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(root, num_bytes,
 631                                           BTRFS_QGROUP_RSV_META_PREALLOC, true);
 632                 if (ret < 0)
 633                         return ret;
 634                 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 635                                           BTRFS_RESERVE_NO_FLUSH);
 636                 /*
 637                  * Since we're under a transaction reserve_metadata_bytes could
 638                  * try to commit the transaction which will make it return
 639                  * EAGAIN to make us stop the transaction we have, so return
 640                  * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 641                  */
 642                 if (ret == -EAGAIN) {
 643                         ret = -ENOSPC;
 644                         btrfs_qgroup_free_meta_prealloc(root, num_bytes);
 645                 }
 646                 if (!ret) {
 647                         node->bytes_reserved = num_bytes;
 648                         trace_btrfs_space_reservation(fs_info,
 649                                                       "delayed_inode",
 650                                                       btrfs_ino(inode),
 651                                                       num_bytes, 1);
 652                 } else {
 653                         btrfs_qgroup_free_meta_prealloc(root, num_bytes);
 654                 }
 655                 return ret;
 656         }
 657
 658         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
 659         if (!ret) {
 660                 trace_btrfs_space_reservation(fs_info, "delayed_inode",
 661                                               btrfs_ino(inode), num_bytes, 1);
 662                 node->bytes_reserved = num_bytes;
 663         }
 664
 665         return ret;
 666 }
 667
 668 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
 669                                                 struct btrfs_delayed_node *node,
 670                                                 bool qgroup_free)
 671 {
 672         struct btrfs_block_rsv *rsv;
 673
 674         if (!node->bytes_reserved)
 675                 return;
 676
 677         rsv = &fs_info->delayed_block_rsv;
 678         trace_btrfs_space_reservation(fs_info, "delayed_inode",
 679                                       node->inode_id, node->bytes_reserved, 0);
 680         btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
 681         if (qgroup_free)
 682                 btrfs_qgroup_free_meta_prealloc(node->root,
 683                                 node->bytes_reserved);
 684         else
 685                 btrfs_qgroup_convert_reserved_meta(node->root,
 686                                 node->bytes_reserved);
 687         node->bytes_reserved = 0;
 688 }
 689
 690 /*
 691  * This helper will insert some continuous items into the same leaf according
 692  * to the free space of the leaf.
 693  */
 694 static int btrfs_batch_insert_items(struct btrfs_root *root,
 695                                     struct btrfs_path *path,
 696                                     struct btrfs_delayed_item *item)
 697 {
 698         struct btrfs_delayed_item *curr, *next;
 699         int free_space;
 700         int total_data_size = 0, total_size = 0;
 701         struct extent_buffer *leaf;
 702         char *data_ptr;
 703         struct btrfs_key *keys;
 704         u32 *data_size;
 705         struct list_head head;
 706         int slot;
 707         int nitems;
 708         int i;
 709         int ret = 0;
 710
 711         BUG_ON(!path->nodes[0]);
 712
 713         leaf = path->nodes[0];
 714         free_space = btrfs_leaf_free_space(leaf);
 715         INIT_LIST_HEAD(&head);
 716
 717         next = item;
 718         nitems = 0;
 719
 720         /*
 721          * count the number of the continuous items that we can insert in batch
 722          */
 723         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 724                free_space) {
 725                 total_data_size += next->data_len;
 726                 total_size += next->data_len + sizeof(struct btrfs_item);
 727                 list_add_tail(&next->tree_list, &head);
 728                 nitems++;
 729
 730                 curr = next;
 731                 next = __btrfs_next_delayed_item(curr);
 732                 if (!next)
 733                         break;
 734
 735                 if (!btrfs_is_continuous_delayed_item(curr, next))
 736                         break;
 737         }
 738
 739         if (!nitems) {
 740                 ret = 0;
 741                 goto out;
 742         }
 743
 744         /*
 745          * we need allocate some memory space, but it might cause the task
 746          * to sleep, so we set all locked nodes in the path to blocking locks
 747          * first.
 748          */
 749         btrfs_set_path_blocking(path);
 750
 751         keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
 752         if (!keys) {
 753                 ret = -ENOMEM;
 754                 goto out;
 755         }
 756
 757         data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
 758         if (!data_size) {
 759                 ret = -ENOMEM;
 760                 goto error;
 761         }
 762
 763         /* get keys of all the delayed items */
 764         i = 0;
 765         list_for_each_entry(next, &head, tree_list) {
 766                 keys[i] = next->key;
 767                 data_size[i] = next->data_len;
 768                 i++;
 769         }
 770
 771         /* insert the keys of the items */
 772         setup_items_for_insert(root, path, keys, data_size, nitems);
 773
 774         /* insert the dir index items */
 775         slot = path->slots[0];
 776         list_for_each_entry_safe(curr, next, &head, tree_list) {
 777                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 778                 write_extent_buffer(leaf, &curr->data,
 779                                     (unsigned long)data_ptr,
 780                                     curr->data_len);
 781                 slot++;
 782
 783                 btrfs_delayed_item_release_metadata(root, curr);
 784
 785                 list_del(&curr->tree_list);
 786                 btrfs_release_delayed_item(curr);
 787         }
 788
 789 error:
 790         kfree(data_size);
 791         kfree(keys);
 792 out:
 793         return ret;
 794 }
 795
 796 /*
 797  * This helper can just do simple insertion that needn't extend item for new
 798  * data, such as directory name index insertion, inode insertion.
