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 #include "ioctl.h"
  19
  20 void btrfs_fsinfo_release(struct kobject *obj)
  21 {
  22         struct btrfs_fs_info *fsinfo = container_of(obj,
  23                                             struct btrfs_fs_info, kobj);
  24         kfree(fsinfo);
  25 }
  26
  27 struct kobj_type btrfs_fsinfo_ktype = {
  28         .release = btrfs_fsinfo_release,
  29 };
  30
  31 struct btrfs_iget_args {
  32         u64 ino;
  33         struct btrfs_root *root;
  34 };
  35
  36 decl_subsys(btrfs, &btrfs_fsinfo_ktype, NULL);
  37
  38 #define BTRFS_SUPER_MAGIC 0x9123682E
  39
  40 static struct inode_operations btrfs_dir_inode_operations;
  41 static struct inode_operations btrfs_dir_ro_inode_operations;
  42 static struct super_operations btrfs_super_ops;
  43 static struct file_operations btrfs_dir_file_operations;
  44 static struct inode_operations btrfs_file_inode_operations;
  45 static struct address_space_operations btrfs_aops;
  46 static struct file_operations btrfs_file_operations;
  47
  48 static void btrfs_read_locked_inode(struct inode *inode)
  49 {
  50         struct btrfs_path *path;
  51         struct btrfs_inode_item *inode_item;
  52         struct btrfs_root *root = BTRFS_I(inode)->root;
  53         struct btrfs_key location;
  54         int ret;
  55
  56         path = btrfs_alloc_path();
  57         BUG_ON(!path);
  58         btrfs_init_path(path);
  59         mutex_lock(&root->fs_info->fs_mutex);
  60
  61         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
  62         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
  63         if (ret) {
  64                 btrfs_free_path(path);
  65                 goto make_bad;
  66         }
  67         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
  68                                   path->slots[0],
  69                                   struct btrfs_inode_item);
  70
  71         inode->i_mode = btrfs_inode_mode(inode_item);
  72         inode->i_nlink = btrfs_inode_nlink(inode_item);
  73         inode->i_uid = btrfs_inode_uid(inode_item);
  74         inode->i_gid = btrfs_inode_gid(inode_item);
  75         inode->i_size = btrfs_inode_size(inode_item);
  76         inode->i_atime.tv_sec = btrfs_timespec_sec(&inode_item->atime);
  77         inode->i_atime.tv_nsec = btrfs_timespec_nsec(&inode_item->atime);
  78         inode->i_mtime.tv_sec = btrfs_timespec_sec(&inode_item->mtime);
  79         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(&inode_item->mtime);
  80         inode->i_ctime.tv_sec = btrfs_timespec_sec(&inode_item->ctime);
  81         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(&inode_item->ctime);
  82         inode->i_blocks = btrfs_inode_nblocks(inode_item);
  83         inode->i_generation = btrfs_inode_generation(inode_item);
  84
  85         btrfs_free_path(path);
  86         inode_item = NULL;
  87
  88         mutex_unlock(&root->fs_info->fs_mutex);
  89
  90         switch (inode->i_mode & S_IFMT) {
  91 #if 0
  92         default:
  93                 init_special_inode(inode, inode->i_mode,
  94                                    btrfs_inode_rdev(inode_item));
  95                 break;
  96 #endif
  97         case S_IFREG:
  98                 inode->i_mapping->a_ops = &btrfs_aops;
  99                 inode->i_fop = &btrfs_file_operations;
 100                 inode->i_op = &btrfs_file_inode_operations;
 101                 break;
 102         case S_IFDIR:
 103                 inode->i_fop = &btrfs_dir_file_operations;
 104                 if (root == root->fs_info->tree_root)
 105                         inode->i_op = &btrfs_dir_ro_inode_operations;
 106                 else
 107                         inode->i_op = &btrfs_dir_inode_operations;
 108                 break;
 109         case S_IFLNK:
 110                 // inode->i_op = &page_symlink_inode_operations;
 111                 break;
 112         }
 113         return;
 114
 115 make_bad:
 116         btrfs_release_path(root, path);
 117         btrfs_free_path(path);
 118         mutex_unlock(&root->fs_info->fs_mutex);
 119         make_bad_inode(inode);
 120 }
 121
 122 static int btrfs_unlink_trans(struct btrfs_trans_handle *trans,
 123                               struct btrfs_root *root,
 124                               struct inode *dir,
 125                               struct dentry *dentry)
 126 {
 127         struct btrfs_path *path;
 128         const char *name = dentry->d_name.name;
 129         int name_len = dentry->d_name.len;
 130         int ret;
 131         u64 objectid;
 132         struct btrfs_dir_item *di;
 133
 134         path = btrfs_alloc_path();
 135         BUG_ON(!path);
 136         btrfs_init_path(path);
 137         ret = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
 138                                     name, name_len, -1);
 139         if (ret < 0)
 140                 goto err;
 141         if (ret > 0) {
 142                 ret = -ENOENT;
 143                 goto err;
 144         }
 145         di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 146                             struct btrfs_dir_item);
 147         objectid = btrfs_disk_key_objectid(&di->location);
 148
 149         ret = btrfs_del_item(trans, root, path);
 150         BUG_ON(ret);
 151
 152         btrfs_release_path(root, path);
 153         dentry->d_inode->i_ctime = dir->i_ctime;
 154 err:
 155         btrfs_release_path(root, path);
 156         btrfs_free_path(path);
 157         if (ret == 0) {
 158                 inode_dec_link_count(dentry->d_inode);
 159                 dir->i_size -= name_len * 2;
 160                 mark_inode_dirty(dir);
 161         }
 162         return ret;
 163 }
 164
 165 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 166 {
 167         struct btrfs_root *root;
 168         struct btrfs_trans_handle *trans;
 169         int ret;
 170
 171         root = BTRFS_I(dir)->root;
 172         mutex_lock(&root->fs_info->fs_mutex);
 173         trans = btrfs_start_transaction(root, 1);
 174         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 175         btrfs_end_transaction(trans, root);
 176         mutex_unlock(&root->fs_info->fs_mutex);
 177         return ret;
 178 }
 179
 180 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 181 {
 182         struct inode *inode = dentry->d_inode;
 183         int err;
 184         int ret;
 185         struct btrfs_root *root = BTRFS_I(dir)->root;
 186         struct btrfs_path *path;
 187         struct btrfs_key key;
 188         struct btrfs_trans_handle *trans;
 189         struct btrfs_key found_key;
 190         int found_type;
 191         struct btrfs_leaf *leaf;
 192         char *goodnames = "..";
 193
 194         path = btrfs_alloc_path();
 195         BUG_ON(!path);
 196         btrfs_init_path(path);
 197         mutex_lock(&root->fs_info->fs_mutex);
 198         trans = btrfs_start_transaction(root, 1);
 199         key.objectid = inode->i_ino;
 200         key.offset = (u64)-1;
 201         key.flags = (u32)-1;
 202         while(1) {
 203                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 204                 if (ret < 0) {
 205                         err = ret;
 206                         goto out;
 207                 }
 208                 BUG_ON(ret == 0);
 209                 if (path->slots[0] == 0) {
 210                         err = -ENOENT;
 211                         goto out;
 212                 }
 213                 path->slots[0]--;
 214                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 215                 btrfs_disk_key_to_cpu(&found_key,
 216                                       &leaf->items[path->slots[0]].key);
 217                 found_type = btrfs_key_type(&found_key);
 218                 if (found_key.objectid != inode->i_ino) {
 219                         err = -ENOENT;
 220                         goto out;
 221                 }
 222                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
 223                      found_type != BTRFS_DIR_INDEX_KEY) ||
 224                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
 225                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
 226                         err = -ENOTEMPTY;
 227                         goto out;
 228                 }
 229                 ret = btrfs_del_item(trans, root, path);
 230                 BUG_ON(ret);
 231
 232                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
 233                         break;
 234                 btrfs_release_path(root, path);
 235         }
 236         ret = 0;
 237         btrfs_release_path(root, path);
 238
 239         /* now the directory is empty */
 240         err = btrfs_unlink_trans(trans, root, dir, dentry);
 241         if (!err) {
 242                 inode->i_size = 0;
 243         }
 244 out:
 245         btrfs_release_path(root, path);
 246         btrfs_free_path(path);
 247         mutex_unlock(&root->fs_info->fs_mutex);
 248         ret = btrfs_end_transaction(trans, root);
 249         if (ret && !err)
 250                 err = ret;
 251         return err;
 252 }
 253
 254 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 255                             struct btrfs_root *root,
 256                             struct inode *inode)
 257 {
 258         struct btrfs_path *path;
 259         int ret;
 260
 261         clear_inode(inode);
 262
 263         path = btrfs_alloc_path();
 264         BUG_ON(!path);
 265         btrfs_init_path(path);
 266         ret = btrfs_lookup_inode(trans, root, path,
 267                                  &BTRFS_I(inode)->location, -1);
 268         BUG_ON(ret);
 269         ret = btrfs_del_item(trans, root, path);
 270         BUG_ON(ret);
 271         btrfs_free_path(path);
 272         return ret;
 273 }
 274
 275 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 276                                    struct btrfs_root *root,
 277                                    struct inode *inode)
 278 {
 279         int ret;
 280         struct btrfs_path *path;
 281         struct btrfs_key key;
 282         struct btrfs_disk_key *found_key;
 283         struct btrfs_leaf *leaf;
 284         struct btrfs_file_extent_item *fi = NULL;
 285         u64 extent_start = 0;
 286         u64 extent_num_blocks = 0;
 287         int found_extent;
 288
 289         path = btrfs_alloc_path();
 290         BUG_ON(!path);
 291         /* FIXME, add redo link to tree so we don't leak on crash */
 292         key.objectid = inode->i_ino;
 293         key.offset = (u64)-1;
 294         key.flags = 0;
 295         /*
 296          * use BTRFS_CSUM_ITEM_KEY because it is larger than inline keys
 297          * or extent data
 298          */
 299         btrfs_set_key_type(&key, BTRFS_CSUM_ITEM_KEY);
 300         while(1) {
 301                 btrfs_init_path(path);
 302                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 303                 if (ret < 0) {
 304                         goto error;
 305                 }
 306                 if (ret > 0) {
 307                         BUG_ON(path->slots[0] == 0);
 308                         path->slots[0]--;
 309                 }
 310                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 311                 found_key = &leaf->items[path->slots[0]].key;
 312                 if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
 313                         break;
 314                 if (btrfs_disk_key_type(found_key) != BTRFS_CSUM_ITEM_KEY &&
