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_offset(&item->key) >
 486                     root->fs_info->highest_inode) {
 487 printk("stopping at highest inode %Lu\n", root->fs_info->highest_inode);
 488                         break;
 489                 }
 490                 if (btrfs_disk_key_type(&item->key) != BTRFS_DIR_INDEX_KEY)
 491                         continue;
 492                 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
 493                         continue;
 494                 filp->f_pos = btrfs_disk_key_offset(&item->key);
 495                 advance = 1;
 496                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 497                 over = filldir(dirent, (const char *)(di + 1),
 498                                btrfs_dir_name_len(di),
 499                                btrfs_disk_key_offset(&item->key),
 500                                btrfs_dir_objectid(di), d_type);
 501                 if (over)
 502                         goto nopos;
 503         }
 504         filp->f_pos++;
 505 nopos:
 506         ret = 0;
 507 err:
 508         btrfs_release_path(root, path);
 509         btrfs_free_path(path);
 510         mutex_unlock(&root->fs_info->fs_mutex);
 511         return ret;
 512 }
 513
 514 static void btrfs_put_super (struct super_block * sb)
 515 {
 516         struct btrfs_root *root = btrfs_sb(sb);
 517         int ret;
 518
 519         ret = close_ctree(root);
 520         if (ret) {
 521                 printk("close ctree returns %d\n", ret);
 522         }
 523         sb->s_fs_info = NULL;
 524 }
 525
 526 static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 527 {
 528         struct inode * inode;
 529         struct dentry * root_dentry;
 530         struct btrfs_super_block *disk_super;
 531         struct btrfs_root *root;
 532
 533         sb->s_maxbytes = MAX_LFS_FILESIZE;
 534         sb->s_magic = BTRFS_SUPER_MAGIC;
 535         sb->s_op = &btrfs_super_ops;
 536         sb->s_time_gran = 1;
 537
 538         root = open_ctree(sb);
 539
 540         if (!root) {
 541                 printk("btrfs: open_ctree failed\n");
 542                 return -EIO;
 543         }
 544         sb->s_fs_info = root;
 545         disk_super = root->fs_info->disk_super;
 546         printk("read in super total blocks %Lu root %Lu\n",
 547                btrfs_super_total_blocks(disk_super),
 548                btrfs_super_root_dir(disk_super));
 549
 550         inode = iget_locked(sb, btrfs_super_root_dir(disk_super));
 551         if (!inode)
 552                 return -ENOMEM;
 553         if (inode->i_state & I_NEW) {
 554                 btrfs_read_locked_inode(inode);
 555                 unlock_new_inode(inode);
 556         }
 557
 558         root_dentry = d_alloc_root(inode);
 559         if (!root_dentry) {
 560                 iput(inode);
 561                 return -ENOMEM;
 562         }
 563         sb->s_root = root_dentry;
 564
 565         return 0;
 566 }
 567
 568 static void fill_inode_item(struct btrfs_inode_item *item,
 569                             struct inode *inode)
 570 {
 571         btrfs_set_inode_uid(item, inode->i_uid);
 572         btrfs_set_inode_gid(item, inode->i_gid);
 573         btrfs_set_inode_size(item, inode->i_size);
 574         btrfs_set_inode_mode(item, inode->i_mode);
 575         btrfs_set_inode_nlink(item, inode->i_nlink);
 576         btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
 577         btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
 578         btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
 579         btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
 580         btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
 581         btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
 582         btrfs_set_inode_nblocks(item, inode->i_blocks);
 583         btrfs_set_inode_generation(item, inode->i_generation);
 584         check_inode(inode);
 585 }
 586
 587 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
 588                               struct btrfs_root *root,
 589                               struct inode *inode)
 590 {
 591         struct btrfs_inode_item *inode_item;
 592         struct btrfs_path *path;
 593         int ret;
 594
 595         path = btrfs_alloc_path();
 596         BUG_ON(!path);
 597         btrfs_init_path(path);
 598
 599         ret = btrfs_lookup_inode(trans, root, path, inode->i_ino, 1);
 600         if (ret) {
 601                 if (ret > 0)
 602                         ret = -ENOENT;
 603                 goto failed;
 604         }
 605
 606         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 607                                   path->slots[0],
 608                                   struct btrfs_inode_item);
 609
 610         fill_inode_item(inode_item, inode);
 611         btrfs_mark_buffer_dirty(path->nodes[0]);
 612 failed:
 613         btrfs_release_path(root, path);
 614         btrfs_free_path(path);
 615         check_inode(inode);
 616         return 0;
 617 }
 618
 619 static int btrfs_write_inode(struct inode *inode, int wait)
 620 {
 621         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 622         struct btrfs_trans_handle *trans;
 623         int ret;
 624
 625         mutex_lock(&root->fs_info->fs_mutex);
 626         trans = btrfs_start_transaction(root, 1);
 627         ret = btrfs_update_inode(trans, root, inode);
 628         if (wait)
 629                 btrfs_commit_transaction(trans, root);
 630         else
 631                 btrfs_end_transaction(trans, root);
 632         mutex_unlock(&root->fs_info->fs_mutex);
 633         check_inode(inode);
 634         return ret;
 635 }
 636
 637 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 638                                      struct inode *dir, int mode)
 639 {
 640         struct inode *inode;
 641         struct btrfs_inode_item inode_item;
 642         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 643         struct btrfs_key key;
 644         int ret;
 645         u64 objectid;
 646
 647         inode = new_inode(dir->i_sb);
 648         if (!inode)
 649                 return ERR_PTR(-ENOMEM);
 650
 651         check_inode(inode);
 652         ret = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 653         BUG_ON(ret);
