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->delayed_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->delayed_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         src_rsv = trans->block_rsv;
 628         dst_rsv = &root->fs_info->delayed_block_rsv;
 629
 630         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 631
 632         /*
 633          * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 634          * which doesn't reserve space for speed.  This is a problem since we
 635          * still need to reserve space for this update, so try to reserve the
 636          * space.
 637          *
 638          * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 639          * we're accounted for.
 640          */
 641         if (!trans->bytes_reserved &&
 642             src_rsv != &root->fs_info->delalloc_block_rsv) {
 643                 ret = btrfs_block_rsv_add_noflush(root, dst_rsv, num_bytes);
 644                 /*
 645                  * Since we're under a transaction reserve_metadata_bytes could
 646                  * try to commit the transaction which will make it return
 647                  * EAGAIN to make us stop the transaction we have, so return
 648                  * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 649                  */
 650                 if (ret == -EAGAIN)
 651                         ret = -ENOSPC;
 652                 if (!ret)
 653                         node->bytes_reserved = num_bytes;
 654                 return ret;
 655         }
 656
 657         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 658         if (!ret)
 659                 node->bytes_reserved = num_bytes;
 660
 661         return ret;
 662 }
 663
 664 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
 665                                                 struct btrfs_delayed_node *node)
 666 {
 667         struct btrfs_block_rsv *rsv;
 668
 669         if (!node->bytes_reserved)
 670                 return;
 671
 672         rsv = &root->fs_info->delayed_block_rsv;
 673         btrfs_block_rsv_release(root, rsv,
 674                                 node->bytes_reserved);
 675         node->bytes_reserved = 0;
 676 }
 677
 678 /*
 679  * This helper will insert some continuous items into the same leaf according
 680  * to the free space of the leaf.
 681  */
 682 static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans,
 683                                 struct btrfs_root *root,
 684                                 struct btrfs_path *path,
 685                                 struct btrfs_delayed_item *item)
 686 {
 687         struct btrfs_delayed_item *curr, *next;
 688         int free_space;
 689         int total_data_size = 0, total_size = 0;
 690         struct extent_buffer *leaf;
 691         char *data_ptr;
 692         struct btrfs_key *keys;
 693         u32 *data_size;
 694         struct list_head head;
 695         int slot;
 696         int nitems;
 697         int i;
 698         int ret = 0;
 699
 700         BUG_ON(!path->nodes[0]);
 701
 702         leaf = path->nodes[0];
 703         free_space = btrfs_leaf_free_space(root, leaf);
 704         INIT_LIST_HEAD(&head);
 705
 706         next = item;
 707         nitems = 0;
 708
 709         /*
 710          * count the number of the continuous items that we can insert in batch
 711          */
 712         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 713                free_space) {
 714                 total_data_size += next->data_len;
 715                 total_size += next->data_len + sizeof(struct btrfs_item);
 716                 list_add_tail(&next->tree_list, &head);
 717                 nitems++;
 718
 719                 curr = next;
 720                 next = __btrfs_next_delayed_item(curr);
 721                 if (!next)
 722                         break;
 723
 724                 if (!btrfs_is_continuous_delayed_item(curr, next))
 725                         break;
 726         }
 727
 728         if (!nitems) {
 729                 ret = 0;
 730                 goto out;
 731         }
 732
 733         /*
 734          * we need allocate some memory space, but it might cause the task
 735          * to sleep, so we set all locked nodes in the path to blocking locks
 736          * first.
 737          */
 738         btrfs_set_path_blocking(path);
 739
 740         keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS);
 741         if (!keys) {
 742                 ret = -ENOMEM;
 743                 goto out;
 744         }
 745
 746         data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS);
 747         if (!data_size) {
 748                 ret = -ENOMEM;
 749                 goto error;
 750         }
 751
 752         /* get keys of all the delayed items */
 753         i = 0;
 754         list_for_each_entry(next, &head, tree_list) {
 755                 keys[i] = next->key;
 756                 data_size[i] = next->data_len;
 757                 i++;
 758         }
 759
 760         /* reset all the locked nodes in the patch to spinning locks. */
 761         btrfs_clear_path_blocking(path, NULL, 0);
 762
 763         /* insert the keys of the items */
 764         ret = setup_items_for_insert(trans, root, path, keys, data_size,
 765                                      total_data_size, total_size, nitems);
 766         if (ret)
 767                 goto error;
 768
 769         /* insert the dir index items */
 770         slot = path->slots[0];
 771         list_for_each_entry_safe(curr, next, &head, tree_list) {
 772                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 773                 write_extent_buffer(leaf, &curr->data,
 774                                     (unsigned long)data_ptr,
 775                                     curr->data_len);
 776                 slot++;
 777
 778                 btrfs_delayed_item_release_metadata(root, curr);
 779
 780                 list_del(&curr->tree_list);
 781                 btrfs_release_delayed_item(curr);
 782         }
 783
 784 error:
 785         kfree(data_size);
 786         kfree(keys);
 787 out:
 788         return ret;
 789 }
 790
 791 /*
 792  * This helper can just do simple insertion that needn't extend item for new
 793  * data, such as directory name index insertion, inode insertion.
