Pileus Git - ~andy/linux/blob - 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                 /* can be accessed and cached in the inode */
 112                 atomic_add(2, &node->refs);
 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         /* cached in the btrfs inode and can be accessed */
 142         atomic_add(2, &node->refs);
 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 (!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 && delayed_node->inode_dirty) {
1008                 BUG_ON(!delayed_node->root);
1009                 delayed_node->inode_dirty = 0;
1010                 delayed_node->count--;
1011
1012                 delayed_root = delayed_node->root->fs_info->delayed_root;
1013                 finish_one_item(delayed_root);
1014         }
1015 }
1016
1017 static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1018                                         struct btrfs_root *root,
1019                                         struct btrfs_path *path,
1020                                         struct btrfs_delayed_node *node)
1021 {
1022         struct btrfs_key key;
1023         struct btrfs_inode_item *inode_item;
1024         struct extent_buffer *leaf;
1025         int ret;
1026
1027         key.objectid = node->inode_id;
1028         btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
1029         key.offset = 0;
1030
1031         ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1032         if (ret > 0) {
1033                 btrfs_release_path(path);
1034                 return -ENOENT;
1035         } else if (ret < 0) {
1036                 return ret;
1037         }
1038
1039         btrfs_unlock_up_safe(path, 1);
1040         leaf = path->nodes[0];
1041         inode_item = btrfs_item_ptr(leaf, path->slots[0],
1042                                     struct btrfs_inode_item);
1043         write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1044                             sizeof(struct btrfs_inode_item));
1045         btrfs_mark_buffer_dirty(leaf);
1046         btrfs_release_path(path);
1047
1048         btrfs_delayed_inode_release_metadata(root, node);
1049         btrfs_release_delayed_inode(node);
1050
1051         return 0;
1052 }
1053
1054 static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1055                                              struct btrfs_root *root,
1056                                              struct btrfs_path *path,
1057                                              struct btrfs_delayed_node *node)
1058 {
1059         int ret;
1060
1061         mutex_lock(&node->mutex);
1062         if (!node->inode_dirty) {
1063                 mutex_unlock(&node->mutex);
1064                 return 0;
1065         }
1066
1067         ret = __btrfs_update_delayed_inode(trans, root, path, node);
1068         mutex_unlock(&node->mutex);
1069         return ret;
1070 }
1071
1072 static inline int
1073 __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1074                                    struct btrfs_path *path,
1075                                    struct btrfs_delayed_node *node)
1076 {
1077         int ret;
1078
1079         ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1080         if (ret)
1081                 return ret;
1082
1083         ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1084         if (ret)
1085                 return ret;
1086
1087         ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1088         return ret;
1089 }
1090
1091 /*
1092  * Called when committing the transaction.
1093  * Returns 0 on success.
1094  * Returns < 0 on error and returns with an aborted transaction with any
1095  * outstanding delayed items cleaned up.
1096  */
1097 static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1098                                      struct btrfs_root *root, int nr)
1099 {
1100         struct btrfs_delayed_root *delayed_root;
1101         struct btrfs_delayed_node *curr_node, *prev_node;
1102         struct btrfs_path *path;
1103         struct btrfs_block_rsv *block_rsv;
1104         int ret = 0;
1105         bool count = (nr > 0);
1106
1107         if (trans->aborted)
1108                 return -EIO;
1109
1110         path = btrfs_alloc_path();
1111         if (!path)
1112                 return -ENOMEM;
1113         path->leave_spinning = 1;
1114
1115         block_rsv = trans->block_rsv;
1116         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1117
1118         delayed_root = btrfs_get_delayed_root(root);
1119
1120         curr_node = btrfs_first_delayed_node(delayed_root);
1121         while (curr_node && (!count || (count && nr--))) {
1122                 ret = __btrfs_commit_inode_delayed_items(trans, path,
1123                                                          curr_node);
1124                 if (ret) {
1125                         btrfs_release_delayed_node(curr_node);
1126                         curr_node = NULL;
1127                         btrfs_abort_transaction(trans, root, ret);
1128                         break;
1129                 }
1130
1131                 prev_node = curr_node;
1132                 curr_node = btrfs_next_delayed_node(curr_node);
1133                 btrfs_release_delayed_node(prev_node);
1134         }
1135
1136         if (curr_node)
1137                 btrfs_release_delayed_node(curr_node);
1138         btrfs_free_path(path);
1139         trans->block_rsv = block_rsv;
1140
1141         return ret;
1142 }
1143
1144 int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1145                             struct btrfs_root *root)
