fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/smp_lock.h>
  30 #include <linux/backing-dev.h>
  31 #include <linux/mpage.h>
  32 #include <linux/swap.h>
  33 #include <linux/writeback.h>
  34 #include <linux/statfs.h>
  35 #include <linux/compat.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/version.h>
  38 #include <linux/xattr.h>
  39 #include "ctree.h"
  40 #include "disk-io.h"
  41 #include "transaction.h"
  42 #include "btrfs_inode.h"
  43 #include "ioctl.h"
  44 #include "print-tree.h"
  45 #include "volumes.h"
  46
  47 struct btrfs_iget_args {
  48         u64 ino;
  49         struct btrfs_root *root;
  50 };
  51
  52 static struct inode_operations btrfs_dir_inode_operations;
  53 static struct inode_operations btrfs_symlink_inode_operations;
  54 static struct inode_operations btrfs_dir_ro_inode_operations;
  55 static struct inode_operations btrfs_special_inode_operations;
  56 static struct inode_operations btrfs_file_inode_operations;
  57 static struct address_space_operations btrfs_aops;
  58 static struct address_space_operations btrfs_symlink_aops;
  59 static struct file_operations btrfs_dir_file_operations;
  60 static struct extent_io_ops btrfs_extent_io_ops;
  61
  62 static struct kmem_cache *btrfs_inode_cachep;
  63 struct kmem_cache *btrfs_trans_handle_cachep;
  64 struct kmem_cache *btrfs_transaction_cachep;
  65 struct kmem_cache *btrfs_bit_radix_cachep;
  66 struct kmem_cache *btrfs_path_cachep;
  67
  68 #define S_SHIFT 12
  69 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  70         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  71         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  72         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  73         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  74         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  75         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  76         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  77 };
  78
  79 int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
  80                            int for_del)
  81 {
  82         u64 total = btrfs_super_total_bytes(&root->fs_info->super_copy);
  83         u64 used = btrfs_super_bytes_used(&root->fs_info->super_copy);
  84         u64 thresh;
  85         unsigned long flags;
  86         int ret = 0;
  87
  88         if (for_del)
  89                 thresh = total * 90;
  90         else
  91                 thresh = total * 85;
  92
  93         do_div(thresh, 100);
  94
  95         spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
  96         if (used + root->fs_info->delalloc_bytes + num_required > thresh)
  97                 ret = -ENOSPC;
  98         spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
  99         return ret;
 100 }
 101
 102 static int cow_file_range(struct inode *inode, u64 start, u64 end)
 103 {
 104         struct btrfs_root *root = BTRFS_I(inode)->root;
 105         struct btrfs_trans_handle *trans;
 106         u64 alloc_hint = 0;
 107         u64 num_bytes;
 108         u64 cur_alloc_size;
 109         u64 blocksize = root->sectorsize;
 110         u64 orig_start = start;
 111         u64 orig_num_bytes;
 112         struct btrfs_key ins;
 113         int ret;
 114
 115         trans = btrfs_start_transaction(root, 1);
 116         BUG_ON(!trans);
 117         btrfs_set_trans_block_group(trans, inode);
 118
 119         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 120         num_bytes = max(blocksize,  num_bytes);
 121         ret = btrfs_drop_extents(trans, root, inode,
 122                                  start, start + num_bytes, start, &alloc_hint);
 123         orig_num_bytes = num_bytes;
 124
 125         if (alloc_hint == EXTENT_MAP_INLINE)
 126                 goto out;
 127
 128         BUG_ON(num_bytes > btrfs_super_total_bytes(&root->fs_info->super_copy));
 129
 130         while(num_bytes > 0) {
 131                 cur_alloc_size = min(num_bytes, root->fs_info->max_extent);
 132                 ret = btrfs_alloc_extent(trans, root, cur_alloc_size,
 133                                          root->sectorsize,
 134                                          root->root_key.objectid,
 135                                          trans->transid,
 136                                          inode->i_ino, start, 0,
 137                                          alloc_hint, (u64)-1, &ins, 1);
 138                 if (ret) {
 139                         WARN_ON(1);
 140                         goto out;
 141                 }
 142                 cur_alloc_size = ins.offset;
 143                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
 144                                                start, ins.objectid, ins.offset,
 145                                                ins.offset, 0);
 146                 inode->i_blocks += ins.offset >> 9;
 147                 btrfs_check_file(root, inode);
 148                 if (num_bytes < cur_alloc_size) {
 149                         printk("num_bytes %Lu cur_alloc %Lu\n", num_bytes,
 150                                cur_alloc_size);
 151                         break;
 152                 }
 153                 num_bytes -= cur_alloc_size;
 154                 alloc_hint = ins.objectid + ins.offset;
 155                 start += cur_alloc_size;
 156         }
 157         btrfs_drop_extent_cache(inode, orig_start,
 158                                 orig_start + orig_num_bytes - 1);
 159         btrfs_add_ordered_inode(inode);
 160         btrfs_update_inode(trans, root, inode);
 161 out:
 162         btrfs_end_transaction(trans, root);
 163         return ret;
 164 }
 165
 166 static int run_delalloc_nocow(struct inode *inode, u64 start, u64 end)
 167 {
 168         u64 extent_start;
 169         u64 extent_end;
 170         u64 bytenr;
 171         u64 cow_end;
 172         u64 loops = 0;
 173         u64 total_fs_bytes;
 174         struct btrfs_root *root = BTRFS_I(inode)->root;
 175         struct extent_buffer *leaf;
 176         int found_type;
 177         struct btrfs_path *path;
 178         struct btrfs_file_extent_item *item;
 179         int ret;
 180         int err;
 181         struct btrfs_key found_key;
 182
 183         total_fs_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy);
 184         path = btrfs_alloc_path();
 185         BUG_ON(!path);
 186 again:
 187         ret = btrfs_lookup_file_extent(NULL, root, path,
 188                                        inode->i_ino, start, 0);
 189         if (ret < 0) {
 190                 btrfs_free_path(path);
 191                 return ret;
 192         }
 193
 194         cow_end = end;
 195         if (ret != 0) {
 196                 if (path->slots[0] == 0)
 197                         goto not_found;
 198                 path->slots[0]--;
 199         }
 200
 201         leaf = path->nodes[0];
 202         item = btrfs_item_ptr(leaf, path->slots[0],
 203                               struct btrfs_file_extent_item);
 204
 205         /* are we inside the extent that was found? */
 206         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 207         found_type = btrfs_key_type(&found_key);
 208         if (found_key.objectid != inode->i_ino ||
 209             found_type != BTRFS_EXTENT_DATA_KEY) {
 210                 goto not_found;
 211         }
 212
 213         found_type = btrfs_file_extent_type(leaf, item);
 214         extent_start = found_key.offset;
 215         if (found_type == BTRFS_FILE_EXTENT_REG) {
 216                 u64 extent_num_bytes;
 217
 218                 extent_num_bytes = btrfs_file_extent_num_bytes(leaf, item);
 219                 extent_end = extent_start + extent_num_bytes;
 220                 err = 0;
 221
 222                 if (loops && start != extent_start)
 223                         goto not_found;
 224
 225                 if (start < extent_start || start >= extent_end)
 226                         goto not_found;
 227
 228                 cow_end = min(end, extent_end - 1);
 229                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
 230                 if (bytenr == 0)
 231                         goto not_found;
 232
 233                 /*
 234                  * we may be called by the resizer, make sure we're inside
 235                  * the limits of the FS
 236                  */
 237                 if (bytenr + extent_num_bytes > total_fs_bytes)
 238                         goto not_found;
 239
 240                 if (btrfs_count_snapshots_in_path(root, path, bytenr) != 1) {
 241                         goto not_found;
 242                 }
 243
 244                 start = extent_end;
 245         } else {
 246                 goto not_found;
 247         }
 248 loop:
 249         if (start > end) {
 250                 btrfs_free_path(path);
 251                 return 0;
 252         }
 253         btrfs_release_path(root, path);
 254         loops++;
 255         goto again;
 256
 257 not_found:
 258         cow_file_range(inode, start, cow_end);
 259         start = cow_end + 1;
 260         goto loop;
 261 }
 262
 263 static int run_delalloc_range(struct inode *inode, u64 start, u64 end)
 264 {
 265         struct btrfs_root *root = BTRFS_I(inode)->root;
 266         int ret;
 267         mutex_lock(&root->fs_info->fs_mutex);
 268         if (btrfs_test_opt(root, NODATACOW) ||
 269             btrfs_test_flag(inode, NODATACOW))
 270                 ret = run_delalloc_nocow(inode, start, end);
 271         else
 272                 ret = cow_file_range(inode, start, end);
 273
 274         mutex_unlock(&root->fs_info->fs_mutex);
 275         return ret;
 276 }
 277
 278 int btrfs_set_bit_hook(struct inode *inode, u64 start, u64 end,
 279                        unsigned long old, unsigned long bits)
 280 {
 281         unsigned long flags;
 282         if (!(old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 283                 struct btrfs_root *root = BTRFS_I(inode)->root;
 284                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 285                 BTRFS_I(inode)->delalloc_bytes += end - start + 1;
 286                 root->fs_info->delalloc_bytes += end - start + 1;
 287                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 288         }
 289         return 0;
 290 }
 291
 292 int btrfs_clear_bit_hook(struct inode *inode, u64 start, u64 end,
 293                          unsigned long old, unsigned long bits)
 294 {
 295         if ((old & EXTENT_DELALLOC) && (bits & EXTENT_DELALLOC)) {
 296                 struct btrfs_root *root = BTRFS_I(inode)->root;
 297                 unsigned long flags;
 298
 299                 spin_lock_irqsave(&root->fs_info->delalloc_lock, flags);
 300                 if (end - start + 1 > root->fs_info->delalloc_bytes) {
 301                         printk("warning: delalloc account %Lu %Lu\n",
 302                                end - start + 1, root->fs_info->delalloc_bytes);
 303                         root->fs_info->delalloc_bytes = 0;
 304                         BTRFS_I(inode)->delalloc_bytes = 0;
 305                 } else {
 306                         root->fs_info->delalloc_bytes -= end - start + 1;
 307                         BTRFS_I(inode)->delalloc_bytes -= end - start + 1;
 308                 }
 309                 spin_unlock_irqrestore(&root->fs_info->delalloc_lock, flags);
 310         }
 311         return 0;
 312 }
 313
 314 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 315                          size_t size, struct bio *bio)
 316 {
 317         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 318         struct btrfs_mapping_tree *map_tree;
 319         u64 logical = bio->bi_sector << 9;
 320         u64 length = 0;
 321         u64 map_length;
 322         int ret;
 323
 324         length = bio->bi_size;
 325         map_tree = &root->fs_info->mapping_tree;
 326         map_length = length;
 327         ret = btrfs_map_block(map_tree, READ, logical,
 328                               &map_length, NULL, 0);
 329
 330         if (map_length < length + size) {
 331                 return 1;
 332         }
 333         return 0;
 334 }
 335
 336 int __btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 337                           int mirror_num)
 338 {
 339         struct btrfs_root *root = BTRFS_I(inode)->root;
 340         struct btrfs_trans_handle *trans;
 341         int ret = 0;
 342         char *sums = NULL;
 343
 344         ret = btrfs_csum_one_bio(root, bio, &sums);
 345         BUG_ON(ret);
 346
 347         mutex_lock(&root->fs_info->fs_mutex);
 348         trans = btrfs_start_transaction(root, 1);
 349
 350         btrfs_set_trans_block_group(trans, inode);
 351         btrfs_csum_file_blocks(trans, root, inode, bio, sums);
 352
 353         ret = btrfs_end_transaction(trans, root);
 354         BUG_ON(ret);
 355         mutex_unlock(&root->fs_info->fs_mutex);
 356
 357         kfree(sums);
 358
 359         return btrfs_map_bio(root, rw, bio, mirror_num);
 360 }
 361
 362 int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 363                           int mirror_num)
 364 {
 365         struct btrfs_root *root = BTRFS_I(inode)->root;
 366         int ret = 0;
 367
 368         if (!(rw & (1 << BIO_RW))) {
 369                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
 370                 BUG_ON(ret);
 371                 goto mapit;
 372         }
 373
 374         if (btrfs_test_opt(root, NODATASUM) ||
 375             btrfs_test_flag(inode, NODATASUM)) {
 376                 goto mapit;
 377         }
 378
 379         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 380                                    inode, rw, bio, mirror_num,
 381                                    __btrfs_submit_bio_hook);
 382 mapit:
 383         return btrfs_map_bio(root, rw, bio, mirror_num);
 384 }
 385
 386 int btrfs_readpage_io_hook(struct page *page, u64 start, u64 end)
 387 {
 388         int ret = 0;
 389         struct inode *inode = page->mapping->host;
 390         struct btrfs_root *root = BTRFS_I(inode)->root;
 391         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 392         struct btrfs_csum_item *item;
 393         struct btrfs_path *path = NULL;
 394         u32 csum;
 395
 396         if (btrfs_test_opt(root, NODATASUM) ||
 397             btrfs_test_flag(inode, NODATASUM))
 398                 return 0;
 399
 400         mutex_lock(&root->fs_info->fs_mutex);
 401         path = btrfs_alloc_path();
 402         item = btrfs_lookup_csum(NULL, root, path, inode->i_ino, start, 0);
 403         if (IS_ERR(item)) {
 404                 ret = PTR_ERR(item);
 405                 /* a csum that isn't present is a preallocated region. */
 406                 if (ret == -ENOENT || ret == -EFBIG)
 407                         ret = 0;
 408                 csum = 0;
 409                 printk("no csum found for inode %lu start %Lu\n", inode->i_ino, start);
 410                 goto out;
 411         }
 412         read_extent_buffer(path->nodes[0], &csum, (unsigned long)item,
 413                            BTRFS_CRC32_SIZE);
 414         set_state_private(io_tree, start, csum);
 415 out:
 416         if (path)
 417                 btrfs_free_path(path);
 418         mutex_unlock(&root->fs_info->fs_mutex);
 419         return ret;
 420 }
 421
 422 struct io_failure_record {
 423         struct page *page;
 424         u64 start;
 425         u64 len;
 426         u64 logical;
 427         int last_mirror;
 428 };
 429
 430 int btrfs_readpage_io_failed_hook(struct bio *failed_bio,
 431                                   struct page *page, u64 start, u64 end,
 432                                   struct extent_state *state)
 433 {
 434         struct io_failure_record *failrec = NULL;
 435         u64 private;
 436         struct extent_map *em;
 437         struct inode *inode = page->mapping->host;
 438         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
 439         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 440         struct bio *bio;
 441         int num_copies;
 442         int ret;
 443         u64 logical;
 444
 445         ret = get_state_private(failure_tree, start, &private);
 446         if (ret) {
 447                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
 448                 if (!failrec)
 449                         return -ENOMEM;
 450                 failrec->start = start;
 451                 failrec->len = end - start + 1;
 452                 failrec->last_mirror = 0;
 453
 454                 spin_lock(&em_tree->lock);
 455                 em = lookup_extent_mapping(em_tree, start, failrec->len);
 456                 if (em->start > start || em->start + em->len < start) {
 457                         free_extent_map(em);
 458                         em = NULL;
 459                 }
 460                 spin_unlock(&em_tree->lock);
 461
 462                 if (!em || IS_ERR(em)) {
 463                         kfree(failrec);
 464                         return -EIO;
 465                 }
 466                 logical = start - em->start;
 467                 logical = em->block_start + logical;
 468                 failrec->logical = logical;
 469                 free_extent_map(em);
 470                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
 471                                 EXTENT_DIRTY, GFP_NOFS);
 472                 set_state_private(failure_tree, start,
 473                                  (u64)(unsigned long)failrec);
 474         } else {
