1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Copyright (C) 2009 Oracle. All rights reserved. 4 */ 5 6 #include <linux/sched.h> 7 #include <linux/slab.h> 8 #include <linux/sort.h> 9 #include "messages.h" 10 #include "ctree.h" 11 #include "delayed-ref.h" 12 #include "transaction.h" 13 #include "qgroup.h" 14 #include "space-info.h" 15 #include "tree-mod-log.h" 16 #include "fs.h" 17 18 struct kmem_cache *btrfs_delayed_ref_head_cachep; 19 struct kmem_cache *btrfs_delayed_tree_ref_cachep; 20 struct kmem_cache *btrfs_delayed_data_ref_cachep; 21 struct kmem_cache *btrfs_delayed_extent_op_cachep; 22 /* 23 * delayed back reference update tracking. For subvolume trees 24 * we queue up extent allocations and backref maintenance for 25 * delayed processing. This avoids deep call chains where we 26 * add extents in the middle of btrfs_search_slot, and it allows 27 * us to buffer up frequently modified backrefs in an rb tree instead 28 * of hammering updates on the extent allocation tree. 29 */ 30 31 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info) 32 { 33 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 34 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; 35 bool ret = false; 36 u64 reserved; 37 38 spin_lock(&global_rsv->lock); 39 reserved = global_rsv->reserved; 40 spin_unlock(&global_rsv->lock); 41 42 /* 43 * Since the global reserve is just kind of magic we don't really want 44 * to rely on it to save our bacon, so if our size is more than the 45 * delayed_refs_rsv and the global rsv then it's time to think about 46 * bailing. 47 */ 48 spin_lock(&delayed_refs_rsv->lock); 49 reserved += delayed_refs_rsv->reserved; 50 if (delayed_refs_rsv->size >= reserved) 51 ret = true; 52 spin_unlock(&delayed_refs_rsv->lock); 53 return ret; 54 } 55 56 /* 57 * Release a ref head's reservation. 58 * 59 * @fs_info: the filesystem 60 * @nr: number of items to drop 61 * 62 * Drops the delayed ref head's count from the delayed refs rsv and free any 63 * excess reservation we had. 64 */ 65 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr) 66 { 67 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; 68 const u64 num_bytes = btrfs_calc_delayed_ref_bytes(fs_info, nr); 69 u64 released = 0; 70 71 released = btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL); 72 if (released) 73 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", 74 0, released, 0); 75 } 76 77 /* 78 * Adjust the size of the delayed refs rsv. 79 * 80 * This is to be called anytime we may have adjusted trans->delayed_ref_updates, 81 * it'll calculate the additional size and add it to the delayed_refs_rsv. 82 */ 83 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans) 84 { 85 struct btrfs_fs_info *fs_info = trans->fs_info; 86 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; 87 u64 num_bytes; 88 89 if (!trans->delayed_ref_updates) 90 return; 91 92 num_bytes = btrfs_calc_delayed_ref_bytes(fs_info, 93 trans->delayed_ref_updates); 94 95 spin_lock(&delayed_rsv->lock); 96 delayed_rsv->size += num_bytes; 97 delayed_rsv->full = false; 98 spin_unlock(&delayed_rsv->lock); 99 trans->delayed_ref_updates = 0; 100 } 101 102 /* 103 * Transfer bytes to our delayed refs rsv. 104 * 105 * @fs_info: the filesystem 106 * @src: source block rsv to transfer from 107 * @num_bytes: number of bytes to transfer 108 * 109 * This transfers up to the num_bytes amount from the src rsv to the 110 * delayed_refs_rsv. Any extra bytes are returned to the space info. 111 */ 112 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info, 113 struct btrfs_block_rsv *src, 114 u64 num_bytes) 115 { 116 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 117 u64 to_free = 0; 118 119 spin_lock(&src->lock); 120 src->reserved -= num_bytes; 121 src->size -= num_bytes; 122 spin_unlock(&src->lock); 123 124 spin_lock(&delayed_refs_rsv->lock); 125 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) { 126 u64 delta = delayed_refs_rsv->size - 127 delayed_refs_rsv->reserved; 128 if (num_bytes > delta) { 129 to_free = num_bytes - delta; 130 num_bytes = delta; 131 } 132 } else { 133 to_free = num_bytes; 134 num_bytes = 0; 135 } 136 137 if (num_bytes) 138 delayed_refs_rsv->reserved += num_bytes; 139 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size) 140 delayed_refs_rsv->full = true; 141 spin_unlock(&delayed_refs_rsv->lock); 142 143 if (num_bytes) 144 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", 145 0, num_bytes, 1); 146 if (to_free) 147 btrfs_space_info_free_bytes_may_use(fs_info, 148 delayed_refs_rsv->space_info, to_free); 149 } 150 151 /* 152 * Refill based on our delayed refs usage. 