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
  25 #define BTRFS_DELAYED_WRITEBACK         400
  26 #define BTRFS_DELAYED_BACKGROUND        100
  27
  28 static struct kmem_cache *delayed_node_cache;
  29
  30 int __init btrfs_delayed_inode_init(void)
  31 {
  32         delayed_node_cache = kmem_cache_create("delayed_node",
  33                                         sizeof(struct btrfs_delayed_node),
  34                                         0,
  35                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
  36                                         NULL);
  37         if (!delayed_node_cache)
  38                 return -ENOMEM;
  39         return 0;
  40 }
  41
  42 void btrfs_delayed_inode_exit(void)
  43 {
  44         if (delayed_node_cache)
  45                 kmem_cache_destroy(delayed_node_cache);
  46 }
  47
  48 static inline void btrfs_init_delayed_node(
  49                                 struct btrfs_delayed_node *delayed_node,
  50                                 struct btrfs_root *root, u64 inode_id)
  51 {
  52         delayed_node->root = root;
  53         delayed_node->inode_id = inode_id;
  54         atomic_set(&delayed_node->refs, 0);
  55         delayed_node->count = 0;
  56         delayed_node->in_list = 0;
  57         delayed_node->inode_dirty = 0;
  58         delayed_node->ins_root = RB_ROOT;
  59         delayed_node->del_root = RB_ROOT;
  60         mutex_init(&delayed_node->mutex);
  61         delayed_node->index_cnt = 0;
  62         INIT_LIST_HEAD(&delayed_node->n_list);
  63         INIT_LIST_HEAD(&delayed_node->p_list);
  64         delayed_node->bytes_reserved = 0;
  65 }
  66
  67 static inline int btrfs_is_continuous_delayed_item(
  68                                         struct btrfs_delayed_item *item1,
  69                                         struct btrfs_delayed_item *item2)
  70 {
  71         if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  72             item1->key.objectid == item2->key.objectid &&
  73             item1->key.type == item2->key.type &&
  74             item1->key.offset + 1 == item2->key.offset)
  75                 return 1;
  76         return 0;
  77 }
  78
  79 static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
  80                                                         struct btrfs_root *root)
  81 {
  82         return root->fs_info->delayed_root;
  83 }
  84
  85 static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
  86 {
  87         struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
  88         struct btrfs_root *root = btrfs_inode->root;
  89         u64 ino = btrfs_ino(inode);
  90         struct btrfs_delayed_node *node;
  91
  92         node = ACCESS_ONCE(btrfs_inode->delayed_node);
  93         if (node) {
  94                 atomic_inc(&node->refs);
  95                 return node;
  96         }
  97
  98         spin_lock(&root->inode_lock);
  99         node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
 100         if (node) {
 101                 if (btrfs_inode->delayed_node) {
 102                         atomic_inc(&node->refs);        /* can be accessed */
 103                         BUG_ON(btrfs_inode->delayed_node != node);
 104                         spin_unlock(&root->inode_lock);
 105                         return node;
 106                 }
 107                 btrfs_inode->delayed_node = node;
 108                 atomic_inc(&node->refs);        /* can be accessed */
 109                 atomic_inc(&node->refs);        /* cached in the inode */
 110                 spin_unlock(&root->inode_lock);
 111                 return node;
 112         }
 113         spin_unlock(&root->inode_lock);
 114
 115         return NULL;
 116 }
 117
 118 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 119                                                         struct inode *inode)
 120 {
 121         struct btrfs_delayed_node *node;
 122         struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
 123         struct btrfs_root *root = btrfs_inode->root;
 124         u64 ino = btrfs_ino(inode);
 125         int ret;
 126
 127 again:
 128         node = btrfs_get_delayed_node(inode);
 129         if (node)
 130                 return node;
 131
 132         node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
 133         if (!node)
 134                 return ERR_PTR(-ENOMEM);
 135         btrfs_init_delayed_node(node, root, ino);
 136
 137         atomic_inc(&node->refs);        /* cached in the btrfs inode */
 138         atomic_inc(&node->refs);        /* can be accessed */
 139
 140         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 141         if (ret) {
 142                 kmem_cache_free(delayed_node_cache, node);
 143                 return ERR_PTR(ret);
 144         }
 145
 146         spin_lock(&root->inode_lock);
 147         ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 148         if (ret == -EEXIST) {
 149                 kmem_cache_free(delayed_node_cache, node);
 150                 spin_unlock(&root->inode_lock);
 151                 radix_tree_preload_end();
 152                 goto again;
 153         }
 154         btrfs_inode->delayed_node = node;
 155         spin_unlock(&root->inode_lock);
 156         radix_tree_preload_end();
 157
 158         return node;
 159 }
 160
 161 /*
 162  * Call it when holding delayed_node->mutex
 163  *
 164  * If mod = 1, add this node into the prepared list.
 165  */
 166 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 167                                      struct btrfs_delayed_node *node,
 168                                      int mod)
 169 {
 170         spin_lock(&root->lock);
 171         if (node->in_list) {
 172                 if (!list_empty(&node->p_list))
 173                         list_move_tail(&node->p_list, &root->prepare_list);
 174                 else if (mod)
 175                         list_add_tail(&node->p_list, &root->prepare_list);
 176         } else {
 177                 list_add_tail(&node->n_list, &root->node_list);
 178                 list_add_tail(&node->p_list, &root->prepare_list);
 179                 atomic_inc(&node->refs);        /* inserted into list */
 180                 root->nodes++;
 181                 node->in_list = 1;
 182         }
 183         spin_unlock(&root->lock);
 184 }
 185
 186 /* Call it when holding delayed_node->mutex */
 187 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 188                                        struct btrfs_delayed_node *node)
 189 {
 190         spin_lock(&root->lock);
 191         if (node->in_list) {
 192                 root->nodes--;
 193                 atomic_dec(&node->refs);        /* not in the list */
 194                 list_del_init(&node->n_list);
 195                 if (!list_empty(&node->p_list))
 196                         list_del_init(&node->p_list);
 197                 node->in_list = 0;
 198         }
 199         spin_unlock(&root->lock);
 200 }
 201
 202 struct btrfs_delayed_node *btrfs_first_delayed_node(
 203                         struct btrfs_delayed_root *delayed_root)
 204 {
 205         struct list_head *p;
 206         struct btrfs_delayed_node *node = NULL;
 207
 208         spin_lock(&delayed_root->lock);
 209         if (list_empty(&delayed_root->node_list))
 210                 goto out;
 211
 212         p = delayed_root->node_list.next;
 213         node = list_entry(p, struct btrfs_delayed_node, n_list);
 214         atomic_inc(&node->refs);
 215 out:
 216         spin_unlock(&delayed_root->lock);
 217
 218         return node;
 219 }
 220
 221 struct btrfs_delayed_node *btrfs_next_delayed_node(
 222                                                 struct btrfs_delayed_node *node)
 223 {
 224         struct btrfs_delayed_root *delayed_root;
 225         struct list_head *p;
 226         struct btrfs_delayed_node *next = NULL;
 227
 228         delayed_root = node->root->fs_info->delayed_root;
