Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2009 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/sched.h>
   7#include <linux/pagemap.h>
   8#include <linux/writeback.h>
   9#include <linux/blkdev.h>
  10#include <linux/rbtree.h>
  11#include <linux/slab.h>
 
  12#include "ctree.h"
  13#include "disk-io.h"
  14#include "transaction.h"
  15#include "volumes.h"
  16#include "locking.h"
  17#include "btrfs_inode.h"
  18#include "async-thread.h"
  19#include "free-space-cache.h"
  20#include "inode-map.h"
  21#include "qgroup.h"
  22#include "print-tree.h"
  23#include "delalloc-space.h"
  24#include "block-group.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  25
  26/*
  27 * backref_node, mapping_node and tree_block start with this
  28 */
  29struct tree_entry {
  30	struct rb_node rb_node;
  31	u64 bytenr;
  32};
  33
  34/*
  35 * present a tree block in the backref cache
  36 */
  37struct backref_node {
  38	struct rb_node rb_node;
  39	u64 bytenr;
  40
  41	u64 new_bytenr;
  42	/* objectid of tree block owner, can be not uptodate */
  43	u64 owner;
  44	/* link to pending, changed or detached list */
  45	struct list_head list;
  46	/* list of upper level blocks reference this block */
  47	struct list_head upper;
  48	/* list of child blocks in the cache */
  49	struct list_head lower;
  50	/* NULL if this node is not tree root */
  51	struct btrfs_root *root;
  52	/* extent buffer got by COW the block */
  53	struct extent_buffer *eb;
  54	/* level of tree block */
  55	unsigned int level:8;
  56	/* is the block in non-reference counted tree */
  57	unsigned int cowonly:1;
  58	/* 1 if no child node in the cache */
  59	unsigned int lowest:1;
  60	/* is the extent buffer locked */
  61	unsigned int locked:1;
  62	/* has the block been processed */
  63	unsigned int processed:1;
  64	/* have backrefs of this block been checked */
  65	unsigned int checked:1;
  66	/*
  67	 * 1 if corresponding block has been cowed but some upper
  68	 * level block pointers may not point to the new location
  69	 */
  70	unsigned int pending:1;
  71	/*
  72	 * 1 if the backref node isn't connected to any other
  73	 * backref node.
  74	 */
  75	unsigned int detached:1;
  76};
  77
  78/*
  79 * present a block pointer in the backref cache
  80 */
  81struct backref_edge {
  82	struct list_head list[2];
  83	struct backref_node *node[2];
  84};
  85
  86#define LOWER	0
  87#define UPPER	1
  88#define RELOCATION_RESERVED_NODES	256
  89
  90struct backref_cache {
  91	/* red black tree of all backref nodes in the cache */
  92	struct rb_root rb_root;
  93	/* for passing backref nodes to btrfs_reloc_cow_block */
  94	struct backref_node *path[BTRFS_MAX_LEVEL];
  95	/*
  96	 * list of blocks that have been cowed but some block
  97	 * pointers in upper level blocks may not reflect the
  98	 * new location
  99	 */
 100	struct list_head pending[BTRFS_MAX_LEVEL];
 101	/* list of backref nodes with no child node */
 102	struct list_head leaves;
 103	/* list of blocks that have been cowed in current transaction */
 104	struct list_head changed;
 105	/* list of detached backref node. */
 106	struct list_head detached;
 107
 108	u64 last_trans;
 109
 110	int nr_nodes;
 111	int nr_edges;
 112};
 113
 114/*
 115 * map address of tree root to tree
 116 */
 117struct mapping_node {
 118	struct rb_node rb_node;
 119	u64 bytenr;
 
 
 120	void *data;
 121};
 122
 123struct mapping_tree {
 124	struct rb_root rb_root;
 125	spinlock_t lock;
 126};
 127
 128/*
 129 * present a tree block to process
 130 */
 131struct tree_block {
 132	struct rb_node rb_node;
 133	u64 bytenr;
 
 
 
 134	struct btrfs_key key;
 135	unsigned int level:8;
 136	unsigned int key_ready:1;
 137};
 138
 139#define MAX_EXTENTS 128
 140
 141struct file_extent_cluster {
 142	u64 start;
 143	u64 end;
 144	u64 boundary[MAX_EXTENTS];
 145	unsigned int nr;
 
 
 
 
 
 
 
