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