 315                     btrfs_disk_key_type(found_key) != BTRFS_INLINE_DATA_KEY &&
 316                     btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY)
 317                         break;
 318                 if (btrfs_disk_key_offset(found_key) < inode->i_size)
 319                         break;
 320                 if (btrfs_disk_key_type(found_key) == BTRFS_EXTENT_DATA_KEY) {
 321                         fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 322                                             path->slots[0],
 323                                             struct btrfs_file_extent_item);
 324                         extent_start = btrfs_file_extent_disk_blocknr(fi);
 325                         extent_num_blocks =
 326                                 btrfs_file_extent_disk_num_blocks(fi);
 327                         /* FIXME blocksize != 4096 */
 328                         inode->i_blocks -=
 329                                 btrfs_file_extent_num_blocks(fi) << 3;
 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_I(inode)->root;
 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                                struct btrfs_key *location)
 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_I(dir)->root;
 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                 location->objectid = 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         btrfs_disk_key_to_cpu(location, &di->location);
 398 out:
 399         btrfs_release_path(root, path);
 400         btrfs_free_path(path);
 401         return ret;
 402 }
 403
 404 int fixup_tree_root_location(struct btrfs_root *root,
 405                              struct btrfs_key *location,
 406                              struct btrfs_root **sub_root)
 407 {
 408         struct btrfs_path *path;
 409         struct btrfs_root_item *ri;
 410
 411         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 412                 return 0;
 413         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 414                 return 0;
 415
 416         path = btrfs_alloc_path();
 417         BUG_ON(!path);
 418         mutex_lock(&root->fs_info->fs_mutex);
 419
 420         *sub_root = btrfs_read_fs_root(root->fs_info, location);
 421         if (IS_ERR(*sub_root))
 422                 return PTR_ERR(*sub_root);
 423
 424         ri = &(*sub_root)->root_item;
 425         location->objectid = btrfs_root_dirid(ri);
 426         location->flags = 0;
 427         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 428         location->offset = 0;
 429
 430         btrfs_free_path(path);
 431         mutex_unlock(&root->fs_info->fs_mutex);
 432         return 0;
 433 }
 434
 435 int btrfs_init_locked_inode(struct inode *inode, void *p)
 436 {
 437         struct btrfs_iget_args *args = p;
 438         inode->i_ino = args->ino;
 439         BTRFS_I(inode)->root = args->root;
 440         return 0;
 441 }
 442
 443 int btrfs_find_actor(struct inode *inode, void *opaque)
 444 {
 445         struct btrfs_iget_args *args = opaque;
 446         return (args->ino == inode->i_ino &&
 447                 args->root == BTRFS_I(inode)->root);
 448 }
 449
 450 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 451                                 struct btrfs_root *root)
 452 {
 453         struct inode *inode;
 454         struct btrfs_iget_args args;
 455         args.ino = objectid;
 456         args.root = root;
 457
 458         inode = iget5_locked(s, objectid, btrfs_find_actor,
 459                              btrfs_init_locked_inode,
 460                              (void *)&args);
 461         return inode;
 462 }
 463
 464 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 465                                    struct nameidata *nd)
 466 {
 467         struct inode * inode;
 468         struct btrfs_inode *bi = BTRFS_I(dir);
 469         struct btrfs_root *root = bi->root;
 470         struct btrfs_root *sub_root = root;
 471         struct btrfs_key location;
 472         int ret;
 473
 474         if (dentry->d_name.len > BTRFS_NAME_LEN)
 475                 return ERR_PTR(-ENAMETOOLONG);
 476         mutex_lock(&root->fs_info->fs_mutex);
 477         ret = btrfs_inode_by_name(dir, dentry, &location);
 478         mutex_unlock(&root->fs_info->fs_mutex);
 479         if (ret < 0)
 480                 return ERR_PTR(ret);
 481         inode = NULL;
 482         if (location.objectid) {
 483                 ret = fixup_tree_root_location(root, &location, &sub_root);
 484                 if (ret < 0)
 485                         return ERR_PTR(ret);
 486                 if (ret > 0)
 487                         return ERR_PTR(-ENOENT);
 488                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 489                                           sub_root);
 490                 if (!inode)
 491                         return ERR_PTR(-EACCES);
 492                 if (inode->i_state & I_NEW) {
 493                         if (sub_root != root) {
 494 printk("adding new root for inode %lu root %p (found %p)\n", inode->i_ino, sub_root, BTRFS_I(inode)->root);
 495                                 igrab(inode);
 496                                 sub_root->inode = inode;
 497                         }
 498                         BTRFS_I(inode)->root = sub_root;
 499                         memcpy(&BTRFS_I(inode)->location, &location,
 500                                sizeof(location));
 501                         btrfs_read_locked_inode(inode);
 502                         unlock_new_inode(inode);
 503                 }
 504         }
 505         return d_splice_alias(inode, dentry);
 506 }
 507
 508 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 509 {
 510         struct inode *inode = filp->f_path.dentry->d_inode;
 511         struct btrfs_root *root = BTRFS_I(inode)->root;
 512         struct btrfs_item *item;
 513         struct btrfs_dir_item *di;
 514         struct btrfs_key key;
 515         struct btrfs_path *path;
 516         int ret;
 517         u32 nritems;
 518         struct btrfs_leaf *leaf;
 519         int slot;
 520         int advance;
 521         unsigned char d_type = DT_UNKNOWN;
 522         int over = 0;
 523         int key_type = BTRFS_DIR_ITEM_KEY;
 524
 525         /* FIXME, use a real flag for deciding about the key type */
 526         if (root->fs_info->tree_root == root)
 527                 key_type = BTRFS_DIR_ITEM_KEY;
 528         mutex_lock(&root->fs_info->fs_mutex);
 529         key.objectid = inode->i_ino;
 530         key.flags = 0;
 531         btrfs_set_key_type(&key, key_type);
 532         key.offset = filp->f_pos;
 533         path = btrfs_alloc_path();
 534         btrfs_init_path(path);
 535         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 536         if (ret < 0)
 537                 goto err;
 538         advance = 0;
 539         while(1) {
 540                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 541                 nritems = btrfs_header_nritems(&leaf->header);
 542                 slot = path->slots[0];
 543                 if (advance || slot >= nritems) {
 544                         if (slot >= nritems -1) {
 545                                 ret = btrfs_next_leaf(root, path);
 546                                 if (ret)
 547                                         break;
 548                                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 549                                 nritems = btrfs_header_nritems(&leaf->header);
 550                                 slot = path->slots[0];
 551                         } else {
 552                                 slot++;
 553                                 path->slots[0]++;
 554                         }
 555                 }
 556                 advance = 1;
 557                 item = leaf->items + slot;
 558                 if (btrfs_disk_key_objectid(&item->key) != key.objectid)
 559                         break;
 560                 if (btrfs_disk_key_type(&item->key) != key_type)
 561                         break;
 562                 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
 563                         continue;
 564                 filp->f_pos = btrfs_disk_key_offset(&item->key);
 565                 advance = 1;
 566                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 567                 over = filldir(dirent, (const char *)(di + 1),
 568                                btrfs_dir_name_len(di),
 569                                btrfs_disk_key_offset(&item->key),
 570                                btrfs_disk_key_objectid(&di->location), d_type);
 571                 if (over)
 572                         goto nopos;
 573         }
 574         filp->f_pos++;
 575 nopos:
 576         ret = 0;
 577 err:
 578         btrfs_release_path(root, path);
 579         btrfs_free_path(path);
 580         mutex_unlock(&root->fs_info->fs_mutex);
 581         return ret;
 582 }
 583
 584 static void btrfs_put_super (struct super_block * sb)
 585 {
 586         struct btrfs_root *root = btrfs_sb(sb);
 587         int ret;
 588
 589         ret = close_ctree(root);
 590         if (ret) {
 591                 printk("close ctree returns %d\n", ret);
 592         }
 593         sb->s_fs_info = NULL;
 594 }
 595
 596 static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 597 {
 598         struct inode * inode;
 599         struct dentry * root_dentry;
 600         struct btrfs_super_block *disk_super;
 601         struct btrfs_root *tree_root;
 602         struct btrfs_inode *bi;
 603
 604         sb->s_maxbytes = MAX_LFS_FILESIZE;
 605         sb->s_magic = BTRFS_SUPER_MAGIC;
 606         sb->s_op = &btrfs_super_ops;
 607         sb->s_time_gran = 1;
 608
 609         tree_root = open_ctree(sb);
 610
 611         if (!tree_root) {
 612                 printk("btrfs: open_ctree failed\n");
 613                 return -EIO;
 614         }
 615         sb->s_fs_info = tree_root;
 616         disk_super = tree_root->fs_info->disk_super;
 617         printk("read in super total blocks %Lu root %Lu\n",
 618                btrfs_super_total_blocks(disk_super),
 619                btrfs_super_root_dir(disk_super));
 620
 621         inode = btrfs_iget_locked(sb, btrfs_super_root_dir(disk_super),
 622                                   tree_root);
 623         bi = BTRFS_I(inode);
 624         bi->location.objectid = inode->i_ino;
 625         bi->location.offset = 0;
 626         bi->location.flags = 0;
 627         bi->root = tree_root;
 628         btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
 629
 630         if (!inode)
 631                 return -ENOMEM;
 632         if (inode->i_state & I_NEW) {
 633                 btrfs_read_locked_inode(inode);
 634                 unlock_new_inode(inode);
 635         }
 636
 637         root_dentry = d_alloc_root(inode);
 638         if (!root_dentry) {
 639                 iput(inode);
 640                 return -ENOMEM;
 641         }
 642         sb->s_root = root_dentry;
 643
 644         return 0;
 645 }
 646
 647 static void fill_inode_item(struct btrfs_inode_item *item,
 648                             struct inode *inode)
 649 {
 650         btrfs_set_inode_uid(item, inode->i_uid);
 651         btrfs_set_inode_gid(item, inode->i_gid);
 652         btrfs_set_inode_size(item, inode->i_size);
 653         btrfs_set_inode_mode(item, inode->i_mode);
 654         btrfs_set_inode_nlink(item, inode->i_nlink);
 655         btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
