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                 found_extent = 0;
 321                 if (btrfs_disk_key_type(found_key) == BTRFS_EXTENT_DATA_KEY) {
 322                         fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 323                                             path->slots[0],
 324                                             struct btrfs_file_extent_item);
 325                         if (btrfs_file_extent_type(fi) !=
 326                             BTRFS_FILE_EXTENT_INLINE) {
 327                                 extent_start =
 328                                         btrfs_file_extent_disk_blocknr(fi);
 329                                 extent_num_blocks =
 330                                         btrfs_file_extent_disk_num_blocks(fi);
 331                                 /* FIXME blocksize != 4096 */
 332                                 inode->i_blocks -=
 333                                         btrfs_file_extent_num_blocks(fi) << 3;
 334                                 found_extent = 1;
 335                         }
 336                 }
 337                 ret = btrfs_del_item(trans, root, path);
 338                 BUG_ON(ret);
 339                 btrfs_release_path(root, path);
 340                 if (found_extent) {
 341                         ret = btrfs_free_extent(trans, root, extent_start,
 342                                                 extent_num_blocks, 0);
 343                         BUG_ON(ret);
 344                 }
 345         }
 346         ret = 0;
 347 error:
 348         btrfs_release_path(root, path);
 349         btrfs_free_path(path);
 350         return ret;
 351 }
 352
 353 static void btrfs_delete_inode(struct inode *inode)
 354 {
 355         struct btrfs_trans_handle *trans;
 356         struct btrfs_root *root = BTRFS_I(inode)->root;
 357         int ret;
 358
 359         truncate_inode_pages(&inode->i_data, 0);
 360         if (is_bad_inode(inode)) {
 361                 goto no_delete;
 362         }
 363         inode->i_size = 0;
 364         mutex_lock(&root->fs_info->fs_mutex);
 365         trans = btrfs_start_transaction(root, 1);
 366         if (S_ISREG(inode->i_mode)) {
 367                 ret = btrfs_truncate_in_trans(trans, root, inode);
 368                 BUG_ON(ret);
 369         }
 370         btrfs_free_inode(trans, root, inode);
 371         btrfs_end_transaction(trans, root);
 372         mutex_unlock(&root->fs_info->fs_mutex);
 373         return;
 374 no_delete:
 375         clear_inode(inode);
 376 }
 377
 378 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 379                                struct btrfs_key *location)
 380 {
 381         const char *name = dentry->d_name.name;
 382         int namelen = dentry->d_name.len;
 383         struct btrfs_dir_item *di;
 384         struct btrfs_path *path;
 385         struct btrfs_root *root = BTRFS_I(dir)->root;
 386         int ret;
 387
 388         path = btrfs_alloc_path();
 389         BUG_ON(!path);
 390         btrfs_init_path(path);
 391         ret = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
 392                                     namelen, 0);
 393         if (ret || !btrfs_match_dir_item_name(root, path, name, namelen)) {
 394                 location->objectid = 0;
 395                 ret = 0;
 396                 goto out;
 397         }
 398         di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 399                             struct btrfs_dir_item);
 400         btrfs_disk_key_to_cpu(location, &di->location);
 401 out:
 402         btrfs_release_path(root, path);
 403         btrfs_free_path(path);
 404         return ret;
 405 }
 406
 407 int fixup_tree_root_location(struct btrfs_root *root,
 408                              struct btrfs_key *location,
 409                              struct btrfs_root **sub_root)
 410 {
 411         struct btrfs_path *path;
 412         struct btrfs_root_item *ri;
 413
 414         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 415                 return 0;
 416         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 417                 return 0;
 418
 419         path = btrfs_alloc_path();
 420         BUG_ON(!path);
 421         mutex_lock(&root->fs_info->fs_mutex);
 422
 423         *sub_root = btrfs_read_fs_root(root->fs_info, location);
 424         if (IS_ERR(*sub_root))
 425                 return PTR_ERR(*sub_root);
 426
 427         ri = &(*sub_root)->root_item;
 428         location->objectid = btrfs_root_dirid(ri);
 429         location->flags = 0;
 430         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 431         location->offset = 0;
 432
 433         btrfs_free_path(path);
 434         mutex_unlock(&root->fs_info->fs_mutex);
 435         return 0;
 436 }
 437
 438 int btrfs_init_locked_inode(struct inode *inode, void *p)
 439 {
 440         struct btrfs_iget_args *args = p;
 441         inode->i_ino = args->ino;
 442         BTRFS_I(inode)->root = args->root;
 443         return 0;
 444 }
 445
 446 int btrfs_find_actor(struct inode *inode, void *opaque)
 447 {
 448         struct btrfs_iget_args *args = opaque;
 449         return (args->ino == inode->i_ino &&
 450                 args->root == BTRFS_I(inode)->root);
 451 }
 452
 453 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 454                                 struct btrfs_root *root)
 455 {
 456         struct inode *inode;
 457         struct btrfs_iget_args args;
 458         args.ino = objectid;
 459         args.root = root;
 460
 461         inode = iget5_locked(s, objectid, btrfs_find_actor,
 462                              btrfs_init_locked_inode,
 463                              (void *)&args);
 464         return inode;
 465 }
 466
 467 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 468                                    struct nameidata *nd)
 469 {
 470         struct inode * inode;
 471         struct btrfs_inode *bi = BTRFS_I(dir);
 472         struct btrfs_root *root = bi->root;
 473         struct btrfs_root *sub_root = root;
 474         struct btrfs_key location;
 475         int ret;
 476
 477         if (dentry->d_name.len > BTRFS_NAME_LEN)
 478                 return ERR_PTR(-ENAMETOOLONG);
 479         mutex_lock(&root->fs_info->fs_mutex);
 480         ret = btrfs_inode_by_name(dir, dentry, &location);
 481         mutex_unlock(&root->fs_info->fs_mutex);
 482         if (ret < 0)
 483                 return ERR_PTR(ret);
 484         inode = NULL;
 485         if (location.objectid) {
 486                 ret = fixup_tree_root_location(root, &location, &sub_root);
 487                 if (ret < 0)
 488                         return ERR_PTR(ret);
 489                 if (ret > 0)
 490                         return ERR_PTR(-ENOENT);
 491                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 492                                           sub_root);
 493                 if (!inode)
 494                         return ERR_PTR(-EACCES);
 495                 if (inode->i_state & I_NEW) {
 496                         if (sub_root != root) {
 497 printk("adding new root for inode %lu root %p (found %p)\n", inode->i_ino, sub_root, BTRFS_I(inode)->root);
 498                                 igrab(inode);
 499                                 sub_root->inode = inode;
 500                         }
 501                         BTRFS_I(inode)->root = sub_root;
 502                         memcpy(&BTRFS_I(inode)->location, &location,
 503                                sizeof(location));
 504                         btrfs_read_locked_inode(inode);
 505                         unlock_new_inode(inode);
 506                 }
 507         }
 508         return d_splice_alias(inode, dentry);
 509 }
 510
 511 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 512 {
 513         struct inode *inode = filp->f_path.dentry->d_inode;
 514         struct btrfs_root *root = BTRFS_I(inode)->root;
 515         struct btrfs_item *item;
 516         struct btrfs_dir_item *di;
 517         struct btrfs_key key;
 518         struct btrfs_path *path;
 519         int ret;
 520         u32 nritems;
 521         struct btrfs_leaf *leaf;
 522         int slot;
 523         int advance;
 524         unsigned char d_type = DT_UNKNOWN;
 525         int over = 0;
 526         int key_type = BTRFS_DIR_ITEM_KEY;
 527
 528         /* FIXME, use a real flag for deciding about the key type */
 529         if (root->fs_info->tree_root == root)
 530                 key_type = BTRFS_DIR_ITEM_KEY;
 531         mutex_lock(&root->fs_info->fs_mutex);
 532         key.objectid = inode->i_ino;
 533         key.flags = 0;
 534         btrfs_set_key_type(&key, key_type);
 535         key.offset = filp->f_pos;
 536         path = btrfs_alloc_path();
 537         btrfs_init_path(path);
 538         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 539         if (ret < 0)
 540                 goto err;
 541         advance = 0;
 542         while(1) {
 543                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 544                 nritems = btrfs_header_nritems(&leaf->header);
 545                 slot = path->slots[0];
 546                 if (advance || slot >= nritems) {
 547                         if (slot >= nritems -1) {
 548                                 ret = btrfs_next_leaf(root, path);
 549                                 if (ret)
 550                                         break;
 551                                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 552                                 nritems = btrfs_header_nritems(&leaf->header);
 553                                 slot = path->slots[0];
 554                         } else {
 555                                 slot++;
 556                                 path->slots[0]++;
 557                         }
 558                 }
 559                 advance = 1;
 560                 item = leaf->items + slot;
 561                 if (btrfs_disk_key_objectid(&item->key) != key.objectid)
 562                         break;
 563                 if (btrfs_disk_key_type(&item->key) != key_type)
 564                         break;
 565                 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
 566                         continue;
 567                 filp->f_pos = btrfs_disk_key_offset(&item->key);
 568                 advance = 1;
 569                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 570                 over = filldir(dirent, (const char *)(di + 1),
 571                                btrfs_dir_name_len(di),
 572                                btrfs_disk_key_offset(&item->key),
 573                                btrfs_disk_key_objectid(&di->location), d_type);
 574                 if (over)
 575                         goto nopos;
 576         }
 577         filp->f_pos++;
 578 nopos:
 579         ret = 0;
 580 err:
 581         btrfs_release_path(root, path);
 582         btrfs_free_path(path);
 583         mutex_unlock(&root->fs_info->fs_mutex);
 584         return ret;
 585 }
 586
 587 static void btrfs_put_super (struct super_block * sb)
 588 {
 589         struct btrfs_root *root = btrfs_sb(sb);
 590         int ret;
 591
 592         ret = close_ctree(root);
 593         if (ret) {
 594                 printk("close ctree returns %d\n", ret);
 595         }
 596         sb->s_fs_info = NULL;
 597 }
 598
 599 static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 600 {
 601         struct inode * inode;
 602         struct dentry * root_dentry;
 603         struct btrfs_super_block *disk_super;
 604         struct btrfs_root *tree_root;
 605         struct btrfs_inode *bi;
 606
 607         sb->s_maxbytes = MAX_LFS_FILESIZE;
 608         sb->s_magic = BTRFS_SUPER_MAGIC;
 609         sb->s_op = &btrfs_super_ops;
 610         sb->s_time_gran = 1;
 611
