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->in_list = 0;
  59         delayed_node->inode_dirty = 0;
  60         delayed_node->ins_root = RB_ROOT;
  61         delayed_node->del_root = RB_ROOT;
  62         mutex_init(&delayed_node->mutex);
  63         delayed_node->index_cnt = 0;
  64         INIT_LIST_HEAD(&delayed_node->n_list);
  65         INIT_LIST_HEAD(&delayed_node->p_list);
  66         delayed_node->bytes_reserved = 0;
  67         memset(&delayed_node->inode_item, 0, sizeof(delayed_node->inode_item));
  68 }
  69
  70 static inline int btrfs_is_continuous_delayed_item(
  71                                         struct btrfs_delayed_item *item1,
  72                                         struct btrfs_delayed_item *item2)
  73 {
  74         if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  75             item1->key.objectid == item2->key.objectid &&
  76             item1->key.type == item2->key.type &&
  77             item1->key.offset + 1 == item2->key.offset)
  78                 return 1;
  79         return 0;
  80 }
  81
  82 static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
  83                                                         struct btrfs_root *root)
  84 {
  85         return root->fs_info->delayed_root;
  86 }
  87
  88 static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
  89 {
  90         struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
  91         struct btrfs_root *root = btrfs_inode->root;
  92         u64 ino = btrfs_ino(inode);
  93         struct btrfs_delayed_node *node;
  94
  95         node = ACCESS_ONCE(btrfs_inode->delayed_node);
  96         if (node) {
  97                 atomic_inc(&node->refs);
  98                 return node;
  99         }
 100
 101         spin_lock(&root->inode_lock);
 102         node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
 103         if (node) {
 104                 if (btrfs_inode->delayed_node) {
 105                         atomic_inc(&node->refs);        /* can be accessed */
 106                         BUG_ON(btrfs_inode->delayed_node != node);
 107                         spin_unlock(&root->inode_lock);
 108                         return node;
 109                 }
 110                 btrfs_inode->delayed_node = node;
 111                 atomic_inc(&node->refs);        /* can be accessed */
 112                 atomic_inc(&node->refs);        /* cached in the inode */
 113                 spin_unlock(&root->inode_lock);
 114                 return node;
 115         }
 116         spin_unlock(&root->inode_lock);
 117
 118         return NULL;
 119 }
 120
 121 /* Will return either the node or PTR_ERR(-ENOMEM) */
 122 static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 123                                                         struct inode *inode)
 124 {
 125         struct btrfs_delayed_node *node;
 126         struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
 127         struct btrfs_root *root = btrfs_inode->root;
 128         u64 ino = btrfs_ino(inode);
 129         int ret;
 130
 131 again:
 132         node = btrfs_get_delayed_node(inode);
 133         if (node)
 134                 return node;
 135
 136         node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
 137         if (!node)
 138                 return ERR_PTR(-ENOMEM);
 139         btrfs_init_delayed_node(node, root, ino);
 140
 141         atomic_inc(&node->refs);        /* cached in the btrfs inode */
 142         atomic_inc(&node->refs);        /* can be accessed */
 143
 144         ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 145         if (ret) {
 146                 kmem_cache_free(delayed_node_cache, node);
 147                 return ERR_PTR(ret);
 148         }
 149
 150         spin_lock(&root->inode_lock);
 151         ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 152         if (ret == -EEXIST) {
 153                 kmem_cache_free(delayed_node_cache, node);
 154                 spin_unlock(&root->inode_lock);
 155                 radix_tree_preload_end();
 156                 goto again;
 157         }
 158         btrfs_inode->delayed_node = node;
 159         spin_unlock(&root->inode_lock);
 160         radix_tree_preload_end();
 161
 162         return node;
 163 }
 164
 165 /*
 166  * Call it when holding delayed_node->mutex
 167  *
 168  * If mod = 1, add this node into the prepared list.