1146 {
1147         return __btrfs_run_delayed_items(trans, root, -1);
1148 }
1149
1150 int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans,
1151                                struct btrfs_root *root, int nr)
1152 {
1153         return __btrfs_run_delayed_items(trans, root, nr);
1154 }
1155
1156 int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1157                                      struct inode *inode)
1158 {
1159         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1160         struct btrfs_path *path;
1161         struct btrfs_block_rsv *block_rsv;
1162         int ret;
1163
1164         if (!delayed_node)
1165                 return 0;
1166
1167         mutex_lock(&delayed_node->mutex);
1168         if (!delayed_node->count) {
1169                 mutex_unlock(&delayed_node->mutex);
1170                 btrfs_release_delayed_node(delayed_node);
1171                 return 0;
1172         }
1173         mutex_unlock(&delayed_node->mutex);
1174
1175         path = btrfs_alloc_path();
1176         if (!path) {
1177                 btrfs_release_delayed_node(delayed_node);
1178                 return -ENOMEM;
1179         }
1180         path->leave_spinning = 1;
1181
1182         block_rsv = trans->block_rsv;
1183         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1184
1185         ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1186
1187         btrfs_release_delayed_node(delayed_node);
1188         btrfs_free_path(path);
1189         trans->block_rsv = block_rsv;
1190
1191         return ret;
1192 }
1193
1194 int btrfs_commit_inode_delayed_inode(struct inode *inode)
1195 {
1196         struct btrfs_trans_handle *trans;
1197         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1198         struct btrfs_path *path;
1199         struct btrfs_block_rsv *block_rsv;
1200         int ret;
1201
1202         if (!delayed_node)
1203                 return 0;
1204
1205         mutex_lock(&delayed_node->mutex);
1206         if (!delayed_node->inode_dirty) {
1207                 mutex_unlock(&delayed_node->mutex);
1208                 btrfs_release_delayed_node(delayed_node);
1209                 return 0;
1210         }
1211         mutex_unlock(&delayed_node->mutex);
1212
1213         trans = btrfs_join_transaction(delayed_node->root);
1214         if (IS_ERR(trans)) {
1215                 ret = PTR_ERR(trans);
1216                 goto out;
1217         }
1218
1219         path = btrfs_alloc_path();
1220         if (!path) {
1221                 ret = -ENOMEM;
1222                 goto trans_out;
1223         }
1224         path->leave_spinning = 1;
1225
1226         block_rsv = trans->block_rsv;
1227         trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1228
1229         mutex_lock(&delayed_node->mutex);
1230         if (delayed_node->inode_dirty)
1231                 ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1232                                                    path, delayed_node);
1233         else
1234                 ret = 0;
1235         mutex_unlock(&delayed_node->mutex);
1236
1237         btrfs_free_path(path);
1238         trans->block_rsv = block_rsv;
1239 trans_out:
1240         btrfs_end_transaction(trans, delayed_node->root);
1241         btrfs_btree_balance_dirty(delayed_node->root);
1242 out:
1243         btrfs_release_delayed_node(delayed_node);
1244
1245         return ret;
1246 }
1247
1248 void btrfs_remove_delayed_node(struct inode *inode)
1249 {
1250         struct btrfs_delayed_node *delayed_node;
1251
1252         delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1253         if (!delayed_node)
1254                 return;
1255
1256         BTRFS_I(inode)->delayed_node = NULL;
1257         btrfs_release_delayed_node(delayed_node);
1258 }
1259
1260 struct btrfs_async_delayed_work {
1261         struct btrfs_delayed_root *delayed_root;
1262         int nr;
1263         struct btrfs_work work;
1264 };
1265
1266 static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1267 {
1268         struct btrfs_async_delayed_work *async_work;
1269         struct btrfs_delayed_root *delayed_root;
1270         struct btrfs_trans_handle *trans;
1271         struct btrfs_path *path;
1272         struct btrfs_delayed_node *delayed_node = NULL;
1273         struct btrfs_root *root;
1274         struct btrfs_block_rsv *block_rsv;
1275         int total_done = 0;
1276
1277         async_work = container_of(work, struct btrfs_async_delayed_work, work);
1278         delayed_root = async_work->delayed_root;
1279
1280         path = btrfs_alloc_path();
1281         if (!path)
1282                 goto out;
1283
1284 again:
1285         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND / 2)
1286                 goto free_path;
1287
1288         delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1289         if (!delayed_node)
1290                 goto free_path;
1291
1292         path->leave_spinning = 1;
1293         root = delayed_node->root;
1294
1295         trans = btrfs_join_transaction(root);
1296         if (IS_ERR(trans))
1297                 goto release_path;
1298
1299         block_rsv = trans->block_rsv;
1300         trans->block_rsv = &root->fs_info->delayed_block_rsv;
1301
1302         __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1303         /*
1304          * Maybe new delayed items have been inserted, so we need requeue
1305          * the work. Besides that, we must dequeue the empty delayed nodes
1306          * to avoid the race between delayed items balance and the worker.
