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("btrfs_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         memset(&delayed_node->inode_item, 0, sizeof(delayed_node->inode_item));
  66 }
  67
  68 static inline int btrfs_is_continuous_delayed_item(
  69                                         struct btrfs_delayed_item *item1,
  70                                         struct btrfs_delayed_item *item2)
  71 {
  72         if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  73             item1->key.objectid == item2->key.objectid &&
  74             item1->key.type == item2->key.type &&
  75             item1->key.offset + 1 == item2->key.offset)
  76                 return 1;
  77         return 0;
  78 }
  79
  80 static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
  81                                                         struct btrfs_root *root)
  82 {
  83         return root->fs_info->delayed_root;
  84 }
  85
  86 static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
  87 {
  88         struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
  89         struct btrfs_root *root = btrfs_inode->root;
  90         u64 ino = btrfs_ino(inode);
  91         struct btrfs_delayed_node *node;
  92
  93         node = ACCESS_ONCE(btrfs_inode->delayed_node);
  94         if (node) {
  95                 atomic_inc(&node->refs);
  96                 return node;
  97         }
  98
  99         spin_lock(&root->inode_lock);
 100         node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
 101         if (node) {
 102                 if (btrfs_inode->delayed_node) {
 103                         atomic_inc(&node->refs);        /* can be accessed */
 104                         BUG_ON(btrfs_inode->delayed_node != node);
 105                         spin_unlock(&root->inode_lock);
 106                         return node;
 107                 }
 108                 btrfs_inode->delayed_node = node;
 109                 atomic_inc(&node->refs);        /* can be accessed */
 110                 atomic_inc(&node->refs);        /* cached in the inode */
 111                 spin_unlock(&root->inode_lock);
 112                 return node;
 113         }
 114         spin_unlock(&root->inode_lock);
 115
 116         return NULL;
 117 }
 118
 119 /* Will return either the node or PTR_ERR(-ENOMEM) */
 120 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 121                                                         struct inode *inode)
 122 {
 123         struct btrfs_delayed_node *node;
 124         struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
 125         struct btrfs_root *root = btrfs_inode->root;
 126         u64 ino = btrfs_ino(inode);
 127         int ret;
 128
 129 again:
 130         node = btrfs_get_delayed_node(inode);
 131         if (node)
 132                 return node;
 133
 134         node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
 135         if (!node)
 136                 return ERR_PTR(-ENOMEM);
 137         btrfs_init_delayed_node(node, root, ino);
 138
 139         atomic_inc(&node->refs);        /* cached in the btrfs inode */
 140         atomic_inc(&node->refs);        /* can be accessed */
 141
 142         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 143         if (ret) {
 144                 kmem_cache_free(delayed_node_cache, node);
 145                 return ERR_PTR(ret);
 146         }
 147
 148         spin_lock(&root->inode_lock);
 149         ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 150         if (ret == -EEXIST) {
 151                 kmem_cache_free(delayed_node_cache, node);
 152                 spin_unlock(&root->inode_lock);
 153                 radix_tree_preload_end();
 154                 goto again;
 155         }
 156         btrfs_inode->delayed_node = node;
 157         spin_unlock(&root->inode_lock);
 158         radix_tree_preload_end();
 159
 160         return node;
 161 }
 162
 163 /*
 164  * Call it when holding delayed_node->mutex
 165  *
 166  * If mod = 1, add this node into the prepared list.
 167  */
 168 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 169                                      struct btrfs_delayed_node *node,
 170                                      int mod)
 171 {
 172         spin_lock(&root->lock);
 173         if (node->in_list) {
 174                 if (!list_empty(&node->p_list))
 175                         list_move_tail(&node->p_list, &root->prepare_list);
 176                 else if (mod)
 177                         list_add_tail(&node->p_list, &root->prepare_list);
 178         } else {
 179                 list_add_tail(&node->n_list, &root->node_list);
 180                 list_add_tail(&node->p_list, &root->prepare_list);
 181                 atomic_inc(&node->refs);        /* inserted into list */
 182                 root->nodes++;
 183                 node->in_list = 1;
 184         }
 185         spin_unlock(&root->lock);
 186 }
 187
 188 /* Call it when holding delayed_node->mutex */
 189 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 190                                        struct btrfs_delayed_node *node)
 191 {
 192         spin_lock(&root->lock);
 193         if (node->in_list) {
 194                 root->nodes--;
 195                 atomic_dec(&node->refs);        /* not in the list */
 196                 list_del_init(&node->n_list);
 197                 if (!list_empty(&node->p_list))
 198                         list_del_init(&node->p_list);
 199                 node->in_list = 0;
 200         }
 201         spin_unlock(&root->lock);
 202 }
 203
 204 struct btrfs_delayed_node *btrfs_first_delayed_node(
 205                         struct btrfs_delayed_root *delayed_root)
 206 {
 207         struct list_head *p;
 208         struct btrfs_delayed_node *node = NULL;
 209
 210         spin_lock(&delayed_root->lock);
 211         if (list_empty(&delayed_root->node_list))
 212                 goto out;
 213
 214         p = delayed_root->node_list.next;
 215         node = list_entry(p, struct btrfs_delayed_node, n_list);
 216         atomic_inc(&node->refs);
 217 out:
 218         spin_unlock(&delayed_root->lock);
 219
 220         return node;
 221 }
 222
 223 struct btrfs_delayed_node *btrfs_next_delayed_node(
 224                                                 struct btrfs_delayed_node *node)
 225 {
 226         struct btrfs_delayed_root *delayed_root;
 227         struct list_head *p;
 228         struct btrfs_delayed_node *next = NULL;
 229
 230         delayed_root = node->root->fs_info->delayed_root;
 231         spin_lock(&delayed_root->lock);
 232         if (!node->in_list) {   /* not in the list */
 233                 if (list_empty(&delayed_root->node_list))
 234                         goto out;
 235                 p = delayed_root->node_list.next;
 236         } else if (list_is_last(&node->n_list, &delayed_root->node_list))
 237                 goto out;
 238         else
 239                 p = node->n_list.next;
 240
 241         next = list_entry(p, struct btrfs_delayed_node, n_list);
 242         atomic_inc(&next->refs);
 243 out:
 244         spin_unlock(&delayed_root->lock);
 245
 246         return next;
 247 }
 248
 249 static void __btrfs_release_delayed_node(
 250                                 struct btrfs_delayed_node *delayed_node,
 251                                 int mod)
 252 {
 253         struct btrfs_delayed_root *delayed_root;
 254
 255         if (!delayed_node)
 256                 return;
 257
 258         delayed_root = delayed_node->root->fs_info->delayed_root;
 259
 260         mutex_lock(&delayed_node->mutex);
 261         if (delayed_node->count)
 262                 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
 263         else
 264                 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 265         mutex_unlock(&delayed_node->mutex);
 266
 267         if (atomic_dec_and_test(&delayed_node->refs)) {
 268                 struct btrfs_root *root = delayed_node->root;
 269                 spin_lock(&root->inode_lock);
 270                 if (atomic_read(&delayed_node->refs) == 0) {
 271                         radix_tree_delete(&root->delayed_nodes_tree,
 272                                           delayed_node->inode_id);
 273                         kmem_cache_free(delayed_node_cache, delayed_node);
 274                 }
 275                 spin_unlock(&root->inode_lock);
 276         }
 277 }
 278
 279 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
 280 {
 281         __btrfs_release_delayed_node(node, 0);
 282 }
 283
 284 struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
 285                                         struct btrfs_delayed_root *delayed_root)
 286 {
 287         struct list_head *p;
 288         struct btrfs_delayed_node *node = NULL;
 289
 290         spin_lock(&delayed_root->lock);
 291         if (list_empty(&delayed_root->prepare_list))
 292                 goto out;
 293
 294         p = delayed_root->prepare_list.next;
 295         list_del_init(p);
 296         node = list_entry(p, struct btrfs_delayed_node, p_list);
 297         atomic_inc(&node->refs);
 298 out:
 299         spin_unlock(&delayed_root->lock);
 300
 301         return node;
 302 }
 303
 304 static inline void btrfs_release_prepared_delayed_node(
 305                                         struct btrfs_delayed_node *node)
 306 {
 307         __btrfs_release_delayed_node(node, 1);
 308 }
 309
 310 struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
 311 {
 312         struct btrfs_delayed_item *item;
 313         item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
 314         if (item) {
 315                 item->data_len = data_len;
 316                 item->ins_or_del = 0;
 317                 item->bytes_reserved = 0;
 318                 item->delayed_node = NULL;
 319                 atomic_set(&item->refs, 1);
 320         }
 321         return item;
 322 }
 323
 324 /*
 325  * __btrfs_lookup_delayed_item - look up the delayed item by key
 326  * @delayed_node: pointer to the delayed node
 327  * @key:          the key to look up
 328  * @prev:         used to store the prev item if the right item isn't found
 329  * @next:         used to store the next item if the right item isn't found
 330  *
 331  * Note: if we don't find the right item, we will return the prev item and
 332  * the next item.
