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 "ctree.h" 10 #include "delayed-ref.h" 11 #include "transaction.h" 12 #include "qgroup.h" 13 #include "space-info.h" 14 15 struct kmem_cache *btrfs_delayed_ref_head_cachep; 16 struct kmem_cache *btrfs_delayed_tree_ref_cachep; 17 struct kmem_cache *btrfs_delayed_data_ref_cachep; 18 struct kmem_cache *btrfs_delayed_extent_op_cachep; 19 /* 20 * delayed back reference update tracking. For subvolume trees 21 * we queue up extent allocations and backref maintenance for 22 * delayed processing. This avoids deep call chains where we 23 * add extents in the middle of btrfs_search_slot, and it allows 24 * us to buffer up frequently modified backrefs in an rb tree instead 25 * of hammering updates on the extent allocation tree. 26 */ 27 28 bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info) 29 { 30 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 31 struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; 32 bool ret = false; 33 u64 reserved; 34 35 spin_lock(&global_rsv->lock); 36 reserved = global_rsv->reserved; 37 spin_unlock(&global_rsv->lock); 38 39 /* 40 * Since the global reserve is just kind of magic we don't really want 41 * to rely on it to save our bacon, so if our size is more than the 42 * delayed_refs_rsv and the global rsv then it's time to think about 43 * bailing. 44 */ 45 spin_lock(&delayed_refs_rsv->lock); 46 reserved += delayed_refs_rsv->reserved; 47 if (delayed_refs_rsv->size >= reserved) 48 ret = true; 49 spin_unlock(&delayed_refs_rsv->lock); 50 return ret; 51 } 52 53 int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans) 54 { 55 u64 num_entries = 56 atomic_read(&trans->transaction->delayed_refs.num_entries); 57 u64 avg_runtime; 58 u64 val; 59 60 smp_mb(); 61 avg_runtime = trans->fs_info->avg_delayed_ref_runtime; 62 val = num_entries * avg_runtime; 63 if (val >= NSEC_PER_SEC) 64 return 1; 65 if (val >= NSEC_PER_SEC / 2) 66 return 2; 67 68 return btrfs_check_space_for_delayed_refs(trans->fs_info); 69 } 70 71 /** 72 * Release a ref head's reservation 73 * 74 * @fs_info: the filesystem 75 * @nr: number of items to drop 76 * 77 * This drops the delayed ref head's count from the delayed refs rsv and frees 78 * any excess reservation we had. 79 */ 80 void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr) 81 { 82 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; 83 u64 num_bytes = btrfs_calc_insert_metadata_size(fs_info, nr); 84 u64 released = 0; 85 86 released = btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL); 87 if (released) 88 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", 89 0, released, 0); 90 } 91 92 /* 93 * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv 94 * @trans - the trans that may have generated delayed refs 95 * 96 * This is to be called anytime we may have adjusted trans->delayed_ref_updates, 97 * it'll calculate the additional size and add it to the delayed_refs_rsv. 98 */ 99 void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans) 100 { 101 struct btrfs_fs_info *fs_info = trans->fs_info; 102 struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; 103 u64 num_bytes; 104 105 if (!trans->delayed_ref_updates) 106 return; 107 108 num_bytes = btrfs_calc_insert_metadata_size(fs_info, 109 trans->delayed_ref_updates); 110 spin_lock(&delayed_rsv->lock); 111 delayed_rsv->size += num_bytes; 112 delayed_rsv->full = 0; 113 spin_unlock(&delayed_rsv->lock); 114 trans->delayed_ref_updates = 0; 115 } 116 117 /** 118 * Transfer bytes to our delayed refs rsv 119 * 120 * @fs_info: the filesystem 121 * @src: source block rsv to transfer from 122 * @num_bytes: number of bytes to transfer 123 * 124 * This transfers up to the num_bytes amount from the src rsv to the 125 * delayed_refs_rsv. Any extra bytes are returned to the space info. 126 */ 127 void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info, 128 struct btrfs_block_rsv *src, 129 u64 num_bytes) 130 { 131 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 132 u64 to_free = 0; 133 134 spin_lock(&src->lock); 135 src->reserved -= num_bytes; 136 src->size -= num_bytes; 137 spin_unlock(&src->lock); 138 139 spin_lock(&delayed_refs_rsv->lock); 140 if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) { 141 u64 delta = delayed_refs_rsv->size - 142 delayed_refs_rsv->reserved; 143 if (num_bytes > delta) { 144 to_free = num_bytes - delta; 145 num_bytes = delta; 146 } 147 } else { 148 to_free = num_bytes; 149 num_bytes = 0; 150 } 151 152 if (num_bytes) 153 delayed_refs_rsv->reserved += num_bytes; 154 if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size) 155 delayed_refs_rsv->full = 1; 156 spin_unlock(&delayed_refs_rsv->lock); 157 158 if (num_bytes) 159 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", 160 0, num_bytes, 1); 161 if (to_free) 162 btrfs_space_info_free_bytes_may_use(fs_info, 163 delayed_refs_rsv->space_info, to_free); 164 } 165 166 /** 167 * Refill based on our delayed refs usage 168 * 169 * @fs_info: the filesystem 170 * @flush: control how we can flush for this reservation. 