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                 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
 600                                               item->key.objectid,
 601                                               num_bytes, 1);
 602                 item->bytes_reserved = num_bytes;
 603         }
 604
 605         return ret;
 606 }
 607
 608 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 609                                                 struct btrfs_delayed_item *item)
 610 {
 611         struct btrfs_block_rsv *rsv;
 612
 613         if (!item->bytes_reserved)
 614                 return;
 615
 616         rsv = &root->fs_info->delayed_block_rsv;
 617         trace_btrfs_space_reservation(root->fs_info, "delayed_item",
 618                                       item->key.objectid, item->bytes_reserved,
 619                                       0);
 620         btrfs_block_rsv_release(root, rsv,
 621                                 item->bytes_reserved);
 622 }
 623
 624 static int btrfs_delayed_inode_reserve_metadata(
 625                                         struct btrfs_trans_handle *trans,
 626                                         struct btrfs_root *root,
 627                                         struct inode *inode,
 628                                         struct btrfs_delayed_node *node)
 629 {
 630         struct btrfs_block_rsv *src_rsv;
 631         struct btrfs_block_rsv *dst_rsv;
 632         u64 num_bytes;
 633         int ret;
 634         bool release = false;
 635
 636         src_rsv = trans->block_rsv;
 637         dst_rsv = &root->fs_info->delayed_block_rsv;
 638
 639         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 640
 641         /*
 642          * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 643          * which doesn't reserve space for speed.  This is a problem since we
 644          * still need to reserve space for this update, so try to reserve the
 645          * space.
 646          *
 647          * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 648          * we're accounted for.
 649          */
 650         if (!src_rsv || (!trans->bytes_reserved &&
 651             src_rsv != &root->fs_info->delalloc_block_rsv)) {
 652                 ret = btrfs_block_rsv_add_noflush(root, dst_rsv, num_bytes);
 653                 /*
 654                  * Since we're under a transaction reserve_metadata_bytes could
 655                  * try to commit the transaction which will make it return
 656                  * EAGAIN to make us stop the transaction we have, so return
 657                  * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 658                  */
 659                 if (ret == -EAGAIN)
 660                         ret = -ENOSPC;
 661                 if (!ret) {
 662                         node->bytes_reserved = num_bytes;
 663                         trace_btrfs_space_reservation(root->fs_info,
 664                                                       "delayed_inode",
 665                                                       btrfs_ino(inode),
 666                                                       num_bytes, 1);
 667                 }
 668                 return ret;
 669         } else if (src_rsv == &root->fs_info->delalloc_block_rsv) {
 670                 spin_lock(&BTRFS_I(inode)->lock);
 671                 if (BTRFS_I(inode)->delalloc_meta_reserved) {
 672                         BTRFS_I(inode)->delalloc_meta_reserved = 0;
 673                         spin_unlock(&BTRFS_I(inode)->lock);
 674                         release = true;
 675                         goto migrate;
 676                 }
 677                 spin_unlock(&BTRFS_I(inode)->lock);
 678
 679                 /* Ok we didn't have space pre-reserved.  This shouldn't happen
 680                  * too often but it can happen if we do delalloc to an existing
 681                  * inode which gets dirtied because of the time update, and then
 682                  * isn't touched again until after the transaction commits and
 683                  * then we try to write out the data.  First try to be nice and
 684                  * reserve something strictly for us.  If not be a pain and try
 685                  * to steal from the delalloc block rsv.
 686                  */
 687                 ret = btrfs_block_rsv_add_noflush(root, dst_rsv, num_bytes);
 688                 if (!ret)
 689                         goto out;
 690
 691                 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 692                 if (!ret)
 693                         goto out;
 694
 695                 /*
 696                  * Ok this is a problem, let's just steal from the global rsv
 697                  * since this really shouldn't happen that often.
 698                  */
 699                 WARN_ON(1);
 700                 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
 701                                               dst_rsv, num_bytes);
 702                 goto out;
 703         }
 704
 705 migrate:
 706         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 707
 708 out:
 709         /*
 710          * Migrate only takes a reservation, it doesn't touch the size of the
 711          * block_rsv.  This is to simplify people who don't normally have things
 712          * migrated from their block rsv.  If they go to release their
 713          * reservation, that will decrease the size as well, so if migrate
 714          * reduced size we'd end up with a negative size.  But for the
 715          * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
 716          * but we could in fact do this reserve/migrate dance several times
 717          * between the time we did the original reservation and we'd clean it
 718          * up.  So to take care of this, release the space for the meta
 719          * reservation here.  I think it may be time for a documentation page on
 720          * how block rsvs. work.
