SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/inode.c

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