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 <[email protected]>
   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_delayed_node *node)
 606 {
 607         struct btrfs_fs_info *fs_info = root->fs_info;
 608         struct btrfs_block_rsv *src_rsv;
 609         struct btrfs_block_rsv *dst_rsv;
 610         u64 num_bytes;
 611         int ret;
 612
 613         src_rsv = trans->block_rsv;
 614         dst_rsv = &fs_info->delayed_block_rsv;
 615
 616         num_bytes = btrfs_calc_metadata_size(fs_info, 1);
 617
 618         /*
 619          * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 620          * which doesn't reserve space for speed.  This is a problem since we
 621          * still need to reserve space for this update, so try to reserve the
 622          * space.
 623          *
 624          * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 625          * we always reserve enough to update the inode item.
 626          */
 627         if (!src_rsv || (!trans->bytes_reserved &&
 628                          src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
 629                 ret = btrfs_qgroup_reserve_meta(root, num_bytes,
 630                                           BTRFS_QGROUP_RSV_META_PREALLOC, 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                 /* NO_FLUSH could only fail with -ENOSPC */
 636                 ASSERT(ret == 0 || ret == -ENOSPC);
 637                 if (ret)
 638                         btrfs_qgroup_free_meta_prealloc(root, num_bytes);
 639         } else {
 640                 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
 641         }
 642
 643         if (!ret) {
 644                 trace_btrfs_space_reservation(fs_info, "delayed_inode",
 645                                               node->inode_id, num_bytes, 1);
 646                 node->bytes_reserved = num_bytes;
 647         }
 648
 649         return ret;
 650 }
 651
 652 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
 653                                                 struct btrfs_delayed_node *node,
 654                                                 bool qgroup_free)
 655 {
 656         struct btrfs_block_rsv *rsv;
 657
 658         if (!node->bytes_reserved)
 659                 return;
 660
 661         rsv = &fs_info->delayed_block_rsv;
 662         trace_btrfs_space_reservation(fs_info, "delayed_inode",
 663                                       node->inode_id, node->bytes_reserved, 0);
 664         btrfs_block_rsv_release(fs_info, rsv, node->bytes_reserved, NULL);
 665         if (qgroup_free)
 666                 btrfs_qgroup_free_meta_prealloc(node->root,
 667                                 node->bytes_reserved);
 668         else
 669                 btrfs_qgroup_convert_reserved_meta(node->root,
 670                                 node->bytes_reserved);
 671         node->bytes_reserved = 0;
 672 }
 673
 674 /*
 675  * This helper will insert some continuous items into the same leaf according
 676  * to the free space of the leaf.
 677  */
 678 static int btrfs_batch_insert_items(struct btrfs_root *root,
 679                                     struct btrfs_path *path,
 680                                     struct btrfs_delayed_item *item)
 681 {
 682         struct btrfs_delayed_item *curr, *next;
 683         int free_space;
 684         int total_data_size = 0, total_size = 0;
 685         struct extent_buffer *leaf;
 686         char *data_ptr;
 687         struct btrfs_key *keys;
 688         u32 *data_size;
 689         struct list_head head;
 690         int slot;
 691         int nitems;
 692         int i;
 693         int ret = 0;
 694
 695         BUG_ON(!path->nodes[0]);
 696
 697         leaf = path->nodes[0];
 698         free_space = btrfs_leaf_free_space(leaf);
 699         INIT_LIST_HEAD(&head);
 700
 701         next = item;
 702         nitems = 0;
 703
 704         /*
 705          * count the number of the continuous items that we can insert in batch
 706          */
 707         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 708                free_space) {
 709                 total_data_size += next->data_len;
 710                 total_size += next->data_len + sizeof(struct btrfs_item);
 711                 list_add_tail(&next->tree_list, &head);
 712                 nitems++;
 713
 714                 curr = next;
 715                 next = __btrfs_next_delayed_item(curr);
 716                 if (!next)
 717                         break;
 718
 719                 if (!btrfs_is_continuous_delayed_item(curr, next))
 720                         break;
 721         }
 722
 723         if (!nitems) {
 724                 ret = 0;
 725                 goto out;
 726         }
 727
 728         keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
 729         if (!keys) {
 730                 ret = -ENOMEM;
 731                 goto out;
 732         }
 733
 734         data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
 735         if (!data_size) {
 736                 ret = -ENOMEM;
 737                 goto error;
 738         }
 739
 740         /* get keys of all the delayed items */
 741         i = 0;
 742         list_for_each_entry(next, &head, tree_list) {
 743                 keys[i] = next->key;
 744                 data_size[i] = next->data_len;
 745                 i++;
 746         }
 747
 748         /* insert the keys of the items */
 749         setup_items_for_insert(root, path, keys, data_size, nitems);
 750
 751         /* insert the dir index items */
 752         slot = path->slots[0];
 753         list_for_each_entry_safe(curr, next, &head, tree_list) {
 754                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 755                 write_extent_buffer(leaf, &curr->data,
 756                                     (unsigned long)data_ptr,
 757                                     curr->data_len);
 758                 slot++;
 759
 760                 btrfs_delayed_item_release_metadata(root, curr);
 761
 762                 list_del(&curr->tree_list);
 763                 btrfs_release_delayed_item(curr);
 764         }
 765
 766 error:
 767         kfree(data_size);
 768         kfree(keys);
 769 out:
 770         return ret;
 771 }
 772
 773 /*
 774  * This helper can just do simple insertion that needn't extend item for new
 775  * data, such as directory name index insertion, inode insertion.
