fs/btrfs/delayed-inode.c

   1 /*
   2  * Copyright (C) 2011 Fujitsu.  All rights reserved.
   3  * Written by Miao Xie <miaox@cn.fujitsu.com>
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public
   7  * License v2 as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12  * General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public
  15  * License along with this program; if not, write to the
  16  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  17  * Boston, MA 021110-1307, USA.
  18  */
  19
  20 #include <linux/slab.h>
  21 #include <linux/iversion.h>
  22 #include "delayed-inode.h"
  23 #include "disk-io.h"
  24 #include "transaction.h"
  25 #include "ctree.h"
  26 #include "qgroup.h"
  27
  28 #define BTRFS_DELAYED_WRITEBACK         512
  29 #define BTRFS_DELAYED_BACKGROUND        128
  30 #define BTRFS_DELAYED_BATCH             16
  31
  32 static struct kmem_cache *delayed_node_cache;
  33
  34 int __init btrfs_delayed_inode_init(void)
  35 {
  36         delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
  37                                         sizeof(struct btrfs_delayed_node),
  38                                         0,
  39                                         SLAB_MEM_SPREAD,
  40                                         NULL);
  41         if (!delayed_node_cache)
  42                 return -ENOMEM;
  43         return 0;
  44 }
  45
  46 void __cold btrfs_delayed_inode_exit(void)
  47 {
  48         kmem_cache_destroy(delayed_node_cache);
  49 }
  50
  51 static inline void btrfs_init_delayed_node(
  52                                 struct btrfs_delayed_node *delayed_node,
  53                                 struct btrfs_root *root, u64 inode_id)
  54 {
  55         delayed_node->root = root;
  56         delayed_node->inode_id = inode_id;
  57         refcount_set(&delayed_node->refs, 0);
  58         delayed_node->ins_root = RB_ROOT;
  59         delayed_node->del_root = RB_ROOT;
  60         mutex_init(&delayed_node->mutex);
  61         INIT_LIST_HEAD(&delayed_node->n_list);
  62         INIT_LIST_HEAD(&delayed_node->p_list);
  63 }
  64
  65 static inline int btrfs_is_continuous_delayed_item(
  66                                         struct btrfs_delayed_item *item1,
  67                                         struct btrfs_delayed_item *item2)
  68 {
  69         if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  70             item1->key.objectid == item2->key.objectid &&
  71             item1->key.type == item2->key.type &&
  72             item1->key.offset + 1 == item2->key.offset)
  73                 return 1;
  74         return 0;
  75 }
  76
  77 static struct btrfs_delayed_node *btrfs_get_delayed_node(
  78                 struct btrfs_inode *btrfs_inode)
  79 {
  80         struct btrfs_root *root = btrfs_inode->root;
  81         u64 ino = btrfs_ino(btrfs_inode);
  82         struct btrfs_delayed_node *node;
  83
  84         node = READ_ONCE(btrfs_inode->delayed_node);
  85         if (node) {
  86                 refcount_inc(&node->refs);
  87                 return node;
  88         }
  89
  90         spin_lock(&root->inode_lock);
  91         node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
  92
  93         if (node) {
  94                 if (btrfs_inode->delayed_node) {
  95                         refcount_inc(&node->refs);      /* can be accessed */
  96                         BUG_ON(btrfs_inode->delayed_node != node);
  97                         spin_unlock(&root->inode_lock);
  98                         return node;
  99                 }
 100
 101                 /*
 102                  * It's possible that we're racing into the middle of removing
 103                  * this node from the radix tree.  In this case, the refcount
 104                  * was zero and it should never go back to one.  Just return
 105                  * NULL like it was never in the radix at all; our release
 106                  * function is in the process of removing it.
 107                  *
 108                  * Some implementations of refcount_inc refuse to bump the
 109                  * refcount once it has hit zero.  If we don't do this dance
 110                  * here, refcount_inc() may decide to just WARN_ONCE() instead
 111                  * of actually bumping the refcount.
 112                  *
 113                  * If this node is properly in the radix, we want to bump the
 114                  * refcount twice, once for the inode and once for this get
 115                  * operation.
 116                  */
 117                 if (refcount_inc_not_zero(&node->refs)) {
 118                         refcount_inc(&node->refs);
 119                         btrfs_inode->delayed_node = node;
 120                 } else {
 121                         node = NULL;
 122                 }
 123
 124                 spin_unlock(&root->inode_lock);
 125                 return node;
 126         }
 127         spin_unlock(&root->inode_lock);
 128
 129         return NULL;
 130 }
 131
 132 /* Will return either the node or PTR_ERR(-ENOMEM) */
 133 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 134                 struct btrfs_inode *btrfs_inode)
 135 {
 136         struct btrfs_delayed_node *node;
 137         struct btrfs_root *root = btrfs_inode->root;
 138         u64 ino = btrfs_ino(btrfs_inode);
 139         int ret;
 140
 141 again:
 142         node = btrfs_get_delayed_node(btrfs_inode);
 143         if (node)
 144                 return node;
 145
 146         node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
 147         if (!node)
 148                 return ERR_PTR(-ENOMEM);
 149         btrfs_init_delayed_node(node, root, ino);
 150
 151         /* cached in the btrfs inode and can be accessed */
 152         refcount_set(&node->refs, 2);
 153
 154         ret = radix_tree_preload(GFP_NOFS);
 155         if (ret) {
 156                 kmem_cache_free(delayed_node_cache, node);
 157                 return ERR_PTR(ret);
 158         }
 159
 160         spin_lock(&root->inode_lock);
 161         ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 162         if (ret == -EEXIST) {
 163                 spin_unlock(&root->inode_lock);
 164                 kmem_cache_free(delayed_node_cache, node);
 165                 radix_tree_preload_end();
 166                 goto again;
 167         }
 168         btrfs_inode->delayed_node = node;
 169         spin_unlock(&root->inode_lock);
 170         radix_tree_preload_end();
 171
 172         return node;
 173 }
 174
 175 /*
 176  * Call it when holding delayed_node->mutex
 177  *
 178  * If mod = 1, add this node into the prepared list.