 656         btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
 657         btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
 658         btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
 659         btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
 660         btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
 661         btrfs_set_inode_nblocks(item, inode->i_blocks);
 662         btrfs_set_inode_generation(item, inode->i_generation);
 663 }
 664
 665 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
 666                               struct btrfs_root *root,
 667                               struct inode *inode)
 668 {
 669         struct btrfs_inode_item *inode_item;
 670         struct btrfs_path *path;
 671         int ret;
 672
 673         path = btrfs_alloc_path();
 674         BUG_ON(!path);
 675         btrfs_init_path(path);
 676         ret = btrfs_lookup_inode(trans, root, path,
 677                                  &BTRFS_I(inode)->location, 1);
 678         if (ret) {
 679                 if (ret > 0)
 680                         ret = -ENOENT;
 681                 goto failed;
 682         }
 683
 684         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 685                                   path->slots[0],
 686                                   struct btrfs_inode_item);
 687
 688         fill_inode_item(inode_item, inode);
 689         btrfs_mark_buffer_dirty(path->nodes[0]);
 690         ret = 0;
 691 failed:
 692         btrfs_release_path(root, path);
 693         btrfs_free_path(path);
 694         return ret;
 695 }
 696
 697 static int btrfs_write_inode(struct inode *inode, int wait)
 698 {
 699         struct btrfs_root *root = BTRFS_I(inode)->root;
 700         struct btrfs_trans_handle *trans;
 701         int ret;
 702
 703         mutex_lock(&root->fs_info->fs_mutex);
 704         trans = btrfs_start_transaction(root, 1);
 705         ret = btrfs_update_inode(trans, root, inode);
 706         if (wait)
 707                 btrfs_commit_transaction(trans, root);
 708         else
 709                 btrfs_end_transaction(trans, root);
 710         mutex_unlock(&root->fs_info->fs_mutex);
 711         return ret;
 712 }
 713
 714 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 715                                      struct btrfs_root *root,
 716                                      u64 objectid, int mode)
 717 {
 718         struct inode *inode;
 719         struct btrfs_inode_item inode_item;
 720         struct btrfs_key *location;
 721         int ret;
 722
 723         inode = new_inode(root->fs_info->sb);
 724         if (!inode)
 725                 return ERR_PTR(-ENOMEM);
 726
 727         BTRFS_I(inode)->root = root;
 728
 729         inode->i_uid = current->fsuid;
 730         inode->i_gid = current->fsgid;
 731         inode->i_mode = mode;
 732         inode->i_ino = objectid;
 733         inode->i_blocks = 0;
 734         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 735         fill_inode_item(&inode_item, inode);
 736         location = &BTRFS_I(inode)->location;
 737         location->objectid = objectid;
 738         location->flags = 0;
 739         location->offset = 0;
 740         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 741
 742         ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
 743         BUG_ON(ret);
 744
 745         insert_inode_hash(inode);
 746         return inode;
 747 }
 748
 749 static int btrfs_add_link(struct btrfs_trans_handle *trans,
 750                             struct dentry *dentry, struct inode *inode)
 751 {
 752         int ret;
 753         struct btrfs_key key;
 754         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
 755         key.objectid = inode->i_ino;
 756         key.flags = 0;
 757         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 758         key.offset = 0;
 759
 760         ret = btrfs_insert_dir_item(trans, root,
 761                                     dentry->d_name.name, dentry->d_name.len,
 762                                     dentry->d_parent->d_inode->i_ino,
 763                                     &key, 0);
 764         if (ret == 0) {
 765                 dentry->d_parent->d_inode->i_size += dentry->d_name.len * 2;
 766                 ret = btrfs_update_inode(trans, root,
 767                                          dentry->d_parent->d_inode);
 768         }
 769         return ret;
 770 }
 771
 772 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
 773                             struct dentry *dentry, struct inode *inode)
 774 {
 775         int err = btrfs_add_link(trans, dentry, inode);
 776         if (!err) {
 777                 d_instantiate(dentry, inode);
 778                 return 0;
 779         }
 780         if (err > 0)
 781                 err = -EEXIST;
 782         return err;
 783 }
 784
 785 static int btrfs_create(struct inode *dir, struct dentry *dentry,
 786                         int mode, struct nameidata *nd)
 787 {
 788         struct btrfs_trans_handle *trans;
 789         struct btrfs_root *root = BTRFS_I(dir)->root;
 790         struct inode *inode;
 791         int err;
 792         int drop_inode = 0;
 793         u64 objectid;
 794
 795         mutex_lock(&root->fs_info->fs_mutex);
 796         trans = btrfs_start_transaction(root, 1);
 797
 798         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 799         if (err) {
 800                 err = -ENOSPC;
 801                 goto out_unlock;
 802         }
 803
 804         inode = btrfs_new_inode(trans, root, objectid, mode);
 805         err = PTR_ERR(inode);
 806         if (IS_ERR(inode))
 807                 goto out_unlock;
 808         // FIXME mark the inode dirty
 809         err = btrfs_add_nondir(trans, dentry, inode);
 810         if (err)
 811                 drop_inode = 1;
 812         else {
 813                 inode->i_mapping->a_ops = &btrfs_aops;
 814                 inode->i_fop = &btrfs_file_operations;
 815                 inode->i_op = &btrfs_file_inode_operations;
 816         }
 817         dir->i_sb->s_dirt = 1;
 818 out_unlock:
 819         btrfs_end_transaction(trans, root);
 820         mutex_unlock(&root->fs_info->fs_mutex);
 821
 822         if (drop_inode) {
 823                 inode_dec_link_count(inode);
 824                 iput(inode);
 825         }
 826         return err;
 827 }
 828
 829 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
 830                                 struct btrfs_root *root,
 831                                 u64 objectid, u64 dirid)
 832 {
 833         int ret;
 834         char buf[2];
 835         struct btrfs_key key;
 836
 837         buf[0] = '.';
 838         buf[1] = '.';
 839
 840         key.objectid = objectid;
 841         key.offset = 0;
 842         key.flags = 0;
 843         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 844
 845         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
 846                                     &key, 1);
 847         if (ret)
 848                 goto error;
 849         key.objectid = dirid;
 850         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
 851                                     &key, 1);
 852         if (ret)
 853                 goto error;
 854 error:
 855         return ret;
 856 }
 857
 858 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 859 {
 860         struct inode *inode;
 861         struct btrfs_trans_handle *trans;
 862         struct btrfs_root *root = BTRFS_I(dir)->root;
 863         int err = 0;
 864         int drop_on_err = 0;
 865         u64 objectid;
 866
 867         mutex_lock(&root->fs_info->fs_mutex);
 868         trans = btrfs_start_transaction(root, 1);
 869         if (IS_ERR(trans)) {
 870                 err = PTR_ERR(trans);
 871                 goto out_unlock;
 872         }
 873
 874         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 875         if (err) {
 876                 err = -ENOSPC;
 877                 goto out_unlock;
 878         }
 879
 880         inode = btrfs_new_inode(trans, root, objectid, S_IFDIR | mode);
 881         if (IS_ERR(inode)) {
 882                 err = PTR_ERR(inode);
 883                 goto out_fail;
 884         }
 885         drop_on_err = 1;
 886         inode->i_op = &btrfs_dir_inode_operations;
 887         inode->i_fop = &btrfs_dir_file_operations;
 888
 889         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
 890         if (err)
 891                 goto out_fail;
 892
 893         inode->i_size = 6;
 894         err = btrfs_update_inode(trans, root, inode);
 895         if (err)
 896                 goto out_fail;
 897         err = btrfs_add_link(trans, dentry, inode);
 898         if (err)
 899                 goto out_fail;
 900         d_instantiate(dentry, inode);
 901         drop_on_err = 0;
 902
 903 out_fail:
 904         btrfs_end_transaction(trans, root);
 905 out_unlock:
 906         mutex_unlock(&root->fs_info->fs_mutex);
 907         if (drop_on_err)
 908                 iput(inode);
 909         return err;
 910 }
 911
 912 static int btrfs_sync_fs(struct super_block *sb, int wait)
 913 {
 914         struct btrfs_trans_handle *trans;
 915         struct btrfs_root *root;
 916         int ret;
 917         root = btrfs_sb(sb);
 918
 919         sb->s_dirt = 0;
 920         if (!wait) {
 921                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 922                 return 0;
 923         }
 924         filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
 925         mutex_lock(&root->fs_info->fs_mutex);
 926         trans = btrfs_start_transaction(root, 1);
 927         ret = btrfs_commit_transaction(trans, root);
 928         sb->s_dirt = 0;
 929         BUG_ON(ret);
 930 printk("btrfs sync_fs\n");
 931         mutex_unlock(&root->fs_info->fs_mutex);
 932         return 0;
 933 }
 934
 935 #if 0
 936 static int btrfs_get_block_inline(struct inode *inode, sector_t iblock,
 937                            struct buffer_head *result, int create)
 938 {
 939         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 940         struct btrfs_path *path;
 941         struct btrfs_key key;
 942         struct btrfs_leaf *leaf;
 943         int num_bytes = result->b_size;
 944         int item_size;
 945         int ret;
 946         u64 pos;
 947         char *ptr;
 948         int copy_size;
 949         int err = 0;
 950         char *safe_ptr;
 951         char *data_ptr;
 952
 953         path = btrfs_alloc_path();
 954         BUG_ON(!path);
 955
 956         WARN_ON(create);
 957         if (create) {
 958                 return 0;
 959         }
 960         pos = iblock << inode->i_blkbits;
 961         key.objectid = inode->i_ino;
 962         key.flags = 0;
 963         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
 964         ptr = kmap(result->b_page);
 965         safe_ptr = ptr;
 966         ptr += (pos & (PAGE_CACHE_SIZE -1));
 967 again:
 968         key.offset = pos;
 969         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 970         if (ret) {
 971                 if (ret < 0)
 972                         err = ret;
 973                 else
 974                         err = 0;
 975                 goto out;
 976         }
 977         leaf = btrfs_buffer_leaf(path->nodes[0]);
 978         item_size = btrfs_item_size(leaf->items + path->slots[0]);
 979         copy_size = min(num_bytes, item_size);
 980         data_ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 981         WARN_ON(safe_ptr + PAGE_CACHE_SIZE < ptr + copy_size);
