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