 654
 655         inode->i_uid = current->fsuid;
 656         inode->i_gid = current->fsgid;
 657         inode->i_mode = mode;
 658         inode->i_ino = objectid;
 659         inode->i_blocks = 0;
 660         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME_SEC;
 661         fill_inode_item(&inode_item, inode);
 662
 663         key.objectid = objectid;
 664         key.flags = 0;
 665         key.offset = 0;
 666         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 667         ret = btrfs_insert_inode_map(trans, root, objectid, &key);
 668         BUG_ON(ret);
 669
 670         ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
 671         BUG_ON(ret);
 672
 673         insert_inode_hash(inode);
 674         check_inode(inode);
 675         check_inode(dir);
 676         return inode;
 677 }
 678
 679 static int btrfs_add_link(struct btrfs_trans_handle *trans,
 680                             struct dentry *dentry, struct inode *inode)
 681 {
 682         int ret;
 683         ret = btrfs_insert_dir_item(trans, btrfs_sb(inode->i_sb),
 684                                     dentry->d_name.name, dentry->d_name.len,
 685                                     dentry->d_parent->d_inode->i_ino,
 686                                     inode->i_ino, 0);
 687         if (ret == 0) {
 688                 dentry->d_parent->d_inode->i_size += dentry->d_name.len * 2;
 689                 ret = btrfs_update_inode(trans, btrfs_sb(inode->i_sb),
 690                                          dentry->d_parent->d_inode);
 691         }
 692         check_inode(inode);
 693         check_inode(dentry->d_parent->d_inode);
 694         return ret;
 695 }
 696
 697 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
 698                             struct dentry *dentry, struct inode *inode)
 699 {
 700         int err = btrfs_add_link(trans, dentry, inode);
 701         if (!err) {
 702                 d_instantiate(dentry, inode);
 703                 return 0;
 704         }
 705         if (err > 0)
 706                 err = -EEXIST;
 707         check_inode(inode);
 708         return err;
 709 }
 710
 711 static int btrfs_create(struct inode *dir, struct dentry *dentry,
 712                         int mode, struct nameidata *nd)
 713 {
 714         struct btrfs_trans_handle *trans;
 715         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 716         struct inode *inode;
 717         int err;
 718         int drop_inode = 0;
 719
 720         mutex_lock(&root->fs_info->fs_mutex);
 721         trans = btrfs_start_transaction(root, 1);
 722         inode = btrfs_new_inode(trans, dir, mode);
 723         err = PTR_ERR(inode);
 724         if (IS_ERR(inode))
 725                 goto out_unlock;
 726         // FIXME mark the inode dirty
 727         err = btrfs_add_nondir(trans, dentry, inode);
 728         if (err)
 729                 drop_inode = 1;
 730         else {
 731                 inode->i_mapping->a_ops = &btrfs_aops;
 732                 inode->i_fop = &btrfs_file_operations;
 733                 inode->i_op = &btrfs_file_inode_operations;
 734         }
 735         dir->i_sb->s_dirt = 1;
 736 out_unlock:
 737         btrfs_end_transaction(trans, root);
 738         mutex_unlock(&root->fs_info->fs_mutex);
 739         check_inode(inode);
 740         check_inode(dir);
 741
 742         if (drop_inode) {
 743                 inode_dec_link_count(inode);
 744                 iput(inode);
 745         }
 746         return err;
 747 }
 748
 749 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
 750                                 struct inode *inode, struct inode *dir)
 751 {
 752         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 753         int ret;
 754         char buf[2];
 755         buf[0] = '.';
 756         buf[1] = '.';
 757
 758         ret = btrfs_insert_dir_item(trans, root, buf, 1, inode->i_ino,
 759                                     inode->i_ino, 1);
 760         if (ret)
 761                 goto error;
 762         ret = btrfs_insert_dir_item(trans, root, buf, 2, inode->i_ino,
 763                                     dir->i_ino, 1);
 764         if (ret)
 765                 goto error;
 766         inode->i_size = 6;
 767         ret = btrfs_update_inode(trans, root, inode);
 768 error:
 769         return ret;
 770 }
 771
 772 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 773 {
 774         struct inode *inode;
 775         struct btrfs_trans_handle *trans;
 776         struct btrfs_root *root = btrfs_sb(dir->i_sb);
 777         int err = 0;
 778         int drop_on_err = 0;
 779
 780         mutex_lock(&root->fs_info->fs_mutex);
 781         trans = btrfs_start_transaction(root, 1);
 782         if (IS_ERR(trans)) {
 783                 err = PTR_ERR(trans);
 784                 goto out_unlock;
 785         }
 786         inode = btrfs_new_inode(trans, dir, S_IFDIR | mode);
 787         if (IS_ERR(inode)) {
 788                 err = PTR_ERR(inode);
 789                 goto out_fail;
 790         }
 791         drop_on_err = 1;
 792         inode->i_op = &btrfs_dir_inode_operations;
 793         inode->i_fop = &btrfs_dir_file_operations;
 794
 795         err = btrfs_make_empty_dir(trans, inode, dir);
 796         if (err)
 797                 goto out_fail;
 798         err = btrfs_add_link(trans, dentry, inode);
 799         if (err)
 800                 goto out_fail;
 801         d_instantiate(dentry, inode);
 802         drop_on_err = 0;
 803
 804 out_fail:
 805         btrfs_end_transaction(trans, root);
 806 out_unlock:
 807         mutex_unlock(&root->fs_info->fs_mutex);
 808         if (drop_on_err)
 809                 iput(inode);
 810         return err;
 811 }
 812
 813 static int btrfs_sync_fs(struct super_block *sb, int wait)
 814 {
 815         struct btrfs_trans_handle *trans;
 816         struct btrfs_root *root;
 817         int ret;
 818         root = btrfs_sb(sb);
 819
 820         sb->s_dirt = 0;
 821         if (!wait) {
 822                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 823                 return 0;
 824         }
 825         filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
 826         mutex_lock(&root->fs_info->fs_mutex);
 827         trans = btrfs_start_transaction(root, 1);
