fs/btrfs/extent-tree.c

   1 #include <linux/module.h>
   2 #include "ctree.h"
   3 #include "disk-io.h"
   4 #include "print-tree.h"
   5 #include "transaction.h"
   6
   7 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
   8                             *orig_root, u64 num_blocks, u64 search_start, u64
   9                             search_end, struct btrfs_key *ins);
  10 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  11                                  btrfs_root *extent_root);
  12 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
  13                                btrfs_root *extent_root);
  14
  15 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
  16                                 struct btrfs_root *root,
  17                                 u64 blocknr, u64 num_blocks)
  18 {
  19         struct btrfs_path *path;
  20         int ret;
  21         struct btrfs_key key;
  22         struct btrfs_leaf *l;
  23         struct btrfs_extent_item *item;
  24         struct btrfs_key ins;
  25         u32 refs;
  26
  27         find_free_extent(trans, root->fs_info->extent_root, 0, 0, (u64)-1,
  28                          &ins);
  29         path = btrfs_alloc_path();
  30         BUG_ON(!path);
  31         btrfs_init_path(path);
  32         key.objectid = blocknr;
  33         key.flags = 0;
  34         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  35         key.offset = num_blocks;
  36         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
  37                                 0, 1);
  38         if (ret != 0) {
  39 printk("can't find block %Lu %Lu\n", blocknr, num_blocks);
  40                 BUG();
  41         }
  42         BUG_ON(ret != 0);
  43         l = btrfs_buffer_leaf(path->nodes[0]);
  44         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
  45         refs = btrfs_extent_refs(item);
  46         btrfs_set_extent_refs(item, refs + 1);
  47         btrfs_mark_buffer_dirty(path->nodes[0]);
  48
  49         btrfs_release_path(root->fs_info->extent_root, path);
  50         btrfs_free_path(path);
  51         finish_current_insert(trans, root->fs_info->extent_root);
  52         del_pending_extents(trans, root->fs_info->extent_root);
  53         return 0;
  54 }
  55
  56 static int lookup_extent_ref(struct btrfs_trans_handle *trans,
  57                              struct btrfs_root *root, u64 blocknr,
  58                              u64 num_blocks, u32 *refs)
  59 {
  60         struct btrfs_path *path;
  61         int ret;
  62         struct btrfs_key key;
  63         struct btrfs_leaf *l;
  64         struct btrfs_extent_item *item;
  65
  66         path = btrfs_alloc_path();
  67         btrfs_init_path(path);
  68         key.objectid = blocknr;
  69         key.offset = num_blocks;
  70         key.flags = 0;
  71         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  72         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path,
  73                                 0, 0);
  74         if (ret != 0)
  75                 BUG();
  76         l = btrfs_buffer_leaf(path->nodes[0]);
  77         item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item);
  78         *refs = btrfs_extent_refs(item);
  79         btrfs_release_path(root->fs_info->extent_root, path);
  80         btrfs_free_path(path);
  81         return 0;
  82 }
  83
  84 int btrfs_inc_root_ref(struct btrfs_trans_handle *trans,
  85                        struct btrfs_root *root)
  86 {
  87         return btrfs_inc_extent_ref(trans, root, bh_blocknr(root->node), 1);
  88 }
  89
  90 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  91                   struct buffer_head *buf)
  92 {
  93         u64 blocknr;
  94         struct btrfs_node *buf_node;
  95         struct btrfs_leaf *buf_leaf;
  96         struct btrfs_disk_key *key;
  97         struct btrfs_file_extent_item *fi;
  98         int i;
  99         int leaf;
 100         int ret;
 101
 102         if (!root->ref_cows)
 103                 return 0;
 104         buf_node = btrfs_buffer_node(buf);
 105         leaf = btrfs_is_leaf(buf_node);
 106         buf_leaf = btrfs_buffer_leaf(buf);
 107         for (i = 0; i < btrfs_header_nritems(&buf_node->header); i++) {
 108                 if (leaf) {
 109                         key = &buf_leaf->items[i].key;
 110                         if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
 111                                 continue;