 169  */
 170 static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 171                                      struct btrfs_delayed_node *node,
 172                                      int mod)
 173 {
 174         spin_lock(&root->lock);
 175         if (node->in_list) {
 176                 if (!list_empty(&node->p_list))
 177                         list_move_tail(&node->p_list, &root->prepare_list);
 178                 else if (mod)
 179                         list_add_tail(&node->p_list, &root->prepare_list);
 180         } else {
 181                 list_add_tail(&node->n_list, &root->node_list);
 182                 list_add_tail(&node->p_list, &root->prepare_list);
 183                 atomic_inc(&node->refs);        /* inserted into list */
 184                 root->nodes++;
 185                 node->in_list = 1;
 186         }
 187         spin_unlock(&root->lock);
 188 }
 189
 190 /* Call it when holding delayed_node->mutex */
 191 static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 192                                        struct btrfs_delayed_node *node)
 193 {
 194         spin_lock(&root->lock);
 195         if (node->in_list) {
 196                 root->nodes--;
 197                 atomic_dec(&node->refs);        /* not in the list */
 198                 list_del_init(&node->n_list);
 199                 if (!list_empty(&node->p_list))
 200                         list_del_init(&node->p_list);
 201                 node->in_list = 0;
 202         }
 203         spin_unlock(&root->lock);
 204 }
 205
 206 static struct btrfs_delayed_node *btrfs_first_delayed_node(
 207                         struct btrfs_delayed_root *delayed_root)
 208 {
 209         struct list_head *p;
 210         struct btrfs_delayed_node *node = NULL;
 211
 212         spin_lock(&delayed_root->lock);
 213         if (list_empty(&delayed_root->node_list))
 214                 goto out;
 215
 216         p = delayed_root->node_list.next;
 217         node = list_entry(p, struct btrfs_delayed_node, n_list);
 218         atomic_inc(&node->refs);
 219 out:
 220         spin_unlock(&delayed_root->lock);
 221
 222         return node;
 223 }
 224
 225 static struct btrfs_delayed_node *btrfs_next_delayed_node(
 226                                                 struct btrfs_delayed_node *node)
 227 {
 228         struct btrfs_delayed_root *delayed_root;
 229         struct list_head *p;
 230         struct btrfs_delayed_node *next = NULL;
 231
 232         delayed_root = node->root->fs_info->delayed_root;
 233         spin_lock(&delayed_root->lock);
 234         if (!node->in_list) {   /* 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 (!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                 WARN_ON(1);
 660                 ret = btrfs_block_rsv_migrate(&root->fs_info->global_block_rsv,
 661                                               dst_rsv, num_bytes);
 662                 goto out;
 663         }
 664
 665 migrate:
 666         ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 667
 668 out:
 669         /*
 670          * Migrate only takes a reservation, it doesn't touch the size of the
 671          * block_rsv.  This is to simplify people who don't normally have things
 672          * migrated from their block rsv.  If they go to release their
 673          * reservation, that will decrease the size as well, so if migrate
 674          * reduced size we'd end up with a negative size.  But for the
 675          * delalloc_meta_reserved stuff we will only know to drop 1 reservation,
 676          * but we could in fact do this reserve/migrate dance several times
 677          * between the time we did the original reservation and we'd clean it
 678          * up.  So to take care of this, release the space for the meta
 679          * reservation here.  I think it may be time for a documentation page on
 680          * how block rsvs. work.
 681          */
 682         if (!ret) {
 683                 trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
 684                                               btrfs_ino(inode), num_bytes, 1);
 685                 node->bytes_reserved = num_bytes;
 686         }
 687
 688         if (release) {
 689                 trace_btrfs_space_reservation(root->fs_info, "delalloc",
 690                                               btrfs_ino(inode), num_bytes, 0);
 691                 btrfs_block_rsv_release(root, src_rsv, num_bytes);
 692         }
 693
 694         return ret;
 695 }
 696
 697 static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
 698                                                 struct btrfs_delayed_node *node)
 699 {
 700         struct btrfs_block_rsv *rsv;
 701
 702         if (!node->bytes_reserved)
 703                 return;
 704
 705         rsv = &root->fs_info->delayed_block_rsv;
 706         trace_btrfs_space_reservation(root->fs_info, "delayed_inode",
 707                                       node->inode_id, node->bytes_reserved, 0);
 708         btrfs_block_rsv_release(root, rsv,
 709                                 node->bytes_reserved);
 710         node->bytes_reserved = 0;
 711 }
 712
 713 /*
 714  * This helper will insert some continuous items into the same leaf according
 715  * to the free space of the leaf.
 716  */
 717 static int btrfs_batch_insert_items(struct btrfs_root *root,
 718                                     struct btrfs_path *path,
 719                                     struct btrfs_delayed_item *item)
 720 {
 721         struct btrfs_delayed_item *curr, *next;
 722         int free_space;
 723         int total_data_size = 0, total_size = 0;
 724         struct extent_buffer *leaf;
 725         char *data_ptr;
 726         struct btrfs_key *keys;
 727         u32 *data_size;
 728         struct list_head head;
 729         int slot;
 730         int nitems;
 731         int i;
 732         int ret = 0;
 733
 734         BUG_ON(!path->nodes[0]);
 735
 736         leaf = path->nodes[0];
 737         free_space = btrfs_leaf_free_space(root, leaf);
 738         INIT_LIST_HEAD(&head);
 739
 740         next = item;
 741         nitems = 0;
 742
 743         /*
 744          * count the number of the continuous items that we can insert in batch
 745          */
 746         while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 747                free_space) {
 748                 total_data_size += next->data_len;
 749                 total_size += next->data_len + sizeof(struct btrfs_item);
 750                 list_add_tail(&next->tree_list, &head);
 751                 nitems++;
 752
 753                 curr = next;
 754                 next = __btrfs_next_delayed_item(curr);
 755                 if (!next)
 756                         break;
 757
 758                 if (!btrfs_is_continuous_delayed_item(curr, next))
 759                         break;
 760         }
 761
 762         if (!nitems) {
 763                 ret = 0;
 764                 goto out;
 765         }
 766
 767         /*
 768          * we need allocate some memory space, but it might cause the task
 769          * to sleep, so we set all locked nodes in the path to blocking locks
 770          * first.
