fs/btrfs/delayed-inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2011 Fujitsu.  All rights reserved.
   4  * Written by Miao Xie <miaox@cn.fujitsu.com>
   5  */
   6
   7 #include <linux/slab.h>
   8 #include <linux/iversion.h>
   9 #include <linux/sched/mm.h>
  10 #include "misc.h"
  11 #include "delayed-inode.h"
  12 #include "disk-io.h"
  13 #include "transaction.h"
  14 #include "ctree.h"
  15 #include "qgroup.h"
  16 #include "locking.h"
  17
  18 #define BTRFS_DELAYED_WRITEBACK         512
  19 #define BTRFS_DELAYED_BACKGROUND        128
  20 #define BTRFS_DELAYED_BATCH             16
  21
  22 static struct kmem_cache *delayed_node_cache;
  23
  24 int __init btrfs_delayed_inode_init(void)
  25 {
  26         delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
  27                                         sizeof(struct btrfs_delayed_node),
  28                                         0,
  29                                         SLAB_MEM_SPREAD,
  30                                         NULL);
  31         if (!delayed_node_cache)
  32                 return -ENOMEM;
  33         return 0;
  34 }
  35
  36 void __cold btrfs_delayed_inode_exit(void)
  37 {
  38         kmem_cache_destroy(delayed_node_cache);
  39 }
  40
  41 static inline void btrfs_init_delayed_node(
  42                                 struct btrfs_delayed_node *delayed_node,
  43                                 struct btrfs_root *root, u64 inode_id)
  44 {
  45         delayed_node->root = root;
  46         delayed_node->inode_id = inode_id;
  47         refcount_set(&delayed_node->refs, 0);
  48         delayed_node->ins_root = RB_ROOT_CACHED;
  49         delayed_node->del_root = RB_ROOT_CACHED;
  50         mutex_init(&delayed_node->mutex);
  51         INIT_LIST_HEAD(&delayed_node->n_list);
  52         INIT_LIST_HEAD(&delayed_node->p_list);
  53 }
  54
  55 static inline int btrfs_is_continuous_delayed_item(
  56                                         struct btrfs_delayed_item *item1,
  57                                         struct btrfs_delayed_item *item2)
  58 {
  59         if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  60             item1->key.objectid == item2->key.objectid &&
  61             item1->key.type == item2->key.type &&
  62             item1->key.offset + 1 == item2->key.offset)
  63                 return 1;
  64         return 0;
  65 }
  66
  67 static struct btrfs_delayed_node *btrfs_get_delayed_node(
  68                 struct btrfs_inode *btrfs_inode)
  69 {
  70         struct btrfs_root *root = btrfs_inode->root;
  71         u64 ino = btrfs_ino(btrfs_inode);
  72         struct btrfs_delayed_node *node;
  73
  74         node = READ_ONCE(btrfs_inode->delayed_node);
  75         if (node) {
  76                 refcount_inc(&node->refs);
  77                 return node;
  78         }
  79
  80         spin_lock(&root->inode_lock);
  81         node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
  82
  83         if (node) {
  84                 if (btrfs_inode->delayed_node) {
  85                         refcount_inc(&node->refs);      /* can be accessed */
  86                         BUG_ON(btrfs_inode->delayed_node != node);
  87                         spin_unlock(&root->inode_lock);
  88                         return node;
  89                 }
  90
  91                 /*
  92                  * It's possible that we're racing into the middle of removing
  93                  * this node from the radix tree.  In this case, the refcount
  94                  * was zero and it should never go back to one.  Just return
  95                  * NULL like it was never in the radix at all; our release
  96                  * function is in the process of removing it.
  97                  *
  98                  * Some implementations of refcount_inc refuse to bump the
  99                  * refcount once it has hit zero.  If we don't do this dance
 100                  * here, refcount_inc() may decide to just WARN_ONCE() instead
 101                  * of actually bumping the refcount.
 102                  *
 103                  * If this node is properly in the radix, we want to bump the
 104                  * refcount twice, once for the inode and once for this get
 105                  * operation.
 106                  */
 107                 if (refcount_inc_not_zero(&node->refs)) {
 108                         refcount_inc(&node->refs);
 109                         btrfs_inode->delayed_node = node;
 110                 } else {
 111                         node = NULL;
 112                 }
 113
 114                 spin_unlock(&root->inode_lock);
 115                 return node;
 116         }
 117         spin_unlock(&root->inode_lock);
 118
 119         return NULL;
 120 }
 121
 122 /* Will return either the node or PTR_ERR(-ENOMEM) */
 123 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 124                 struct btrfs_inode *btrfs_inode)
 125 {
 126         struct btrfs_delayed_node *node;
 127         struct btrfs_root *root = btrfs_inode->root;
 128         u64 ino = btrfs_ino(btrfs_inode);
 129         int ret;
 130
 131 again:
 132         node = btrfs_get_delayed_node(btrfs_inode);
 133         if (node)
 134                 return node;
 135
 136         node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
 137         if (!node)
 138                 return ERR_PTR(-ENOMEM);
 139         btrfs_init_delayed_node(node, root, ino);
 140
 141         /* cached in the btrfs inode and can be accessed */
 142         refcount_set(&node->refs, 2);
 143
 144         ret = radix_tree_preload(GFP_NOFS);
 145         if (ret) {
 146                 kmem_cache_free(delayed_node_cache, node);
 147                 return ERR_PTR(ret);
 148         }
 149
 150         spin_lock(&root->inode_lock);
 151         ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 152         if (ret == -EEXIST) {
 153                 spin_unlock(&root->inode_lock);
 154                 kmem_cache_free(delayed_node_cache, node);
 155                 radix_tree_preload_end();
 156                 goto again;
 157         }
 158         btrfs_inode->delayed_node = node;
 159         spin_unlock(&root->inode_lock);
 160         radix_tree_preload_end();
 161
 162         return node;
 163 }
 164
 165 /*
 166  * Call it when holding delayed_node->mutex
 167  *
 168  * If mod = 1, add this node into the prepared list.
