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