 794  */
 795 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 796                                      struct btrfs_root *root,
 797                                      struct btrfs_path *path,
 798                                      struct btrfs_delayed_item *delayed_item)
 799 {
 800         struct extent_buffer *leaf;
 801         struct btrfs_item *item;
 802         char *ptr;
 803         int ret;
 804
 805         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 806                                       delayed_item->data_len);
 807         if (ret < 0 && ret != -EEXIST)
 808                 return ret;
 809
 810         leaf = path->nodes[0];
 811
 812         item = btrfs_item_nr(leaf, path->slots[0]);
 813         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 814
 815         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 816                             delayed_item->data_len);
 817         btrfs_mark_buffer_dirty(leaf);
 818
 819         btrfs_delayed_item_release_metadata(root, delayed_item);
 820         return 0;
 821 }
 822
 823 /*
 824  * we insert an item first, then if there are some continuous items, we try
 825  * to insert those items into the same leaf.
 826  */
 827 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 828                                       struct btrfs_path *path,
 829                                       struct btrfs_root *root,
 830                                       struct btrfs_delayed_node *node)
 831 {
 832         struct btrfs_delayed_item *curr, *prev;
 833         int ret = 0;
 834
 835 do_again:
 836         mutex_lock(&node->mutex);
 837         curr = __btrfs_first_delayed_insertion_item(node);
 838         if (!curr)
 839                 goto insert_end;
 840
 841         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 842         if (ret < 0) {
 843                 btrfs_release_path(path);
 844                 goto insert_end;
 845         }
 846
 847         prev = curr;
 848         curr = __btrfs_next_delayed_item(prev);
 849         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 850                 /* insert the continuous items into the same leaf */
 851                 path->slots[0]++;
 852                 btrfs_batch_insert_items(trans, root, path, curr);
 853         }
 854         btrfs_release_delayed_item(prev);
 855         btrfs_mark_buffer_dirty(path->nodes[0]);
 856
 857         btrfs_release_path(path);
 858         mutex_unlock(&node->mutex);
 859         goto do_again;
 860
 861 insert_end:
 862         mutex_unlock(&node->mutex);
 863         return ret;
 864 }
 865
 866 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 867                                     struct btrfs_root *root,
 868                                     struct btrfs_path *path,
 869                                     struct btrfs_delayed_item *item)
 870 {
 871         struct btrfs_delayed_item *curr, *next;
 872         struct extent_buffer *leaf;
 873         struct btrfs_key key;
 874         struct list_head head;
 875         int nitems, i, last_item;
 876         int ret = 0;
 877
 878         BUG_ON(!path->nodes[0]);
 879
 880         leaf = path->nodes[0];
 881
 882         i = path->slots[0];
 883         last_item = btrfs_header_nritems(leaf) - 1;
 884         if (i > last_item)
 885                 return -ENOENT; /* FIXME: Is errno suitable? */
 886
 887         next = item;
 888         INIT_LIST_HEAD(&head);
 889         btrfs_item_key_to_cpu(leaf, &key, i);
 890         nitems = 0;
 891         /*
 892          * count the number of the dir index items that we can delete in batch
 893          */
 894         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 895                 list_add_tail(&next->tree_list, &head);
 896                 nitems++;
 897
 898                 curr = next;
 899                 next = __btrfs_next_delayed_item(curr);
 900                 if (!next)
 901                         break;
 902
 903                 if (!btrfs_is_continuous_delayed_item(curr, next))
 904                         break;
 905
 906                 i++;
 907                 if (i > last_item)
 908                         break;
 909                 btrfs_item_key_to_cpu(leaf, &key, i);
 910         }
 911
 912         if (!nitems)
 913                 return 0;
 914
 915         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 916         if (ret)
 917                 goto out;
 918
 919         list_for_each_entry_safe(curr, next, &head, tree_list) {
 920                 btrfs_delayed_item_release_metadata(root, curr);
 921                 list_del(&curr->tree_list);
 922                 btrfs_release_delayed_item(curr);
 923         }
 924
 925 out:
 926         return ret;
 927 }
 928
 929 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 930                                       struct btrfs_path *path,
 931                                       struct btrfs_root *root,
 932                                       struct btrfs_delayed_node *node)
 933 {
 934         struct btrfs_delayed_item *curr, *prev;
 935         int ret = 0;
 936
 937 do_again:
 938         mutex_lock(&node->mutex);
 939         curr = __btrfs_first_delayed_deletion_item(node);
 940         if (!curr)
 941                 goto delete_fail;
 942
 943         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 944         if (ret < 0)
 945                 goto delete_fail;
 946         else if (ret > 0) {
 947                 /*
 948                  * can't find the item which the node points to, so this node
 949                  * is invalid, just drop it.