 169  */
 170 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 171                                      struct btrfs_delayed_node *node,
 172                                      int mod)
 173 {
 174         spin_lock(&root->lock);
 175         if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 176                 if (!list_empty(&node->p_list))
 177                         list_move_tail(&node->p_list, &root->prepare_list);
 178                 else if (mod)
 179                         list_add_tail(&node->p_list, &root->prepare_list);
 180         } else {
 181                 list_add_tail(&node->n_list, &root->node_list);
 182                 list_add_tail(&node->p_list, &root->prepare_list);
 183                 refcount_inc(&node->refs);      /* inserted into list */
 184                 root->nodes++;
 185                 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 186         }
 187         spin_unlock(&root->lock);
 188 }
 189
 190 /* Call it when holding delayed_node->mutex */
 191 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 192                                        struct btrfs_delayed_node *node)
 193 {
 194         spin_lock(&root->lock);
 195         if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 196                 root->nodes--;
 197                 refcount_dec(&node->refs);      /* not in the list */
 198                 list_del_init(&node->n_list);
 199                 if (!list_empty(&node->p_list))
 200                         list_del_init(&node->p_list);
 201                 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 202         }
 203         spin_unlock(&root->lock);
 204 }
 205
 206 static struct btrfs_delayed_node *btrfs_first_delayed_node(
 207                         struct btrfs_delayed_root *delayed_root)
 208 {
 209         struct list_head *p;
 210         struct btrfs_delayed_node *node = NULL;
 211
 212         spin_lock(&delayed_root->lock);
 213         if (list_empty(&delayed_root->node_list))
 214                 goto out;
 215
 216         p = delayed_root->node_list.next;
 217         node = list_entry(p, struct btrfs_delayed_node, n_list);
 218         refcount_inc(&node->refs);
 219 out:
 220         spin_unlock(&delayed_root->lock);
 221
 222         return node;
 223 }
 224
 225 static struct btrfs_delayed_node *btrfs_next_delayed_node(
 226                                                 struct btrfs_delayed_node *node)
 227 {
 228         struct btrfs_delayed_root *delayed_root;
 229         struct list_head *p;
 230         struct btrfs_delayed_node *next = NULL;
 231
 232         delayed_root = node->root->fs_info->delayed_root;
 233         spin_lock(&delayed_root->lock);
 234         if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 235                 /* not in the list */
 236                 if (list_empty(&delayed_root->node_list))
 237                         goto out;
 238                 p = delayed_root->node_list.next;
 239         } else if (list_is_last(&node->n_list, &delayed_root->node_list))
 240                 goto out;
 241         else
 242                 p = node->n_list.next;
 243
 244         next = list_entry(p, struct btrfs_delayed_node, n_list);
 245         refcount_inc(&next->refs);
 246 out:
 247         spin_unlock(&delayed_root->lock);
 248
 249         return next;
 250 }
 251
 252 static void __btrfs_release_delayed_node(
 253                                 struct btrfs_delayed_node *delayed_node,
 254                                 int mod)
 255 {
 256         struct btrfs_delayed_root *delayed_root;
 257
 258         if (!delayed_node)
 259                 return;
 260
 261         delayed_root = delayed_node->root->fs_info->delayed_root;
 262
 263         mutex_lock(&delayed_node->mutex);
 264         if (delayed_node->count)
 265                 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
 266         else
 267                 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 268         mutex_unlock(&delayed_node->mutex);
 269
 270         if (refcount_dec_and_test(&delayed_node->refs)) {
 271                 struct btrfs_root *root = delayed_node->root;
 272
 273                 spin_lock(&root->inode_lock);
 274                 /*
 275                  * Once our refcount goes to zero, nobody is allowed to bump it
 276                  * back up.  We can delete it now.
 277                  */
 278                 ASSERT(refcount_read(&delayed_node->refs) == 0);
 279                 radix_tree_delete(&root->delayed_nodes_tree,
 280                                   delayed_node->inode_id);
 281                 spin_unlock(&root->inode_lock);
 282                 kmem_cache_free(delayed_node_cache, delayed_node);
 283         }
 284 }
 285
 286 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
 287 {
 288         __btrfs_release_delayed_node(node, 0);
 289 }
 290
 291 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
 292                                         struct btrfs_delayed_root *delayed_root)
 293 {
 294         struct list_head *p;
 295         struct btrfs_delayed_node *node = NULL;
 296
 297         spin_lock(&delayed_root->lock);
 298         if (list_empty(&delayed_root->prepare_list))
 299                 goto out;
 300
 301         p = delayed_root->prepare_list.next;
 302         list_del_init(p);
 303         node = list_entry(p, struct btrfs_delayed_node, p_list);
 304         refcount_inc(&node->refs);
 305 out:
 306         spin_unlock(&delayed_root->lock);
 307
 308         return node;
 309 }
 310
 311 static inline void btrfs_release_prepared_delayed_node(
 312                                         struct btrfs_delayed_node *node)
 313 {
 314         __btrfs_release_delayed_node(node, 1);
 315 }
 316
 317 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
 318 {
 319         struct btrfs_delayed_item *item;
 320         item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
 321         if (item) {
 322                 item->data_len = data_len;
 323                 item->ins_or_del = 0;
 324                 item->bytes_reserved = 0;
 325                 item->delayed_node = NULL;
 326                 refcount_set(&item->refs, 1);
 327         }
 328         return item;
 329 }
 330
 331 /*
 332  * __btrfs_lookup_delayed_item - look up the delayed item by key
 333  * @delayed_node: pointer to the delayed node
 334  * @key:          the key to look up
 335  * @prev:         used to store the prev item if the right item isn't found
 336  * @next:         used to store the next item if the right item isn't found
 337  *
 338  * Note: if we don't find the right item, we will return the prev item and
 339  * the next item.
 340  */
 341 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 342                                 struct rb_root *root,
 343                                 struct btrfs_key *key,
 344                                 struct btrfs_delayed_item **prev,
 345                                 struct btrfs_delayed_item **next)
 346 {
 347         struct rb_node *node, *prev_node = NULL;
 348         struct btrfs_delayed_item *delayed_item = NULL;
 349         int ret = 0;
 350
 351         node = root->rb_node;
 352
 353         while (node) {
 354                 delayed_item = rb_entry(node, struct btrfs_delayed_item,
 355                                         rb_node);
 356                 prev_node = node;
 357                 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
 358                 if (ret < 0)
 359                         node = node->rb_right;
 360                 else if (ret > 0)
 361                         node = node->rb_left;
 362                 else
 363                         return delayed_item;
 364         }
 365
 366         if (prev) {
 367                 if (!prev_node)
 368                         *prev = NULL;
 369                 else if (ret < 0)
 370                         *prev = delayed_item;
 371                 else if ((node = rb_prev(prev_node)) != NULL) {
 372                         *prev = rb_entry(node, struct btrfs_delayed_item,
 373                                          rb_node);
 374                 } else
 375                         *prev = NULL;
 376         }
 377
 378         if (next) {
 379                 if (!prev_node)
 380                         *next = NULL;
 381                 else if (ret > 0)
 382                         *next = delayed_item;
 383                 else if ((node = rb_next(prev_node)) != NULL) {
 384                         *next = rb_entry(node, struct btrfs_delayed_item,
 385                                          rb_node);
 386                 } else
 387                         *next = NULL;
 388         }
 389         return NULL;
 390 }
 391
 392 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
 393                                         struct btrfs_delayed_node *delayed_node,
 394                                         struct btrfs_key *key)
 395 {
 396         return __btrfs_lookup_delayed_item(&delayed_node->ins_root.rb_root, key,
 397                                            NULL, NULL);
 398 }
 399
 400 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
 401                                     struct btrfs_delayed_item *ins,
 402                                     int action)
 403 {
 404         struct rb_node **p, *node;
 405         struct rb_node *parent_node = NULL;
 406         struct rb_root_cached *root;
 407         struct btrfs_delayed_item *item;
 408         int cmp;
 409         bool leftmost = true;
 410
 411         if (action == BTRFS_DELAYED_INSERTION_ITEM)
 412                 root = &delayed_node->ins_root;
 413         else if (action == BTRFS_DELAYED_DELETION_ITEM)
 414                 root = &delayed_node->del_root;
 415         else
 416                 BUG();
 417         p = &root->rb_root.rb_node;
 418         node = &ins->rb_node;
 419
 420         while (*p) {
 421                 parent_node = *p;
 422                 item = rb_entry(parent_node, struct btrfs_delayed_item,
 423                                  rb_node);
 424
 425                 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
 426                 if (cmp < 0) {
 427                         p = &(*p)->rb_right;
 428                         leftmost = false;
 429                 } else if (cmp > 0) {
 430                         p = &(*p)->rb_left;
 431                 } else {
 432                         return -EEXIST;
