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