 771          */
 772         btrfs_set_path_blocking(path);
 773
 774         keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS);
 775         if (!keys) {
 776                 ret = -ENOMEM;
 777                 goto out;
 778         }
 779
 780         data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS);
 781         if (!data_size) {
 782                 ret = -ENOMEM;
 783                 goto error;
 784         }
 785
 786         /* get keys of all the delayed items */
 787         i = 0;
 788         list_for_each_entry(next, &head, tree_list) {
 789                 keys[i] = next->key;
 790                 data_size[i] = next->data_len;
 791                 i++;
 792         }
 793
 794         /* reset all the locked nodes in the patch to spinning locks. */
 795         btrfs_clear_path_blocking(path, NULL, 0);
 796
 797         /* insert the keys of the items */
 798         setup_items_for_insert(root, path, keys, data_size,
 799                                total_data_size, total_size, nitems);
 800
 801         /* insert the dir index items */
 802         slot = path->slots[0];
 803         list_for_each_entry_safe(curr, next, &head, tree_list) {
 804                 data_ptr = btrfs_item_ptr(leaf, slot, char);
 805                 write_extent_buffer(leaf, &curr->data,
 806                                     (unsigned long)data_ptr,
 807                                     curr->data_len);
 808                 slot++;
 809
 810                 btrfs_delayed_item_release_metadata(root, curr);
 811
 812                 list_del(&curr->tree_list);
 813                 btrfs_release_delayed_item(curr);
 814         }
 815
 816 error:
 817         kfree(data_size);
 818         kfree(keys);
 819 out:
 820         return ret;
 821 }
 822
 823 /*
 824  * This helper can just do simple insertion that needn't extend item for new
 825  * data, such as directory name index insertion, inode insertion.
 826  */
 827 static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 828                                      struct btrfs_root *root,
 829                                      struct btrfs_path *path,
 830                                      struct btrfs_delayed_item *delayed_item)
 831 {
 832         struct extent_buffer *leaf;
 833         char *ptr;
 834         int ret;
 835
 836         ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 837                                       delayed_item->data_len);
 838         if (ret < 0 && ret != -EEXIST)
 839                 return ret;
 840
 841         leaf = path->nodes[0];
 842
 843         ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 844
 845         write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 846                             delayed_item->data_len);
 847         btrfs_mark_buffer_dirty(leaf);
 848
 849         btrfs_delayed_item_release_metadata(root, delayed_item);
 850         return 0;
 851 }
 852
 853 /*
 854  * we insert an item first, then if there are some continuous items, we try
 855  * to insert those items into the same leaf.
 856  */
 857 static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 858                                       struct btrfs_path *path,
 859                                       struct btrfs_root *root,
 860                                       struct btrfs_delayed_node *node)
 861 {
 862         struct btrfs_delayed_item *curr, *prev;
 863         int ret = 0;
 864
 865 do_again:
 866         mutex_lock(&node->mutex);
 867         curr = __btrfs_first_delayed_insertion_item(node);
 868         if (!curr)
 869                 goto insert_end;
 870
 871         ret = btrfs_insert_delayed_item(trans, root, path, curr);
 872         if (ret < 0) {
 873                 btrfs_release_path(path);
 874                 goto insert_end;
 875         }
 876
 877         prev = curr;
 878         curr = __btrfs_next_delayed_item(prev);
 879         if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 880                 /* insert the continuous items into the same leaf */
 881                 path->slots[0]++;
 882                 btrfs_batch_insert_items(root, path, curr);
 883         }
 884         btrfs_release_delayed_item(prev);
 885         btrfs_mark_buffer_dirty(path->nodes[0]);
 886
 887         btrfs_release_path(path);
 888         mutex_unlock(&node->mutex);
 889         goto do_again;
 890
 891 insert_end:
 892         mutex_unlock(&node->mutex);
 893         return ret;
 894 }
 895
 896 static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 897                                     struct btrfs_root *root,
 898                                     struct btrfs_path *path,
 899                                     struct btrfs_delayed_item *item)
 900 {
 901         struct btrfs_delayed_item *curr, *next;
 902         struct extent_buffer *leaf;
 903         struct btrfs_key key;
 904         struct list_head head;
 905         int nitems, i, last_item;
 906         int ret = 0;
 907
 908         BUG_ON(!path->nodes[0]);
 909
 910         leaf = path->nodes[0];
 911
 912         i = path->slots[0];
 913         last_item = btrfs_header_nritems(leaf) - 1;
 914         if (i > last_item)
 915                 return -ENOENT; /* FIXME: Is errno suitable? */
 916
 917         next = item;
 918         INIT_LIST_HEAD(&head);
 919         btrfs_item_key_to_cpu(leaf, &key, i);
 920         nitems = 0;
 921         /*
 922          * count the number of the dir index items that we can delete in batch
 923          */
 924         while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 925                 list_add_tail(&next->tree_list, &head);
 926                 nitems++;
 927
 928                 curr = next;
 929                 next = __btrfs_next_delayed_item(curr);
 930                 if (!next)
 931                         break;
 932
 933                 if (!btrfs_is_continuous_delayed_item(curr, next))
 934                         break;
 935
 936                 i++;
 937                 if (i > last_item)
 938                         break;
 939                 btrfs_item_key_to_cpu(leaf, &key, i);
 940         }
 941
 942         if (!nitems)
 943                 return 0;
 944
 945         ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 946         if (ret)
 947                 goto out;
 948
 949         list_for_each_entry_safe(curr, next, &head, tree_list) {
 950                 btrfs_delayed_item_release_metadata(root, curr);
 951                 list_del(&curr->tree_list);
 952                 btrfs_release_delayed_item(curr);
 953         }
 954
 955 out:
 956         return ret;
 957 }
 958
 959 static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 960                                       struct btrfs_path *path,
 961                                       struct btrfs_root *root,
 962                                       struct btrfs_delayed_node *node)
 963 {
 964         struct btrfs_delayed_item *curr, *prev;
 965         int ret = 0;
 966
 967 do_again:
 968         mutex_lock(&node->mutex);
 969         curr = __btrfs_first_delayed_deletion_item(node);
 970         if (!curr)
 971                 goto delete_fail;
 972
 973         ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 974         if (ret < 0)
 975                 goto delete_fail;
 976         else if (ret > 0) {
 977                 /*
 978                  * can't find the item which the node points to, so this node
 979                  * is invalid, just drop it.