 333  */
 334 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 335                                 struct rb_root *root,
 336                                 struct btrfs_key *key,
 337                                 struct btrfs_delayed_item **prev,
 338                                 struct btrfs_delayed_item **next)
 339 {
 340         struct rb_node *node, *prev_node = NULL;
 341         struct btrfs_delayed_item *delayed_item = NULL;
 342         int ret = 0;
 343
 344         node = root->rb_node;
 345
 346         while (node) {
 347                 delayed_item = rb_entry(node, struct btrfs_delayed_item,
 348                                         rb_node);
 349                 prev_node = node;
 350                 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
 351                 if (ret < 0)
 352                         node = node->rb_right;
 353                 else if (ret > 0)
 354                         node = node->rb_left;
 355                 else
 356                         return delayed_item;
 357         }
 358
 359         if (prev) {
 360                 if (!prev_node)
 361                         *prev = NULL;
 362                 else if (ret < 0)
 363                         *prev = delayed_item;
 364                 else if ((node = rb_prev(prev_node)) != NULL) {
 365                         *prev = rb_entry(node, struct btrfs_delayed_item,
 366                                          rb_node);
 367                 } else
 368                         *prev = NULL;
 369         }
 370
 371         if (next) {
 372                 if (!prev_node)
 373                         *next = NULL;
 374                 else if (ret > 0)
 375                         *next = delayed_item;
 376                 else if ((node = rb_next(prev_node)) != NULL) {
 377                         *next = rb_entry(node, struct btrfs_delayed_item,
 378                                          rb_node);
 379                 } else
 380                         *next = NULL;
 381         }
 382         return NULL;
 383 }
 384
 385 struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
 386                                         struct btrfs_delayed_node *delayed_node,
 387                                         struct btrfs_key *key)
 388 {
 389         struct btrfs_delayed_item *item;
 390
 391         item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 392                                            NULL, NULL);
 393         return item;
 394 }
 395
 396 struct btrfs_delayed_item *__btrfs_lookup_delayed_deletion_item(
 397                                         struct btrfs_delayed_node *delayed_node,
 398                                         struct btrfs_key *key)
 399 {
 400         struct btrfs_delayed_item *item;
 401
 402         item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
 403                                            NULL, NULL);
 404         return item;
 405 }
 406
 407 struct btrfs_delayed_item *__btrfs_search_delayed_insertion_item(
 408                                         struct btrfs_delayed_node *delayed_node,
 409                                         struct btrfs_key *key)
 410 {
 411         struct btrfs_delayed_item *item, *next;
 412
 413         item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 414                                            NULL, &next);
 415         if (!item)
 416                 item = next;
 417
 418         return item;
 419 }
 420
 421 struct btrfs_delayed_item *__btrfs_search_delayed_deletion_item(
 422                                         struct btrfs_delayed_node *delayed_node,
 423                                         struct btrfs_key *key)
 424 {
 425         struct btrfs_delayed_item *item, *next;
 426
 427         item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
 428                                            NULL, &next);
 429         if (!item)
 430                 item = next;
 431
 432         return item;
 433 }
 434
 435 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
 436                                     struct btrfs_delayed_item *ins,
 437                                     int action)
 438 {
 439         struct rb_node **p, *node;
 440         struct rb_node *parent_node = NULL;
 441         struct rb_root *root;
 442         struct btrfs_delayed_item *item;
 443         int cmp;
 444
 445         if (action == BTRFS_DELAYED_INSERTION_ITEM)
 446                 root = &delayed_node->ins_root;
 447         else if (action == BTRFS_DELAYED_DELETION_ITEM)
 448                 root = &delayed_node->del_root;
 449         else
 450                 BUG();
 451         p = &root->rb_node;
 452         node = &ins->rb_node;
 453
 454         while (*p) {
 455                 parent_node = *p;
 456                 item = rb_entry(parent_node, struct btrfs_delayed_item,
 457                                  rb_node);
 458
 459                 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
 460                 if (cmp < 0)
 461                         p = &(*p)->rb_right;
 462                 else if (cmp > 0)
 463                         p = &(*p)->rb_left;
 464                 else
 465                         return -EEXIST;
 466         }
 467
 468         rb_link_node(node, parent_node, p);
 469         rb_insert_color(node, root);
 470         ins->delayed_node = delayed_node;
 471         ins->ins_or_del = action;
 472
 473         if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
 474             action == BTRFS_DELAYED_INSERTION_ITEM &&
 475             ins->key.offset >= delayed_node->index_cnt)
 476                         delayed_node->index_cnt = ins->key.offset + 1;
 477
 478         delayed_node->count++;
 479         atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
 480         return 0;
 481 }
 482
 483 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
 484                                               struct btrfs_delayed_item *item)
 485 {
 486         return __btrfs_add_delayed_item(node, item,
 487                                         BTRFS_DELAYED_INSERTION_ITEM);
 488 }
 489
 490 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
 491                                              struct btrfs_delayed_item *item)
 492 {
 493         return __btrfs_add_delayed_item(node, item,
 494                                         BTRFS_DELAYED_DELETION_ITEM);
 495 }
 496
 497 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 498 {
 499         struct rb_root *root;
 500         struct btrfs_delayed_root *delayed_root;
 501
 502         delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
 503
 504         BUG_ON(!delayed_root);
 505         BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
 506                delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
 507
 508         if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
 509                 root = &delayed_item->delayed_node->ins_root;
 510         else
 511                 root = &delayed_item->delayed_node->del_root;
 512
 513         rb_erase(&delayed_item->rb_node, root);
 514         delayed_item->delayed_node->count--;
 515         if (atomic_dec_return(&delayed_root->items) <
 516             BTRFS_DELAYED_BACKGROUND &&
 517             waitqueue_active(&delayed_root->wait))
 518                 wake_up(&delayed_root->wait);
 519 }
 520
 521 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 522 {
 523         if (item) {
 524                 __btrfs_remove_delayed_item(item);
 525                 if (atomic_dec_and_test(&item->refs))
 526                         kfree(item);
 527         }
 528 }
 529
 530 struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 531                                         struct btrfs_delayed_node *delayed_node)
 532 {
 533         struct rb_node *p;
 534         struct btrfs_delayed_item *item = NULL;
 535
 536         p = rb_first(&delayed_node->ins_root);
 537         if (p)
 538                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 539
 540         return item;
 541 }
 542
