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         ret = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
 154                                           objectid, -1);
 155         BUG_ON(ret);
 156         ret = btrfs_del_item(trans, root, path);
 157         BUG_ON(ret);
 158         dentry->d_inode->i_ctime = dir->i_ctime;
 159 err:
 160         btrfs_release_path(root, path);
 161         btrfs_free_path(path);
 162         if (ret == 0) {
 163                 inode_dec_link_count(dentry->d_inode);
 164                 dir->i_size -= name_len * 2;
 165                 mark_inode_dirty(dir);
 166         }
 167         return ret;
 168 }
 169
 170 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
 171 {
 172         struct btrfs_root *root;
 173         struct btrfs_trans_handle *trans;
 174         int ret;
 175
 176         root = BTRFS_I(dir)->root;
 177         mutex_lock(&root->fs_info->fs_mutex);
 178         trans = btrfs_start_transaction(root, 1);
 179         ret = btrfs_unlink_trans(trans, root, dir, dentry);
 180         btrfs_end_transaction(trans, root);
 181         mutex_unlock(&root->fs_info->fs_mutex);
 182         return ret;
 183 }
 184
 185 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
 186 {
 187         struct inode *inode = dentry->d_inode;
 188         int err;
 189         int ret;
 190         struct btrfs_root *root = BTRFS_I(dir)->root;
 191         struct btrfs_path *path;
 192         struct btrfs_key key;
 193         struct btrfs_trans_handle *trans;
 194         struct btrfs_key found_key;
 195         int found_type;
 196         struct btrfs_leaf *leaf;
 197         char *goodnames = "..";
 198
 199         path = btrfs_alloc_path();
 200         BUG_ON(!path);
 201         btrfs_init_path(path);
 202         mutex_lock(&root->fs_info->fs_mutex);
 203         trans = btrfs_start_transaction(root, 1);
 204         key.objectid = inode->i_ino;
 205         key.offset = (u64)-1;
 206         key.flags = (u32)-1;
 207         while(1) {
 208                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 209                 if (ret < 0) {
 210                         err = ret;
 211                         goto out;
 212                 }
 213                 BUG_ON(ret == 0);
 214                 if (path->slots[0] == 0) {
 215                         err = -ENOENT;
 216                         goto out;
 217                 }
 218                 path->slots[0]--;
 219                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 220                 btrfs_disk_key_to_cpu(&found_key,
 221                                       &leaf->items[path->slots[0]].key);
 222                 found_type = btrfs_key_type(&found_key);
 223                 if (found_key.objectid != inode->i_ino) {
 224                         err = -ENOENT;
 225                         goto out;
 226                 }
 227                 if ((found_type != BTRFS_DIR_ITEM_KEY &&
 228                      found_type != BTRFS_DIR_INDEX_KEY) ||
 229                     (!btrfs_match_dir_item_name(root, path, goodnames, 2) &&
 230                     !btrfs_match_dir_item_name(root, path, goodnames, 1))) {
 231                         err = -ENOTEMPTY;
 232                         goto out;
 233                 }
 234                 ret = btrfs_del_item(trans, root, path);
 235                 BUG_ON(ret);
 236
 237                 if (found_type == BTRFS_DIR_ITEM_KEY && found_key.offset == 1)
 238                         break;
 239                 btrfs_release_path(root, path);
 240         }
 241         ret = 0;
 242         btrfs_release_path(root, path);
 243
 244         /* now the directory is empty */
 245         err = btrfs_unlink_trans(trans, root, dir, dentry);
 246         if (!err) {
 247                 inode->i_size = 0;
 248         }
 249 out:
 250         btrfs_release_path(root, path);
 251         btrfs_free_path(path);
 252         mutex_unlock(&root->fs_info->fs_mutex);
 253         ret = btrfs_end_transaction(trans, root);
 254         if (ret && !err)
 255                 err = ret;
 256         return err;
 257 }
 258
 259 static int btrfs_free_inode(struct btrfs_trans_handle *trans,
 260                             struct btrfs_root *root,
 261                             struct inode *inode)
 262 {
 263         struct btrfs_path *path;
 264         int ret;
 265
 266         clear_inode(inode);
 267
 268         path = btrfs_alloc_path();
 269         BUG_ON(!path);
 270         btrfs_init_path(path);
 271         ret = btrfs_lookup_inode(trans, root, path,
 272                                  &BTRFS_I(inode)->location, -1);
 273         BUG_ON(ret);
 274         ret = btrfs_del_item(trans, root, path);
 275         BUG_ON(ret);
 276         btrfs_free_path(path);
 277         return ret;
 278 }
 279
 280 static int btrfs_truncate_in_trans(struct btrfs_trans_handle *trans,
 281                                    struct btrfs_root *root,
 282                                    struct inode *inode)
 283 {
 284         int ret;
 285         struct btrfs_path *path;
 286         struct btrfs_key key;
 287         struct btrfs_disk_key *found_key;
 288         struct btrfs_leaf *leaf;
 289         struct btrfs_file_extent_item *fi = NULL;
 290         u64 extent_start = 0;
 291         u64 extent_num_blocks = 0;
 292         int found_extent;
 293
 294         path = btrfs_alloc_path();
 295         BUG_ON(!path);
 296         /* FIXME, add redo link to tree so we don't leak on crash */
 297         key.objectid = inode->i_ino;
 298         key.offset = (u64)-1;
 299         key.flags = 0;
 300         /*
 301          * use BTRFS_CSUM_ITEM_KEY because it is larger than inline keys
 302          * or extent data
 303          */
 304         btrfs_set_key_type(&key, BTRFS_CSUM_ITEM_KEY);
 305         while(1) {
 306                 btrfs_init_path(path);
 307                 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 308                 if (ret < 0) {
 309                         goto error;
 310                 }
 311                 if (ret > 0) {
 312                         BUG_ON(path->slots[0] == 0);
 313                         path->slots[0]--;
 314                 }
 315                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 316                 found_key = &leaf->items[path->slots[0]].key;
 317                 if (btrfs_disk_key_objectid(found_key) != inode->i_ino)
 318                         break;
 319                 if (btrfs_disk_key_type(found_key) != BTRFS_CSUM_ITEM_KEY &&
 320                     btrfs_disk_key_type(found_key) != BTRFS_INLINE_DATA_KEY &&
 321                     btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY)
 322                         break;
 323                 if (btrfs_disk_key_offset(found_key) < inode->i_size)
 324                         break;
 325                 if (btrfs_disk_key_type(found_key) == BTRFS_EXTENT_DATA_KEY) {
 326                         fi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 327                                             path->slots[0],
 328                                             struct btrfs_file_extent_item);
 329                         extent_start = btrfs_file_extent_disk_blocknr(fi);
 330                         extent_num_blocks =
 331                                 btrfs_file_extent_disk_num_blocks(fi);
 332                         inode->i_blocks -=
 333                                 btrfs_file_extent_num_blocks(fi) >> 9;
 334                         found_extent = 1;
 335                 } else {
 336                         found_extent = 0;
 337                 }
 338                 ret = btrfs_del_item(trans, root, path);
 339                 BUG_ON(ret);
 340                 btrfs_release_path(root, path);
 341                 if (found_extent) {
 342                         ret = btrfs_free_extent(trans, root, extent_start,
 343                                                 extent_num_blocks, 0);
 344                         BUG_ON(ret);
 345                 }
 346         }
 347         ret = 0;
 348 error:
 349         btrfs_release_path(root, path);
 350         btrfs_free_path(path);
 351         return ret;
 352 }
 353
 354 static void btrfs_delete_inode(struct inode *inode)
 355 {
 356         struct btrfs_trans_handle *trans;
 357         struct btrfs_root *root = BTRFS_I(inode)->root;
 358         int ret;
 359
 360         truncate_inode_pages(&inode->i_data, 0);
 361         if (is_bad_inode(inode)) {
 362                 goto no_delete;
 363         }
 364         inode->i_size = 0;
 365         mutex_lock(&root->fs_info->fs_mutex);
 366         trans = btrfs_start_transaction(root, 1);
 367         if (S_ISREG(inode->i_mode)) {
 368                 ret = btrfs_truncate_in_trans(trans, root, inode);
 369                 BUG_ON(ret);
 370         }
 371         btrfs_free_inode(trans, root, inode);
 372         btrfs_end_transaction(trans, root);
 373         mutex_unlock(&root->fs_info->fs_mutex);
 374         return;
 375 no_delete:
 376         clear_inode(inode);
 377 }
 378
 379 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
 380                                struct btrfs_key *location)
 381 {
 382         const char *name = dentry->d_name.name;
 383         int namelen = dentry->d_name.len;
 384         struct btrfs_dir_item *di;
 385         struct btrfs_path *path;
 386         struct btrfs_root *root = BTRFS_I(dir)->root;
 387         int ret;
 388
 389         path = btrfs_alloc_path();
 390         BUG_ON(!path);
 391         btrfs_init_path(path);
 392         ret = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
 393                                     namelen, 0);
 394         if (ret || !btrfs_match_dir_item_name(root, path, name, namelen)) {
 395                 location->objectid = 0;
 396                 ret = 0;
 397                 goto out;
 398         }
 399         di = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 400                             struct btrfs_dir_item);
 401         btrfs_disk_key_to_cpu(location, &di->location);
 402 out:
 403         btrfs_release_path(root, path);
 404         btrfs_free_path(path);
 405         return ret;
 406 }
 407
 408 int fixup_tree_root_location(struct btrfs_root *root,
 409                              struct btrfs_key *location,
 410                              struct btrfs_root **sub_root)
 411 {
 412         struct btrfs_path *path;
 413         struct btrfs_root_item *ri;
 414
 415         if (btrfs_key_type(location) != BTRFS_ROOT_ITEM_KEY)
 416                 return 0;
 417         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
 418                 return 0;
 419
 420         path = btrfs_alloc_path();
 421         BUG_ON(!path);
 422         mutex_lock(&root->fs_info->fs_mutex);
 423
 424         *sub_root = btrfs_read_fs_root(root->fs_info, location);
 425         if (IS_ERR(*sub_root))
 426                 return PTR_ERR(*sub_root);
 427
 428         ri = &(*sub_root)->root_item;
 429         location->objectid = btrfs_root_dirid(ri);
 430         location->flags = 0;
 431         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 432         location->offset = 0;
 433
 434         btrfs_free_path(path);
 435         mutex_unlock(&root->fs_info->fs_mutex);
 436         return 0;
 437 }
 438
 439 int btrfs_init_locked_inode(struct inode *inode, void *p)
 440 {
 441         struct btrfs_iget_args *args = p;
 442         inode->i_ino = args->ino;
 443         BTRFS_I(inode)->root = args->root;
 444         return 0;
 445 }
 446
 447 int btrfs_find_actor(struct inode *inode, void *opaque)
 448 {
 449         struct btrfs_iget_args *args = opaque;
 450         return (args->ino == inode->i_ino &&
 451                 args->root == BTRFS_I(inode)->root);
 452 }
 453