 612         tree_root = open_ctree(sb);
 613
 614         if (!tree_root) {
 615                 printk("btrfs: open_ctree failed\n");
 616                 return -EIO;
 617         }
 618         sb->s_fs_info = tree_root;
 619         disk_super = tree_root->fs_info->disk_super;
 620         printk("read in super total blocks %Lu root %Lu\n",
 621                btrfs_super_total_blocks(disk_super),
 622                btrfs_super_root_dir(disk_super));
 623
 624         inode = btrfs_iget_locked(sb, btrfs_super_root_dir(disk_super),
 625                                   tree_root);
 626         bi = BTRFS_I(inode);
 627         bi->location.objectid = inode->i_ino;
 628         bi->location.offset = 0;
 629         bi->location.flags = 0;
 630         bi->root = tree_root;
 631         btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
 632
 633         if (!inode)
 634                 return -ENOMEM;
 635         if (inode->i_state & I_NEW) {
 636                 btrfs_read_locked_inode(inode);
 637                 unlock_new_inode(inode);
 638         }
 639
 640         root_dentry = d_alloc_root(inode);
 641         if (!root_dentry) {
 642                 iput(inode);
 643                 return -ENOMEM;
 644         }
 645         sb->s_root = root_dentry;
 646
 647         return 0;
 648 }
 649
 650 static void fill_inode_item(struct btrfs_inode_item *item,
 651                             struct inode *inode)
 652 {
 653         btrfs_set_inode_uid(item, inode->i_uid);
 654         btrfs_set_inode_gid(item, inode->i_gid);
 655         btrfs_set_inode_size(item, inode->i_size);
 656         btrfs_set_inode_mode(item, inode->i_mode);
 657         btrfs_set_inode_nlink(item, inode->i_nlink);
 658         btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
 659         btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
 660         btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
 661         btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
 662         btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
 663         btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
 664         btrfs_set_inode_nblocks(item, inode->i_blocks);
 665         btrfs_set_inode_generation(item, inode->i_generation);
 666 }
 667
 668 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
 669                               struct btrfs_root *root,
 670                               struct inode *inode)
 671 {
 672         struct btrfs_inode_item *inode_item;
 673         struct btrfs_path *path;
 674         int ret;
 675
 676         path = btrfs_alloc_path();
 677         BUG_ON(!path);
 678         btrfs_init_path(path);
 679         ret = btrfs_lookup_inode(trans, root, path,
 680                                  &BTRFS_I(inode)->location, 1);
 681         if (ret) {
 682                 if (ret > 0)
 683                         ret = -ENOENT;
 684                 goto failed;
 685         }
 686
 687         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 688                                   path->slots[0],
 689                                   struct btrfs_inode_item);
 690
 691         fill_inode_item(inode_item, inode);
 692         btrfs_mark_buffer_dirty(path->nodes[0]);
 693         ret = 0;
 694 failed:
 695         btrfs_release_path(root, path);
 696         btrfs_free_path(path);
 697         return ret;
 698 }
 699
 700 static int btrfs_write_inode(struct inode *inode, int wait)
 701 {
 702         struct btrfs_root *root = BTRFS_I(inode)->root;
 703         struct btrfs_trans_handle *trans;
 704         int ret;
 705
 706         mutex_lock(&root->fs_info->fs_mutex);
 707         trans = btrfs_start_transaction(root, 1);
 708         ret = btrfs_update_inode(trans, root, inode);
 709         if (wait)
 710                 btrfs_commit_transaction(trans, root);
 711         else
 712                 btrfs_end_transaction(trans, root);
 713         mutex_unlock(&root->fs_info->fs_mutex);
 714         return ret;
 715 }
 716
 717 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 718                                      struct btrfs_root *root,
 719                                      u64 objectid, int mode)
 720 {
 721         struct inode *inode;
 722         struct btrfs_inode_item inode_item;
 723         struct btrfs_key *location;
 724         int ret;
 725
 726         inode = new_inode(root->fs_info->sb);
 727         if (!inode)
 728                 return ERR_PTR(-ENOMEM);
 729
 730         BTRFS_I(inode)->root = root;
 731
 732         inode->i_uid = current->fsuid;
 733         inode->i_gid = current->fsgid;
 734         inode->i_mode = mode;
 735         inode->i_ino = objectid;
 736         inode->i_blocks = 0;
 737         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 738         fill_inode_item(&inode_item, inode);
 739         location = &BTRFS_I(inode)->location;
 740         location->objectid = objectid;
 741         location->flags = 0;
 742         location->offset = 0;
 743         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 744
 745         ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
 746         BUG_ON(ret);
 747
 748         insert_inode_hash(inode);
 749         return inode;
 750 }
 751
 752 static int btrfs_add_link(struct btrfs_trans_handle *trans,
 753                             struct dentry *dentry, struct inode *inode)
 754 {
 755         int ret;
 756         struct btrfs_key key;
 757         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
 758         key.objectid = inode->i_ino;
 759         key.flags = 0;
 760         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 761         key.offset = 0;
 762
 763         ret = btrfs_insert_dir_item(trans, root,
 764                                     dentry->d_name.name, dentry->d_name.len,
 765                                     dentry->d_parent->d_inode->i_ino,
 766                                     &key, 0);
 767         if (ret == 0) {
 768                 dentry->d_parent->d_inode->i_size += dentry->d_name.len * 2;
 769                 ret = btrfs_update_inode(trans, root,
 770                                          dentry->d_parent->d_inode);
 771         }
 772         return ret;
 773 }
 774
 775 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
 776                             struct dentry *dentry, struct inode *inode)
 777 {
 778         int err = btrfs_add_link(trans, dentry, inode);
 779         if (!err) {
 780                 d_instantiate(dentry, inode);
 781                 return 0;
 782         }
 783         if (err > 0)
 784                 err = -EEXIST;
 785         return err;
 786 }
 787
 788 static int btrfs_create(struct inode *dir, struct dentry *dentry,
 789                         int mode, struct nameidata *nd)
 790 {
 791         struct btrfs_trans_handle *trans;
 792         struct btrfs_root *root = BTRFS_I(dir)->root;
 793         struct inode *inode;
 794         int err;
 795         int drop_inode = 0;
 796         u64 objectid;
 797
 798         mutex_lock(&root->fs_info->fs_mutex);
 799         trans = btrfs_start_transaction(root, 1);
 800
 801         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 802         if (err) {
 803                 err = -ENOSPC;
 804                 goto out_unlock;
 805         }
 806
 807         inode = btrfs_new_inode(trans, root, objectid, mode);
 808         err = PTR_ERR(inode);
 809         if (IS_ERR(inode))
 810                 goto out_unlock;
 811         // FIXME mark the inode dirty
 812         err = btrfs_add_nondir(trans, dentry, inode);
 813         if (err)
 814                 drop_inode = 1;
 815         else {
 816                 inode->i_mapping->a_ops = &btrfs_aops;
 817                 inode->i_fop = &btrfs_file_operations;
 818                 inode->i_op = &btrfs_file_inode_operations;
 819         }
 820         dir->i_sb->s_dirt = 1;
 821 out_unlock:
 822         btrfs_end_transaction(trans, root);
 823         mutex_unlock(&root->fs_info->fs_mutex);
 824
 825         if (drop_inode) {
 826                 inode_dec_link_count(inode);
 827                 iput(inode);
 828         }
 829         return err;
 830 }
 831
 832 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
 833                                 struct btrfs_root *root,
 834                                 u64 objectid, u64 dirid)
 835 {
 836         int ret;
 837         char buf[2];
 838         struct btrfs_key key;
 839
 840         buf[0] = '.';
 841         buf[1] = '.';
 842
 843         key.objectid = objectid;
 844         key.offset = 0;
 845         key.flags = 0;
 846         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 847
 848         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
 849                                     &key, 1);
 850         if (ret)
 851                 goto error;
 852         key.objectid = dirid;
 853         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
 854                                     &key, 1);
 855         if (ret)
 856                 goto error;
 857 error:
 858         return ret;
 859 }
 860
 861 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 862 {
 863         struct inode *inode;
 864         struct btrfs_trans_handle *trans;
 865         struct btrfs_root *root = BTRFS_I(dir)->root;
 866         int err = 0;
 867         int drop_on_err = 0;
 868         u64 objectid;
 869
 870         mutex_lock(&root->fs_info->fs_mutex);
 871         trans = btrfs_start_transaction(root, 1);
 872         if (IS_ERR(trans)) {
 873                 err = PTR_ERR(trans);
 874                 goto out_unlock;
 875         }
 876
 877         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 878         if (err) {
 879                 err = -ENOSPC;
 880                 goto out_unlock;
 881         }
 882
 883         inode = btrfs_new_inode(trans, root, objectid, S_IFDIR | mode);
 884         if (IS_ERR(inode)) {
 885                 err = PTR_ERR(inode);
 886                 goto out_fail;
 887         }
 888         drop_on_err = 1;
 889         inode->i_op = &btrfs_dir_inode_operations;
 890         inode->i_fop = &btrfs_dir_file_operations;
 891
 892         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
 893         if (err)
 894                 goto out_fail;
 895
 896         inode->i_size = 6;
 897         err = btrfs_update_inode(trans, root, inode);
 898         if (err)
 899                 goto out_fail;
 900         err = btrfs_add_link(trans, dentry, inode);
 901         if (err)
 902                 goto out_fail;
 903         d_instantiate(dentry, inode);
 904         drop_on_err = 0;
 905
 906 out_fail:
 907         btrfs_end_transaction(trans, root);
 908 out_unlock:
 909         mutex_unlock(&root->fs_info->fs_mutex);
 910         if (drop_on_err)
 911                 iput(inode);
 912         return err;
 913 }
 914
 915 static int btrfs_sync_fs(struct super_block *sb, int wait)
 916 {
 917         struct btrfs_trans_handle *trans;
 918         struct btrfs_root *root;
 919         int ret;
 920         root = btrfs_sb(sb);
 921
 922         sb->s_dirt = 0;
 923         if (!wait) {
 924                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 925                 return 0;
 926         }
 927         filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
 928         mutex_lock(&root->fs_info->fs_mutex);
 929         trans = btrfs_start_transaction(root, 1);
 930         ret = btrfs_commit_transaction(trans, root);
 931         sb->s_dirt = 0;
 932         BUG_ON(ret);
 933 printk("btrfs sync_fs\n");
 934         mutex_unlock(&root->fs_info->fs_mutex);
 935         return 0;
 936 }
 937
 938 #if 0
 939 static int btrfs_get_block_inline(struct inode *inode, sector_t iblock,
 940                            struct buffer_head *result, int create)
 941 {