 543 struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 544                                         struct btrfs_delayed_node *delayed_node)
 545 {
 546         struct rb_node *p;
 547         struct btrfs_delayed_item *item = NULL;
 548
 549         p = rb_first(&delayed_node->del_root);
 550         if (p)
 551                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 552
 553         return item;
 554 }
 555
 556 struct btrfs_delayed_item *__btrfs_next_delayed_item(
 557                                                 struct btrfs_delayed_item *item)
 558 {
 559         struct rb_node *p;
 560         struct btrfs_delayed_item *next = NULL;
 561
 562         p = rb_next(&item->rb_node);
 563         if (p)
 564                 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
 565
 566         return next;
 567 }
 568
 569 static inline struct btrfs_root *btrfs_get_fs_root(struct btrfs_root *root,
 570                                                    u64 root_id)
 571 {
 572         struct btrfs_key root_key;
 573
 574         if (root->objectid == root_id)
 575                 return root;
 576
 577         root_key.objectid = root_id;
 578         root_key.type = BTRFS_ROOT_ITEM_KEY;
 579         root_key.offset = (u64)-1;
 580         return btrfs_read_fs_root_no_name(root->fs_info, &root_key);
 581 }
 582
 583 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 584                                                struct btrfs_root *root,
 585                                                struct btrfs_delayed_item *item)
 586 {
 587         struct btrfs_block_rsv *src_rsv;
 588         struct btrfs_block_rsv *dst_rsv;
 589         u64 num_bytes;
 590         int ret;
 591
 592         if (!trans->bytes_reserved)
 593                 return 0;
 594
 595         src_rsv = trans->block_rsv;
 596         dst_rsv = &root->fs_info->delayed_block_rsv;
 597
 598         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 599         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 600         if (!ret) {
 601                 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
 602                                               item->key.objectid,
 603                                               num_bytes, 1);
 604                 item->bytes_reserved = num_bytes;
 605         }
 606
 607         return ret;
 608 }
 609
 610 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 611                                                 struct btrfs_delayed_item *item)
 612 {
 613         struct btrfs_block_rsv *rsv;
 614
 615         if (!item->bytes_reserved)
 616                 return;
 617
 618         rsv = &root->fs_info->delayed_block_rsv;
 619         trace_btrfs_space_reservation(root->fs_info, "delayed_item",
 620                                       item->key.objectid, item->bytes_reserved,
 621                                       0);
 622         btrfs_block_rsv_release(root, rsv,
 623                                 item->bytes_reserved);
 624 }
 625
 626 static int btrfs_delayed_inode_reserve_metadata(
 627                                         struct btrfs_trans_handle *trans,
 628                                         struct btrfs_root *root,
 629                                         struct inode *inode,
 630                                         struct btrfs_delayed_node *node)
 631 {
 632         struct btrfs_block_rsv *src_rsv;
 633         struct btrfs_block_rsv *dst_rsv;
 634         u64 num_bytes;
 635         int ret;
 636         bool release = false;
 637
 638         src_rsv = trans->block_rsv;
 639         dst_rsv = &root->fs_info->delayed_block_rsv;
 640
 641         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 642
 643         /*
 644          * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 645          * which doesn't reserve space for speed.  This is a problem since we
 646          * still need to reserve space for this update, so try to reserve the
 647          * space.
 648          *
 649          * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 650          * we're accounted for.
 651          */
 652         if (!src_rsv || (!trans->bytes_reserved &&
 653                          src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
 654                 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 655                                           BTRFS_RESERVE_NO_FLUSH);
 656                 /*
 657                  * Since we're under a transaction reserve_metadata_bytes could
 658                  * try to commit the transaction which will make it return
 659                  * EAGAIN to make us stop the transaction we have, so return
 660                  * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 661                  */
 662                 if (ret == -EAGAIN)
 663                         ret = -ENOSPC;
 664                 if (!ret) {
 665                         node->bytes_reserved = num_bytes;
 666                         trace_btrfs_space_reservation(root->fs_info,
 667                                                       "delayed_inode",
 668                                                       btrfs_ino(inode),
 669                                                       num_bytes, 1);
 670                 }
 671                 return ret;
 672         } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
 673                 spin_lock(&BTRFS_I(inode)->lock);
 674                 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
 675                                        &BTRFS_I(inode)->runtime_flags)) {
 676                         spin_unlock(&BTRFS_I(inode)->lock);
 677                         release = true;
 678                         goto migrate;
 679                 }
 680                 spin_unlock(&BTRFS_I(inode)->lock);
 681
 682                 /* Ok we didn't have space pre-reserved.  This shouldn't happen
 683                  * too often but it can happen if we do delalloc to an existing
 684                  * inode which gets dirtied because of the time update, and then
 685                  * isn't touched again until after the transaction commits and
 686                  * then we try to write out the data.  First try to be nice and
 687                  * reserve something strictly for us.  If not be a pain and try
 688                  * to steal from the delalloc block rsv.
 689                  */
 690                 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 691                                           BTRFS_RESERVE_NO_FLUSH);
 692                 if (!ret)
 693                         goto out;
 694
 695                 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 696                 if (!ret)
 697                         goto out;
 698
 699                 /*
 700                  * Ok this is a problem, let's just steal from the global rsv
 701                  * since this really shouldn't happen that often.
 702                  */
 703                 WARN_ON(1);
 704                 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
 705                                               dst_rsv, num_bytes);
 706                 goto out;
 707         }
 708
 709 migrate:
 710         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 711
 712 out:
 713         /*
 714          * Migrate only takes a reservation, it doesn't touch the size of the
 715          * block_rsv.  This is to simplify people who don't normally have things
 716          * migrated from their block rsv.  If they go to release their
 717          * reservation, that will decrease the size as well, so if migrate
 718          * reduced size we'd end up with a negative size.  But for the
 719          * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
 720          * but we could in fact do this reserve/migrate dance several times
 721          * between the time we did the original reservation and we'd clean it
 722          * up.  So to take care of this, release the space for the meta
 723          * reservation here.  I think it may be time for a documentation page on
 724          * how block rsvs. work.