 433                 }
 434         }
 435
 436         rb_link_node(node, parent_node, p);
 437         rb_insert_color_cached(node, root, leftmost);
 438         ins->delayed_node = delayed_node;
 439         ins->ins_or_del = action;
 440
 441         if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
 442             action == BTRFS_DELAYED_INSERTION_ITEM &&
 443             ins->key.offset >= delayed_node->index_cnt)
 444                         delayed_node->index_cnt = ins->key.offset + 1;
 445
 446         delayed_node->count++;
 447         atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
 448         return 0;
 449 }
 450
 451 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
 452                                               struct btrfs_delayed_item *item)
 453 {
 454         return __btrfs_add_delayed_item(node, item,
 455                                         BTRFS_DELAYED_INSERTION_ITEM);
 456 }
 457
 458 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
 459                                              struct btrfs_delayed_item *item)
 460 {
 461         return __btrfs_add_delayed_item(node, item,
 462                                         BTRFS_DELAYED_DELETION_ITEM);
 463 }
 464
 465 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
 466 {
 467         int seq = atomic_inc_return(&delayed_root->items_seq);
 468
 469         /* atomic_dec_return implies a barrier */
 470         if ((atomic_dec_return(&delayed_root->items) <
 471             BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
 472                 cond_wake_up_nomb(&delayed_root->wait);
 473 }
 474
 475 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 476 {
 477         struct rb_root_cached *root;
 478         struct btrfs_delayed_root *delayed_root;
 479
 480         /* Not associated with any delayed_node */
 481         if (!delayed_item->delayed_node)
 482                 return;
 483         delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
 484
 485         BUG_ON(!delayed_root);
 486         BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
 487                delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
 488
 489         if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
 490                 root = &delayed_item->delayed_node->ins_root;
 491         else
 492                 root = &delayed_item->delayed_node->del_root;
 493
 494         rb_erase_cached(&delayed_item->rb_node, root);
 495         delayed_item->delayed_node->count--;
 496
 497         finish_one_item(delayed_root);
 498 }
 499
 500 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 501 {
 502         if (item) {
 503                 __btrfs_remove_delayed_item(item);
 504                 if (refcount_dec_and_test(&item->refs))
 505                         kfree(item);
 506         }
 507 }
 508
 509 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 510                                         struct btrfs_delayed_node *delayed_node)
 511 {
 512         struct rb_node *p;
 513         struct btrfs_delayed_item *item = NULL;
 514
 515         p = rb_first_cached(&delayed_node->ins_root);
 516         if (p)
 517                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 518
 519         return item;
 520 }
 521
 522 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 523                                         struct btrfs_delayed_node *delayed_node)
 524 {
 525         struct rb_node *p;
 526         struct btrfs_delayed_item *item = NULL;
 527
 528         p = rb_first_cached(&delayed_node->del_root);
 529         if (p)
 530                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 531
 532         return item;
 533 }
 534
 535 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
 536                                                 struct btrfs_delayed_item *item)
 537 {
 538         struct rb_node *p;
 539         struct btrfs_delayed_item *next = NULL;
 540
 541         p = rb_next(&item->rb_node);
 542         if (p)
 543                 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
 544
 545         return next;
 546 }
 547
 548 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 549                                                struct btrfs_root *root,
 550                                                struct btrfs_delayed_item *item)
 551 {
 552         struct btrfs_block_rsv *src_rsv;
 553         struct btrfs_block_rsv *dst_rsv;
 554         struct btrfs_fs_info *fs_info = root->fs_info;
 555         u64 num_bytes;
 556         int ret;
 557
 558         if (!trans->bytes_reserved)
 559                 return 0;
 560
 561         src_rsv = trans->block_rsv;
 562         dst_rsv = &fs_info->delayed_block_rsv;
 563
 564         num_bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
 565
 566         /*
 567          * Here we migrate space rsv from transaction rsv, since have already
 568          * reserved space when starting a transaction.  So no need to reserve
 569          * qgroup space here.
 570          */
 571         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
 572         if (!ret) {
 573                 trace_btrfs_space_reservation(fs_info, "delayed_item",
 574                                               item->key.objectid,
 575                                               num_bytes, 1);
 576                 item->bytes_reserved = num_bytes;
 577         }
 578
 579         return ret;
 580 }
 581
 582 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 583                                                 struct btrfs_delayed_item *item)
 584 {
 585         struct btrfs_block_rsv *rsv;
 586         struct btrfs_fs_info *fs_info = root->fs_info;
 587
 588         if (!item->bytes_reserved)
 589                 return;
 590
 591         rsv = &fs_info->delayed_block_rsv;
 592         /*
 593          * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
 594          * to release/reserve qgroup space.
 595          */
 596         trace_btrfs_space_reservation(fs_info, "delayed_item",
 597                                       item->key.objectid, item->bytes_reserved,
 598                                       0);
 599         btrfs_block_rsv_release(fs_info, rsv, item->bytes_reserved, NULL);
 600 }
 601
 602 static int btrfs_delayed_inode_reserve_metadata(
 603                                         struct btrfs_trans_handle *trans,
 604                                         struct btrfs_root *root,
 605                                         struct btrfs_inode *inode,
 606                                         struct btrfs_delayed_node *node)
 607 {
 608         struct btrfs_fs_info *fs_info = root->fs_info;
 609         struct btrfs_block_rsv *src_rsv;
 610         struct btrfs_block_rsv *dst_rsv;
 611         u64 num_bytes;
 612         int ret;
 613
 614         src_rsv = trans->block_rsv;
 615         dst_rsv = &fs_info->delayed_block_rsv;
 616
 617         num_bytes = btrfs_calc_metadata_size(fs_info, 1);
 618
 619         /*
 620          * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 621          * which doesn't reserve space for speed.  This is a problem since we
 622          * still need to reserve space for this update, so try to reserve the
 623          * space.
 624          *
 625          * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 626          * we always reserve enough to update the inode item.
 627          */
 628         if (!src_rsv || (!trans->bytes_reserved &&
 629                          src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
 630                 ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
 631                 if (ret < 0)
 632                         return ret;
 633                 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 634                                           BTRFS_RESERVE_NO_FLUSH);
 635                 /*
 636                  * Since we're under a transaction reserve_metadata_bytes could
 637                  * try to commit the transaction which will make it return
 638                  * EAGAIN to make us stop the transaction we have, so return
 639                  * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 640                  */
 641                 if (ret == -EAGAIN) {
 642                         ret = -ENOSPC;
 643                         btrfs_qgroup_free_meta_prealloc(root, num_bytes);
 644                 }
 645                 if (!ret) {
 646                         node->bytes_reserved = num_bytes;
 647                         trace_btrfs_space_reservation(fs_info,
 648                                                       "delayed_inode",
 649                                                       btrfs_ino(inode),
 650                                                       num_bytes, 1);
 651                 } else {
 652                         btrfs_qgroup_free_meta_prealloc(root, fs_info->nodesize);
 653                 }
 654                 return ret;
 655         }
 656
 657         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
 658         if (!ret) {
 659                 trace_btrfs_space_reservation(fs_info, "delayed_inode",
 660                                               btrfs_ino(inode), num_bytes, 1);
 661                 node->bytes_reserved = num_bytes;
 662         }
 663
 664         return ret;
 665 }
 666
 667 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
 668                                                 struct btrfs_delayed_node *node,
 669                                                 bool qgroup_free)
 670 {
 671         struct btrfs_block_rsv *rsv;
 672
 673         if (!node->bytes_reserved)
 674                 return;
 675
 676         rsv = &fs_info->delayed_block_rsv;
 677         trace_btrfs_space_reservation(fs_info, "delayed_inode",
 678                                       node->inode_id, node->bytes_reserved, 0);
 679         btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
 680         if (qgroup_free)
 681                 btrfs_qgroup_free_meta_prealloc(node->root,
 682                                 node->bytes_reserved);
 683         else
 684                 btrfs_qgroup_convert_reserved_meta(node->root,
 685                                 node->bytes_reserved);
 686         node->bytes_reserved = 0;
 687 }
 688
 689 /*
 690  * This helper will insert some continuous items into the same leaf according
 691  * to the free space of the leaf.