 799  */
 800 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 801                                      struct btrfs_root *root,
 802                                      struct btrfs_path *path,
 803                                      struct btrfs_delayed_item *delayed_item)
 804 {
 805         struct extent_buffer *leaf;
 806         unsigned int nofs_flag;
 807         char *ptr;
 808         int ret;
 809
 810         nofs_flag = memalloc_nofs_save();
 811         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 812                                       delayed_item->data_len);
 813         memalloc_nofs_restore(nofs_flag);
 814         if (ret < 0 && ret != -EEXIST)
 815                 return ret;
 816
 817         leaf = path->nodes[0];
 818
 819         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 820
 821         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 822                             delayed_item->data_len);
 823         btrfs_mark_buffer_dirty(leaf);
 824
 825         btrfs_delayed_item_release_metadata(root, delayed_item);
 826         return 0;
 827 }
 828
 829 /*
 830  * we insert an item first, then if there are some continuous items, we try
 831  * to insert those items into the same leaf.
 832  */
 833 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 834                                       struct btrfs_path *path,
 835                                       struct btrfs_root *root,
 836                                       struct btrfs_delayed_node *node)
 837 {
 838         struct btrfs_delayed_item *curr, *prev;
 839         int ret = 0;
 840
 841 do_again:
 842         mutex_lock(&node->mutex);
 843         curr = __btrfs_first_delayed_insertion_item(node);
 844         if (!curr)
 845                 goto insert_end;
 846
 847         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 848         if (ret < 0) {
 849                 btrfs_release_path(path);
 850                 goto insert_end;
 851         }
 852
 853         prev = curr;
 854         curr = __btrfs_next_delayed_item(prev);
 855         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 856                 /* insert the continuous items into the same leaf */
 857                 path->slots[0]++;
 858                 btrfs_batch_insert_items(root, path, curr);
 859         }
 860         btrfs_release_delayed_item(prev);
 861         btrfs_mark_buffer_dirty(path->nodes[0]);
 862
 863         btrfs_release_path(path);
 864         mutex_unlock(&node->mutex);
 865         goto do_again;
 866
 867 insert_end:
 868         mutex_unlock(&node->mutex);
 869         return ret;
 870 }
 871
 872 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 873                                     struct btrfs_root *root,
 874                                     struct btrfs_path *path,
 875                                     struct btrfs_delayed_item *item)
 876 {
 877         struct btrfs_delayed_item *curr, *next;
 878         struct extent_buffer *leaf;
 879         struct btrfs_key key;
 880         struct list_head head;
 881         int nitems, i, last_item;
 882         int ret = 0;
 883
 884         BUG_ON(!path->nodes[0]);
 885
 886         leaf = path->nodes[0];
 887
 888         i = path->slots[0];
 889         last_item = btrfs_header_nritems(leaf) - 1;
 890         if (i > last_item)
 891                 return -ENOENT; /* FIXME: Is errno suitable? */
 892
 893         next = item;
 894         INIT_LIST_HEAD(&head);
 895         btrfs_item_key_to_cpu(leaf, &key, i);
 896         nitems = 0;
 897         /*
 898          * count the number of the dir index items that we can delete in batch
 899          */
 900         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 901                 list_add_tail(&next->tree_list, &head);
 902                 nitems++;
 903
 904                 curr = next;
 905                 next = __btrfs_next_delayed_item(curr);
 906                 if (!next)
 907                         break;
 908
 909                 if (!btrfs_is_continuous_delayed_item(curr, next))
 910                         break;
 911
 912                 i++;
 913                 if (i > last_item)
 914                         break;
 915                 btrfs_item_key_to_cpu(leaf, &key, i);
 916         }
 917
 918         if (!nitems)
 919                 return 0;
 920
 921         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 922         if (ret)
 923                 goto out;
 924
 925         list_for_each_entry_safe(curr, next, &head, tree_list) {
 926                 btrfs_delayed_item_release_metadata(root, curr);
 927                 list_del(&curr->tree_list);
 928                 btrfs_release_delayed_item(curr);
 929         }
 930
 931 out:
 932         return ret;
 933 }
 934
 935 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 936                                       struct btrfs_path *path,
 937                                       struct btrfs_root *root,
 938                                       struct btrfs_delayed_node *node)
 939 {
 940         struct btrfs_delayed_item *curr, *prev;
 941         unsigned int nofs_flag;
 942         int ret = 0;
 943
 944 do_again:
 945         mutex_lock(&node->mutex);
 946         curr = __btrfs_first_delayed_deletion_item(node);
 947         if (!curr)
 948                 goto delete_fail;
 949
 950         nofs_flag = memalloc_nofs_save();
 951         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 952         memalloc_nofs_restore(nofs_flag);
 953         if (ret < 0)
 954                 goto delete_fail;
 955         else if (ret > 0) {
 956                 /*
 957                  * can't find the item which the node points to, so this node
 958                  * is invalid, just drop it.