 725          */
 726         if (!ret) {
 727                 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
 728                                               btrfs_ino(inode), num_bytes, 1);
 729                 node->bytes_reserved = num_bytes;
 730         }
 731
 732         if (release) {
 733                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
 734                                               btrfs_ino(inode), num_bytes, 0);
 735                 btrfs_block_rsv_release(root, src_rsv, num_bytes);
 736         }
 737
 738         return ret;
 739 }
 740
 741 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
 742                                                 struct btrfs_delayed_node *node)
 743 {
 744         struct btrfs_block_rsv *rsv;
 745
 746         if (!node->bytes_reserved)
 747                 return;
 748
 749         rsv = &root->fs_info->delayed_block_rsv;
 750         trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
 751                                       node->inode_id, node->bytes_reserved, 0);
 752         btrfs_block_rsv_release(root, rsv,
 753                                 node->bytes_reserved);
 754         node->bytes_reserved = 0;
 755 }
 756
 757 /*
 758  * This helper will insert some continuous items into the same leaf according
 759  * to the free space of the leaf.
 760  */
 761 static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans,
 762                                 struct btrfs_root *root,
 763                                 struct btrfs_path *path,
 764                                 struct btrfs_delayed_item *item)
 765 {
 766         struct btrfs_delayed_item *curr, *next;
 767         int free_space;
 768         int total_data_size = 0, total_size = 0;
 769         struct extent_buffer *leaf;
 770         char *data_ptr;
 771         struct btrfs_key *keys;
 772         u32 *data_size;
 773         struct list_head head;
 774         int slot;
 775         int nitems;
 776         int i;
 777         int ret = 0;
 778
 779         BUG_ON(!path->nodes[0]);
 780
 781         leaf = path->nodes[0];
 782         free_space = btrfs_leaf_free_space(root, leaf);
 783         INIT_LIST_HEAD(&head);
 784
 785         next = item;
 786         nitems = 0;
 787
 788         /*
 789          * count the number of the continuous items that we can insert in batch
 790          */
 791         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 792                free_space) {
 793                 total_data_size += next->data_len;
 794                 total_size += next->data_len + sizeof(struct btrfs_item);
 795                 list_add_tail(&next->tree_list, &head);
 796                 nitems++;
 797
 798                 curr = next;
 799                 next = __btrfs_next_delayed_item(curr);
 800                 if (!next)
 801                         break;
 802
 803                 if (!btrfs_is_continuous_delayed_item(curr, next))
 804                         break;
 805         }
 806
 807         if (!nitems) {
 808                 ret = 0;
 809                 goto out;
 810         }
 811
 812         /*
 813          * we need allocate some memory space, but it might cause the task
 814          * to sleep, so we set all locked nodes in the path to blocking locks
 815          * first.
 816          */
 817         btrfs_set_path_blocking(path);
 818
 819         keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS);
 820         if (!keys) {
 821                 ret = -ENOMEM;
 822                 goto out;
 823         }
 824
 825         data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS);
 826         if (!data_size) {
 827                 ret = -ENOMEM;
 828                 goto error;
 829         }
 830
 831         /* get keys of all the delayed items */
 832         i = 0;
 833         list_for_each_entry(next, &head, tree_list) {
 834                 keys[i] = next->key;
 835                 data_size[i] = next->data_len;
 836                 i++;
 837         }
 838
 839         /* reset all the locked nodes in the patch to spinning locks. */
 840         btrfs_clear_path_blocking(path, NULL, 0);
 841
 842         /* insert the keys of the items */
 843         setup_items_for_insert(trans, root, path, keys, data_size,
 844                                total_data_size, total_size, nitems);
 845
 846         /* insert the dir index items */
 847         slot = path->slots[0];
 848         list_for_each_entry_safe(curr, next, &head, tree_list) {
 849                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 850                 write_extent_buffer(leaf, &curr->data,
 851                                     (unsigned long)data_ptr,
 852                                     curr->data_len);
 853                 slot++;
 854
 855                 btrfs_delayed_item_release_metadata(root, curr);
 856
 857                 list_del(&curr->tree_list);
 858                 btrfs_release_delayed_item(curr);
 859         }
 860
 861 error:
 862         kfree(data_size);
 863         kfree(keys);
 864 out:
 865         return ret;
 866 }
 867
 868 /*
 869  * This helper can just do simple insertion that needn't extend item for new
 870  * data, such as directory name index insertion, inode insertion.
 871  */
 872 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 873                                      struct btrfs_root *root,
 874                                      struct btrfs_path *path,
 875                                      struct btrfs_delayed_item *delayed_item)
 876 {
 877         struct extent_buffer *leaf;
 878         struct btrfs_item *item;
 879         char *ptr;
 880         int ret;
 881
 882         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 883                                       delayed_item->data_len);
 884         if (ret < 0 && ret != -EEXIST)
 885                 return ret;
 886
 887         leaf = path->nodes[0];
 888
 889         item = btrfs_item_nr(leaf, path->slots[0]);
 890         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 891
 892         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 893                             delayed_item->data_len);
 894         btrfs_mark_buffer_dirty(leaf);
 895
 896         btrfs_delayed_item_release_metadata(root, delayed_item);
 897         return 0;
 898 }
 899
 900 /*
 901  * we insert an item first, then if there are some continuous items, we try
 902  * to insert those items into the same leaf.
 903  */
 904 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 905                                       struct btrfs_path *path,
 906                                       struct btrfs_root *root,
 907                                       struct btrfs_delayed_node *node)
 908 {
 909         struct btrfs_delayed_item *curr, *prev;
 910         int ret = 0;
 911
 912 do_again:
 913         mutex_lock(&node->mutex);
 914         curr = __btrfs_first_delayed_insertion_item(node);
 915         if (!curr)
 916                 goto insert_end;
 917
 918         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 919         if (ret < 0) {
 920                 btrfs_release_path(path);
 921                 goto insert_end;
 922         }
 923
 924         prev = curr;
 925         curr = __btrfs_next_delayed_item(prev);
 926         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 927                 /* insert the continuous items into the same leaf */
 928                 path->slots[0]++;
 929                 btrfs_batch_insert_items(trans, root, path, curr);
 930         }
 931         btrfs_release_delayed_item(prev);
 932         btrfs_mark_buffer_dirty(path->nodes[0]);
 933
 934         btrfs_release_path(path);
 935         mutex_unlock(&node->mutex);
 936         goto do_again;
 937
 938 insert_end:
 939         mutex_unlock(&node->mutex);
 940         return ret;
 941 }
 942
 943 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 944                                     struct btrfs_root *root,
 945                                     struct btrfs_path *path,
 946                                     struct btrfs_delayed_item *item)
 947 {
 948         struct btrfs_delayed_item *curr, *next;
 949         struct extent_buffer *leaf;
 950         struct btrfs_key key;
 951         struct list_head head;
 952         int nitems, i, last_item;
 953         int ret = 0;
 954
 955         BUG_ON(!path->nodes[0]);
 956
 957         leaf = path->nodes[0];
 958
 959         i = path->slots[0];
 960         last_item = btrfs_header_nritems(leaf) - 1;
 961         if (i > last_item)
 962                 return -ENOENT; /* FIXME: Is errno suitable? */
 963
 964         next = item;
 965         INIT_LIST_HEAD(&head);
 966         btrfs_item_key_to_cpu(leaf, &key, i);
 967         nitems = 0;
 968         /*
 969          * count the number of the dir index items that we can delete in batch
 970          */
 971         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 972                 list_add_tail(&next->tree_list, &head);
 973                 nitems++;
 974
 975                 curr = next;
 976                 next = __btrfs_next_delayed_item(curr);
 977                 if (!next)
 978                         break;
 979
 980                 if (!btrfs_is_continuous_delayed_item(curr, next))
 981                         break;
 982
 983                 i++;
 984                 if (i > last_item)
 985                         break;
 986                 btrfs_item_key_to_cpu(leaf, &key, i);
 987         }
 988
 989         if (!nitems)
 990                 return 0;
 991
 992         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 993         if (ret)
 994                 goto out;
 995
 996         list_for_each_entry_safe(curr, next, &head, tree_list) {
 997                 btrfs_delayed_item_release_metadata(root, curr);
 998                 list_del(&curr->tree_list);
 999                 btrfs_release_delayed_item(curr);
1000         }
1001
1002 out:
1003         return ret;
1004 }
1005
1006 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
1007                                       struct btrfs_path *path,
1008                                       struct btrfs_root *root,
1009                                       struct btrfs_delayed_node *node)
1010 {
1011         struct btrfs_delayed_item *curr, *prev;
1012         int ret = 0;
1013
1014 do_again:
1015         mutex_lock(&node->mutex);
1016         curr = __btrfs_first_delayed_deletion_item(node);
1017         if (!curr)
1018                 goto delete_fail;
1019
1020         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
1021         if (ret < 0)
1022                 goto delete_fail;
1023         else if (ret > 0) {
1024                 /*
1025                  * can't find the item which the node points to, so this node
1026                  * is invalid, just drop it.