 692  */
 693 static int btrfs_batch_insert_items(struct btrfs_root *root,
 694                                     struct btrfs_path *path,
 695                                     struct btrfs_delayed_item *item)
 696 {
 697         struct btrfs_delayed_item *curr, *next;
 698         int free_space;
 699         int total_data_size = 0, total_size = 0;
 700         struct extent_buffer *leaf;
 701         char *data_ptr;
 702         struct btrfs_key *keys;
 703         u32 *data_size;
 704         struct list_head head;
 705         int slot;
 706         int nitems;
 707         int i;
 708         int ret = 0;
 709
 710         BUG_ON(!path->nodes[0]);
 711
 712         leaf = path->nodes[0];
 713         free_space = btrfs_leaf_free_space(leaf);
 714         INIT_LIST_HEAD(&head);
 715
 716         next = item;
 717         nitems = 0;
 718
 719         /*
 720          * count the number of the continuous items that we can insert in batch
 721          */
 722         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 723                free_space) {
 724                 total_data_size += next->data_len;
 725                 total_size += next->data_len + sizeof(struct btrfs_item);
 726                 list_add_tail(&next->tree_list, &head);
 727                 nitems++;
 728
 729                 curr = next;
 730                 next = __btrfs_next_delayed_item(curr);
 731                 if (!next)
 732                         break;
 733
 734                 if (!btrfs_is_continuous_delayed_item(curr, next))
 735                         break;
 736         }
 737
 738         if (!nitems) {
 739                 ret = 0;
 740                 goto out;
 741         }
 742
 743         /*
 744          * we need allocate some memory space, but it might cause the task
 745          * to sleep, so we set all locked nodes in the path to blocking locks
 746          * first.
 747          */
 748         btrfs_set_path_blocking(path);
 749
 750         keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
 751         if (!keys) {
 752                 ret = -ENOMEM;
 753                 goto out;
 754         }
 755
 756         data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
 757         if (!data_size) {
 758                 ret = -ENOMEM;
 759                 goto error;
 760         }
 761
 762         /* get keys of all the delayed items */
 763         i = 0;
 764         list_for_each_entry(next, &head, tree_list) {
 765                 keys[i] = next->key;
 766                 data_size[i] = next->data_len;
 767                 i++;
 768         }
 769
 770         /* insert the keys of the items */
 771         setup_items_for_insert(root, path, keys, data_size, nitems);
 772
 773         /* insert the dir index items */
 774         slot = path->slots[0];
 775         list_for_each_entry_safe(curr, next, &head, tree_list) {
 776                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 777                 write_extent_buffer(leaf, &curr->data,
 778                                     (unsigned long)data_ptr,
 779                                     curr->data_len);
 780                 slot++;
 781
 782                 btrfs_delayed_item_release_metadata(root, curr);
 783
 784                 list_del(&curr->tree_list);
 785                 btrfs_release_delayed_item(curr);
 786         }
 787
 788 error:
 789         kfree(data_size);
 790         kfree(keys);
 791 out:
 792         return ret;
 793 }
 794
 795 /*
 796  * This helper can just do simple insertion that needn't extend item for new
 797  * data, such as directory name index insertion, inode insertion.
 798  */
 799 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 800                                      struct btrfs_root *root,
 801                                      struct btrfs_path *path,
 802                                      struct btrfs_delayed_item *delayed_item)
 803 {
 804         struct extent_buffer *leaf;
 805         unsigned int nofs_flag;
 806         char *ptr;
 807         int ret;
 808
 809         nofs_flag = memalloc_nofs_save();
 810         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 811                                       delayed_item->data_len);
 812         memalloc_nofs_restore(nofs_flag);
 813         if (ret < 0 && ret != -EEXIST)
 814                 return ret;
 815
 816         leaf = path->nodes[0];
 817
 818         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 819
 820         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 821                             delayed_item->data_len);
 822         btrfs_mark_buffer_dirty(leaf);
 823
 824         btrfs_delayed_item_release_metadata(root, delayed_item);
 825         return 0;
 826 }
 827
 828 /*
 829  * we insert an item first, then if there are some continuous items, we try
 830  * to insert those items into the same leaf.
 831  */
 832 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 833                                       struct btrfs_path *path,
 834                                       struct btrfs_root *root,
 835                                       struct btrfs_delayed_node *node)
 836 {
 837         struct btrfs_delayed_item *curr, *prev;
 838         int ret = 0;
 839
 840 do_again:
 841         mutex_lock(&node->mutex);
 842         curr = __btrfs_first_delayed_insertion_item(node);
 843         if (!curr)
 844                 goto insert_end;
 845
 846         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 847         if (ret < 0) {
 848                 btrfs_release_path(path);
 849                 goto insert_end;
 850         }
 851
 852         prev = curr;
 853         curr = __btrfs_next_delayed_item(prev);
 854         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 855                 /* insert the continuous items into the same leaf */
 856                 path->slots[0]++;
 857                 btrfs_batch_insert_items(root, path, curr);
 858         }
 859         btrfs_release_delayed_item(prev);
 860         btrfs_mark_buffer_dirty(path->nodes[0]);
 861
 862         btrfs_release_path(path);
 863         mutex_unlock(&node->mutex);
 864         goto do_again;
 865
 866 insert_end:
 867         mutex_unlock(&node->mutex);
 868         return ret;
 869 }
 870
 871 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 872                                     struct btrfs_root *root,
 873                                     struct btrfs_path *path,
 874                                     struct btrfs_delayed_item *item)
 875 {
 876         struct btrfs_delayed_item *curr, *next;
 877         struct extent_buffer *leaf;
 878         struct btrfs_key key;
 879         struct list_head head;
 880         int nitems, i, last_item;
 881         int ret = 0;
 882
 883         BUG_ON(!path->nodes[0]);
 884
 885         leaf = path->nodes[0];
 886
 887         i = path->slots[0];
 888         last_item = btrfs_header_nritems(leaf) - 1;
 889         if (i > last_item)
 890                 return -ENOENT; /* FIXME: Is errno suitable? */
 891
 892         next = item;
 893         INIT_LIST_HEAD(&head);
 894         btrfs_item_key_to_cpu(leaf, &key, i);
 895         nitems = 0;
 896         /*
 897          * count the number of the dir index items that we can delete in batch
 898          */
 899         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 900                 list_add_tail(&next->tree_list, &head);
 901                 nitems++;
 902
 903                 curr = next;
 904                 next = __btrfs_next_delayed_item(curr);
 905                 if (!next)
 906                         break;
 907
 908                 if (!btrfs_is_continuous_delayed_item(curr, next))
 909                         break;
 910
 911                 i++;
 912                 if (i > last_item)
 913                         break;
 914                 btrfs_item_key_to_cpu(leaf, &key, i);
 915         }
 916
 917         if (!nitems)
 918                 return 0;
 919
 920         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 921         if (ret)
 922                 goto out;
 923
 924         list_for_each_entry_safe(curr, next, &head, tree_list) {
 925                 btrfs_delayed_item_release_metadata(root, curr);
 926                 list_del(&curr->tree_list);
 927                 btrfs_release_delayed_item(curr);
 928         }
 929
 930 out:
 931         return ret;
 932 }
 933
 934 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 935                                       struct btrfs_path *path,
 936                                       struct btrfs_root *root,
 937                                       struct btrfs_delayed_node *node)
 938 {
 939         struct btrfs_delayed_item *curr, *prev;
 940         unsigned int nofs_flag;
 941         int ret = 0;
 942
 943 do_again:
 944         mutex_lock(&node->mutex);
 945         curr = __btrfs_first_delayed_deletion_item(node);
 946         if (!curr)
 947                 goto delete_fail;
 948
 949         nofs_flag = memalloc_nofs_save();
 950         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 951         memalloc_nofs_restore(nofs_flag);
 952         if (ret < 0)
 953                 goto delete_fail;
 954         else if (ret > 0) {
 955                 /*
 956                  * can't find the item which the node points to, so this node
 957                  * is invalid, just drop it.