 146};
 147
 148struct reloc_control {
 149	/* block group to relocate */
 150	struct btrfs_block_group_cache *block_group;
 151	/* extent tree */
 152	struct btrfs_root *extent_root;
 153	/* inode for moving data */
 154	struct inode *data_inode;
 155
 156	struct btrfs_block_rsv *block_rsv;
 157
 158	struct backref_cache backref_cache;
 159
 160	struct file_extent_cluster cluster;
 161	/* tree blocks have been processed */
 162	struct extent_io_tree processed_blocks;
 163	/* map start of tree root to corresponding reloc tree */
 164	struct mapping_tree reloc_root_tree;
 165	/* list of reloc trees */
 166	struct list_head reloc_roots;
 167	/* list of subvolume trees that get relocated */
 168	struct list_head dirty_subvol_roots;
 169	/* size of metadata reservation for merging reloc trees */
 170	u64 merging_rsv_size;
 171	/* size of relocated tree nodes */
 172	u64 nodes_relocated;
 173	/* reserved size for block group relocation*/
 174	u64 reserved_bytes;
 175
 176	u64 search_start;
 177	u64 extents_found;
 178
 179	unsigned int stage:8;
 180	unsigned int create_reloc_tree:1;
 181	unsigned int merge_reloc_tree:1;
 182	unsigned int found_file_extent:1;
 183};
 184
 185/* stages of data relocation */
 186#define MOVE_DATA_EXTENTS	0
 187#define UPDATE_DATA_PTRS	1
 188
 189static void remove_backref_node(struct backref_cache *cache,
 190				struct backref_node *node);
 191static void __mark_block_processed(struct reloc_control *rc,
 192				   struct backref_node *node);
 193
 194static void mapping_tree_init(struct mapping_tree *tree)
 195{
 196	tree->rb_root = RB_ROOT;
 197	spin_lock_init(&tree->lock);
 198}
 199
 200static void backref_cache_init(struct backref_cache *cache)
 201{
 202	int i;
 203	cache->rb_root = RB_ROOT;
 204	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
 205		INIT_LIST_HEAD(&cache->pending[i]);
 206	INIT_LIST_HEAD(&cache->changed);
 207	INIT_LIST_HEAD(&cache->detached);
 208	INIT_LIST_HEAD(&cache->leaves);
 209}
 210
 211static void backref_cache_cleanup(struct backref_cache *cache)
 212{
 213	struct backref_node *node;
 214	int i;
 215
 216	while (!list_empty(&cache->detached)) {
 217		node = list_entry(cache->detached.next,
 218				  struct backref_node, list);
 219		remove_backref_node(cache, node);
 220	}
 221
 222	while (!list_empty(&cache->leaves)) {
 223		node = list_entry(cache->leaves.next,
 224				  struct backref_node, lower);
 225		remove_backref_node(cache, node);
 226	}
 227
 228	cache->last_trans = 0;
 229
 230	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
 231		ASSERT(list_empty(&cache->pending[i]));
 232	ASSERT(list_empty(&cache->changed));
 233	ASSERT(list_empty(&cache->detached));
 234	ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
 235	ASSERT(!cache->nr_nodes);
 236	ASSERT(!cache->nr_edges);
 237}
 238
 239static struct backref_node *alloc_backref_node(struct backref_cache *cache)
 240{
 241	struct backref_node *node;
 242
 243	node = kzalloc(sizeof(*node), GFP_NOFS);
 244	if (node) {
 245		INIT_LIST_HEAD(&node->list);
 246		INIT_LIST_HEAD(&node->upper);
 247		INIT_LIST_HEAD(&node->lower);
 248		RB_CLEAR_NODE(&node->rb_node);
 249		cache->nr_nodes++;
 250	}
 251	return node;
 252}
 253
 254static void free_backref_node(struct backref_cache *cache,
 255			      struct backref_node *node)
 256{
 257	if (node) {
 258		cache->nr_nodes--;
 259		kfree(node);
 260	}
 261}
 262
 263static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
 264{
 265	struct backref_edge *edge;
 266
 267	edge = kzalloc(sizeof(*edge), GFP_NOFS);
 268	if (edge)
 269		cache->nr_edges++;
 270	return edge;
 271}
 272
 273static void free_backref_edge(struct backref_cache *cache,
 274			      struct backref_edge *edge)
 275{
 276	if (edge) {
 277		cache->nr_edges--;
 278		kfree(edge);
 279	}
 280}
 281
 282static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
 283				   struct rb_node *node)
 284{
 285	struct rb_node **p = &root->rb_node;
 286	struct rb_node *parent = NULL;
 287	struct tree_entry *entry;
 288
 289	while (*p) {
 290		parent = *p;
 291		entry = rb_entry(parent, struct tree_entry, rb_node);
 292
 293		if (bytenr < entry->bytenr)
 294			p = &(*p)->rb_left;
 295		else if (bytenr > entry->bytenr)
 296			p = &(*p)->rb_right;
 297		else
 298			return parent;
 299	}
 300
 301	rb_link_node(node, parent, p);
 302	rb_insert_color(node, root);
 303	return NULL;
 304}
 305
 306static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
 307{
 308	struct rb_node *n = root->rb_node;
 309	struct tree_entry *entry;
 310
 311	while (n) {
 312		entry = rb_entry(n, struct tree_entry, rb_node);
 313
 314		if (bytenr < entry->bytenr)
 315			n = n->rb_left;
 316		else if (bytenr > entry->bytenr)
 317			n = n->rb_right;
 318		else
 319			return n;
 320	}
 321	return NULL;
 322}
 323
 324static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
 325{
 326
 327	struct btrfs_fs_info *fs_info = NULL;
 328	struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
 329					      rb_node);
 330	if (bnode->root)
 331		fs_info = bnode->root->fs_info;
 332	btrfs_panic(fs_info, errno,
 333		    "Inconsistency in backref cache found at offset %llu",
 334		    bytenr);
 335}
 336
 337/*
 338 * walk up backref nodes until reach node presents tree root
 339 */
 340static struct backref_node *walk_up_backref(struct backref_node *node,
 341					    struct backref_edge *edges[],
 342					    int *index)
 343{
 344	struct backref_edge *edge;
 345	int idx = *index;
 346
 347	while (!list_empty(&node->upper)) {
 348		edge = list_entry(node->upper.next,
 349				  struct backref_edge, list[LOWER]);
 350		edges[idx++] = edge;
 351		node = edge->node[UPPER];
 352	}
 353	BUG_ON(node->detached);
 354	*index = idx;
 355	return node;
 356}
 357
 358/*
 359 * walk down backref nodes to find start of next reference path
 360 */
 361static struct backref_node *walk_down_backref(struct backref_edge *edges[],
 362					      int *index)
 363{
 364	struct backref_edge *edge;
 365	struct backref_node *lower;
 366	int idx = *index;
 367
 368	while (idx > 0) {
 369		edge = edges[idx - 1];
 370		lower = edge->node[LOWER];
 371		if (list_is_last(&edge->list[LOWER], &lower->upper)) {
 372			idx--;
 373			continue;
 374		}
 375		edge = list_entry(edge->list[LOWER].next,
 376				  struct backref_edge, list[LOWER]);
 377		edges[idx - 1] = edge;
 378		*index = idx;
 379		return edge->node[UPPER];
 380	}
 381	*index = 0;
 382	return NULL;
 383}
 384
 385static void unlock_node_buffer(struct backref_node *node)
 386{
 387	if (node->locked) {
 388		btrfs_tree_unlock(node->eb);
 389		node->locked = 0;
 390	}
 391}
 392
 393static void drop_node_buffer(struct backref_node *node)
 394{
 395	if (node->eb) {
 396		unlock_node_buffer(node);
 397		free_extent_buffer(node->eb);
 398		node->eb = NULL;
 399	}
 400}
 401
 402static void drop_backref_node(struct backref_cache *tree,
 403			      struct backref_node *node)
 404{
 405	BUG_ON(!list_empty(&node->upper));
 406
 407	drop_node_buffer(node);
 408	list_del(&node->list);
 409	list_del(&node->lower);
 410	if (!RB_EMPTY_NODE(&node->rb_node))
 411		rb_erase(&node->rb_node, &tree->rb_root);
 412	free_backref_node(tree, node);
 413}
 414
 415/*
 416 * remove a backref node from the backref cache
 417 */
 418static void remove_backref_node(struct backref_cache *cache,
 419				struct backref_node *node)
 420{
 421	struct backref_node *upper;
 422	struct backref_edge *edge;
 423
 424	if (!node)
 425		return;
 426
 427	BUG_ON(!node->lowest && !node->detached);
 428	while (!list_empty(&node->upper)) {
 429		edge = list_entry(node->upper.next, struct backref_edge,
 430				  list[LOWER]);
 431		upper = edge->node[UPPER];
 432		list_del(&edge->list[LOWER]);
 433		list_del(&edge->list[UPPER]);
 434		free_backref_edge(cache, edge);
 435
 436		if (RB_EMPTY_NODE(&upper->rb_node)) {
 437			BUG_ON(!list_empty(&node->upper));
 438			drop_backref_node(cache, node);
 439			node = upper;
 440			node->lowest = 1;
 441			continue;
 442		}
 443		/*
 444		 * add the node to leaf node list if no other
 445		 * child block cached.
 446		 */
 447		if (list_empty(&upper->lower)) {
 448			list_add_tail(&upper->lower, &cache->leaves);
 449			upper->lowest = 1;
 450		}
 451	}
 452
 453	drop_backref_node(cache, node);
 454}
 455
 456static void update_backref_node(struct backref_cache *cache,
 457				struct backref_node *node, u64 bytenr)
 458{
 459	struct rb_node *rb_node;
 460	rb_erase(&node->rb_node, &cache->rb_root);
 461	node->bytenr = bytenr;
 462	rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
 463	if (rb_node)
 464		backref_tree_panic(rb_node, -EEXIST, bytenr);
 465}
 466
 467/*
 468 * update backref cache after a transaction commit
 469 */
 470static int update_backref_cache(struct btrfs_trans_handle *trans,
 471				struct backref_cache *cache)
 472{
 473	struct backref_node *node;
 474	int level = 0;
 475
 476	if (cache->last_trans == 0) {
 477		cache->last_trans = trans->transid;
 478		return 0;
 479	}
 480
 481	if (cache->last_trans == trans->transid)
 482		return 0;
 483
 484	/*
 485	 * detached nodes are used to avoid unnecessary backref
 486	 * lookup. transaction commit changes the extent tree.
 487	 * so the detached nodes are no longer useful.
 488	 */
 489	while (!list_empty(&cache->detached)) {
 490		node = list_entry(cache->detached.next,
 491				  struct backref_node, list);
 492		remove_backref_node(cache, node);
 493	}
 494
 495	while (!list_empty(&cache->changed)) {
 496		node = list_entry(cache->changed.next,
 497				  struct backref_node, list);
 498		list_del_init(&node->list);
 499		BUG_ON(node->pending);
 500		update_backref_node(cache, node, node->new_bytenr);
 501	}
 502
 503	/*
 504	 * some nodes can be left in the pending list if there were
 505	 * errors during processing the pending nodes.
 506	 */
 507	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
 508		list_for_each_entry(node, &cache->pending[level], list) {
 509			BUG_ON(!node->pending);
 510			if (node->bytenr == node->new_bytenr)
 511				continue;
 512			update_backref_node(cache, node, node->new_bytenr);
 513		}
 514	}
 515
 516	cache->last_trans = 0;
 517	return 1;
 518}
 519
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 520
 521static int should_ignore_root(struct btrfs_root *root)
 522{
 523	struct btrfs_root *reloc_root;
 524
 525	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 526		return 0;
 
