extent-tree.c

   1 #include <stdio.h>
   2 #include <stdlib.h>
   3 #include "kerncompat.h"
   4 #include "radix-tree.h"
   5 #include "ctree.h"
   6 #include "disk-io.h"
   7 #include "print-tree.h"
   8 #include "transaction.h"
   9
  10 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
  11                             *orig_root, u64 num_blocks, u64 search_start, u64
  12                             search_end, struct btrfs_key *ins);
  13 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
  14                                  btrfs_root *extent_root);
  15 static int run_pending(struct btrfs_trans_handle *trans, struct btrfs_root
  16                        *extent_root);
  17
  18 /*
  19  * pending extents are blocks that we're trying to allocate in the extent
  20  * map while trying to grow the map because of other allocations.  To avoid
  21  * recursing, they are tagged in the radix tree and cleaned up after
  22  * other allocations are done.  The pending tag is also used in the same
  23  * manner for deletes.
  24  */
  25 #define CTREE_EXTENT_PENDING_DEL 0
  26
  27 static int inc_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
  28                          *root, u64 blocknr)
  29 {
  30         struct btrfs_path path;
  31         int ret;
  32         struct btrfs_key key;
  33         struct btrfs_leaf *l;
  34         struct btrfs_extent_item *item;
  35         struct btrfs_key ins;
  36         u32 refs;
  37
  38         find_free_extent(trans, root->fs_info->extent_root, 0, 0, (u64)-1,
  39                          &ins);
  40         btrfs_init_path(&path);
  41         key.objectid = blocknr;
  42         key.flags = 0;
  43         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  44         key.offset = 1;
  45         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, &path,
  46                                 0, 1);
  47         if (ret != 0)
  48                 BUG();
  49         BUG_ON(ret != 0);
  50         l = &path.nodes[0]->leaf;
  51         item = btrfs_item_ptr(l, path.slots[0], struct btrfs_extent_item);
  52         refs = btrfs_extent_refs(item);
  53         btrfs_set_extent_refs(item, refs + 1);
  54
  55         BUG_ON(list_empty(&path.nodes[0]->dirty));
  56         btrfs_release_path(root->fs_info->extent_root, &path);
  57         finish_current_insert(trans, root->fs_info->extent_root);
  58         run_pending(trans, root->fs_info->extent_root);
  59         return 0;
  60 }
  61
  62 static int lookup_block_ref(struct btrfs_trans_handle *trans, struct btrfs_root
  63                             *root, u64 blocknr, u32 *refs)
  64 {
  65         struct btrfs_path path;
  66         int ret;
  67         struct btrfs_key key;
  68         struct btrfs_leaf *l;
  69         struct btrfs_extent_item *item;
  70         btrfs_init_path(&path);
  71         key.objectid = blocknr;
  72         key.offset = 1;
  73         key.flags = 0;
  74         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
  75         ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, &path,
  76                                 0, 0);
  77         if (ret != 0)
  78                 BUG();
  79         l = &path.nodes[0]->leaf;
  80         item = btrfs_item_ptr(l, path.slots[0], struct btrfs_extent_item);
  81         *refs = btrfs_extent_refs(item);
  82         btrfs_release_path(root->fs_info->extent_root, &path);
  83         return 0;
  84 }
  85
  86 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
  87                   struct btrfs_buffer *buf)
  88 {
  89         u64 blocknr;
  90         int i;
  91
  92         if (!root->ref_cows)
  93                 return 0;
  94         if (btrfs_is_leaf(&buf->node))
  95                 return 0;
  96
  97         for (i = 0; i < btrfs_header_nritems(&buf->node.header); i++) {
  98                 blocknr = btrfs_node_blockptr(&buf->node, i);
  99                 inc_block_ref(trans, root, blocknr);
 100         }
 101         return 0;
 102 }
 103
 104 int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, struct
 105                                btrfs_root *root)
 106 {
 107         unsigned long gang[8];
 108         u64 first = 0;
 109         int ret;
 110         int i;
 111
 112         while(1) {
 113                 ret = radix_tree_gang_lookup(&root->fs_info->pinned_radix,
 114                                              (void **)gang, 0,
 115                                              ARRAY_SIZE(gang));
 116                 if (!ret)
 117                         break;
 118                 if (!first)
 119                         first = gang[0];