 982         memcpy(ptr, data_ptr, copy_size);
 983         pos += copy_size;
 984         num_bytes -= copy_size;
 985         WARN_ON(num_bytes < 0);
 986         ptr += copy_size;
 987         btrfs_release_path(root, path);
 988         if (num_bytes != 0) {
 989                 if (pos >= i_size_read(inode))
 990                         memset(ptr, 0, num_bytes);
 991                 else
 992                         goto again;
 993         }
 994         set_buffer_uptodate(result);
 995         map_bh(result, inode->i_sb, 0);
 996         err = 0;
 997 out:
 998         btrfs_free_path(path);
 999         kunmap(result->b_page);
1000         return err;
1001 }
1002 #endif
1003
1004 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
1005                            struct buffer_head *result, int create)
1006 {
1007         int ret;
1008         int err = 0;
1009         u64 blocknr;
1010         u64 extent_start = 0;
1011         u64 extent_end = 0;
1012         u64 objectid = inode->i_ino;
1013         struct btrfs_path *path;
1014         struct btrfs_root *root = BTRFS_I(inode)->root;
1015         struct btrfs_trans_handle *trans = NULL;
1016         struct btrfs_file_extent_item *item;
1017         struct btrfs_leaf *leaf;
1018         struct btrfs_disk_key *found_key;
1019
1020         path = btrfs_alloc_path();
1021         BUG_ON(!path);
1022         btrfs_init_path(path);
1023         if (create) {
1024                 trans = btrfs_start_transaction(root, 1);
1025                 WARN_ON(1);
1026         }
1027
1028         ret = btrfs_lookup_file_extent(trans, root, path,
1029                                        inode->i_ino,
1030                                        iblock << inode->i_blkbits, create);
1031         if (ret < 0) {
1032                 err = ret;
1033                 goto out;
1034         }
1035
1036         if (ret != 0) {
1037                 if (path->slots[0] == 0) {
1038                         btrfs_release_path(root, path);
1039                         goto allocate;
1040                 }
1041                 path->slots[0]--;
1042         }
1043
1044         item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
1045                               struct btrfs_file_extent_item);
1046         leaf = btrfs_buffer_leaf(path->nodes[0]);
1047         blocknr = btrfs_file_extent_disk_blocknr(item);
1048         blocknr += btrfs_file_extent_offset(item);
1049
1050         /* exact match found, use it, FIXME, deal with extents
1051          * other than the page size
1052          */
1053         if (0 && ret == 0) {
1054                 err = 0;
1055                 if (create &&
1056                     btrfs_file_extent_generation(item) != trans->transid) {
1057                         struct btrfs_key ins;
1058                         ret = btrfs_alloc_extent(trans, root, 1,
1059                                                  blocknr, (u64)-1, &ins);
1060                         BUG_ON(ret);
1061                         btrfs_set_file_extent_disk_blocknr(item, ins.objectid);
1062                         mark_buffer_dirty(path->nodes[0]);
1063                         ret = btrfs_free_extent(trans, root,
1064                                                 blocknr, 1, 0);
1065                         BUG_ON(ret);
1066                         blocknr = ins.objectid;
1067
1068                 }
1069                 btrfs_map_bh_to_logical(root, result, blocknr);
1070                 goto out;
1071         }
1072
1073         /* are we inside the extent that was found? */
1074         found_key = &leaf->items[path->slots[0]].key;
1075         if (btrfs_disk_key_objectid(found_key) != objectid ||
1076             btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY) {
1077                 extent_end = 0;
1078                 extent_start = 0;
1079                 btrfs_release_path(root, path);
1080                 goto allocate;
1081         }
1082
1083         extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
1084         extent_start = extent_start >> inode->i_blkbits;
1085         extent_end = extent_start + btrfs_file_extent_num_blocks(item);
1086         if (iblock >= extent_start && iblock < extent_end) {
1087                 err = 0;
1088                 btrfs_map_bh_to_logical(root, result, blocknr + iblock -
1089                                         extent_start);
1090                 goto out;
1091         }
1092 allocate:
1093         /* ok, create a new extent */
1094         if (!create) {
1095                 err = 0;
1096                 goto out;
1097         }
1098 #if 0
1099         ret = btrfs_alloc_file_extent(trans, root, objectid,
1100                                       iblock << inode->i_blkbits,
1101                                       1, extent_end, &blocknr);
1102         if (ret) {
1103                 err = ret;
1104                 goto out;
1105         }
1106         inode->i_blocks += inode->i_sb->s_blocksize >> 9;
1107         set_buffer_new(result);
1108         map_bh(result, inode->i_sb, blocknr);
1109
1110         btrfs_map_bh_to_logical(root, result, blocknr);
1111 #endif
1112 out:
1113         btrfs_release_path(root, path);
1114         btrfs_free_path(path);
1115         if (trans)
1116                 btrfs_end_transaction(trans, root);
1117         return err;
1118 }
1119
1120 static int btrfs_get_block(struct inode *inode, sector_t iblock,
1121                            struct buffer_head *result, int create)
1122 {
1123         int err;
1124         struct btrfs_root *root = BTRFS_I(inode)->root;
1125         mutex_lock(&root->fs_info->fs_mutex);
1126         err = btrfs_get_block_lock(inode, iblock, result, create);
1127         // err = btrfs_get_block_inline(inode, iblock, result, create);
1128         mutex_unlock(&root->fs_info->fs_mutex);
1129         return err;
1130 }
1131
1132 static int btrfs_prepare_write(struct file *file, struct page *page,
1133                                unsigned from, unsigned to)
1134 {
1135         return nobh_prepare_write(page, from, to, btrfs_get_block);
1136 }
1137 static int btrfs_commit_write(struct file *file, struct page *page,
1138                                unsigned from, unsigned to)
1139 {
1140         return nobh_commit_write(file, page, from, to);
1141 }
1142
1143 static void btrfs_write_super(struct super_block *sb)
1144 {
1145         btrfs_sync_fs(sb, 1);
1146 }
1147
1148 static int btrfs_readpage(struct file *file, struct page *page)
1149 {
1150         return mpage_readpage(page, btrfs_get_block);
1151 }
1152
1153 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1154 {
1155         return nobh_writepage(page, btrfs_get_block, wbc);
1156 }
1157
1158 static void btrfs_truncate(struct inode *inode)
1159 {
1160         struct btrfs_root *root = BTRFS_I(inode)->root;
1161         int ret;
1162         struct btrfs_trans_handle *trans;
1163
1164         if (!S_ISREG(inode->i_mode))
1165                 return;
1166         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1167                 return;
1168
1169         nobh_truncate_page(inode->i_mapping, inode->i_size);
1170
1171         /* FIXME, add redo link to tree so we don't leak on crash */
1172         mutex_lock(&root->fs_info->fs_mutex);
1173         trans = btrfs_start_transaction(root, 1);
1174         ret = btrfs_truncate_in_trans(trans, root, inode);
1175         BUG_ON(ret);
1176         ret = btrfs_end_transaction(trans, root);
1177         BUG_ON(ret);
1178         mutex_unlock(&root->fs_info->fs_mutex);
1179         mark_inode_dirty(inode);
1180 }
1181
1182 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
1183                                 struct page **prepared_pages,
1184                                 const char __user * buf)
1185 {
1186         long page_fault = 0;
1187         int i;
1188         int offset = pos & (PAGE_CACHE_SIZE - 1);
1189
1190         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
1191                 size_t count = min_t(size_t,
1192                                      PAGE_CACHE_SIZE - offset, write_bytes);
1193                 struct page *page = prepared_pages[i];
1194                 fault_in_pages_readable(buf, count);
1195
1196                 /* Copy data from userspace to the current page */
1197                 kmap(page);
1198                 page_fault = __copy_from_user(page_address(page) + offset,
1199                                               buf, count);
1200                 /* Flush processor's dcache for this page */
1201                 flush_dcache_page(page);
1202                 kunmap(page);
1203                 buf += count;
1204                 write_bytes -= count;
1205
1206                 if (page_fault)
1207                         break;
1208         }
1209         return page_fault ? -EFAULT : 0;
1210 }
1211
1212 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
1213 {
1214         size_t i;
1215         for (i = 0; i < num_pages; i++) {
1216                 if (!pages[i])
1217                         break;
1218                 unlock_page(pages[i]);
1219                 mark_page_accessed(pages[i]);
1220                 page_cache_release(pages[i]);
1221         }
1222 }
1223 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
1224                                    struct btrfs_root *root,
1225                                    struct file *file,
1226                                    struct page **pages,
1227                                    size_t num_pages,
1228                                    loff_t pos,
1229                                    size_t write_bytes)
1230 {
1231         int i;
1232         int offset;
1233         int err = 0;
1234         int ret;
1235         int this_write;
1236         struct inode *inode = file->f_path.dentry->d_inode;
1237
1238         for (i = 0; i < num_pages; i++) {
1239                 offset = pos & (PAGE_CACHE_SIZE -1);
1240                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1241                 /* FIXME, one block at a time */
1242
1243                 mutex_lock(&root->fs_info->fs_mutex);
1244                 trans = btrfs_start_transaction(root, 1);
1245                 btrfs_csum_file_block(trans, root, inode->i_ino,
1246                                       pages[i]->index << PAGE_CACHE_SHIFT,
1247                                       kmap(pages[i]), PAGE_CACHE_SIZE);
1248                 kunmap(pages[i]);
1249                 SetPageChecked(pages[i]);
1250                 ret = btrfs_end_transaction(trans, root);
1251                 BUG_ON(ret);
1252                 mutex_unlock(&root->fs_info->fs_mutex);
1253
1254                 ret = nobh_commit_write(file, pages[i], offset,
1255                                          offset + this_write);
1256                 pos += this_write;
1257                 if (ret) {
1258                         err = ret;
1259                         goto failed;
1260                 }
1261                 WARN_ON(this_write > write_bytes);
1262                 write_bytes -= this_write;
1263         }
1264 failed:
1265         return err;
1266 }
1267
1268 static int drop_extents(struct btrfs_trans_handle *trans,
1269                           struct btrfs_root *root,
1270                           struct inode *inode,
1271                           u64 start, u64 end)
1272 {
1273         int ret;
1274         struct btrfs_key key;
1275         struct btrfs_leaf *leaf;
1276         int slot;
1277         struct btrfs_file_extent_item *extent;
1278         u64 extent_end;
1279         int keep;
1280         struct btrfs_file_extent_item old;