 950                  */
 951                 prev = curr;
 952                 curr = __btrfs_next_delayed_item(prev);
 953                 btrfs_release_delayed_item(prev);
 954                 ret = 0;
 955                 btrfs_release_path(path);
 956                 if (curr)
 957                         goto do_again;
 958                 else
 959                         goto delete_fail;
 960         }
 961
 962         btrfs_batch_delete_items(trans, root, path, curr);
 963         btrfs_release_path(path);
 964         mutex_unlock(&node->mutex);
 965         goto do_again;
 966
 967 delete_fail:
 968         btrfs_release_path(path);
 969         mutex_unlock(&node->mutex);
 970         return ret;
 971 }
 972
 973 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 974 {
 975         struct btrfs_delayed_root *delayed_root;
 976
 977         if (delayed_node && delayed_node->inode_dirty) {
 978                 BUG_ON(!delayed_node->root);
 979                 delayed_node->inode_dirty = 0;
 980                 delayed_node->count--;
 981
 982                 delayed_root = delayed_node->root->fs_info->delayed_root;
 983                 atomic_dec(&delayed_root->items);
 984                 if (atomic_read(&delayed_root->items) <
 985                     BTRFS_DELAYED_BACKGROUND &&
 986                     waitqueue_active(&delayed_root->wait))
 987                         wake_up(&delayed_root->wait);
 988         }
 989 }
 990
 991 static int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
 992                                       struct btrfs_root *root,
 993                                       struct btrfs_path *path,
 994                                       struct btrfs_delayed_node *node)
 995 {
 996         struct btrfs_key key;
 997         struct btrfs_inode_item *inode_item;
 998         struct extent_buffer *leaf;
 999         int ret;
1000
1001         mutex_lock(&node->mutex);
1002         if (!node->inode_dirty) {
1003                 mutex_unlock(&node->mutex);
1004                 return 0;
1005         }
1006
1007         key.objectid = node->inode_id;
1008         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1009         key.offset = 0;
1010         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1011         if (ret > 0) {
1012                 btrfs_release_path(path);
1013                 mutex_unlock(&node->mutex);
1014                 return -ENOENT;
1015         } else if (ret < 0) {
1016                 mutex_unlock(&node->mutex);
1017                 return ret;
1018         }
1019
1020         btrfs_unlock_up_safe(path, 1);
1021         leaf = path->nodes[0];
1022         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1023                                     struct btrfs_inode_item);
1024         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1025                             sizeof(struct btrfs_inode_item));
1026         btrfs_mark_buffer_dirty(leaf);
1027         btrfs_release_path(path);
1028
1029         btrfs_delayed_inode_release_metadata(root, node);
1030         btrfs_release_delayed_inode(node);
1031         mutex_unlock(&node->mutex);
1032
1033         return 0;
1034 }
1035
1036 /* Called when committing the transaction. */
1037 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1038                             struct btrfs_root *root)
1039 {
1040         struct btrfs_delayed_root *delayed_root;
1041         struct btrfs_delayed_node *curr_node, *prev_node;
1042         struct btrfs_path *path;
1043         struct btrfs_block_rsv *block_rsv;
1044         int ret = 0;
1045
1046         path = btrfs_alloc_path();
1047         if (!path)
1048                 return -ENOMEM;
1049         path->leave_spinning = 1;
1050
1051         block_rsv = trans->block_rsv;
1052         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1053
1054         delayed_root = btrfs_get_delayed_root(root);
1055
1056         curr_node = btrfs_first_delayed_node(delayed_root);
1057         while (curr_node) {
1058                 root = curr_node->root;
1059                 ret = btrfs_insert_delayed_items(trans, path, root,
1060                                                  curr_node);
1061                 if (!ret)
1062                         ret = btrfs_delete_delayed_items(trans, path, root,
1063                                                          curr_node);
1064                 if (!ret)
1065                         ret = btrfs_update_delayed_inode(trans, root, path,
1066                                                          curr_node);
1067                 if (ret) {
1068                         btrfs_release_delayed_node(curr_node);
1069                         break;
1070                 }
1071
1072                 prev_node = curr_node;
1073                 curr_node = btrfs_next_delayed_node(curr_node);
1074                 btrfs_release_delayed_node(prev_node);
1075         }
1076
1077         btrfs_free_path(path);
