fs/btrfs/delayed-ref.c

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