 942         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 943         struct btrfs_path *path;
 944         struct btrfs_key key;
 945         struct btrfs_leaf *leaf;
 946         int num_bytes = result->b_size;
 947         int item_size;
 948         int ret;
 949         u64 pos;
 950         char *ptr;
 951         int copy_size;
 952         int err = 0;
 953         char *safe_ptr;
 954         char *data_ptr;
 955
 956         path = btrfs_alloc_path();
 957         BUG_ON(!path);
 958
 959         WARN_ON(create);
 960         if (create) {
 961                 return 0;
 962         }
 963         pos = iblock << inode->i_blkbits;
 964         key.objectid = inode->i_ino;
 965         key.flags = 0;
 966         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
 967         ptr = kmap(result->b_page);
 968         safe_ptr = ptr;
 969         ptr += (pos & (PAGE_CACHE_SIZE -1));
 970 again:
 971         key.offset = pos;
 972         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 973         if (ret) {
 974                 if (ret < 0)
 975                         err = ret;
 976                 else
 977                         err = 0;
 978                 goto out;
 979         }
 980         leaf = btrfs_buffer_leaf(path->nodes[0]);
 981         item_size = btrfs_item_size(leaf->items + path->slots[0]);
 982         copy_size = min(num_bytes, item_size);
 983         data_ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 984         WARN_ON(safe_ptr + PAGE_CACHE_SIZE < ptr + copy_size);
 985         memcpy(ptr, data_ptr, copy_size);
 986         pos += copy_size;
 987         num_bytes -= copy_size;
 988         WARN_ON(num_bytes < 0);
 989         ptr += copy_size;
 990         btrfs_release_path(root, path);
 991         if (num_bytes != 0) {
 992                 if (pos >= i_size_read(inode))
 993                         memset(ptr, 0, num_bytes);
 994                 else
 995                         goto again;
 996         }
 997         set_buffer_uptodate(result);
 998         map_bh(result, inode->i_sb, 0);
 999         err = 0;
1000 out:
1001         btrfs_free_path(path);
1002         kunmap(result->b_page);
1003         return err;
1004 }
1005 #endif
1006
1007 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
1008                            struct buffer_head *result, int create)
1009 {
1010         int ret;
1011         int err = 0;
1012         u64 blocknr;
1013         u64 extent_start = 0;
1014         u64 extent_end = 0;
1015         u64 objectid = inode->i_ino;
1016         u32 found_type;
1017         struct btrfs_path *path;
1018         struct btrfs_root *root = BTRFS_I(inode)->root;
1019         struct btrfs_file_extent_item *item;
1020         struct btrfs_leaf *leaf;
1021         struct btrfs_disk_key *found_key;
1022
1023         path = btrfs_alloc_path();
1024         BUG_ON(!path);
1025         btrfs_init_path(path);
1026         if (create) {
1027                 WARN_ON(1);
1028         }
1029
1030         ret = btrfs_lookup_file_extent(NULL, root, path,
1031                                        inode->i_ino,
1032                                        iblock << inode->i_blkbits, 0);
1033         if (ret < 0) {
1034                 err = ret;
1035                 goto out;
1036         }
1037
1038         if (ret != 0) {
1039                 if (path->slots[0] == 0) {
1040                         btrfs_release_path(root, path);
1041                         goto out;
1042                 }
1043                 path->slots[0]--;
1044         }
1045
1046         item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
1047                               struct btrfs_file_extent_item);
1048         leaf = btrfs_buffer_leaf(path->nodes[0]);
1049         blocknr = btrfs_file_extent_disk_blocknr(item);
1050         blocknr += btrfs_file_extent_offset(item);
1051
1052         /* are we inside the extent that was found? */
1053         found_key = &leaf->items[path->slots[0]].key;
1054         found_type = btrfs_disk_key_type(found_key);
1055         if (btrfs_disk_key_objectid(found_key) != objectid ||
1056             found_type != BTRFS_EXTENT_DATA_KEY) {
1057                 extent_end = 0;
1058                 extent_start = 0;
1059                 btrfs_release_path(root, path);
1060                 goto out;
1061         }
1062         found_type = btrfs_file_extent_type(item);
1063         extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
1064         if (found_type == BTRFS_FILE_EXTENT_REG) {
1065                 extent_start = extent_start >> inode->i_blkbits;
1066                 extent_end = extent_start + btrfs_file_extent_num_blocks(item);
1067                 if (iblock >= extent_start && iblock < extent_end) {
1068                         err = 0;
1069                         btrfs_map_bh_to_logical(root, result, blocknr +
1070                                                 iblock - extent_start);
1071                         goto out;
1072                 }
1073         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1074                 char *ptr;
1075                 char *map;
1076                 u32 size;
1077                 size = btrfs_file_extent_inline_len(leaf->items +
1078                                                     path->slots[0]);
1079                 extent_end = (extent_start + size) >> inode->i_blkbits;
1080                 extent_start >>= inode->i_blkbits;
1081                 if (iblock < extent_start || iblock > extent_end) {
1082                         goto out;
1083                 }
1084                 ptr = btrfs_file_extent_inline_start(item);
1085                 map = kmap(result->b_page);
1086                 memcpy(map, ptr, size);
1087                 memset(map + size, 0, PAGE_CACHE_SIZE - size);
1088                 flush_dcache_page(result->b_page);
1089                 kunmap(result->b_page);
1090                 set_buffer_uptodate(result);
1091                 SetPageChecked(result->b_page);
1092                 btrfs_map_bh_to_logical(root, result, 0);
1093         }
1094 out:
1095         btrfs_release_path(root, path);
1096         btrfs_free_path(path);
1097         return err;
1098 }
1099
1100 static int btrfs_get_block(struct inode *inode, sector_t iblock,
1101                            struct buffer_head *result, int create)
1102 {
1103         int err;
1104         struct btrfs_root *root = BTRFS_I(inode)->root;
1105         mutex_lock(&root->fs_info->fs_mutex);
1106         err = btrfs_get_block_lock(inode, iblock, result, create);
1107         mutex_unlock(&root->fs_info->fs_mutex);
1108         return err;
1109 }
1110
1111 static int btrfs_prepare_write(struct file *file, struct page *page,
1112                                unsigned from, unsigned to)
1113 {
1114         return nobh_prepare_write(page, from, to, btrfs_get_block);
1115 }
1116
1117 static void btrfs_write_super(struct super_block *sb)
1118 {
1119         btrfs_sync_fs(sb, 1);
1120 }
1121
1122 static int btrfs_readpage(struct file *file, struct page *page)
1123 {
1124         return mpage_readpage(page, btrfs_get_block);
1125 }
1126
1127 /*
1128  * While block_write_full_page is writing back the dirty buffers under
1129  * the page lock, whoever dirtied the buffers may decide to clean them
1130  * again at any time.  We handle that by only looking at the buffer
1131  * state inside lock_buffer().
1132  *
1133  * If block_write_full_page() is called for regular writeback
1134  * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1135  * locked buffer.   This only can happen if someone has written the buffer
1136  * directly, with submit_bh().  At the address_space level PageWriteback
1137  * prevents this contention from occurring.
1138  */
1139 static int __btrfs_write_full_page(struct inode *inode, struct page *page,
1140                                    struct writeback_control *wbc)
1141 {
1142         int err;
1143         sector_t block;
1144         sector_t last_block;
1145         struct buffer_head *bh, *head;
1146         const unsigned blocksize = 1 << inode->i_blkbits;
1147         int nr_underway = 0;
1148
1149         BUG_ON(!PageLocked(page));
1150
1151         last_block = (i_size_read(inode) - 1) >> inode->i_blkbits;
1152
1153         if (!page_has_buffers(page)) {
1154                 create_empty_buffers(page, blocksize,
1155                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
1156         }
1157
1158         /*
1159          * Be very careful.  We have no exclusion from __set_page_dirty_buffers
1160          * here, and the (potentially unmapped) buffers may become dirty at
1161          * any time.  If a buffer becomes dirty here after we've inspected it
1162          * then we just miss that fact, and the page stays dirty.
1163          *
1164          * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1165          * handle that here by just cleaning them.
1166          */
1167
1168         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
1169         head = page_buffers(page);
1170         bh = head;
1171
1172         /*
1173          * Get all the dirty buffers mapped to disk addresses and
1174          * handle any aliases from the underlying blockdev's mapping.
1175          */
1176         do {
1177                 if (block > last_block) {
1178                         /*
1179                          * mapped buffers outside i_size will occur, because
1180                          * this page can be outside i_size when there is a
1181                          * truncate in progress.
1182                          */
1183                         /*
1184                          * The buffer was zeroed by block_write_full_page()
1185                          */
1186                         clear_buffer_dirty(bh);
1187                         set_buffer_uptodate(bh);
1188                 } else if (!buffer_mapped(bh) && buffer_dirty(bh)) {
1189                         WARN_ON(bh->b_size != blocksize);
1190                         err = btrfs_get_block(inode, block, bh, 0);
1191                         if (err)
1192                                 goto recover;
1193                         if (buffer_new(bh)) {
1194                                 /* blockdev mappings never come here */
1195                                 clear_buffer_new(bh);
1196                                 unmap_underlying_metadata(bh->b_bdev,
1197                                                         bh->b_blocknr);
1198                         }
1199                 }
1200                 bh = bh->b_this_page;
1201                 block++;
1202         } while (bh != head);
1203
1204         do {
1205                 if (!buffer_mapped(bh))
1206                         continue;
1207                 /*
1208                  * If it's a fully non-blocking write attempt and we cannot
1209                  * lock the buffer then redirty the page.  Note that this can
1210                  * potentially cause a busy-wait loop from pdflush and kswapd
1211                  * activity, but those code paths have their own higher-level
1212                  * throttling.
1213                  */
1214                 if (wbc->sync_mode != WB_SYNC_NONE || !wbc->nonblocking) {
1215                         lock_buffer(bh);
1216                 } else if (test_set_buffer_locked(bh)) {
1217                         redirty_page_for_writepage(wbc, page);
1218                         continue;
1219                 }
1220                 if (test_clear_buffer_dirty(bh) && bh->b_blocknr != 0) {
1221                         mark_buffer_async_write(bh);
1222                 } else {
1223                         unlock_buffer(bh);
1224                 }
1225         } while ((bh = bh->b_this_page) != head);
1226
1227         /*
1228          * The page and its buffers are protected by PageWriteback(), so we can
1229          * drop the bh refcounts early.