1307          * The race like this:
1308          *      Task1                           Worker thread
1309          *                                      count == 0, needn't requeue
1310          *                                        also needn't insert the
1311          *                                        delayed node into prepare
1312          *                                        list again.
1313          *      add lots of delayed items
1314          *      queue the delayed node
1315          *        already in the list,
1316          *        and not in the prepare
1317          *        list, it means the delayed
1318          *        node is being dealt with
1319          *        by the worker.
1320          *      do delayed items balance
1321          *        the delayed node is being
1322          *        dealt with by the worker
1323          *        now, just wait.
1324          *                                      the worker goto idle.
1325          * Task1 will sleep until the transaction is commited.
1326          */
1327         mutex_lock(&delayed_node->mutex);
1328         btrfs_dequeue_delayed_node(root->fs_info->delayed_root, delayed_node);
1329         mutex_unlock(&delayed_node->mutex);
1330
1331         trans->block_rsv = block_rsv;
1332         btrfs_end_transaction_dmeta(trans, root);
1333         btrfs_btree_balance_dirty_nodelay(root);
1334
1335 release_path:
1336         btrfs_release_path(path);
1337         total_done++;
1338
1339         btrfs_release_prepared_delayed_node(delayed_node);
1340         if (async_work->nr == 0 || total_done < async_work->nr)
1341                 goto again;
1342
1343 free_path:
1344         btrfs_free_path(path);
1345 out:
1346         wake_up(&delayed_root->wait);
1347         kfree(async_work);
1348 }
1349
1350
1351 static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1352                                      struct btrfs_root *root, int nr)
1353 {
1354         struct btrfs_async_delayed_work *async_work;
1355
1356         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1357                 return 0;
1358
1359         async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1360         if (!async_work)
1361                 return -ENOMEM;
1362
1363         async_work->delayed_root = delayed_root;
1364         async_work->work.func = btrfs_async_run_delayed_root;
1365         async_work->work.flags = 0;
1366         async_work->nr = nr;
1367
1368         btrfs_queue_worker(&root->fs_info->delayed_workers, &async_work->work);
1369         return 0;
1370 }
1371
1372 void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1373 {
1374         struct btrfs_delayed_root *delayed_root;
1375         delayed_root = btrfs_get_delayed_root(root);
1376         WARN_ON(btrfs_first_delayed_node(delayed_root));
1377 }
1378
1379 static int refs_newer(struct btrfs_delayed_root *delayed_root,
1380                       int seq, int count)
1381 {
1382         int val = atomic_read(&delayed_root->items_seq);
1383
1384         if (val < seq || val >= seq + count)
1385                 return 1;
1386         return 0;
1387 }
1388
1389 void btrfs_balance_delayed_items(struct btrfs_root *root)
1390 {
1391         struct btrfs_delayed_root *delayed_root;
1392         int seq;
1393
1394         delayed_root = btrfs_get_delayed_root(root);
1395
1396         if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1397                 return;
1398
1399         seq = atomic_read(&delayed_root->items_seq);
1400
1401         if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1402                 int ret;
1403                 DEFINE_WAIT(__wait);
1404
1405                 ret = btrfs_wq_run_delayed_node(delayed_root, root, 0);
1406                 if (ret)
1407                         return;
1408
1409                 while (1) {
1410                         prepare_to_wait(&delayed_root->wait, &__wait,
1411                                         TASK_INTERRUPTIBLE);
1412
1413                         if (refs_newer(delayed_root, seq,
1414                                        BTRFS_DELAYED_BATCH) ||
1415                             atomic_read(&delayed_root->items) <
1416                             BTRFS_DELAYED_BACKGROUND) {
1417                                 break;
1418                         }
1419                         if (!signal_pending(current))
1420                                 schedule();
1421                         else
1422                                 break;
1423                 }
1424                 finish_wait(&delayed_root->wait, &__wait);