 112                         fi = btrfs_item_ptr(buf_leaf, i,
 113                                             struct btrfs_file_extent_item);
 114                         if (btrfs_file_extent_type(fi) ==
 115                             BTRFS_FILE_EXTENT_INLINE)
 116                                 continue;
 117                         ret = btrfs_inc_extent_ref(trans, root,
 118                                     btrfs_file_extent_disk_blocknr(fi),
 119                                     btrfs_file_extent_disk_num_blocks(fi));
 120                         BUG_ON(ret);
 121                 } else {
 122                         blocknr = btrfs_node_blockptr(buf_node, i);
 123                         ret = btrfs_inc_extent_ref(trans, root, blocknr, 1);
 124                         BUG_ON(ret);
 125                 }
 126         }
 127         return 0;
 128 }
 129
 130 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, struct
 131                                btrfs_root *root)
 132 {
 133         unsigned long gang[8];
 134         u64 first = 0;
 135         int ret;
 136         int i;
 137         struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix;
 138
 139         while(1) {
 140                 ret = find_first_radix_bit(pinned_radix, gang,
 141                                            ARRAY_SIZE(gang));
 142                 if (!ret)
 143                         break;
 144                 if (!first)
 145                         first = gang[0];
 146                 for (i = 0; i < ret; i++) {
 147                         clear_radix_bit(pinned_radix, gang[i]);
 148                 }
 149         }
 150         if (root->fs_info->last_insert.objectid > first)
 151                 root->fs_info->last_insert.objectid = first;
 152         root->fs_info->last_insert.offset = 0;
 153         return 0;
 154 }
 155
 156 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
 157                                  btrfs_root *extent_root)
 158 {
 159         struct btrfs_key ins;
 160         struct btrfs_extent_item extent_item;
 161         int i;
 162         int ret;
 163         u64 super_blocks_used;
 164         struct btrfs_fs_info *info = extent_root->fs_info;
 165
 166         btrfs_set_extent_refs(&extent_item, 1);
 167         ins.offset = 1;
 168         ins.flags = 0;
 169         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
 170         btrfs_set_extent_owner(&extent_item, extent_root->root_key.objectid);
 171
 172         for (i = 0; i < extent_root->fs_info->extent_tree_insert_nr; i++) {
 173                 ins.objectid = extent_root->fs_info->extent_tree_insert[i];
 174                 super_blocks_used = btrfs_super_blocks_used(info->disk_super);
 175                 btrfs_set_super_blocks_used(info->disk_super,
 176                                             super_blocks_used + 1);
 177                 ret = btrfs_insert_item(trans, extent_root, &ins, &extent_item,
 178                                         sizeof(extent_item));
 179                 BUG_ON(ret);
 180         }
 181         extent_root->fs_info->extent_tree_insert_nr = 0;
 182         extent_root->fs_info->extent_tree_prealloc_nr = 0;
 183         return 0;
 184 }
 185
 186 static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending)
 187 {
 188         int err;
 189         struct btrfs_header *header;
 190         struct buffer_head *bh;
 191
 192         if (!pending) {
 193                 bh = btrfs_find_tree_block(root, blocknr);
 194                 if (bh) {
 195                         if (buffer_uptodate(bh)) {
 196                                 u64 transid =
 197                                     root->fs_info->running_transaction->transid;
 198                                 header = btrfs_buffer_header(bh);
 199                                 if (btrfs_header_generation(header) ==
 200                                     transid) {
 201                                         btrfs_block_release(root, bh);
 202                                         return 0;
 203                                 }
 204                         }
 205                         btrfs_block_release(root, bh);
 206                 }
 207                 err = set_radix_bit(&root->fs_info->pinned_radix, blocknr);
 208         } else {
 209                 err = set_radix_bit(&root->fs_info->pending_del_radix, blocknr);
 210         }
 211         BUG_ON(err);
 212         return 0;
 213 }
 214
 215 /*
 216  * remove an extent from the root, returns 0 on success