 776  */
 777 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 778                                      struct btrfs_root *root,
 779                                      struct btrfs_path *path,
 780                                      struct btrfs_delayed_item *delayed_item)
 781 {
 782         struct extent_buffer *leaf;
 783         unsigned int nofs_flag;
 784         char *ptr;
 785         int ret;
 786
 787         nofs_flag = memalloc_nofs_save();
 788         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 789                                       delayed_item->data_len);
 790         memalloc_nofs_restore(nofs_flag);
 791         if (ret < 0 && ret != -EEXIST)
 792                 return ret;
 793
 794         leaf = path->nodes[0];
 795
 796         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 797
 798         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 799                             delayed_item->data_len);
 800         btrfs_mark_buffer_dirty(leaf);
 801
 802         btrfs_delayed_item_release_metadata(root, delayed_item);
 803         return 0;
 804 }
 805
 806 /*
 807  * we insert an item first, then if there are some continuous items, we try
 808  * to insert those items into the same leaf.
 809  */
 810 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 811                                       struct btrfs_path *path,
 812                                       struct btrfs_root *root,
 813                                       struct btrfs_delayed_node *node)
 814 {
 815         struct btrfs_delayed_item *curr, *prev;
 816         int ret = 0;
 817
 818 do_again:
 819         mutex_lock(&node->mutex);
 820         curr = __btrfs_first_delayed_insertion_item(node);
 821         if (!curr)
 822                 goto insert_end;
 823
 824         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 825         if (ret < 0) {
 826                 btrfs_release_path(path);
 827                 goto insert_end;
 828         }
 829
 830         prev = curr;
 831         curr = __btrfs_next_delayed_item(prev);
 832         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 833                 /* insert the continuous items into the same leaf */
 834                 path->slots[0]++;
 835                 btrfs_batch_insert_items(root, path, curr);
 836         }
 837         btrfs_release_delayed_item(prev);
 838         btrfs_mark_buffer_dirty(path->nodes[0]);
 839
 840         btrfs_release_path(path);
 841         mutex_unlock(&node->mutex);
 842         goto do_again;
 843
 844 insert_end:
 845         mutex_unlock(&node->mutex);
 846         return ret;
 847 }
 848
 849 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 850                                     struct btrfs_root *root,
 851                                     struct btrfs_path *path,
 852                                     struct btrfs_delayed_item *item)
 853 {
 854         struct btrfs_delayed_item *curr, *next;
 855         struct extent_buffer *leaf;
 856         struct btrfs_key key;
 857         struct list_head head;
 858         int nitems, i, last_item;
 859         int ret = 0;
 860
 861         BUG_ON(!path->nodes[0]);
 862
 863         leaf = path->nodes[0];
 864
 865         i = path->slots[0];
 866         last_item = btrfs_header_nritems(leaf) - 1;
 867         if (i > last_item)
 868                 return -ENOENT; /* FIXME: Is errno suitable? */
 869
 870         next = item;
 871         INIT_LIST_HEAD(&head);
 872         btrfs_item_key_to_cpu(leaf, &key, i);
 873         nitems = 0;
 874         /*
 875          * count the number of the dir index items that we can delete in batch
 876          */
 877         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 878                 list_add_tail(&next->tree_list, &head);
 879                 nitems++;
 880
 881                 curr = next;
 882                 next = __btrfs_next_delayed_item(curr);
 883                 if (!next)
 884                         break;
 885
 886                 if (!btrfs_is_continuous_delayed_item(curr, next))
 887                         break;
 888
 889                 i++;
 890                 if (i > last_item)
 891                         break;
 892                 btrfs_item_key_to_cpu(leaf, &key, i);
 893         }
 894
 895         if (!nitems)
 896                 return 0;
 897
 898         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 899         if (ret)
 900                 goto out;
 901
 902         list_for_each_entry_safe(curr, next, &head, tree_list) {
 903                 btrfs_delayed_item_release_metadata(root, curr);
 904                 list_del(&curr->tree_list);
 905                 btrfs_release_delayed_item(curr);
 906         }
 907
 908 out:
 909         return ret;
 910 }
 911
 912 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 913                                       struct btrfs_path *path,
 914                                       struct btrfs_root *root,
 915                                       struct btrfs_delayed_node *node)
 916 {
 917         struct btrfs_delayed_item *curr, *prev;
 918         unsigned int nofs_flag;
 919         int ret = 0;
 920
 921 do_again:
 922         mutex_lock(&node->mutex);
 923         curr = __btrfs_first_delayed_deletion_item(node);
 924         if (!curr)
 925                 goto delete_fail;
 926
 927         nofs_flag = memalloc_nofs_save();
 928         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 929         memalloc_nofs_restore(nofs_flag);
 930         if (ret < 0)
 931                 goto delete_fail;
 932         else if (ret > 0) {
 933                 /*
 934                  * can't find the item which the node points to, so this node
 935                  * is invalid, just drop it.