1425         }
1426
1427         btrfs_wq_run_delayed_node(delayed_root, root, BTRFS_DELAYED_BATCH);
1428 }
1429
1430 /* Will return 0 or -ENOMEM */
1431 int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1432                                    struct btrfs_root *root, const char *name,
1433                                    int name_len, struct inode *dir,
1434                                    struct btrfs_disk_key *disk_key, u8 type,
1435                                    u64 index)
1436 {
1437         struct btrfs_delayed_node *delayed_node;
1438         struct btrfs_delayed_item *delayed_item;
1439         struct btrfs_dir_item *dir_item;
1440         int ret;
1441
1442         delayed_node = btrfs_get_or_create_delayed_node(dir);
1443         if (IS_ERR(delayed_node))
1444                 return PTR_ERR(delayed_node);
1445
1446         delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1447         if (!delayed_item) {
1448                 ret = -ENOMEM;
1449                 goto release_node;
1450         }
1451
1452         delayed_item->key.objectid = btrfs_ino(dir);
1453         btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1454         delayed_item->key.offset = index;
1455
1456         dir_item = (struct btrfs_dir_item *)delayed_item->data;
1457         dir_item->location = *disk_key;
1458         btrfs_set_stack_dir_transid(dir_item, trans->transid);
1459         btrfs_set_stack_dir_data_len(dir_item, 0);
1460         btrfs_set_stack_dir_name_len(dir_item, name_len);
1461         btrfs_set_stack_dir_type(dir_item, type);
1462         memcpy((char *)(dir_item + 1), name, name_len);
1463
1464         ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1465         /*
1466          * we have reserved enough space when we start a new transaction,
1467          * so reserving metadata failure is impossible
1468          */
1469         BUG_ON(ret);
1470
1471
1472         mutex_lock(&delayed_node->mutex);
1473         ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1474         if (unlikely(ret)) {
1475                 printk(KERN_ERR "err add delayed dir index item(name: %.*s) "
1476                                 "into the insertion tree of the delayed node"
1477                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1478                                 name_len, name, delayed_node->root->objectid,
1479                                 delayed_node->inode_id, ret);
1480                 BUG();
1481         }
1482         mutex_unlock(&delayed_node->mutex);
1483
1484 release_node:
1485         btrfs_release_delayed_node(delayed_node);
1486         return ret;
1487 }
1488
1489 static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1490                                                struct btrfs_delayed_node *node,
1491                                                struct btrfs_key *key)
1492 {
1493         struct btrfs_delayed_item *item;
1494
1495         mutex_lock(&node->mutex);
1496         item = __btrfs_lookup_delayed_insertion_item(node, key);
1497         if (!item) {
1498                 mutex_unlock(&node->mutex);
1499                 return 1;
1500         }
1501
1502         btrfs_delayed_item_release_metadata(root, item);
1503         btrfs_release_delayed_item(item);
1504         mutex_unlock(&node->mutex);
1505         return 0;
1506 }
1507
1508 int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1509                                    struct btrfs_root *root, struct inode *dir,
1510                                    u64 index)
1511 {
1512         struct btrfs_delayed_node *node;
1513         struct btrfs_delayed_item *item;
1514         struct btrfs_key item_key;
1515         int ret;
1516
1517         node = btrfs_get_or_create_delayed_node(dir);
1518         if (IS_ERR(node))
1519                 return PTR_ERR(node);
1520
1521         item_key.objectid = btrfs_ino(dir);
1522         btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1523         item_key.offset = index;
1524
1525         ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
1526         if (!ret)
1527                 goto end;
1528
1529         item = btrfs_alloc_delayed_item(0);
1530         if (!item) {
1531                 ret = -ENOMEM;
1532                 goto end;
1533         }
1534
1535         item->key = item_key;
1536
1537         ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1538         /*
1539          * we have reserved enough space when we start a new transaction,
1540          * so reserving metadata failure is impossible.