 179  */
 180 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 181                                      struct btrfs_delayed_node *node,
 182                                      int mod)
 183 {
 184         spin_lock(&root->lock);
 185         if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 186                 if (!list_empty(&node->p_list))
 187                         list_move_tail(&node->p_list, &root->prepare_list);
 188                 else if (mod)
 189                         list_add_tail(&node->p_list, &root->prepare_list);
 190         } else {
 191                 list_add_tail(&node->n_list, &root->node_list);
 192                 list_add_tail(&node->p_list, &root->prepare_list);
 193                 refcount_inc(&node->refs);      /* inserted into list */
 194                 root->nodes++;
 195                 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 196         }
 197         spin_unlock(&root->lock);
 198 }
 199
 200 /* Call it when holding delayed_node->mutex */
 201 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 202                                        struct btrfs_delayed_node *node)
 203 {
 204         spin_lock(&root->lock);
 205         if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 206                 root->nodes--;
 207                 refcount_dec(&node->refs);      /* not in the list */
 208                 list_del_init(&node->n_list);
 209                 if (!list_empty(&node->p_list))
 210                         list_del_init(&node->p_list);
 211                 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 212         }
 213         spin_unlock(&root->lock);
 214 }
 215
 216 static struct btrfs_delayed_node *btrfs_first_delayed_node(
 217                         struct btrfs_delayed_root *delayed_root)
 218 {
 219         struct list_head *p;
 220         struct btrfs_delayed_node *node = NULL;
 221
 222         spin_lock(&delayed_root->lock);
 223         if (list_empty(&delayed_root->node_list))
 224                 goto out;
 225
 226         p = delayed_root->node_list.next;
 227         node = list_entry(p, struct btrfs_delayed_node, n_list);
 228         refcount_inc(&node->refs);
 229 out:
 230         spin_unlock(&delayed_root->lock);
 231
 232         return node;
 233 }
 234
 235 static struct btrfs_delayed_node *btrfs_next_delayed_node(
 236                                                 struct btrfs_delayed_node *node)
 237 {
 238         struct btrfs_delayed_root *delayed_root;
 239         struct list_head *p;
 240         struct btrfs_delayed_node *next = NULL;
 241
 242         delayed_root = node->root->fs_info->delayed_root;
 243         spin_lock(&delayed_root->lock);
 244         if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 245                 /* not in the list */
 246                 if (list_empty(&delayed_root->node_list))
 247                         goto out;
 248                 p = delayed_root->node_list.next;
 249         } else if (list_is_last(&node->n_list, &delayed_root->node_list))
 250                 goto out;
 251         else
 252                 p = node->n_list.next;
 253
 254         next = list_entry(p, struct btrfs_delayed_node, n_list);
 255         refcount_inc(&next->refs);
 256 out:
 257         spin_unlock(&delayed_root->lock);
 258
 259         return next;
 260 }
 261
 262 static void __btrfs_release_delayed_node(
 263                                 struct btrfs_delayed_node *delayed_node,
 264                                 int mod)
 265 {
 266         struct btrfs_delayed_root *delayed_root;
 267
 268         if (!delayed_node)
 269                 return;
 270
 271         delayed_root = delayed_node->root->fs_info->delayed_root;
 272
 273         mutex_lock(&delayed_node->mutex);
 274         if (delayed_node->count)
 275                 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
 276         else
 277                 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 278         mutex_unlock(&delayed_node->mutex);
 279
 280         if (refcount_dec_and_test(&delayed_node->refs)) {
 281                 struct btrfs_root *root = delayed_node->root;
 282
 283                 spin_lock(&root->inode_lock);
 284                 /*
 285                  * Once our refcount goes to zero, nobody is allowed to bump it
 286                  * back up.  We can delete it now.
 287                  */
 288                 ASSERT(refcount_read(&delayed_node->refs) == 0);
 289                 radix_tree_delete(&root->delayed_nodes_tree,
 290                                   delayed_node->inode_id);
 291                 spin_unlock(&root->inode_lock);
 292                 kmem_cache_free(delayed_node_cache, delayed_node);
 293         }
 294 }
 295
 296 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
 297 {
 298         __btrfs_release_delayed_node(node, 0);
 299 }
 300
 301 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
 302                                         struct btrfs_delayed_root *delayed_root)
 303 {
 304         struct list_head *p;
 305         struct btrfs_delayed_node *node = NULL;
 306
 307         spin_lock(&delayed_root->lock);
 308         if (list_empty(&delayed_root->prepare_list))
 309                 goto out;
 310
 311         p = delayed_root->prepare_list.next;
 312         list_del_init(p);
 313         node = list_entry(p, struct btrfs_delayed_node, p_list);
 314         refcount_inc(&node->refs);
 315 out:
 316         spin_unlock(&delayed_root->lock);
 317
 318         return node;
 319 }
 320
 321 static inline void btrfs_release_prepared_delayed_node(
 322                                         struct btrfs_delayed_node *node)
 323 {
 324         __btrfs_release_delayed_node(node, 1);
 325 }
 326
 327 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
 328 {
 329         struct btrfs_delayed_item *item;
 330         item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
 331         if (item) {
 332                 item->data_len = data_len;
 333                 item->ins_or_del = 0;
 334                 item->bytes_reserved = 0;
 335                 item->delayed_node = NULL;
 336                 refcount_set(&item->refs, 1);
 337         }
 338         return item;
 339 }
 340
 341 /*
 342  * __btrfs_lookup_delayed_item - look up the delayed item by key
 343  * @delayed_node: pointer to the delayed node
 344  * @key:          the key to look up
 345  * @prev:         used to store the prev item if the right item isn't found
 346  * @next:         used to store the next item if the right item isn't found
 347  *
 348  * Note: if we don't find the right item, we will return the prev item and
 349  * the next item.
 350  */
 351 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 352                                 struct rb_root *root,
 353                                 struct btrfs_key *key,
 354                                 struct btrfs_delayed_item **prev,
 355                                 struct btrfs_delayed_item **next)
 356 {
 357         struct rb_node *node, *prev_node = NULL;
 358         struct btrfs_delayed_item *delayed_item = NULL;
 359         int ret = 0;
 360
 361         node = root->rb_node;
 362
 363         while (node) {
 364                 delayed_item = rb_entry(node, struct btrfs_delayed_item,
 365                                         rb_node);
 366                 prev_node = node;
 367                 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
 368                 if (ret < 0)
 369                         node = node->rb_right;
 370                 else if (ret > 0)
 371                         node = node->rb_left;
 372                 else
 373                         return delayed_item;
 374         }
 375
 376         if (prev) {
 377                 if (!prev_node)
 378                         *prev = NULL;
 379                 else if (ret < 0)
 380                         *prev = delayed_item;
 381                 else if ((node = rb_prev(prev_node)) != NULL) {
 382                         *prev = rb_entry(node, struct btrfs_delayed_item,
 383                                          rb_node);
 384                 } else
 385                         *prev = NULL;
 386         }
 387
 388         if (next) {
 389                 if (!prev_node)
 390                         *next = NULL;
 391                 else if (ret > 0)
 392                         *next = delayed_item;
 393                 else if ((node = rb_next(prev_node)) != NULL) {
 394                         *next = rb_entry(node, struct btrfs_delayed_item,
 395                                          rb_node);
 396                 } else
 397                         *next = NULL;
 398         }
 399         return NULL;
 400 }
 401
 402 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
 403                                         struct btrfs_delayed_node *delayed_node,
 404                                         struct btrfs_key *key)
 405 {
 406         return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 407                                            NULL, NULL);
 408 }
 409
 410 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
 411                                     struct btrfs_delayed_item *ins,
 412                                     int action)
 413 {
 414         struct rb_node **p, *node;
 415         struct rb_node *parent_node = NULL;
 416         struct rb_root *root;
 417         struct btrfs_delayed_item *item;
 418         int cmp;
 419
 420         if (action == BTRFS_DELAYED_INSERTION_ITEM)
 421                 root = &delayed_node->ins_root;
 422         else if (action == BTRFS_DELAYED_DELETION_ITEM)
 423                 root = &delayed_node->del_root;
 424         else
 425                 BUG();
 426         p = &root->rb_node;
 427         node = &ins->rb_node;
 428
 429         while (*p) {
 430                 parent_node = *p;
 431                 item = rb_entry(parent_node, struct btrfs_delayed_item,