 980                  */
 981                 prev = curr;
 982                 curr = __btrfs_next_delayed_item(prev);
 983                 btrfs_release_delayed_item(prev);
 984                 ret = 0;
 985                 btrfs_release_path(path);
 986                 if (curr) {
 987                         mutex_unlock(&node->mutex);
 988                         goto do_again;
 989                 } else
 990                         goto delete_fail;
 991         }
 992
 993         btrfs_batch_delete_items(trans, root, path, curr);
 994         btrfs_release_path(path);
 995         mutex_unlock(&node->mutex);
 996         goto do_again;
 997
 998 delete_fail:
 999         btrfs_release_path(path);
1000         mutex_unlock(&node->mutex);
1001         return ret;
1002 }
1003
1004 static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
1005 {
1006         struct btrfs_delayed_root *delayed_root;
1007
1008         if (delayed_node && delayed_node->inode_dirty) {
1009                 BUG_ON(!delayed_node->root);
1010                 delayed_node->inode_dirty = 0;
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 (!node->inode_dirty) {
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                 return -ENOMEM;
1179         path->leave_spinning = 1;
1180
1181         block_rsv = trans->block_rsv;
1182         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1183
1184         ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1185
1186         btrfs_release_delayed_node(delayed_node);
1187         btrfs_free_path(path);
1188         trans->block_rsv = block_rsv;
1189
1190         return ret;
1191 }
1192
1193 int btrfs_commit_inode_delayed_inode(struct inode *inode)
1194 {
1195         struct btrfs_trans_handle *trans;
1196         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1197         struct btrfs_path *path;
1198         struct btrfs_block_rsv *block_rsv;
1199         int ret;
1200
1201         if (!delayed_node)
1202                 return 0;
1203
1204         mutex_lock(&delayed_node->mutex);
1205         if (!delayed_node->inode_dirty) {
1206                 mutex_unlock(&delayed_node->mutex);
1207                 btrfs_release_delayed_node(delayed_node);
1208                 return 0;
1209         }
1210         mutex_unlock(&delayed_node->mutex);
1211
1212         trans = btrfs_join_transaction(delayed_node->root);
1213         if (IS_ERR(trans)) {
1214                 ret = PTR_ERR(trans);
1215                 goto out;
1216         }
1217
1218         path = btrfs_alloc_path();
1219         if (!path) {
1220                 ret = -ENOMEM;
1221                 goto trans_out;
1222         }
1223         path->leave_spinning = 1;
1224
1225         block_rsv = trans->block_rsv;
1226         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1227
1228         mutex_lock(&delayed_node->mutex);
1229         if (delayed_node->inode_dirty)
1230                 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1231                                                    path, delayed_node);
1232         else
1233                 ret = 0;
1234         mutex_unlock(&delayed_node->mutex);
1235
1236         btrfs_free_path(path);
1237         trans->block_rsv = block_rsv;
1238 trans_out:
1239         btrfs_end_transaction(trans, delayed_node->root);
1240         btrfs_btree_balance_dirty(delayed_node->root);
1241 out:
1242         btrfs_release_delayed_node(delayed_node);
1243
1244         return ret;
1245 }
1246
1247 void btrfs_remove_delayed_node(struct inode *inode)
1248 {
1249         struct btrfs_delayed_node *delayed_node;
1250
1251         delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1252         if (!delayed_node)
1253                 return;
1254
1255         BTRFS_I(inode)->delayed_node = NULL;
1256         btrfs_release_delayed_node(delayed_node);
1257 }
1258
1259 struct btrfs_async_delayed_work {
1260         struct btrfs_delayed_root *delayed_root;
1261         int nr;
1262         struct btrfs_work work;
1263 };
1264
1265 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1266 {
1267         struct btrfs_async_delayed_work *async_work;
1268         struct btrfs_delayed_root *delayed_root;
1269         struct btrfs_trans_handle *trans;
1270         struct btrfs_path *path;
1271         struct btrfs_delayed_node *delayed_node = NULL;
1272         struct btrfs_root *root;
1273         struct btrfs_block_rsv *block_rsv;
1274         int total_done = 0;
1275
1276         async_work = container_of(work, struct btrfs_async_delayed_work, work);
1277         delayed_root = async_work->delayed_root;
1278
1279         path = btrfs_alloc_path();
1280         if (!path)
1281                 goto out;
1282
1283 again:
1284         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1285                 goto free_path;
1286
1287         delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1288         if (!delayed_node)
1289                 goto free_path;
1290
1291         path->leave_spinning = 1;
1292         root = delayed_node->root;
1293
1294         trans = btrfs_join_transaction(root);
1295         if (IS_ERR(trans))
1296                 goto release_path;
1297
1298         block_rsv = trans->block_rsv;
1299         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1300
1301         __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1302         /*
1303          * Maybe new delayed items have been inserted, so we need requeue
1304          * the work. Besides that, we must dequeue the empty delayed nodes
1305          * to avoid the race between delayed items balance and the worker.