 
 
 
 527
 528	reloc_root = root->reloc_root;
 529	if (!reloc_root)
 530		return 0;
 531
 532	if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
 533	    root->fs_info->running_transaction->transid - 1)
 534		return 0;
 535	/*
 536	 * if there is reloc tree and it was created in previous
 537	 * transaction backref lookup can find the reloc tree,
 538	 * so backref node for the fs tree root is useless for
 539	 * relocation.
 540	 */
 541	return 1;
 542}
 
 543/*
 544 * find reloc tree by address of tree root
 545 */
 546static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
 547					  u64 bytenr)
 548{
 
 549	struct rb_node *rb_node;
 550	struct mapping_node *node;
 551	struct btrfs_root *root = NULL;
 552
 
 553	spin_lock(&rc->reloc_root_tree.lock);
 554	rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
 555	if (rb_node) {
 556		node = rb_entry(rb_node, struct mapping_node, rb_node);
 557		root = (struct btrfs_root *)node->data;
 558	}
 559	spin_unlock(&rc->reloc_root_tree.lock);
 560	return root;
 561}
 562
 563static int is_cowonly_root(u64 root_objectid)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 564{
 565	if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
 566	    root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
 567	    root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
 568	    root_objectid == BTRFS_DEV_TREE_OBJECTID ||
 569	    root_objectid == BTRFS_TREE_LOG_OBJECTID ||
 570	    root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
 571	    root_objectid == BTRFS_UUID_TREE_OBJECTID ||
 572	    root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
 573	    root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
 574		return 1;
 575	return 0;
 576}
 577
 578static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
 579					u64 root_objectid)
 580{
 581	struct btrfs_key key;
 582
 583	key.objectid = root_objectid;
 584	key.type = BTRFS_ROOT_ITEM_KEY;
 585	if (is_cowonly_root(root_objectid))
 586		key.offset = 0;
 587	else
 588		key.offset = (u64)-1;
 589
 590	return btrfs_get_fs_root(fs_info, &key, false);
 591}
 592
 593static noinline_for_stack
 594int find_inline_backref(struct extent_buffer *leaf, int slot,
 595			unsigned long *ptr, unsigned long *end)
 596{
 597	struct btrfs_key key;
 598	struct btrfs_extent_item *ei;
 599	struct btrfs_tree_block_info *bi;
 600	u32 item_size;
 601
 602	btrfs_item_key_to_cpu(leaf, &key, slot);
 
 
 
 603
 604	item_size = btrfs_item_size_nr(leaf, slot);
 605	if (item_size < sizeof(*ei)) {
 606		btrfs_print_v0_err(leaf->fs_info);
 607		btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL);
 608		return 1;
 609	}
 610	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 611	WARN_ON(!(btrfs_extent_flags(leaf, ei) &
 612		  BTRFS_EXTENT_FLAG_TREE_BLOCK));
 613
 614	if (key.type == BTRFS_EXTENT_ITEM_KEY &&
 615	    item_size <= sizeof(*ei) + sizeof(*bi)) {
 616		WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
 617		return 1;
 618	}
 619	if (key.type == BTRFS_METADATA_ITEM_KEY &&
 620	    item_size <= sizeof(*ei)) {
 621		WARN_ON(item_size < sizeof(*ei));
 622		return 1;
 623	}
 624
 625	if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 626		bi = (struct btrfs_tree_block_info *)(ei + 1);
 627		*ptr = (unsigned long)(bi + 1);
 628	} else {
 629		*ptr = (unsigned long)(ei + 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 630	}
 631	*end = (unsigned long)ei + item_size;
 632	return 0;
 633}
 634
 635/*
 636 * build backref tree for a given tree block. root of the backref tree
 637 * corresponds the tree block, leaves of the backref tree correspond
 638 * roots of b-trees that reference the tree block.
 639 *
 640 * the basic idea of this function is check backrefs of a given block
 641 * to find upper level blocks that reference the block, and then check
 642 * backrefs of these upper level blocks recursively. the recursion stop
 643 * when tree root is reached or backrefs for the block is cached.
 644 *
 645 * NOTE: if we find backrefs for a block are cached, we know backrefs
 646 * for all upper level blocks that directly/indirectly reference the
 647 * block are also cached.
 648 */
 649static noinline_for_stack
 650struct backref_node *build_backref_tree(struct reloc_control *rc,
 651					struct btrfs_key *node_key,
 652					int level, u64 bytenr)
 653{
 654	struct backref_cache *cache = &rc->backref_cache;
 655	struct btrfs_path *path1; /* For searching extent root */
 656	struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */
 657	struct extent_buffer *eb;
 658	struct btrfs_root *root;
 659	struct backref_node *cur;
 660	struct backref_node *upper;
 661	struct backref_node *lower;
 662	struct backref_node *node = NULL;
 663	struct backref_node *exist = NULL;
 664	struct backref_edge *edge;
 665	struct rb_node *rb_node;
 666	struct btrfs_key key;
 667	unsigned long end;
 668	unsigned long ptr;
 669	LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */
 670	LIST_HEAD(useless);
 671	int cowonly;
 672	int ret;
 673	int err = 0;
 674	bool need_check = true;
 675
 676	path1 = btrfs_alloc_path();
 677	path2 = btrfs_alloc_path();
 678	if (!path1 || !path2) {
 
