fs/btrfs/super.c

   1 #include <linux/module.h>
   2 #include <linux/buffer_head.h>
   3 #include <linux/fs.h>
   4 #include <linux/pagemap.h>
   5 #include <linux/highmem.h>
   6 #include <linux/time.h>
   7 #include <linux/init.h>
   8 #include <linux/string.h>
   9 #include <linux/smp_lock.h>
  10 #include <linux/backing-dev.h>
  11 #include <linux/mpage.h>
  12 #include <linux/swap.h>
  13 #include <linux/writeback.h>
  14 #include "ctree.h"
  15 #include "disk-io.h"
  16 #include "transaction.h"
  17 #include "btrfs_inode.h"
  18
  19 #define BTRFS_SUPER_MAGIC 0x9123682E
  20
  21 static struct inode_operations btrfs_dir_inode_operations;
  22 static struct super_operations btrfs_super_ops;
  23 static struct file_operations btrfs_dir_file_operations;
  24 static struct inode_operations btrfs_file_inode_operations;
  25 static struct address_space_operations btrfs_aops;
  26 static struct file_operations btrfs_file_operations;
  27
  28 static int check_inode(struct inode *inode)
  29 {
  30         struct btrfs_inode *ei = BTRFS_I(inode);
  31         WARN_ON(ei->magic != 0xDEADBEEF);
  32         WARN_ON(ei->magic2 != 0xDEADBEAF);
  33         return 0;
  34 }
  35
  36 static void btrfs_read_locked_inode(struct inode *inode)
  37 {
  38         struct btrfs_path *path;
  39         struct btrfs_inode_item *inode_item;
  40         struct btrfs_root *root = btrfs_sb(inode->i_sb);
  41         int ret;
  42
  43         path = btrfs_alloc_path();
  44         BUG_ON(!path);
  45         btrfs_init_path(path);
  46         mutex_lock(&root->fs_info->fs_mutex);
  47
  48         check_inode(inode);
  49         ret = btrfs_lookup_inode(NULL, root, path, inode->i_ino, 0);
  50         if (ret) {
  51                 btrfs_release_path(root, path);
  52                 btrfs_free_path(path);
  53                 mutex_unlock(&root->fs_info->fs_mutex);
  54                 make_bad_inode(inode);
  55                 return;
  56         }
  57         check_inode(inode);
  58         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
  59                                   path->slots[0],
  60                                   struct btrfs_inode_item);
  61
  62         inode->i_mode = btrfs_inode_mode(inode_item);
  63         inode->i_nlink = btrfs_inode_nlink(inode_item);
  64         inode->i_uid = btrfs_inode_uid(inode_item);
  65         inode->i_gid = btrfs_inode_gid(inode_item);
  66         inode->i_size = btrfs_inode_size(inode_item);
  67         inode->i_atime.tv_sec = btrfs_timespec_sec(&inode_item->atime);
  68         inode->i_atime.tv_nsec = btrfs_timespec_nsec(&inode_item->atime);
  69         inode->i_mtime.tv_sec = btrfs_timespec_sec(&inode_item->mtime);
  70         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(&inode_item->mtime);
  71         inode->i_ctime.tv_sec = btrfs_timespec_sec(&inode_item->ctime);
  72         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(&inode_item->ctime);
  73         inode->i_blocks = btrfs_inode_nblocks(inode_item);
  74         inode->i_generation = btrfs_inode_generation(inode_item);
  75
  76         btrfs_release_path(root, path);
  77         btrfs_free_path(path);
  78         inode_item = NULL;
  79
  80         mutex_unlock(&root->fs_info->fs_mutex);
  81         check_inode(inode);
  82         switch (inode->i_mode & S_IFMT) {
  83 #if 0
  84         default:
  85                 init_special_inode(inode, inode->i_mode,
  86                                    btrfs_inode_rdev(inode_item));
  87                 break;
  88 #endif
  89         case S_IFREG:
  90                 inode->i_mapping->a_ops = &btrfs_aops;
  91                 inode->i_fop = &btrfs_file_operations;
  92                 inode->i_op = &btrfs_file_inode_operations;
  93                 break;
  94         case S_IFDIR:
  95                 inode->i_op = &btrfs_dir_inode_operations;
  96                 inode->i_fop = &btrfs_dir_file_operations;
  97                 break;
  98         case S_IFLNK:
  99                 // inode->i_op = &page_symlink_inode_operations;
 100                 break;
 101         }
 102         check_inode(inode);
 103         return;
 104 }
 105
 106 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 107                               struct btrfs_root *root,
 108                               struct inode *dir,
 109                               struct dentry *dentry)
 110 {
 111         struct btrfs_path *path;
 112         const char *name = dentry->d_name.name;
 113         int name_len = dentry->d_name.len;
 114         int ret;
 115         u64 objectid;
 116         struct btrfs_dir_item *di;
 117
 118         path = btrfs_alloc_path();
 119         BUG_ON(!path);
 120         btrfs_init_path(path);
 121         ret = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 122                                     name, name_len, -1);
 123         if (ret < 0)
 124                 goto err;
 125         if (ret > 0) {
 126                 ret = -ENOENT;
 127                 goto err;
 128         }
 129         di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 130                             struct btrfs_dir_item);
 131         objectid = btrfs_dir_objectid(di);
 132
 133         ret = btrfs_del_item(trans, root, path);
 134         BUG_ON(ret);
 135
 136         btrfs_release_path(root, path);
 137         ret = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 138                                           objectid, -1);
 139         BUG_ON(ret);
 140         ret = btrfs_del_item(trans, root, path);
 141         BUG_ON(ret);
 142         dentry->d_inode->i_ctime = dir->i_ctime;
 143 err:
 144         btrfs_release_path(root, path);
 145         btrfs_free_path(path);
 146         if (ret == 0) {
 147                 inode_dec_link_count(dentry->d_inode);
 148                 dir->i_size -= name_len * 2;
 149                 mark_inode_dirty(dir);
 150         }
 151         return ret;
 152 }
 153
 154 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 155 {
 156         struct btrfs_root *root;
 157         struct btrfs_trans_handle *trans;
 158         int ret;
 159
 160         root = btrfs_sb(dir->i_sb);
 161         mutex_lock(&root->fs_info->fs_mutex);
 162         trans = btrfs_start_transaction(root, 1);
 163         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 164         btrfs_end_transaction(trans, root);
 165         mutex_unlock(&root->fs_info->fs_mutex);
 166         return ret;
 167 }
 168
 169 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 170 {
 171         struct inode *inode = dentry->d_inode;
 172         int err;
 173         int ret;
 174         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 175         struct btrfs_path *path;
 176         struct btrfs_key key;
 177         struct btrfs_trans_handle *trans;
 178         struct btrfs_key found_key;
 179         int found_type;
 180         struct btrfs_leaf *leaf;
 181         char *goodnames = "..";
 182
 183         path = btrfs_alloc_path();
 184         BUG_ON(!path);
 185         btrfs_init_path(path);
 186         mutex_lock(&root->fs_info->fs_mutex);
 187         trans = btrfs_start_transaction(root, 1);
 188         key.objectid = inode->i_ino;
 189         key.offset = (u64)-1;
 190         key.flags = (u32)-1;
 191         while(1) {
 192                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 193                 if (ret < 0) {
 194                         err = ret;
 195                         goto out;
 196                 }
 197                 BUG_ON(ret == 0);
 198                 if (path->slots[0] == 0) {
 199                         err = -ENOENT;
 200                         goto out;
 201                 }
 202                 path->slots[0]--;
 203                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 204                 btrfs_disk_key_to_cpu(&found_key,
 205                                       &leaf->items[path->slots[0]].key);
 206                 found_type = btrfs_key_type(&found_key);
 207                 if (found_key.objectid != inode->i_ino) {
 208                         err = -ENOENT;
 209                         goto out;
 210                 }
 211                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
 212                      found_type != BTRFS_DIR_INDEX_KEY) ||
 213                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
