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 "delayed-inode.h"
  22 #include "disk-io.h"
  23 #include "transaction.h"
  24 #include "ctree.h"
  25
  26 #define BTRFS_DELAYED_WRITEBACK         512
  27 #define BTRFS_DELAYED_BACKGROUND        128
  28 #define BTRFS_DELAYED_BATCH             16
  29
  30 static struct kmem_cache *delayed_node_cache;
  31
  32 int __init btrfs_delayed_inode_init(void)
  33 {
  34         delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
  35                                         sizeof(struct btrfs_delayed_node),
  36                                         0,
  37                                         SLAB_MEM_SPREAD,
  38                                         NULL);
  39         if (!delayed_node_cache)
  40                 return -ENOMEM;
  41         return 0;
  42 }
  43
  44 void btrfs_delayed_inode_exit(void)
  45 {
  46         kmem_cache_destroy(delayed_node_cache);
  47 }
  48
  49 static inline void btrfs_init_delayed_node(
  50                                 struct btrfs_delayed_node *delayed_node,
  51                                 struct btrfs_root *root, u64 inode_id)
  52 {
  53         delayed_node->root = root;
  54         delayed_node->inode_id = inode_id;
  55         refcount_set(&delayed_node->refs, 0);
  56         delayed_node->ins_root = RB_ROOT;
  57         delayed_node->del_root = RB_ROOT;
  58         mutex_init(&delayed_node->mutex);
  59         INIT_LIST_HEAD(&delayed_node->n_list);
  60         INIT_LIST_HEAD(&delayed_node->p_list);
  61 }
  62
  63 static inline int btrfs_is_continuous_delayed_item(
  64                                         struct btrfs_delayed_item *item1,
  65                                         struct btrfs_delayed_item *item2)
  66 {
  67         if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  68             item1->key.objectid == item2->key.objectid &&
  69             item1->key.type == item2->key.type &&
  70             item1->key.offset + 1 == item2->key.offset)
  71                 return 1;
  72         return 0;
  73 }
  74
  75 static struct btrfs_delayed_node *btrfs_get_delayed_node(
  76                 struct btrfs_inode *btrfs_inode)
  77 {
  78         struct btrfs_root *root = btrfs_inode->root;
  79         u64 ino = btrfs_ino(btrfs_inode);
  80         struct btrfs_delayed_node *node;
  81
  82         node = READ_ONCE(btrfs_inode->delayed_node);
  83         if (node) {
  84                 refcount_inc(&node->refs);
  85                 return node;
  86         }
  87
  88         spin_lock(&root->inode_lock);
  89         node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
  90
  91         if (node) {
  92                 if (btrfs_inode->delayed_node) {
  93                         refcount_inc(&node->refs);      /* can be accessed */
  94                         BUG_ON(btrfs_inode->delayed_node != node);
  95                         spin_unlock(&root->inode_lock);
  96                         return node;
  97                 }
  98
  99                 /*
 100                  * It's possible that we're racing into the middle of removing
 101                  * this node from the radix tree.  In this case, the refcount
 102                  * was zero and it should never go back to one.  Just return
 103                  * NULL like it was never in the radix at all; our release
 104                  * function is in the process of removing it.
 105                  *
 106                  * Some implementations of refcount_inc refuse to bump the
 107                  * refcount once it has hit zero.  If we don't do this dance
 108                  * here, refcount_inc() may decide to just WARN_ONCE() instead
 109                  * of actually bumping the refcount.
 110                  *
 111                  * If this node is properly in the radix, we want to bump the
 112                  * refcount twice, once for the inode and once for this get
 113                  * operation.
 114                  */
 115                 if (refcount_inc_not_zero(&node->refs)) {
 116                         refcount_inc(&node->refs);
 117                         btrfs_inode->delayed_node = node;
 118                 } else {
 119                         node = NULL;
 120                 }
 121
 122                 spin_unlock(&root->inode_lock);
 123                 return node;
 124         }
 125         spin_unlock(&root->inode_lock);
 126
 127         return NULL;
 128 }
 129
 130 /* Will return either the node or PTR_ERR(-ENOMEM) */
 131 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 132                 struct btrfs_inode *btrfs_inode)
 133 {
 134         struct btrfs_delayed_node *node;
 135         struct btrfs_root *root = btrfs_inode->root;
 136         u64 ino = btrfs_ino(btrfs_inode);
 137         int ret;
 138
 139 again:
 140         node = btrfs_get_delayed_node(btrfs_inode);
 141         if (node)
 142                 return node;
 143
 144         node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
 145         if (!node)
 146                 return ERR_PTR(-ENOMEM);
 147         btrfs_init_delayed_node(node, root, ino);
 148
 149         /* cached in the btrfs inode and can be accessed */
 150         refcount_set(&node->refs, 2);
 151
 152         ret = radix_tree_preload(GFP_NOFS);
 153         if (ret) {
 154                 kmem_cache_free(delayed_node_cache, node);
 155                 return ERR_PTR(ret);
 156         }
 157
 158         spin_lock(&root->inode_lock);
 159         ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 160         if (ret == -EEXIST) {
 161                 spin_unlock(&root->inode_lock);
 162                 kmem_cache_free(delayed_node_cache, node);
 163                 radix_tree_preload_end();
 164                 goto again;
 165         }
 166         btrfs_inode->delayed_node = node;
 167         spin_unlock(&root->inode_lock);
 168         radix_tree_preload_end();
 169
 170         return node;
 171 }
 172
 173 /*
 174  * Call it when holding delayed_node->mutex
 175  *
 176  * If mod = 1, add this node into the prepared list.
 177  */
 178 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 179                                      struct btrfs_delayed_node *node,
 180                                      int mod)
 181 {
 182         spin_lock(&root->lock);
 183         if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 184                 if (!list_empty(&node->p_list))
 185                         list_move_tail(&node->p_list, &root->prepare_list);
 186                 else if (mod)
 187                         list_add_tail(&node->p_list, &root->prepare_list);
 188         } else {
 189                 list_add_tail(&node->n_list, &root->node_list);
 190                 list_add_tail(&node->p_list, &root->prepare_list);
 191                 refcount_inc(&node->refs);      /* inserted into list */
 192                 root->nodes++;
 193                 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 194         }
 195         spin_unlock(&root->lock);
 196 }
 197
 198 /* Call it when holding delayed_node->mutex */
 199 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 200                                        struct btrfs_delayed_node *node)
 201 {
 202         spin_lock(&root->lock);
 203         if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 204                 root->nodes--;
 205                 refcount_dec(&node->refs);      /* not in the list */
 206                 list_del_init(&node->n_list);
 207                 if (!list_empty(&node->p_list))
 208                         list_del_init(&node->p_list);
 209                 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 210         }
 211         spin_unlock(&root->lock);
 212 }
 213
 214 static struct btrfs_delayed_node *btrfs_first_delayed_node(
 215                         struct btrfs_delayed_root *delayed_root)
 216 {
 217         struct list_head *p;
 218         struct btrfs_delayed_node *node = NULL;
 219
 220         spin_lock(&delayed_root->lock);
 221         if (list_empty(&delayed_root->node_list))
 222                 goto out;
 223
 224         p = delayed_root->node_list.next;
 225         node = list_entry(p, struct btrfs_delayed_node, n_list);
 226         refcount_inc(&node->refs);
 227 out:
 228         spin_unlock(&delayed_root->lock);
 229
 230         return node;
 231 }
 232
 233 static struct btrfs_delayed_node *btrfs_next_delayed_node(
 234                                                 struct btrfs_delayed_node *node)
 235 {
 236         struct btrfs_delayed_root *delayed_root;
 237         struct list_head *p;
 238         struct btrfs_delayed_node *next = NULL;
 239
 240         delayed_root = node->root->fs_info->delayed_root;
 241         spin_lock(&delayed_root->lock);
 242         if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 243                 /* not in the list */
 244                 if (list_empty(&delayed_root->node_list))
 245                         goto out;
 246                 p = delayed_root->node_list.next;
 247         } else if (list_is_last(&node->n_list, &delayed_root->node_list))
 248                 goto out;
 249         else
 250                 p = node->n_list.next;
 251
 252         next = list_entry(p, struct btrfs_delayed_node, n_list);
 253         refcount_inc(&next->refs);
 254 out:
 255         spin_unlock(&delayed_root->lock);
 256
 257         return next;
 258 }
 259
 260 static void __btrfs_release_delayed_node(
 261                                 struct btrfs_delayed_node *delayed_node,
 262                                 int mod)
 263 {
 264         struct btrfs_delayed_root *delayed_root;
 265
 266         if (!delayed_node)
 267                 return;
 268
 269         delayed_root = delayed_node->root->fs_info->delayed_root;
 270
 271         mutex_lock(&delayed_node->mutex);
 272         if (delayed_node->count)
 273                 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
 274         else
 275                 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 276         mutex_unlock(&delayed_node->mutex);
 277
 278         if (refcount_dec_and_test(&delayed_node->refs)) {
 279                 struct btrfs_root *root = delayed_node->root;
 280
 281                 spin_lock(&root->inode_lock);
 282                 /*
 283                  * Once our refcount goes to zero, nobody is allowed to bump it
 284                  * back up.  We can delete it now.
