Linux Audio

Check our new training course

Open-source upstreaming

Need help get the support for your hardware in upstream Linux?
Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2011 Fujitsu.  All rights reserved.
   4 * Written by Miao Xie <miaox@cn.fujitsu.com>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   5 */
   6
   7#include <linux/slab.h>
   8#include <linux/iversion.h>
   9#include "misc.h"
  10#include "delayed-inode.h"
  11#include "disk-io.h"
  12#include "transaction.h"
  13#include "ctree.h"
  14#include "qgroup.h"
  15
  16#define BTRFS_DELAYED_WRITEBACK		512
  17#define BTRFS_DELAYED_BACKGROUND	128
  18#define BTRFS_DELAYED_BATCH		16
  19
  20static struct kmem_cache *delayed_node_cache;
  21
  22int __init btrfs_delayed_inode_init(void)
  23{
  24	delayed_node_cache = kmem_cache_create("btrfs_delayed_node",
  25					sizeof(struct btrfs_delayed_node),
  26					0,
  27					SLAB_MEM_SPREAD,
  28					NULL);
  29	if (!delayed_node_cache)
  30		return -ENOMEM;
  31	return 0;
  32}
  33
  34void __cold btrfs_delayed_inode_exit(void)
  35{
  36	kmem_cache_destroy(delayed_node_cache);
 
  37}
  38
  39static inline void btrfs_init_delayed_node(
  40				struct btrfs_delayed_node *delayed_node,
  41				struct btrfs_root *root, u64 inode_id)
  42{
  43	delayed_node->root = root;
  44	delayed_node->inode_id = inode_id;
  45	refcount_set(&delayed_node->refs, 0);
  46	delayed_node->ins_root = RB_ROOT_CACHED;
  47	delayed_node->del_root = RB_ROOT_CACHED;
 
 
 
  48	mutex_init(&delayed_node->mutex);
 
  49	INIT_LIST_HEAD(&delayed_node->n_list);
  50	INIT_LIST_HEAD(&delayed_node->p_list);
 
  51}
  52
  53static inline int btrfs_is_continuous_delayed_item(
  54					struct btrfs_delayed_item *item1,
  55					struct btrfs_delayed_item *item2)
  56{
  57	if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  58	    item1->key.objectid == item2->key.objectid &&
  59	    item1->key.type == item2->key.type &&
  60	    item1->key.offset + 1 == item2->key.offset)
  61		return 1;
  62	return 0;
  63}
  64
  65static struct btrfs_delayed_node *btrfs_get_delayed_node(
  66		struct btrfs_inode *btrfs_inode)
 
 
 
 
 
  67{
 
  68	struct btrfs_root *root = btrfs_inode->root;
  69	u64 ino = btrfs_ino(btrfs_inode);
  70	struct btrfs_delayed_node *node;
  71
  72	node = READ_ONCE(btrfs_inode->delayed_node);
  73	if (node) {
  74		refcount_inc(&node->refs);
  75		return node;
  76	}
  77
  78	spin_lock(&root->inode_lock);
  79	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
  80
  81	if (node) {
  82		if (btrfs_inode->delayed_node) {
  83			refcount_inc(&node->refs);	/* can be accessed */
  84			BUG_ON(btrfs_inode->delayed_node != node);
  85			spin_unlock(&root->inode_lock);
  86			return node;
  87		}
  88
  89		/*
  90		 * It's possible that we're racing into the middle of removing
  91		 * this node from the radix tree.  In this case, the refcount
  92		 * was zero and it should never go back to one.  Just return
  93		 * NULL like it was never in the radix at all; our release
  94		 * function is in the process of removing it.
  95		 *
  96		 * Some implementations of refcount_inc refuse to bump the
  97		 * refcount once it has hit zero.  If we don't do this dance
  98		 * here, refcount_inc() may decide to just WARN_ONCE() instead
  99		 * of actually bumping the refcount.
 100		 *
 101		 * If this node is properly in the radix, we want to bump the
 102		 * refcount twice, once for the inode and once for this get
 103		 * operation.
 104		 */
 105		if (refcount_inc_not_zero(&node->refs)) {
 106			refcount_inc(&node->refs);
 107			btrfs_inode->delayed_node = node;
 108		} else {
 109			node = NULL;
 110		}
 111
 112		spin_unlock(&root->inode_lock);
 113		return node;
 114	}
 115	spin_unlock(&root->inode_lock);
 116
 117	return NULL;
 118}
 119
 120/* Will return either the node or PTR_ERR(-ENOMEM) */
 121static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 122		struct btrfs_inode *btrfs_inode)
 123{
 124	struct btrfs_delayed_node *node;
 
 125	struct btrfs_root *root = btrfs_inode->root;
 126	u64 ino = btrfs_ino(btrfs_inode);
 127	int ret;
 128
 129again:
 130	node = btrfs_get_delayed_node(btrfs_inode);
 131	if (node)
 132		return node;
 133
 134	node = kmem_cache_zalloc(delayed_node_cache, GFP_NOFS);
 135	if (!node)
 136		return ERR_PTR(-ENOMEM);
 137	btrfs_init_delayed_node(node, root, ino);
 138
 139	/* cached in the btrfs inode and can be accessed */
 140	refcount_set(&node->refs, 2);
 141
 142	ret = radix_tree_preload(GFP_NOFS);
 143	if (ret) {
 144		kmem_cache_free(delayed_node_cache, node);
 145		return ERR_PTR(ret);
 146	}
 147
 148	spin_lock(&root->inode_lock);
 149	ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 150	if (ret == -EEXIST) {
 151		spin_unlock(&root->inode_lock);
 152		kmem_cache_free(delayed_node_cache, node);
 
 153		radix_tree_preload_end();
 154		goto again;
 155	}
 156	btrfs_inode->delayed_node = node;
 157	spin_unlock(&root->inode_lock);
 158	radix_tree_preload_end();
 159
 160	return node;
 161}
 162
 163/*
 164 * Call it when holding delayed_node->mutex
 165 *
 166 * If mod = 1, add this node into the prepared list.
 167 */
 168static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 169				     struct btrfs_delayed_node *node,
 170				     int mod)
 171{
 172	spin_lock(&root->lock);
 173	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 174		if (!list_empty(&node->p_list))
 175			list_move_tail(&node->p_list, &root->prepare_list);
 176		else if (mod)
 177			list_add_tail(&node->p_list, &root->prepare_list);
 178	} else {
 179		list_add_tail(&node->n_list, &root->node_list);
 180		list_add_tail(&node->p_list, &root->prepare_list);
 181		refcount_inc(&node->refs);	/* inserted into list */
 182		root->nodes++;
 183		set_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 184	}
 185	spin_unlock(&root->lock);
 186}
 187
 188/* Call it when holding delayed_node->mutex */
 189static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 190				       struct btrfs_delayed_node *node)
 191{
 192	spin_lock(&root->lock);
 193	if (test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 194		root->nodes--;
 195		refcount_dec(&node->refs);	/* not in the list */
 196		list_del_init(&node->n_list);
 197		if (!list_empty(&node->p_list))
 198			list_del_init(&node->p_list);
 199		clear_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags);
 200	}
 201	spin_unlock(&root->lock);
 202}
 203
 204static struct btrfs_delayed_node *btrfs_first_delayed_node(
 205			struct btrfs_delayed_root *delayed_root)
 206{
 207	struct list_head *p;
 208	struct btrfs_delayed_node *node = NULL;
 209
 210	spin_lock(&delayed_root->lock);
 211	if (list_empty(&delayed_root->node_list))
 212		goto out;
 213
 214	p = delayed_root->node_list.next;
 215	node = list_entry(p, struct btrfs_delayed_node, n_list);
 216	refcount_inc(&node->refs);
 217out:
 218	spin_unlock(&delayed_root->lock);
 219
 220	return node;
 221}
 222
 223static struct btrfs_delayed_node *btrfs_next_delayed_node(
 224						struct btrfs_delayed_node *node)
 225{
 226	struct btrfs_delayed_root *delayed_root;
 227	struct list_head *p;
 228	struct btrfs_delayed_node *next = NULL;
 229
 230	delayed_root = node->root->fs_info->delayed_root;
 231	spin_lock(&delayed_root->lock);
 232	if (!test_bit(BTRFS_DELAYED_NODE_IN_LIST, &node->flags)) {
 233		/* not in the list */
 234		if (list_empty(&delayed_root->node_list))
 235			goto out;
 236		p = delayed_root->node_list.next;
 237	} else if (list_is_last(&node->n_list, &delayed_root->node_list))
 238		goto out;
 239	else
 240		p = node->n_list.next;
 241
 242	next = list_entry(p, struct btrfs_delayed_node, n_list);
 243	refcount_inc(&next->refs);
 244out:
 245	spin_unlock(&delayed_root->lock);
 246
 247	return next;
 248}
 249
 250static void __btrfs_release_delayed_node(
 251				struct btrfs_delayed_node *delayed_node,
 252				int mod)
 253{
 254	struct btrfs_delayed_root *delayed_root;
 255
 256	if (!delayed_node)
 257		return;
 258
 259	delayed_root = delayed_node->root->fs_info->delayed_root;
 260
 261	mutex_lock(&delayed_node->mutex);
 262	if (delayed_node->count)
 263		btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
 264	else
 265		btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 266	mutex_unlock(&delayed_node->mutex);
 267
 268	if (refcount_dec_and_test(&delayed_node->refs)) {
 269		struct btrfs_root *root = delayed_node->root;
 270
 271		spin_lock(&root->inode_lock);
 272		/*
 273		 * Once our refcount goes to zero, nobody is allowed to bump it
 274		 * back up.  We can delete it now.
 275		 */
 276		ASSERT(refcount_read(&delayed_node->refs) == 0);
 277		radix_tree_delete(&root->delayed_nodes_tree,
 278				  delayed_node->inode_id);
 279		spin_unlock(&root->inode_lock);
 280		kmem_cache_free(delayed_node_cache, delayed_node);
 281	}
 282}
 283
 284static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
 285{
 286	__btrfs_release_delayed_node(node, 0);
 287}
 288
 289static struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
 290					struct btrfs_delayed_root *delayed_root)
 291{
 292	struct list_head *p;
 293	struct btrfs_delayed_node *node = NULL;
 294
 295	spin_lock(&delayed_root->lock);
 296	if (list_empty(&delayed_root->prepare_list))
 297		goto out;
 298
 299	p = delayed_root->prepare_list.next;
 300	list_del_init(p);
 301	node = list_entry(p, struct btrfs_delayed_node, p_list);
 302	refcount_inc(&node->refs);
 303out:
 304	spin_unlock(&delayed_root->lock);
 305
 306	return node;
 307}
 308
 309static inline void btrfs_release_prepared_delayed_node(
 310					struct btrfs_delayed_node *node)
 311{
 312	__btrfs_release_delayed_node(node, 1);
 313}
 314
 315static struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
 316{
 317	struct btrfs_delayed_item *item;
 318	item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
 319	if (item) {
 320		item->data_len = data_len;
 321		item->ins_or_del = 0;
 322		item->bytes_reserved = 0;
 323		item->delayed_node = NULL;
 324		refcount_set(&item->refs, 1);
 325	}
 326	return item;
 327}
 328
 329/*
 330 * __btrfs_lookup_delayed_item - look up the delayed item by key
 331 * @delayed_node: pointer to the delayed node
 332 * @key:	  the key to look up
 333 * @prev:	  used to store the prev item if the right item isn't found
 334 * @next:	  used to store the next item if the right item isn't found
 335 *
 336 * Note: if we don't find the right item, we will return the prev item and
 337 * the next item.
 338 */
 339static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 340				struct rb_root *root,
 341				struct btrfs_key *key,
 342				struct btrfs_delayed_item **prev,
 343				struct btrfs_delayed_item **next)
 344{
 345	struct rb_node *node, *prev_node = NULL;
 346	struct btrfs_delayed_item *delayed_item = NULL;
 347	int ret = 0;
 348
 349	node = root->rb_node;
 350
 351	while (node) {
 352		delayed_item = rb_entry(node, struct btrfs_delayed_item,
 353					rb_node);
 354		prev_node = node;
 355		ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
 356		if (ret < 0)
 357			node = node->rb_right;
 358		else if (ret > 0)
 359			node = node->rb_left;
 360		else
 361			return delayed_item;
 362	}
 363
 364	if (prev) {
 365		if (!prev_node)
 366			*prev = NULL;
 367		else if (ret < 0)
 368			*prev = delayed_item;
 369		else if ((node = rb_prev(prev_node)) != NULL) {
 370			*prev = rb_entry(node, struct btrfs_delayed_item,
 371					 rb_node);
 372		} else
 373			*prev = NULL;
 374	}
 375
 376	if (next) {
 377		if (!prev_node)
 378			*next = NULL;
 379		else if (ret > 0)
 380			*next = delayed_item;
 381		else if ((node = rb_next(prev_node)) != NULL) {
 382			*next = rb_entry(node, struct btrfs_delayed_item,
 383					 rb_node);
 384		} else
 385			*next = NULL;
 386	}
 387	return NULL;
 388}
 389
 390static struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
 
 
 
 
 
 
 
 
 
 
 
 391					struct btrfs_delayed_node *delayed_node,
 392					struct btrfs_key *key)
 393{
 394	return __btrfs_lookup_delayed_item(&delayed_node->ins_root.rb_root, key,
 
 
 395					   NULL, NULL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 396}
 397
 398static 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_cached *root;
 405	struct btrfs_delayed_item *item;
 406	int cmp;
 407	bool leftmost = true;
 408
 409	if (action == BTRFS_DELAYED_INSERTION_ITEM)
 410		root = &delayed_node->ins_root;
 411	else if (action == BTRFS_DELAYED_DELETION_ITEM)
 412		root = &delayed_node->del_root;
 413	else
 414		BUG();
 415	p = &root->rb_root.rb_node;
 416	node = &ins->rb_node;
 417
 418	while (*p) {
 419		parent_node = *p;
 420		item = rb_entry(parent_node, struct btrfs_delayed_item,
 421				 rb_node);
 422
 423		cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
 424		if (cmp < 0) {
 425			p = &(*p)->rb_right;
 426			leftmost = false;
 427		} else if (cmp > 0) {
 428			p = &(*p)->rb_left;
 429		} else {
 430			return -EEXIST;
 431		}
 432	}
 433
 434	rb_link_node(node, parent_node, p);
 435	rb_insert_color_cached(node, root, leftmost);
 436	ins->delayed_node = delayed_node;
 437	ins->ins_or_del = action;
 438
 439	if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
 440	    action == BTRFS_DELAYED_INSERTION_ITEM &&
 441	    ins->key.offset >= delayed_node->index_cnt)
 442			delayed_node->index_cnt = ins->key.offset + 1;
 443
 444	delayed_node->count++;
 445	atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
 446	return 0;
 447}
 448
 449static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
 450					      struct btrfs_delayed_item *item)
 451{
 452	return __btrfs_add_delayed_item(node, item,
 453					BTRFS_DELAYED_INSERTION_ITEM);
 454}
 455
 456static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
 457					     struct btrfs_delayed_item *item)
 458{
 459	return __btrfs_add_delayed_item(node, item,
 460					BTRFS_DELAYED_DELETION_ITEM);
 461}
 462
 463static void finish_one_item(struct btrfs_delayed_root *delayed_root)
 464{
 465	int seq = atomic_inc_return(&delayed_root->items_seq);
 466
 467	/* atomic_dec_return implies a barrier */
 468	if ((atomic_dec_return(&delayed_root->items) <
 469	    BTRFS_DELAYED_BACKGROUND || seq % BTRFS_DELAYED_BATCH == 0))
 470		cond_wake_up_nomb(&delayed_root->wait);
 471}
 472
 473static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 474{
 475	struct rb_root_cached *root;
 476	struct btrfs_delayed_root *delayed_root;
 477
 478	/* Not associated with any delayed_node */
 479	if (!delayed_item->delayed_node)
 480		return;
 481	delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
 482
 483	BUG_ON(!delayed_root);
 484	BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
 485	       delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
 486
 487	if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
 488		root = &delayed_item->delayed_node->ins_root;
 489	else
 490		root = &delayed_item->delayed_node->del_root;
 491
 492	rb_erase_cached(&delayed_item->rb_node, root);
 493	delayed_item->delayed_node->count--;
 494
 495	finish_one_item(delayed_root);
 