 214                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
 215                         err = -ENOTEMPTY;
 216                         goto out;
 217                 }
 218                 ret = btrfs_del_item(trans, root, path);
 219                 BUG_ON(ret);
 220
 221                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
 222                         break;
 223                 btrfs_release_path(root, path);
 224         }
 225         ret = 0;
 226         btrfs_release_path(root, path);
 227
 228         /* now the directory is empty */
 229         err = btrfs_unlink_trans(trans, root, dir, dentry);
 230         if (!err) {
 231                 inode->i_size = 0;
 232         }
 233 out:
 234         btrfs_release_path(root, path);
 235         btrfs_free_path(path);
 236         mutex_unlock(&root->fs_info->fs_mutex);
 237         ret = btrfs_end_transaction(trans, root);
 238         if (ret && !err)
 239                 err = ret;
 240         return err;
 241 }
 242
 243 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 244                             struct btrfs_root *root,
 245                             struct inode *inode)
 246 {
 247         u64 objectid = inode->i_ino;
 248         struct btrfs_path *path;
 249         struct btrfs_inode_map_item *map;
 250         struct btrfs_key stat_data_key;
 251         int ret;
 252
 253         clear_inode(inode);
 254
 255         path = btrfs_alloc_path();
 256         BUG_ON(!path);
 257         btrfs_init_path(path);
 258         ret = btrfs_lookup_inode_map(trans, root, path, objectid, -1);
 259         if (ret) {
 260                 if (ret > 0)
 261                         ret = -ENOENT;
 262                 goto error;
 263         }
 264         map = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 265                             struct btrfs_inode_map_item);
 266         btrfs_disk_key_to_cpu(&stat_data_key, &map->key);
 267         ret = btrfs_del_item(trans, root->fs_info->inode_root, path);
 268         BUG_ON(ret);
 269         btrfs_release_path(root, path);
 270
 271         ret = btrfs_lookup_inode(trans, root, path, objectid, -1);
 272         BUG_ON(ret);
 273         ret = btrfs_del_item(trans, root, path);
 274         BUG_ON(ret);
 275 error:
 276         btrfs_release_path(root, path);
 277         btrfs_free_path(path);
 278         return ret;
 279 }
 280
 281 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 282                                    struct btrfs_root *root,
 283                                    struct inode *inode)
 284 {
 285         int ret;
 286         struct btrfs_path *path;
 287         struct btrfs_key key;
 288         struct btrfs_disk_key *found_key;
 289         struct btrfs_leaf *leaf;
 290         struct btrfs_file_extent_item *fi = NULL;
 291         u64 extent_start = 0;
 292         u64 extent_num_blocks = 0;
 293         int found_extent;
 294
 295         path = btrfs_alloc_path();
 296         BUG_ON(!path);
 297         /* FIXME, add redo link to tree so we don't leak on crash */
 298         key.objectid = inode->i_ino;
 299         key.offset = (u64)-1;
 300         key.flags = 0;
 301         /*
 302          * use BTRFS_CSUM_ITEM_KEY because it is larger than inline keys
 303          * or extent data
 304          */
 305         btrfs_set_key_type(&key, BTRFS_CSUM_ITEM_KEY);
 306         while(1) {
 307                 btrfs_init_path(path);
 308                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 309                 if (ret < 0) {
 310                         goto error;
 311                 }
 312                 if (ret > 0) {
 313                         BUG_ON(path->slots[0] == 0);
 314                         path->slots[0]--;
 315                 }
 316                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 317                 found_key = &leaf->items[path->slots[0]].key;
 318                 if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
 319                         break;
 320                 if (btrfs_disk_key_type(found_key) != BTRFS_CSUM_ITEM_KEY &&
 321                     btrfs_disk_key_type(found_key) != BTRFS_INLINE_DATA_KEY &&
 322                     btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY)
 323                         break;
 324                 if (btrfs_disk_key_offset(found_key) < inode->i_size)
 325                         break;
 326                 if (btrfs_disk_key_type(found_key) == BTRFS_EXTENT_DATA_KEY) {
 327                         fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 328                                             path->slots[0],
 329                                             struct btrfs_file_extent_item);
 330                         extent_start = btrfs_file_extent_disk_blocknr(fi);
 331                         extent_num_blocks =
 332                                 btrfs_file_extent_disk_num_blocks(fi);
 333                         inode->i_blocks -=
 334                                 btrfs_file_extent_num_blocks(fi) >> 9;
 335                         found_extent = 1;
 336                 } else {
 337                         found_extent = 0;
 338                 }
 339                 ret = btrfs_del_item(trans, root, path);
 340                 BUG_ON(ret);
 341                 btrfs_release_path(root, path);
 342                 if (found_extent) {
 343                         ret = btrfs_free_extent(trans, root, extent_start,
 344                                                 extent_num_blocks, 0);
 345                         BUG_ON(ret);
 346                 }
 347         }
 348         ret = 0;
 349 error:
 350         btrfs_release_path(root, path);
 351         btrfs_free_path(path);
 352         return ret;
 353 }
 354
 355 static void btrfs_delete_inode(struct inode *inode)
 356 {
 357         struct btrfs_trans_handle *trans;
 358         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 359         int ret;
 360
 361         truncate_inode_pages(&inode->i_data, 0);
 362         if (is_bad_inode(inode)) {
 363                 goto no_delete;
 364         }
 365         inode->i_size = 0;
 366         mutex_lock(&root->fs_info->fs_mutex);
 367         trans = btrfs_start_transaction(root, 1);
 368         if (S_ISREG(inode->i_mode)) {
 369                 ret = btrfs_truncate_in_trans(trans, root, inode);
 370                 BUG_ON(ret);
 371         }
 372         btrfs_free_inode(trans, root, inode);
 373         btrfs_end_transaction(trans, root);
 374         mutex_unlock(&root->fs_info->fs_mutex);
 375         return;
 376 no_delete:
 377         clear_inode(inode);
 378 }
 379
 380 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 381                               ino_t *ino)
 382 {
 383         const char *name = dentry->d_name.name;
 384         int namelen = dentry->d_name.len;
 385         struct btrfs_dir_item *di;
 386         struct btrfs_path *path;
 387         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 388         int ret;
 389
 390         path = btrfs_alloc_path();
 391         BUG_ON(!path);
 392         btrfs_init_path(path);
 393         ret = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
 394                                     namelen, 0);
 395         if (ret || !btrfs_match_dir_item_name(root, path, name, namelen)) {
 396                 *ino = 0;
 397                 ret = 0;
 398                 goto out;
 399         }
 400         di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 401                             struct btrfs_dir_item);
 402         *ino = btrfs_dir_objectid(di);
 403 out:
 404         btrfs_release_path(root, path);
 405         btrfs_free_path(path);
 406         check_inode(dir);
 407         return ret;
 408 }
 409
 410 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 411                                    struct nameidata *nd)
 412 {
 413         struct inode * inode;
 414         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 415         ino_t ino;
 416         int ret;
 417
 418         if (dentry->d_name.len > BTRFS_NAME_LEN)
 419                 return ERR_PTR(-ENAMETOOLONG);
 420         mutex_lock(&root->fs_info->fs_mutex);
 421         ret = btrfs_inode_by_name(dir, dentry, &ino);
 422         mutex_unlock(&root->fs_info->fs_mutex);
 423         if (ret < 0)