 721          */
 722         if (!ret) {
 723                 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
 724                                               btrfs_ino(inode), num_bytes, 1);
 725                 node->bytes_reserved = num_bytes;
 726         }
 727
 728         if (release) {
 729                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
 730                                               btrfs_ino(inode), num_bytes, 0);
 731                 btrfs_block_rsv_release(root, src_rsv, num_bytes);
 732         }
 733
 734         return ret;
 735 }
 736
 737 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
 738                                                 struct btrfs_delayed_node *node)
 739 {
 740         struct btrfs_block_rsv *rsv;
 741
 742         if (!node->bytes_reserved)
 743                 return;
 744
 745         rsv = &root->fs_info->delayed_block_rsv;
 746         trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
 747                                       node->inode_id, node->bytes_reserved, 0);
 748         btrfs_block_rsv_release(root, rsv,
 749                                 node->bytes_reserved);
 750         node->bytes_reserved = 0;
 751 }
 752
 753 /*
 754  * This helper will insert some continuous items into the same leaf according
 755  * to the free space of the leaf.
 756  */
 757 static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans,
 758                                 struct btrfs_root *root,
 759                                 struct btrfs_path *path,
 760                                 struct btrfs_delayed_item *item)
 761 {
 762         struct btrfs_delayed_item *curr, *next;
 763         int free_space;
 764         int total_data_size = 0, total_size = 0;
 765         struct extent_buffer *leaf;
 766         char *data_ptr;
 767         struct btrfs_key *keys;
 768         u32 *data_size;
 769         struct list_head head;
 770         int slot;
 771         int nitems;
 772         int i;
 773         int ret = 0;
 774
 775         BUG_ON(!path->nodes[0]);
 776
 777         leaf = path->nodes[0];
 778         free_space = btrfs_leaf_free_space(root, leaf);
 779         INIT_LIST_HEAD(&head);
 780
 781         next = item;
 782         nitems = 0;
 783
 784         /*
 785          * count the number of the continuous items that we can insert in batch
 786          */
 787         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 788                free_space) {
 789                 total_data_size += next->data_len;
 790                 total_size += next->data_len + sizeof(struct btrfs_item);
 791                 list_add_tail(&next->tree_list, &head);
 792                 nitems++;
 793
 794                 curr = next;
 795                 next = __btrfs_next_delayed_item(curr);
 796                 if (!next)
 797                         break;
 798
 799                 if (!btrfs_is_continuous_delayed_item(curr, next))
 800                         break;
 801         }
 802
 803         if (!nitems) {
 804                 ret = 0;
 805                 goto out;
 806         }
 807
 808         /*
 809          * we need allocate some memory space, but it might cause the task
 810          * to sleep, so we set all locked nodes in the path to blocking locks
 811          * first.
 812          */
 813         btrfs_set_path_blocking(path);
 814
 815         keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS);
 816         if (!keys) {
 817                 ret = -ENOMEM;
 818                 goto out;
 819         }
 820
 821         data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS);
 822         if (!data_size) {
 823                 ret = -ENOMEM;
 824                 goto error;
 825         }
 826
 827         /* get keys of all the delayed items */
 828         i = 0;
 829         list_for_each_entry(next, &head, tree_list) {
 830                 keys[i] = next->key;
 831                 data_size[i] = next->data_len;
 832                 i++;
 833         }
 834
 835         /* reset all the locked nodes in the patch to spinning locks. */
 836         btrfs_clear_path_blocking(path, NULL, 0);
 837
 838         /* insert the keys of the items */
 839         ret = setup_items_for_insert(trans, root, path, keys, data_size,
 840                                      total_data_size, total_size, nitems);
 841         if (ret)
 842                 goto error;
 843
 844         /* insert the dir index items */
 845         slot = path->slots[0];
 846         list_for_each_entry_safe(curr, next, &head, tree_list) {
 847                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 848                 write_extent_buffer(leaf, &curr->data,
 849                                     (unsigned long)data_ptr,
 850                                     curr->data_len);
 851                 slot++;
 852
 853                 btrfs_delayed_item_release_metadata(root, curr);
 854
 855                 list_del(&curr->tree_list);
 856                 btrfs_release_delayed_item(curr);
 857         }
 858
 859 error:
 860         kfree(data_size);
 861         kfree(keys);
 862 out:
 863         return ret;
 864 }
 865
 866 /*
 867  * This helper can just do simple insertion that needn't extend item for new
 868  * data, such as directory name index insertion, inode insertion.
 869  */
 870 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 871                                      struct btrfs_root *root,
 872                                      struct btrfs_path *path,
 873                                      struct btrfs_delayed_item *delayed_item)
 874 {
 875         struct extent_buffer *leaf;
 876         struct btrfs_item *item;
 877         char *ptr;
 878         int ret;
 879
 880         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 881                                       delayed_item->data_len);
 882         if (ret < 0 && ret != -EEXIST)
 883                 return ret;
 884
 885         leaf = path->nodes[0];
 886
 887         item = btrfs_item_nr(leaf, path->slots[0]);
 888         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 889
 890         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 891                             delayed_item->data_len);
 892         btrfs_mark_buffer_dirty(leaf);
 893
 894         btrfs_delayed_item_release_metadata(root, delayed_item);
 895         return 0;
 896 }
 897
 898 /*
 899  * we insert an item first, then if there are some continuous items, we try
 900  * to insert those items into the same leaf.