1281         struct btrfs_path *path;
1282         u64 search_start = start;
1283         int bookend;
1284         path = btrfs_alloc_path();
1285         if (!path)
1286                 return -ENOMEM;
1287         while(1) {
1288                 btrfs_release_path(root, path);
1289                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1290                                                search_start, -1);
1291                 if (ret < 0)
1292                         goto out;
1293                 if (ret > 0) {
1294                         if (path->slots[0] == 0) {
1295                                 ret = -ENOENT;
1296                                 goto out;
1297                         }
1298                         path->slots[0]--;
1299                 }
1300                 keep = 0;
1301                 bookend = 0;
1302                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1303                 slot = path->slots[0];
1304                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
1305                 extent = btrfs_item_ptr(leaf, slot,
1306                                         struct btrfs_file_extent_item);
1307                 extent_end = key.offset +
1308                         (btrfs_file_extent_num_blocks(extent) <<
1309                          inode->i_blkbits);
1310                 if (key.offset >= end || key.objectid != inode->i_ino) {
1311                         ret = 0;
1312                         goto out;
1313                 }
1314                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1315                         goto out;
1316                 if (search_start >= extent_end)
1317                         goto out;
1318                 search_start = extent_end;
1319
1320                 if (end < extent_end && end >= key.offset) {
1321                         memcpy(&old, extent, sizeof(old));
1322                         ret = btrfs_inc_extent_ref(trans, root,
1323                                    btrfs_file_extent_disk_blocknr(&old),
1324                                    btrfs_file_extent_disk_num_blocks(&old));
1325                         BUG_ON(ret);
1326                         bookend = 1;
1327                 }
1328
1329                 if (start > key.offset) {
1330                         u64 new_num;
1331                         u64 old_num;
1332                         /* truncate existing extent */
1333                         keep = 1;
1334                         WARN_ON(start & (root->blocksize - 1));
1335                         new_num = (start - key.offset) >> inode->i_blkbits;
1336                         old_num = btrfs_file_extent_num_blocks(extent);
1337                         inode->i_blocks -= (old_num - new_num) << 3;
1338                         btrfs_set_file_extent_num_blocks(extent, new_num);
1339                         mark_buffer_dirty(path->nodes[0]);
1340                 }
1341                 if (!keep) {
1342                         u64 disk_blocknr;
1343                         u64 disk_num_blocks;
1344                         disk_blocknr = btrfs_file_extent_disk_blocknr(extent);
1345                         disk_num_blocks =
1346                                 btrfs_file_extent_disk_num_blocks(extent);
1347                         ret = btrfs_del_item(trans, root, path);
1348                         BUG_ON(ret);
1349                         inode->i_blocks -=
1350                                 btrfs_file_extent_num_blocks(extent) << 3;
1351                         btrfs_release_path(root, path);
1352                         ret = btrfs_free_extent(trans, root, disk_blocknr,
1353                                                 disk_num_blocks, 0);
1354
1355                         BUG_ON(ret);
1356                         if (!bookend && search_start >= end) {
1357                                 ret = 0;
1358                                 goto out;
1359                         }
1360                         if (!bookend)
1361                                 continue;
1362                 }
1363                 if (bookend) {
1364                         /* create bookend */
1365                         struct btrfs_key ins;
1366                         ins.objectid = inode->i_ino;
1367                         ins.offset = end;
1368                         ins.flags = 0;
1369                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
1370
1371                         btrfs_release_path(root, path);
1372                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
1373                                                       sizeof(*extent));
1374                         BUG_ON(ret);
1375                         extent = btrfs_item_ptr(
1376                                     btrfs_buffer_leaf(path->nodes[0]),
1377                                     path->slots[0],
1378                                     struct btrfs_file_extent_item);
1379                         btrfs_set_file_extent_disk_blocknr(extent,
1380                                     btrfs_file_extent_disk_blocknr(&old));
1381                         btrfs_set_file_extent_disk_num_blocks(extent,
1382                                     btrfs_file_extent_disk_num_blocks(&old));
1383
1384                         btrfs_set_file_extent_offset(extent,
1385                                     btrfs_file_extent_offset(&old) +
1386                                     ((end - key.offset) >> inode->i_blkbits));
1387                         WARN_ON(btrfs_file_extent_num_blocks(&old) <
1388                                 (end - key.offset) >> inode->i_blkbits);
1389                         btrfs_set_file_extent_num_blocks(extent,
1390                                     btrfs_file_extent_num_blocks(&old) -
1391                                     ((end - key.offset) >> inode->i_blkbits));
1392
1393                         btrfs_set_file_extent_generation(extent,
1394                                     btrfs_file_extent_generation(&old));
1395                         btrfs_mark_buffer_dirty(path->nodes[0]);
1396                         inode->i_blocks +=
1397                                 btrfs_file_extent_num_blocks(extent) << 3;
1398                         ret = 0;
1399                         goto out;
1400                 }
1401         }
1402 out:
1403         btrfs_free_path(path);
1404         return ret;
1405 }
1406
1407 static int prepare_pages(struct btrfs_root *root,
1408                          struct file *file,
1409                          struct page **pages,
1410                          size_t num_pages,
1411                          loff_t pos,
1412                          unsigned long first_index,
1413                          unsigned long last_index,
1414                          size_t write_bytes,
1415                          u64 alloc_extent_start)
1416 {
1417         int i;
1418         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1419         struct inode *inode = file->f_path.dentry->d_inode;
1420         int offset;
1421         int err = 0;
1422         int this_write;
1423         struct buffer_head *bh;
1424         struct buffer_head *head;
1425         loff_t isize = i_size_read(inode);
1426
1427         memset(pages, 0, num_pages * sizeof(struct page *));
1428
1429         for (i = 0; i < num_pages; i++) {
1430                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
1431                 if (!pages[i]) {
1432                         err = -ENOMEM;
1433                         goto failed_release;
1434                 }
1435                 offset = pos & (PAGE_CACHE_SIZE -1);
1436                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1437                 create_empty_buffers(pages[i], root->fs_info->sb->s_blocksize,
1438                                      (1 << BH_Uptodate));
1439                 head = page_buffers(pages[i]);
1440                 bh = head;
1441                 do {
1442                         err = btrfs_map_bh_to_logical(root, bh,
1443                                                       alloc_extent_start);
1444                         BUG_ON(err);
1445                         if (err)
1446                                 goto failed_truncate;
1447                         bh = bh->b_this_page;
1448                         alloc_extent_start++;
1449                 } while (bh != head);
1450                 pos += this_write;
1451                 WARN_ON(this_write > write_bytes);
1452                 write_bytes -= this_write;
1453         }
1454         return 0;
1455
1456 failed_release:
1457         btrfs_drop_pages(pages, num_pages);
1458         return err;
1459
1460 failed_truncate:
1461         btrfs_drop_pages(pages, num_pages);
1462         if (pos > isize)
1463                 vmtruncate(inode, isize);
1464         return err;
1465 }
1466
1467 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1468                                 size_t count, loff_t *ppos)
1469 {
1470         loff_t pos;
1471         size_t num_written = 0;
1472         int err = 0;
1473         int ret = 0;
1474         struct inode *inode = file->f_path.dentry->d_inode;
1475         struct btrfs_root *root = BTRFS_I(inode)->root;
1476         struct page *pages[8];
1477         struct page *pinned[2] = { NULL, NULL };
1478         unsigned long first_index;
1479         unsigned long last_index;
1480         u64 start_pos;
1481         u64 num_blocks;
1482         u64 alloc_extent_start;
1483         struct btrfs_trans_handle *trans;
1484         struct btrfs_key ins;
1485
1486         if (file->f_flags & O_DIRECT)
1487                 return -EINVAL;
1488         pos = *ppos;
1489         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1490         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1491         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1492         if (err)
1493                 goto out;
1494         if (count == 0)
1495                 goto out;
1496         err = remove_suid(file->f_path.dentry);
1497         if (err)
1498                 goto out;
1499         file_update_time(file);
1500
1501         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1502         num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
1503                         inode->i_blkbits;
1504
1505         mutex_lock(&inode->i_mutex);
1506         first_index = pos >> PAGE_CACHE_SHIFT;
1507         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1508
1509         if ((first_index << PAGE_CACHE_SHIFT) < inode->i_size &&
1510             (pos & (PAGE_CACHE_SIZE - 1))) {
1511                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1512                 if (!PageUptodate(pinned[0])) {
1513                         ret = mpage_readpage(pinned[0], btrfs_get_block);
1514                         BUG_ON(ret);
1515                 } else {
1516                         unlock_page(pinned[0]);
1517                 }
1518         }
1519         if (first_index != last_index &&
1520             (last_index << PAGE_CACHE_SHIFT) < inode->i_size &&
1521             (count & (PAGE_CACHE_SIZE - 1))) {
1522                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1523                 if (!PageUptodate(pinned[1])) {
1524                         ret = mpage_readpage(pinned[1], btrfs_get_block);
1525                         BUG_ON(ret);
1526                 } else {
1527                         unlock_page(pinned[1]);
1528                 }
1529         }
1530
1531         mutex_lock(&root->fs_info->fs_mutex);