 285                  */
 286                 ASSERT(refcount_read(&delayed_node->refs) == 0);
 287                 radix_tree_delete(&root->delayed_nodes_tree,
 288                                   delayed_node->inode_id);
 289                 spin_unlock(&root->inode_lock);
 290                 kmem_cache_free(delayed_node_cache, delayed_node);
 291         }
 292 }
 293
 294 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
 295 {
 296         __btrfs_release_delayed_node(node, 0);
 297 }
 298
 299 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
 300                                         struct btrfs_delayed_root *delayed_root)
 301 {
 302         struct list_head *p;
 303         struct btrfs_delayed_node *node = NULL;
 304
 305         spin_lock(&delayed_root->lock);
 306         if (list_empty(&delayed_root->prepare_list))
 307                 goto out;
 308
 309         p = delayed_root->prepare_list.next;
 310         list_del_init(p);
 311         node = list_entry(p, struct btrfs_delayed_node, p_list);
 312         refcount_inc(&node->refs);
 313 out:
 314         spin_unlock(&delayed_root->lock);
 315
 316         return node;
 317 }
 318
 319 static inline void btrfs_release_prepared_delayed_node(
 320                                         struct btrfs_delayed_node *node)
 321 {
 322         __btrfs_release_delayed_node(node, 1);
 323 }
 324
 325 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
 326 {
 327         struct btrfs_delayed_item *item;
 328         item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
 329         if (item) {
 330                 item->data_len = data_len;
 331                 item->ins_or_del = 0;
 332                 item->bytes_reserved = 0;
 333                 item->delayed_node = NULL;
 334                 refcount_set(&item->refs, 1);
 335         }
 336         return item;
 337 }
 338
 339 /*
 340  * __btrfs_lookup_delayed_item - look up the delayed item by key
 341  * @delayed_node: pointer to the delayed node
 342  * @key:          the key to look up
 343  * @prev:         used to store the prev item if the right item isn't found
 344  * @next:         used to store the next item if the right item isn't found
 345  *
 346  * Note: if we don't find the right item, we will return the prev item and
 347  * the next item.
 348  */
 349 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 350                                 struct rb_root *root,
 351                                 struct btrfs_key *key,
 352                                 struct btrfs_delayed_item **prev,
 353                                 struct btrfs_delayed_item **next)
 354 {
 355         struct rb_node *node, *prev_node = NULL;
 356         struct btrfs_delayed_item *delayed_item = NULL;
 357         int ret = 0;
 358
 359         node = root->rb_node;
 360
 361         while (node) {
 362                 delayed_item = rb_entry(node, struct btrfs_delayed_item,
 363                                         rb_node);
 364                 prev_node = node;
 365                 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
 366                 if (ret < 0)
 367                         node = node->rb_right;
 368                 else if (ret > 0)
 369                         node = node->rb_left;
 370                 else
 371                         return delayed_item;
 372         }
 373
 374         if (prev) {
 375                 if (!prev_node)
 376                         *prev = NULL;
 377                 else if (ret < 0)
 378                         *prev = delayed_item;
 379                 else if ((node = rb_prev(prev_node)) != NULL) {
 380                         *prev = rb_entry(node, struct btrfs_delayed_item,
 381                                          rb_node);
 382                 } else
 383                         *prev = NULL;
 384         }
 385
 386         if (next) {
 387                 if (!prev_node)
 388                         *next = NULL;
 389                 else if (ret > 0)
 390                         *next = delayed_item;
 391                 else if ((node = rb_next(prev_node)) != NULL) {
 392                         *next = rb_entry(node, struct btrfs_delayed_item,
 393                                          rb_node);
 394                 } else
 395                         *next = NULL;
 396         }
 397         return NULL;
 398 }
 399
 400 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
 401                                         struct btrfs_delayed_node *delayed_node,
 402                                         struct btrfs_key *key)
 403 {
 404         return __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 405                                            NULL, NULL);
 406 }
 407
 408 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
 409                                     struct btrfs_delayed_item *ins,
 410                                     int action)
 411 {
 412         struct rb_node **p, *node;
 413         struct rb_node *parent_node = NULL;
 414         struct rb_root *root;
 415         struct btrfs_delayed_item *item;
 416         int cmp;
 417
 418         if (action == BTRFS_DELAYED_INSERTION_ITEM)
 419                 root = &delayed_node->ins_root;
 420         else if (action == BTRFS_DELAYED_DELETION_ITEM)
 421                 root = &delayed_node->del_root;
 422         else
 423                 BUG();
 424         p = &root->rb_node;
 425         node = &ins->rb_node;
 426
 427         while (*p) {
 428                 parent_node = *p;
 429                 item = rb_entry(parent_node, struct btrfs_delayed_item,
 430                                  rb_node);
 431
 432                 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
 433                 if (cmp < 0)
 434                         p = &(*p)->rb_right;
 435                 else if (cmp > 0)
 436                         p = &(*p)->rb_left;
 437                 else
 438                         return -EEXIST;
 439         }
 440
 441         rb_link_node(node, parent_node, p);
 442         rb_insert_color(node, root);
 443         ins->delayed_node = delayed_node;
 444         ins->ins_or_del = action;
 445
 446         if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
 447             action == BTRFS_DELAYED_INSERTION_ITEM &&
 448             ins->key.offset >= delayed_node->index_cnt)
 449                         delayed_node->index_cnt = ins->key.offset + 1;
 450
 451         delayed_node->count++;
 452         atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
 453         return 0;
 454 }
 455
 456 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
 457                                               struct btrfs_delayed_item *item)
 458 {
 459         return __btrfs_add_delayed_item(node, item,
 460                                         BTRFS_DELAYED_INSERTION_ITEM);
 461 }
 462
 463 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
 464                                              struct btrfs_delayed_item *item)
 465 {
 466         return __btrfs_add_delayed_item(node, item,
 467                                         BTRFS_DELAYED_DELETION_ITEM);
 468 }
 469
 470 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
 471 {
 472         int seq = atomic_inc_return(&delayed_root->items_seq);
 473
 474         /*
 475          * atomic_dec_return implies a barrier for waitqueue_active
 476          */
 477         if ((atomic_dec_return(&delayed_root->items) <
 478             BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
 479             waitqueue_active(&delayed_root->wait))
 480                 wake_up(&delayed_root->wait);
 481 }
 482
 483 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 484 {
 485         struct rb_root *root;
 486         struct btrfs_delayed_root *delayed_root;
 487
 488         delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
 489
 490         BUG_ON(!delayed_root);
 491         BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
 492                delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
 493
 494         if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
 495                 root = &delayed_item->delayed_node->ins_root;
 496         else
 497                 root = &delayed_item->delayed_node->del_root;
 498
 499         rb_erase(&delayed_item->rb_node, root);
 500         delayed_item->delayed_node->count--;
 501
 502         finish_one_item(delayed_root);
 503 }
 504
 505 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 506 {
 507         if (item) {
 508                 __btrfs_remove_delayed_item(item);
 509                 if (refcount_dec_and_test(&item->refs))
 510                         kfree(item);
 511         }
 512 }
 513
 514 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 515                                         struct btrfs_delayed_node *delayed_node)