 901  */
 902 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 903                                       struct btrfs_path *path,
 904                                       struct btrfs_root *root,
 905                                       struct btrfs_delayed_node *node)
 906 {
 907         struct btrfs_delayed_item *curr, *prev;
 908         int ret = 0;
 909
 910 do_again:
 911         mutex_lock(&node->mutex);
 912         curr = __btrfs_first_delayed_insertion_item(node);
 913         if (!curr)
 914                 goto insert_end;
 915
 916         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 917         if (ret < 0) {
 918                 btrfs_release_path(path);
 919                 goto insert_end;
 920         }
 921
 922         prev = curr;
 923         curr = __btrfs_next_delayed_item(prev);
 924         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 925                 /* insert the continuous items into the same leaf */
 926                 path->slots[0]++;
 927                 btrfs_batch_insert_items(trans, root, path, curr);
 928         }
 929         btrfs_release_delayed_item(prev);
 930         btrfs_mark_buffer_dirty(path->nodes[0]);
 931
 932         btrfs_release_path(path);
 933         mutex_unlock(&node->mutex);
 934         goto do_again;
 935
 936 insert_end:
 937         mutex_unlock(&node->mutex);
 938         return ret;
 939 }
 940
 941 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 942                                     struct btrfs_root *root,
 943                                     struct btrfs_path *path,
 944                                     struct btrfs_delayed_item *item)
 945 {
 946         struct btrfs_delayed_item *curr, *next;
 947         struct extent_buffer *leaf;
 948         struct btrfs_key key;
 949         struct list_head head;
 950         int nitems, i, last_item;
 951         int ret = 0;
 952
 953         BUG_ON(!path->nodes[0]);
 954
 955         leaf = path->nodes[0];
 956
 957         i = path->slots[0];
 958         last_item = btrfs_header_nritems(leaf) - 1;
 959         if (i > last_item)
 960                 return -ENOENT; /* FIXME: Is errno suitable? */
 961
 962         next = item;
 963         INIT_LIST_HEAD(&head);
 964         btrfs_item_key_to_cpu(leaf, &key, i);
 965         nitems = 0;
 966         /*
 967          * count the number of the dir index items that we can delete in batch
 968          */
 969         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 970                 list_add_tail(&next->tree_list, &head);
 971                 nitems++;
 972
 973                 curr = next;
 974                 next = __btrfs_next_delayed_item(curr);
 975                 if (!next)
 976                         break;
 977
 978                 if (!btrfs_is_continuous_delayed_item(curr, next))
 979                         break;
 980
 981                 i++;
 982                 if (i > last_item)
 983                         break;
 984                 btrfs_item_key_to_cpu(leaf, &key, i);
 985         }
 986
 987         if (!nitems)
 988                 return 0;
 989
 990         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 991         if (ret)
 992                 goto out;
 993
 994         list_for_each_entry_safe(curr, next, &head, tree_list) {
 995                 btrfs_delayed_item_release_metadata(root, curr);
 996                 list_del(&curr->tree_list);
 997                 btrfs_release_delayed_item(curr);
 998         }
 999
1000 out:
1001         return ret;
1002 }
1003
1004 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
1005                                       struct btrfs_path *path,
1006                                       struct btrfs_root *root,
1007                                       struct btrfs_delayed_node *node)
1008 {
1009         struct btrfs_delayed_item *curr, *prev;
1010         int ret = 0;
1011
1012 do_again:
1013         mutex_lock(&node->mutex);
1014         curr = __btrfs_first_delayed_deletion_item(node);
1015         if (!curr)
1016                 goto delete_fail;
1017
1018         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
1019         if (ret < 0)
1020                 goto delete_fail;
1021         else if (ret > 0) {
1022                 /*
1023                  * can't find the item which the node points to, so this node
1024                  * is invalid, just drop it.