 424                 return ERR_PTR(ret);
 425         inode = NULL;
 426         if (ino) {
 427                 inode = iget(dir->i_sb, ino);
 428                 if (!inode)
 429                         return ERR_PTR(-EACCES);
 430                 check_inode(inode);
 431         }
 432         check_inode(dir);
 433         return d_splice_alias(inode, dentry);
 434 }
 435
 436 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 437 {
 438         struct inode *inode = filp->f_path.dentry->d_inode;
 439         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 440         struct btrfs_item *item;
 441         struct btrfs_dir_item *di;
 442         struct btrfs_key key;
 443         struct btrfs_path *path;
 444         int ret;
 445         u32 nritems;
 446         struct btrfs_leaf *leaf;
 447         int slot;
 448         int advance;
 449         unsigned char d_type = DT_UNKNOWN;
 450         int over = 0;
 451
 452         mutex_lock(&root->fs_info->fs_mutex);
 453         key.objectid = inode->i_ino;
 454         key.flags = 0;
 455         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
 456         key.offset = filp->f_pos;
 457         path = btrfs_alloc_path();
 458         btrfs_init_path(path);
 459         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 460         if (ret < 0) {
 461                 goto err;
 462         }
 463         advance = 0;
 464         while(1) {
 465                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 466                 nritems = btrfs_header_nritems(&leaf->header);
 467                 slot = path->slots[0];
 468                 if (advance || slot >= nritems) {
 469                         if (slot >= nritems -1) {
 470                                 ret = btrfs_next_leaf(root, path);
 471                                 if (ret)
 472                                         break;
 473                                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 474                                 nritems = btrfs_header_nritems(&leaf->header);
 475                                 slot = path->slots[0];
 476                         } else {
 477                                 slot++;
 478                                 path->slots[0]++;
 479                         }
 480                 }
 481                 advance = 1;
 482                 item = leaf->items + slot;
 483                 if (btrfs_disk_key_objectid(&item->key) != key.objectid)
 484                         break;
 485                 if (btrfs_disk_key_type(&item->key) != BTRFS_DIR_INDEX_KEY)
 486                         continue;
 487                 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
 488                         continue;
 489
 490                 advance = 1;
 491                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 492                 over = filldir(dirent, (const char *)(di + 1),
 493                                btrfs_dir_name_len(di),
 494                                btrfs_disk_key_offset(&item->key),
 495                                btrfs_dir_objectid(di), d_type);
 496                 if (over) {
 497                         filp->f_pos = btrfs_disk_key_offset(&item->key);
 498                         break;
 499                 }
 500                 filp->f_pos = btrfs_disk_key_offset(&item->key) + 1;
 501         }
 502         ret = 0;
 503 err:
 504         btrfs_release_path(root, path);
 505         btrfs_free_path(path);
 506         mutex_unlock(&root->fs_info->fs_mutex);
 507         return ret;
 508 }
 509
 510 static void btrfs_put_super (struct super_block * sb)
 511 {
 512         struct btrfs_root *root = btrfs_sb(sb);
 513         int ret;
 514
 515         ret = close_ctree(root);
 516         if (ret) {
 517                 printk("close ctree returns %d\n", ret);
 518         }
 519         sb->s_fs_info = NULL;
 520 }
 521
 522 static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 523 {
 524         struct inode * inode;
 525         struct dentry * root_dentry;
 526         struct btrfs_super_block *disk_super;
 527         struct btrfs_root *root;
 528
 529         sb->s_maxbytes = MAX_LFS_FILESIZE;
 530         sb->s_magic = BTRFS_SUPER_MAGIC;
 531         sb->s_op = &btrfs_super_ops;
 532         sb->s_time_gran = 1;
 533
 534         root = open_ctree(sb);
 535
 536         if (!root) {
 537                 printk("btrfs: open_ctree failed\n");
 538                 return -EIO;
 539         }
 540         sb->s_fs_info = root;
 541         disk_super = root->fs_info->disk_super;
 542         printk("read in super total blocks %Lu root %Lu\n",
 543                btrfs_super_total_blocks(disk_super),
 544                btrfs_super_root_dir(disk_super));
 545
 546         inode = iget_locked(sb, btrfs_super_root_dir(disk_super));
 547         if (!inode)
 548                 return -ENOMEM;
 549         if (inode->i_state & I_NEW) {
 550                 btrfs_read_locked_inode(inode);
 551                 unlock_new_inode(inode);
 552         }
 553
 554         root_dentry = d_alloc_root(inode);
 555         if (!root_dentry) {
 556                 iput(inode);
 557                 return -ENOMEM;
 558         }
 559         sb->s_root = root_dentry;
 560
 561         return 0;
 562 }
 563
 564 static void fill_inode_item(struct btrfs_inode_item *item,
 565                             struct inode *inode)
 566 {
 567         btrfs_set_inode_uid(item, inode->i_uid);
 568         btrfs_set_inode_gid(item, inode->i_gid);
 569         btrfs_set_inode_size(item, inode->i_size);
 570         btrfs_set_inode_mode(item, inode->i_mode);
 571         btrfs_set_inode_nlink(item, inode->i_nlink);
 572         btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
 573         btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
 574         btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
 575         btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
 576         btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
 577         btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
 578         btrfs_set_inode_nblocks(item, inode->i_blocks);
 579         btrfs_set_inode_generation(item, inode->i_generation);
 580         check_inode(inode);
 581 }
 582
 583 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
 584                               struct btrfs_root *root,
 585                               struct inode *inode)
 586 {
 587         struct btrfs_inode_item *inode_item;
 588         struct btrfs_path *path;
 589         int ret;
 590
 591         path = btrfs_alloc_path();
 592         BUG_ON(!path);
 593         btrfs_init_path(path);
 594
 595         ret = btrfs_lookup_inode(trans, root, path, inode->i_ino, 1);
 596         if (ret) {
 597                 if (ret > 0)
 598                         ret = -ENOENT;
 599                 goto failed;
 600         }
 601
 602         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 603                                   path->slots[0],
 604                                   struct btrfs_inode_item);
 605
 606         fill_inode_item(inode_item, inode);
 607         btrfs_mark_buffer_dirty(path->nodes[0]);
 608 failed:
 609         btrfs_release_path(root, path);
 610         btrfs_free_path(path);
 611         check_inode(inode);
 612         return 0;
 613 }
 614
 615 static int btrfs_write_inode(struct inode *inode, int wait)
 616 {
 617         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 618         struct btrfs_trans_handle *trans;
 619         int ret;
 620
 621         mutex_lock(&root->fs_info->fs_mutex);
 622         trans = btrfs_start_transaction(root, 1);
 623         ret = btrfs_update_inode(trans, root, inode);
 624         if (wait)
 625                 btrfs_commit_transaction(trans, root);
 626         else
 627                 btrfs_end_transaction(trans, root);
 628         mutex_unlock(&root->fs_info->fs_mutex);
 629         check_inode(inode);
 630         return ret;
 631 }
 632
 633 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 634                                      struct inode *dir, int mode)
 635 {
 636         struct inode *inode;
 637         struct btrfs_inode_item inode_item;
 638         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 639         struct btrfs_key key;
 640         int ret;
 641         u64 objectid;