1078         trans->block_rsv = block_rsv;
1079         return ret;
1080 }
1081
1082 static int __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1083                                               struct btrfs_delayed_node *node)
1084 {
1085         struct btrfs_path *path;
1086         struct btrfs_block_rsv *block_rsv;
1087         int ret;
1088
1089         path = btrfs_alloc_path();
1090         if (!path)
1091                 return -ENOMEM;
1092         path->leave_spinning = 1;
1093
1094         block_rsv = trans->block_rsv;
1095         trans->block_rsv = &node->root->fs_info->delayed_block_rsv;
1096
1097         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1098         if (!ret)
1099                 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1100         if (!ret)
1101                 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1102         btrfs_free_path(path);
1103
1104         trans->block_rsv = block_rsv;
1105         return ret;
1106 }
1107
1108 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1109                                      struct inode *inode)
1110 {
1111         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1112         int ret;
1113
1114         if (!delayed_node)
1115                 return 0;
1116
1117         mutex_lock(&delayed_node->mutex);
1118         if (!delayed_node->count) {
1119                 mutex_unlock(&delayed_node->mutex);
1120                 btrfs_release_delayed_node(delayed_node);
1121                 return 0;
1122         }
1123         mutex_unlock(&delayed_node->mutex);
1124
1125         ret = __btrfs_commit_inode_delayed_items(trans, delayed_node);
1126         btrfs_release_delayed_node(delayed_node);
1127         return ret;
1128 }
1129
1130 void btrfs_remove_delayed_node(struct inode *inode)
1131 {
1132         struct btrfs_delayed_node *delayed_node;
1133
1134         delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1135         if (!delayed_node)
1136                 return;
1137
1138         BTRFS_I(inode)->delayed_node = NULL;
1139         btrfs_release_delayed_node(delayed_node);
1140 }
1141
1142 struct btrfs_async_delayed_node {
1143         struct btrfs_root *root;
1144         struct btrfs_delayed_node *delayed_node;
1145         struct btrfs_work work;
1146 };
1147
1148 static void btrfs_async_run_delayed_node_done(struct btrfs_work *work)
1149 {
1150         struct btrfs_async_delayed_node *async_node;
1151         struct btrfs_trans_handle *trans;
1152         struct btrfs_path *path;
1153         struct btrfs_delayed_node *delayed_node = NULL;
1154         struct btrfs_root *root;
1155         struct btrfs_block_rsv *block_rsv;
1156         unsigned long nr = 0;
1157         int need_requeue = 0;
1158         int ret;
1159
1160         async_node = container_of(work, struct btrfs_async_delayed_node, work);
1161
1162         path = btrfs_alloc_path();
1163         if (!path)
1164                 goto out;
1165         path->leave_spinning = 1;
1166
1167         delayed_node = async_node->delayed_node;
1168         root = delayed_node->root;
1169
1170         trans = btrfs_join_transaction(root);
1171         if (IS_ERR(trans))
1172                 goto free_path;
1173
1174         block_rsv = trans->block_rsv;
1175         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1176
1177         ret = btrfs_insert_delayed_items(trans, path, root, delayed_node);
1178         if (!ret)
1179                 ret = btrfs_delete_delayed_items(trans, path, root,
1180                                                  delayed_node);
1181
1182         if (!ret)
1183                 btrfs_update_delayed_inode(trans, root, path, delayed_node);
1184
1185         /*
1186          * Maybe new delayed items have been inserted, so we need requeue
1187          * the work. Besides that, we must dequeue the empty delayed nodes
1188          * to avoid the race between delayed items balance and the worker.
1189          * The race like this:
1190          *      Task1                           Worker thread
1191          *                                      count == 0, needn't requeue
1192          *                                        also needn't insert the
1193          *                                        delayed node into prepare
1194          *                                        list again.
1195          *      add lots of delayed items
1196          *      queue the delayed node
1197          *        already in the list,
1198          *        and not in the prepare
1199          *        list, it means the delayed
1200          *        node is being dealt with
1201          *        by the worker.
1202          *      do delayed items balance
1203          *        the delayed node is being
1204          *        dealt with by the worker
1205          *        now, just wait.
1206          *                                      the worker goto idle.
1207          * Task1 will sleep until the transaction is commited.