1027                  */
1028                 prev = curr;
1029                 curr = __btrfs_next_delayed_item(prev);
1030                 btrfs_release_delayed_item(prev);
1031                 ret = 0;
1032                 btrfs_release_path(path);
1033                 if (curr) {
1034                         mutex_unlock(&node->mutex);
1035                         goto do_again;
1036                 } else
1037                         goto delete_fail;
1038         }
1039
1040         btrfs_batch_delete_items(trans, root, path, curr);
1041         btrfs_release_path(path);
1042         mutex_unlock(&node->mutex);
1043         goto do_again;
1044
1045 delete_fail:
1046         btrfs_release_path(path);
1047         mutex_unlock(&node->mutex);
1048         return ret;
1049 }
1050
1051 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1052 {
1053         struct btrfs_delayed_root *delayed_root;
1054
1055         if (delayed_node && delayed_node->inode_dirty) {
1056                 BUG_ON(!delayed_node->root);
1057                 delayed_node->inode_dirty = 0;
1058                 delayed_node->count--;
1059
1060                 delayed_root = delayed_node->root->fs_info->delayed_root;
1061                 if (atomic_dec_return(&delayed_root->items) <
1062                     BTRFS_DELAYED_BACKGROUND &&
1063                     waitqueue_active(&delayed_root->wait))
1064                         wake_up(&delayed_root->wait);
1065         }
1066 }
1067
1068 static int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1069                                       struct btrfs_root *root,
1070                                       struct btrfs_path *path,
1071                                       struct btrfs_delayed_node *node)
1072 {
1073         struct btrfs_key key;
1074         struct btrfs_inode_item *inode_item;
1075         struct extent_buffer *leaf;
1076         int ret;
1077
1078         mutex_lock(&node->mutex);
1079         if (!node->inode_dirty) {
1080                 mutex_unlock(&node->mutex);
1081                 return 0;
1082         }
1083
1084         key.objectid = node->inode_id;
1085         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1086         key.offset = 0;
1087         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1088         if (ret > 0) {
1089                 btrfs_release_path(path);
1090                 mutex_unlock(&node->mutex);
1091                 return -ENOENT;
1092         } else if (ret < 0) {
1093                 mutex_unlock(&node->mutex);
1094                 return ret;
1095         }
1096
1097         btrfs_unlock_up_safe(path, 1);
1098         leaf = path->nodes[0];
1099         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1100                                     struct btrfs_inode_item);
1101         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1102                             sizeof(struct btrfs_inode_item));
1103         btrfs_mark_buffer_dirty(leaf);
1104         btrfs_release_path(path);
1105
1106         btrfs_delayed_inode_release_metadata(root, node);
1107         btrfs_release_delayed_inode(node);
1108         mutex_unlock(&node->mutex);
1109
1110         return 0;
1111 }
1112
1113 /*
1114  * Called when committing the transaction.
1115  * Returns 0 on success.
1116  * Returns < 0 on error and returns with an aborted transaction with any
1117  * outstanding delayed items cleaned up.
1118  */
1119 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1120                                      struct btrfs_root *root, int nr)
1121 {
1122         struct btrfs_root *curr_root = root;
1123         struct btrfs_delayed_root *delayed_root;
1124         struct btrfs_delayed_node *curr_node, *prev_node;
1125         struct btrfs_path *path;
1126         struct btrfs_block_rsv *block_rsv;
1127         int ret = 0;
1128         bool count = (nr > 0);
1129
1130         if (trans->aborted)
1131                 return -EIO;
1132
1133         path = btrfs_alloc_path();
1134         if (!path)
1135                 return -ENOMEM;
1136         path->leave_spinning = 1;
1137
1138         block_rsv = trans->block_rsv;
1139         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1140
1141         delayed_root = btrfs_get_delayed_root(root);
1142
1143         curr_node = btrfs_first_delayed_node(delayed_root);
1144         while (curr_node && (!count || (count && nr--))) {
1145                 curr_root = curr_node->root;
1146                 ret = btrfs_insert_delayed_items(trans, path, curr_root,
1147                                                  curr_node);
1148                 if (!ret)
1149                         ret = btrfs_delete_delayed_items(trans, path,
1150                                                 curr_root, curr_node);
1151                 if (!ret)
1152                         ret = btrfs_update_delayed_inode(trans, curr_root,
1153                                                 path, curr_node);
1154                 if (ret) {
1155                         btrfs_release_delayed_node(curr_node);
1156                         curr_node = NULL;
1157                         btrfs_abort_transaction(trans, root, ret);
1158                         break;
1159                 }
1160
1161                 prev_node = curr_node;
1162                 curr_node = btrfs_next_delayed_node(curr_node);
1163                 btrfs_release_delayed_node(prev_node);
1164         }
1165
1166         if (curr_node)
1167                 btrfs_release_delayed_node(curr_node);
1168         btrfs_free_path(path);
1169         trans->block_rsv = block_rsv;
1170
1171         return ret;
1172 }
1173
1174 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1175                             struct btrfs_root *root)
1176 {
1177         return __btrfs_run_delayed_items(trans, root, -1);
1178 }
1179
1180 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1181                                struct btrfs_root *root, int nr)
1182 {
1183         return __btrfs_run_delayed_items(trans, root, nr);
1184 }
1185
1186 static int __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1187                                               struct btrfs_delayed_node *node)
1188 {
1189         struct btrfs_path *path;
1190         struct btrfs_block_rsv *block_rsv;
1191         int ret;
1192
1193         path = btrfs_alloc_path();
1194         if (!path)
1195                 return -ENOMEM;
1196         path->leave_spinning = 1;
1197
1198         block_rsv = trans->block_rsv;
1199         trans->block_rsv = &node->root->fs_info->delayed_block_rsv;
1200
1201         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1202         if (!ret)
1203                 ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1204         if (!ret)
1205                 ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1206         btrfs_free_path(path);
1207
1208         trans->block_rsv = block_rsv;
1209         return ret;
1210 }
1211
1212 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1213                                      struct inode *inode)
1214 {
1215         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1216         int ret;
1217
1218         if (!delayed_node)
1219                 return 0;
1220
1221         mutex_lock(&delayed_node->mutex);
1222         if (!delayed_node->count) {
1223                 mutex_unlock(&delayed_node->mutex);
1224                 btrfs_release_delayed_node(delayed_node);
1225                 return 0;
1226         }
1227         mutex_unlock(&delayed_node->mutex);
1228
1229         ret = __btrfs_commit_inode_delayed_items(trans, delayed_node);
1230         btrfs_release_delayed_node(delayed_node);
1231         return ret;
1232 }
1233
1234 void btrfs_remove_delayed_node(struct inode *inode)
1235 {
1236         struct btrfs_delayed_node *delayed_node;
1237
1238         delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1239         if (!delayed_node)
1240                 return;
1241
