fs/btrfs/inode.c

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