 959                  */
 960                 prev = curr;
 961                 curr = __btrfs_next_delayed_item(prev);
 962                 btrfs_release_delayed_item(prev);
 963                 ret = 0;
 964                 btrfs_release_path(path);
 965                 if (curr) {
 966                         mutex_unlock(&node->mutex);
 967                         goto do_again;
 968                 } else
 969                         goto delete_fail;
 970         }
 971
 972         btrfs_batch_delete_items(trans, root, path, curr);
 973         btrfs_release_path(path);
 974         mutex_unlock(&node->mutex);
 975         goto do_again;
 976
 977 delete_fail:
 978         btrfs_release_path(path);
 979         mutex_unlock(&node->mutex);
 980         return ret;
 981 }
 982
 983 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 984 {
 985         struct btrfs_delayed_root *delayed_root;
 986
 987         if (delayed_node &&
 988             test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
 989                 BUG_ON(!delayed_node->root);
 990                 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
 991                 delayed_node->count--;
 992
 993                 delayed_root = delayed_node->root->fs_info->delayed_root;
 994                 finish_one_item(delayed_root);
 995         }
 996 }
 997
 998 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
 999 {
1000         struct btrfs_delayed_root *delayed_root;
1001
1002         ASSERT(delayed_node->root);
1003         clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1004         delayed_node->count--;
1005
1006         delayed_root = delayed_node->root->fs_info->delayed_root;
1007         finish_one_item(delayed_root);
1008 }
1009
1010 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1011                                         struct btrfs_root *root,
1012                                         struct btrfs_path *path,
1013                                         struct btrfs_delayed_node *node)
1014 {
1015         struct btrfs_fs_info *fs_info = root->fs_info;
1016         struct btrfs_key key;
1017         struct btrfs_inode_item *inode_item;
1018         struct extent_buffer *leaf;
1019         unsigned int nofs_flag;
1020         int mod;
1021         int ret;
1022
1023         key.objectid = node->inode_id;
1024         key.type = BTRFS_INODE_ITEM_KEY;
1025         key.offset = 0;
1026
1027         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1028                 mod = -1;
1029         else
1030                 mod = 1;
1031
1032         nofs_flag = memalloc_nofs_save();
1033         ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1034         memalloc_nofs_restore(nofs_flag);
1035         if (ret > 0)
1036                 ret = -ENOENT;
1037         if (ret < 0)
1038                 goto out;
1039
1040         leaf = path->nodes[0];
1041         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1042                                     struct btrfs_inode_item);
1043         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1044                             sizeof(struct btrfs_inode_item));
1045         btrfs_mark_buffer_dirty(leaf);
1046
1047         if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1048                 goto no_iref;
1049
1050         path->slots[0]++;
1051         if (path->slots[0] >= btrfs_header_nritems(leaf))
1052                 goto search;
1053 again:
1054         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1055         if (key.objectid != node->inode_id)
1056                 goto out;
1057
1058         if (key.type != BTRFS_INODE_REF_KEY &&
1059             key.type != BTRFS_INODE_EXTREF_KEY)
1060                 goto out;
1061
1062         /*
1063          * Delayed iref deletion is for the inode who has only one link,
1064          * so there is only one iref. The case that several irefs are
1065          * in the same item doesn't exist.
1066          */
1067         btrfs_del_item(trans, root, path);
1068 out:
1069         btrfs_release_delayed_iref(node);
1070 no_iref:
1071         btrfs_release_path(path);
1072 err_out:
1073         btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1074         btrfs_release_delayed_inode(node);
1075
1076         /*
1077          * If we fail to update the delayed inode we need to abort the
1078          * transaction, because we could leave the inode with the improper
1079          * counts behind.
1080          */
1081         if (ret && ret != -ENOENT)
1082                 btrfs_abort_transaction(trans, ret);
1083
1084         return ret;
1085
1086 search:
1087         btrfs_release_path(path);
1088
1089         key.type = BTRFS_INODE_EXTREF_KEY;
1090         key.offset = -1;
1091
1092         nofs_flag = memalloc_nofs_save();
1093         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1094         memalloc_nofs_restore(nofs_flag);
1095         if (ret < 0)
1096                 goto err_out;
1097         ASSERT(ret);
1098
1099         ret = 0;
1100         leaf = path->nodes[0];
1101         path->slots[0]--;
1102         goto again;
1103 }
1104
1105 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1106                                              struct btrfs_root *root,
1107                                              struct btrfs_path *path,
1108                                              struct btrfs_delayed_node *node)
1109 {
1110         int ret;
1111
1112         mutex_lock(&node->mutex);
1113         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1114                 mutex_unlock(&node->mutex);
1115                 return 0;
1116         }
1117
1118         ret = __btrfs_update_delayed_inode(trans, root, path, node);
1119         mutex_unlock(&node->mutex);
1120         return ret;
1121 }
1122
1123 static inline int
1124 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1125                                    struct btrfs_path *path,
1126                                    struct btrfs_delayed_node *node)
1127 {
1128         int ret;
1129
1130         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1131         if (ret)
1132                 return ret;
1133
1134         ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1135         if (ret)
1136                 return ret;
1137
1138         ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1139         return ret;
1140 }
1141
1142 /*
1143  * Called when committing the transaction.
1144  * Returns 0 on success.
1145  * Returns < 0 on error and returns with an aborted transaction with any
1146  * outstanding delayed items cleaned up.