 475                 failrec = (struct io_failure_record *)(unsigned long)private;
 476         }
 477         num_copies = btrfs_num_copies(
 478                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
 479                               failrec->logical, failrec->len);
 480         failrec->last_mirror++;
 481         if (!state) {
 482                 spin_lock_irq(&BTRFS_I(inode)->io_tree.lock);
 483                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
 484                                                     failrec->start,
 485                                                     EXTENT_LOCKED);
 486                 if (state && state->start != failrec->start)
 487                         state = NULL;
 488                 spin_unlock_irq(&BTRFS_I(inode)->io_tree.lock);
 489         }
 490         if (!state || failrec->last_mirror > num_copies) {
 491                 set_state_private(failure_tree, failrec->start, 0);
 492                 clear_extent_bits(failure_tree, failrec->start,
 493                                   failrec->start + failrec->len - 1,
 494                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
 495                 kfree(failrec);
 496                 return -EIO;
 497         }
 498         bio = bio_alloc(GFP_NOFS, 1);
 499         bio->bi_private = state;
 500         bio->bi_end_io = failed_bio->bi_end_io;
 501         bio->bi_sector = failrec->logical >> 9;
 502         bio->bi_bdev = failed_bio->bi_bdev;
 503         bio->bi_size = 0;
 504         bio_add_page(bio, page, failrec->len, start - page_offset(page));
 505         btrfs_submit_bio_hook(inode, READ, bio, failrec->last_mirror);
 506         return 0;
 507 }
 508
 509 int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 510                                struct extent_state *state)
 511 {
 512         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
 513         struct inode *inode = page->mapping->host;
 514         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
 515         char *kaddr;
 516         u64 private = ~(u32)0;
 517         int ret;
 518         struct btrfs_root *root = BTRFS_I(inode)->root;
 519         u32 csum = ~(u32)0;
 520         unsigned long flags;
 521
 522         if (btrfs_test_opt(root, NODATASUM) ||
 523             btrfs_test_flag(inode, NODATASUM))
 524                 return 0;
 525         if (state && state->start == start) {
 526                 private = state->private;
 527                 ret = 0;
 528         } else {
 529                 ret = get_state_private(io_tree, start, &private);
 530         }
 531         local_irq_save(flags);
 532         kaddr = kmap_atomic(page, KM_IRQ0);
 533         if (ret) {
 534                 goto zeroit;
 535         }
 536         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
 537         btrfs_csum_final(csum, (char *)&csum);
 538         if (csum != private) {
 539                 goto zeroit;
 540         }
 541         kunmap_atomic(kaddr, KM_IRQ0);
 542         local_irq_restore(flags);
 543
 544         /* if the io failure tree for this inode is non-empty,
 545          * check to see if we've recovered from a failed IO
 546          */
 547         private = 0;
 548         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
 549                              (u64)-1, 1, EXTENT_DIRTY)) {
 550                 u64 private_failure;
 551                 struct io_failure_record *failure;
 552                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
 553                                         start, &private_failure);
 554                 if (ret == 0) {
 555                         failure = (struct io_failure_record *)(unsigned long)
 556                                    private_failure;
 557                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
 558                                           failure->start, 0);
 559                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
 560                                           failure->start,
 561                                           failure->start + failure->len - 1,
 562                                           EXTENT_DIRTY | EXTENT_LOCKED,
 563                                           GFP_NOFS);
 564                         kfree(failure);
 565                 }
 566         }
 567         return 0;
 568
 569 zeroit:
 570         printk("btrfs csum failed ino %lu off %llu csum %u private %Lu\n",
 571                page->mapping->host->i_ino, (unsigned long long)start, csum,
 572                private);
 573         memset(kaddr + offset, 1, end - start + 1);
 574         flush_dcache_page(page);
 575         kunmap_atomic(kaddr, KM_IRQ0);
 576         local_irq_restore(flags);
 577         if (private == 0)
 578                 return 0;
 579         return -EIO;
 580 }
 581
 582 void btrfs_read_locked_inode(struct inode *inode)
 583 {
 584         struct btrfs_path *path;
 585         struct extent_buffer *leaf;
 586         struct btrfs_inode_item *inode_item;
 587         struct btrfs_timespec *tspec;
 588         struct btrfs_root *root = BTRFS_I(inode)->root;
 589         struct btrfs_key location;
 590         u64 alloc_group_block;
 591         u32 rdev;
 592         int ret;
 593
 594         path = btrfs_alloc_path();
 595         BUG_ON(!path);
 596         mutex_lock(&root->fs_info->fs_mutex);
 597         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
 598
 599         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
 600         if (ret)
 601                 goto make_bad;
 602
 603         leaf = path->nodes[0];
 604         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 605                                     struct btrfs_inode_item);
 606
 607         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
 608         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
 609         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
 610         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
 611         inode->i_size = btrfs_inode_size(leaf, inode_item);
 612
 613         tspec = btrfs_inode_atime(inode_item);
 614         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 615         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 616
 617         tspec = btrfs_inode_mtime(inode_item);
 618         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 619         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 620
 621         tspec = btrfs_inode_ctime(inode_item);
 622         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
 623         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
 624
 625         inode->i_blocks = btrfs_inode_nblocks(leaf, inode_item);
 626         inode->i_generation = btrfs_inode_generation(leaf, inode_item);
 627         inode->i_rdev = 0;
 628         rdev = btrfs_inode_rdev(leaf, inode_item);
 629
 630         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
 631         BTRFS_I(inode)->block_group = btrfs_lookup_block_group(root->fs_info,
 632                                                        alloc_group_block);
 633         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
 634         if (!BTRFS_I(inode)->block_group) {
 635                 BTRFS_I(inode)->block_group = btrfs_find_block_group(root,
 636                                                  NULL, 0,
 637                                                  BTRFS_BLOCK_GROUP_METADATA, 0);
 638         }
 639         btrfs_free_path(path);
 640         inode_item = NULL;
 641
 642         mutex_unlock(&root->fs_info->fs_mutex);
 643
 644         switch (inode->i_mode & S_IFMT) {
 645         case S_IFREG:
 646                 inode->i_mapping->a_ops = &btrfs_aops;
 647                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
 648                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
 649                 inode->i_fop = &btrfs_file_operations;
 650                 inode->i_op = &btrfs_file_inode_operations;
 651                 break;
 652         case S_IFDIR:
 653                 inode->i_fop = &btrfs_dir_file_operations;
 654                 if (root == root->fs_info->tree_root)
 655                         inode->i_op = &btrfs_dir_ro_inode_operations;
 656                 else
 657                         inode->i_op = &btrfs_dir_inode_operations;
 658                 break;
 659         case S_IFLNK:
 660                 inode->i_op = &btrfs_symlink_inode_operations;
 661                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
 662                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
 663                 break;
 664         default:
 665                 init_special_inode(inode, inode->i_mode, rdev);
 666                 break;
 667         }
 668         return;
 669
 670 make_bad:
 671         btrfs_release_path(root, path);
 672         btrfs_free_path(path);
 673         mutex_unlock(&root->fs_info->fs_mutex);
 674         make_bad_inode(inode);
 675 }
 676
 677 static void fill_inode_item(struct extent_buffer *leaf,
 678                             struct btrfs_inode_item *item,
 679                             struct inode *inode)
 680 {
 681         btrfs_set_inode_uid(leaf, item, inode->i_uid);
 682         btrfs_set_inode_gid(leaf, item, inode->i_gid);
 683         btrfs_set_inode_size(leaf, item, inode->i_size);
 684         btrfs_set_inode_mode(leaf, item, inode->i_mode);
 685         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
 686
 687         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
 688                                inode->i_atime.tv_sec);
 689         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
 690                                 inode->i_atime.tv_nsec);
 691
 692         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
 693                                inode->i_mtime.tv_sec);
 694         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
 695                                 inode->i_mtime.tv_nsec);
 696
 697         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
 698                                inode->i_ctime.tv_sec);
 699         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
 700                                 inode->i_ctime.tv_nsec);
 701
 702         btrfs_set_inode_nblocks(leaf, item, inode->i_blocks);
 703         btrfs_set_inode_generation(leaf, item, inode->i_generation);
 704         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
 705         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
 706         btrfs_set_inode_block_group(leaf, item,
 707                                     BTRFS_I(inode)->block_group->key.objectid);
 708 }
 709
 710 int btrfs_update_inode(struct btrfs_trans_handle *trans,
 711                               struct btrfs_root *root,
 712                               struct inode *inode)
 713 {
 714         struct btrfs_inode_item *inode_item;
 715         struct btrfs_path *path;
 716         struct extent_buffer *leaf;
 717         int ret;
 718
 719         path = btrfs_alloc_path();
 720         BUG_ON(!path);
 721         ret = btrfs_lookup_inode(trans, root, path,
 722                                  &BTRFS_I(inode)->location, 1);
 723         if (ret) {
 724                 if (ret > 0)
 725                         ret = -ENOENT;
 726                 goto failed;
 727         }
 728
 729         leaf = path->nodes[0];
 730         inode_item = btrfs_item_ptr(leaf, path->slots[0],
 731                                   struct btrfs_inode_item);
 732
 733         fill_inode_item(leaf, inode_item, inode);
 734         btrfs_mark_buffer_dirty(leaf);
 735         btrfs_set_inode_last_trans(trans, inode);
 736         ret = 0;
 737 failed:
 738         btrfs_release_path(root, path);
 739         btrfs_free_path(path);
 740         return ret;
 741 }
 742
 743
 744 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 745                               struct btrfs_root *root,
 746                               struct inode *dir,
 747                               struct dentry *dentry)
 748 {
 749         struct btrfs_path *path;
 750         const char *name = dentry->d_name.name;
 751         int name_len = dentry->d_name.len;
 752         int ret = 0;
 753         struct extent_buffer *leaf;
 754         struct btrfs_dir_item *di;
 755         struct btrfs_key key;
 756
 757         path = btrfs_alloc_path();
 758         if (!path) {
 759                 ret = -ENOMEM;
 760                 goto err;
 761         }
 762
 763         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 764                                     name, name_len, -1);
 765         if (IS_ERR(di)) {
 766                 ret = PTR_ERR(di);
 767                 goto err;
 768         }
 769         if (!di) {
 770                 ret = -ENOENT;
 771                 goto err;
 772         }
 773         leaf = path->nodes[0];
 774         btrfs_dir_item_key_to_cpu(leaf, di, &key);
 775         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 776         if (ret)
 777                 goto err;
 778         btrfs_release_path(root, path);
 779
 780         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 781                                          key.objectid, name, name_len, -1);
 782         if (IS_ERR(di)) {
 783                 ret = PTR_ERR(di);
 784                 goto err;
 785         }
 786         if (!di) {
 787                 ret = -ENOENT;
 788                 goto err;
 789         }
 790         ret = btrfs_delete_one_dir_name(trans, root, path, di);
 791
 792         dentry->d_inode->i_ctime = dir->i_ctime;
 793         ret = btrfs_del_inode_ref(trans, root, name, name_len,
 794                                   dentry->d_inode->i_ino,
 795                                   dentry->d_parent->d_inode->i_ino);
 796         if (ret) {
 797                 printk("failed to delete reference to %.*s, "
 798                        "inode %lu parent %lu\n", name_len, name,
 799                        dentry->d_inode->i_ino,
 800                        dentry->d_parent->d_inode->i_ino);
 801         }
 802 err:
 803         btrfs_free_path(path);
 804         if (!ret) {
 805                 dir->i_size -= name_len * 2;
 806                 dir->i_mtime = dir->i_ctime = CURRENT_TIME;
 807                 btrfs_update_inode(trans, root, dir);
 808 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
 809                 dentry->d_inode->i_nlink--;
 810 #else
 811                 drop_nlink(dentry->d_inode);
 812 #endif
 813                 ret = btrfs_update_inode(trans, root, dentry->d_inode);
 814                 dir->i_sb->s_dirt = 1;
 815         }
 816         return ret;
 817 }
 818
 819 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 820 {
 821         struct btrfs_root *root;
 822         struct btrfs_trans_handle *trans;
 823         struct inode *inode = dentry->d_inode;
 824         int ret;
 825         unsigned long nr = 0;
 826
 827         root = BTRFS_I(dir)->root;
 828         mutex_lock(&root->fs_info->fs_mutex);
 829
 830         ret = btrfs_check_free_space(root, 1, 1);
 831         if (ret)
 832                 goto fail;
 833
 834         trans = btrfs_start_transaction(root, 1);
 835
 836         btrfs_set_trans_block_group(trans, dir);
 837         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 838         nr = trans->blocks_used;
 839
 840         if (inode->i_nlink == 0) {
 841                 int found;
 842                 /* if the inode isn't linked anywhere,
 843                  * we don't need to worry about
 844                  * data=ordered
 845                  */
 846                 found = btrfs_del_ordered_inode(inode);
 847                 if (found == 1) {
 848                         atomic_dec(&inode->i_count);
 849                 }
 850         }
 851
 852         btrfs_end_transaction(trans, root);
 853 fail:
 854         mutex_unlock(&root->fs_info->fs_mutex);
 855         btrfs_btree_balance_dirty(root, nr);
 856         btrfs_throttle(root);
 857         return ret;
 858 }
 859
 860 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 861 {
 862         struct inode *inode = dentry->d_inode;
 863         int err = 0;
 864         int ret;
 865         struct btrfs_root *root = BTRFS_I(dir)->root;
 866         struct btrfs_trans_handle *trans;
 867         unsigned long nr = 0;
 868
 869         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
 870                 return -ENOTEMPTY;
 871
 872         mutex_lock(&root->fs_info->fs_mutex);
 873         ret = btrfs_check_free_space(root, 1, 1);
 874         if (ret)
 875                 goto fail;
 876
 877         trans = btrfs_start_transaction(root, 1);
 878         btrfs_set_trans_block_group(trans, dir);
 879
 880         /* now the directory is empty */
 881         err = btrfs_unlink_trans(trans, root, dir, dentry);
 882         if (!err) {
 883                 inode->i_size = 0;
 884         }
 885
 886         nr = trans->blocks_used;
 887         ret = btrfs_end_transaction(trans, root);
 888 fail:
 889         mutex_unlock(&root->fs_info->fs_mutex);
 890         btrfs_btree_balance_dirty(root, nr);
 891         btrfs_throttle(root);
 892
 893         if (ret && !err)
 894                 err = ret;
 895         return err;
 896 }
 897
 898 /*
 899  * this can truncate away extent items, csum items and directory items.
 900  * It starts at a high offset and removes keys until it can't find
 901  * any higher than i_size.
 902  *
 903  * csum items that cross the new i_size are truncated to the new size
 904  * as well.