 120                 for (i = 0; i < ret; i++) {
 121                         radix_tree_delete(&root->fs_info->pinned_radix,
 122                                           gang[i]);
 123                 }
 124         }
 125         root->fs_info->last_insert.objectid = first;
 126         root->fs_info->last_insert.offset = 0;
 127         return 0;
 128 }
 129
 130 static int finish_current_insert(struct btrfs_trans_handle *trans, struct
 131                                  btrfs_root *extent_root)
 132 {
 133         struct btrfs_key ins;
 134         struct btrfs_extent_item extent_item;
 135         int i;
 136         int ret;
 137         u64 super_blocks_used;
 138         struct btrfs_fs_info *info = extent_root->fs_info;
 139
 140         btrfs_set_extent_refs(&extent_item, 1);
 141         ins.offset = 1;
 142         ins.flags = 0;
 143         btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY);
 144
 145         for (i = 0; i < extent_root->fs_info->current_insert.flags; i++) {
 146                 ins.objectid = extent_root->fs_info->current_insert.objectid +
 147                                 i;
 148                 super_blocks_used = btrfs_super_blocks_used(info->disk_super);
 149                 btrfs_set_super_blocks_used(info->disk_super,
 150                                             super_blocks_used + 1);
 151                 ret = btrfs_insert_item(trans, extent_root, &ins, &extent_item,
 152                                         sizeof(extent_item));
 153                 BUG_ON(ret);
 154         }
 155         extent_root->fs_info->current_insert.offset = 0;
 156         return 0;
 157 }
 158
 159 /*
 160  * remove an extent from the root, returns 0 on success
 161  */
 162 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 163                          *root, u64 blocknr, u64 num_blocks, int pin)
 164 {
 165         struct btrfs_path path;
 166         struct btrfs_key key;
 167         struct btrfs_fs_info *info = root->fs_info;
 168         struct btrfs_root *extent_root = info->extent_root;
 169         int ret;
 170         struct btrfs_extent_item *ei;
 171         struct btrfs_key ins;
 172         u32 refs;
 173
 174         BUG_ON(pin && num_blocks != 1);
 175         key.objectid = blocknr;
 176         key.flags = 0;
 177         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 178         key.offset = num_blocks;
 179
 180         find_free_extent(trans, root, 0, 0, (u64)-1, &ins);
 181         btrfs_init_path(&path);
 182         ret = btrfs_search_slot(trans, extent_root, &key, &path, -1, 1);
 183         if (ret) {
 184                 printf("failed to find %Lu\n", key.objectid);
 185                 btrfs_print_tree(extent_root, extent_root->node);
 186                 printf("failed to find %Lu\n", key.objectid);
 187                 BUG();
 188         }
 189         ei = btrfs_item_ptr(&path.nodes[0]->leaf, path.slots[0],
 190                             struct btrfs_extent_item);
 191         BUG_ON(ei->refs == 0);
 192         refs = btrfs_extent_refs(ei) - 1;
 193         btrfs_set_extent_refs(ei, refs);
 194         if (refs == 0) {
 195                 u64 super_blocks_used;
 196                 if (pin) {
 197                         int err;
 198                         radix_tree_preload(GFP_KERNEL);
 199                         err = radix_tree_insert(&info->pinned_radix,
 200                                                 blocknr, (void *)blocknr);
 201                         BUG_ON(err);
 202                         radix_tree_preload_end();
 203                 }
 204                 super_blocks_used = btrfs_super_blocks_used(info->disk_super);
 205                 btrfs_set_super_blocks_used(info->disk_super,
 206                                             super_blocks_used - num_blocks);
 207                 ret = btrfs_del_item(trans, extent_root, &path);
 208                 if (!pin && extent_root->fs_info->last_insert.objectid >
 209                     blocknr)
 210                         extent_root->fs_info->last_insert.objectid = blocknr;
 211                 if (ret)
 212                         BUG();
 213         }
 214         btrfs_release_path(extent_root, &path);
 215         finish_current_insert(trans, extent_root);
 216         return ret;
 217 }
 218
 219 /*
 220  * find all the blocks marked as pending in the radix tree and remove
 221  * them from the extent map
 222  */
 223 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 224                                btrfs_root *extent_root)
 225 {
 226         int ret;
 227         struct btrfs_buffer *gang[4];
 228         int i;
 229
 230         while(1) {
