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