 905  */
 906 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 907                                    struct btrfs_root *root,
 908                                    struct inode *inode,
 909                                    u32 min_type)
 910 {
 911         int ret;
 912         struct btrfs_path *path;
 913         struct btrfs_key key;
 914         struct btrfs_key found_key;
 915         u32 found_type;
 916         struct extent_buffer *leaf;
 917         struct btrfs_file_extent_item *fi;
 918         u64 extent_start = 0;
 919         u64 extent_num_bytes = 0;
 920         u64 item_end = 0;
 921         u64 root_gen = 0;
 922         u64 root_owner = 0;
 923         int found_extent;
 924         int del_item;
 925         int pending_del_nr = 0;
 926         int pending_del_slot = 0;
 927         int extent_type = -1;
 928         u64 mask = root->sectorsize - 1;
 929
 930         btrfs_drop_extent_cache(inode, inode->i_size & (~mask), (u64)-1);
 931         path = btrfs_alloc_path();
 932         path->reada = -1;
 933         BUG_ON(!path);
 934
 935         /* FIXME, add redo link to tree so we don't leak on crash */
 936         key.objectid = inode->i_ino;
 937         key.offset = (u64)-1;
 938         key.type = (u8)-1;
 939
 940         btrfs_init_path(path);
 941 search_again:
 942         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 943         if (ret < 0) {
 944                 goto error;
 945         }
 946         if (ret > 0) {
 947                 BUG_ON(path->slots[0] == 0);
 948                 path->slots[0]--;
 949         }
 950
 951         while(1) {
 952                 fi = NULL;
 953                 leaf = path->nodes[0];
 954                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
 955                 found_type = btrfs_key_type(&found_key);
 956
 957                 if (found_key.objectid != inode->i_ino)
 958                         break;
 959
 960                 if (found_type < min_type)
 961                         break;
 962
 963                 item_end = found_key.offset;
 964                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
 965                         fi = btrfs_item_ptr(leaf, path->slots[0],
 966                                             struct btrfs_file_extent_item);
 967                         extent_type = btrfs_file_extent_type(leaf, fi);
 968                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
 969                                 item_end +=
 970                                     btrfs_file_extent_num_bytes(leaf, fi);
 971                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 972                                 struct btrfs_item *item = btrfs_item_nr(leaf,
 973                                                                 path->slots[0]);
 974                                 item_end += btrfs_file_extent_inline_len(leaf,
 975                                                                          item);
 976                         }
 977                         item_end--;
 978                 }
 979                 if (found_type == BTRFS_CSUM_ITEM_KEY) {
 980                         ret = btrfs_csum_truncate(trans, root, path,
 981                                                   inode->i_size);
 982                         BUG_ON(ret);
 983                 }
 984                 if (item_end < inode->i_size) {
 985                         if (found_type == BTRFS_DIR_ITEM_KEY) {
 986                                 found_type = BTRFS_INODE_ITEM_KEY;
 987                         } else if (found_type == BTRFS_EXTENT_ITEM_KEY) {
 988                                 found_type = BTRFS_CSUM_ITEM_KEY;
 989                         } else if (found_type == BTRFS_EXTENT_DATA_KEY) {
 990                                 found_type = BTRFS_XATTR_ITEM_KEY;
 991                         } else if (found_type == BTRFS_XATTR_ITEM_KEY) {
 992                                 found_type = BTRFS_INODE_REF_KEY;
 993                         } else if (found_type) {
 994                                 found_type--;
 995                         } else {
 996                                 break;
 997                         }
 998                         btrfs_set_key_type(&key, found_type);
 999                         goto next;
1000                 }
1001                 if (found_key.offset >= inode->i_size)
1002                         del_item = 1;
1003                 else
1004                         del_item = 0;
1005                 found_extent = 0;
1006
1007                 /* FIXME, shrink the extent if the ref count is only 1 */
1008                 if (found_type != BTRFS_EXTENT_DATA_KEY)
1009                         goto delete;
1010
1011                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
1012                         u64 num_dec;
1013                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
1014                         if (!del_item) {
1015                                 u64 orig_num_bytes =
1016                                         btrfs_file_extent_num_bytes(leaf, fi);
1017                                 extent_num_bytes = inode->i_size -
1018                                         found_key.offset + root->sectorsize - 1;
1019                                 extent_num_bytes = extent_num_bytes &
1020                                         ~((u64)root->sectorsize - 1);
1021                                 btrfs_set_file_extent_num_bytes(leaf, fi,
1022                                                          extent_num_bytes);
1023                                 num_dec = (orig_num_bytes -
1024                                            extent_num_bytes);
1025                                 if (extent_start != 0)
1026                                         dec_i_blocks(inode, num_dec);
1027                                 btrfs_mark_buffer_dirty(leaf);
1028                         } else {
1029                                 extent_num_bytes =
1030                                         btrfs_file_extent_disk_num_bytes(leaf,
1031                                                                          fi);
1032                                 /* FIXME blocksize != 4096 */
1033                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
1034                                 if (extent_start != 0) {
1035                                         found_extent = 1;
1036                                         dec_i_blocks(inode, num_dec);
1037                                 }
1038                                 root_gen = btrfs_header_generation(leaf);
1039                                 root_owner = btrfs_header_owner(leaf);
1040                         }
1041                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1042                         if (!del_item) {
1043                                 u32 newsize = inode->i_size - found_key.offset;
1044                                 dec_i_blocks(inode, item_end + 1 -
1045                                             found_key.offset - newsize);
1046                                 newsize =
1047                                     btrfs_file_extent_calc_inline_size(newsize);
1048                                 ret = btrfs_truncate_item(trans, root, path,
1049                                                           newsize, 1);
1050                                 BUG_ON(ret);
1051                         } else {
1052                                 dec_i_blocks(inode, item_end + 1 -
1053                                              found_key.offset);
1054                         }
1055                 }
1056 delete:
1057                 if (del_item) {
1058                         if (!pending_del_nr) {
1059                                 /* no pending yet, add ourselves */
1060                                 pending_del_slot = path->slots[0];
1061                                 pending_del_nr = 1;
1062                         } else if (pending_del_nr &&
1063                                    path->slots[0] + 1 == pending_del_slot) {
1064                                 /* hop on the pending chunk */
1065                                 pending_del_nr++;
1066                                 pending_del_slot = path->slots[0];
1067                         } else {
1068                                 printk("bad pending slot %d pending_del_nr %d pending_del_slot %d\n", path->slots[0], pending_del_nr, pending_del_slot);
1069                         }
1070                 } else {
1071                         break;
1072                 }
1073                 if (found_extent) {
1074                         ret = btrfs_free_extent(trans, root, extent_start,
1075                                                 extent_num_bytes,
1076                                                 root_owner,
1077                                                 root_gen, inode->i_ino,
1078                                                 found_key.offset, 0);
1079                         BUG_ON(ret);
1080                 }
1081 next:
1082                 if (path->slots[0] == 0) {
1083                         if (pending_del_nr)
1084                                 goto del_pending;
1085                         btrfs_release_path(root, path);
1086                         goto search_again;
1087                 }
1088
1089                 path->slots[0]--;
1090                 if (pending_del_nr &&
1091                     path->slots[0] + 1 != pending_del_slot) {
1092                         struct btrfs_key debug;
1093 del_pending:
1094                         btrfs_item_key_to_cpu(path->nodes[0], &debug,
1095                                               pending_del_slot);
1096                         ret = btrfs_del_items(trans, root, path,
1097                                               pending_del_slot,
1098                                               pending_del_nr);
1099                         BUG_ON(ret);
1100                         pending_del_nr = 0;
1101                         btrfs_release_path(root, path);
1102                         goto search_again;
1103                 }
1104         }
1105         ret = 0;
1106 error:
1107         if (pending_del_nr) {
1108                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
1109                                       pending_del_nr);
1110         }
1111         btrfs_release_path(root, path);
1112         btrfs_free_path(path);
1113         inode->i_sb->s_dirt = 1;
1114         return ret;
1115 }
1116
1117 static int btrfs_cow_one_page(struct inode *inode, struct page *page,
1118                               size_t zero_start)
1119 {
1120         char *kaddr;
1121         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1122         u64 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1123         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
1124         int ret = 0;
1125
1126         WARN_ON(!PageLocked(page));
1127         set_page_extent_mapped(page);
1128
1129         lock_extent(io_tree, page_start, page_end, GFP_NOFS);
1130         set_extent_delalloc(&BTRFS_I(inode)->io_tree, page_start,
1131                             page_end, GFP_NOFS);
1132
1133         if (zero_start != PAGE_CACHE_SIZE) {
1134                 kaddr = kmap(page);
1135                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
1136                 flush_dcache_page(page);
1137                 kunmap(page);
1138         }
1139         set_page_dirty(page);
1140         unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
1141
1142         return ret;
1143 }
1144
1145 /*
1146  * taken from block_truncate_page, but does cow as it zeros out
1147  * any bytes left in the last page in the file.
1148  */
1149 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
1150 {
1151         struct inode *inode = mapping->host;
1152         struct btrfs_root *root = BTRFS_I(inode)->root;
1153         u32 blocksize = root->sectorsize;
1154         pgoff_t index = from >> PAGE_CACHE_SHIFT;
1155         unsigned offset = from & (PAGE_CACHE_SIZE-1);
1156         struct page *page;
1157         int ret = 0;
1158         u64 page_start;
1159
1160         if ((offset & (blocksize - 1)) == 0)
1161                 goto out;
1162
1163         ret = -ENOMEM;
1164         page = grab_cache_page(mapping, index);
1165         if (!page)
1166                 goto out;
1167         if (!PageUptodate(page)) {
1168                 ret = btrfs_readpage(NULL, page);
1169                 lock_page(page);
1170                 if (!PageUptodate(page)) {
1171                         ret = -EIO;
1172                         goto out;
1173                 }
1174         }
1175         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
1176
1177         ret = btrfs_cow_one_page(inode, page, offset);
1178
1179         unlock_page(page);
1180         page_cache_release(page);
1181 out:
1182         return ret;
1183 }
1184
1185 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
1186 {
1187         struct inode *inode = dentry->d_inode;
1188         int err;
1189
1190         err = inode_change_ok(inode, attr);
1191         if (err)
1192                 return err;
1193
1194         if (S_ISREG(inode->i_mode) &&
1195             attr->ia_valid & ATTR_SIZE && attr->ia_size > inode->i_size) {
1196                 struct btrfs_trans_handle *trans;
1197                 struct btrfs_root *root = BTRFS_I(inode)->root;
1198                 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1199
1200                 u64 mask = root->sectorsize - 1;
1201                 u64 hole_start = (inode->i_size + mask) & ~mask;
1202                 u64 block_end = (attr->ia_size + mask) & ~mask;
1203                 u64 hole_size;
1204                 u64 alloc_hint = 0;
1205
1206                 if (attr->ia_size <= hole_start)
1207                         goto out;
1208
1209                 mutex_lock(&root->fs_info->fs_mutex);
1210                 err = btrfs_check_free_space(root, 1, 0);
1211                 mutex_unlock(&root->fs_info->fs_mutex);
1212                 if (err)
1213                         goto fail;
1214
1215                 btrfs_truncate_page(inode->i_mapping, inode->i_size);
1216
1217                 lock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1218                 hole_size = block_end - hole_start;
1219
1220                 mutex_lock(&root->fs_info->fs_mutex);
1221                 trans = btrfs_start_transaction(root, 1);
1222                 btrfs_set_trans_block_group(trans, inode);
1223                 err = btrfs_drop_extents(trans, root, inode,
1224                                          hole_start, block_end, hole_start,
1225                                          &alloc_hint);
1226
1227                 if (alloc_hint != EXTENT_MAP_INLINE) {
1228                         err = btrfs_insert_file_extent(trans, root,
1229                                                        inode->i_ino,
1230                                                        hole_start, 0, 0,
1231                                                        hole_size, 0);
1232                         btrfs_drop_extent_cache(inode, hole_start,
1233                                                 (u64)-1);
1234                         btrfs_check_file(root, inode);
1235                 }
1236                 btrfs_end_transaction(trans, root);
1237                 mutex_unlock(&root->fs_info->fs_mutex);
1238                 unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
1239                 if (err)
1240                         return err;
1241         }
1242 out:
1243         err = inode_setattr(inode, attr);
1244 fail:
1245         return err;
1246 }
1247
1248 void btrfs_put_inode(struct inode *inode)
1249 {
1250         int ret;
1251
1252         if (!BTRFS_I(inode)->ordered_trans) {
1253                 return;
1254         }
1255
1256         if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY) ||
1257             mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))
1258                 return;
1259
1260         ret = btrfs_del_ordered_inode(inode);
1261         if (ret == 1) {
1262                 atomic_dec(&inode->i_count);
1263         }
1264 }
1265
1266 void btrfs_delete_inode(struct inode *inode)
1267 {
1268         struct btrfs_trans_handle *trans;
1269         struct btrfs_root *root = BTRFS_I(inode)->root;
1270         unsigned long nr;
1271         int ret;
1272
1273         truncate_inode_pages(&inode->i_data, 0);
1274         if (is_bad_inode(inode)) {
1275                 goto no_delete;
1276         }
1277
1278         inode->i_size = 0;
1279         mutex_lock(&root->fs_info->fs_mutex);
1280         trans = btrfs_start_transaction(root, 1);
1281
1282         btrfs_set_trans_block_group(trans, inode);
1283         ret = btrfs_truncate_in_trans(trans, root, inode, 0);
1284         if (ret)
1285                 goto no_delete_lock;
1286
1287         nr = trans->blocks_used;
1288         clear_inode(inode);
1289
1290         btrfs_end_transaction(trans, root);
1291         mutex_unlock(&root->fs_info->fs_mutex);
1292         btrfs_btree_balance_dirty(root, nr);
1293         btrfs_throttle(root);
1294         return;
1295
1296 no_delete_lock:
1297         nr = trans->blocks_used;
1298         btrfs_end_transaction(trans, root);
1299         mutex_unlock(&root->fs_info->fs_mutex);
1300         btrfs_btree_balance_dirty(root, nr);
1301         btrfs_throttle(root);
1302 no_delete:
1303         clear_inode(inode);
1304 }
1305
1306 /*
1307  * this returns the key found in the dir entry in the location pointer.
1308  * If no dir entries were found, location->objectid is 0.
1309  */
1310 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
1311                                struct btrfs_key *location)
1312 {
1313         const char *name = dentry->d_name.name;
1314         int namelen = dentry->d_name.len;
1315         struct btrfs_dir_item *di;
1316         struct btrfs_path *path;
1317         struct btrfs_root *root = BTRFS_I(dir)->root;
1318         int ret = 0;
1319
1320         if (namelen == 1 && strcmp(name, ".") == 0) {
1321                 location->objectid = dir->i_ino;
1322                 location->type = BTRFS_INODE_ITEM_KEY;
1323                 location->offset = 0;
1324                 return 0;
1325         }
1326         path = btrfs_alloc_path();
1327         BUG_ON(!path);
1328
1329         if (namelen == 2 && strcmp(name, "..") == 0) {
1330                 struct btrfs_key key;
1331                 struct extent_buffer *leaf;
1332                 u32 nritems;
1333                 int slot;
1334
1335                 key.objectid = dir->i_ino;
1336                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1337                 key.offset = 0;
1338                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1339                 BUG_ON(ret == 0);
1340                 ret = 0;
1341
1342                 leaf = path->nodes[0];
1343                 slot = path->slots[0];
1344                 nritems = btrfs_header_nritems(leaf);
1345                 if (slot >= nritems)
1346                         goto out_err;
1347
1348                 btrfs_item_key_to_cpu(leaf, &key, slot);
1349                 if (key.objectid != dir->i_ino ||
1350                     key.type != BTRFS_INODE_REF_KEY) {
1351                         goto out_err;
1352                 }
1353                 location->objectid = key.offset;
1354                 location->type = BTRFS_INODE_ITEM_KEY;
1355                 location->offset = 0;
1356                 goto out;
1357         }
1358
1359         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
1360                                     namelen, 0);
1361         if (IS_ERR(di))
1362                 ret = PTR_ERR(di);
1363         if (!di || IS_ERR(di)) {
1364                 goto out_err;
1365         }
1366         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
1367 out:
1368         btrfs_free_path(path);
1369         return ret;
1370 out_err:
1371         location->objectid = 0;
1372         goto out;
1373 }
1374
1375 /*
1376  * when we hit a tree root in a directory, the btrfs part of the inode
1377  * needs to be changed to reflect the root directory of the tree root.  This
1378  * is kind of like crossing a mount point.