 516 {
 517         struct rb_node *p;
 518         struct btrfs_delayed_item *item = NULL;
 519
 520         p = rb_first(&delayed_node->ins_root);
 521         if (p)
 522                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 523
 524         return item;
 525 }
 526
 527 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 528                                         struct btrfs_delayed_node *delayed_node)
 529 {
 530         struct rb_node *p;
 531         struct btrfs_delayed_item *item = NULL;
 532
 533         p = rb_first(&delayed_node->del_root);
 534         if (p)
 535                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 536
 537         return item;
 538 }
 539
 540 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
 541                                                 struct btrfs_delayed_item *item)
 542 {
 543         struct rb_node *p;
 544         struct btrfs_delayed_item *next = NULL;
 545
 546         p = rb_next(&item->rb_node);
 547         if (p)
 548                 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
 549
 550         return next;
 551 }
 552
 553 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 554                                                struct btrfs_fs_info *fs_info,
 555                                                struct btrfs_delayed_item *item)
 556 {
 557         struct btrfs_block_rsv *src_rsv;
 558         struct btrfs_block_rsv *dst_rsv;
 559         u64 num_bytes;
 560         int ret;
 561
 562         if (!trans->bytes_reserved)
 563                 return 0;
 564
 565         src_rsv = trans->block_rsv;
 566         dst_rsv = &fs_info->delayed_block_rsv;
 567
 568         num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
 569         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
 570         if (!ret) {
 571                 trace_btrfs_space_reservation(fs_info, "delayed_item",
 572                                               item->key.objectid,
 573                                               num_bytes, 1);
 574                 item->bytes_reserved = num_bytes;
 575         }
 576
 577         return ret;
 578 }
 579
 580 static void btrfs_delayed_item_release_metadata(struct btrfs_fs_info *fs_info,
 581                                                 struct btrfs_delayed_item *item)
 582 {
 583         struct btrfs_block_rsv *rsv;
 584
 585         if (!item->bytes_reserved)
 586                 return;
 587
 588         rsv = &fs_info->delayed_block_rsv;
 589         trace_btrfs_space_reservation(fs_info, "delayed_item",
 590                                       item->key.objectid, item->bytes_reserved,
 591                                       0);
 592         btrfs_block_rsv_release(fs_info, rsv,
 593                                 item->bytes_reserved);
 594 }
 595
 596 static int btrfs_delayed_inode_reserve_metadata(
 597                                         struct btrfs_trans_handle *trans,
 598                                         struct btrfs_root *root,
 599                                         struct btrfs_inode *inode,
 600                                         struct btrfs_delayed_node *node)
 601 {
 602         struct btrfs_fs_info *fs_info = root->fs_info;
 603         struct btrfs_block_rsv *src_rsv;
 604         struct btrfs_block_rsv *dst_rsv;
 605         u64 num_bytes;
 606         int ret;
 607         bool release = false;
 608
 609         src_rsv = trans->block_rsv;
 610         dst_rsv = &fs_info->delayed_block_rsv;
 611
 612         num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1);
 613
 614         /*
 615          * If our block_rsv is the delalloc block reserve then check and see if
 616          * we have our extra reservation for updating the inode.  If not fall
 617          * through and try to reserve space quickly.
 618          *
 619          * We used to try and steal from the delalloc block rsv or the global
 620          * reserve, but we'd steal a full reservation, which isn't kind.  We are
 621          * here through delalloc which means we've likely just cowed down close
 622          * to the leaf that contains the inode, so we would steal less just
 623          * doing the fallback inode update, so if we do end up having to steal
 624          * from the global block rsv we hopefully only steal one or two blocks
 625          * worth which is less likely to hurt us.
 626          */
 627         if (src_rsv && src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
 628                 spin_lock(&inode->lock);
 629                 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
 630                                        &inode->runtime_flags))
 631                         release = true;
 632                 else
 633                         src_rsv = NULL;
 634                 spin_unlock(&inode->lock);
 635         }
 636
 637         /*
 638          * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 639          * which doesn't reserve space for speed.  This is a problem since we
 640          * still need to reserve space for this update, so try to reserve the
 641          * space.
 642          *
 643          * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 644          * we're accounted for.
 645          */
 646         if (!src_rsv || (!trans->bytes_reserved &&
 647                          src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
 648                 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 649                                           BTRFS_RESERVE_NO_FLUSH);
 650                 /*
 651                  * Since we're under a transaction reserve_metadata_bytes could
 652                  * try to commit the transaction which will make it return
 653                  * EAGAIN to make us stop the transaction we have, so return
 654                  * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 655                  */
 656                 if (ret == -EAGAIN)
 657                         ret = -ENOSPC;
 658                 if (!ret) {
 659                         node->bytes_reserved = num_bytes;
 660                         trace_btrfs_space_reservation(fs_info,
 661                                                       "delayed_inode",
 662                                                       btrfs_ino(inode),
 663                                                       num_bytes, 1);
 664                 }
 665                 return ret;
 666         }
 667
 668         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1);
 669
 670         /*
 671          * Migrate only takes a reservation, it doesn't touch the size of the
 672          * block_rsv.  This is to simplify people who don't normally have things
 673          * migrated from their block rsv.  If they go to release their
 674          * reservation, that will decrease the size as well, so if migrate
 675          * reduced size we'd end up with a negative size.  But for the
 676          * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
 677          * but we could in fact do this reserve/migrate dance several times
 678          * between the time we did the original reservation and we'd clean it
 679          * up.  So to take care of this, release the space for the meta
 680          * reservation here.  I think it may be time for a documentation page on
 681          * how block rsvs. work.
 682          */
 683         if (!ret) {
 684                 trace_btrfs_space_reservation(fs_info, "delayed_inode",
 685                                               btrfs_ino(inode), num_bytes, 1);
 686                 node->bytes_reserved = num_bytes;
 687         }
 688
 689         if (release) {
 690                 trace_btrfs_space_reservation(fs_info, "delalloc",
 691                                               btrfs_ino(inode), num_bytes, 0);
 692                 btrfs_block_rsv_release(fs_info, src_rsv, num_bytes);
 693         }
 694
 695         return ret;
 696 }
 697
 698 static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
 699                                                 struct btrfs_delayed_node *node)
 700 {
 701         struct btrfs_block_rsv *rsv;
 702
 703         if (!node->bytes_reserved)
 704                 return;
 705
 706         rsv = &fs_info->delayed_block_rsv;
 707         trace_btrfs_space_reservation(fs_info, "delayed_inode",
 708                                       node->inode_id, node->bytes_reserved, 0);
 709         btrfs_block_rsv_release(fs_info, rsv,
 710                                 node->bytes_reserved);
 711         node->bytes_reserved = 0;
 712 }
 713
 714 /*
 715  * This helper will insert some continuous items into the same leaf according
 716  * to the free space of the leaf.