 936                  */
 937                 prev = curr;
 938                 curr = __btrfs_next_delayed_item(prev);
 939                 btrfs_release_delayed_item(prev);
 940                 ret = 0;
 941                 btrfs_release_path(path);
 942                 if (curr) {
 943                         mutex_unlock(&node->mutex);
 944                         goto do_again;
 945                 } else
 946                         goto delete_fail;
 947         }
 948
 949         btrfs_batch_delete_items(trans, root, path, curr);
 950         btrfs_release_path(path);
 951         mutex_unlock(&node->mutex);
 952         goto do_again;
 953
 954 delete_fail:
 955         btrfs_release_path(path);
 956         mutex_unlock(&node->mutex);
 957         return ret;
 958 }
 959
 960 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 961 {
 962         struct btrfs_delayed_root *delayed_root;
 963
 964         if (delayed_node &&
 965             test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
 966                 BUG_ON(!delayed_node->root);
 967                 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
 968                 delayed_node->count--;
 969
 970                 delayed_root = delayed_node->root->fs_info->delayed_root;
 971                 finish_one_item(delayed_root);
 972         }
 973 }
 974
 975 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
 976 {
 977
 978         if (test_and_clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags)) {
 979                 struct btrfs_delayed_root *delayed_root;
 980
 981                 ASSERT(delayed_node->root);
 982                 delayed_node->count--;
 983
 984                 delayed_root = delayed_node->root->fs_info->delayed_root;
 985                 finish_one_item(delayed_root);
 986         }
 987 }
 988
 989 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
 990                                         struct btrfs_root *root,
 991                                         struct btrfs_path *path,
 992                                         struct btrfs_delayed_node *node)
 993 {
 994         struct btrfs_fs_info *fs_info = root->fs_info;
 995         struct btrfs_key key;
 996         struct btrfs_inode_item *inode_item;
 997         struct extent_buffer *leaf;
 998         unsigned int nofs_flag;
 999         int mod;
1000         int ret;
1001
1002         key.objectid = node->inode_id;
1003         key.type = BTRFS_INODE_ITEM_KEY;
1004         key.offset = 0;
1005
1006         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1007                 mod = -1;
1008         else
1009                 mod = 1;
1010
1011         nofs_flag = memalloc_nofs_save();
1012         ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1013         memalloc_nofs_restore(nofs_flag);
1014         if (ret > 0)
1015                 ret = -ENOENT;
1016         if (ret < 0)
1017                 goto out;
1018
1019         leaf = path->nodes[0];
1020         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1021                                     struct btrfs_inode_item);
1022         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1023                             sizeof(struct btrfs_inode_item));
1024         btrfs_mark_buffer_dirty(leaf);
1025
1026         if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1027                 goto out;
1028
1029         path->slots[0]++;
1030         if (path->slots[0] >= btrfs_header_nritems(leaf))
1031                 goto search;
1032 again:
1033         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1034         if (key.objectid != node->inode_id)
1035                 goto out;
1036
1037         if (key.type != BTRFS_INODE_REF_KEY &&
1038             key.type != BTRFS_INODE_EXTREF_KEY)
1039                 goto out;
1040
1041         /*
1042          * Delayed iref deletion is for the inode who has only one link,
1043          * so there is only one iref. The case that several irefs are
1044          * in the same item doesn't exist.
1045          */
1046         btrfs_del_item(trans, root, path);
1047 out:
1048         btrfs_release_delayed_iref(node);
1049         btrfs_release_path(path);
1050 err_out:
1051         btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1052         btrfs_release_delayed_inode(node);
1053
1054         /*
1055          * If we fail to update the delayed inode we need to abort the
1056          * transaction, because we could leave the inode with the improper
1057          * counts behind.
1058          */
1059         if (ret && ret != -ENOENT)
1060                 btrfs_abort_transaction(trans, ret);
1061
1062         return ret;
1063
1064 search:
1065         btrfs_release_path(path);
1066
1067         key.type = BTRFS_INODE_EXTREF_KEY;
1068         key.offset = -1;
1069
1070         nofs_flag = memalloc_nofs_save();
1071         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1072         memalloc_nofs_restore(nofs_flag);
1073         if (ret < 0)
1074                 goto err_out;
1075         ASSERT(ret);
1076
1077         ret = 0;
1078         leaf = path->nodes[0];
1079         path->slots[0]--;
1080         goto again;
1081 }
1082
1083 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1084                                              struct btrfs_root *root,
1085                                              struct btrfs_path *path,
1086                                              struct btrfs_delayed_node *node)
1087 {
1088         int ret;
1089
1090         mutex_lock(&node->mutex);
1091         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1092                 mutex_unlock(&node->mutex);
1093                 return 0;
1094         }
1095
1096         ret = __btrfs_update_delayed_inode(trans, root, path, node);
1097         mutex_unlock(&node->mutex);
1098         return ret;
1099 }
1100
1101 static inline int
1102 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1103                                    struct btrfs_path *path,
1104                                    struct btrfs_delayed_node *node)
1105 {
1106         int ret;
1107
1108         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1109         if (ret)
1110                 return ret;
1111
1112         ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1113         if (ret)
1114                 return ret;
1115
1116         ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1117         return ret;
1118 }
1119
1120 /*
1121  * Called when committing the transaction.
1122  * Returns 0 on success.
1123  * Returns < 0 on error and returns with an aborted transaction with any
1124  * outstanding delayed items cleaned up.
