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
 138         btrfs_set_extent_refs(&extent_item, 1);
 139         btrfs_set_extent_owner(&extent_item,
 140                 btrfs_header_parentid(&extent_root->node->node.header));
 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                 ret = btrfs_insert_item(trans, extent_root, &ins, &extent_item,
 149                                         sizeof(extent_item));
 150                 BUG_ON(ret);
 151         }
 152         extent_root->fs_info->current_insert.offset = 0;
 153         return 0;
 154 }
 155
 156 /*
 157  * remove an extent from the root, returns 0 on success
 158  */
 159 static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 160                          *root, u64 blocknr, u64 num_blocks, int pin)
 161 {
 162         struct btrfs_path path;
 163         struct btrfs_key key;
 164         struct btrfs_root *extent_root = root->fs_info->extent_root;
 165         int ret;
 166         struct btrfs_extent_item *ei;
 167         struct btrfs_key ins;
 168         u32 refs;
 169
 170         BUG_ON(pin && num_blocks != 1);
 171         key.objectid = blocknr;
 172         key.flags = 0;
 173         btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
 174         key.offset = num_blocks;
 175
 176         find_free_extent(trans, root, 0, 0, (u64)-1, &ins);
 177         btrfs_init_path(&path);
 178         ret = btrfs_search_slot(trans, extent_root, &key, &path, -1, 1);
 179         if (ret) {
 180                 printf("failed to find %Lu\n", key.objectid);
 181                 btrfs_print_tree(extent_root, extent_root->node);
 182                 printf("failed to find %Lu\n", key.objectid);
 183                 BUG();
 184         }
 185         ei = btrfs_item_ptr(&path.nodes[0]->leaf, path.slots[0],
 186                             struct btrfs_extent_item);
 187         BUG_ON(ei->refs == 0);
 188         refs = btrfs_extent_refs(ei) - 1;
 189         btrfs_set_extent_refs(ei, refs);
 190         if (refs == 0) {
 191                 if (pin) {
 192                         int err;
 193                         radix_tree_preload(GFP_KERNEL);
 194                         err = radix_tree_insert(
 195                                         &extent_root->fs_info->pinned_radix,
 196                                         blocknr, (void *)blocknr);
 197                         BUG_ON(err);
 198                         radix_tree_preload_end();
 199                 }
 200                 ret = btrfs_del_item(trans, extent_root, &path);
 201                 if (!pin && extent_root->fs_info->last_insert.objectid >
 202                     blocknr)
 203                         extent_root->fs_info->last_insert.objectid = blocknr;
 204                 if (ret)
 205                         BUG();
 206         }
 207         btrfs_release_path(extent_root, &path);
 208         finish_current_insert(trans, extent_root);
 209         return ret;
 210 }
 211
 212 /*
 213  * find all the blocks marked as pending in the radix tree and remove
 214  * them from the extent map
 215  */
 216 static int del_pending_extents(struct btrfs_trans_handle *trans, struct
 217                                btrfs_root *extent_root)
 218 {
 219         int ret;
 220         struct btrfs_buffer *gang[4];
 221         int i;
 222
 223         while(1) {
 224                 ret = radix_tree_gang_lookup_tag(
 225                                         &extent_root->fs_info->cache_radix,
 226                                         (void **)gang, 0,
 227                                         ARRAY_SIZE(gang),
 228                                         CTREE_EXTENT_PENDING_DEL);
 229                 if (!ret)
 230                         break;
 231                 for (i = 0; i < ret; i++) {
 232                         ret = __free_extent(trans, extent_root,
 233                                             gang[i]->blocknr, 1, 1);
 234                         radix_tree_tag_clear(&extent_root->fs_info->cache_radix,