1541          */
1542         BUG_ON(ret);
1543
1544         mutex_lock(&node->mutex);
1545         ret = __btrfs_add_delayed_deletion_item(node, item);
1546         if (unlikely(ret)) {
1547                 printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1548                                 "into the deletion tree of the delayed node"
1549                                 "(root id: %llu, inode id: %llu, errno: %d)\n",
1550                                 index, node->root->objectid, node->inode_id,
1551                                 ret);
1552                 BUG();
1553         }
1554         mutex_unlock(&node->mutex);
1555 end:
1556         btrfs_release_delayed_node(node);
1557         return ret;
1558 }
1559
1560 int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1561 {
1562         struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1563
1564         if (!delayed_node)
1565                 return -ENOENT;
1566
1567         /*
1568          * Since we have held i_mutex of this directory, it is impossible that
1569          * a new directory index is added into the delayed node and index_cnt
1570          * is updated now. So we needn't lock the delayed node.
1571          */
1572         if (!delayed_node->index_cnt) {
1573                 btrfs_release_delayed_node(delayed_node);
1574                 return -EINVAL;
1575         }
1576
1577         BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1578         btrfs_release_delayed_node(delayed_node);
1579         return 0;
1580 }
1581
1582 void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1583                              struct list_head *del_list)
1584 {
1585         struct btrfs_delayed_node *delayed_node;
1586         struct btrfs_delayed_item *item;
1587
1588         delayed_node = btrfs_get_delayed_node(inode);
1589         if (!delayed_node)
1590                 return;
1591
1592         mutex_lock(&delayed_node->mutex);
1593         item = __btrfs_first_delayed_insertion_item(delayed_node);
1594         while (item) {
1595                 atomic_inc(&item->refs);
1596                 list_add_tail(&item->readdir_list, ins_list);
1597                 item = __btrfs_next_delayed_item(item);
1598         }
1599
1600         item = __btrfs_first_delayed_deletion_item(delayed_node);
1601         while (item) {
1602                 atomic_inc(&item->refs);
1603                 list_add_tail(&item->readdir_list, del_list);
1604                 item = __btrfs_next_delayed_item(item);
1605         }
1606         mutex_unlock(&delayed_node->mutex);
1607         /*
1608          * This delayed node is still cached in the btrfs inode, so refs
1609          * must be > 1 now, and we needn't check it is going to be freed
1610          * or not.
1611          *
1612          * Besides that, this function is used to read dir, we do not
1613          * insert/delete delayed items in this period. So we also needn't
1614          * requeue or dequeue this delayed node.
1615          */
1616         atomic_dec(&delayed_node->refs);
1617 }
1618
1619 void btrfs_put_delayed_items(struct list_head *ins_list,
1620                              struct list_head *del_list)
1621 {
1622         struct btrfs_delayed_item *curr, *next;
1623
1624         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1625                 list_del(&curr->readdir_list);
1626                 if (atomic_dec_and_test(&curr->refs))
1627                         kfree(curr);
1628         }
1629
1630         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1631                 list_del(&curr->readdir_list);
1632                 if (atomic_dec_and_test(&curr->refs))
1633                         kfree(curr);
1634         }
1635 }
1636
1637 int btrfs_should_delete_dir_index(struct list_head *del_list,
1638                                   u64 index)
1639 {
1640         struct btrfs_delayed_item *curr, *next;
1641         int ret;
1642
1643         if (list_empty(del_list))
1644                 return 0;
1645
1646         list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1647                 if (curr->key.offset > index)
1648                         break;
1649
1650                 list_del(&curr->readdir_list);
1651                 ret = (curr->key.offset == index);
1652
1653                 if (atomic_dec_and_test(&curr->refs))
1654                         kfree(curr);
1655
1656                 if (ret)
1657                         return 1;
1658                 else
1659                         continue;
1660         }
1661         return 0;
1662 }
1663
1664 /*
1665  * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1666  *
1667  */
1668 int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
1669                                     struct list_head *ins_list)
1670 {
1671         struct btrfs_dir_item *di;
1672         struct btrfs_delayed_item *curr, *next;
1673         struct btrfs_key location;
1674         char *name;
1675         int name_len;
1676         int over = 0;
1677         unsigned char d_type;
1678
1679         if (list_empty(ins_list))
1680                 return 0;
1681
1682         /*
1683          * Changing the data of the delayed item is impossible. So
1684          * we needn't lock them. And we have held i_mutex of the
1685          * directory, nobody can delete any directory indexes now.