1379  */
1380 static int fixup_tree_root_location(struct btrfs_root *root,
1381                              struct btrfs_key *location,
1382                              struct btrfs_root **sub_root,
1383                              struct dentry *dentry)
1384 {
1385         struct btrfs_path *path;
1386         struct btrfs_root_item *ri;
1387
1388         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
1389                 return 0;
1390         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1391                 return 0;
1392
1393         path = btrfs_alloc_path();
1394         BUG_ON(!path);
1395         mutex_lock(&root->fs_info->fs_mutex);
1396
1397         *sub_root = btrfs_read_fs_root(root->fs_info, location,
1398                                         dentry->d_name.name,
1399                                         dentry->d_name.len);
1400         if (IS_ERR(*sub_root))
1401                 return PTR_ERR(*sub_root);
1402
1403         ri = &(*sub_root)->root_item;
1404         location->objectid = btrfs_root_dirid(ri);
1405         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1406         location->offset = 0;
1407
1408         btrfs_free_path(path);
1409         mutex_unlock(&root->fs_info->fs_mutex);
1410         return 0;
1411 }
1412
1413 static int btrfs_init_locked_inode(struct inode *inode, void *p)
1414 {
1415         struct btrfs_iget_args *args = p;
1416         inode->i_ino = args->ino;
1417         BTRFS_I(inode)->root = args->root;
1418         BTRFS_I(inode)->delalloc_bytes = 0;
1419         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1420         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1421                              inode->i_mapping, GFP_NOFS);
1422         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1423                              inode->i_mapping, GFP_NOFS);
1424         atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1425         return 0;
1426 }
1427
1428 static int btrfs_find_actor(struct inode *inode, void *opaque)
1429 {
1430         struct btrfs_iget_args *args = opaque;
1431         return (args->ino == inode->i_ino &&
1432                 args->root == BTRFS_I(inode)->root);
1433 }
1434
1435 struct inode *btrfs_ilookup(struct super_block *s, u64 objectid,
1436                             u64 root_objectid)
1437 {
1438         struct btrfs_iget_args args;
1439         args.ino = objectid;
1440         args.root = btrfs_lookup_fs_root(btrfs_sb(s)->fs_info, root_objectid);
1441
1442         if (!args.root)
1443                 return NULL;
1444
1445         return ilookup5(s, objectid, btrfs_find_actor, (void *)&args);
1446 }
1447
1448 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
1449                                 struct btrfs_root *root)
1450 {
1451         struct inode *inode;
1452         struct btrfs_iget_args args;
1453         args.ino = objectid;
1454         args.root = root;
1455
1456         inode = iget5_locked(s, objectid, btrfs_find_actor,
1457                              btrfs_init_locked_inode,
1458                              (void *)&args);
1459         return inode;
1460 }
1461
1462 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
1463                                    struct nameidata *nd)
1464 {
1465         struct inode * inode;
1466         struct btrfs_inode *bi = BTRFS_I(dir);
1467         struct btrfs_root *root = bi->root;
1468         struct btrfs_root *sub_root = root;
1469         struct btrfs_key location;
1470         int ret;
1471
1472         if (dentry->d_name.len > BTRFS_NAME_LEN)
1473                 return ERR_PTR(-ENAMETOOLONG);
1474
1475         mutex_lock(&root->fs_info->fs_mutex);
1476         ret = btrfs_inode_by_name(dir, dentry, &location);
1477         mutex_unlock(&root->fs_info->fs_mutex);
1478
1479         if (ret < 0)
1480                 return ERR_PTR(ret);
1481
1482         inode = NULL;
1483         if (location.objectid) {
1484                 ret = fixup_tree_root_location(root, &location, &sub_root,
1485                                                 dentry);
1486                 if (ret < 0)
1487                         return ERR_PTR(ret);
1488                 if (ret > 0)
1489                         return ERR_PTR(-ENOENT);
1490                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
1491                                           sub_root);
1492                 if (!inode)
1493                         return ERR_PTR(-EACCES);
1494                 if (inode->i_state & I_NEW) {
1495                         /* the inode and parent dir are two different roots */
1496                         if (sub_root != root) {
1497                                 igrab(inode);
1498                                 sub_root->inode = inode;
1499                         }
1500                         BTRFS_I(inode)->root = sub_root;
1501                         memcpy(&BTRFS_I(inode)->location, &location,
1502                                sizeof(location));
1503                         btrfs_read_locked_inode(inode);
1504                         unlock_new_inode(inode);
1505                 }
1506         }
1507         return d_splice_alias(inode, dentry);
1508 }
1509
1510 static unsigned char btrfs_filetype_table[] = {
1511         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
1512 };
1513
1514 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
1515 {
1516         struct inode *inode = filp->f_dentry->d_inode;
1517         struct btrfs_root *root = BTRFS_I(inode)->root;
1518         struct btrfs_item *item;
1519         struct btrfs_dir_item *di;
1520         struct btrfs_key key;
1521         struct btrfs_key found_key;
1522         struct btrfs_path *path;
1523         int ret;
1524         u32 nritems;
1525         struct extent_buffer *leaf;
1526         int slot;
1527         int advance;
1528         unsigned char d_type;
1529         int over = 0;
1530         u32 di_cur;
1531         u32 di_total;
1532         u32 di_len;
1533         int key_type = BTRFS_DIR_INDEX_KEY;
1534         char tmp_name[32];
1535         char *name_ptr;
1536         int name_len;
1537
1538         /* FIXME, use a real flag for deciding about the key type */
1539         if (root->fs_info->tree_root == root)
1540                 key_type = BTRFS_DIR_ITEM_KEY;
1541
1542         /* special case for "." */
1543         if (filp->f_pos == 0) {
1544                 over = filldir(dirent, ".", 1,
1545                                1, inode->i_ino,
1546                                DT_DIR);
1547                 if (over)
1548                         return 0;
1549                 filp->f_pos = 1;
1550         }
1551
1552         mutex_lock(&root->fs_info->fs_mutex);
1553         key.objectid = inode->i_ino;
1554         path = btrfs_alloc_path();
1555         path->reada = 2;
1556
1557         /* special case for .., just use the back ref */
1558         if (filp->f_pos == 1) {
1559                 btrfs_set_key_type(&key, BTRFS_INODE_REF_KEY);
1560                 key.offset = 0;
1561                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1562                 BUG_ON(ret == 0);
1563                 leaf = path->nodes[0];
1564                 slot = path->slots[0];
1565                 nritems = btrfs_header_nritems(leaf);
1566                 if (slot >= nritems) {
1567                         btrfs_release_path(root, path);
1568                         goto read_dir_items;
1569                 }
1570                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1571                 btrfs_release_path(root, path);
1572                 if (found_key.objectid != key.objectid ||
1573                     found_key.type != BTRFS_INODE_REF_KEY)
1574                         goto read_dir_items;
1575                 over = filldir(dirent, "..", 2,
1576                                2, found_key.offset, DT_DIR);
1577                 if (over)
1578                         goto nopos;
1579                 filp->f_pos = 2;
1580         }
1581
1582 read_dir_items:
1583         btrfs_set_key_type(&key, key_type);
1584         key.offset = filp->f_pos;
1585
1586         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1587         if (ret < 0)
1588                 goto err;
1589         advance = 0;
1590         while(1) {
1591                 leaf = path->nodes[0];
1592                 nritems = btrfs_header_nritems(leaf);
1593                 slot = path->slots[0];
1594                 if (advance || slot >= nritems) {
1595                         if (slot >= nritems -1) {
1596                                 ret = btrfs_next_leaf(root, path);
1597                                 if (ret)
1598                                         break;
1599                                 leaf = path->nodes[0];
1600                                 nritems = btrfs_header_nritems(leaf);
1601                                 slot = path->slots[0];
1602                         } else {
1603                                 slot++;
1604                                 path->slots[0]++;
1605                         }
1606                 }
1607                 advance = 1;
1608                 item = btrfs_item_nr(leaf, slot);
1609                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1610
1611                 if (found_key.objectid != key.objectid)
1612                         break;
1613                 if (btrfs_key_type(&found_key) != key_type)
1614                         break;
1615                 if (found_key.offset < filp->f_pos)
1616                         continue;
1617
1618                 filp->f_pos = found_key.offset;
1619                 advance = 1;
1620                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
1621                 di_cur = 0;
1622                 di_total = btrfs_item_size(leaf, item);
1623                 while(di_cur < di_total) {
1624                         struct btrfs_key location;
1625
1626                         name_len = btrfs_dir_name_len(leaf, di);
1627                         if (name_len < 32) {
1628                                 name_ptr = tmp_name;
1629                         } else {
1630                                 name_ptr = kmalloc(name_len, GFP_NOFS);
1631                                 BUG_ON(!name_ptr);
1632                         }
1633                         read_extent_buffer(leaf, name_ptr,
1634                                            (unsigned long)(di + 1), name_len);
1635
1636                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
1637                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
1638                         over = filldir(dirent, name_ptr, name_len,
1639                                        found_key.offset,
1640                                        location.objectid,
1641                                        d_type);
1642
1643                         if (name_ptr != tmp_name)
1644                                 kfree(name_ptr);
1645
1646                         if (over)
1647                                 goto nopos;
1648                         di_len = btrfs_dir_name_len(leaf, di) +
1649                                 btrfs_dir_data_len(leaf, di) +sizeof(*di);
1650                         di_cur += di_len;
1651                         di = (struct btrfs_dir_item *)((char *)di + di_len);
1652                 }
1653         }
1654         if (key_type == BTRFS_DIR_INDEX_KEY)
1655                 filp->f_pos = INT_LIMIT(typeof(filp->f_pos));
1656         else
1657                 filp->f_pos++;
1658 nopos:
1659         ret = 0;
1660 err:
1661         btrfs_release_path(root, path);
1662         btrfs_free_path(path);
1663         mutex_unlock(&root->fs_info->fs_mutex);
1664         return ret;
1665 }
1666
1667 int btrfs_write_inode(struct inode *inode, int wait)
1668 {
1669         struct btrfs_root *root = BTRFS_I(inode)->root;
1670         struct btrfs_trans_handle *trans;
1671         int ret = 0;
1672
1673         if (wait) {
1674                 mutex_lock(&root->fs_info->fs_mutex);
1675                 trans = btrfs_start_transaction(root, 1);
1676                 btrfs_set_trans_block_group(trans, inode);
1677                 ret = btrfs_commit_transaction(trans, root);
1678                 mutex_unlock(&root->fs_info->fs_mutex);
1679         }
1680         return ret;
1681 }
1682
1683 /*
1684  * This is somewhat expensive, updating the tree every time the
1685  * inode changes.  But, it is most likely to find the inode in cache.
1686  * FIXME, needs more benchmarking...there are no reasons other than performance
1687  * to keep or drop this code.
1688  */
1689 void btrfs_dirty_inode(struct inode *inode)
1690 {
1691         struct btrfs_root *root = BTRFS_I(inode)->root;
1692         struct btrfs_trans_handle *trans;
1693
1694         mutex_lock(&root->fs_info->fs_mutex);
1695         trans = btrfs_start_transaction(root, 1);
1696         btrfs_set_trans_block_group(trans, inode);
1697         btrfs_update_inode(trans, root, inode);
1698         btrfs_end_transaction(trans, root);
1699         mutex_unlock(&root->fs_info->fs_mutex);
1700 }
1701
1702 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
1703                                      struct btrfs_root *root,
1704                                      const char *name, int name_len,
1705                                      u64 ref_objectid,
1706                                      u64 objectid,
1707                                      struct btrfs_block_group_cache *group,
1708                                      int mode)
1709 {
1710         struct inode *inode;
1711         struct btrfs_inode_item *inode_item;
1712         struct btrfs_block_group_cache *new_inode_group;
1713         struct btrfs_key *location;
1714         struct btrfs_path *path;
1715         struct btrfs_inode_ref *ref;
1716         struct btrfs_key key[2];
1717         u32 sizes[2];
1718         unsigned long ptr;
1719         int ret;
1720         int owner;
1721
1722         path = btrfs_alloc_path();
1723         BUG_ON(!path);
1724
1725         inode = new_inode(root->fs_info->sb);
1726         if (!inode)
1727                 return ERR_PTR(-ENOMEM);
1728
1729         extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1730         extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1731                              inode->i_mapping, GFP_NOFS);
1732         extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1733                              inode->i_mapping, GFP_NOFS);
1734         atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1735         BTRFS_I(inode)->delalloc_bytes = 0;
1736         BTRFS_I(inode)->root = root;
1737
1738         if (mode & S_IFDIR)
1739                 owner = 0;
1740         else
1741                 owner = 1;
1742         new_inode_group = btrfs_find_block_group(root, group, 0,
1743                                        BTRFS_BLOCK_GROUP_METADATA, owner);
1744         if (!new_inode_group) {
1745                 printk("find_block group failed\n");
1746                 new_inode_group = group;
1747         }
1748         BTRFS_I(inode)->block_group = new_inode_group;
1749         BTRFS_I(inode)->flags = 0;
1750
1751         key[0].objectid = objectid;
1752         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
1753         key[0].offset = 0;
1754
1755         key[1].objectid = objectid;
1756         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
1757         key[1].offset = ref_objectid;
1758
1759         sizes[0] = sizeof(struct btrfs_inode_item);
1760         sizes[1] = name_len + sizeof(*ref);
1761
1762         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
1763         if (ret != 0)
1764                 goto fail;
1765
1766         if (objectid > root->highest_inode)
1767                 root->highest_inode = objectid;
1768
1769         inode->i_uid = current->fsuid;
1770         inode->i_gid = current->fsgid;
1771         inode->i_mode = mode;
1772         inode->i_ino = objectid;
1773         inode->i_blocks = 0;
1774         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
1775         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
1776                                   struct btrfs_inode_item);
1777         fill_inode_item(path->nodes[0], inode_item, inode);
1778
1779         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
1780                              struct btrfs_inode_ref);
1781         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
1782         ptr = (unsigned long)(ref + 1);
1783         write_extent_buffer(path->nodes[0], name, ptr, name_len);
1784
1785         btrfs_mark_buffer_dirty(path->nodes[0]);
1786         btrfs_free_path(path);
1787
1788         location = &BTRFS_I(inode)->location;
1789         location->objectid = objectid;
1790         location->offset = 0;
1791         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
1792
1793         insert_inode_hash(inode);
1794         return inode;
1795 fail:
1796         btrfs_free_path(path);
1797         return ERR_PTR(ret);
1798 }
1799
1800 static inline u8 btrfs_inode_type(struct inode *inode)
1801 {
1802         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
1803 }
1804
1805 static int btrfs_add_link(struct btrfs_trans_handle *trans,
1806                             struct dentry *dentry, struct inode *inode,
1807                             int add_backref)
1808 {
1809         int ret;
1810         struct btrfs_key key;
1811         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
1812         struct inode *parent_inode;
1813
1814         key.objectid = inode->i_ino;
1815         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1816         key.offset = 0;
1817