1125  */
1126 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1127 {
1128         struct btrfs_fs_info *fs_info = trans->fs_info;
1129         struct btrfs_delayed_root *delayed_root;
1130         struct btrfs_delayed_node *curr_node, *prev_node;
1131         struct btrfs_path *path;
1132         struct btrfs_block_rsv *block_rsv;
1133         int ret = 0;
1134         bool count = (nr > 0);
1135
1136         if (TRANS_ABORTED(trans))
1137                 return -EIO;
1138
1139         path = btrfs_alloc_path();
1140         if (!path)
1141                 return -ENOMEM;
1142
1143         block_rsv = trans->block_rsv;
1144         trans->block_rsv = &fs_info->delayed_block_rsv;
1145
1146         delayed_root = fs_info->delayed_root;
1147
1148         curr_node = btrfs_first_delayed_node(delayed_root);
1149         while (curr_node && (!count || nr--)) {
1150                 ret = __btrfs_commit_inode_delayed_items(trans, path,
1151                                                          curr_node);
1152                 if (ret) {
1153                         btrfs_release_delayed_node(curr_node);
1154                         curr_node = NULL;
1155                         btrfs_abort_transaction(trans, ret);
1156                         break;
1157                 }
1158
1159                 prev_node = curr_node;
1160                 curr_node = btrfs_next_delayed_node(curr_node);
1161                 btrfs_release_delayed_node(prev_node);
1162         }
1163
1164         if (curr_node)
1165                 btrfs_release_delayed_node(curr_node);
1166         btrfs_free_path(path);
1167         trans->block_rsv = block_rsv;
1168
1169         return ret;
1170 }
1171
1172 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1173 {
1174         return __btrfs_run_delayed_items(trans, -1);
1175 }
1176
1177 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1178 {
1179         return __btrfs_run_delayed_items(trans, nr);
1180 }
1181
1182 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1183                                      struct btrfs_inode *inode)
1184 {
1185         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1186         struct btrfs_path *path;
1187         struct btrfs_block_rsv *block_rsv;
1188         int ret;
1189
1190         if (!delayed_node)
1191                 return 0;
1192
1193         mutex_lock(&delayed_node->mutex);
1194         if (!delayed_node->count) {
1195                 mutex_unlock(&delayed_node->mutex);
1196                 btrfs_release_delayed_node(delayed_node);
1197                 return 0;
1198         }
1199         mutex_unlock(&delayed_node->mutex);
1200
1201         path = btrfs_alloc_path();
1202         if (!path) {
1203                 btrfs_release_delayed_node(delayed_node);
1204                 return -ENOMEM;
1205         }
1206
1207         block_rsv = trans->block_rsv;
1208         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1209
1210         ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1211
1212         btrfs_release_delayed_node(delayed_node);
1213         btrfs_free_path(path);
1214         trans->block_rsv = block_rsv;
1215
1216         return ret;
1217 }
1218
1219 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1220 {
1221         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1222         struct btrfs_trans_handle *trans;
1223         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1224         struct btrfs_path *path;
1225         struct btrfs_block_rsv *block_rsv;
1226         int ret;
1227
1228         if (!delayed_node)
1229                 return 0;
1230
1231         mutex_lock(&delayed_node->mutex);
1232         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1233                 mutex_unlock(&delayed_node->mutex);
1234                 btrfs_release_delayed_node(delayed_node);
1235                 return 0;
1236         }
1237         mutex_unlock(&delayed_node->mutex);
1238
1239         trans = btrfs_join_transaction(delayed_node->root);
1240         if (IS_ERR(trans)) {
1241                 ret = PTR_ERR(trans);
1242                 goto out;
1243         }
1244
1245         path = btrfs_alloc_path();
1246         if (!path) {
1247                 ret = -ENOMEM;
1248                 goto trans_out;
1249         }
1250
1251         block_rsv = trans->block_rsv;
1252         trans->block_rsv = &fs_info->delayed_block_rsv;
1253
1254         mutex_lock(&delayed_node->mutex);
1255         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1256                 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1257                                                    path, delayed_node);
1258         else
1259                 ret = 0;
1260         mutex_unlock(&delayed_node->mutex);
1261
1262         btrfs_free_path(path);
1263         trans->block_rsv = block_rsv;
1264 trans_out:
1265         btrfs_end_transaction(trans);
1266         btrfs_btree_balance_dirty(fs_info);
1267 out:
1268         btrfs_release_delayed_node(delayed_node);
1269
1270         return ret;
1271 }
1272
1273 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1274 {
1275         struct btrfs_delayed_node *delayed_node;
1276
1277         delayed_node = READ_ONCE(inode->delayed_node);
1278         if (!delayed_node)
1279                 return;
1280
1281         inode->delayed_node = NULL;
1282         btrfs_release_delayed_node(delayed_node);
1283 }
1284
1285 struct btrfs_async_delayed_work {
1286         struct btrfs_delayed_root *delayed_root;
1287         int nr;
1288         struct btrfs_work work;
1289 };
1290
1291 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1292 {
1293         struct btrfs_async_delayed_work *async_work;
1294         struct btrfs_delayed_root *delayed_root;
1295         struct btrfs_trans_handle *trans;
1296         struct btrfs_path *path;
1297         struct btrfs_delayed_node *delayed_node = NULL;
1298         struct btrfs_root *root;
1299         struct btrfs_block_rsv *block_rsv;
1300         int total_done = 0;
1301
1302         async_work = container_of(work, struct btrfs_async_delayed_work, work);
1303         delayed_root = async_work->delayed_root;
1304
1305         path = btrfs_alloc_path();
1306         if (!path)
1307                 goto out;
1308
1309         do {
1310                 if (atomic_read(&delayed_root->items) <
1311                     BTRFS_DELAYED_BACKGROUND / 2)
1312                         break;
1313
1314                 delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1315                 if (!delayed_node)
1316                         break;
1317
1318                 root = delayed_node->root;
1319
1320                 trans = btrfs_join_transaction(root);
1321                 if (IS_ERR(trans)) {
1322                         btrfs_release_path(path);
1323                         btrfs_release_prepared_delayed_node(delayed_node);
1324                         total_done++;
1325                         continue;
1326                 }
1327
1328                 block_rsv = trans->block_rsv;
1329                 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1330