1306          * The race like this:
1307          *      Task1                           Worker thread
1308          *                                      count == 0, needn't requeue
1309          *                                        also needn't insert the
1310          *                                        delayed node into prepare
1311          *                                        list again.
1312          *      add lots of delayed items
1313          *      queue the delayed node
1314          *        already in the list,
1315          *        and not in the prepare
1316          *        list, it means the delayed
1317          *        node is being dealt with
1318          *        by the worker.
1319          *      do delayed items balance
1320          *        the delayed node is being
1321          *        dealt with by the worker
1322          *        now, just wait.
1323          *                                      the worker goto idle.
1324          * Task1 will sleep until the transaction is commited.
1325          */
1326         mutex_lock(&delayed_node->mutex);
1327         btrfs_dequeue_delayed_node(root->fs_info->delayed_root, delayed_node);
1328         mutex_unlock(&delayed_node->mutex);
1329
1330         trans->block_rsv = block_rsv;
1331         btrfs_end_transaction_dmeta(trans, root);
1332         btrfs_btree_balance_dirty_nodelay(root);
1333
1334 release_path:
1335         btrfs_release_path(path);
1336         total_done++;
1337
1338         btrfs_release_prepared_delayed_node(delayed_node);
1339         if (async_work->nr == 0 || total_done < async_work->nr)
1340                 goto again;
1341
1342 free_path:
1343         btrfs_free_path(path);
1344 out:
1345         wake_up(&delayed_root->wait);
1346         kfree(async_work);
1347 }
1348
1349
1350 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1351                                      struct btrfs_root *root, int nr)
1352 {
1353         struct btrfs_async_delayed_work *async_work;
1354
1355         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1356                 return 0;
1357
1358         async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1359         if (!async_work)
1360                 return -ENOMEM;
1361
1362         async_work->delayed_root = delayed_root;
1363         async_work->work.func = btrfs_async_run_delayed_root;
1364         async_work->work.flags = 0;
1365         async_work->nr = nr;
1366
1367         btrfs_queue_worker(&root->fs_info->delayed_workers, &async_work->work);
1368         return 0;
1369 }
1370
1371 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1372 {
1373         struct btrfs_delayed_root *delayed_root;
1374         delayed_root = btrfs_get_delayed_root(root);
1375         WARN_ON(btrfs_first_delayed_node(delayed_root));
1376 }
1377
1378 static int refs_newer(struct btrfs_delayed_root *delayed_root,
1379                       int seq, int count)
1380 {
1381         int val = atomic_read(&delayed_root->items_seq);
1382
1383         if (val < seq || val >= seq + count)
1384                 return 1;
1385         return 0;
1386 }
1387
1388 void btrfs_balance_delayed_items(struct btrfs_root *root)
1389 {
1390         struct btrfs_delayed_root *delayed_root;
1391         int seq;
1392
1393         delayed_root = btrfs_get_delayed_root(root);
1394
1395         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1396                 return;
1397
1398         seq = atomic_read(&delayed_root->items_seq);
1399
1400         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1401                 int ret;
1402                 DEFINE_WAIT(__wait);
1403
1404                 ret = btrfs_wq_run_delayed_node(delayed_root, root, 0);
1405                 if (ret)
1406                         return;
1407
1408                 while (1) {
1409                         prepare_to_wait(&delayed_root->wait, &__wait,
1410                                         TASK_INTERRUPTIBLE);
1411
1412                         if (refs_newer(delayed_root, seq,
1413                                        BTRFS_DELAYED_BATCH) ||
1414                             atomic_read(&delayed_root->items) <
1415                             BTRFS_DELAYED_BACKGROUND) {
1416                                 break;
1417                         }
1418                         if (!signal_pending(current))
1419                                 schedule();
1420                         else
1421                                 break;
1422                 }
1423                 finish_wait(&delayed_root->wait, &__wait);
1424         }
1425
1426         btrfs_wq_run_delayed_node(delayed_root, root, BTRFS_DELAYED_BATCH);
1427 }
1428
1429 /* Will return 0 or -ENOMEM */
1430 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1431                                    struct btrfs_root *root, const char *name,
1432                                    int name_len, struct inode *dir,
1433                                    struct btrfs_disk_key *disk_key, u8 type,
1434                                    u64 index)
1435 {
1436         struct btrfs_delayed_node *delayed_node;