 229         spin_lock(&delayed_root->lock);
 230         if (!node->in_list) {   /* not in the list */
 231                 if (list_empty(&delayed_root->node_list))
 232                         goto out;
 233                 p = delayed_root->node_list.next;
 234         } else if (list_is_last(&node->n_list, &delayed_root->node_list))
 235                 goto out;
 236         else
 237                 p = node->n_list.next;
 238
 239         next = list_entry(p, struct btrfs_delayed_node, n_list);
 240         atomic_inc(&next->refs);
 241 out:
 242         spin_unlock(&delayed_root->lock);
 243
 244         return next;
 245 }
 246
 247 static void __btrfs_release_delayed_node(
 248                                 struct btrfs_delayed_node *delayed_node,
 249                                 int mod)
 250 {
 251         struct btrfs_delayed_root *delayed_root;
 252
 253         if (!delayed_node)
 254                 return;
 255
 256         delayed_root = delayed_node->root->fs_info->delayed_root;
 257
 258         mutex_lock(&delayed_node->mutex);
 259         if (delayed_node->count)
 260                 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
 261         else
 262                 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 263         mutex_unlock(&delayed_node->mutex);
 264
 265         if (atomic_dec_and_test(&delayed_node->refs)) {
 266                 struct btrfs_root *root = delayed_node->root;
 267                 spin_lock(&root->inode_lock);
 268                 if (atomic_read(&delayed_node->refs) == 0) {
 269                         radix_tree_delete(&root->delayed_nodes_tree,
 270                                           delayed_node->inode_id);
 271                         kmem_cache_free(delayed_node_cache, delayed_node);
 272                 }
 273                 spin_unlock(&root->inode_lock);
 274         }
 275 }
 276
 277 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
 278 {
 279         __btrfs_release_delayed_node(node, 0);
 280 }
 281
 282 struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
 283                                         struct btrfs_delayed_root *delayed_root)
 284 {
 285         struct list_head *p;
 286         struct btrfs_delayed_node *node = NULL;
 287
 288         spin_lock(&delayed_root->lock);
 289         if (list_empty(&delayed_root->prepare_list))
 290                 goto out;
 291
 292         p = delayed_root->prepare_list.next;
 293         list_del_init(p);
 294         node = list_entry(p, struct btrfs_delayed_node, p_list);
 295         atomic_inc(&node->refs);
 296 out:
 297         spin_unlock(&delayed_root->lock);
 298
 299         return node;
 300 }
 301
 302 static inline void btrfs_release_prepared_delayed_node(
 303                                         struct btrfs_delayed_node *node)
 304 {
 305         __btrfs_release_delayed_node(node, 1);
 306 }
 307
 308 struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
 309 {
 310         struct btrfs_delayed_item *item;
 311         item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
 312         if (item) {
 313                 item->data_len = data_len;
 314                 item->ins_or_del = 0;
 315                 item->bytes_reserved = 0;
 316                 item->delayed_node = NULL;
 317                 atomic_set(&item->refs, 1);
 318         }
 319         return item;
 320 }
 321
 322 /*
 323  * __btrfs_lookup_delayed_item - look up the delayed item by key
 324  * @delayed_node: pointer to the delayed node
 325  * @key:          the key to look up
 326  * @prev:         used to store the prev item if the right item isn't found
 327  * @next:         used to store the next item if the right item isn't found
 328  *
 329  * Note: if we don't find the right item, we will return the prev item and
 330  * the next item.
 331  */
 332 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 333                                 struct rb_root *root,
 334                                 struct btrfs_key *key,
 335                                 struct btrfs_delayed_item **prev,
 336                                 struct btrfs_delayed_item **next)
 337 {
 338         struct rb_node *node, *prev_node = NULL;
 339         struct btrfs_delayed_item *delayed_item = NULL;
 340         int ret = 0;
 341
 342         node = root->rb_node;
 343
 344         while (node) {
 345                 delayed_item = rb_entry(node, struct btrfs_delayed_item,
 346                                         rb_node);
 347                 prev_node = node;
 348                 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
 349                 if (ret < 0)
 350                         node = node->rb_right;
 351                 else if (ret > 0)
 352                         node = node->rb_left;
 353                 else
 354                         return delayed_item;
 355         }
 356
 357         if (prev) {
 358                 if (!prev_node)
 359                         *prev = NULL;
 360                 else if (ret < 0)
 361                         *prev = delayed_item;
 362                 else if ((node = rb_prev(prev_node)) != NULL) {
 363                         *prev = rb_entry(node, struct btrfs_delayed_item,
 364                                          rb_node);
 365                 } else
 366                         *prev = NULL;
 367         }
 368
 369         if (next) {
 370                 if (!prev_node)
 371                         *next = NULL;
 372                 else if (ret > 0)
 373                         *next = delayed_item;
 374                 else if ((node = rb_next(prev_node)) != NULL) {
 375                         *next = rb_entry(node, struct btrfs_delayed_item,
 376                                          rb_node);
 377                 } else
 378                         *next = NULL;
 379         }
 380         return NULL;
 381 }
 382
 383 struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
 384                                         struct btrfs_delayed_node *delayed_node,
 385                                         struct btrfs_key *key)
 386 {
 387         struct btrfs_delayed_item *item;
 388
 389         item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 390                                            NULL, NULL);
 391         return item;
 392 }
 393
 394 struct btrfs_delayed_item *__btrfs_lookup_delayed_deletion_item(
 395                                         struct btrfs_delayed_node *delayed_node,
 396                                         struct btrfs_key *key)
 397 {
 398         struct btrfs_delayed_item *item;
 399
 400         item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
 401                                            NULL, NULL);
 402         return item;
 403 }
 404
 405 struct btrfs_delayed_item *__btrfs_search_delayed_insertion_item(
 406                                         struct btrfs_delayed_node *delayed_node,
 407                                         struct btrfs_key *key)
 408 {
 409         struct btrfs_delayed_item *item, *next;
 410
 411         item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 412                                            NULL, &next);
 413         if (!item)
 414                 item = next;
 415
 416         return item;
 417 }
 418
 419 struct btrfs_delayed_item *__btrfs_search_delayed_deletion_item(
 420                                         struct btrfs_delayed_node *delayed_node,
 421                                         struct btrfs_key *key)
 422 {
 423         struct btrfs_delayed_item *item, *next;
 424
 425         item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
 426                                            NULL, &next);
 427         if (!item)
 428                 item = next;
 429
 430         return item;
 431 }
 432
 433 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
 434                                     struct btrfs_delayed_item *ins,
 435                                     int action)