 454 struct inode *btrfs_iget_locked(struct super_block *s, u64 objectid,
 455                                 struct btrfs_root *root)
 456 {
 457         struct inode *inode;
 458         struct btrfs_iget_args args;
 459         args.ino = objectid;
 460         args.root = root;
 461
 462         inode = iget5_locked(s, objectid, btrfs_find_actor,
 463                              btrfs_init_locked_inode,
 464                              (void *)&args);
 465         return inode;
 466 }
 467
 468 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
 469                                    struct nameidata *nd)
 470 {
 471         struct inode * inode;
 472         struct btrfs_inode *bi = BTRFS_I(dir);
 473         struct btrfs_root *root = bi->root;
 474         struct btrfs_root *sub_root = root;
 475         struct btrfs_key location;
 476         int ret;
 477
 478         if (dentry->d_name.len > BTRFS_NAME_LEN)
 479                 return ERR_PTR(-ENAMETOOLONG);
 480         mutex_lock(&root->fs_info->fs_mutex);
 481         ret = btrfs_inode_by_name(dir, dentry, &location);
 482         mutex_unlock(&root->fs_info->fs_mutex);
 483         if (ret < 0)
 484                 return ERR_PTR(ret);
 485         inode = NULL;
 486         if (location.objectid) {
 487                 ret = fixup_tree_root_location(root, &location, &sub_root);
 488                 if (ret < 0)
 489                         return ERR_PTR(ret);
 490                 if (ret > 0)
 491                         return ERR_PTR(-ENOENT);
 492                 inode = btrfs_iget_locked(dir->i_sb, location.objectid,
 493                                           sub_root);
 494                 if (!inode)
 495                         return ERR_PTR(-EACCES);
 496                 if (inode->i_state & I_NEW) {
 497                         if (sub_root != root) {
 498 printk("adding new root for inode %lu root %p (found %p)\n", inode->i_ino, sub_root, BTRFS_I(inode)->root);
 499                                 igrab(inode);
 500                                 sub_root->inode = inode;
 501                         }
 502                         BTRFS_I(inode)->root = sub_root;
 503                         memcpy(&BTRFS_I(inode)->location, &location,
 504                                sizeof(location));
 505                         btrfs_read_locked_inode(inode);
 506                         unlock_new_inode(inode);
 507                 }
 508         }
 509         return d_splice_alias(inode, dentry);
 510 }
 511
 512 static int btrfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
 513 {
 514         struct inode *inode = filp->f_path.dentry->d_inode;
 515         struct btrfs_root *root = BTRFS_I(inode)->root;
 516         struct btrfs_item *item;
 517         struct btrfs_dir_item *di;
 518         struct btrfs_key key;
 519         struct btrfs_path *path;
 520         int ret;
 521         u32 nritems;
 522         struct btrfs_leaf *leaf;
 523         int slot;
 524         int advance;
 525         unsigned char d_type = DT_UNKNOWN;
 526         int over = 0;
 527         int key_type = BTRFS_DIR_INDEX_KEY;
 528
 529         /* FIXME, use a real flag for deciding about the key type */
 530         if (root->fs_info->tree_root == root)
 531                 key_type = BTRFS_DIR_ITEM_KEY;
 532         mutex_lock(&root->fs_info->fs_mutex);
 533         key.objectid = inode->i_ino;
 534         key.flags = 0;
 535         btrfs_set_key_type(&key, key_type);
 536         key.offset = filp->f_pos;
 537         path = btrfs_alloc_path();
 538         btrfs_init_path(path);
 539         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 540         if (ret < 0)
 541                 goto err;
 542         advance = 0;
 543         while(1) {
 544                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 545                 nritems = btrfs_header_nritems(&leaf->header);
 546                 slot = path->slots[0];
 547                 if (advance || slot >= nritems) {
 548                         if (slot >= nritems -1) {
 549                                 ret = btrfs_next_leaf(root, path);
 550                                 if (ret)
 551                                         break;
 552                                 leaf = btrfs_buffer_leaf(path->nodes[0]);
 553                                 nritems = btrfs_header_nritems(&leaf->header);
 554                                 slot = path->slots[0];
 555                         } else {
 556                                 slot++;
 557                                 path->slots[0]++;
 558                         }
 559                 }
 560                 advance = 1;
 561                 item = leaf->items + slot;
 562                 if (btrfs_disk_key_objectid(&item->key) != key.objectid)
 563                         break;
 564                 if (key_type == BTRFS_DIR_INDEX_KEY &&
 565                     btrfs_disk_key_offset(&item->key) > root->highest_inode)
 566                         break;
 567                 if (btrfs_disk_key_type(&item->key) != key_type)
 568                         continue;
 569                 if (btrfs_disk_key_offset(&item->key) < filp->f_pos)
 570                         continue;
 571                 filp->f_pos = btrfs_disk_key_offset(&item->key);
 572                 advance = 1;
 573                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
 574                 over = filldir(dirent, (const char *)(di + 1),
 575                                btrfs_dir_name_len(di),
 576                                btrfs_disk_key_offset(&item->key),
 577                                btrfs_disk_key_objectid(&di->location), d_type);
 578                 if (over)
 579                         goto nopos;
 580         }
 581         filp->f_pos++;
 582 nopos:
 583         ret = 0;
 584 err:
 585         btrfs_release_path(root, path);
 586         btrfs_free_path(path);
 587         mutex_unlock(&root->fs_info->fs_mutex);
 588         return ret;
 589 }
 590
 591 static void btrfs_put_super (struct super_block * sb)
 592 {
 593         struct btrfs_root *root = btrfs_sb(sb);
 594         int ret;
 595
 596         ret = close_ctree(root);
 597         if (ret) {
 598                 printk("close ctree returns %d\n", ret);
 599         }
 600         sb->s_fs_info = NULL;
 601 }
 602
 603 static int btrfs_fill_super(struct super_block * sb, void * data, int silent)
 604 {
 605         struct inode * inode;
 606         struct dentry * root_dentry;
 607         struct btrfs_super_block *disk_super;
 608         struct btrfs_root *tree_root;
 609         struct btrfs_inode *bi;
 610
 611         sb->s_maxbytes = MAX_LFS_FILESIZE;
 612         sb->s_magic = BTRFS_SUPER_MAGIC;
 613         sb->s_op = &btrfs_super_ops;
 614         sb->s_time_gran = 1;
 615
 616         tree_root = open_ctree(sb);
 617
 618         if (!tree_root) {
 619                 printk("btrfs: open_ctree failed\n");
 620                 return -EIO;
 621         }
 622         sb->s_fs_info = tree_root;
 623         disk_super = tree_root->fs_info->disk_super;
 624         printk("read in super total blocks %Lu root %Lu\n",
 625                btrfs_super_total_blocks(disk_super),
 626                btrfs_super_root_dir(disk_super));
 627
 628         inode = btrfs_iget_locked(sb, btrfs_super_root_dir(disk_super),
 629                                   tree_root);
 630         bi = BTRFS_I(inode);
 631         bi->location.objectid = inode->i_ino;
 632         bi->location.offset = 0;
 633         bi->location.flags = 0;
 634         bi->root = tree_root;
 635         btrfs_set_key_type(&bi->location, BTRFS_INODE_ITEM_KEY);
 636
 637         if (!inode)
 638                 return -ENOMEM;
 639         if (inode->i_state & I_NEW) {
 640                 btrfs_read_locked_inode(inode);
 641                 unlock_new_inode(inode);
 642         }
 643
 644         root_dentry = d_alloc_root(inode);
 645         if (!root_dentry) {
 646                 iput(inode);
 647                 return -ENOMEM;
 648         }
 649         sb->s_root = root_dentry;
 650
 651         return 0;
 652 }
 653
 654 static void fill_inode_item(struct btrfs_inode_item *item,
 655                             struct inode *inode)
 656 {
 657         btrfs_set_inode_uid(item, inode->i_uid);
 658         btrfs_set_inode_gid(item, inode->i_gid);
 659         btrfs_set_inode_size(item, inode->i_size);
 660         btrfs_set_inode_mode(item, inode->i_mode);
 661         btrfs_set_inode_nlink(item, inode->i_nlink);
 662         btrfs_set_timespec_sec(&item->atime, inode->i_atime.tv_sec);
 663         btrfs_set_timespec_nsec(&item->atime, inode->i_atime.tv_nsec);
 664         btrfs_set_timespec_sec(&item->mtime, inode->i_mtime.tv_sec);
 665         btrfs_set_timespec_nsec(&item->mtime, inode->i_mtime.tv_nsec);
 666         btrfs_set_timespec_sec(&item->ctime, inode->i_ctime.tv_sec);
 667         btrfs_set_timespec_nsec(&item->ctime, inode->i_ctime.tv_nsec);
 668         btrfs_set_inode_nblocks(item, inode->i_blocks);
 669         btrfs_set_inode_generation(item, inode->i_generation);
 670 }
 671
 672 static int btrfs_update_inode(struct btrfs_trans_handle *trans,
 673                               struct btrfs_root *root,
 674                               struct inode *inode)
 675 {
 676         struct btrfs_inode_item *inode_item;
 677         struct btrfs_path *path;
 678         int ret;
 679
 680         path = btrfs_alloc_path();
 681         BUG_ON(!path);
 682         btrfs_init_path(path);
 683         ret = btrfs_lookup_inode(trans, root, path,
 684                                  &BTRFS_I(inode)->location, 1);
 685         if (ret) {
 686                 if (ret > 0)
 687                         ret = -ENOENT;
 688                 goto failed;
 689         }
 690
 691         inode_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
 692                                   path->slots[0],
 693                                   struct btrfs_inode_item);
 694
 695         fill_inode_item(inode_item, inode);
 696         btrfs_mark_buffer_dirty(path->nodes[0]);
 697         ret = 0;
 698 failed:
 699         btrfs_release_path(root, path);
 700         btrfs_free_path(path);
 701         return ret;
 702 }
 703
 704 static int btrfs_write_inode(struct inode *inode, int wait)
 705 {
 706         struct btrfs_root *root = BTRFS_I(inode)->root;
 707         struct btrfs_trans_handle *trans;
 708         int ret;
 709
 710         mutex_lock(&root->fs_info->fs_mutex);
 711         trans = btrfs_start_transaction(root, 1);
 712         ret = btrfs_update_inode(trans, root, inode);
 713         if (wait)
 714                 btrfs_commit_transaction(trans, root);
 715         else
 716                 btrfs_end_transaction(trans, root);
 717         mutex_unlock(&root->fs_info->fs_mutex);
 718         return ret;
 719 }
 720
 721 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
 722                                      struct btrfs_root *root,
 723                                      u64 objectid, int mode)
 724 {
 725         struct inode *inode;
 726         struct btrfs_inode_item inode_item;
 727         struct btrfs_key *location;
 728         int ret;
 729
 730         inode = new_inode(root->fs_info->sb);
 731         if (!inode)
 732                 return ERR_PTR(-ENOMEM);
 733
 734         BTRFS_I(inode)->root = root;
 735
 736         inode->i_uid = current->fsuid;
 737         inode->i_gid = current->fsgid;
 738         inode->i_mode = mode;
 739         inode->i_ino = objectid;
 740         inode->i_blocks = 0;
 741         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
 742         fill_inode_item(&inode_item, inode);
 743         location = &BTRFS_I(inode)->location;
 744         location->objectid = objectid;
 745         location->flags = 0;