 231                 ret = radix_tree_gang_lookup_tag(
 232                                         &extent_root->fs_info->cache_radix,
 233                                         (void **)gang, 0,
 234                                         ARRAY_SIZE(gang),
 235                                         CTREE_EXTENT_PENDING_DEL);
 236                 if (!ret)
 237                         break;
 238                 for (i = 0; i < ret; i++) {
 239                         ret = __free_extent(trans, extent_root,
 240                                             gang[i]->blocknr, 1, 1);
 241                         radix_tree_tag_clear(&extent_root->fs_info->cache_radix,
 242                                              gang[i]->blocknr,
 243                                              CTREE_EXTENT_PENDING_DEL);
 244                         btrfs_block_release(extent_root, gang[i]);
 245                 }
 246         }
 247         return 0;
 248 }
 249
 250 static int run_pending(struct btrfs_trans_handle *trans, struct btrfs_root
 251                        *extent_root)
 252 {
 253         while(radix_tree_tagged(&extent_root->fs_info->cache_radix,
 254                                 CTREE_EXTENT_PENDING_DEL))
 255                 del_pending_extents(trans, extent_root);
 256         return 0;
 257 }
 258
 259
 260 /*
 261  * remove an extent from the root, returns 0 on success
 262  */
 263 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 264                       *root, u64 blocknr, u64 num_blocks, int pin)
 265 {
 266         struct btrfs_root *extent_root = root->fs_info->extent_root;
 267         struct btrfs_buffer *t;
 268         int pending_ret;
 269         int ret;
 270
 271         if (root == extent_root) {
 272                 t = find_tree_block(root, blocknr);
 273                 radix_tree_tag_set(&root->fs_info->cache_radix, blocknr,
 274                                    CTREE_EXTENT_PENDING_DEL);
 275                 return 0;
 276         }
 277         ret = __free_extent(trans, root, blocknr, num_blocks, pin);
 278         pending_ret = run_pending(trans, root->fs_info->extent_root);
 279         return ret ? ret : pending_ret;
 280 }
 281
 282 /*
 283  * walks the btree of allocated extents and find a hole of a given size.
 284  * The key ins is changed to record the hole:
 285  * ins->objectid == block start
 286  * ins->flags = BTRFS_EXTENT_ITEM_KEY
 287  * ins->offset == number of blocks
 288  * Any available blocks before search_start are skipped.
 289  */
 290 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 291                             *orig_root, u64 num_blocks, u64 search_start, u64
 292                             search_end, struct btrfs_key *ins)
 293 {
 294         struct btrfs_path path;
 295         struct btrfs_key key;
 296         int ret;
 297         u64 hole_size = 0;
 298         int slot = 0;
 299         u64 last_block;
 300         u64 test_block;
 301         int start_found;
 302         struct btrfs_leaf *l;
 303         struct btrfs_root * root = orig_root->fs_info->extent_root;
 304         int total_needed = num_blocks;
 305
 306         total_needed += (btrfs_header_level(&root->node->node.header) + 1) * 3;
 307         if (root->fs_info->last_insert.objectid > search_start)
 308                 search_start = root->fs_info->last_insert.objectid;
 309
 310         ins->flags = 0;
 311         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
 312
 313 check_failed:
 314         btrfs_init_path(&path);
 315         ins->objectid = search_start;
 316         ins->offset = 0;
 317         start_found = 0;
 318         ret = btrfs_search_slot(trans, root, ins, &path, 0, 0);
 319         if (ret < 0)
 320                 goto error;
 321
 322         if (path.slots[0] > 0)
 323                 path.slots[0]--;
 324
 325         while (1) {
 326                 l = &path.nodes[0]->leaf;
 327                 slot = path.slots[0];
 328                 if (slot >= btrfs_header_nritems(&l->header)) {
 329                         ret = btrfs_next_leaf(root, &path);
 330                         if (ret == 0)
 331                                 continue;
 332                         if (ret < 0)
 333                                 goto error;
 334                         if (!start_found) {
 335                                 ins->objectid = search_start;
 336                                 ins->offset = (u64)-1;
 337                                 start_found = 1;
 338                                 goto check_pending;
 339                         }
 340                         ins->objectid = last_block > search_start ?