 
 496}
 497
 498static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 499{
 500	if (item) {
 501		__btrfs_remove_delayed_item(item);
 502		if (refcount_dec_and_test(&item->refs))
 503			kfree(item);
 504	}
 505}
 506
 507static struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 508					struct btrfs_delayed_node *delayed_node)
 509{
 510	struct rb_node *p;
 511	struct btrfs_delayed_item *item = NULL;
 512
 513	p = rb_first_cached(&delayed_node->ins_root);
 514	if (p)
 515		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 516
 517	return item;
 518}
 519
 520static struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 521					struct btrfs_delayed_node *delayed_node)
 522{
 523	struct rb_node *p;
 524	struct btrfs_delayed_item *item = NULL;
 525
 526	p = rb_first_cached(&delayed_node->del_root);
 527	if (p)
 528		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 529
 530	return item;
 531}
 532
 533static struct btrfs_delayed_item *__btrfs_next_delayed_item(
 534						struct btrfs_delayed_item *item)
 535{
 536	struct rb_node *p;
 537	struct btrfs_delayed_item *next = NULL;
 538
 539	p = rb_next(&item->rb_node);
 540	if (p)
 541		next = rb_entry(p, struct btrfs_delayed_item, rb_node);
 542
 543	return next;
 544}
 545
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 546static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 547					       struct btrfs_root *root,
 548					       struct btrfs_delayed_item *item)
 549{
 550	struct btrfs_block_rsv *src_rsv;
 551	struct btrfs_block_rsv *dst_rsv;
 552	struct btrfs_fs_info *fs_info = root->fs_info;
 553	u64 num_bytes;
 554	int ret;
 555
 556	if (!trans->bytes_reserved)
 557		return 0;
 558
 559	src_rsv = trans->block_rsv;
 560	dst_rsv = &fs_info->delayed_block_rsv;
 561
 562	num_bytes = btrfs_calc_insert_metadata_size(fs_info, 1);
 563
 564	/*
 565	 * Here we migrate space rsv from transaction rsv, since have already
 566	 * reserved space when starting a transaction.  So no need to reserve
 567	 * qgroup space here.
 568	 */
 569	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
 570	if (!ret) {
 571		trace_btrfs_space_reservation(fs_info, "delayed_item",
 572					      item->key.objectid,
 573					      num_bytes, 1);
 574		item->bytes_reserved = num_bytes;
 575	}
 576
 577	return ret;
 578}
 579
 580static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 581						struct btrfs_delayed_item *item)
 582{
 583	struct btrfs_block_rsv *rsv;
 584	struct btrfs_fs_info *fs_info = root->fs_info;
 585
 586	if (!item->bytes_reserved)
 587		return;
 588
 589	rsv = &fs_info->delayed_block_rsv;
 590	/*
 591	 * Check btrfs_delayed_item_reserve_metadata() to see why we don't need
 592	 * to release/reserve qgroup space.
 593	 */
 594	trace_btrfs_space_reservation(fs_info, "delayed_item",
 595				      item->key.objectid, item->bytes_reserved,
 596				      0);
 597	btrfs_block_rsv_release(fs_info, rsv,
 598				item->bytes_reserved);
 599}
 600
 601static int btrfs_delayed_inode_reserve_metadata(
 602					struct btrfs_trans_handle *trans,
 603					struct btrfs_root *root,
 604					struct btrfs_inode *inode,
 605					struct btrfs_delayed_node *node)
 606{
 607	struct btrfs_fs_info *fs_info = root->fs_info;
 608	struct btrfs_block_rsv *src_rsv;
 609	struct btrfs_block_rsv *dst_rsv;
 610	u64 num_bytes;
 611	int ret;
 612
 613	src_rsv = trans->block_rsv;
 614	dst_rsv = &fs_info->delayed_block_rsv;
 615
 616	num_bytes = btrfs_calc_metadata_size(fs_info, 1);
 617
 618	/*
 619	 * btrfs_dirty_inode will update the inode under btrfs_join_transaction
 620	 * which doesn't reserve space for speed.  This is a problem since we
 621	 * still need to reserve space for this update, so try to reserve the
 622	 * space.
 623	 *
 624	 * Now if src_rsv == delalloc_block_rsv we'll let it just steal since
 625	 * we always reserve enough to update the inode item.
 626	 */
 627	if (!src_rsv || (!trans->bytes_reserved &&
 628			 src_rsv->type != BTRFS_BLOCK_RSV_DELALLOC)) {
 629		ret = btrfs_qgroup_reserve_meta_prealloc(root,
 630				fs_info->nodesize, true);
 631		if (ret < 0)
 632			return ret;
 633		ret = btrfs_block_rsv_add(root, dst_rsv, num_bytes,
 634					  BTRFS_RESERVE_NO_FLUSH);
 635		/*
 636		 * Since we're under a transaction reserve_metadata_bytes could
 637		 * try to commit the transaction which will make it return
 638		 * EAGAIN to make us stop the transaction we have, so return
 639		 * ENOSPC instead so that btrfs_dirty_inode knows what to do.
 640		 */
 641		if (ret == -EAGAIN) {
 642			ret = -ENOSPC;
 643			btrfs_qgroup_free_meta_prealloc(root, num_bytes);
 644		}
 645		if (!ret) {
 646			node->bytes_reserved = num_bytes;
 647			trace_btrfs_space_reservation(fs_info,
 648						      "delayed_inode",
 649						      btrfs_ino(inode),
 650						      num_bytes, 1);
 651		} else {
 652			btrfs_qgroup_free_meta_prealloc(root, fs_info->nodesize);
 653		}
 654		return ret;
 655	}
 656
 657	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, true);
 658	if (!ret) {
 659		trace_btrfs_space_reservation(fs_info, "delayed_inode",
 660					      btrfs_ino(inode), num_bytes, 1);
 661		node->bytes_reserved = num_bytes;
 662	}
 663
 664	return ret;
 665}
 666
 667static void btrfs_delayed_inode_release_metadata(struct btrfs_fs_info *fs_info,
 668						struct btrfs_delayed_node *node,
 669						bool qgroup_free)
 670{
 671	struct btrfs_block_rsv *rsv;
 672
 673	if (!node->bytes_reserved)
 674		return;
 675
 676	rsv = &fs_info->delayed_block_rsv;
 677	trace_btrfs_space_reservation(fs_info, "delayed_inode",
 678				      node->inode_id, node->bytes_reserved, 0);
 679	btrfs_block_rsv_release(fs_info, rsv,
 680				node->bytes_reserved);
 681	if (qgroup_free)
 682		btrfs_qgroup_free_meta_prealloc(node->root,
 683				node->bytes_reserved);
 684	else
 685		btrfs_qgroup_convert_reserved_meta(node->root,
 686				node->bytes_reserved);
 687	node->bytes_reserved = 0;
 688}
 689
 690/*
 691 * This helper will insert some continuous items into the same leaf according
 692 * to the free space of the leaf.
 693 */
 694static int btrfs_batch_insert_items(struct btrfs_root *root,
 695				    struct btrfs_path *path,
 696				    struct btrfs_delayed_item *item)
 
 697{
 698	struct btrfs_delayed_item *curr, *next;
 699	int free_space;
 700	int total_data_size = 0, total_size = 0;
 701	struct extent_buffer *leaf;
 702	char *data_ptr;
 703	struct btrfs_key *keys;
 704	u32 *data_size;
 705	struct list_head head;
 706	int slot;
 707	int nitems;
 708	int i;
 709	int ret = 0;
 710
 711	BUG_ON(!path->nodes[0]);
 712
 713	leaf = path->nodes[0];
 714	free_space = btrfs_leaf_free_space(leaf);
 715	INIT_LIST_HEAD(&head);
 716
 717	next = item;
 718	nitems = 0;
 719
 720	/*
 721	 * count the number of the continuous items that we can insert in batch
 722	 */
 723	while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 724	       free_space) {
 725		total_data_size += next->data_len;
 726		total_size += next->data_len + sizeof(struct btrfs_item);
 727		list_add_tail(&next->tree_list, &head);
 728		nitems++;
 729
 730		curr = next;
 731		next = __btrfs_next_delayed_item(curr);
 732		if (!next)
 733			break;
 734
 735		if (!btrfs_is_continuous_delayed_item(curr, next))
 736			break;
 737	}
 738
 739	if (!nitems) {
 740		ret = 0;
 741		goto out;
 742	}
 743
 744	/*
 745	 * we need allocate some memory space, but it might cause the task
 746	 * to sleep, so we set all locked nodes in the path to blocking locks
 747	 * first.
 748	 */
 749	btrfs_set_path_blocking(path);
 750
 751	keys = kmalloc_array(nitems, sizeof(struct btrfs_key), GFP_NOFS);
 752	if (!keys) {
 753		ret = -ENOMEM;
 754		goto out;
 755	}
 756
 757	data_size = kmalloc_array(nitems, sizeof(u32), GFP_NOFS);
 758	if (!data_size) {
 759		ret = -ENOMEM;
 760		goto error;
 761	}
 762
 763	/* get keys of all the delayed items */
 764	i = 0;
 765	list_for_each_entry(next, &head, tree_list) {
 766		keys[i] = next->key;
 767		data_size[i] = next->data_len;
 768		i++;
 769	}
 770
 
 
 
 771	/* insert the keys of the items */
 772	setup_items_for_insert(root, path, keys, data_size,
 773			       total_data_size, total_size, nitems);
 
 
 774
 775	/* insert the dir index items */
 776	slot = path->slots[0];
 777	list_for_each_entry_safe(curr, next, &head, tree_list) {
 778		data_ptr = btrfs_item_ptr(leaf, slot, char);
 779		write_extent_buffer(leaf, &curr->data,
 780				    (unsigned long)data_ptr,
 781				    curr->data_len);
 782		slot++;
 783
 784		btrfs_delayed_item_release_metadata(root, curr);
 785
 786		list_del(&curr->tree_list);
 787		btrfs_release_delayed_item(curr);
 788	}
 789
 790error:
 791	kfree(data_size);
 792	kfree(keys);
 793out:
 794	return ret;
 795}
 796
 797/*
 798 * This helper can just do simple insertion that needn't extend item for new
 799 * data, such as directory name index insertion, inode insertion.
 800 */
 801static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 802				     struct btrfs_root *root,
 803				     struct btrfs_path *path,
 804				     struct btrfs_delayed_item *delayed_item)
 805{
 806	struct extent_buffer *leaf;
 
 807	char *ptr;
 808	int ret;
 809
 810	ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 811				      delayed_item->data_len);
 812	if (ret < 0 && ret != -EEXIST)
 813		return ret;
 814
 815	leaf = path->nodes[0];
 816
 
 817	ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 818
 819	write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 820			    delayed_item->data_len);
 821	btrfs_mark_buffer_dirty(leaf);
 822
 823	btrfs_delayed_item_release_metadata(root, delayed_item);
 824	return 0;
 825}
 826
 827/*
 828 * we insert an item first, then if there are some continuous items, we try
 829 * to insert those items into the same leaf.
 830 */
 831static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 832				      struct btrfs_path *path,
 833				      struct btrfs_root *root,
 834				      struct btrfs_delayed_node *node)
 835{
 836	struct btrfs_delayed_item *curr, *prev;
 837	int ret = 0;
 838
 839do_again:
 840	mutex_lock(&node->mutex);
 841	curr = __btrfs_first_delayed_insertion_item(node);
 842	if (!curr)
 843		goto insert_end;
 844
 845	ret = btrfs_insert_delayed_item(trans, root, path, curr);
 846	if (ret < 0) {
 847		btrfs_release_path(path);
 848		goto insert_end;
 849	}
 850
 851	prev = curr;
 852	curr = __btrfs_next_delayed_item(prev);
 853	if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 854		/* insert the continuous items into the same leaf */
 855		path->slots[0]++;
 856		btrfs_batch_insert_items(root, path, curr);
 857	}
 858	btrfs_release_delayed_item(prev);
 859	btrfs_mark_buffer_dirty(path->nodes[0]);
 860
 861	btrfs_release_path(path);
 862	mutex_unlock(&node->mutex);
 863	goto do_again;
 864
 865insert_end:
 866	mutex_unlock(&node->mutex);
 867	return ret;
 868}
 869
 870static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 871				    struct btrfs_root *root,
 872				    struct btrfs_path *path,
 873				    struct btrfs_delayed_item *item)
 874{
 875	struct btrfs_delayed_item *curr, *next;
 876	struct extent_buffer *leaf;
 877	struct btrfs_key key;
 878	struct list_head head;
 879	int nitems, i, last_item;
 880	int ret = 0;
 881
 882	BUG_ON(!path->nodes[0]);
 883
 884	leaf = path->nodes[0];
 885
 886	i = path->slots[0];
 887	last_item = btrfs_header_nritems(leaf) - 1;
 888	if (i > last_item)
 889		return -ENOENT;	/* FIXME: Is errno suitable? */
 890
 891	next = item;
 892	INIT_LIST_HEAD(&head);
 893	btrfs_item_key_to_cpu(leaf, &key, i);
 894	nitems = 0;
 895	/*
 896	 * count the number of the dir index items that we can delete in batch
 897	 */
 898	while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 899		list_add_tail(&next->tree_list, &head);
 900		nitems++;
 901
 902		curr = next;
 903		next = __btrfs_next_delayed_item(curr);
 904		if (!next)
 905			break;
 906
 907		if (!btrfs_is_continuous_delayed_item(curr, next))
 908			break;
 909
 910		i++;
 911		if (i > last_item)
 912			break;
 913		btrfs_item_key_to_cpu(leaf, &key, i);
 914	}
 915
 916	if (!nitems)
 917		return 0;
 918
 919	ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 920	if (ret)
 921		goto out;
 922
 923	list_for_each_entry_safe(curr, next, &head, tree_list) {
 924		btrfs_delayed_item_release_metadata(root, curr);
 925		list_del(&curr->tree_list);
 926		btrfs_release_delayed_item(curr);
 927	}
 928
 929out:
 930	return ret;
 931}
 932
 933static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 934				      struct btrfs_path *path,
 935				      struct btrfs_root *root,
 936				      struct btrfs_delayed_node *node)
 937{
 938	struct btrfs_delayed_item *curr, *prev;
 939	int ret = 0;
 940
 941do_again:
 942	mutex_lock(&node->mutex);
 943	curr = __btrfs_first_delayed_deletion_item(node);
 944	if (!curr)
 945		goto delete_fail;
 946
 947	ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 948	if (ret < 0)
 949		goto delete_fail;
 950	else if (ret > 0) {
 951		/*
 952		 * can't find the item which the node points to, so this node
 953		 * is invalid, just drop it.
 954		 */
 955		prev = curr;
 956		curr = __btrfs_next_delayed_item(prev);
 957		btrfs_release_delayed_item(prev);
 958		ret = 0;
 959		btrfs_release_path(path);
 960		if (curr) {
 961			mutex_unlock(&node->mutex);
 962			goto do_again;
 963		} else
 964			goto delete_fail;
 965	}
 966
 967	btrfs_batch_delete_items(trans, root, path, curr);
 968	btrfs_release_path(path);
 969	mutex_unlock(&node->mutex);
 970	goto do_again;
 971
 972delete_fail:
 973	btrfs_release_path(path);
 974	mutex_unlock(&node->mutex);
 975	return ret;
 976}
 977
 978static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 979{
 980	struct btrfs_delayed_root *delayed_root;
 981
 982	if (delayed_node &&
 983	    test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
 984		BUG_ON(!delayed_node->root);
 985		clear_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
 986		delayed_node->count--;
 987
 988		delayed_root = delayed_node->root->fs_info->delayed_root;
 989		finish_one_item(delayed_root);
 
 
 
 
 990	}
 991}
 992
 993static void btrfs_release_delayed_iref(struct btrfs_delayed_node *delayed_node)
 994{
 995	struct btrfs_delayed_root *delayed_root;
 996
 997	ASSERT(delayed_node->root);
 998	clear_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
 999	delayed_node->count--;
1000
1001	delayed_root = delayed_node->root->fs_info->delayed_root;
1002	finish_one_item(delayed_root);
1003}
1004
1005static int __btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1006					struct btrfs_root *root,
1007					struct btrfs_path *path,
1008					struct btrfs_delayed_node *node)
1009{
1010	struct btrfs_fs_info *fs_info = root->fs_info;
1011	struct btrfs_key key;
1012	struct btrfs_inode_item *inode_item;
1013	struct extent_buffer *leaf;
1014	int mod;
1015	int ret;
1016
 
 
 
 
 
 
1017	key.objectid = node->inode_id;
1018	key.type = BTRFS_INODE_ITEM_KEY;
1019	key.offset = 0;
1020
1021	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1022		mod = -1;
1023	else
1024		mod = 1;
1025
1026	ret = btrfs_lookup_inode(trans, root, path, &key, mod);
1027	if (ret > 0) {
1028		btrfs_release_path(path);
 