 436 {
 437         struct rb_node **p, *node;
 438         struct rb_node *parent_node = NULL;
 439         struct rb_root *root;
 440         struct btrfs_delayed_item *item;
 441         int cmp;
 442
 443         if (action == BTRFS_DELAYED_INSERTION_ITEM)
 444                 root = &delayed_node->ins_root;
 445         else if (action == BTRFS_DELAYED_DELETION_ITEM)
 446                 root = &delayed_node->del_root;
 447         else
 448                 BUG();
 449         p = &root->rb_node;
 450         node = &ins->rb_node;
 451
 452         while (*p) {
 453                 parent_node = *p;
 454                 item = rb_entry(parent_node, struct btrfs_delayed_item,
 455                                  rb_node);
 456
 457                 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
 458                 if (cmp < 0)
 459                         p = &(*p)->rb_right;
 460                 else if (cmp > 0)
 461                         p = &(*p)->rb_left;
 462                 else
 463                         return -EEXIST;
 464         }
 465
 466         rb_link_node(node, parent_node, p);
 467         rb_insert_color(node, root);
 468         ins->delayed_node = delayed_node;
 469         ins->ins_or_del = action;
 470
 471         if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
 472             action == BTRFS_DELAYED_INSERTION_ITEM &&
 473             ins->key.offset >= delayed_node->index_cnt)
 474                         delayed_node->index_cnt = ins->key.offset + 1;
 475
 476         delayed_node->count++;
 477         atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
 478         return 0;
 479 }
 480
 481 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
 482                                               struct btrfs_delayed_item *item)
 483 {
 484         return __btrfs_add_delayed_item(node, item,
 485                                         BTRFS_DELAYED_INSERTION_ITEM);
 486 }
 487
 488 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
 489                                              struct btrfs_delayed_item *item)
 490 {
 491         return __btrfs_add_delayed_item(node, item,
 492                                         BTRFS_DELAYED_DELETION_ITEM);
 493 }
 494
 495 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 496 {
 497         struct rb_root *root;
 498         struct btrfs_delayed_root *delayed_root;
 499
 500         delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
 501
 502         BUG_ON(!delayed_root);
 503         BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
 504                delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
 505
 506         if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
 507                 root = &delayed_item->delayed_node->ins_root;
 508         else
 509                 root = &delayed_item->delayed_node->del_root;
 510
 511         rb_erase(&delayed_item->rb_node, root);
 512         delayed_item->delayed_node->count--;
 513         atomic_dec(&delayed_root->items);
 514         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND &&
 515             waitqueue_active(&delayed_root->wait))
 516                 wake_up(&delayed_root->wait);
 517 }
 518
 519 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 520 {
 521         if (item) {
 522                 __btrfs_remove_delayed_item(item);
 523                 if (atomic_dec_and_test(&item->refs))
 524                         kfree(item);
 525         }
 526 }
 527
 528 struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 529                                         struct btrfs_delayed_node *delayed_node)
 530 {
 531         struct rb_node *p;
 532         struct btrfs_delayed_item *item = NULL;
 533
 534         p = rb_first(&delayed_node->ins_root);
 535         if (p)
 536                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 537
 538         return item;
 539 }
 540
 541 struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 542                                         struct btrfs_delayed_node *delayed_node)
 543 {
 544         struct rb_node *p;
 545         struct btrfs_delayed_item *item = NULL;
 546
 547         p = rb_first(&delayed_node->del_root);
 548         if (p)
 549                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 550
 551         return item;
 552 }
 553
 554 struct btrfs_delayed_item *__btrfs_next_delayed_item(
 555                                                 struct btrfs_delayed_item *item)
 556 {
 557         struct rb_node *p;
 558         struct btrfs_delayed_item *next = NULL;
 559
 560         p = rb_next(&item->rb_node);
 561         if (p)
 562                 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
 563
 564         return next;
 565 }
 566
 567 static inline struct btrfs_root *btrfs_get_fs_root(struct btrfs_root *root,
 568                                                    u64 root_id)
 569 {
 570         struct btrfs_key root_key;
 571
 572         if (root->objectid == root_id)
 573                 return root;
 574
 575         root_key.objectid = root_id;
 576         root_key.type = BTRFS_ROOT_ITEM_KEY;
 577         root_key.offset = (u64)-1;
 578         return btrfs_read_fs_root_no_name(root->fs_info, &root_key);
 579 }
 580
 581 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 582                                                struct btrfs_root *root,
 583                                                struct btrfs_delayed_item *item)
 584 {
 585         struct btrfs_block_rsv *src_rsv;
 586         struct btrfs_block_rsv *dst_rsv;
 587         u64 num_bytes;
 588         int ret;
 589
 590         if (!trans->bytes_reserved)
 591                 return 0;
 592
 593         src_rsv = trans->block_rsv;
 594         dst_rsv = &root->fs_info->global_block_rsv;
 595
 596         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 597         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 598         if (!ret)
 599                 item->bytes_reserved = num_bytes;
 600
 601         return ret;
 602 }
 603
 604 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 605                                                 struct btrfs_delayed_item *item)
 606 {
 607         struct btrfs_block_rsv *rsv;
 608
 609         if (!item->bytes_reserved)
 610                 return;
 611
 612         rsv = &root->fs_info->global_block_rsv;
 613         btrfs_block_rsv_release(root, rsv,
 614                                 item->bytes_reserved);
 615 }
 616
 617 static int btrfs_delayed_inode_reserve_metadata(
 618                                         struct btrfs_trans_handle *trans,
 619                                         struct btrfs_root *root,
 620                                         struct btrfs_delayed_node *node)
 621 {
 622         struct btrfs_block_rsv *src_rsv;
 623         struct btrfs_block_rsv *dst_rsv;
 624         u64 num_bytes;
 625         int ret;
 626
 627         if (!trans->bytes_reserved)
 628                 return 0;
 629
 630         src_rsv = trans->block_rsv;
 631         dst_rsv = &root->fs_info->global_block_rsv;
 632
 633         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 634         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 635         if (!ret)
 636                 node->bytes_reserved = num_bytes;
 637
 638         return ret;
 639 }
 640
 641 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
 642                                                 struct btrfs_delayed_node *node)
 643 {
 644         struct btrfs_block_rsv *rsv;
 645
 646         if (!node->bytes_reserved)
 647                 return;
 648
 649         rsv = &root->fs_info->global_block_rsv;
 650         btrfs_block_rsv_release(root, rsv,
 651                                 node->bytes_reserved);
 652         node->bytes_reserved = 0;
 653 }
 654
 655 /*
 656  * This helper will insert some continuous items into the same leaf according
 657  * to the free space of the leaf.