 432                                  rb_node);
 433
 434                 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
 435                 if (cmp < 0)
 436                         p = &(*p)->rb_right;
 437                 else if (cmp > 0)
 438                         p = &(*p)->rb_left;
 439                 else
 440                         return -EEXIST;
 441         }
 442
 443         rb_link_node(node, parent_node, p);
 444         rb_insert_color(node, root);
 445         ins->delayed_node = delayed_node;
 446         ins->ins_or_del = action;
 447
 448         if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
 449             action == BTRFS_DELAYED_INSERTION_ITEM &&
 450             ins->key.offset >= delayed_node->index_cnt)
 451                         delayed_node->index_cnt = ins->key.offset + 1;
 452
 453         delayed_node->count++;
 454         atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
 455         return 0;
 456 }
 457
 458 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
 459                                               struct btrfs_delayed_item *item)
 460 {
 461         return __btrfs_add_delayed_item(node, item,
 462                                         BTRFS_DELAYED_INSERTION_ITEM);
 463 }
 464
 465 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
 466                                              struct btrfs_delayed_item *item)
 467 {
 468         return __btrfs_add_delayed_item(node, item,
 469                                         BTRFS_DELAYED_DELETION_ITEM);
 470 }
 471
 472 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
 473 {
 474         int seq = atomic_inc_return(&delayed_root->items_seq);
 475
 476         /*
 477          * atomic_dec_return implies a barrier for waitqueue_active
 478          */
 479         if ((atomic_dec_return(&delayed_root->items) <
 480             BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
 481             waitqueue_active(&delayed_root->wait))
 482                 wake_up(&delayed_root->wait);
 483 }
 484
 485 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 486 {
 487         struct rb_root *root;
 488         struct btrfs_delayed_root *delayed_root;
 489
 490         delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
 491
 492         BUG_ON(!delayed_root);
 493         BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
 494                delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
 495
 496         if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
 497                 root = &delayed_item->delayed_node->ins_root;
 498         else
 499                 root = &delayed_item->delayed_node->del_root;
 500
 501         rb_erase(&delayed_item->rb_node, root);
 502         delayed_item->delayed_node->count--;
 503
 504         finish_one_item(delayed_root);
 505 }
 506
 507 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 508 {
 509         if (item) {
 510                 __btrfs_remove_delayed_item(item);
 511                 if (refcount_dec_and_test(&item->refs))
 512                         kfree(item);
 513         }
 514 }
 515
 516 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 517                                         struct btrfs_delayed_node *delayed_node)
 518 {
 519         struct rb_node *p;
 520         struct btrfs_delayed_item *item = NULL;
 521
 522         p = rb_first(&delayed_node->ins_root);
 523         if (p)
 524                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 525
 526         return item;
 527 }
 528
 529 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 530                                         struct btrfs_delayed_node *delayed_node)
 531 {
 532         struct rb_node *p;
 533         struct btrfs_delayed_item *item = NULL;
 534
 535         p = rb_first(&delayed_node->del_root);
 536         if (p)
 537                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 538
 539         return item;
 540 }
 541
 542 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
 543                                                 struct btrfs_delayed_item *item)
 544 {
 545         struct rb_node *p;
 546         struct btrfs_delayed_item *next = NULL;
 547
 548         p = rb_next(&item->rb_node);
 549         if (p)
 550                 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
 551
 552         return next;
 553 }
 554
 555 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 556                                                struct btrfs_root *root,
 557                                                struct btrfs_delayed_item *item)
 558 {
 559         struct btrfs_block_rsv *src_rsv;
 560         struct btrfs_block_rsv *dst_rsv;
 561         struct btrfs_fs_info *fs_info = root->fs_info;
 562         u64 num_bytes;
 563         int ret;
 564
 565         if (!trans->bytes_reserved)
 566                 return 0;
 567
 568         src_rsv = trans->block_rsv;
 569         dst_rsv = &fs_info->delayed_block_rsv;
 570
 571         num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
 572         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
 573         if (!ret) {
 574                 trace_btrfs_space_reservation(fs_info, "delayed_item",
 575                                               item->key.objectid,
 576                                               num_bytes, 1);
 577                 item->bytes_reserved = num_bytes;
 578         }
 579
 580         return ret;
 581 }
 582
 583 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 584                                                 struct btrfs_delayed_item *item)
 585 {
 586         struct btrfs_block_rsv *rsv;
 587         struct btrfs_fs_info *fs_info = root->fs_info;
 588
 589         if (!item->bytes_reserved)
 590                 return;
 591
 592         rsv = &fs_info->delayed_block_rsv;
 593         btrfs_qgroup_convert_reserved_meta(root, item->bytes_reserved);
 594         trace_btrfs_space_reservation(fs_info, "delayed_item",
 595                                       item->key.objectid, item->bytes_reserved,
 596                                       0);
 597         btrfs_block_rsv_release(fs_info, rsv,
 598                                 item->bytes_reserved);
 599 }
 600
 601 static int btrfs_delayed_inode_reserve_metadata(
 602                                         struct btrfs_trans_handle *trans,
 603                                         struct btrfs_root *root,
 604                                         struct btrfs_inode *inode,
 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_trans_metadata_size(fs_info, 1);
 617
 618         ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true);
 619         if (ret < 0)
 620                 return ret;
 621         /*
 622          * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 623          * which doesn't reserve space for speed.  This is a problem since we
 624          * still need to reserve space for this update, so try to reserve the
 625          * space.
 626          *
 627          * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 628          * we always reserve enough to update the inode item.
 629          */
 630         if (!src_rsv || (!trans->bytes_reserved &&
 631                          src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
 632                 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 633                                           BTRFS_RESERVE_NO_FLUSH);
 634                 /*
 635                  * Since we're under a transaction reserve_metadata_bytes could
 636                  * try to commit the transaction which will make it return
 637                  * EAGAIN to make us stop the transaction we have, so return
 638                  * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 639                  */
 640                 if (ret == -EAGAIN) {
 641                         ret = -ENOSPC;
 642                         btrfs_qgroup_free_meta_prealloc(root, num_bytes);
 643                 }
 644                 if (!ret) {
 645                         node->bytes_reserved = num_bytes;
 646                         trace_btrfs_space_reservation(fs_info,
 647                                                       "delayed_inode",
 648                                                       btrfs_ino(inode),
 649                                                       num_bytes, 1);
 650                 }
 651                 return ret;
 652         }
 653
 654         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
 655         if (!ret) {
 656                 trace_btrfs_space_reservation(fs_info, "delayed_inode",
 657                                               btrfs_ino(inode), num_bytes, 1);
 658                 node->bytes_reserved = num_bytes;
 659         }
 660
 661         return ret;
 662 }
 663
 664 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
 665                                                 struct btrfs_delayed_node *node,
 666                                                 bool qgroup_free)
 667 {
 668         struct btrfs_block_rsv *rsv;
 669
 670         if (!node->bytes_reserved)
 671                 return;
 672
 673         rsv = &fs_info->delayed_block_rsv;
 674         trace_btrfs_space_reservation(fs_info, "delayed_inode",
 675                                       node->inode_id, node->bytes_reserved, 0);
 676         btrfs_block_rsv_release(fs_info, rsv,
 677                                 node->bytes_reserved);
 678         if (qgroup_free)
 679                 btrfs_qgroup_free_meta_prealloc(node->root,
 680                                 node->bytes_reserved);
 681         else
 682                 btrfs_qgroup_convert_reserved_meta(node->root,
 683                                 node->bytes_reserved);
 684         node->bytes_reserved = 0;
 685 }
 686
 687 /*
 688  * This helper will insert some continuous items into the same leaf according
 689  * to the free space of the leaf.