1818         ret = btrfs_insert_dir_item(trans, root,
1819                                     dentry->d_name.name, dentry->d_name.len,
1820                                     dentry->d_parent->d_inode->i_ino,
1821                                     &key, btrfs_inode_type(inode));
1822         if (ret == 0) {
1823                 if (add_backref) {
1824                         ret = btrfs_insert_inode_ref(trans, root,
1825                                              dentry->d_name.name,
1826                                              dentry->d_name.len,
1827                                              inode->i_ino,
1828                                              dentry->d_parent->d_inode->i_ino);
1829                 }
1830                 parent_inode = dentry->d_parent->d_inode;
1831                 parent_inode->i_size += dentry->d_name.len * 2;
1832                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
1833                 ret = btrfs_update_inode(trans, root,
1834                                          dentry->d_parent->d_inode);
1835         }
1836         return ret;
1837 }
1838
1839 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
1840                             struct dentry *dentry, struct inode *inode,
1841                             int backref)
1842 {
1843         int err = btrfs_add_link(trans, dentry, inode, backref);
1844         if (!err) {
1845                 d_instantiate(dentry, inode);
1846                 return 0;
1847         }
1848         if (err > 0)
1849                 err = -EEXIST;
1850         return err;
1851 }
1852
1853 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
1854                         int mode, dev_t rdev)
1855 {
1856         struct btrfs_trans_handle *trans;
1857         struct btrfs_root *root = BTRFS_I(dir)->root;
1858         struct inode *inode = NULL;
1859         int err;
1860         int drop_inode = 0;
1861         u64 objectid;
1862         unsigned long nr = 0;
1863
1864         if (!new_valid_dev(rdev))
1865                 return -EINVAL;
1866
1867         mutex_lock(&root->fs_info->fs_mutex);
1868         err = btrfs_check_free_space(root, 1, 0);
1869         if (err)
1870                 goto fail;
1871
1872         trans = btrfs_start_transaction(root, 1);
1873         btrfs_set_trans_block_group(trans, dir);
1874
1875         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1876         if (err) {
1877                 err = -ENOSPC;
1878                 goto out_unlock;
1879         }
1880
1881         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1882                                 dentry->d_name.len,
1883                                 dentry->d_parent->d_inode->i_ino, objectid,
1884                                 BTRFS_I(dir)->block_group, mode);
1885         err = PTR_ERR(inode);
1886         if (IS_ERR(inode))
1887                 goto out_unlock;
1888
1889         btrfs_set_trans_block_group(trans, inode);
1890         err = btrfs_add_nondir(trans, dentry, inode, 0);
1891         if (err)
1892                 drop_inode = 1;
1893         else {
1894                 inode->i_op = &btrfs_special_inode_operations;
1895                 init_special_inode(inode, inode->i_mode, rdev);
1896                 btrfs_update_inode(trans, root, inode);
1897         }
1898         dir->i_sb->s_dirt = 1;
1899         btrfs_update_inode_block_group(trans, inode);
1900         btrfs_update_inode_block_group(trans, dir);
1901 out_unlock:
1902         nr = trans->blocks_used;
1903         btrfs_end_transaction(trans, root);
1904 fail:
1905         mutex_unlock(&root->fs_info->fs_mutex);
1906
1907         if (drop_inode) {
1908                 inode_dec_link_count(inode);
1909                 iput(inode);
1910         }
1911         btrfs_btree_balance_dirty(root, nr);
1912         btrfs_throttle(root);
1913         return err;
1914 }
1915
1916 static int btrfs_create(struct inode *dir, struct dentry *dentry,
1917                         int mode, struct nameidata *nd)
1918 {
1919         struct btrfs_trans_handle *trans;
1920         struct btrfs_root *root = BTRFS_I(dir)->root;
1921         struct inode *inode = NULL;
1922         int err;
1923         int drop_inode = 0;
1924         unsigned long nr = 0;
1925         u64 objectid;
1926
1927         mutex_lock(&root->fs_info->fs_mutex);
1928         err = btrfs_check_free_space(root, 1, 0);
1929         if (err)
1930                 goto fail;
1931         trans = btrfs_start_transaction(root, 1);
1932         btrfs_set_trans_block_group(trans, dir);
1933
1934         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
1935         if (err) {
1936                 err = -ENOSPC;
1937                 goto out_unlock;
1938         }
1939
1940         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
1941                                 dentry->d_name.len,
1942                                 dentry->d_parent->d_inode->i_ino,
1943                                 objectid, BTRFS_I(dir)->block_group, mode);
1944         err = PTR_ERR(inode);
1945         if (IS_ERR(inode))
1946                 goto out_unlock;
1947
1948         btrfs_set_trans_block_group(trans, inode);
1949         err = btrfs_add_nondir(trans, dentry, inode, 0);
1950         if (err)
1951                 drop_inode = 1;
1952         else {
1953                 inode->i_mapping->a_ops = &btrfs_aops;
1954                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
1955                 inode->i_fop = &btrfs_file_operations;
1956                 inode->i_op = &btrfs_file_inode_operations;
1957                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
1958                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
1959                                      inode->i_mapping, GFP_NOFS);
1960                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
1961                                      inode->i_mapping, GFP_NOFS);
1962                 BTRFS_I(inode)->delalloc_bytes = 0;
1963                 atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
1964                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
1965         }
1966         dir->i_sb->s_dirt = 1;
1967         btrfs_update_inode_block_group(trans, inode);
1968         btrfs_update_inode_block_group(trans, dir);
1969 out_unlock:
1970         nr = trans->blocks_used;
1971         btrfs_end_transaction(trans, root);
1972 fail:
1973         mutex_unlock(&root->fs_info->fs_mutex);
1974
1975         if (drop_inode) {
1976                 inode_dec_link_count(inode);
1977                 iput(inode);
1978         }
1979         btrfs_btree_balance_dirty(root, nr);
1980         btrfs_throttle(root);
1981         return err;
1982 }
1983
1984 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
1985                       struct dentry *dentry)
1986 {
1987         struct btrfs_trans_handle *trans;
1988         struct btrfs_root *root = BTRFS_I(dir)->root;
1989         struct inode *inode = old_dentry->d_inode;
1990         unsigned long nr = 0;
1991         int err;
1992         int drop_inode = 0;
1993
1994         if (inode->i_nlink == 0)
1995                 return -ENOENT;
1996
1997 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
1998         inode->i_nlink++;
1999 #else
2000         inc_nlink(inode);
2001 #endif
2002         mutex_lock(&root->fs_info->fs_mutex);
2003         err = btrfs_check_free_space(root, 1, 0);
2004         if (err)
2005                 goto fail;
2006         trans = btrfs_start_transaction(root, 1);
2007
2008         btrfs_set_trans_block_group(trans, dir);
2009         atomic_inc(&inode->i_count);
2010         err = btrfs_add_nondir(trans, dentry, inode, 1);
2011
2012         if (err)
2013                 drop_inode = 1;
2014
2015         dir->i_sb->s_dirt = 1;
2016         btrfs_update_inode_block_group(trans, dir);
2017         err = btrfs_update_inode(trans, root, inode);
2018
2019         if (err)
2020                 drop_inode = 1;
2021
2022         nr = trans->blocks_used;
2023         btrfs_end_transaction(trans, root);
2024 fail:
2025         mutex_unlock(&root->fs_info->fs_mutex);
2026
2027         if (drop_inode) {
2028                 inode_dec_link_count(inode);
2029                 iput(inode);
2030         }
2031         btrfs_btree_balance_dirty(root, nr);
2032         btrfs_throttle(root);
2033         return err;
2034 }
2035
2036 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2037 {
2038         struct inode *inode = NULL;
2039         struct btrfs_trans_handle *trans;
2040         struct btrfs_root *root = BTRFS_I(dir)->root;
2041         int err = 0;
2042         int drop_on_err = 0;
2043         u64 objectid = 0;
2044         unsigned long nr = 1;
2045
2046         mutex_lock(&root->fs_info->fs_mutex);
2047         err = btrfs_check_free_space(root, 1, 0);
2048         if (err)
2049                 goto out_unlock;
2050
2051         trans = btrfs_start_transaction(root, 1);
2052         btrfs_set_trans_block_group(trans, dir);
2053
2054         if (IS_ERR(trans)) {
2055                 err = PTR_ERR(trans);
2056                 goto out_unlock;
2057         }
2058
2059         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
2060         if (err) {
2061                 err = -ENOSPC;
2062                 goto out_unlock;
2063         }
2064
2065         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
2066                                 dentry->d_name.len,
2067                                 dentry->d_parent->d_inode->i_ino, objectid,
2068                                 BTRFS_I(dir)->block_group, S_IFDIR | mode);
2069         if (IS_ERR(inode)) {
2070                 err = PTR_ERR(inode);
2071                 goto out_fail;
2072         }
2073
2074         drop_on_err = 1;
2075         inode->i_op = &btrfs_dir_inode_operations;
2076         inode->i_fop = &btrfs_dir_file_operations;
2077         btrfs_set_trans_block_group(trans, inode);
2078
2079         inode->i_size = 0;
2080         err = btrfs_update_inode(trans, root, inode);
2081         if (err)
2082                 goto out_fail;
2083
2084         err = btrfs_add_link(trans, dentry, inode, 0);
2085         if (err)
2086                 goto out_fail;
2087
2088         d_instantiate(dentry, inode);
2089         drop_on_err = 0;
2090         dir->i_sb->s_dirt = 1;
2091         btrfs_update_inode_block_group(trans, inode);
2092         btrfs_update_inode_block_group(trans, dir);
2093
2094 out_fail:
2095         nr = trans->blocks_used;
2096         btrfs_end_transaction(trans, root);
2097
2098 out_unlock:
2099         mutex_unlock(&root->fs_info->fs_mutex);
2100         if (drop_on_err)
2101                 iput(inode);
2102         btrfs_btree_balance_dirty(root, nr);
2103         btrfs_throttle(root);
2104         return err;
2105 }
2106
2107 static int merge_extent_mapping(struct extent_map_tree *em_tree,
2108                                 struct extent_map *existing,
2109                                 struct extent_map *em)
2110 {
2111         u64 start_diff;
2112         u64 new_end;
2113         int ret = 0;
2114         int real_blocks = existing->block_start < EXTENT_MAP_LAST_BYTE;
2115
2116         if (real_blocks && em->block_start >= EXTENT_MAP_LAST_BYTE)
2117                 goto invalid;
2118
2119         if (!real_blocks && em->block_start != existing->block_start)
2120                 goto invalid;
2121
2122         new_end = max(existing->start + existing->len, em->start + em->len);
2123
2124         if (existing->start >= em->start) {
2125                 if (em->start + em->len < existing->start)
2126                         goto invalid;
2127
2128                 start_diff = existing->start - em->start;
2129                 if (real_blocks && em->block_start + start_diff !=
2130                     existing->block_start)
2131                         goto invalid;
2132
2133                 em->len = new_end - em->start;
2134
2135                 remove_extent_mapping(em_tree, existing);
2136                 /* free for the tree */
2137                 free_extent_map(existing);
2138                 ret = add_extent_mapping(em_tree, em);
2139
2140         } else if (em->start > existing->start) {
2141
2142                 if (existing->start + existing->len < em->start)
2143                         goto invalid;
2144
2145                 start_diff = em->start - existing->start;
2146                 if (real_blocks && existing->block_start + start_diff !=
2147                     em->block_start)
2148                         goto invalid;
2149
2150                 remove_extent_mapping(em_tree, existing);
2151                 em->block_start = existing->block_start;
2152                 em->start = existing->start;
2153                 em->len = new_end - existing->start;
2154                 free_extent_map(existing);
2155
2156                 ret = add_extent_mapping(em_tree, em);
2157         } else {
2158                 goto invalid;
2159         }
2160         return ret;
2161
2162 invalid:
2163         printk("invalid extent map merge [%Lu %Lu %Lu] [%Lu %Lu %Lu]\n",
2164                existing->start, existing->len, existing->block_start,
2165                em->start, em->len, em->block_start);
2166         return -EIO;
2167 }
2168
2169 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
2170                                     size_t pg_offset, u64 start, u64 len,
2171                                     int create)
2172 {
2173         int ret;
2174         int err = 0;
2175         u64 bytenr;
2176         u64 extent_start = 0;
2177         u64 extent_end = 0;
2178         u64 objectid = inode->i_ino;
2179         u32 found_type;
2180         struct btrfs_path *path;
2181         struct btrfs_root *root = BTRFS_I(inode)->root;
2182         struct btrfs_file_extent_item *item;
2183         struct extent_buffer *leaf;
2184         struct btrfs_key found_key;
2185         struct extent_map *em = NULL;
2186         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
2187         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2188         struct btrfs_trans_handle *trans = NULL;
2189
2190         path = btrfs_alloc_path();
2191         BUG_ON(!path);
2192         mutex_lock(&root->fs_info->fs_mutex);
2193
2194 again:
2195         spin_lock(&em_tree->lock);
2196         em = lookup_extent_mapping(em_tree, start, len);
2197         if (em)
2198                 em->bdev = root->fs_info->fs_devices->latest_bdev;
2199         spin_unlock(&em_tree->lock);
2200
2201         if (em) {
2202                 if (em->start > start || em->start + em->len <= start)
2203                         free_extent_map(em);
2204                 else if (em->block_start == EXTENT_MAP_INLINE && page)
2205                         free_extent_map(em);
2206                 else
2207                         goto out;
2208         }
2209         em = alloc_extent_map(GFP_NOFS);
2210         if (!em) {
2211                 err = -ENOMEM;
2212                 goto out;
2213         }
2214
2215         em->start = EXTENT_MAP_HOLE;
2216         em->len = (u64)-1;
2217         em->bdev = root->fs_info->fs_devices->latest_bdev;
2218         ret = btrfs_lookup_file_extent(trans, root, path,
2219                                        objectid, start, trans != NULL);
2220         if (ret < 0) {
2221                 err = ret;
2222                 goto out;
2223         }
2224
2225         if (ret != 0) {
2226                 if (path->slots[0] == 0)
2227                         goto not_found;
2228                 path->slots[0]--;
2229         }
2230
2231         leaf = path->nodes[0];
2232         item = btrfs_item_ptr(leaf, path->slots[0],
2233                               struct btrfs_file_extent_item);
2234         /* are we inside the extent that was found? */
2235         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2236         found_type = btrfs_key_type(&found_key);
2237         if (found_key.objectid != objectid ||
2238             found_type != BTRFS_EXTENT_DATA_KEY) {
2239                 goto not_found;
2240         }
2241
2242         found_type = btrfs_file_extent_type(leaf, item);
2243         extent_start = found_key.offset;
2244         if (found_type == BTRFS_FILE_EXTENT_REG) {
2245                 extent_end = extent_start +
2246                        btrfs_file_extent_num_bytes(leaf, item);
2247                 err = 0;
2248                 if (start < extent_start || start >= extent_end) {
2249                         em->start = start;
2250                         if (start < extent_start) {
2251                                 if (start + len <= extent_start)
2252                                         goto not_found;
2253                                 em->len = extent_end - extent_start;
2254                         } else {
2255                                 em->len = len;
2256                         }
2257                         goto not_found_em;
2258                 }
2259                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
2260                 if (bytenr == 0) {
2261                         em->start = extent_start;
2262                         em->len = extent_end - extent_start;
2263                         em->block_start = EXTENT_MAP_HOLE;
2264                         goto insert;
2265                 }
2266                 bytenr += btrfs_file_extent_offset(leaf, item);
2267                 em->block_start = bytenr;
2268                 em->start = extent_start;
2269                 em->len = extent_end - extent_start;
2270                 goto insert;
2271         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
2272                 u64 page_start;
2273                 unsigned long ptr;
2274                 char *map;