 958                  */
 959                 prev = curr;
 960                 curr = __btrfs_next_delayed_item(prev);
 961                 btrfs_release_delayed_item(prev);
 962                 ret = 0;
 963                 btrfs_release_path(path);
 964                 if (curr) {
 965                         mutex_unlock(&node->mutex);
 966                         goto do_again;
 967                 } else
 968                         goto delete_fail;
 969         }
 970
 971         btrfs_batch_delete_items(trans, root, path, curr);
 972         btrfs_release_path(path);
 973         mutex_unlock(&node->mutex);
 974         goto do_again;
 975
 976 delete_fail:
 977         btrfs_release_path(path);
 978         mutex_unlock(&node->mutex);
 979         return ret;
 980 }
 981
 982 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 983 {
 984         struct btrfs_delayed_root *delayed_root;
 985
 986         if (delayed_node &&
 987             test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
 988                 BUG_ON(!delayed_node->root);
 989                 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
 990                 delayed_node->count--;
 991
 992                 delayed_root = delayed_node->root->fs_info->delayed_root;
 993                 finish_one_item(delayed_root);
 994         }
 995 }
 996
 997 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
 998 {
 999         struct btrfs_delayed_root *delayed_root;
1000
1001         ASSERT(delayed_node->root);
1002         clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1003         delayed_node->count--;
1004
1005         delayed_root = delayed_node->root->fs_info->delayed_root;
1006         finish_one_item(delayed_root);
1007 }
1008
1009 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1010                                         struct btrfs_root *root,
1011                                         struct btrfs_path *path,
1012                                         struct btrfs_delayed_node *node)
1013 {
1014         struct btrfs_fs_info *fs_info = root->fs_info;
1015         struct btrfs_key key;
1016         struct btrfs_inode_item *inode_item;
1017         struct extent_buffer *leaf;
1018         unsigned int nofs_flag;
1019         int mod;
1020         int ret;
1021
1022         key.objectid = node->inode_id;
1023         key.type = BTRFS_INODE_ITEM_KEY;
1024         key.offset = 0;
1025
1026         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1027                 mod = -1;
1028         else
1029                 mod = 1;
1030
1031         nofs_flag = memalloc_nofs_save();
1032         ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1033         memalloc_nofs_restore(nofs_flag);
1034         if (ret > 0) {
1035                 btrfs_release_path(path);
1036                 return -ENOENT;
1037         } else if (ret < 0) {
1038                 return ret;
1039         }
1040
1041         leaf = path->nodes[0];
1042         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1043                                     struct btrfs_inode_item);
1044         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1045                             sizeof(struct btrfs_inode_item));
1046         btrfs_mark_buffer_dirty(leaf);
1047
1048         if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1049                 goto no_iref;
1050
1051         path->slots[0]++;
1052         if (path->slots[0] >= btrfs_header_nritems(leaf))
1053                 goto search;
1054 again:
1055         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1056         if (key.objectid != node->inode_id)
1057                 goto out;
1058
1059         if (key.type != BTRFS_INODE_REF_KEY &&
1060             key.type != BTRFS_INODE_EXTREF_KEY)
1061                 goto out;
1062
1063         /*
1064          * Delayed iref deletion is for the inode who has only one link,
1065          * so there is only one iref. The case that several irefs are
1066          * in the same item doesn't exist.
1067          */
1068         btrfs_del_item(trans, root, path);
1069 out:
1070         btrfs_release_delayed_iref(node);
1071 no_iref:
1072         btrfs_release_path(path);
1073 err_out:
1074         btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1075         btrfs_release_delayed_inode(node);
1076
1077         return ret;
1078
1079 search:
1080         btrfs_release_path(path);
1081
1082         key.type = BTRFS_INODE_EXTREF_KEY;
1083         key.offset = -1;
1084
1085         nofs_flag = memalloc_nofs_save();
1086         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1087         memalloc_nofs_restore(nofs_flag);
1088         if (ret < 0)
1089                 goto err_out;
1090         ASSERT(ret);
1091
1092         ret = 0;
1093         leaf = path->nodes[0];
1094         path->slots[0]--;
1095         goto again;
1096 }
1097
1098 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1099                                              struct btrfs_root *root,
1100                                              struct btrfs_path *path,
1101                                              struct btrfs_delayed_node *node)
1102 {
1103         int ret;
1104
1105         mutex_lock(&node->mutex);
1106         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1107                 mutex_unlock(&node->mutex);
1108                 return 0;
1109         }
1110
1111         ret = __btrfs_update_delayed_inode(trans, root, path, node);
1112         mutex_unlock(&node->mutex);
1113         return ret;
1114 }
1115
1116 static inline int
1117 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1118                                    struct btrfs_path *path,
1119                                    struct btrfs_delayed_node *node)
1120 {
1121         int ret;
1122
1123         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1124         if (ret)
1125                 return ret;
1126
1127         ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1128         if (ret)
1129                 return ret;
1130
1131         ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1132         return ret;
1133 }
1134
1135 /*
1136  * Called when committing the transaction.
1137  * Returns 0 on success.
1138  * Returns < 0 on error and returns with an aborted transaction with any
1139  * outstanding delayed items cleaned up.
1140  */
1141 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1142 {
1143         struct btrfs_fs_info *fs_info = trans->fs_info;
1144         struct btrfs_delayed_root *delayed_root;
1145         struct btrfs_delayed_node *curr_node, *prev_node;
1146         struct btrfs_path *path;
1147         struct btrfs_block_rsv *block_rsv;
1148         int ret = 0;
1149         bool count = (nr > 0);
1150
1151         if (TRANS_ABORTED(trans))
1152                 return -EIO;
1153
1154         path = btrfs_alloc_path();
1155         if (!path)
1156                 return -ENOMEM;
1157         path->leave_spinning = 1;
1158
1159         block_rsv = trans->block_rsv;
1160         trans->block_rsv = &fs_info->delayed_block_rsv;
1161
1162         delayed_root = fs_info->delayed_root;
1163
1164         curr_node = btrfs_first_delayed_node(delayed_root);
1165         while (curr_node && (!count || (count && nr--))) {
1166                 ret = __btrfs_commit_inode_delayed_items(trans, path,
1167                                                          curr_node);
1168                 if (ret) {
1169                         btrfs_release_delayed_node(curr_node);
1170                         curr_node = NULL;
1171                         btrfs_abort_transaction(trans, ret);
1172                         break;
1173                 }
1174
1175                 prev_node = curr_node;
1176                 curr_node = btrfs_next_delayed_node(curr_node);
1177                 btrfs_release_delayed_node(prev_node);
1178         }
1179
1180         if (curr_node)
1181                 btrfs_release_delayed_node(curr_node);
1182         btrfs_free_path(path);
1183         trans->block_rsv = block_rsv;
1184
1185         return ret;
1186 }
1187
1188 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1189 {
1190         return __btrfs_run_delayed_items(trans, -1);
1191 }
1192
1193 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1194 {
1195         return __btrfs_run_delayed_items(trans, nr);
1196 }
1197
1198 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1199                                      struct btrfs_inode *inode)
1200 {
1201         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1202         struct btrfs_path *path;
1203         struct btrfs_block_rsv *block_rsv;
1204         int ret;
1205
1206         if (!delayed_node)
1207                 return 0;
1208
1209         mutex_lock(&delayed_node->mutex);
1210         if (!delayed_node->count) {
1211                 mutex_unlock(&delayed_node->mutex);
1212                 btrfs_release_delayed_node(delayed_node);
1213                 return 0;
1214         }
1215         mutex_unlock(&delayed_node->mutex);
1216
1217         path = btrfs_alloc_path();
1218         if (!path) {
1219                 btrfs_release_delayed_node(delayed_node);
1220                 return -ENOMEM;
1221         }
1222         path->leave_spinning = 1;
1223
1224         block_rsv = trans->block_rsv;
1225         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1226
1227         ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1228
1229         btrfs_release_delayed_node(delayed_node);
1230         btrfs_free_path(path);
1231         trans->block_rsv = block_rsv;
1232
1233         return ret;
1234 }
1235
1236 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1237 {
1238         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1239         struct btrfs_trans_handle *trans;
1240         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1241         struct btrfs_path *path;
1242         struct btrfs_block_rsv *block_rsv;
1243         int ret;
1244
1245         if (!delayed_node)
1246                 return 0;
1247
1248         mutex_lock(&delayed_node->mutex);
1249         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1250                 mutex_unlock(&delayed_node->mutex);
1251                 btrfs_release_delayed_node(delayed_node);
1252                 return 0;
1253         }