1242         BTRFS_I(inode)->delayed_node = NULL;
1243         btrfs_release_delayed_node(delayed_node);
1244 }
1245
1246 struct btrfs_async_delayed_node {
1247         struct btrfs_root *root;
1248         struct btrfs_delayed_node *delayed_node;
1249         struct btrfs_work work;
1250 };
1251
1252 static void btrfs_async_run_delayed_node_done(struct btrfs_work *work)
1253 {
1254         struct btrfs_async_delayed_node *async_node;
1255         struct btrfs_trans_handle *trans;
1256         struct btrfs_path *path;
1257         struct btrfs_delayed_node *delayed_node = NULL;
1258         struct btrfs_root *root;
1259         struct btrfs_block_rsv *block_rsv;
1260         int need_requeue = 0;
1261         int ret;
1262
1263         async_node = container_of(work, struct btrfs_async_delayed_node, work);
1264
1265         path = btrfs_alloc_path();
1266         if (!path)
1267                 goto out;
1268         path->leave_spinning = 1;
1269
1270         delayed_node = async_node->delayed_node;
1271         root = delayed_node->root;
1272
1273         trans = btrfs_join_transaction(root);
1274         if (IS_ERR(trans))
1275                 goto free_path;
1276
1277         block_rsv = trans->block_rsv;
1278         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1279
1280         ret = btrfs_insert_delayed_items(trans, path, root, delayed_node);
1281         if (!ret)
1282                 ret = btrfs_delete_delayed_items(trans, path, root,
1283                                                  delayed_node);
1284
1285         if (!ret)
1286                 btrfs_update_delayed_inode(trans, root, path, delayed_node);
1287
1288         /*
1289          * Maybe new delayed items have been inserted, so we need requeue
1290          * the work. Besides that, we must dequeue the empty delayed nodes
1291          * to avoid the race between delayed items balance and the worker.
1292          * The race like this:
1293          *      Task1                           Worker thread
1294          *                                      count == 0, needn't requeue
1295          *                                        also needn't insert the
1296          *                                        delayed node into prepare
1297          *                                        list again.
1298          *      add lots of delayed items
1299          *      queue the delayed node
1300          *        already in the list,
1301          *        and not in the prepare
1302          *        list, it means the delayed
1303          *        node is being dealt with
1304          *        by the worker.
1305          *      do delayed items balance
1306          *        the delayed node is being
1307          *        dealt with by the worker
1308          *        now, just wait.
1309          *                                      the worker goto idle.
1310          * Task1 will sleep until the transaction is commited.
1311          */
1312         mutex_lock(&delayed_node->mutex);
1313         if (delayed_node->count)
1314                 need_requeue = 1;
1315         else
1316                 btrfs_dequeue_delayed_node(root->fs_info->delayed_root,
1317                                            delayed_node);
1318         mutex_unlock(&delayed_node->mutex);
1319
1320         trans->block_rsv = block_rsv;
1321         btrfs_end_transaction_dmeta(trans, root);
1322         btrfs_btree_balance_dirty_nodelay(root);
1323 free_path:
1324         btrfs_free_path(path);
1325 out:
1326         if (need_requeue)
1327                 btrfs_requeue_work(&async_node->work);
1328         else {
1329                 btrfs_release_prepared_delayed_node(delayed_node);
1330                 kfree(async_node);
1331         }
1332 }
1333
1334 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1335                                      struct btrfs_root *root, int all)
1336 {
1337         struct btrfs_async_delayed_node *async_node;
1338         struct btrfs_delayed_node *curr;
1339         int count = 0;
1340
1341 again:
1342         curr = btrfs_first_prepared_delayed_node(delayed_root);
1343         if (!curr)
1344                 return 0;
1345
1346         async_node = kmalloc(sizeof(*async_node), GFP_NOFS);
1347         if (!async_node) {
1348                 btrfs_release_prepared_delayed_node(curr);
1349                 return -ENOMEM;
1350         }
1351
1352         async_node->root = root;
1353         async_node->delayed_node = curr;
1354
1355         async_node->work.func = btrfs_async_run_delayed_node_done;
1356         async_node->work.flags = 0;
1357
1358         btrfs_queue_worker(&root->fs_info->delayed_workers, &async_node->work);
1359         count++;
1360
1361         if (all || count < 4)
1362                 goto again;
1363
1364         return 0;
1365 }
1366
1367 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1368 {
1369         struct btrfs_delayed_root *delayed_root;
1370         delayed_root = btrfs_get_delayed_root(root);
1371         WARN_ON(btrfs_first_delayed_node(delayed_root));
1372 }
1373
1374 void btrfs_balance_delayed_items(struct btrfs_root *root)
1375 {
1376         struct btrfs_delayed_root *delayed_root;
1377
1378         delayed_root = btrfs_get_delayed_root(root);
1379
1380         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1381                 return;
1382
1383         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1384                 int ret;
1385                 ret = btrfs_wq_run_delayed_node(delayed_root, root, 1);
1386                 if (ret)
1387                         return;
1388
1389                 wait_event_interruptible_timeout(
1390                                 delayed_root->wait,
1391                                 (atomic_read(&delayed_root->items) <
1392                                  BTRFS_DELAYED_BACKGROUND),
1393                                 HZ);
1394                 return;
1395         }
1396
1397         btrfs_wq_run_delayed_node(delayed_root, root, 0);
1398 }
1399
1400 /* Will return 0 or -ENOMEM */
1401 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1402                                    struct btrfs_root *root, const char *name,
1403                                    int name_len, struct inode *dir,
1404                                    struct btrfs_disk_key *disk_key, u8 type,
1405                                    u64 index)
1406 {
1407         struct btrfs_delayed_node *delayed_node;
1408         struct btrfs_delayed_item *delayed_item;
1409         struct btrfs_dir_item *dir_item;
1410         int ret;
1411
1412         delayed_node = btrfs_get_or_create_delayed_node(dir);
1413         if (IS_ERR(delayed_node))
1414                 return PTR_ERR(delayed_node);
1415
1416         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1417         if (!delayed_item) {
1418                 ret = -ENOMEM;
1419                 goto release_node;
1420         }
1421
1422         delayed_item->key.objectid = btrfs_ino(dir);
1423         btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1424         delayed_item->key.offset = index;
1425
1426         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1427         dir_item->location = *disk_key;
1428         dir_item->transid = cpu_to_le64(trans->transid);
1429         dir_item->data_len = 0;
1430         dir_item->name_len = cpu_to_le16(name_len);
1431         dir_item->type = type;
1432         memcpy((char *)(dir_item + 1), name, name_len);
1433
1434         ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1435         /*
1436          * we have reserved enough space when we start a new transaction,
1437          * so reserving metadata failure is impossible
1438          */
1439         BUG_ON(ret);
1440
1441
1442         mutex_lock(&delayed_node->mutex);
1443         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1444         if (unlikely(ret)) {
1445                 printk(KERN_ERR "err add delayed dir index item(name: %s) into "