1147  */
1148 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1149 {
1150         struct btrfs_fs_info *fs_info = trans->fs_info;
1151         struct btrfs_delayed_root *delayed_root;
1152         struct btrfs_delayed_node *curr_node, *prev_node;
1153         struct btrfs_path *path;
1154         struct btrfs_block_rsv *block_rsv;
1155         int ret = 0;
1156         bool count = (nr > 0);
1157
1158         if (TRANS_ABORTED(trans))
1159                 return -EIO;
1160
1161         path = btrfs_alloc_path();
1162         if (!path)
1163                 return -ENOMEM;
1164         path->leave_spinning = 1;
1165
1166         block_rsv = trans->block_rsv;
1167         trans->block_rsv = &fs_info->delayed_block_rsv;
1168
1169         delayed_root = fs_info->delayed_root;
1170
1171         curr_node = btrfs_first_delayed_node(delayed_root);
1172         while (curr_node && (!count || (count && nr--))) {
1173                 ret = __btrfs_commit_inode_delayed_items(trans, path,
1174                                                          curr_node);
1175                 if (ret) {
1176                         btrfs_abort_transaction(trans, ret);
1177                         break;
1178                 }
1179
1180                 prev_node = curr_node;
1181                 curr_node = btrfs_next_delayed_node(curr_node);
1182                 /*
1183                  * See the comment below about releasing path before releasing
1184                  * node. If the commit of delayed items was successful the path
1185                  * should always be released, but in case of an error, it may
1186                  * point to locked extent buffers (a leaf at the very least).
1187                  */
1188                 ASSERT(path->nodes[0] == NULL);
1189                 btrfs_release_delayed_node(prev_node);
1190         }
1191
1192         /*
1193          * Release the path to avoid a potential deadlock and lockdep splat when
1194          * releasing the delayed node, as that requires taking the delayed node's
1195          * mutex. If another task starts running delayed items before we take
1196          * the mutex, it will first lock the mutex and then it may try to lock
1197          * the same btree path (leaf).
1198          */
1199         btrfs_free_path(path);
1200
1201         if (curr_node)
1202                 btrfs_release_delayed_node(curr_node);
1203         trans->block_rsv = block_rsv;
1204
1205         return ret;
1206 }
1207
1208 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1209 {
1210         return __btrfs_run_delayed_items(trans, -1);
1211 }
1212
1213 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1214 {
1215         return __btrfs_run_delayed_items(trans, nr);
1216 }
1217
1218 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1219                                      struct btrfs_inode *inode)
1220 {
1221         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1222         struct btrfs_path *path;
1223         struct btrfs_block_rsv *block_rsv;
1224         int ret;
1225
1226         if (!delayed_node)
1227                 return 0;
1228
1229         mutex_lock(&delayed_node->mutex);
1230         if (!delayed_node->count) {
1231                 mutex_unlock(&delayed_node->mutex);
1232                 btrfs_release_delayed_node(delayed_node);
1233                 return 0;
1234         }
1235         mutex_unlock(&delayed_node->mutex);
1236
1237         path = btrfs_alloc_path();
1238         if (!path) {
1239                 btrfs_release_delayed_node(delayed_node);
1240                 return -ENOMEM;
1241         }
1242         path->leave_spinning = 1;
1243
1244         block_rsv = trans->block_rsv;
1245         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1246
1247         ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1248
1249         btrfs_release_delayed_node(delayed_node);
1250         btrfs_free_path(path);
1251         trans->block_rsv = block_rsv;
1252
1253         return ret;
1254 }
1255
1256 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1257 {
1258         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1259         struct btrfs_trans_handle *trans;
1260         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1261         struct btrfs_path *path;
1262         struct btrfs_block_rsv *block_rsv;
1263         int ret;
1264
1265         if (!delayed_node)
1266                 return 0;
1267
1268         mutex_lock(&delayed_node->mutex);
1269         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1270                 mutex_unlock(&delayed_node->mutex);
1271                 btrfs_release_delayed_node(delayed_node);
1272                 return 0;
1273         }
1274         mutex_unlock(&delayed_node->mutex);
1275
1276         trans = btrfs_join_transaction(delayed_node->root);
1277         if (IS_ERR(trans)) {
1278                 ret = PTR_ERR(trans);
1279                 goto out;
1280         }
1281
1282         path = btrfs_alloc_path();
1283         if (!path) {
1284                 ret = -ENOMEM;
1285                 goto trans_out;
1286         }
1287         path->leave_spinning = 1;
1288
1289         block_rsv = trans->block_rsv;
1290         trans->block_rsv = &fs_info->delayed_block_rsv;
1291
1292         mutex_lock(&delayed_node->mutex);
1293         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1294                 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1295                                                    path, delayed_node);
1296         else
1297                 ret = 0;
1298         mutex_unlock(&delayed_node->mutex);
1299
1300         btrfs_free_path(path);
1301         trans->block_rsv = block_rsv;
1302 trans_out:
1303         btrfs_end_transaction(trans);
1304         btrfs_btree_balance_dirty(fs_info);
1305 out:
1306         btrfs_release_delayed_node(delayed_node);
1307
1308         return ret;
1309 }
1310
1311 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1312 {
1313         struct btrfs_delayed_node *delayed_node;
1314
1315         delayed_node = READ_ONCE(inode->delayed_node);
1316         if (!delayed_node)
1317                 return;
1318
1319         inode->delayed_node = NULL;
1320         btrfs_release_delayed_node(delayed_node);
1321 }
1322
1323 struct btrfs_async_delayed_work {
1324         struct btrfs_delayed_root *delayed_root;
1325         int nr;
1326         struct btrfs_work work;
1327 };
1328
1329 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1330 {
1331         struct btrfs_async_delayed_work *async_work;
1332         struct btrfs_delayed_root *delayed_root;
1333         struct btrfs_trans_handle *trans;
1334         struct btrfs_path *path;
1335         struct btrfs_delayed_node *delayed_node = NULL;
1336         struct btrfs_root *root;
1337         struct btrfs_block_rsv *block_rsv;
1338         int total_done = 0;
1339
1340         async_work = container_of(work, struct btrfs_async_delayed_work, work);
1341         delayed_root = async_work->delayed_root;
1342
1343         path = btrfs_alloc_path();
1344         if (!path)
1345                 goto out;
1346
1347         do {
1348                 if (atomic_read(&delayed_root->items) <
1349                     BTRFS_DELAYED_BACKGROUND / 2)
1350                         break;
1351
1352                 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1353                 if (!delayed_node)
1354                         break;
1355
1356                 path->leave_spinning = 1;
1357                 root = delayed_node->root;