1029		return -ENOENT;
1030	} else if (ret < 0) {
 
1031		return ret;
1032	}
1033
 
1034	leaf = path->nodes[0];
1035	inode_item = btrfs_item_ptr(leaf, path->slots[0],
1036				    struct btrfs_inode_item);
1037	write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1038			    sizeof(struct btrfs_inode_item));
1039	btrfs_mark_buffer_dirty(leaf);
1040
1041	if (!test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &node->flags))
1042		goto no_iref;
1043
1044	path->slots[0]++;
1045	if (path->slots[0] >= btrfs_header_nritems(leaf))
1046		goto search;
1047again:
1048	btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1049	if (key.objectid != node->inode_id)
1050		goto out;
1051
1052	if (key.type != BTRFS_INODE_REF_KEY &&
1053	    key.type != BTRFS_INODE_EXTREF_KEY)
1054		goto out;
1055
1056	/*
1057	 * Delayed iref deletion is for the inode who has only one link,
1058	 * so there is only one iref. The case that several irefs are
1059	 * in the same item doesn't exist.
1060	 */
1061	btrfs_del_item(trans, root, path);
1062out:
1063	btrfs_release_delayed_iref(node);
1064no_iref:
1065	btrfs_release_path(path);
1066err_out:
1067	btrfs_delayed_inode_release_metadata(fs_info, node, (ret < 0));
1068	btrfs_release_delayed_inode(node);
1069
1070	return ret;
1071
1072search:
1073	btrfs_release_path(path);
1074
1075	key.type = BTRFS_INODE_EXTREF_KEY;
1076	key.offset = -1;
1077	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1078	if (ret < 0)
1079		goto err_out;
1080	ASSERT(ret);
1081
1082	ret = 0;
1083	leaf = path->nodes[0];
1084	path->slots[0]--;
1085	goto again;
1086}
1087
1088static inline int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
1089					     struct btrfs_root *root,
1090					     struct btrfs_path *path,
1091					     struct btrfs_delayed_node *node)
1092{
1093	int ret;
1094
1095	mutex_lock(&node->mutex);
1096	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &node->flags)) {
1097		mutex_unlock(&node->mutex);
1098		return 0;
1099	}
1100
1101	ret = __btrfs_update_delayed_inode(trans, root, path, node);
1102	mutex_unlock(&node->mutex);
1103	return ret;
1104}
1105
1106static inline int
1107__btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1108				   struct btrfs_path *path,
1109				   struct btrfs_delayed_node *node)
1110{
1111	int ret;
1112
1113	ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1114	if (ret)
1115		return ret;
1116
1117	ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1118	if (ret)
1119		return ret;
1120
1121	ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1122	return ret;
1123}
1124
1125/*
1126 * Called when committing the transaction.
1127 * Returns 0 on success.
1128 * Returns < 0 on error and returns with an aborted transaction with any
1129 * outstanding delayed items cleaned up.
1130 */
1131static int __btrfs_run_delayed_items(struct btrfs_trans_handle *trans, int nr)
1132{
1133	struct btrfs_fs_info *fs_info = trans->fs_info;
1134	struct btrfs_delayed_root *delayed_root;
1135	struct btrfs_delayed_node *curr_node, *prev_node;
1136	struct btrfs_path *path;
1137	struct btrfs_block_rsv *block_rsv;
1138	int ret = 0;
1139	bool count = (nr > 0);
1140
1141	if (trans->aborted)
1142		return -EIO;
1143
1144	path = btrfs_alloc_path();
1145	if (!path)
1146		return -ENOMEM;
1147	path->leave_spinning = 1;
1148
1149	block_rsv = trans->block_rsv;
1150	trans->block_rsv = &fs_info->delayed_block_rsv;
1151
1152	delayed_root = fs_info->delayed_root;
1153
1154	curr_node = btrfs_first_delayed_node(delayed_root);
1155	while (curr_node && (!count || (count && nr--))) {
1156		ret = __btrfs_commit_inode_delayed_items(trans, path,
 
 
 
 
 
 
 
1157							 curr_node);
1158		if (ret) {
1159			btrfs_release_delayed_node(curr_node);
1160			curr_node = NULL;
1161			btrfs_abort_transaction(trans, ret);
1162			break;
1163		}
1164
1165		prev_node = curr_node;
1166		curr_node = btrfs_next_delayed_node(curr_node);
1167		btrfs_release_delayed_node(prev_node);
1168	}
1169
1170	if (curr_node)
1171		btrfs_release_delayed_node(curr_node);
1172	btrfs_free_path(path);
1173	trans->block_rsv = block_rsv;
1174
1175	return ret;
1176}
1177
1178int btrfs_run_delayed_items(struct btrfs_trans_handle *trans)
1179{
1180	return __btrfs_run_delayed_items(trans, -1);
1181}
1182
1183int btrfs_run_delayed_items_nr(struct btrfs_trans_handle *trans, int nr)
1184{
1185	return __btrfs_run_delayed_items(trans, nr);
1186}
1187
1188int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1189				     struct btrfs_inode *inode)
1190{
1191	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1192	struct btrfs_path *path;
1193	struct btrfs_block_rsv *block_rsv;
1194	int ret;
1195
1196	if (!delayed_node)
1197		return 0;
1198
1199	mutex_lock(&delayed_node->mutex);
1200	if (!delayed_node->count) {
1201		mutex_unlock(&delayed_node->mutex);
1202		btrfs_release_delayed_node(delayed_node);
1203		return 0;
1204	}
1205	mutex_unlock(&delayed_node->mutex);
1206
1207	path = btrfs_alloc_path();
1208	if (!path) {
1209		btrfs_release_delayed_node(delayed_node);
1210		return -ENOMEM;
1211	}
1212	path->leave_spinning = 1;
1213
1214	block_rsv = trans->block_rsv;
1215	trans->block_rsv = &delayed_node->root->fs_info->delayed_block_rsv;
1216
1217	ret = __btrfs_commit_inode_delayed_items(trans, path, delayed_node);
1218
1219	btrfs_release_delayed_node(delayed_node);
 
 
1220	btrfs_free_path(path);
1221	trans->block_rsv = block_rsv;
1222
 
1223	return ret;
1224}
1225
1226int btrfs_commit_inode_delayed_inode(struct btrfs_inode *inode)
 
1227{
1228	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1229	struct btrfs_trans_handle *trans;
1230	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1231	struct btrfs_path *path;
1232	struct btrfs_block_rsv *block_rsv;
1233	int ret;
1234
1235	if (!delayed_node)
1236		return 0;
1237
1238	mutex_lock(&delayed_node->mutex);
1239	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1240		mutex_unlock(&delayed_node->mutex);
1241		btrfs_release_delayed_node(delayed_node);
1242		return 0;
1243	}
1244	mutex_unlock(&delayed_node->mutex);
1245
1246	trans = btrfs_join_transaction(delayed_node->root);
1247	if (IS_ERR(trans)) {
1248		ret = PTR_ERR(trans);
1249		goto out;
1250	}
1251
1252	path = btrfs_alloc_path();
1253	if (!path) {
1254		ret = -ENOMEM;
1255		goto trans_out;
1256	}
1257	path->leave_spinning = 1;
1258
1259	block_rsv = trans->block_rsv;
1260	trans->block_rsv = &fs_info->delayed_block_rsv;
1261
1262	mutex_lock(&delayed_node->mutex);
1263	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags))
1264		ret = __btrfs_update_delayed_inode(trans, delayed_node->root,
1265						   path, delayed_node);
1266	else
1267		ret = 0;
1268	mutex_unlock(&delayed_node->mutex);
1269
1270	btrfs_free_path(path);
1271	trans->block_rsv = block_rsv;
1272trans_out:
1273	btrfs_end_transaction(trans);
1274	btrfs_btree_balance_dirty(fs_info);
1275out:
1276	btrfs_release_delayed_node(delayed_node);
1277
1278	return ret;
1279}
1280
1281void btrfs_remove_delayed_node(struct btrfs_inode *inode)
1282{
1283	struct btrfs_delayed_node *delayed_node;
1284
1285	delayed_node = READ_ONCE(inode->delayed_node);
1286	if (!delayed_node)
1287		return;
1288
1289	inode->delayed_node = NULL;
1290	btrfs_release_delayed_node(delayed_node);
1291}
1292
1293struct btrfs_async_delayed_work {
1294	struct btrfs_delayed_root *delayed_root;
1295	int nr;
1296	struct btrfs_work work;
1297};
1298
1299static void btrfs_async_run_delayed_root(struct btrfs_work *work)
1300{
1301	struct btrfs_async_delayed_work *async_work;
1302	struct btrfs_delayed_root *delayed_root;
1303	struct btrfs_trans_handle *trans;
1304	struct btrfs_path *path;
1305	struct btrfs_delayed_node *delayed_node = NULL;
1306	struct btrfs_root *root;
1307	struct btrfs_block_rsv *block_rsv;
1308	int total_done = 0;
 
 
1309
1310	async_work = container_of(work, struct btrfs_async_delayed_work, work);
1311	delayed_root = async_work->delayed_root;
1312
1313	path = btrfs_alloc_path();
1314	if (!path)
1315		goto out;
 
1316
1317	do {
1318		if (atomic_read(&delayed_root->items) <
1319		    BTRFS_DELAYED_BACKGROUND / 2)
1320			break;
1321
1322		delayed_node = btrfs_first_prepared_delayed_node(delayed_root);
1323		if (!delayed_node)
1324			break;
1325
1326		path->leave_spinning = 1;
1327		root = delayed_node->root;
1328
1329		trans = btrfs_join_transaction(root);
1330		if (IS_ERR(trans)) {
1331			btrfs_release_path(path);
1332			btrfs_release_prepared_delayed_node(delayed_node);
1333			total_done++;
1334			continue;
1335		}
1336
1337		block_rsv = trans->block_rsv;
1338		trans->block_rsv = &root->fs_info->delayed_block_rsv;
1339
1340		__btrfs_commit_inode_delayed_items(trans, path, delayed_node);
 
 
 
1341
1342		trans->block_rsv = block_rsv;
1343		btrfs_end_transaction(trans);
1344		btrfs_btree_balance_dirty_nodelay(root->fs_info);
1345
1346		btrfs_release_path(path);
1347		btrfs_release_prepared_delayed_node(delayed_node);
1348		total_done++;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1349
1350	} while ((async_work->nr == 0 && total_done < BTRFS_DELAYED_WRITEBACK)
1351		 || total_done < async_work->nr);
1352
 
 
 
 
1353	btrfs_free_path(path);
1354out:
1355	wake_up(&delayed_root->wait);
1356	kfree(async_work);
 
 
 
 
1357}
1358
1359
1360static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1361				     struct btrfs_fs_info *fs_info, int nr)
1362{
1363	struct btrfs_async_delayed_work *async_work;
 
 
 
 
 
 
 
1364
1365	async_work = kmalloc(sizeof(*async_work), GFP_NOFS);
1366	if (!async_work)
 
1367		return -ENOMEM;
 
 
 
 
 
 
 
1368
1369	async_work->delayed_root = delayed_root;
1370	btrfs_init_work(&async_work->work, btrfs_delayed_meta_helper,
1371			btrfs_async_run_delayed_root, NULL, NULL);
1372	async_work->nr = nr;
 
1373
1374	btrfs_queue_work(fs_info->delayed_workers, &async_work->work);
1375	return 0;
1376}
1377
1378void btrfs_assert_delayed_root_empty(struct btrfs_fs_info *fs_info)
1379{
1380	WARN_ON(btrfs_first_delayed_node(fs_info->delayed_root));
 
 
1381}
1382
1383static int could_end_wait(struct btrfs_delayed_root *delayed_root, int seq)
1384{
1385	int val = atomic_read(&delayed_root->items_seq);
1386
1387	if (val < seq || val >= seq + BTRFS_DELAYED_BATCH)
1388		return 1;
1389
1390	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
1391		return 1;
1392
1393	return 0;
1394}
1395
1396void btrfs_balance_delayed_items(struct btrfs_fs_info *fs_info)
1397{
1398	struct btrfs_delayed_root *delayed_root = fs_info->delayed_root;
1399
1400	if ((atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND) ||
1401		btrfs_workqueue_normal_congested(fs_info->delayed_workers))
1402		return;
1403
1404	if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
1405		int seq;
1406		int ret;
1407
1408		seq = atomic_read(&delayed_root->items_seq);
1409
1410		ret = btrfs_wq_run_delayed_node(delayed_root, fs_info, 0);
1411		if (ret)
1412			return;
1413
1414		wait_event_interruptible(delayed_root->wait,
1415					 could_end_wait(delayed_root, seq));
 
 
 
1416		return;
1417	}
1418
1419	btrfs_wq_run_delayed_node(delayed_root, fs_info, BTRFS_DELAYED_BATCH);
1420}
1421
1422/* Will return 0 or -ENOMEM */
1423int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1424				   const char *name, int name_len,
1425				   struct btrfs_inode *dir,
1426				   struct btrfs_disk_key *disk_key, u8 type,
1427				   u64 index)
1428{
1429	struct btrfs_delayed_node *delayed_node;
1430	struct btrfs_delayed_item *delayed_item;
1431	struct btrfs_dir_item *dir_item;
1432	int ret;
1433
1434	delayed_node = btrfs_get_or_create_delayed_node(dir);
1435	if (IS_ERR(delayed_node))
1436		return PTR_ERR(delayed_node);
1437
1438	delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1439	if (!delayed_item) {
1440		ret = -ENOMEM;
1441		goto release_node;
1442	}
1443
 
 
 
 
 
 
 
1444	delayed_item->key.objectid = btrfs_ino(dir);
1445	delayed_item->key.type = BTRFS_DIR_INDEX_KEY;
1446	delayed_item->key.offset = index;
1447
1448	dir_item = (struct btrfs_dir_item *)delayed_item->data;
1449	dir_item->location = *disk_key;
1450	btrfs_set_stack_dir_transid(dir_item, trans->transid);
1451	btrfs_set_stack_dir_data_len(dir_item, 0);
1452	btrfs_set_stack_dir_name_len(dir_item, name_len);
1453	btrfs_set_stack_dir_type(dir_item, type);
1454	memcpy((char *)(dir_item + 1), name, name_len);
1455
1456	ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, delayed_item);
1457	/*
1458	 * we have reserved enough space when we start a new transaction,
1459	 * so reserving metadata failure is impossible
1460	 */
1461	BUG_ON(ret);
1462
1463	mutex_lock(&delayed_node->mutex);
1464	ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1465	if (unlikely(ret)) {
1466		btrfs_err(trans->fs_info,
1467			  "err add delayed dir index item(name: %.*s) into the insertion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1468			  name_len, name, delayed_node->root->root_key.objectid,
1469			  delayed_node->inode_id, ret);
 
 
 
1470		BUG();
1471	}
1472	mutex_unlock(&delayed_node->mutex);
1473
1474release_node:
1475	btrfs_release_delayed_node(delayed_node);
1476	return ret;
1477}
1478
1479static int btrfs_delete_delayed_insertion_item(struct btrfs_fs_info *fs_info,
1480					       struct btrfs_delayed_node *node,
1481					       struct btrfs_key *key)
1482{
1483	struct btrfs_delayed_item *item;
1484
1485	mutex_lock(&node->mutex);
1486	item = __btrfs_lookup_delayed_insertion_item(node, key);
1487	if (!item) {
1488		mutex_unlock(&node->mutex);
1489		return 1;
1490	}
1491
1492	btrfs_delayed_item_release_metadata(node->root, item);
1493	btrfs_release_delayed_item(item);
1494	mutex_unlock(&node->mutex);
1495	return 0;
1496}
1497
1498int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1499				   struct btrfs_inode *dir, u64 index)
 