 746         location->offset = 0;
 747         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
 748
 749         ret = btrfs_insert_inode(trans, root, objectid, &inode_item);
 750         BUG_ON(ret);
 751
 752         insert_inode_hash(inode);
 753         return inode;
 754 }
 755
 756 static int btrfs_add_link(struct btrfs_trans_handle *trans,
 757                             struct dentry *dentry, struct inode *inode)
 758 {
 759         int ret;
 760         struct btrfs_key key;
 761         struct btrfs_root *root = BTRFS_I(dentry->d_parent->d_inode)->root;
 762         key.objectid = inode->i_ino;
 763         key.flags = 0;
 764         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 765         key.offset = 0;
 766
 767         ret = btrfs_insert_dir_item(trans, root,
 768                                     dentry->d_name.name, dentry->d_name.len,
 769                                     dentry->d_parent->d_inode->i_ino,
 770                                     &key, 0);
 771         if (ret == 0) {
 772                 dentry->d_parent->d_inode->i_size += dentry->d_name.len * 2;
 773                 ret = btrfs_update_inode(trans, root,
 774                                          dentry->d_parent->d_inode);
 775         }
 776         return ret;
 777 }
 778
 779 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
 780                             struct dentry *dentry, struct inode *inode)
 781 {
 782         int err = btrfs_add_link(trans, dentry, inode);
 783         if (!err) {
 784                 d_instantiate(dentry, inode);
 785                 return 0;
 786         }
 787         if (err > 0)
 788                 err = -EEXIST;
 789         return err;
 790 }
 791
 792 static int btrfs_create(struct inode *dir, struct dentry *dentry,
 793                         int mode, struct nameidata *nd)
 794 {
 795         struct btrfs_trans_handle *trans;
 796         struct btrfs_root *root = BTRFS_I(dir)->root;
 797         struct inode *inode;
 798         int err;
 799         int drop_inode = 0;
 800         u64 objectid;
 801
 802         mutex_lock(&root->fs_info->fs_mutex);
 803         trans = btrfs_start_transaction(root, 1);
 804
 805         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 806         if (err) {
 807                 err = -ENOSPC;
 808                 goto out_unlock;
 809         }
 810
 811         inode = btrfs_new_inode(trans, root, objectid, mode);
 812         err = PTR_ERR(inode);
 813         if (IS_ERR(inode))
 814                 goto out_unlock;
 815         // FIXME mark the inode dirty
 816         err = btrfs_add_nondir(trans, dentry, inode);
 817         if (err)
 818                 drop_inode = 1;
 819         else {
 820                 inode->i_mapping->a_ops = &btrfs_aops;
 821                 inode->i_fop = &btrfs_file_operations;
 822                 inode->i_op = &btrfs_file_inode_operations;
 823         }
 824         dir->i_sb->s_dirt = 1;
 825 out_unlock:
 826         btrfs_end_transaction(trans, root);
 827         mutex_unlock(&root->fs_info->fs_mutex);
 828
 829         if (drop_inode) {
 830                 inode_dec_link_count(inode);
 831                 iput(inode);
 832         }
 833         return err;
 834 }
 835
 836 static int btrfs_make_empty_dir(struct btrfs_trans_handle *trans,
 837                                 struct btrfs_root *root,
 838                                 u64 objectid, u64 dirid)
 839 {
 840         int ret;
 841         char buf[2];
 842         struct btrfs_key key;
 843
 844         buf[0] = '.';
 845         buf[1] = '.';
 846
 847         key.objectid = objectid;
 848         key.offset = 0;
 849         key.flags = 0;
 850         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 851
 852         ret = btrfs_insert_dir_item(trans, root, buf, 1, objectid,
 853                                     &key, 1);
 854         if (ret)
 855                 goto error;
 856         key.objectid = dirid;
 857         ret = btrfs_insert_dir_item(trans, root, buf, 2, objectid,
 858                                     &key, 1);
 859         if (ret)
 860                 goto error;
 861 error:
 862         return ret;
 863 }
 864
 865 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 866 {
 867         struct inode *inode;
 868         struct btrfs_trans_handle *trans;
 869         struct btrfs_root *root = BTRFS_I(dir)->root;
 870         int err = 0;
 871         int drop_on_err = 0;
 872         u64 objectid;
 873
 874         mutex_lock(&root->fs_info->fs_mutex);
 875         trans = btrfs_start_transaction(root, 1);
 876         if (IS_ERR(trans)) {
 877                 err = PTR_ERR(trans);
 878                 goto out_unlock;
 879         }
 880
 881         err = btrfs_find_free_objectid(trans, root, dir->i_ino, &objectid);
 882         if (err) {
 883                 err = -ENOSPC;
 884                 goto out_unlock;
 885         }
 886
 887         inode = btrfs_new_inode(trans, root, objectid, S_IFDIR | mode);
 888         if (IS_ERR(inode)) {
 889                 err = PTR_ERR(inode);
 890                 goto out_fail;
 891         }
 892         drop_on_err = 1;
 893         inode->i_op = &btrfs_dir_inode_operations;
 894         inode->i_fop = &btrfs_dir_file_operations;
 895
 896         err = btrfs_make_empty_dir(trans, root, inode->i_ino, dir->i_ino);
 897         if (err)
 898                 goto out_fail;
 899
 900         inode->i_size = 6;
 901         err = btrfs_update_inode(trans, root, inode);
 902         if (err)
 903                 goto out_fail;
 904         err = btrfs_add_link(trans, dentry, inode);
 905         if (err)
 906                 goto out_fail;
 907         d_instantiate(dentry, inode);
 908         drop_on_err = 0;
 909
 910 out_fail:
 911         btrfs_end_transaction(trans, root);
 912 out_unlock:
 913         mutex_unlock(&root->fs_info->fs_mutex);
 914         if (drop_on_err)
 915                 iput(inode);
 916         return err;
 917 }
 918
 919 static int btrfs_sync_fs(struct super_block *sb, int wait)
 920 {
 921         struct btrfs_trans_handle *trans;
 922         struct btrfs_root *root;
 923         int ret;
 924         root = btrfs_sb(sb);
 925
 926         sb->s_dirt = 0;
 927         if (!wait) {
 928                 filemap_flush(root->fs_info->btree_inode->i_mapping);
 929                 return 0;
 930         }
 931         filemap_write_and_wait(root->fs_info->btree_inode->i_mapping);
 932         mutex_lock(&root->fs_info->fs_mutex);
 933         trans = btrfs_start_transaction(root, 1);
 934         ret = btrfs_commit_transaction(trans, root);
 935         sb->s_dirt = 0;
 936         BUG_ON(ret);
 937 printk("btrfs sync_fs\n");
 938         mutex_unlock(&root->fs_info->fs_mutex);
 939         return 0;
 940 }
 941
 942 #if 0
 943 static int btrfs_get_block_inline(struct inode *inode, sector_t iblock,
 944                            struct buffer_head *result, int create)
 945 {
 946         struct btrfs_root *root = btrfs_sb(inode->i_sb);
 947         struct btrfs_path *path;
 948         struct btrfs_key key;
 949         struct btrfs_leaf *leaf;
 950         int num_bytes = result->b_size;
 951         int item_size;
 952         int ret;
 953         u64 pos;
 954         char *ptr;
 955         int copy_size;
 956         int err = 0;
 957         char *safe_ptr;
 958         char *data_ptr;
 959
 960         path = btrfs_alloc_path();
 961         BUG_ON(!path);
 962
 963         WARN_ON(create);
 964         if (create) {
 965                 return 0;
 966         }
 967         pos = iblock << inode->i_blkbits;
 968         key.objectid = inode->i_ino;
 969         key.flags = 0;
 970         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
 971         ptr = kmap(result->b_page);
 972         safe_ptr = ptr;
 973         ptr += (pos & (PAGE_CACHE_SIZE -1));
 974 again:
 975         key.offset = pos;
 976         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 977         if (ret) {
 978                 if (ret < 0)
 979                         err = ret;
 980                 else
 981                         err = 0;
 982                 goto out;
 983         }
 984         leaf = btrfs_buffer_leaf(path->nodes[0]);
 985         item_size = btrfs_item_size(leaf->items + path->slots[0]);
 986         copy_size = min(num_bytes, item_size);
 987         data_ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 988         WARN_ON(safe_ptr + PAGE_CACHE_SIZE < ptr + copy_size);
 989         memcpy(ptr, data_ptr, copy_size);
 990         pos += copy_size;
 991         num_bytes -= copy_size;
 992         WARN_ON(num_bytes < 0);
 993         ptr += copy_size;
 994         btrfs_release_path(root, path);
 995         if (num_bytes != 0) {
 996                 if (pos >= i_size_read(inode))
 997                         memset(ptr, 0, num_bytes);
 998                 else
 999                         goto again;
1000         }
1001         set_buffer_uptodate(result);
1002         map_bh(result, inode->i_sb, 0);
1003         err = 0;
1004 out:
1005         btrfs_free_path(path);
1006         kunmap(result->b_page);
1007         return err;
1008 }
1009 #endif
1010
1011 static int btrfs_get_block_lock(struct inode *inode, sector_t iblock,
1012                            struct buffer_head *result, int create)
1013 {
1014         int ret;
1015         int err = 0;
1016         u64 blocknr;
1017         u64 extent_start = 0;
1018         u64 extent_end = 0;
1019         u64 objectid = inode->i_ino;
1020         struct btrfs_path *path;
1021         struct btrfs_root *root = BTRFS_I(inode)->root;
1022         struct btrfs_trans_handle *trans = NULL;
1023         struct btrfs_file_extent_item *item;
1024         struct btrfs_leaf *leaf;
1025         struct btrfs_disk_key *found_key;
1026
1027         path = btrfs_alloc_path();
1028         BUG_ON(!path);
1029         btrfs_init_path(path);
1030         if (create) {
1031                 trans = btrfs_start_transaction(root, 1);
1032                 WARN_ON(1);
1033         }
1034
1035         ret = btrfs_lookup_file_extent(trans, root, path,
1036                                        inode->i_ino,
1037                                        iblock << inode->i_blkbits, create);
1038         if (ret < 0) {
1039                 err = ret;
1040                 goto out;
1041         }
1042
1043         if (ret != 0) {
1044                 if (path->slots[0] == 0) {
1045                         btrfs_release_path(root, path);
1046                         goto allocate;
1047                 }
1048                 path->slots[0]--;
1049         }
1050
1051         item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
1052                               struct btrfs_file_extent_item);
1053         leaf = btrfs_buffer_leaf(path->nodes[0]);
1054         blocknr = btrfs_file_extent_disk_blocknr(item);
1055         blocknr += btrfs_file_extent_offset(item);
1056
1057         /* exact match found, use it, FIXME, deal with extents
1058          * other than the page size
1059          */
1060         if (0 && ret == 0) {
1061                 err = 0;
1062                 if (create &&
1063                     btrfs_file_extent_generation(item) != trans->transid) {
1064                         struct btrfs_key ins;
1065                         ret = btrfs_alloc_extent(trans, root, 1,
1066                                                  blocknr, (u64)-1, &ins);
1067                         BUG_ON(ret);
1068                         btrfs_set_file_extent_disk_blocknr(item, ins.objectid);
1069                         mark_buffer_dirty(path->nodes[0]);
1070                         ret = btrfs_free_extent(trans, root,
1071                                                 blocknr, 1, 0);
1072                         BUG_ON(ret);