2275                 size_t size;
2276                 size_t extent_offset;
2277                 size_t copy_size;
2278
2279                 size = btrfs_file_extent_inline_len(leaf, btrfs_item_nr(leaf,
2280                                                     path->slots[0]));
2281                 extent_end = (extent_start + size + root->sectorsize - 1) &
2282                         ~((u64)root->sectorsize - 1);
2283                 if (start < extent_start || start >= extent_end) {
2284                         em->start = start;
2285                         if (start < extent_start) {
2286                                 if (start + len <= extent_start)
2287                                         goto not_found;
2288                                 em->len = extent_end - extent_start;
2289                         } else {
2290                                 em->len = len;
2291                         }
2292                         goto not_found_em;
2293                 }
2294                 em->block_start = EXTENT_MAP_INLINE;
2295
2296                 if (!page) {
2297                         em->start = extent_start;
2298                         em->len = size;
2299                         goto out;
2300                 }
2301
2302                 page_start = page_offset(page) + pg_offset;
2303                 extent_offset = page_start - extent_start;
2304                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
2305                                 size - extent_offset);
2306                 em->start = extent_start + extent_offset;
2307                 em->len = (copy_size + root->sectorsize - 1) &
2308                         ~((u64)root->sectorsize - 1);
2309                 map = kmap(page);
2310                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
2311                 if (create == 0 && !PageUptodate(page)) {
2312                         read_extent_buffer(leaf, map + pg_offset, ptr,
2313                                            copy_size);
2314                         flush_dcache_page(page);
2315                 } else if (create && PageUptodate(page)) {
2316                         if (!trans) {
2317                                 kunmap(page);
2318                                 free_extent_map(em);
2319                                 em = NULL;
2320                                 btrfs_release_path(root, path);
2321                                 trans = btrfs_start_transaction(root, 1);
2322                                 goto again;
2323                         }
2324                         write_extent_buffer(leaf, map + pg_offset, ptr,
2325                                             copy_size);
2326                         btrfs_mark_buffer_dirty(leaf);
2327                 }
2328                 kunmap(page);
2329                 set_extent_uptodate(io_tree, em->start,
2330                                     extent_map_end(em) - 1, GFP_NOFS);
2331                 goto insert;
2332         } else {
2333                 printk("unkknown found_type %d\n", found_type);
2334                 WARN_ON(1);
2335         }
2336 not_found:
2337         em->start = start;
2338         em->len = len;
2339 not_found_em:
2340         em->block_start = EXTENT_MAP_HOLE;
2341 insert:
2342         btrfs_release_path(root, path);
2343         if (em->start > start || extent_map_end(em) <= start) {
2344                 printk("bad extent! em: [%Lu %Lu] passed [%Lu %Lu]\n", em->start, em->len, start, len);
2345                 err = -EIO;
2346                 goto out;
2347         }
2348
2349         err = 0;
2350         spin_lock(&em_tree->lock);
2351         ret = add_extent_mapping(em_tree, em);
2352         /* it is possible that someone inserted the extent into the tree
2353          * while we had the lock dropped.  It is also possible that
2354          * an overlapping map exists in the tree
2355          */
2356         if (ret == -EEXIST) {
2357                 struct extent_map *existing;
2358                 existing = lookup_extent_mapping(em_tree, start, len);
2359                 if (existing && (existing->start > start ||
2360                     existing->start + existing->len <= start)) {
2361                         free_extent_map(existing);
2362                         existing = NULL;
2363                 }
2364                 if (!existing) {
2365                         existing = lookup_extent_mapping(em_tree, em->start,
2366                                                          em->len);
2367                         if (existing) {
2368                                 err = merge_extent_mapping(em_tree, existing,
2369                                                            em);
2370                                 free_extent_map(existing);
2371                                 if (err) {
2372                                         free_extent_map(em);
2373                                         em = NULL;
2374                                 }
2375                         } else {
2376                                 err = -EIO;
2377                                 printk("failing to insert %Lu %Lu\n",
2378                                        start, len);
2379                                 free_extent_map(em);
2380                                 em = NULL;
2381                         }
2382                 } else {
2383                         free_extent_map(em);
2384                         em = existing;
2385                 }
2386         }
2387         spin_unlock(&em_tree->lock);
2388 out:
2389         btrfs_free_path(path);
2390         if (trans) {
2391                 ret = btrfs_end_transaction(trans, root);
2392                 if (!err)
2393                         err = ret;
2394         }
2395         mutex_unlock(&root->fs_info->fs_mutex);
2396         if (err) {
2397                 free_extent_map(em);
2398                 WARN_ON(1);
2399                 return ERR_PTR(err);
2400         }
2401         return em;
2402 }
2403
2404 #if 0 /* waiting for O_DIRECT reads */
2405 static int btrfs_get_block(struct inode *inode, sector_t iblock,
2406                         struct buffer_head *bh_result, int create)
2407 {
2408         struct extent_map *em;
2409         u64 start = (u64)iblock << inode->i_blkbits;
2410         struct btrfs_multi_bio *multi = NULL;
2411         struct btrfs_root *root = BTRFS_I(inode)->root;
2412         u64 len;
2413         u64 logical;
2414         u64 map_length;
2415         int ret = 0;
2416
2417         em = btrfs_get_extent(inode, NULL, 0, start, bh_result->b_size, 0);
2418
2419         if (!em || IS_ERR(em))
2420                 goto out;
2421
2422         if (em->start > start || em->start + em->len <= start) {
2423             goto out;
2424         }
2425
2426         if (em->block_start == EXTENT_MAP_INLINE) {
2427                 ret = -EINVAL;
2428                 goto out;
2429         }
2430
2431         len = em->start + em->len - start;
2432         len = min_t(u64, len, INT_LIMIT(typeof(bh_result->b_size)));
2433
2434         if (em->block_start == EXTENT_MAP_HOLE ||
2435             em->block_start == EXTENT_MAP_DELALLOC) {
2436                 bh_result->b_size = len;
2437                 goto out;
2438         }
2439
2440         logical = start - em->start;
2441         logical = em->block_start + logical;
2442
2443         map_length = len;
2444         ret = btrfs_map_block(&root->fs_info->mapping_tree, READ,
2445                               logical, &map_length, &multi, 0);
2446         BUG_ON(ret);
2447         bh_result->b_blocknr = multi->stripes[0].physical >> inode->i_blkbits;
2448         bh_result->b_size = min(map_length, len);
2449
2450         bh_result->b_bdev = multi->stripes[0].dev->bdev;
2451         set_buffer_mapped(bh_result);
2452         kfree(multi);
2453 out:
2454         free_extent_map(em);
2455         return ret;
2456 }
2457 #endif
2458
2459 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
2460                         const struct iovec *iov, loff_t offset,
2461                         unsigned long nr_segs)
2462 {
2463         return -EINVAL;
2464 #if 0
2465         struct file *file = iocb->ki_filp;
2466         struct inode *inode = file->f_mapping->host;
2467
2468         if (rw == WRITE)
2469                 return -EINVAL;
2470
2471         return blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
2472                                   offset, nr_segs, btrfs_get_block, NULL);
2473 #endif
2474 }
2475
2476 static sector_t btrfs_bmap(struct address_space *mapping, sector_t iblock)
2477 {
2478         return extent_bmap(mapping, iblock, btrfs_get_extent);
2479 }
2480
2481 int btrfs_readpage(struct file *file, struct page *page)
2482 {
2483         struct extent_io_tree *tree;
2484         tree = &BTRFS_I(page->mapping->host)->io_tree;
2485         return extent_read_full_page(tree, page, btrfs_get_extent);
2486 }
2487
2488 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
2489 {
2490         struct extent_io_tree *tree;
2491
2492
2493         if (current->flags & PF_MEMALLOC) {
2494                 redirty_page_for_writepage(wbc, page);
2495                 unlock_page(page);
2496                 return 0;
2497         }
2498         tree = &BTRFS_I(page->mapping->host)->io_tree;
2499         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
2500 }
2501
2502 static int btrfs_writepages(struct address_space *mapping,
2503                             struct writeback_control *wbc)
2504 {
2505         struct extent_io_tree *tree;
2506         tree = &BTRFS_I(mapping->host)->io_tree;
2507         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
2508 }
2509
2510 static int
2511 btrfs_readpages(struct file *file, struct address_space *mapping,
2512                 struct list_head *pages, unsigned nr_pages)
2513 {
2514         struct extent_io_tree *tree;
2515         tree = &BTRFS_I(mapping->host)->io_tree;
2516         return extent_readpages(tree, mapping, pages, nr_pages,
2517                                 btrfs_get_extent);
2518 }
2519
2520 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
2521 {
2522         struct extent_io_tree *tree;
2523         struct extent_map_tree *map;
2524         int ret;
2525
2526         tree = &BTRFS_I(page->mapping->host)->io_tree;
2527         map = &BTRFS_I(page->mapping->host)->extent_tree;
2528         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
2529         if (ret == 1) {
2530                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
2531                 ClearPagePrivate(page);
2532                 set_page_private(page, 0);
2533                 page_cache_release(page);
2534         }
2535         return ret;
2536 }
2537
2538 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
2539 {
2540         struct extent_io_tree *tree;
2541
2542         tree = &BTRFS_I(page->mapping->host)->io_tree;
2543         extent_invalidatepage(tree, page, offset);
2544         btrfs_releasepage(page, GFP_NOFS);
2545         if (PagePrivate(page)) {
2546                 invalidate_extent_lru(tree, page_offset(page), PAGE_CACHE_SIZE);
2547                 ClearPagePrivate(page);
2548                 set_page_private(page, 0);
2549                 page_cache_release(page);
2550         }
2551 }
2552
2553 /*
2554  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
2555  * called from a page fault handler when a page is first dirtied. Hence we must
2556  * be careful to check for EOF conditions here. We set the page up correctly
2557  * for a written page which means we get ENOSPC checking when writing into
2558  * holes and correct delalloc and unwritten extent mapping on filesystems that
2559  * support these features.
2560  *
2561  * We are not allowed to take the i_mutex here so we have to play games to
2562  * protect against truncate races as the page could now be beyond EOF.  Because
2563  * vmtruncate() writes the inode size before removing pages, once we have the
2564  * page lock we can determine safely if the page is beyond EOF. If it is not
2565  * beyond EOF, then the page is guaranteed safe against truncation until we
2566  * unlock the page.
2567  */
2568 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
2569 {
2570         struct inode *inode = fdentry(vma->vm_file)->d_inode;
2571         struct btrfs_root *root = BTRFS_I(inode)->root;
2572         unsigned long end;
2573         loff_t size;
2574         int ret;
2575         u64 page_start;
2576
2577         mutex_lock(&root->fs_info->fs_mutex);
2578         ret = btrfs_check_free_space(root, PAGE_CACHE_SIZE, 0);
2579         mutex_unlock(&root->fs_info->fs_mutex);
2580         if (ret)
2581                 goto out;
2582
2583         ret = -EINVAL;
2584
2585         lock_page(page);
2586         wait_on_page_writeback(page);
2587         size = i_size_read(inode);
2588         page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2589
2590         if ((page->mapping != inode->i_mapping) ||
2591             (page_start > size)) {
2592                 /* page got truncated out from underneath us */
2593                 goto out_unlock;
2594         }
2595
2596         /* page is wholly or partially inside EOF */
2597         if (page_start + PAGE_CACHE_SIZE > size)
2598                 end = size & ~PAGE_CACHE_MASK;
2599         else
2600                 end = PAGE_CACHE_SIZE;
2601
2602         ret = btrfs_cow_one_page(inode, page, end);
2603
2604 out_unlock:
2605         unlock_page(page);
2606 out:
2607         return ret;
2608 }
2609
2610 static void btrfs_truncate(struct inode *inode)
2611 {
2612         struct btrfs_root *root = BTRFS_I(inode)->root;
2613         int ret;
2614         struct btrfs_trans_handle *trans;
2615         unsigned long nr;
2616
2617         if (!S_ISREG(inode->i_mode))
2618                 return;
2619         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2620                 return;
2621
2622         btrfs_truncate_page(inode->i_mapping, inode->i_size);
2623
2624         mutex_lock(&root->fs_info->fs_mutex);
2625         trans = btrfs_start_transaction(root, 1);
2626         btrfs_set_trans_block_group(trans, inode);
2627
2628         /* FIXME, add redo link to tree so we don't leak on crash */
2629         ret = btrfs_truncate_in_trans(trans, root, inode,
2630                                       BTRFS_EXTENT_DATA_KEY);
2631         btrfs_update_inode(trans, root, inode);
2632         nr = trans->blocks_used;
2633
2634         ret = btrfs_end_transaction(trans, root);
2635         BUG_ON(ret);
2636         mutex_unlock(&root->fs_info->fs_mutex);
2637         btrfs_btree_balance_dirty(root, nr);
2638         btrfs_throttle(root);
2639 }
2640
2641 static int noinline create_subvol(struct btrfs_root *root, char *name,
2642                                   int namelen)
2643 {
2644         struct btrfs_trans_handle *trans;
2645         struct btrfs_key key;
2646         struct btrfs_root_item root_item;
2647         struct btrfs_inode_item *inode_item;
2648         struct extent_buffer *leaf;
2649         struct btrfs_root *new_root = root;
2650         struct inode *inode;
2651         struct inode *dir;
2652         int ret;
2653         int err;
2654         u64 objectid;
2655         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2656         unsigned long nr = 1;
2657
2658         mutex_lock(&root->fs_info->fs_mutex);
2659         ret = btrfs_check_free_space(root, 1, 0);
2660         if (ret)
2661                 goto fail_commit;
2662
2663         trans = btrfs_start_transaction(root, 1);
2664         BUG_ON(!trans);
2665
2666         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2667                                        0, &objectid);
2668         if (ret)
2669                 goto fail;
2670
2671         leaf = __btrfs_alloc_free_block(trans, root, root->leafsize,
2672                                         objectid, trans->transid, 0, 0,
2673                                         0, 0);
2674         if (IS_ERR(leaf))
2675                 return PTR_ERR(leaf);
2676
2677         btrfs_set_header_nritems(leaf, 0);
2678         btrfs_set_header_level(leaf, 0);
2679         btrfs_set_header_bytenr(leaf, leaf->start);
2680         btrfs_set_header_generation(leaf, trans->transid);
2681         btrfs_set_header_owner(leaf, objectid);
2682
2683         write_extent_buffer(leaf, root->fs_info->fsid,
2684                             (unsigned long)btrfs_header_fsid(leaf),
2685                             BTRFS_FSID_SIZE);
2686         btrfs_mark_buffer_dirty(leaf);
2687
2688         inode_item = &root_item.inode;
2689         memset(inode_item, 0, sizeof(*inode_item));
2690         inode_item->generation = cpu_to_le64(1);
2691         inode_item->size = cpu_to_le64(3);
2692         inode_item->nlink = cpu_to_le32(1);
2693         inode_item->nblocks = cpu_to_le64(1);
2694         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
2695
2696         btrfs_set_root_bytenr(&root_item, leaf->start);
2697         btrfs_set_root_level(&root_item, 0);
2698         btrfs_set_root_refs(&root_item, 1);
2699         btrfs_set_root_used(&root_item, 0);
2700
2701         memset(&root_item.drop_progress, 0, sizeof(root_item.drop_progress));
2702         root_item.drop_level = 0;
2703
2704         free_extent_buffer(leaf);
2705         leaf = NULL;
2706
2707         btrfs_set_root_dirid(&root_item, new_dirid);
2708
2709         key.objectid = objectid;
2710         key.offset = 1;
2711         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2712         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2713                                 &root_item);
2714         if (ret)
2715                 goto fail;
2716
2717         /*