 642
 643         inode = new_inode(dir->i_sb);
 644         if (!inode)
 645                 return ERR_PTR(-ENOMEM);
 646
 647         check_inode(inode);
 648         ret = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 649         BUG_ON(ret);
 650
 651         inode->i_uid = current->fsuid;
 652         inode->i_gid = current->fsgid;
 653         inode->i_mode = mode;
 654         inode->i_ino = objectid;
 655         inode->i_blocks = 0;
 656         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
 657         fill_inode_item(&inode_item, inode);
 658
 659         key.objectid = objectid;
 660         key.flags = 0;
 661         key.offset = 0;
 662         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 663         ret = btrfs_insert_inode_map(trans, root, objectid, &key);
 664         BUG_ON(ret);
 665
 666         ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
 667         BUG_ON(ret);
 668
 669         insert_inode_hash(inode);
 670         check_inode(inode);
 671         check_inode(dir);
 672         return inode;
 673 }
 674
 675 static int btrfs_add_link(struct btrfs_trans_handle *trans,
 676                             struct dentry *dentry, struct inode *inode)
 677 {
 678         int ret;
 679         ret = btrfs_insert_dir_item(trans, btrfs_sb(inode->i_sb),
 680                                     dentry->d_name.name, dentry->d_name.len,
 681                                     dentry->d_parent->d_inode->i_ino,
 682                                     inode->i_ino, 0);
 683         if (ret == 0) {
 684                 dentry->d_parent->d_inode->i_size += dentry->d_name.len * 2;
 685                 ret = btrfs_update_inode(trans, btrfs_sb(inode->i_sb),
 686                                          dentry->d_parent->d_inode);
 687         }
 688         check_inode(inode);
 689         check_inode(dentry->d_parent->d_inode);
 690         return ret;
 691 }
 692
 693 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
 694                             struct dentry *dentry, struct inode *inode)
 695 {
 696         int err = btrfs_add_link(trans, dentry, inode);
 697         if (!err) {
 698                 d_instantiate(dentry, inode);
 699                 return 0;
 700         }
 701         if (err > 0)
 702                 err = -EEXIST;
 703         check_inode(inode);
 704         return err;
 705 }
 706
 707 static int btrfs_create(struct inode *dir, struct dentry *dentry,
 708                         int mode, struct nameidata *nd)
 709 {
 710         struct btrfs_trans_handle *trans;
 711         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 712         struct inode *inode;
 713         int err;
 714         int drop_inode = 0;
 715
 716         mutex_lock(&root->fs_info->fs_mutex);
 717         trans = btrfs_start_transaction(root, 1);
 718         inode = btrfs_new_inode(trans, dir, mode);
 719         err = PTR_ERR(inode);
 720         if (IS_ERR(inode))
 721                 goto out_unlock;
 722         // FIXME mark the inode dirty
 723         err = btrfs_add_nondir(trans, dentry, inode);
 724         if (err)
 725                 drop_inode = 1;
 726         else {
 727                 inode->i_mapping->a_ops = &btrfs_aops;
 728                 inode->i_fop = &btrfs_file_operations;
 729                 inode->i_op = &btrfs_file_inode_operations;
 730         }
 731         dir->i_sb->s_dirt = 1;
 732 out_unlock:
 733         btrfs_end_transaction(trans, root);
 734         mutex_unlock(&root->fs_info->fs_mutex);
 735         check_inode(inode);
 736         check_inode(dir);
 737
 738         if (drop_inode) {
 739                 inode_dec_link_count(inode);
 740                 iput(inode);
 741         }
 742         return err;
 743 }
 744
 745 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
 746                                 struct inode *inode, struct inode *dir)
 747 {
 748         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 749         int ret;
 750         char buf[2];
 751         buf[0] = '.';
 752         buf[1] = '.';
 753
 754         ret = btrfs_insert_dir_item(trans, root, buf, 1, inode->i_ino,
 755                                     inode->i_ino, 1);
 756         if (ret)
 757                 goto error;
 758         ret = btrfs_insert_dir_item(trans, root, buf, 2, inode->i_ino,
 759                                     dir->i_ino, 1);
 760         if (ret)
 761                 goto error;
 762         inode->i_size = 6;
 763         ret = btrfs_update_inode(trans, root, inode);
 764 error:
 765         return ret;
 766 }
 767
 768 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 769 {
 770         struct inode *inode;
 771         struct btrfs_trans_handle *trans;
 772         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 773         int err = 0;
 774         int drop_on_err = 0;
 775
 776         mutex_lock(&root->fs_info->fs_mutex);
 777         trans = btrfs_start_transaction(root, 1);
 778         if (IS_ERR(trans)) {
 779                 err = PTR_ERR(trans);
 780                 goto out_unlock;
 781         }
 782         inode = btrfs_new_inode(trans, dir, S_IFDIR | mode);
 783         if (IS_ERR(inode)) {
 784                 err = PTR_ERR(inode);
 785                 goto out_fail;
 786         }
 787         drop_on_err = 1;
 788         inode->i_op = &btrfs_dir_inode_operations;
 789         inode->i_fop = &btrfs_dir_file_operations;
 790
 791         err = btrfs_make_empty_dir(trans, inode, dir);
 792         if (err)
 793                 goto out_fail;
 794         err = btrfs_add_link(trans, dentry, inode);
 795         if (err)
 796                 goto out_fail;
 797         d_instantiate(dentry, inode);
 798         drop_on_err = 0;
 799
 800 out_fail:
 801         btrfs_end_transaction(trans, root);
 802 out_unlock:
 803         mutex_unlock(&root->fs_info->fs_mutex);
 804         if (drop_on_err)
 805                 iput(inode);
 806         return err;
 807 }
 808
 809 static int btrfs_sync_fs(struct super_block *sb, int wait)
 810 {
 811         struct btrfs_trans_handle *trans;
 812         struct btrfs_root *root;
 813         int ret;
 814         root = btrfs_sb(sb);
 815
 816         sb->s_dirt = 0;
 817         if (!wait) {
 818                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 819                 return 0;
 820         }
 821         filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
 822         mutex_lock(&root->fs_info->fs_mutex);
 823         trans = btrfs_start_transaction(root, 1);
 824         ret = btrfs_commit_transaction(trans, root);
 825         sb->s_dirt = 0;
 826         BUG_ON(ret);
 827 printk("btrfs sync_fs\n");
 828         mutex_unlock(&root->fs_info->fs_mutex);
 829         return 0;
 830 }
 831
 832 #if 0
 833 static int btrfs_get_block_inline(struct inode *inode, sector_t iblock,
 834                            struct buffer_head *result, int create)
 835 {
 836         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 837         struct btrfs_path *path;
 838         struct btrfs_key key;
 839         struct btrfs_leaf *leaf;
 840         int num_bytes = result->b_size;
 841         int item_size;
 842         int ret;
 843         u64 pos;
 844         char *ptr;
 845         int copy_size;
 846         int err = 0;
 847         char *safe_ptr;
 848         char *data_ptr;
 849
 850         path = btrfs_alloc_path();
 851         BUG_ON(!path);
 852
 853         WARN_ON(create);
 854         if (create) {
 855                 return 0;
 856         }
 857         pos = iblock << inode->i_blkbits;
 858         key.objectid = inode->i_ino;
 859         key.flags = 0;
 860         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
 861         ptr = kmap(result->b_page);
 862         safe_ptr = ptr;
 863         ptr += (pos & (PAGE_CACHE_SIZE -1));
 864 again:
 865         key.offset = pos;
 866         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 867         if (ret) {
 868                 if (ret < 0)
 869                         err = ret;
 870                 else
 871                         err = 0;
 872                 goto out;
 873         }
 874         leaf = btrfs_buffer_leaf(path->nodes[0]);
 875         item_size = btrfs_item_size(leaf->items + path->slots[0]);
 876         copy_size = min(num_bytes, item_size);