1230          */
1231         BUG_ON(PageWriteback(page));
1232         set_page_writeback(page);
1233
1234         do {
1235                 struct buffer_head *next = bh->b_this_page;
1236                 if (buffer_async_write(bh)) {
1237                         submit_bh(WRITE, bh);
1238                         nr_underway++;
1239                 }
1240                 bh = next;
1241         } while (bh != head);
1242         unlock_page(page);
1243
1244         err = 0;
1245 done:
1246         if (nr_underway == 0) {
1247                 /*
1248                  * The page was marked dirty, but the buffers were
1249                  * clean.  Someone wrote them back by hand with
1250                  * ll_rw_block/submit_bh.  A rare case.
1251                  */
1252                 int uptodate = 1;
1253                 do {
1254                         if (!buffer_uptodate(bh)) {
1255                                 uptodate = 0;
1256                                 break;
1257                         }
1258                         bh = bh->b_this_page;
1259                 } while (bh != head);
1260                 if (uptodate)
1261                         SetPageUptodate(page);
1262                 end_page_writeback(page);
1263                 /*
1264                  * The page and buffer_heads can be released at any time from
1265                  * here on.
1266                  */
1267                 wbc->pages_skipped++;   /* We didn't write this page */
1268         }
1269         return err;
1270
1271 recover:
1272         /*
1273          * ENOSPC, or some other error.  We may already have added some
1274          * blocks to the file, so we need to write these out to avoid
1275          * exposing stale data.
1276          * The page is currently locked and not marked for writeback
1277          */
1278         bh = head;
1279         /* Recovery: lock and submit the mapped buffers */
1280         do {
1281                 if (buffer_mapped(bh) && buffer_dirty(bh)) {
1282                         lock_buffer(bh);
1283                         mark_buffer_async_write(bh);
1284                 } else {
1285                         /*
1286                          * The buffer may have been set dirty during
1287                          * attachment to a dirty page.
1288                          */
1289                         clear_buffer_dirty(bh);
1290                 }
1291         } while ((bh = bh->b_this_page) != head);
1292         SetPageError(page);
1293         BUG_ON(PageWriteback(page));
1294         set_page_writeback(page);
1295         do {
1296                 struct buffer_head *next = bh->b_this_page;
1297                 if (buffer_async_write(bh)) {
1298                         clear_buffer_dirty(bh);
1299                         submit_bh(WRITE, bh);
1300                         nr_underway++;
1301                 }
1302                 bh = next;
1303         } while (bh != head);
1304         unlock_page(page);
1305         goto done;
1306 }
1307
1308 /*
1309  * The generic ->writepage function for buffer-backed address_spaces
1310  */
1311 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1312 {
1313         struct inode * const inode = page->mapping->host;
1314         loff_t i_size = i_size_read(inode);
1315         const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
1316         unsigned offset;
1317         void *kaddr;
1318
1319         /* Is the page fully inside i_size? */
1320         if (page->index < end_index)
1321                 return __btrfs_write_full_page(inode, page, wbc);
1322
1323         /* Is the page fully outside i_size? (truncate in progress) */
1324         offset = i_size & (PAGE_CACHE_SIZE-1);
1325         if (page->index >= end_index+1 || !offset) {
1326                 /*
1327                  * The page may have dirty, unmapped buffers.  For example,
1328                  * they may have been added in ext3_writepage().  Make them
1329                  * freeable here, so the page does not leak.
1330                  */
1331                 block_invalidatepage(page, 0);
1332                 unlock_page(page);
1333                 return 0; /* don't care */
1334         }
1335
1336         /*
1337          * The page straddles i_size.  It must be zeroed out on each and every
1338          * writepage invokation because it may be mmapped.  "A file is mapped
1339          * in multiples of the page size.  For a file that is not a multiple of
1340          * the  page size, the remaining memory is zeroed when mapped, and
1341          * writes to that region are not written out to the file."
1342          */
1343         kaddr = kmap_atomic(page, KM_USER0);
1344         memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
1345         flush_dcache_page(page);
1346         kunmap_atomic(kaddr, KM_USER0);
1347         return __btrfs_write_full_page(inode, page, wbc);
1348 }
1349
1350 static void btrfs_truncate(struct inode *inode)
1351 {
1352         struct btrfs_root *root = BTRFS_I(inode)->root;
1353         int ret;
1354         struct btrfs_trans_handle *trans;
1355
1356         if (!S_ISREG(inode->i_mode))
1357                 return;
1358         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1359                 return;
1360
1361         nobh_truncate_page(inode->i_mapping, inode->i_size);
1362
1363         /* FIXME, add redo link to tree so we don't leak on crash */
1364         mutex_lock(&root->fs_info->fs_mutex);
1365         trans = btrfs_start_transaction(root, 1);
1366         ret = btrfs_truncate_in_trans(trans, root, inode);
1367         BUG_ON(ret);
1368         ret = btrfs_end_transaction(trans, root);
1369         BUG_ON(ret);
1370         mutex_unlock(&root->fs_info->fs_mutex);
1371         mark_inode_dirty(inode);
1372 }
1373
1374 /*
1375  * Make sure any changes to nobh_commit_write() are reflected in
1376  * nobh_truncate_page(), since it doesn't call commit_write().
1377  */
1378 static int btrfs_commit_write(struct file *file, struct page *page,
1379                               unsigned from, unsigned to)
1380 {
1381         struct inode *inode = page->mapping->host;
1382         struct buffer_head *bh;
1383         loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1384
1385         SetPageUptodate(page);
1386         bh = page_buffers(page);
1387         if (buffer_mapped(bh) && bh->b_blocknr != 0) {
1388                 set_page_dirty(page);
1389         }
1390         if (pos > inode->i_size) {
1391                 i_size_write(inode, pos);
1392                 mark_inode_dirty(inode);
1393         }
1394         return 0;
1395 }
1396
1397 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
1398                                 struct page **prepared_pages,
1399                                 const char __user * buf)
1400 {
1401         long page_fault = 0;
1402         int i;
1403         int offset = pos & (PAGE_CACHE_SIZE - 1);
1404
1405         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
1406                 size_t count = min_t(size_t,
1407                                      PAGE_CACHE_SIZE - offset, write_bytes);
1408                 struct page *page = prepared_pages[i];
1409                 fault_in_pages_readable(buf, count);
1410
1411                 /* Copy data from userspace to the current page */
1412                 kmap(page);
1413                 page_fault = __copy_from_user(page_address(page) + offset,
1414                                               buf, count);
1415                 /* Flush processor's dcache for this page */
1416                 flush_dcache_page(page);
1417                 kunmap(page);
1418                 buf += count;
1419                 write_bytes -= count;
1420
1421                 if (page_fault)
1422                         break;
1423         }
1424         return page_fault ? -EFAULT : 0;
1425 }
1426
1427 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
1428 {
1429         size_t i;
1430         for (i = 0; i < num_pages; i++) {
1431                 if (!pages[i])
1432                         break;
1433                 unlock_page(pages[i]);
1434                 mark_page_accessed(pages[i]);
1435                 page_cache_release(pages[i]);
1436         }
1437 }
1438 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
1439                                    struct btrfs_root *root,
1440                                    struct file *file,
1441                                    struct page **pages,
1442                                    size_t num_pages,
1443                                    loff_t pos,
1444                                    size_t write_bytes)
1445 {
1446         int i;
1447         int offset;
1448         int err = 0;
1449         int ret;
1450         int this_write;
1451         struct inode *inode = file->f_path.dentry->d_inode;
1452         struct buffer_head *bh;
1453         struct btrfs_file_extent_item *ei;
1454
1455         for (i = 0; i < num_pages; i++) {
1456                 offset = pos & (PAGE_CACHE_SIZE -1);
1457                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1458                 /* FIXME, one block at a time */
1459
1460                 mutex_lock(&root->fs_info->fs_mutex);
1461                 trans = btrfs_start_transaction(root, 1);
1462
1463                 bh = page_buffers(pages[i]);
1464                 if (buffer_mapped(bh) && bh->b_blocknr == 0) {
1465                         struct btrfs_key key;
1466                         struct btrfs_path *path;
1467                         char *ptr;
1468                         u32 datasize;
1469
1470                         path = btrfs_alloc_path();
1471                         BUG_ON(!path);
1472                         key.objectid = inode->i_ino;
1473                         key.offset = pages[i]->index << PAGE_CACHE_SHIFT;
1474                         key.flags = 0;
1475                         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
1476                         BUG_ON(write_bytes >= PAGE_CACHE_SIZE);
1477                         datasize = offset +
1478                                 btrfs_file_extent_calc_inline_size(write_bytes);
1479                         ret = btrfs_insert_empty_item(trans, root, path, &key,
1480                                                       datasize);
1481                         BUG_ON(ret);
1482                         ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1483                                path->slots[0], struct btrfs_file_extent_item);
1484                         btrfs_set_file_extent_generation(ei, trans->transid);
1485                         btrfs_set_file_extent_type(ei,
1486                                                    BTRFS_FILE_EXTENT_INLINE);
1487                         ptr = btrfs_file_extent_inline_start(ei);
1488                         memcpy(ptr, bh->b_data, offset + write_bytes);
1489                         mark_buffer_dirty(path->nodes[0]);
1490                         btrfs_free_path(path);
1491                 } else {
1492                         btrfs_csum_file_block(trans, root, inode->i_ino,
1493                                       pages[i]->index << PAGE_CACHE_SHIFT,
1494                                       kmap(pages[i]), PAGE_CACHE_SIZE);
1495                         kunmap(pages[i]);
1496                 }
1497                 SetPageChecked(pages[i]);
1498                 ret = btrfs_end_transaction(trans, root);
1499                 BUG_ON(ret);
1500                 mutex_unlock(&root->fs_info->fs_mutex);
1501
1502                 ret = btrfs_commit_write(file, pages[i], offset,
1503                                          offset + this_write);
1504                 pos += this_write;
1505                 if (ret) {
1506                         err = ret;
1507                         goto failed;
1508                 }
1509                 WARN_ON(this_write > write_bytes);
1510                 write_bytes -= this_write;
1511         }
1512 failed:
1513         return err;
1514 }
1515
1516 static int drop_extents(struct btrfs_trans_handle *trans,
1517                           struct btrfs_root *root,
1518                           struct inode *inode,
1519                           u64 start, u64 end)
1520 {
1521         int ret;
1522         struct btrfs_key key;
1523         struct btrfs_leaf *leaf;
1524         int slot;
1525         struct btrfs_file_extent_item *extent;
1526         u64 extent_end = 0;
1527         int keep;
1528         struct btrfs_file_extent_item old;
1529         struct btrfs_path *path;
1530         u64 search_start = start;
1531         int bookend;
1532         int found_type;
1533         int found_extent;
1534         int found_inline;
1535
1536         path = btrfs_alloc_path();
1537         if (!path)
1538                 return -ENOMEM;
1539         while(1) {
