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 #include "ctree.h"
  25
  26 #define BTRFS_DELAYED_WRITEBACK         512
  27 #define BTRFS_DELAYED_BACKGROUND        128
  28 #define BTRFS_DELAYED_BATCH             16
  29
  30 static struct kmem_cache *delayed_node_cache;
  31
  32 int __init btrfs_delayed_inode_init(void)
  33 {
  34         delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
  35                                         sizeof(struct btrfs_delayed_node),
  36                                         0,
  37                                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
  38                                         NULL);
  39         if (!delayed_node_cache)
  40                 return -ENOMEM;
  41         return 0;
  42 }
  43
  44 void btrfs_delayed_inode_exit(void)
  45 {
  46         if (delayed_node_cache)
  47                 kmem_cache_destroy(delayed_node_cache);
  48 }
  49
  50 static inline void btrfs_init_delayed_node(
  51                                 struct btrfs_delayed_node *delayed_node,
  52                                 struct btrfs_root *root, u64 inode_id)
  53 {
  54         delayed_node->root = root;
  55         delayed_node->inode_id = inode_id;
  56         atomic_set(&delayed_node->refs, 0);
  57         delayed_node->count = 0;
  58         delayed_node->flags = 0;
  59         delayed_node->ins_root = RB_ROOT;
  60         delayed_node->del_root = RB_ROOT;
  61         mutex_init(&delayed_node->mutex);
  62         delayed_node->index_cnt = 0;
  63         INIT_LIST_HEAD(&delayed_node->n_list);
  64         INIT_LIST_HEAD(&delayed_node->p_list);
  65         delayed_node->bytes_reserved = 0;
  66         memset(&delayed_node->inode_item, 0, sizeof(delayed_node->inode_item));
  67 }
  68
  69 static inline int btrfs_is_continuous_delayed_item(
  70                                         struct btrfs_delayed_item *item1,
  71                                         struct btrfs_delayed_item *item2)
  72 {
  73         if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  74             item1->key.objectid == item2->key.objectid &&
  75             item1->key.type == item2->key.type &&
  76             item1->key.offset + 1 == item2->key.offset)
  77                 return 1;
  78         return 0;
  79 }
  80
  81 static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
  82                                                         struct btrfs_root *root)
  83 {
  84         return root->fs_info->delayed_root;
  85 }
  86
  87 static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
  88 {
  89         struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
  90         struct btrfs_root *root = btrfs_inode->root;
  91         u64 ino = btrfs_ino(inode);
  92         struct btrfs_delayed_node *node;
  93
  94         node = ACCESS_ONCE(btrfs_inode->delayed_node);
  95         if (node) {
  96                 atomic_inc(&node->refs);
  97                 return node;
  98         }
  99
 100         spin_lock(&root->inode_lock);
 101         node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
 102         if (node) {
 103                 if (btrfs_inode->delayed_node) {
 104                         atomic_inc(&node->refs);        /* can be accessed */
 105                         BUG_ON(btrfs_inode->delayed_node != node);
 106                         spin_unlock(&root->inode_lock);
 107                         return node;
 108                 }
 109                 btrfs_inode->delayed_node = node;
 110                 /* can be accessed and cached in the inode */
 111                 atomic_add(2, &node->refs);
 112                 spin_unlock(&root->inode_lock);
 113                 return node;
 114         }
 115         spin_unlock(&root->inode_lock);
 116
 117         return NULL;
 118 }
 119
 120 /* Will return either the node or PTR_ERR(-ENOMEM) */
 121 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 122                                                         struct inode *inode)
 123 {
 124         struct btrfs_delayed_node *node;
 125         struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
 126         struct btrfs_root *root = btrfs_inode->root;
 127         u64 ino = btrfs_ino(inode);
 128         int ret;
 129
 130 again:
 131         node = btrfs_get_delayed_node(inode);
 132         if (node)
 133                 return node;
 134
 135         node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
 136         if (!node)
 137                 return ERR_PTR(-ENOMEM);
 138         btrfs_init_delayed_node(node, root, ino);
 139
 140         /* cached in the btrfs inode and can be accessed */
 141         atomic_add(2, &node->refs);
 142
 143         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 144         if (ret) {
 145                 kmem_cache_free(delayed_node_cache, node);
 146                 return ERR_PTR(ret);
 147         }
 148
 149         spin_lock(&root->inode_lock);
 150         ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 151         if (ret == -EEXIST) {
 152                 kmem_cache_free(delayed_node_cache, node);
 153                 spin_unlock(&root->inode_lock);
 154                 radix_tree_preload_end();
 155                 goto again;
 156         }
 157         btrfs_inode->delayed_node = node;
 158         spin_unlock(&root->inode_lock);
 159         radix_tree_preload_end();
 160
 161         return node;
 162 }
 163
 164 /*
 165  * Call it when holding delayed_node->mutex
 166  *
 167  * If mod = 1, add this node into the prepared list.
 168  */
 169 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 170                                      struct btrfs_delayed_node *node,
 171                                      int mod)
 172 {
 173         spin_lock(&root->lock);
 174         if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 175                 if (!list_empty(&node->p_list))
 176                         list_move_tail(&node->p_list, &root->prepare_list);
 177                 else if (mod)
 178                         list_add_tail(&node->p_list, &root->prepare_list);
 179         } else {
 180                 list_add_tail(&node->n_list, &root->node_list);
 181                 list_add_tail(&node->p_list, &root->prepare_list);
 182                 atomic_inc(&node->refs);        /* inserted into list */
 183                 root->nodes++;
 184                 set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 185         }
 186         spin_unlock(&root->lock);
 187 }
 188
 189 /* Call it when holding delayed_node->mutex */
 190 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 191                                        struct btrfs_delayed_node *node)
 192 {
 193         spin_lock(&root->lock);
 194         if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 195                 root->nodes--;
 196                 atomic_dec(&node->refs);        /* not in the list */
 197                 list_del_init(&node->n_list);
 198                 if (!list_empty(&node->p_list))
 199                         list_del_init(&node->p_list);
 200                 clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 201         }
 202         spin_unlock(&root->lock);
 203 }
 204
 205 static struct btrfs_delayed_node *btrfs_first_delayed_node(
 206                         struct btrfs_delayed_root *delayed_root)
 207 {
 208         struct list_head *p;
 209         struct btrfs_delayed_node *node = NULL;
 210
 211         spin_lock(&delayed_root->lock);
 212         if (list_empty(&delayed_root->node_list))
 213                 goto out;
 214
 215         p = delayed_root->node_list.next;
 216         node = list_entry(p, struct btrfs_delayed_node, n_list);
 217         atomic_inc(&node->refs);
 218 out:
 219         spin_unlock(&delayed_root->lock);
 220
 221         return node;
 222 }
 223
 224 static struct btrfs_delayed_node *btrfs_next_delayed_node(
 225                                                 struct btrfs_delayed_node *node)
 226 {
 227         struct btrfs_delayed_root *delayed_root;
 228         struct list_head *p;
 229         struct btrfs_delayed_node *next = NULL;
 230
 231         delayed_root = node->root->fs_info->delayed_root;
 232         spin_lock(&delayed_root->lock);
 233         if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 234                 /* not in the list */
 235                 if (list_empty(&delayed_root->node_list))
 236                         goto out;
 237                 p = delayed_root->node_list.next;
 238         } else if (list_is_last(&node->n_list, &delayed_root->node_list))
 239                 goto out;
 240         else
 241                 p = node->n_list.next;
 242
 243         next = list_entry(p, struct btrfs_delayed_node, n_list);
 244         atomic_inc(&next->refs);
 245 out:
 246         spin_unlock(&delayed_root->lock);
 247