1437         struct btrfs_delayed_item *delayed_item;
1438         struct btrfs_dir_item *dir_item;
1439         int ret;
1440
1441         delayed_node = btrfs_get_or_create_delayed_node(dir);
1442         if (IS_ERR(delayed_node))
1443                 return PTR_ERR(delayed_node);
1444
1445         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1446         if (!delayed_item) {
1447                 ret = -ENOMEM;
1448                 goto release_node;
1449         }
1450
1451         delayed_item->key.objectid = btrfs_ino(dir);
1452         btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1453         delayed_item->key.offset = index;
1454
1455         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1456         dir_item->location = *disk_key;
1457         btrfs_set_stack_dir_transid(dir_item, trans->transid);
1458         btrfs_set_stack_dir_data_len(dir_item, 0);
1459         btrfs_set_stack_dir_name_len(dir_item, name_len);
1460         btrfs_set_stack_dir_type(dir_item, type);
1461         memcpy((char *)(dir_item + 1), name, name_len);
1462
1463         ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1464         /*
1465          * we have reserved enough space when we start a new transaction,
1466          * so reserving metadata failure is impossible
1467          */
1468         BUG_ON(ret);
1469
1470
1471         mutex_lock(&delayed_node->mutex);
1472         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1473         if (unlikely(ret)) {
1474                 printk(KERN_ERR "err add delayed dir index item(name: %.*s) "
1475                                 "into the insertion tree of the delayed node"
1476                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1477                                 name_len, name, delayed_node->root->objectid,
1478                                 delayed_node->inode_id, ret);
1479                 BUG();
1480         }
1481         mutex_unlock(&delayed_node->mutex);
1482
1483 release_node:
1484         btrfs_release_delayed_node(delayed_node);
1485         return ret;
1486 }
1487
1488 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1489                                                struct btrfs_delayed_node *node,
1490                                                struct btrfs_key *key)
1491 {
1492         struct btrfs_delayed_item *item;
1493
1494         mutex_lock(&node->mutex);
1495         item = __btrfs_lookup_delayed_insertion_item(node, key);
1496         if (!item) {
1497                 mutex_unlock(&node->mutex);
1498                 return 1;
1499         }
1500
1501         btrfs_delayed_item_release_metadata(root, item);
1502         btrfs_release_delayed_item(item);
1503         mutex_unlock(&node->mutex);
1504         return 0;
1505 }
1506
1507 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1508                                    struct btrfs_root *root, struct inode *dir,
1509                                    u64 index)
1510 {
1511         struct btrfs_delayed_node *node;
1512         struct btrfs_delayed_item *item;
1513         struct btrfs_key item_key;
1514         int ret;
1515
1516         node = btrfs_get_or_create_delayed_node(dir);
1517         if (IS_ERR(node))
1518                 return PTR_ERR(node);
1519
1520         item_key.objectid = btrfs_ino(dir);
1521         btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1522         item_key.offset = index;
1523
1524         ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1525         if (!ret)
1526                 goto end;
1527
1528         item = btrfs_alloc_delayed_item(0);
1529         if (!item) {
1530                 ret = -ENOMEM;
1531                 goto end;
1532         }
1533
1534         item->key = item_key;
1535
1536         ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1537         /*
1538          * we have reserved enough space when we start a new transaction,
1539          * so reserving metadata failure is impossible.
1540          */
1541         BUG_ON(ret);
1542
1543         mutex_lock(&node->mutex);
1544         ret = __btrfs_add_delayed_deletion_item(node, item);
1545         if (unlikely(ret)) {
1546                 printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1547                                 "into the deletion tree of the delayed node"
1548                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1549                                 index, node->root->objectid, node->inode_id,
1550                                 ret);
1551                 BUG();
1552         }
1553         mutex_unlock(&node->mutex);
1554 end:
1555         btrfs_release_delayed_node(node);
1556         return ret;
1557 }
1558
1559 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1560 {
1561         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1562
1563         if (!delayed_node)
1564                 return -ENOENT;
1565
1566         /*
1567          * Since we have held i_mutex of this directory, it is impossible that
1568          * a new directory index is added into the delayed node and index_cnt
1569          * is updated now. So we needn't lock the delayed node.