171 * 172 * This will refill the delayed block_rsv up to 1 items size worth of space and 173 * will return -ENOSPC if we can't make the reservation. 174 */ 175 int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info, 176 enum btrfs_reserve_flush_enum flush) 177 { 178 struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; 179 u64 limit = btrfs_calc_insert_metadata_size(fs_info, 1); 180 u64 num_bytes = 0; 181 int ret = -ENOSPC; 182 183 spin_lock(&block_rsv->lock); 184 if (block_rsv->reserved < block_rsv->size) { 185 num_bytes = block_rsv->size - block_rsv->reserved; 186 num_bytes = min(num_bytes, limit); 187 } 188 spin_unlock(&block_rsv->lock); 189 190 if (!num_bytes) 191 return 0; 192 193 ret = btrfs_reserve_metadata_bytes(fs_info->extent_root, block_rsv, 194 num_bytes, flush); 195 if (ret) 196 return ret; 197 btrfs_block_rsv_add_bytes(block_rsv, num_bytes, 0); 198 trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", 199 0, num_bytes, 1); 200 return 0; 201 } 202 203 /* 204 * compare two delayed tree backrefs with same bytenr and type 205 */ 206 static int comp_tree_refs(struct btrfs_delayed_tree_ref *ref1, 207 struct btrfs_delayed_tree_ref *ref2) 208 { 209 if (ref1->node.type == BTRFS_TREE_BLOCK_REF_KEY) { 210 if (ref1->root < ref2->root) 211 return -1; 212 if (ref1->root > ref2->root) 213 return 1; 214 } else { 215 if (ref1->parent < ref2->parent) 216 return -1; 217 if (ref1->parent > ref2->parent) 218 return 1; 219 } 220 return 0; 221 } 222 223 /* 224 * compare two delayed data backrefs with same bytenr and type 225 */ 226 static int comp_data_refs(struct btrfs_delayed_data_ref *ref1, 227 struct btrfs_delayed_data_ref *ref2) 228 { 229 if (ref1->node.type == BTRFS_EXTENT_DATA_REF_KEY) { 230 if (ref1->root < ref2->root) 231 return -1; 232 if (ref1->root > ref2->root) 233 return 1; 234 if (ref1->objectid < ref2->objectid) 235 return -1; 236 if (ref1->objectid > ref2->objectid) 237 return 1; 238 if (ref1->offset < ref2->offset) 239 return -1; 240 if (ref1->offset > ref2->offset) 241 return 1; 242 } else { 243 if (ref1->parent < ref2->parent) 244 return -1; 245 if (ref1->parent > ref2->parent) 246 return 1; 247 } 248 return 0; 249 } 250 251 static int comp_refs(struct btrfs_delayed_ref_node *ref1, 252 struct btrfs_delayed_ref_node *ref2, 253 bool check_seq) 254 { 255 int ret = 0; 256 257 if (ref1->type < ref2->type) 258 return -1; 259 if (ref1->type > ref2->type) 260 return 1; 261 if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY || 262 ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) 263 ret = comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref1), 264 btrfs_delayed_node_to_tree_ref(ref2)); 265 else 266 ret = comp_data_refs(btrfs_delayed_node_to_data_ref(ref1), 267 btrfs_delayed_node_to_data_ref(ref2)); 268 if (ret) 269 return ret; 270 if (check_seq) { 271 if (ref1->seq < ref2->seq) 272 return -1; 273 if (ref1->seq > ref2->seq) 274 return 1; 275 } 276 return 0; 277 } 278 279 /* insert a new ref to head ref rbtree */ 280 static struct btrfs_delayed_ref_head *htree_insert(struct rb_root_cached *root, 281 struct rb_node *node) 282 { 283 struct rb_node **p = &root->rb_root.rb_node; 284 struct rb_node *parent_node = NULL; 285 struct btrfs_delayed_ref_head *entry; 286 struct btrfs_delayed_ref_head *ins; 287 u64 bytenr; 288 bool leftmost = true; 289 290 ins = rb_entry(node, struct btrfs_delayed_ref_head, href_node); 291 bytenr = ins->bytenr; 292 while (*p) { 293 parent_node = *p; 294 entry = rb_entry(parent_node, struct btrfs_delayed_ref_head, 295 href_node); 296 297 if (bytenr < entry->bytenr) { 298 p = &(*p)->rb_left; 299 } else if (bytenr > entry->bytenr) { 300 p = &(*p)->rb_right; 301 leftmost = false; 302 } else { 303 return entry; 304 } 305 } 306 307 rb_link_node(node, parent_node, p); 308 rb_insert_color_cached(node, root, leftmost); 309 return NULL; 310 } 311 312 static struct btrfs_delayed_ref_node* tree_insert(struct rb_root_cached *root, 313 struct btrfs_delayed_ref_node *ins) 314 { 315 