 690  */
 691 static int btrfs_batch_insert_items(struct btrfs_root *root,
 692                                     struct btrfs_path *path,
 693                                     struct btrfs_delayed_item *item)
 694 {
 695         struct btrfs_fs_info *fs_info = root->fs_info;
 696         struct btrfs_delayed_item *curr, *next;
 697         int free_space;
 698         int total_data_size = 0, total_size = 0;
 699         struct extent_buffer *leaf;
 700         char *data_ptr;
 701         struct btrfs_key *keys;
 702         u32 *data_size;
 703         struct list_head head;
 704         int slot;
 705         int nitems;
 706         int i;
 707         int ret = 0;
 708
 709         BUG_ON(!path->nodes[0]);
 710
 711         leaf = path->nodes[0];
 712         free_space = btrfs_leaf_free_space(fs_info, leaf);
 713         INIT_LIST_HEAD(&head);
 714
 715         next = item;
 716         nitems = 0;
 717
 718         /*
 719          * count the number of the continuous items that we can insert in batch
 720          */
 721         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 722                free_space) {
 723                 total_data_size += next->data_len;
 724                 total_size += next->data_len + sizeof(struct btrfs_item);
 725                 list_add_tail(&next->tree_list, &head);
 726                 nitems++;
 727
 728                 curr = next;
 729                 next = __btrfs_next_delayed_item(curr);
 730                 if (!next)
 731                         break;
 732
 733                 if (!btrfs_is_continuous_delayed_item(curr, next))
 734                         break;
 735         }
 736
 737         if (!nitems) {
 738                 ret = 0;
 739                 goto out;
 740         }
 741
 742         /*
 743          * we need allocate some memory space, but it might cause the task
 744          * to sleep, so we set all locked nodes in the path to blocking locks
 745          * first.
 746          */
 747         btrfs_set_path_blocking(path);
 748
 749         keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
 750         if (!keys) {
 751                 ret = -ENOMEM;
 752                 goto out;
 753         }
 754
 755         data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
 756         if (!data_size) {
 757                 ret = -ENOMEM;
 758                 goto error;
 759         }
 760
 761         /* get keys of all the delayed items */
 762         i = 0;
 763         list_for_each_entry(next, &head, tree_list) {
 764                 keys[i] = next->key;
 765                 data_size[i] = next->data_len;
 766                 i++;
 767         }
 768
 769         /* reset all the locked nodes in the patch to spinning locks. */
 770         btrfs_clear_path_blocking(path, NULL, 0);
 771
 772         /* insert the keys of the items */
 773         setup_items_for_insert(root, path, keys, data_size,
 774                                total_data_size, total_size, nitems);
 775
 776         /* insert the dir index items */
 777         slot = path->slots[0];
 778         list_for_each_entry_safe(curr, next, &head, tree_list) {
 779                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 780                 write_extent_buffer(leaf, &curr->data,
 781                                     (unsigned long)data_ptr,
 782                                     curr->data_len);
 783                 slot++;
 784
 785                 btrfs_delayed_item_release_metadata(root, curr);
 786
 787                 list_del(&curr->tree_list);
 788                 btrfs_release_delayed_item(curr);
 789         }
 790
 791 error:
 792         kfree(data_size);
 793         kfree(keys);
 794 out:
 795         return ret;
 796 }
 797
 798 /*
 799  * This helper can just do simple insertion that needn't extend item for new
 800  * data, such as directory name index insertion, inode insertion.
 801  */
 802 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 803                                      struct btrfs_root *root,
 804                                      struct btrfs_path *path,
 805                                      struct btrfs_delayed_item *delayed_item)
 806 {
 807         struct extent_buffer *leaf;
 808         char *ptr;
 809         int ret;
 810
 811         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 812                                       delayed_item->data_len);
 813         if (ret < 0 && ret != -EEXIST)
 814                 return ret;
 815
 816         leaf = path->nodes[0];
 817
 818         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 819
 820         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 821                             delayed_item->data_len);
 822         btrfs_mark_buffer_dirty(leaf);
 823
 824         btrfs_delayed_item_release_metadata(root, delayed_item);
 825         return 0;
 826 }
 827
 828 /*
 829  * we insert an item first, then if there are some continuous items, we try
 830  * to insert those items into the same leaf.
 831  */
 832 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 833                                       struct btrfs_path *path,
 834                                       struct btrfs_root *root,
 835                                       struct btrfs_delayed_node *node)
 836 {
 837         struct btrfs_delayed_item *curr, *prev;
 838         int ret = 0;
 839
 840 do_again:
 841         mutex_lock(&node->mutex);
 842         curr = __btrfs_first_delayed_insertion_item(node);
 843         if (!curr)
 844                 goto insert_end;
 845
 846         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 847         if (ret < 0) {
 848                 btrfs_release_path(path);
 849                 goto insert_end;
 850         }
 851
 852         prev = curr;
 853         curr = __btrfs_next_delayed_item(prev);
 854         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 855                 /* insert the continuous items into the same leaf */
 856                 path->slots[0]++;
 857                 btrfs_batch_insert_items(root, path, curr);
 858         }
 859         btrfs_release_delayed_item(prev);
 860         btrfs_mark_buffer_dirty(path->nodes[0]);
 861
 862         btrfs_release_path(path);
 863         mutex_unlock(&node->mutex);
 864         goto do_again;
 865
 866 insert_end:
 867         mutex_unlock(&node->mutex);
 868         return ret;
 869 }
 870
 871 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 872                                     struct btrfs_root *root,
 873                                     struct btrfs_path *path,
 874                                     struct btrfs_delayed_item *item)
 875 {
 876         struct btrfs_delayed_item *curr, *next;
 877         struct extent_buffer *leaf;
 878         struct btrfs_key key;
 879         struct list_head head;
 880         int nitems, i, last_item;
 881         int ret = 0;
 882
 883         BUG_ON(!path->nodes[0]);
 884
 885         leaf = path->nodes[0];
 886
 887         i = path->slots[0];
 888         last_item = btrfs_header_nritems(leaf) - 1;
 889         if (i > last_item)
 890                 return -ENOENT; /* FIXME: Is errno suitable? */
 891
 892         next = item;
 893         INIT_LIST_HEAD(&head);
 894         btrfs_item_key_to_cpu(leaf, &key, i);
 895         nitems = 0;
 896         /*
 897          * count the number of the dir index items that we can delete in batch
 898          */
 899         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 900                 list_add_tail(&next->tree_list, &head);
 901                 nitems++;
 902
 903                 curr = next;
 904                 next = __btrfs_next_delayed_item(curr);
 905                 if (!next)
 906                         break;
 907
 908                 if (!btrfs_is_continuous_delayed_item(curr, next))
 909                         break;
 910
 911                 i++;
 912                 if (i > last_item)
 913                         break;
 914                 btrfs_item_key_to_cpu(leaf, &key, i);
 915         }
 916
 917         if (!nitems)
 918                 return 0;
 919
 920         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 921         if (ret)
 922                 goto out;
 923
 924         list_for_each_entry_safe(curr, next, &head, tree_list) {
 925                 btrfs_delayed_item_release_metadata(root, curr);
 926                 list_del(&curr->tree_list);
 927                 btrfs_release_delayed_item(curr);
 928         }
 929
 930 out:
 931         return ret;
 932 }
 933
 934 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 935                                       struct btrfs_path *path,
 936                                       struct btrfs_root *root,
 937                                       struct btrfs_delayed_node *node)
 938 {
 939         struct btrfs_delayed_item *curr, *prev;
 940         int ret = 0;
 941
 942 do_again:
 943         mutex_lock(&node->mutex);
 944         curr = __btrfs_first_delayed_deletion_item(node);
 945         if (!curr)
 946                 goto delete_fail;
 947
 948         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 949         if (ret < 0)
 950                 goto delete_fail;
 951         else if (ret > 0) {
 952                 /*
 953                  * can't find the item which the node points to, so this node
 954                  * is invalid, just drop it.