1358
1359                 trans = btrfs_join_transaction(root);
1360                 if (IS_ERR(trans)) {
1361                         btrfs_release_path(path);
1362                         btrfs_release_prepared_delayed_node(delayed_node);
1363                         total_done++;
1364                         continue;
1365                 }
1366
1367                 block_rsv = trans->block_rsv;
1368                 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1369
1370                 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1371
1372                 trans->block_rsv = block_rsv;
1373                 btrfs_end_transaction(trans);
1374                 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1375
1376                 btrfs_release_path(path);
1377                 btrfs_release_prepared_delayed_node(delayed_node);
1378                 total_done++;
1379
1380         } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1381                  || total_done < async_work->nr);
1382
1383         btrfs_free_path(path);
1384 out:
1385         wake_up(&delayed_root->wait);
1386         kfree(async_work);
1387 }
1388
1389
1390 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1391                                      struct btrfs_fs_info *fs_info, int nr)
1392 {
1393         struct btrfs_async_delayed_work *async_work;
1394
1395         async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1396         if (!async_work)
1397                 return -ENOMEM;
1398
1399         async_work->delayed_root = delayed_root;
1400         btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1401                         NULL);
1402         async_work->nr = nr;
1403
1404         btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1405         return 0;
1406 }
1407
1408 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1409 {
1410         WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1411 }
1412
1413 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1414 {
1415         int val = atomic_read(&delayed_root->items_seq);
1416
1417         if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1418                 return 1;
1419
1420         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1421                 return 1;
1422
1423         return 0;
1424 }
1425
1426 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1427 {
1428         struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1429
1430         if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1431                 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1432                 return;
1433
1434         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1435                 int seq;
1436                 int ret;
1437
1438                 seq = atomic_read(&delayed_root->items_seq);
1439
1440                 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1441                 if (ret)
1442                         return;
1443
1444                 wait_event_interruptible(delayed_root->wait,
1445                                          could_end_wait(delayed_root, seq));
1446                 return;
1447         }
1448
1449         btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1450 }
1451
1452 /* Will return 0 or -ENOMEM */
1453 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1454                                    const char *name, int name_len,
1455                                    struct btrfs_inode *dir,
1456                                    struct btrfs_disk_key *disk_key, u8 type,
1457                                    u64 index)
1458 {
1459         struct btrfs_delayed_node *delayed_node;
1460         struct btrfs_delayed_item *delayed_item;
1461         struct btrfs_dir_item *dir_item;
1462         int ret;
1463
1464         delayed_node = btrfs_get_or_create_delayed_node(dir);
1465         if (IS_ERR(delayed_node))
1466                 return PTR_ERR(delayed_node);
1467
1468         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1469         if (!delayed_item) {
1470                 ret = -ENOMEM;
1471                 goto release_node;
1472         }
1473
1474         delayed_item->key.objectid = btrfs_ino(dir);
1475         delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1476         delayed_item->key.offset = index;
1477
1478         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1479         dir_item->location = *disk_key;
1480         btrfs_set_stack_dir_transid(dir_item, trans->transid);
1481         btrfs_set_stack_dir_data_len(dir_item, 0);
1482         btrfs_set_stack_dir_name_len(dir_item, name_len);
1483         btrfs_set_stack_dir_type(dir_item, type);
1484         memcpy((char *)(dir_item + 1), name, name_len);
1485
1486         ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1487         /*
1488          * we have reserved enough space when we start a new transaction,
1489          * so reserving metadata failure is impossible
1490          */
1491         BUG_ON(ret);
1492
1493         mutex_lock(&delayed_node->mutex);
1494         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1495         if (unlikely(ret)) {
1496                 btrfs_err(trans->fs_info,
1497                           "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1498                           name_len, name, delayed_node->root->root_key.objectid,
1499                           delayed_node->inode_id, ret);
1500                 BUG();
1501         }
1502         mutex_unlock(&delayed_node->mutex);
1503
1504 release_node:
1505         btrfs_release_delayed_node(delayed_node);
1506         return ret;
1507 }
1508
1509 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1510                                                struct btrfs_delayed_node *node,
1511                                                struct btrfs_key *key)
1512 {
1513         struct btrfs_delayed_item *item;
1514
1515         mutex_lock(&node->mutex);
1516         item = __btrfs_lookup_delayed_insertion_item(node, key);
1517         if (!item) {
1518                 mutex_unlock(&node->mutex);
1519                 return 1;
1520         }
1521
1522         btrfs_delayed_item_release_metadata(node->root, item);
1523         btrfs_release_delayed_item(item);
1524         mutex_unlock(&node->mutex);
1525         return 0;
1526 }
1527
1528 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1529                                    struct btrfs_inode *dir, u64 index)
1530 {
1531         struct btrfs_delayed_node *node;
1532         struct btrfs_delayed_item *item;
1533         struct btrfs_key item_key;
1534         int ret;
1535
1536         node = btrfs_get_or_create_delayed_node(dir);
1537         if (IS_ERR(node))
1538                 return PTR_ERR(node);
1539
1540         item_key.objectid = btrfs_ino(dir);
1541         item_key.type = BTRFS_DIR_INDEX_KEY;
1542         item_key.offset = index;
1543
1544         ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1545                                                   &item_key);
1546         if (!ret)
1547                 goto end;
1548
1549         item = btrfs_alloc_delayed_item(0);
1550         if (!item) {
1551                 ret = -ENOMEM;
1552                 goto end;
1553         }
1554
1555         item->key = item_key;
1556
1557         ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1558         /*
1559          * we have reserved enough space when we start a new transaction,
1560          * so reserving metadata failure is impossible.