 235                                              gang[i]->blocknr,
 236                                              CTREE_EXTENT_PENDING_DEL);
 237                         btrfs_block_release(extent_root, gang[i]);
 238                 }
 239         }
 240         return 0;
 241 }
 242
 243 static int run_pending(struct btrfs_trans_handle *trans, struct btrfs_root
 244                        *extent_root)
 245 {
 246         while(radix_tree_tagged(&extent_root->fs_info->cache_radix,
 247                                 CTREE_EXTENT_PENDING_DEL))
 248                 del_pending_extents(trans, extent_root);
 249         return 0;
 250 }
 251
 252
 253 /*
 254  * remove an extent from the root, returns 0 on success
 255  */
 256 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 257                       *root, u64 blocknr, u64 num_blocks, int pin)
 258 {
 259         struct btrfs_root *extent_root = root->fs_info->extent_root;
 260         struct btrfs_buffer *t;
 261         int pending_ret;
 262         int ret;
 263
 264         if (root == extent_root) {
 265                 t = find_tree_block(root, blocknr);
 266                 radix_tree_tag_set(&root->fs_info->cache_radix, blocknr,
 267                                    CTREE_EXTENT_PENDING_DEL);
 268                 return 0;
 269         }
 270         ret = __free_extent(trans, root, blocknr, num_blocks, pin);
 271         pending_ret = run_pending(trans, root->fs_info->extent_root);
 272         return ret ? ret : pending_ret;
 273 }
 274
 275 /*
 276  * walks the btree of allocated extents and find a hole of a given size.
 277  * The key ins is changed to record the hole:
 278  * ins->objectid == block start
 279  * ins->flags = BTRFS_EXTENT_ITEM_KEY
 280  * ins->offset == number of blocks
 281  * Any available blocks before search_start are skipped.
 282  */
 283 static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 284                             *orig_root, u64 num_blocks, u64 search_start, u64
 285                             search_end, struct btrfs_key *ins)
 286 {
 287         struct btrfs_path path;
 288         struct btrfs_key key;
 289         int ret;
 290         u64 hole_size = 0;
 291         int slot = 0;
 292         u64 last_block;
 293         u64 test_block;
 294         int start_found;
 295         struct btrfs_leaf *l;
 296         struct btrfs_root * root = orig_root->fs_info->extent_root;
 297         int total_needed = num_blocks;
 298
 299         total_needed += (btrfs_header_level(&root->node->node.header) + 1) * 3;
 300         if (root->fs_info->last_insert.objectid > search_start)
 301                 search_start = root->fs_info->last_insert.objectid;
 302
 303         ins->flags = 0;
 304         btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
 305
 306 check_failed:
 307         btrfs_init_path(&path);
 308         ins->objectid = search_start;
 309         ins->offset = 0;
 310         start_found = 0;
 311         ret = btrfs_search_slot(trans, root, ins, &path, 0, 0);
 312         if (ret < 0)
 313                 goto error;
 314
 315         if (path.slots[0] > 0)
 316                 path.slots[0]--;
 317
 318         while (1) {
 319                 l = &path.nodes[0]->leaf;
 320                 slot = path.slots[0];
 321                 if (slot >= btrfs_header_nritems(&l->header)) {
 322                         ret = btrfs_next_leaf(root, &path);
 323                         if (ret == 0)
 324                                 continue;
 325                         if (ret < 0)
 326                                 goto error;
 327                         if (!start_found) {
 328                                 ins->objectid = search_start;
 329                                 ins->offset = (u64)-1;
 330                                 start_found = 1;
 331                                 goto check_pending;
 332                         }
 333                         ins->objectid = last_block > search_start ?