 828         ret = btrfs_commit_transaction(trans, root);
 829         sb->s_dirt = 0;
 830         BUG_ON(ret);
 831 printk("btrfs sync_fs\n");
 832         mutex_unlock(&root->fs_info->fs_mutex);
 833         return 0;
 834 }
 835
 836 #if 0
 837 static int btrfs_get_block_inline(struct inode *inode, sector_t iblock,
 838                            struct buffer_head *result, int create)
 839 {
 840         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 841         struct btrfs_path *path;
 842         struct btrfs_key key;
 843         struct btrfs_leaf *leaf;
 844         int num_bytes = result->b_size;
 845         int item_size;
 846         int ret;
 847         u64 pos;
 848         char *ptr;
 849         int copy_size;
 850         int err = 0;
 851         char *safe_ptr;
 852         char *data_ptr;
 853
 854         path = btrfs_alloc_path();
 855         BUG_ON(!path);
 856
 857         WARN_ON(create);
 858         if (create) {
 859                 return 0;
 860         }
 861         pos = iblock << inode->i_blkbits;
 862         key.objectid = inode->i_ino;
 863         key.flags = 0;
 864         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
 865         ptr = kmap(result->b_page);
 866         safe_ptr = ptr;
 867         ptr += (pos & (PAGE_CACHE_SIZE -1));
 868 again:
 869         key.offset = pos;
 870         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 871         if (ret) {
 872                 if (ret < 0)
 873                         err = ret;
 874                 else
 875                         err = 0;
 876                 goto out;
 877         }
 878         leaf = btrfs_buffer_leaf(path->nodes[0]);
 879         item_size = btrfs_item_size(leaf->items + path->slots[0]);
 880         copy_size = min(num_bytes, item_size);
 881         data_ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 882         WARN_ON(safe_ptr + PAGE_CACHE_SIZE < ptr + copy_size);
 883         memcpy(ptr, data_ptr, copy_size);
 884         pos += copy_size;
 885         num_bytes -= copy_size;
 886         WARN_ON(num_bytes < 0);
 887         ptr += copy_size;
 888         btrfs_release_path(root, path);
 889         if (num_bytes != 0) {
 890                 if (pos >= i_size_read(inode))
 891                         memset(ptr, 0, num_bytes);
 892                 else
 893                         goto again;
 894         }
 895         set_buffer_uptodate(result);
 896         map_bh(result, inode->i_sb, 0);
 897         err = 0;
 898 out:
 899         btrfs_free_path(path);
 900         kunmap(result->b_page);
 901         return err;
 902 }
 903 #endif
 904
 905 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
 906                            struct buffer_head *result, int create)
 907 {
 908         int ret;
 909         int err = 0;
 910         u64 blocknr;
 911         u64 extent_start = 0;
 912         u64 extent_end = 0;
 913         u64 objectid = inode->i_ino;
 914         struct btrfs_path *path;
 915         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 916         struct btrfs_trans_handle *trans = NULL;
 917         struct btrfs_file_extent_item *item;
 918         struct btrfs_leaf *leaf;
 919         struct btrfs_disk_key *found_key;
 920
 921         path = btrfs_alloc_path();
 922         BUG_ON(!path);
 923         btrfs_init_path(path);
 924         if (create)
 925                 trans = btrfs_start_transaction(root, 1);
 926
 927         ret = btrfs_lookup_file_extent(trans, root, path,
 928                                        inode->i_ino,
 929                                        iblock << inode->i_blkbits, 0);
 930         if (ret < 0) {
 931                 err = ret;
 932                 goto out;
 933         }
 934
 935         if (ret != 0) {
 936                 if (path->slots[0] == 0) {
 937                         btrfs_release_path(root, path);
 938                         goto allocate;
 939                 }
 940                 path->slots[0]--;
 941         }
 942
 943         item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 944                               struct btrfs_file_extent_item);
 945         leaf = btrfs_buffer_leaf(path->nodes[0]);
 946         blocknr = btrfs_file_extent_disk_blocknr(item);
 947         blocknr += btrfs_file_extent_offset(item);
 948
 949         /* exact match found, use it */
 950         if (ret == 0) {
 951                 err = 0;
 952                 map_bh(result, inode->i_sb, blocknr);
 953                 goto out;
 954         }
 955
 956         /* are we inside the extent that was found? */
 957         found_key = &leaf->items[path->slots[0]].key;
 958         if (btrfs_disk_key_objectid(found_key) != objectid ||
 959             btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY) {
 960                 extent_end = 0;
 961                 extent_start = 0;
 962                 btrfs_release_path(root, path);
 963                 goto allocate;
 964         }
 965
 966         extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
 967         extent_start = extent_start >> inode->i_blkbits;
 968         extent_start += btrfs_file_extent_offset(item);
 969         extent_end = extent_start + btrfs_file_extent_num_blocks(item);
 970         if (iblock >= extent_start && iblock < extent_end) {
 971                 err = 0;
 972                 map_bh(result, inode->i_sb, blocknr + iblock - extent_start);
 973                 goto out;
 974         }
 975 allocate:
 976         /* ok, create a new extent */
 977         if (!create) {
 978                 err = 0;
 979                 goto out;
 980         }
 981         ret = btrfs_alloc_file_extent(trans, root, objectid,
 982                                       iblock << inode->i_blkbits,
 983                                       1, extent_end, &blocknr);
 984         if (ret) {
 985                 err = ret;
 986                 goto out;
 987         }
 988         inode->i_blocks += inode->i_sb->s_blocksize >> 9;
 989         set_buffer_new(result);
 990         map_bh(result, inode->i_sb, blocknr);
 991
 992 out:
 993         btrfs_release_path(root, path);
 994         btrfs_free_path(path);
 995         if (trans)
 996                 btrfs_end_transaction(trans, root);
 997         return err;
 998 }
 999
