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