1331                 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1332
1333                 trans->block_rsv = block_rsv;
1334                 btrfs_end_transaction(trans);
1335                 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1336
1337                 btrfs_release_path(path);
1338                 btrfs_release_prepared_delayed_node(delayed_node);
1339                 total_done++;
1340
1341         } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1342                  || total_done < async_work->nr);
1343
1344         btrfs_free_path(path);
1345 out:
1346         wake_up(&delayed_root->wait);
1347         kfree(async_work);
1348 }
1349
1350
1351 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1352                                      struct btrfs_fs_info *fs_info, int nr)
1353 {
1354         struct btrfs_async_delayed_work *async_work;
1355
1356         async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1357         if (!async_work)
1358                 return -ENOMEM;
1359
1360         async_work->delayed_root = delayed_root;
1361         btrfs_init_work(&async_work->work, btrfs_async_run_delayed_root, NULL,
1362                         NULL);
1363         async_work->nr = nr;
1364
1365         btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1366         return 0;
1367 }
1368
1369 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1370 {
1371         WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1372 }
1373
1374 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1375 {
1376         int val = atomic_read(&delayed_root->items_seq);
1377
1378         if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1379                 return 1;
1380
1381         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1382                 return 1;
1383
1384         return 0;
1385 }
1386
1387 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1388 {
1389         struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1390
1391         if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1392                 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1393                 return;
1394
1395         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1396                 int seq;
1397                 int ret;
1398
1399                 seq = atomic_read(&delayed_root->items_seq);
1400
1401                 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1402                 if (ret)
1403                         return;
1404
1405                 wait_event_interruptible(delayed_root->wait,
1406                                          could_end_wait(delayed_root, seq));
1407                 return;
1408         }
1409
1410         btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1411 }
1412
1413 /* Will return 0 or -ENOMEM */
1414 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1415                                    const char *name, int name_len,
1416                                    struct btrfs_inode *dir,
1417                                    struct btrfs_disk_key *disk_key, u8 type,
1418                                    u64 index)
1419 {
1420         struct btrfs_delayed_node *delayed_node;
1421         struct btrfs_delayed_item *delayed_item;
1422         struct btrfs_dir_item *dir_item;
1423         int ret;
1424
1425         delayed_node = btrfs_get_or_create_delayed_node(dir);
1426         if (IS_ERR(delayed_node))
1427                 return PTR_ERR(delayed_node);
1428
1429         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1430         if (!delayed_item) {
1431                 ret = -ENOMEM;
1432                 goto release_node;
1433         }
1434
1435         delayed_item->key.objectid = btrfs_ino(dir);
1436         delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1437         delayed_item->key.offset = index;
1438
1439         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1440         dir_item->location = *disk_key;
1441         btrfs_set_stack_dir_transid(dir_item, trans->transid);
1442         btrfs_set_stack_dir_data_len(dir_item, 0);
1443         btrfs_set_stack_dir_name_len(dir_item, name_len);
1444         btrfs_set_stack_dir_type(dir_item, type);
1445         memcpy((char *)(dir_item + 1), name, name_len);
1446
1447         ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1448         /*
1449          * we have reserved enough space when we start a new transaction,
1450          * so reserving metadata failure is impossible
1451          */
1452         BUG_ON(ret);
1453
1454         mutex_lock(&delayed_node->mutex);
1455         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1456         if (unlikely(ret)) {
1457                 btrfs_err(trans->fs_info,
1458                           "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1459                           name_len, name, delayed_node->root->root_key.objectid,
1460                           delayed_node->inode_id, ret);
1461                 BUG();
1462         }
1463         mutex_unlock(&delayed_node->mutex);
1464
1465 release_node:
1466         btrfs_release_delayed_node(delayed_node);
1467         return ret;
1468 }
1469
1470 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1471                                                struct btrfs_delayed_node *node,
1472                                                struct btrfs_key *key)
1473 {
1474         struct btrfs_delayed_item *item;
1475
1476         mutex_lock(&node->mutex);
1477         item = __btrfs_lookup_delayed_insertion_item(node, key);
1478         if (!item) {
1479                 mutex_unlock(&node->mutex);
1480                 return 1;
1481         }
1482
1483         btrfs_delayed_item_release_metadata(node->root, item);
1484         btrfs_release_delayed_item(item);
1485         mutex_unlock(&node->mutex);
1486         return 0;
1487 }
1488
1489 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1490                                    struct btrfs_inode *dir, u64 index)
1491 {
1492         struct btrfs_delayed_node *node;
1493         struct btrfs_delayed_item *item;
1494         struct btrfs_key item_key;
1495         int ret;
1496
1497         node = btrfs_get_or_create_delayed_node(dir);
1498         if (IS_ERR(node))
1499                 return PTR_ERR(node);
1500
1501         item_key.objectid = btrfs_ino(dir);
1502         item_key.type = BTRFS_DIR_INDEX_KEY;
1503         item_key.offset = index;
1504
1505         ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1506                                                   &item_key);
1507         if (!ret)
1508                 goto end;
1509
1510         item = btrfs_alloc_delayed_item(0);
1511         if (!item) {
1512                 ret = -ENOMEM;
1513                 goto end;
1514         }
1515
1516         item->key = item_key;
1517
1518         ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1519         /*
1520          * we have reserved enough space when we start a new transaction,
1521          * so reserving metadata failure is impossible.