1561          */
1562         if (ret < 0) {
1563                 btrfs_err(trans->fs_info,
1564 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1565                 btrfs_release_delayed_item(item);
1566                 goto end;
1567         }
1568
1569         mutex_lock(&node->mutex);
1570         ret = __btrfs_add_delayed_deletion_item(node, item);
1571         if (unlikely(ret)) {
1572                 btrfs_err(trans->fs_info,
1573                           "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1574                           index, node->root->root_key.objectid,
1575                           node->inode_id, ret);
1576                 btrfs_delayed_item_release_metadata(dir->root, item);
1577                 btrfs_release_delayed_item(item);
1578         }
1579         mutex_unlock(&node->mutex);
1580 end:
1581         btrfs_release_delayed_node(node);
1582         return ret;
1583 }
1584
1585 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1586 {
1587         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1588
1589         if (!delayed_node)
1590                 return -ENOENT;
1591
1592         /*
1593          * Since we have held i_mutex of this directory, it is impossible that
1594          * a new directory index is added into the delayed node and index_cnt
1595          * is updated now. So we needn't lock the delayed node.
1596          */
1597         if (!delayed_node->index_cnt) {
1598                 btrfs_release_delayed_node(delayed_node);
1599                 return -EINVAL;
1600         }
1601
1602         inode->index_cnt = delayed_node->index_cnt;
1603         btrfs_release_delayed_node(delayed_node);
1604         return 0;
1605 }
1606
1607 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1608                                      struct list_head *ins_list,
1609                                      struct list_head *del_list)
1610 {
1611         struct btrfs_delayed_node *delayed_node;
1612         struct btrfs_delayed_item *item;
1613
1614         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1615         if (!delayed_node)
1616                 return false;
1617
1618         /*
1619          * We can only do one readdir with delayed items at a time because of
1620          * item->readdir_list.
1621          */
1622         inode_unlock_shared(inode);
1623         inode_lock(inode);
1624
1625         mutex_lock(&delayed_node->mutex);
1626         item = __btrfs_first_delayed_insertion_item(delayed_node);
1627         while (item) {
1628                 refcount_inc(&item->refs);
1629                 list_add_tail(&item->readdir_list, ins_list);
1630                 item = __btrfs_next_delayed_item(item);
1631         }
1632
1633         item = __btrfs_first_delayed_deletion_item(delayed_node);
1634         while (item) {
1635                 refcount_inc(&item->refs);
1636                 list_add_tail(&item->readdir_list, del_list);
1637                 item = __btrfs_next_delayed_item(item);
1638         }
1639         mutex_unlock(&delayed_node->mutex);
1640         /*
1641          * This delayed node is still cached in the btrfs inode, so refs
1642          * must be > 1 now, and we needn't check it is going to be freed
1643          * or not.
1644          *
1645          * Besides that, this function is used to read dir, we do not
1646          * insert/delete delayed items in this period. So we also needn't
1647          * requeue or dequeue this delayed node.
1648          */
1649         refcount_dec(&delayed_node->refs);
1650
1651         return true;
1652 }
1653
1654 void btrfs_readdir_put_delayed_items(struct inode *inode,
1655                                      struct list_head *ins_list,
1656                                      struct list_head *del_list)
1657 {
1658         struct btrfs_delayed_item *curr, *next;
1659
1660         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1661                 list_del(&curr->readdir_list);
1662                 if (refcount_dec_and_test(&curr->refs))
1663                         kfree(curr);
1664         }
1665
1666         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1667                 list_del(&curr->readdir_list);
1668                 if (refcount_dec_and_test(&curr->refs))
1669                         kfree(curr);
1670         }
1671
1672         /*
1673          * The VFS is going to do up_read(), so we need to downgrade back to a
1674          * read lock.