1000 static int btrfs_get_block(struct inode *inode, sector_t iblock,
1001                            struct buffer_head *result, int create)
1002 {
1003         int err;
1004         struct btrfs_root *root = btrfs_sb(inode->i_sb);
1005         mutex_lock(&root->fs_info->fs_mutex);
1006         err = btrfs_get_block_lock(inode, iblock, result, create);
1007         // err = btrfs_get_block_inline(inode, iblock, result, create);
1008         mutex_unlock(&root->fs_info->fs_mutex);
1009         return err;
1010 }
1011
1012 static int btrfs_prepare_write(struct file *file, struct page *page,
1013                                unsigned from, unsigned to)
1014 {
1015         return nobh_prepare_write(page, from, to, btrfs_get_block);
1016 }
1017 static int btrfs_commit_write(struct file *file, struct page *page,
1018                                unsigned from, unsigned to)
1019 {
1020         return nobh_commit_write(file, page, from, to);
1021 }
1022
1023 static void btrfs_write_super(struct super_block *sb)
1024 {
1025         btrfs_sync_fs(sb, 1);
1026 }
1027
1028 static int btrfs_readpage(struct file *file, struct page *page)
1029 {
1030         return mpage_readpage(page, btrfs_get_block);
1031 }
1032
1033 static int btrfs_readpages(struct file *file, struct address_space *mapping,
1034                            struct list_head *pages, unsigned nr_pages)
1035 {
1036         return mpage_readpages(mapping, pages, nr_pages, btrfs_get_block);
1037 }
1038
1039 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1040 {
1041         return nobh_writepage(page, btrfs_get_block, wbc);
1042 }
1043
1044 static void btrfs_truncate(struct inode *inode)
1045 {
1046         struct btrfs_root *root = btrfs_sb(inode->i_sb);
1047         int ret;
1048         struct btrfs_trans_handle *trans;
1049
1050         if (!S_ISREG(inode->i_mode))
1051                 return;
1052         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1053                 return;
1054
1055         nobh_truncate_page(inode->i_mapping, inode->i_size);
1056
1057         /* FIXME, add redo link to tree so we don't leak on crash */
1058         mutex_lock(&root->fs_info->fs_mutex);
1059         trans = btrfs_start_transaction(root, 1);
1060         ret = btrfs_truncate_in_trans(trans, root, inode);
1061         BUG_ON(ret);
1062         ret = btrfs_end_transaction(trans, root);
1063         BUG_ON(ret);
1064         mutex_unlock(&root->fs_info->fs_mutex);
1065         mark_inode_dirty(inode);
1066 }
1067
1068 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
1069                                 struct page **prepared_pages,
1070                                 const char __user * buf)
1071 {
1072         long page_fault = 0;
1073         int i;
1074         int offset = pos & (PAGE_CACHE_SIZE - 1);
1075
1076         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
1077                 size_t count = min_t(size_t,
1078                                      PAGE_CACHE_SIZE - offset, write_bytes);
1079                 struct page *page = prepared_pages[i];
1080                 fault_in_pages_readable(buf, count);
1081
1082                 /* Copy data from userspace to the current page */
1083                 kmap(page);
1084                 page_fault = __copy_from_user(page_address(page) + offset,
1085                                               buf, count);
1086                 /* Flush processor's dcache for this page */
1087                 flush_dcache_page(page);
1088                 kunmap(page);
1089                 buf += count;
1090                 write_bytes -= count;
1091
1092                 if (page_fault)
1093                         break;
1094         }
1095         return page_fault ? -EFAULT : 0;
1096 }
1097
1098 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
1099 {
1100         size_t i;
1101         for (i = 0; i < num_pages; i++) {
1102                 if (!pages[i])
1103                         break;
1104                 unlock_page(pages[i]);
1105                 mark_page_accessed(pages[i]);
1106                 page_cache_release(pages[i]);
1107         }
1108 }
1109 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
1110                                    struct btrfs_root *root,
1111                                    struct file *file,
1112                                    struct page **pages,
1113                                    size_t num_pages,
1114                                    loff_t pos,
1115                                    size_t write_bytes)
1116 {
1117         int i;
1118         int offset;
1119         int err = 0;
1120         int ret;
1121         int this_write;
1122         struct inode *inode = file->f_path.dentry->d_inode;
1123
1124         for (i = 0; i < num_pages; i++) {
1125                 offset = pos & (PAGE_CACHE_SIZE -1);
1126                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1127                 /* FIXME, one block at a time */
1128
1129                 mutex_lock(&root->fs_info->fs_mutex);
1130                 trans = btrfs_start_transaction(root, 1);
1131                 btrfs_csum_file_block(trans, root, inode->i_ino,
1132                                       pages[i]->index << PAGE_CACHE_SHIFT,
1133                                       kmap(pages[i]), PAGE_CACHE_SIZE);
1134                 kunmap(pages[i]);
1135                 SetPageChecked(pages[i]);
1136                 ret = btrfs_end_transaction(trans, root);
1137                 BUG_ON(ret);
1138                 mutex_unlock(&root->fs_info->fs_mutex);
1139
1140                 ret = nobh_commit_write(file, pages[i], offset,
1141                                          offset + this_write);
1142                 pos += this_write;
1143                 if (ret) {
1144                         err = ret;
1145                         goto failed;
1146                 }
1147                 WARN_ON(this_write > write_bytes);
1148                 write_bytes -= this_write;
1149         }
1150 failed:
1151         return err;
1152 }
1153
1154 static int prepare_pages(struct btrfs_trans_handle *trans,
1155                          struct btrfs_root *root,
1156                          struct file *file,
1157                          struct page **pages,
1158                          size_t num_pages,