1540                 btrfs_release_path(root, path);
1541                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1542                                                search_start, -1);
1543                 if (ret < 0)
1544                         goto out;
1545                 if (ret > 0) {
1546                         if (path->slots[0] == 0) {
1547                                 ret = 0;
1548                                 goto out;
1549                         }
1550                         path->slots[0]--;
1551                 }
1552                 keep = 0;
1553                 bookend = 0;
1554                 found_extent = 0;
1555                 found_inline = 0;
1556                 extent = NULL;
1557                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1558                 slot = path->slots[0];
1559                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
1560                 if (key.offset >= end || key.objectid != inode->i_ino) {
1561                         ret = 0;
1562                         goto out;
1563                 }
1564                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) {
1565                         ret = 0;
1566                         goto out;
1567                 }
1568                 extent = btrfs_item_ptr(leaf, slot,
1569                                         struct btrfs_file_extent_item);
1570                 found_type = btrfs_file_extent_type(extent);
1571                 if (found_type == BTRFS_FILE_EXTENT_REG) {
1572                         extent_end = key.offset +
1573                                 (btrfs_file_extent_num_blocks(extent) <<
1574                                  inode->i_blkbits);
1575                         found_extent = 1;
1576                 } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
1577                         found_inline = 1;
1578                         extent_end = key.offset +
1579                              btrfs_file_extent_inline_len(leaf->items + slot);
1580                 }
1581
1582                 if (!found_extent && !found_inline) {
1583                         ret = 0;
1584                         goto out;
1585                 }
1586
1587                 if (search_start >= extent_end) {
1588                         ret = 0;
1589                         goto out;
1590                 }
1591
1592                 search_start = extent_end;
1593
1594                 if (end < extent_end && end >= key.offset) {
1595                         if (found_extent) {
1596                                 memcpy(&old, extent, sizeof(old));
1597                                 ret = btrfs_inc_extent_ref(trans, root,
1598                                       btrfs_file_extent_disk_blocknr(&old),
1599                                       btrfs_file_extent_disk_num_blocks(&old));
1600                                 BUG_ON(ret);
1601                         }
1602                         WARN_ON(found_inline);
1603                         bookend = 1;
1604                 }
1605
1606                 if (start > key.offset) {
1607                         u64 new_num;
1608                         u64 old_num;
1609                         /* truncate existing extent */
1610                         keep = 1;
1611                         WARN_ON(start & (root->blocksize - 1));
1612                         if (found_extent) {
1613                                 new_num = (start - key.offset) >>
1614                                         inode->i_blkbits;
1615                                 old_num = btrfs_file_extent_num_blocks(extent);
1616                                 inode->i_blocks -= (old_num - new_num) << 3;
1617                                 btrfs_set_file_extent_num_blocks(extent,
1618                                                                  new_num);
1619                                 mark_buffer_dirty(path->nodes[0]);
1620                         } else {
1621                                 WARN_ON(1);
1622                                 /*
1623                                 ret = btrfs_truncate_item(trans, root, path,
1624                                                           start - key.offset);
1625                                 BUG_ON(ret);
1626                                 */
1627                         }
1628                 }
1629                 if (!keep) {
1630                         u64 disk_blocknr = 0;
1631                         u64 disk_num_blocks = 0;
1632                         u64 extent_num_blocks = 0;
1633                         if (found_extent) {
1634                                 disk_blocknr =
1635                                       btrfs_file_extent_disk_blocknr(extent);
1636                                 disk_num_blocks =
1637                                       btrfs_file_extent_disk_num_blocks(extent);
1638                                 extent_num_blocks =
1639                                       btrfs_file_extent_num_blocks(extent);
1640                         }
1641                         ret = btrfs_del_item(trans, root, path);
1642                         BUG_ON(ret);
1643                         btrfs_release_path(root, path);
1644                         if (found_extent) {
1645                                 inode->i_blocks -=
1646                                 btrfs_file_extent_num_blocks(extent) << 3;
1647                                 ret = btrfs_free_extent(trans, root,
1648                                                         disk_blocknr,
1649                                                         disk_num_blocks, 0);
1650                         }
1651
1652                         BUG_ON(ret);
1653                         if (!bookend && search_start >= end) {
1654                                 ret = 0;
1655                                 goto out;
1656                         }
1657                         if (!bookend)
1658                                 continue;
1659                 }
1660                 if (bookend && found_extent) {
1661                         /* create bookend */
1662                         struct btrfs_key ins;
1663                         ins.objectid = inode->i_ino;
1664                         ins.offset = end;
1665                         ins.flags = 0;
1666                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
1667
1668                         btrfs_release_path(root, path);
1669                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
1670                                                       sizeof(*extent));
1671                         BUG_ON(ret);
1672                         extent = btrfs_item_ptr(
1673                                     btrfs_buffer_leaf(path->nodes[0]),
1674                                     path->slots[0],
1675                                     struct btrfs_file_extent_item);
1676                         btrfs_set_file_extent_disk_blocknr(extent,
1677                                     btrfs_file_extent_disk_blocknr(&old));
1678                         btrfs_set_file_extent_disk_num_blocks(extent,
1679                                     btrfs_file_extent_disk_num_blocks(&old));
1680
1681                         btrfs_set_file_extent_offset(extent,
1682                                     btrfs_file_extent_offset(&old) +
1683                                     ((end - key.offset) >> inode->i_blkbits));
1684                         WARN_ON(btrfs_file_extent_num_blocks(&old) <
1685                                 (end - key.offset) >> inode->i_blkbits);
1686                         btrfs_set_file_extent_num_blocks(extent,
1687                                     btrfs_file_extent_num_blocks(&old) -
1688                                     ((end - key.offset) >> inode->i_blkbits));
1689
1690                         btrfs_set_file_extent_type(extent,
1691                                                    BTRFS_FILE_EXTENT_REG);
1692                         btrfs_set_file_extent_generation(extent,
1693                                     btrfs_file_extent_generation(&old));
1694                         btrfs_mark_buffer_dirty(path->nodes[0]);
1695                         inode->i_blocks +=
1696                                 btrfs_file_extent_num_blocks(extent) << 3;
1697                         ret = 0;
1698                         goto out;
1699                 }
1700         }
1701 out:
1702         btrfs_free_path(path);
1703         return ret;
1704 }
1705
1706 static int prepare_pages(struct btrfs_root *root,
1707                          struct file *file,
1708                          struct page **pages,
1709                          size_t num_pages,
1710                          loff_t pos,
1711                          unsigned long first_index,
1712                          unsigned long last_index,
1713                          size_t write_bytes,
1714                          u64 alloc_extent_start)
1715 {
1716         int i;
1717         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1718         struct inode *inode = file->f_path.dentry->d_inode;
1719         int offset;
1720         int err = 0;
1721         int this_write;
1722         struct buffer_head *bh;
1723         struct buffer_head *head;
1724         loff_t isize = i_size_read(inode);
1725
1726         memset(pages, 0, num_pages * sizeof(struct page *));
1727
1728         for (i = 0; i < num_pages; i++) {
1729                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
1730                 if (!pages[i]) {
1731                         err = -ENOMEM;
1732                         goto failed_release;
1733                 }
1734                 offset = pos & (PAGE_CACHE_SIZE -1);
1735                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1736                 create_empty_buffers(pages[i], root->fs_info->sb->s_blocksize,
1737                                      (1 << BH_Uptodate));
1738                 head = page_buffers(pages[i]);
1739                 bh = head;
1740                 do {
1741                         err = btrfs_map_bh_to_logical(root, bh,
1742                                                       alloc_extent_start);
1743                         BUG_ON(err);
1744                         if (err)
1745                                 goto failed_truncate;
1746                         bh = bh->b_this_page;
1747                         if (alloc_extent_start)
1748                                 alloc_extent_start++;
1749                 } while (bh != head);
1750                 pos += this_write;
1751                 WARN_ON(this_write > write_bytes);
1752                 write_bytes -= this_write;
1753         }
1754         return 0;
1755
1756 failed_release:
1757         btrfs_drop_pages(pages, num_pages);
1758         return err;
1759
1760 failed_truncate:
1761         btrfs_drop_pages(pages, num_pages);
1762         if (pos > isize)
1763                 vmtruncate(inode, isize);
1764         return err;
1765 }
1766
1767 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1768                                 size_t count, loff_t *ppos)
1769 {
1770         loff_t pos;
1771         size_t num_written = 0;
1772         int err = 0;
1773         int ret = 0;
1774         struct inode *inode = file->f_path.dentry->d_inode;
1775         struct btrfs_root *root = BTRFS_I(inode)->root;
1776         struct page *pages[8];
1777         struct page *pinned[2] = { NULL, NULL };
1778         unsigned long first_index;
1779         unsigned long last_index;
1780         u64 start_pos;
1781         u64 num_blocks;
1782         u64 alloc_extent_start;
1783         struct btrfs_trans_handle *trans;
1784         struct btrfs_key ins;
1785
1786         if (file->f_flags & O_DIRECT)
1787                 return -EINVAL;
1788         pos = *ppos;
1789         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1790         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1791         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1792         if (err)
1793                 goto out;
1794         if (count == 0)
1795                 goto out;
1796         err = remove_suid(file->f_path.dentry);
1797         if (err)
1798                 goto out;
1799         file_update_time(file);
1800
1801         start_pos = pos & ~((u64)PAGE_CACHE_SIZE - 1);
1802         num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
1803                         inode->i_blkbits;
1804
1805         mutex_lock(&inode->i_mutex);
1806         first_index = pos >> PAGE_CACHE_SHIFT;
1807         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1808
1809         if ((first_index << PAGE_CACHE_SHIFT) < inode->i_size &&
1810             (pos & (PAGE_CACHE_SIZE - 1))) {
1811                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
1812                 if (!PageUptodate(pinned[0])) {