 
 679		err = -ENOMEM;
 680		goto out;
 681	}
 682	path1->reada = READA_FORWARD;
 683	path2->reada = READA_FORWARD;
 684
 685	node = alloc_backref_node(cache);
 686	if (!node) {
 687		err = -ENOMEM;
 688		goto out;
 689	}
 690
 691	node->bytenr = bytenr;
 692	node->level = level;
 693	node->lowest = 1;
 694	cur = node;
 695again:
 696	end = 0;
 697	ptr = 0;
 698	key.objectid = cur->bytenr;
 699	key.type = BTRFS_METADATA_ITEM_KEY;
 700	key.offset = (u64)-1;
 701
 702	path1->search_commit_root = 1;
 703	path1->skip_locking = 1;
 704	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
 705				0, 0);
 706	if (ret < 0) {
 707		err = ret;
 708		goto out;
 709	}
 710	ASSERT(ret);
 711	ASSERT(path1->slots[0]);
 712
 713	path1->slots[0]--;
 714
 715	WARN_ON(cur->checked);
 716	if (!list_empty(&cur->upper)) {
 717		/*
 718		 * the backref was added previously when processing
 719		 * backref of type BTRFS_TREE_BLOCK_REF_KEY
 720		 */
 721		ASSERT(list_is_singular(&cur->upper));
 722		edge = list_entry(cur->upper.next, struct backref_edge,
 723				  list[LOWER]);
 724		ASSERT(list_empty(&edge->list[UPPER]));
 725		exist = edge->node[UPPER];
 726		/*
 727		 * add the upper level block to pending list if we need
 728		 * check its backrefs
 729		 */
 730		if (!exist->checked)
 731			list_add_tail(&edge->list[UPPER], &list);
 732	} else {
 733		exist = NULL;
 734	}
 735
 736	while (1) {
 737		cond_resched();
 738		eb = path1->nodes[0];
 739
 740		if (ptr >= end) {
 741			if (path1->slots[0] >= btrfs_header_nritems(eb)) {
 742				ret = btrfs_next_leaf(rc->extent_root, path1);
 743				if (ret < 0) {
 744					err = ret;
 745					goto out;
 746				}
 747				if (ret > 0)
 748					break;
 749				eb = path1->nodes[0];
 750			}
 751
 752			btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
 753			if (key.objectid != cur->bytenr) {
 754				WARN_ON(exist);
 755				break;
 756			}
 757
 758			if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 759			    key.type == BTRFS_METADATA_ITEM_KEY) {
 760				ret = find_inline_backref(eb, path1->slots[0],
 761							  &ptr, &end);
 762				if (ret)
 763					goto next;
 764			}
 765		}
 766
 767		if (ptr < end) {
 768			/* update key for inline back ref */
 769			struct btrfs_extent_inline_ref *iref;
 770			int type;
 771			iref = (struct btrfs_extent_inline_ref *)ptr;
 772			type = btrfs_get_extent_inline_ref_type(eb, iref,
 773							BTRFS_REF_TYPE_BLOCK);
 774			if (type == BTRFS_REF_TYPE_INVALID) {
 775				err = -EUCLEAN;
 776				goto out;
 777			}
 778			key.type = type;
 779			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
 780
 781			WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
 782				key.type != BTRFS_SHARED_BLOCK_REF_KEY);
 783		}
 784
 785		/*
 786		 * Parent node found and matches current inline ref, no need to
 787		 * rebuild this node for this inline ref.
 788		 */
 789		if (exist &&
 790		    ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
 791		      exist->owner == key.offset) ||
 792		     (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
 793		      exist->bytenr == key.offset))) {
 794			exist = NULL;
 795			goto next;
 796		}
 797
 798		/* SHARED_BLOCK_REF means key.offset is the parent bytenr */
 799		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
 800			if (key.objectid == key.offset) {
 801				/*
 802				 * Only root blocks of reloc trees use backref
 803				 * pointing to itself.
 804				 */
 805				root = find_reloc_root(rc, cur->bytenr);
 806				ASSERT(root);
 807				cur->root = root;
 808				break;
 809			}
 810
 811			edge = alloc_backref_edge(cache);
 812			if (!edge) {
 813				err = -ENOMEM;
 814				goto out;
 815			}
 816			rb_node = tree_search(&cache->rb_root, key.offset);
 817			if (!rb_node) {
 818				upper = alloc_backref_node(cache);
 819				if (!upper) {
 820					free_backref_edge(cache, edge);
 821					err = -ENOMEM;
 822					goto out;
 823				}
 824				upper->bytenr = key.offset;
 825				upper->level = cur->level + 1;
 826				/*
 827				 *  backrefs for the upper level block isn't
 828				 *  cached, add the block to pending list
 829				 */
 830				list_add_tail(&edge->list[UPPER], &list);
 831			} else {
 832				upper = rb_entry(rb_node, struct backref_node,
 833						 rb_node);
 834				ASSERT(upper->checked);
 835				INIT_LIST_HEAD(&edge->list[UPPER]);
 836			}
 837			list_add_tail(&edge->list[LOWER], &cur->upper);
 838			edge->node[LOWER] = cur;
 839			edge->node[UPPER] = upper;
 840
 841			goto next;
 842		} else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
 843			err = -EINVAL;
 844			btrfs_print_v0_err(rc->extent_root->fs_info);
 845			btrfs_handle_fs_error(rc->extent_root->fs_info, err,
 846					      NULL);
 847			goto out;
 848		} else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
 849			goto next;
 850		}
 851
 852		/*
 853		 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
 854		 * means the root objectid. We need to search the tree to get
 855		 * its parent bytenr.
 856		 */
 857		root = read_fs_root(rc->extent_root->fs_info, key.offset);
 858		if (IS_ERR(root)) {
 859			err = PTR_ERR(root);
 860			goto out;
 861		}
 862
 863		if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 864			cur->cowonly = 1;
 865
 866		if (btrfs_root_level(&root->root_item) == cur->level) {
 867			/* tree root */
 868			ASSERT(btrfs_root_bytenr(&root->root_item) ==
 869			       cur->bytenr);
 870			if (should_ignore_root(root))
 871				list_add(&cur->list, &useless);