1254         mutex_unlock(&delayed_node->mutex);
1255
1256         trans = btrfs_join_transaction(delayed_node->root);
1257         if (IS_ERR(trans)) {
1258                 ret = PTR_ERR(trans);
1259                 goto out;
1260         }
1261
1262         path = btrfs_alloc_path();
1263         if (!path) {
1264                 ret = -ENOMEM;
1265                 goto trans_out;
1266         }
1267         path->leave_spinning = 1;
1268
1269         block_rsv = trans->block_rsv;
1270         trans->block_rsv = &fs_info->delayed_block_rsv;
1271
1272         mutex_lock(&delayed_node->mutex);
1273         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1274                 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1275                                                    path, delayed_node);
1276         else
1277                 ret = 0;
1278         mutex_unlock(&delayed_node->mutex);
1279
1280         btrfs_free_path(path);
1281         trans->block_rsv = block_rsv;
1282 trans_out:
1283         btrfs_end_transaction(trans);
1284         btrfs_btree_balance_dirty(fs_info);
1285 out:
1286         btrfs_release_delayed_node(delayed_node);
1287
1288         return ret;
1289 }
1290
1291 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1292 {
1293         struct btrfs_delayed_node *delayed_node;
1294
1295         delayed_node = READ_ONCE(inode->delayed_node);
1296         if (!delayed_node)
1297                 return;
1298
1299         inode->delayed_node = NULL;
1300         btrfs_release_delayed_node(delayed_node);
1301 }
1302
1303 struct btrfs_async_delayed_work {
1304         struct btrfs_delayed_root *delayed_root;
1305         int nr;
1306         struct btrfs_work work;
1307 };
1308
1309 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1310 {
1311         struct btrfs_async_delayed_work *async_work;
1312         struct btrfs_delayed_root *delayed_root;
1313         struct btrfs_trans_handle *trans;
1314         struct btrfs_path *path;
1315         struct btrfs_delayed_node *delayed_node = NULL;
1316         struct btrfs_root *root;
1317         struct btrfs_block_rsv *block_rsv;
1318         int total_done = 0;
1319
1320         async_work = container_of(work, struct btrfs_async_delayed_work, work);
1321         delayed_root = async_work->delayed_root;
1322
1323         path = btrfs_alloc_path();
1324         if (!path)
1325                 goto out;
1326
1327         do {
1328                 if (atomic_read(&delayed_root->items) <
1329                     BTRFS_DELAYED_BACKGROUND / 2)
1330                         break;
1331
1332                 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1333                 if (!delayed_node)
1334                         break;
1335
1336                 path->leave_spinning = 1;
1337                 root = delayed_node->root;
1338
1339                 trans = btrfs_join_transaction(root);
1340                 if (IS_ERR(trans)) {
1341                         btrfs_release_path(path);
1342                         btrfs_release_prepared_delayed_node(delayed_node);
1343                         total_done++;
1344                         continue;
1345                 }
1346
1347                 block_rsv = trans->block_rsv;
1348                 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1349
1350                 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1351
1352                 trans->block_rsv = block_rsv;
1353                 btrfs_end_transaction(trans);
1354                 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1355
1356                 btrfs_release_path(path);
1357                 btrfs_release_prepared_delayed_node(delayed_node);
1358                 total_done++;
1359
1360         } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1361                  || total_done < async_work->nr);
1362
1363         btrfs_free_path(path);
1364 out:
1365         wake_up(&delayed_root->wait);
1366         kfree(async_work);
1367 }
1368
1369
1370 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1371                                      struct btrfs_fs_info *fs_info, int nr)
1372 {
1373         struct btrfs_async_delayed_work *async_work;
1374
1375         async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1376         if (!async_work)
1377                 return -ENOMEM;
1378
1379         async_work->delayed_root = delayed_root;
1380         btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1381                         NULL);
1382         async_work->nr = nr;
1383
1384         btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1385         return 0;
1386 }
1387
1388 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1389 {
1390         WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1391 }
1392
1393 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1394 {
1395         int val = atomic_read(&delayed_root->items_seq);
1396
1397         if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1398                 return 1;
1399
1400         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1401                 return 1;
1402
1403         return 0;
1404 }
1405
1406 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1407 {
1408         struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1409
1410         if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1411                 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1412                 return;
1413
1414         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1415                 int seq;
1416                 int ret;
1417
1418                 seq = atomic_read(&delayed_root->items_seq);
1419
1420                 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1421                 if (ret)
1422                         return;
1423
1424                 wait_event_interruptible(delayed_root->wait,
1425                                          could_end_wait(delayed_root, seq));
1426                 return;
1427         }
1428
1429         btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1430 }
1431
1432 /* Will return 0 or -ENOMEM */
1433 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1434                                    const char *name, int name_len,
1435                                    struct btrfs_inode *dir,
1436                                    struct btrfs_disk_key *disk_key, u8 type,
1437                                    u64 index)
1438 {
1439         struct btrfs_delayed_node *delayed_node;
1440         struct btrfs_delayed_item *delayed_item;
1441         struct btrfs_dir_item *dir_item;
1442         int ret;
1443
1444         delayed_node = btrfs_get_or_create_delayed_node(dir);
1445         if (IS_ERR(delayed_node))
1446                 return PTR_ERR(delayed_node);
1447
1448         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1449         if (!delayed_item) {
1450                 ret = -ENOMEM;
1451                 goto release_node;
1452         }
1453
1454         delayed_item->key.objectid = btrfs_ino(dir);
1455         delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1456         delayed_item->key.offset = index;
1457
1458         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1459         dir_item->location = *disk_key;
1460         btrfs_set_stack_dir_transid(dir_item, trans->transid);
1461         btrfs_set_stack_dir_data_len(dir_item, 0);
1462         btrfs_set_stack_dir_name_len(dir_item, name_len);
1463         btrfs_set_stack_dir_type(dir_item, type);
1464         memcpy((char *)(dir_item + 1), name, name_len);
1465
1466         ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1467         /*
1468          * we have reserved enough space when we start a new transaction,
1469          * so reserving metadata failure is impossible
1470          */
1471         BUG_ON(ret);
1472
1473         mutex_lock(&delayed_node->mutex);
1474         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1475         if (unlikely(ret)) {
1476                 btrfs_err(trans->fs_info,
1477                           "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1478                           name_len, name, delayed_node->root->root_key.objectid,
1479                           delayed_node->inode_id, ret);
1480                 BUG();
1481         }
1482         mutex_unlock(&delayed_node->mutex);
1483
1484 release_node:
1485         btrfs_release_delayed_node(delayed_node);
1486         return ret;
1487 }
1488
1489 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1490                                                struct btrfs_delayed_node *node,
1491                                                struct btrfs_key *key)
1492 {
1493         struct btrfs_delayed_item *item;
1494
1495         mutex_lock(&node->mutex);
1496         item = __btrfs_lookup_delayed_insertion_item(node, key);
1497         if (!item) {
1498                 mutex_unlock(&node->mutex);
1499                 return 1;
1500         }
1501
1502         btrfs_delayed_item_release_metadata(node->root, item);
1503         btrfs_release_delayed_item(item);
1504         mutex_unlock(&node->mutex);
1505         return 0;
1506 }
1507
1508 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1509                                    struct btrfs_inode *dir, u64 index)
1510 {
1511         struct btrfs_delayed_node *node;
1512         struct btrfs_delayed_item *item;
1513         struct btrfs_key item_key;
1514         int ret;
1515
1516         node = btrfs_get_or_create_delayed_node(dir);
1517         if (IS_ERR(node))
1518                 return PTR_ERR(node);
1519
1520         item_key.objectid = btrfs_ino(dir);
1521         item_key.type = BTRFS_DIR_INDEX_KEY;
1522         item_key.offset = index;
1523
1524         ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1525                                                   &item_key);
1526         if (!ret)
1527                 goto end;
1528
1529         item = btrfs_alloc_delayed_item(0);
1530         if (!item) {
1531                 ret = -ENOMEM;
1532                 goto end;
1533         }
1534
1535         item->key = item_key;
1536
1537         ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1538         /*
1539          * we have reserved enough space when we start a new transaction,
1540          * so reserving metadata failure is impossible.