1813                         ret = mpage_readpage(pinned[0], btrfs_get_block);
1814                         BUG_ON(ret);
1815                 } else {
1816                         unlock_page(pinned[0]);
1817                 }
1818         }
1819         if (first_index != last_index &&
1820             (last_index << PAGE_CACHE_SHIFT) < inode->i_size &&
1821             (count & (PAGE_CACHE_SIZE - 1))) {
1822                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
1823                 if (!PageUptodate(pinned[1])) {
1824                         ret = mpage_readpage(pinned[1], btrfs_get_block);
1825                         BUG_ON(ret);
1826                 } else {
1827                         unlock_page(pinned[1]);
1828                 }
1829         }
1830
1831         mutex_lock(&root->fs_info->fs_mutex);
1832         trans = btrfs_start_transaction(root, 1);
1833         if (!trans) {
1834                 err = -ENOMEM;
1835                 mutex_unlock(&root->fs_info->fs_mutex);
1836                 goto out_unlock;
1837         }
1838         /* FIXME blocksize != 4096 */
1839         inode->i_blocks += num_blocks << 3;
1840         if (start_pos < inode->i_size) {
1841                 /* FIXME blocksize != pagesize */
1842                 ret = drop_extents(trans, root, inode,
1843                                    start_pos,
1844                                    (pos + count + root->blocksize -1) &
1845                                    ~((u64)root->blocksize - 1));
1846                 BUG_ON(ret);
1847         }
1848         if (inode->i_size >= PAGE_CACHE_SIZE || pos + count < inode->i_size ||
1849             pos + count - start_pos > BTRFS_MAX_INLINE_DATA_SIZE(root)) {
1850                 ret = btrfs_alloc_extent(trans, root, num_blocks, 1,
1851                                  (u64)-1, &ins);
1852                 BUG_ON(ret);
1853                 ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
1854                                        start_pos, ins.objectid, ins.offset);
1855                 BUG_ON(ret);
1856         } else {
1857                 ins.offset = 0;
1858                 ins.objectid = 0;
1859         }
1860         BUG_ON(ret);
1861         alloc_extent_start = ins.objectid;
1862         ret = btrfs_end_transaction(trans, root);
1863         mutex_unlock(&root->fs_info->fs_mutex);
1864
1865         while(count > 0) {
1866                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1867                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1868                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1869                                         PAGE_CACHE_SHIFT;
1870
1871                 memset(pages, 0, sizeof(pages));
1872                 ret = prepare_pages(root, file, pages, num_pages,
1873                                     pos, first_index, last_index,
1874                                     write_bytes, alloc_extent_start);
1875                 BUG_ON(ret);
1876
1877                 /* FIXME blocks != pagesize */
1878                 if (alloc_extent_start)
1879                         alloc_extent_start += num_pages;
1880                 ret = btrfs_copy_from_user(pos, num_pages,
1881                                            write_bytes, pages, buf);
1882                 BUG_ON(ret);
1883
1884                 ret = dirty_and_release_pages(NULL, root, file, pages,
1885                                               num_pages, pos, write_bytes);
1886                 BUG_ON(ret);
1887                 btrfs_drop_pages(pages, num_pages);
1888
1889                 buf += write_bytes;
1890                 count -= write_bytes;
1891                 pos += write_bytes;
1892                 num_written += write_bytes;
1893
1894                 balance_dirty_pages_ratelimited(inode->i_mapping);
1895                 cond_resched();
1896         }
1897 out_unlock:
1898         mutex_unlock(&inode->i_mutex);
1899 out:
1900         if (pinned[0])
1901                 page_cache_release(pinned[0]);
1902         if (pinned[1])
1903                 page_cache_release(pinned[1]);
1904         *ppos = pos;
1905         current->backing_dev_info = NULL;
1906         mark_inode_dirty(inode);
1907         return num_written ? num_written : err;
1908 }
1909
1910 #if 0
1911 static ssize_t inline_one_page(struct btrfs_root *root, struct inode *inode,
1912                            struct page *page, loff_t pos,
1913                            size_t offset, size_t write_bytes)
1914 {
1915         struct btrfs_path *path;
1916         struct btrfs_trans_handle *trans;
1917         struct btrfs_key key;
1918         struct btrfs_leaf *leaf;
1919         struct btrfs_key found_key;
1920         int ret;
1921         size_t copy_size = 0;
1922         char *dst = NULL;
1923         int err = 0;
1924         size_t num_written = 0;
1925
1926         path = btrfs_alloc_path();
1927         BUG_ON(!path);
1928         mutex_lock(&root->fs_info->fs_mutex);
1929         trans = btrfs_start_transaction(root, 1);
1930         key.objectid = inode->i_ino;
1931         key.flags = 0;
1932         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
1933
1934 again:
1935         key.offset = pos;
1936         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1937         if (ret < 0) {
1938                 err = ret;
1939                 goto out;
1940         }
1941         if (ret == 0) {
1942                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1943                 btrfs_disk_key_to_cpu(&found_key,
1944                                       &leaf->items[path->slots[0]].key);
1945                 copy_size = btrfs_item_size(leaf->items + path->slots[0]);
1946                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1947                 copy_size = min(write_bytes, copy_size);
1948                 goto copyit;
1949         } else {
1950                 int slot = path->slots[0];
1951                 if (slot > 0) {
1952                         slot--;
1953                 }
1954                 // FIXME find max key
1955                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1956                 btrfs_disk_key_to_cpu(&found_key,
1957                                       &leaf->items[slot].key);
1958                 if (found_key.objectid != inode->i_ino)
1959                         goto insert;
1960                 if (btrfs_key_type(&found_key) != BTRFS_INLINE_DATA_KEY)
1961                         goto insert;
1962                 copy_size = btrfs_item_size(leaf->items + slot);
1963                 if (found_key.offset + copy_size <= pos)
1964                         goto insert;
1965                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1966                 dst += pos - found_key.offset;
1967                 copy_size = copy_size - (pos - found_key.offset);
1968                 BUG_ON(copy_size < 0);
1969                 copy_size = min(write_bytes, copy_size);
1970                 WARN_ON(copy_size == 0);
1971                 goto copyit;
1972         }
1973 insert:
1974         btrfs_release_path(root, path);
1975         copy_size = min(write_bytes,
1976                         (size_t)BTRFS_LEAF_DATA_SIZE(root) -
1977                         sizeof(struct btrfs_item) * 4);
1978         ret = btrfs_insert_empty_item(trans, root, path, &key, copy_size);
1979         BUG_ON(ret);
1980         dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1981                              path->slots[0], char);
1982 copyit:
1983         WARN_ON(copy_size == 0);
1984         WARN_ON(dst + copy_size >
1985                 btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1986                                                  path->slots[0], char) +
1987                 btrfs_item_size(btrfs_buffer_leaf(path->nodes[0])->items +
1988                                                   path->slots[0]));
1989         btrfs_memcpy(root, path->nodes[0]->b_data, dst,
1990                      page_address(page) + offset, copy_size);
1991         mark_buffer_dirty(path->nodes[0]);
1992         btrfs_release_path(root, path);
1993         pos += copy_size;
1994         offset += copy_size;
1995         num_written += copy_size;
1996         write_bytes -= copy_size;
1997         if (write_bytes)
1998                 goto again;
1999 out:
2000         btrfs_free_path(path);
2001         ret = btrfs_end_transaction(trans, root);
2002         BUG_ON(ret);
2003         mutex_unlock(&root->fs_info->fs_mutex);
2004         return num_written ? num_written : err;
2005 }
2006
2007 static ssize_t btrfs_file_inline_write(struct file *file,
2008                                        const char __user *buf,
2009                                        size_t count, loff_t *ppos)
2010 {
2011         loff_t pos;
2012         size_t num_written = 0;
2013         int err = 0;
2014         int ret = 0;
2015         struct inode *inode = file->f_path.dentry->d_inode;
2016         struct btrfs_root *root = BTRFS_I(inode)->root;
2017         unsigned long page_index;
2018
2019         if (file->f_flags & O_DIRECT)
2020                 return -EINVAL;
2021         pos = *ppos;
2022
2023         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2024         current->backing_dev_info = inode->i_mapping->backing_dev_info;
2025         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2026         if (err)
2027                 goto out;
2028         if (count == 0)
2029                 goto out;
2030         err = remove_suid(file->f_path.dentry);
2031         if (err)
2032                 goto out;
2033         file_update_time(file);
2034         mutex_lock(&inode->i_mutex);
2035         while(count > 0) {
2036                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
2037                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
2038                 struct page *page;
2039
2040                 page_index = pos >> PAGE_CACHE_SHIFT;
2041                 page = grab_cache_page(inode->i_mapping, page_index);
2042                 if (!PageUptodate(page)) {
2043                         ret = mpage_readpage(page, btrfs_get_block);
2044                         BUG_ON(ret);
2045                         lock_page(page);
2046                 }
2047                 ret = btrfs_copy_from_user(pos, 1,
2048                                            write_bytes, &page, buf);
2049                 BUG_ON(ret);
2050                 write_bytes = inline_one_page(root, inode, page, pos,
2051                                       offset, write_bytes);
2052                 SetPageUptodate(page);
2053                 if (write_bytes > 0 && pos + write_bytes > inode->i_size) {
2054                         i_size_write(inode, pos + write_bytes);
2055                         mark_inode_dirty(inode);
2056                 }
2057                 page_cache_release(page);
2058                 unlock_page(page);
2059                 if (write_bytes < 0)
2060                         goto out_unlock;
2061                 buf += write_bytes;
2062                 count -= write_bytes;
2063                 pos += write_bytes;
2064                 num_written += write_bytes;
2065
2066                 balance_dirty_pages_ratelimited(inode->i_mapping);
2067                 cond_resched();
2068         }
2069 out_unlock:
2070         mutex_unlock(&inode->i_mutex);
2071 out:
2072         *ppos = pos;
2073         current->backing_dev_info = NULL;
2074         return num_written ? num_written : err;
2075 }
2076 #endif
2077
2078 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
2079                         unsigned long offset, unsigned long size)
2080 {
2081         char *kaddr;
2082         unsigned long left, count = desc->count;
2083         struct inode *inode = page->mapping->host;
2084
2085         if (size > count)
2086                 size = count;
2087
2088         if (!PageChecked(page)) {
2089                 /* FIXME, do it per block */
2090                 struct btrfs_root *root = BTRFS_I(inode)->root;
2091
2092                 int ret = btrfs_csum_verify_file_block(root,
2093                                   page->mapping->host->i_ino,
2094                                   page->index << PAGE_CACHE_SHIFT,
2095                                   kmap(page), PAGE_CACHE_SIZE);
2096                 if (ret) {
2097                         printk("failed to verify ino %lu page %lu\n",
2098                                page->mapping->host->i_ino,
2099                                page->index);
2100                         memset(page_address(page), 0, PAGE_CACHE_SIZE);
2101                 }
2102                 SetPageChecked(page);
2103                 kunmap(page);
2104         }
2105         /*
2106          * Faults on the destination of a read are common, so do it before
2107          * taking the kmap.
