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