153 * 154 * @fs_info: the filesystem 155 * @flush: control how we can flush for this reservation. 156 * 157 * This will refill the delayed block_rsv up to 1 items size worth of space and 158 * will return -ENOSPC if we can't make the reservation. 159 */ 160 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info, 161 enum btrfs_reserve_flush_enum flush) 162 { 163 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; 164 u64 limit = btrfs_calc_delayed_ref_bytes(fs_info, 1); 165 u64 num_bytes = 0; 166 int ret = -ENOSPC; 167 168 spin_lock(&block_rsv->lock); 169 if (block_rsv->reserved < block_rsv->size) { 170 num_bytes = block_rsv->size - block_rsv->reserved; 171 num_bytes = min(num_bytes, limit); 172 } 173 spin_unlock(&block_rsv->lock); 174 175 if (!num_bytes) 176 return 0; 177 178 ret = btrfs_reserve_metadata_bytes(fs_info, block_rsv, num_bytes, flush); 179 if (ret) 180 return ret; 181 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, false); 182 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", 183 0, num_bytes, 1); 184 return 0; 185 } 186 187 /* 188 * compare two delayed tree backrefs with same bytenr and type 189 */ 190 static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref1, 191 struct btrfs_delayed_tree_ref *ref2) 192 { 193 if (ref1->node.type == BTRFS_TREE_BLOCK_REF_KEY) { 194 if (ref1->root < ref2->root) 195 return -1; 196 if (ref1->root > ref2->root) 197 return 1; 198 } else { 199 if (ref1->parent < ref2->parent) 200 return -1; 201 if (ref1->parent > ref2->parent) 202 return 1; 203 } 204 return 0; 205 } 206 207 /* 208 * compare two delayed data backrefs with same bytenr and type 209 */ 210 static int comp_data_refs(struct btrfs_delayed_data_ref *ref1, 211 struct btrfs_delayed_data_ref *ref2) 212 { 213 if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) { 214 if (ref1->root < ref2->root) 215 return -1; 216 if (ref1->root > ref2->root) 217 return 1; 218 if (ref1->objectid < ref2->objectid) 219 return -1; 220 if (ref1->objectid > ref2->objectid) 221 return 1; 222 if (ref1->offset < ref2->offset) 223 return -1; 224 if (ref1->offset > ref2->offset) 225 return 1; 226 } else { 227 if (ref1->parent < ref2->parent) 228 return -1; 229 if (ref1->parent > ref2->parent) 230 return 1; 231 } 232 return 0; 233 } 234 235 static int comp_refs(struct btrfs_delayed_ref_node *ref1, 236 struct btrfs_delayed_ref_node *ref2, 237 bool check_seq) 238 { 239 int ret = 0; 240 241 if (ref1->type < ref2->type) 242 return -1; 243 if (ref1->type > ref2->type) 244 return 1; 245 if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY || 246 ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) 247 ret = comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref1), 248 btrfs_delayed_node_to_tree_ref(ref2)); 249 else 250 ret = comp_data_refs(btrfs_delayed_node_to_data_ref(ref1), 251 btrfs_delayed_node_to_data_ref(ref2)); 252 if (ret) 253 return ret; 254 if (check_seq) { 255 if (ref1->seq < ref2->seq) 256 return -1; 257 if (ref1->seq > ref2->seq) 258 return 1; 259 } 260 return 0; 261 } 262 263 /* insert a new ref to head ref rbtree */ 264 static struct btrfs_delayed_ref_head *htree_insert(struct rb_root_cached *root, 265 struct rb_node *node) 266 { 267 struct rb_node **p = &root->rb_root.rb_node; 268 struct rb_node *parent_node = NULL; 269 struct btrfs_delayed_ref_head *entry; 270 struct btrfs_delayed_ref_head *ins; 271 u64 bytenr; 272 bool leftmost = true; 273 274 ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node); 275 bytenr = ins->bytenr; 276 while (*p) { 277 parent_node = *p; 278 entry = rb_entry(parent_node, struct btrfs_delayed_ref_head, 279 href_node); 280 281 if (bytenr < entry->bytenr) { 282 p = &(*p)->rb_left; 283 } else if (bytenr > entry->bytenr) { 284 p = &(*p)->rb_right; 285 leftmost = false; 286 } else { 