1675          */
1676         downgrade_write(&inode->i_rwsem);
1677 }
1678
1679 int btrfs_should_delete_dir_index(struct list_head *del_list,
1680                                   u64 index)
1681 {
1682         struct btrfs_delayed_item *curr;
1683         int ret = 0;
1684
1685         list_for_each_entry(curr, del_list, readdir_list) {
1686                 if (curr->key.offset > index)
1687                         break;
1688                 if (curr->key.offset == index) {
1689                         ret = 1;
1690                         break;
1691                 }
1692         }
1693         return ret;
1694 }
1695
1696 /*
1697  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1698  *
1699  */
1700 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1701                                     struct list_head *ins_list)
1702 {
1703         struct btrfs_dir_item *di;
1704         struct btrfs_delayed_item *curr, *next;
1705         struct btrfs_key location;
1706         char *name;
1707         int name_len;
1708         int over = 0;
1709         unsigned char d_type;
1710
1711         if (list_empty(ins_list))
1712                 return 0;
1713
1714         /*
1715          * Changing the data of the delayed item is impossible. So
1716          * we needn't lock them. And we have held i_mutex of the
1717          * directory, nobody can delete any directory indexes now.
1718          */
1719         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1720                 list_del(&curr->readdir_list);
1721
1722                 if (curr->key.offset < ctx->pos) {
1723                         if (refcount_dec_and_test(&curr->refs))
1724                                 kfree(curr);
1725                         continue;
1726                 }
1727
1728                 ctx->pos = curr->key.offset;
1729
1730                 di = (struct btrfs_dir_item *)curr->data;
1731                 name = (char *)(di + 1);
1732                 name_len = btrfs_stack_dir_name_len(di);
1733
1734                 d_type = fs_ftype_to_dtype(di->type);
1735                 btrfs_disk_key_to_cpu(&location, &di->location);
1736
1737                 over = !dir_emit(ctx, name, name_len,
1738                                location.objectid, d_type);
1739
1740                 if (refcount_dec_and_test(&curr->refs))
1741                         kfree(curr);
1742
1743                 if (over)
1744                         return 1;
1745                 ctx->pos++;
1746         }
1747         return 0;
1748 }
1749
1750 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1751                                   struct btrfs_inode_item *inode_item,
1752                                   struct inode *inode)
1753 {
1754         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1755         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1756         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1757         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1758         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1759         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1760         btrfs_set_stack_inode_generation(inode_item,
1761                                          BTRFS_I(inode)->generation);
1762         btrfs_set_stack_inode_sequence(inode_item,
1763                                        inode_peek_iversion(inode));
1764         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1765         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1766         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1767         btrfs_set_stack_inode_block_group(inode_item, 0);
1768
1769         btrfs_set_stack_timespec_sec(&inode_item->atime,
1770                                      inode->i_atime.tv_sec);
1771         btrfs_set_stack_timespec_nsec(&inode_item->atime,
1772                                       inode->i_atime.tv_nsec);
1773
1774         btrfs_set_stack_timespec_sec(&inode_item->mtime,
1775                                      inode->i_mtime.tv_sec);
1776         btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1777                                       inode->i_mtime.tv_nsec);
1778
1779         btrfs_set_stack_timespec_sec(&inode_item->ctime,
1780                                      inode->i_ctime.tv_sec);
1781         btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1782                                       inode->i_ctime.tv_nsec);
1783
1784         btrfs_set_stack_timespec_sec(&inode_item->otime,
1785                                      BTRFS_I(inode)->i_otime.tv_sec);
1786         btrfs_set_stack_timespec_nsec(&inode_item->otime,
1787                                      BTRFS_I(inode)->i_otime.tv_nsec);
1788 }
1789
1790 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1791 {
1792         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1793         struct btrfs_delayed_node *delayed_node;
1794         struct btrfs_inode_item *inode_item;
1795
1796         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1797         if (!delayed_node)
1798                 return -ENOENT;
1799
1800         mutex_lock(&delayed_node->mutex);
1801         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1802                 mutex_unlock(&delayed_node->mutex);
1803                 btrfs_release_delayed_node(delayed_node);
1804                 return -ENOENT;
1805         }
1806
1807         inode_item = &delayed_node->inode_item;
1808
1809         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1810         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1811         btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1812         btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1813                         round_up(i_size_read(inode), fs_info->sectorsize));
1814         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1815         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1816         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1817         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1818         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1819
1820         inode_set_iversion_queried(inode,
1821                                    btrfs_stack_inode_sequence(inode_item));
1822         inode->i_rdev = 0;
1823         *rdev = btrfs_stack_inode_rdev(inode_item);
1824         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1825
1826         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1827         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1828
1829         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1830         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1831
1832         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1833         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1834
1835         BTRFS_I(inode)->i_otime.tv_sec =
1836                 btrfs_stack_timespec_sec(&inode_item->otime);
1837         BTRFS_I(inode)->i_otime.tv_nsec =
1838                 btrfs_stack_timespec_nsec(&inode_item->otime);
1839
1840         inode->i_generation = BTRFS_I(inode)->generation;
1841         BTRFS_I(inode)->index_cnt = (u64)-1;
1842
1843         mutex_unlock(&delayed_node->mutex);
1844         btrfs_release_delayed_node(delayed_node);
1845         return 0;
1846 }
1847
1848 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1849                                struct btrfs_root *root, struct inode *inode)
1850 {
1851         struct btrfs_delayed_node *delayed_node;
1852         int ret = 0;
1853
1854         delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1855         if (IS_ERR(delayed_node))
1856                 return PTR_ERR(delayed_node);
1857
1858         mutex_lock(&delayed_node->mutex);
1859         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1860                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1861                 goto release_node;
1862         }
1863
1864         ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1865                                                    delayed_node);
1866         if (ret)
1867                 goto release_node;
1868
1869         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1870         set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1871         delayed_node->count++;
1872         atomic_inc(&root->fs_info->delayed_root->items);
1873 release_node:
1874         mutex_unlock(&delayed_node->mutex);
1875         btrfs_release_delayed_node(delayed_node);
1876         return ret;
1877 }
1878
1879 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1880 {
1881         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1882         struct btrfs_delayed_node *delayed_node;
1883
1884         /*
1885          * we don't do delayed inode updates during log recovery because it
1886          * leads to enospc problems.  This means we also can't do
1887          * delayed inode refs
1888          */
1889         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1890                 return -EAGAIN;
1891
1892         delayed_node = btrfs_get_or_create_delayed_node(inode);
1893         if (IS_ERR(delayed_node))
1894                 return PTR_ERR(delayed_node);
1895
1896         /*
1897          * We don't reserve space for inode ref deletion is because:
1898          * - We ONLY do async inode ref deletion for the inode who has only
1899          *   one link(i_nlink == 1), it means there is only one inode ref.