struct rb_node **p = &root->rb_root.rb_node; 316 struct rb_node *node = &ins->ref_node; 317 struct rb_node *parent_node = NULL; 318 struct btrfs_delayed_ref_node *entry; 319 bool leftmost = true; 320 321 while (*p) { 322 int comp; 323 324 parent_node = *p; 325 entry = rb_entry(parent_node, struct btrfs_delayed_ref_node, 326 ref_node); 327 comp = comp_refs(ins, entry, true); 328 if (comp < 0) { 329 p = &(*p)->rb_left; 330 } else if (comp > 0) { 331 p = &(*p)->rb_right; 332 leftmost = false; 333 } else { 334 return entry; 335 } 336 } 337 338 rb_link_node(node, parent_node, p); 339 rb_insert_color_cached(node, root, leftmost); 340 return NULL; 341 } 342 343 static struct btrfs_delayed_ref_head *find_first_ref_head( 344 struct btrfs_delayed_ref_root *dr) 345 { 346 struct rb_node *n; 347 struct btrfs_delayed_ref_head *entry; 348 349 n = rb_first_cached(&dr->href_root); 350 if (!n) 351 return NULL; 352 353 entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); 354 355 return entry; 356 } 357 358 /* 359 * Find a head entry based on bytenr. This returns the delayed ref head if it 360 * was able to find one, or NULL if nothing was in that spot. If return_bigger 361 * is given, the next bigger entry is returned if no exact match is found. 362 */ 363 static struct btrfs_delayed_ref_head *find_ref_head( 364 struct btrfs_delayed_ref_root *dr, u64 bytenr, 365 bool return_bigger) 366 { 367 struct rb_root *root = &dr->href_root.rb_root; 368 struct rb_node *n; 369 struct btrfs_delayed_ref_head *entry; 370 371 n = root->rb_node; 372 entry = NULL; 373 while (n) { 374 entry = rb_entry(n, struct btrfs_delayed_ref_head, href_node); 375 376 if (bytenr < entry->bytenr) 377 n = n->rb_left; 378 else if (bytenr > entry->bytenr) 379 n = n->rb_right; 380 else 381 return entry; 382 } 383 if (entry && return_bigger) { 384 if (bytenr > entry->bytenr) { 385 n = rb_next(&entry->href_node); 386 if (!n) 387 return NULL; 388 entry = rb_entry(n, struct btrfs_delayed_ref_head, 389 href_node); 390 } 391 return entry; 392 } 393 return NULL; 394 } 395 396 int btrfs_delayed_ref_lock(struct btrfs_delayed_ref_root *delayed_refs, 397 struct btrfs_delayed_ref_head *head) 398 { 399 lockdep_assert_held(&delayed_refs->lock); 400 if (mutex_trylock(&head->mutex)) 401 return 0; 402 403 refcount_inc(&head->refs); 404 spin_unlock(&delayed_refs->lock); 405 406 mutex_lock(&head->mutex); 407 spin_lock(&delayed_refs->lock); 408 if (RB_EMPTY_NODE(&head->href_node)) { 409 mutex_unlock(&head->mutex); 410 btrfs_put_delayed_ref_head(head); 411 return -EAGAIN; 412 } 413 btrfs_put_delayed_ref_head(head); 414 return 0; 415 } 416 417 static inline void drop_delayed_ref(struct btrfs_trans_handle *trans, 418 struct btrfs_delayed_ref_root *delayed_refs, 419 struct btrfs_delayed_ref_head *head, 420 struct btrfs_delayed_ref_node *ref) 421 { 422 lockdep_assert_held(&head->lock); 423 rb_erase_cached(&ref->ref_node, &head->ref_tree); 424 RB_CLEAR_NODE(&ref->ref_node); 425 if (!list_empty(&ref->add_list)) 426 list_del(&ref->add_list); 427 ref->in_tree = 0; 428 btrfs_put_delayed_ref(ref); 429 atomic_dec(&delayed_refs->num_entries); 430 } 431 432 static bool merge_ref(struct btrfs_trans_handle *trans, 433 struct btrfs_delayed_ref_root *delayed_refs, 434 struct btrfs_delayed_ref_head *head, 435 struct btrfs_delayed_ref_node *ref, 436 u64 seq) 437 { 438 struct btrfs_delayed_ref_node *next; 439 struct rb_node *node = rb_next(&ref->ref_node); 440 bool done = false; 441 442 while (!done && node) { 443 int mod; 444 445 next = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); 446 node = rb_next(node); 447 if (seq && next->seq >= seq) 448 break; 449 if (comp_refs(ref, next, false)) 450 break; 451 452 if (ref->action == next->action) { 453 mod = next->ref_mod; 454 } else { 455 if (ref->ref_mod < next->ref_mod) { 456 swap(ref, next); 457 done = true; 458 } 459 mod = -next->ref_mod; 460 } 461 462 drop_delayed_ref(trans, delayed_refs, head, next); 463 ref->ref_mod += mod; 464 if (ref->ref_mod == 0) { 465 drop_delayed_ref(trans, delayed_refs, head, ref); 466 done = true; 467 } else { 468 /* 469 * Can't have multiples of the same ref on a tree block. 