287 return entry; 288 } 289 } 290 291 rb_link_node(node, parent_node, p); 292 rb_insert_color_cached(node, root, leftmost); 293 return NULL; 294 } 295 296 static struct btrfs_delayed_ref_node* tree_insert(struct rb_root_cached *root, 297 struct btrfs_delayed_ref_node *ins) 298 { 299 struct rb_node **p = &root->rb_root.rb_node; 300 struct rb_node *node = &ins->ref_node; 301 struct rb_node *parent_node = NULL; 302 struct btrfs_delayed_ref_node *entry; 303 bool leftmost = true; 304 305 while (*p) { 306 int comp; 307 308 parent_node = *p; 309 entry = rb_entry(parent_node, struct btrfs_delayed_ref_node, 310 ref_node); 311 comp = comp_refs(ins, entry, true); 312 if (comp < 0) { 313 p = &(*p)->rb_left; 314 } else if (comp > 0) { 315 p = &(*p)->rb_right; 316 leftmost = false; 317 } else { 318 return entry; 319 } 320 } 321 322 rb_link_node(node, parent_node, p); 323 rb_insert_color_cached(node, root, leftmost); 324 return NULL; 325 } 326 327 static struct btrfs_delayed_ref_head *find_first_ref_head( 328 struct btrfs_delayed_ref_root *dr) 329 { 330 struct rb_node *n; 331 struct btrfs_delayed_ref_head *entry; 332 333 n = rb_first_cached(&dr->href_root); 334 if (!n) 335 return NULL; 336 337 entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); 338 339 return entry; 340 } 341 342 /* 343 * Find a head entry based on bytenr. This returns the delayed ref head if it 344 * was able to find one, or NULL if nothing was in that spot. If return_bigger 345 * is given, the next bigger entry is returned if no exact match is found. 346 */ 347 static struct btrfs_delayed_ref_head *find_ref_head( 348 struct btrfs_delayed_ref_root *dr, u64 bytenr, 349 bool return_bigger) 350 { 351 struct rb_root *root = &dr->href_root.rb_root; 352 struct rb_node *n; 353 struct btrfs_delayed_ref_head *entry; 354 355 n = root->rb_node; 356 entry = NULL; 357 while (n) { 358 entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); 359 360 if (bytenr < entry->bytenr) 361 n = n->rb_left; 362 else if (bytenr > entry->bytenr) 363 n = n->rb_right; 364 else 365 return entry; 366 } 367 if (entry && return_bigger) { 368 if (bytenr > entry->bytenr) { 369 n = rb_next(&entry->href_node); 370 if (!n) 371 return NULL; 372 entry = rb_entry(n, struct btrfs_delayed_ref_head, 373 href_node); 374 } 375 return entry; 376 } 377 return NULL; 378 } 379 380 int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs, 381 struct btrfs_delayed_ref_head *head) 382 { 383 lockdep_assert_held(&delayed_refs->lock); 384 if (mutex_trylock(&head->mutex)) 385 return 0; 386 387 refcount_inc(&head->refs); 388 spin_unlock(&delayed_refs->lock); 389 390 mutex_lock(&head->mutex); 391 spin_lock(&delayed_refs->lock); 392 if (RB_EMPTY_NODE(&head->href_node)) { 393 mutex_unlock(&head->mutex); 394 btrfs_put_delayed_ref_head(head); 395 return -EAGAIN; 396 } 397 btrfs_put_delayed_ref_head(head); 398 return 0; 399 } 400 401 static inline void drop_delayed_ref(struct btrfs_delayed_ref_root *delayed_refs, 402 struct btrfs_delayed_ref_head *head, 403 struct btrfs_delayed_ref_node *ref) 404 { 405 lockdep_assert_held(&head->lock); 406 rb_erase_cached(&ref->ref_node, &head->ref_tree); 407 RB_CLEAR_NODE(&ref->ref_node); 408 if (!list_empty(&ref->add_list)) 409 list_del(&ref->add_list); 410 btrfs_put_delayed_ref(ref); 411 atomic_dec(&delayed_refs->num_entries); 412 } 413 414 static bool merge_ref(struct btrfs_delayed_ref_root *delayed_refs, 415 struct btrfs_delayed_ref_head *head, 416 struct btrfs_delayed_ref_node *ref, 417 u64 seq) 418 { 419 struct btrfs_delayed_ref_node *next; 420 struct rb_node *node = rb_next(&ref->ref_node); 421 bool done = false; 422 423 while (!done && node) { 424 int mod; 425 426 next = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); 427 node = rb_next(node); 428 if (seq && next->seq >= seq) 429 break; 430 if (comp_refs(ref, next, false)) 431 break; 432 433 if (ref->action == next->action) { 434 mod = next->ref_mod; 435 } else { 436 if (ref->ref_mod < next->ref_mod) { 437 swap(ref, next); 438 done = true; 439 } 440 mod = -next->ref_mod; 441 } 442 443 drop_delayed_ref(delayed_refs, head, next); 444 ref->ref_mod += mod; 445 if (ref->ref_mod == 0) { 446 drop_delayed_ref(delayed_refs, head, ref); 447 done = true; 448 } else { 449 /* 450 * Can't have multiples of the same ref on a tree block. 