1073                         blocknr = ins.objectid;
1074
1075                 }
1076                 btrfs_map_bh_to_logical(root, result, blocknr);
1077                 goto out;
1078         }
1079
1080         /* are we inside the extent that was found? */
1081         found_key = &leaf->items[path->slots[0]].key;
1082         if (btrfs_disk_key_objectid(found_key) != objectid ||
1083             btrfs_disk_key_type(found_key) != BTRFS_EXTENT_DATA_KEY) {
1084                 extent_end = 0;
1085                 extent_start = 0;
1086                 btrfs_release_path(root, path);
1087                 goto allocate;
1088         }
1089
1090         extent_start = btrfs_disk_key_offset(&leaf->items[path->slots[0]].key);
1091         extent_start = extent_start >> inode->i_blkbits;
1092         extent_end = extent_start + btrfs_file_extent_num_blocks(item);
1093         if (iblock >= extent_start && iblock < extent_end) {
1094                 err = 0;
1095                 btrfs_map_bh_to_logical(root, result, blocknr + iblock -
1096                                         extent_start);
1097                 goto out;
1098         }
1099 allocate:
1100         /* ok, create a new extent */
1101         if (!create) {
1102                 err = 0;
1103                 goto out;
1104         }
1105 #if 0
1106         ret = btrfs_alloc_file_extent(trans, root, objectid,
1107                                       iblock << inode->i_blkbits,
1108                                       1, extent_end, &blocknr);
1109         if (ret) {
1110                 err = ret;
1111                 goto out;
1112         }
1113         inode->i_blocks += inode->i_sb->s_blocksize >> 9;
1114         set_buffer_new(result);
1115         map_bh(result, inode->i_sb, blocknr);
1116
1117         btrfs_map_bh_to_logical(root, result, blocknr);
1118 #endif
1119 out:
1120         btrfs_release_path(root, path);
1121         btrfs_free_path(path);
1122 printk("mapping iblock %lu to %lu\n", iblock, result->b_blocknr);
1123         if (trans)
1124                 btrfs_end_transaction(trans, root);
1125         return err;
1126 }
1127
1128 static int btrfs_get_block(struct inode *inode, sector_t iblock,
1129                            struct buffer_head *result, int create)
1130 {
1131         int err;
1132         struct btrfs_root *root = BTRFS_I(inode)->root;
1133         mutex_lock(&root->fs_info->fs_mutex);
1134         err = btrfs_get_block_lock(inode, iblock, result, create);
1135         // err = btrfs_get_block_inline(inode, iblock, result, create);
1136         mutex_unlock(&root->fs_info->fs_mutex);
1137         return err;
1138 }
1139
1140 static int btrfs_prepare_write(struct file *file, struct page *page,
1141                                unsigned from, unsigned to)
1142 {
1143         return nobh_prepare_write(page, from, to, btrfs_get_block);
1144 }
1145 static int btrfs_commit_write(struct file *file, struct page *page,
1146                                unsigned from, unsigned to)
1147 {
1148         return nobh_commit_write(file, page, from, to);
1149 }
1150
1151 static void btrfs_write_super(struct super_block *sb)
1152 {
1153         btrfs_sync_fs(sb, 1);
1154 }
1155
1156 static int btrfs_readpage(struct file *file, struct page *page)
1157 {
1158         return mpage_readpage(page, btrfs_get_block);
1159 }
1160
1161 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
1162 {
1163         return nobh_writepage(page, btrfs_get_block, wbc);
1164 }
1165
1166 static void btrfs_truncate(struct inode *inode)
1167 {
1168         struct btrfs_root *root = BTRFS_I(inode)->root;
1169         int ret;
1170         struct btrfs_trans_handle *trans;
1171
1172         if (!S_ISREG(inode->i_mode))
1173                 return;
1174         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
1175                 return;
1176
1177         nobh_truncate_page(inode->i_mapping, inode->i_size);
1178
1179         /* FIXME, add redo link to tree so we don't leak on crash */
1180         mutex_lock(&root->fs_info->fs_mutex);
1181         trans = btrfs_start_transaction(root, 1);
1182         ret = btrfs_truncate_in_trans(trans, root, inode);
1183         BUG_ON(ret);
1184         ret = btrfs_end_transaction(trans, root);
1185         BUG_ON(ret);
1186         mutex_unlock(&root->fs_info->fs_mutex);
1187         mark_inode_dirty(inode);
1188 }
1189
1190 static int btrfs_copy_from_user(loff_t pos, int num_pages, int write_bytes,
1191                                 struct page **prepared_pages,
1192                                 const char __user * buf)
1193 {
1194         long page_fault = 0;
1195         int i;
1196         int offset = pos & (PAGE_CACHE_SIZE - 1);
1197
1198         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
1199                 size_t count = min_t(size_t,
1200                                      PAGE_CACHE_SIZE - offset, write_bytes);
1201                 struct page *page = prepared_pages[i];
1202                 fault_in_pages_readable(buf, count);
1203
1204                 /* Copy data from userspace to the current page */
1205                 kmap(page);
1206                 page_fault = __copy_from_user(page_address(page) + offset,
1207                                               buf, count);
1208                 /* Flush processor's dcache for this page */
1209                 flush_dcache_page(page);
1210                 kunmap(page);
1211                 buf += count;
1212                 write_bytes -= count;
1213
1214                 if (page_fault)
1215                         break;
1216         }
1217         return page_fault ? -EFAULT : 0;
1218 }
1219
1220 static void btrfs_drop_pages(struct page **pages, size_t num_pages)
1221 {
1222         size_t i;
1223         for (i = 0; i < num_pages; i++) {
1224                 if (!pages[i])
1225                         break;
1226                 unlock_page(pages[i]);
1227                 mark_page_accessed(pages[i]);
1228                 page_cache_release(pages[i]);
1229         }
1230 }
1231 static int dirty_and_release_pages(struct btrfs_trans_handle *trans,
1232                                    struct btrfs_root *root,
1233                                    struct file *file,
1234                                    struct page **pages,
1235                                    size_t num_pages,
1236                                    u64 extent_offset,
1237                                    loff_t pos,
1238                                    size_t write_bytes)
1239 {
1240         int i;
1241         int offset;
1242         int err = 0;
1243         int ret;
1244         int this_write;
1245         struct inode *inode = file->f_path.dentry->d_inode;
1246
1247         for (i = 0; i < num_pages; i++) {
1248                 offset = pos & (PAGE_CACHE_SIZE -1);
1249                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1250                 /* FIXME, one block at a time */
1251
1252                 mutex_lock(&root->fs_info->fs_mutex);
1253                 trans = btrfs_start_transaction(root, 1);
1254                 btrfs_csum_file_block(trans, root, inode->i_ino,
1255                                       pages[i]->index << PAGE_CACHE_SHIFT,
1256                                       extent_offset,
1257                                       kmap(pages[i]), PAGE_CACHE_SIZE);
1258                 kunmap(pages[i]);
1259                 SetPageChecked(pages[i]);
1260                 ret = btrfs_end_transaction(trans, root);
1261                 BUG_ON(ret);
1262                 mutex_unlock(&root->fs_info->fs_mutex);
1263
1264                 ret = nobh_commit_write(file, pages[i], offset,
1265                                          offset + this_write);
1266                 pos += this_write;
1267                 if (ret) {
1268                         err = ret;
1269                         goto failed;
1270                 }
1271                 WARN_ON(this_write > write_bytes);
1272                 write_bytes -= this_write;
1273         }
1274 failed:
1275         return err;
1276 }
1277
1278 static int drop_csums(struct btrfs_trans_handle *trans,
1279                           struct btrfs_root *root,
1280                           struct inode *inode,
1281                           u64 start, u64 end)
1282 {
1283         struct btrfs_path *path;
1284         struct btrfs_leaf *leaf;
1285         struct btrfs_key key;
1286         int slot;
1287         struct btrfs_csum_item *item;
1288         char *old_block = NULL;
1289         u64 cur = start;
1290         u64 found_end;
1291         u64 num_csums;
1292         u64 item_size;
1293         int ret;
1294
1295         path = btrfs_alloc_path();
1296         if (!path)
1297                 return -ENOMEM;
1298         while(cur < end) {
1299                 item = btrfs_lookup_csum(trans, root, path,
1300                                          inode->i_ino, cur, 1);
1301                 if (IS_ERR(item)) {
1302                         cur += root->blocksize;
1303                         continue;
1304                 }
1305                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1306                 slot = path->slots[0];
1307                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
1308                 item_size = btrfs_item_size(leaf->items + slot);
1309                 num_csums = item_size / sizeof(struct btrfs_csum_item);
1310                 found_end = key.offset + (num_csums << inode->i_blkbits);
1311                 cur = found_end;
1312
1313                 if (found_end > end) {
1314                         char *src;
1315                         old_block = kmalloc(root->blocksize, GFP_NOFS);
1316                         src = btrfs_item_ptr(leaf, slot, char);
1317                         memcpy(old_block, src, item_size);
1318                 }
1319                 if (key.offset < start) {
1320                         u64 new_size = (start - key.offset) >>
1321                                         inode->i_blkbits;
1322                         new_size *= sizeof(struct btrfs_csum_item);
1323                         ret = btrfs_truncate_item(trans, root, path, new_size);
1324                         BUG_ON(ret);
1325                 } else {
1326                         btrfs_del_item(trans, root, path);
1327                 }
1328                 btrfs_release_path(root, path);
1329                 if (found_end > end) {
1330                         char *dst;
1331                         int i;
1332                         int new_size;
1333
1334                         num_csums = (found_end - end) >> inode->i_blkbits;
1335                         new_size = num_csums * sizeof(struct btrfs_csum_item);
1336                         key.offset = end;
1337                         ret = btrfs_insert_empty_item(trans, root, path,
1338                                                       &key, new_size);
1339                         BUG_ON(ret);
1340                         dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1341                                              path->slots[0], char);
1342                         memcpy(dst, old_block + item_size - new_size,
1343                                new_size);
1344                         item = (struct btrfs_csum_item *)dst;
1345                         for (i = 0; i < num_csums; i++) {
1346                                 btrfs_set_csum_extent_offset(item, end);
1347                                 item++;
1348                         }
1349                         mark_buffer_dirty(path->nodes[0]);
1350                         kfree(old_block);
1351                         break;
1352                 }
1353         }
1354         btrfs_free_path(path);
1355         return 0;
1356 }
1357
1358 static int drop_extents(struct btrfs_trans_handle *trans,