1159                          loff_t pos,
1160                          size_t write_bytes)
1161 {
1162         int i;
1163         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1164         struct inode *inode = file->f_path.dentry->d_inode;
1165         int offset;
1166         int err = 0;
1167         int ret;
1168         int this_write;
1169         loff_t isize = i_size_read(inode);
1170
1171         memset(pages, 0, num_pages * sizeof(struct page *));
1172
1173         for (i = 0; i < num_pages; i++) {
1174                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
1175                 if (!pages[i]) {
1176                         err = -ENOMEM;
1177                         goto failed_release;
1178                 }
1179                 offset = pos & (PAGE_CACHE_SIZE -1);
1180                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1181                 ret = nobh_prepare_write(pages[i], offset,
1182                                          offset + this_write,
1183                                          btrfs_get_block);
1184                 pos += this_write;
1185                 if (ret) {
1186                         err = ret;
1187                         goto failed_truncate;
1188                 }
1189                 WARN_ON(this_write > write_bytes);
1190                 write_bytes -= this_write;
1191         }
1192         return 0;
1193
1194 failed_release:
1195         btrfs_drop_pages(pages, num_pages);
1196         return err;
1197
1198 failed_truncate:
1199         btrfs_drop_pages(pages, num_pages);
1200         if (pos > isize)
1201                 vmtruncate(inode, isize);
1202         return err;
1203 }
1204
1205 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1206                                 size_t count, loff_t *ppos)
1207 {
1208         loff_t pos;
1209         size_t num_written = 0;
1210         int err = 0;
1211         int ret = 0;
1212         struct inode *inode = file->f_path.dentry->d_inode;
1213         struct btrfs_root *root = btrfs_sb(inode->i_sb);
1214         struct page *pages[1];
1215
1216         if (file->f_flags & O_DIRECT)
1217                 return -EINVAL;
1218         pos = *ppos;
1219
1220         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1221         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1222         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1223         if (err)
1224                 goto out;
1225         if (count == 0)
1226                 goto out;
1227         err = remove_suid(file->f_path.dentry);
1228         if (err)
1229                 goto out;
1230         file_update_time(file);
1231         mutex_lock(&inode->i_mutex);
1232         while(count > 0) {
1233                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1234                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1235                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1236                                         PAGE_CACHE_SHIFT;
1237                 ret = prepare_pages(NULL, root, file, pages, num_pages,
1238                                     pos, write_bytes);
1239                 BUG_ON(ret);
1240                 ret = btrfs_copy_from_user(pos, num_pages,
1241                                            write_bytes, pages, buf);
1242                 BUG_ON(ret);
1243
1244                 ret = dirty_and_release_pages(NULL, root, file, pages,
1245                                               num_pages, pos, write_bytes);
1246                 BUG_ON(ret);
1247                 btrfs_drop_pages(pages, num_pages);
1248
1249                 buf += write_bytes;
1250                 count -= write_bytes;
1251                 pos += write_bytes;
1252                 num_written += write_bytes;
1253
1254                 balance_dirty_pages_ratelimited(inode->i_mapping);
1255                 cond_resched();
1256         }
1257         mutex_unlock(&inode->i_mutex);
1258 out:
1259         *ppos = pos;
1260         current->backing_dev_info = NULL;
1261         return num_written ? num_written : err;
1262 }
1263
1264 #if 0
1265 static ssize_t inline_one_page(struct btrfs_root *root, struct inode *inode,
1266                            struct page *page, loff_t pos,
1267                            size_t offset, size_t write_bytes)
1268 {
1269         struct btrfs_path *path;
1270         struct btrfs_trans_handle *trans;
1271         struct btrfs_key key;
1272         struct btrfs_leaf *leaf;
1273         struct btrfs_key found_key;
1274         int ret;
1275         size_t copy_size = 0;
1276         char *dst = NULL;
1277         int err = 0;
1278         size_t num_written = 0;
1279
1280         path = btrfs_alloc_path();
1281         BUG_ON(!path);
1282         mutex_lock(&root->fs_info->fs_mutex);
1283         trans = btrfs_start_transaction(root, 1);
1284         key.objectid = inode->i_ino;
1285         key.flags = 0;
1286         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
1287
1288 again:
1289         key.offset = pos;
1290         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1291         if (ret < 0) {
1292                 err = ret;
1293                 goto out;
1294         }
1295         if (ret == 0) {
1296                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1297                 btrfs_disk_key_to_cpu(&found_key,
1298                                       &leaf->items[path->slots[0]].key);
1299                 copy_size = btrfs_item_size(leaf->items + path->slots[0]);
1300                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1301                 copy_size = min(write_bytes, copy_size);
1302                 goto copyit;
1303         } else {
1304                 int slot = path->slots[0];
1305                 if (slot > 0) {
1306                         slot--;
1307                 }
1308                 // FIXME find max key
1309                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1310                 btrfs_disk_key_to_cpu(&found_key,
1311                                       &leaf->items[slot].key);
1312                 if (found_key.objectid != inode->i_ino)
1313                         goto insert;
1314                 if (btrfs_key_type(&found_key) != BTRFS_INLINE_DATA_KEY)
1315                         goto insert;