 717  */
 718 static int btrfs_batch_insert_items(struct btrfs_root *root,
 719                                     struct btrfs_path *path,
 720                                     struct btrfs_delayed_item *item)
 721 {
 722         struct btrfs_fs_info *fs_info = root->fs_info;
 723         struct btrfs_delayed_item *curr, *next;
 724         int free_space;
 725         int total_data_size = 0, total_size = 0;
 726         struct extent_buffer *leaf;
 727         char *data_ptr;
 728         struct btrfs_key *keys;
 729         u32 *data_size;
 730         struct list_head head;
 731         int slot;
 732         int nitems;
 733         int i;
 734         int ret = 0;
 735
 736         BUG_ON(!path->nodes[0]);
 737
 738         leaf = path->nodes[0];
 739         free_space = btrfs_leaf_free_space(fs_info, leaf);
 740         INIT_LIST_HEAD(&head);
 741
 742         next = item;
 743         nitems = 0;
 744
 745         /*
 746          * count the number of the continuous items that we can insert in batch
 747          */
 748         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 749                free_space) {
 750                 total_data_size += next->data_len;
 751                 total_size += next->data_len + sizeof(struct btrfs_item);
 752                 list_add_tail(&next->tree_list, &head);
 753                 nitems++;
 754
 755                 curr = next;
 756                 next = __btrfs_next_delayed_item(curr);
 757                 if (!next)
 758                         break;
 759
 760                 if (!btrfs_is_continuous_delayed_item(curr, next))
 761                         break;
 762         }
 763
 764         if (!nitems) {
 765                 ret = 0;
 766                 goto out;
 767         }
 768
 769         /*
 770          * we need allocate some memory space, but it might cause the task
 771          * to sleep, so we set all locked nodes in the path to blocking locks
 772          * first.
 773          */
 774         btrfs_set_path_blocking(path);
 775
 776         keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
 777         if (!keys) {
 778                 ret = -ENOMEM;
 779                 goto out;
 780         }
 781
 782         data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
 783         if (!data_size) {
 784                 ret = -ENOMEM;
 785                 goto error;
 786         }
 787
 788         /* get keys of all the delayed items */
 789         i = 0;
 790         list_for_each_entry(next, &head, tree_list) {
 791                 keys[i] = next->key;
 792                 data_size[i] = next->data_len;
 793                 i++;
 794         }
 795
 796         /* reset all the locked nodes in the patch to spinning locks. */
 797         btrfs_clear_path_blocking(path, NULL, 0);
 798
 799         /* insert the keys of the items */
 800         setup_items_for_insert(root, path, keys, data_size,
 801                                total_data_size, total_size, nitems);
 802
 803         /* insert the dir index items */
 804         slot = path->slots[0];
 805         list_for_each_entry_safe(curr, next, &head, tree_list) {
 806                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 807                 write_extent_buffer(leaf, &curr->data,
 808                                     (unsigned long)data_ptr,
 809                                     curr->data_len);
 810                 slot++;
 811
 812                 btrfs_delayed_item_release_metadata(fs_info, curr);
 813
 814                 list_del(&curr->tree_list);
 815                 btrfs_release_delayed_item(curr);
 816         }
 817
 818 error:
 819         kfree(data_size);
 820         kfree(keys);
 821 out:
 822         return ret;
 823 }
 824
 825 /*
 826  * This helper can just do simple insertion that needn't extend item for new
 827  * data, such as directory name index insertion, inode insertion.
 828  */
 829 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 830                                      struct btrfs_root *root,
 831                                      struct btrfs_path *path,
 832                                      struct btrfs_delayed_item *delayed_item)
 833 {
 834         struct btrfs_fs_info *fs_info = root->fs_info;
 835         struct extent_buffer *leaf;
 836         char *ptr;
 837         int ret;
 838
 839         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 840                                       delayed_item->data_len);
 841         if (ret < 0 && ret != -EEXIST)
 842                 return ret;
 843
 844         leaf = path->nodes[0];
 845
 846         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 847
 848         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 849                             delayed_item->data_len);
 850         btrfs_mark_buffer_dirty(leaf);
 851
 852         btrfs_delayed_item_release_metadata(fs_info, delayed_item);
 853         return 0;
 854 }
 855
 856 /*
 857  * we insert an item first, then if there are some continuous items, we try
 858  * to insert those items into the same leaf.
 859  */
 860 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 861                                       struct btrfs_path *path,
 862                                       struct btrfs_root *root,
 863                                       struct btrfs_delayed_node *node)
 864 {
 865         struct btrfs_delayed_item *curr, *prev;
 866         int ret = 0;
 867
 868 do_again:
 869         mutex_lock(&node->mutex);
 870         curr = __btrfs_first_delayed_insertion_item(node);
 871         if (!curr)
 872                 goto insert_end;
 873
 874         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 875         if (ret < 0) {
 876                 btrfs_release_path(path);
 877                 goto insert_end;
 878         }
 879
 880         prev = curr;
 881         curr = __btrfs_next_delayed_item(prev);
 882         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 883                 /* insert the continuous items into the same leaf */
 884                 path->slots[0]++;
 885                 btrfs_batch_insert_items(root, path, curr);
 886         }
 887         btrfs_release_delayed_item(prev);
 888         btrfs_mark_buffer_dirty(path->nodes[0]);
 889
 890         btrfs_release_path(path);
 891         mutex_unlock(&node->mutex);
 892         goto do_again;
 893
 894 insert_end:
 895         mutex_unlock(&node->mutex);
 896         return ret;
 897 }
 898
 899 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 900                                     struct btrfs_root *root,
 901                                     struct btrfs_path *path,
 902                                     struct btrfs_delayed_item *item)
 903 {
 904         struct btrfs_fs_info *fs_info = root->fs_info;
 905         struct btrfs_delayed_item *curr, *next;
 906         struct extent_buffer *leaf;
 907         struct btrfs_key key;
 908         struct list_head head;
 909         int nitems, i, last_item;
 910         int ret = 0;
 911
 912         BUG_ON(!path->nodes[0]);
 913
 914         leaf = path->nodes[0];
 915
 916         i = path->slots[0];
 917         last_item = btrfs_header_nritems(leaf) - 1;
 918         if (i > last_item)
 919                 return -ENOENT; /* FIXME: Is errno suitable? */
 920
 921         next = item;
 922         INIT_LIST_HEAD(&head);
 923         btrfs_item_key_to_cpu(leaf, &key, i);
 924         nitems = 0;
 925         /*
 926          * count the number of the dir index items that we can delete in batch
 927          */
 928         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 929                 list_add_tail(&next->tree_list, &head);
 930                 nitems++;
 931
 932                 curr = next;
 933                 next = __btrfs_next_delayed_item(curr);
 934                 if (!next)
 935                         break;
 936
 937                 if (!btrfs_is_continuous_delayed_item(curr, next))
 938                         break;
 939
 940                 i++;
 941                 if (i > last_item)
 942                         break;
 943                 btrfs_item_key_to_cpu(leaf, &key, i);
 944         }
 945
 946         if (!nitems)
 947                 return 0;
 948
 949         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 950         if (ret)
 951                 goto out;
 952
 953         list_for_each_entry_safe(curr, next, &head, tree_list) {
 954                 btrfs_delayed_item_release_metadata(fs_info, curr);
 955                 list_del(&curr->tree_list);
 956                 btrfs_release_delayed_item(curr);
 957         }
 958
 959 out:
 960         return ret;
 961 }
 962
 963 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 964                                       struct btrfs_path *path,
 965                                       struct btrfs_root *root,
 966                                       struct btrfs_delayed_node *node)
 967 {
 968         struct btrfs_delayed_item *curr, *prev;
 969         int ret = 0;
 970
 971 do_again:
 972         mutex_lock(&node->mutex);
 973         curr = __btrfs_first_delayed_deletion_item(node);
 974         if (!curr)
 975                 goto delete_fail;
 976
 977         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 978         if (ret < 0)
 979                 goto delete_fail;
 980         else if (ret > 0) {
 981                 /*
 982                  * can't find the item which the node points to, so this node
 983                  * is invalid, just drop it.