 248         return next;
 249 }
 250
 251 static void __btrfs_release_delayed_node(
 252                                 struct btrfs_delayed_node *delayed_node,
 253                                 int mod)
 254 {
 255         struct btrfs_delayed_root *delayed_root;
 256
 257         if (!delayed_node)
 258                 return;
 259
 260         delayed_root = delayed_node->root->fs_info->delayed_root;
 261
 262         mutex_lock(&delayed_node->mutex);
 263         if (delayed_node->count)
 264                 btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
 265         else
 266                 btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 267         mutex_unlock(&delayed_node->mutex);
 268
 269         if (atomic_dec_and_test(&delayed_node->refs)) {
 270                 struct btrfs_root *root = delayed_node->root;
 271                 spin_lock(&root->inode_lock);
 272                 if (atomic_read(&delayed_node->refs) == 0) {
 273                         radix_tree_delete(&root->delayed_nodes_tree,
 274                                           delayed_node->inode_id);
 275                         kmem_cache_free(delayed_node_cache, delayed_node);
 276                 }
 277                 spin_unlock(&root->inode_lock);
 278         }
 279 }
 280
 281 static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
 282 {
 283         __btrfs_release_delayed_node(node, 0);
 284 }
 285
 286 static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
 287                                         struct btrfs_delayed_root *delayed_root)
 288 {
 289         struct list_head *p;
 290         struct btrfs_delayed_node *node = NULL;
 291
 292         spin_lock(&delayed_root->lock);
 293         if (list_empty(&delayed_root->prepare_list))
 294                 goto out;
 295
 296         p = delayed_root->prepare_list.next;
 297         list_del_init(p);
 298         node = list_entry(p, struct btrfs_delayed_node, p_list);
 299         atomic_inc(&node->refs);
 300 out:
 301         spin_unlock(&delayed_root->lock);
 302
 303         return node;
 304 }
 305
 306 static inline void btrfs_release_prepared_delayed_node(
 307                                         struct btrfs_delayed_node *node)
 308 {
 309         __btrfs_release_delayed_node(node, 1);
 310 }
 311
 312 static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
 313 {
 314         struct btrfs_delayed_item *item;
 315         item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
 316         if (item) {
 317                 item->data_len = data_len;
 318                 item->ins_or_del = 0;
 319                 item->bytes_reserved = 0;
 320                 item->delayed_node = NULL;
 321                 atomic_set(&item->refs, 1);
 322         }
 323         return item;
 324 }
 325
 326 /*
 327  * __btrfs_lookup_delayed_item - look up the delayed item by key
 328  * @delayed_node: pointer to the delayed node
 329  * @key:          the key to look up
 330  * @prev:         used to store the prev item if the right item isn't found
 331  * @next:         used to store the next item if the right item isn't found
 332  *
 333  * Note: if we don't find the right item, we will return the prev item and
 334  * the next item.
 335  */
 336 static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 337                                 struct rb_root *root,
 338                                 struct btrfs_key *key,
 339                                 struct btrfs_delayed_item **prev,
 340                                 struct btrfs_delayed_item **next)
 341 {
 342         struct rb_node *node, *prev_node = NULL;
 343         struct btrfs_delayed_item *delayed_item = NULL;
 344         int ret = 0;
 345
 346         node = root->rb_node;
 347
 348         while (node) {
 349                 delayed_item = rb_entry(node, struct btrfs_delayed_item,
 350                                         rb_node);
 351                 prev_node = node;
 352                 ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
 353                 if (ret < 0)
 354                         node = node->rb_right;
 355                 else if (ret > 0)
 356                         node = node->rb_left;
 357                 else
 358                         return delayed_item;
 359         }
 360
 361         if (prev) {
 362                 if (!prev_node)
 363                         *prev = NULL;
 364                 else if (ret < 0)
 365                         *prev = delayed_item;
 366                 else if ((node = rb_prev(prev_node)) != NULL) {
 367                         *prev = rb_entry(node, struct btrfs_delayed_item,
 368                                          rb_node);
 369                 } else
 370                         *prev = NULL;
 371         }
 372
 373         if (next) {
 374                 if (!prev_node)
 375                         *next = NULL;
 376                 else if (ret > 0)
 377                         *next = delayed_item;
 378                 else if ((node = rb_next(prev_node)) != NULL) {
 379                         *next = rb_entry(node, struct btrfs_delayed_item,
 380                                          rb_node);
 381                 } else
 382                         *next = NULL;
 383         }
 384         return NULL;
 385 }
 386
 387 static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
 388                                         struct btrfs_delayed_node *delayed_node,
 389                                         struct btrfs_key *key)
 390 {
 391         struct btrfs_delayed_item *item;
 392
 393         item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 394                                            NULL, NULL);
 395         return item;
 396 }
 397
 398 static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
 399                                     struct btrfs_delayed_item *ins,
 400                                     int action)
 401 {
 402         struct rb_node **p, *node;
 403         struct rb_node *parent_node = NULL;
 404         struct rb_root *root;
 405         struct btrfs_delayed_item *item;
 406         int cmp;
 407
 408         if (action == BTRFS_DELAYED_INSERTION_ITEM)
 409                 root = &delayed_node->ins_root;
 410         else if (action == BTRFS_DELAYED_DELETION_ITEM)
 411                 root = &delayed_node->del_root;
 412         else
 413                 BUG();
 414         p = &root->rb_node;
 415         node = &ins->rb_node;
 416
 417         while (*p) {
 418                 parent_node = *p;
 419                 item = rb_entry(parent_node, struct btrfs_delayed_item,
 420                                  rb_node);
 421
 422                 cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
 423                 if (cmp < 0)
 424                         p = &(*p)->rb_right;
 425                 else if (cmp > 0)
 426                         p = &(*p)->rb_left;
 427                 else
 428                         return -EEXIST;
 429         }
 430
 431         rb_link_node(node, parent_node, p);
 432         rb_insert_color(node, root);
 433         ins->delayed_node = delayed_node;
 434         ins->ins_or_del = action;
 435
 436         if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
 437             action == BTRFS_DELAYED_INSERTION_ITEM &&
 438             ins->key.offset >= delayed_node->index_cnt)
 439                         delayed_node->index_cnt = ins->key.offset + 1;
 440
 441         delayed_node->count++;
 442         atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
 443         return 0;
 444 }
 445
 446 static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
 447                                               struct btrfs_delayed_item *item)
 448 {
 449         return __btrfs_add_delayed_item(node, item,
 450                                         BTRFS_DELAYED_INSERTION_ITEM);
 451 }
 452
 453 static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
 454                                              struct btrfs_delayed_item *item)
 455 {
 456         return __btrfs_add_delayed_item(node, item,
 457                                         BTRFS_DELAYED_DELETION_ITEM);
 458 }
 459
 460 static void finish_one_item(struct btrfs_delayed_root *delayed_root)
 461 {
 462         int seq = atomic_inc_return(&delayed_root->items_seq);
 463         if ((atomic_dec_return(&delayed_root->items) <
 464             BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0) &&
 465             waitqueue_active(&delayed_root->wait))
 466                 wake_up(&delayed_root->wait);
 467 }
 468
 469 static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 470 {
 471         struct rb_root *root;
 472         struct btrfs_delayed_root *delayed_root;
 473
 474         delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
 475
 476         BUG_ON(!delayed_root);