451 */ 452 WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY || 453 ref->type == BTRFS_SHARED_BLOCK_REF_KEY); 454 } 455 } 456 457 return done; 458 } 459 460 void btrfs_merge_delayed_refs(struct btrfs_fs_info *fs_info, 461 struct btrfs_delayed_ref_root *delayed_refs, 462 struct btrfs_delayed_ref_head *head) 463 { 464 struct btrfs_delayed_ref_node *ref; 465 struct rb_node *node; 466 u64 seq = 0; 467 468 lockdep_assert_held(&head->lock); 469 470 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) 471 return; 472 473 /* We don't have too many refs to merge for data. */ 474 if (head->is_data) 475 return; 476 477 seq = btrfs_tree_mod_log_lowest_seq(fs_info); 478 again: 479 for (node = rb_first_cached(&head->ref_tree); node; 480 node = rb_next(node)) { 481 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); 482 if (seq && ref->seq >= seq) 483 continue; 484 if (merge_ref(delayed_refs, head, ref, seq)) 485 goto again; 486 } 487 } 488 489 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq) 490 { 491 int ret = 0; 492 u64 min_seq = btrfs_tree_mod_log_lowest_seq(fs_info); 493 494 if (min_seq != 0 && seq >= min_seq) { 495 btrfs_debug(fs_info, 496 "holding back delayed_ref %llu, lowest is %llu", 497 seq, min_seq); 498 ret = 1; 499 } 500 501 return ret; 502 } 503 504 struct btrfs_delayed_ref_head *btrfs_select_ref_head( 505 struct btrfs_delayed_ref_root *delayed_refs) 506 { 507 struct btrfs_delayed_ref_head *head; 508 509 lockdep_assert_held(&delayed_refs->lock); 510 again: 511 head = find_ref_head(delayed_refs, delayed_refs->run_delayed_start, 512 true); 513 if (!head && delayed_refs->run_delayed_start != 0) { 514 delayed_refs->run_delayed_start = 0; 515 head = find_first_ref_head(delayed_refs); 516 } 517 if (!head) 518 return NULL; 519 520 while (head->processing) { 521 struct rb_node *node; 522 523 node = rb_next(&head->href_node); 524 if (!node) { 525 if (delayed_refs->run_delayed_start == 0) 526 return NULL; 527 delayed_refs->run_delayed_start = 0; 528 goto again; 529 } 530 head = rb_entry(node, struct btrfs_delayed_ref_head, 531 href_node); 532 } 533 534 head->processing = true; 535 WARN_ON(delayed_refs->num_heads_ready == 0); 536 delayed_refs->num_heads_ready--; 537 delayed_refs->run_delayed_start = head->bytenr + 538 head->num_bytes; 539 return head; 540 } 541 542 void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs, 543 struct btrfs_delayed_ref_head *head) 544 { 545 lockdep_assert_held(&delayed_refs->lock); 546 lockdep_assert_held(&head->lock); 547 548 rb_erase_cached(&head->href_node, &delayed_refs->href_root); 549 RB_CLEAR_NODE(&head->href_node); 550 atomic_dec(&delayed_refs->num_entries); 551 delayed_refs->num_heads--; 552 if (!head->processing) 553 delayed_refs->num_heads_ready--; 554 } 555 556 /* 557 * Helper to insert the ref_node to the tail or merge with tail. 558 * 559 * Return false if the ref was inserted. 560 * Return true if the ref was merged into an existing one (and therefore can be 561 * freed by the caller). 562 */ 563 static bool insert_delayed_ref(struct btrfs_delayed_ref_root *root, 564 struct btrfs_delayed_ref_head *href, 565 struct btrfs_delayed_ref_node *ref) 566 { 567 struct btrfs_delayed_ref_node *exist; 568 int mod; 569 570 spin_lock(&href->lock); 571 exist = tree_insert(&href->ref_tree, ref); 572 if (!exist) { 573 if (ref->action == BTRFS_ADD_DELAYED_REF) 574 list_add_tail(&ref->add_list, &href->ref_add_list); 575 atomic_inc(&root->num_entries); 576 spin_unlock(&href->lock); 577 return false; 578 } 579 580 /* Now we are sure we can merge */ 581 if (exist->action == ref->action) { 582 mod = ref->ref_mod; 583 } else { 584 /* Need to change action */ 585 if (exist->ref_mod < ref->ref_mod) { 586 exist->action = ref->action; 587 mod = -exist->ref_mod; 588 exist->ref_mod = ref->ref_mod; 589 if (ref->action == BTRFS_ADD_DELAYED_REF) 590 list_add_tail(&exist->add_list, 591 &href->ref_add_list); 592 else if (ref->action == BTRFS_DROP_DELAYED_REF) { 593 ASSERT(!list_empty(&exist->add_list)); 594 list_del(&exist->add_list); 595 } else { 596 ASSERT(0); 597 } 598 } else 599 mod = -ref->ref_mod; 600 } 601 exist->ref_mod += mod; 602 603 /* remove existing tail if its ref_mod is zero */ 604 if (exist->ref_mod == 0) 605 drop_delayed_ref(root, href, exist); 606 spin_unlock(&href->lock); 607 return true; 608 } 609 610 /* 611 * helper function to update the accounting in the head ref 612 * existing and update must have the same bytenr 613 */ 614 static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans, 615 struct btrfs_delayed_ref_head *existing, 616 struct btrfs_delayed_ref_head *update) 617 { 618 struct btrfs_delayed_ref_root *delayed_refs = 619 &trans->transaction->delayed_refs; 620 struct btrfs_fs_info *fs_info = trans->fs_info; 621 int old_ref_mod; 622 623 BUG_ON(existing->is_data != update->is_data); 624 625 spin_lock(&existing->lock); 626 if (update->must_insert_reserved) { 627 /* if the extent was freed and then 628 * reallocated before the delayed ref 629 * entries were processed, we can end up 630 * with an existing head ref without 631 * the must_insert_reserved flag set. 632 * Set it again here 633 */ 634 existing->must_insert_reserved = update->must_insert_reserved; 635 636 /* 637 * update the num_bytes so we make sure the accounting 638 * is done correctly 639 */ 640 existing->num_bytes = update->num_bytes; 641 642 } 643 644 if (update->extent_op) { 645 if (!existing->extent_op) { 646 existing->extent_op = update->extent_op; 647 } else { 648 if (update->extent_op->update_key) { 649 memcpy(&existing->extent_op->key, 650 &update->extent_op->key, 651 sizeof(update->extent_op->key)); 652 existing->extent_op->update_key = true; 653 } 654 if (update->extent_op->update_flags) { 655 existing->extent_op->flags_to_set |= 656 update->extent_op->flags_to_set; 657 existing->extent_op->update_flags = true; 658 } 659 btrfs_free_delayed_extent_op(update->extent_op); 660 } 661 } 662 /* 663 * update the reference mod on the head to reflect this new operation, 664 * only need the lock for this case cause we could be processing it 665 * currently, for refs we just added we know we're a-ok. 666 */ 667 old_ref_mod = existing->total_ref_mod; 668 existing->ref_mod += update->ref_mod; 669 existing->total_ref_mod += update->ref_mod; 670 671 /* 672 * If we are going to from a positive ref mod to a negative or vice 673 * versa we need to make sure to adjust pending_csums accordingly. 674 */ 675 if (existing->is_data) { 676 u64 csum_leaves = 677 btrfs_csum_bytes_to_leaves(fs_info, 678 existing->num_bytes); 679 680 if (existing->total_ref_mod >= 0 && old_ref_mod < 0) { 681 delayed_refs->pending_csums -= existing->num_bytes; 682 btrfs_delayed_refs_rsv_release(fs_info, csum_leaves); 683 } 684 if (existing->total_ref_mod < 0 && old_ref_mod >= 0) { 685 delayed_refs->pending_csums += existing->num_bytes; 686 trans->delayed_ref_updates += csum_leaves; 687 } 688 } 689 690 spin_unlock(&existing->lock); 691 } 692 693 static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref, 694 struct btrfs_qgroup_extent_record *qrecord, 695 u64 bytenr, u64 num_bytes, u64 ref_root, 696 u64 reserved, int action, bool is_data, 697 bool is_system) 698 { 699 int count_mod = 1; 700 bool must_insert_reserved = false; 701 702 /* If reserved is provided, it must be a data extent. */ 703 BUG_ON(!is_data && reserved); 704 705 switch (action) { 706 case BTRFS_UPDATE_DELAYED_HEAD: 707 count_mod = 0; 708 break; 709 case BTRFS_DROP_DELAYED_REF: 710 /* 711 * The head node stores the sum of all the mods, so dropping a ref 712 * should drop the sum in the head node by one. 713 */ 714 count_mod = -1; 715 break; 716 case BTRFS_ADD_DELAYED_EXTENT: 717 /* 718 * BTRFS_ADD_DELAYED_EXTENT means that we need to update the 719 * reserved accounting when the extent is finally added, or if a 720 * later modification deletes the delayed ref without ever 721 * inserting the extent into the extent allocation tree. 722 * ref->must_insert_reserved is the flag used to record that 723 * accounting mods are required. 724 * 725 * Once we record must_insert_reserved, switch the action to 726 * BTRFS_ADD_DELAYED_REF because other special casing is not 727 * required. 