1208          */
1209         mutex_lock(&delayed_node->mutex);
1210         if (delayed_node->count)
1211                 need_requeue = 1;
1212         else
1213                 btrfs_dequeue_delayed_node(root->fs_info->delayed_root,
1214                                            delayed_node);
1215         mutex_unlock(&delayed_node->mutex);
1216
1217         nr = trans->blocks_used;
1218
1219         trans->block_rsv = block_rsv;
1220         btrfs_end_transaction_dmeta(trans, root);
1221         __btrfs_btree_balance_dirty(root, nr);
1222 free_path:
1223         btrfs_free_path(path);
1224 out:
1225         if (need_requeue)
1226                 btrfs_requeue_work(&async_node->work);
1227         else {
1228                 btrfs_release_prepared_delayed_node(delayed_node);
1229                 kfree(async_node);
1230         }
1231 }
1232
1233 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1234                                      struct btrfs_root *root, int all)
1235 {
1236         struct btrfs_async_delayed_node *async_node;
1237         struct btrfs_delayed_node *curr;
1238         int count = 0;
1239
1240 again:
1241         curr = btrfs_first_prepared_delayed_node(delayed_root);
1242         if (!curr)
1243                 return 0;
1244
1245         async_node = kmalloc(sizeof(*async_node), GFP_NOFS);
1246         if (!async_node) {
1247                 btrfs_release_prepared_delayed_node(curr);
1248                 return -ENOMEM;
1249         }
1250
1251         async_node->root = root;
1252         async_node->delayed_node = curr;
1253
1254         async_node->work.func = btrfs_async_run_delayed_node_done;
1255         async_node->work.flags = 0;
1256
1257         btrfs_queue_worker(&root->fs_info->delayed_workers, &async_node->work);
1258         count++;
1259
1260         if (all || count < 4)
1261                 goto again;
1262
1263         return 0;
1264 }
1265
1266 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1267 {
1268         struct btrfs_delayed_root *delayed_root;
1269         delayed_root = btrfs_get_delayed_root(root);
1270         WARN_ON(btrfs_first_delayed_node(delayed_root));
1271 }
1272
1273 void btrfs_balance_delayed_items(struct btrfs_root *root)
1274 {
1275         struct btrfs_delayed_root *delayed_root;
1276
1277         delayed_root = btrfs_get_delayed_root(root);
1278
1279         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1280                 return;
1281
1282         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1283                 int ret;
1284                 ret = btrfs_wq_run_delayed_node(delayed_root, root, 1);
1285                 if (ret)
1286                         return;
1287
1288                 wait_event_interruptible_timeout(
1289                                 delayed_root->wait,
1290                                 (atomic_read(&delayed_root->items) <
1291                                  BTRFS_DELAYED_BACKGROUND),
1292                                 HZ);
1293                 return;
1294         }
1295
1296         btrfs_wq_run_delayed_node(delayed_root, root, 0);
1297 }
1298
1299 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1300                                    struct btrfs_root *root, const char *name,
1301                                    int name_len, struct inode *dir,
1302                                    struct btrfs_disk_key *disk_key, u8 type,
1303                                    u64 index)
1304 {
1305         struct btrfs_delayed_node *delayed_node;
1306         struct btrfs_delayed_item *delayed_item;
1307         struct btrfs_dir_item *dir_item;
1308         int ret;
1309
1310         delayed_node = btrfs_get_or_create_delayed_node(dir);
1311         if (IS_ERR(delayed_node))
1312                 return PTR_ERR(delayed_node);
1313
1314         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1315         if (!delayed_item) {
1316                 ret = -ENOMEM;
1317                 goto release_node;
1318         }
1319
1320         ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1321         /*
1322          * we have reserved enough space when we start a new transaction,
1323          * so reserving metadata failure is impossible
1324          */
1325         BUG_ON(ret);
1326
1327         delayed_item->key.objectid = btrfs_ino(dir);
1328         btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1329         delayed_item->key.offset = index;
1330
1331         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1332         dir_item->location = *disk_key;
1333         dir_item->transid = cpu_to_le64(trans->transid);
1334         dir_item->data_len = 0;
1335         dir_item->name_len = cpu_to_le16(name_len);
1336         dir_item->type = type;
1337         memcpy((char *)(dir_item + 1), name, name_len);
1338
1339         mutex_lock(&delayed_node->mutex);
1340         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1341         if (unlikely(ret)) {
1342                 printk(KERN_ERR "err add delayed dir index item(name: %s) into "
1343                                 "the insertion tree of the delayed node"
1344                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1345                                 name,
1346                                 (unsigned long long)delayed_node->root->objectid,