1025                  */
1026                 prev = curr;
1027                 curr = __btrfs_next_delayed_item(prev);
1028                 btrfs_release_delayed_item(prev);
1029                 ret = 0;
1030                 btrfs_release_path(path);
1031                 if (curr)
1032                         goto do_again;
1033                 else
1034                         goto delete_fail;
1035         }
1036
1037         btrfs_batch_delete_items(trans, root, path, curr);
1038         btrfs_release_path(path);
1039         mutex_unlock(&node->mutex);
1040         goto do_again;
1041
1042 delete_fail:
1043         btrfs_release_path(path);
1044         mutex_unlock(&node->mutex);
1045         return ret;
1046 }
1047
1048 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1049 {
1050         struct btrfs_delayed_root *delayed_root;
1051
1052         if (delayed_node && delayed_node->inode_dirty) {
1053                 BUG_ON(!delayed_node->root);
1054                 delayed_node->inode_dirty = 0;
1055                 delayed_node->count--;
1056
1057                 delayed_root = delayed_node->root->fs_info->delayed_root;
1058                 atomic_dec(&delayed_root->items);
1059                 if (atomic_read(&delayed_root->items) <
1060                     BTRFS_DELAYED_BACKGROUND &&
1061                     waitqueue_active(&delayed_root->wait))
1062                         wake_up(&delayed_root->wait);
1063         }
1064 }
1065
1066 static int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1067                                       struct btrfs_root *root,
1068                                       struct btrfs_path *path,
1069                                       struct btrfs_delayed_node *node)
1070 {
1071         struct btrfs_key key;
1072         struct btrfs_inode_item *inode_item;
1073         struct extent_buffer *leaf;
1074         int ret;
1075
1076         mutex_lock(&node->mutex);
1077         if (!node->inode_dirty) {
1078                 mutex_unlock(&node->mutex);
1079                 return 0;
1080         }
1081
1082         key.objectid = node->inode_id;
1083         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1084         key.offset = 0;
1085         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1086         if (ret > 0) {
1087                 btrfs_release_path(path);
1088                 mutex_unlock(&node->mutex);
1089                 return -ENOENT;
1090         } else if (ret < 0) {
1091                 mutex_unlock(&node->mutex);
1092                 return ret;
1093         }
1094
1095         btrfs_unlock_up_safe(path, 1);
1096         leaf = path->nodes[0];
1097         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1098                                     struct btrfs_inode_item);
1099         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1100                             sizeof(struct btrfs_inode_item));
1101         btrfs_mark_buffer_dirty(leaf);
1102         btrfs_release_path(path);
1103
1104         btrfs_delayed_inode_release_metadata(root, node);
1105         btrfs_release_delayed_inode(node);
1106         mutex_unlock(&node->mutex);
1107
1108         return 0;
1109 }
1110
1111 /* Called when committing the transaction. */
1112 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1113                             struct btrfs_root *root)
1114 {
1115         struct btrfs_delayed_root *delayed_root;
1116         struct btrfs_delayed_node *curr_node, *prev_node;
1117         struct btrfs_path *path;
1118         struct btrfs_block_rsv *block_rsv;
1119         int ret = 0;
1120
1121         path = btrfs_alloc_path();
1122         if (!path)
1123                 return -ENOMEM;
1124         path->leave_spinning = 1;
1125
1126         block_rsv = trans->block_rsv;
1127         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1128
1129         delayed_root = btrfs_get_delayed_root(root);
1130
1131         curr_node = btrfs_first_delayed_node(delayed_root);
1132         while (curr_node) {
1133                 root = curr_node->root;
1134                 ret = btrfs_insert_delayed_items(trans, path, root,
1135                                                  curr_node);
1136                 if (!ret)
1137                         ret = btrfs_delete_delayed_items(trans, path, root,
1138                                                          curr_node);
1139                 if (!ret)
1140                         ret = btrfs_update_delayed_inode(trans, root, path,
1141                                                          curr_node);
1142                 if (ret) {
1143                         btrfs_release_delayed_node(curr_node);
1144                         break;
1145                 }
1146
1147                 prev_node = curr_node;
1148                 curr_node = btrfs_next_delayed_node(curr_node);
1149                 btrfs_release_delayed_node(prev_node);
1150         }
1151
1152         btrfs_free_path(path);
1153         trans->block_rsv = block_rsv;
1154         return ret;
1155 }
1156
1157 static int __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1158                                               struct btrfs_delayed_node *node)
1159 {
1160         struct btrfs_path *path;
1161         struct btrfs_block_rsv *block_rsv;
1162         int ret;
1163
1164         path = btrfs_alloc_path();
1165         if (!path)
1166                 return -ENOMEM;
1167         path->leave_spinning = 1;
1168
1169         block_rsv = trans->block_rsv;
1170         trans->block_rsv = &node->root->fs_info->delayed_block_rsv;
1171
1172         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1173         if (!ret)
1174                 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1175         if (!ret)
1176                 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1177         btrfs_free_path(path);
1178
1179         trans->block_rsv = block_rsv;
1180         return ret;
1181 }
1182
1183 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1184                                      struct inode *inode)
1185 {
1186         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1187         int ret;
1188
1189         if (!delayed_node)
1190                 return 0;
1191
1192         mutex_lock(&delayed_node->mutex);
1193         if (!delayed_node->count) {
1194                 mutex_unlock(&delayed_node->mutex);
1195                 btrfs_release_delayed_node(delayed_node);
1196                 return 0;
1197         }
1198         mutex_unlock(&delayed_node->mutex);
1199
1200         ret = __btrfs_commit_inode_delayed_items(trans, delayed_node);
1201         btrfs_release_delayed_node(delayed_node);
1202         return ret;
1203 }
1204
1205 void btrfs_remove_delayed_node(struct inode *inode)
1206 {
1207         struct btrfs_delayed_node *delayed_node;
1208
1209         delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1210         if (!delayed_node)
1211                 return;
1212
1213         BTRFS_I(inode)->delayed_node = NULL;
1214         btrfs_release_delayed_node(delayed_node);
1215 }
1216
1217 struct btrfs_async_delayed_node {
1218         struct btrfs_root *root;
1219         struct btrfs_delayed_node *delayed_node;
1220         struct btrfs_work work;
1221 };
1222
1223 static void btrfs_async_run_delayed_node_done(struct btrfs_work *work)
1224 {
1225         struct btrfs_async_delayed_node *async_node;
1226         struct btrfs_trans_handle *trans;
1227         struct btrfs_path *path;
1228         struct btrfs_delayed_node *delayed_node = NULL;
1229         struct btrfs_root *root;
1230         struct btrfs_block_rsv *block_rsv;
1231         unsigned long nr = 0;
1232         int need_requeue = 0;
1233         int ret;
1234
1235         async_node = container_of(work, struct btrfs_async_delayed_node, work);
1236
1237         path = btrfs_alloc_path();
1238         if (!path)
1239                 goto out;
1240         path->leave_spinning = 1;
1241
1242         delayed_node = async_node->delayed_node;
1243         root = delayed_node->root;
1244
1245         trans = btrfs_join_transaction(root);
1246         if (IS_ERR(trans))
1247                 goto free_path;
1248
1249         block_rsv = trans->block_rsv;
1250         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1251
1252         ret = btrfs_insert_delayed_items(trans, path, root, delayed_node);
1253         if (!ret)
1254                 ret = btrfs_delete_delayed_items(trans, path, root,
1255                                                  delayed_node);
1256
1257         if (!ret)
1258                 btrfs_update_delayed_inode(trans, root, path, delayed_node);
1259
1260         /*
1261          * Maybe new delayed items have been inserted, so we need requeue
1262          * the work. Besides that, we must dequeue the empty delayed nodes
1263          * to avoid the race between delayed items balance and the worker.