2718          * insert the directory item
2719          */
2720         key.offset = (u64)-1;
2721         dir = root->fs_info->sb->s_root->d_inode;
2722         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2723                                     name, namelen, dir->i_ino, &key,
2724                                     BTRFS_FT_DIR);
2725         if (ret)
2726                 goto fail;
2727
2728         ret = btrfs_insert_inode_ref(trans, root->fs_info->tree_root,
2729                              name, namelen, objectid,
2730                              root->fs_info->sb->s_root->d_inode->i_ino);
2731         if (ret)
2732                 goto fail;
2733
2734         ret = btrfs_commit_transaction(trans, root);
2735         if (ret)
2736                 goto fail_commit;
2737
2738         new_root = btrfs_read_fs_root(root->fs_info, &key, name, namelen);
2739         BUG_ON(!new_root);
2740
2741         trans = btrfs_start_transaction(new_root, 1);
2742         BUG_ON(!trans);
2743
2744         inode = btrfs_new_inode(trans, new_root, "..", 2, new_dirid,
2745                                 new_dirid,
2746                                 BTRFS_I(dir)->block_group, S_IFDIR | 0700);
2747         if (IS_ERR(inode))
2748                 goto fail;
2749         inode->i_op = &btrfs_dir_inode_operations;
2750         inode->i_fop = &btrfs_dir_file_operations;
2751         new_root->inode = inode;
2752
2753         ret = btrfs_insert_inode_ref(trans, new_root, "..", 2, new_dirid,
2754                                      new_dirid);
2755         inode->i_nlink = 1;
2756         inode->i_size = 0;
2757         ret = btrfs_update_inode(trans, new_root, inode);
2758         if (ret)
2759                 goto fail;
2760 fail:
2761         nr = trans->blocks_used;
2762         err = btrfs_commit_transaction(trans, new_root);
2763         if (err && !ret)
2764                 ret = err;
2765 fail_commit:
2766         mutex_unlock(&root->fs_info->fs_mutex);
2767         btrfs_btree_balance_dirty(root, nr);
2768         btrfs_throttle(root);
2769         return ret;
2770 }
2771
2772 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2773 {
2774         struct btrfs_pending_snapshot *pending_snapshot;
2775         struct btrfs_trans_handle *trans;
2776         int ret;
2777         int err;
2778         unsigned long nr = 0;
2779
2780         if (!root->ref_cows)
2781                 return -EINVAL;
2782
2783         mutex_lock(&root->fs_info->fs_mutex);
2784         ret = btrfs_check_free_space(root, 1, 0);
2785         if (ret)
2786                 goto fail_unlock;
2787
2788         pending_snapshot = kmalloc(sizeof(*pending_snapshot), GFP_NOFS);
2789         if (!pending_snapshot) {
2790                 ret = -ENOMEM;
2791                 goto fail_unlock;
2792         }
2793         pending_snapshot->name = kmalloc(namelen + 1, GFP_NOFS);
2794         if (!pending_snapshot->name) {
2795                 ret = -ENOMEM;
2796                 kfree(pending_snapshot);
2797                 goto fail_unlock;
2798         }
2799         memcpy(pending_snapshot->name, name, namelen);
2800         pending_snapshot->name[namelen] = '\0';
2801         trans = btrfs_start_transaction(root, 1);
2802         BUG_ON(!trans);
2803         pending_snapshot->root = root;
2804         list_add(&pending_snapshot->list,
2805                  &trans->transaction->pending_snapshots);
2806         ret = btrfs_update_inode(trans, root, root->inode);
2807         err = btrfs_commit_transaction(trans, root);
2808
2809 fail_unlock:
2810         mutex_unlock(&root->fs_info->fs_mutex);
2811         btrfs_btree_balance_dirty(root, nr);
2812         btrfs_throttle(root);
2813         return ret;
2814 }
2815
2816 unsigned long btrfs_force_ra(struct address_space *mapping,
2817                               struct file_ra_state *ra, struct file *file,
2818                               pgoff_t offset, pgoff_t last_index)
2819 {
2820         pgoff_t req_size = last_index - offset + 1;
2821
2822 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
2823         offset = page_cache_readahead(mapping, ra, file, offset, req_size);
2824         return offset;
2825 #else
2826         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
2827         return offset + req_size;
2828 #endif
2829 }
2830
2831 int btrfs_defrag_file(struct file *file) {
2832         struct inode *inode = fdentry(file)->d_inode;
2833         struct btrfs_root *root = BTRFS_I(inode)->root;
2834         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2835         struct page *page;
2836         unsigned long last_index;
2837         unsigned long ra_pages = root->fs_info->bdi.ra_pages;
2838         unsigned long total_read = 0;
2839         u64 page_start;
2840         u64 page_end;
2841         unsigned long i;
2842         int ret;
2843
2844         mutex_lock(&root->fs_info->fs_mutex);
2845         ret = btrfs_check_free_space(root, inode->i_size, 0);
2846         mutex_unlock(&root->fs_info->fs_mutex);
2847         if (ret)
2848                 return -ENOSPC;
2849
2850         mutex_lock(&inode->i_mutex);
2851         last_index = inode->i_size >> PAGE_CACHE_SHIFT;
2852         for (i = 0; i <= last_index; i++) {
2853                 if (total_read % ra_pages == 0) {
2854                         btrfs_force_ra(inode->i_mapping, &file->f_ra, file, i,
2855                                        min(last_index, i + ra_pages - 1));
2856                 }
2857                 total_read++;
2858                 page = grab_cache_page(inode->i_mapping, i);
2859                 if (!page)
2860                         goto out_unlock;
2861                 if (!PageUptodate(page)) {
2862                         btrfs_readpage(NULL, page);
2863                         lock_page(page);
2864                         if (!PageUptodate(page)) {
2865                                 unlock_page(page);
2866                                 page_cache_release(page);
2867                                 goto out_unlock;
2868                         }
2869                 }
2870
2871 #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,18)
2872                 ClearPageDirty(page);
2873 #else
2874                 cancel_dirty_page(page, PAGE_CACHE_SIZE);
2875 #endif
2876                 wait_on_page_writeback(page);
2877                 set_page_extent_mapped(page);
2878
2879                 page_start = (u64)page->index << PAGE_CACHE_SHIFT;
2880                 page_end = page_start + PAGE_CACHE_SIZE - 1;
2881
2882                 lock_extent(io_tree, page_start, page_end, GFP_NOFS);
2883                 set_extent_delalloc(io_tree, page_start,
2884                                     page_end, GFP_NOFS);
2885
2886                 unlock_extent(io_tree, page_start, page_end, GFP_NOFS);
2887                 set_page_dirty(page);
2888                 unlock_page(page);
2889                 page_cache_release(page);
2890                 balance_dirty_pages_ratelimited_nr(inode->i_mapping, 1);
2891         }
2892
2893 out_unlock:
2894         mutex_unlock(&inode->i_mutex);
2895         return 0;
2896 }
2897
2898 static int btrfs_ioctl_resize(struct btrfs_root *root, void __user *arg)
2899 {
2900         u64 new_size;
2901         u64 old_size;
2902         u64 devid = 1;
2903         struct btrfs_ioctl_vol_args *vol_args;
2904         struct btrfs_trans_handle *trans;
2905         struct btrfs_device *device = NULL;
2906         char *sizestr;
2907         char *devstr = NULL;
2908         int ret = 0;
2909         int namelen;
2910         int mod = 0;
2911
2912         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
2913
2914         if (!vol_args)
2915                 return -ENOMEM;
2916
2917         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
2918                 ret = -EFAULT;
2919                 goto out;
2920         }
2921         namelen = strlen(vol_args->name);
2922         if (namelen > BTRFS_VOL_NAME_MAX) {
2923                 ret = -EINVAL;
2924                 goto out;
2925         }
2926
2927         mutex_lock(&root->fs_info->fs_mutex);
2928         sizestr = vol_args->name;
2929         devstr = strchr(sizestr, ':');
2930         if (devstr) {
2931                 char *end;
2932                 sizestr = devstr + 1;
2933                 *devstr = '\0';
2934                 devstr = vol_args->name;
2935                 devid = simple_strtoull(devstr, &end, 10);
2936 printk("resizing devid %Lu\n", devid);
2937         }
2938         device = btrfs_find_device(root, devid, NULL);
2939         if (!device) {
2940                 printk("resizer unable to find device %Lu\n", devid);
2941                 ret = -EINVAL;
2942                 goto out_unlock;
2943         }
2944         if (!strcmp(sizestr, "max"))
2945                 new_size = device->bdev->bd_inode->i_size;
2946         else {
2947                 if (sizestr[0] == '-') {
2948                         mod = -1;
2949                         sizestr++;
2950                 } else if (sizestr[0] == '+') {
2951                         mod = 1;
2952                         sizestr++;
2953                 }
2954                 new_size = btrfs_parse_size(sizestr);
2955                 if (new_size == 0) {
2956                         ret = -EINVAL;
2957                         goto out_unlock;
2958                 }
2959         }
2960
2961         old_size = device->total_bytes;
2962
2963         if (mod < 0) {
2964                 if (new_size > old_size) {
2965                         ret = -EINVAL;
2966                         goto out_unlock;
2967                 }
2968                 new_size = old_size - new_size;
2969         } else if (mod > 0) {
2970                 new_size = old_size + new_size;
2971         }
2972
2973         if (new_size < 256 * 1024 * 1024) {
2974                 ret = -EINVAL;
2975                 goto out_unlock;
2976         }
2977         if (new_size > device->bdev->bd_inode->i_size) {
2978                 ret = -EFBIG;
2979                 goto out_unlock;
2980         }
2981
2982         do_div(new_size, root->sectorsize);
2983         new_size *= root->sectorsize;
2984
2985 printk("new size for %s is %llu\n", device->name, (unsigned long long)new_size);
2986
2987         if (new_size > old_size) {
2988                 trans = btrfs_start_transaction(root, 1);
2989                 ret = btrfs_grow_device(trans, device, new_size);
2990                 btrfs_commit_transaction(trans, root);
2991         } else {
2992                 ret = btrfs_shrink_device(device, new_size);
2993         }
2994
2995 out_unlock:
2996         mutex_unlock(&root->fs_info->fs_mutex);
2997 out:
2998         kfree(vol_args);
2999         return ret;
3000 }
3001
3002 static int noinline btrfs_ioctl_snap_create(struct btrfs_root *root,
3003                                             void __user *arg)
3004 {
3005         struct btrfs_ioctl_vol_args *vol_args;
3006         struct btrfs_dir_item *di;
3007         struct btrfs_path *path;
3008         u64 root_dirid;
3009         int namelen;
3010         int ret;
3011
3012         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
3013
3014         if (!vol_args)
3015                 return -ENOMEM;
3016
3017         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
3018                 ret = -EFAULT;
3019                 goto out;
3020         }
3021
3022         namelen = strlen(vol_args->name);
3023         if (namelen > BTRFS_VOL_NAME_MAX) {
3024                 ret = -EINVAL;
3025                 goto out;
3026         }
3027         if (strchr(vol_args->name, '/')) {
3028                 ret = -EINVAL;
3029                 goto out;
3030         }
3031
3032         path = btrfs_alloc_path();
3033         if (!path) {
3034                 ret = -ENOMEM;
3035                 goto out;
3036         }
3037
3038         root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
3039         mutex_lock(&root->fs_info->fs_mutex);
3040         di = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
3041                             path, root_dirid,
3042                             vol_args->name, namelen, 0);
3043         mutex_unlock(&root->fs_info->fs_mutex);
3044         btrfs_free_path(path);
3045
3046         if (di && !IS_ERR(di)) {
3047                 ret = -EEXIST;
3048                 goto out;
3049         }
3050
3051         if (IS_ERR(di)) {
3052                 ret = PTR_ERR(di);
3053                 goto out;
3054         }
3055
3056         if (root == root->fs_info->tree_root)
3057                 ret = create_subvol(root, vol_args->name, namelen);
3058         else
3059                 ret = create_snapshot(root, vol_args->name, namelen);
3060 out:
3061         kfree(vol_args);
3062         return ret;
3063 }
3064
3065 static int btrfs_ioctl_defrag(struct file *file)
3066 {
3067         struct inode *inode = fdentry(file)->d_inode;
3068         struct btrfs_root *root = BTRFS_I(inode)->root;
3069
3070         switch (inode->i_mode & S_IFMT) {
3071         case S_IFDIR:
3072                 mutex_lock(&root->fs_info->fs_mutex);
3073                 btrfs_defrag_root(root, 0);
3074                 btrfs_defrag_root(root->fs_info->extent_root, 0);
3075                 mutex_unlock(&root->fs_info->fs_mutex);
3076                 break;
3077         case S_IFREG:
3078                 btrfs_defrag_file(file);
3079                 break;
3080         }
3081
3082         return 0;
3083 }
3084
3085 long btrfs_ioctl_add_dev(struct btrfs_root *root, void __user *arg)
3086 {
3087         struct btrfs_ioctl_vol_args *vol_args;
3088         int ret;
3089
3090         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
3091
3092         if (!vol_args)
3093                 return -ENOMEM;
3094
3095         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
3096                 ret = -EFAULT;
3097                 goto out;
3098         }
3099         ret = btrfs_init_new_device(root, vol_args->name);
3100
3101 out:
3102         kfree(vol_args);
3103         return ret;
3104 }
3105
3106 long btrfs_ioctl_rm_dev(struct btrfs_root *root, void __user *arg)
3107 {
3108         struct btrfs_ioctl_vol_args *vol_args;
3109         int ret;
3110
3111         vol_args = kmalloc(sizeof(*vol_args), GFP_NOFS);
3112
3113         if (!vol_args)
3114                 return -ENOMEM;
3115
3116         if (copy_from_user(vol_args, arg, sizeof(*vol_args))) {
3117                 ret = -EFAULT;
3118                 goto out;
3119         }
3120         ret = btrfs_rm_device(root, vol_args->name);
3121
3122 out:
3123         kfree(vol_args);
3124         return ret;
3125 }
3126
3127 int dup_item_to_inode(struct btrfs_trans_handle *trans,
3128                        struct btrfs_root *root,
3129                        struct btrfs_path *path,
3130                        struct extent_buffer *leaf,
3131                        int slot,
3132                        struct btrfs_key *key,
3133                        u64 destino)
3134 {
3135         char *dup;
3136         int len = btrfs_item_size_nr(leaf, slot);
3137         struct btrfs_key ckey = *key;
3138         int ret = 0;
3139
3140         dup = kmalloc(len, GFP_NOFS);
3141         if (!dup)
3142                 return -ENOMEM;
3143
3144         read_extent_buffer(leaf, dup, btrfs_item_ptr_offset(leaf, slot), len);
3145         btrfs_release_path(root, path);
3146
3147         ckey.objectid = destino;
3148         ret = btrfs_insert_item(trans, root, &ckey, dup, len);
3149         kfree(dup);
3150         return ret;
3151 }
3152
3153 long btrfs_ioctl_clone(struct file *file, unsigned long src_fd)
3154 {
3155         struct inode *inode = fdentry(file)->d_inode;
3156         struct btrfs_root *root = BTRFS_I(inode)->root;
3157         struct file *src_file;
3158         struct inode *src;
3159         struct btrfs_trans_handle *trans;
3160         int ret;
3161         u64 pos;
3162         struct btrfs_path *path;
3163         struct btrfs_key key;
3164         struct extent_buffer *leaf;
3165         u32 nritems;
3166         int slot;
3167
3168         src_file = fget(src_fd);
3169         if (!src_file)
3170                 return -EBADF;
3171         src = src_file->f_dentry->d_inode;
3172
3173         ret = -EXDEV;
3174         if (src->i_sb != inode->i_sb)
3175                 goto out_fput;
3176
3177         if (inode < src) {
3178                 mutex_lock(&inode->i_mutex);
3179                 mutex_lock(&src->i_mutex);
3180         } else {
3181                 mutex_lock(&src->i_mutex);
3182                 mutex_lock(&inode->i_mutex);
3183         }
3184
3185         ret = -ENOTEMPTY;
3186         if (inode->i_size)
3187                 goto out_unlock;
3188
3189         /* do any pending delalloc/csum calc on src, one way or
3190            another, and lock file content */
3191         while (1) {
3192                 filemap_write_and_wait(src->i_mapping);
3193                 lock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
3194                 if (BTRFS_I(src)->delalloc_bytes == 0)
3195                         break;
3196                 unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
3197         }
3198
3199         mutex_lock(&root->fs_info->fs_mutex);
3200         trans = btrfs_start_transaction(root, 0);
3201         path = btrfs_alloc_path();
3202         if (!path) {
3203                 ret = -ENOMEM;
3204                 goto out;
3205         }
3206         key.offset = 0;
3207         key.type = BTRFS_EXTENT_DATA_KEY;
3208         key.objectid = src->i_ino;
3209         pos = 0;
3210         path->reada = 2;
3211
3212         while (1) {
3213                 /*
3214                  * note the key will change type as we walk through the
3215                  * tree.