1347                                 (unsigned long long)delayed_node->inode_id,
1348                                 ret);
1349                 BUG();
1350         }
1351         mutex_unlock(&delayed_node->mutex);
1352
1353 release_node:
1354         btrfs_release_delayed_node(delayed_node);
1355         return ret;
1356 }
1357
1358 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1359                                                struct btrfs_delayed_node *node,
1360                                                struct btrfs_key *key)
1361 {
1362         struct btrfs_delayed_item *item;
1363
1364         mutex_lock(&node->mutex);
1365         item = __btrfs_lookup_delayed_insertion_item(node, key);
1366         if (!item) {
1367                 mutex_unlock(&node->mutex);
1368                 return 1;
1369         }
1370
1371         btrfs_delayed_item_release_metadata(root, item);
1372         btrfs_release_delayed_item(item);
1373         mutex_unlock(&node->mutex);
1374         return 0;
1375 }
1376
1377 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1378                                    struct btrfs_root *root, struct inode *dir,
1379                                    u64 index)
1380 {
1381         struct btrfs_delayed_node *node;
1382         struct btrfs_delayed_item *item;
1383         struct btrfs_key item_key;
1384         int ret;
1385
1386         node = btrfs_get_or_create_delayed_node(dir);
1387         if (IS_ERR(node))
1388                 return PTR_ERR(node);
1389
1390         item_key.objectid = btrfs_ino(dir);
1391         btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1392         item_key.offset = index;
1393
1394         ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1395         if (!ret)
1396                 goto end;
1397
1398         item = btrfs_alloc_delayed_item(0);
1399         if (!item) {
1400                 ret = -ENOMEM;
1401                 goto end;
1402         }
1403
1404         item->key = item_key;
1405
1406         ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1407         /*
1408          * we have reserved enough space when we start a new transaction,
1409          * so reserving metadata failure is impossible.
1410          */
1411         BUG_ON(ret);
1412
1413         mutex_lock(&node->mutex);
1414         ret = __btrfs_add_delayed_deletion_item(node, item);
1415         if (unlikely(ret)) {
1416                 printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1417                                 "into the deletion tree of the delayed node"
1418                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1419                                 (unsigned long long)index,
1420                                 (unsigned long long)node->root->objectid,
1421                                 (unsigned long long)node->inode_id,
1422                                 ret);
1423                 BUG();
1424         }
1425         mutex_unlock(&node->mutex);
1426 end:
1427         btrfs_release_delayed_node(node);
1428         return ret;
1429 }
1430
1431 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1432 {
1433         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1434
1435         if (!delayed_node)
1436                 return -ENOENT;
1437
1438         /*
1439          * Since we have held i_mutex of this directory, it is impossible that
1440          * a new directory index is added into the delayed node and index_cnt
1441          * is updated now. So we needn't lock the delayed node.
1442          */
1443         if (!delayed_node->index_cnt) {
1444                 btrfs_release_delayed_node(delayed_node);
1445                 return -EINVAL;
1446         }
1447
1448         BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1449         btrfs_release_delayed_node(delayed_node);
1450         return 0;
1451 }
1452
1453 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1454                              struct list_head *del_list)
1455 {
1456         struct btrfs_delayed_node *delayed_node;
1457         struct btrfs_delayed_item *item;
1458
1459         delayed_node = btrfs_get_delayed_node(inode);
1460         if (!delayed_node)
1461                 return;
1462
1463         mutex_lock(&delayed_node->mutex);
1464         item = __btrfs_first_delayed_insertion_item(delayed_node);
1465         while (item) {
1466                 atomic_inc(&item->refs);
1467                 list_add_tail(&item->readdir_list, ins_list);
1468                 item = __btrfs_next_delayed_item(item);
1469         }
1470
1471         item = __btrfs_first_delayed_deletion_item(delayed_node);
1472         while (item) {
1473                 atomic_inc(&item->refs);
1474                 list_add_tail(&item->readdir_list, del_list);
1475                 item = __btrfs_next_delayed_item(item);
1476         }
1477         mutex_unlock(&delayed_node->mutex);
1478         /*
1479          * This delayed node is still cached in the btrfs inode, so refs
1480          * must be > 1 now, and we needn't check it is going to be freed
1481          * or not.
1482          *
1483          * Besides that, this function is used to read dir, we do not
1484          * insert/delete delayed items in this period. So we also needn't
1485          * requeue or dequeue this delayed node.