 334                                         last_block : search_start;
 335                         ins->offset = (u64)-1;
 336                         goto check_pending;
 337                 }
 338                 btrfs_disk_key_to_cpu(&key, &l->items[slot].key);
 339                 if (key.objectid >= search_start) {
 340                         if (start_found) {
 341                                 if (last_block < search_start)
 342                                         last_block = search_start;
 343                                 hole_size = key.objectid - last_block;
 344                                 if (hole_size > total_needed) {
 345                                         ins->objectid = last_block;
 346                                         ins->offset = hole_size;
 347                                         goto check_pending;
 348                                 }
 349                         }
 350                 }
 351                 start_found = 1;
 352                 last_block = key.objectid + key.offset;
 353                 path.slots[0]++;
 354         }
 355         // FIXME -ENOSPC
 356 check_pending:
 357         /* we have to make sure we didn't find an extent that has already
 358          * been allocated by the map tree or the original allocation
 359          */
 360         btrfs_release_path(root, &path);
 361         BUG_ON(ins->objectid < search_start);
 362         for (test_block = ins->objectid;
 363              test_block < ins->objectid + total_needed; test_block++) {
 364                 if (radix_tree_lookup(&root->fs_info->pinned_radix,
 365                                       test_block)) {
 366                         search_start = test_block + 1;
 367                         goto check_failed;
 368                 }
 369         }
 370         BUG_ON(root->fs_info->current_insert.offset);
 371         root->fs_info->current_insert.offset = total_needed - num_blocks;
 372         root->fs_info->current_insert.objectid = ins->objectid + num_blocks;
 373         root->fs_info->current_insert.flags = 0;
 374         root->fs_info->last_insert.objectid = ins->objectid;
 375         ins->offset = num_blocks;
 376         return 0;
 377 error:
 378         btrfs_release_path(root, &path);
 379         return ret;
 380 }
 381
 382 /*
 383  * finds a free extent and does all the dirty work required for allocation
 384  * returns the key for the extent through ins, and a tree buffer for
 385  * the first block of the extent through buf.
 386  *
 387  * returns 0 if everything worked, non-zero otherwise.
 388  */
 389 static int alloc_extent(struct btrfs_trans_handle *trans, struct btrfs_root
 390                         *root, u64 num_blocks, u64 search_start, u64
 391                         search_end, u64 owner, struct btrfs_key *ins)
 392 {
 393         int ret;
 394         int pending_ret;
 395         struct btrfs_root *extent_root = root->fs_info->extent_root;
 396         struct btrfs_extent_item extent_item;
 397
 398         btrfs_set_extent_refs(&extent_item, 1);
 399         btrfs_set_extent_owner(&extent_item, owner);
 400
 401         if (root == extent_root) {
 402                 BUG_ON(extent_root->fs_info->current_insert.offset == 0);
 403                 BUG_ON(num_blocks != 1);
 404                 BUG_ON(extent_root->fs_info->current_insert.flags ==
 405                        extent_root->fs_info->current_insert.offset);
 406                 ins->offset = 1;
 407                 ins->objectid = extent_root->fs_info->current_insert.objectid +
 408                                 extent_root->fs_info->current_insert.flags++;
 409                 return 0;
 410         }
 411         ret = find_free_extent(trans, root, num_blocks, search_start,
 412                                search_end, ins);
 413         if (ret)
 414                 return ret;
 415
 416         ret = btrfs_insert_item(trans, extent_root, ins, &extent_item,
 417                                 sizeof(extent_item));
 418
 419         finish_current_insert(trans, extent_root);
 420         pending_ret = run_pending(trans, extent_root);
 421         if (ret)
 422                 return ret;
 423         if (pending_ret)
 424                 return pending_ret;
 425         return 0;
 426 }
 427
 428 /*
 429  * helper function to allocate a block for a given tree
 430  * returns the tree buffer or NULL.
 431  */
 432 struct btrfs_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 433                                             struct btrfs_root *root)
 434 {
 435         struct btrfs_key ins;
 436         int ret;
 437         struct btrfs_buffer *buf;
 438
 439         ret = alloc_extent(trans, root, 1, 0, (unsigned long)-1,
 440                            btrfs_header_parentid(&root->node->node.header),
 441                            &ins);
 442         if (ret) {
 443                 BUG();
 444                 return NULL;
 445         }
 446         buf = find_tree_block(root, ins.objectid);
 447         dirty_tree_block(trans, root, buf);
 448         return buf;
 449 }
 450
 451 /*
 452  * helper function for drop_snapshot, this walks down the tree dropping ref
 453  * counts as it goes.