 658  */
 659 static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans,
 660                                 struct btrfs_root *root,
 661                                 struct btrfs_path *path,
 662                                 struct btrfs_delayed_item *item)
 663 {
 664         struct btrfs_delayed_item *curr, *next;
 665         int free_space;
 666         int total_data_size = 0, total_size = 0;
 667         struct extent_buffer *leaf;
 668         char *data_ptr;
 669         struct btrfs_key *keys;
 670         u32 *data_size;
 671         struct list_head head;
 672         int slot;
 673         int nitems;
 674         int i;
 675         int ret = 0;
 676
 677         BUG_ON(!path->nodes[0]);
 678
 679         leaf = path->nodes[0];
 680         free_space = btrfs_leaf_free_space(root, leaf);
 681         INIT_LIST_HEAD(&head);
 682
 683         next = item;
 684         nitems = 0;
 685
 686         /*
 687          * count the number of the continuous items that we can insert in batch
 688          */
 689         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 690                free_space) {
 691                 total_data_size += next->data_len;
 692                 total_size += next->data_len + sizeof(struct btrfs_item);
 693                 list_add_tail(&next->tree_list, &head);
 694                 nitems++;
 695
 696                 curr = next;
 697                 next = __btrfs_next_delayed_item(curr);
 698                 if (!next)
 699                         break;
 700
 701                 if (!btrfs_is_continuous_delayed_item(curr, next))
 702                         break;
 703         }
 704
 705         if (!nitems) {
 706                 ret = 0;
 707                 goto out;
 708         }
 709
 710         /*
 711          * we need allocate some memory space, but it might cause the task
 712          * to sleep, so we set all locked nodes in the path to blocking locks
 713          * first.
 714          */
 715         btrfs_set_path_blocking(path);
 716
 717         keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS);
 718         if (!keys) {
 719                 ret = -ENOMEM;
 720                 goto out;
 721         }
 722
 723         data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS);
 724         if (!data_size) {
 725                 ret = -ENOMEM;
 726                 goto error;
 727         }
 728
 729         /* get keys of all the delayed items */
 730         i = 0;
 731         list_for_each_entry(next, &head, tree_list) {
 732                 keys[i] = next->key;
 733                 data_size[i] = next->data_len;
 734                 i++;
 735         }
 736
 737         /* reset all the locked nodes in the patch to spinning locks. */
 738         btrfs_clear_path_blocking(path, NULL);
 739
 740         /* insert the keys of the items */
 741         ret = setup_items_for_insert(trans, root, path, keys, data_size,
 742                                      total_data_size, total_size, nitems);
 743         if (ret)
 744                 goto error;
 745
 746         /* insert the dir index items */
 747         slot = path->slots[0];
 748         list_for_each_entry_safe(curr, next, &head, tree_list) {
 749                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 750                 write_extent_buffer(leaf, &curr->data,
 751                                     (unsigned long)data_ptr,
 752                                     curr->data_len);
 753                 slot++;
 754
 755                 btrfs_delayed_item_release_metadata(root, curr);
 756
 757                 list_del(&curr->tree_list);
 758                 btrfs_release_delayed_item(curr);
 759         }
 760
 761 error:
 762         kfree(data_size);
 763         kfree(keys);
 764 out:
 765         return ret;
 766 }
 767
 768 /*
 769  * This helper can just do simple insertion that needn't extend item for new
 770  * data, such as directory name index insertion, inode insertion.
 771  */
 772 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 773                                      struct btrfs_root *root,
 774                                      struct btrfs_path *path,
 775                                      struct btrfs_delayed_item *delayed_item)
 776 {
 777         struct extent_buffer *leaf;
 778         struct btrfs_item *item;
 779         char *ptr;
 780         int ret;
 781
 782         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 783                                       delayed_item->data_len);
 784         if (ret < 0 && ret != -EEXIST)
 785                 return ret;
 786
 787         leaf = path->nodes[0];
 788
 789         item = btrfs_item_nr(leaf, path->slots[0]);
 790         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 791
 792         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 793                             delayed_item->data_len);
 794         btrfs_mark_buffer_dirty(leaf);
 795
 796         btrfs_delayed_item_release_metadata(root, delayed_item);
 797         return 0;
 798 }
 799
 800 /*
 801  * we insert an item first, then if there are some continuous items, we try
 802  * to insert those items into the same leaf.
 803  */
 804 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 805                                       struct btrfs_path *path,
 806                                       struct btrfs_root *root,
 807                                       struct btrfs_delayed_node *node)
 808 {
 809         struct btrfs_delayed_item *curr, *prev;
 810         int ret = 0;
 811
 812 do_again:
 813         mutex_lock(&node->mutex);
 814         curr = __btrfs_first_delayed_insertion_item(node);
 815         if (!curr)
 816                 goto insert_end;
 817
 818         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 819         if (ret < 0) {
 820                 btrfs_release_path(path);
 821                 goto insert_end;
 822         }
 823
 824         prev = curr;
 825         curr = __btrfs_next_delayed_item(prev);
 826         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 827                 /* insert the continuous items into the same leaf */
 828                 path->slots[0]++;
 829                 btrfs_batch_insert_items(trans, root, path, curr);
 830         }
 831         btrfs_release_delayed_item(prev);
 832         btrfs_mark_buffer_dirty(path->nodes[0]);
 833
 834         btrfs_release_path(path);
 835         mutex_unlock(&node->mutex);
 836         goto do_again;
 837
 838 insert_end:
 839         mutex_unlock(&node->mutex);
 840         return ret;
 841 }
 842
 843 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 844                                     struct btrfs_root *root,
 845                                     struct btrfs_path *path,
 846                                     struct btrfs_delayed_item *item)
 847 {
 848         struct btrfs_delayed_item *curr, *next;
 849         struct extent_buffer *leaf;
 850         struct btrfs_key key;
 851         struct list_head head;
 852         int nitems, i, last_item;
 853         int ret = 0;
 854
 855         BUG_ON(!path->nodes[0]);
 856
 857         leaf = path->nodes[0];
 858
 859         i = path->slots[0];
 860         last_item = btrfs_header_nritems(leaf) - 1;
 861         if (i > last_item)
 862                 return -ENOENT; /* FIXME: Is errno suitable? */
 863
 864         next = item;
 865         INIT_LIST_HEAD(&head);
 866         btrfs_item_key_to_cpu(leaf, &key, i);
 867         nitems = 0;
 868         /*
 869          * count the number of the dir index items that we can delete in batch
 870          */
 871         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 872                 list_add_tail(&next->tree_list, &head);
 873                 nitems++;
 874
 875                 curr = next;
 876                 next = __btrfs_next_delayed_item(curr);
 877                 if (!next)
 878                         break;
 879
 880                 if (!btrfs_is_continuous_delayed_item(curr, next))
 881                         break;
 882
 883                 i++;
 884                 if (i > last_item)
 885                         break;
 886                 btrfs_item_key_to_cpu(leaf, &key, i);
 887         }
 888
 889         if (!nitems)
 890                 return 0;
 891
 892         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 893         if (ret)
 894                 goto out;
 895
 896         list_for_each_entry_safe(curr, next, &head, tree_list) {
 897                 btrfs_delayed_item_release_metadata(root, curr);
 898                 list_del(&curr->tree_list);
 899                 btrfs_release_delayed_item(curr);
 900         }
 901
 902 out:
 903         return ret;
 904 }
 905
 906 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 907                                       struct btrfs_path *path,
 908                                       struct btrfs_root *root,
 909                                       struct btrfs_delayed_node *node)
 910 {
 911         struct btrfs_delayed_item *curr, *prev;
 912         int ret = 0;
 913
 914 do_again:
 915         mutex_lock(&node->mutex);
 916         curr = __btrfs_first_delayed_deletion_item(node);
 917         if (!curr)
 918                 goto delete_fail;
 919
 920         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 921         if (ret < 0)
 922                 goto delete_fail;
 923         else if (ret > 0) {
 924                 /*
 925                  * can't find the item which the node points to, so this node
 926                  * is invalid, just drop it.