1570          */
1571         if (!delayed_node->index_cnt) {
1572                 btrfs_release_delayed_node(delayed_node);
1573                 return -EINVAL;
1574         }
1575
1576         BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1577         btrfs_release_delayed_node(delayed_node);
1578         return 0;
1579 }
1580
1581 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1582                              struct list_head *del_list)
1583 {
1584         struct btrfs_delayed_node *delayed_node;
1585         struct btrfs_delayed_item *item;
1586
1587         delayed_node = btrfs_get_delayed_node(inode);
1588         if (!delayed_node)
1589                 return;
1590
1591         mutex_lock(&delayed_node->mutex);
1592         item = __btrfs_first_delayed_insertion_item(delayed_node);
1593         while (item) {
1594                 atomic_inc(&item->refs);
1595                 list_add_tail(&item->readdir_list, ins_list);
1596                 item = __btrfs_next_delayed_item(item);
1597         }
1598
1599         item = __btrfs_first_delayed_deletion_item(delayed_node);
1600         while (item) {
1601                 atomic_inc(&item->refs);
1602                 list_add_tail(&item->readdir_list, del_list);
1603                 item = __btrfs_next_delayed_item(item);
1604         }
1605         mutex_unlock(&delayed_node->mutex);
1606         /*
1607          * This delayed node is still cached in the btrfs inode, so refs
1608          * must be > 1 now, and we needn't check it is going to be freed
1609          * or not.
1610          *
1611          * Besides that, this function is used to read dir, we do not
1612          * insert/delete delayed items in this period. So we also needn't
1613          * requeue or dequeue this delayed node.
1614          */
1615         atomic_dec(&delayed_node->refs);
1616 }
1617
1618 void btrfs_put_delayed_items(struct list_head *ins_list,
1619                              struct list_head *del_list)
1620 {
1621         struct btrfs_delayed_item *curr, *next;
1622
1623         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1624                 list_del(&curr->readdir_list);
1625                 if (atomic_dec_and_test(&curr->refs))
1626                         kfree(curr);
1627         }
1628
1629         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1630                 list_del(&curr->readdir_list);
1631                 if (atomic_dec_and_test(&curr->refs))
1632                         kfree(curr);
1633         }
1634 }
1635
1636 int btrfs_should_delete_dir_index(struct list_head *del_list,
1637                                   u64 index)
1638 {
1639         struct btrfs_delayed_item *curr, *next;
1640         int ret;
1641
1642         if (list_empty(del_list))
1643                 return 0;
1644
1645         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1646                 if (curr->key.offset > index)
1647                         break;
1648
1649                 list_del(&curr->readdir_list);
1650                 ret = (curr->key.offset == index);
1651
1652                 if (atomic_dec_and_test(&curr->refs))
1653                         kfree(curr);
1654
1655                 if (ret)
1656                         return 1;
1657                 else
1658                         continue;
1659         }
1660         return 0;
1661 }
1662
1663 /*
1664  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1665  *
1666  */
1667 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1668                                     struct list_head *ins_list)
1669 {
1670         struct btrfs_dir_item *di;
1671         struct btrfs_delayed_item *curr, *next;
1672         struct btrfs_key location;
1673         char *name;
1674         int name_len;
1675         int over = 0;
1676         unsigned char d_type;
1677
1678         if (list_empty(ins_list))
1679                 return 0;
1680
1681         /*
1682          * Changing the data of the delayed item is impossible. So
1683          * we needn't lock them. And we have held i_mutex of the
1684          * directory, nobody can delete any directory indexes now.
1685          */
1686         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1687                 list_del(&curr->readdir_list);
1688
1689                 if (curr->key.offset < ctx->pos) {
1690                         if (atomic_dec_and_test(&curr->refs))
1691                                 kfree(curr);
1692                         continue;
1693                 }
1694
1695                 ctx->pos = curr->key.offset;
1696
1697                 di = (struct btrfs_dir_item *)curr->data;
1698                 name = (char *)(di + 1);
1699                 name_len = btrfs_stack_dir_name_len(di);
1700
1701                 d_type = btrfs_filetype_table[di->type];
1702                 btrfs_disk_key_to_cpu(&location, &di->location);
1703
1704                 over = !dir_emit(ctx, name, name_len,
1705                                location.objectid, d_type);
1706
1707                 if (atomic_dec_and_test(&curr->refs))
1708                         kfree(curr);
1709
1710                 if (over)
1711                         return 1;
1712         }
1713         return 0;
1714 }
1715
1716 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1717                                   struct btrfs_inode_item *inode_item,
1718                                   struct inode *inode)
1719 {
1720         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1721         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1722         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1723         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1724         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1725         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1726         btrfs_set_stack_inode_generation(inode_item,
1727                                          BTRFS_I(inode)->generation);
1728         btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1729         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1730         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1731         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1732         btrfs_set_stack_inode_block_group(inode_item, 0);
1733
1734         btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1735                                      inode->i_atime.tv_sec);
1736         btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1737                                       inode->i_atime.tv_nsec);
1738
1739         btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1740                                      inode->i_mtime.tv_sec);
1741         btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1742                                       inode->i_mtime.tv_nsec);
1743
1744         btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1745                                      inode->i_ctime.tv_sec);