1532         trans = btrfs_start_transaction(root, 1);
1533         if (!trans) {
1534                 err = -ENOMEM;
1535                 mutex_unlock(&root->fs_info->fs_mutex);
1536                 goto out_unlock;
1537         }
1538         /* FIXME blocksize != 4096 */
1539         inode->i_blocks += num_blocks << 3;
1540         if (start_pos < inode->i_size) {
1541                 /* FIXME blocksize != pagesize */
1542                 ret = drop_extents(trans, root, inode,
1543                                    start_pos,
1544                                    (pos + count + root->blocksize -1) &
1545                                    ~((u64)root->blocksize - 1));
1546         }
1547         ret = btrfs_alloc_extent(trans, root, num_blocks, 1,
1548                                  (u64)-1, &ins);
1549         BUG_ON(ret);
1550         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
1551                                        start_pos, ins.objectid, ins.offset);
1552         BUG_ON(ret);
1553         alloc_extent_start = ins.objectid;
1554         ret = btrfs_end_transaction(trans, root);
1555         mutex_unlock(&root->fs_info->fs_mutex);
1556
1557         while(count > 0) {
1558                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1559                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1560                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1561                                         PAGE_CACHE_SHIFT;
1562
1563                 memset(pages, 0, sizeof(pages));
1564                 ret = prepare_pages(root, file, pages, num_pages,
1565                                     pos, first_index, last_index,
1566                                     write_bytes, alloc_extent_start);
1567                 BUG_ON(ret);
1568
1569                 /* FIXME blocks != pagesize */
1570                 alloc_extent_start += num_pages;
1571                 ret = btrfs_copy_from_user(pos, num_pages,
1572                                            write_bytes, pages, buf);
1573                 BUG_ON(ret);
1574
1575                 ret = dirty_and_release_pages(NULL, root, file, pages,
1576                                               num_pages, pos, write_bytes);
1577                 BUG_ON(ret);
1578                 btrfs_drop_pages(pages, num_pages);
1579
1580                 buf += write_bytes;
1581                 count -= write_bytes;
1582                 pos += write_bytes;
1583                 num_written += write_bytes;
1584
1585                 balance_dirty_pages_ratelimited(inode->i_mapping);
1586                 cond_resched();
1587         }
1588 out_unlock:
1589         mutex_unlock(&inode->i_mutex);
1590 out:
1591         if (pinned[0])
1592                 page_cache_release(pinned[0]);
1593         if (pinned[1])
1594                 page_cache_release(pinned[1]);
1595         *ppos = pos;
1596         current->backing_dev_info = NULL;
1597         mark_inode_dirty(inode);
1598         return num_written ? num_written : err;
1599 }
1600
1601 #if 0
1602 static ssize_t inline_one_page(struct btrfs_root *root, struct inode *inode,
1603                            struct page *page, loff_t pos,
1604                            size_t offset, size_t write_bytes)
1605 {
1606         struct btrfs_path *path;
1607         struct btrfs_trans_handle *trans;
1608         struct btrfs_key key;
1609         struct btrfs_leaf *leaf;
1610         struct btrfs_key found_key;
1611         int ret;
1612         size_t copy_size = 0;
1613         char *dst = NULL;
1614         int err = 0;
1615         size_t num_written = 0;
1616
1617         path = btrfs_alloc_path();
1618         BUG_ON(!path);
1619         mutex_lock(&root->fs_info->fs_mutex);
1620         trans = btrfs_start_transaction(root, 1);
1621         key.objectid = inode->i_ino;
1622         key.flags = 0;
1623         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
1624
1625 again:
1626         key.offset = pos;
1627         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1628         if (ret < 0) {
1629                 err = ret;
1630                 goto out;
1631         }
1632         if (ret == 0) {
1633                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1634                 btrfs_disk_key_to_cpu(&found_key,
1635                                       &leaf->items[path->slots[0]].key);
1636                 copy_size = btrfs_item_size(leaf->items + path->slots[0]);
1637                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1638                 copy_size = min(write_bytes, copy_size);
1639                 goto copyit;
1640         } else {
1641                 int slot = path->slots[0];
1642                 if (slot > 0) {
1643                         slot--;
1644                 }
1645                 // FIXME find max key
1646                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1647                 btrfs_disk_key_to_cpu(&found_key,
1648                                       &leaf->items[slot].key);
1649                 if (found_key.objectid != inode->i_ino)
1650                         goto insert;
1651                 if (btrfs_key_type(&found_key) != BTRFS_INLINE_DATA_KEY)
1652                         goto insert;
1653                 copy_size = btrfs_item_size(leaf->items + slot);
1654                 if (found_key.offset + copy_size <= pos)
1655                         goto insert;
1656                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1657                 dst += pos - found_key.offset;
1658                 copy_size = copy_size - (pos - found_key.offset);
1659                 BUG_ON(copy_size < 0);
1660                 copy_size = min(write_bytes, copy_size);
1661                 WARN_ON(copy_size == 0);
1662                 goto copyit;
1663         }
1664 insert:
1665         btrfs_release_path(root, path);
1666         copy_size = min(write_bytes,
1667                         (size_t)BTRFS_LEAF_DATA_SIZE(root) -
1668                         sizeof(struct btrfs_item) * 4);
1669         ret = btrfs_insert_empty_item(trans, root, path, &key, copy_size);
1670         BUG_ON(ret);
1671         dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1672                              path->slots[0], char);
1673 copyit:
1674         WARN_ON(copy_size == 0);
1675         WARN_ON(dst + copy_size >
1676                 btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1677                                                  path->slots[0], char) +
1678                 btrfs_item_size(btrfs_buffer_leaf(path->nodes[0])->items +
1679                                                   path->slots[0]));
1680         btrfs_memcpy(root, path->nodes[0]->b_data, dst,
1681                      page_address(page) + offset, copy_size);
1682         mark_buffer_dirty(path->nodes[0]);
1683         btrfs_release_path(root, path);
1684         pos += copy_size;
1685         offset += copy_size;
1686         num_written += copy_size;
1687         write_bytes -= copy_size;
1688         if (write_bytes)
1689                 goto again;
1690 out:
1691         btrfs_free_path(path);
1692         ret = btrfs_end_transaction(trans, root);
1693         BUG_ON(ret);
1694         mutex_unlock(&root->fs_info->fs_mutex);
1695         return num_written ? num_written : err;
1696 }
1697
1698 static ssize_t btrfs_file_inline_write(struct file *file,
1699                                        const char __user *buf,
1700                                        size_t count, loff_t *ppos)
1701 {
1702         loff_t pos;
1703         size_t num_written = 0;
1704         int err = 0;
1705         int ret = 0;
1706         struct inode *inode = file->f_path.dentry->d_inode;
1707         struct btrfs_root *root = BTRFS_I(inode)->root;
1708         unsigned long page_index;
1709
1710         if (file->f_flags & O_DIRECT)
1711                 return -EINVAL;
1712         pos = *ppos;
1713
1714         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1715         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1716         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1717         if (err)
1718                 goto out;
1719         if (count == 0)
1720                 goto out;
1721         err = remove_suid(file->f_path.dentry);
1722         if (err)
1723                 goto out;
1724         file_update_time(file);
1725         mutex_lock(&inode->i_mutex);
1726         while(count > 0) {
1727                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1728                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1729                 struct page *page;
1730
1731                 page_index = pos >> PAGE_CACHE_SHIFT;
1732                 page = grab_cache_page(inode->i_mapping, page_index);
1733                 if (!PageUptodate(page)) {
1734                         ret = mpage_readpage(page, btrfs_get_block);
1735                         BUG_ON(ret);
1736                         lock_page(page);
1737                 }
1738                 ret = btrfs_copy_from_user(pos, 1,
1739                                            write_bytes, &page, buf);
1740                 BUG_ON(ret);
1741                 write_bytes = inline_one_page(root, inode, page, pos,
1742                                       offset, write_bytes);
1743                 SetPageUptodate(page);
1744                 if (write_bytes > 0 && pos + write_bytes > inode->i_size) {
1745                         i_size_write(inode, pos + write_bytes);
1746                         mark_inode_dirty(inode);
1747                 }
1748                 page_cache_release(page);
1749                 unlock_page(page);
1750                 if (write_bytes < 0)
1751                         goto out_unlock;
1752                 buf += write_bytes;
1753                 count -= write_bytes;
1754                 pos += write_bytes;
1755                 num_written += write_bytes;
1756
1757                 balance_dirty_pages_ratelimited(inode->i_mapping);
1758                 cond_resched();
1759         }
1760 out_unlock:
1761         mutex_unlock(&inode->i_mutex);
1762 out:
1763         *ppos = pos;
1764         current->backing_dev_info = NULL;
1765         return num_written ? num_written : err;
1766 }
1767 #endif
1768
1769 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
1770                         unsigned long offset, unsigned long size)
1771 {
1772         char *kaddr;
1773         unsigned long left, count = desc->count;
1774         struct inode *inode = page->mapping->host;
1775
1776         if (size > count)
1777                 size = count;
1778
1779         if (!PageChecked(page)) {
1780                 /* FIXME, do it per block */
1781                 struct btrfs_root *root = BTRFS_I(inode)->root;
1782                 int ret = btrfs_csum_verify_file_block(root,
1783                                           page->mapping->host->i_ino,
1784                                           page->index << PAGE_CACHE_SHIFT,
1785                                           kmap(page), PAGE_CACHE_SIZE);
1786                 if (ret) {
1787                         printk("failed to verify ino %lu page %lu\n",
1788                                page->mapping->host->i_ino,
1789                                page->index);
1790                         memset(page_address(page), 0, PAGE_CACHE_SIZE);
1791                 }
1792                 SetPageChecked(page);
1793                 kunmap(page);
1794         }
1795         /*
1796          * Faults on the destination of a read are common, so do it before
1797          * taking the kmap.