1486          */
1487         atomic_dec(&delayed_node->refs);
1488 }
1489
1490 void btrfs_put_delayed_items(struct list_head *ins_list,
1491                              struct list_head *del_list)
1492 {
1493         struct btrfs_delayed_item *curr, *next;
1494
1495         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1496                 list_del(&curr->readdir_list);
1497                 if (atomic_dec_and_test(&curr->refs))
1498                         kfree(curr);
1499         }
1500
1501         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1502                 list_del(&curr->readdir_list);
1503                 if (atomic_dec_and_test(&curr->refs))
1504                         kfree(curr);
1505         }
1506 }
1507
1508 int btrfs_should_delete_dir_index(struct list_head *del_list,
1509                                   u64 index)
1510 {
1511         struct btrfs_delayed_item *curr, *next;
1512         int ret;
1513
1514         if (list_empty(del_list))
1515                 return 0;
1516
1517         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1518                 if (curr->key.offset > index)
1519                         break;
1520
1521                 list_del(&curr->readdir_list);
1522                 ret = (curr->key.offset == index);
1523
1524                 if (atomic_dec_and_test(&curr->refs))
1525                         kfree(curr);
1526
1527                 if (ret)
1528                         return 1;
1529                 else
1530                         continue;
1531         }
1532         return 0;
1533 }
1534
1535 /*
1536  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1537  *
1538  */
1539 int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent,
1540                                     filldir_t filldir,
1541                                     struct list_head *ins_list)
1542 {
1543         struct btrfs_dir_item *di;
1544         struct btrfs_delayed_item *curr, *next;
1545         struct btrfs_key location;
1546         char *name;
1547         int name_len;
1548         int over = 0;
1549         unsigned char d_type;
1550
1551         if (list_empty(ins_list))
1552                 return 0;
1553
1554         /*
1555          * Changing the data of the delayed item is impossible. So
1556          * we needn't lock them. And we have held i_mutex of the
1557          * directory, nobody can delete any directory indexes now.
1558          */
1559         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1560                 list_del(&curr->readdir_list);
1561
1562                 if (curr->key.offset < filp->f_pos) {
1563                         if (atomic_dec_and_test(&curr->refs))
1564                                 kfree(curr);
1565                         continue;
1566                 }
1567
1568                 filp->f_pos = curr->key.offset;
1569
1570                 di = (struct btrfs_dir_item *)curr->data;
1571                 name = (char *)(di + 1);
1572                 name_len = le16_to_cpu(di->name_len);
1573
1574                 d_type = btrfs_filetype_table[di->type];
1575                 btrfs_disk_key_to_cpu(&location, &di->location);
1576
1577                 over = filldir(dirent, name, name_len, curr->key.offset,
1578                                location.objectid, d_type);
1579
1580                 if (atomic_dec_and_test(&curr->refs))
1581                         kfree(curr);
1582
1583                 if (over)
1584                         return 1;
1585         }
1586         return 0;
1587 }
1588
1589 BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item,
1590                          generation, 64);
1591 BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item,
1592                          sequence, 64);
1593 BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item,
1594                          transid, 64);
1595 BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64);
1596 BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item,
1597                          nbytes, 64);
1598 BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item,
1599                          block_group, 64);
1600 BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32);
1601 BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32);
1602 BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32);
1603 BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32);
1604 BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64);
1605 BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64);
1606
1607 BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64);
1608 BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32);
1609
1610 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1611                                   struct btrfs_inode_item *inode_item,
1612                                   struct inode *inode)
1613 {
1614         btrfs_set_stack_inode_uid(inode_item, inode->i_uid);
1615         btrfs_set_stack_inode_gid(inode_item, inode->i_gid);
1616         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1617         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1618         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1619         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1620         btrfs_set_stack_inode_generation(inode_item,
1621                                          BTRFS_I(inode)->generation);
1622         btrfs_set_stack_inode_sequence(inode_item, BTRFS_I(inode)->sequence);
1623         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1624         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1625         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1626         btrfs_set_stack_inode_block_group(inode_item, 0);
1627
1628         btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1629                                      inode->i_atime.tv_sec);
1630         btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1631                                       inode->i_atime.tv_nsec);
1632
1633         btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1634                                      inode->i_mtime.tv_sec);
1635         btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1636                                       inode->i_mtime.tv_nsec);
1637