2108          */
2109         if (!fault_in_pages_writeable(desc->arg.buf, size)) {
2110                 kaddr = kmap_atomic(page, KM_USER0);
2111                 left = __copy_to_user_inatomic(desc->arg.buf,
2112                                                 kaddr + offset, size);
2113                 kunmap_atomic(kaddr, KM_USER0);
2114                 if (left == 0)
2115                         goto success;
2116         }
2117
2118         /* Do it the slow way */
2119         kaddr = kmap(page);
2120         left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
2121         kunmap(page);
2122
2123         if (left) {
2124                 size -= left;
2125                 desc->error = -EFAULT;
2126         }
2127 success:
2128         desc->count = count - size;
2129         desc->written += size;
2130         desc->arg.buf += size;
2131         return size;
2132 }
2133
2134 /**
2135  * btrfs_file_aio_read - filesystem read routine
2136  * @iocb:       kernel I/O control block
2137  * @iov:        io vector request
2138  * @nr_segs:    number of segments in the iovec
2139  * @pos:        current file position
2140  */
2141 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
2142                                    unsigned long nr_segs, loff_t pos)
2143 {
2144         struct file *filp = iocb->ki_filp;
2145         ssize_t retval;
2146         unsigned long seg;
2147         size_t count;
2148         loff_t *ppos = &iocb->ki_pos;
2149
2150         count = 0;
2151         for (seg = 0; seg < nr_segs; seg++) {
2152                 const struct iovec *iv = &iov[seg];
2153
2154                 /*
2155                  * If any segment has a negative length, or the cumulative
2156                  * length ever wraps negative then return -EINVAL.
2157                  */
2158                 count += iv->iov_len;
2159                 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
2160                         return -EINVAL;
2161                 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
2162                         continue;
2163                 if (seg == 0)
2164                         return -EFAULT;
2165                 nr_segs = seg;
2166                 count -= iv->iov_len;   /* This segment is no good */
2167                 break;
2168         }
2169         retval = 0;
2170         if (count) {
2171                 for (seg = 0; seg < nr_segs; seg++) {
2172                         read_descriptor_t desc;
2173
2174                         desc.written = 0;
2175                         desc.arg.buf = iov[seg].iov_base;
2176                         desc.count = iov[seg].iov_len;
2177                         if (desc.count == 0)
2178                                 continue;
2179                         desc.error = 0;
2180                         do_generic_file_read(filp, ppos, &desc,
2181                                              btrfs_read_actor);
2182                         retval += desc.written;
2183                         if (desc.error) {
2184                                 retval = retval ?: desc.error;
2185                                 break;
2186                         }
2187                 }
2188         }
2189         return retval;
2190 }
2191
2192 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
2193 {
2194         struct btrfs_trans_handle *trans;
2195         struct btrfs_key key;
2196         struct btrfs_root_item root_item;
2197         struct btrfs_inode_item *inode_item;
2198         struct buffer_head *subvol;
2199         struct btrfs_leaf *leaf;
2200         struct btrfs_root *new_root;
2201         struct inode *inode;
2202         int ret;
2203         u64 objectid;
2204         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
2205
2206         mutex_lock(&root->fs_info->fs_mutex);
2207         trans = btrfs_start_transaction(root, 1);
2208         BUG_ON(!trans);
2209
2210         subvol = btrfs_alloc_free_block(trans, root);
2211         leaf = btrfs_buffer_leaf(subvol);
2212         btrfs_set_header_nritems(&leaf->header, 0);
2213         btrfs_set_header_level(&leaf->header, 0);
2214         btrfs_set_header_blocknr(&leaf->header, bh_blocknr(subvol));
2215         btrfs_set_header_generation(&leaf->header, trans->transid);
2216         memcpy(leaf->header.fsid, root->fs_info->disk_super->fsid,
2217                sizeof(leaf->header.fsid));
2218
2219         inode_item = &root_item.inode;
2220         memset(inode_item, 0, sizeof(*inode_item));
2221         btrfs_set_inode_generation(inode_item, 1);
2222         btrfs_set_inode_size(inode_item, 3);
2223         btrfs_set_inode_nlink(inode_item, 1);
2224         btrfs_set_inode_nblocks(inode_item, 1);
2225         btrfs_set_inode_mode(inode_item, S_IFDIR | 0755);
2226
2227         btrfs_set_root_blocknr(&root_item, bh_blocknr(subvol));
2228         btrfs_set_root_refs(&root_item, 1);
2229
2230         mark_buffer_dirty(subvol);
2231         brelse(subvol);
2232         subvol = NULL;
2233
2234         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2235                                        0, &objectid);
2236         BUG_ON(ret);
2237
2238         btrfs_set_root_dirid(&root_item, new_dirid);
2239
2240         key.objectid = objectid;
2241         key.offset = 1;
2242         key.flags = 0;
2243         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2244         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2245                                 &root_item);
2246         BUG_ON(ret);
2247
2248         /*
2249          * insert the directory item
2250          */
2251         key.offset = (u64)-1;
2252         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2253                                     name, namelen,
2254                                     root->fs_info->sb->s_root->d_inode->i_ino,
2255                                     &key, 0);
2256         BUG_ON(ret);
2257
2258         ret = btrfs_commit_transaction(trans, root);
2259         BUG_ON(ret);
2260
2261         new_root = btrfs_read_fs_root(root->fs_info, &key);
2262         BUG_ON(!new_root);
2263
2264         trans = btrfs_start_transaction(new_root, 1);
2265         BUG_ON(!trans);
2266
2267         inode = btrfs_new_inode(trans, new_root, new_dirid, S_IFDIR | 0700);
2268         inode->i_op = &btrfs_dir_inode_operations;
2269         inode->i_fop = &btrfs_dir_file_operations;
2270
2271         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
2272         BUG_ON(ret);
2273
2274         inode->i_nlink = 1;
2275         inode->i_size = 6;
2276         ret = btrfs_update_inode(trans, new_root, inode);
2277         BUG_ON(ret);
2278
2279         ret = btrfs_commit_transaction(trans, new_root);
2280         BUG_ON(ret);
2281
2282         iput(inode);
2283
2284         mutex_unlock(&root->fs_info->fs_mutex);
2285         return 0;
2286 }
2287
2288 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2289 {
2290         struct btrfs_trans_handle *trans;
2291         struct btrfs_key key;
2292         struct btrfs_root_item new_root_item;
2293         int ret;
2294         u64 objectid;
2295
2296         if (!root->ref_cows)
2297                 return -EINVAL;
2298
2299         mutex_lock(&root->fs_info->fs_mutex);
2300         trans = btrfs_start_transaction(root, 1);
2301         BUG_ON(!trans);
2302
2303         ret = btrfs_update_inode(trans, root, root->inode);
2304         BUG_ON(ret);
2305
2306         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2307                                        0, &objectid);
2308         BUG_ON(ret);
2309
2310         memcpy(&new_root_item, &root->root_item,
2311                sizeof(new_root_item));
2312
2313         key.objectid = objectid;
2314         key.offset = 1;
2315         key.flags = 0;
2316         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2317         btrfs_set_root_blocknr(&new_root_item, bh_blocknr(root->node));
2318
2319         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2320                                 &new_root_item);
2321         BUG_ON(ret);
2322
2323         /*
2324          * insert the directory item
2325          */
2326         key.offset = (u64)-1;
2327         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2328                                     name, namelen,
2329                                     root->fs_info->sb->s_root->d_inode->i_ino,
2330                                     &key, 0);
2331
2332         BUG_ON(ret);
2333
2334         ret = btrfs_inc_root_ref(trans, root);
2335         BUG_ON(ret);
2336
2337         ret = btrfs_commit_transaction(trans, root);
2338         BUG_ON(ret);
2339         mutex_unlock(&root->fs_info->fs_mutex);
2340         return 0;
2341 }
2342
2343 static int add_disk(struct btrfs_root *root, char *name, int namelen)
2344 {
2345         struct block_device *bdev;
2346         struct btrfs_path *path;
2347         struct super_block *sb = root->fs_info->sb;
2348         struct btrfs_root *dev_root = root->fs_info->dev_root;
2349         struct btrfs_trans_handle *trans;
2350         struct btrfs_device_item *dev_item;
2351         struct btrfs_key key;
2352         u16 item_size;
2353         u64 num_blocks;
2354         u64 new_blocks;
2355         u64 device_id;
2356         int ret;
2357
2358 printk("adding disk %s\n", name);
2359         path = btrfs_alloc_path();
2360         if (!path)
2361                 return -ENOMEM;
2362         num_blocks = btrfs_super_total_blocks(root->fs_info->disk_super);
2363         bdev = open_bdev_excl(name, O_RDWR, sb);
2364         if (IS_ERR(bdev)) {
2365                 ret = PTR_ERR(bdev);
2366 printk("open bdev excl failed ret %d\n", ret);
2367                 goto out_nolock;
2368         }
2369         set_blocksize(bdev, sb->s_blocksize);
2370         new_blocks = bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2371         key.objectid = num_blocks;
2372         key.offset = new_blocks;
2373         key.flags = 0;
2374         btrfs_set_key_type(&key, BTRFS_DEV_ITEM_KEY);
2375
2376         mutex_lock(&dev_root->fs_info->fs_mutex);
2377         trans = btrfs_start_transaction(dev_root, 1);
2378         item_size = sizeof(*dev_item) + namelen;
2379 printk("insert empty on %Lu %Lu %u size %d\n", num_blocks, new_blocks, key.flags, item_size);
2380         ret = btrfs_insert_empty_item(trans, dev_root, path, &key, item_size);
2381         if (ret) {
2382 printk("insert failed %d\n", ret);
2383                 close_bdev_excl(bdev);
2384                 if (ret > 0)
2385                         ret = -EEXIST;
2386                 goto out;
2387         }
2388         dev_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
2389                                   path->slots[0], struct btrfs_device_item);
2390         btrfs_set_device_pathlen(dev_item, namelen);
2391         memcpy(dev_item + 1, name, namelen);
2392
2393         device_id = btrfs_super_last_device_id(root->fs_info->disk_super) + 1;