1359                           struct btrfs_root *root,
1360                           struct inode *inode,
1361                           u64 start, u64 end)
1362 {
1363         int ret;
1364         struct btrfs_key key;
1365         struct btrfs_leaf *leaf;
1366         int slot;
1367         struct btrfs_file_extent_item *extent;
1368         u64 extent_end;
1369         int keep;
1370         struct btrfs_file_extent_item old;
1371         struct btrfs_path *path;
1372         u64 search_start = start;
1373         int bookend;
1374
1375         path = btrfs_alloc_path();
1376         if (!path)
1377                 return -ENOMEM;
1378 search_again:
1379 printk("drop extent inode %lu start %Lu end %Lu\n", inode->i_ino, start, end);
1380         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1381                                        search_start, -1);
1382         if (ret != 0) {
1383 printk("lookup failed\n");
1384                 goto out;
1385         }
1386         while(1) {
1387                 keep = 0;
1388                 bookend = 0;
1389                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1390                 slot = path->slots[0];
1391                 btrfs_disk_key_to_cpu(&key, &leaf->items[slot].key);
1392
1393 printk("found key %Lu %Lu %u\n", key.objectid, key.offset, key.flags);
1394
1395                 extent = btrfs_item_ptr(leaf, slot,
1396                                         struct btrfs_file_extent_item);
1397                 extent_end = key.offset +
1398                         (btrfs_file_extent_num_blocks(extent) <<
1399                          inode->i_blkbits);
1400 printk("extent end is %Lu\n", extent_end);
1401                 if (key.offset >= end || key.objectid != inode->i_ino) {
1402                         ret = 0;
1403                         goto out;
1404                 }
1405                 if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
1406                         goto next_leaf;
1407
1408                 if (end < extent_end && end >= key.offset) {
1409                         memcpy(&old, extent, sizeof(old));
1410                         ret = btrfs_inc_extent_ref(trans, root,
1411                                    btrfs_file_extent_disk_blocknr(&old),
1412                                    btrfs_file_extent_disk_num_blocks(&old));
1413                         BUG_ON(ret);
1414                         bookend = 1;
1415                 }
1416
1417                 if (start > key.offset) {
1418                         u64 new_num;
1419                         /* truncate existing extent */
1420                         keep = 1;
1421                         WARN_ON(start & (root->blocksize - 1));
1422                         new_num = (start - key.offset) >> inode->i_blkbits;
1423 printk("truncating existing extent, was %Lu ", btrfs_file_extent_num_blocks(extent));
1424                         btrfs_set_file_extent_num_blocks(extent, new_num);
1425 printk("now %Lu\n", btrfs_file_extent_num_blocks(extent));
1426
1427                         mark_buffer_dirty(path->nodes[0]);
1428                 }
1429                 if (!keep) {
1430                         u64 disk_blocknr;
1431                         u64 disk_num_blocks;
1432 printk("del old\n");
1433                         disk_blocknr = btrfs_file_extent_disk_blocknr(extent);
1434                         disk_num_blocks =
1435                                 btrfs_file_extent_disk_num_blocks(extent);
1436                         search_start = key.offset +
1437                                 (btrfs_file_extent_num_blocks(extent) <<
1438                                 inode->i_blkbits);
1439                         ret = btrfs_del_item(trans, root, path);
1440                         BUG_ON(ret);
1441                         btrfs_release_path(root, path);
1442
1443                         ret = btrfs_free_extent(trans, root, disk_blocknr,
1444                                                 disk_num_blocks, 0);
1445
1446                         BUG_ON(ret);
1447                         if (!bookend && search_start >= end) {
1448                                 ret = 0;
1449                                 goto out;
1450                         }
1451                         if (!bookend)
1452                                 goto search_again;
1453                 }
1454                 if (bookend) {
1455                         /* create bookend */
1456                         struct btrfs_key ins;
1457 printk("bookend! extent end %Lu\n", extent_end);
1458                         ins.objectid = inode->i_ino;
1459                         ins.offset = end;
1460                         ins.flags = 0;
1461                         btrfs_set_key_type(&ins, BTRFS_EXTENT_DATA_KEY);
1462
1463                         btrfs_release_path(root, path);
1464                         ret = drop_csums(trans, root, inode, start, end);
1465                         BUG_ON(ret);
1466                         ret = btrfs_insert_empty_item(trans, root, path, &ins,
1467                                                       sizeof(*extent));
1468                         BUG_ON(ret);
1469                         extent = btrfs_item_ptr(
1470                                     btrfs_buffer_leaf(path->nodes[0]),
1471                                     path->slots[0],
1472                                     struct btrfs_file_extent_item);
1473                         btrfs_set_file_extent_disk_blocknr(extent,
1474                                     btrfs_file_extent_disk_blocknr(&old));
1475                         btrfs_set_file_extent_disk_num_blocks(extent,
1476                                     btrfs_file_extent_disk_num_blocks(&old));
1477
1478                         btrfs_set_file_extent_offset(extent,
1479                                     btrfs_file_extent_offset(&old) +
1480                                     ((end - key.offset) >> inode->i_blkbits));
1481                         WARN_ON(btrfs_file_extent_num_blocks(&old) <
1482                                 (end - key.offset) >> inode->i_blkbits);
1483                         btrfs_set_file_extent_num_blocks(extent,
1484                                     btrfs_file_extent_num_blocks(&old) -
1485                                     ((end - key.offset) >> inode->i_blkbits));
1486
1487                         btrfs_set_file_extent_generation(extent,
1488                                     btrfs_file_extent_generation(&old));
1489 printk("new bookend at offset %Lu, file_extent_offset %Lu, file_extent_num_blocks %Lu\n", end, btrfs_file_extent_offset(extent), btrfs_file_extent_num_blocks(extent));
1490                         btrfs_mark_buffer_dirty(path->nodes[0]);
1491                         ret = 0;
1492                         goto out_nocsum;
1493                 }
1494 next_leaf:
1495                 if (slot >= btrfs_header_nritems(&leaf->header) - 1) {
1496                         ret = btrfs_next_leaf(root, path);
1497                         if (ret) {
1498                                 ret = 0;
1499                                 goto out;
1500                         }
1501                 } else {
1502                         path->slots[0]++;
1503                 }
1504         }
1505
1506 out:
1507         ret = drop_csums(trans, root, inode, start, end);
1508         BUG_ON(ret);
1509
1510 out_nocsum:
1511         btrfs_free_path(path);
1512         return ret;
1513 }
1514
1515 static int prepare_pages(struct btrfs_root *root,
1516                          struct file *file,
1517                          struct page **pages,
1518                          size_t num_pages,
1519                          loff_t pos,
1520                          unsigned long first_index,
1521                          unsigned long last_index,
1522                          size_t write_bytes,
1523                          u64 alloc_extent_start)
1524 {
1525         int i;
1526         unsigned long index = pos >> PAGE_CACHE_SHIFT;
1527         struct inode *inode = file->f_path.dentry->d_inode;
1528         int offset;
1529         int err = 0;
1530         int this_write;
1531         struct buffer_head *bh;
1532         struct buffer_head *head;
1533         loff_t isize = i_size_read(inode);
1534
1535         memset(pages, 0, num_pages * sizeof(struct page *));
1536
1537         for (i = 0; i < num_pages; i++) {
1538                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
1539                 if (!pages[i]) {
1540                         err = -ENOMEM;
1541                         goto failed_release;
1542                 }
1543                 offset = pos & (PAGE_CACHE_SIZE -1);
1544                 this_write = min(PAGE_CACHE_SIZE - offset, write_bytes);
1545 #if 0
1546                 if ((pages[i]->index == first_index ||
1547                     pages[i]->index == last_index) && pos < isize &&
1548                     !PageUptodate(pages[i])) {
1549                         ret = mpage_readpage(pages[i], btrfs_get_block);
1550                         BUG_ON(ret);
1551                         lock_page(pages[i]);
1552                 }
1553 #endif
1554                 create_empty_buffers(pages[i], root->fs_info->sb->s_blocksize,
1555                                      (1 << BH_Uptodate));
1556                 head = page_buffers(pages[i]);
1557                 bh = head;
1558                 do {
1559 printk("mapping page %lu to block %Lu\n", pages[i]->index, alloc_extent_start);
1560                         err = btrfs_map_bh_to_logical(root, bh,
1561                                                       alloc_extent_start);
1562                         BUG_ON(err);
1563                         if (err)
1564                                 goto failed_truncate;
1565                         bh = bh->b_this_page;
1566                         alloc_extent_start++;
1567                 } while (bh != head);
1568                 pos += this_write;
1569                 WARN_ON(this_write > write_bytes);
1570                 write_bytes -= this_write;
1571         }
1572         return 0;
1573
1574 failed_release:
1575         btrfs_drop_pages(pages, num_pages);
1576         return err;
1577
1578 failed_truncate:
1579         btrfs_drop_pages(pages, num_pages);
1580         if (pos > isize)
1581                 vmtruncate(inode, isize);
1582         return err;
1583 }
1584
1585 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
1586                                 size_t count, loff_t *ppos)
1587 {
1588         loff_t pos;
1589         size_t num_written = 0;
1590         int err = 0;
1591         int ret = 0;
1592         struct inode *inode = file->f_path.dentry->d_inode;
1593         struct btrfs_root *root = BTRFS_I(inode)->root;
1594         struct page *pages[8];
1595         unsigned long first_index;
1596         unsigned long last_index;
1597         u64 start_pos;
1598         u64 num_blocks;
1599         u64 alloc_extent_start;
1600         u64 orig_extent_start;
1601         struct btrfs_trans_handle *trans;
1602         struct btrfs_key ins;
1603
1604         if (file->f_flags & O_DIRECT)
1605                 return -EINVAL;
1606         pos = *ppos;
1607
1608         start_pos = pos & ~(root->blocksize - 1);
1609         /* FIXME */
1610         if (start_pos != pos)
1611                 return -EINVAL;
1612         num_blocks = (count + pos - start_pos + root->blocksize - 1) >>
1613                         inode->i_blkbits;
1614
1615         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1616         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1617         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1618         if (err)
1619                 goto out;
1620         if (count == 0)
1621                 goto out;
1622         err = remove_suid(file->f_path.dentry);