 955                  */
 956                 prev = curr;
 957                 curr = __btrfs_next_delayed_item(prev);
 958                 btrfs_release_delayed_item(prev);
 959                 ret = 0;
 960                 btrfs_release_path(path);
 961                 if (curr) {
 962                         mutex_unlock(&node->mutex);
 963                         goto do_again;
 964                 } else
 965                         goto delete_fail;
 966         }
 967
 968         btrfs_batch_delete_items(trans, root, path, curr);
 969         btrfs_release_path(path);
 970         mutex_unlock(&node->mutex);
 971         goto do_again;
 972
 973 delete_fail:
 974         btrfs_release_path(path);
 975         mutex_unlock(&node->mutex);
 976         return ret;
 977 }
 978
 979 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 980 {
 981         struct btrfs_delayed_root *delayed_root;
 982
 983         if (delayed_node &&
 984             test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
 985                 BUG_ON(!delayed_node->root);
 986                 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
 987                 delayed_node->count--;
 988
 989                 delayed_root = delayed_node->root->fs_info->delayed_root;
 990                 finish_one_item(delayed_root);
 991         }
 992 }
 993
 994 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
 995 {
 996         struct btrfs_delayed_root *delayed_root;
 997
 998         ASSERT(delayed_node->root);
 999         clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1000         delayed_node->count--;
1001
1002         delayed_root = delayed_node->root->fs_info->delayed_root;
1003         finish_one_item(delayed_root);
1004 }
1005
1006 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1007                                         struct btrfs_root *root,
1008                                         struct btrfs_path *path,
1009                                         struct btrfs_delayed_node *node)
1010 {
1011         struct btrfs_fs_info *fs_info = root->fs_info;
1012         struct btrfs_key key;
1013         struct btrfs_inode_item *inode_item;
1014         struct extent_buffer *leaf;
1015         int mod;
1016         int ret;
1017
1018         key.objectid = node->inode_id;
1019         key.type = BTRFS_INODE_ITEM_KEY;
1020         key.offset = 0;
1021
1022         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1023                 mod = -1;
1024         else
1025                 mod = 1;
1026
1027         ret = btrfs_lookup_inode(trans, root, path, &key, mod);
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         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1079         if (ret < 0)
1080                 goto err_out;
1081         ASSERT(ret);
1082
1083         ret = 0;
1084         leaf = path->nodes[0];
1085         path->slots[0]--;
1086         goto again;
1087 }
1088
1089 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1090                                              struct btrfs_root *root,
1091                                              struct btrfs_path *path,
1092                                              struct btrfs_delayed_node *node)
1093 {
1094         int ret;
1095
1096         mutex_lock(&node->mutex);
1097         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1098                 mutex_unlock(&node->mutex);
1099                 return 0;
1100         }
1101
1102         ret = __btrfs_update_delayed_inode(trans, root, path, node);
1103         mutex_unlock(&node->mutex);
1104         return ret;
1105 }
1106
1107 static inline int
1108 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1109                                    struct btrfs_path *path,
1110                                    struct btrfs_delayed_node *node)
1111 {
1112         int ret;
1113
1114         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1115         if (ret)
1116                 return ret;
1117
1118         ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1119         if (ret)
1120                 return ret;
1121
1122         ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1123         return ret;
1124 }
1125
1126 /*
1127  * Called when committing the transaction.
1128  * Returns 0 on success.
1129  * Returns < 0 on error and returns with an aborted transaction with any
1130  * outstanding delayed items cleaned up.
1131  */
1132 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1133 {
1134         struct btrfs_fs_info *fs_info = trans->fs_info;
1135         struct btrfs_delayed_root *delayed_root;
1136         struct btrfs_delayed_node *curr_node, *prev_node;
1137         struct btrfs_path *path;
1138         struct btrfs_block_rsv *block_rsv;
1139         int ret = 0;
1140         bool count = (nr > 0);
1141
1142         if (trans->aborted)
1143                 return -EIO;
1144
1145         path = btrfs_alloc_path();
1146         if (!path)
1147                 return -ENOMEM;
1148         path->leave_spinning = 1;
1149
1150         block_rsv = trans->block_rsv;
1151         trans->block_rsv = &fs_info->delayed_block_rsv;
1152
1153         delayed_root = fs_info->delayed_root;
1154
1155         curr_node = btrfs_first_delayed_node(delayed_root);
1156         while (curr_node && (!count || (count && nr--))) {
1157                 ret = __btrfs_commit_inode_delayed_items(trans, path,
1158                                                          curr_node);
1159                 if (ret) {
1160                         btrfs_release_delayed_node(curr_node);
1161                         curr_node = NULL;
1162                         btrfs_abort_transaction(trans, ret);
1163                         break;
1164                 }
1165
1166                 prev_node = curr_node;
1167                 curr_node = btrfs_next_delayed_node(curr_node);
1168                 btrfs_release_delayed_node(prev_node);
1169         }
1170
1171         if (curr_node)
1172                 btrfs_release_delayed_node(curr_node);
1173         btrfs_free_path(path);
1174         trans->block_rsv = block_rsv;
1175
1176         return ret;
1177 }
1178
1179 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1180 {
1181         return __btrfs_run_delayed_items(trans, -1);
1182 }
1183
1184 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1185 {
1186         return __btrfs_run_delayed_items(trans, nr);
1187 }
1188
1189 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1190                                      struct btrfs_inode *inode)
1191 {
1192         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1193         struct btrfs_path *path;
1194         struct btrfs_block_rsv *block_rsv;
1195         int ret;
1196
1197         if (!delayed_node)
1198                 return 0;
1199
1200         mutex_lock(&delayed_node->mutex);
1201         if (!delayed_node->count) {
1202                 mutex_unlock(&delayed_node->mutex);
1203                 btrfs_release_delayed_node(delayed_node);
1204                 return 0;
1205         }
1206         mutex_unlock(&delayed_node->mutex);
1207
1208         path = btrfs_alloc_path();
1209         if (!path) {
1210                 btrfs_release_delayed_node(delayed_node);
1211                 return -ENOMEM;
1212         }
1213         path->leave_spinning = 1;
1214
1215         block_rsv = trans->block_rsv;
1216         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1217
1218         ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1219
1220         btrfs_release_delayed_node(delayed_node);
1221         btrfs_free_path(path);
1222         trans->block_rsv = block_rsv;
1223
1224         return ret;
1225 }
1226
1227 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1228 {
1229         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1230         struct btrfs_trans_handle *trans;
1231         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1232         struct btrfs_path *path;
1233         struct btrfs_block_rsv *block_rsv;
1234         int ret;
1235
1236         if (!delayed_node)
1237                 return 0;
1238
1239         mutex_lock(&delayed_node->mutex);
1240         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1241                 mutex_unlock(&delayed_node->mutex);
