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