1264          * The race like this:
1265          *      Task1                           Worker thread
1266          *                                      count == 0, needn't requeue
1267          *                                        also needn't insert the
1268          *                                        delayed node into prepare
1269          *                                        list again.
1270          *      add lots of delayed items
1271          *      queue the delayed node
1272          *        already in the list,
1273          *        and not in the prepare
1274          *        list, it means the delayed
1275          *        node is being dealt with
1276          *        by the worker.
1277          *      do delayed items balance
1278          *        the delayed node is being
1279          *        dealt with by the worker
1280          *        now, just wait.
1281          *                                      the worker goto idle.
1282          * Task1 will sleep until the transaction is commited.
1283          */
1284         mutex_lock(&delayed_node->mutex);
1285         if (delayed_node->count)
1286                 need_requeue = 1;
1287         else
1288                 btrfs_dequeue_delayed_node(root->fs_info->delayed_root,
1289                                            delayed_node);
1290         mutex_unlock(&delayed_node->mutex);
1291
1292         nr = trans->blocks_used;
1293
1294         trans->block_rsv = block_rsv;
1295         btrfs_end_transaction_dmeta(trans, root);
1296         __btrfs_btree_balance_dirty(root, nr);
1297 free_path:
1298         btrfs_free_path(path);
1299 out:
1300         if (need_requeue)
1301                 btrfs_requeue_work(&async_node->work);
1302         else {
1303                 btrfs_release_prepared_delayed_node(delayed_node);
1304                 kfree(async_node);
1305         }
1306 }
1307
1308 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1309                                      struct btrfs_root *root, int all)
1310 {
1311         struct btrfs_async_delayed_node *async_node;
1312         struct btrfs_delayed_node *curr;
1313         int count = 0;
1314
1315 again:
1316         curr = btrfs_first_prepared_delayed_node(delayed_root);
1317         if (!curr)
1318                 return 0;
1319
1320         async_node = kmalloc(sizeof(*async_node), GFP_NOFS);
1321         if (!async_node) {
1322                 btrfs_release_prepared_delayed_node(curr);
1323                 return -ENOMEM;
1324         }
1325
1326         async_node->root = root;
1327         async_node->delayed_node = curr;
1328
1329         async_node->work.func = btrfs_async_run_delayed_node_done;
1330         async_node->work.flags = 0;
1331
1332         btrfs_queue_worker(&root->fs_info->delayed_workers, &async_node->work);
1333         count++;
1334
1335         if (all || count < 4)
1336                 goto again;
1337
1338         return 0;
1339 }
1340
1341 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1342 {
1343         struct btrfs_delayed_root *delayed_root;
1344         delayed_root = btrfs_get_delayed_root(root);
1345         WARN_ON(btrfs_first_delayed_node(delayed_root));
1346 }
1347
1348 void btrfs_balance_delayed_items(struct btrfs_root *root)
1349 {
1350         struct btrfs_delayed_root *delayed_root;
1351
1352         delayed_root = btrfs_get_delayed_root(root);
1353
1354         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1355                 return;
1356
1357         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1358                 int ret;
1359                 ret = btrfs_wq_run_delayed_node(delayed_root, root, 1);
1360                 if (ret)
1361                         return;
1362
1363                 wait_event_interruptible_timeout(
1364                                 delayed_root->wait,
1365                                 (atomic_read(&delayed_root->items) <
1366                                  BTRFS_DELAYED_BACKGROUND),
1367                                 HZ);
1368                 return;
1369         }
1370
1371         btrfs_wq_run_delayed_node(delayed_root, root, 0);
1372 }
1373
1374 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1375                                    struct btrfs_root *root, const char *name,
1376                                    int name_len, struct inode *dir,
1377                                    struct btrfs_disk_key *disk_key, u8 type,
1378                                    u64 index)
1379 {
1380         struct btrfs_delayed_node *delayed_node;
1381         struct btrfs_delayed_item *delayed_item;
1382         struct btrfs_dir_item *dir_item;
1383         int ret;
1384
1385         delayed_node = btrfs_get_or_create_delayed_node(dir);
1386         if (IS_ERR(delayed_node))
1387                 return PTR_ERR(delayed_node);
1388
1389         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1390         if (!delayed_item) {
1391                 ret = -ENOMEM;
1392                 goto release_node;
1393         }
1394
1395         delayed_item->key.objectid = btrfs_ino(dir);
1396         btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1397         delayed_item->key.offset = index;
1398
1399         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1400         dir_item->location = *disk_key;
1401         dir_item->transid = cpu_to_le64(trans->transid);
1402         dir_item->data_len = 0;
1403         dir_item->name_len = cpu_to_le16(name_len);
1404         dir_item->type = type;
1405         memcpy((char *)(dir_item + 1), name, name_len);
1406
1407         ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1408         /*
1409          * we have reserved enough space when we start a new transaction,