 872			else
 873				cur->root = root;
 874			break;
 875		}
 876
 877		level = cur->level + 1;
 878
 879		/* Search the tree to find parent blocks referring the block. */
 880		path2->search_commit_root = 1;
 881		path2->skip_locking = 1;
 882		path2->lowest_level = level;
 883		ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
 884		path2->lowest_level = 0;
 885		if (ret < 0) {
 886			err = ret;
 887			goto out;
 888		}
 889		if (ret > 0 && path2->slots[level] > 0)
 890			path2->slots[level]--;
 891
 892		eb = path2->nodes[level];
 893		if (btrfs_node_blockptr(eb, path2->slots[level]) !=
 894		    cur->bytenr) {
 895			btrfs_err(root->fs_info,
 896	"couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
 897				  cur->bytenr, level - 1,
 898				  root->root_key.objectid,
 899				  node_key->objectid, node_key->type,
 900				  node_key->offset);
 901			err = -ENOENT;
 902			goto out;
 903		}
 904		lower = cur;
 905		need_check = true;
 906
 907		/* Add all nodes and edges in the path */
 908		for (; level < BTRFS_MAX_LEVEL; level++) {
 909			if (!path2->nodes[level]) {
 910				ASSERT(btrfs_root_bytenr(&root->root_item) ==
 911				       lower->bytenr);
 912				if (should_ignore_root(root))
 913					list_add(&lower->list, &useless);
 914				else
 915					lower->root = root;
 916				break;
 917			}
 918
 919			edge = alloc_backref_edge(cache);
 920			if (!edge) {
 921				err = -ENOMEM;
 922				goto out;
 923			}
 924
 925			eb = path2->nodes[level];
 926			rb_node = tree_search(&cache->rb_root, eb->start);
 927			if (!rb_node) {
 928				upper = alloc_backref_node(cache);
 929				if (!upper) {
 930					free_backref_edge(cache, edge);
 931					err = -ENOMEM;
 932					goto out;
 933				}
 934				upper->bytenr = eb->start;
 935				upper->owner = btrfs_header_owner(eb);
 936				upper->level = lower->level + 1;
 937				if (!test_bit(BTRFS_ROOT_REF_COWS,
 938					      &root->state))
 939					upper->cowonly = 1;
 940
 941				/*
 942				 * if we know the block isn't shared
 943				 * we can void checking its backrefs.
 944				 */
 945				if (btrfs_block_can_be_shared(root, eb))
 946					upper->checked = 0;
 947				else
 948					upper->checked = 1;
 949
 950				/*
 951				 * add the block to pending list if we
 952				 * need check its backrefs, we only do this once
 953				 * while walking up a tree as we will catch
 954				 * anything else later on.
 955				 */
 956				if (!upper->checked && need_check) {
 957					need_check = false;
 958					list_add_tail(&edge->list[UPPER],
 959						      &list);
 960				} else {
 961					if (upper->checked)
 962						need_check = true;
 963					INIT_LIST_HEAD(&edge->list[UPPER]);
 964				}
 965			} else {
 966				upper = rb_entry(rb_node, struct backref_node,
 967						 rb_node);
 968				ASSERT(upper->checked);
 969				INIT_LIST_HEAD(&edge->list[UPPER]);
 970				if (!upper->owner)
 971					upper->owner = btrfs_header_owner(eb);
 972			}
 973			list_add_tail(&edge->list[LOWER], &lower->upper);
 974			edge->node[LOWER] = lower;
 975			edge->node[UPPER] = upper;
 976
 977			if (rb_node)
 978				break;
 979			lower = upper;
 980			upper = NULL;
 981		}
 982		btrfs_release_path(path2);
 983next:
 984		if (ptr < end) {
 985			ptr += btrfs_extent_inline_ref_size(key.type);
 986			if (ptr >= end) {
 987				WARN_ON(ptr > end);
 988				ptr = 0;
 989				end = 0;
 990			}
 991		}
 992		if (ptr >= end)
 993			path1->slots[0]++;
 994	}
 995	btrfs_release_path(path1);
 996
 997	cur->checked = 1;
 998	WARN_ON(exist);
 999
1000	/* the pending list isn't empty, take the first block to process */
1001	if (!list_empty(&list)) {
1002		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1003		list_del_init(&edge->list[UPPER]);
1004		cur = edge->node[UPPER];
1005		goto again;
1006	}
1007
1008	/*
1009	 * everything goes well, connect backref nodes and insert backref nodes
1010	 * into the cache.
1011	 */
1012	ASSERT(node->checked);
1013	cowonly = node->cowonly;
1014	if (!cowonly) {
1015		rb_node = tree_insert(&cache->rb_root, node->bytenr,
1016				      &node->rb_node);
1017		if (rb_node)
1018			backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1019		list_add_tail(&node->lower, &cache->leaves);
1020	}
1021
1022	list_for_each_entry(edge, &node->upper, list[LOWER])
1023		list_add_tail(&edge->list[UPPER], &list);
1024
1025	while (!list_empty(&list)) {
1026		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1027		list_del_init(&edge->list[UPPER]);
1028		upper = edge->node[UPPER];
1029		if (upper->detached) {
1030			list_del(&edge->list[LOWER]);
1031			lower = edge->node[LOWER];
1032			free_backref_edge(cache, edge);
1033			if (list_empty(&lower->upper))
1034				list_add(&lower->list, &useless);
1035			continue;
1036		}
1037
1038		if (!RB_EMPTY_NODE(&upper->rb_node)) {
1039			if (upper->lowest) {
1040				list_del_init(&upper->lower);
1041				upper->lowest = 0;
1042			}
1043
1044			list_add_tail(&edge->list[UPPER], &upper->lower);
1045			continue;
1046		}
1047
1048		if (!upper->checked) {
1049			/*
1050			 * Still want to blow up for developers since this is a
1051			 * logic bug.
1052			 */
1053			ASSERT(0);
1054			err = -EINVAL;
1055			goto out;
1056		}
1057		if (cowonly != upper->cowonly) {
1058			ASSERT(0);
1059			err = -EINVAL;
1060			goto out;
1061		}
1062
1063		if (!cowonly) {
1064			rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1065					      &upper->rb_node);
1066			if (rb_node)
1067				backref_tree_panic(rb_node, -EEXIST,
1068						   upper->bytenr);
1069		}
 