2394         btrfs_set_super_last_device_id(root->fs_info->disk_super, device_id);
2395         btrfs_set_device_id(dev_item, device_id);
2396         mark_buffer_dirty(path->nodes[0]);
2397
2398         ret = btrfs_insert_dev_radix(root, bdev, device_id, num_blocks,
2399                                      new_blocks);
2400
2401         if (!ret) {
2402                 btrfs_set_super_total_blocks(root->fs_info->disk_super,
2403                                              num_blocks + new_blocks);
2404                 i_size_write(root->fs_info->btree_inode,
2405                              (num_blocks + new_blocks) <<
2406                              root->fs_info->btree_inode->i_blkbits);
2407         }
2408
2409 out:
2410         ret = btrfs_commit_transaction(trans, dev_root);
2411         BUG_ON(ret);
2412         mutex_unlock(&root->fs_info->fs_mutex);
2413 out_nolock:
2414         btrfs_free_path(path);
2415
2416         return ret;
2417 }
2418
2419 static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
2420                        cmd, unsigned long arg)
2421 {
2422         struct btrfs_root *root = BTRFS_I(inode)->root;
2423         struct btrfs_ioctl_vol_args vol_args;
2424         int ret = 0;
2425         int namelen;
2426         struct btrfs_path *path;
2427         u64 root_dirid;
2428
2429         switch (cmd) {
2430         case BTRFS_IOC_SNAP_CREATE:
2431                 if (copy_from_user(&vol_args,
2432                                    (struct btrfs_ioctl_vol_args __user *)arg,
2433                                    sizeof(vol_args)))
2434                         return -EFAULT;
2435                 namelen = strlen(vol_args.name);
2436                 if (namelen > BTRFS_VOL_NAME_MAX)
2437                         return -EINVAL;
2438                 path = btrfs_alloc_path();
2439                 if (!path)
2440                         return -ENOMEM;
2441                 root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2442                 mutex_lock(&root->fs_info->fs_mutex);
2443                 ret = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2444                                     path, root_dirid,
2445                                     vol_args.name, namelen, 0);
2446                 mutex_unlock(&root->fs_info->fs_mutex);
2447                 btrfs_free_path(path);
2448                 if (ret == 0)
2449                         return -EEXIST;
2450
2451                 if (root == root->fs_info->tree_root)
2452                         ret = create_subvol(root, vol_args.name, namelen);
2453                 else
2454                         ret = create_snapshot(root, vol_args.name, namelen);
2455                 WARN_ON(ret);
2456                 break;
2457         case BTRFS_IOC_ADD_DISK:
2458                 if (copy_from_user(&vol_args,
2459                                    (struct btrfs_ioctl_vol_args __user *)arg,
2460                                    sizeof(vol_args)))
2461                         return -EFAULT;
2462                 namelen = strlen(vol_args.name);
2463                 if (namelen > BTRFS_VOL_NAME_MAX)
2464                         return -EINVAL;
2465                 vol_args.name[namelen] = '\0';
2466                 ret = add_disk(root, vol_args.name, namelen);
2467                 break;
2468         default:
2469                 return -ENOTTY;
2470         }
2471         return ret;
2472 }
2473
2474 static struct kmem_cache *btrfs_inode_cachep;
2475 struct kmem_cache *btrfs_trans_handle_cachep;
2476 struct kmem_cache *btrfs_transaction_cachep;
2477 struct kmem_cache *btrfs_bit_radix_cachep;
2478 struct kmem_cache *btrfs_path_cachep;
2479
2480 /*
2481  * Called inside transaction, so use GFP_NOFS
2482  */
2483 static struct inode *btrfs_alloc_inode(struct super_block *sb)
2484 {
2485         struct btrfs_inode *ei;
2486
2487         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2488         if (!ei)
2489                 return NULL;
2490         return &ei->vfs_inode;
2491 }
2492
2493 static void btrfs_destroy_inode(struct inode *inode)
2494 {
2495         WARN_ON(!list_empty(&inode->i_dentry));
2496         WARN_ON(inode->i_data.nrpages);
2497
2498         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2499 }
2500
2501 static void init_once(void * foo, struct kmem_cache * cachep,
2502                       unsigned long flags)
2503 {
2504         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2505
2506         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2507             SLAB_CTOR_CONSTRUCTOR) {
2508                 inode_init_once(&ei->vfs_inode);
2509         }
2510 }
2511
2512 static int init_inodecache(void)
2513 {
2514         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
2515                                              sizeof(struct btrfs_inode),
2516                                              0, (SLAB_RECLAIM_ACCOUNT|
2517                                                 SLAB_MEM_SPREAD),
2518                                              init_once, NULL);
2519         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
2520                                              sizeof(struct btrfs_trans_handle),
2521                                              0, (SLAB_RECLAIM_ACCOUNT|
2522                                                 SLAB_MEM_SPREAD),
2523                                              NULL, NULL);
2524         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
2525                                              sizeof(struct btrfs_transaction),
2526                                              0, (SLAB_RECLAIM_ACCOUNT|
2527                                                 SLAB_MEM_SPREAD),
2528                                              NULL, NULL);
2529         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
2530                                              sizeof(struct btrfs_transaction),
2531                                              0, (SLAB_RECLAIM_ACCOUNT|
2532                                                 SLAB_MEM_SPREAD),
2533                                              NULL, NULL);
2534         btrfs_bit_radix_cachep = kmem_cache_create("btrfs_radix",
2535                                              256,
2536                                              0, (SLAB_RECLAIM_ACCOUNT|
2537                                                 SLAB_MEM_SPREAD |
2538                                                 SLAB_DESTROY_BY_RCU),
2539                                              NULL, NULL);
2540         if (btrfs_inode_cachep == NULL || btrfs_trans_handle_cachep == NULL ||
2541             btrfs_transaction_cachep == NULL || btrfs_bit_radix_cachep == NULL)
2542                 return -ENOMEM;
2543         return 0;
2544 }
2545
2546 static void destroy_inodecache(void)
2547 {
2548         kmem_cache_destroy(btrfs_inode_cachep);
2549         kmem_cache_destroy(btrfs_trans_handle_cachep);
2550         kmem_cache_destroy(btrfs_transaction_cachep);
2551         kmem_cache_destroy(btrfs_bit_radix_cachep);
2552         kmem_cache_destroy(btrfs_path_cachep);
2553 }
2554
2555 static int btrfs_get_sb(struct file_system_type *fs_type,
2556         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2557 {
2558         return get_sb_bdev(fs_type, flags, dev_name, data,
2559                            btrfs_fill_super, mnt);
2560 }
2561
2562
2563 static int btrfs_getattr(struct vfsmount *mnt,
2564                          struct dentry *dentry, struct kstat *stat)
2565 {
2566         struct inode *inode = dentry->d_inode;
2567         generic_fillattr(inode, stat);
2568         stat->blksize = 256 * 1024;
2569         return 0;
2570 }
2571
2572 static struct file_system_type btrfs_fs_type = {
2573         .owner          = THIS_MODULE,
2574         .name           = "btrfs",
2575         .get_sb         = btrfs_get_sb,
2576         .kill_sb        = kill_block_super,
2577         .fs_flags       = FS_REQUIRES_DEV,
2578 };
2579
2580 static struct super_operations btrfs_super_ops = {
2581         .statfs         = simple_statfs,
2582         .delete_inode   = btrfs_delete_inode,
2583         .put_super      = btrfs_put_super,
2584         .read_inode     = btrfs_read_locked_inode,
2585         .write_super    = btrfs_write_super,
2586         .sync_fs        = btrfs_sync_fs,
2587         .write_inode    = btrfs_write_inode,
2588         .alloc_inode    = btrfs_alloc_inode,
2589         .destroy_inode  = btrfs_destroy_inode,
2590 };
2591
2592 static struct inode_operations btrfs_dir_inode_operations = {
2593         .lookup         = btrfs_lookup,
2594         .create         = btrfs_create,
2595         .unlink         = btrfs_unlink,
2596         .mkdir          = btrfs_mkdir,
2597         .rmdir          = btrfs_rmdir,
2598 };
2599
2600 static struct inode_operations btrfs_dir_ro_inode_operations = {
2601         .lookup         = btrfs_lookup,
2602 };
2603
2604 static struct file_operations btrfs_dir_file_operations = {
2605         .llseek         = generic_file_llseek,
2606         .read           = generic_read_dir,
2607         .readdir        = btrfs_readdir,
2608         .ioctl          = btrfs_ioctl,
2609 };
2610
2611 static struct address_space_operations btrfs_aops = {
2612         .readpage       = btrfs_readpage,
2613         .writepage      = btrfs_writepage,
2614         .sync_page      = block_sync_page,
2615         .prepare_write  = btrfs_prepare_write,
2616         .commit_write   = btrfs_commit_write,
2617 };
2618
2619 static struct inode_operations btrfs_file_inode_operations = {
2620         .truncate       = btrfs_truncate,
2621         .getattr        = btrfs_getattr,
2622 };
2623
2624 static struct file_operations btrfs_file_operations = {
2625         .llseek         = generic_file_llseek,
2626         .read           = do_sync_read,
2627         .aio_read       = btrfs_file_aio_read,
2628         .write          = btrfs_file_write,
2629         .mmap           = generic_file_mmap,
2630         .open           = generic_file_open,
2631         .ioctl          = btrfs_ioctl,
2632 };
2633
2634 static int __init init_btrfs_fs(void)
2635 {
2636         int err;
2637         printk("btrfs loaded!\n");
2638         err = init_inodecache();
2639         if (err)
2640                 return err;
2641         kset_set_kset_s(&btrfs_subsys, fs_subsys);
2642         err = subsystem_register(&btrfs_subsys);
2643         if (err)
2644                 goto out;
2645         return register_filesystem(&btrfs_fs_type);
2646 out:
2647         destroy_inodecache();
2648         return err;
2649 }
2650
2651 static void __exit exit_btrfs_fs(void)
2652 {
2653         destroy_inodecache();
2654         unregister_filesystem(&btrfs_fs_type);
2655         subsystem_unregister(&btrfs_subsys);
2656         printk("btrfs unloaded\n");
2657 }
2658
2659 module_init(init_btrfs_fs)
2660 module_exit(exit_btrfs_fs)
2661
2662 MODULE_LICENSE("GPL");