 984                  */
 985                 prev = curr;
 986                 curr = __btrfs_next_delayed_item(prev);
 987                 btrfs_release_delayed_item(prev);
 988                 ret = 0;
 989                 btrfs_release_path(path);
 990                 if (curr) {
 991                         mutex_unlock(&node->mutex);
 992                         goto do_again;
 993                 } else
 994                         goto delete_fail;
 995         }
 996
 997         btrfs_batch_delete_items(trans, root, path, curr);
 998         btrfs_release_path(path);
 999         mutex_unlock(&node->mutex);
1000         goto do_again;
1001
1002 delete_fail:
1003         btrfs_release_path(path);
1004         mutex_unlock(&node->mutex);
1005         return ret;
1006 }
1007
1008 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1009 {
1010         struct btrfs_delayed_root *delayed_root;
1011
1012         if (delayed_node &&
1013             test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1014                 BUG_ON(!delayed_node->root);
1015                 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1016                 delayed_node->count--;
1017
1018                 delayed_root = delayed_node->root->fs_info->delayed_root;
1019                 finish_one_item(delayed_root);
1020         }
1021 }
1022
1023 static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
1024 {
1025         struct btrfs_delayed_root *delayed_root;
1026
1027         ASSERT(delayed_node->root);
1028         clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1029         delayed_node->count--;
1030
1031         delayed_root = delayed_node->root->fs_info->delayed_root;
1032         finish_one_item(delayed_root);
1033 }
1034
1035 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1036                                         struct btrfs_root *root,
1037                                         struct btrfs_path *path,
1038                                         struct btrfs_delayed_node *node)
1039 {
1040         struct btrfs_fs_info *fs_info = root->fs_info;
1041         struct btrfs_key key;
1042         struct btrfs_inode_item *inode_item;
1043         struct extent_buffer *leaf;
1044         int mod;
1045         int ret;
1046
1047         key.objectid = node->inode_id;
1048         key.type = BTRFS_INODE_ITEM_KEY;
1049         key.offset = 0;
1050
1051         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1052                 mod = -1;
1053         else
1054                 mod = 1;
1055
1056         ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1057         if (ret > 0) {
1058                 btrfs_release_path(path);
1059                 return -ENOENT;
1060         } else if (ret < 0) {
1061                 return ret;
1062         }
1063
1064         leaf = path->nodes[0];
1065         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1066                                     struct btrfs_inode_item);
1067         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1068                             sizeof(struct btrfs_inode_item));
1069         btrfs_mark_buffer_dirty(leaf);
1070
1071         if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1072                 goto no_iref;
1073
1074         path->slots[0]++;
1075         if (path->slots[0] >= btrfs_header_nritems(leaf))
1076                 goto search;
1077 again:
1078         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1079         if (key.objectid != node->inode_id)
1080                 goto out;
1081
1082         if (key.type != BTRFS_INODE_REF_KEY &&
1083             key.type != BTRFS_INODE_EXTREF_KEY)
1084                 goto out;
1085
1086         /*
1087          * Delayed iref deletion is for the inode who has only one link,
1088          * so there is only one iref. The case that several irefs are
1089          * in the same item doesn't exist.
1090          */
1091         btrfs_del_item(trans, root, path);
1092 out:
1093         btrfs_release_delayed_iref(node);
1094 no_iref:
1095         btrfs_release_path(path);
1096 err_out:
1097         btrfs_delayed_inode_release_metadata(fs_info, node);
1098         btrfs_release_delayed_inode(node);
1099
1100         return ret;
1101
1102 search:
1103         btrfs_release_path(path);
1104
1105         key.type = BTRFS_INODE_EXTREF_KEY;
1106         key.offset = -1;
1107         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1108         if (ret < 0)
1109                 goto err_out;
1110         ASSERT(ret);
1111
1112         ret = 0;
1113         leaf = path->nodes[0];
1114         path->slots[0]--;
1115         goto again;
1116 }
1117
1118 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1119                                              struct btrfs_root *root,
1120                                              struct btrfs_path *path,
1121                                              struct btrfs_delayed_node *node)
1122 {
1123         int ret;
1124
1125         mutex_lock(&node->mutex);
1126         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1127                 mutex_unlock(&node->mutex);
1128                 return 0;
1129         }
1130
1131         ret = __btrfs_update_delayed_inode(trans, root, path, node);
1132         mutex_unlock(&node->mutex);
1133         return ret;
1134 }
1135
1136 static inline int
1137 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1138                                    struct btrfs_path *path,
1139                                    struct btrfs_delayed_node *node)
1140 {
1141         int ret;
1142
1143         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1144         if (ret)
1145                 return ret;
1146
1147         ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1148         if (ret)
1149                 return ret;
1150
1151         ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1152         return ret;
1153 }
1154
1155 /*
1156  * Called when committing the transaction.
1157  * Returns 0 on success.
1158  * Returns < 0 on error and returns with an aborted transaction with any
1159  * outstanding delayed items cleaned up.