1070
1071		list_add_tail(&edge->list[UPPER], &upper->lower);
1072
1073		list_for_each_entry(edge, &upper->upper, list[LOWER])
1074			list_add_tail(&edge->list[UPPER], &list);
 
1075	}
1076	/*
1077	 * process useless backref nodes. backref nodes for tree leaves
1078	 * are deleted from the cache. backref nodes for upper level
1079	 * tree blocks are left in the cache to avoid unnecessary backref
1080	 * lookup.
1081	 */
1082	while (!list_empty(&useless)) {
1083		upper = list_entry(useless.next, struct backref_node, list);
1084		list_del_init(&upper->list);
1085		ASSERT(list_empty(&upper->upper));
1086		if (upper == node)
1087			node = NULL;
1088		if (upper->lowest) {
1089			list_del_init(&upper->lower);
1090			upper->lowest = 0;
1091		}
1092		while (!list_empty(&upper->lower)) {
1093			edge = list_entry(upper->lower.next,
1094					  struct backref_edge, list[UPPER]);
1095			list_del(&edge->list[UPPER]);
1096			list_del(&edge->list[LOWER]);
1097			lower = edge->node[LOWER];
1098			free_backref_edge(cache, edge);
1099
1100			if (list_empty(&lower->upper))
1101				list_add(&lower->list, &useless);
1102		}
1103		__mark_block_processed(rc, upper);
1104		if (upper->level > 0) {
1105			list_add(&upper->list, &cache->detached);
1106			upper->detached = 1;
1107		} else {
1108			rb_erase(&upper->rb_node, &cache->rb_root);
1109			free_backref_node(cache, upper);
1110		}
1111	}
1112out:
1113	btrfs_free_path(path1);
1114	btrfs_free_path(path2);
 
1115	if (err) {
1116		while (!list_empty(&useless)) {
1117			lower = list_entry(useless.next,
1118					   struct backref_node, list);
1119			list_del_init(&lower->list);
1120		}
1121		while (!list_empty(&list)) {
1122			edge = list_first_entry(&list, struct backref_edge,
1123						list[UPPER]);
1124			list_del(&edge->list[UPPER]);
1125			list_del(&edge->list[LOWER]);
1126			lower = edge->node[LOWER];
1127			upper = edge->node[UPPER];
1128			free_backref_edge(cache, edge);
1129
1130			/*
1131			 * Lower is no longer linked to any upper backref nodes
1132			 * and isn't in the cache, we can free it ourselves.
1133			 */
1134			if (list_empty(&lower->upper) &&
1135			    RB_EMPTY_NODE(&lower->rb_node))
1136				list_add(&lower->list, &useless);
1137
1138			if (!RB_EMPTY_NODE(&upper->rb_node))
1139				continue;
1140
1141			/* Add this guy's upper edges to the list to process */
1142			list_for_each_entry(edge, &upper->upper, list[LOWER])
1143				list_add_tail(&edge->list[UPPER], &list);
1144			if (list_empty(&upper->upper))
1145				list_add(&upper->list, &useless);
1146		}
1147
1148		while (!list_empty(&useless)) {
1149			lower = list_entry(useless.next,
1150					   struct backref_node, list);
1151			list_del_init(&lower->list);
1152			if (lower == node)
1153				node = NULL;
1154			free_backref_node(cache, lower);
1155		}
1156
1157		free_backref_node(cache, node);
1158		return ERR_PTR(err);
1159	}
1160	ASSERT(!node || !node->detached);
 
 
1161	return node;
1162}
1163
1164/*
1165 * helper to add backref node for the newly created snapshot.
1166 * the backref node is created by cloning backref node that
1167 * corresponds to root of source tree
1168 */
1169static int clone_backref_node(struct btrfs_trans_handle *trans,
1170			      struct reloc_control *rc,
1171			      struct btrfs_root *src,
1172			      struct btrfs_root *dest)
1173{
1174	struct btrfs_root *reloc_root = src->reloc_root;
1175	struct backref_cache *cache = &rc->backref_cache;
1176	struct backref_node *node = NULL;
1177	struct backref_node *new_node;
1178	struct backref_edge *edge;
1179	struct backref_edge *new_edge;
1180	struct rb_node *rb_node;
1181
1182	if (cache->last_trans > 0)
1183		update_backref_cache(trans, cache);
1184
1185	rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1186	if (rb_node) {
1187		node = rb_entry(rb_node, struct backref_node, rb_node);
1188		if (node->detached)
1189			node = NULL;
1190		else
1191			BUG_ON(node->new_bytenr != reloc_root->node->start);
1192	}
1193
1194	if (!node) {
1195		rb_node = tree_search(&cache->rb_root,
1196				      reloc_root->commit_root->start);
1197		if (rb_node) {
1198			node = rb_entry(rb_node, struct backref_node,
1199					rb_node);
1200			BUG_ON(node->detached);
1201		}
1202	}
1203
1204	if (!node)
1205		return 0;
1206
1207	new_node = alloc_backref_node(cache);
 
1208	if (!new_node)
1209		return -ENOMEM;
1210
1211	new_node->bytenr = dest->node->start;
1212	new_node->level = node->level;
1213	new_node->lowest = node->lowest;
1214	new_node->checked = 1;
1215	new_node->root = dest;
 