470 */ 471 WARN_ON(ref->type == BTRFS_TREE_BLOCK_REF_KEY || 472 ref->type == BTRFS_SHARED_BLOCK_REF_KEY); 473 } 474 } 475 476 return done; 477 } 478 479 void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, 480 struct btrfs_delayed_ref_root *delayed_refs, 481 struct btrfs_delayed_ref_head *head) 482 { 483 struct btrfs_fs_info *fs_info = trans->fs_info; 484 struct btrfs_delayed_ref_node *ref; 485 struct rb_node *node; 486 u64 seq = 0; 487 488 lockdep_assert_held(&head->lock); 489 490 if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) 491 return; 492 493 /* We don't have too many refs to merge for data. */ 494 if (head->is_data) 495 return; 496 497 read_lock(&fs_info->tree_mod_log_lock); 498 if (!list_empty(&fs_info->tree_mod_seq_list)) { 499 struct seq_list *elem; 500 501 elem = list_first_entry(&fs_info->tree_mod_seq_list, 502 struct seq_list, list); 503 seq = elem->seq; 504 } 505 read_unlock(&fs_info->tree_mod_log_lock); 506 507 again: 508 for (node = rb_first_cached(&head->ref_tree); node; 509 node = rb_next(node)) { 510 ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); 511 if (seq && ref->seq >= seq) 512 continue; 513 if (merge_ref(trans, delayed_refs, head, ref, seq)) 514 goto again; 515 } 516 } 517 518 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, u64 seq) 519 { 520 struct seq_list *elem; 521 int ret = 0; 522 523 read_lock(&fs_info->tree_mod_log_lock); 524 if (!list_empty(&fs_info->tree_mod_seq_list)) { 525 elem = list_first_entry(&fs_info->tree_mod_seq_list, 526 struct seq_list, list); 527 if (seq >= elem->seq) { 528 btrfs_debug(fs_info, 529 "holding back delayed_ref %#x.%x, lowest is %#x.%x", 530 (u32)(seq >> 32), (u32)seq, 531 (u32)(elem->seq >> 32), (u32)elem->seq); 532 ret = 1; 533 } 534 } 535 536 read_unlock(&fs_info->tree_mod_log_lock); 537 return ret; 538 } 539 540 struct btrfs_delayed_ref_head *btrfs_select_ref_head( 541 struct btrfs_delayed_ref_root *delayed_refs) 542 { 543 struct btrfs_delayed_ref_head *head; 544 545 again: 546 head = find_ref_head(delayed_refs, delayed_refs->run_delayed_start, 547 true); 548 if (!head && delayed_refs->run_delayed_start != 0) { 549 delayed_refs->run_delayed_start = 0; 550 head = find_first_ref_head(delayed_refs); 551 } 552 if (!head) 553 return NULL; 554 555 while (head->processing) { 556 struct rb_node *node; 557 558 node = rb_next(&head->href_node); 559 if (!node) { 560 if (delayed_refs->run_delayed_start == 0) 561 return NULL; 562 delayed_refs->run_delayed_start = 0; 563 goto again; 564 } 565 head = rb_entry(node, struct btrfs_delayed_ref_head, 566 href_node); 567 } 568 569 head->processing = 1; 570 WARN_ON(delayed_refs->num_heads_ready == 0); 571 delayed_refs->num_heads_ready--; 572 delayed_refs->run_delayed_start = head->bytenr + 573 head->num_bytes; 574 return head; 575 } 576 577 void btrfs_delete_ref_head(struct btrfs_delayed_ref_root *delayed_refs, 578 struct btrfs_delayed_ref_head *head) 579 { 580 lockdep_assert_held(&delayed_refs->lock); 581 lockdep_assert_held(&head->lock); 582 583 rb_erase_cached(&head->href_node, &delayed_refs->href_root); 584 RB_CLEAR_NODE(&head->href_node); 585 atomic_dec(&delayed_refs->num_entries); 586 delayed_refs->num_heads--; 587 if (head->processing == 0) 588 delayed_refs->num_heads_ready--; 589 } 590 591 /* 592 * Helper to insert the ref_node to the tail or merge with tail. 593 * 594 * Return 0 for insert. 595 * Return >0 for merge. 596 */ 597 static int insert_delayed_ref(struct btrfs_trans_handle *trans, 598 struct btrfs_delayed_ref_root *root, 599 struct btrfs_delayed_ref_head *href, 600 struct btrfs_delayed_ref_node *ref) 601 { 602 struct btrfs_delayed_ref_node *exist; 603 int mod; 604 int ret = 0; 605 606 spin_lock(&href->lock); 607 exist = tree_insert(&href->ref_tree, ref); 608 if (!exist) 609 goto inserted; 610 611 /* Now we are sure we can merge */ 612 ret = 1; 613 if (exist->action == ref->action) { 614 mod = ref->ref_mod; 615 } else { 616 /* Need to change action */ 617 if (exist->ref_mod < ref->ref_mod) { 618 exist->action = ref->action; 619 mod = -exist->ref_mod; 620 exist->ref_mod = ref->ref_mod; 621 if (ref->action == BTRFS_ADD_DELAYED_REF) 622 list_add_tail(&exist->add_list, 623 &href->ref_add_list); 624 else if (ref->action == BTRFS_DROP_DELAYED_REF) { 625 ASSERT(!list_empty(&exist->add_list)); 626 list_del(&exist->add_list); 627 } else { 628 ASSERT(0); 629 } 630 } else 631 mod = -ref->ref_mod; 632 } 633 exist->ref_mod += mod; 634 635 /* remove existing tail if its ref_mod is zero */ 636 if (exist->ref_mod == 0) 637 drop_delayed_ref(trans, root, href, exist); 638 spin_unlock(&href->lock); 639 return ret; 640 inserted: 641 if (ref->action == BTRFS_ADD_DELAYED_REF) 642 list_add_tail(&ref->add_list, &href->ref_add_list); 643 atomic_inc(&root->num_entries); 644 spin_unlock(&href->lock); 645 return ret; 646 } 647 648 /* 649 * helper function to update the accounting in the head ref 650 * existing