1242                 btrfs_release_delayed_node(delayed_node);
1243                 return 0;
1244         }
1245         mutex_unlock(&delayed_node->mutex);
1246
1247         trans = btrfs_join_transaction(delayed_node->root);
1248         if (IS_ERR(trans)) {
1249                 ret = PTR_ERR(trans);
1250                 goto out;
1251         }
1252
1253         path = btrfs_alloc_path();
1254         if (!path) {
1255                 ret = -ENOMEM;
1256                 goto trans_out;
1257         }
1258         path->leave_spinning = 1;
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                 path->leave_spinning = 1;
1328                 root = delayed_node->root;
1329
1330                 trans = btrfs_join_transaction(root);
1331                 if (IS_ERR(trans)) {
1332                         btrfs_release_path(path);
1333                         btrfs_release_prepared_delayed_node(delayed_node);
1334                         total_done++;
1335                         continue;
1336                 }
1337
1338                 block_rsv = trans->block_rsv;
1339                 trans->block_rsv = &root->fs_info->delayed_block_rsv;
1340
1341                 __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1342
1343                 trans->block_rsv = block_rsv;
1344                 btrfs_end_transaction(trans);
1345                 btrfs_btree_balance_dirty_nodelay(root->fs_info);
1346
1347                 btrfs_release_path(path);
1348                 btrfs_release_prepared_delayed_node(delayed_node);
1349                 total_done++;
1350
1351         } while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1352                  || total_done < async_work->nr);
1353
1354         btrfs_free_path(path);
1355 out:
1356         wake_up(&delayed_root->wait);
1357         kfree(async_work);
1358 }
1359
1360
1361 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1362                                      struct btrfs_fs_info *fs_info, int nr)
1363 {
1364         struct btrfs_async_delayed_work *async_work;
1365
1366         async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1367         if (!async_work)
1368                 return -ENOMEM;
1369
1370         async_work->delayed_root = delayed_root;
1371         btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1372                         btrfs_async_run_delayed_root, NULL, NULL);
1373         async_work->nr = nr;
1374
1375         btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1376         return 0;
1377 }
1378
1379 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1380 {
1381         WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1382 }
1383
1384 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1385 {
1386         int val = atomic_read(&delayed_root->items_seq);
1387
1388         if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1389                 return 1;
1390
1391         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1392                 return 1;
1393
1394         return 0;
1395 }
1396
1397 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1398 {
1399         struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1400
1401         if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1402                 btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1403                 return;
1404
1405         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1406                 int seq;
1407                 int ret;
1408
1409                 seq = atomic_read(&delayed_root->items_seq);
1410
1411                 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1412                 if (ret)
1413                         return;
1414
1415                 wait_event_interruptible(delayed_root->wait,
1416                                          could_end_wait(delayed_root, seq));
1417                 return;
1418         }
1419
1420         btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1421 }
1422
1423 /* Will return 0 or -ENOMEM */
1424 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1425                                    struct btrfs_fs_info *fs_info,
1426                                    const char *name, int name_len,
1427                                    struct btrfs_inode *dir,
1428                                    struct btrfs_disk_key *disk_key, u8 type,
1429                                    u64 index)
1430 {
1431         struct btrfs_delayed_node *delayed_node;
1432         struct btrfs_delayed_item *delayed_item;
1433         struct btrfs_dir_item *dir_item;
1434         int ret;
1435
1436         delayed_node = btrfs_get_or_create_delayed_node(dir);
1437         if (IS_ERR(delayed_node))
1438                 return PTR_ERR(delayed_node);
1439
1440         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1441         if (!delayed_item) {
1442                 ret = -ENOMEM;
1443                 goto release_node;
1444         }
1445
1446         delayed_item->key.objectid = btrfs_ino(dir);
1447         delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1448         delayed_item->key.offset = index;
1449
1450         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1451         dir_item->location = *disk_key;
1452         btrfs_set_stack_dir_transid(dir_item, trans->transid);
1453         btrfs_set_stack_dir_data_len(dir_item, 0);
1454         btrfs_set_stack_dir_name_len(dir_item, name_len);
1455         btrfs_set_stack_dir_type(dir_item, type);
1456         memcpy((char *)(dir_item + 1), name, name_len);
1457
1458         ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1459         /*
1460          * we have reserved enough space when we start a new transaction,
1461          * so reserving metadata failure is impossible
1462          */
1463         BUG_ON(ret);
1464
1465
1466         mutex_lock(&delayed_node->mutex);
1467         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1468         if (unlikely(ret)) {
1469                 btrfs_err(fs_info,
1470                           "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1471                           name_len, name, delayed_node->root->objectid,
1472                           delayed_node->inode_id, ret);
1473                 BUG();
1474         }
1475         mutex_unlock(&delayed_node->mutex);
1476
1477 release_node:
1478         btrfs_release_delayed_node(delayed_node);
1479         return ret;
1480 }
1481
1482 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1483                                                struct btrfs_delayed_node *node,
1484                                                struct btrfs_key *key)
1485 {
1486         struct btrfs_delayed_item *item;
1487
1488         mutex_lock(&node->mutex);
1489         item = __btrfs_lookup_delayed_insertion_item(node, key);
1490         if (!item) {
1491                 mutex_unlock(&node->mutex);
1492                 return 1;
1493         }
1494
1495         btrfs_delayed_item_release_metadata(node->root, item);
1496         btrfs_release_delayed_item(item);
1497         mutex_unlock(&node->mutex);
1498         return 0;
1499 }
1500
1501 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1502                                    struct btrfs_fs_info *fs_info,
1503                                    struct btrfs_inode *dir, u64 index)
1504 {
1505         struct btrfs_delayed_node *node;
1506         struct btrfs_delayed_item *item;
1507         struct btrfs_key item_key;
1508         int ret;
1509
1510         node = btrfs_get_or_create_delayed_node(dir);
1511         if (IS_ERR(node))
1512                 return PTR_ERR(node);
1513
1514         item_key.objectid = btrfs_ino(dir);
1515         item_key.type = BTRFS_DIR_INDEX_KEY;
1516         item_key.offset = index;
1517
1518         ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key);
1519         if (!ret)
1520                 goto end;
1521
1522         item = btrfs_alloc_delayed_item(0);
1523         if (!item) {
1524                 ret = -ENOMEM;
1525                 goto end;
1526         }
1527
1528         item->key = item_key;
1529
1530         ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1531         /*
1532          * we have reserved enough space when we start a new transaction,
1533          * so reserving metadata failure is impossible.