 454  */
 455 static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root
 456                           *root, struct btrfs_path *path, int *level)
 457 {
 458         struct btrfs_buffer *next;
 459         struct btrfs_buffer *cur;
 460         u64 blocknr;
 461         int ret;
 462         u32 refs;
 463
 464         ret = lookup_block_ref(trans, root, path->nodes[*level]->blocknr,
 465                                &refs);
 466         BUG_ON(ret);
 467         if (refs > 1)
 468                 goto out;
 469         /*
 470          * walk down to the last node level and free all the leaves
 471          */
 472         while(*level > 0) {
 473                 cur = path->nodes[*level];
 474                 if (path->slots[*level] >=
 475                     btrfs_header_nritems(&cur->node.header))
 476                         break;
 477                 blocknr = btrfs_node_blockptr(&cur->node, path->slots[*level]);
 478                 ret = lookup_block_ref(trans, root, blocknr, &refs);
 479                 if (refs != 1 || *level == 1) {
 480                         path->slots[*level]++;
 481                         ret = btrfs_free_extent(trans, root, blocknr, 1, 1);
 482                         BUG_ON(ret);
 483                         continue;
 484                 }
 485                 BUG_ON(ret);
 486                 next = read_tree_block(root, blocknr);
 487                 if (path->nodes[*level-1])
 488                         btrfs_block_release(root, path->nodes[*level-1]);
 489                 path->nodes[*level-1] = next;
 490                 *level = btrfs_header_level(&next->node.header);
 491                 path->slots[*level] = 0;
 492         }
 493 out:
 494         ret = btrfs_free_extent(trans, root, path->nodes[*level]->blocknr, 1,
 495                                 1);
 496         btrfs_block_release(root, path->nodes[*level]);
 497         path->nodes[*level] = NULL;
 498         *level += 1;
 499         BUG_ON(ret);
 500         return 0;
 501 }
 502
 503 /*
 504  * helper for dropping snapshots.  This walks back up the tree in the path
 505  * to find the first node higher up where we haven't yet gone through
 506  * all the slots
 507  */
 508 static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root
 509                         *root, struct btrfs_path *path, int *level)
 510 {
 511         int i;
 512         int slot;
 513         int ret;
 514         for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) {
 515                 slot = path->slots[i];
 516                 if (slot <
 517                     btrfs_header_nritems(&path->nodes[i]->node.header)- 1) {
 518                         path->slots[i]++;
 519                         *level = i;
 520                         return 0;
 521                 } else {
 522                         ret = btrfs_free_extent(trans, root,
 523                                                 path->nodes[*level]->blocknr,
 524                                                 1, 1);
 525                         btrfs_block_release(root, path->nodes[*level]);
 526                         path->nodes[*level] = NULL;
 527                         *level = i + 1;
 528                         BUG_ON(ret);
 529                 }
 530         }
 531         return 1;
 532 }
 533
 534 /*
 535  * drop the reference count on the tree rooted at 'snap'.  This traverses
 536  * the tree freeing any blocks that have a ref count of zero after being
 537  * decremented.
 538  */
 539 int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
 540                         *root, struct btrfs_buffer *snap)
 541 {
 542         int ret = 0;
 543         int wret;
 544         int level;
 545         struct btrfs_path path;
 546         int i;
 547         int orig_level;
 548
 549         btrfs_init_path(&path);
 550
 551         level = btrfs_header_level(&snap->node.header);
 552         orig_level = level;
 553         path.nodes[level] = snap;
 554         path.slots[level] = 0;
 555         while(1) {
 556                 wret = walk_down_tree(trans, root, &path, &level);
 557                 if (wret > 0)
 558                         break;
 559                 if (wret < 0)
 560                         ret = wret;
 561
 562                 wret = walk_up_tree(trans, root, &path, &level);
 563                 if (wret > 0)
 564                         break;
 565                 if (wret < 0)
 566                         ret = wret;
 567         }
 568         for (i = 0; i <= orig_level; i++) {
 569                 if (path.nodes[i]) {
 570                         btrfs_block_release(root, path.nodes[i]);
 571                 }
 572         }
 573         return ret;
 574 }