1686          */
1687         list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1688                 list_del(&curr->readdir_list);
1689
1690                 if (curr->key.offset < ctx->pos) {
1691                         if (atomic_dec_and_test(&curr->refs))
1692                                 kfree(curr);
1693                         continue;
1694                 }
1695
1696                 ctx->pos = curr->key.offset;
1697
1698                 di = (struct btrfs_dir_item *)curr->data;
1699                 name = (char *)(di + 1);
1700                 name_len = btrfs_stack_dir_name_len(di);
1701
1702                 d_type = btrfs_filetype_table[di->type];
1703                 btrfs_disk_key_to_cpu(&location, &di->location);
1704
1705                 over = !dir_emit(ctx, name, name_len,
1706                                location.objectid, d_type);
1707
1708                 if (atomic_dec_and_test(&curr->refs))
1709                         kfree(curr);
1710
1711                 if (over)
1712                         return 1;
1713         }
1714         return 0;
1715 }
1716
1717 static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1718                                   struct btrfs_inode_item *inode_item,
1719                                   struct inode *inode)
1720 {
1721         btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1722         btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1723         btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1724         btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1725         btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1726         btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1727         btrfs_set_stack_inode_generation(inode_item,
1728                                          BTRFS_I(inode)->generation);
1729         btrfs_set_stack_inode_sequence(inode_item, inode->i_version);
1730         btrfs_set_stack_inode_transid(inode_item, trans->transid);
1731         btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1732         btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1733         btrfs_set_stack_inode_block_group(inode_item, 0);
1734
1735         btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1736                                      inode->i_atime.tv_sec);
1737         btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1738                                       inode->i_atime.tv_nsec);
1739
1740         btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1741                                      inode->i_mtime.tv_sec);
1742         btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1743                                       inode->i_mtime.tv_nsec);
1744
1745         btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1746                                      inode->i_ctime.tv_sec);
1747         btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1748                                       inode->i_ctime.tv_nsec);
1749 }
1750
1751 int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1752 {
1753         struct btrfs_delayed_node *delayed_node;
1754         struct btrfs_inode_item *inode_item;
1755         struct btrfs_timespec *tspec;
1756
1757         delayed_node = btrfs_get_delayed_node(inode);
1758         if (!delayed_node)
1759                 return -ENOENT;
1760
1761         mutex_lock(&delayed_node->mutex);
1762         if (!delayed_node->inode_dirty) {
1763                 mutex_unlock(&delayed_node->mutex);
1764                 btrfs_release_delayed_node(delayed_node);
1765                 return -ENOENT;
1766         }
1767
1768         inode_item = &delayed_node->inode_item;
1769
1770         i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1771         i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1772         btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1773         inode->i_mode = btrfs_stack_inode_mode(inode_item);
1774         set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1775         inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1776         BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1777         inode->i_version = btrfs_stack_inode_sequence(inode_item);
1778         inode->i_rdev = 0;
1779         *rdev = btrfs_stack_inode_rdev(inode_item);
1780         BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1781
1782         tspec = btrfs_inode_atime(inode_item);
1783         inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1784         inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1785
1786         tspec = btrfs_inode_mtime(inode_item);
1787         inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1788         inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1789
1790         tspec = btrfs_inode_ctime(inode_item);