3216                  */
3217                 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
3218                 if (ret < 0)
3219                         goto out;
3220
3221                 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
3222                         ret = btrfs_next_leaf(root, path);
3223                         if (ret < 0)
3224                                 goto out;
3225                         if (ret > 0)
3226                                 break;
3227                 }
3228                 leaf = path->nodes[0];
3229                 slot = path->slots[0];
3230                 btrfs_item_key_to_cpu(leaf, &key, slot);
3231                 nritems = btrfs_header_nritems(leaf);
3232
3233                 if (btrfs_key_type(&key) > BTRFS_CSUM_ITEM_KEY ||
3234                     key.objectid != src->i_ino)
3235                         break;
3236
3237                 if (btrfs_key_type(&key) == BTRFS_EXTENT_DATA_KEY) {
3238                         struct btrfs_file_extent_item *extent;
3239                         int found_type;
3240                         pos = key.offset;
3241                         extent = btrfs_item_ptr(leaf, slot,
3242                                                 struct btrfs_file_extent_item);
3243                         found_type = btrfs_file_extent_type(leaf, extent);
3244                         if (found_type == BTRFS_FILE_EXTENT_REG) {
3245                                 u64 len = btrfs_file_extent_num_bytes(leaf,
3246                                                                       extent);
3247                                 u64 ds = btrfs_file_extent_disk_bytenr(leaf,
3248                                                                        extent);
3249                                 u64 dl = btrfs_file_extent_disk_num_bytes(leaf,
3250                                                                  extent);
3251                                 u64 off = btrfs_file_extent_offset(leaf,
3252                                                                    extent);
3253                                 btrfs_insert_file_extent(trans, root,
3254                                                          inode->i_ino, pos,
3255                                                          ds, dl, len, off);
3256                                 /* ds == 0 means there's a hole */
3257                                 if (ds != 0) {
3258                                         btrfs_inc_extent_ref(trans, root,
3259                                                      ds, dl,
3260                                                      root->root_key.objectid,
3261                                                      trans->transid,
3262                                                      inode->i_ino, pos);
3263                                 }
3264                                 pos = key.offset + len;
3265                         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
3266                                 ret = dup_item_to_inode(trans, root, path,
3267                                                         leaf, slot, &key,
3268                                                         inode->i_ino);
3269                                 if (ret)
3270                                         goto out;
3271                                 pos = key.offset + btrfs_item_size_nr(leaf,
3272                                                                       slot);
3273                         }
3274                 } else if (btrfs_key_type(&key) == BTRFS_CSUM_ITEM_KEY) {
3275                         ret = dup_item_to_inode(trans, root, path, leaf,
3276                                                 slot, &key, inode->i_ino);
3277
3278                         if (ret)
3279                                 goto out;
3280                 }
3281                 key.offset++;
3282                 btrfs_release_path(root, path);
3283         }
3284
3285         ret = 0;
3286 out:
3287         btrfs_free_path(path);
3288
3289         inode->i_blocks = src->i_blocks;
3290         i_size_write(inode, src->i_size);
3291         btrfs_update_inode(trans, root, inode);
3292
3293         unlock_extent(&BTRFS_I(src)->io_tree, 0, (u64)-1, GFP_NOFS);
3294
3295         btrfs_end_transaction(trans, root);
3296         mutex_unlock(&root->fs_info->fs_mutex);
3297
3298 out_unlock:
3299         mutex_unlock(&src->i_mutex);
3300         mutex_unlock(&inode->i_mutex);
3301 out_fput:
3302         fput(src_file);
3303         return ret;
3304 }
3305
3306 long btrfs_ioctl(struct file *file, unsigned int
3307                 cmd, unsigned long arg)
3308 {
3309         struct btrfs_root *root = BTRFS_I(fdentry(file)->d_inode)->root;
3310
3311         switch (cmd) {
3312         case BTRFS_IOC_SNAP_CREATE:
3313                 return btrfs_ioctl_snap_create(root, (void __user *)arg);
3314         case BTRFS_IOC_DEFRAG:
3315                 return btrfs_ioctl_defrag(file);
3316         case BTRFS_IOC_RESIZE:
3317                 return btrfs_ioctl_resize(root, (void __user *)arg);
3318         case BTRFS_IOC_ADD_DEV:
3319                 return btrfs_ioctl_add_dev(root, (void __user *)arg);
3320         case BTRFS_IOC_RM_DEV:
3321                 return btrfs_ioctl_rm_dev(root, (void __user *)arg);
3322         case BTRFS_IOC_BALANCE:
3323                 return btrfs_balance(root->fs_info->dev_root);
3324         case BTRFS_IOC_CLONE:
3325                 return btrfs_ioctl_clone(file, arg);
3326         }
3327
3328         return -ENOTTY;
3329 }
3330
3331 /*
3332  * Called inside transaction, so use GFP_NOFS
3333  */
3334 struct inode *btrfs_alloc_inode(struct super_block *sb)
3335 {
3336         struct btrfs_inode *ei;
3337
3338         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
3339         if (!ei)
3340                 return NULL;
3341         ei->last_trans = 0;
3342         ei->ordered_trans = 0;
3343         return &ei->vfs_inode;
3344 }
3345
3346 void btrfs_destroy_inode(struct inode *inode)
3347 {
3348         WARN_ON(!list_empty(&inode->i_dentry));
3349         WARN_ON(inode->i_data.nrpages);
3350
3351         btrfs_drop_extent_cache(inode, 0, (u64)-1);
3352         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
3353 }
3354
3355 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3356 static void init_once(struct kmem_cache * cachep, void *foo)
3357 #else
3358 static void init_once(void * foo, struct kmem_cache * cachep,
3359                       unsigned long flags)
3360 #endif
3361 {
3362         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
3363
3364         inode_init_once(&ei->vfs_inode);
3365 }
3366
3367 void btrfs_destroy_cachep(void)
3368 {
3369         if (btrfs_inode_cachep)
3370                 kmem_cache_destroy(btrfs_inode_cachep);
3371         if (btrfs_trans_handle_cachep)
3372                 kmem_cache_destroy(btrfs_trans_handle_cachep);
3373         if (btrfs_transaction_cachep)
3374                 kmem_cache_destroy(btrfs_transaction_cachep);
3375         if (btrfs_bit_radix_cachep)
3376                 kmem_cache_destroy(btrfs_bit_radix_cachep);
3377         if (btrfs_path_cachep)
3378                 kmem_cache_destroy(btrfs_path_cachep);
3379 }
3380
3381 struct kmem_cache *btrfs_cache_create(const char *name, size_t size,
3382                                        unsigned long extra_flags,
3383 #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23)
3384                                        void (*ctor)(struct kmem_cache *, void *)
3385 #else
3386                                        void (*ctor)(void *, struct kmem_cache *,
3387                                                     unsigned long)
3388 #endif
3389                                      )
3390 {
3391         return kmem_cache_create(name, size, 0, (SLAB_RECLAIM_ACCOUNT |
3392                                  SLAB_MEM_SPREAD | extra_flags), ctor
3393 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,23)
3394                                  ,NULL
3395 #endif
3396                                 );
3397 }
3398
3399 int btrfs_init_cachep(void)
3400 {
3401         btrfs_inode_cachep = btrfs_cache_create("btrfs_inode_cache",
3402                                           sizeof(struct btrfs_inode),
3403                                           0, init_once);
3404         if (!btrfs_inode_cachep)
3405                 goto fail;
3406         btrfs_trans_handle_cachep =
3407                         btrfs_cache_create("btrfs_trans_handle_cache",
3408                                            sizeof(struct btrfs_trans_handle),
3409                                            0, NULL);
3410         if (!btrfs_trans_handle_cachep)
3411                 goto fail;
3412         btrfs_transaction_cachep = btrfs_cache_create("btrfs_transaction_cache",
3413                                              sizeof(struct btrfs_transaction),
3414                                              0, NULL);
3415         if (!btrfs_transaction_cachep)
3416                 goto fail;
3417         btrfs_path_cachep = btrfs_cache_create("btrfs_path_cache",
3418                                          sizeof(struct btrfs_path),
3419                                          0, NULL);
3420         if (!btrfs_path_cachep)
3421                 goto fail;
3422         btrfs_bit_radix_cachep = btrfs_cache_create("btrfs_radix", 256,
3423                                               SLAB_DESTROY_BY_RCU, NULL);
3424         if (!btrfs_bit_radix_cachep)
3425                 goto fail;
3426         return 0;
3427 fail:
3428         btrfs_destroy_cachep();
3429         return -ENOMEM;
3430 }
3431
3432 static int btrfs_getattr(struct vfsmount *mnt,
3433                          struct dentry *dentry, struct kstat *stat)
3434 {
3435         struct inode *inode = dentry->d_inode;
3436         generic_fillattr(inode, stat);
3437         stat->blksize = PAGE_CACHE_SIZE;
3438         stat->blocks = inode->i_blocks + (BTRFS_I(inode)->delalloc_bytes >> 9);
3439         return 0;
3440 }
3441
3442 static int btrfs_rename(struct inode * old_dir, struct dentry *old_dentry,
3443                            struct inode * new_dir,struct dentry *new_dentry)
3444 {
3445         struct btrfs_trans_handle *trans;
3446         struct btrfs_root *root = BTRFS_I(old_dir)->root;
3447         struct inode *new_inode = new_dentry->d_inode;
3448         struct inode *old_inode = old_dentry->d_inode;
3449         struct timespec ctime = CURRENT_TIME;
3450         struct btrfs_path *path;
3451         int ret;
3452
3453         if (S_ISDIR(old_inode->i_mode) && new_inode &&
3454             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE) {
3455                 return -ENOTEMPTY;
3456         }
3457
3458         mutex_lock(&root->fs_info->fs_mutex);
3459         ret = btrfs_check_free_space(root, 1, 0);
3460         if (ret)
3461                 goto out_unlock;
3462
3463         trans = btrfs_start_transaction(root, 1);
3464
3465         btrfs_set_trans_block_group(trans, new_dir);
3466         path = btrfs_alloc_path();
3467         if (!path) {
3468                 ret = -ENOMEM;
3469                 goto out_fail;
3470         }
3471
3472         old_dentry->d_inode->i_nlink++;
3473         old_dir->i_ctime = old_dir->i_mtime = ctime;
3474         new_dir->i_ctime = new_dir->i_mtime = ctime;
3475         old_inode->i_ctime = ctime;
3476
3477         ret = btrfs_unlink_trans(trans, root, old_dir, old_dentry);
3478         if (ret)
3479                 goto out_fail;
3480
3481         if (new_inode) {
3482                 new_inode->i_ctime = CURRENT_TIME;
3483                 ret = btrfs_unlink_trans(trans, root, new_dir, new_dentry);
3484                 if (ret)
3485                         goto out_fail;
3486         }
3487         ret = btrfs_add_link(trans, new_dentry, old_inode, 1);
3488         if (ret)
3489                 goto out_fail;
3490
3491 out_fail:
3492         btrfs_free_path(path);
3493         btrfs_end_transaction(trans, root);
3494 out_unlock:
3495         mutex_unlock(&root->fs_info->fs_mutex);
3496         return ret;
3497 }
3498
3499 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
3500                          const char *symname)
3501 {
3502         struct btrfs_trans_handle *trans;
3503         struct btrfs_root *root = BTRFS_I(dir)->root;
3504         struct btrfs_path *path;
3505         struct btrfs_key key;
3506         struct inode *inode = NULL;
3507         int err;
3508         int drop_inode = 0;
3509         u64 objectid;
3510         int name_len;
3511         int datasize;
3512         unsigned long ptr;
3513         struct btrfs_file_extent_item *ei;
3514         struct extent_buffer *leaf;
3515         unsigned long nr = 0;
3516
3517         name_len = strlen(symname) + 1;
3518         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
3519                 return -ENAMETOOLONG;
3520
3521         mutex_lock(&root->fs_info->fs_mutex);
3522         err = btrfs_check_free_space(root, 1, 0);
3523         if (err)
3524                 goto out_fail;
3525
3526         trans = btrfs_start_transaction(root, 1);
3527         btrfs_set_trans_block_group(trans, dir);
3528
3529         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
3530         if (err) {
3531                 err = -ENOSPC;
3532                 goto out_unlock;
3533         }
3534
3535         inode = btrfs_new_inode(trans, root, dentry->d_name.name,
3536                                 dentry->d_name.len,
3537                                 dentry->d_parent->d_inode->i_ino, objectid,
3538                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO);
3539         err = PTR_ERR(inode);
3540         if (IS_ERR(inode))
3541                 goto out_unlock;
3542
3543         btrfs_set_trans_block_group(trans, inode);
3544         err = btrfs_add_nondir(trans, dentry, inode, 0);
3545         if (err)
3546                 drop_inode = 1;
3547         else {
3548                 inode->i_mapping->a_ops = &btrfs_aops;
3549                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3550                 inode->i_fop = &btrfs_file_operations;
3551                 inode->i_op = &btrfs_file_inode_operations;
3552                 extent_map_tree_init(&BTRFS_I(inode)->extent_tree, GFP_NOFS);
3553                 extent_io_tree_init(&BTRFS_I(inode)->io_tree,
3554                                      inode->i_mapping, GFP_NOFS);
3555                 extent_io_tree_init(&BTRFS_I(inode)->io_failure_tree,
3556                                      inode->i_mapping, GFP_NOFS);
3557                 BTRFS_I(inode)->delalloc_bytes = 0;
3558                 atomic_set(&BTRFS_I(inode)->ordered_writeback, 0);
3559                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3560         }
3561         dir->i_sb->s_dirt = 1;
3562         btrfs_update_inode_block_group(trans, inode);
3563         btrfs_update_inode_block_group(trans, dir);
3564         if (drop_inode)
3565                 goto out_unlock;
3566
3567         path = btrfs_alloc_path();
3568         BUG_ON(!path);
3569         key.objectid = inode->i_ino;
3570         key.offset = 0;
3571         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
3572         datasize = btrfs_file_extent_calc_inline_size(name_len);
3573         err = btrfs_insert_empty_item(trans, root, path, &key,
3574                                       datasize);
3575         if (err) {
3576                 drop_inode = 1;
3577                 goto out_unlock;
3578         }
3579         leaf = path->nodes[0];
3580         ei = btrfs_item_ptr(leaf, path->slots[0],
3581                             struct btrfs_file_extent_item);
3582         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
3583         btrfs_set_file_extent_type(leaf, ei,
3584                                    BTRFS_FILE_EXTENT_INLINE);
3585         ptr = btrfs_file_extent_inline_start(ei);
3586         write_extent_buffer(leaf, symname, ptr, name_len);
3587         btrfs_mark_buffer_dirty(leaf);
3588         btrfs_free_path(path);
3589
3590         inode->i_op = &btrfs_symlink_inode_operations;
3591         inode->i_mapping->a_ops = &btrfs_symlink_aops;
3592         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3593         inode->i_size = name_len - 1;
3594         err = btrfs_update_inode(trans, root, inode);
3595         if (err)
3596                 drop_inode = 1;
3597
3598 out_unlock:
3599         nr = trans->blocks_used;
3600         btrfs_end_transaction(trans, root);
3601 out_fail:
3602         mutex_unlock(&root->fs_info->fs_mutex);
3603         if (drop_inode) {
3604                 inode_dec_link_count(inode);
3605                 iput(inode);
3606         }
3607         btrfs_btree_balance_dirty(root, nr);
3608         btrfs_throttle(root);
3609         return err;
3610 }
3611
3612 static int btrfs_permission(struct inode *inode, int mask,
3613                             struct nameidata *nd)
3614 {
3615         if (btrfs_test_flag(inode, READONLY) && (mask & MAY_WRITE))
3616                 return -EACCES;
3617         return generic_permission(inode, mask, NULL);
3618 }
3619
3620 static struct inode_operations btrfs_dir_inode_operations = {
3621         .lookup         = btrfs_lookup,
3622         .create         = btrfs_create,
3623         .unlink         = btrfs_unlink,
3624         .link           = btrfs_link,
3625         .mkdir          = btrfs_mkdir,
3626         .rmdir          = btrfs_rmdir,
3627         .rename         = btrfs_rename,
3628         .symlink        = btrfs_symlink,
3629         .setattr        = btrfs_setattr,
3630         .mknod          = btrfs_mknod,
3631         .setxattr       = generic_setxattr,
3632         .getxattr       = generic_getxattr,
3633         .listxattr      = btrfs_listxattr,
3634         .removexattr    = generic_removexattr,
3635         .permission     = btrfs_permission,
3636 };
3637 static struct inode_operations btrfs_dir_ro_inode_operations = {
3638         .lookup         = btrfs_lookup,
3639         .permission     = btrfs_permission,
3640 };
3641 static struct file_operations btrfs_dir_file_operations = {
3642         .llseek         = generic_file_llseek,
3643         .read           = generic_read_dir,
3644         .readdir        = btrfs_readdir,
3645         .unlocked_ioctl = btrfs_ioctl,
3646 #ifdef CONFIG_COMPAT
3647         .compat_ioctl   = btrfs_ioctl,
3648 #endif
3649 };
3650
3651 static struct extent_io_ops btrfs_extent_io_ops = {
3652         .fill_delalloc = run_delalloc_range,
3653         .submit_bio_hook = btrfs_submit_bio_hook,
3654         .merge_bio_hook = btrfs_merge_bio_hook,
3655         .readpage_io_hook = btrfs_readpage_io_hook,
3656         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
3657         .readpage_io_failed_hook = btrfs_readpage_io_failed_hook,
3658         .set_bit_hook = btrfs_set_bit_hook,
3659         .clear_bit_hook = btrfs_clear_bit_hook,
3660 };
3661
3662 static struct address_space_operations btrfs_aops = {
3663         .readpage       = btrfs_readpage,
3664         .writepage      = btrfs_writepage,
3665         .writepages     = btrfs_writepages,
3666         .readpages      = btrfs_readpages,
3667         .sync_page      = block_sync_page,
3668         .bmap           = btrfs_bmap,
3669         .direct_IO      = btrfs_direct_IO,
3670         .invalidatepage = btrfs_invalidatepage,
3671         .releasepage    = btrfs_releasepage,
3672         .set_page_dirty = __set_page_dirty_nobuffers,
3673 };
3674
3675 static struct address_space_operations btrfs_symlink_aops = {
3676         .readpage       = btrfs_readpage,
3677         .writepage      = btrfs_writepage,
3678         .invalidatepage = btrfs_invalidatepage,
3679         .releasepage    = btrfs_releasepage,
3680 };
3681
3682 static struct inode_operations btrfs_file_inode_operations = {
3683         .truncate       = btrfs_truncate,
3684         .getattr        = btrfs_getattr,
3685         .setattr        = btrfs_setattr,
3686         .setxattr       = generic_setxattr,
3687         .getxattr       = generic_getxattr,
3688         .listxattr      = btrfs_listxattr,
3689         .removexattr    = generic_removexattr,
3690         .permission     = btrfs_permission,
3691 };
3692 static struct inode_operations btrfs_special_inode_operations = {
3693         .getattr        = btrfs_getattr,
3694         .setattr        = btrfs_setattr,
3695         .permission     = btrfs_permission,
3696 };
3697 static struct inode_operations btrfs_symlink_inode_operations = {
3698         .readlink       = generic_readlink,
3699         .follow_link    = page_follow_link_light,
3700         .put_link       = page_put_link,
3701         .permission     = btrfs_permission,
3702 };