1534          */
1535         BUG_ON(ret);
1536
1537         mutex_lock(&node->mutex);
1538         ret = __btrfs_add_delayed_deletion_item(node, item);
1539         if (unlikely(ret)) {
1540                 btrfs_err(fs_info,
1541                           "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1542                           index, node->root->objectid, node->inode_id, ret);
1543                 BUG();
1544         }
1545         mutex_unlock(&node->mutex);
1546 end:
1547         btrfs_release_delayed_node(node);
1548         return ret;
1549 }
1550
1551 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1552 {
1553         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1554
1555         if (!delayed_node)
1556                 return -ENOENT;
1557
1558         /*
1559          * Since we have held i_mutex of this directory, it is impossible that
1560          * a new directory index is added into the delayed node and index_cnt
1561          * is updated now. So we needn't lock the delayed node.
1562          */
1563         if (!delayed_node->index_cnt) {
1564                 btrfs_release_delayed_node(delayed_node);
1565                 return -EINVAL;
1566         }
1567
1568         inode->index_cnt = delayed_node->index_cnt;
1569         btrfs_release_delayed_node(delayed_node);
1570         return 0;
1571 }
1572
1573 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1574                                      struct list_head *ins_list,
1575                                      struct list_head *del_list)
1576 {
1577         struct btrfs_delayed_node *delayed_node;
1578         struct btrfs_delayed_item *item;
1579
1580         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1581         if (!delayed_node)
1582                 return false;
1583
1584         /*
1585          * We can only do one readdir with delayed items at a time because of
1586          * item->readdir_list.
1587          */
1588         inode_unlock_shared(inode);
1589         inode_lock(inode);
1590
1591         mutex_lock(&delayed_node->mutex);
1592         item = __btrfs_first_delayed_insertion_item(delayed_node);
1593         while (item) {
1594                 refcount_inc(&item->refs);
1595                 list_add_tail(&item->readdir_list, ins_list);
1596                 item = __btrfs_next_delayed_item(item);
1597         }
1598
1599         item = __btrfs_first_delayed_deletion_item(delayed_node);
1600         while (item) {
1601                 refcount_inc(&item->refs);
1602                 list_add_tail(&item->readdir_list, del_list);
1603                 item = __btrfs_next_delayed_item(item);
1604         }
1605         mutex_unlock(&delayed_node->mutex);
1606         /*
1607          * This delayed node is still cached in the btrfs inode, so refs
1608          * must be > 1 now, and we needn't check it is going to be freed
1609          * or not.
1610          *
1611          * Besides that, this function is used to read dir, we do not
1612          * insert/delete delayed items in this period. So we also needn't
1613          * requeue or dequeue this delayed node.
1614          */
1615         refcount_dec(&delayed_node->refs);
1616
1617         return true;
1618 }
1619
1620 void btrfs_readdir_put_delayed_items(struct inode *inode,
1621                                      struct list_head *ins_list,
1622                                      struct list_head *del_list)
1623 {
1624         struct btrfs_delayed_item *curr, *next;
1625
1626         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1627                 list_del(&curr->readdir_list);
1628                 if (refcount_dec_and_test(&curr->refs))
1629                         kfree(curr);
1630         }
1631
1632         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1633                 list_del(&curr->readdir_list);
1634                 if (refcount_dec_and_test(&curr->refs))
1635                         kfree(curr);
1636         }
1637
1638         /*
1639          * The VFS is going to do up_read(), so we need to downgrade back to a
1640          * read lock.
1641          */
1642         downgrade_write(&inode->i_rwsem);
1643 }
1644
1645 int btrfs_should_delete_dir_index(struct list_head *del_list,
1646                                   u64 index)
1647 {
1648         struct btrfs_delayed_item *curr;
1649         int ret = 0;
1650
1651         list_for_each_entry(curr, del_list, readdir_list) {
1652                 if (curr->key.offset > index)
1653                         break;
1654                 if (curr->key.offset == index) {
1655                         ret = 1;
1656                         break;
1657                 }
1658         }
1659         return ret;
1660 }
1661
1662 /*
1663  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1664  *
1665  */
1666 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1667                                     struct list_head *ins_list)
1668 {
1669         struct btrfs_dir_item *di;
1670         struct btrfs_delayed_item *curr, *next;
1671         struct btrfs_key location;
1672         char *name;
1673         int name_len;
1674         int over = 0;
1675         unsigned char d_type;
1676
1677         if (list_empty(ins_list))
1678                 return 0;
1679
1680         /*
1681          * Changing the data of the delayed item is impossible. So
1682          * we needn't lock them. And we have held i_mutex of the
1683          * directory, nobody can delete any directory indexes now.