1500{
1501	struct btrfs_delayed_node *node;
1502	struct btrfs_delayed_item *item;
1503	struct btrfs_key item_key;
1504	int ret;
1505
1506	node = btrfs_get_or_create_delayed_node(dir);
1507	if (IS_ERR(node))
1508		return PTR_ERR(node);
1509
1510	item_key.objectid = btrfs_ino(dir);
1511	item_key.type = BTRFS_DIR_INDEX_KEY;
1512	item_key.offset = index;
1513
1514	ret = btrfs_delete_delayed_insertion_item(trans->fs_info, node,
1515						  &item_key);
1516	if (!ret)
1517		goto end;
1518
1519	item = btrfs_alloc_delayed_item(0);
1520	if (!item) {
1521		ret = -ENOMEM;
1522		goto end;
1523	}
1524
1525	item->key = item_key;
1526
1527	ret = btrfs_delayed_item_reserve_metadata(trans, dir->root, item);
1528	/*
1529	 * we have reserved enough space when we start a new transaction,
1530	 * so reserving metadata failure is impossible.
1531	 */
1532	if (ret < 0) {
1533		btrfs_err(trans->fs_info,
1534"metadata reservation failed for delayed dir item deltiona, should have been reserved");
1535		btrfs_release_delayed_item(item);
1536		goto end;
1537	}
1538
1539	mutex_lock(&node->mutex);
1540	ret = __btrfs_add_delayed_deletion_item(node, item);
1541	if (unlikely(ret)) {
1542		btrfs_err(trans->fs_info,
1543			  "err add delayed dir index item(index: %llu) into the deletion tree of the delayed node(root id: %llu, inode id: %llu, errno: %d)",
1544			  index, node->root->root_key.objectid,
1545			  node->inode_id, ret);
1546		btrfs_delayed_item_release_metadata(dir->root, item);
1547		btrfs_release_delayed_item(item);
 
 
1548	}
1549	mutex_unlock(&node->mutex);
1550end:
1551	btrfs_release_delayed_node(node);
1552	return ret;
1553}
1554
1555int btrfs_inode_delayed_dir_index_count(struct btrfs_inode *inode)
1556{
1557	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1558
1559	if (!delayed_node)
1560		return -ENOENT;
1561
1562	/*
1563	 * Since we have held i_mutex of this directory, it is impossible that
1564	 * a new directory index is added into the delayed node and index_cnt
1565	 * is updated now. So we needn't lock the delayed node.
1566	 */
1567	if (!delayed_node->index_cnt) {
1568		btrfs_release_delayed_node(delayed_node);
1569		return -EINVAL;
1570	}
1571
1572	inode->index_cnt = delayed_node->index_cnt;
1573	btrfs_release_delayed_node(delayed_node);
1574	return 0;
1575}
1576
1577bool btrfs_readdir_get_delayed_items(struct inode *inode,
1578				     struct list_head *ins_list,
1579				     struct list_head *del_list)
1580{
1581	struct btrfs_delayed_node *delayed_node;
1582	struct btrfs_delayed_item *item;
1583
1584	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1585	if (!delayed_node)
1586		return false;
1587
1588	/*
1589	 * We can only do one readdir with delayed items at a time because of
1590	 * item->readdir_list.
1591	 */
1592	inode_unlock_shared(inode);
1593	inode_lock(inode);
1594
1595	mutex_lock(&delayed_node->mutex);
1596	item = __btrfs_first_delayed_insertion_item(delayed_node);
1597	while (item) {
1598		refcount_inc(&item->refs);
1599		list_add_tail(&item->readdir_list, ins_list);
1600		item = __btrfs_next_delayed_item(item);
1601	}
1602
1603	item = __btrfs_first_delayed_deletion_item(delayed_node);
1604	while (item) {
1605		refcount_inc(&item->refs);
1606		list_add_tail(&item->readdir_list, del_list);
1607		item = __btrfs_next_delayed_item(item);
1608	}
1609	mutex_unlock(&delayed_node->mutex);
1610	/*
1611	 * This delayed node is still cached in the btrfs inode, so refs
1612	 * must be > 1 now, and we needn't check it is going to be freed
1613	 * or not.
1614	 *
1615	 * Besides that, this function is used to read dir, we do not
1616	 * insert/delete delayed items in this period. So we also needn't
1617	 * requeue or dequeue this delayed node.
1618	 */
1619	refcount_dec(&delayed_node->refs);
1620
1621	return true;
1622}
1623
1624void btrfs_readdir_put_delayed_items(struct inode *inode,
1625				     struct list_head *ins_list,
1626				     struct list_head *del_list)
1627{
1628	struct btrfs_delayed_item *curr, *next;
1629
1630	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1631		list_del(&curr->readdir_list);
1632		if (refcount_dec_and_test(&curr->refs))
1633			kfree(curr);
1634	}
1635
1636	list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1637		list_del(&curr->readdir_list);
1638		if (refcount_dec_and_test(&curr->refs))
1639			kfree(curr);
1640	}
1641
1642	/*
1643	 * The VFS is going to do up_read(), so we need to downgrade back to a
1644	 * read lock.
1645	 */
1646	downgrade_write(&inode->i_rwsem);
1647}
1648
1649int btrfs_should_delete_dir_index(struct list_head *del_list,
1650				  u64 index)
1651{
1652	struct btrfs_delayed_item *curr;
1653	int ret = 0;
 
 
 
1654
1655	list_for_each_entry(curr, del_list, readdir_list) {
1656		if (curr->key.offset > index)
1657			break;
1658		if (curr->key.offset == index) {
1659			ret = 1;
1660			break;
1661		}
 
 
 
 
 
 
 
1662	}
1663	return ret;
1664}
1665
1666/*
1667 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1668 *
1669 */
1670int btrfs_readdir_delayed_dir_index(struct dir_context *ctx,
 
1671				    struct list_head *ins_list)
1672{
1673	struct btrfs_dir_item *di;
1674	struct btrfs_delayed_item *curr, *next;
1675	struct btrfs_key location;
1676	char *name;
1677	int name_len;
1678	int over = 0;
1679	unsigned char d_type;
1680
1681	if (list_empty(ins_list))
1682		return 0;
1683
1684	/*
1685	 * Changing the data of the delayed item is impossible. So
1686	 * we needn't lock them. And we have held i_mutex of the
1687	 * directory, nobody can delete any directory indexes now.
1688	 */
1689	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1690		list_del(&curr->readdir_list);
1691
1692		if (curr->key.offset < ctx->pos) {
1693			if (refcount_dec_and_test(&curr->refs))
1694				kfree(curr);
1695			continue;
1696		}
1697
1698		ctx->pos = curr->key.offset;
1699
1700		di = (struct btrfs_dir_item *)curr->data;
1701		name = (char *)(di + 1);
1702		name_len = btrfs_stack_dir_name_len(di);
1703
1704		d_type = fs_ftype_to_dtype(di->type);
1705		btrfs_disk_key_to_cpu(&location, &di->location);
1706
1707		over = !dir_emit(ctx, name, name_len,
1708			       location.objectid, d_type);
1709
1710		if (refcount_dec_and_test(&curr->refs))
1711			kfree(curr);
1712
1713		if (over)
1714			return 1;
1715		ctx->pos++;
1716	}
1717	return 0;
1718}
1719
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1720static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1721				  struct btrfs_inode_item *inode_item,
1722				  struct inode *inode)
1723{
1724	btrfs_set_stack_inode_uid(inode_item, i_uid_read(inode));
1725	btrfs_set_stack_inode_gid(inode_item, i_gid_read(inode));
1726	btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1727	btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1728	btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1729	btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1730	btrfs_set_stack_inode_generation(inode_item,
1731					 BTRFS_I(inode)->generation);
1732	btrfs_set_stack_inode_sequence(inode_item,
1733				       inode_peek_iversion(inode));
1734	btrfs_set_stack_inode_transid(inode_item, trans->transid);
1735	btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1736	btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1737	btrfs_set_stack_inode_block_group(inode_item, 0);
1738
1739	btrfs_set_stack_timespec_sec(&inode_item->atime,
1740				     inode->i_atime.tv_sec);
1741	btrfs_set_stack_timespec_nsec(&inode_item->atime,
1742				      inode->i_atime.tv_nsec);
1743
1744	btrfs_set_stack_timespec_sec(&inode_item->mtime,
1745				     inode->i_mtime.tv_sec);
1746	btrfs_set_stack_timespec_nsec(&inode_item->mtime,
1747				      inode->i_mtime.tv_nsec);
1748
1749	btrfs_set_stack_timespec_sec(&inode_item->ctime,
1750				     inode->i_ctime.tv_sec);
1751	btrfs_set_stack_timespec_nsec(&inode_item->ctime,
1752				      inode->i_ctime.tv_nsec);
1753
1754	btrfs_set_stack_timespec_sec(&inode_item->otime,
1755				     BTRFS_I(inode)->i_otime.tv_sec);
1756	btrfs_set_stack_timespec_nsec(&inode_item->otime,
1757				     BTRFS_I(inode)->i_otime.tv_nsec);
1758}
1759
1760int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1761{
1762	struct btrfs_delayed_node *delayed_node;
1763	struct btrfs_inode_item *inode_item;
 
1764
1765	delayed_node = btrfs_get_delayed_node(BTRFS_I(inode));
1766	if (!delayed_node)
1767		return -ENOENT;
1768
1769	mutex_lock(&delayed_node->mutex);
1770	if (!test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1771		mutex_unlock(&delayed_node->mutex);
1772		btrfs_release_delayed_node(delayed_node);
1773		return -ENOENT;
1774	}
1775
1776	inode_item = &delayed_node->inode_item;
1777
1778	i_uid_write(inode, btrfs_stack_inode_uid(inode_item));
1779	i_gid_write(inode, btrfs_stack_inode_gid(inode_item));
1780	btrfs_i_size_write(BTRFS_I(inode), btrfs_stack_inode_size(inode_item));
1781	inode->i_mode = btrfs_stack_inode_mode(inode_item);
1782	set_nlink(inode, btrfs_stack_inode_nlink(inode_item));
1783	inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1784	BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1785        BTRFS_I(inode)->last_trans = btrfs_stack_inode_transid(inode_item);
1786
1787	inode_set_iversion_queried(inode,
1788				   btrfs_stack_inode_sequence(inode_item));
1789	inode->i_rdev = 0;
1790	*rdev = btrfs_stack_inode_rdev(inode_item);
1791	BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1792
1793	inode->i_atime.tv_sec = btrfs_stack_timespec_sec(&inode_item->atime);
1794	inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->atime);
1795
1796	inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(&inode_item->mtime);
1797	inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->mtime);
1798
1799	inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(&inode_item->ctime);
1800	inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(&inode_item->ctime);
1801
1802	BTRFS_I(inode)->i_otime.tv_sec =
1803		btrfs_stack_timespec_sec(&inode_item->otime);
1804	BTRFS_I(inode)->i_otime.tv_nsec =
1805		btrfs_stack_timespec_nsec(&inode_item->otime);
1806
1807	inode->i_generation = BTRFS_I(inode)->generation;
1808	BTRFS_I(inode)->index_cnt = (u64)-1;
1809
1810	mutex_unlock(&delayed_node->mutex);
1811	btrfs_release_delayed_node(delayed_node);
1812	return 0;
1813}
1814
1815int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1816			       struct btrfs_root *root, struct inode *inode)
1817{
1818	struct btrfs_delayed_node *delayed_node;
1819	int ret = 0;
1820
1821	delayed_node = btrfs_get_or_create_delayed_node(BTRFS_I(inode));
1822	if (IS_ERR(delayed_node))
1823		return PTR_ERR(delayed_node);
1824
1825	mutex_lock(&delayed_node->mutex);
1826	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1827		fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1828		goto release_node;
1829	}
1830
1831	ret = btrfs_delayed_inode_reserve_metadata(trans, root, BTRFS_I(inode),
1832						   delayed_node);
1833	if (ret)
1834		goto release_node;
 
 
1835
1836	fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1837	set_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags);
1838	delayed_node->count++;
1839	atomic_inc(&root->fs_info->delayed_root->items);
1840release_node:
1841	mutex_unlock(&delayed_node->mutex);
1842	btrfs_release_delayed_node(delayed_node);
1843	return ret;
1844}
1845
1846int btrfs_delayed_delete_inode_ref(struct btrfs_inode *inode)
1847{
1848	struct btrfs_fs_info *fs_info = inode->root->fs_info;
1849	struct btrfs_delayed_node *delayed_node;
1850
1851	/*
1852	 * we don't do delayed inode updates during log recovery because it
1853	 * leads to enospc problems.  This means we also can't do
1854	 * delayed inode refs
1855	 */
1856	if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags))
1857		return -EAGAIN;
1858
1859	delayed_node = btrfs_get_or_create_delayed_node(inode);
1860	if (IS_ERR(delayed_node))
1861		return PTR_ERR(delayed_node);
1862
1863	/*
1864	 * We don't reserve space for inode ref deletion is because:
1865	 * - We ONLY do async inode ref deletion for the inode who has only
1866	 *   one link(i_nlink == 1), it means there is only one inode ref.
1867	 *   And in most case, the inode ref and the inode item are in the
1868	 *   same leaf, and we will deal with them at the same time.
1869	 *   Since we are sure we will reserve the space for the inode item,
1870	 *   it is unnecessary to reserve space for inode ref deletion.
1871	 * - If the inode ref and the inode item are not in the same leaf,
1872	 *   We also needn't worry about enospc problem, because we reserve
1873	 *   much more space for the inode update than it needs.
1874	 * - At the worst, we can steal some space from the global reservation.
1875	 *   It is very rare.
1876	 */
1877	mutex_lock(&delayed_node->mutex);
1878	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1879		goto release_node;
1880
1881	set_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags);
1882	delayed_node->count++;
1883	atomic_inc(&fs_info->delayed_root->items);
1884release_node:
1885	mutex_unlock(&delayed_node->mutex);
1886	btrfs_release_delayed_node(delayed_node);
1887	return 0;
1888}
1889
1890static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1891{
1892	struct btrfs_root *root = delayed_node->root;
1893	struct btrfs_fs_info *fs_info = root->fs_info;
1894	struct btrfs_delayed_item *curr_item, *prev_item;
1895
1896	mutex_lock(&delayed_node->mutex);
1897	curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1898	while (curr_item) {
1899		btrfs_delayed_item_release_metadata(root, curr_item);
1900		prev_item = curr_item;
1901		curr_item = __btrfs_next_delayed_item(prev_item);
1902		btrfs_release_delayed_item(prev_item);
1903	}
1904
1905	curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1906	while (curr_item) {
1907		btrfs_delayed_item_release_metadata(root, curr_item);
1908		prev_item = curr_item;
1909		curr_item = __btrfs_next_delayed_item(prev_item);
1910		btrfs_release_delayed_item(prev_item);
1911	}
1912
1913	if (test_bit(BTRFS_DELAYED_NODE_DEL_IREF, &delayed_node->flags))
1914		btrfs_release_delayed_iref(delayed_node);
1915
1916	if (test_bit(BTRFS_DELAYED_NODE_INODE_DIRTY, &delayed_node->flags)) {
1917		btrfs_delayed_inode_release_metadata(fs_info, delayed_node, false);
1918		btrfs_release_delayed_inode(delayed_node);
1919	}
1920	mutex_unlock(&delayed_node->mutex);
1921}
1922
1923void btrfs_kill_delayed_inode_items(struct btrfs_inode *inode)
1924{
1925	struct btrfs_delayed_node *delayed_node;
1926
1927	delayed_node = btrfs_get_delayed_node(inode);
1928	if (!delayed_node)
1929		return;
1930
1931	__btrfs_kill_delayed_node(delayed_node);
1932	btrfs_release_delayed_node(delayed_node);
1933}
1934
1935void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1936{
1937	u64 inode_id = 0;
1938	struct btrfs_delayed_node *delayed_nodes[8];
1939	int i, n;
1940
1941	while (1) {
1942		spin_lock(&root->inode_lock);
1943		n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1944					   (void **)delayed_nodes, inode_id,
1945					   ARRAY_SIZE(delayed_nodes));
1946		if (!n) {
1947			spin_unlock(&root->inode_lock);
1948			break;
1949		}
1950
1951		inode_id = delayed_nodes[n - 1]->inode_id + 1;
1952
1953		for (i = 0; i < n; i++)
1954			refcount_inc(&delayed_nodes[i]->refs);
1955		spin_unlock(&root->inode_lock);
1956
1957		for (i = 0; i < n; i++) {
1958			__btrfs_kill_delayed_node(delayed_nodes[i]);
1959			btrfs_release_delayed_node(delayed_nodes[i]);
1960		}
1961	}
1962}
1963
1964void btrfs_destroy_delayed_inodes(struct btrfs_fs_info *fs_info)
1965{
1966	struct btrfs_delayed_node *curr_node, *prev_node;
1967
1968	curr_node = btrfs_first_delayed_node(fs_info->delayed_root);
1969	while (curr_node) {
1970		__btrfs_kill_delayed_node(curr_node);
1971
1972		prev_node = curr_node;
1973		curr_node = btrfs_next_delayed_node(curr_node);
1974		btrfs_release_delayed_node(prev_node);
1975	}
1976}
1977
v3.1
 