1160  */
1161 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1162                                      struct btrfs_fs_info *fs_info, int nr)
1163 {
1164         struct btrfs_delayed_root *delayed_root;
1165         struct btrfs_delayed_node *curr_node, *prev_node;
1166         struct btrfs_path *path;
1167         struct btrfs_block_rsv *block_rsv;
1168         int ret = 0;
1169         bool count = (nr > 0);
1170
1171         if (trans->aborted)
1172                 return -EIO;
1173
1174         path = btrfs_alloc_path();
1175         if (!path)
1176                 return -ENOMEM;
1177         path->leave_spinning = 1;
1178
1179         block_rsv = trans->block_rsv;
1180         trans->block_rsv = &fs_info->delayed_block_rsv;
1181
1182         delayed_root = fs_info->delayed_root;
1183
1184         curr_node = btrfs_first_delayed_node(delayed_root);
1185         while (curr_node && (!count || (count && nr--))) {
1186                 ret = __btrfs_commit_inode_delayed_items(trans, path,
1187                                                          curr_node);
1188                 if (ret) {
1189                         btrfs_release_delayed_node(curr_node);
1190                         curr_node = NULL;
1191                         btrfs_abort_transaction(trans, ret);
1192                         break;
1193                 }
1194
1195                 prev_node = curr_node;
1196                 curr_node = btrfs_next_delayed_node(curr_node);
1197                 btrfs_release_delayed_node(prev_node);
1198         }
1199
1200         if (curr_node)
1201                 btrfs_release_delayed_node(curr_node);
1202         btrfs_free_path(path);
1203         trans->block_rsv = block_rsv;
1204
1205         return ret;
1206 }
1207
1208 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1209                             struct btrfs_fs_info *fs_info)
1210 {
1211         return __btrfs_run_delayed_items(trans, fs_info, -1);
1212 }
1213
1214 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1215                                struct btrfs_fs_info *fs_info, int nr)
1216 {
1217         return __btrfs_run_delayed_items(trans, fs_info, nr);
1218 }
1219
1220 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1221                                      struct btrfs_inode *inode)
1222 {
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 (!delayed_node->count) {
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         path = btrfs_alloc_path();
1240         if (!path) {
1241                 btrfs_release_delayed_node(delayed_node);
1242                 return -ENOMEM;
1243         }
1244         path->leave_spinning = 1;
1245
1246         block_rsv = trans->block_rsv;
1247         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1248
1249         ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1250
1251         btrfs_release_delayed_node(delayed_node);
1252         btrfs_free_path(path);
1253         trans->block_rsv = block_rsv;
1254
1255         return ret;
1256 }
1257
1258 int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
1259 {
1260         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1261         struct btrfs_trans_handle *trans;
1262         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1263         struct btrfs_path *path;
1264         struct btrfs_block_rsv *block_rsv;
1265         int ret;
1266
1267         if (!delayed_node)
1268                 return 0;
1269
1270         mutex_lock(&delayed_node->mutex);
1271         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1272                 mutex_unlock(&delayed_node->mutex);
1273                 btrfs_release_delayed_node(delayed_node);
1274                 return 0;
1275         }
1276         mutex_unlock(&delayed_node->mutex);
1277
1278         trans = btrfs_join_transaction(delayed_node->root);
1279         if (IS_ERR(trans)) {
1280                 ret = PTR_ERR(trans);
1281                 goto out;
1282         }
1283
1284         path = btrfs_alloc_path();
1285         if (!path) {
1286                 ret = -ENOMEM;
1287                 goto trans_out;
1288         }
1289         path->leave_spinning = 1;
1290
1291         block_rsv = trans->block_rsv;
1292         trans->block_rsv = &fs_info->delayed_block_rsv;
1293
1294         mutex_lock(&delayed_node->mutex);
1295         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1296                 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1297                                                    path, delayed_node);
1298         else
1299                 ret = 0;
1300         mutex_unlock(&delayed_node->mutex);
1301
1302         btrfs_free_path(path);
1303         trans->block_rsv = block_rsv;
1304 trans_out:
1305         btrfs_end_transaction(trans);
1306         btrfs_btree_balance_dirty(fs_info);
1307 out:
1308         btrfs_release_delayed_node(delayed_node);
1309
1310         return ret;
1311 }
1312
1313 void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1314 {
1315         struct btrfs_delayed_node *delayed_node;
1316
1317         delayed_node = READ_ONCE(inode->delayed_node);
1318         if (!delayed_node)
1319                 return;
1320
1321         inode->delayed_node = NULL;
1322         btrfs_release_delayed_node(delayed_node);
1323 }
1324
1325 struct btrfs_async_delayed_work {
1326         struct btrfs_delayed_root *delayed_root;
1327         int nr;
1328         struct btrfs_work work;
1329 };
1330
1331 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1332 {
1333         struct btrfs_async_delayed_work *async_work;
1334         struct btrfs_delayed_root *delayed_root;
1335         struct btrfs_trans_handle *trans;
1336         struct btrfs_path *path;
1337         struct btrfs_delayed_node *delayed_node = NULL;
1338         struct btrfs_root *root;
1339         struct btrfs_block_rsv *block_rsv;
1340         int total_done = 0;
1341
1342         async_work = container_of(work, struct btrfs_async_delayed_work, work);
1343         delayed_root = async_work->delayed_root;
1344
1345         path = btrfs_alloc_path();
1346         if (!path)
1347                 goto out;
1348
1349 again:
1350         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1351                 goto free_path;
1352
1353         delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1354         if (!delayed_node)
1355                 goto free_path;
1356
1357         path->leave_spinning = 1;
1358         root = delayed_node->root;
1359
1360         trans = btrfs_join_transaction(root);
1361         if (IS_ERR(trans))
1362                 goto release_path;
1363
1364         block_rsv = trans->block_rsv;
1365         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1366
1367         __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1368
1369         trans->block_rsv = block_rsv;
1370         btrfs_end_transaction(trans);
1371         btrfs_btree_balance_dirty_nodelay(root->fs_info);
1372
1373 release_path:
1374         btrfs_release_path(path);
1375         total_done++;
1376
1377         btrfs_release_prepared_delayed_node(delayed_node);
1378         if ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK) ||
1379             total_done < async_work->nr)
1380                 goto again;
1381
1382 free_path:
1383         btrfs_free_path(path);
1384 out:
1385         wake_up(&delayed_root->wait);
1386         kfree(async_work);
1387 }
1388
1389
1390 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1391                                      struct btrfs_fs_info *fs_info, int nr)
1392 {
1393         struct btrfs_async_delayed_work *async_work;
1394
1395         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND ||
1396             btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1397                 return 0;
1398
1399         async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1400         if (!async_work)
1401                 return -ENOMEM;
1402
1403         async_work->delayed_root = delayed_root;
1404         btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1405                         btrfs_async_run_delayed_root, NULL, NULL);
1406         async_work->nr = nr;
1407
1408         btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1409         return 0;
1410 }
1411
1412 void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1413 {
1414         WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
1415 }
1416
1417 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1418 {
1419         int val = atomic_read(&delayed_root->items_seq);
1420
1421         if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1422                 return 1;
1423
1424         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1425                 return 1;
1426
1427         return 0;
1428 }
1429
1430 void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1431 {
1432         struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1433
1434         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1435                 return;
1436
1437         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1438                 int seq;
1439                 int ret;
1440
1441                 seq = atomic_read(&delayed_root->items_seq);
1442
1443                 ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1444                 if (ret)
1445                         return;
1446
1447                 wait_event_interruptible(delayed_root->wait,
1448                                          could_end_wait(delayed_root, seq));
1449                 return;
1450         }
1451
1452         btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1453 }
1454
1455 /* Will return 0 or -ENOMEM */
1456 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1457                                    struct btrfs_fs_info *fs_info,
1458                                    const char *name, int name_len,
1459                                    struct btrfs_inode *dir,
1460                                    struct btrfs_disk_key *disk_key, u8 type,
1461                                    u64 index)
1462 {
1463         struct btrfs_delayed_node *delayed_node;
1464         struct btrfs_delayed_item *delayed_item;
1465         struct btrfs_dir_item *dir_item;
1466         int ret;
1467
1468         delayed_node = btrfs_get_or_create_delayed_node(dir);
1469         if (IS_ERR(delayed_node))
1470                 return PTR_ERR(delayed_node);
1471
1472         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1473         if (!delayed_item) {
1474                 ret = -ENOMEM;
1475                 goto release_node;
1476         }
1477
1478         delayed_item->key.objectid = btrfs_ino(dir);
1479         delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1480         delayed_item->key.offset = index;
1481
1482         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1483         dir_item->location = *disk_key;
1484         btrfs_set_stack_dir_transid(dir_item, trans->transid);
1485         btrfs_set_stack_dir_data_len(dir_item, 0);
1486         btrfs_set_stack_dir_name_len(dir_item, name_len);
1487         btrfs_set_stack_dir_type(dir_item, type);
1488         memcpy((char *)(dir_item + 1), name, name_len);
1489
1490         ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, delayed_item);
1491         /*
1492          * we have reserved enough space when we start a new transaction,
1493          * so reserving metadata failure is impossible
1494          */
1495         BUG_ON(ret);
1496
1497
1498         mutex_lock(&delayed_node->mutex);
1499         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1500         if (unlikely(ret)) {
1501                 btrfs_err(fs_info,
1502                           "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1503                           name_len, name, delayed_node->root->objectid,
1504                           delayed_node->inode_id, ret);
1505                 BUG();
1506         }
1507         mutex_unlock(&delayed_node->mutex);
1508
1509 release_node:
1510         btrfs_release_delayed_node(delayed_node);
1511         return ret;
1512 }
1513
1514 static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1515                                                struct btrfs_delayed_node *node,
1516                                                struct btrfs_key *key)
1517 {
1518         struct btrfs_delayed_item *item;
1519
1520         mutex_lock(&node->mutex);
1521         item = __btrfs_lookup_delayed_insertion_item(node, key);
1522         if (!item) {
1523                 mutex_unlock(&node->mutex);
1524                 return 1;
1525         }
1526
1527         btrfs_delayed_item_release_metadata(fs_info, item);
1528         btrfs_release_delayed_item(item);
1529         mutex_unlock(&node->mutex);
1530         return 0;
1531 }
1532
1533 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1534                                    struct btrfs_fs_info *fs_info,
1535                                    struct btrfs_inode *dir, u64 index)
1536 {
1537         struct btrfs_delayed_node *node;
1538         struct btrfs_delayed_item *item;
1539         struct btrfs_key item_key;
1540         int ret;
1541
1542         node = btrfs_get_or_create_delayed_node(dir);
1543         if (IS_ERR(node))
1544                 return PTR_ERR(node);
1545
1546         item_key.objectid = btrfs_ino(dir);
1547         item_key.type = BTRFS_DIR_INDEX_KEY;
1548         item_key.offset = index;
1549
1550         ret = btrfs_delete_delayed_insertion_item(fs_info, node, &item_key);
1551         if (!ret)
1552                 goto end;
1553
1554         item = btrfs_alloc_delayed_item(0);
1555         if (!item) {
1556                 ret = -ENOMEM;
1557                 goto end;
1558         }
1559
1560         item->key = item_key;
1561
1562         ret = btrfs_delayed_item_reserve_metadata(trans, fs_info, item);
1563         /*
1564          * we have reserved enough space when we start a new transaction,
1565          * so reserving metadata failure is impossible.