 341                                         last_block : search_start;
 342                         ins->offset = (u64)-1;
 343                         goto check_pending;
 344                 }
 345                 btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
 346                 if (key.objectid >= search_start) {
 347                         if (start_found) {
 348                                 if (last_block < search_start)
 349                                         last_block = search_start;
 350                                 hole_size = key.objectid - last_block;
 351                                 if (hole_size > total_needed) {
 352                                         ins->objectid = last_block;
 353                                         ins->offset = hole_size;
 354                                         goto check_pending;
 355                                 }
 356                         }
 357                 }
 358                 start_found = 1;
 359                 last_block = key.objectid + key.offset;
 360                 path.slots[0]++;
 361         }
 362         // FIXME -ENOSPC
 363 check_pending:
 364         /* we have to make sure we didn't find an extent that has already
 365          * been allocated by the map tree or the original allocation
 366          */
 367         btrfs_release_path(root, &path);
 368         BUG_ON(ins->objectid < search_start);
 369         for (test_block = ins->objectid;
 370              test_block < ins->objectid + total_needed; test_block++) {
 371                 if (radix_tree_lookup(&root->fs_info->pinned_radix,
 372                                       test_block)) {
 373                         search_start = test_block + 1;
 374                         goto check_failed;
 375                 }
 376         }
 377         BUG_ON(root->fs_info->current_insert.offset);
 378         root->fs_info->current_insert.offset = total_needed - num_blocks;
 379         root->fs_info->current_insert.objectid = ins->objectid + num_blocks;
 380         root->fs_info->current_insert.flags = 0;
 381         root->fs_info->last_insert.objectid = ins->objectid;
 382         ins->offset = num_blocks;
 383         return 0;
 384 error:
 385         btrfs_release_path(root, &path);
 386         return ret;
 387 }
 388
 389 /*
 390  * finds a free extent and does all the dirty work required for allocation
 391  * returns the key for the extent through ins, and a tree buffer for
 392  * the first block of the extent through buf.
 393  *
 394  * returns 0 if everything worked, non-zero otherwise.
 395  */
 396 static int alloc_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 397                         *root, u64 owner, u8 type, u64 num_blocks,
 398                         u64 search_start, u64
 399                         search_end, struct btrfs_key *ins)
 400 {
 401         int ret;
 402         int pending_ret;
 403         u64 super_blocks_used;
 404         struct btrfs_fs_info *info = root->fs_info;
 405         struct btrfs_root *extent_root = info->extent_root;
 406         struct btrfs_extent_item extent_item;
 407
 408         btrfs_set_extent_refs(&extent_item, 1);
 409         btrfs_set_extent_owner(&extent_item, owner);
 410         btrfs_set_extent_type(&extent_item, type);
 411
 412         if (root == extent_root) {
 413                 BUG_ON(extent_root->fs_info->current_insert.offset == 0);
 414                 BUG_ON(num_blocks != 1);
 415                 BUG_ON(extent_root->fs_info->current_insert.flags ==
 416                        extent_root->fs_info->current_insert.offset);
 417                 ins->offset = 1;
 418                 ins->objectid = extent_root->fs_info->current_insert.objectid +
 419                                 extent_root->fs_info->current_insert.flags++;
 420                 return 0;
 421         }
 422         ret = find_free_extent(trans, root, num_blocks, search_start,
 423                                search_end, ins);
 424         if (ret)
 425                 return ret;
 426
 427         super_blocks_used = btrfs_super_blocks_used(info->disk_super);
 428         btrfs_set_super_blocks_used(info->disk_super, super_blocks_used +
 429                                     num_blocks);
 430         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
 431                                 sizeof(extent_item));
 432
 433         finish_current_insert(trans, extent_root);
 434         pending_ret = run_pending(trans, extent_root);
 435         if (ret)
 436                 return ret;
 437         if (pending_ret)
 438                 return pending_ret;
 439         return 0;
 440 }
 441
 442 /*
 443  * helper function to allocate a block for a given tree
 444  * returns the tree buffer or NULL.
 445  */
 446 struct btrfs_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 447                                             struct btrfs_root *root)
 448 {
 449         struct btrfs_key ins;
 450         int ret;
 451         struct btrfs_buffer *buf;
 452
 453         ret = alloc_extent(trans, root, root->root_key.objectid,
 454                            BTRFS_EXTENT_TREE, 1, 0, (unsigned long)-1, &ins);
 455         if (ret) {
 456                 BUG();
 457                 return NULL;
 458         }
 459         buf = find_tree_block(root, ins.objectid);
 460         dirty_tree_block(trans, root, buf);
 461         btrfs_set_header_generation(&buf->node.header,
 462                                     root->root_key.offset + 1);
 463         memcpy(buf->node.header.fsid, root->fs_info->disk_super->fsid,
 464                sizeof(buf->node.header.fsid));
 465         return buf;
 466 }
 467
 468 /*
 469  * helper function for drop_snapshot, this walks down the tree dropping ref
 470  * counts as it goes.