 927                  */
 928                 prev = curr;
 929                 curr = __btrfs_next_delayed_item(prev);
 930                 btrfs_release_delayed_item(prev);
 931                 ret = 0;
 932                 btrfs_release_path(path);
 933                 if (curr)
 934                         goto do_again;
 935                 else
 936                         goto delete_fail;
 937         }
 938
 939         btrfs_batch_delete_items(trans, root, path, curr);
 940         btrfs_release_path(path);
 941         mutex_unlock(&node->mutex);
 942         goto do_again;
 943
 944 delete_fail:
 945         btrfs_release_path(path);
 946         mutex_unlock(&node->mutex);
 947         return ret;
 948 }
 949
 950 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 951 {
 952         struct btrfs_delayed_root *delayed_root;
 953
 954         if (delayed_node && delayed_node->inode_dirty) {
 955                 BUG_ON(!delayed_node->root);
 956                 delayed_node->inode_dirty = 0;
 957                 delayed_node->count--;
 958
 959                 delayed_root = delayed_node->root->fs_info->delayed_root;
 960                 atomic_dec(&delayed_root->items);
 961                 if (atomic_read(&delayed_root->items) <
 962                     BTRFS_DELAYED_BACKGROUND &&
 963                     waitqueue_active(&delayed_root->wait))
 964                         wake_up(&delayed_root->wait);
 965         }
 966 }
 967
 968 static int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
 969                                       struct btrfs_root *root,
 970                                       struct btrfs_path *path,
 971                                       struct btrfs_delayed_node *node)
 972 {
 973         struct btrfs_key key;
 974         struct btrfs_inode_item *inode_item;
 975         struct extent_buffer *leaf;
 976         int ret;
 977
 978         mutex_lock(&node->mutex);
 979         if (!node->inode_dirty) {
 980                 mutex_unlock(&node->mutex);
 981                 return 0;
 982         }
 983
 984         key.objectid = node->inode_id;
 985         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 986         key.offset = 0;
 987         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
 988         if (ret > 0) {
 989                 btrfs_release_path(path);
 990                 mutex_unlock(&node->mutex);
 991                 return -ENOENT;
 992         } else if (ret < 0) {
 993                 mutex_unlock(&node->mutex);
 994                 return ret;
 995         }
 996
 997         btrfs_unlock_up_safe(path, 1);
 998         leaf = path->nodes[0];
 999         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1000                                     struct btrfs_inode_item);
1001         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1002                             sizeof(struct btrfs_inode_item));
1003         btrfs_mark_buffer_dirty(leaf);
1004         btrfs_release_path(path);
1005
1006         btrfs_delayed_inode_release_metadata(root, node);
1007         btrfs_release_delayed_inode(node);
1008         mutex_unlock(&node->mutex);
1009
1010         return 0;
1011 }
1012
1013 /* Called when committing the transaction. */
1014 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1015                             struct btrfs_root *root)
1016 {
1017         struct btrfs_delayed_root *delayed_root;
1018         struct btrfs_delayed_node *curr_node, *prev_node;
1019         struct btrfs_path *path;
1020         struct btrfs_block_rsv *block_rsv;
1021         int ret = 0;
1022
1023         path = btrfs_alloc_path();
1024         if (!path)
1025                 return -ENOMEM;
1026         path->leave_spinning = 1;
1027
1028         block_rsv = trans->block_rsv;
1029         trans->block_rsv = &root->fs_info->global_block_rsv;
1030
1031         delayed_root = btrfs_get_delayed_root(root);
1032
1033         curr_node = btrfs_first_delayed_node(delayed_root);
1034         while (curr_node) {
1035                 root = curr_node->root;
1036                 ret = btrfs_insert_delayed_items(trans, path, root,
1037                                                  curr_node);
1038                 if (!ret)
1039                         ret = btrfs_delete_delayed_items(trans, path, root,
1040                                                          curr_node);
1041                 if (!ret)
1042                         ret = btrfs_update_delayed_inode(trans, root, path,
1043                                                          curr_node);
1044                 if (ret) {
1045                         btrfs_release_delayed_node(curr_node);
1046                         break;
1047                 }
1048
1049                 prev_node = curr_node;
1050                 curr_node = btrfs_next_delayed_node(curr_node);
1051                 btrfs_release_delayed_node(prev_node);
1052         }
1053
1054         btrfs_free_path(path);
1055         trans->block_rsv = block_rsv;
1056         return ret;
1057 }
1058
1059 static int __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1060                                               struct btrfs_delayed_node *node)
1061 {
1062         struct btrfs_path *path;
1063         struct btrfs_block_rsv *block_rsv;
1064         int ret;
1065
1066         path = btrfs_alloc_path();
1067         if (!path)
1068                 return -ENOMEM;
1069         path->leave_spinning = 1;
1070
1071         block_rsv = trans->block_rsv;
1072         trans->block_rsv = &node->root->fs_info->global_block_rsv;
1073
1074         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1075         if (!ret)
1076                 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1077         if (!ret)
1078                 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1079         btrfs_free_path(path);
1080
1081         trans->block_rsv = block_rsv;
1082         return ret;
1083 }
1084
1085 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1086                                      struct inode *inode)
1087 {
1088         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1089         int ret;
1090
1091         if (!delayed_node)
1092                 return 0;
1093
1094         mutex_lock(&delayed_node->mutex);
1095         if (!delayed_node->count) {
1096                 mutex_unlock(&delayed_node->mutex);
1097                 btrfs_release_delayed_node(delayed_node);
1098                 return 0;
1099         }
1100         mutex_unlock(&delayed_node->mutex);
1101
1102         ret = __btrfs_commit_inode_delayed_items(trans, delayed_node);
1103         btrfs_release_delayed_node(delayed_node);
1104         return ret;
1105 }
1106
1107 void btrfs_remove_delayed_node(struct inode *inode)
1108 {
1109         struct btrfs_delayed_node *delayed_node;
1110
1111         delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1112         if (!delayed_node)
1113                 return;
1114
1115         BTRFS_I(inode)->delayed_node = NULL;
1116         btrfs_release_delayed_node(delayed_node);
1117 }
1118
1119 struct btrfs_async_delayed_node {
1120         struct btrfs_root *root;
1121         struct btrfs_delayed_node *delayed_node;
1122         struct btrfs_work work;
1123 };
1124
1125 static void btrfs_async_run_delayed_node_done(struct btrfs_work *work)
1126 {
1127         struct btrfs_async_delayed_node *async_node;
1128         struct btrfs_trans_handle *trans;
1129         struct btrfs_path *path;
1130         struct btrfs_delayed_node *delayed_node = NULL;
1131         struct btrfs_root *root;
1132         struct btrfs_block_rsv *block_rsv;
1133         unsigned long nr = 0;
1134         int need_requeue = 0;
1135         int ret;
1136
1137         async_node = container_of(work, struct btrfs_async_delayed_node, work);
1138
1139         path = btrfs_alloc_path();
1140         if (!path)
1141                 goto out;
1142         path->leave_spinning = 1;
1143
1144         delayed_node = async_node->delayed_node;
1145         root = delayed_node->root;
1146
1147         trans = btrfs_join_transaction(root);
1148         if (IS_ERR(trans))
1149                 goto free_path;
1150
1151         block_rsv = trans->block_rsv;
1152         trans->block_rsv = &root->fs_info->global_block_rsv;
1153
1154         ret = btrfs_insert_delayed_items(trans, path, root, delayed_node);
1155         if (!ret)
1156                 ret = btrfs_delete_delayed_items(trans, path, root,
1157                                                  delayed_node);
1158
1159         if (!ret)
1160                 btrfs_update_delayed_inode(trans, root, path, delayed_node);
1161
1162         /*
1163          * Maybe new delayed items have been inserted, so we need requeue
1164          * the work. Besides that, we must dequeue the empty delayed nodes
1165          * to avoid the race between delayed items balance and the worker.