 877         data_ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 878         WARN_ON(safe_ptr + PAGE_CACHE_SIZE < ptr + copy_size);
 879         memcpy(ptr, data_ptr, copy_size);
 880         pos += copy_size;
 881         num_bytes -= copy_size;
 882         WARN_ON(num_bytes < 0);
 883         ptr += copy_size;
 884         btrfs_release_path(root, path);
 885         if (num_bytes != 0) {
 886                 if (pos >= i_size_read(inode))
 887                         memset(ptr, 0, num_bytes);
 888                 else
 889                         goto again;
 890         }
 891         set_buffer_uptodate(result);
 892         map_bh(result, inode->i_sb, 0);
 893         err = 0;
 894 out:
 895         btrfs_free_path(path);
 896         kunmap(result->b_page);
 897         return err;
 898 }
 899 #endif
 900
 901 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
 902                            struct buffer_head *result, int create)
 903 {
 904         int ret;
 905         int err = 0;
 906         u64 blocknr;
 907         u64 extent_start = 0;
 908         u64 extent_end = 0;
 909         u64 objectid = inode->i_ino;
 910         struct btrfs_path *path;
 911         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 912         struct btrfs_trans_handle *trans = NULL;
 913         struct btrfs_file_extent_item *item;
 914         struct btrfs_leaf *leaf;
 915         struct btrfs_disk_key *found_key;
 916
 917         path = btrfs_alloc_path();
 918         BUG_ON(!path);
 919         btrfs_init_path(path);
 920         if (create)
 921                 trans = btrfs_start_transaction(root, 1);
 922
 923         ret = btrfs_lookup_file_extent(trans, root, path,
 924                                        inode->i_ino,
 925                                        iblock << inode->i_blkbits, 0);
 926         if (ret < 0) {
 927                 err = ret;
 928                 goto out;
 929         }
 930
 931         if (ret != 0) {
 932                 if (path->slots[0] == 0) {
 933                         btrfs_release_path(root, path);
 934                         goto allocate;
 935                 }
 936                 path->slots[0]--;
 937         }
 938
 939         item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 940                               struct btrfs_file_extent_item);
 941         leaf = btrfs_buffer_leaf(path->nodes[0]);
 942         blocknr = btrfs_file_extent_disk_blocknr(item);
 943         blocknr += btrfs_file_extent_offset(item);
 944
 945         /* exact match found, use it */
 946         if (ret == 0) {
 947                 err = 0;
 948                 map_bh(result, inode->i_sb, blocknr);
 949                 goto out;
 950         }
 951
 952         /* are we inside the extent that was found? */
 953         found_key = &leaf->items[path->slots[0]].key;
 954         if (btrfs_disk_key_objectid(found_key) != objectid ||
 955             btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY) {
 956                 extent_end = 0;
 957                 extent_start = 0;
 958                 btrfs_release_path(root, path);
 959                 goto allocate;
 960         }
 961
 962         extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
 963         extent_start = extent_start >> inode->i_blkbits;
 964         extent_start += btrfs_file_extent_offset(item);
 965         extent_end = extent_start + btrfs_file_extent_num_blocks(item);
 966         if (iblock >= extent_start && iblock < extent_end) {
 967                 err = 0;
 968                 map_bh(result, inode->i_sb, blocknr + iblock - extent_start);
 969                 goto out;
 970         }
 971 allocate:
 972         /* ok, create a new extent */
 973         if (!create) {
 974                 err = 0;
 975                 goto out;
 976         }
 977         ret = btrfs_alloc_file_extent(trans, root, objectid,
 978                                       iblock << inode->i_blkbits,
 979                                       1, extent_end, &blocknr);
 980         if (ret) {
 981                 err = ret;
 982                 goto out;
 983         }
 984         inode->i_blocks += inode->i_sb->s_blocksize >> 9;
 985         set_buffer_new(result);
 986         map_bh(result, inode->i_sb, blocknr);
 987
 988 out:
 989         btrfs_release_path(root, path);
 990         btrfs_free_path(path);
 991         if (trans)
 992                 btrfs_end_transaction(trans, root);
 993         return err;
 994 }
 995
 996 static int btrfs_get_block(struct inode *inode, sector_t iblock,
 997                            struct buffer_head *result, int create)
 998 {
 999         int err;
1000         struct btrfs_root *root = btrfs_sb(inode->i_sb);
1001         mutex_lock(&root->fs_info->fs_mutex);
1002         err = btrfs_get_block_lock(inode, iblock, result, create);
1003         // err = btrfs_get_block_inline(inode, iblock, result, create);
1004         mutex_unlock(&root->fs_info->fs_mutex);
1005         return err;
1006 }
1007
1008 static int btrfs_prepare_write(struct file *file, struct page *page,
1009                                unsigned from, unsigned to)
1010 {
1011         return nobh_prepare_write(page, from, to, btrfs_get_block);
1012 }
1013 static int btrfs_commit_write(struct file *file, struct page *page,
1014                                unsigned from, unsigned to)
1015 {
1016         return nobh_commit_write(file, page, from, to);
1017 }
1018
1019 static void btrfs_write_super(struct super_block *sb)
1020 {
1021         btrfs_sync_fs(sb, 1);
1022 }
1023
1024 static int btrfs_readpage(struct file *file, struct page *page)
1025 {
1026         return mpage_readpage(page, btrfs_get_block);
1027 }
1028
1029 static int btrfs_readpages(struct file *file, struct address_space *mapping,
1030                            struct list_head *pages, unsigned nr_pages)
1031 {
1032         return mpage_readpages(mapping, pages, nr_pages, btrfs_get_block);
1033 }
1034
1035 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1036 {
1037         return nobh_writepage(page, btrfs_get_block, wbc);
1038 }
1039
1040 static void btrfs_truncate(struct inode *inode)
1041 {
1042         struct btrfs_root *root = btrfs_sb(inode->i_sb);
1043         int ret;
1044         struct btrfs_trans_handle *trans;
1045
1046         if (!S_ISREG(inode->i_mode))
1047                 return;
1048         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1049                 return;
1050
1051         nobh_truncate_page(inode->i_mapping, inode->i_size);
1052
1053         /* FIXME, add redo link to tree so we don't leak on crash */
1054         mutex_lock(&root->fs_info->fs_mutex);
1055         trans = btrfs_start_transaction(root, 1);
1056         ret = btrfs_truncate_in_trans(trans, root, inode);
1057         BUG_ON(ret);
1058         ret = btrfs_end_transaction(trans, root);
1059         BUG_ON(ret);
1060         mutex_unlock(&root->fs_info->fs_mutex);
1061         mark_inode_dirty(inode);
1062 }
1063
1064 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
1065                                 struct page **prepared_pages,
1066                                 const char __user * buf)
1067 {
1068         long page_fault = 0;
1069         int i;
1070         int offset = pos & (PAGE_CACHE_SIZE - 1);
1071
1072         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
1073                 size_t count = min_t(size_t,
1074                                      PAGE_CACHE_SIZE - offset, write_bytes);
1075                 struct page *page = prepared_pages[i];
1076                 fault_in_pages_readable(buf, count);
1077
1078                 /* Copy data from userspace to the current page */
1079                 kmap(page);
1080                 page_fault = __copy_from_user(page_address(page) + offset,
1081                                               buf, count);
1082                 /* Flush processor's dcache for this page */
1083                 flush_dcache_page(page);
1084                 kunmap(page);
1085                 buf += count;
1086                 write_bytes -= count;
1087
1088                 if (page_fault)
1089                         break;
1090         }
1091         return page_fault ? -EFAULT : 0;
1092 }
1093
1094 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
1095 {
1096         size_t i;
1097         for (i = 0; i < num_pages; i++) {
1098                 if (!pages[i])
1099                         break;