and update must have the same bytenr 651 */ 652 static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans, 653 struct btrfs_delayed_ref_head *existing, 654 struct btrfs_delayed_ref_head *update) 655 { 656 struct btrfs_delayed_ref_root *delayed_refs = 657 &trans->transaction->delayed_refs; 658 struct btrfs_fs_info *fs_info = trans->fs_info; 659 u64 flags = btrfs_ref_head_to_space_flags(existing); 660 int old_ref_mod; 661 662 BUG_ON(existing->is_data != update->is_data); 663 664 spin_lock(&existing->lock); 665 if (update->must_insert_reserved) { 666 /* if the extent was freed and then 667 * reallocated before the delayed ref 668 * entries were processed, we can end up 669 * with an existing head ref without 670 * the must_insert_reserved flag set. 671 * Set it again here 672 */ 673 existing->must_insert_reserved = update->must_insert_reserved; 674 675 /* 676 * update the num_bytes so we make sure the accounting 677 * is done correctly 678 */ 679 existing->num_bytes = update->num_bytes; 680 681 } 682 683 if (update->extent_op) { 684 if (!existing->extent_op) { 685 existing->extent_op = update->extent_op; 686 } else { 687 if (update->extent_op->update_key) { 688 memcpy(&existing->extent_op->key, 689 &update->extent_op->key, 690 sizeof(update->extent_op->key)); 691 existing->extent_op->update_key = true; 692 } 693 if (update->extent_op->update_flags) { 694 existing->extent_op->flags_to_set |= 695 update->extent_op->flags_to_set; 696 existing->extent_op->update_flags = true; 697 } 698 btrfs_free_delayed_extent_op(update->extent_op); 699 } 700 } 701 /* 702 * update the reference mod on the head to reflect this new operation, 703 * only need the lock for this case cause we could be processing it 704 * currently, for refs we just added we know we're a-ok. 705 */ 706 old_ref_mod = existing->total_ref_mod; 707 existing->ref_mod += update->ref_mod; 708 existing->total_ref_mod += update->ref_mod; 709 710 /* 711 * If we are going to from a positive ref mod to a negative or vice 712 * versa we need to make sure to adjust pending_csums accordingly. 713 */ 714 if (existing->is_data) { 715 u64 csum_leaves = 716 btrfs_csum_bytes_to_leaves(fs_info, 717 existing->num_bytes); 718 719 if (existing->total_ref_mod >= 0 && old_ref_mod < 0) { 720 delayed_refs->pending_csums -= existing->num_bytes; 721 btrfs_delayed_refs_rsv_release(fs_info, csum_leaves); 722 } 723 if (existing->total_ref_mod < 0 && old_ref_mod >= 0) { 724 delayed_refs->pending_csums += existing->num_bytes; 725 trans->delayed_ref_updates += csum_leaves; 726 } 727 } 728 729 /* 730 * This handles the following conditions: 731 * 732 * 1. We had a ref mod of 0 or more and went negative, indicating that 733 * we may be freeing space, so add our space to the 734 * total_bytes_pinned counter. 735 * 2. We were negative and went to 0 or positive, so no longer can say 736 * that the space would be pinned, decrement our counter from the 737 * total_bytes_pinned counter. 738 * 3. We are now at 0 and have ->must_insert_reserved set, which means 739 * this was a new allocation and then we dropped it, and thus must 740 * add our space to the total_bytes_pinned counter. 741 */ 742 if (existing->total_ref_mod < 0 && old_ref_mod >= 0) 743 btrfs_mod_total_bytes_pinned(fs_info, flags, existing->num_bytes); 744 else if (existing->total_ref_mod >= 0 && old_ref_mod < 0) 745 btrfs_mod_total_bytes_pinned(fs_info, flags, -existing->num_bytes); 746 else if (existing->total_ref_mod == 0 && existing->must_insert_reserved) 747 btrfs_mod_total_bytes_pinned(fs_info, flags, existing->num_bytes); 748 749 spin_unlock(&existing->lock); 750 } 751 752 static void init_delayed_ref_head(struct btrfs_delayed_ref_head *head_ref, 753 struct btrfs_qgroup_extent_record *qrecord, 754 u64 bytenr, u64 num_bytes, u64 ref_root, 755 u64 reserved, int action, bool is_data, 756 bool is_system) 757 { 758 int count_mod = 1; 759 int must_insert_reserved = 0; 760 761 /* If reserved is provided, it must be a data extent. */ 762 BUG_ON(!is_data && reserved); 763 764 /* 765 * The head node stores the sum of all the mods, so dropping a ref 766 * should drop the sum in the head node by one. 767 */ 768 if (action == BTRFS_UPDATE_DELAYED_HEAD) 769 count_mod = 0; 770 else if (action == BTRFS_DROP_DELAYED_REF) 771 count_mod = -1; 772 773 /* 774 * BTRFS_ADD_DELAYED_EXTENT means that we need to update the reserved 775 * accounting when the extent is finally added, or if a later 776 * modification deletes the delayed ref without ever inserting the 777 * extent into the extent allocation tree. ref->must_insert_reserved 778 * is the flag used to record that accounting mods are required. 