1166          * The race like this:
1167          *      Task1                           Worker thread
1168          *                                      count == 0, needn't requeue
1169          *                                        also needn't insert the
1170          *                                        delayed node into prepare
1171          *                                        list again.
1172          *      add lots of delayed items
1173          *      queue the delayed node
1174          *        already in the list,
1175          *        and not in the prepare
1176          *        list, it means the delayed
1177          *        node is being dealt with
1178          *        by the worker.
1179          *      do delayed items balance
1180          *        the delayed node is being
1181          *        dealt with by the worker
1182          *        now, just wait.
1183          *                                      the worker goto idle.
1184          * Task1 will sleep until the transaction is commited.
1185          */
1186         mutex_lock(&delayed_node->mutex);
1187         if (delayed_node->count)
1188                 need_requeue = 1;
1189         else
1190                 btrfs_dequeue_delayed_node(root->fs_info->delayed_root,
1191                                            delayed_node);
1192         mutex_unlock(&delayed_node->mutex);
1193
1194         nr = trans->blocks_used;
1195
1196         trans->block_rsv = block_rsv;
1197         btrfs_end_transaction_dmeta(trans, root);
1198         __btrfs_btree_balance_dirty(root, nr);
1199 free_path:
1200         btrfs_free_path(path);
1201 out:
1202         if (need_requeue)
1203                 btrfs_requeue_work(&async_node->work);
1204         else {
1205                 btrfs_release_prepared_delayed_node(delayed_node);
1206                 kfree(async_node);
1207         }
1208 }
1209
1210 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1211                                      struct btrfs_root *root, int all)
1212 {
1213         struct btrfs_async_delayed_node *async_node;
1214         struct btrfs_delayed_node *curr;
1215         int count = 0;
1216
1217 again:
1218         curr = btrfs_first_prepared_delayed_node(delayed_root);
1219         if (!curr)
1220                 return 0;
1221
1222         async_node = kmalloc(sizeof(*async_node), GFP_NOFS);
1223         if (!async_node) {
1224                 btrfs_release_prepared_delayed_node(curr);
1225                 return -ENOMEM;
1226         }
1227
1228         async_node->root = root;
1229         async_node->delayed_node = curr;
1230
1231         async_node->work.func = btrfs_async_run_delayed_node_done;
1232         async_node->work.flags = 0;
1233
1234         btrfs_queue_worker(&root->fs_info->delayed_workers, &async_node->work);
1235         count++;
1236
1237         if (all || count < 4)
1238                 goto again;
1239
1240         return 0;
1241 }
1242
1243 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1244 {
1245         struct btrfs_delayed_root *delayed_root;
1246         delayed_root = btrfs_get_delayed_root(root);
1247         WARN_ON(btrfs_first_delayed_node(delayed_root));
1248 }
1249
1250 void btrfs_balance_delayed_items(struct btrfs_root *root)
1251 {
1252         struct btrfs_delayed_root *delayed_root;
1253
1254         delayed_root = btrfs_get_delayed_root(root);
1255
1256         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1257                 return;
1258
1259         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1260                 int ret;
1261                 ret = btrfs_wq_run_delayed_node(delayed_root, root, 1);
1262                 if (ret)
1263                         return;
1264
1265                 wait_event_interruptible_timeout(
1266                                 delayed_root->wait,
1267                                 (atomic_read(&delayed_root->items) <
1268                                  BTRFS_DELAYED_BACKGROUND),
1269                                 HZ);
1270                 return;
1271         }
1272
1273         btrfs_wq_run_delayed_node(delayed_root, root, 0);
1274 }
1275
1276 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1277                                    struct btrfs_root *root, const char *name,
1278                                    int name_len, struct inode *dir,
1279                                    struct btrfs_disk_key *disk_key, u8 type,
1280                                    u64 index)
1281 {
1282         struct btrfs_delayed_node *delayed_node;
1283         struct btrfs_delayed_item *delayed_item;
1284         struct btrfs_dir_item *dir_item;
1285         int ret;
1286
1287         delayed_node = btrfs_get_or_create_delayed_node(dir);
1288         if (IS_ERR(delayed_node))
1289                 return PTR_ERR(delayed_node);
1290
1291         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1292         if (!delayed_item) {
1293                 ret = -ENOMEM;
1294                 goto release_node;
1295         }
1296
1297         ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1298         /*
1299          * we have reserved enough space when we start a new transaction,
1300          * so reserving metadata failure is impossible
1301          */
1302         BUG_ON(ret);
1303
1304         delayed_item->key.objectid = btrfs_ino(dir);
1305         btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1306         delayed_item->key.offset = index;
1307
1308         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1309         dir_item->location = *disk_key;
1310         dir_item->transid = cpu_to_le64(trans->transid);
1311         dir_item->data_len = 0;
1312         dir_item->name_len = cpu_to_le16(name_len);
1313         dir_item->type = type;
1314         memcpy((char *)(dir_item + 1), name, name_len);
1315
1316         mutex_lock(&delayed_node->mutex);
1317         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1318         if (unlikely(ret)) {
1319                 printk(KERN_ERR "err add delayed dir index item(name: %s) into "
1320                                 "the insertion tree of the delayed node"
1321                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1322                                 name,
1323                                 (unsigned long long)delayed_node->root->objectid,
1324                                 (unsigned long long)delayed_node->inode_id,
1325                                 ret);
1326                 BUG();
1327         }
1328         mutex_unlock(&delayed_node->mutex);
1329
1330 release_node:
1331         btrfs_release_delayed_node(delayed_node);
1332         return ret;
1333 }
1334
1335 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1336                                                struct btrfs_delayed_node *node,
1337                                                struct btrfs_key *key)
1338 {
1339         struct btrfs_delayed_item *item;
1340
1341         mutex_lock(&node->mutex);
1342         item = __btrfs_lookup_delayed_insertion_item(node, key);
1343         if (!item) {
1344                 mutex_unlock(&node->mutex);
1345                 return 1;
1346         }
1347
1348         btrfs_delayed_item_release_metadata(root, item);
1349         btrfs_release_delayed_item(item);
1350         mutex_unlock(&node->mutex);
1351         return 0;
1352 }
1353
1354 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1355                                    struct btrfs_root *root, struct inode *dir,
1356                                    u64 index)
1357 {
1358         struct btrfs_delayed_node *node;
1359         struct btrfs_delayed_item *item;
1360         struct btrfs_key item_key;
1361         int ret;
1362
1363         node = btrfs_get_or_create_delayed_node(dir);
1364         if (IS_ERR(node))
1365                 return PTR_ERR(node);
1366
1367         item_key.objectid = btrfs_ino(dir);
1368         btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1369         item_key.offset = index;
1370
1371         ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1372         if (!ret)
1373                 goto end;
1374
1375         item = btrfs_alloc_delayed_item(0);
1376         if (!item) {
1377                 ret = -ENOMEM;
1378                 goto end;
1379         }
1380
1381         item->key = item_key;
1382
1383         ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1384         /*
1385          * we have reserved enough space when we start a new transaction,
1386          * so reserving metadata failure is impossible.