1316                 copy_size = btrfs_item_size(leaf->items + slot);
1317                 if (found_key.offset + copy_size <= pos)
1318                         goto insert;
1319                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1320                 dst += pos - found_key.offset;
1321                 copy_size = copy_size - (pos - found_key.offset);
1322                 BUG_ON(copy_size < 0);
1323                 copy_size = min(write_bytes, copy_size);
1324                 WARN_ON(copy_size == 0);
1325                 goto copyit;
1326         }
1327 insert:
1328         btrfs_release_path(root, path);
1329         copy_size = min(write_bytes,
1330                         (size_t)BTRFS_LEAF_DATA_SIZE(root) -
1331                         sizeof(struct btrfs_item) * 4);
1332         ret = btrfs_insert_empty_item(trans, root, path, &key, copy_size);
1333         BUG_ON(ret);
1334         dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1335                              path->slots[0], char);
1336 copyit:
1337         WARN_ON(copy_size == 0);
1338         WARN_ON(dst + copy_size >
1339                 btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1340                                                  path->slots[0], char) +
1341                 btrfs_item_size(btrfs_buffer_leaf(path->nodes[0])->items +
1342                                                   path->slots[0]));
1343         btrfs_memcpy(root, path->nodes[0]->b_data, dst,
1344                      page_address(page) + offset, copy_size);
1345         mark_buffer_dirty(path->nodes[0]);
1346         btrfs_release_path(root, path);
1347         pos += copy_size;
1348         offset += copy_size;
1349         num_written += copy_size;
1350         write_bytes -= copy_size;
1351         if (write_bytes)
1352                 goto again;
1353 out:
1354         btrfs_free_path(path);
1355         ret = btrfs_end_transaction(trans, root);
1356         BUG_ON(ret);
1357         mutex_unlock(&root->fs_info->fs_mutex);
1358         return num_written ? num_written : err;
1359 }
1360
1361 static ssize_t btrfs_file_inline_write(struct file *file,
1362                                        const char __user *buf,
1363                                        size_t count, loff_t *ppos)
1364 {
1365         loff_t pos;
1366         size_t num_written = 0;
1367         int err = 0;
1368         int ret = 0;
1369         struct inode *inode = file->f_path.dentry->d_inode;
1370         struct btrfs_root *root = btrfs_sb(inode->i_sb);
1371         unsigned long page_index;
1372
1373         if (file->f_flags & O_DIRECT)
1374                 return -EINVAL;
1375         pos = *ppos;
1376
1377         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1378         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1379         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1380         if (err)
1381                 goto out;
1382         if (count == 0)
1383                 goto out;
1384         err = remove_suid(file->f_path.dentry);
1385         if (err)
1386                 goto out;
1387         file_update_time(file);
1388         mutex_lock(&inode->i_mutex);
1389         while(count > 0) {
1390                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1391                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1392                 struct page *page;
1393
1394                 page_index = pos >> PAGE_CACHE_SHIFT;
1395                 page = grab_cache_page(inode->i_mapping, page_index);
1396                 if (!PageUptodate(page)) {
1397                         ret = mpage_readpage(page, btrfs_get_block);
1398                         BUG_ON(ret);
1399                         lock_page(page);
1400                 }
1401                 ret = btrfs_copy_from_user(pos, 1,
1402                                            write_bytes, &page, buf);
1403                 BUG_ON(ret);
1404                 write_bytes = inline_one_page(root, inode, page, pos,
1405                                       offset, write_bytes);
1406                 SetPageUptodate(page);
1407                 if (write_bytes > 0 && pos + write_bytes > inode->i_size) {
1408                         i_size_write(inode, pos + write_bytes);
1409                         mark_inode_dirty(inode);
1410                 }
1411                 page_cache_release(page);
1412                 unlock_page(page);
1413                 if (write_bytes < 0)
1414                         goto out_unlock;
1415                 buf += write_bytes;
1416                 count -= write_bytes;
1417                 pos += write_bytes;
1418                 num_written += write_bytes;
1419
1420                 balance_dirty_pages_ratelimited(inode->i_mapping);
1421                 cond_resched();
1422         }
1423 out_unlock:
1424         mutex_unlock(&inode->i_mutex);
1425 out:
1426         *ppos = pos;
1427         current->backing_dev_info = NULL;
1428         return num_written ? num_written : err;
1429 }
1430 #endif
1431
1432 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
1433                         unsigned long offset, unsigned long size)
1434 {
1435         char *kaddr;
1436         unsigned long left, count = desc->count;
1437
1438         if (size > count)
1439                 size = count;
1440
1441         if (!PageChecked(page)) {
1442                 /* FIXME, do it per block */
1443                 struct btrfs_root *root = btrfs_sb(page->mapping->host->i_sb);
1444                 int ret = btrfs_csum_verify_file_block(root,
1445                                           page->mapping->host->i_ino,
1446                                           page->index << PAGE_CACHE_SHIFT,
1447                                           kmap(page), PAGE_CACHE_SIZE);
1448                 if (ret) {
1449                         printk("failed to verify ino %lu page %lu\n",
1450                                page->mapping->host->i_ino,
1451                                page->index);
1452                         memset(page_address(page), 0, PAGE_CACHE_SIZE);
1453                 }
1454                 SetPageChecked(page);
1455                 kunmap(page);
1456         }
1457         /*
1458          * Faults on the destination of a read are common, so do it before
1459          * taking the kmap.