1100                 unlock_page(pages[i]);
1101                 mark_page_accessed(pages[i]);
1102                 page_cache_release(pages[i]);
1103         }
1104 }
1105 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
1106                                    struct btrfs_root *root,
1107                                    struct file *file,
1108                                    struct page **pages,
1109                                    size_t num_pages,
1110                                    loff_t pos,
1111                                    size_t write_bytes)
1112 {
1113         int i;
1114         int offset;
1115         int err = 0;
1116         int ret;
1117         int this_write;
1118         struct inode *inode = file->f_path.dentry->d_inode;
1119
1120         for (i = 0; i < num_pages; i++) {
1121                 offset = pos & (PAGE_CACHE_SIZE -1);
1122                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1123                 /* FIXME, one block at a time */
1124
1125                 mutex_lock(&root->fs_info->fs_mutex);
1126                 trans = btrfs_start_transaction(root, 1);
1127                 btrfs_csum_file_block(trans, root, inode->i_ino,
1128                                       pages[i]->index << PAGE_CACHE_SHIFT,
1129                                       kmap(pages[i]), PAGE_CACHE_SIZE);
1130                 kunmap(pages[i]);
1131                 SetPageChecked(pages[i]);
1132                 ret = btrfs_end_transaction(trans, root);
1133                 BUG_ON(ret);
1134                 mutex_unlock(&root->fs_info->fs_mutex);
1135
1136                 ret = nobh_commit_write(file, pages[i], offset,
1137                                          offset + this_write);
1138                 pos += this_write;
1139                 if (ret) {
1140                         err = ret;
1141                         goto failed;
1142                 }
1143                 WARN_ON(this_write > write_bytes);
1144                 write_bytes -= this_write;
1145         }
1146 failed:
1147         return err;
1148 }
1149
1150 static int prepare_pages(struct btrfs_trans_handle *trans,
1151                          struct btrfs_root *root,
1152                          struct file *file,
1153                          struct page **pages,
1154                          size_t num_pages,
1155                          loff_t pos,
1156                          size_t write_bytes)
1157 {
1158         int i;
1159         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1160         struct inode *inode = file->f_path.dentry->d_inode;
1161         int offset;
1162         int err = 0;
1163         int ret;
1164         int this_write;
1165         loff_t isize = i_size_read(inode);
1166
1167         memset(pages, 0, num_pages * sizeof(struct page *));
1168
1169         for (i = 0; i < num_pages; i++) {
1170                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
1171                 if (!pages[i]) {
1172                         err = -ENOMEM;
1173                         goto failed_release;
1174                 }
1175                 offset = pos & (PAGE_CACHE_SIZE -1);
1176                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1177                 ret = nobh_prepare_write(pages[i], offset,
1178                                          offset + this_write,
1179                                          btrfs_get_block);
1180                 pos += this_write;
1181                 if (ret) {
1182                         err = ret;
1183                         goto failed_truncate;
1184                 }
1185                 WARN_ON(this_write > write_bytes);
1186                 write_bytes -= this_write;
1187         }
1188         return 0;
1189
1190 failed_release:
1191         btrfs_drop_pages(pages, num_pages);
1192         return err;
1193
1194 failed_truncate:
1195         btrfs_drop_pages(pages, num_pages);
1196         if (pos > isize)
1197                 vmtruncate(inode, isize);
1198         return err;
1199 }
1200
1201 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1202                                 size_t count, loff_t *ppos)
1203 {
1204         loff_t pos;
1205         size_t num_written = 0;
1206         int err = 0;
1207         int ret = 0;
1208         struct inode *inode = file->f_path.dentry->d_inode;
1209         struct btrfs_root *root = btrfs_sb(inode->i_sb);
1210         struct page *pages[1];
1211
1212         if (file->f_flags & O_DIRECT)
1213                 return -EINVAL;
1214         pos = *ppos;
1215
1216         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1217         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1218         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1219         if (err)
1220                 goto out;
1221         if (count == 0)
1222                 goto out;
1223         err = remove_suid(file->f_path.dentry);
1224         if (err)
1225                 goto out;
1226         file_update_time(file);
1227         mutex_lock(&inode->i_mutex);
1228         while(count > 0) {
1229                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1230                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1231                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1232                                         PAGE_CACHE_SHIFT;
1233                 ret = prepare_pages(NULL, root, file, pages, num_pages,
1234                                     pos, write_bytes);
1235                 BUG_ON(ret);
1236                 ret = btrfs_copy_from_user(pos, num_pages,
1237                                            write_bytes, pages, buf);
1238                 BUG_ON(ret);
1239
1240                 ret = dirty_and_release_pages(NULL, root, file, pages,
1241                                               num_pages, pos, write_bytes);
1242                 BUG_ON(ret);
1243                 btrfs_drop_pages(pages, num_pages);
1244
1245                 buf += write_bytes;
1246                 count -= write_bytes;
1247                 pos += write_bytes;
1248                 num_written += write_bytes;
1249
1250                 balance_dirty_pages_ratelimited(inode->i_mapping);
1251                 cond_resched();
1252         }
1253         mutex_unlock(&inode->i_mutex);
1254 out:
1255         *ppos = pos;
1256         current->backing_dev_info = NULL;
1257         return num_written ? num_written : err;
1258 }
1259
1260 #if 0
1261 static ssize_t inline_one_page(struct btrfs_root *root, struct inode *inode,
1262                            struct page *page, loff_t pos,
1263                            size_t offset, size_t write_bytes)
1264 {
1265         struct btrfs_path *path;
1266         struct btrfs_trans_handle *trans;
1267         struct btrfs_key key;
1268         struct btrfs_leaf *leaf;
1269         struct btrfs_key found_key;
1270         int ret;
1271         size_t copy_size = 0;
1272         char *dst = NULL;
1273         int err = 0;
1274         size_t num_written = 0;
1275
1276         path = btrfs_alloc_path();
1277         BUG_ON(!path);
1278         mutex_lock(&root->fs_info->fs_mutex);
1279         trans = btrfs_start_transaction(root, 1);
1280         key.objectid = inode->i_ino;
1281         key.flags = 0;
1282         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
1283
1284 again:
1285         key.offset = pos;
1286         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1287         if (ret < 0) {
1288                 err = ret;
1289                 goto out;
1290         }
1291         if (ret == 0) {
1292                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1293                 btrfs_disk_key_to_cpu(&found_key,
1294                                       &leaf->items[path->slots[0]].key);
1295                 copy_size = btrfs_item_size(leaf->items + path->slots[0]);
1296                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1297                 copy_size = min(write_bytes, copy_size);
1298                 goto copyit;
1299         } else {
1300                 int slot = path->slots[0];
1301                 if (slot > 0) {
1302                         slot--;
1303                 }
1304                 // FIXME find max key
1305                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1306                 btrfs_disk_key_to_cpu(&found_key,