 477         BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
 478                delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
 479
 480         if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
 481                 root = &delayed_item->delayed_node->ins_root;
 482         else
 483                 root = &delayed_item->delayed_node->del_root;
 484
 485         rb_erase(&delayed_item->rb_node, root);
 486         delayed_item->delayed_node->count--;
 487
 488         finish_one_item(delayed_root);
 489 }
 490
 491 static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 492 {
 493         if (item) {
 494                 __btrfs_remove_delayed_item(item);
 495                 if (atomic_dec_and_test(&item->refs))
 496                         kfree(item);
 497         }
 498 }
 499
 500 static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 501                                         struct btrfs_delayed_node *delayed_node)
 502 {
 503         struct rb_node *p;
 504         struct btrfs_delayed_item *item = NULL;
 505
 506         p = rb_first(&delayed_node->ins_root);
 507         if (p)
 508                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 509
 510         return item;
 511 }
 512
 513 static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 514                                         struct btrfs_delayed_node *delayed_node)
 515 {
 516         struct rb_node *p;
 517         struct btrfs_delayed_item *item = NULL;
 518
 519         p = rb_first(&delayed_node->del_root);
 520         if (p)
 521                 item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 522
 523         return item;
 524 }
 525
 526 static struct btrfs_delayed_item *__btrfs_next_delayed_item(
 527                                                 struct btrfs_delayed_item *item)
 528 {
 529         struct rb_node *p;
 530         struct btrfs_delayed_item *next = NULL;
 531
 532         p = rb_next(&item->rb_node);
 533         if (p)
 534                 next = rb_entry(p, struct btrfs_delayed_item, rb_node);
 535
 536         return next;
 537 }
 538
 539 static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 540                                                struct btrfs_root *root,
 541                                                struct btrfs_delayed_item *item)
 542 {
 543         struct btrfs_block_rsv *src_rsv;
 544         struct btrfs_block_rsv *dst_rsv;
 545         u64 num_bytes;
 546         int ret;
 547
 548         if (!trans->bytes_reserved)
 549                 return 0;
 550
 551         src_rsv = trans->block_rsv;
 552         dst_rsv = &root->fs_info->delayed_block_rsv;
 553
 554         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 555         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 556         if (!ret) {
 557                 trace_btrfs_space_reservation(root->fs_info, "delayed_item",
 558                                               item->key.objectid,
 559                                               num_bytes, 1);
 560                 item->bytes_reserved = num_bytes;
 561         }
 562
 563         return ret;
 564 }
 565
 566 static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 567                                                 struct btrfs_delayed_item *item)
 568 {
 569         struct btrfs_block_rsv *rsv;
 570
 571         if (!item->bytes_reserved)
 572                 return;
 573
 574         rsv = &root->fs_info->delayed_block_rsv;
 575         trace_btrfs_space_reservation(root->fs_info, "delayed_item",
 576                                       item->key.objectid, item->bytes_reserved,
 577                                       0);
 578         btrfs_block_rsv_release(root, rsv,
 579                                 item->bytes_reserved);
 580 }
 581
 582 static int btrfs_delayed_inode_reserve_metadata(
 583                                         struct btrfs_trans_handle *trans,
 584                                         struct btrfs_root *root,
 585                                         struct inode *inode,
 586                                         struct btrfs_delayed_node *node)
 587 {
 588         struct btrfs_block_rsv *src_rsv;
 589         struct btrfs_block_rsv *dst_rsv;
 590         u64 num_bytes;
 591         int ret;
 592         bool release = false;
 593
 594         src_rsv = trans->block_rsv;
 595         dst_rsv = &root->fs_info->delayed_block_rsv;
 596
 597         num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 598
 599         /*
 600          * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 601          * which doesn't reserve space for speed.  This is a problem since we
 602          * still need to reserve space for this update, so try to reserve the
 603          * space.
 604          *
 605          * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 606          * we're accounted for.
 607          */
 608         if (!src_rsv || (!trans->bytes_reserved &&
 609                          src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
 610                 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 611                                           BTRFS_RESERVE_NO_FLUSH);
 612                 /*
 613                  * Since we're under a transaction reserve_metadata_bytes could
 614                  * try to commit the transaction which will make it return
 615                  * EAGAIN to make us stop the transaction we have, so return
 616                  * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 617                  */
 618                 if (ret == -EAGAIN)
 619                         ret = -ENOSPC;
 620                 if (!ret) {
 621                         node->bytes_reserved = num_bytes;
 622                         trace_btrfs_space_reservation(root->fs_info,
 623                                                       "delayed_inode",
 624                                                       btrfs_ino(inode),
 625                                                       num_bytes, 1);
 626                 }
 627                 return ret;
 628         } else if (src_rsv->type == BTRFS_BLOCK_RSV_DELALLOC) {
 629                 spin_lock(&BTRFS_I(inode)->lock);
 630                 if (test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED,
 631                                        &BTRFS_I(inode)->runtime_flags)) {
 632                         spin_unlock(&BTRFS_I(inode)->lock);
 633                         release = true;
 634                         goto migrate;
 635                 }
 636                 spin_unlock(&BTRFS_I(inode)->lock);
 637
 638                 /* Ok we didn't have space pre-reserved.  This shouldn't happen
 639                  * too often but it can happen if we do delalloc to an existing
 640                  * inode which gets dirtied because of the time update, and then
 641                  * isn't touched again until after the transaction commits and
 642                  * then we try to write out the data.  First try to be nice and
 643                  * reserve something strictly for us.  If not be a pain and try
 644                  * to steal from the delalloc block rsv.
 645                  */
 646                 ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 647                                           BTRFS_RESERVE_NO_FLUSH);
 648                 if (!ret)
 649                         goto out;
 650
 651                 ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 652                 if (!WARN_ON(ret))
 653                         goto out;
 654
 655                 /*
 656                  * Ok this is a problem, let's just steal from the global rsv
 657                  * since this really shouldn't happen that often.
 658                  */
 659                 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
 660                                               dst_rsv, num_bytes);
 661                 goto out;
 662         }
 663
 664 migrate:
 665         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 666
 667 out:
 668         /*
 669          * Migrate only takes a reservation, it doesn't touch the size of the
 670          * block_rsv.  This is to simplify people who don't normally have things
 671          * migrated from their block rsv.  If they go to release their
 672          * reservation, that will decrease the size as well, so if migrate
 673          * reduced size we'd end up with a negative size.  But for the
 674          * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
 675          * but we could in fact do this reserve/migrate dance several times
 676          * between the time we did the original reservation and we'd clean it
 677          * up.  So to take care of this, release the space for the meta
 678          * reservation here.  I think it may be time for a documentation page on
 679          * how block rsvs. work.