   1/*
   2 * Copyright (C) 2011 Fujitsu.  All rights reserved.
   3 * Written by Miao Xie <miaox@cn.fujitsu.com>
   4 *
   5 * This program is free software; you can redistribute it and/or
   6 * modify it under the terms of the GNU General Public
   7 * License v2 as published by the Free Software Foundation.
   8 *
   9 * This program is distributed in the hope that it will be useful,
  10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  12 * General Public License for more details.
  13 *
  14 * You should have received a copy of the GNU General Public
  15 * License along with this program; if not, write to the
  16 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  17 * Boston, MA 021110-1307, USA.
  18 */
  19
  20#include <linux/slab.h>
 
 
  21#include "delayed-inode.h"
  22#include "disk-io.h"
  23#include "transaction.h"
 
 
  24
  25#define BTRFS_DELAYED_WRITEBACK		400
  26#define BTRFS_DELAYED_BACKGROUND	100
 
  27
  28static struct kmem_cache *delayed_node_cache;
  29
  30int __init btrfs_delayed_inode_init(void)
  31{
  32	delayed_node_cache = kmem_cache_create("delayed_node",
  33					sizeof(struct btrfs_delayed_node),
  34					0,
  35					SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
  36					NULL);
  37	if (!delayed_node_cache)
  38		return -ENOMEM;
  39	return 0;
  40}
  41
  42void btrfs_delayed_inode_exit(void)
  43{
  44	if (delayed_node_cache)
  45		kmem_cache_destroy(delayed_node_cache);
  46}
  47
  48static inline void btrfs_init_delayed_node(
  49				struct btrfs_delayed_node *delayed_node,
  50				struct btrfs_root *root, u64 inode_id)
  51{
  52	delayed_node->root = root;
  53	delayed_node->inode_id = inode_id;
  54	atomic_set(&delayed_node->refs, 0);
  55	delayed_node->count = 0;
  56	delayed_node->in_list = 0;
  57	delayed_node->inode_dirty = 0;
  58	delayed_node->ins_root = RB_ROOT;
  59	delayed_node->del_root = RB_ROOT;
  60	mutex_init(&delayed_node->mutex);
  61	delayed_node->index_cnt = 0;
  62	INIT_LIST_HEAD(&delayed_node->n_list);
  63	INIT_LIST_HEAD(&delayed_node->p_list);
  64	delayed_node->bytes_reserved = 0;
  65}
  66
  67static inline int btrfs_is_continuous_delayed_item(
  68					struct btrfs_delayed_item *item1,
  69					struct btrfs_delayed_item *item2)
  70{
  71	if (item1->key.type == BTRFS_DIR_INDEX_KEY &&
  72	    item1->key.objectid == item2->key.objectid &&
  73	    item1->key.type == item2->key.type &&
  74	    item1->key.offset + 1 == item2->key.offset)
  75		return 1;
  76	return 0;
  77}
  78
  79static inline struct btrfs_delayed_root *btrfs_get_delayed_root(
  80							struct btrfs_root *root)
  81{
  82	return root->fs_info->delayed_root;
  83}
  84
  85static struct btrfs_delayed_node *btrfs_get_delayed_node(struct inode *inode)
  86{
  87	struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
  88	struct btrfs_root *root = btrfs_inode->root;
  89	u64 ino = btrfs_ino(inode);
  90	struct btrfs_delayed_node *node;
  91
  92	node = ACCESS_ONCE(btrfs_inode->delayed_node);
  93	if (node) {
  94		atomic_inc(&node->refs);
  95		return node;
  96	}
  97
  98	spin_lock(&root->inode_lock);
  99	node = radix_tree_lookup(&root->delayed_nodes_tree, ino);
 
 100	if (node) {
 101		if (btrfs_inode->delayed_node) {
 102			atomic_inc(&node->refs);	/* can be accessed */
 103			BUG_ON(btrfs_inode->delayed_node != node);
 104			spin_unlock(&root->inode_lock);
 105			return node;
 106		}
 107		btrfs_inode->delayed_node = node;
 108		atomic_inc(&node->refs);	/* can be accessed */
 109		atomic_inc(&node->refs);	/* cached in the inode */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 110		spin_unlock(&root->inode_lock);
 111		return node;
 112	}
 113	spin_unlock(&root->inode_lock);
 114
 115	return NULL;
 116}
 117
 
 118static struct btrfs_delayed_node *btrfs_get_or_create_delayed_node(
 119							struct inode *inode)
 120{
 121	struct btrfs_delayed_node *node;
 122	struct btrfs_inode *btrfs_inode = BTRFS_I(inode);
 123	struct btrfs_root *root = btrfs_inode->root;
 124	u64 ino = btrfs_ino(inode);
 125	int ret;
 126
 127again:
 128	node = btrfs_get_delayed_node(inode);
 129	if (node)
 130		return node;
 131
 132	node = kmem_cache_alloc(delayed_node_cache, GFP_NOFS);
 133	if (!node)
 134		return ERR_PTR(-ENOMEM);
 135	btrfs_init_delayed_node(node, root, ino);
 136
 137	atomic_inc(&node->refs);	/* cached in the btrfs inode */
 138	atomic_inc(&node->refs);	/* can be accessed */
 139
 140	ret = radix_tree_preload(GFP_NOFS & ~__GFP_HIGHMEM);
 141	if (ret) {
 142		kmem_cache_free(delayed_node_cache, node);
 143		return ERR_PTR(ret);
 144	}
 145
 146	spin_lock(&root->inode_lock);
 147	ret = radix_tree_insert(&root->delayed_nodes_tree, ino, node);
 148	if (ret == -EEXIST) {
 
 149		kmem_cache_free(delayed_node_cache, node);
 150		spin_unlock(&root->inode_lock);
 151		radix_tree_preload_end();
 152		goto again;
 153	}
 154	btrfs_inode->delayed_node = node;
 155	spin_unlock(&root->inode_lock);
 156	radix_tree_preload_end();
 157
 158	return node;
 159}
 160
 161/*
 162 * Call it when holding delayed_node->mutex
 163 *
 164 * If mod = 1, add this node into the prepared list.
 165 */
 166static void btrfs_queue_delayed_node(struct btrfs_delayed_root *root,
 167				     struct btrfs_delayed_node *node,
 168				     int mod)
 169{
 170	spin_lock(&root->lock);
 171	if (node->in_list) {
 172		if (!list_empty(&node->p_list))
 173			list_move_tail(&node->p_list, &root->prepare_list);
 174		else if (mod)
 175			list_add_tail(&node->p_list, &root->prepare_list);
 176	} else {
 177		list_add_tail(&node->n_list, &root->node_list);
 178		list_add_tail(&node->p_list, &root->prepare_list);
 179		atomic_inc(&node->refs);	/* inserted into list */
 180		root->nodes++;
 181		node->in_list = 1;
 182	}
 183	spin_unlock(&root->lock);
 184}
 185
 186/* Call it when holding delayed_node->mutex */
 187static void btrfs_dequeue_delayed_node(struct btrfs_delayed_root *root,
 188				       struct btrfs_delayed_node *node)
 189{
 190	spin_lock(&root->lock);
 191	if (node->in_list) {
 192		root->nodes--;
 193		atomic_dec(&node->refs);	/* not in the list */
 194		list_del_init(&node->n_list);
 195		if (!list_empty(&node->p_list))
 196			list_del_init(&node->p_list);
 197		node->in_list = 0;
 198	}
 199	spin_unlock(&root->lock);
 200}
 201
 202struct btrfs_delayed_node *btrfs_first_delayed_node(
 203			struct btrfs_delayed_root *delayed_root)
 204{
 205	struct list_head *p;
 206	struct btrfs_delayed_node *node = NULL;
 207
 208	spin_lock(&delayed_root->lock);
 209	if (list_empty(&delayed_root->node_list))
 210		goto out;
 211
 212	p = delayed_root->node_list.next;
 213	node = list_entry(p, struct btrfs_delayed_node, n_list);
 214	atomic_inc(&node->refs);
 215out:
 216	spin_unlock(&delayed_root->lock);
 217
 218	return node;
 219}
 220
 221struct btrfs_delayed_node *btrfs_next_delayed_node(
 222						struct btrfs_delayed_node *node)
 223{
 224	struct btrfs_delayed_root *delayed_root;
 225	struct list_head *p;
 226	struct btrfs_delayed_node *next = NULL;
 227
 228	delayed_root = node->root->fs_info->delayed_root;
 229	spin_lock(&delayed_root->lock);
 230	if (!node->in_list) {	/* not in the list */
 
 231		if (list_empty(&delayed_root->node_list))
 232			goto out;
 233		p = delayed_root->node_list.next;
 234	} else if (list_is_last(&node->n_list, &delayed_root->node_list))
 235		goto out;
 236	else
 237		p = node->n_list.next;
 238
 239	next = list_entry(p, struct btrfs_delayed_node, n_list);
 240	atomic_inc(&next->refs);
 241out:
 242	spin_unlock(&delayed_root->lock);
 243
 244	return next;
 245}
 246
 247static void __btrfs_release_delayed_node(
 248				struct btrfs_delayed_node *delayed_node,
 249				int mod)
 250{
 251	struct btrfs_delayed_root *delayed_root;
 252
 253	if (!delayed_node)
 254		return;
 255
 256	delayed_root = delayed_node->root->fs_info->delayed_root;
 257
 258	mutex_lock(&delayed_node->mutex);
 259	if (delayed_node->count)
 260		btrfs_queue_delayed_node(delayed_root, delayed_node, mod);
 261	else
 262		btrfs_dequeue_delayed_node(delayed_root, delayed_node);
 263	mutex_unlock(&delayed_node->mutex);
 264
 265	if (atomic_dec_and_test(&delayed_node->refs)) {
 266		struct btrfs_root *root = delayed_node->root;
 
 267		spin_lock(&root->inode_lock);
 268		if (atomic_read(&delayed_node->refs) == 0) {
 269			radix_tree_delete(&root->delayed_nodes_tree,
 270					  delayed_node->inode_id);
 271			kmem_cache_free(delayed_node_cache, delayed_node);
 272		}
 
 
 273		spin_unlock(&root->inode_lock);
 
 274	}
 275}
 276
 277static inline void btrfs_release_delayed_node(struct btrfs_delayed_node *node)
 278{
 279	__btrfs_release_delayed_node(node, 0);
 280}
 281
 282struct btrfs_delayed_node *btrfs_first_prepared_delayed_node(
 283					struct btrfs_delayed_root *delayed_root)
 284{
 285	struct list_head *p;
 286	struct btrfs_delayed_node *node = NULL;
 287
 288	spin_lock(&delayed_root->lock);
 289	if (list_empty(&delayed_root->prepare_list))
 290		goto out;
 291
 292	p = delayed_root->prepare_list.next;
 293	list_del_init(p);
 294	node = list_entry(p, struct btrfs_delayed_node, p_list);
 295	atomic_inc(&node->refs);
 296out:
 297	spin_unlock(&delayed_root->lock);
 298
 299	return node;
 300}
 301
 302static inline void btrfs_release_prepared_delayed_node(
 303					struct btrfs_delayed_node *node)
 304{
 305	__btrfs_release_delayed_node(node, 1);
 306}
 307
 308struct btrfs_delayed_item *btrfs_alloc_delayed_item(u32 data_len)
 309{
 310	struct btrfs_delayed_item *item;
 311	item = kmalloc(sizeof(*item) + data_len, GFP_NOFS);
 312	if (item) {
 313		item->data_len = data_len;
 314		item->ins_or_del = 0;
 315		item->bytes_reserved = 0;
 316		item->delayed_node = NULL;
 317		atomic_set(&item->refs, 1);
 318	}
 319	return item;
 320}
 321
 322/*
 323 * __btrfs_lookup_delayed_item - look up the delayed item by key
 324 * @delayed_node: pointer to the delayed node
 325 * @key:	  the key to look up
 326 * @prev:	  used to store the prev item if the right item isn't found
 327 * @next:	  used to store the next item if the right item isn't found
 328 *
 329 * Note: if we don't find the right item, we will return the prev item and
 330 * the next item.
 331 */
 332static struct btrfs_delayed_item *__btrfs_lookup_delayed_item(
 333				struct rb_root *root,
 334				struct btrfs_key *key,
 335				struct btrfs_delayed_item **prev,
 336				struct btrfs_delayed_item **next)
 337{
 338	struct rb_node *node, *prev_node = NULL;
 339	struct btrfs_delayed_item *delayed_item = NULL;
 340	int ret = 0;
 341
 342	node = root->rb_node;
 343
 344	while (node) {
 345		delayed_item = rb_entry(node, struct btrfs_delayed_item,
 346					rb_node);
 347		prev_node = node;
 348		ret = btrfs_comp_cpu_keys(&delayed_item->key, key);
 349		if (ret < 0)
 350			node = node->rb_right;
 351		else if (ret > 0)
 352			node = node->rb_left;
 353		else
 354			return delayed_item;
 355	}
 356
 357	if (prev) {
 358		if (!prev_node)
 359			*prev = NULL;
 360		else if (ret < 0)
 361			*prev = delayed_item;
 362		else if ((node = rb_prev(prev_node)) != NULL) {
 363			*prev = rb_entry(node, struct btrfs_delayed_item,
 364					 rb_node);
 365		} else
 366			*prev = NULL;
 367	}
 368
 369	if (next) {
 370		if (!prev_node)
 371			*next = NULL;
 372		else if (ret > 0)
 373			*next = delayed_item;
 374		else if ((node = rb_next(prev_node)) != NULL) {
 375			*next = rb_entry(node, struct btrfs_delayed_item,
 376					 rb_node);
 377		} else
 378			*next = NULL;
 379	}
 380	return NULL;
 381}
 382
 383struct btrfs_delayed_item *__btrfs_lookup_delayed_insertion_item(
 384					struct btrfs_delayed_node *delayed_node,
 385					struct btrfs_key *key)
 386{
 387	struct btrfs_delayed_item *item;
 388
 389	item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 390					   NULL, NULL);
 391	return item;
 392}
 393
 394struct btrfs_delayed_item *__btrfs_lookup_delayed_deletion_item(
 395					struct btrfs_delayed_node *delayed_node,
 396					struct btrfs_key *key)
 397{
 398	struct btrfs_delayed_item *item;
 399
 400	item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
 401					   NULL, NULL);
 402	return item;
 403}
 404
 405struct btrfs_delayed_item *__btrfs_search_delayed_insertion_item(
 406					struct btrfs_delayed_node *delayed_node,
 407					struct btrfs_key *key)
 408{
 409	struct btrfs_delayed_item *item, *next;
 410
 411	item = __btrfs_lookup_delayed_item(&delayed_node->ins_root, key,
 412					   NULL, &next);
 413	if (!item)
 414		item = next;
 415
 416	return item;
 417}
 418
 419struct btrfs_delayed_item *__btrfs_search_delayed_deletion_item(
 420					struct btrfs_delayed_node *delayed_node,
 421					struct btrfs_key *key)
 422{
 423	struct btrfs_delayed_item *item, *next;
 424
 425	item = __btrfs_lookup_delayed_item(&delayed_node->del_root, key,
 426					   NULL, &next);
 427	if (!item)
 428		item = next;
 429
 430	return item;
 431}
 432
 433static int __btrfs_add_delayed_item(struct btrfs_delayed_node *delayed_node,
 434				    struct btrfs_delayed_item *ins,
 435				    int action)
 436{
 437	struct rb_node **p, *node;
 438	struct rb_node *parent_node = NULL;
 439	struct rb_root *root;
 440	struct btrfs_delayed_item *item;
 441	int cmp;
 
 442
 443	if (action == BTRFS_DELAYED_INSERTION_ITEM)
 444		root = &delayed_node->ins_root;
 445	else if (action == BTRFS_DELAYED_DELETION_ITEM)
 446		root = &delayed_node->del_root;
 447	else
 448		BUG();
 449	p = &root->rb_node;
 450	node = &ins->rb_node;
 451
 452	while (*p) {
 453		parent_node = *p;
 454		item = rb_entry(parent_node, struct btrfs_delayed_item,
 455				 rb_node);
 456
 457		cmp = btrfs_comp_cpu_keys(&item->key, &ins->key);
 458		if (cmp < 0)
 459			p = &(*p)->rb_right;
 460		else if (cmp > 0)
 
 461			p = &(*p)->rb_left;
 462		else
 463			return -EEXIST;
 