1746         btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1747                                       inode->i_ctime.tv_nsec);
1748 }
1749
1750 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1751 {
1752         struct btrfs_delayed_node *delayed_node;
1753         struct btrfs_inode_item *inode_item;
1754         struct btrfs_timespec *tspec;
1755
1756         delayed_node = btrfs_get_delayed_node(inode);
1757         if (!delayed_node)
1758                 return -ENOENT;
1759
1760         mutex_lock(&delayed_node->mutex);
1761         if (!delayed_node->inode_dirty) {
1762                 mutex_unlock(&delayed_node->mutex);
1763                 btrfs_release_delayed_node(delayed_node);
1764                 return -ENOENT;
1765         }
1766
1767         inode_item = &delayed_node->inode_item;
1768
1769         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1770         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1771         btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1772         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1773         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1774         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1775         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1776         inode->i_version = btrfs_stack_inode_sequence(inode_item);
1777         inode->i_rdev = 0;
1778         *rdev = btrfs_stack_inode_rdev(inode_item);
1779         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1780
1781         tspec = btrfs_inode_atime(inode_item);
1782         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1783         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1784
1785         tspec = btrfs_inode_mtime(inode_item);
1786         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1787         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1788
1789         tspec = btrfs_inode_ctime(inode_item);
1790         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1791         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1792
1793         inode->i_generation = BTRFS_I(inode)->generation;
1794         BTRFS_I(inode)->index_cnt = (u64)-1;
1795
1796         mutex_unlock(&delayed_node->mutex);
1797         btrfs_release_delayed_node(delayed_node);
1798         return 0;
1799 }
1800
1801 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1802                                struct btrfs_root *root, struct inode *inode)
1803 {
1804         struct btrfs_delayed_node *delayed_node;
1805         int ret = 0;
1806
1807         delayed_node = btrfs_get_or_create_delayed_node(inode);
1808         if (IS_ERR(delayed_node))
1809                 return PTR_ERR(delayed_node);
1810
1811         mutex_lock(&delayed_node->mutex);
1812         if (delayed_node->inode_dirty) {
1813                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1814                 goto release_node;
1815         }
1816
1817         ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1818                                                    delayed_node);
1819         if (ret)
1820                 goto release_node;
1821
1822         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1823         delayed_node->inode_dirty = 1;
1824         delayed_node->count++;
1825         atomic_inc(&root->fs_info->delayed_root->items);
1826 release_node:
1827         mutex_unlock(&delayed_node->mutex);
1828         btrfs_release_delayed_node(delayed_node);
1829         return ret;
1830 }
1831
1832 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1833 {
1834         struct btrfs_root *root = delayed_node->root;
1835         struct btrfs_delayed_item *curr_item, *prev_item;
1836
1837         mutex_lock(&delayed_node->mutex);
1838         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1839         while (curr_item) {
1840                 btrfs_delayed_item_release_metadata(root, curr_item);
1841                 prev_item = curr_item;
1842                 curr_item = __btrfs_next_delayed_item(prev_item);
1843                 btrfs_release_delayed_item(prev_item);
1844         }
1845
1846         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1847         while (curr_item) {
1848                 btrfs_delayed_item_release_metadata(root, curr_item);
1849                 prev_item = curr_item;
1850                 curr_item = __btrfs_next_delayed_item(prev_item);
1851                 btrfs_release_delayed_item(prev_item);
1852         }
1853
1854         if (delayed_node->inode_dirty) {
1855                 btrfs_delayed_inode_release_metadata(root, delayed_node);
1856                 btrfs_release_delayed_inode(delayed_node);
1857         }
1858         mutex_unlock(&delayed_node->mutex);
1859 }
1860
1861 void btrfs_kill_delayed_inode_items(struct inode *inode)
1862 {
1863         struct btrfs_delayed_node *delayed_node;
1864
1865         delayed_node = btrfs_get_delayed_node(inode);
1866         if (!delayed_node)
1867                 return;
1868
1869         __btrfs_kill_delayed_node(delayed_node);
1870         btrfs_release_delayed_node(delayed_node);
1871 }
1872
1873 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1874 {
1875         u64 inode_id = 0;
1876         struct btrfs_delayed_node *delayed_nodes[8];
1877         int i, n;
1878
1879         while (1) {
1880                 spin_lock(&root->inode_lock);
1881                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1882                                            (void **)delayed_nodes, inode_id,
1883                                            ARRAY_SIZE(delayed_nodes));
1884                 if (!n) {
1885                         spin_unlock(&root->inode_lock);
1886                         break;
1887                 }
1888
1889                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1890
1891                 for (i = 0; i < n; i++)
1892                         atomic_inc(&delayed_nodes[i]->refs);
1893                 spin_unlock(&root->inode_lock);
1894
1895                 for (i = 0; i < n; i++) {
1896                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1897                         btrfs_release_delayed_node(delayed_nodes[i]);
1898                 }
1899         }
1900 }
1901
1902 void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1903 {
1904         struct btrfs_delayed_root *delayed_root;
1905         struct btrfs_delayed_node *curr_node, *prev_node;
1906
1907         delayed_root = btrfs_get_delayed_root(root);
1908
1909         curr_node = btrfs_first_delayed_node(delayed_root);
1910         while (curr_node) {
1911                 __btrfs_kill_delayed_node(curr_node);
1912
1913                 prev_node = curr_node;
1914                 curr_node = btrfs_next_delayed_node(curr_node);
1915                 btrfs_release_delayed_node(prev_node);
1916         }
1917 }
1918