1387          */
1388         BUG_ON(ret);
1389
1390         mutex_lock(&node->mutex);
1391         ret = __btrfs_add_delayed_deletion_item(node, item);
1392         if (unlikely(ret)) {
1393                 printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1394                                 "into the deletion tree of the delayed node"
1395                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1396                                 (unsigned long long)index,
1397                                 (unsigned long long)node->root->objectid,
1398                                 (unsigned long long)node->inode_id,
1399                                 ret);
1400                 BUG();
1401         }
1402         mutex_unlock(&node->mutex);
1403 end:
1404         btrfs_release_delayed_node(node);
1405         return ret;
1406 }
1407
1408 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1409 {
1410         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1411
1412         if (!delayed_node)
1413                 return -ENOENT;
1414
1415         /*
1416          * Since we have held i_mutex of this directory, it is impossible that
1417          * a new directory index is added into the delayed node and index_cnt
1418          * is updated now. So we needn't lock the delayed node.
1419          */
1420         if (!delayed_node->index_cnt) {
1421                 btrfs_release_delayed_node(delayed_node);
1422                 return -EINVAL;
1423         }
1424
1425         BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1426         btrfs_release_delayed_node(delayed_node);
1427         return 0;
1428 }
1429
1430 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1431                              struct list_head *del_list)
1432 {
1433         struct btrfs_delayed_node *delayed_node;
1434         struct btrfs_delayed_item *item;
1435
1436         delayed_node = btrfs_get_delayed_node(inode);
1437         if (!delayed_node)
1438                 return;
1439
1440         mutex_lock(&delayed_node->mutex);
1441         item = __btrfs_first_delayed_insertion_item(delayed_node);
1442         while (item) {
1443                 atomic_inc(&item->refs);
1444                 list_add_tail(&item->readdir_list, ins_list);
1445                 item = __btrfs_next_delayed_item(item);
1446         }
1447
1448         item = __btrfs_first_delayed_deletion_item(delayed_node);
1449         while (item) {
1450                 atomic_inc(&item->refs);
1451                 list_add_tail(&item->readdir_list, del_list);
1452                 item = __btrfs_next_delayed_item(item);
1453         }
1454         mutex_unlock(&delayed_node->mutex);
1455         /*
1456          * This delayed node is still cached in the btrfs inode, so refs
1457          * must be > 1 now, and we needn't check it is going to be freed
1458          * or not.
1459          *
1460          * Besides that, this function is used to read dir, we do not
1461          * insert/delete delayed items in this period. So we also needn't
1462          * requeue or dequeue this delayed node.
1463          */
1464         atomic_dec(&delayed_node->refs);
1465 }
1466
1467 void btrfs_put_delayed_items(struct list_head *ins_list,
1468                              struct list_head *del_list)
1469 {
1470         struct btrfs_delayed_item *curr, *next;
1471
1472         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1473                 list_del(&curr->readdir_list);
1474                 if (atomic_dec_and_test(&curr->refs))
1475                         kfree(curr);
1476         }
1477
1478         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1479                 list_del(&curr->readdir_list);
1480                 if (atomic_dec_and_test(&curr->refs))
1481                         kfree(curr);
1482         }
1483 }
1484
1485 int btrfs_should_delete_dir_index(struct list_head *del_list,
1486                                   u64 index)
1487 {
1488         struct btrfs_delayed_item *curr, *next;
1489         int ret;
1490
1491         if (list_empty(del_list))
1492                 return 0;
1493
1494         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1495                 if (curr->key.offset > index)
1496                         break;
1497
1498                 list_del(&curr->readdir_list);
1499                 ret = (curr->key.offset == index);
1500
1501                 if (atomic_dec_and_test(&curr->refs))
1502                         kfree(curr);
1503
1504                 if (ret)
1505                         return 1;
1506                 else
1507                         continue;
1508         }
1509         return 0;
1510 }
1511
1512 /*
1513  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1514  *
1515  */
1516 int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent,
1517                                     filldir_t filldir,
1518                                     struct list_head *ins_list)
1519 {
1520         struct btrfs_dir_item *di;
1521         struct btrfs_delayed_item *curr, *next;
1522         struct btrfs_key location;
1523         char *name;
1524         int name_len;
1525         int over = 0;
1526         unsigned char d_type;
1527
1528         if (list_empty(ins_list))
1529                 return 0;
1530
1531         /*
1532          * Changing the data of the delayed item is impossible. So
1533          * we needn't lock them. And we have held i_mutex of the
1534          * directory, nobody can delete any directory indexes now.