1684          */
1685         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1686                 list_del(&curr->readdir_list);
1687
1688                 if (curr->key.offset < ctx->pos) {
1689                         if (refcount_dec_and_test(&curr->refs))
1690                                 kfree(curr);
1691                         continue;
1692                 }
1693
1694                 ctx->pos = curr->key.offset;
1695
1696                 di = (struct btrfs_dir_item *)curr->data;
1697                 name = (char *)(di + 1);
1698                 name_len = btrfs_stack_dir_name_len(di);
1699
1700                 d_type = btrfs_filetype_table[di->type];
1701                 btrfs_disk_key_to_cpu(&location, &di->location);
1702
1703                 over = !dir_emit(ctx, name, name_len,
1704                                location.objectid, d_type);
1705
1706                 if (refcount_dec_and_test(&curr->refs))
1707                         kfree(curr);
1708
1709                 if (over)
1710                         return 1;
1711                 ctx->pos++;
1712         }
1713         return 0;
1714 }
1715
1716 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1717                                   struct btrfs_inode_item *inode_item,
1718                                   struct inode *inode)
1719 {
1720         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1721         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1722         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1723         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1724         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1725         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1726         btrfs_set_stack_inode_generation(inode_item,
1727                                          BTRFS_I(inode)->generation);
1728         btrfs_set_stack_inode_sequence(inode_item,
1729                                        inode_peek_iversion(inode));
1730         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1731         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1732         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1733         btrfs_set_stack_inode_block_group(inode_item, 0);
1734
1735         btrfs_set_stack_timespec_sec(&inode_item->atime,
1736                                      inode->i_atime.tv_sec);
1737         btrfs_set_stack_timespec_nsec(&inode_item->atime,
1738                                       inode->i_atime.tv_nsec);
1739
1740         btrfs_set_stack_timespec_sec(&inode_item->mtime,
1741                                      inode->i_mtime.tv_sec);
1742         btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1743                                       inode->i_mtime.tv_nsec);
1744
1745         btrfs_set_stack_timespec_sec(&inode_item->ctime,
1746                                      inode->i_ctime.tv_sec);
1747         btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1748                                       inode->i_ctime.tv_nsec);
1749
1750         btrfs_set_stack_timespec_sec(&inode_item->otime,
1751                                      BTRFS_I(inode)->i_otime.tv_sec);
1752         btrfs_set_stack_timespec_nsec(&inode_item->otime,
1753                                      BTRFS_I(inode)->i_otime.tv_nsec);
1754 }
1755
1756 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1757 {
1758         struct btrfs_delayed_node *delayed_node;
1759         struct btrfs_inode_item *inode_item;
1760
1761         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1762         if (!delayed_node)
1763                 return -ENOENT;
1764
1765         mutex_lock(&delayed_node->mutex);
1766         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1767                 mutex_unlock(&delayed_node->mutex);
1768                 btrfs_release_delayed_node(delayed_node);
1769                 return -ENOENT;
1770         }
1771
1772         inode_item = &delayed_node->inode_item;
1773
1774         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1775         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1776         btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1777         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1778         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1779         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1780         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1781         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1782
1783         inode_set_iversion_queried(inode,
1784                                    btrfs_stack_inode_sequence(inode_item));
1785         inode->i_rdev = 0;
1786         *rdev = btrfs_stack_inode_rdev(inode_item);
1787         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1788
1789         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1790         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1791
1792         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1793         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1794
1795         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1796         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1797
1798         BTRFS_I(inode)->i_otime.tv_sec =
1799                 btrfs_stack_timespec_sec(&inode_item->otime);
1800         BTRFS_I(inode)->i_otime.tv_nsec =
1801                 btrfs_stack_timespec_nsec(&inode_item->otime);
1802
1803         inode->i_generation = BTRFS_I(inode)->generation;
1804         BTRFS_I(inode)->index_cnt = (u64)-1;
1805
1806         mutex_unlock(&delayed_node->mutex);
1807         btrfs_release_delayed_node(delayed_node);
1808         return 0;
1809 }
1810
1811 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1812                                struct btrfs_root *root, struct inode *inode)
1813 {
1814         struct btrfs_delayed_node *delayed_node;
1815         int ret = 0;
1816
1817         delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1818         if (IS_ERR(delayed_node))
1819                 return PTR_ERR(delayed_node);
1820
1821         mutex_lock(&delayed_node->mutex);
1822         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1823                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1824                 goto release_node;
1825         }
1826
1827         ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1828                                                    delayed_node);
1829         if (ret)
1830                 goto release_node;
1831
1832         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1833         set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1834         delayed_node->count++;
1835         atomic_inc(&root->fs_info->delayed_root->items);
1836 release_node:
1837         mutex_unlock(&delayed_node->mutex);
1838         btrfs_release_delayed_node(delayed_node);
1839         return ret;
1840 }
1841
1842 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1843 {
1844         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1845         struct btrfs_delayed_node *delayed_node;
1846
1847         /*
1848          * we don't do delayed inode updates during log recovery because it
1849          * leads to enospc problems.  This means we also can't do
1850          * delayed inode refs
1851          */
1852         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1853                 return -EAGAIN;
1854
1855         delayed_node = btrfs_get_or_create_delayed_node(inode);
1856         if (IS_ERR(delayed_node))
1857                 return PTR_ERR(delayed_node);
1858
1859         /*
1860          * We don't reserve space for inode ref deletion is because:
1861          * - We ONLY do async inode ref deletion for the inode who has only
1862          *   one link(i_nlink == 1), it means there is only one inode ref.
1863          *   And in most case, the inode ref and the inode item are in the
1864          *   same leaf, and we will deal with them at the same time.
1865          *   Since we are sure we will reserve the space for the inode item,
1866          *   it is unnecessary to reserve space for inode ref deletion.
1867          * - If the inode ref and the inode item are not in the same leaf,
1868          *   We also needn't worry about enospc problem, because we reserve
1869          *   much more space for the inode update than it needs.
1870          * - At the worst, we can steal some space from the global reservation.
1871          *   It is very rare.
1872          */
1873         mutex_lock(&delayed_node->mutex);
1874         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1875                 goto release_node;
1876
1877         set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1878         delayed_node->count++;
1879         atomic_inc(&fs_info->delayed_root->items);
1880 release_node:
1881         mutex_unlock(&delayed_node->mutex);
1882         btrfs_release_delayed_node(delayed_node);
1883         return 0;
1884 }
1885
1886 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1887 {
1888         struct btrfs_root *root = delayed_node->root;
1889         struct btrfs_fs_info *fs_info = root->fs_info;
1890         struct btrfs_delayed_item *curr_item, *prev_item;
1891
1892         mutex_lock(&delayed_node->mutex);
1893         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1894         while (curr_item) {
1895                 btrfs_delayed_item_release_metadata(root, curr_item);
1896                 prev_item = curr_item;
1897                 curr_item = __btrfs_next_delayed_item(prev_item);
1898                 btrfs_release_delayed_item(prev_item);
1899         }
1900
1901         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1902         while (curr_item) {
1903                 btrfs_delayed_item_release_metadata(root, curr_item);
1904                 prev_item = curr_item;
1905                 curr_item = __btrfs_next_delayed_item(prev_item);
1906                 btrfs_release_delayed_item(prev_item);
1907         }
1908
1909         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1910                 btrfs_release_delayed_iref(delayed_node);
1911
1912         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1913                 btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1914                 btrfs_release_delayed_inode(delayed_node);
1915         }
1916         mutex_unlock(&delayed_node->mutex);
1917 }
1918
1919 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1920 {
1921         struct btrfs_delayed_node *delayed_node;
1922
1923         delayed_node = btrfs_get_delayed_node(inode);
1924         if (!delayed_node)
1925                 return;
1926
1927         __btrfs_kill_delayed_node(delayed_node);
1928         btrfs_release_delayed_node(delayed_node);
1929 }
1930
1931 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1932 {
1933         u64 inode_id = 0;
1934         struct btrfs_delayed_node *delayed_nodes[8];
1935         int i, n;
1936
1937         while (1) {
1938                 spin_lock(&root->inode_lock);
1939                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1940                                            (void **)delayed_nodes, inode_id,
1941                                            ARRAY_SIZE(delayed_nodes));
1942                 if (!n) {
1943                         spin_unlock(&root->inode_lock);
1944                         break;
1945                 }
1946
1947                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1948
1949                 for (i = 0; i < n; i++)
1950                         refcount_inc(&delayed_nodes[i]->refs);
1951                 spin_unlock(&root->inode_lock);
1952
1953                 for (i = 0; i < n; i++) {
1954                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1955                         btrfs_release_delayed_node(delayed_nodes[i]);
1956                 }
1957         }
1958 }
1959
1960 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1961 {
1962         struct btrfs_delayed_node *curr_node, *prev_node;
1963
1964         curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1965         while (curr_node) {
1966                 __btrfs_kill_delayed_node(curr_node);
1967
1968                 prev_node = curr_node;
1969                 curr_node = btrfs_next_delayed_node(curr_node);
1970                 btrfs_release_delayed_node(prev_node);
1971         }
1972 }
1973