779 * 780 * Once we record must_insert_reserved, switch the action to 781 * BTRFS_ADD_DELAYED_REF because other special casing is not required. 782 */ 783 if (action == BTRFS_ADD_DELAYED_EXTENT) 784 must_insert_reserved = 1; 785 else 786 must_insert_reserved = 0; 787 788 refcount_set(&head_ref->refs, 1); 789 head_ref->bytenr = bytenr; 790 head_ref->num_bytes = num_bytes; 791 head_ref->ref_mod = count_mod; 792 head_ref->must_insert_reserved = must_insert_reserved; 793 head_ref->is_data = is_data; 794 head_ref->is_system = is_system; 795 head_ref->ref_tree = RB_ROOT_CACHED; 796 INIT_LIST_HEAD(&head_ref->ref_add_list); 797 RB_CLEAR_NODE(&head_ref->href_node); 798 head_ref->processing = 0; 799 head_ref->total_ref_mod = count_mod; 800 spin_lock_init(&head_ref->lock); 801 mutex_init(&head_ref->mutex); 802 803 if (qrecord) { 804 if (ref_root && reserved) { 805 qrecord->data_rsv = reserved; 806 qrecord->data_rsv_refroot = ref_root; 807 } 808 qrecord->bytenr = bytenr; 809 qrecord->num_bytes = num_bytes; 810 qrecord->old_roots = NULL; 811 } 812 } 813 814 /* 815 * helper function to actually insert a head node into the rbtree. 816 * this does all the dirty work in terms of maintaining the correct 817 * overall modification count. 818 */ 819 static noinline struct btrfs_delayed_ref_head * 820 add_delayed_ref_head(struct btrfs_trans_handle *trans, 821 struct btrfs_delayed_ref_head *head_ref, 822 struct btrfs_qgroup_extent_record *qrecord, 823 int action, int *qrecord_inserted_ret) 824 { 825 struct btrfs_delayed_ref_head *existing; 826 struct btrfs_delayed_ref_root *delayed_refs; 827 int qrecord_inserted = 0; 828 829 delayed_refs = &trans->transaction->delayed_refs; 830 831 /* Record qgroup extent info if provided */ 832 if (qrecord) { 833 if (btrfs_qgroup_trace_extent_nolock(trans->fs_info, 834 delayed_refs, qrecord)) 835 kfree(qrecord); 836 else 837 qrecord_inserted = 1; 838 } 839 840 trace_add_delayed_ref_head(trans->fs_info, head_ref, action); 841 842 existing = htree_insert(&delayed_refs->href_root, 843 &head_ref->href_node); 844 if (existing) { 845 update_existing_head_ref(trans, existing, head_ref); 846 /* 847 * we've updated the existing ref, free the newly 848 * allocated ref 849 */ 850 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); 851 head_ref = existing; 852 } else { 853 u64 flags = btrfs_ref_head_to_space_flags(head_ref); 854 855 if (head_ref->is_data && head_ref->ref_mod < 0) { 856 delayed_refs->pending_csums += head_ref->num_bytes; 857 trans->delayed_ref_updates += 858 btrfs_csum_bytes_to_leaves(trans->fs_info, 859 head_ref->num_bytes); 860 } 861 if (head_ref->ref_mod < 0) 862 btrfs_mod_total_bytes_pinned(trans->fs_info, flags, 863 head_ref->num_bytes); 864 delayed_refs->num_heads++; 865 delayed_refs->num_heads_ready++; 866 atomic_inc(&delayed_refs->num_entries); 867 trans->delayed_ref_updates++; 868 } 869 if (qrecord_inserted_ret) 870 *qrecord_inserted_ret = qrecord_inserted; 871 872 return head_ref; 873 } 874 875 /* 876 * init_delayed_ref_common - Initialize the structure which represents a 877 * modification to a an extent. 878 * 879 * @fs_info: Internal to the mounted filesystem mount structure. 880 * 881 * @ref: The structure which is going to be initialized. 882 * 883 * @bytenr: The logical address of the extent for which a modification is 884 * going to be recorded. 885 * 886 * @num_bytes: Size of the extent whose modification is being recorded. 887 * 888 * @ref_root: The id of the root where this modification has originated, this 889 * can be either one of the well-known metadata trees or the 890 * subvolume id which references this extent. 891 * 892 * @action: Can be one of BTRFS_ADD_DELAYED_REF/BTRFS_DROP_DELAYED_REF or 893 * BTRFS_ADD_DELAYED_EXTENT 894 * 895 * @ref_type: Holds the type of the extent which is being recorded, can be 896 * one of BTRFS_SHARED_BLOCK_REF_KEY/BTRFS_TREE_BLOCK_REF_KEY 897 * when recording a metadata extent or BTRFS_SHARED_DATA_REF_KEY/ 898 * BTRFS_EXTENT_DATA_REF_KEY when recording data extent 899 */ 900 static void init_delayed_ref_common(struct btrfs_fs_info *fs_info, 901 struct btrfs_delayed_ref_node *ref, 902 u64 bytenr, u64 num_bytes, u64 ref_root, 903 int action, u8 ref_type) 904 { 905 u64 seq = 0; 906 907 if (action == BTRFS_ADD_DELAYED_EXTENT) 908 action = BTRFS_ADD_DELAYED_REF; 909 910 if (is_fstree(ref_root)) 911 seq = atomic64_read(&fs_info->tree_mod_seq); 912 913 refcount_set(&ref->refs, 1); 914 ref->bytenr = bytenr; 915 ref->num_bytes = num_bytes; 916 ref->ref_mod = 1; 917 ref->action = action; 918 ref->is_head = 0; 919 ref->in_tree = 1; 920 ref->seq = seq; 921 ref->type = ref_type; 922 RB_CLEAR_NODE(&ref->ref_node); 923 INIT_LIST_HEAD(&ref->add_list); 924 } 925 926 /* 927 * add a delayed tree ref. This does all of the accounting required 928 * to make sure the delayed ref is eventually processed before this 929 * transaction commits. 