 680          */
 681         if (!ret) {
 682                 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
 683                                               btrfs_ino(inode), num_bytes, 1);
 684                 node->bytes_reserved = num_bytes;
 685         }
 686
 687         if (release) {
 688                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
 689                                               btrfs_ino(inode), num_bytes, 0);
 690                 btrfs_block_rsv_release(root, src_rsv, num_bytes);
 691         }
 692
 693         return ret;
 694 }
 695
 696 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
 697                                                 struct btrfs_delayed_node *node)
 698 {
 699         struct btrfs_block_rsv *rsv;
 700
 701         if (!node->bytes_reserved)
 702                 return;
 703
 704         rsv = &root->fs_info->delayed_block_rsv;
 705         trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
 706                                       node->inode_id, node->bytes_reserved, 0);
 707         btrfs_block_rsv_release(root, rsv,
 708                                 node->bytes_reserved);
 709         node->bytes_reserved = 0;
 710 }
 711
 712 /*
 713  * This helper will insert some continuous items into the same leaf according
 714  * to the free space of the leaf.
 715  */
 716 static int btrfs_batch_insert_items(struct btrfs_root *root,
 717                                     struct btrfs_path *path,
 718                                     struct btrfs_delayed_item *item)
 719 {
 720         struct btrfs_delayed_item *curr, *next;
 721         int free_space;
 722         int total_data_size = 0, total_size = 0;
 723         struct extent_buffer *leaf;
 724         char *data_ptr;
 725         struct btrfs_key *keys;
 726         u32 *data_size;
 727         struct list_head head;
 728         int slot;
 729         int nitems;
 730         int i;
 731         int ret = 0;
 732
 733         BUG_ON(!path->nodes[0]);
 734
 735         leaf = path->nodes[0];
 736         free_space = btrfs_leaf_free_space(root, leaf);
 737         INIT_LIST_HEAD(&head);
 738
 739         next = item;
 740         nitems = 0;
 741
 742         /*
 743          * count the number of the continuous items that we can insert in batch
 744          */
 745         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 746                free_space) {
 747                 total_data_size += next->data_len;
 748                 total_size += next->data_len + sizeof(struct btrfs_item);
 749                 list_add_tail(&next->tree_list, &head);
 750                 nitems++;
 751
 752                 curr = next;
 753                 next = __btrfs_next_delayed_item(curr);
 754                 if (!next)
 755                         break;
 756
 757                 if (!btrfs_is_continuous_delayed_item(curr, next))
 758                         break;
 759         }
 760
 761         if (!nitems) {
 762                 ret = 0;
 763                 goto out;
 764         }
 765
 766         /*
 767          * we need allocate some memory space, but it might cause the task
 768          * to sleep, so we set all locked nodes in the path to blocking locks
 769          * first.
 770          */
 771         btrfs_set_path_blocking(path);
 772
 773         keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
 774         if (!keys) {
 775                 ret = -ENOMEM;
 776                 goto out;
 777         }
 778
 779         data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
 780         if (!data_size) {
 781                 ret = -ENOMEM;
 782                 goto error;
 783         }
 784
 785         /* get keys of all the delayed items */
 786         i = 0;
 787         list_for_each_entry(next, &head, tree_list) {
 788                 keys[i] = next->key;
 789                 data_size[i] = next->data_len;
 790                 i++;
 791         }
 792
 793         /* reset all the locked nodes in the patch to spinning locks. */
 794         btrfs_clear_path_blocking(path, NULL, 0);
 795
 796         /* insert the keys of the items */
 797         setup_items_for_insert(root, path, keys, data_size,
 798                                total_data_size, total_size, nitems);
 799
 800         /* insert the dir index items */
 801         slot = path->slots[0];
 802         list_for_each_entry_safe(curr, next, &head, tree_list) {
 803                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 804                 write_extent_buffer(leaf, &curr->data,
 805                                     (unsigned long)data_ptr,
 806                                     curr->data_len);
 807                 slot++;
 808
 809                 btrfs_delayed_item_release_metadata(root, curr);
 810
 811                 list_del(&curr->tree_list);
 812                 btrfs_release_delayed_item(curr);
 813         }
 814
 815 error:
 816         kfree(data_size);
 817         kfree(keys);
 818 out:
 819         return ret;
 820 }
 821
 822 /*
 823  * This helper can just do simple insertion that needn't extend item for new
 824  * data, such as directory name index insertion, inode insertion.
 825  */
 826 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 827                                      struct btrfs_root *root,
 828                                      struct btrfs_path *path,
 829                                      struct btrfs_delayed_item *delayed_item)
 830 {
 831         struct extent_buffer *leaf;
 832         char *ptr;
 833         int ret;
 834
 835         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 836                                       delayed_item->data_len);
 837         if (ret < 0 && ret != -EEXIST)
 838                 return ret;
 839
 840         leaf = path->nodes[0];
 841
 842         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 843
 844         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 845                             delayed_item->data_len);
 846         btrfs_mark_buffer_dirty(leaf);
 847
 848         btrfs_delayed_item_release_metadata(root, delayed_item);
 849         return 0;
 850 }
 851
 852 /*
 853  * we insert an item first, then if there are some continuous items, we try
 854  * to insert those items into the same leaf.
 855  */
 856 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 857                                       struct btrfs_path *path,
 858                                       struct btrfs_root *root,
 859                                       struct btrfs_delayed_node *node)
 860 {
 861         struct btrfs_delayed_item *curr, *prev;
 862         int ret = 0;
 863
 864 do_again:
 865         mutex_lock(&node->mutex);
 866         curr = __btrfs_first_delayed_insertion_item(node);
 867         if (!curr)
 868                 goto insert_end;
 869
 870         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 871         if (ret < 0) {
 872                 btrfs_release_path(path);
 873                 goto insert_end;
 874         }
 875
 876         prev = curr;
 877         curr = __btrfs_next_delayed_item(prev);
 878         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 879                 /* insert the continuous items into the same leaf */
 880                 path->slots[0]++;
 881                 btrfs_batch_insert_items(root, path, curr);
 882         }
 883         btrfs_release_delayed_item(prev);
 884         btrfs_mark_buffer_dirty(path->nodes[0]);
 885
 886         btrfs_release_path(path);
 887         mutex_unlock(&node->mutex);
 888         goto do_again;
 889
 890 insert_end:
 891         mutex_unlock(&node->mutex);
 892         return ret;
 893 }
 894
 895 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 896                                     struct btrfs_root *root,
 897                                     struct btrfs_path *path,
 898                                     struct btrfs_delayed_item *item)
 899 {
 900         struct btrfs_delayed_item *curr, *next;
 901         struct extent_buffer *leaf;
 902         struct btrfs_key key;
 903         struct list_head head;
 904         int nitems, i, last_item;
 905         int ret = 0;
 906
 907         BUG_ON(!path->nodes[0]);
 908
 909         leaf = path->nodes[0];
 910
 911         i = path->slots[0];
 912         last_item = btrfs_header_nritems(leaf) - 1;
 913         if (i > last_item)
 914                 return -ENOENT; /* FIXME: Is errno suitable? */
 915
 916         next = item;
 917         INIT_LIST_HEAD(&head);
 918         btrfs_item_key_to_cpu(leaf, &key, i);
 919         nitems = 0;
 920         /*
 921          * count the number of the dir index items that we can delete in batch
 922          */
 923         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 924                 list_add_tail(&next->tree_list, &head);
 925                 nitems++;
 926
 927                 curr = next;
 928                 next = __btrfs_next_delayed_item(curr);
 929                 if (!next)
 930                         break;
 931
 932                 if (!btrfs_is_continuous_delayed_item(curr, next))
 933                         break;
 934
 935                 i++;
 936                 if (i > last_item)
 937                         break;
 938                 btrfs_item_key_to_cpu(leaf, &key, i);
 939         }
 940
 941         if (!nitems)
 942                 return 0;
 943
 944         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 945         if (ret)
 946                 goto out;
 947
 948         list_for_each_entry_safe(curr, next, &head, tree_list) {
 949                 btrfs_delayed_item_release_metadata(root, curr);
 950                 list_del(&curr->tree_list);
 951                 btrfs_release_delayed_item(curr);
 952         }
 953
 954 out:
 955         return ret;
 956 }
 957
 958 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 959                                       struct btrfs_path *path,
 960                                       struct btrfs_root *root,
 961                                       struct btrfs_delayed_node *node)
 962 {
 963         struct btrfs_delayed_item *curr, *prev;
 964         int ret = 0;
 965
 966 do_again:
 967         mutex_lock(&node->mutex);
 968         curr = __btrfs_first_delayed_deletion_item(node);
 969         if (!curr)
 970                 goto delete_fail;
 971
 972         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 973         if (ret < 0)
 974                 goto delete_fail;
 975         else if (ret > 0) {
 976                 /*
 977                  * can't find the item which the node points to, so this node
 978                  * is invalid, just drop it.