1307                                       &leaf->items[slot].key);
1308                 if (found_key.objectid != inode->i_ino)
1309                         goto insert;
1310                 if (btrfs_key_type(&found_key) != BTRFS_INLINE_DATA_KEY)
1311                         goto insert;
1312                 copy_size = btrfs_item_size(leaf->items + slot);
1313                 if (found_key.offset + copy_size <= pos)
1314                         goto insert;
1315                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1316                 dst += pos - found_key.offset;
1317                 copy_size = copy_size - (pos - found_key.offset);
1318                 BUG_ON(copy_size < 0);
1319                 copy_size = min(write_bytes, copy_size);
1320                 WARN_ON(copy_size == 0);
1321                 goto copyit;
1322         }
1323 insert:
1324         btrfs_release_path(root, path);
1325         copy_size = min(write_bytes,
1326                         (size_t)BTRFS_LEAF_DATA_SIZE(root) -
1327                         sizeof(struct btrfs_item) * 4);
1328         ret = btrfs_insert_empty_item(trans, root, path, &key, copy_size);
1329         BUG_ON(ret);
1330         dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1331                              path->slots[0], char);
1332 copyit:
1333         WARN_ON(copy_size == 0);
1334         WARN_ON(dst + copy_size >
1335                 btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1336                                                  path->slots[0], char) +
1337                 btrfs_item_size(btrfs_buffer_leaf(path->nodes[0])->items +
1338                                                   path->slots[0]));
1339         btrfs_memcpy(root, path->nodes[0]->b_data, dst,
1340                      page_address(page) + offset, copy_size);
1341         mark_buffer_dirty(path->nodes[0]);
1342         btrfs_release_path(root, path);
1343         pos += copy_size;
1344         offset += copy_size;
1345         num_written += copy_size;
1346         write_bytes -= copy_size;
1347         if (write_bytes)
1348                 goto again;
1349 out:
1350         btrfs_free_path(path);
1351         ret = btrfs_end_transaction(trans, root);
1352         BUG_ON(ret);
1353         mutex_unlock(&root->fs_info->fs_mutex);
1354         return num_written ? num_written : err;
1355 }
1356
1357 static ssize_t btrfs_file_inline_write(struct file *file,
1358                                        const char __user *buf,
1359                                        size_t count, loff_t *ppos)
1360 {
1361         loff_t pos;
1362         size_t num_written = 0;
1363         int err = 0;
1364         int ret = 0;
1365         struct inode *inode = file->f_path.dentry->d_inode;
1366         struct btrfs_root *root = btrfs_sb(inode->i_sb);
1367         unsigned long page_index;
1368
1369         if (file->f_flags & O_DIRECT)
1370                 return -EINVAL;
1371         pos = *ppos;
1372
1373         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1374         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1375         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1376         if (err)
1377                 goto out;
1378         if (count == 0)
1379                 goto out;
1380         err = remove_suid(file->f_path.dentry);
1381         if (err)
1382                 goto out;
1383         file_update_time(file);
1384         mutex_lock(&inode->i_mutex);
1385         while(count > 0) {
1386                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1387                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1388                 struct page *page;
1389
1390                 page_index = pos >> PAGE_CACHE_SHIFT;
1391                 page = grab_cache_page(inode->i_mapping, page_index);
1392                 if (!PageUptodate(page)) {
1393                         ret = mpage_readpage(page, btrfs_get_block);
1394                         BUG_ON(ret);
1395                         lock_page(page);
1396                 }
1397                 ret = btrfs_copy_from_user(pos, 1,
1398                                            write_bytes, &page, buf);
1399                 BUG_ON(ret);
1400                 write_bytes = inline_one_page(root, inode, page, pos,
1401                                       offset, write_bytes);
1402                 SetPageUptodate(page);
1403                 if (write_bytes > 0 && pos + write_bytes > inode->i_size) {
1404                         i_size_write(inode, pos + write_bytes);
1405                         mark_inode_dirty(inode);
1406                 }
1407                 page_cache_release(page);
1408                 unlock_page(page);
1409                 if (write_bytes < 0)
1410                         goto out_unlock;
1411                 buf += write_bytes;
1412                 count -= write_bytes;
1413                 pos += write_bytes;
1414                 num_written += write_bytes;
1415
1416                 balance_dirty_pages_ratelimited(inode->i_mapping);
1417                 cond_resched();
1418         }
1419 out_unlock:
1420         mutex_unlock(&inode->i_mutex);
1421 out:
1422         *ppos = pos;
1423         current->backing_dev_info = NULL;
1424         return num_written ? num_written : err;
1425 }
1426 #endif
1427
1428 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
1429                         unsigned long offset, unsigned long size)
1430 {
1431         char *kaddr;
1432         unsigned long left, count = desc->count;
1433
1434         if (size > count)
1435                 size = count;
1436
1437         if (!PageChecked(page)) {
1438                 /* FIXME, do it per block */
1439                 struct btrfs_root *root = btrfs_sb(page->mapping->host->i_sb);
1440                 int ret = btrfs_csum_verify_file_block(root,
1441                                           page->mapping->host->i_ino,
1442                                           page->index << PAGE_CACHE_SHIFT,
1443                                           kmap(page), PAGE_CACHE_SIZE);
1444                 if (ret) {
1445                         printk("failed to verify ino %lu page %lu\n",
1446                                page->mapping->host->i_ino,
1447                                page->index);
1448                         memset(page_address(page), 0, PAGE_CACHE_SIZE);
1449                 }
1450                 SetPageChecked(page);
1451                 kunmap(page);
1452         }
1453         /*
1454          * Faults on the destination of a read are common, so do it before
1455          * taking the kmap.
1456          */
1457         if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1458                 kaddr = kmap_atomic(page, KM_USER0);
1459                 left = __copy_to_user_inatomic(desc->arg.buf,
1460                                                 kaddr + offset, size);
1461                 kunmap_atomic(kaddr, KM_USER0);
1462                 if (left == 0)
1463                         goto success;
1464         }
1465
1466         /* Do it the slow way */
1467         kaddr = kmap(page);
1468         left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1469         kunmap(page);
1470
1471         if (left) {
1472                 size -= left;
1473                 desc->error = -EFAULT;
1474         }
1475 success:
1476         desc->count = count - size;
1477         desc->written += size;
1478         desc->arg.buf += size;
1479         return size;
1480 }
1481
1482 /**
1483  * btrfs_file_aio_read - filesystem read routine
1484  * @iocb:       kernel I/O control block
1485  * @iov:        io vector request
1486  * @nr_segs:    number of segments in the iovec
1487  * @pos:        current file position
1488  */
1489 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1490                                    unsigned long nr_segs, loff_t pos)
1491 {
1492         struct file *filp = iocb->ki_filp;
1493         ssize_t retval;
1494         unsigned long seg;
1495         size_t count;
1496         loff_t *ppos = &iocb->ki_pos;
1497
1498         count = 0;
1499         for (seg = 0; seg < nr_segs; seg++) {
1500                 const struct iovec *iv = &iov[seg];
1501
1502                 /*
1503                  * If any segment has a negative length, or the cumulative
1504                  * length ever wraps negative then return -EINVAL.