1522          */
1523         if (ret < 0) {
1524                 btrfs_err(trans->fs_info,
1525 "metadata reservation failed for delayed dir item deltiona, should have been reserved");
1526                 btrfs_release_delayed_item(item);
1527                 goto end;
1528         }
1529
1530         mutex_lock(&node->mutex);
1531         ret = __btrfs_add_delayed_deletion_item(node, item);
1532         if (unlikely(ret)) {
1533                 btrfs_err(trans->fs_info,
1534                           "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1535                           index, node->root->root_key.objectid,
1536                           node->inode_id, ret);
1537                 btrfs_delayed_item_release_metadata(dir->root, item);
1538                 btrfs_release_delayed_item(item);
1539         }
1540         mutex_unlock(&node->mutex);
1541 end:
1542         btrfs_release_delayed_node(node);
1543         return ret;
1544 }
1545
1546 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1547 {
1548         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1549
1550         if (!delayed_node)
1551                 return -ENOENT;
1552
1553         /*
1554          * Since we have held i_mutex of this directory, it is impossible that
1555          * a new directory index is added into the delayed node and index_cnt
1556          * is updated now. So we needn't lock the delayed node.
1557          */
1558         if (!delayed_node->index_cnt) {
1559                 btrfs_release_delayed_node(delayed_node);
1560                 return -EINVAL;
1561         }
1562
1563         inode->index_cnt = delayed_node->index_cnt;
1564         btrfs_release_delayed_node(delayed_node);
1565         return 0;
1566 }
1567
1568 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1569                                      struct list_head *ins_list,
1570                                      struct list_head *del_list)
1571 {
1572         struct btrfs_delayed_node *delayed_node;
1573         struct btrfs_delayed_item *item;
1574
1575         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1576         if (!delayed_node)
1577                 return false;
1578
1579         /*
1580          * We can only do one readdir with delayed items at a time because of
1581          * item->readdir_list.
1582          */
1583         btrfs_inode_unlock(inode, BTRFS_ILOCK_SHARED);
1584         btrfs_inode_lock(inode, 0);
1585
1586         mutex_lock(&delayed_node->mutex);
1587         item = __btrfs_first_delayed_insertion_item(delayed_node);
1588         while (item) {
1589                 refcount_inc(&item->refs);
1590                 list_add_tail(&item->readdir_list, ins_list);
1591                 item = __btrfs_next_delayed_item(item);
1592         }
1593
1594         item = __btrfs_first_delayed_deletion_item(delayed_node);
1595         while (item) {
1596                 refcount_inc(&item->refs);
1597                 list_add_tail(&item->readdir_list, del_list);
1598                 item = __btrfs_next_delayed_item(item);
1599         }
1600         mutex_unlock(&delayed_node->mutex);
1601         /*
1602          * This delayed node is still cached in the btrfs inode, so refs
1603          * must be > 1 now, and we needn't check it is going to be freed
1604          * or not.
1605          *
1606          * Besides that, this function is used to read dir, we do not
1607          * insert/delete delayed items in this period. So we also needn't
1608          * requeue or dequeue this delayed node.
1609          */
1610         refcount_dec(&delayed_node->refs);
1611
1612         return true;
1613 }
1614
1615 void btrfs_readdir_put_delayed_items(struct inode *inode,
1616                                      struct list_head *ins_list,
1617                                      struct list_head *del_list)
1618 {
1619         struct btrfs_delayed_item *curr, *next;
1620
1621         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1622                 list_del(&curr->readdir_list);
1623                 if (refcount_dec_and_test(&curr->refs))
1624                         kfree(curr);
1625         }
1626
1627         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1628                 list_del(&curr->readdir_list);
1629                 if (refcount_dec_and_test(&curr->refs))
1630                         kfree(curr);
1631         }
1632
1633         /*
1634          * The VFS is going to do up_read(), so we need to downgrade back to a
1635          * read lock.
