepage      = btrfs_writepage,
3272         .invalidatepage = btrfs_invalidatepage,
3273         .releasepage    = btrfs_releasepage,
3274 };
3275
3276 static struct inode_operations btrfs_file_inode_operations = {
3277         .truncate       = btrfs_truncate,
3278         .getattr        = btrfs_getattr,
3279         .setattr        = btrfs_setattr,
3280         .setxattr       = generic_setxattr,
3281         .getxattr       = generic_getxattr,
3282         .listxattr      = btrfs_listxattr,
3283         .removexattr    = generic_removexattr,
3284         .permission     = btrfs_permission,
3285 };
3286 static struct inode_operations btrfs_special_inode_operations = {
3287         .getattr        = btrfs_getattr,
3288         .setattr        = btrfs_setattr,
3289         .permission     = btrfs_permission,
3290 };
3291 static struct inode_operations btrfs_symlink_inode_operations = {
3292         .readlink       = generic_readlink,
3293         .follow_link    = page_follow_link_light,
3294         .put_link       = page_put_link,
3295         .permission     = btrfs_permission,
3296 };