930 */ 931 int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans, 932 struct btrfs_ref *generic_ref, 933 struct btrfs_delayed_extent_op *extent_op) 934 { 935 struct btrfs_fs_info *fs_info = trans->fs_info; 936 struct btrfs_delayed_tree_ref *ref; 937 struct btrfs_delayed_ref_head *head_ref; 938 struct btrfs_delayed_ref_root *delayed_refs; 939 struct btrfs_qgroup_extent_record *record = NULL; 940 int qrecord_inserted; 941 bool is_system; 942 int action = generic_ref->action; 943 int level = generic_ref->tree_ref.level; 944 int ret; 945 u64 bytenr = generic_ref->bytenr; 946 u64 num_bytes = generic_ref->len; 947 u64 parent = generic_ref->parent; 948 u8 ref_type; 949 950 is_system = (generic_ref->real_root == BTRFS_CHUNK_TREE_OBJECTID); 951 952 ASSERT(generic_ref->type == BTRFS_REF_METADATA && generic_ref->action); 953 BUG_ON(extent_op && extent_op->is_data); 954 ref = kmem_cache_alloc(btrfs_delayed_tree_ref_cachep, GFP_NOFS); 955 if (!ref) 956 return -ENOMEM; 957 958 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); 959 if (!head_ref) { 960 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); 961 return -ENOMEM; 962 } 963 964 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) && 965 is_fstree(generic_ref->real_root) && 966 is_fstree(generic_ref->tree_ref.root) && 967 !generic_ref->skip_qgroup) { 968 record = kzalloc(sizeof(*record), GFP_NOFS); 969 if (!record) { 970 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); 971 kmem_cache_free(btrfs_delayed_ref_head_cachep, head_ref); 972 return -ENOMEM; 973 } 974 } 975 976 if (parent) 977 ref_type = BTRFS_SHARED_BLOCK_REF_KEY; 978 else 979 ref_type = BTRFS_TREE_BLOCK_REF_KEY; 980 981 init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes, 982 generic_ref->tree_ref.root, action, ref_type); 983 ref->root = generic_ref->tree_ref.root; 984 ref->parent = parent; 985 ref->level = level; 986 987 init_delayed_ref_head(head_ref, record, bytenr, num_bytes, 988 generic_ref->tree_ref.root, 0, action, false, 989 is_system); 990 head_ref->extent_op = extent_op; 991 992 delayed_refs = &trans->transaction->delayed_refs; 993 spin_lock(&delayed_refs->lock); 994 995 /* 996 * insert both the head node and the new ref without dropping 997 * the spin lock 998 */ 999 head_ref = add_delayed_ref_head(trans, head_ref, record, 1000 action, &qrecord_inserted); 1001 1002 ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node); 1003 spin_unlock(&delayed_refs->lock); 1004 1005 /* 1006 * Need to update the delayed_refs_rsv with any changes we may have 1007 * made. 1008 */ 1009 btrfs_update_delayed_refs_rsv(trans); 1010 1011 trace_add_delayed_tree_ref(fs_info, &ref->node, ref, 1012 action == BTRFS_ADD_DELAYED_EXTENT ? 1013 BTRFS_ADD_DELAYED_REF : action); 1014 if (ret > 0) 1015 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); 1016 1017 if (qrecord_inserted) 1018 btrfs_qgroup_trace_extent_post(fs_info, record); 1019 1020 return 0; 1021 } 1022 1023 /* 1024 * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref. 1025 */ 1026 int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans, 1027 struct btrfs_ref *generic_ref, 1028 u64 reserved) 1029 { 1030 struct btrfs_fs_info *fs_info = trans->fs_info; 1031 struct btrfs_delayed_data_ref *ref; 1032 struct btrfs_delayed_ref_head *head_ref; 1033 struct btrfs_delayed_ref_root *delayed_refs; 1034 struct btrfs_qgroup_extent_record *record = NULL; 1035 int qrecord_inserted; 1036 int action = generic_ref->action; 1037 int ret; 1038 u64 bytenr = generic_ref->bytenr; 1039 u64 num_bytes = generic_ref->len; 1040 u64 parent = generic_ref->parent; 1041 u64 ref_root = generic_ref->data_ref.ref_root; 1042 u64 owner = generic_ref->data_ref.ino; 1043 u64 offset = generic_ref->data_ref.offset; 1044 u8 ref_type; 1045 1046 ASSERT(generic_ref->type == BTRFS_REF_DATA && action); 1047 ref = kmem_cache_alloc(btrfs_delayed_data_ref_cachep, GFP_NOFS); 1048 if (!ref) 1049 return -ENOMEM; 1050 1051 if (parent) 1052 ref_type = BTRFS_SHARED_DATA_REF_KEY; 1053 else 1054 ref_type = BTRFS_EXTENT_DATA_REF_KEY; 1055 