1446                                 "the insertion tree of the delayed node"
1447                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1448                                 name,
1449                                 (unsigned long long)delayed_node->root->objectid,
1450                                 (unsigned long long)delayed_node->inode_id,
1451                                 ret);
1452                 BUG();
1453         }
1454         mutex_unlock(&delayed_node->mutex);
1455
1456 release_node:
1457         btrfs_release_delayed_node(delayed_node);
1458         return ret;
1459 }
1460
1461 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1462                                                struct btrfs_delayed_node *node,
1463                                                struct btrfs_key *key)
1464 {
1465         struct btrfs_delayed_item *item;
1466
1467         mutex_lock(&node->mutex);
1468         item = __btrfs_lookup_delayed_insertion_item(node, key);
1469         if (!item) {
1470                 mutex_unlock(&node->mutex);
1471                 return 1;
1472         }
1473
1474         btrfs_delayed_item_release_metadata(root, item);
1475         btrfs_release_delayed_item(item);
1476         mutex_unlock(&node->mutex);
1477         return 0;
1478 }
1479
1480 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1481                                    struct btrfs_root *root, struct inode *dir,
1482                                    u64 index)
1483 {
1484         struct btrfs_delayed_node *node;
1485         struct btrfs_delayed_item *item;
1486         struct btrfs_key item_key;
1487         int ret;
1488
1489         node = btrfs_get_or_create_delayed_node(dir);
1490         if (IS_ERR(node))
1491                 return PTR_ERR(node);
1492
1493         item_key.objectid = btrfs_ino(dir);
1494         btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1495         item_key.offset = index;
1496
1497         ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1498         if (!ret)
1499                 goto end;
1500
1501         item = btrfs_alloc_delayed_item(0);
1502         if (!item) {
1503                 ret = -ENOMEM;
1504                 goto end;
1505         }
1506
1507         item->key = item_key;
1508
1509         ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1510         /*
1511          * we have reserved enough space when we start a new transaction,
1512          * so reserving metadata failure is impossible.
1513          */
1514         BUG_ON(ret);
1515
1516         mutex_lock(&node->mutex);
1517         ret = __btrfs_add_delayed_deletion_item(node, item);
1518         if (unlikely(ret)) {
1519                 printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1520                                 "into the deletion tree of the delayed node"
1521                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1522                                 (unsigned long long)index,
1523                                 (unsigned long long)node->root->objectid,
1524                                 (unsigned long long)node->inode_id,
1525                                 ret);
1526                 BUG();
1527         }
1528         mutex_unlock(&node->mutex);
1529 end:
1530         btrfs_release_delayed_node(node);
1531         return ret;
1532 }
1533
1534 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1535 {
1536         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1537
1538         if (!delayed_node)
1539                 return -ENOENT;
1540
1541         /*
1542          * Since we have held i_mutex of this directory, it is impossible that
1543          * a new directory index is added into the delayed node and index_cnt
1544          * is updated now. So we needn't lock the delayed node.
1545          */
1546         if (!delayed_node->index_cnt) {
1547                 btrfs_release_delayed_node(delayed_node);
1548                 return -EINVAL;
1549         }
1550
1551         BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1552         btrfs_release_delayed_node(delayed_node);
1553         return 0;
1554 }
1555
1556 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1557                              struct list_head *del_list)
1558 {
1559         struct btrfs_delayed_node *delayed_node;
1560         struct btrfs_delayed_item *item;
1561
1562         delayed_node = btrfs_get_delayed_node(inode);
1563         if (!delayed_node)
1564                 return;
1565
1566         mutex_lock(&delayed_node->mutex);
1567         item = __btrfs_first_delayed_insertion_item(delayed_node);
1568         while (item) {
1569                 atomic_inc(&item->refs);
1570                 list_add_tail(&item->readdir_list, ins_list);
1571                 item = __btrfs_next_delayed_item(item);
1572         }
1573
1574         item = __btrfs_first_delayed_deletion_item(delayed_node);
1575         while (item) {
1576                 atomic_inc(&item->refs);
1577                 list_add_tail(&item->readdir_list, del_list);
1578                 item = __btrfs_next_delayed_item(item);
1579         }
1580         mutex_unlock(&delayed_node->mutex);
1581         /*
1582          * This delayed node is still cached in the btrfs inode, so refs
1583          * must be > 1 now, and we needn't check it is going to be freed
1584          * or not.
1585          *
1586          * Besides that, this function is used to read dir, we do not
1587          * insert/delete delayed items in this period. So we also needn't
1588          * requeue or dequeue this delayed node.
1589          */
1590         atomic_dec(&delayed_node->refs);
1591 }
1592
1593 void btrfs_put_delayed_items(struct list_head *ins_list,
1594                              struct list_head *del_list)
1595 {
1596         struct btrfs_delayed_item *curr, *next;
1597
1598         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1599                 list_del(&curr->readdir_list);
1600                 if (atomic_dec_and_test(&curr->refs))
1601                         kfree(curr);
1602         }
1603
1604         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1605                 list_del(&curr->readdir_list);
1606                 if (atomic_dec_and_test(&curr->refs))
1607                         kfree(curr);
1608         }
1609 }
1610
1611 int btrfs_should_delete_dir_index(struct list_head *del_list,
1612                                   u64 index)
1613 {
1614         struct btrfs_delayed_item *curr, *next;
1615         int ret;
1616
1617         if (list_empty(del_list))
1618                 return 0;
1619
1620         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1621                 if (curr->key.offset > index)
1622                         break;
1623
1624                 list_del(&curr->readdir_list);
1625                 ret = (curr->key.offset == index);
1626
1627                 if (atomic_dec_and_test(&curr->refs))
1628                         kfree(curr);
1629
1630                 if (ret)
1631                         return 1;
1632                 else
1633                         continue;
1634         }
1635         return 0;
1636 }
1637
1638 /*
1639  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1640  *
1641  */
1642 int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent,
1643                                     filldir_t filldir,
1644                                     struct list_head *ins_list)
1645 {
1646         struct btrfs_dir_item *di;
1647         struct btrfs_delayed_item *curr, *next;
1648         struct btrfs_key location;
1649         char *name;
1650         int name_len;
1651         int over = 0;
1652         unsigned char d_type;
1653
1654         if (list_empty(ins_list))
1655                 return 0;
1656
1657         /*
1658          * Changing the data of the delayed item is impossible. So
1659          * we needn't lock them. And we have held i_mutex of the
1660          * directory, nobody can delete any directory indexes now.