1566          */
1567         BUG_ON(ret);
1568
1569         mutex_lock(&node->mutex);
1570         ret = __btrfs_add_delayed_deletion_item(node, item);
1571         if (unlikely(ret)) {
1572                 btrfs_err(fs_info,
1573                           "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1574                           index, node->root->objectid, node->inode_id, ret);
1575                 BUG();
1576         }
1577         mutex_unlock(&node->mutex);
1578 end:
1579         btrfs_release_delayed_node(node);
1580         return ret;
1581 }
1582
1583 int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1584 {
1585         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1586
1587         if (!delayed_node)
1588                 return -ENOENT;
1589
1590         /*
1591          * Since we have held i_mutex of this directory, it is impossible that
1592          * a new directory index is added into the delayed node and index_cnt
1593          * is updated now. So we needn't lock the delayed node.
1594          */
1595         if (!delayed_node->index_cnt) {
1596                 btrfs_release_delayed_node(delayed_node);
1597                 return -EINVAL;
1598         }
1599
1600         inode->index_cnt = delayed_node->index_cnt;
1601         btrfs_release_delayed_node(delayed_node);
1602         return 0;
1603 }
1604
1605 bool btrfs_readdir_get_delayed_items(struct inode *inode,
1606                                      struct list_head *ins_list,
1607                                      struct list_head *del_list)
1608 {
1609         struct btrfs_delayed_node *delayed_node;
1610         struct btrfs_delayed_item *item;
1611
1612         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1613         if (!delayed_node)
1614                 return false;
1615
1616         /*
1617          * We can only do one readdir with delayed items at a time because of
1618          * item->readdir_list.
1619          */
1620         inode_unlock_shared(inode);
1621         inode_lock(inode);
1622
1623         mutex_lock(&delayed_node->mutex);
1624         item = __btrfs_first_delayed_insertion_item(delayed_node);
1625         while (item) {
1626                 refcount_inc(&item->refs);
1627                 list_add_tail(&item->readdir_list, ins_list);
1628                 item = __btrfs_next_delayed_item(item);
1629         }
1630
1631         item = __btrfs_first_delayed_deletion_item(delayed_node);
1632         while (item) {
1633                 refcount_inc(&item->refs);
1634                 list_add_tail(&item->readdir_list, del_list);
1635                 item = __btrfs_next_delayed_item(item);
1636         }
1637         mutex_unlock(&delayed_node->mutex);
1638         /*
1639          * This delayed node is still cached in the btrfs inode, so refs
1640          * must be > 1 now, and we needn't check it is going to be freed
1641          * or not.
1642          *
1643          * Besides that, this function is used to read dir, we do not
1644          * insert/delete delayed items in this period. So we also needn't
1645          * requeue or dequeue this delayed node.
1646          */
1647         refcount_dec(&delayed_node->refs);
1648
1649         return true;
1650 }
1651
1652 void btrfs_readdir_put_delayed_items(struct inode *inode,
1653                                      struct list_head *ins_list,
1654                                      struct list_head *del_list)
1655 {
1656         struct btrfs_delayed_item *curr, *next;
1657
1658         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1659                 list_del(&curr->readdir_list);
1660                 if (refcount_dec_and_test(&curr->refs))
1661                         kfree(curr);
1662         }
1663
1664         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1665                 list_del(&curr->readdir_list);
1666                 if (refcount_dec_and_test(&curr->refs))
1667                         kfree(curr);
1668         }
1669
1670         /*
1671          * The VFS is going to do up_read(), so we need to downgrade back to a
1672          * read lock.
1673          */
1674         downgrade_write(&inode->i_rwsem);
1675 }
1676
1677 int btrfs_should_delete_dir_index(struct list_head *del_list,
1678                                   u64 index)
1679 {
1680         struct btrfs_delayed_item *curr;
1681         int ret = 0;
1682
1683         list_for_each_entry(curr, del_list, readdir_list) {
1684                 if (curr->key.offset > index)
1685                         break;
1686                 if (curr->key.offset == index) {
1687                         ret = 1;
1688                         break;
1689                 }
1690         }
1691         return ret;
1692 }
1693
1694 /*
1695  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1696  *
1697  */
1698 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1699                                     struct list_head *ins_list)
1700 {
1701         struct btrfs_dir_item *di;
1702         struct btrfs_delayed_item *curr, *next;
1703         struct btrfs_key location;
1704         char *name;
1705         int name_len;
1706         int over = 0;
1707         unsigned char d_type;
1708
1709         if (list_empty(ins_list))
1710                 return 0;
1711
1712         /*
1713          * Changing the data of the delayed item is impossible. So
1714          * we needn't lock them. And we have held i_mutex of the
1715          * directory, nobody can delete any directory indexes now.