 979                  */
 980                 prev = curr;
 981                 curr = __btrfs_next_delayed_item(prev);
 982                 btrfs_release_delayed_item(prev);
 983                 ret = 0;
 984                 btrfs_release_path(path);
 985                 if (curr) {
 986                         mutex_unlock(&node->mutex);
 987                         goto do_again;
 988                 } else
 989                         goto delete_fail;
 990         }
 991
 992         btrfs_batch_delete_items(trans, root, path, curr);
 993         btrfs_release_path(path);
 994         mutex_unlock(&node->mutex);
 995         goto do_again;
 996
 997 delete_fail:
 998         btrfs_release_path(path);
 999         mutex_unlock(&node->mutex);
1000         return ret;
1001 }
1002
1003 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1004 {
1005         struct btrfs_delayed_root *delayed_root;
1006
1007         if (delayed_node &&
1008             test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1009                 BUG_ON(!delayed_node->root);
1010                 clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1011                 delayed_node->count--;
1012
1013                 delayed_root = delayed_node->root->fs_info->delayed_root;
1014                 finish_one_item(delayed_root);
1015         }
1016 }
1017
1018 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1019                                         struct btrfs_root *root,
1020                                         struct btrfs_path *path,
1021                                         struct btrfs_delayed_node *node)
1022 {
1023         struct btrfs_key key;
1024         struct btrfs_inode_item *inode_item;
1025         struct extent_buffer *leaf;
1026         int ret;
1027
1028         key.objectid = node->inode_id;
1029         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1030         key.offset = 0;
1031
1032         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1033         if (ret > 0) {
1034                 btrfs_release_path(path);
1035                 return -ENOENT;
1036         } else if (ret < 0) {
1037                 return ret;
1038         }
1039
1040         btrfs_unlock_up_safe(path, 1);
1041         leaf = path->nodes[0];
1042         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1043                                     struct btrfs_inode_item);
1044         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1045                             sizeof(struct btrfs_inode_item));
1046         btrfs_mark_buffer_dirty(leaf);
1047         btrfs_release_path(path);
1048
1049         btrfs_delayed_inode_release_metadata(root, node);
1050         btrfs_release_delayed_inode(node);
1051
1052         return 0;
1053 }
1054
1055 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1056                                              struct btrfs_root *root,
1057                                              struct btrfs_path *path,
1058                                              struct btrfs_delayed_node *node)
1059 {
1060         int ret;
1061
1062         mutex_lock(&node->mutex);
1063         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1064                 mutex_unlock(&node->mutex);
1065                 return 0;
1066         }
1067
1068         ret = __btrfs_update_delayed_inode(trans, root, path, node);
1069         mutex_unlock(&node->mutex);
1070         return ret;
1071 }
1072
1073 static inline int
1074 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1075                                    struct btrfs_path *path,
1076                                    struct btrfs_delayed_node *node)
1077 {
1078         int ret;
1079
1080         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1081         if (ret)
1082                 return ret;
1083
1084         ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1085         if (ret)
1086                 return ret;
1087
1088         ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1089         return ret;
1090 }
1091
1092 /*
1093  * Called when committing the transaction.
1094  * Returns 0 on success.
1095  * Returns < 0 on error and returns with an aborted transaction with any
1096  * outstanding delayed items cleaned up.
1097  */
1098 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1099                                      struct btrfs_root *root, int nr)
1100 {
1101         struct btrfs_delayed_root *delayed_root;
1102         struct btrfs_delayed_node *curr_node, *prev_node;
1103         struct btrfs_path *path;
1104         struct btrfs_block_rsv *block_rsv;
1105         int ret = 0;
1106         bool count = (nr > 0);
1107
1108         if (trans->aborted)
1109                 return -EIO;
1110
1111         path = btrfs_alloc_path();
1112         if (!path)
1113                 return -ENOMEM;
1114         path->leave_spinning = 1;
1115
1116         block_rsv = trans->block_rsv;
1117         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1118
1119         delayed_root = btrfs_get_delayed_root(root);
1120
1121         curr_node = btrfs_first_delayed_node(delayed_root);
1122         while (curr_node && (!count || (count && nr--))) {
1123                 ret = __btrfs_commit_inode_delayed_items(trans, path,
1124                                                          curr_node);
1125                 if (ret) {
1126                         btrfs_release_delayed_node(curr_node);
1127                         curr_node = NULL;
1128                         btrfs_abort_transaction(trans, root, ret);
1129                         break;
1130                 }
1131
1132                 prev_node = curr_node;
1133                 curr_node = btrfs_next_delayed_node(curr_node);
1134                 btrfs_release_delayed_node(prev_node);
1135         }
1136
1137         if (curr_node)
1138                 btrfs_release_delayed_node(curr_node);
1139         btrfs_free_path(path);
1140         trans->block_rsv = block_rsv;
1141
1142         return ret;
1143 }
1144
1145 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1146                             struct btrfs_root *root)
1147 {
1148         return __btrfs_run_delayed_items(trans, root, -1);
1149 }
1150
1151 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1152                                struct btrfs_root *root, int nr)
1153 {
1154         return __btrfs_run_delayed_items(trans, root, nr);
1155 }
1156
1157 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1158                                      struct inode *inode)
1159 {
1160         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1161         struct btrfs_path *path;
1162         struct btrfs_block_rsv *block_rsv;
1163         int ret;
1164
1165         if (!delayed_node)
1166                 return 0;
1167
1168         mutex_lock(&delayed_node->mutex);
1169         if (!delayed_node->count) {
1170                 mutex_unlock(&delayed_node->mutex);
1171                 btrfs_release_delayed_node(delayed_node);
1172                 return 0;
1173         }
1174         mutex_unlock(&delayed_node->mutex);
1175
1176         path = btrfs_alloc_path();
1177         if (!path) {
1178                 btrfs_release_delayed_node(delayed_node);
1179                 return -ENOMEM;
1180         }
1181         path->leave_spinning = 1;
1182
1183         block_rsv = trans->block_rsv;
1184         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1185
1186         ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1187
1188         btrfs_release_delayed_node(delayed_node);
1189         btrfs_free_path(path);
1190         trans->block_rsv = block_rsv;
1191
1192         return ret;
1193 }
1194
1195 int btrfs_commit_inode_delayed_inode(struct inode *inode)
1196 {
1197         struct btrfs_trans_handle *trans;
1198         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1199         struct btrfs_path *path;
1200         struct btrfs_block_rsv *block_rsv;
1201         int ret;
1202
1203         if (!delayed_node)
1204                 return 0;