1541          */
1542         if (ret < 0) {
1543                 btrfs_err(trans->fs_info,
1544 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1545                 btrfs_release_delayed_item(item);
1546                 goto end;
1547         }
1548
1549         mutex_lock(&node->mutex);
1550         ret = __btrfs_add_delayed_deletion_item(node, item);
1551         if (unlikely(ret)) {
1552                 btrfs_err(trans->fs_info,
1553                           "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1554                           index, node->root->root_key.objectid,
1555                           node->inode_id, ret);
1556                 btrfs_delayed_item_release_metadata(dir->root, item);
1557                 btrfs_release_delayed_item(item);
1558         }
1559         mutex_unlock(&node->mutex);
1560 end:
1561         btrfs_release_delayed_node(node);
1562         return ret;
1563 }
1564
1565 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1566 {
1567         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1568
1569         if (!delayed_node)
1570                 return -ENOENT;
1571
1572         /*
1573          * Since we have held i_mutex of this directory, it is impossible that
1574          * a new directory index is added into the delayed node and index_cnt
1575          * is updated now. So we needn't lock the delayed node.
1576          */
1577         if (!delayed_node->index_cnt) {
1578                 btrfs_release_delayed_node(delayed_node);
1579                 return -EINVAL;
1580         }
1581
1582         inode->index_cnt = delayed_node->index_cnt;
1583         btrfs_release_delayed_node(delayed_node);
1584         return 0;
1585 }
1586
1587 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1588                                      struct list_head *ins_list,
1589                                      struct list_head *del_list)
1590 {
1591         struct btrfs_delayed_node *delayed_node;
1592         struct btrfs_delayed_item *item;
1593
1594         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1595         if (!delayed_node)
1596                 return false;
1597
1598         /*
1599          * We can only do one readdir with delayed items at a time because of
1600          * item->readdir_list.
1601          */
1602         inode_unlock_shared(inode);
1603         inode_lock(inode);
1604
1605         mutex_lock(&delayed_node->mutex);
1606         item = __btrfs_first_delayed_insertion_item(delayed_node);
1607         while (item) {
1608                 refcount_inc(&item->refs);
1609                 list_add_tail(&item->readdir_list, ins_list);
1610                 item = __btrfs_next_delayed_item(item);
1611         }
1612
1613         item = __btrfs_first_delayed_deletion_item(delayed_node);
1614         while (item) {
1615                 refcount_inc(&item->refs);
1616                 list_add_tail(&item->readdir_list, del_list);
1617                 item = __btrfs_next_delayed_item(item);
1618         }
1619         mutex_unlock(&delayed_node->mutex);
1620         /*
1621          * This delayed node is still cached in the btrfs inode, so refs
1622          * must be > 1 now, and we needn't check it is going to be freed
1623          * or not.
1624          *
1625          * Besides that, this function is used to read dir, we do not
1626          * insert/delete delayed items in this period. So we also needn't
1627          * requeue or dequeue this delayed node.
1628          */
1629         refcount_dec(&delayed_node->refs);
1630
1631         return true;
1632 }
1633
1634 void btrfs_readdir_put_delayed_items(struct inode *inode,
1635                                      struct list_head *ins_list,
1636                                      struct list_head *del_list)
1637 {
1638         struct btrfs_delayed_item *curr, *next;
1639
1640         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1641                 list_del(&curr->readdir_list);
1642                 if (refcount_dec_and_test(&curr->refs))
1643                         kfree(curr);
1644         }
1645
1646         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1647                 list_del(&curr->readdir_list);
1648                 if (refcount_dec_and_test(&curr->refs))
1649                         kfree(curr);
1650         }
1651
1652         /*
1653          * The VFS is going to do up_read(), so we need to downgrade back to a
1654          * read lock.
1655          */
1656         downgrade_write(&inode->i_rwsem);
1657 }
1658
1659 int btrfs_should_delete_dir_index(struct list_head *del_list,
1660                                   u64 index)
1661 {
1662         struct btrfs_delayed_item *curr;
1663         int ret = 0;
1664
1665         list_for_each_entry(curr, del_list, readdir_list) {
1666                 if (curr->key.offset > index)
1667                         break;
1668                 if (curr->key.offset == index) {
1669                         ret = 1;
1670                         break;
1671                 }
1672         }
1673         return ret;
1674 }
1675
1676 /*
1677  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1678  *
1679  */
1680 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1681                                     struct list_head *ins_list)
1682 {
1683         struct btrfs_dir_item *di;
1684         struct btrfs_delayed_item *curr, *next;
1685         struct btrfs_key location;
1686         char *name;
1687         int name_len;
1688         int over = 0;
1689         unsigned char d_type;
1690
1691         if (list_empty(ins_list))
1692                 return 0;
1693
1694         /*
1695          * Changing the data of the delayed item is impossible. So
1696          * we needn't lock them. And we have held i_mutex of the
1697          * directory, nobody can delete any directory indexes now.