1216
1217	if (!node->lowest) {
1218		list_for_each_entry(edge, &node->lower, list[UPPER]) {
1219			new_edge = alloc_backref_edge(cache);
1220			if (!new_edge)
1221				goto fail;
1222
1223			new_edge->node[UPPER] = new_node;
1224			new_edge->node[LOWER] = edge->node[LOWER];
1225			list_add_tail(&new_edge->list[UPPER],
1226				      &new_node->lower);
1227		}
1228	} else {
1229		list_add_tail(&new_node->lower, &cache->leaves);
1230	}
1231
1232	rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1233			      &new_node->rb_node);
1234	if (rb_node)
1235		backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1236
1237	if (!new_node->lowest) {
1238		list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1239			list_add_tail(&new_edge->list[LOWER],
1240				      &new_edge->node[LOWER]->upper);
1241		}
1242	}
1243	return 0;
1244fail:
1245	while (!list_empty(&new_node->lower)) {
1246		new_edge = list_entry(new_node->lower.next,
1247				      struct backref_edge, list[UPPER]);
1248		list_del(&new_edge->list[UPPER]);
1249		free_backref_edge(cache, new_edge);
1250	}
1251	free_backref_node(cache, new_node);
1252	return -ENOMEM;
1253}
1254
1255/*
1256 * helper to add 'address of tree root -> reloc tree' mapping
1257 */
1258static int __must_check __add_reloc_root(struct btrfs_root *root)
1259{
1260	struct btrfs_fs_info *fs_info = root->fs_info;
1261	struct rb_node *rb_node;
1262	struct mapping_node *node;
1263	struct reloc_control *rc = fs_info->reloc_ctl;
1264
1265	node = kmalloc(sizeof(*node), GFP_NOFS);
1266	if (!node)
1267		return -ENOMEM;
1268
1269	node->bytenr = root->node->start;
1270	node->data = root;
1271
1272	spin_lock(&rc->reloc_root_tree.lock);
1273	rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1274			      node->bytenr, &node->rb_node);
1275	spin_unlock(&rc->reloc_root_tree.lock);
1276	if (rb_node) {
1277		btrfs_panic(fs_info, -EEXIST,
1278			    "Duplicate root found for start=%llu while inserting into relocation tree",
1279			    node->bytenr);
 
1280	}
1281
1282	list_add_tail(&root->root_list, &rc->reloc_roots);
1283	return 0;
1284}
1285
1286/*
1287 * helper to delete the 'address of tree root -> reloc tree'
1288 * mapping
1289 */
1290static void __del_reloc_root(struct btrfs_root *root)
1291{
1292	struct btrfs_fs_info *fs_info = root->fs_info;
1293	struct rb_node *rb_node;
1294	struct mapping_node *node = NULL;
1295	struct reloc_control *rc = fs_info->reloc_ctl;
 
1296
1297	if (rc && root->node) {
1298		spin_lock(&rc->reloc_root_tree.lock);
1299		rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1300				      root->node->start);
1301		if (rb_node) {
1302			node = rb_entry(rb_node, struct mapping_node, rb_node);
1303			rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 
1304		}
1305		spin_unlock(&rc->reloc_root_tree.lock);
1306		if (!node)
1307			return;
1308		BUG_ON((struct btrfs_root *)node->data != root);
1309	}
1310
 
 
 
 
 
 
 
 
1311	spin_lock(&fs_info->trans_lock);
1312	list_del_init(&root->root_list);
 
 
 
1313	spin_unlock(&fs_info->trans_lock);
 
 
1314	kfree(node);
1315}
1316
1317/*
1318 * helper to update the 'address of tree root -> reloc tree'
1319 * mapping
1320 */
1321static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1322{
1323	struct btrfs_fs_info *fs_info = root->fs_info;
1324	struct rb_node *rb_node;
1325	struct mapping_node *node = NULL;
1326	struct reloc_control *rc = fs_info->reloc_ctl;
1327
1328	spin_lock(&rc->reloc_root_tree.lock);
1329	rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1330			      root->node->start);
1331	if (rb_node) {
1332		node = rb_entry(rb_node, struct mapping_node, rb_node);
1333		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1334	}
1335	spin_unlock(&rc->reloc_root_tree.lock);
1336
1337	if (!node)
1338		return 0;
1339	BUG_ON((struct btrfs_root *)node->data != root);
1340
1341	spin_lock(&rc->reloc_root_tree.lock);
1342	node->bytenr = new_bytenr;
1343	rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1344			      node->bytenr, &node->rb_node);
1345	spin_unlock(&rc->reloc_root_tree.lock);
1346	if (rb_node)
1347		backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1348	return 0;
1349}
1350
1351static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1352					struct btrfs_root *root, u64 objectid)
1353{
1354	struct btrfs_fs_info *fs_info = root->fs_info;
1355	struct btrfs_root *reloc_root;
1356	struct extent_buffer *eb;
1357	struct btrfs_root_item *root_item;
1358	struct btrfs_key root_key;
1359	int ret;
 
1360
1361	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1362	BUG_ON(!root_item);
 
1363
1364	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1365	root_key.type = BTRFS_ROOT_ITEM_KEY;
1366	root_key.offset = objectid;
1367
1368	if (root->root_key.objectid == objectid) {
1369		u64 commit_root_gen;
1370
1371		/* called by btrfs_init_reloc_root */
1372		ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1373				      BTRFS_TREE_RELOC_OBJECTID);
1374		BUG_ON(ret);
 
 
1375		/*
1376		 * Set the last_snapshot field to the generation of the commit
1377		 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1378		 * correctly (returns true) when the relocation root is created
1379		 * either inside the critical section of a transaction commit
1380		 * (through transaction.c:qgroup_account_snapshot()) and when
1381		 * it's created before the transaction commit is started.
1382		 */
1383		commit_root_gen = btrfs_header_generation(root->commit_root);
1384		btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1385	} else {
1386		/*
1387		 * called by btrfs_reloc_post_snapshot_hook.
1388		 * the source tree is a reloc tree, all tree blocks
1389		 * modified after it was created have RELOC flag
1390		 * set in their headers. so it's OK to not update
1391		 * the 'last_snapshot'.
1392		 */
1393		ret = btrfs_copy_root(trans, root, root->node, &eb,
1394				      BTRFS_TREE_RELOC_OBJECTID);
1395		BUG_ON(ret);
 
1396	}
1397
 
 
 
 
 
 
1398	memcpy(root_item, &root->root_item, sizeof(*root_item));
1399	btrfs_set_root_bytenr(root_item, eb->start);
1400	btrfs_set_root_level(root_item, btrfs_header_level(eb));
1401	btrfs_set_root_generation(root_item, trans->transid);
1402
1403	if (root->root_key.objectid == objectid) {
1404		btrfs_set_root_refs(root_item, 0);
1405		memset(&root_item->drop_progress, 0,
1406		       sizeof(struct btrfs_disk_key));
1407		root_item->drop_level = 0;
1408	}
1409
1410	btrfs_tree_unlock(eb);
1411	free_extent_buffer(eb);
1412
1413	ret = btrfs_insert_root(trans, fs_info->tree_root,
1414				&root_key, root_item);
1415	BUG_ON(ret);
 
 
1416	kfree(root_item);
1417
1418	reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1419	BUG_ON(IS_ERR(reloc_root));
 
 
 
 
1420	reloc_root->last_trans = trans->transid;
1421	return reloc_root;
 
 
 
 
 
 
1422}
1423
1424/*
1425 * create reloc tree for a given fs tree. reloc tree is just a
1426 * snapshot of the fs tree with special root objectid.
 