1410          * so reserving metadata failure is impossible
1411          */
1412         BUG_ON(ret);
1413
1414
1415         mutex_lock(&delayed_node->mutex);
1416         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1417         if (unlikely(ret)) {
1418                 printk(KERN_ERR "err add delayed dir index item(name: %s) into "
1419                                 "the insertion tree of the delayed node"
1420                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1421                                 name,
1422                                 (unsigned long long)delayed_node->root->objectid,
1423                                 (unsigned long long)delayed_node->inode_id,
1424                                 ret);
1425                 BUG();
1426         }
1427         mutex_unlock(&delayed_node->mutex);
1428
1429 release_node:
1430         btrfs_release_delayed_node(delayed_node);
1431         return ret;
1432 }
1433
1434 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1435                                                struct btrfs_delayed_node *node,
1436                                                struct btrfs_key *key)
1437 {
1438         struct btrfs_delayed_item *item;
1439
1440         mutex_lock(&node->mutex);
1441         item = __btrfs_lookup_delayed_insertion_item(node, key);
1442         if (!item) {
1443                 mutex_unlock(&node->mutex);
1444                 return 1;
1445         }
1446
1447         btrfs_delayed_item_release_metadata(root, item);
1448         btrfs_release_delayed_item(item);
1449         mutex_unlock(&node->mutex);
1450         return 0;
1451 }
1452
1453 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1454                                    struct btrfs_root *root, struct inode *dir,
1455                                    u64 index)
1456 {
1457         struct btrfs_delayed_node *node;
1458         struct btrfs_delayed_item *item;
1459         struct btrfs_key item_key;
1460         int ret;
1461
1462         node = btrfs_get_or_create_delayed_node(dir);
1463         if (IS_ERR(node))
1464                 return PTR_ERR(node);
1465
1466         item_key.objectid = btrfs_ino(dir);
1467         btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1468         item_key.offset = index;
1469
1470         ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1471         if (!ret)
1472                 goto end;
1473
1474         item = btrfs_alloc_delayed_item(0);
1475         if (!item) {
1476                 ret = -ENOMEM;
1477                 goto end;
1478         }
1479
1480         item->key = item_key;
1481
1482         ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1483         /*
1484          * we have reserved enough space when we start a new transaction,
1485          * so reserving metadata failure is impossible.
1486          */
1487         BUG_ON(ret);
1488
1489         mutex_lock(&node->mutex);
1490         ret = __btrfs_add_delayed_deletion_item(node, item);
1491         if (unlikely(ret)) {
1492                 printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1493                                 "into the deletion tree of the delayed node"
1494                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1495                                 (unsigned long long)index,
1496                                 (unsigned long long)node->root->objectid,
1497                                 (unsigned long long)node->inode_id,
1498                                 ret);
1499                 BUG();
1500         }
1501         mutex_unlock(&node->mutex);
1502 end:
1503         btrfs_release_delayed_node(node);
1504         return ret;
1505 }
1506
1507 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1508 {
1509         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1510
1511         if (!delayed_node)
1512                 return -ENOENT;
1513
1514         /*
1515          * Since we have held i_mutex of this directory, it is impossible that
1516          * a new directory index is added into the delayed node and index_cnt
1517          * is updated now. So we needn't lock the delayed node.
1518          */
1519         if (!delayed_node->index_cnt) {
1520                 btrfs_release_delayed_node(delayed_node);
1521                 return -EINVAL;
1522         }
1523
1524         BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1525         btrfs_release_delayed_node(delayed_node);
1526         return 0;
1527 }
1528
1529 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1530                              struct list_head *del_list)
1531 {
1532         struct btrfs_delayed_node *delayed_node;
1533         struct btrfs_delayed_item *item;
1534
1535         delayed_node = btrfs_get_delayed_node(inode);
1536         if (!delayed_node)
1537                 return;
1538
1539         mutex_lock(&delayed_node->mutex);
1540         item = __btrfs_first_delayed_insertion_item(delayed_node);
1541         while (item) {
1542                 atomic_inc(&item->refs);
1543                 list_add_tail(&item->readdir_list, ins_list);
1544                 item = __btrfs_next_delayed_item(item);
1545         }
1546
1547         item = __btrfs_first_delayed_deletion_item(delayed_node);
1548         while (item) {
1549                 atomic_inc(&item->refs);
1550                 list_add_tail(&item->readdir_list, del_list);
1551                 item = __btrfs_next_delayed_item(item);
1552         }
1553         mutex_unlock(&delayed_node->mutex);
1554         /*
1555          * This delayed node is still cached in the btrfs inode, so refs
1556          * must be > 1 now, and we needn't check it is going to be freed
1557          * or not.