1698          */
1699         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1700                 list_del(&curr->readdir_list);
1701
1702                 if (curr->key.offset < ctx->pos) {
1703                         if (refcount_dec_and_test(&curr->refs))
1704                                 kfree(curr);
1705                         continue;
1706                 }
1707
1708                 ctx->pos = curr->key.offset;
1709
1710                 di = (struct btrfs_dir_item *)curr->data;
1711                 name = (char *)(di + 1);
1712                 name_len = btrfs_stack_dir_name_len(di);
1713
1714                 d_type = fs_ftype_to_dtype(di->type);
1715                 btrfs_disk_key_to_cpu(&location, &di->location);
1716
1717                 over = !dir_emit(ctx, name, name_len,
1718                                location.objectid, d_type);
1719
1720                 if (refcount_dec_and_test(&curr->refs))
1721                         kfree(curr);
1722
1723                 if (over)
1724                         return 1;
1725                 ctx->pos++;
1726         }
1727         return 0;
1728 }
1729
1730 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1731                                   struct btrfs_inode_item *inode_item,
1732                                   struct inode *inode)
1733 {
1734         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1735         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1736         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1737         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1738         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1739         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1740         btrfs_set_stack_inode_generation(inode_item,
1741                                          BTRFS_I(inode)->generation);
1742         btrfs_set_stack_inode_sequence(inode_item,
1743                                        inode_peek_iversion(inode));
1744         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1745         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1746         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1747         btrfs_set_stack_inode_block_group(inode_item, 0);
1748
1749         btrfs_set_stack_timespec_sec(&inode_item->atime,
1750                                      inode->i_atime.tv_sec);
1751         btrfs_set_stack_timespec_nsec(&inode_item->atime,
1752                                       inode->i_atime.tv_nsec);
1753
1754         btrfs_set_stack_timespec_sec(&inode_item->mtime,
1755                                      inode->i_mtime.tv_sec);
1756         btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1757                                       inode->i_mtime.tv_nsec);
1758
1759         btrfs_set_stack_timespec_sec(&inode_item->ctime,
1760                                      inode->i_ctime.tv_sec);
1761         btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1762                                       inode->i_ctime.tv_nsec);
1763
1764         btrfs_set_stack_timespec_sec(&inode_item->otime,
1765                                      BTRFS_I(inode)->i_otime.tv_sec);
1766         btrfs_set_stack_timespec_nsec(&inode_item->otime,
1767                                      BTRFS_I(inode)->i_otime.tv_nsec);
1768 }
1769
1770 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1771 {
1772         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1773         struct btrfs_delayed_node *delayed_node;
1774         struct btrfs_inode_item *inode_item;
1775
1776         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1777         if (!delayed_node)
1778                 return -ENOENT;
1779
1780         mutex_lock(&delayed_node->mutex);
1781         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1782                 mutex_unlock(&delayed_node->mutex);
1783                 btrfs_release_delayed_node(delayed_node);
1784                 return -ENOENT;
1785         }
1786
1787         inode_item = &delayed_node->inode_item;
1788
1789         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1790         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1791         btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1792         btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1793                         round_up(i_size_read(inode), fs_info->sectorsize));
1794         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1795         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1796         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1797         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1798         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1799
1800         inode_set_iversion_queried(inode,
1801                                    btrfs_stack_inode_sequence(inode_item));
1802         inode->i_rdev = 0;
1803         *rdev = btrfs_stack_inode_rdev(inode_item);
1804         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1805
1806         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1807         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1808
1809         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1810         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1811
1812         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1813         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1814
1815         BTRFS_I(inode)->i_otime.tv_sec =
1816                 btrfs_stack_timespec_sec(&inode_item->otime);
1817         BTRFS_I(inode)->i_otime.tv_nsec =
1818                 btrfs_stack_timespec_nsec(&inode_item->otime);
1819
1820         inode->i_generation = BTRFS_I(inode)->generation;
1821         BTRFS_I(inode)->index_cnt = (u64)-1;
1822
1823         mutex_unlock(&delayed_node->mutex);
1824         btrfs_release_delayed_node(delayed_node);
1825         return 0;
1826 }
1827
1828 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1829                                struct btrfs_root *root, struct inode *inode)
1830 {
1831         struct btrfs_delayed_node *delayed_node;
1832         int ret = 0;
1833
1834         delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1835         if (IS_ERR(delayed_node))
1836                 return PTR_ERR(delayed_node);
1837
1838         mutex_lock(&delayed_node->mutex);
1839         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1840                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1841                 goto release_node;
1842         }
1843
1844         ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1845                                                    delayed_node);
1846         if (ret)
1847                 goto release_node;
1848
1849         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1850         set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1851         delayed_node->count++;
1852         atomic_inc(&root->fs_info->delayed_root->items);
1853 release_node:
1854         mutex_unlock(&delayed_node->mutex);
1855         btrfs_release_delayed_node(delayed_node);
1856         return ret;
1857 }
1858
1859 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1860 {
1861         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1862         struct btrfs_delayed_node *delayed_node;
1863
1864         /*
1865          * we don't do delayed inode updates during log recovery because it
1866          * leads to enospc problems.  This means we also can't do
1867          * delayed inode refs
1868          */
1869         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1870                 return -EAGAIN;
1871
1872         delayed_node = btrfs_get_or_create_delayed_node(inode);
1873         if (IS_ERR(delayed_node))
1874                 return PTR_ERR(delayed_node);
1875
1876         /*
1877          * We don't reserve space for inode ref deletion is because:
1878          * - We ONLY do async inode ref deletion for the inode who has only
1879          *   one link(i_nlink == 1), it means there is only one inode ref.
1880          *   And in most case, the inode ref and the inode item are in the
1881          *   same leaf, and we will deal with them at the same time.
1882          *   Since we are sure we will reserve the space for the inode item,
1883          *   it is unnecessary to reserve space for inode ref deletion.
1884          * - If the inode ref and the inode item are not in the same leaf,
1885          *   We also needn't worry about enospc problem, because we reserve
1886          *   much more space for the inode update than it needs.
1887          * - At the worst, we can steal some space from the global reservation.
1888          *   It is very rare.
1889          */
1890         mutex_lock(&delayed_node->mutex);
1891         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1892                 goto release_node;
1893
1894         set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1895         delayed_node->count++;
1896         atomic_inc(&fs_info->delayed_root->items);
1897 release_node:
1898         mutex_unlock(&delayed_node->mutex);
1899         btrfs_release_delayed_node(delayed_node);
1900         return 0;
1901 }
1902
1903 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1904 {
1905         struct btrfs_root *root = delayed_node->root;
1906         struct btrfs_fs_info *fs_info = root->fs_info;
1907         struct btrfs_delayed_item *curr_item, *prev_item;
1908
1909         mutex_lock(&delayed_node->mutex);
1910         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1911         while (curr_item) {
1912                 btrfs_delayed_item_release_metadata(root, curr_item);
1913                 prev_item = curr_item;
1914                 curr_item = __btrfs_next_delayed_item(prev_item);
1915                 btrfs_release_delayed_item(prev_item);
1916         }
1917
1918         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1919         while (curr_item) {
1920                 btrfs_delayed_item_release_metadata(root, curr_item);
1921                 prev_item = curr_item;
1922                 curr_item = __btrfs_next_delayed_item(prev_item);
1923                 btrfs_release_delayed_item(prev_item);
1924         }
1925
1926         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1927                 btrfs_release_delayed_iref(delayed_node);
1928
1929         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1930                 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1931                 btrfs_release_delayed_inode(delayed_node);
1932         }
1933         mutex_unlock(&delayed_node->mutex);
1934 }
1935
1936 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1937 {
1938         struct btrfs_delayed_node *delayed_node;
1939
1940         delayed_node = btrfs_get_delayed_node(inode);
1941         if (!delayed_node)
1942                 return;
1943
1944         __btrfs_kill_delayed_node(delayed_node);
1945         btrfs_release_delayed_node(delayed_node);
1946 }
1947
1948 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1949 {
1950         u64 inode_id = 0;
1951         struct btrfs_delayed_node *delayed_nodes[8];
1952         int i, n;
1953
1954         while (1) {
1955                 spin_lock(&root->inode_lock);
1956                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1957                                            (void **)delayed_nodes, inode_id,
1958                                            ARRAY_SIZE(delayed_nodes));
1959                 if (!n) {
1960                         spin_unlock(&root->inode_lock);
1961                         break;
1962                 }
1963
1964                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1965                 for (i = 0; i < n; i++) {
1966                         /*
1967                          * Don't increase refs in case the node is dead and
1968                          * about to be removed from the tree in the loop below
1969                          */
1970                         if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
1971                                 delayed_nodes[i] = NULL;
1972                 }
1973                 spin_unlock(&root->inode_lock);
1974
1975                 for (i = 0; i < n; i++) {
1976                         if (!delayed_nodes[i])
1977                                 continue;
1978                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1979                         btrfs_release_delayed_node(delayed_nodes[i]);
1980                 }
1981         }
1982 }
1983
1984 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1985 {
1986         struct btrfs_delayed_node *curr_node, *prev_node;
1987
1988         curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1989         while (curr_node) {
1990                 __btrfs_kill_delayed_node(curr_node);
1991
1992                 prev_node = curr_node;
1993                 curr_node = btrfs_next_delayed_node(curr_node);
1994                 btrfs_release_delayed_node(prev_node);
1995         }
1996 }
1997