1460          */
1461         if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1462                 kaddr = kmap_atomic(page, KM_USER0);
1463                 left = __copy_to_user_inatomic(desc->arg.buf,
1464                                                 kaddr + offset, size);
1465                 kunmap_atomic(kaddr, KM_USER0);
1466                 if (left == 0)
1467                         goto success;
1468         }
1469
1470         /* Do it the slow way */
1471         kaddr = kmap(page);
1472         left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1473         kunmap(page);
1474
1475         if (left) {
1476                 size -= left;
1477                 desc->error = -EFAULT;
1478         }
1479 success:
1480         desc->count = count - size;
1481         desc->written += size;
1482         desc->arg.buf += size;
1483         return size;
1484 }
1485
1486 /**
1487  * btrfs_file_aio_read - filesystem read routine
1488  * @iocb:       kernel I/O control block
1489  * @iov:        io vector request
1490  * @nr_segs:    number of segments in the iovec
1491  * @pos:        current file position
1492  */
1493 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1494                                    unsigned long nr_segs, loff_t pos)
1495 {
1496         struct file *filp = iocb->ki_filp;
1497         ssize_t retval;
1498         unsigned long seg;
1499         size_t count;
1500         loff_t *ppos = &iocb->ki_pos;
1501
1502         count = 0;
1503         for (seg = 0; seg < nr_segs; seg++) {
1504                 const struct iovec *iv = &iov[seg];
1505
1506                 /*
1507                  * If any segment has a negative length, or the cumulative
1508                  * length ever wraps negative then return -EINVAL.
1509                  */
1510                 count += iv->iov_len;
1511                 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1512                         return -EINVAL;
1513                 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1514                         continue;
1515                 if (seg == 0)
1516                         return -EFAULT;
1517                 nr_segs = seg;
1518                 count -= iv->iov_len;   /* This segment is no good */
1519                 break;
1520         }
1521         retval = 0;
1522         if (count) {
1523                 for (seg = 0; seg < nr_segs; seg++) {
1524                         read_descriptor_t desc;
1525
1526                         desc.written = 0;
1527                         desc.arg.buf = iov[seg].iov_base;
1528                         desc.count = iov[seg].iov_len;
1529                         if (desc.count == 0)
1530                                 continue;
1531                         desc.error = 0;
1532                         do_generic_file_read(filp, ppos, &desc,
1533                                              btrfs_read_actor);
1534                         retval += desc.written;
1535                         if (desc.error) {
1536                                 retval = retval ?: desc.error;
1537                                 break;
1538                         }
1539                 }
1540         }
1541         return retval;
1542 }
1543
1544 static struct kmem_cache *btrfs_inode_cachep;
1545 struct kmem_cache *btrfs_trans_handle_cachep;
1546 struct kmem_cache *btrfs_transaction_cachep;
1547 struct kmem_cache *btrfs_bit_radix_cachep;
1548 struct kmem_cache *btrfs_path_cachep;
1549
1550 /*
1551  * Called inside transaction, so use GFP_NOFS
1552  */
1553 static struct inode *btrfs_alloc_inode(struct super_block *sb)
1554 {
1555         struct btrfs_inode *ei;
1556
1557         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
1558         if (!ei)
1559                 return NULL;
1560         ei->magic = 0xDEADBEEF;
1561         ei->magic2 = 0xDEADBEAF;
1562         return &ei->vfs_inode;
1563 }
1564
1565 static void btrfs_destroy_inode(struct inode *inode)
1566 {
1567         struct btrfs_inode *ei = BTRFS_I(inode);
1568         WARN_ON(ei->magic != 0xDEADBEEF);
1569         WARN_ON(ei->magic2 != 0xDEADBEAF);
1570         WARN_ON(!list_empty(&inode->i_dentry));
1571         WARN_ON(inode->i_data.nrpages);
1572
1573         ei->magic = 0;
1574         ei->magic2 = 0;
1575         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
1576 }
1577
1578 static void init_once(void * foo, struct kmem_cache * cachep,
1579                       unsigned long flags)
1580 {
1581         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
1582
1583         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1584             SLAB_CTOR_CONSTRUCTOR) {
1585                 inode_init_once(&ei->vfs_inode);
1586         }
1587 }
1588
1589 static int init_inodecache(void)
1590 {
1591         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
1592                                              sizeof(struct btrfs_inode),
1593                                              0, (SLAB_RECLAIM_ACCOUNT|
1594                                                 SLAB_MEM_SPREAD),
1595                                              init_once, NULL);