1900          *   And in most case, the inode ref and the inode item are in the
1901          *   same leaf, and we will deal with them at the same time.
1902          *   Since we are sure we will reserve the space for the inode item,
1903          *   it is unnecessary to reserve space for inode ref deletion.
1904          * - If the inode ref and the inode item are not in the same leaf,
1905          *   We also needn't worry about enospc problem, because we reserve
1906          *   much more space for the inode update than it needs.
1907          * - At the worst, we can steal some space from the global reservation.
1908          *   It is very rare.
1909          */
1910         mutex_lock(&delayed_node->mutex);
1911         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1912                 goto release_node;
1913
1914         set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1915         delayed_node->count++;
1916         atomic_inc(&fs_info->delayed_root->items);
1917 release_node:
1918         mutex_unlock(&delayed_node->mutex);
1919         btrfs_release_delayed_node(delayed_node);
1920         return 0;
1921 }
1922
1923 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1924 {
1925         struct btrfs_root *root = delayed_node->root;
1926         struct btrfs_fs_info *fs_info = root->fs_info;
1927         struct btrfs_delayed_item *curr_item, *prev_item;
1928
1929         mutex_lock(&delayed_node->mutex);
1930         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1931         while (curr_item) {
1932                 btrfs_delayed_item_release_metadata(root, curr_item);
1933                 prev_item = curr_item;
1934                 curr_item = __btrfs_next_delayed_item(prev_item);
1935                 btrfs_release_delayed_item(prev_item);
1936         }
1937
1938         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1939         while (curr_item) {
1940                 btrfs_delayed_item_release_metadata(root, curr_item);
1941                 prev_item = curr_item;
1942                 curr_item = __btrfs_next_delayed_item(prev_item);
1943                 btrfs_release_delayed_item(prev_item);
1944         }
1945
1946         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1947                 btrfs_release_delayed_iref(delayed_node);
1948
1949         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1950                 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1951                 btrfs_release_delayed_inode(delayed_node);
1952         }
1953         mutex_unlock(&delayed_node->mutex);
1954 }
1955
1956 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1957 {
1958         struct btrfs_delayed_node *delayed_node;
1959
1960         delayed_node = btrfs_get_delayed_node(inode);
1961         if (!delayed_node)
1962                 return;
1963
1964         __btrfs_kill_delayed_node(delayed_node);
1965         btrfs_release_delayed_node(delayed_node);
1966 }
1967
1968 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1969 {
1970         u64 inode_id = 0;
1971         struct btrfs_delayed_node *delayed_nodes[8];
1972         int i, n;
1973
1974         while (1) {
1975                 spin_lock(&root->inode_lock);
1976                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1977                                            (void **)delayed_nodes, inode_id,
1978                                            ARRAY_SIZE(delayed_nodes));
1979                 if (!n) {
1980                         spin_unlock(&root->inode_lock);
1981                         break;
1982                 }
1983
1984                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1985                 for (i = 0; i < n; i++) {
1986                         /*
1987                          * Don't increase refs in case the node is dead and
1988                          * about to be removed from the tree in the loop below
1989                          */
1990                         if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
1991                                 delayed_nodes[i] = NULL;
1992                 }
1993                 spin_unlock(&root->inode_lock);
1994
1995                 for (i = 0; i < n; i++) {
1996                         if (!delayed_nodes[i])
1997                                 continue;
1998                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1999                         btrfs_release_delayed_node(delayed_nodes[i]);
2000                 }
2001         }
2002 }
2003
2004 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
2005 {
2006         struct btrfs_delayed_node *curr_node, *prev_node;
2007
2008         curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
2009         while (curr_node) {
2010                 __btrfs_kill_delayed_node(curr_node);
2011
2012                 prev_node = curr_node;
2013                 curr_node = btrfs_next_delayed_node(curr_node);
2014                 btrfs_release_delayed_node(prev_node);
2015         }
2016 }
2017