1535          */
1536         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1537                 list_del(&curr->readdir_list);
1538
1539                 if (curr->key.offset < filp->f_pos) {
1540                         if (atomic_dec_and_test(&curr->refs))
1541                                 kfree(curr);
1542                         continue;
1543                 }
1544
1545                 filp->f_pos = curr->key.offset;
1546
1547                 di = (struct btrfs_dir_item *)curr->data;
1548                 name = (char *)(di + 1);
1549                 name_len = le16_to_cpu(di->name_len);
1550
1551                 d_type = btrfs_filetype_table[di->type];
1552                 btrfs_disk_key_to_cpu(&location, &di->location);
1553
1554                 over = filldir(dirent, name, name_len, curr->key.offset,
1555                                location.objectid, d_type);
1556
1557                 if (atomic_dec_and_test(&curr->refs))
1558                         kfree(curr);
1559
1560                 if (over)
1561                         return 1;
1562         }
1563         return 0;
1564 }
1565
1566 BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item,
1567                          generation, 64);
1568 BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item,
1569                          sequence, 64);
1570 BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item,
1571                          transid, 64);
1572 BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64);
1573 BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item,
1574                          nbytes, 64);
1575 BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item,
1576                          block_group, 64);
1577 BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32);
1578 BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32);
1579 BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32);
1580 BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32);
1581 BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64);
1582 BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64);
1583
1584 BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64);
1585 BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32);
1586
1587 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1588                                   struct btrfs_inode_item *inode_item,
1589                                   struct inode *inode)
1590 {
1591         btrfs_set_stack_inode_uid(inode_item, inode->i_uid);
1592         btrfs_set_stack_inode_gid(inode_item, inode->i_gid);
1593         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1594         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1595         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1596         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1597         btrfs_set_stack_inode_generation(inode_item,
1598                                          BTRFS_I(inode)->generation);
1599         btrfs_set_stack_inode_sequence(inode_item, BTRFS_I(inode)->sequence);
1600         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1601         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1602         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1603         btrfs_set_stack_inode_block_group(inode_item, 0);
1604
1605         btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1606                                      inode->i_atime.tv_sec);
1607         btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1608                                       inode->i_atime.tv_nsec);
1609
1610         btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1611                                      inode->i_mtime.tv_sec);
1612         btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1613                                       inode->i_mtime.tv_nsec);
1614
1615         btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1616                                      inode->i_ctime.tv_sec);
1617         btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1618                                       inode->i_ctime.tv_nsec);
1619 }
1620
1621 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1622 {
1623         struct btrfs_delayed_node *delayed_node;
1624         struct btrfs_inode_item *inode_item;
1625         struct btrfs_timespec *tspec;
1626
1627         delayed_node = btrfs_get_delayed_node(inode);
1628         if (!delayed_node)
1629                 return -ENOENT;
1630
1631         mutex_lock(&delayed_node->mutex);
1632         if (!delayed_node->inode_dirty) {
1633                 mutex_unlock(&delayed_node->mutex);
1634                 btrfs_release_delayed_node(delayed_node);
1635                 return -ENOENT;
1636         }
1637
1638         inode_item = &delayed_node->inode_item;
1639
1640         inode->i_uid = btrfs_stack_inode_uid(inode_item);
1641         inode->i_gid = btrfs_stack_inode_gid(inode_item);
1642         btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1643         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1644         inode->i_nlink = btrfs_stack_inode_nlink(inode_item);
1645         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1646         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1647         BTRFS_I(inode)->sequence = btrfs_stack_inode_sequence(inode_item);
1648         inode->i_rdev = 0;
1649         *rdev = btrfs_stack_inode_rdev(inode_item);
1650         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1651
1652         tspec = btrfs_inode_atime(inode_item);
1653         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1654         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1655
1656         tspec = btrfs_inode_mtime(inode_item);
1657         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1658         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1659
1660         tspec = btrfs_inode_ctime(inode_item);
1661         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1662         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1663
1664         inode->i_generation = BTRFS_I(inode)->generation;
1665         BTRFS_I(inode)->index_cnt = (u64)-1;
1666
1667         mutex_unlock(&delayed_node->mutex);
1668         btrfs_release_delayed_node(delayed_node);
1669         return 0;
1670 }
1671
1672 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1673                                struct btrfs_root *root, struct inode *inode)
1674 {
1675         struct btrfs_delayed_node *delayed_node;
1676         int ret = 0;
1677
1678         delayed_node = btrfs_get_or_create_delayed_node(inode);
1679         if (IS_ERR(delayed_node))
1680                 return PTR_ERR(delayed_node);
1681
1682         mutex_lock(&delayed_node->mutex);
1683         if (delayed_node->inode_dirty) {
1684                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1685                 goto release_node;
1686         }
1687
1688         ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1689         /*
1690          * we must reserve enough space when we start a new transaction,
1691          * so reserving metadata failure is impossible
1692          */
1693         BUG_ON(ret);
1694
1695         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1696         delayed_node->inode_dirty = 1;
1697         delayed_node->count++;
1698         atomic_inc(&root->fs_info->delayed_root->items);
1699 release_node:
1700         mutex_unlock(&delayed_node->mutex);
1701         btrfs_release_delayed_node(delayed_node);
1702         return ret;
1703 }
1704
1705 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1706 {
1707         struct btrfs_root *root = delayed_node->root;
1708         struct btrfs_delayed_item *curr_item, *prev_item;
1709
1710         mutex_lock(&delayed_node->mutex);
1711         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1712         while (curr_item) {
1713                 btrfs_delayed_item_release_metadata(root, curr_item);
1714                 prev_item = curr_item;
1715                 curr_item = __btrfs_next_delayed_item(prev_item);
1716                 btrfs_release_delayed_item(prev_item);
1717         }
1718
1719         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1720         while (curr_item) {
1721                 btrfs_delayed_item_release_metadata(root, curr_item);
1722                 prev_item = curr_item;
1723                 curr_item = __btrfs_next_delayed_item(prev_item);
1724                 btrfs_release_delayed_item(prev_item);
1725         }
1726
1727         if (delayed_node->inode_dirty) {
1728                 btrfs_delayed_inode_release_metadata(root, delayed_node);
1729                 btrfs_release_delayed_inode(delayed_node);
1730         }
1731         mutex_unlock(&delayed_node->mutex);
1732 }
1733
1734 void btrfs_kill_delayed_inode_items(struct inode *inode)
1735 {
1736         struct btrfs_delayed_node *delayed_node;
1737
1738         delayed_node = btrfs_get_delayed_node(inode);
1739         if (!delayed_node)
1740                 return;
1741
1742         __btrfs_kill_delayed_node(delayed_node);
1743         btrfs_release_delayed_node(delayed_node);
1744 }
1745
1746 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1747 {
1748         u64 inode_id = 0;
1749         struct btrfs_delayed_node *delayed_nodes[8];
1750         int i, n;
1751
1752         while (1) {
1753                 spin_lock(&root->inode_lock);
1754                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1755                                            (void **)delayed_nodes, inode_id,
1756                                            ARRAY_SIZE(delayed_nodes));
1757                 if (!n) {
1758                         spin_unlock(&root->inode_lock);
1759                         break;
1760                 }
1761
1762                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1763
1764                 for (i = 0; i < n; i++)
1765                         atomic_inc(&delayed_nodes[i]->refs);
1766                 spin_unlock(&root->inode_lock);
1767
1768                 for (i = 0; i < n; i++) {
1769                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1770                         btrfs_release_delayed_node(delayed_nodes[i]);
1771                 }
1772         }
1773 }