1791         inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1792         inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1793
1794         inode->i_generation = BTRFS_I(inode)->generation;
1795         BTRFS_I(inode)->index_cnt = (u64)-1;
1796
1797         mutex_unlock(&delayed_node->mutex);
1798         btrfs_release_delayed_node(delayed_node);
1799         return 0;
1800 }
1801
1802 int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1803                                struct btrfs_root *root, struct inode *inode)
1804 {
1805         struct btrfs_delayed_node *delayed_node;
1806         int ret = 0;
1807
1808         delayed_node = btrfs_get_or_create_delayed_node(inode);
1809         if (IS_ERR(delayed_node))
1810                 return PTR_ERR(delayed_node);
1811
1812         mutex_lock(&delayed_node->mutex);
1813         if (delayed_node->inode_dirty) {
1814                 fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1815                 goto release_node;
1816         }
1817
1818         ret = btrfs_delayed_inode_reserve_metadata(trans, root, inode,
1819                                                    delayed_node);
1820         if (ret)
1821                 goto release_node;
1822
1823         fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1824         delayed_node->inode_dirty = 1;
1825         delayed_node->count++;
1826         atomic_inc(&root->fs_info->delayed_root->items);
1827 release_node:
1828         mutex_unlock(&delayed_node->mutex);
1829         btrfs_release_delayed_node(delayed_node);
1830         return ret;
1831 }
1832
1833 static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1834 {
1835         struct btrfs_root *root = delayed_node->root;
1836         struct btrfs_delayed_item *curr_item, *prev_item;
1837
1838         mutex_lock(&delayed_node->mutex);
1839         curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1840         while (curr_item) {
1841                 btrfs_delayed_item_release_metadata(root, curr_item);
1842                 prev_item = curr_item;
1843                 curr_item = __btrfs_next_delayed_item(prev_item);
1844                 btrfs_release_delayed_item(prev_item);
1845         }
1846
1847         curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1848         while (curr_item) {
1849                 btrfs_delayed_item_release_metadata(root, curr_item);
1850                 prev_item = curr_item;
1851                 curr_item = __btrfs_next_delayed_item(prev_item);
1852                 btrfs_release_delayed_item(prev_item);
1853         }
1854
1855         if (delayed_node->inode_dirty) {
1856                 btrfs_delayed_inode_release_metadata(root, delayed_node);
1857                 btrfs_release_delayed_inode(delayed_node);
1858         }
1859         mutex_unlock(&delayed_node->mutex);
1860 }
1861
1862 void btrfs_kill_delayed_inode_items(struct inode *inode)
1863 {
1864         struct btrfs_delayed_node *delayed_node;
1865
1866         delayed_node = btrfs_get_delayed_node(inode);
1867         if (!delayed_node)
1868                 return;
1869
1870         __btrfs_kill_delayed_node(delayed_node);
1871         btrfs_release_delayed_node(delayed_node);
1872 }
1873
1874 void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1875 {
1876         u64 inode_id = 0;
1877         struct btrfs_delayed_node *delayed_nodes[8];
1878         int i, n;
1879
1880         while (1) {
1881                 spin_lock(&root->inode_lock);
1882                 n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1883                                            (void **)delayed_nodes, inode_id,
1884                                            ARRAY_SIZE(delayed_nodes));
1885                 if (!n) {
1886                         spin_unlock(&root->inode_lock);
1887                         break;
1888                 }
1889
1890                 inode_id = delayed_nodes[n - 1]->inode_id + 1;
1891
1892                 for (i = 0; i < n; i++)
1893                         atomic_inc(&delayed_nodes[i]->refs);
1894                 spin_unlock(&root->inode_lock);
1895
1896                 for (i = 0; i < n; i++) {
1897                         __btrfs_kill_delayed_node(delayed_nodes[i]);
1898                         btrfs_release_delayed_node(delayed_nodes[i]);
1899                 }
1900         }
1901 }
1902
1903 void btrfs_destroy_delayed_inodes(struct btrfs_root *root)
1904 {
1905         struct btrfs_delayed_root *delayed_root;
1906         struct btrfs_delayed_node *curr_node, *prev_node;
1907
1908         delayed_root = btrfs_get_delayed_root(root);
1909
1910         curr_node = btrfs_first_delayed_node(delayed_root);
1911         while (curr_node) {
1912                 __btrfs_kill_delayed_node(curr_node);
1913
1914                 prev_node = curr_node;
1915                 curr_node = btrfs_next_delayed_node(curr_node);
1916                 btrfs_release_delayed_node(prev_node);
1917         }
1918 }
1919