1505                  */
1506                 count += iv->iov_len;
1507                 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1508                         return -EINVAL;
1509                 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1510                         continue;
1511                 if (seg == 0)
1512                         return -EFAULT;
1513                 nr_segs = seg;
1514                 count -= iv->iov_len;   /* This segment is no good */
1515                 break;
1516         }
1517         retval = 0;
1518         if (count) {
1519                 for (seg = 0; seg < nr_segs; seg++) {
1520                         read_descriptor_t desc;
1521
1522                         desc.written = 0;
1523                         desc.arg.buf = iov[seg].iov_base;
1524                         desc.count = iov[seg].iov_len;
1525                         if (desc.count == 0)
1526                                 continue;
1527                         desc.error = 0;
1528                         do_generic_file_read(filp, ppos, &desc,
1529                                              btrfs_read_actor);
1530                         retval += desc.written;
1531                         if (desc.error) {
1532                                 retval = retval ?: desc.error;
1533                                 break;
1534                         }
1535                 }
1536         }
1537         return retval;
1538 }
1539
1540 static struct kmem_cache *btrfs_inode_cachep;
1541 struct kmem_cache *btrfs_trans_handle_cachep;
1542 struct kmem_cache *btrfs_transaction_cachep;
1543 struct kmem_cache *btrfs_bit_radix_cachep;
1544 struct kmem_cache *btrfs_path_cachep;
1545
1546 /*
1547  * Called inside transaction, so use GFP_NOFS
1548  */
1549 static struct inode *btrfs_alloc_inode(struct super_block *sb)
1550 {
1551         struct btrfs_inode *ei;
1552
1553         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
1554         if (!ei)
1555                 return NULL;
1556         ei->magic = 0xDEADBEEF;
1557         ei->magic2 = 0xDEADBEAF;
1558         return &ei->vfs_inode;
1559 }
1560
1561 static void btrfs_destroy_inode(struct inode *inode)
1562 {
1563         struct btrfs_inode *ei = BTRFS_I(inode);
1564         WARN_ON(ei->magic != 0xDEADBEEF);
1565         WARN_ON(ei->magic2 != 0xDEADBEAF);
1566         WARN_ON(!list_empty(&inode->i_dentry));
1567         WARN_ON(inode->i_data.nrpages);
1568
1569         ei->magic = 0;
1570         ei->magic2 = 0;
1571         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
1572 }
1573
1574 static void init_once(void * foo, struct kmem_cache * cachep,
1575                       unsigned long flags)
1576 {
1577         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
1578
1579         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1580             SLAB_CTOR_CONSTRUCTOR) {
1581                 inode_init_once(&ei->vfs_inode);
1582         }
1583 }
1584
1585 static int init_inodecache(void)
1586 {
1587         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
1588                                              sizeof(struct btrfs_inode),
1589                                              0, (SLAB_RECLAIM_ACCOUNT|
1590                                                 SLAB_MEM_SPREAD),
1591                                              init_once, NULL);
1592         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
1593                                              sizeof(struct btrfs_trans_handle),
1594                                              0, (SLAB_RECLAIM_ACCOUNT|
1595                                                 SLAB_MEM_SPREAD),
1596                                              NULL, NULL);
1597         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
1598                                              sizeof(struct btrfs_transaction),
1599                                              0, (SLAB_RECLAIM_ACCOUNT|
1600                                                 SLAB_MEM_SPREAD),
1601                                              NULL, NULL);
1602         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
1603                                              sizeof(struct btrfs_transaction),
1604                                              0, (SLAB_RECLAIM_ACCOUNT|
1605                                                 SLAB_MEM_SPREAD),
1606                                              NULL, NULL);
1607         btrfs_bit_radix_cachep = kmem_cache_create("btrfs_radix",
1608                                              256,
1609                                              0, (SLAB_RECLAIM_ACCOUNT|
1610                                                 SLAB_MEM_SPREAD |
1611                                                 SLAB_DESTROY_BY_RCU),
1612                                              NULL, NULL);
1613         if (btrfs_inode_cachep == NULL || btrfs_trans_handle_cachep == NULL ||
1614             btrfs_transaction_cachep == NULL || btrfs_bit_radix_cachep == NULL)
1615                 return -ENOMEM;
1616         return 0;
1617 }
1618
1619 static void destroy_inodecache(void)
1620 {
1621         kmem_cache_destroy(btrfs_inode_cachep);
1622         kmem_cache_destroy(btrfs_trans_handle_cachep);
1623         kmem_cache_destroy(btrfs_transaction_cachep);
1624         kmem_cache_destroy(btrfs_bit_radix_cachep);
1625         kmem_cache_destroy(btrfs_path_cachep);
1626 }
1627
1628 static int btrfs_get_sb(struct file_system_type *fs_type,
1629         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1630 {
1631         return get_sb_bdev(fs_type, flags, dev_name, data,
1632                            btrfs_fill_super, mnt);
1633 }
1634
1635 static struct file_system_type btrfs_fs_type = {
1636         .owner          = THIS_MODULE,
1637         .name           = "btrfs",
1638         .get_sb         = btrfs_get_sb,
1639         .kill_sb        = kill_block_super,
1640         .fs_flags       = FS_REQUIRES_DEV,
1641 };
1642
1643 static struct super_operations btrfs_super_ops = {
1644         .statfs         = simple_statfs,
1645         .delete_inode   = btrfs_delete_inode,
1646         .put_super      = btrfs_put_super,
1647         .read_inode     = btrfs_read_locked_inode,
1648         .write_super    = btrfs_write_super,
1649         .sync_fs        = btrfs_sync_fs,
1650         .write_inode    = btrfs_write_inode,
1651         .alloc_inode    = btrfs_alloc_inode,
1652         .destroy_inode  = btrfs_destroy_inode,
1653 };
1654
1655 static struct inode_operations btrfs_dir_inode_operations = {
1656         .lookup         = btrfs_lookup,
1657         .create         = btrfs_create,
1658         .unlink         = btrfs_unlink,
1659         .mkdir          = btrfs_mkdir,
1660         .rmdir          = btrfs_rmdir,
1661 };
1662
1663 static struct file_operations btrfs_dir_file_operations = {
1664         .llseek         = generic_file_llseek,
1665         .read           = generic_read_dir,
1666         .readdir        = btrfs_readdir,
1667 };
1668
1669 static struct address_space_operations btrfs_aops = {
1670         .readpage       = btrfs_readpage,
1671         .readpages      = btrfs_readpages,
1672         .writepage      = btrfs_writepage,
1673         .sync_page      = block_sync_page,
1674         .prepare_write  = btrfs_prepare_write,
1675         .commit_write   = btrfs_commit_write,
1676 };
1677
1678 static struct inode_operations btrfs_file_inode_operations = {
1679         .truncate       = btrfs_truncate,
1680 };
1681
1682 static struct file_operations btrfs_file_operations = {
1683         .llseek         = generic_file_llseek,
1684         .read           = do_sync_read,
1685         .aio_read       = btrfs_file_aio_read,
1686         .write          = btrfs_file_write,
1687         .mmap           = generic_file_mmap,
1688         .open           = generic_file_open,
1689 };
1690
1691 static int __init init_btrfs_fs(void)
1692 {
1693         int err;
1694         printk("btrfs loaded!\n");
1695         err = init_inodecache();
1696         if (err)
1697                 return err;
1698         return register_filesystem(&btrfs_fs_type);
1699 }
1700
1701 static void __exit exit_btrfs_fs(void)
1702 {
1703         destroy_inodecache();
1704         unregister_filesystem(&btrfs_fs_type);
1705         printk("btrfs unloaded\n");
1706 }
1707
1708 module_init(init_btrfs_fs)
1709 module_exit(exit_btrfs_fs)
1710
1711 MODULE_LICENSE("GPL");