 217  */
 218 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 219                          *root, u64 blocknr, u64 num_blocks, int pin)
 220 {
 221         struct btrfs_path *path;
 222         struct btrfs_key key;
 223         struct btrfs_fs_info *info = root->fs_info;
 224         struct btrfs_root *extent_root = info->extent_root;
 225         int ret;
 226         struct btrfs_extent_item *ei;
 227         struct btrfs_key ins;
 228         u32 refs;
 229
 230         key.objectid = blocknr;
 231         key.flags = 0;
 232         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 233         key.offset = num_blocks;
 234
 235         find_free_extent(trans, root, 0, 0, (u64)-1, &ins);
 236         path = btrfs_alloc_path();
 237         BUG_ON(!path);
 238         btrfs_init_path(path);
 239
 240         ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1);
 241         if (ret) {
 242                 printk("failed to find %Lu\n", key.objectid);
 243                 btrfs_print_tree(extent_root, extent_root->node);
 244                 printk("failed to find %Lu\n", key.objectid);
 245                 BUG();
 246         }
 247         ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0],
 248                             struct btrfs_extent_item);
 249         BUG_ON(ei->refs == 0);
 250         refs = btrfs_extent_refs(ei) - 1;
 251         btrfs_set_extent_refs(ei, refs);
 252         btrfs_mark_buffer_dirty(path->nodes[0]);
 253         if (refs == 0) {
 254                 u64 super_blocks_used;
 255
 256                 if (pin) {
 257                         ret = pin_down_block(root, blocknr, 0);
 258                         BUG_ON(ret);
 259                 }
 260
 261                 super_blocks_used = btrfs_super_blocks_used(info->disk_super);
 262                 btrfs_set_super_blocks_used(info->disk_super,
 263                                             super_blocks_used - num_blocks);
 264                 ret = btrfs_del_item(trans, extent_root, path);
 265                 if (ret)
 266                         BUG();
 267         }
 268         btrfs_release_path(extent_root, path);
 269         btrfs_free_path(path);
 270         finish_current_insert(trans, extent_root);
 271         return ret;
 272 }
 273
 274 /*
 275  * find all the blocks marked as pending in the radix tree and remove
 276  * them from the extent map
 277  */
 278 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 279                                btrfs_root *extent_root)
 280 {
 281         int ret;
 282         int wret;
 283         int err = 0;
 284         unsigned long gang[4];
 285         int i;
 286         struct radix_tree_root *pending_radix;
 287         struct radix_tree_root *pinned_radix;
 288
 289         pending_radix = &extent_root->fs_info->pending_del_radix;
 290         pinned_radix = &extent_root->fs_info->pinned_radix;
 291
 292         while(1) {
 293                 ret = find_first_radix_bit(pending_radix, gang,
 294                                            ARRAY_SIZE(gang));
 295                 if (!ret)
 296                         break;
 297                 for (i = 0; i < ret; i++) {
 298                         wret = set_radix_bit(pinned_radix, gang[i]);
 299                         BUG_ON(wret);
 300                         wret = clear_radix_bit(pending_radix, gang[i]);
 301                         BUG_ON(wret);
 302                         wret = __free_extent(trans, extent_root,
 303                                              gang[i], 1, 0);
 304                         if (wret)
 305                                 err = wret;
 306                 }
 307         }
 308         return err;
 309 }
 310
 311 /*
 312  * remove an extent from the root, returns 0 on success
 313  */
 314 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 315                       *root, u64 blocknr, u64 num_blocks, int pin)
 316 {
 317         struct btrfs_root *extent_root = root->fs_info->extent_root;
 318         int pending_ret;
 319         int ret;
 320
 321         if (root == extent_root) {
 322                 pin_down_block(root, blocknr, 1);
 323                 return 0;
 324         }
 325         ret = __free_extent(trans, root, blocknr, num_blocks, pin);
 326         pending_ret = del_pending_extents(trans, root->fs_info->extent_root);
 327         return ret ? ret : pending_ret;
 328 }
 329
 330 /*
 331  * walks the btree of allocated extents and find a hole of a given size.