1716          */
1717         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1718                 list_del(&curr->readdir_list);
1719
1720                 if (curr->key.offset < ctx->pos) {
1721                         if (refcount_dec_and_test(&curr->refs))
1722                                 kfree(curr);
1723                         continue;
1724                 }
1725
1726                 ctx->pos = curr->key.offset;
1727
1728                 di = (struct btrfs_dir_item *)curr->data;
1729                 name = (char *)(di + 1);
1730                 name_len = btrfs_stack_dir_name_len(di);
1731
1732                 d_type = btrfs_filetype_table[di->type];
1733                 btrfs_disk_key_to_cpu(&location, &di->location);
1734
1735                 over = !dir_emit(ctx, name, name_len,
1736                                location.objectid, d_type);
1737
1738                 if (refcount_dec_and_test(&curr->refs))
1739                         kfree(curr);
1740
1741                 if (over)
1742                         return 1;
1743                 ctx->pos++;
1744         }
1745         return 0;
1746 }
1747
1748 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1749                                   struct btrfs_inode_item *inode_item,
1750                                   struct inode *inode)
1751 {
1752         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1753         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1754         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1755         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1756         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1757         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1758         btrfs_set_stack_inode_generation(inode_item,
1759                                          BTRFS_I(inode)->generation);
1760         btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1761         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1762         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1763         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1764         btrfs_set_stack_inode_block_group(inode_item, 0);
1765
1766         btrfs_set_stack_timespec_sec(&inode_item->atime,
1767                                      inode->i_atime.tv_sec);
1768         btrfs_set_stack_timespec_nsec(&inode_item->atime,
1769                                       inode->i_atime.tv_nsec);
1770
1771         btrfs_set_stack_timespec_sec(&inode_item->mtime,
1772                                      inode->i_mtime.tv_sec);
1773         btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1774                                       inode->i_mtime.tv_nsec);
1775
1776         btrfs_set_stack_timespec_sec(&inode_item->ctime,
1777                                      inode->i_ctime.tv_sec);
1778         btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1779                                       inode->i_ctime.tv_nsec);
1780
1781         btrfs_set_stack_timespec_sec(&inode_item->otime,
1782                                      BTRFS_I(inode)->i_otime.tv_sec);
1783         btrfs_set_stack_timespec_nsec(&inode_item->otime,
1784                                      BTRFS_I(inode)->i_otime.tv_nsec);
1785 }
1786
1787 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1788 {
1789         struct btrfs_delayed_node *delayed_node;
1790         struct btrfs_inode_item *inode_item;
1791
1792         delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1793         if (!delayed_node)
1794                 return -ENOENT;
1795
1796         mutex_lock(&delayed_node->mutex);
1797         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1798                 mutex_unlock(&delayed_node->mutex);
1799                 btrfs_release_delayed_node(delayed_node);
1800                 return -ENOENT;
1801         }
1802
1803         inode_item = &delayed_node->inode_item;
1804
1805         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1806         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1807         btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1808         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1809         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1810         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1811         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1812         BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1813
1814         inode->i_version = btrfs_stack_inode_sequence(inode_item);
1815         inode->i_rdev = 0;
1816         *rdev = btrfs_stack_inode_rdev(inode_item);
1817         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1818
1819         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1820         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1821
1822         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1823         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1824
1825         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1826         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1827
1828         BTRFS_I(inode)->i_otime.tv_sec =
1829                 btrfs_stack_timespec_sec(&inode_item->otime);
1830         BTRFS_I(inode)->i_otime.tv_nsec =
1831                 btrfs_stack_timespec_nsec(&inode_item->otime);
1832
1833         inode->i_generation = BTRFS_I(inode)->generation;
1834         BTRFS_I(inode)->index_cnt = (u64)-1;
1835
1836         mutex_unlock(&delayed_node->mutex);
1837         btrfs_release_delayed_node(delayed_node);
1838         return 0;
1839 }
1840
1841 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1842                                struct btrfs_root *root, struct inode *inode)
1843 {
1844         struct btrfs_delayed_node *delayed_node;
1845         int ret = 0;
1846
1847         delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1848         if (IS_ERR(delayed_node))
1849                 return PTR_ERR(delayed_node);
1850
1851         mutex_lock(&delayed_node->mutex);
1852         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1853                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1854                 goto release_node;
1855         }
1856
1857         ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1858                                                    delayed_node);
1859         if (ret)
1860                 goto release_node;
1861
1862         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1863         set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1864         delayed_node->count++;
1865         atomic_inc(&root->fs_info->delayed_root->items);
1866 release_node:
1867         mutex_unlock(&delayed_node->mutex);
1868         btrfs_release_delayed_node(delayed_node);
1869         return ret;
1870 }
1871
1872 int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1873 {
1874         struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb);
1875         struct btrfs_delayed_node *delayed_node;
1876
1877         /*
1878          * we don't do delayed inode updates during log recovery because it
1879          * leads to enospc problems.  This means we also can't do
1880          * delayed inode refs
1881          */
1882         if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1883                 return -EAGAIN;
1884
1885         delayed_node = btrfs_get_or_create_delayed_node(inode);
1886         if (IS_ERR(delayed_node))
1887                 return PTR_ERR(delayed_node);
1888
1889         /*
1890          * We don't reserve space for inode ref deletion is because:
1891          * - We ONLY do async inode ref deletion for the inode who has only
1892          *   one link(i_nlink == 1), it means there is only one inode ref.
1893          *   And in most case, the inode ref and the inode item are in the
1894          *   same leaf, and we will deal with them at the same time.
1895          *   Since we are sure we will reserve the space for the inode item,
1896          *   it is unnecessary to reserve space for inode ref deletion.
1897          * - If the inode ref and the inode item are not in the same leaf,
1898          *   We also needn't worry about enospc problem, because we reserve
1899          *   much more space for the inode update than it needs.
1900          * - At the worst, we can steal some space from the global reservation.
1901          *   It is very rare.
1902          */
1903         mutex_lock(&delayed_node->mutex);
1904         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1905                 goto release_node;
1906
1907         set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1908         delayed_node->count++;
1909         atomic_inc(&fs_info->delayed_root->items);
1910 release_node:
1911         mutex_unlock(&delayed_node->mutex);
1912         btrfs_release_delayed_node(delayed_node);
1913         return 0;
1914 }
1915
1916 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1917 {
1918         struct btrfs_root *root = delayed_node->root;
1919         struct btrfs_fs_info *fs_info = root->fs_info;
1920         struct btrfs_delayed_item *curr_item, *prev_item;
1921
1922         mutex_lock(&delayed_node->mutex);
1923         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1924         while (curr_item) {
1925                 btrfs_delayed_item_release_metadata(fs_info, curr_item);
1926                 prev_item = curr_item;
1927                 curr_item = __btrfs_next_delayed_item(prev_item);
1928                 btrfs_release_delayed_item(prev_item);
1929         }
1930
1931         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1932         while (curr_item) {
1933                 btrfs_delayed_item_release_metadata(fs_info, curr_item);
1934                 prev_item = curr_item;
1935                 curr_item = __btrfs_next_delayed_item(prev_item);
1936                 btrfs_release_delayed_item(prev_item);
1937         }
1938
1939         if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1940                 btrfs_release_delayed_iref(delayed_node);
1941
1942         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1943                 btrfs_delayed_inode_release_metadata(fs_info, delayed_node);
1944                 btrfs_release_delayed_inode(delayed_node);
1945         }
1946         mutex_unlock(&delayed_node->mutex);
1947 }
1948
1949 void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1950 {
1951         struct btrfs_delayed_node *delayed_node;
1952
1953         delayed_node = btrfs_get_delayed_node(inode);
1954         if (!delayed_node)
1955                 return;
1956
1957         __btrfs_kill_delayed_node(delayed_node);
1958         btrfs_release_delayed_node(delayed_node);
1959 }
1960
1961 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1962 {
1963         u64 inode_id = 0;
1964         struct btrfs_delayed_node *delayed_nodes[8];
1965         int i, n;
1966
1967         while (1) {
1968                 spin_lock(&root->inode_lock);
1969                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1970                                            (void **)delayed_nodes, inode_id,
1971                                            ARRAY_SIZE(delayed_nodes));
1972                 if (!n) {
1973                         spin_unlock(&root->inode_lock);
1974                         break;
1975                 }
1976
1977                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1978
1979                 for (i = 0; i < n; i++)
1980                         refcount_inc(&delayed_nodes[i]->refs);
1981                 spin_unlock(&root->inode_lock);
1982
1983                 for (i = 0; i < n; i++) {
1984                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1985                         btrfs_release_delayed_node(delayed_nodes[i]);
1986                 }
1987         }
1988 }
1989
1990 void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1991 {
1992         struct btrfs_delayed_node *curr_node, *prev_node;
1993
1994         curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1995         while (curr_node) {
1996                 __btrfs_kill_delayed_node(curr_node);
1997
1998                 prev_node = curr_node;
1999                 curr_node = btrfs_next_delayed_node(curr_node);
2000                 btrfs_release_delayed_node(prev_node);
2001         }
2002 }
2003