1623         if (err)
1624                 goto out;
1625         file_update_time(file);
1626         mutex_lock(&inode->i_mutex);
1627         first_index = pos >> PAGE_CACHE_SHIFT;
1628         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
1629
1630         mutex_lock(&root->fs_info->fs_mutex);
1631         trans = btrfs_start_transaction(root, 1);
1632         if (!trans) {
1633                 err = -ENOMEM;
1634                 mutex_unlock(&root->fs_info->fs_mutex);
1635                 goto out_unlock;
1636         }
1637         if (start_pos < inode->i_size) {
1638                 ret = drop_extents(trans, root, inode,
1639                                    start_pos,
1640                                    (pos + count + root->blocksize -1) &
1641                                    ~(root->blocksize - 1));
1642         }
1643         orig_extent_start = start_pos;
1644         ret = btrfs_alloc_extent(trans, root, num_blocks, 1,
1645                                  (u64)-1, &ins);
1646         BUG_ON(ret);
1647         ret = btrfs_insert_file_extent(trans, root, inode->i_ino,
1648                                        start_pos, ins.objectid, ins.offset);
1649         BUG_ON(ret);
1650         alloc_extent_start = ins.objectid;
1651         ret = btrfs_end_transaction(trans, root);
1652         mutex_unlock(&root->fs_info->fs_mutex);
1653
1654         while(count > 0) {
1655                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1656                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1657                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
1658                                         PAGE_CACHE_SHIFT;
1659 printk("num_pages is %lu\n", num_pages);
1660
1661                 memset(pages, 0, sizeof(pages));
1662                 ret = prepare_pages(root, file, pages, num_pages,
1663                                     pos, first_index, last_index,
1664                                     write_bytes, alloc_extent_start);
1665                 BUG_ON(ret);
1666
1667                 /* FIXME blocks != pagesize */
1668                 alloc_extent_start += num_pages;
1669                 ret = btrfs_copy_from_user(pos, num_pages,
1670                                            write_bytes, pages, buf);
1671                 BUG_ON(ret);
1672
1673 printk("2num_pages is %lu\n", num_pages);
1674                 ret = dirty_and_release_pages(NULL, root, file, pages,
1675                                               num_pages, orig_extent_start,
1676                                               pos, write_bytes);
1677                 BUG_ON(ret);
1678                 btrfs_drop_pages(pages, num_pages);
1679
1680                 buf += write_bytes;
1681                 count -= write_bytes;
1682                 pos += write_bytes;
1683                 num_written += write_bytes;
1684
1685                 balance_dirty_pages_ratelimited(inode->i_mapping);
1686                 cond_resched();
1687         }
1688 out_unlock:
1689         mutex_unlock(&inode->i_mutex);
1690 out:
1691         *ppos = pos;
1692         current->backing_dev_info = NULL;
1693         return num_written ? num_written : err;
1694 }
1695
1696 #if 0
1697 static ssize_t inline_one_page(struct btrfs_root *root, struct inode *inode,
1698                            struct page *page, loff_t pos,
1699                            size_t offset, size_t write_bytes)
1700 {
1701         struct btrfs_path *path;
1702         struct btrfs_trans_handle *trans;
1703         struct btrfs_key key;
1704         struct btrfs_leaf *leaf;
1705         struct btrfs_key found_key;
1706         int ret;
1707         size_t copy_size = 0;
1708         char *dst = NULL;
1709         int err = 0;
1710         size_t num_written = 0;
1711
1712         path = btrfs_alloc_path();
1713         BUG_ON(!path);
1714         mutex_lock(&root->fs_info->fs_mutex);
1715         trans = btrfs_start_transaction(root, 1);
1716         key.objectid = inode->i_ino;
1717         key.flags = 0;
1718         btrfs_set_key_type(&key, BTRFS_INLINE_DATA_KEY);
1719
1720 again:
1721         key.offset = pos;
1722         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1723         if (ret < 0) {
1724                 err = ret;
1725                 goto out;
1726         }
1727         if (ret == 0) {
1728                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1729                 btrfs_disk_key_to_cpu(&found_key,
1730                                       &leaf->items[path->slots[0]].key);
1731                 copy_size = btrfs_item_size(leaf->items + path->slots[0]);
1732                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1733                 copy_size = min(write_bytes, copy_size);
1734                 goto copyit;
1735         } else {
1736                 int slot = path->slots[0];
1737                 if (slot > 0) {
1738                         slot--;
1739                 }
1740                 // FIXME find max key
1741                 leaf = btrfs_buffer_leaf(path->nodes[0]);
1742                 btrfs_disk_key_to_cpu(&found_key,
1743                                       &leaf->items[slot].key);
1744                 if (found_key.objectid != inode->i_ino)
1745                         goto insert;
1746                 if (btrfs_key_type(&found_key) != BTRFS_INLINE_DATA_KEY)
1747                         goto insert;
1748                 copy_size = btrfs_item_size(leaf->items + slot);
1749                 if (found_key.offset + copy_size <= pos)
1750                         goto insert;
1751                 dst = btrfs_item_ptr(leaf, path->slots[0], char);
1752                 dst += pos - found_key.offset;
1753                 copy_size = copy_size - (pos - found_key.offset);
1754                 BUG_ON(copy_size < 0);
1755                 copy_size = min(write_bytes, copy_size);
1756                 WARN_ON(copy_size == 0);
1757                 goto copyit;
1758         }
1759 insert:
1760         btrfs_release_path(root, path);
1761         copy_size = min(write_bytes,
1762                         (size_t)BTRFS_LEAF_DATA_SIZE(root) -
1763                         sizeof(struct btrfs_item) * 4);
1764         ret = btrfs_insert_empty_item(trans, root, path, &key, copy_size);
1765         BUG_ON(ret);
1766         dst = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1767                              path->slots[0], char);
1768 copyit:
1769         WARN_ON(copy_size == 0);
1770         WARN_ON(dst + copy_size >
1771                 btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
1772                                                  path->slots[0], char) +
1773                 btrfs_item_size(btrfs_buffer_leaf(path->nodes[0])->items +
1774                                                   path->slots[0]));
1775         btrfs_memcpy(root, path->nodes[0]->b_data, dst,
1776                      page_address(page) + offset, copy_size);
1777         mark_buffer_dirty(path->nodes[0]);
1778         btrfs_release_path(root, path);
1779         pos += copy_size;
1780         offset += copy_size;
1781         num_written += copy_size;
1782         write_bytes -= copy_size;
1783         if (write_bytes)
1784                 goto again;
1785 out:
1786         btrfs_free_path(path);
1787         ret = btrfs_end_transaction(trans, root);
1788         BUG_ON(ret);
1789         mutex_unlock(&root->fs_info->fs_mutex);
1790         return num_written ? num_written : err;
1791 }
1792
1793 static ssize_t btrfs_file_inline_write(struct file *file,
1794                                        const char __user *buf,
1795                                        size_t count, loff_t *ppos)
1796 {
1797         loff_t pos;
1798         size_t num_written = 0;
1799         int err = 0;
1800         int ret = 0;
1801         struct inode *inode = file->f_path.dentry->d_inode;
1802         struct btrfs_root *root = BTRFS_I(inode)->root;
1803         unsigned long page_index;
1804
1805         if (file->f_flags & O_DIRECT)
1806                 return -EINVAL;
1807         pos = *ppos;
1808
1809         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
1810         current->backing_dev_info = inode->i_mapping->backing_dev_info;
1811         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
1812         if (err)
1813                 goto out;
1814         if (count == 0)
1815                 goto out;
1816         err = remove_suid(file->f_path.dentry);
1817         if (err)
1818                 goto out;
1819         file_update_time(file);
1820         mutex_lock(&inode->i_mutex);
1821         while(count > 0) {
1822                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
1823                 size_t write_bytes = min(count, PAGE_CACHE_SIZE - offset);
1824                 struct page *page;
1825
1826                 page_index = pos >> PAGE_CACHE_SHIFT;
1827                 page = grab_cache_page(inode->i_mapping, page_index);
1828                 if (!PageUptodate(page)) {
1829                         ret = mpage_readpage(page, btrfs_get_block);
1830                         BUG_ON(ret);
1831                         lock_page(page);
1832                 }
1833                 ret = btrfs_copy_from_user(pos, 1,
1834                                            write_bytes, &page, buf);
1835                 BUG_ON(ret);
1836                 write_bytes = inline_one_page(root, inode, page, pos,
1837                                       offset, write_bytes);
1838                 SetPageUptodate(page);
1839                 if (write_bytes > 0 && pos + write_bytes > inode->i_size) {
1840                         i_size_write(inode, pos + write_bytes);
1841                         mark_inode_dirty(inode);
1842                 }
1843                 page_cache_release(page);
1844                 unlock_page(page);
1845                 if (write_bytes < 0)
1846                         goto out_unlock;
1847                 buf += write_bytes;
1848                 count -= write_bytes;
1849                 pos += write_bytes;
1850                 num_written += write_bytes;
1851
1852                 balance_dirty_pages_ratelimited(inode->i_mapping);
1853                 cond_resched();
1854         }
1855 out_unlock:
1856         mutex_unlock(&inode->i_mutex);
1857 out:
1858         *ppos = pos;
1859         current->backing_dev_info = NULL;
1860         return num_written ? num_written : err;
1861 }
1862 #endif
1863
1864 static int btrfs_read_actor(read_descriptor_t *desc, struct page *page,
1865                         unsigned long offset, unsigned long size)
1866 {
1867         char *kaddr;
1868         unsigned long left, count = desc->count;
1869         struct inode *inode = page->mapping->host;
1870
1871         if (size > count)
1872                 size = count;
1873
1874         if (!PageChecked(page)) {
1875                 /* FIXME, do it per block */
1876                 struct btrfs_root *root = BTRFS_I(inode)->root;
1877                 int ret = btrfs_csum_verify_file_block(root,
1878                                           page->mapping->host->i_ino,
1879                                           page->index << PAGE_CACHE_SHIFT,
1880                                           kmap(page), PAGE_CACHE_SIZE);
1881                 if (ret) {
1882                         printk("failed to verify ino %lu page %lu\n",
1883                                page->mapping->host->i_ino,
1884                                page->index);
1885                         memset(page_address(page), 0, PAGE_CACHE_SIZE);
1886                 }
1887                 SetPageChecked(page);
1888                 kunmap(page);
1889         }
1890         /*
1891          * Faults on the destination of a read are common, so do it before
1892          * taking the kmap.