 332  * The key ins is changed to record the hole:
 333  * ins->objectid == block start
 334  * ins->flags = BTRFS_EXTENT_ITEM_KEY
 335  * ins->offset == number of blocks
 336  * Any available blocks before search_start are skipped.
 337  */
 338 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 339                             *orig_root, u64 num_blocks, u64 search_start, u64
 340                             search_end, struct btrfs_key *ins)
 341 {
 342         struct btrfs_path *path;
 343         struct btrfs_key key;
 344         int ret;
 345         u64 hole_size = 0;
 346         int slot = 0;
 347         u64 last_block = 0;
 348         u64 test_block;
 349         int start_found;
 350         struct btrfs_leaf *l;
 351         struct btrfs_root * root = orig_root->fs_info->extent_root;
 352         struct btrfs_fs_info *info = root->fs_info;
 353         int total_needed = num_blocks;
 354         int total_found = 0;
 355         int fill_prealloc = 0;
 356         int level;
 357
 358         path = btrfs_alloc_path();
 359         ins->flags = 0;
 360         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
 361
 362         level = btrfs_header_level(btrfs_buffer_header(root->node));
 363         if (num_blocks == 0) {
 364                 fill_prealloc = 1;
 365                 num_blocks = 1;
 366                 total_needed = min(level + 2, BTRFS_MAX_LEVEL) * 3;
 367         }
 368         if (info->last_insert.objectid == 0 && search_end == (u64)-1) {
 369                 struct btrfs_disk_key *last_key;
 370                 btrfs_init_path(path);
 371                 ins->objectid = (u64)-1;
 372                 ins->offset = (u64)-1;
 373                 ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
 374                 if (ret < 0)
 375                         goto error;
 376                 BUG_ON(ret == 0);
 377                 if (path->slots[0] > 0)
 378                         path->slots[0]--;
 379                 l = btrfs_buffer_leaf(path->nodes[0]);
 380                 last_key = &l->items[path->slots[0]].key;
 381                 search_start = btrfs_disk_key_objectid(last_key);
 382         }
 383         if (info->last_insert.objectid > search_start)
 384                 search_start = info->last_insert.objectid;
 385
 386 check_failed:
 387         btrfs_init_path(path);
 388         ins->objectid = search_start;
 389         ins->offset = 0;
 390         start_found = 0;
 391         ret = btrfs_search_slot(trans, root, ins, path, 0, 0);
 392         if (ret < 0)
 393                 goto error;
 394
 395         if (path->slots[0] > 0)
 396                 path->slots[0]--;
 397
 398         while (1) {
 399                 l = btrfs_buffer_leaf(path->nodes[0]);
 400                 slot = path->slots[0];
 401                 if (slot >= btrfs_header_nritems(&l->header)) {
 402                         if (fill_prealloc) {
 403                                 info->extent_tree_prealloc_nr = 0;
 404                                 total_found = 0;
 405                         }
 406                         ret = btrfs_next_leaf(root, path);
 407                         if (ret == 0)
 408                                 continue;
 409                         if (ret < 0)
 410                                 goto error;
 411                         if (!start_found) {
 412                                 ins->objectid = search_start;
 413                                 ins->offset = (u64)-1 - search_start;
 414                                 start_found = 1;
 415                                 goto check_pending;
 416                         }
 417                         ins->objectid = last_block > search_start ?