1205
1206         mutex_lock(&delayed_node->mutex);
1207         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1208                 mutex_unlock(&delayed_node->mutex);
1209                 btrfs_release_delayed_node(delayed_node);
1210                 return 0;
1211         }
1212         mutex_unlock(&delayed_node->mutex);
1213
1214         trans = btrfs_join_transaction(delayed_node->root);
1215         if (IS_ERR(trans)) {
1216                 ret = PTR_ERR(trans);
1217                 goto out;
1218         }
1219
1220         path = btrfs_alloc_path();
1221         if (!path) {
1222                 ret = -ENOMEM;
1223                 goto trans_out;
1224         }
1225         path->leave_spinning = 1;
1226
1227         block_rsv = trans->block_rsv;
1228         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1229
1230         mutex_lock(&delayed_node->mutex);
1231         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1232                 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1233                                                    path, delayed_node);
1234         else
1235                 ret = 0;
1236         mutex_unlock(&delayed_node->mutex);
1237
1238         btrfs_free_path(path);
1239         trans->block_rsv = block_rsv;
1240 trans_out:
1241         btrfs_end_transaction(trans, delayed_node->root);
1242         btrfs_btree_balance_dirty(delayed_node->root);
1243 out:
1244         btrfs_release_delayed_node(delayed_node);
1245
1246         return ret;
1247 }
1248
1249 void btrfs_remove_delayed_node(struct inode *inode)
1250 {
1251         struct btrfs_delayed_node *delayed_node;
1252
1253         delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1254         if (!delayed_node)
1255                 return;
1256
1257         BTRFS_I(inode)->delayed_node = NULL;
1258         btrfs_release_delayed_node(delayed_node);
1259 }
1260
1261 struct btrfs_async_delayed_work {
1262         struct btrfs_delayed_root *delayed_root;
1263         int nr;
1264         struct btrfs_work work;
1265 };
1266
1267 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1268 {
1269         struct btrfs_async_delayed_work *async_work;
1270         struct btrfs_delayed_root *delayed_root;
1271         struct btrfs_trans_handle *trans;
1272         struct btrfs_path *path;
1273         struct btrfs_delayed_node *delayed_node = NULL;
1274         struct btrfs_root *root;
1275         struct btrfs_block_rsv *block_rsv;
1276         int total_done = 0;
1277
1278         async_work = container_of(work, struct btrfs_async_delayed_work, work);
1279         delayed_root = async_work->delayed_root;
1280
1281         path = btrfs_alloc_path();
1282         if (!path)
1283                 goto out;
1284
1285 again:
1286         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1287                 goto free_path;
1288
1289         delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1290         if (!delayed_node)
1291                 goto free_path;
1292
1293         path->leave_spinning = 1;
1294         root = delayed_node->root;
1295
1296         trans = btrfs_join_transaction(root);
1297         if (IS_ERR(trans))
1298                 goto release_path;
1299
1300         block_rsv = trans->block_rsv;
1301         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1302
1303         __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1304
1305         trans->block_rsv = block_rsv;
1306         btrfs_end_transaction(trans, root);
1307         btrfs_btree_balance_dirty_nodelay(root);
1308
1309 release_path:
1310         btrfs_release_path(path);
1311         total_done++;
1312
1313         btrfs_release_prepared_delayed_node(delayed_node);
1314         if (async_work->nr == 0 || total_done < async_work->nr)
1315                 goto again;
1316
1317 free_path:
1318         btrfs_free_path(path);
1319 out:
1320         wake_up(&delayed_root->wait);
1321         kfree(async_work);
1322 }
1323
1324
1325 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1326                                      struct btrfs_root *root, int nr)
1327 {
1328         struct btrfs_async_delayed_work *async_work;
1329
1330         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1331                 return 0;
1332
1333         async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1334         if (!async_work)
1335                 return -ENOMEM;
1336
1337         async_work->delayed_root = delayed_root;
1338         async_work->work.func = btrfs_async_run_delayed_root;
1339         async_work->work.flags = 0;
1340         async_work->nr = nr;
1341
1342         btrfs_queue_worker(&root->fs_info->delayed_workers, &async_work->work);
1343         return 0;
1344 }
1345
1346 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1347 {
1348         struct btrfs_delayed_root *delayed_root;
1349         delayed_root = btrfs_get_delayed_root(root);
1350         WARN_ON(btrfs_first_delayed_node(delayed_root));
1351 }
1352
1353 static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1354 {
1355         int val = atomic_read(&delayed_root->items_seq);
1356
1357         if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1358                 return 1;
1359
1360         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1361                 return 1;
1362
1363         return 0;
1364 }
1365
1366 void btrfs_balance_delayed_items(struct btrfs_root *root)
1367 {
1368         struct btrfs_delayed_root *delayed_root;
1369
1370         delayed_root = btrfs_get_delayed_root(root);
1371
1372         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1373                 return;
1374
1375         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1376                 int seq;
1377                 int ret;
1378
1379                 seq = atomic_read(&delayed_root->items_seq);
1380
1381                 ret = btrfs_wq_run_delayed_node(delayed_root, root, 0);
1382                 if (ret)
1383                         return;
1384
1385                 wait_event_interruptible(delayed_root->wait,
1386                                          could_end_wait(delayed_root, seq));
1387                 return;
1388         }
1389
1390         btrfs_wq_run_delayed_node(delayed_root, root, BTRFS_DELAYED_BATCH);
1391 }
1392
1393 /* Will return 0 or -ENOMEM */
1394 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1395                                    struct btrfs_root *root, const char *name,
1396                                    int name_len, struct inode *dir,
1397                                    struct btrfs_disk_key *disk_key, u8 type,
1398                                    u64 index)
1399 {
1400         struct btrfs_delayed_node *delayed_node;
1401         struct btrfs_delayed_item *delayed_item;
1402         struct btrfs_dir_item *dir_item;
1403         int ret;
1404
1405         delayed_node = btrfs_get_or_create_delayed_node(dir);
1406         if (IS_ERR(delayed_node))
1407                 return PTR_ERR(delayed_node);
1408
1409         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1410         if (!delayed_item) {
1411                 ret = -ENOMEM;
1412                 goto release_node;
1413         }
1414
1415         delayed_item->key.objectid = btrfs_ino(dir);
1416         btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1417         delayed_item->key.offset = index;
1418
1419         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1420         dir_item->location = *disk_key;
1421         btrfs_set_stack_dir_transid(dir_item, trans->transid);
1422         btrfs_set_stack_dir_data_len(dir_item, 0);
1423         btrfs_set_stack_dir_name_len(dir_item, name_len);
1424         btrfs_set_stack_dir_type(dir_item, type);
1425         memcpy((char *)(dir_item + 1), name, name_len);
1426
1427         ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1428         /*
1429          * we have reserved enough space when we start a new transaction,
1430          * so reserving metadata failure is impossible