1893          */
1894         if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1895                 kaddr = kmap_atomic(page, KM_USER0);
1896                 left = __copy_to_user_inatomic(desc->arg.buf,
1897                                                 kaddr + offset, size);
1898                 kunmap_atomic(kaddr, KM_USER0);
1899                 if (left == 0)
1900                         goto success;
1901         }
1902
1903         /* Do it the slow way */
1904         kaddr = kmap(page);
1905         left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1906         kunmap(page);
1907
1908         if (left) {
1909                 size -= left;
1910                 desc->error = -EFAULT;
1911         }
1912 success:
1913         desc->count = count - size;
1914         desc->written += size;
1915         desc->arg.buf += size;
1916         return size;
1917 }
1918
1919 /**
1920  * btrfs_file_aio_read - filesystem read routine
1921  * @iocb:       kernel I/O control block
1922  * @iov:        io vector request
1923  * @nr_segs:    number of segments in the iovec
1924  * @pos:        current file position
1925  */
1926 static ssize_t btrfs_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1927                                    unsigned long nr_segs, loff_t pos)
1928 {
1929         struct file *filp = iocb->ki_filp;
1930         ssize_t retval;
1931         unsigned long seg;
1932         size_t count;
1933         loff_t *ppos = &iocb->ki_pos;
1934
1935         count = 0;
1936         for (seg = 0; seg < nr_segs; seg++) {
1937                 const struct iovec *iv = &iov[seg];
1938
1939                 /*
1940                  * If any segment has a negative length, or the cumulative
1941                  * length ever wraps negative then return -EINVAL.
1942                  */
1943                 count += iv->iov_len;
1944                 if (unlikely((ssize_t)(count|iv->iov_len) < 0))
1945                         return -EINVAL;
1946                 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
1947                         continue;
1948                 if (seg == 0)
1949                         return -EFAULT;
1950                 nr_segs = seg;
1951                 count -= iv->iov_len;   /* This segment is no good */
1952                 break;
1953         }
1954         retval = 0;
1955         if (count) {
1956                 for (seg = 0; seg < nr_segs; seg++) {
1957                         read_descriptor_t desc;
1958
1959                         desc.written = 0;
1960                         desc.arg.buf = iov[seg].iov_base;
1961                         desc.count = iov[seg].iov_len;
1962                         if (desc.count == 0)
1963                                 continue;
1964                         desc.error = 0;
1965                         do_generic_file_read(filp, ppos, &desc,
1966                                              btrfs_read_actor);
1967                         retval += desc.written;
1968                         if (desc.error) {
1969                                 retval = retval ?: desc.error;
1970                                 break;
1971                         }
1972                 }
1973         }
1974         return retval;
1975 }
1976
1977 static int create_subvol(struct btrfs_root *root, char *name, int namelen)
1978 {
1979         struct btrfs_trans_handle *trans;
1980         struct btrfs_key key;
1981         struct btrfs_root_item root_item;
1982         struct btrfs_inode_item *inode_item;
1983         struct buffer_head *subvol;
1984         struct btrfs_leaf *leaf;
1985         struct btrfs_root *new_root;
1986         struct inode *inode;
1987         int ret;
1988         u64 objectid;
1989         u64 new_dirid = BTRFS_FIRST_FREE_OBJECTID;
1990
1991         mutex_lock(&root->fs_info->fs_mutex);
1992         trans = btrfs_start_transaction(root, 1);
1993         BUG_ON(!trans);
1994
1995         subvol = btrfs_alloc_free_block(trans, root);
1996         leaf = btrfs_buffer_leaf(subvol);
1997         btrfs_set_header_nritems(&leaf->header, 0);
1998         btrfs_set_header_level(&leaf->header, 0);
1999         btrfs_set_header_blocknr(&leaf->header, bh_blocknr(subvol));
2000         btrfs_set_header_generation(&leaf->header, trans->transid);
2001         memcpy(leaf->header.fsid, root->fs_info->disk_super->fsid,
2002                sizeof(leaf->header.fsid));
2003
2004         inode_item = &root_item.inode;
2005         memset(inode_item, 0, sizeof(*inode_item));
2006         btrfs_set_inode_generation(inode_item, 1);
2007         btrfs_set_inode_size(inode_item, 3);
2008         btrfs_set_inode_nlink(inode_item, 1);
2009         btrfs_set_inode_nblocks(inode_item, 1);
2010         btrfs_set_inode_mode(inode_item, S_IFDIR | 0755);
2011
2012         btrfs_set_root_blocknr(&root_item, bh_blocknr(subvol));
2013         btrfs_set_root_refs(&root_item, 1);
2014
2015         mark_buffer_dirty(subvol);
2016         brelse(subvol);
2017         subvol = NULL;
2018
2019         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2020                                        0, &objectid);
2021         BUG_ON(ret);
2022
2023         btrfs_set_root_dirid(&root_item, new_dirid);
2024
2025         key.objectid = objectid;
2026         key.offset = 1;
2027         key.flags = 0;
2028         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2029         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2030                                 &root_item);
2031         BUG_ON(ret);
2032
2033         /*
2034          * insert the directory item
2035          */
2036         key.offset = (u64)-1;
2037         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2038                                     name, namelen,
2039                                     root->fs_info->sb->s_root->d_inode->i_ino,
2040                                     &key, 0);
2041         BUG_ON(ret);
2042
2043         ret = btrfs_commit_transaction(trans, root);
2044         BUG_ON(ret);
2045
2046         new_root = btrfs_read_fs_root(root->fs_info, &key);
2047         BUG_ON(!new_root);
2048
2049         trans = btrfs_start_transaction(new_root, 1);
2050         BUG_ON(!trans);
2051
2052         inode = btrfs_new_inode(trans, new_root, new_dirid, S_IFDIR | 0700);
2053         inode->i_op = &btrfs_dir_inode_operations;
2054         inode->i_fop = &btrfs_dir_file_operations;
2055
2056         ret = btrfs_make_empty_dir(trans, new_root, new_dirid, new_dirid);
2057         BUG_ON(ret);
2058
2059         inode->i_nlink = 1;
2060         inode->i_size = 6;
2061         ret = btrfs_update_inode(trans, new_root, inode);
2062         BUG_ON(ret);
2063
2064         ret = btrfs_commit_transaction(trans, new_root);
2065         BUG_ON(ret);
2066
2067         iput(inode);
2068
2069         mutex_unlock(&root->fs_info->fs_mutex);
2070         return 0;
2071 }
2072
2073 static int create_snapshot(struct btrfs_root *root, char *name, int namelen)
2074 {
2075         struct btrfs_trans_handle *trans;
2076         struct btrfs_key key;
2077         struct btrfs_root_item new_root_item;
2078         int ret;
2079         u64 objectid;
2080
2081         if (!root->ref_cows)
2082                 return -EINVAL;
2083
2084         mutex_lock(&root->fs_info->fs_mutex);
2085         trans = btrfs_start_transaction(root, 1);
2086         BUG_ON(!trans);
2087
2088         ret = btrfs_update_inode(trans, root, root->inode);
2089         BUG_ON(ret);
2090
2091         ret = btrfs_find_free_objectid(trans, root->fs_info->tree_root,
2092                                        0, &objectid);
2093         BUG_ON(ret);
2094
2095         memcpy(&new_root_item, &root->root_item,
2096                sizeof(new_root_item));
2097
2098         key.objectid = objectid;
2099         key.offset = 1;
2100         key.flags = 0;
2101         btrfs_set_key_type(&key, BTRFS_ROOT_ITEM_KEY);
2102         btrfs_set_root_blocknr(&new_root_item, bh_blocknr(root->node));
2103
2104         ret = btrfs_insert_root(trans, root->fs_info->tree_root, &key,
2105                                 &new_root_item);
2106         BUG_ON(ret);
2107
2108         /*
2109          * insert the directory item
2110          */
2111         key.offset = (u64)-1;
2112         ret = btrfs_insert_dir_item(trans, root->fs_info->tree_root,
2113                                     name, namelen,
2114                                     root->fs_info->sb->s_root->d_inode->i_ino,
2115                                     &key, 0);
2116
2117         BUG_ON(ret);
2118
2119         ret = btrfs_inc_root_ref(trans, root);
2120         BUG_ON(ret);
2121
2122         ret = btrfs_commit_transaction(trans, root);
2123         BUG_ON(ret);
2124         mutex_unlock(&root->fs_info->fs_mutex);
2125         return 0;
2126 }
2127
2128 static int add_disk(struct btrfs_root *root, char *name, int namelen)
2129 {
2130         struct block_device *bdev;
2131         struct btrfs_path *path;
2132         struct super_block *sb = root->fs_info->sb;
2133         struct btrfs_root *dev_root = root->fs_info->dev_root;
2134         struct btrfs_trans_handle *trans;
2135         struct btrfs_device_item *dev_item;
2136         struct btrfs_key key;
2137         u16 item_size;
2138         u64 num_blocks;
2139         u64 new_blocks;
2140         u64 device_id;
2141         int ret;
2142
2143 printk("adding disk %s\n", name);
2144         path = btrfs_alloc_path();
2145         if (!path)
2146                 return -ENOMEM;
2147         num_blocks = btrfs_super_total_blocks(root->fs_info->disk_super);
2148         bdev = open_bdev_excl(name, O_RDWR, sb);
2149         if (IS_ERR(bdev)) {
2150                 ret = PTR_ERR(bdev);
2151 printk("open bdev excl failed ret %d\n", ret);
2152                 goto out_nolock;
2153         }
2154         set_blocksize(bdev, sb->s_blocksize);
2155         new_blocks = bdev->bd_inode->i_size >> sb->s_blocksize_bits;
2156         key.objectid = num_blocks;
2157         key.offset = new_blocks;
2158         key.flags = 0;
2159         btrfs_set_key_type(&key, BTRFS_DEV_ITEM_KEY);
2160
2161         mutex_lock(&dev_root->fs_info->fs_mutex);
2162         trans = btrfs_start_transaction(dev_root, 1);
2163         item_size = sizeof(*dev_item) + namelen;
2164 printk("insert empty on %Lu %Lu %u size %d\n", num_blocks, new_blocks, key.flags, item_size);
2165         ret = btrfs_insert_empty_item(trans, dev_root, path, &key, item_size);
2166         if (ret) {
2167 printk("insert failed %d\n", ret);
2168                 close_bdev_excl(bdev);
2169                 if (ret > 0)
2170                         ret = -EEXIST;
2171                 goto out;
2172         }
2173         dev_item = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]),
2174                                   path->slots[0], struct btrfs_device_item);