 418                                         last_block : search_start;
 419                         ins->offset = (u64)-1 - ins->objectid;
 420                         goto check_pending;
 421                 }
 422                 btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
 423                 if (key.objectid >= search_start) {
 424                         if (start_found) {
 425                                 if (last_block < search_start)
 426                                         last_block = search_start;
 427                                 hole_size = key.objectid - last_block;
 428                                 if (hole_size > num_blocks) {
 429                                         ins->objectid = last_block;
 430                                         ins->offset = hole_size;
 431                                         goto check_pending;
 432                                 }
 433                         }
 434                 }
 435                 start_found = 1;
 436                 last_block = key.objectid + key.offset;
 437                 path->slots[0]++;
 438         }
 439         // FIXME -ENOSPC
 440 check_pending:
 441         /* we have to make sure we didn't find an extent that has already
 442          * been allocated by the map tree or the original allocation
 443          */
 444         btrfs_release_path(root, path);
 445         BUG_ON(ins->objectid < search_start);
 446         for (test_block = ins->objectid;
 447              test_block < ins->objectid + num_blocks; test_block++) {
 448                 if (test_radix_bit(&info->pinned_radix, test_block)) {
 449                         search_start = test_block + 1;
 450                         goto check_failed;
 451                 }
 452         }
 453         if (!fill_prealloc && info->extent_tree_insert_nr) {
 454                 u64 last =
 455                   info->extent_tree_insert[info->extent_tree_insert_nr - 1];
 456                 if (ins->objectid + num_blocks >
 457                     info->extent_tree_insert[0] &&
 458                     ins->objectid <= last) {
 459                         search_start = last + 1;
 460                         WARN_ON(1);
 461                         goto check_failed;
 462                 }
 463         }
 464         if (!fill_prealloc && info->extent_tree_prealloc_nr) {
 465                 u64 first =
 466                   info->extent_tree_prealloc[info->extent_tree_prealloc_nr - 1];
 467                 if (ins->objectid + num_blocks > first &&
 468                     ins->objectid <= info->extent_tree_prealloc[0]) {
 469                         search_start = info->extent_tree_prealloc[0] + 1;
 470                         WARN_ON(1);
 471                         goto check_failed;
 472                 }
 473         }
 474         if (fill_prealloc) {
 475                 int nr;
 476                 test_block = ins->objectid;
 477                 while(test_block < ins->objectid + ins->offset &&
 478                       total_found < total_needed) {
 479                         nr = total_needed - total_found - 1;
 480                         BUG_ON(nr < 0);
 481                         root->fs_info->extent_tree_prealloc[nr] =
 482                                 test_block;
 483                         total_found++;
 484                         test_block++;
 485                 }
 486                 if (total_found < total_needed) {
 487                         search_start = test_block;
 488                         goto check_failed;
 489                 }
 490                 root->fs_info->extent_tree_prealloc_nr = total_found;
 491         }
 492         root->fs_info->last_insert.objectid = ins->objectid;
 493         ins->offset = num_blocks;
 494         btrfs_free_path(path);
 495         return 0;
 496 error:
 497         btrfs_release_path(root, path);
 498         btrfs_free_path(path);
 499         return ret;
 500 }
 501
 502 /*
 503  * finds a free extent and does all the dirty work required for allocation
 504  * returns the key for the extent through ins, and a tree buffer for
 505  * the first block of the extent through buf.
 506  *
 507  * returns 0 if everything worked, non-zero otherwise.
 508  */
 509 int btrfs_alloc_extent(struct btrfs_trans_handle *trans,
 510                        struct btrfs_root *root, u64 owner,
 511                        u64 num_blocks, u64 search_start,
 512                        u64 search_end, struct btrfs_key *ins)
 513 {
 514         int ret;
 515         int pending_ret;
 516         u64 super_blocks_used;
 517         struct btrfs_fs_info *info = root->fs_info;
 518         struct btrfs_root *extent_root = info->extent_root;
 519         struct btrfs_extent_item extent_item;
 520         struct btrfs_key prealloc_key;
 521
 522         btrfs_set_extent_refs(&extent_item, 1);
 523         btrfs_set_extent_owner(&extent_item, owner);
 524
 525         if (root == extent_root) {
 526                 int nr;
 527                 BUG_ON(info->extent_tree_prealloc_nr == 0);
 528                 BUG_ON(num_blocks != 1);
 529                 ins->offset = 1;
 530                 info->extent_tree_prealloc_nr--;
 531                 nr = info->extent_tree_prealloc_nr;
 532                 ins->objectid = info->extent_tree_prealloc[nr];
 533                 info->extent_tree_insert[info->extent_tree_insert_nr++] =
 534                         ins->objectid;
 535                 return 0;
 536         }
 537         /* do the real allocation */
 538         ret = find_free_extent(trans, root, num_blocks, search_start,
 539                                search_end, ins);
 540         if (ret)
 541                 return ret;
 542
 543         /* then do prealloc for the extent tree */
 544         ret = find_free_extent(trans, root, 0, ins->objectid + ins->offset,
 545                                search_end, &prealloc_key);
 546         if (ret)
 547                 return ret;
 548
 549         super_blocks_used = btrfs_super_blocks_used(info->disk_super);
 550         btrfs_set_super_blocks_used(info->disk_super, super_blocks_used +
 551                                     num_blocks);
 552         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
 553                                 sizeof(extent_item));
 554
 555         finish_current_insert(trans, extent_root);
 556         pending_ret = del_pending_extents(trans, extent_root);
 557         if (ret)
 558                 return ret;
 559         if (pending_ret)
 560                 return pending_ret;
 561         return 0;
 562 }
 563
 564 /*
 565  * helper function to allocate a block for a given tree
 566  * returns the tree buffer or NULL.