1596         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
1597                                              sizeof(struct btrfs_trans_handle),
1598                                              0, (SLAB_RECLAIM_ACCOUNT|
1599                                                 SLAB_MEM_SPREAD),
1600                                              NULL, NULL);
1601         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
1602                                              sizeof(struct btrfs_transaction),
1603                                              0, (SLAB_RECLAIM_ACCOUNT|
1604                                                 SLAB_MEM_SPREAD),
1605                                              NULL, NULL);
1606         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
1607                                              sizeof(struct btrfs_transaction),
1608                                              0, (SLAB_RECLAIM_ACCOUNT|
1609                                                 SLAB_MEM_SPREAD),
1610                                              NULL, NULL);
1611         btrfs_bit_radix_cachep = kmem_cache_create("btrfs_radix",
1612                                              256,
1613                                              0, (SLAB_RECLAIM_ACCOUNT|
1614                                                 SLAB_MEM_SPREAD |
1615                                                 SLAB_DESTROY_BY_RCU),
1616                                              NULL, NULL);
1617         if (btrfs_inode_cachep == NULL || btrfs_trans_handle_cachep == NULL ||
1618             btrfs_transaction_cachep == NULL || btrfs_bit_radix_cachep == NULL)
1619                 return -ENOMEM;
1620         return 0;
1621 }
1622
1623 static void destroy_inodecache(void)
1624 {
1625         kmem_cache_destroy(btrfs_inode_cachep);
1626         kmem_cache_destroy(btrfs_trans_handle_cachep);
1627         kmem_cache_destroy(btrfs_transaction_cachep);
1628         kmem_cache_destroy(btrfs_bit_radix_cachep);
1629         kmem_cache_destroy(btrfs_path_cachep);
1630 }
1631
1632 static int btrfs_get_sb(struct file_system_type *fs_type,
1633         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
1634 {
1635         return get_sb_bdev(fs_type, flags, dev_name, data,
1636                            btrfs_fill_super, mnt);
1637 }
1638
1639 static struct file_system_type btrfs_fs_type = {
1640         .owner          = THIS_MODULE,
1641         .name           = "btrfs",
1642         .get_sb         = btrfs_get_sb,
1643         .kill_sb        = kill_block_super,
1644         .fs_flags       = FS_REQUIRES_DEV,
1645 };
1646
1647 static struct super_operations btrfs_super_ops = {
1648         .statfs         = simple_statfs,
1649         .delete_inode   = btrfs_delete_inode,
1650         .put_super      = btrfs_put_super,
1651         .read_inode     = btrfs_read_locked_inode,
1652         .write_super    = btrfs_write_super,
1653         .sync_fs        = btrfs_sync_fs,
1654         .write_inode    = btrfs_write_inode,
1655         .alloc_inode    = btrfs_alloc_inode,
1656         .destroy_inode  = btrfs_destroy_inode,
1657 };
1658
1659 static struct inode_operations btrfs_dir_inode_operations = {
1660         .lookup         = btrfs_lookup,
1661         .create         = btrfs_create,
1662         .unlink         = btrfs_unlink,
1663         .mkdir          = btrfs_mkdir,
1664         .rmdir          = btrfs_rmdir,
1665 };
1666
1667 static struct file_operations btrfs_dir_file_operations = {
1668         .llseek         = generic_file_llseek,
1669         .read           = generic_read_dir,
1670         .readdir        = btrfs_readdir,
1671 };
1672
1673 static struct address_space_operations btrfs_aops = {
1674         .readpage       = btrfs_readpage,
1675         .readpages      = btrfs_readpages,
1676         .writepage      = btrfs_writepage,
1677         .sync_page      = block_sync_page,
1678         .prepare_write  = btrfs_prepare_write,
1679         .commit_write   = btrfs_commit_write,
1680 };
1681
1682 static struct inode_operations btrfs_file_inode_operations = {
1683         .truncate       = btrfs_truncate,
1684 };
1685
1686 static struct file_operations btrfs_file_operations = {
1687         .llseek         = generic_file_llseek,
1688         .read           = do_sync_read,
1689         .aio_read       = btrfs_file_aio_read,
1690         .write          = btrfs_file_write,
1691         .mmap           = generic_file_mmap,
1692         .open           = generic_file_open,
1693 };
1694
1695 static int __init init_btrfs_fs(void)
1696 {
1697         int err;
1698         printk("btrfs loaded!\n");
1699         err = init_inodecache();
1700         if (err)
1701                 return err;
1702         return register_filesystem(&btrfs_fs_type);
1703 }
1704
1705 static void __exit exit_btrfs_fs(void)
1706 {
1707         destroy_inodecache();
1708         unregister_filesystem(&btrfs_fs_type);
1709         printk("btrfs unloaded\n");
1710 }
1711
1712 module_init(init_btrfs_fs)
1713 module_exit(exit_btrfs_fs)
1714
1715 MODULE_LICENSE("GPL");