1661          */
1662         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1663                 list_del(&curr->readdir_list);
1664
1665                 if (curr->key.offset < filp->f_pos) {
1666                         if (atomic_dec_and_test(&curr->refs))
1667                                 kfree(curr);
1668                         continue;
1669                 }
1670
1671                 filp->f_pos = curr->key.offset;
1672
1673                 di = (struct btrfs_dir_item *)curr->data;
1674                 name = (char *)(di + 1);
1675                 name_len = le16_to_cpu(di->name_len);
1676
1677                 d_type = btrfs_filetype_table[di->type];
1678                 btrfs_disk_key_to_cpu(&location, &di->location);
1679
1680                 over = filldir(dirent, name, name_len, curr->key.offset,
1681                                location.objectid, d_type);
1682
1683                 if (atomic_dec_and_test(&curr->refs))
1684                         kfree(curr);
1685
1686                 if (over)
1687                         return 1;
1688         }
1689         return 0;
1690 }
1691
1692 BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item,
1693                          generation, 64);
1694 BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item,
1695                          sequence, 64);
1696 BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item,
1697                          transid, 64);
1698 BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64);
1699 BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item,
1700                          nbytes, 64);
1701 BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item,
1702                          block_group, 64);
1703 BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32);
1704 BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32);
1705 BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32);
1706 BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32);
1707 BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64);
1708 BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64);
1709
1710 BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64);
1711 BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32);
1712
1713 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1714                                   struct btrfs_inode_item *inode_item,
1715                                   struct inode *inode)
1716 {
1717         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1718         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1719         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1720         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1721         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1722         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1723         btrfs_set_stack_inode_generation(inode_item,
1724                                          BTRFS_I(inode)->generation);
1725         btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1726         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1727         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1728         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1729         btrfs_set_stack_inode_block_group(inode_item, 0);
1730
1731         btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1732                                      inode->i_atime.tv_sec);
1733         btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1734                                       inode->i_atime.tv_nsec);
1735
1736         btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1737                                      inode->i_mtime.tv_sec);
1738         btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1739                                       inode->i_mtime.tv_nsec);
1740
1741         btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1742                                      inode->i_ctime.tv_sec);
1743         btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1744                                       inode->i_ctime.tv_nsec);
1745 }
1746
1747 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1748 {
1749         struct btrfs_delayed_node *delayed_node;
1750         struct btrfs_inode_item *inode_item;
1751         struct btrfs_timespec *tspec;
1752
1753         delayed_node = btrfs_get_delayed_node(inode);
1754         if (!delayed_node)
1755                 return -ENOENT;
1756
1757         mutex_lock(&delayed_node->mutex);
1758         if (!delayed_node->inode_dirty) {
1759                 mutex_unlock(&delayed_node->mutex);
1760                 btrfs_release_delayed_node(delayed_node);
1761                 return -ENOENT;
1762         }
1763
1764         inode_item = &delayed_node->inode_item;
1765
1766         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1767         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1768         btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1769         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1770         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1771         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1772         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1773         inode->i_version = btrfs_stack_inode_sequence(inode_item);
1774         inode->i_rdev = 0;
1775         *rdev = btrfs_stack_inode_rdev(inode_item);
1776         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1777
1778         tspec = btrfs_inode_atime(inode_item);
1779         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1780         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1781
1782         tspec = btrfs_inode_mtime(inode_item);
1783         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1784         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1785
1786         tspec = btrfs_inode_ctime(inode_item);
1787         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1788         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1789
1790         inode->i_generation = BTRFS_I(inode)->generation;
1791         BTRFS_I(inode)->index_cnt = (u64)-1;
1792
1793         mutex_unlock(&delayed_node->mutex);
1794         btrfs_release_delayed_node(delayed_node);
1795         return 0;
1796 }
1797
1798 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1799                                struct btrfs_root *root, struct inode *inode)
1800 {
1801         struct btrfs_delayed_node *delayed_node;
1802         int ret = 0;
1803
1804         delayed_node = btrfs_get_or_create_delayed_node(inode);
1805         if (IS_ERR(delayed_node))
1806                 return PTR_ERR(delayed_node);
1807
1808         mutex_lock(&delayed_node->mutex);
1809         if (delayed_node->inode_dirty) {
1810                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1811                 goto release_node;
1812         }
1813
1814         ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1815                                                    delayed_node);
1816         if (ret)
1817                 goto release_node;
1818
1819         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1820         delayed_node->inode_dirty = 1;
1821         delayed_node->count++;
1822         atomic_inc(&root->fs_info->delayed_root->items);
1823 release_node:
1824         mutex_unlock(&delayed_node->mutex);
1825         btrfs_release_delayed_node(delayed_node);
1826         return ret;
1827 }
1828
1829 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1830 {
1831         struct btrfs_root *root = delayed_node->root;
1832         struct btrfs_delayed_item *curr_item, *prev_item;
1833
1834         mutex_lock(&delayed_node->mutex);
1835         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1836         while (curr_item) {
1837                 btrfs_delayed_item_release_metadata(root, curr_item);
1838                 prev_item = curr_item;
1839                 curr_item = __btrfs_next_delayed_item(prev_item);
1840                 btrfs_release_delayed_item(prev_item);
1841         }
1842
1843         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1844         while (curr_item) {
1845                 btrfs_delayed_item_release_metadata(root, curr_item);
1846                 prev_item = curr_item;
1847                 curr_item = __btrfs_next_delayed_item(prev_item);
1848                 btrfs_release_delayed_item(prev_item);
1849         }
1850
1851         if (delayed_node->inode_dirty) {
1852                 btrfs_delayed_inode_release_metadata(root, delayed_node);
1853                 btrfs_release_delayed_inode(delayed_node);
1854         }
1855         mutex_unlock(&delayed_node->mutex);
1856 }
1857
1858 void btrfs_kill_delayed_inode_items(struct inode *inode)
1859 {
1860         struct btrfs_delayed_node *delayed_node;
1861
1862         delayed_node = btrfs_get_delayed_node(inode);
1863         if (!delayed_node)
1864                 return;
1865
1866         __btrfs_kill_delayed_node(delayed_node);
1867         btrfs_release_delayed_node(delayed_node);
1868 }
1869
1870 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1871 {
1872         u64 inode_id = 0;
1873         struct btrfs_delayed_node *delayed_nodes[8];
1874         int i, n;
1875
1876         while (1) {
1877                 spin_lock(&root->inode_lock);
1878                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1879                                            (void **)delayed_nodes, inode_id,
1880                                            ARRAY_SIZE(delayed_nodes));
1881                 if (!n) {
1882                         spin_unlock(&root->inode_lock);
1883                         break;
1884                 }
1885
1886                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1887
1888                 for (i = 0; i < n; i++)
1889                         atomic_inc(&delayed_nodes[i]->refs);
1890                 spin_unlock(&root->inode_lock);
1891
1892                 for (i = 0; i < n; i++) {
1893                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1894                         btrfs_release_delayed_node(delayed_nodes[i]);
1895                 }
1896         }
1897 }
1898
1899 void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1900 {
1901         struct btrfs_delayed_root *delayed_root;
1902         struct btrfs_delayed_node *curr_node, *prev_node;
1903
1904         delayed_root = btrfs_get_delayed_root(root);
1905
1906         curr_node = btrfs_first_delayed_node(delayed_root);
1907         while (curr_node) {
1908                 __btrfs_kill_delayed_node(curr_node);
1909
1910                 prev_node = curr_node;
1911                 curr_node = btrfs_next_delayed_node(curr_node);
1912                 btrfs_release_delayed_node(prev_node);
1913         }
1914 }
1915