1636          */
1637         downgrade_write(&inode->i_rwsem);
1638 }
1639
1640 int btrfs_should_delete_dir_index(struct list_head *del_list,
1641                                   u64 index)
1642 {
1643         struct btrfs_delayed_item *curr;
1644         int ret = 0;
1645
1646         list_for_each_entry(curr, del_list, readdir_list) {
1647                 if (curr->key.offset > index)
1648                         break;
1649                 if (curr->key.offset == index) {
1650                         ret = 1;
1651                         break;
1652                 }
1653         }
1654         return ret;
1655 }
1656
1657 /*
1658  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1659  *
1660  */
1661 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1662                                     struct list_head *ins_list)
1663 {
1664         struct btrfs_dir_item *di;
1665         struct btrfs_delayed_item *curr, *next;
1666         struct btrfs_key location;
1667         char *name;
1668         int name_len;
1669         int over = 0;
1670         unsigned char d_type;
1671
1672         if (list_empty(ins_list))
1673                 return 0;
1674
1675         /*
1676          * Changing the data of the delayed item is impossible. So
1677          * we needn't lock them. And we have held i_mutex of the
1678          * directory, nobody can delete any directory indexes now.
1679          */
1680         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1681                 list_del(&curr->readdir_list);
1682
1683                 if (curr->key.offset < ctx->pos) {
1684                         if (refcount_dec_and_test(&curr->refs))
1685                                 kfree(curr);
1686                         continue;
1687                 }
1688
1689                 ctx->pos = curr->key.offset;
1690
1691                 di = (struct btrfs_dir_item *)curr->data;
1692                 name = (char *)(di + 1);
1693                 name_len = btrfs_stack_dir_name_len(di);
1694
1695                 d_type = fs_ftype_to_dtype(di->type);
1696                 btrfs_disk_key_to_cpu(&location, &di->location);
1697
1698                 over = !dir_emit(ctx, name, name_len,
1699                                location.objectid, d_type);
1700
1701                 if (refcount_dec_and_test(&curr->refs))
1702                         kfree(curr);
1703
1704                 if (over)
1705                         return 1;
1706                 ctx->pos++;
1707         }
1708         return 0;
1709 }
1710
1711 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1712                                   struct btrfs_inode_item *inode_item,
1713                                   struct inode *inode)
1714 {
1715         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1716         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1717         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1718         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1719         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1720         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1721         btrfs_set_stack_inode_generation(inode_item,
1722                                          BTRFS_I(inode)->generation);
1723         btrfs_set_stack_inode_sequence(inode_item,
1724                                        inode_peek_iversion(inode));
1725         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1726         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1727         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1728         btrfs_set_stack_inode_block_group(inode_item, 0);
1729
1730         btrfs_set_stack_timespec_sec(&inode_item->atime,
1731                                      inode->i_atime.tv_sec);
1732         btrfs_set_stack_timespec_nsec(&inode_item->atime,
1733                                       inode->i_atime.tv_nsec);
1734
1735         btrfs_set_stack_timespec_sec(&inode_item->mtime,
1736                                      inode->i_mtime.tv_sec);
1737         btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1738                                       inode->i_mtime.tv_nsec);
1739
1740         btrfs_set_stack_timespec_sec(&inode_item->ctime,
1741                                      inode->i_ctime.tv_sec);
1742         btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1743                                       inode->i_ctime.tv_nsec);
1744
1745         btrfs_set_stack_timespec_sec(&inode_item->otime,
1746                                      BTRFS_I(inode)->i_otime.tv_sec);
1747         btrfs_set_stack_timespec_nsec(&inode_item->otime,
1748                                      BTRFS_I(inode)->i_otime.tv_nsec);
1749 }
1750
1751 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1752 {
1753         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1754         struct btrfs_delayed_node *delayed_node;
1755         struct btrfs_inode_item *inode_item;
1756
1757         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1758         if (!delayed_node)
1759                 return -ENOENT;
1760
1761         mutex_lock(&delayed_node->mutex);
1762         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1763                 mutex_unlock(&delayed_node->mutex);
1764                 btrfs_release_delayed_node(delayed_node);
1765                 return -ENOENT;
1766         }
1767
1768         inode_item = &delayed_node->inode_item;
1769
1770         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1771         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1772         btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1773         btrfs_inode_set_file_extent_range(BTRFS_I(inode), 0,
1774                         round_up(i_size_read(inode), fs_info->sectorsize));
1775         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1776         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1777         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1778         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1779         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1780
1781         inode_set_iversion_queried(inode,
1782                                    btrfs_stack_inode_sequence(inode_item));
1783         inode->i_rdev = 0;
1784         *rdev = btrfs_stack_inode_rdev(inode_item);
1785         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1786
1787         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1788         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1789
1790         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1791         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1792
1793         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1794         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1795
1796         BTRFS_I(inode)->i_otime.tv_sec =
1797                 btrfs_stack_timespec_sec(&inode_item->otime);
1798         BTRFS_I(inode)->i_otime.tv_nsec =
1799                 btrfs_stack_timespec_nsec(&inode_item->otime);
1800
1801         inode->i_generation = BTRFS_I(inode)->generation;
1802         BTRFS_I(inode)->index_cnt = (u64)-1;
1803
1804         mutex_unlock(&delayed_node->mutex);
1805         btrfs_release_delayed_node(delayed_node);
1806         return 0;
1807 }
1808
1809 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1810                                struct btrfs_root *root,
1811                                struct btrfs_inode *inode)
1812 {
1813         struct btrfs_delayed_node *delayed_node;
1814         int ret = 0;
1815
1816         delayed_node = btrfs_get_or_create_delayed_node(inode);
1817         if (IS_ERR(delayed_node))
1818                 return PTR_ERR(delayed_node);
1819
1820         mutex_lock(&delayed_node->mutex);
1821         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1822                 fill_stack_inode_item(trans, &delayed_node->inode_item,
1823                                       &inode->vfs_inode);
1824                 goto release_node;
1825         }
1826
1827         ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1828         if (ret)
1829                 goto release_node;
1830
1831         fill_stack_inode_item(trans, &delayed_node->inode_item, &inode->vfs_inode);
1832         set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1833         delayed_node->count++;
1834         atomic_inc(&root->fs_info->delayed_root->items);
1835 release_node:
1836         mutex_unlock(&delayed_node->mutex);
1837         btrfs_release_delayed_node(delayed_node);
1838         return ret;
1839 }
1840
1841 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1842 {
1843         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1844         struct btrfs_delayed_node *delayed_node;
1845
1846         /*
1847          * we don't do delayed inode updates during log recovery because it
1848          * leads to enospc problems.  This means we also can't do
1849          * delayed inode refs
1850          */
1851         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1852                 return -EAGAIN;
1853
1854         delayed_node = btrfs_get_or_create_delayed_node(inode);
1855         if (IS_ERR(delayed_node))
1856                 return PTR_ERR(delayed_node);
1857
1858         /*
1859          * We don't reserve space for inode ref deletion is because:
1860          * - We ONLY do async inode ref deletion for the inode who has only
1861          *   one link(i_nlink == 1), it means there is only one inode ref.