1558          *
1559          * Besides that, this function is used to read dir, we do not
1560          * insert/delete delayed items in this period. So we also needn't
1561          * requeue or dequeue this delayed node.
1562          */
1563         atomic_dec(&delayed_node->refs);
1564 }
1565
1566 void btrfs_put_delayed_items(struct list_head *ins_list,
1567                              struct list_head *del_list)
1568 {
1569         struct btrfs_delayed_item *curr, *next;
1570
1571         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1572                 list_del(&curr->readdir_list);
1573                 if (atomic_dec_and_test(&curr->refs))
1574                         kfree(curr);
1575         }
1576
1577         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1578                 list_del(&curr->readdir_list);
1579                 if (atomic_dec_and_test(&curr->refs))
1580                         kfree(curr);
1581         }
1582 }
1583
1584 int btrfs_should_delete_dir_index(struct list_head *del_list,
1585                                   u64 index)
1586 {
1587         struct btrfs_delayed_item *curr, *next;
1588         int ret;
1589
1590         if (list_empty(del_list))
1591                 return 0;
1592
1593         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1594                 if (curr->key.offset > index)
1595                         break;
1596
1597                 list_del(&curr->readdir_list);
1598                 ret = (curr->key.offset == index);
1599
1600                 if (atomic_dec_and_test(&curr->refs))
1601                         kfree(curr);
1602
1603                 if (ret)
1604                         return 1;
1605                 else
1606                         continue;
1607         }
1608         return 0;
1609 }
1610
1611 /*
1612  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1613  *
1614  */
1615 int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent,
1616                                     filldir_t filldir,
1617                                     struct list_head *ins_list)
1618 {
1619         struct btrfs_dir_item *di;
1620         struct btrfs_delayed_item *curr, *next;
1621         struct btrfs_key location;
1622         char *name;
1623         int name_len;
1624         int over = 0;
1625         unsigned char d_type;
1626
1627         if (list_empty(ins_list))
1628                 return 0;
1629
1630         /*
1631          * Changing the data of the delayed item is impossible. So
1632          * we needn't lock them. And we have held i_mutex of the
1633          * directory, nobody can delete any directory indexes now.
1634          */
1635         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1636                 list_del(&curr->readdir_list);
1637
1638                 if (curr->key.offset < filp->f_pos) {
1639                         if (atomic_dec_and_test(&curr->refs))
1640                                 kfree(curr);
1641                         continue;
1642                 }
1643
1644                 filp->f_pos = curr->key.offset;
1645
1646                 di = (struct btrfs_dir_item *)curr->data;
1647                 name = (char *)(di + 1);
1648                 name_len = le16_to_cpu(di->name_len);
1649
1650                 d_type = btrfs_filetype_table[di->type];
1651                 btrfs_disk_key_to_cpu(&location, &di->location);
1652
1653                 over = filldir(dirent, name, name_len, curr->key.offset,
1654                                location.objectid, d_type);
1655
1656                 if (atomic_dec_and_test(&curr->refs))
1657                         kfree(curr);
1658
1659                 if (over)
1660                         return 1;
1661         }
1662         return 0;
1663 }
1664
1665 BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item,
1666                          generation, 64);
1667 BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item,
1668                          sequence, 64);
1669 BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item,
1670                          transid, 64);
1671 BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64);
1672 BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item,
1673                          nbytes, 64);
1674 BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item,
1675                          block_group, 64);
1676 BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32);
1677 BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32);
1678 BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32);
1679 BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32);
1680 BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64);
1681 BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64);
1682
1683 BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64);
1684 BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32);
1685
1686 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1687                                   struct btrfs_inode_item *inode_item,
1688                                   struct inode *inode)
1689 {
1690         btrfs_set_stack_inode_uid(inode_item, inode->i_uid);
1691         btrfs_set_stack_inode_gid(inode_item, inode->i_gid);
1692         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1693         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1694         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1695         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1696         btrfs_set_stack_inode_generation(inode_item,
1697                                          BTRFS_I(inode)->generation);
1698         btrfs_set_stack_inode_sequence(inode_item, BTRFS_I(inode)->sequence);
1699         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1700         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1701         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1702         btrfs_set_stack_inode_block_group(inode_item, 0);
1703
1704         btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1705                                      inode->i_atime.tv_sec);
1706         btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1707                                       inode->i_atime.tv_nsec);
1708
1709         btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1710                                      inode->i_mtime.tv_sec);
1711         btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1712                                       inode->i_mtime.tv_nsec);