728 */ 729 must_insert_reserved = true; 730 break; 731 } 732 733 refcount_set(&head_ref->refs, 1); 734 head_ref->bytenr = bytenr; 735 head_ref->num_bytes = num_bytes; 736 head_ref->ref_mod = count_mod; 737 head_ref->must_insert_reserved = must_insert_reserved; 738 head_ref->is_data = is_data; 739 head_ref->is_system = is_system; 740 head_ref->ref_tree = RB_ROOT_CACHED; 741 INIT_LIST_HEAD(&head_ref->ref_add_list); 742 RB_CLEAR_NODE(&head_ref->href_node); 743 head_ref->processing = false; 744 head_ref->total_ref_mod = count_mod; 745 spin_lock_init(&head_ref->lock); 746 mutex_init(&head_ref->mutex); 747 748 if (qrecord) { 749 if (ref_root && reserved) { 750 qrecord->data_rsv = reserved; 751 qrecord->data_rsv_refroot = ref_root; 752 } 753 qrecord->bytenr = bytenr; 754 qrecord->num_bytes = num_bytes; 755 qrecord->old_roots = NULL; 756 } 757 } 758 759 /* 760 * helper function to actually insert a head node into the rbtree. 761 * this does all the dirty work in terms of maintaining the correct 762 * overall modification count. 763 */ 764 static noinline struct btrfs_delayed_ref_head * 765 add_delayed_ref_head(struct btrfs_trans_handle *trans, 766 struct btrfs_delayed_ref_head *head_ref, 767 struct btrfs_qgroup_extent_record *qrecord, 768 int action, bool *qrecord_inserted_ret) 769 { 770 struct btrfs_delayed_ref_head *existing; 771 struct btrfs_delayed_ref_root *delayed_refs; 772 bool qrecord_inserted = false; 773 774 delayed_refs = &trans->transaction->delayed_refs; 775 776 /* Record qgroup extent info if provided */ 777 if (qrecord) { 778 if (btrfs_qgroup_trace_extent_nolock(trans->fs_info, 779 delayed_refs, qrecord)) 780 kfree(qrecord); 781 else 782 qrecord_inserted = true; 783 } 784 785 trace_add_delayed_ref_head(trans->fs_info, head_ref, action); 786 787 existing = htree_insert(&delayed_refs->href_root, 788 &head_ref->href_node); 789 if (existing) { 790 update_existing_head_ref(trans, existing, head_ref); 791 /* 792 * we've updated the existing ref, free the newly 793 * allocated ref 794 */ 795 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); 796 head_ref = existing; 797 } else { 798 if (head_ref->is_data && head_ref->ref_mod < 0) { 799 delayed_refs->pending_csums += head_ref->num_bytes; 800 trans->delayed_ref_updates += 801 btrfs_csum_bytes_to_leaves(trans->fs_info, 802 head_ref->num_bytes); 803 } 804 delayed_refs->num_heads++; 805 delayed_refs->num_heads_ready++; 806 atomic_inc(&delayed_refs->num_entries); 807 trans->delayed_ref_updates++; 808 } 809 if (qrecord_inserted_ret) 810 *qrecord_inserted_ret = qrecord_inserted; 811 812 return head_ref; 813 } 814 815 /* 816 * init_delayed_ref_common - Initialize the structure which represents a 817 * modification to a an extent. 818 * 819 * @fs_info: Internal to the mounted filesystem mount structure. 820 * 821 * @ref: The structure which is going to be initialized. 822 * 823 * @bytenr: The logical address of the extent for which a modification is 824 * going to be recorded. 825 * 826 * @num_bytes: Size of the extent whose modification is being recorded. 827 * 828 * @ref_root: The id of the root where this modification has originated, this 829 * can be either one of the well-known metadata trees or the 830 * subvolume id which references this extent. 831 * 832 * @action: Can be one of BTRFS_ADD_DELAYED_REF/BTRFS_DROP_DELAYED_REF or 833 * BTRFS_ADD_DELAYED_EXTENT 834 * 835 * @ref_type: Holds the type of the extent which is being recorded, can be 836 * one of BTRFS_SHARED_BLOCK_REF_KEY/BTRFS_TREE_BLOCK_REF_KEY 837 * when recording a metadata extent or BTRFS_SHARED_DATA_REF_KEY/ 838 * BTRFS_EXTENT_DATA_REF_KEY when recording data extent 839 */ 840 static void init_delayed_ref_common(struct btrfs_fs_info *fs_info, 841 struct btrfs_delayed_ref_node *ref, 842 u64 bytenr, u64 num_bytes, u64 ref_root, 843 int action, u8 ref_type) 844 { 845 u64 seq = 0; 846 847 if (action == BTRFS_ADD_DELAYED_EXTENT) 848 action = BTRFS_ADD_DELAYED_REF; 849 850 if (is_fstree(ref_root)) 851 seq = atomic64_read(&fs_info->tree_mod_seq); 852 853 refcount_set(&ref->refs, 1); 854 ref->bytenr = bytenr; 855 ref->num_bytes = num_bytes; 856 ref->ref_mod = 1; 857 ref->action = action; 858 ref->seq = seq; 859 ref->type = ref_type; 860 RB_CLEAR_NODE(&ref->ref_node); 861 INIT_LIST_HEAD(&ref->add_list); 862 } 863 864 /* 865 * add a delayed tree ref. This does all of the accounting required 866 * to make sure the delayed ref is eventually processed before this 867 * transaction commits. 