2175         btrfs_set_device_pathlen(dev_item, namelen);
2176         memcpy(dev_item + 1, name, namelen);
2177
2178         device_id = btrfs_super_last_device_id(root->fs_info->disk_super) + 1;
2179         btrfs_set_super_last_device_id(root->fs_info->disk_super, device_id);
2180         btrfs_set_device_id(dev_item, device_id);
2181         mark_buffer_dirty(path->nodes[0]);
2182
2183         ret = btrfs_insert_dev_radix(root, bdev, device_id, num_blocks,
2184                                      new_blocks);
2185
2186         if (!ret) {
2187                 btrfs_set_super_total_blocks(root->fs_info->disk_super,
2188                                              num_blocks + new_blocks);
2189                 i_size_write(root->fs_info->btree_inode,
2190                              (num_blocks + new_blocks) <<
2191                              root->fs_info->btree_inode->i_blkbits);
2192         }
2193
2194 out:
2195         ret = btrfs_commit_transaction(trans, dev_root);
2196         BUG_ON(ret);
2197         mutex_unlock(&root->fs_info->fs_mutex);
2198 out_nolock:
2199         btrfs_free_path(path);
2200
2201         return ret;
2202 }
2203
2204 static int btrfs_ioctl(struct inode *inode, struct file *filp, unsigned int
2205                        cmd, unsigned long arg)
2206 {
2207         struct btrfs_root *root = BTRFS_I(inode)->root;
2208         struct btrfs_ioctl_vol_args vol_args;
2209         int ret = 0;
2210         int namelen;
2211         struct btrfs_path *path;
2212         u64 root_dirid;
2213
2214         switch (cmd) {
2215         case BTRFS_IOC_SNAP_CREATE:
2216                 if (copy_from_user(&vol_args,
2217                                    (struct btrfs_ioctl_vol_args __user *)arg,
2218                                    sizeof(vol_args)))
2219                         return -EFAULT;
2220                 namelen = strlen(vol_args.name);
2221                 if (namelen > BTRFS_VOL_NAME_MAX)
2222                         return -EINVAL;
2223                 path = btrfs_alloc_path();
2224                 if (!path)
2225                         return -ENOMEM;
2226                 root_dirid = root->fs_info->sb->s_root->d_inode->i_ino,
2227                 mutex_lock(&root->fs_info->fs_mutex);
2228                 ret = btrfs_lookup_dir_item(NULL, root->fs_info->tree_root,
2229                                     path, root_dirid,
2230                                     vol_args.name, namelen, 0);
2231                 mutex_unlock(&root->fs_info->fs_mutex);
2232                 btrfs_free_path(path);
2233                 if (ret == 0)
2234                         return -EEXIST;
2235
2236                 if (root == root->fs_info->tree_root)
2237                         ret = create_subvol(root, vol_args.name, namelen);
2238                 else
2239                         ret = create_snapshot(root, vol_args.name, namelen);
2240                 WARN_ON(ret);
2241                 break;
2242         case BTRFS_IOC_ADD_DISK:
2243                 if (copy_from_user(&vol_args,
2244                                    (struct btrfs_ioctl_vol_args __user *)arg,
2245                                    sizeof(vol_args)))
2246                         return -EFAULT;
2247                 namelen = strlen(vol_args.name);
2248                 if (namelen > BTRFS_VOL_NAME_MAX)
2249                         return -EINVAL;
2250                 vol_args.name[namelen] = '\0';
2251                 ret = add_disk(root, vol_args.name, namelen);
2252                 break;
2253         default:
2254                 return -ENOTTY;
2255         }
2256         return ret;
2257 }
2258
2259 static struct kmem_cache *btrfs_inode_cachep;
2260 struct kmem_cache *btrfs_trans_handle_cachep;
2261 struct kmem_cache *btrfs_transaction_cachep;
2262 struct kmem_cache *btrfs_bit_radix_cachep;
2263 struct kmem_cache *btrfs_path_cachep;
2264
2265 /*
2266  * Called inside transaction, so use GFP_NOFS
2267  */
2268 static struct inode *btrfs_alloc_inode(struct super_block *sb)
2269 {
2270         struct btrfs_inode *ei;
2271
2272         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
2273         if (!ei)
2274                 return NULL;
2275         return &ei->vfs_inode;
2276 }
2277
2278 static void btrfs_destroy_inode(struct inode *inode)
2279 {
2280         WARN_ON(!list_empty(&inode->i_dentry));
2281         WARN_ON(inode->i_data.nrpages);
2282
2283         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
2284 }
2285
2286 static void init_once(void * foo, struct kmem_cache * cachep,
2287                       unsigned long flags)
2288 {
2289         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
2290
2291         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2292             SLAB_CTOR_CONSTRUCTOR) {
2293                 inode_init_once(&ei->vfs_inode);
2294         }
2295 }
2296
2297 static int init_inodecache(void)
2298 {
2299         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
2300                                              sizeof(struct btrfs_inode),
2301                                              0, (SLAB_RECLAIM_ACCOUNT|
2302                                                 SLAB_MEM_SPREAD),
2303                                              init_once, NULL);
2304         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
2305                                              sizeof(struct btrfs_trans_handle),
2306                                              0, (SLAB_RECLAIM_ACCOUNT|
2307                                                 SLAB_MEM_SPREAD),
2308                                              NULL, NULL);
2309         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
2310                                              sizeof(struct btrfs_transaction),
2311                                              0, (SLAB_RECLAIM_ACCOUNT|
2312                                                 SLAB_MEM_SPREAD),
2313                                              NULL, NULL);
2314         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
2315                                              sizeof(struct btrfs_transaction),
2316                                              0, (SLAB_RECLAIM_ACCOUNT|
2317                                                 SLAB_MEM_SPREAD),
2318                                              NULL, NULL);
2319         btrfs_bit_radix_cachep = kmem_cache_create("btrfs_radix",
2320                                              256,
2321                                              0, (SLAB_RECLAIM_ACCOUNT|
2322                                                 SLAB_MEM_SPREAD |
2323                                                 SLAB_DESTROY_BY_RCU),
2324                                              NULL, NULL);
2325         if (btrfs_inode_cachep == NULL || btrfs_trans_handle_cachep == NULL ||
2326             btrfs_transaction_cachep == NULL || btrfs_bit_radix_cachep == NULL)
2327                 return -ENOMEM;
2328         return 0;
2329 }
2330
2331 static void destroy_inodecache(void)
2332 {
2333         kmem_cache_destroy(btrfs_inode_cachep);
2334         kmem_cache_destroy(btrfs_trans_handle_cachep);
2335         kmem_cache_destroy(btrfs_transaction_cachep);
2336         kmem_cache_destroy(btrfs_bit_radix_cachep);
2337         kmem_cache_destroy(btrfs_path_cachep);
2338 }
2339
2340 static int btrfs_get_sb(struct file_system_type *fs_type,
2341         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2342 {
2343         return get_sb_bdev(fs_type, flags, dev_name, data,
2344                            btrfs_fill_super, mnt);
2345 }
2346
2347 static struct file_system_type btrfs_fs_type = {
2348         .owner          = THIS_MODULE,
2349         .name           = "btrfs",
2350         .get_sb         = btrfs_get_sb,
2351         .kill_sb        = kill_block_super,
2352         .fs_flags       = FS_REQUIRES_DEV,
2353 };
2354
2355 static struct super_operations btrfs_super_ops = {
2356         .statfs         = simple_statfs,
2357         .delete_inode   = btrfs_delete_inode,
2358         .put_super      = btrfs_put_super,
2359         .read_inode     = btrfs_read_locked_inode,
2360         .write_super    = btrfs_write_super,
2361         .sync_fs        = btrfs_sync_fs,
2362         .write_inode    = btrfs_write_inode,
2363         .alloc_inode    = btrfs_alloc_inode,
2364         .destroy_inode  = btrfs_destroy_inode,
2365 };
2366
2367 static struct inode_operations btrfs_dir_inode_operations = {
2368         .lookup         = btrfs_lookup,
2369         .create         = btrfs_create,
2370         .unlink         = btrfs_unlink,
2371         .mkdir          = btrfs_mkdir,
2372         .rmdir          = btrfs_rmdir,
2373 };
2374
2375 static struct inode_operations btrfs_dir_ro_inode_operations = {
2376         .lookup         = btrfs_lookup,
2377 };
2378
2379 static struct file_operations btrfs_dir_file_operations = {
2380         .llseek         = generic_file_llseek,
2381         .read           = generic_read_dir,
2382         .readdir        = btrfs_readdir,
2383         .ioctl          = btrfs_ioctl,
2384 };
2385
2386 static struct address_space_operations btrfs_aops = {
2387         .readpage       = btrfs_readpage,
2388         .writepage      = btrfs_writepage,
2389         .sync_page      = block_sync_page,
2390         .prepare_write  = btrfs_prepare_write,
2391         .commit_write   = btrfs_commit_write,
2392 };
2393
2394 static struct inode_operations btrfs_file_inode_operations = {
2395         .truncate       = btrfs_truncate,
2396 };
2397
2398 static struct file_operations btrfs_file_operations = {
2399         .llseek         = generic_file_llseek,
2400         .read           = do_sync_read,
2401         .aio_read       = btrfs_file_aio_read,
2402         .write          = btrfs_file_write,
2403         .mmap           = generic_file_mmap,
2404         .open           = generic_file_open,
2405         .ioctl          = btrfs_ioctl,
2406 };
2407
2408 static int __init init_btrfs_fs(void)
2409 {
2410         int err;
2411         printk("btrfs loaded!\n");
2412         err = init_inodecache();
2413         if (err)
2414                 return err;
2415         kset_set_kset_s(&btrfs_subsys, fs_subsys);
2416         err = subsystem_register(&btrfs_subsys);
2417         if (err)
2418                 goto out;
2419         return register_filesystem(&btrfs_fs_type);
2420 out:
2421         destroy_inodecache();
2422         return err;
2423 }
2424
2425 static void __exit exit_btrfs_fs(void)
2426 {
2427         destroy_inodecache();
2428         unregister_filesystem(&btrfs_fs_type);
2429         subsystem_unregister(&btrfs_subsys);
2430         printk("btrfs unloaded\n");
2431 }
2432
2433 module_init(init_btrfs_fs)
2434 module_exit(exit_btrfs_fs)
2435
2436 MODULE_LICENSE("GPL");