init_delayed_ref_common(fs_info, &ref->node, bytenr, num_bytes, 1056 ref_root, action, ref_type); 1057 ref->root = ref_root; 1058 ref->parent = parent; 1059 ref->objectid = owner; 1060 ref->offset = offset; 1061 1062 1063 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); 1064 if (!head_ref) { 1065 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); 1066 return -ENOMEM; 1067 } 1068 1069 if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags) && 1070 is_fstree(ref_root) && 1071 is_fstree(generic_ref->real_root) && 1072 !generic_ref->skip_qgroup) { 1073 record = kzalloc(sizeof(*record), GFP_NOFS); 1074 if (!record) { 1075 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); 1076 kmem_cache_free(btrfs_delayed_ref_head_cachep, 1077 head_ref); 1078 return -ENOMEM; 1079 } 1080 } 1081 1082 init_delayed_ref_head(head_ref, record, bytenr, num_bytes, ref_root, 1083 reserved, action, true, false); 1084 head_ref->extent_op = NULL; 1085 1086 delayed_refs = &trans->transaction->delayed_refs; 1087 spin_lock(&delayed_refs->lock); 1088 1089 /* 1090 * insert both the head node and the new ref without dropping 1091 * the spin lock 1092 */ 1093 head_ref = add_delayed_ref_head(trans, head_ref, record, 1094 action, &qrecord_inserted); 1095 1096 ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node); 1097 spin_unlock(&delayed_refs->lock); 1098 1099 /* 1100 * Need to update the delayed_refs_rsv with any changes we may have 1101 * made. 1102 */ 1103 btrfs_update_delayed_refs_rsv(trans); 1104 1105 trace_add_delayed_data_ref(trans->fs_info, &ref->node, ref, 1106 action == BTRFS_ADD_DELAYED_EXTENT ? 1107 BTRFS_ADD_DELAYED_REF : action); 1108 if (ret > 0) 1109 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); 1110 1111 1112 if (qrecord_inserted) 1113 return btrfs_qgroup_trace_extent_post(fs_info, record); 1114 return 0; 1115 } 1116 1117 int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans, 1118 u64 bytenr, u64 num_bytes, 1119 struct btrfs_delayed_extent_op *extent_op) 1120 { 1121 struct btrfs_delayed_ref_head *head_ref; 1122 struct btrfs_delayed_ref_root *delayed_refs; 1123 1124 head_ref = kmem_cache_alloc(btrfs_delayed_ref_head_cachep, GFP_NOFS); 1125 if (!head_ref) 1126 return -ENOMEM; 1127 1128 init_delayed_ref_head(head_ref, NULL, bytenr, num_bytes, 0, 0, 1129 BTRFS_UPDATE_DELAYED_HEAD, extent_op->is_data, 1130 false); 1131 head_ref->extent_op = extent_op; 1132 1133 delayed_refs = &trans->transaction->delayed_refs; 1134 spin_lock(&delayed_refs->lock); 1135 1136 add_delayed_ref_head(trans, head_ref, NULL, BTRFS_UPDATE_DELAYED_HEAD, 1137 NULL); 1138 1139 spin_unlock(&delayed_refs->lock); 1140 1141 /* 1142 * Need to update the delayed_refs_rsv with any changes we may have 1143 * made. 1144 */ 1145 btrfs_update_delayed_refs_rsv(trans); 1146 return 0; 1147 } 1148 1149 /* 1150 * This does a simple search for the head node for a given extent. Returns the 1151 * head node if found, or NULL if not. 1152 */ 1153 struct btrfs_delayed_ref_head * 1154 btrfs_find_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, u64 bytenr) 1155 { 1156 lockdep_assert_held(&delayed_refs->lock); 1157 1158 return find_ref_head(delayed_refs, bytenr, false); 1159 } 1160 1161 void __cold btrfs_delayed_ref_exit(void) 1162 { 1163 kmem_cache_destroy(btrfs_delayed_ref_head_cachep); 1164 kmem_cache_destroy(btrfs_delayed_tree_ref_cachep); 1165 kmem_cache_destroy(btrfs_delayed_data_ref_cachep); 1166 kmem_cache_destroy(btrfs_delayed_extent_op_cachep); 1167 } 1168 1169 int __init btrfs_delayed_ref_init(void) 1170 { 1171 btrfs_delayed_ref_head_cachep = kmem_cache_create( 1172 "btrfs_delayed_ref_head", 1173 sizeof(struct btrfs_delayed_ref_head), 0, 1174 SLAB_MEM_SPREAD, NULL); 1175 if (!btrfs_delayed_ref_head_cachep) 1176 goto fail; 1177 1178 btrfs_delayed_tree_ref_cachep = kmem_cache_create( 1179 "btrfs_delayed_tree_ref", 1180 sizeof(struct btrfs_delayed_tree_ref), 0, 1181 SLAB_MEM_SPREAD, NULL); 1182 if (!btrfs_delayed_tree_ref_cachep) 1183 goto fail; 1184 1185 btrfs_delayed_data_ref_cachep = kmem_cache_create( 1186 "btrfs_delayed_data_ref", 1187 sizeof(struct btrfs_delayed_data_ref), 0, 1188 SLAB_MEM_SPREAD, NULL); 1189 if (!btrfs_delayed_data_ref_cachep) 1190 goto fail; 1191 1192 btrfs_delayed_extent_op_cachep = kmem_cache_create( 1193 "btrfs_delayed_extent_op", 1194 sizeof(struct btrfs_delayed_extent_op), 0, 1195 SLAB_MEM_SPREAD, NULL); 1196 if (!btrfs_delayed_extent_op_cachep) 1197 goto fail; 1198 1199 return 0; 1200 fail: 1201 btrfs_delayed_ref_exit(); 1202 return -ENOMEM; 1203 } 1204