 567  */
 568 struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 569                                            struct btrfs_root *root)
 570 {
 571         struct btrfs_key ins;
 572         int ret;
 573         struct buffer_head *buf;
 574
 575         ret = btrfs_alloc_extent(trans, root, root->root_key.objectid,
 576                                  1, 0, (unsigned long)-1, &ins);
 577         if (ret) {
 578                 BUG();
 579                 return NULL;
 580         }
 581         buf = btrfs_find_create_tree_block(root, ins.objectid);
 582         set_buffer_uptodate(buf);
 583         return buf;
 584 }
 585
 586 static int drop_leaf_ref(struct btrfs_trans_handle *trans,
 587                          struct btrfs_root *root, struct buffer_head *cur)
 588 {
 589         struct btrfs_disk_key *key;
 590         struct btrfs_leaf *leaf;
 591         struct btrfs_file_extent_item *fi;
 592         int i;
 593         int nritems;
 594         int ret;
 595
 596         BUG_ON(!btrfs_is_leaf(btrfs_buffer_node(cur)));
 597         leaf = btrfs_buffer_leaf(cur);
 598         nritems = btrfs_header_nritems(&leaf->header);
 599         for (i = 0; i < nritems; i++) {
 600                 key = &leaf->items[i].key;
 601                 if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY)
 602                         continue;
 603                 fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
 604                 if (btrfs_file_extent_type(fi) == BTRFS_FILE_EXTENT_INLINE)
 605                         continue;
 606                 /*
 607                  * FIXME make sure to insert a trans record that
 608                  * repeats the snapshot del on crash
 609                  */
 610                 ret = btrfs_free_extent(trans, root,
 611                                         btrfs_file_extent_disk_blocknr(fi),
 612                                         btrfs_file_extent_disk_num_blocks(fi),
 613                                         0);
 614                 BUG_ON(ret);
 615         }
 616         return 0;
 617 }
 618
 619 /*
 620  * helper function for drop_snapshot, this walks down the tree dropping ref
 621  * counts as it goes.