1713
1714         btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1715                                      inode->i_ctime.tv_sec);
1716         btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1717                                       inode->i_ctime.tv_nsec);
1718 }
1719
1720 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1721 {
1722         struct btrfs_delayed_node *delayed_node;
1723         struct btrfs_inode_item *inode_item;
1724         struct btrfs_timespec *tspec;
1725
1726         delayed_node = btrfs_get_delayed_node(inode);
1727         if (!delayed_node)
1728                 return -ENOENT;
1729
1730         mutex_lock(&delayed_node->mutex);
1731         if (!delayed_node->inode_dirty) {
1732                 mutex_unlock(&delayed_node->mutex);
1733                 btrfs_release_delayed_node(delayed_node);
1734                 return -ENOENT;
1735         }
1736
1737         inode_item = &delayed_node->inode_item;
1738
1739         inode->i_uid = btrfs_stack_inode_uid(inode_item);
1740         inode->i_gid = btrfs_stack_inode_gid(inode_item);
1741         btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1742         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1743         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1744         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1745         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1746         BTRFS_I(inode)->sequence = btrfs_stack_inode_sequence(inode_item);
1747         inode->i_rdev = 0;
1748         *rdev = btrfs_stack_inode_rdev(inode_item);
1749         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1750
1751         tspec = btrfs_inode_atime(inode_item);
1752         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1753         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1754
1755         tspec = btrfs_inode_mtime(inode_item);
1756         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1757         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1758
1759         tspec = btrfs_inode_ctime(inode_item);
1760         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1761         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1762
1763         inode->i_generation = BTRFS_I(inode)->generation;
1764         BTRFS_I(inode)->index_cnt = (u64)-1;
1765
1766         mutex_unlock(&delayed_node->mutex);
1767         btrfs_release_delayed_node(delayed_node);
1768         return 0;
1769 }
1770
1771 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1772                                struct btrfs_root *root, struct inode *inode)
1773 {
1774         struct btrfs_delayed_node *delayed_node;
1775         int ret = 0;
1776
1777         delayed_node = btrfs_get_or_create_delayed_node(inode);
1778         if (IS_ERR(delayed_node))
1779                 return PTR_ERR(delayed_node);
1780
1781         mutex_lock(&delayed_node->mutex);
1782         if (delayed_node->inode_dirty) {
1783                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1784                 goto release_node;
1785         }
1786
1787         ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1788                                                    delayed_node);
1789         if (ret)
1790                 goto release_node;
1791
1792         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1793         delayed_node->inode_dirty = 1;
1794         delayed_node->count++;
1795         atomic_inc(&root->fs_info->delayed_root->items);
1796 release_node:
1797         mutex_unlock(&delayed_node->mutex);
1798         btrfs_release_delayed_node(delayed_node);
1799         return ret;
1800 }
1801
1802 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1803 {
1804         struct btrfs_root *root = delayed_node->root;
1805         struct btrfs_delayed_item *curr_item, *prev_item;
1806
1807         mutex_lock(&delayed_node->mutex);
1808         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1809         while (curr_item) {
1810                 btrfs_delayed_item_release_metadata(root, curr_item);
1811                 prev_item = curr_item;
1812                 curr_item = __btrfs_next_delayed_item(prev_item);
1813                 btrfs_release_delayed_item(prev_item);
1814         }
1815
1816         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1817         while (curr_item) {
1818                 btrfs_delayed_item_release_metadata(root, curr_item);
1819                 prev_item = curr_item;
1820                 curr_item = __btrfs_next_delayed_item(prev_item);
1821                 btrfs_release_delayed_item(prev_item);
1822         }
1823
1824         if (delayed_node->inode_dirty) {
1825                 btrfs_delayed_inode_release_metadata(root, delayed_node);
1826                 btrfs_release_delayed_inode(delayed_node);
1827         }
1828         mutex_unlock(&delayed_node->mutex);
1829 }
1830
1831 void btrfs_kill_delayed_inode_items(struct inode *inode)
1832 {
1833         struct btrfs_delayed_node *delayed_node;
1834
1835         delayed_node = btrfs_get_delayed_node(inode);
1836         if (!delayed_node)
1837                 return;
1838
1839         __btrfs_kill_delayed_node(delayed_node);
1840         btrfs_release_delayed_node(delayed_node);
1841 }
1842
1843 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1844 {
1845         u64 inode_id = 0;
1846         struct btrfs_delayed_node *delayed_nodes[8];
1847         int i, n;
1848
1849         while (1) {
1850                 spin_lock(&root->inode_lock);
1851                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1852                                            (void **)delayed_nodes, inode_id,
1853                                            ARRAY_SIZE(delayed_nodes));
1854                 if (!n) {
1855                         spin_unlock(&root->inode_lock);
1856                         break;
1857                 }
1858
1859                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1860
1861                 for (i = 0; i < n; i++)
1862                         atomic_inc(&delayed_nodes[i]->refs);
1863                 spin_unlock(&root->inode_lock);
1864
1865                 for (i = 0; i < n; i++) {
1866                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1867                         btrfs_release_delayed_node(delayed_nodes[i]);
1868                 }
1869         }
1870 }