 464	}
 465
 466	rb_link_node(node, parent_node, p);
 467	rb_insert_color(node, root);
 468	ins->delayed_node = delayed_node;
 469	ins->ins_or_del = action;
 470
 471	if (ins->key.type == BTRFS_DIR_INDEX_KEY &&
 472	    action == BTRFS_DELAYED_INSERTION_ITEM &&
 473	    ins->key.offset >= delayed_node->index_cnt)
 474			delayed_node->index_cnt = ins->key.offset + 1;
 475
 476	delayed_node->count++;
 477	atomic_inc(&delayed_node->root->fs_info->delayed_root->items);
 478	return 0;
 479}
 480
 481static int __btrfs_add_delayed_insertion_item(struct btrfs_delayed_node *node,
 482					      struct btrfs_delayed_item *item)
 483{
 484	return __btrfs_add_delayed_item(node, item,
 485					BTRFS_DELAYED_INSERTION_ITEM);
 486}
 487
 488static int __btrfs_add_delayed_deletion_item(struct btrfs_delayed_node *node,
 489					     struct btrfs_delayed_item *item)
 490{
 491	return __btrfs_add_delayed_item(node, item,
 492					BTRFS_DELAYED_DELETION_ITEM);
 493}
 494
 
 
 
 
 
 
 
 
 
 
 495static void __btrfs_remove_delayed_item(struct btrfs_delayed_item *delayed_item)
 496{
 497	struct rb_root *root;
 498	struct btrfs_delayed_root *delayed_root;
 499
 
 
 
 500	delayed_root = delayed_item->delayed_node->root->fs_info->delayed_root;
 501
 502	BUG_ON(!delayed_root);
 503	BUG_ON(delayed_item->ins_or_del != BTRFS_DELAYED_DELETION_ITEM &&
 504	       delayed_item->ins_or_del != BTRFS_DELAYED_INSERTION_ITEM);
 505
 506	if (delayed_item->ins_or_del == BTRFS_DELAYED_INSERTION_ITEM)
 507		root = &delayed_item->delayed_node->ins_root;
 508	else
 509		root = &delayed_item->delayed_node->del_root;
 510
 511	rb_erase(&delayed_item->rb_node, root);
 512	delayed_item->delayed_node->count--;
 513	atomic_dec(&delayed_root->items);
 514	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND &&
 515	    waitqueue_active(&delayed_root->wait))
 516		wake_up(&delayed_root->wait);
 517}
 518
 519static void btrfs_release_delayed_item(struct btrfs_delayed_item *item)
 520{
 521	if (item) {
 522		__btrfs_remove_delayed_item(item);
 523		if (atomic_dec_and_test(&item->refs))
 524			kfree(item);
 525	}
 526}
 527
 528struct btrfs_delayed_item *__btrfs_first_delayed_insertion_item(
 529					struct btrfs_delayed_node *delayed_node)
 530{
 531	struct rb_node *p;
 532	struct btrfs_delayed_item *item = NULL;
 533
 534	p = rb_first(&delayed_node->ins_root);
 535	if (p)
 536		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 537
 538	return item;
 539}
 540
 541struct btrfs_delayed_item *__btrfs_first_delayed_deletion_item(
 542					struct btrfs_delayed_node *delayed_node)
 543{
 544	struct rb_node *p;
 545	struct btrfs_delayed_item *item = NULL;
 546
 547	p = rb_first(&delayed_node->del_root);
 548	if (p)
 549		item = rb_entry(p, struct btrfs_delayed_item, rb_node);
 550
 551	return item;
 552}
 553
 554struct btrfs_delayed_item *__btrfs_next_delayed_item(
 555						struct btrfs_delayed_item *item)
 556{
 557	struct rb_node *p;
 558	struct btrfs_delayed_item *next = NULL;
 559
 560	p = rb_next(&item->rb_node);
 561	if (p)
 562		next = rb_entry(p, struct btrfs_delayed_item, rb_node);
 563
 564	return next;
 565}
 566
 567static inline struct btrfs_root *btrfs_get_fs_root(struct btrfs_root *root,
 568						   u64 root_id)
 569{
 570	struct btrfs_key root_key;
 571
 572	if (root->objectid == root_id)
 573		return root;
 574
 575	root_key.objectid = root_id;
 576	root_key.type = BTRFS_ROOT_ITEM_KEY;
 577	root_key.offset = (u64)-1;
 578	return btrfs_read_fs_root_no_name(root->fs_info, &root_key);
 579}
 580
 581static int btrfs_delayed_item_reserve_metadata(struct btrfs_trans_handle *trans,
 582					       struct btrfs_root *root,
 583					       struct btrfs_delayed_item *item)
 584{
 585	struct btrfs_block_rsv *src_rsv;
 586	struct btrfs_block_rsv *dst_rsv;
 
 587	u64 num_bytes;
 588	int ret;
 589
 590	if (!trans->bytes_reserved)
 591		return 0;
 592
 593	src_rsv = trans->block_rsv;
 594	dst_rsv = &root->fs_info->global_block_rsv;
 
 
 595
 596	num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 597	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 598	if (!ret)
 
 
 
 
 
 
 
 599		item->bytes_reserved = num_bytes;
 
 600
 601	return ret;
 602}
 603
 604static void btrfs_delayed_item_release_metadata(struct btrfs_root *root,
 605						struct btrfs_delayed_item *item)
 606{
 607	struct btrfs_block_rsv *rsv;
 
 608
 609	if (!item->bytes_reserved)
 610		return;
 611
 612	rsv = &root->fs_info->global_block_rsv;
 613	btrfs_block_rsv_release(root, rsv,
 
 
 
 
 
 
 
 614				item->bytes_reserved);
 615}
 616
 617static int btrfs_delayed_inode_reserve_metadata(
 618					struct btrfs_trans_handle *trans,
 619					struct btrfs_root *root,
 
 620					struct btrfs_delayed_node *node)
 621{
 
 622	struct btrfs_block_rsv *src_rsv;
 623	struct btrfs_block_rsv *dst_rsv;
 624	u64 num_bytes;
 625	int ret;
 626
 627	if (!trans->bytes_reserved)
 628		return 0;
 
 
 629
 630	src_rsv = trans->block_rsv;
 631	dst_rsv = &root->fs_info->global_block_rsv;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 632
 633	num_bytes = btrfs_calc_trans_metadata_size(root, 1);
 634	ret = btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes);
 635	if (!ret)
 
 636		node->bytes_reserved = num_bytes;
 
 637
 638	return ret;
 639}
 640
 641static void btrfs_delayed_inode_release_metadata(struct btrfs_root *root,
 642						struct btrfs_delayed_node *node)
 
 643{
 644	struct btrfs_block_rsv *rsv;
 645
 646	if (!node->bytes_reserved)
 647		return;
 648
 649	rsv = &root->fs_info->global_block_rsv;
 650	btrfs_block_rsv_release(root, rsv,
 
 
 
 
 
 
 
 
 651				node->bytes_reserved);
 652	node->bytes_reserved = 0;
 653}
 654
 655/*
 656 * This helper will insert some continuous items into the same leaf according
 657 * to the free space of the leaf.
 658 */
 659static int btrfs_batch_insert_items(struct btrfs_trans_handle *trans,
 660				struct btrfs_root *root,
 661				struct btrfs_path *path,
 662				struct btrfs_delayed_item *item)
 663{
 664	struct btrfs_delayed_item *curr, *next;
 665	int free_space;
 666	int total_data_size = 0, total_size = 0;
 667	struct extent_buffer *leaf;
 668	char *data_ptr;
 669	struct btrfs_key *keys;
 670	u32 *data_size;
 671	struct list_head head;
 672	int slot;
 673	int nitems;
 674	int i;
 675	int ret = 0;
 676
 677	BUG_ON(!path->nodes[0]);
 678
 679	leaf = path->nodes[0];
 680	free_space = btrfs_leaf_free_space(root, leaf);
 681	INIT_LIST_HEAD(&head);
 682
 683	next = item;
 684	nitems = 0;
 685
 686	/*
 687	 * count the number of the continuous items that we can insert in batch
 688	 */
 689	while (total_size + next->data_len + sizeof(struct btrfs_item) <=
 690	       free_space) {
 691		total_data_size += next->data_len;
 692		total_size += next->data_len + sizeof(struct btrfs_item);
 693		list_add_tail(&next->tree_list, &head);
 694		nitems++;
 695
 696		curr = next;
 697		next = __btrfs_next_delayed_item(curr);
 698		if (!next)
 699			break;
 700
 701		if (!btrfs_is_continuous_delayed_item(curr, next))
 702			break;
 703	}
 704
 705	if (!nitems) {
 706		ret = 0;
 707		goto out;
 708	}
 709
 710	/*
 711	 * we need allocate some memory space, but it might cause the task
 712	 * to sleep, so we set all locked nodes in the path to blocking locks
 713	 * first.
 714	 */
 715	btrfs_set_path_blocking(path);
 716
 717	keys = kmalloc(sizeof(struct btrfs_key) * nitems, GFP_NOFS);
 718	if (!keys) {
 719		ret = -ENOMEM;
 720		goto out;
 721	}
 722
 723	data_size = kmalloc(sizeof(u32) * nitems, GFP_NOFS);
 724	if (!data_size) {
 725		ret = -ENOMEM;
 726		goto error;
 727	}
 728
 729	/* get keys of all the delayed items */
 730	i = 0;
 731	list_for_each_entry(next, &head, tree_list) {
 732		keys[i] = next->key;
 733		data_size[i] = next->data_len;
 734		i++;
 735	}
 736
 737	/* reset all the locked nodes in the patch to spinning locks. */
 738	btrfs_clear_path_blocking(path, NULL, 0);
 739
 740	/* insert the keys of the items */
 741	ret = setup_items_for_insert(trans, root, path, keys, data_size,
 742				     total_data_size, total_size, nitems);
 743	if (ret)
 744		goto error;
 745
 746	/* insert the dir index items */
 747	slot = path->slots[0];
 748	list_for_each_entry_safe(curr, next, &head, tree_list) {
 749		data_ptr = btrfs_item_ptr(leaf, slot, char);
 750		write_extent_buffer(leaf, &curr->data,
 751				    (unsigned long)data_ptr,
 752				    curr->data_len);
 753		slot++;
 754
 755		btrfs_delayed_item_release_metadata(root, curr);
 756
 757		list_del(&curr->tree_list);
 758		btrfs_release_delayed_item(curr);
 759	}
 760
 761error:
 762	kfree(data_size);
 763	kfree(keys);
 764out:
 765	return ret;
 766}
 767
 768/*
 769 * This helper can just do simple insertion that needn't extend item for new
 770 * data, such as directory name index insertion, inode insertion.
 771 */
 772static int btrfs_insert_delayed_item(struct btrfs_trans_handle *trans,
 773				     struct btrfs_root *root,
 774				     struct btrfs_path *path,
 775				     struct btrfs_delayed_item *delayed_item)
 776{
 777	struct extent_buffer *leaf;
 778	struct btrfs_item *item;
 779	char *ptr;
 780	int ret;
 781
 782	ret = btrfs_insert_empty_item(trans, root, path, &delayed_item->key,
 783				      delayed_item->data_len);
 784	if (ret < 0 && ret != -EEXIST)
 785		return ret;
 786
 787	leaf = path->nodes[0];
 788
 789	item = btrfs_item_nr(leaf, path->slots[0]);
 790	ptr = btrfs_item_ptr(leaf, path->slots[0], char);
 791
 792	write_extent_buffer(leaf, delayed_item->data, (unsigned long)ptr,
 793			    delayed_item->data_len);
 794	btrfs_mark_buffer_dirty(leaf);
 795
 796	btrfs_delayed_item_release_metadata(root, delayed_item);
 797	return 0;
 798}
 799
 800/*
 801 * we insert an item first, then if there are some continuous items, we try
 802 * to insert those items into the same leaf.
 803 */
 804static int btrfs_insert_delayed_items(struct btrfs_trans_handle *trans,
 805				      struct btrfs_path *path,
 806				      struct btrfs_root *root,
 807				      struct btrfs_delayed_node *node)
 808{
 809	struct btrfs_delayed_item *curr, *prev;
 810	int ret = 0;
 811
 812do_again:
 813	mutex_lock(&node->mutex);
 814	curr = __btrfs_first_delayed_insertion_item(node);
 815	if (!curr)
 816		goto insert_end;
 817
 818	ret = btrfs_insert_delayed_item(trans, root, path, curr);
 819	if (ret < 0) {
 820		btrfs_release_path(path);
 821		goto insert_end;
 822	}
 823
 824	prev = curr;
 825	curr = __btrfs_next_delayed_item(prev);
 826	if (curr && btrfs_is_continuous_delayed_item(prev, curr)) {
 827		/* insert the continuous items into the same leaf */
 828		path->slots[0]++;
 829		btrfs_batch_insert_items(trans, root, path, curr);
 830	}
 831	btrfs_release_delayed_item(prev);
 832	btrfs_mark_buffer_dirty(path->nodes[0]);
 833
 834	btrfs_release_path(path);
 835	mutex_unlock(&node->mutex);
 836	goto do_again;
 837
 838insert_end:
 839	mutex_unlock(&node->mutex);
 840	return ret;
 841}
 842
 843static int btrfs_batch_delete_items(struct btrfs_trans_handle *trans,
 844				    struct btrfs_root *root,
 845				    struct btrfs_path *path,
 846				    struct btrfs_delayed_item *item)
 847{
 848	struct btrfs_delayed_item *curr, *next;
 849	struct extent_buffer *leaf;
 850	struct btrfs_key key;
 851	struct list_head head;
 852	int nitems, i, last_item;
 853	int ret = 0;
 854
 855	BUG_ON(!path->nodes[0]);
 856
 857	leaf = path->nodes[0];
 858
 859	i = path->slots[0];
 860	last_item = btrfs_header_nritems(leaf) - 1;
 861	if (i > last_item)
 862		return -ENOENT;	/* FIXME: Is errno suitable? */
 863
 864	next = item;
 865	INIT_LIST_HEAD(&head);
 866	btrfs_item_key_to_cpu(leaf, &key, i);
 867	nitems = 0;
 868	/*
 869	 * count the number of the dir index items that we can delete in batch
 870	 */
 871	while (btrfs_comp_cpu_keys(&next->key, &key) == 0) {
 872		list_add_tail(&next->tree_list, &head);
 873		nitems++;
 874
 875		curr = next;
 876		next = __btrfs_next_delayed_item(curr);
 877		if (!next)
 878			break;
 879
 880		if (!btrfs_is_continuous_delayed_item(curr, next))
 881			break;
 882
 883		i++;
 884		if (i > last_item)
 885			break;
 886		btrfs_item_key_to_cpu(leaf, &key, i);
 887	}
 888
 889	if (!nitems)
 890		return 0;
 891
 892	ret = btrfs_del_items(trans, root, path, path->slots[0], nitems);
 893	if (ret)
 894		goto out;
 895
 896	list_for_each_entry_safe(curr, next, &head, tree_list) {
 897		btrfs_delayed_item_release_metadata(root, curr);
 898		list_del(&curr->tree_list);
 899		btrfs_release_delayed_item(curr);
 900	}
 901
 902out:
 903	return ret;
 904}
 905
 906static int btrfs_delete_delayed_items(struct btrfs_trans_handle *trans,
 907				      struct btrfs_path *path,
 908				      struct btrfs_root *root,
 909				      struct btrfs_delayed_node *node)
 910{
 911	struct btrfs_delayed_item *curr, *prev;
 912	int ret = 0;
 913
 914do_again:
 915	mutex_lock(&node->mutex);
 916	curr = __btrfs_first_delayed_deletion_item(node);
 917	if (!curr)
 918		goto delete_fail;
 919
 920	ret = btrfs_search_slot(trans, root, &curr->key, path, -1, 1);
 921	if (ret < 0)
 922		goto delete_fail;
 923	else if (ret > 0) {
 924		/*
 925		 * can't find the item which the node points to, so this node
 926		 * is invalid, just drop it.
 927		 */
 928		prev = curr;
 929		curr = __btrfs_next_delayed_item(prev);
 930		btrfs_release_delayed_item(prev);
 931		ret = 0;
 932		btrfs_release_path(path);
 933		if (curr)
 
 934			goto do_again;
 935		else
 936			goto delete_fail;
 937	}
 938
 939	btrfs_batch_delete_items(trans, root, path, curr);
 940	btrfs_release_path(path);
 941	mutex_unlock(&node->mutex);
 942	goto do_again;
 943
 944delete_fail:
 945	btrfs_release_path(path);
 946	mutex_unlock(&node->mutex);
 947	return ret;
 948}
 949
 950static void btrfs_release_delayed_inode(struct btrfs_delayed_node *delayed_node)
 951{
 952	struct btrfs_delayed_root *delayed_root;
 953
 954	if (delayed_node && delayed_node->inode_dirty) {
 