1798          */
1799         if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1800                 kaddr = kmap_atomic(page, KM_USER0);
1801                 left = __copy_to_user_inatomic(desc->arg.buf,
1802                                                 kaddr + offset, size);
1803                 kunmap_atomic(kaddr, KM_USER0);
1804                 if (left == 0)
1805                         goto success;
1806         }
1807
1808         /* Do it the slow way */
1809         kaddr = kmap(page);
1810         left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1811         kunmap(page);
1812
1813         if (left) {
1814                 size -= left;
1815                 desc->error = -EFAULT;
1816         }
1817 success:
1818         desc->count = count - size;
1819         desc->written += size;
1820         desc->arg.buf += size;
1821         return size;
1822 }
1823
1824 /**
1825  * btrfs_file_aio_read - filesystem read routine
1826  * @iocb:       kernel I/O control block
1827  * @iov:        io vector request
1828  * @nr_segs:    number of segments in the iovec
1829  * @pos:        current file position
1830  */
1831 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1832                                    unsigned long nr_segs, loff_t pos)
1833 {
1834         struct file *filp = iocb->ki_filp;
1835         ssize_t retval;
1836         unsigned long seg;
1837         size_t count;
1838         loff_t *ppos = &iocb->ki_pos;
1839
1840         count = 0;
1841         for (seg = 0; seg < nr_segs; seg++) {
1842                 const struct iovec *iv = &iov[seg];
1843
1844                 /*
1845                  * If any segment has a negative length, or the cumulative
1846                  * length ever wraps negative then return -EINVAL.
1847                  */
1848                 count += iv->iov_len;
1849                 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1850                         return -EINVAL;
1851                 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1852                         continue;
1853                 if (seg == 0)
1854                         return -EFAULT;
1855                 nr_segs = seg;
1856                 count -= iv->iov_len;   /* This segment is no good */
1857                 break;
1858         }
1859         retval = 0;
1860         if (count) {
1861                 for (seg = 0; seg < nr_segs; seg++) {
1862                         read_descriptor_t desc;
1863
1864                         desc.written = 0;
1865                         desc.arg.buf = iov[seg].iov_base;
1866                         desc.count = iov[seg].iov_len;
1867                         if (desc.count == 0)
1868                                 continue;
1869                         desc.error = 0;
1870                         do_generic_file_read(filp, ppos, &desc,
1871                                              btrfs_read_actor);
1872                         retval += desc.written;
1873                         if (desc.error) {
1874                                 retval = retval ?: desc.error;
1875                                 break;
1876                         }
1877                 }
1878         }
1879         return retval;
1880 }
1881
1882 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
1883 {
1884         struct btrfs_trans_handle *trans;
1885         struct btrfs_key key;
1886         struct btrfs_root_item root_item;
1887         struct btrfs_inode_item *inode_item;
1888         struct buffer_head *subvol;
1889         struct btrfs_leaf *leaf;
1890         struct btrfs_root *new_root;
1891         struct inode *inode;
1892         int ret;
1893         u64 objectid;
1894         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
1895
1896         mutex_lock(&root->fs_info->fs_mutex);
1897         trans = btrfs_start_transaction(root, 1);
1898         BUG_ON(!trans);
1899
1900         subvol = btrfs_alloc_free_block(trans, root);
1901         leaf = btrfs_buffer_leaf(subvol);
1902         btrfs_set_header_nritems(&leaf->header, 0);
1903         btrfs_set_header_level(&leaf->header, 0);
1904         btrfs_set_header_blocknr(&leaf->header, bh_blocknr(subvol));
1905         btrfs_set_header_generation(&leaf->header, trans->transid);
1906         memcpy(leaf->header.fsid, root->fs_info->disk_super->fsid,
1907                sizeof(leaf->header.fsid));
1908
1909         inode_item = &root_item.inode;
1910         memset(inode_item, 0, sizeof(*inode_item));
1911         btrfs_set_inode_generation(inode_item, 1);
1912         btrfs_set_inode_size(inode_item, 3);
1913         btrfs_set_inode_nlink(inode_item, 1);
1914         btrfs_set_inode_nblocks(inode_item, 1);
1915         btrfs_set_inode_mode(inode_item, S_IFDIR | 0755);
1916
1917         btrfs_set_root_blocknr(&root_item, bh_blocknr(subvol));
1918         btrfs_set_root_refs(&root_item, 1);
1919
1920         mark_buffer_dirty(subvol);
1921         brelse(subvol);
1922         subvol = NULL;
1923
1924         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1925                                        0, &objectid);
1926         BUG_ON(ret);
1927
1928         btrfs_set_root_dirid(&root_item, new_dirid);
1929
1930         key.objectid = objectid;
1931         key.offset = 1;
1932         key.flags = 0;
1933         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
1934         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
1935                                 &root_item);
1936         BUG_ON(ret);
1937
1938         /*
1939          * insert the directory item
1940          */
1941         key.offset = (u64)-1;
1942         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
1943                                     name, namelen,
1944                                     root->fs_info->sb->s_root->d_inode->i_ino,
1945                                     &key, 0);
1946         BUG_ON(ret);
1947
1948         ret = btrfs_commit_transaction(trans, root);
1949         BUG_ON(ret);
1950
1951         new_root = btrfs_read_fs_root(root->fs_info, &key);
1952         BUG_ON(!new_root);
1953
1954         trans = btrfs_start_transaction(new_root, 1);
1955         BUG_ON(!trans);
1956
1957         inode = btrfs_new_inode(trans, new_root, new_dirid, S_IFDIR | 0700);
1958         inode->i_op = &btrfs_dir_inode_operations;
1959         inode->i_fop = &btrfs_dir_file_operations;
1960
1961         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
1962         BUG_ON(ret);
1963
1964         inode->i_nlink = 1;
1965         inode->i_size = 6;
1966         ret = btrfs_update_inode(trans, new_root, inode);
1967         BUG_ON(ret);
1968
1969         ret = btrfs_commit_transaction(trans, new_root);
1970         BUG_ON(ret);
1971
1972         iput(inode);
1973
1974         mutex_unlock(&root->fs_info->fs_mutex);
1975         return 0;
1976 }
1977
1978 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
1979 {
1980         struct btrfs_trans_handle *trans;
1981         struct btrfs_key key;
1982         struct btrfs_root_item new_root_item;
1983         int ret;
1984         u64 objectid;
1985
1986         if (!root->ref_cows)
1987                 return -EINVAL;
1988
1989         mutex_lock(&root->fs_info->fs_mutex);
1990         trans = btrfs_start_transaction(root, 1);
1991         BUG_ON(!trans);
1992
1993         ret = btrfs_update_inode(trans, root, root->inode);
1994         BUG_ON(ret);
1995
1996         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
1997                                        0, &objectid);
1998         BUG_ON(ret);
1999
2000         memcpy(&new_root_item, &root->root_item,
2001                sizeof(new_root_item));
2002
2003         key.objectid = objectid;
2004         key.offset = 1;
2005         key.flags = 0;
2006         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2007         btrfs_set_root_blocknr(&new_root_item, bh_blocknr(root->node));
2008
2009         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2010                                 &new_root_item);
2011         BUG_ON(ret);
2012
2013         /*
2014          * insert the directory item
2015          */
2016         key.offset = (u64)-1;
2017         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2018                                     name, namelen,
2019                                     root->fs_info->sb->s_root->d_inode->i_ino,
2020                                     &key, 0);
2021
2022         BUG_ON(ret);
2023
2024         ret = btrfs_inc_root_ref(trans, root);
2025         BUG_ON(ret);
2026
2027         ret = btrfs_commit_transaction(trans, root);
2028         BUG_ON(ret);
2029         mutex_unlock(&root->fs_info->fs_mutex);
2030         return 0;
2031 }
2032
2033 static int add_disk(struct btrfs_root *root, char *name, int namelen)
2034 {
2035         struct block_device *bdev;
2036         struct btrfs_path *path;
2037         struct super_block *sb = root->fs_info->sb;
2038         struct btrfs_root *dev_root = root->fs_info->dev_root;
2039         struct btrfs_trans_handle *trans;
2040         struct btrfs_device_item *dev_item;
2041         struct btrfs_key key;
2042         u16 item_size;
2043         u64 num_blocks;
2044         u64 new_blocks;
2045         u64 device_id;
2046         int ret;
2047
2048 printk("adding disk %s\n", name);
2049         path = btrfs_alloc_path();
2050         if (!path)
2051                 return -ENOMEM;
2052         num_blocks = btrfs_super_total_blocks(root->fs_info->disk_super);
2053         bdev = open_bdev_excl(name, O_RDWR, sb);
2054         if (IS_ERR(bdev)) {
2055                 ret = PTR_ERR(bdev);
2056 printk("open bdev excl failed ret %d\n", ret);
2057                 goto out_nolock;
2058         }
2059         set_blocksize(bdev, sb->s_blocksize);
2060         new_blocks = bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2061         key.objectid = num_blocks;
2062         key.offset = new_blocks;
2063         key.flags = 0;
2064         btrfs_set_key_type(&key, BTRFS_DEV_ITEM_KEY);
2065
2066         mutex_lock(&dev_root->fs_info->fs_mutex);
2067         trans = btrfs_start_transaction(dev_root, 1);
2068         item_size = sizeof(*dev_item) + namelen;
2069 printk("insert empty on %Lu %Lu %u size %d\n", num_blocks, new_blocks, key.flags, item_size);
2070         ret = btrfs_insert_empty_item(trans, dev_root, path, &key, item_size);
2071         if (ret) {
2072 printk("insert failed %d\n", ret);
2073                 close_bdev_excl(bdev);
2074                 if (ret > 0)
2075                         ret = -EEXIST;
2076                 goto out;
2077         }
2078         dev_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
2079                                   path->slots[0], struct btrfs_device_item);