868 */ 869 int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans, 870 struct btrfs_ref *generic_ref, 871 struct btrfs_delayed_extent_op *extent_op) 872 { 873 struct btrfs_fs_info *fs_info = trans->fs_info; 874 struct btrfs_delayed_tree_ref *ref; 875 struct btrfs_delayed_ref_head *head_ref; 876 struct btrfs_delayed_ref_root *delayed_refs; 877 struct btrfs_qgroup_extent_record *record = NULL; 878 bool qrecord_inserted; 879 bool is_system; 880 bool merged; 881 int action = generic_ref->action; 882 int level = generic_ref->tree_ref.level; 883 u64 bytenr = generic_ref->bytenr; 884 u64 num_bytes = generic_ref->len; 885 u64 parent = generic_ref->parent; 886 u8 ref_type; 887 888 is_system = (generic_ref->tree_ref.owning_root == BTRFS_CHUNK_TREE_OBJECTID); 889 890 ASSERT(generic_ref->type == BTRFS_REF_METADATA && generic_ref->action); 891 ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS); 892 if (!ref) 893 return -ENOMEM; 894 895 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); 896 if (!head_ref) { 897 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); 898 return -ENOMEM; 899 } 900 901 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) && 902 !generic_ref->skip_qgroup) { 903 record = kzalloc(sizeof(*record), GFP_NOFS); 904 if (!record) { 905 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); 906 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); 907 return -ENOMEM; 908 } 909 } 910 911 if (parent) 912 ref_type = BTRFS_SHARED_BLOCK_REF_KEY; 913 else 914 ref_type = BTRFS_TREE_BLOCK_REF_KEY; 915 916 init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes, 917 generic_ref->tree_ref.owning_root, action, 918 ref_type); 919 ref->root = generic_ref->tree_ref.owning_root; 920 ref->parent = parent; 921 ref->level = level; 922 923 init_delayed_ref_head(head_ref, record, bytenr, num_bytes, 924 generic_ref->tree_ref.owning_root, 0, action, 925 false, is_system); 926 head_ref->extent_op = extent_op; 927 928 delayed_refs = &trans->transaction->delayed_refs; 929 spin_lock(&delayed_refs->lock); 930 931 /* 932 * insert both the head node and the new ref without dropping 933 * the spin lock 934 */ 935 head_ref = add_delayed_ref_head(trans, head_ref, record, 936 action, &qrecord_inserted); 937 938 merged = insert_delayed_ref(delayed_refs, head_ref, &ref->node); 939 spin_unlock(&delayed_refs->lock); 940 941 /* 942 * Need to update the delayed_refs_rsv with any changes we may have 943 * made. 944 */ 945 btrfs_update_delayed_refs_rsv(trans); 946 947 trace_add_delayed_tree_ref(fs_info, &ref->node, ref, 948 action == BTRFS_ADD_DELAYED_EXTENT ? 949 BTRFS_ADD_DELAYED_REF : action); 950 if (merged) 951 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); 952 953 if (qrecord_inserted) 954 btrfs_qgroup_trace_extent_post(trans, record); 955 956 return 0; 957 } 958 959 /* 960 * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref. 961 */ 962 int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans, 963 struct btrfs_ref *generic_ref, 964 u64 reserved) 965 { 966 struct btrfs_fs_info *fs_info = trans->fs_info; 967 struct btrfs_delayed_data_ref *ref; 968 struct btrfs_delayed_ref_head *head_ref; 969 struct btrfs_delayed_ref_root *delayed_refs; 970 struct btrfs_qgroup_extent_record *record = NULL; 971 bool qrecord_inserted; 972 int action = generic_ref->action; 973 bool merged; 974 u64 bytenr = generic_ref->bytenr; 975 u64 num_bytes = generic_ref->len; 976 u64 parent = generic_ref->parent; 977 u64 ref_root = generic_ref->data_ref.owning_root; 978 u64 owner = generic_ref->data_ref.ino; 979 u64 offset = generic_ref->data_ref.offset; 980 u8 ref_type; 981 982 ASSERT(generic_ref->type == BTRFS_REF_DATA && action); 983 ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS); 984 if (!ref) 985 return -ENOMEM; 986 987 if (parent) 988 ref_type = BTRFS_SHARED_DATA_REF_KEY; 989 else 990 ref_type = BTRFS_EXTENT_DATA_REF_KEY; 991 init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes, 992 ref_root, action, ref_type); 993 ref->root = ref_root; 994 ref->parent = parent; 995 ref->objectid = owner; 996 ref->offset = offset; 997 998 