 622  */
 623 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
 624                           *root, struct btrfs_path *path, int *level)
 625 {
 626         struct buffer_head *next;
 627         struct buffer_head *cur;
 628         u64 blocknr;
 629         int ret;
 630         u32 refs;
 631
 632         WARN_ON(*level < 0);
 633         WARN_ON(*level >= BTRFS_MAX_LEVEL);
 634         ret = lookup_extent_ref(trans, root, bh_blocknr(path->nodes[*level]),
 635                                1, &refs);
 636         BUG_ON(ret);
 637         if (refs > 1)
 638                 goto out;
 639         /*
 640          * walk down to the last node level and free all the leaves
 641          */
 642         while(*level >= 0) {
 643                 WARN_ON(*level < 0);
 644                 WARN_ON(*level >= BTRFS_MAX_LEVEL);
 645                 cur = path->nodes[*level];
 646                 if (btrfs_header_level(btrfs_buffer_header(cur)) != *level)
 647                         WARN_ON(1);
 648                 if (path->slots[*level] >=
 649                     btrfs_header_nritems(btrfs_buffer_header(cur)))
 650                         break;
 651                 if (*level == 0) {
 652                         ret = drop_leaf_ref(trans, root, cur);
 653                         BUG_ON(ret);
 654                         break;
 655                 }
 656                 blocknr = btrfs_node_blockptr(btrfs_buffer_node(cur),
 657                                               path->slots[*level]);
 658                 ret = lookup_extent_ref(trans, root, blocknr, 1, &refs);
 659                 BUG_ON(ret);
 660                 if (refs != 1) {
 661                         path->slots[*level]++;
 662                         ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
 663                         BUG_ON(ret);
 664                         continue;
 665                 }
 666                 next = read_tree_block(root, blocknr);
 667                 WARN_ON(*level <= 0);
 668                 if (path->nodes[*level-1])
 669                         btrfs_block_release(root, path->nodes[*level-1]);
 670                 path->nodes[*level-1] = next;
 671                 *level = btrfs_header_level(btrfs_buffer_header(next));
 672                 path->slots[*level] = 0;
 673         }
 674 out:
 675         WARN_ON(*level < 0);
 676         WARN_ON(*level >= BTRFS_MAX_LEVEL);
 677         ret = btrfs_free_extent(trans, root,
 678                                 bh_blocknr(path->nodes[*level]), 1, 1);
 679         btrfs_block_release(root, path->nodes[*level]);
 680         path->nodes[*level] = NULL;
 681         *level += 1;
 682         BUG_ON(ret);
 683         return 0;
 684 }
 685
 686 /*
 687  * helper for dropping snapshots.  This walks back up the tree in the path
 688  * to find the first node higher up where we haven't yet gone through
 689  * all the slots
 690  */
 691 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
 692                         *root, struct btrfs_path *path, int *level)
 693 {
 694         int i;
 695         int slot;
 696         int ret;
 697         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
 698                 slot = path->slots[i];
 699                 if (slot < btrfs_header_nritems(
 700                     btrfs_buffer_header(path->nodes[i])) - 1) {
 701                         path->slots[i]++;
 702                         *level = i;
 703                         return 0;
 704                 } else {
 705                         ret = btrfs_free_extent(trans, root,
 706                                                 bh_blocknr(path->nodes[*level]),
 707                                                 1, 1);
 708                         BUG_ON(ret);
 709                         btrfs_block_release(root, path->nodes[*level]);
 710                         path->nodes[*level] = NULL;
 711                         *level = i + 1;
 712                 }
 713         }
 714         return 1;
 715 }
 716
 717 /*
 718  * drop the reference count on the tree rooted at 'snap'.  This traverses
 719  * the tree freeing any blocks that have a ref count of zero after being
 720  * decremented.
 721  */
 722 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
 723                         *root, struct buffer_head *snap)
 724 {
 725         int ret = 0;
 726         int wret;
 727         int level;
 728         struct btrfs_path *path;
 729         int i;
 730         int orig_level;
 731
 732         path = btrfs_alloc_path();
 733         BUG_ON(!path);
 734         btrfs_init_path(path);
 735
 736         level = btrfs_header_level(btrfs_buffer_header(snap));
 737         orig_level = level;
 738         path->nodes[level] = snap;
 739         path->slots[level] = 0;
 740         while(1) {
 741                 wret = walk_down_tree(trans, root, path, &level);
 742                 if (wret > 0)
 743                         break;
 744                 if (wret < 0)
 745                         ret = wret;
 746
 747                 wret = walk_up_tree(trans, root, path, &level);
 748                 if (wret > 0)
 749                         break;
 750                 if (wret < 0)
 751                         ret = wret;
 752         }
 753         for (i = 0; i <= orig_level; i++) {
 754                 if (path->nodes[i]) {
 755                         btrfs_block_release(root, path->nodes[i]);
 756                 }
 757         }
 758         btrfs_free_path(path);
 759         return ret;
 760 }