1431          */
1432         BUG_ON(ret);
1433
1434
1435         mutex_lock(&delayed_node->mutex);
1436         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1437         if (unlikely(ret)) {
1438                 btrfs_err(root->fs_info, "err add delayed dir index item(name: %.*s) "
1439                                 "into the insertion tree of the delayed node"
1440                                 "(root id: %llu, inode id: %llu, errno: %d)",
1441                                 name_len, name, delayed_node->root->objectid,
1442                                 delayed_node->inode_id, ret);
1443                 BUG();
1444         }
1445         mutex_unlock(&delayed_node->mutex);
1446
1447 release_node:
1448         btrfs_release_delayed_node(delayed_node);
1449         return ret;
1450 }
1451
1452 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1453                                                struct btrfs_delayed_node *node,
1454                                                struct btrfs_key *key)
1455 {
1456         struct btrfs_delayed_item *item;
1457
1458         mutex_lock(&node->mutex);
1459         item = __btrfs_lookup_delayed_insertion_item(node, key);
1460         if (!item) {
1461                 mutex_unlock(&node->mutex);
1462                 return 1;
1463         }
1464
1465         btrfs_delayed_item_release_metadata(root, item);
1466         btrfs_release_delayed_item(item);
1467         mutex_unlock(&node->mutex);
1468         return 0;
1469 }
1470
1471 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1472                                    struct btrfs_root *root, struct inode *dir,
1473                                    u64 index)
1474 {
1475         struct btrfs_delayed_node *node;
1476         struct btrfs_delayed_item *item;
1477         struct btrfs_key item_key;
1478         int ret;
1479
1480         node = btrfs_get_or_create_delayed_node(dir);
1481         if (IS_ERR(node))
1482                 return PTR_ERR(node);
1483
1484         item_key.objectid = btrfs_ino(dir);
1485         btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1486         item_key.offset = index;
1487
1488         ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1489         if (!ret)
1490                 goto end;
1491
1492         item = btrfs_alloc_delayed_item(0);
1493         if (!item) {
1494                 ret = -ENOMEM;
1495                 goto end;
1496         }
1497
1498         item->key = item_key;
1499
1500         ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1501         /*
1502          * we have reserved enough space when we start a new transaction,
1503          * so reserving metadata failure is impossible.
1504          */
1505         BUG_ON(ret);
1506
1507         mutex_lock(&node->mutex);
1508         ret = __btrfs_add_delayed_deletion_item(node, item);
1509         if (unlikely(ret)) {
1510                 btrfs_err(root->fs_info, "err add delayed dir index item(index: %llu) "
1511                                 "into the deletion tree of the delayed node"
1512                                 "(root id: %llu, inode id: %llu, errno: %d)",
1513                                 index, node->root->objectid, node->inode_id,
1514                                 ret);
1515                 BUG();
1516         }
1517         mutex_unlock(&node->mutex);
1518 end:
1519         btrfs_release_delayed_node(node);
1520         return ret;
1521 }
1522
1523 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1524 {
1525         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1526
1527         if (!delayed_node)
1528                 return -ENOENT;
1529
1530         /*
1531          * Since we have held i_mutex of this directory, it is impossible that
1532          * a new directory index is added into the delayed node and index_cnt
1533          * is updated now. So we needn't lock the delayed node.
1534          */
1535         if (!delayed_node->index_cnt) {
1536                 btrfs_release_delayed_node(delayed_node);
1537                 return -EINVAL;
1538         }
1539
1540         BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1541         btrfs_release_delayed_node(delayed_node);
1542         return 0;
1543 }
1544
1545 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1546                              struct list_head *del_list)
1547 {
1548         struct btrfs_delayed_node *delayed_node;
1549         struct btrfs_delayed_item *item;
1550
1551         delayed_node = btrfs_get_delayed_node(inode);
1552         if (!delayed_node)
1553                 return;
1554
1555         mutex_lock(&delayed_node->mutex);
1556         item = __btrfs_first_delayed_insertion_item(delayed_node);
1557         while (item) {
1558                 atomic_inc(&item->refs);
1559                 list_add_tail(&item->readdir_list, ins_list);
1560                 item = __btrfs_next_delayed_item(item);
1561         }
1562
1563         item = __btrfs_first_delayed_deletion_item(delayed_node);
1564         while (item) {
1565                 atomic_inc(&item->refs);
1566                 list_add_tail(&item->readdir_list, del_list);
1567                 item = __btrfs_next_delayed_item(item);
1568         }
1569         mutex_unlock(&delayed_node->mutex);
1570         /*
1571          * This delayed node is still cached in the btrfs inode, so refs
1572          * must be > 1 now, and we needn't check it is going to be freed
1573          * or not.
1574          *
1575          * Besides that, this function is used to read dir, we do not
1576          * insert/delete delayed items in this period. So we also needn't
1577          * requeue or dequeue this delayed node.
1578          */
1579         atomic_dec(&delayed_node->refs);
1580 }
1581
1582 void btrfs_put_delayed_items(struct list_head *ins_list,
1583                              struct list_head *del_list)
1584 {
1585         struct btrfs_delayed_item *curr, *next;
1586
1587         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1588                 list_del(&curr->readdir_list);
1589                 if (atomic_dec_and_test(&curr->refs))
1590                         kfree(curr);
1591         }
1592
1593         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1594                 list_del(&curr->readdir_list);
1595                 if (atomic_dec_and_test(&curr->refs))
1596                         kfree(curr);
1597         }
1598 }
1599
1600 int btrfs_should_delete_dir_index(struct list_head *del_list,
1601                                   u64 index)
1602 {
1603         struct btrfs_delayed_item *curr, *next;
1604         int ret;
1605
1606         if (list_empty(del_list))
1607                 return 0;
1608
1609         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1610                 if (curr->key.offset > index)
1611                         break;
1612
1613                 list_del(&curr->readdir_list);
1614                 ret = (curr->key.offset == index);
1615
1616                 if (atomic_dec_and_test(&curr->refs))
1617                         kfree(curr);
1618
1619                 if (ret)
1620                         return 1;
1621                 else
1622                         continue;
1623         }
1624         return 0;
1625 }
1626
1627 /*
1628  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1629  *
1630  */
1631 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1632                                     struct list_head *ins_list)
1633 {
1634         struct btrfs_dir_item *di;
1635         struct btrfs_delayed_item *curr, *next;
1636         struct btrfs_key location;
1637         char *name;
1638         int name_len;
1639         int over = 0;
1640         unsigned char d_type;
1641
1642         if (list_empty(ins_list))
1643                 return 0;
1644
1645         /*
1646          * Changing the data of the delayed item is impossible. So
1647          * we needn't lock them. And we have held i_mutex of the
1648          * directory, nobody can delete any directory indexes now.