 955		BUG_ON(!delayed_node->root);
 956		delayed_node->inode_dirty = 0;
 957		delayed_node->count--;
 958
 959		delayed_root = delayed_node->root->fs_info->delayed_root;
 960		atomic_dec(&delayed_root->items);
 961		if (atomic_read(&delayed_root->items) <
 962		    BTRFS_DELAYED_BACKGROUND &&
 963		    waitqueue_active(&delayed_root->wait))
 964			wake_up(&delayed_root->wait);
 965	}
 966}
 967
 968static int btrfs_update_delayed_inode(struct btrfs_trans_handle *trans,
 969				      struct btrfs_root *root,
 970				      struct btrfs_path *path,
 971				      struct btrfs_delayed_node *node)
 
 
 
 
 
 
 
 
 
 
 
 
 972{
 
 973	struct btrfs_key key;
 974	struct btrfs_inode_item *inode_item;
 975	struct extent_buffer *leaf;
 
 976	int ret;
 977
 978	mutex_lock(&node->mutex);
 979	if (!node->inode_dirty) {
 980		mutex_unlock(&node->mutex);
 981		return 0;
 982	}
 983
 984	key.objectid = node->inode_id;
 985	btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
 986	key.offset = 0;
 987	ret = btrfs_lookup_inode(trans, root, path, &key, 1);
 
 
 
 
 
 
 988	if (ret > 0) {
 989		btrfs_release_path(path);
 990		mutex_unlock(&node->mutex);
 991		return -ENOENT;
 992	} else if (ret < 0) {
 993		mutex_unlock(&node->mutex);
 994		return ret;
 995	}
 996
 997	btrfs_unlock_up_safe(path, 1);
 998	leaf = path->nodes[0];
 999	inode_item = btrfs_item_ptr(leaf, path->slots[0],
1000				    struct btrfs_inode_item);
1001	write_extent_buffer(leaf, &node->inode_item, (unsigned long)inode_item,
1002			    sizeof(struct btrfs_inode_item));
1003	btrfs_mark_buffer_dirty(leaf);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1004	btrfs_release_path(path);
1005
1006	btrfs_delayed_inode_release_metadata(root, node);
1007	btrfs_release_delayed_inode(node);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1008	mutex_unlock(&node->mutex);
 
 
1009
1010	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1011}
1012
1013/* Called when committing the transaction. */
1014int btrfs_run_delayed_items(struct btrfs_trans_handle *trans,
1015			    struct btrfs_root *root)
 
 
 
 
1016{
 
1017	struct btrfs_delayed_root *delayed_root;
1018	struct btrfs_delayed_node *curr_node, *prev_node;
1019	struct btrfs_path *path;
1020	struct btrfs_block_rsv *block_rsv;
1021	int ret = 0;
 
 
 
 
1022
1023	path = btrfs_alloc_path();
1024	if (!path)
1025		return -ENOMEM;
1026	path->leave_spinning = 1;
1027
1028	block_rsv = trans->block_rsv;
1029	trans->block_rsv = &root->fs_info->global_block_rsv;
1030
1031	delayed_root = btrfs_get_delayed_root(root);
1032
1033	curr_node = btrfs_first_delayed_node(delayed_root);
1034	while (curr_node) {
1035		root = curr_node->root;
1036		ret = btrfs_insert_delayed_items(trans, path, root,
1037						 curr_node);
1038		if (!ret)
1039			ret = btrfs_delete_delayed_items(trans, path, root,
1040							 curr_node);
1041		if (!ret)
1042			ret = btrfs_update_delayed_inode(trans, root, path,
1043							 curr_node);
1044		if (ret) {
1045			btrfs_release_delayed_node(curr_node);
 
 
1046			break;
1047		}
1048
1049		prev_node = curr_node;
1050		curr_node = btrfs_next_delayed_node(curr_node);
1051		btrfs_release_delayed_node(prev_node);
1052	}
1053
 
 
1054	btrfs_free_path(path);
1055	trans->block_rsv = block_rsv;
 
1056	return ret;
1057}
1058
1059static int __btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1060					      struct btrfs_delayed_node *node)
 
 
 
 
 
 
 
 
 
 
1061{
 
1062	struct btrfs_path *path;
1063	struct btrfs_block_rsv *block_rsv;
1064	int ret;
1065
 
 
 
 
 
 
 
 
 
 
 
1066	path = btrfs_alloc_path();
1067	if (!path)
 
1068		return -ENOMEM;
 
1069	path->leave_spinning = 1;
1070
1071	block_rsv = trans->block_rsv;
1072	trans->block_rsv = &node->root->fs_info->global_block_rsv;
1073
1074	ret = btrfs_insert_delayed_items(trans, path, node->root, node);
1075	if (!ret)
1076		ret = btrfs_delete_delayed_items(trans, path, node->root, node);
1077	if (!ret)
1078		ret = btrfs_update_delayed_inode(trans, node->root, path, node);
1079	btrfs_free_path(path);
 
1080
1081	trans->block_rsv = block_rsv;
1082	return ret;
1083}
1084
1085int btrfs_commit_inode_delayed_items(struct btrfs_trans_handle *trans,
1086				     struct inode *inode)
1087{
 
 
1088	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
 
 
1089	int ret;
1090
1091	if (!delayed_node)
1092		return 0;
1093
1094	mutex_lock(&delayed_node->mutex);
1095	if (!delayed_node->count) {
1096		mutex_unlock(&delayed_node->mutex);
1097		btrfs_release_delayed_node(delayed_node);
1098		return 0;
1099	}
1100	mutex_unlock(&delayed_node->mutex);
1101
1102	ret = __btrfs_commit_inode_delayed_items(trans, delayed_node);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1103	btrfs_release_delayed_node(delayed_node);
 
1104	return ret;
1105}
1106
1107void btrfs_remove_delayed_node(struct inode *inode)
1108{
1109	struct btrfs_delayed_node *delayed_node;
1110
1111	delayed_node = ACCESS_ONCE(BTRFS_I(inode)->delayed_node);
1112	if (!delayed_node)
1113		return;
1114
1115	BTRFS_I(inode)->delayed_node = NULL;
1116	btrfs_release_delayed_node(delayed_node);
1117}
1118
1119struct btrfs_async_delayed_node {
1120	struct btrfs_root *root;
1121	struct btrfs_delayed_node *delayed_node;
1122	struct btrfs_work work;
1123};
1124
1125static void btrfs_async_run_delayed_node_done(struct btrfs_work *work)
1126{
1127	struct btrfs_async_delayed_node *async_node;
 
1128	struct btrfs_trans_handle *trans;
1129	struct btrfs_path *path;
1130	struct btrfs_delayed_node *delayed_node = NULL;
1131	struct btrfs_root *root;
1132	struct btrfs_block_rsv *block_rsv;
1133	unsigned long nr = 0;
1134	int need_requeue = 0;
1135	int ret;
1136
1137	async_node = container_of(work, struct btrfs_async_delayed_node, work);
 
1138
1139	path = btrfs_alloc_path();
1140	if (!path)
1141		goto out;
1142	path->leave_spinning = 1;
1143
1144	delayed_node = async_node->delayed_node;
1145	root = delayed_node->root;
 
 
 
 
 
 
 
 
 
1146
1147	trans = btrfs_join_transaction(root);
1148	if (IS_ERR(trans))
1149		goto free_path;
 
 
 
 
1150
1151	block_rsv = trans->block_rsv;
1152	trans->block_rsv = &root->fs_info->global_block_rsv;
1153
1154	ret = btrfs_insert_delayed_items(trans, path, root, delayed_node);
1155	if (!ret)
1156		ret = btrfs_delete_delayed_items(trans, path, root,
1157						 delayed_node);
1158
1159	if (!ret)
1160		btrfs_update_delayed_inode(trans, root, path, delayed_node);
 
1161
1162	/*
1163	 * Maybe new delayed items have been inserted, so we need requeue
1164	 * the work. Besides that, we must dequeue the empty delayed nodes
1165	 * to avoid the race between delayed items balance and the worker.
1166	 * The race like this:
1167	 * 	Task1				Worker thread
1168	 * 					count == 0, needn't requeue
1169	 * 					  also needn't insert the
1170	 * 					  delayed node into prepare
1171	 * 					  list again.
1172	 * 	add lots of delayed items
1173	 * 	queue the delayed node
1174	 * 	  already in the list,
1175	 * 	  and not in the prepare
1176	 * 	  list, it means the delayed
1177	 * 	  node is being dealt with
1178	 * 	  by the worker.
1179	 * 	do delayed items balance
1180	 * 	  the delayed node is being
1181	 * 	  dealt with by the worker
1182	 * 	  now, just wait.
1183	 * 	  				the worker goto idle.
1184	 * Task1 will sleep until the transaction is commited.
1185	 */
1186	mutex_lock(&delayed_node->mutex);
1187	if (delayed_node->count)
1188		need_requeue = 1;
1189	else
1190		btrfs_dequeue_delayed_node(root->fs_info->delayed_root,
1191					   delayed_node);
1192	mutex_unlock(&delayed_node->mutex);
1193
1194	nr = trans->blocks_used;
 
1195
1196	trans->block_rsv = block_rsv;
1197	btrfs_end_transaction_dmeta(trans, root);
1198	__btrfs_btree_balance_dirty(root, nr);
1199free_path:
1200	btrfs_free_path(path);
1201out:
1202	if (need_requeue)
1203		btrfs_requeue_work(&async_node->work);
1204	else {
1205		btrfs_release_prepared_delayed_node(delayed_node);
1206		kfree(async_node);
1207	}
1208}
1209
 
1210static int btrfs_wq_run_delayed_node(struct btrfs_delayed_root *delayed_root,
1211				     struct btrfs_root *root, int all)
1212{
1213	struct btrfs_async_delayed_node *async_node;
1214	struct btrfs_delayed_node *curr;
1215	int count = 0;
1216
1217again:
1218	curr = btrfs_first_prepared_delayed_node(delayed_root);
1219	if (!curr)
1220		return 0;
1221
1222	async_node = kmalloc(sizeof(*async_node), GFP_NOFS);
1223	if (!async_node) {
1224		btrfs_release_prepared_delayed_node(curr);
1225		return -ENOMEM;
1226	}
1227
1228	async_node->root = root;
1229	async_node->delayed_node = curr;
1230
1231	async_node->work.func = btrfs_async_run_delayed_node_done;
1232	async_node->work.flags = 0;
1233
1234	btrfs_queue_worker(&root->fs_info->delayed_workers, &async_node->work);
1235	count++;
1236
1237	if (all || count < 4)
1238		goto again;
1239
 
1240	return 0;
1241}
1242
1243void btrfs_assert_delayed_root_empty(struct btrfs_root *root)
1244{
1245	struct btrfs_delayed_root *delayed_root;
1246	delayed_root = btrfs_get_delayed_root(root);
1247	WARN_ON(btrfs_first_delayed_node(delayed_root));
1248}
1249
1250void btrfs_balance_delayed_items(struct btrfs_root *root)
1251{
1252	struct btrfs_delayed_root *delayed_root;
1253
1254	delayed_root = btrfs_get_delayed_root(root);
 
1255
1256	if (atomic_read(&delayed_root->items) < BTRFS_DELAYED_BACKGROUND)
 
 
 
 
 
 
 
 
 
 
 
1257		return;
1258
1259	if (atomic_read(&delayed_root->items) >= BTRFS_DELAYED_WRITEBACK) {
 
1260		int ret;
1261		ret = btrfs_wq_run_delayed_node(delayed_root, root, 1);
 
 
 
1262		if (ret)
1263			return;
1264
1265		wait_event_interruptible_timeout(
1266				delayed_root->wait,
1267				(atomic_read(&delayed_root->items) <
1268				 BTRFS_DELAYED_BACKGROUND),
1269				HZ);
1270		return;
1271	}
1272
1273	btrfs_wq_run_delayed_node(delayed_root, root, 0);
1274}
1275
 
1276int btrfs_insert_delayed_dir_index(struct btrfs_trans_handle *trans,
1277				   struct btrfs_root *root, const char *name,
1278				   int name_len, struct inode *dir,
1279				   struct btrfs_disk_key *disk_key, u8 type,
1280				   u64 index)
1281{
1282	struct btrfs_delayed_node *delayed_node;
1283	struct btrfs_delayed_item *delayed_item;
1284	struct btrfs_dir_item *dir_item;
1285	int ret;
1286
1287	delayed_node = btrfs_get_or_create_delayed_node(dir);
1288	if (IS_ERR(delayed_node))
1289		return PTR_ERR(delayed_node);
1290
1291	delayed_item = btrfs_alloc_delayed_item(sizeof(*dir_item) + name_len);
1292	if (!delayed_item) {
1293		ret = -ENOMEM;
1294		goto release_node;
1295	}
1296
1297	ret = btrfs_delayed_item_reserve_metadata(trans, root, delayed_item);
1298	/*
1299	 * we have reserved enough space when we start a new transaction,
1300	 * so reserving metadata failure is impossible
1301	 */
1302	BUG_ON(ret);
1303
1304	delayed_item->key.objectid = btrfs_ino(dir);
1305	btrfs_set_key_type(&delayed_item->key, BTRFS_DIR_INDEX_KEY);
1306	delayed_item->key.offset = index;
1307
1308	dir_item = (struct btrfs_dir_item *)delayed_item->data;
1309	dir_item->location = *disk_key;
1310	dir_item->transid = cpu_to_le64(trans->transid);
1311	dir_item->data_len = 0;
1312	dir_item->name_len = cpu_to_le16(name_len);
1313	dir_item->type = type;
1314	memcpy((char *)(dir_item + 1), name, name_len);
1315
 
 
 
 
 
 
 
1316	mutex_lock(&delayed_node->mutex);
1317	ret = __btrfs_add_delayed_insertion_item(delayed_node, delayed_item);
1318	if (unlikely(ret)) {
1319		printk(KERN_ERR "err add delayed dir index item(name: %s) into "
1320				"the insertion tree of the delayed node"
1321				"(root id: %llu, inode id: %llu, errno: %d)\n",
1322				name,
1323				(unsigned long long)delayed_node->root->objectid,
1324				(unsigned long long)delayed_node->inode_id,
1325				ret);
1326		BUG();
1327	}
1328	mutex_unlock(&delayed_node->mutex);
1329
1330release_node:
1331	btrfs_release_delayed_node(delayed_node);
1332	return ret;
1333}
1334
1335static int btrfs_delete_delayed_insertion_item(struct btrfs_root *root,
1336					       struct btrfs_delayed_node *node,
1337					       struct btrfs_key *key)
1338{
1339	struct btrfs_delayed_item *item;
1340
1341	mutex_lock(&node->mutex);
1342	item = __btrfs_lookup_delayed_insertion_item(node, key);
1343	if (!item) {
1344		mutex_unlock(&node->mutex);
1345		return 1;
1346	}
1347
1348	btrfs_delayed_item_release_metadata(root, item);
1349	btrfs_release_delayed_item(item);
1350	mutex_unlock(&node->mutex);
1351	return 0;
1352}
1353
1354int btrfs_delete_delayed_dir_index(struct btrfs_trans_handle *trans,
1355				   struct btrfs_root *root, struct inode *dir,
1356				   u64 index)
1357{
1358	struct btrfs_delayed_node *node;
1359	struct btrfs_delayed_item *item;
1360	struct btrfs_key item_key;
1361	int ret;
1362
1363	node = btrfs_get_or_create_delayed_node(dir);
1364	if (IS_ERR(node))
1365		return PTR_ERR(node);
1366
1367	item_key.objectid = btrfs_ino(dir);
1368	btrfs_set_key_type(&item_key, BTRFS_DIR_INDEX_KEY);
1369	item_key.offset = index;
1370
1371	ret = btrfs_delete_delayed_insertion_item(root, node, &item_key);
 
1372	if (!ret)
1373		goto end;
1374
1375	item = btrfs_alloc_delayed_item(0);
1376	if (!item) {
1377		ret = -ENOMEM;
1378		goto end;
1379	}
1380
1381	item->key = item_key;
1382
1383	ret = btrfs_delayed_item_reserve_metadata(trans, root, item);
1384	/*
1385	 * we have reserved enough space when we start a new transaction,
1386	 * so reserving metadata failure is impossible.
1387	 */
1388	BUG_ON(ret);
 