2080         btrfs_set_device_pathlen(dev_item, namelen);
2081         memcpy(dev_item + 1, name, namelen);
2082
2083         device_id = btrfs_super_last_device_id(root->fs_info->disk_super) + 1;
2084         btrfs_set_super_last_device_id(root->fs_info->disk_super, device_id);
2085         btrfs_set_device_id(dev_item, device_id);
2086         mark_buffer_dirty(path->nodes[0]);
2087
2088         ret = btrfs_insert_dev_radix(root, bdev, device_id, num_blocks,
2089                                      new_blocks);
2090
2091         if (!ret) {
2092                 btrfs_set_super_total_blocks(root->fs_info->disk_super,
2093                                              num_blocks + new_blocks);
2094                 i_size_write(root->fs_info->btree_inode,
2095                              (num_blocks + new_blocks) <<
2096                              root->fs_info->btree_inode->i_blkbits);
2097         }
2098
2099 out:
2100         ret = btrfs_commit_transaction(trans, dev_root);
2101         BUG_ON(ret);
2102         mutex_unlock(&root->fs_info->fs_mutex);
2103 out_nolock:
2104         btrfs_free_path(path);
2105
2106         return ret;
2107 }
2108
2109 static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
2110                        cmd, unsigned long arg)
2111 {
2112         struct btrfs_root *root = BTRFS_I(inode)->root;
2113         struct btrfs_ioctl_vol_args vol_args;
2114         int ret = 0;
2115         int namelen;
2116         struct btrfs_path *path;
2117         u64 root_dirid;
2118
2119         switch (cmd) {
2120         case BTRFS_IOC_SNAP_CREATE:
2121                 if (copy_from_user(&vol_args,
2122                                    (struct btrfs_ioctl_vol_args __user *)arg,
2123                                    sizeof(vol_args)))
2124                         return -EFAULT;
2125                 namelen = strlen(vol_args.name);
2126                 if (namelen > BTRFS_VOL_NAME_MAX)
2127                         return -EINVAL;
2128                 path = btrfs_alloc_path();
2129                 if (!path)
2130                         return -ENOMEM;
2131                 root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2132                 mutex_lock(&root->fs_info->fs_mutex);
2133                 ret = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2134                                     path, root_dirid,
2135                                     vol_args.name, namelen, 0);
2136                 mutex_unlock(&root->fs_info->fs_mutex);
2137                 btrfs_free_path(path);
2138                 if (ret == 0)
2139                         return -EEXIST;
2140
2141                 if (root == root->fs_info->tree_root)
2142                         ret = create_subvol(root, vol_args.name, namelen);
2143                 else
2144                         ret = create_snapshot(root, vol_args.name, namelen);
2145                 WARN_ON(ret);
2146                 break;
2147         case BTRFS_IOC_ADD_DISK:
2148                 if (copy_from_user(&vol_args,
2149                                    (struct btrfs_ioctl_vol_args __user *)arg,
2150                                    sizeof(vol_args)))
2151                         return -EFAULT;
2152                 namelen = strlen(vol_args.name);
2153                 if (namelen > BTRFS_VOL_NAME_MAX)
2154                         return -EINVAL;
2155                 vol_args.name[namelen] = '\0';
2156                 ret = add_disk(root, vol_args.name, namelen);
2157                 break;
2158         default:
2159                 return -ENOTTY;
2160         }
2161         return ret;
2162 }
2163
2164 static struct kmem_cache *btrfs_inode_cachep;
2165 struct kmem_cache *btrfs_trans_handle_cachep;
2166 struct kmem_cache *btrfs_transaction_cachep;
2167 struct kmem_cache *btrfs_bit_radix_cachep;
2168 struct kmem_cache *btrfs_path_cachep;
2169
2170 /*
2171  * Called inside transaction, so use GFP_NOFS
2172  */
2173 static struct inode *btrfs_alloc_inode(struct super_block *sb)
2174 {
2175         struct btrfs_inode *ei;
2176
2177         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2178         if (!ei)
2179                 return NULL;
2180         return &ei->vfs_inode;
2181 }
2182
2183 static void btrfs_destroy_inode(struct inode *inode)
2184 {
2185         WARN_ON(!list_empty(&inode->i_dentry));
2186         WARN_ON(inode->i_data.nrpages);
2187
2188         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2189 }
2190
2191 static void init_once(void * foo, struct kmem_cache * cachep,
2192                       unsigned long flags)
2193 {
2194         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2195
2196         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2197             SLAB_CTOR_CONSTRUCTOR) {
2198                 inode_init_once(&ei->vfs_inode);
2199         }
2200 }
2201
2202 static int init_inodecache(void)
2203 {
2204         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
2205                                              sizeof(struct btrfs_inode),
2206                                              0, (SLAB_RECLAIM_ACCOUNT|
2207                                                 SLAB_MEM_SPREAD),
2208                                              init_once, NULL);
2209         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
2210                                              sizeof(struct btrfs_trans_handle),
2211                                              0, (SLAB_RECLAIM_ACCOUNT|
2212                                                 SLAB_MEM_SPREAD),
2213                                              NULL, NULL);
2214         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
2215                                              sizeof(struct btrfs_transaction),
2216                                              0, (SLAB_RECLAIM_ACCOUNT|
2217                                                 SLAB_MEM_SPREAD),
2218                                              NULL, NULL);
2219         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
2220                                              sizeof(struct btrfs_transaction),
2221                                              0, (SLAB_RECLAIM_ACCOUNT|
2222                                                 SLAB_MEM_SPREAD),
2223                                              NULL, NULL);
2224         btrfs_bit_radix_cachep = kmem_cache_create("btrfs_radix",
2225                                              256,
2226                                              0, (SLAB_RECLAIM_ACCOUNT|
2227                                                 SLAB_MEM_SPREAD |
2228                                                 SLAB_DESTROY_BY_RCU),
2229                                              NULL, NULL);
2230         if (btrfs_inode_cachep == NULL || btrfs_trans_handle_cachep == NULL ||
2231             btrfs_transaction_cachep == NULL || btrfs_bit_radix_cachep == NULL)
2232                 return -ENOMEM;
2233         return 0;
2234 }
2235
2236 static void destroy_inodecache(void)
2237 {
2238         kmem_cache_destroy(btrfs_inode_cachep);
2239         kmem_cache_destroy(btrfs_trans_handle_cachep);
2240         kmem_cache_destroy(btrfs_transaction_cachep);
2241         kmem_cache_destroy(btrfs_bit_radix_cachep);
2242         kmem_cache_destroy(btrfs_path_cachep);
2243 }
2244
2245 static int btrfs_get_sb(struct file_system_type *fs_type,
2246         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2247 {
2248         return get_sb_bdev(fs_type, flags, dev_name, data,
2249                            btrfs_fill_super, mnt);
2250 }
2251
2252 static struct file_system_type btrfs_fs_type = {
2253         .owner          = THIS_MODULE,
2254         .name           = "btrfs",
2255         .get_sb         = btrfs_get_sb,
2256         .kill_sb        = kill_block_super,
2257         .fs_flags       = FS_REQUIRES_DEV,
2258 };
2259
2260 static struct super_operations btrfs_super_ops = {
2261         .statfs         = simple_statfs,
2262         .delete_inode   = btrfs_delete_inode,
2263         .put_super      = btrfs_put_super,
2264         .read_inode     = btrfs_read_locked_inode,
2265         .write_super    = btrfs_write_super,
2266         .sync_fs        = btrfs_sync_fs,
2267         .write_inode    = btrfs_write_inode,
2268         .alloc_inode    = btrfs_alloc_inode,
2269         .destroy_inode  = btrfs_destroy_inode,
2270 };
2271
2272 static struct inode_operations btrfs_dir_inode_operations = {
2273         .lookup         = btrfs_lookup,
2274         .create         = btrfs_create,
2275         .unlink         = btrfs_unlink,
2276         .mkdir          = btrfs_mkdir,
2277         .rmdir          = btrfs_rmdir,
2278 };
2279
2280 static struct inode_operations btrfs_dir_ro_inode_operations = {
2281         .lookup         = btrfs_lookup,
2282 };
2283
2284 static struct file_operations btrfs_dir_file_operations = {
2285         .llseek         = generic_file_llseek,
2286         .read           = generic_read_dir,
2287         .readdir        = btrfs_readdir,
2288         .ioctl          = btrfs_ioctl,
2289 };
2290
2291 static struct address_space_operations btrfs_aops = {
2292         .readpage       = btrfs_readpage,
2293         .writepage      = btrfs_writepage,
2294         .sync_page      = block_sync_page,
2295         .prepare_write  = btrfs_prepare_write,
2296         .commit_write   = btrfs_commit_write,
2297 };
2298
2299 static struct inode_operations btrfs_file_inode_operations = {
2300         .truncate       = btrfs_truncate,
2301 };
2302
2303 static struct file_operations btrfs_file_operations = {
2304         .llseek         = generic_file_llseek,
2305         .read           = do_sync_read,
2306         .aio_read       = btrfs_file_aio_read,
2307         .write          = btrfs_file_write,
2308         .mmap           = generic_file_mmap,
2309         .open           = generic_file_open,
2310         .ioctl          = btrfs_ioctl,
2311 };
2312
2313 static int __init init_btrfs_fs(void)
2314 {
2315         int err;
2316         printk("btrfs loaded!\n");
2317         err = init_inodecache();
2318         if (err)
2319                 return err;
2320         kset_set_kset_s(&btrfs_subsys, fs_subsys);
2321         err = subsystem_register(&btrfs_subsys);
2322         if (err)
2323                 goto out;
2324         return register_filesystem(&btrfs_fs_type);
2325 out:
2326         destroy_inodecache();
2327         return err;
2328 }
2329
2330 static void __exit exit_btrfs_fs(void)
2331 {
2332         destroy_inodecache();
2333         unregister_filesystem(&btrfs_fs_type);
2334         subsystem_unregister(&btrfs_subsys);
2335         printk("btrfs unloaded\n");
2336 }
2337
2338 module_init(init_btrfs_fs)
2339 module_exit(exit_btrfs_fs)
2340
2341 MODULE_LICENSE("GPL");