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