 471  */
 472 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
 473                           *root, struct btrfs_path *path, int *level)
 474 {
 475         struct btrfs_buffer *next;
 476         struct btrfs_buffer *cur;
 477         u64 blocknr;
 478         int ret;
 479         u32 refs;
 480
 481         ret = lookup_block_ref(trans, root, path->nodes[*level]->blocknr,
 482                                &refs);
 483         BUG_ON(ret);
 484         if (refs > 1)
 485                 goto out;
 486         /*
 487          * walk down to the last node level and free all the leaves
 488          */
 489         while(*level > 0) {
 490                 cur = path->nodes[*level];
 491                 if (path->slots[*level] >=
 492                     btrfs_header_nritems(&cur->node.header))
 493                         break;
 494                 blocknr = btrfs_node_blockptr(&cur->node, path->slots[*level]);
 495                 ret = lookup_block_ref(trans, root, blocknr, &refs);
 496                 if (refs != 1 || *level == 1) {
 497                         path->slots[*level]++;
 498                         ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
 499                         BUG_ON(ret);
 500                         continue;
 501                 }
 502                 BUG_ON(ret);
 503                 next = read_tree_block(root, blocknr);
 504                 if (path->nodes[*level-1])
 505                         btrfs_block_release(root, path->nodes[*level-1]);
 506                 path->nodes[*level-1] = next;
 507                 *level = btrfs_header_level(&next->node.header);
 508                 path->slots[*level] = 0;
 509         }
 510 out:
 511         ret = btrfs_free_extent(trans, root, path->nodes[*level]->blocknr, 1,
 512                                 1);
 513         btrfs_block_release(root, path->nodes[*level]);
 514         path->nodes[*level] = NULL;
 515         *level += 1;
 516         BUG_ON(ret);
 517         return 0;
 518 }
 519
 520 /*
 521  * helper for dropping snapshots.  This walks back up the tree in the path
 522  * to find the first node higher up where we haven't yet gone through
 523  * all the slots
 524  */
 525 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
 526                         *root, struct btrfs_path *path, int *level)
 527 {
 528         int i;
 529         int slot;
 530         int ret;
 531         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
 532                 slot = path->slots[i];
 533                 if (slot <
 534                     btrfs_header_nritems(&path->nodes[i]->node.header)- 1) {
 535                         path->slots[i]++;
 536                         *level = i;
 537                         return 0;
 538                 } else {
 539                         ret = btrfs_free_extent(trans, root,
 540                                                 path->nodes[*level]->blocknr,
 541                                                 1, 1);
 542                         btrfs_block_release(root, path->nodes[*level]);
 543                         path->nodes[*level] = NULL;
 544                         *level = i + 1;
 545                         BUG_ON(ret);
 546                 }
 547         }
 548         return 1;
 549 }
 550
 551 /*
 552  * drop the reference count on the tree rooted at 'snap'.  This traverses
 553  * the tree freeing any blocks that have a ref count of zero after being
 554  * decremented.
 555  */
 556 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
 557                         *root, struct btrfs_buffer *snap)
 558 {
 559         int ret = 0;
 560         int wret;
 561         int level;
 562         struct btrfs_path path;
 563         int i;
 564         int orig_level;
 565
 566         btrfs_init_path(&path);
 567
 568         level = btrfs_header_level(&snap->node.header);
 569         orig_level = level;
 570         path.nodes[level] = snap;
 571         path.slots[level] = 0;
 572         while(1) {
 573                 wret = walk_down_tree(trans, root, &path, &level);
 574                 if (wret > 0)
 575                         break;
 576                 if (wret < 0)
 577                         ret = wret;
 578
 579                 wret = walk_up_tree(trans, root, &path, &level);
 580                 if (wret > 0)
 581                         break;
 582                 if (wret < 0)
 583                         ret = wret;
 584         }
 585         for (i = 0; i <= orig_level; i++) {
 586                 if (path.nodes[i]) {
 587                         btrfs_block_release(root, path.nodes[i]);
 588                 }
 589         }
 590         return ret;
 591 }