1638         btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1639                                      inode->i_ctime.tv_sec);
1640         btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1641                                       inode->i_ctime.tv_nsec);
1642 }
1643
1644 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1645 {
1646         struct btrfs_delayed_node *delayed_node;
1647         struct btrfs_inode_item *inode_item;
1648         struct btrfs_timespec *tspec;
1649
1650         delayed_node = btrfs_get_delayed_node(inode);
1651         if (!delayed_node)
1652                 return -ENOENT;
1653
1654         mutex_lock(&delayed_node->mutex);
1655         if (!delayed_node->inode_dirty) {
1656                 mutex_unlock(&delayed_node->mutex);
1657                 btrfs_release_delayed_node(delayed_node);
1658                 return -ENOENT;
1659         }
1660
1661         inode_item = &delayed_node->inode_item;
1662
1663         inode->i_uid = btrfs_stack_inode_uid(inode_item);
1664         inode->i_gid = btrfs_stack_inode_gid(inode_item);
1665         btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1666         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1667         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1668         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1669         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1670         BTRFS_I(inode)->sequence = btrfs_stack_inode_sequence(inode_item);
1671         inode->i_rdev = 0;
1672         *rdev = btrfs_stack_inode_rdev(inode_item);
1673         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1674
1675         tspec = btrfs_inode_atime(inode_item);
1676         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1677         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1678
1679         tspec = btrfs_inode_mtime(inode_item);
1680         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1681         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1682
1683         tspec = btrfs_inode_ctime(inode_item);
1684         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1685         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1686
1687         inode->i_generation = BTRFS_I(inode)->generation;
1688         BTRFS_I(inode)->index_cnt = (u64)-1;
1689
1690         mutex_unlock(&delayed_node->mutex);
1691         btrfs_release_delayed_node(delayed_node);
1692         return 0;
1693 }
1694
1695 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1696                                struct btrfs_root *root, struct inode *inode)
1697 {
1698         struct btrfs_delayed_node *delayed_node;
1699         int ret = 0;
1700
1701         delayed_node = btrfs_get_or_create_delayed_node(inode);
1702         if (IS_ERR(delayed_node))
1703                 return PTR_ERR(delayed_node);
1704
1705         mutex_lock(&delayed_node->mutex);
1706         if (delayed_node->inode_dirty) {
1707                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1708                 goto release_node;
1709         }
1710
1711         ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1712         if (ret)
1713                 goto release_node;
1714
1715         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1716         delayed_node->inode_dirty = 1;
1717         delayed_node->count++;
1718         atomic_inc(&root->fs_info->delayed_root->items);
1719 release_node:
1720         mutex_unlock(&delayed_node->mutex);
1721         btrfs_release_delayed_node(delayed_node);
1722         return ret;
1723 }
1724
1725 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1726 {
1727         struct btrfs_root *root = delayed_node->root;
1728         struct btrfs_delayed_item *curr_item, *prev_item;
1729
1730         mutex_lock(&delayed_node->mutex);
1731         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1732         while (curr_item) {
1733                 btrfs_delayed_item_release_metadata(root, curr_item);
1734                 prev_item = curr_item;
1735                 curr_item = __btrfs_next_delayed_item(prev_item);
1736                 btrfs_release_delayed_item(prev_item);
1737         }
1738
1739         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1740         while (curr_item) {
1741                 btrfs_delayed_item_release_metadata(root, curr_item);
1742                 prev_item = curr_item;
1743                 curr_item = __btrfs_next_delayed_item(prev_item);
1744                 btrfs_release_delayed_item(prev_item);
1745         }
1746
1747         if (delayed_node->inode_dirty) {
1748                 btrfs_delayed_inode_release_metadata(root, delayed_node);
1749                 btrfs_release_delayed_inode(delayed_node);
1750         }
1751         mutex_unlock(&delayed_node->mutex);
1752 }
1753
1754 void btrfs_kill_delayed_inode_items(struct inode *inode)
1755 {
1756         struct btrfs_delayed_node *delayed_node;
1757
1758         delayed_node = btrfs_get_delayed_node(inode);
1759         if (!delayed_node)
1760                 return;
1761
1762         __btrfs_kill_delayed_node(delayed_node);
1763         btrfs_release_delayed_node(delayed_node);
1764 }
1765
1766 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1767 {
1768         u64 inode_id = 0;
1769         struct btrfs_delayed_node *delayed_nodes[8];
1770         int i, n;
1771
1772         while (1) {
1773                 spin_lock(&root->inode_lock);
1774                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1775                                            (void **)delayed_nodes, inode_id,
1776                                            ARRAY_SIZE(delayed_nodes));
1777                 if (!n) {
1778                         spin_unlock(&root->inode_lock);
1779                         break;
1780                 }
1781
1782                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1783
1784                 for (i = 0; i < n; i++)
1785                         atomic_inc(&delayed_nodes[i]->refs);
1786                 spin_unlock(&root->inode_lock);
1787
1788                 for (i = 0; i < n; i++) {
1789                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1790                         btrfs_release_delayed_node(delayed_nodes[i]);
1791                 }
1792         }
1793 }