1649          */
1650         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1651                 list_del(&curr->readdir_list);
1652
1653                 if (curr->key.offset < ctx->pos) {
1654                         if (atomic_dec_and_test(&curr->refs))
1655                                 kfree(curr);
1656                         continue;
1657                 }
1658
1659                 ctx->pos = curr->key.offset;
1660
1661                 di = (struct btrfs_dir_item *)curr->data;
1662                 name = (char *)(di + 1);
1663                 name_len = btrfs_stack_dir_name_len(di);
1664
1665                 d_type = btrfs_filetype_table[di->type];
1666                 btrfs_disk_key_to_cpu(&location, &di->location);
1667
1668                 over = !dir_emit(ctx, name, name_len,
1669                                location.objectid, d_type);
1670
1671                 if (atomic_dec_and_test(&curr->refs))
1672                         kfree(curr);
1673
1674                 if (over)
1675                         return 1;
1676         }
1677         return 0;
1678 }
1679
1680 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1681                                   struct btrfs_inode_item *inode_item,
1682                                   struct inode *inode)
1683 {
1684         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1685         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1686         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1687         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1688         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1689         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1690         btrfs_set_stack_inode_generation(inode_item,
1691                                          BTRFS_I(inode)->generation);
1692         btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1693         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1694         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1695         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1696         btrfs_set_stack_inode_block_group(inode_item, 0);
1697
1698         btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1699                                      inode->i_atime.tv_sec);
1700         btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1701                                       inode->i_atime.tv_nsec);
1702
1703         btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1704                                      inode->i_mtime.tv_sec);
1705         btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1706                                       inode->i_mtime.tv_nsec);
1707
1708         btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1709                                      inode->i_ctime.tv_sec);
1710         btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1711                                       inode->i_ctime.tv_nsec);
1712 }
1713
1714 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1715 {
1716         struct btrfs_delayed_node *delayed_node;
1717         struct btrfs_inode_item *inode_item;
1718         struct btrfs_timespec *tspec;
1719
1720         delayed_node = btrfs_get_delayed_node(inode);
1721         if (!delayed_node)
1722                 return -ENOENT;
1723
1724         mutex_lock(&delayed_node->mutex);
1725         if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1726                 mutex_unlock(&delayed_node->mutex);
1727                 btrfs_release_delayed_node(delayed_node);
1728                 return -ENOENT;
1729         }
1730
1731         inode_item = &delayed_node->inode_item;
1732
1733         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1734         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1735         btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1736         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1737         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1738         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1739         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1740         inode->i_version = btrfs_stack_inode_sequence(inode_item);
1741         inode->i_rdev = 0;
1742         *rdev = btrfs_stack_inode_rdev(inode_item);
1743         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1744
1745         tspec = btrfs_inode_atime(inode_item);
1746         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1747         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1748
1749         tspec = btrfs_inode_mtime(inode_item);
1750         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1751         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1752
1753         tspec = btrfs_inode_ctime(inode_item);
1754         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1755         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1756
1757         inode->i_generation = BTRFS_I(inode)->generation;
1758         BTRFS_I(inode)->index_cnt = (u64)-1;
1759
1760         mutex_unlock(&delayed_node->mutex);
1761         btrfs_release_delayed_node(delayed_node);
1762         return 0;
1763 }
1764
1765 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1766                                struct btrfs_root *root, struct inode *inode)
1767 {
1768         struct btrfs_delayed_node *delayed_node;
1769         int ret = 0;
1770
1771         delayed_node = btrfs_get_or_create_delayed_node(inode);
1772         if (IS_ERR(delayed_node))
1773                 return PTR_ERR(delayed_node);
1774
1775         mutex_lock(&delayed_node->mutex);
1776         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1777                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1778                 goto release_node;
1779         }
1780
1781         ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1782                                                    delayed_node);
1783         if (ret)
1784                 goto release_node;
1785
1786         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1787         set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1788         delayed_node->count++;
1789         atomic_inc(&root->fs_info->delayed_root->items);
1790 release_node:
1791         mutex_unlock(&delayed_node->mutex);
1792         btrfs_release_delayed_node(delayed_node);
1793         return ret;
1794 }
1795
1796 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1797 {
1798         struct btrfs_root *root = delayed_node->root;
1799         struct btrfs_delayed_item *curr_item, *prev_item;
1800
1801         mutex_lock(&delayed_node->mutex);
1802         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1803         while (curr_item) {
1804                 btrfs_delayed_item_release_metadata(root, curr_item);
1805                 prev_item = curr_item;
1806                 curr_item = __btrfs_next_delayed_item(prev_item);
1807                 btrfs_release_delayed_item(prev_item);
1808         }
1809
1810         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1811         while (curr_item) {
1812                 btrfs_delayed_item_release_metadata(root, curr_item);
1813                 prev_item = curr_item;
1814                 curr_item = __btrfs_next_delayed_item(prev_item);
1815                 btrfs_release_delayed_item(prev_item);
1816         }
1817
1818         if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1819                 btrfs_delayed_inode_release_metadata(root, delayed_node);
1820                 btrfs_release_delayed_inode(delayed_node);
1821         }
1822         mutex_unlock(&delayed_node->mutex);
1823 }
1824
1825 void btrfs_kill_delayed_inode_items(struct inode *inode)
1826 {
1827         struct btrfs_delayed_node *delayed_node;
1828
1829         delayed_node = btrfs_get_delayed_node(inode);
1830         if (!delayed_node)
1831                 return;
1832
1833         __btrfs_kill_delayed_node(delayed_node);
1834         btrfs_release_delayed_node(delayed_node);
1835 }
1836
1837 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1838 {
1839         u64 inode_id = 0;
1840         struct btrfs_delayed_node *delayed_nodes[8];
1841         int i, n;
1842
1843         while (1) {
1844                 spin_lock(&root->inode_lock);
1845                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1846                                            (void **)delayed_nodes, inode_id,
1847                                            ARRAY_SIZE(delayed_nodes));
1848                 if (!n) {
1849                         spin_unlock(&root->inode_lock);
1850                         break;
1851                 }
1852
1853                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1854
1855                 for (i = 0; i < n; i++)
1856                         atomic_inc(&delayed_nodes[i]->refs);
1857                 spin_unlock(&root->inode_lock);
1858
1859                 for (i = 0; i < n; i++) {
1860                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1861                         btrfs_release_delayed_node(delayed_nodes[i]);
1862                 }
1863         }
1864 }
1865
1866 void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1867 {
1868         struct btrfs_delayed_root *delayed_root;
1869         struct btrfs_delayed_node *curr_node, *prev_node;
1870
1871         delayed_root = btrfs_get_delayed_root(root);
1872
1873         curr_node = btrfs_first_delayed_node(delayed_root);
1874         while (curr_node) {
1875                 __btrfs_kill_delayed_node(curr_node);
1876
1877                 prev_node = curr_node;
1878                 curr_node = btrfs_next_delayed_node(curr_node);
1879                 btrfs_release_delayed_node(prev_node);
1880         }
1881 }
1882