 
 
 
 
1389
1390	mutex_lock(&node->mutex);
1391	ret = __btrfs_add_delayed_deletion_item(node, item);
1392	if (unlikely(ret)) {
1393		printk(KERN_ERR "err add delayed dir index item(index: %llu) "
1394				"into the deletion tree of the delayed node"
1395				"(root id: %llu, inode id: %llu, errno: %d)\n",
1396				(unsigned long long)index,
1397				(unsigned long long)node->root->objectid,
1398				(unsigned long long)node->inode_id,
1399				ret);
1400		BUG();
1401	}
1402	mutex_unlock(&node->mutex);
1403end:
1404	btrfs_release_delayed_node(node);
1405	return ret;
1406}
1407
1408int btrfs_inode_delayed_dir_index_count(struct inode *inode)
1409{
1410	struct btrfs_delayed_node *delayed_node = btrfs_get_delayed_node(inode);
1411
1412	if (!delayed_node)
1413		return -ENOENT;
1414
1415	/*
1416	 * Since we have held i_mutex of this directory, it is impossible that
1417	 * a new directory index is added into the delayed node and index_cnt
1418	 * is updated now. So we needn't lock the delayed node.
1419	 */
1420	if (!delayed_node->index_cnt) {
1421		btrfs_release_delayed_node(delayed_node);
1422		return -EINVAL;
1423	}
1424
1425	BTRFS_I(inode)->index_cnt = delayed_node->index_cnt;
1426	btrfs_release_delayed_node(delayed_node);
1427	return 0;
1428}
1429
1430void btrfs_get_delayed_items(struct inode *inode, struct list_head *ins_list,
1431			     struct list_head *del_list)
 
1432{
1433	struct btrfs_delayed_node *delayed_node;
1434	struct btrfs_delayed_item *item;
1435
1436	delayed_node = btrfs_get_delayed_node(inode);
1437	if (!delayed_node)
1438		return;
 
 
 
 
 
 
 
1439
1440	mutex_lock(&delayed_node->mutex);
1441	item = __btrfs_first_delayed_insertion_item(delayed_node);
1442	while (item) {
1443		atomic_inc(&item->refs);
1444		list_add_tail(&item->readdir_list, ins_list);
1445		item = __btrfs_next_delayed_item(item);
1446	}
1447
1448	item = __btrfs_first_delayed_deletion_item(delayed_node);
1449	while (item) {
1450		atomic_inc(&item->refs);
1451		list_add_tail(&item->readdir_list, del_list);
1452		item = __btrfs_next_delayed_item(item);
1453	}
1454	mutex_unlock(&delayed_node->mutex);
1455	/*
1456	 * This delayed node is still cached in the btrfs inode, so refs
1457	 * must be > 1 now, and we needn't check it is going to be freed
1458	 * or not.
1459	 *
1460	 * Besides that, this function is used to read dir, we do not
1461	 * insert/delete delayed items in this period. So we also needn't
1462	 * requeue or dequeue this delayed node.
1463	 */
1464	atomic_dec(&delayed_node->refs);
 
 
1465}
1466
1467void btrfs_put_delayed_items(struct list_head *ins_list,
1468			     struct list_head *del_list)
 
1469{
1470	struct btrfs_delayed_item *curr, *next;
1471
1472	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1473		list_del(&curr->readdir_list);
1474		if (atomic_dec_and_test(&curr->refs))
1475			kfree(curr);
1476	}
1477
1478	list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1479		list_del(&curr->readdir_list);
1480		if (atomic_dec_and_test(&curr->refs))
1481			kfree(curr);
1482	}
 
 
 
 
 
 
1483}
1484
1485int btrfs_should_delete_dir_index(struct list_head *del_list,
1486				  u64 index)
1487{
1488	struct btrfs_delayed_item *curr, *next;
1489	int ret;
1490
1491	if (list_empty(del_list))
1492		return 0;
1493
1494	list_for_each_entry_safe(curr, next, del_list, readdir_list) {
1495		if (curr->key.offset > index)
1496			break;
1497
1498		list_del(&curr->readdir_list);
1499		ret = (curr->key.offset == index);
1500
1501		if (atomic_dec_and_test(&curr->refs))
1502			kfree(curr);
1503
1504		if (ret)
1505			return 1;
1506		else
1507			continue;
1508	}
1509	return 0;
1510}
1511
1512/*
1513 * btrfs_readdir_delayed_dir_index - read dir info stored in the delayed tree
1514 *
1515 */
1516int btrfs_readdir_delayed_dir_index(struct file *filp, void *dirent,
1517				    filldir_t filldir,
1518				    struct list_head *ins_list)
1519{
1520	struct btrfs_dir_item *di;
1521	struct btrfs_delayed_item *curr, *next;
1522	struct btrfs_key location;
1523	char *name;
1524	int name_len;
1525	int over = 0;
1526	unsigned char d_type;
1527
1528	if (list_empty(ins_list))
1529		return 0;
1530
1531	/*
1532	 * Changing the data of the delayed item is impossible. So
1533	 * we needn't lock them. And we have held i_mutex of the
1534	 * directory, nobody can delete any directory indexes now.
1535	 */
1536	list_for_each_entry_safe(curr, next, ins_list, readdir_list) {
1537		list_del(&curr->readdir_list);
1538
1539		if (curr->key.offset < filp->f_pos) {
1540			if (atomic_dec_and_test(&curr->refs))
1541				kfree(curr);
1542			continue;
1543		}
1544
1545		filp->f_pos = curr->key.offset;
1546
1547		di = (struct btrfs_dir_item *)curr->data;
1548		name = (char *)(di + 1);
1549		name_len = le16_to_cpu(di->name_len);
1550
1551		d_type = btrfs_filetype_table[di->type];
1552		btrfs_disk_key_to_cpu(&location, &di->location);
1553
1554		over = filldir(dirent, name, name_len, curr->key.offset,
1555			       location.objectid, d_type);
1556
1557		if (atomic_dec_and_test(&curr->refs))
1558			kfree(curr);
1559
1560		if (over)
1561			return 1;
 
1562	}
1563	return 0;
1564}
1565
1566BTRFS_SETGET_STACK_FUNCS(stack_inode_generation, struct btrfs_inode_item,
1567			 generation, 64);
1568BTRFS_SETGET_STACK_FUNCS(stack_inode_sequence, struct btrfs_inode_item,
1569			 sequence, 64);
1570BTRFS_SETGET_STACK_FUNCS(stack_inode_transid, struct btrfs_inode_item,
1571			 transid, 64);
1572BTRFS_SETGET_STACK_FUNCS(stack_inode_size, struct btrfs_inode_item, size, 64);
1573BTRFS_SETGET_STACK_FUNCS(stack_inode_nbytes, struct btrfs_inode_item,
1574			 nbytes, 64);
1575BTRFS_SETGET_STACK_FUNCS(stack_inode_block_group, struct btrfs_inode_item,
1576			 block_group, 64);
1577BTRFS_SETGET_STACK_FUNCS(stack_inode_nlink, struct btrfs_inode_item, nlink, 32);
1578BTRFS_SETGET_STACK_FUNCS(stack_inode_uid, struct btrfs_inode_item, uid, 32);
1579BTRFS_SETGET_STACK_FUNCS(stack_inode_gid, struct btrfs_inode_item, gid, 32);
1580BTRFS_SETGET_STACK_FUNCS(stack_inode_mode, struct btrfs_inode_item, mode, 32);
1581BTRFS_SETGET_STACK_FUNCS(stack_inode_rdev, struct btrfs_inode_item, rdev, 64);
1582BTRFS_SETGET_STACK_FUNCS(stack_inode_flags, struct btrfs_inode_item, flags, 64);
1583
1584BTRFS_SETGET_STACK_FUNCS(stack_timespec_sec, struct btrfs_timespec, sec, 64);
1585BTRFS_SETGET_STACK_FUNCS(stack_timespec_nsec, struct btrfs_timespec, nsec, 32);
1586
1587static void fill_stack_inode_item(struct btrfs_trans_handle *trans,
1588				  struct btrfs_inode_item *inode_item,
1589				  struct inode *inode)
1590{
1591	btrfs_set_stack_inode_uid(inode_item, inode->i_uid);
1592	btrfs_set_stack_inode_gid(inode_item, inode->i_gid);
1593	btrfs_set_stack_inode_size(inode_item, BTRFS_I(inode)->disk_i_size);
1594	btrfs_set_stack_inode_mode(inode_item, inode->i_mode);
1595	btrfs_set_stack_inode_nlink(inode_item, inode->i_nlink);
1596	btrfs_set_stack_inode_nbytes(inode_item, inode_get_bytes(inode));
1597	btrfs_set_stack_inode_generation(inode_item,
1598					 BTRFS_I(inode)->generation);
1599	btrfs_set_stack_inode_sequence(inode_item, BTRFS_I(inode)->sequence);
 
1600	btrfs_set_stack_inode_transid(inode_item, trans->transid);
1601	btrfs_set_stack_inode_rdev(inode_item, inode->i_rdev);
1602	btrfs_set_stack_inode_flags(inode_item, BTRFS_I(inode)->flags);
1603	btrfs_set_stack_inode_block_group(inode_item, 0);
1604
1605	btrfs_set_stack_timespec_sec(btrfs_inode_atime(inode_item),
1606				     inode->i_atime.tv_sec);
1607	btrfs_set_stack_timespec_nsec(btrfs_inode_atime(inode_item),
1608				      inode->i_atime.tv_nsec);
1609
1610	btrfs_set_stack_timespec_sec(btrfs_inode_mtime(inode_item),
1611				     inode->i_mtime.tv_sec);
1612	btrfs_set_stack_timespec_nsec(btrfs_inode_mtime(inode_item),
1613				      inode->i_mtime.tv_nsec);
1614
1615	btrfs_set_stack_timespec_sec(btrfs_inode_ctime(inode_item),
1616				     inode->i_ctime.tv_sec);
1617	btrfs_set_stack_timespec_nsec(btrfs_inode_ctime(inode_item),
1618				      inode->i_ctime.tv_nsec);
 
 
 
 
 
1619}
1620
1621int btrfs_fill_inode(struct inode *inode, u32 *rdev)
1622{
1623	struct btrfs_delayed_node *delayed_node;
1624	struct btrfs_inode_item *inode_item;
1625	struct btrfs_timespec *tspec;
1626
1627	delayed_node = btrfs_get_delayed_node(inode);
1628	if (!delayed_node)
1629		return -ENOENT;
1630
1631	mutex_lock(&delayed_node->mutex);
1632	if (!delayed_node->inode_dirty) {
1633		mutex_unlock(&delayed_node->mutex);
1634		btrfs_release_delayed_node(delayed_node);
1635		return -ENOENT;
1636	}
1637
1638	inode_item = &delayed_node->inode_item;
1639
1640	inode->i_uid = btrfs_stack_inode_uid(inode_item);
1641	inode->i_gid = btrfs_stack_inode_gid(inode_item);
1642	btrfs_i_size_write(inode, btrfs_stack_inode_size(inode_item));
1643	inode->i_mode = btrfs_stack_inode_mode(inode_item);
1644	inode->i_nlink = btrfs_stack_inode_nlink(inode_item);
1645	inode_set_bytes(inode, btrfs_stack_inode_nbytes(inode_item));
1646	BTRFS_I(inode)->generation = btrfs_stack_inode_generation(inode_item);
1647	BTRFS_I(inode)->sequence = btrfs_stack_inode_sequence(inode_item);
 
 
 
1648	inode->i_rdev = 0;
1649	*rdev = btrfs_stack_inode_rdev(inode_item);
1650	BTRFS_I(inode)->flags = btrfs_stack_inode_flags(inode_item);
1651
1652	tspec = btrfs_inode_atime(inode_item);
1653	inode->i_atime.tv_sec = btrfs_stack_timespec_sec(tspec);
1654	inode->i_atime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1655
1656	tspec = btrfs_inode_mtime(inode_item);
1657	inode->i_mtime.tv_sec = btrfs_stack_timespec_sec(tspec);
1658	inode->i_mtime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
1659
1660	tspec = btrfs_inode_ctime(inode_item);
1661	inode->i_ctime.tv_sec = btrfs_stack_timespec_sec(tspec);
1662	inode->i_ctime.tv_nsec = btrfs_stack_timespec_nsec(tspec);
 
 
1663
1664	inode->i_generation = BTRFS_I(inode)->generation;
1665	BTRFS_I(inode)->index_cnt = (u64)-1;
1666
1667	mutex_unlock(&delayed_node->mutex);
1668	btrfs_release_delayed_node(delayed_node);
1669	return 0;
1670}
1671
1672int btrfs_delayed_update_inode(struct btrfs_trans_handle *trans,
1673			       struct btrfs_root *root, struct inode *inode)
1674{
1675	struct btrfs_delayed_node *delayed_node;
1676	int ret = 0;
1677
1678	delayed_node = btrfs_get_or_create_delayed_node(inode);
1679	if (IS_ERR(delayed_node))
1680		return PTR_ERR(delayed_node);
1681
1682	mutex_lock(&delayed_node->mutex);
1683	if (delayed_node->inode_dirty) {
1684		fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1685		goto release_node;
1686	}
1687
1688	ret = btrfs_delayed_inode_reserve_metadata(trans, root, delayed_node);
1689	/*
1690	 * we must reserve enough space when we start a new transaction,
1691	 * so reserving metadata failure is impossible
1692	 */
1693	BUG_ON(ret);
1694
1695	fill_stack_inode_item(trans, &delayed_node->inode_item, inode);
1696	delayed_node->inode_dirty = 1;
1697	delayed_node->count++;
1698	atomic_inc(&root->fs_info->delayed_root->items);
1699release_node:
1700	mutex_unlock(&delayed_node->mutex);
1701	btrfs_release_delayed_node(delayed_node);
1702	return ret;
1703}
1704
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1705static void __btrfs_kill_delayed_node(struct btrfs_delayed_node *delayed_node)
1706{
1707	struct btrfs_root *root = delayed_node->root;
 
1708	struct btrfs_delayed_item *curr_item, *prev_item;
1709
1710	mutex_lock(&delayed_node->mutex);
1711	curr_item = __btrfs_first_delayed_insertion_item(delayed_node);
1712	while (curr_item) {
1713		btrfs_delayed_item_release_metadata(root, curr_item);
1714		prev_item = curr_item;
1715		curr_item = __btrfs_next_delayed_item(prev_item);
1716		btrfs_release_delayed_item(prev_item);
1717	}
1718
1719	curr_item = __btrfs_first_delayed_deletion_item(delayed_node);
1720	while (curr_item) {
1721		btrfs_delayed_item_release_metadata(root, curr_item);
1722		prev_item = curr_item;
1723		curr_item = __btrfs_next_delayed_item(prev_item);
1724		btrfs_release_delayed_item(prev_item);
1725	}
1726
1727	if (delayed_node->inode_dirty) {
1728		btrfs_delayed_inode_release_metadata(root, delayed_node);
 
 
 
1729		btrfs_release_delayed_inode(delayed_node);
1730	}
1731	mutex_unlock(&delayed_node->mutex);
1732}
1733
1734void btrfs_kill_delayed_inode_items(struct inode *inode)
1735{
1736	struct btrfs_delayed_node *delayed_node;
1737
1738	delayed_node = btrfs_get_delayed_node(inode);
1739	if (!delayed_node)
1740		return;
1741
1742	__btrfs_kill_delayed_node(delayed_node);
1743	btrfs_release_delayed_node(delayed_node);
1744}
1745
1746void btrfs_kill_all_delayed_nodes(struct btrfs_root *root)
1747{
1748	u64 inode_id = 0;
1749	struct btrfs_delayed_node *delayed_nodes[8];
1750	int i, n;
1751
1752	while (1) {
1753		spin_lock(&root->inode_lock);
1754		n = radix_tree_gang_lookup(&root->delayed_nodes_tree,
1755					   (void **)delayed_nodes, inode_id,
1756					   ARRAY_SIZE(delayed_nodes));
1757		if (!n) {
1758			spin_unlock(&root->inode_lock);
1759			break;
1760		}
1761
1762		inode_id = delayed_nodes[n - 1]->inode_id + 1;
1763
1764		for (i = 0; i < n; i++)
1765			atomic_inc(&delayed_nodes[i]->refs);
1766		spin_unlock(&root->inode_lock);
1767
1768		for (i = 0; i < n; i++) {
1769			__btrfs_kill_delayed_node(delayed_nodes[i]);
1770			btrfs_release_delayed_node(delayed_nodes[i]);
1771		}
1772	}
1773}