1862          *   And in most case, the inode ref and the inode item are in the
1863          *   same leaf, and we will deal with them at the same time.
1864          *   Since we are sure we will reserve the space for the inode item,
1865          *   it is unnecessary to reserve space for inode ref deletion.
1866          * - If the inode ref and the inode item are not in the same leaf,
1867          *   We also needn't worry about enospc problem, because we reserve
1868          *   much more space for the inode update than it needs.
1869          * - At the worst, we can steal some space from the global reservation.
1870          *   It is very rare.
1871          */
1872         mutex_lock(&delayed_node->mutex);
1873         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1874                 goto release_node;
1875
1876         set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1877         delayed_node->count++;
1878         atomic_inc(&fs_info->delayed_root->items);
1879 release_node:
1880         mutex_unlock(&delayed_node->mutex);
1881         btrfs_release_delayed_node(delayed_node);
1882         return 0;
1883 }
1884
1885 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1886 {
1887         struct btrfs_root *root = delayed_node->root;
1888         struct btrfs_fs_info *fs_info = root->fs_info;
1889         struct btrfs_delayed_item *curr_item, *prev_item;
1890
1891         mutex_lock(&delayed_node->mutex);
1892         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1893         while (curr_item) {
1894                 btrfs_delayed_item_release_metadata(root, curr_item);
1895                 prev_item = curr_item;
1896                 curr_item = __btrfs_next_delayed_item(prev_item);
1897                 btrfs_release_delayed_item(prev_item);
1898         }
1899
1900         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1901         while (curr_item) {
1902                 btrfs_delayed_item_release_metadata(root, curr_item);
1903                 prev_item = curr_item;
1904                 curr_item = __btrfs_next_delayed_item(prev_item);
1905                 btrfs_release_delayed_item(prev_item);
1906         }
1907
1908         btrfs_release_delayed_iref(delayed_node);
1909
1910         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1911                 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1912                 btrfs_release_delayed_inode(delayed_node);
1913         }
1914         mutex_unlock(&delayed_node->mutex);
1915 }
1916
1917 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1918 {
1919         struct btrfs_delayed_node *delayed_node;
1920
1921         delayed_node = btrfs_get_delayed_node(inode);
1922         if (!delayed_node)
1923                 return;
1924
1925         __btrfs_kill_delayed_node(delayed_node);
1926         btrfs_release_delayed_node(delayed_node);
1927 }
1928
1929 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1930 {
1931         u64 inode_id = 0;
1932         struct btrfs_delayed_node *delayed_nodes[8];
1933         int i, n;
1934
1935         while (1) {
1936                 spin_lock(&root->inode_lock);
1937                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1938                                            (void **)delayed_nodes, inode_id,
1939                                            ARRAY_SIZE(delayed_nodes));
1940                 if (!n) {
1941                         spin_unlock(&root->inode_lock);
1942                         break;
1943                 }
1944
1945                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1946                 for (i = 0; i < n; i++) {
1947                         /*
1948                          * Don't increase refs in case the node is dead and
1949                          * about to be removed from the tree in the loop below
1950                          */
1951                         if (!refcount_inc_not_zero(&delayed_nodes[i]->refs))
1952                                 delayed_nodes[i] = NULL;
1953                 }
1954                 spin_unlock(&root->inode_lock);
1955
1956                 for (i = 0; i < n; i++) {
1957                         if (!delayed_nodes[i])
1958                                 continue;
1959                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1960                         btrfs_release_delayed_node(delayed_nodes[i]);
1961                 }
1962         }
1963 }
1964
1965 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1966 {
1967         struct btrfs_delayed_node *curr_node, *prev_node;
1968
1969         curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1970         while (curr_node) {
1971                 __btrfs_kill_delayed_node(curr_node);
1972
1973                 prev_node = curr_node;
1974                 curr_node = btrfs_next_delayed_node(curr_node);
1975                 btrfs_release_delayed_node(prev_node);
1976         }
1977 }
1978