999 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); 1000 if (!head_ref) { 1001 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); 1002 return -ENOMEM; 1003 } 1004 1005 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) && 1006 !generic_ref->skip_qgroup) { 1007 record = kzalloc(sizeof(*record), GFP_NOFS); 1008 if (!record) { 1009 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); 1010 kmem_cache_free(btrfs_delayed_ref_head_cachep, 1011 head_ref); 1012 return -ENOMEM; 1013 } 1014 } 1015 1016 init_delayed_ref_head(head_ref, record, bytenr, num_bytes, ref_root, 1017 reserved, action, true, false); 1018 head_ref->extent_op = NULL; 1019 1020 delayed_refs = &trans->transaction->delayed_refs; 1021 spin_lock(&delayed_refs->lock); 1022 1023 /* 1024 * insert both the head node and the new ref without dropping 1025 * the spin lock 1026 */ 1027 head_ref = add_delayed_ref_head(trans, head_ref, record, 1028 action, &qrecord_inserted); 1029 1030 merged = insert_delayed_ref(delayed_refs, head_ref, &ref->node); 1031 spin_unlock(&delayed_refs->lock); 1032 1033 /* 1034 * Need to update the delayed_refs_rsv with any changes we may have 1035 * made. 1036 */ 1037 btrfs_update_delayed_refs_rsv(trans); 1038 1039 trace_add_delayed_data_ref(trans->fs_info, &ref->node, ref, 1040 action == BTRFS_ADD_DELAYED_EXTENT ? 1041 BTRFS_ADD_DELAYED_REF : action); 1042 if (merged) 1043 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); 1044 1045 1046 if (qrecord_inserted) 1047 return btrfs_qgroup_trace_extent_post(trans, record); 1048 return 0; 1049 } 1050 1051 int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans, 1052 u64 bytenr, u64 num_bytes, 1053 struct btrfs_delayed_extent_op *extent_op) 1054 { 1055 struct btrfs_delayed_ref_head *head_ref; 1056 struct btrfs_delayed_ref_root *delayed_refs; 1057 1058 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); 1059 if (!head_ref) 1060 return -ENOMEM; 1061 1062 init_delayed_ref_head(head_ref, NULL, bytenr, num_bytes, 0, 0, 1063 BTRFS_UPDATE_DELAYED_HEAD, false, false); 1064 head_ref->extent_op = extent_op; 1065 1066 delayed_refs = &trans->transaction->delayed_refs; 1067 spin_lock(&delayed_refs->lock); 1068 1069 add_delayed_ref_head(trans, head_ref, NULL, BTRFS_UPDATE_DELAYED_HEAD, 1070 NULL); 1071 1072 spin_unlock(&delayed_refs->lock); 1073 1074 /* 1075 * Need to update the delayed_refs_rsv with any changes we may have 1076 * made. 1077 */ 1078 btrfs_update_delayed_refs_rsv(trans); 1079 return 0; 1080 } 1081 1082 /* 1083 * This does a simple search for the head node for a given extent. Returns the 1084 * head node if found, or NULL if not. 1085 */ 1086 struct btrfs_delayed_ref_head * 1087 btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, u64 bytenr) 1088 { 1089 lockdep_assert_held(&delayed_refs->lock); 1090 1091 return find_ref_head(delayed_refs, bytenr, false); 1092 } 1093 1094 void __cold btrfs_delayed_ref_exit(void) 1095 { 1096 kmem_cache_destroy(btrfs_delayed_ref_head_cachep); 1097 kmem_cache_destroy(btrfs_delayed_tree_ref_cachep); 1098 kmem_cache_destroy(btrfs_delayed_data_ref_cachep); 1099 kmem_cache_destroy(btrfs_delayed_extent_op_cachep); 1100 } 1101 1102 int __init btrfs_delayed_ref_init(void) 1103 { 1104 btrfs_delayed_ref_head_cachep = kmem_cache_create( 1105 "btrfs_delayed_ref_head", 1106 sizeof(struct btrfs_delayed_ref_head), 0, 1107 SLAB_MEM_SPREAD, NULL); 1108 if (!btrfs_delayed_ref_head_cachep) 1109 goto fail; 1110 1111 btrfs_delayed_tree_ref_cachep = kmem_cache_create( 1112 "btrfs_delayed_tree_ref", 1113 sizeof(struct btrfs_delayed_tree_ref), 0, 1114 SLAB_MEM_SPREAD, NULL); 1115 if (!btrfs_delayed_tree_ref_cachep) 1116 goto fail; 1117 1118 btrfs_delayed_data_ref_cachep = kmem_cache_create( 1119 "btrfs_delayed_data_ref", 1120 sizeof(struct btrfs_delayed_data_ref), 0, 1121 SLAB_MEM_SPREAD, NULL); 1122 if (!btrfs_delayed_data_ref_cachep) 1123 goto fail; 1124 1125 btrfs_delayed_extent_op_cachep = kmem_cache_create( 1126 "btrfs_delayed_extent_op", 1127 sizeof(struct btrfs_delayed_extent_op), 0, 1128 SLAB_MEM_SPREAD, NULL); 1129 if (!btrfs_delayed_extent_op_cachep) 1130 goto fail; 1131 1132 return 0; 1133 fail: 1134 btrfs_delayed_ref_exit(); 1135 return -ENOMEM; 1136 } 1137