 
 
1427 */
1428int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1429			  struct btrfs_root *root)
1430{
1431	struct btrfs_fs_info *fs_info = root->fs_info;
1432	struct btrfs_root *reloc_root;
1433	struct reloc_control *rc = fs_info->reloc_ctl;
1434	struct btrfs_block_rsv *rsv;
1435	int clear_rsv = 0;
1436	int ret;
1437
 
 
 
1438	/*
1439	 * The subvolume has reloc tree but the swap is finished, no need to
1440	 * create/update the dead reloc tree
1441	 */
1442	if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
1443		return 0;
1444
 
 
 
 
 
 
 
 
1445	if (root->reloc_root) {
1446		reloc_root = root->reloc_root;
1447		reloc_root->last_trans = trans->transid;
1448		return 0;
1449	}
1450
1451	if (!rc || !rc->create_reloc_tree ||
 
 
 
 
1452	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1453		return 0;
1454
1455	if (!trans->reloc_reserved) {
1456		rsv = trans->block_rsv;
1457		trans->block_rsv = rc->block_rsv;
1458		clear_rsv = 1;
1459	}
1460	reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1461	if (clear_rsv)
1462		trans->block_rsv = rsv;
 
 
1463
1464	ret = __add_reloc_root(reloc_root);
1465	BUG_ON(ret < 0);
1466	root->reloc_root = reloc_root;
 
 
 
 
 
1467	return 0;
1468}
1469
1470/*
1471 * update root item of reloc tree
1472 */
1473int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1474			    struct btrfs_root *root)
1475{
1476	struct btrfs_fs_info *fs_info = root->fs_info;
1477	struct btrfs_root *reloc_root;
1478	struct btrfs_root_item *root_item;
1479	int ret;
1480
1481	if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state) ||
1482	    !root->reloc_root)
1483		goto out;
1484
1485	reloc_root = root->reloc_root;
1486	root_item = &reloc_root->root_item;
1487
 
 
 
 
 
 
 
1488	/* root->reloc_root will stay until current relocation finished */
1489	if (fs_info->reloc_ctl->merge_reloc_tree &&
1490	    btrfs_root_refs(root_item) == 0) {
1491		set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
 
 
 
 
 
1492		__del_reloc_root(reloc_root);
1493	}
1494
1495	if (reloc_root->commit_root != reloc_root->node) {
 
1496		btrfs_set_root_node(root_item, reloc_root->node);
1497		free_extent_buffer(reloc_root->commit_root);
1498		reloc_root->commit_root = btrfs_root_node(reloc_root);
1499	}
1500
1501	ret = btrfs_update_root(trans, fs_info->tree_root,
1502				&reloc_root->root_key, root_item);
1503	BUG_ON(ret);
1504
1505out:
1506	return 0;
1507}
1508
1509/*
1510 * helper to find first cached inode with inode number >= objectid
1511 * in a subvolume
1512 */
1513static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1514{
1515	struct rb_node *node;
1516	struct rb_node *prev;
1517	struct btrfs_inode *entry;
1518	struct inode *inode;
1519
1520	spin_lock(&root->inode_lock);
1521again:
1522	node = root->inode_tree.rb_node;
1523	prev = NULL;
1524	while (node) {
1525		prev = node;
1526		entry = rb_entry(node, struct btrfs_inode, rb_node);
1527
1528		if (objectid < btrfs_ino(entry))
1529			node = node->rb_left;
1530		else if (objectid > btrfs_ino(entry))
1531			node = node->rb_right;
1532		else
1533			break;
1534	}
1535	if (!node) {
1536		while (prev) {
1537			entry = rb_entry(prev, struct btrfs_inode, rb_node);
1538			if (objectid <= btrfs_ino(entry)) {
1539				node = prev;
1540				break;
1541			}
1542			prev = rb_next(prev);
1543		}
1544	}
1545	while (node) {
1546		entry = rb_entry(node, struct btrfs_inode, rb_node);
1547		inode = igrab(&entry->vfs_inode);
1548		if (inode) {
1549			spin_unlock(&root->inode_lock);
1550			return inode;
1551		}
1552
1553		objectid = btrfs_ino(entry) + 1;
1554		if (cond_resched_lock(&root->inode_lock))
1555			goto again;
1556
1557		node = rb_next(node);
1558	}
1559	spin_unlock(&root->inode_lock);
1560	return NULL;
1561}
1562
1563static int in_block_group(u64 bytenr,
1564			  struct btrfs_block_group_cache *block_group)
1565{
1566	if (bytenr >= block_group->key.objectid &&
1567	    bytenr < block_group->key.objectid + block_group->key.offset)
1568		return 1;
1569	return 0;
1570}
1571
1572/*
1573 * get new location of data
1574 */
1575static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1576			    u64 bytenr, u64 num_bytes)
1577{
1578	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1579	struct btrfs_path *path;
1580	struct btrfs_file_extent_item *fi;
1581	struct extent_buffer *leaf;
1582	int ret;
1583
1584	path = btrfs_alloc_path();
1585	if (!path)
1586		return -ENOMEM;
1587
1588	bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1589	ret = btrfs_lookup_file_extent(NULL, root, path,
1590			btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1591	if (ret < 0)
1592		goto out;
1593	if (ret > 0) {
1594		ret = -ENOENT;
1595		goto out;
1596	}
1597
1598	leaf = path->nodes[0];
1599	fi = btrfs_item_ptr(leaf, path->slots[0],
1600			    struct btrfs_file_extent_item);
1601
1602	BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1603	       btrfs_file_extent_compression(leaf, fi) ||
1604	       btrfs_file_extent_encryption(leaf, fi) ||
1605	       btrfs_file_extent_other_encoding(leaf, fi));
1606
1607	if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1608		ret = -EINVAL;
1609		goto out;
1610	}
1611
1612	*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1613	ret = 0;
1614out:
1615	btrfs_free_path(path);
1616	return ret;
1617}
1618
1619/*
1620 * update file extent items in the tree leaf to point to
1621 * the new locations.
1622 */
1623static noinline_for_stack
1624int replace_file_extents(struct btrfs_trans_handle *trans,
1625			 struct reloc_control *rc,
1626			 struct btrfs_root *root,
1627			 struct extent_buffer *leaf)
1628{
1629	struct btrfs_fs_info *fs_info = root->fs_info;
1630	struct btrfs_key key;
1631	struct btrfs_file_extent_item *fi;
1632	struct inode *inode = NULL;
1633	u64 parent;
1634	u64 bytenr;
1635	u64 new_bytenr = 0;
1636	u64 num_bytes;
1637	u64 end;
1638	u32 nritems;
1639	u32 i;
1640	int ret = 0;
1641	int first = 1;
1642	int dirty = 0;
1643
1644	if (rc->stage != UPDATE_DATA_PTRS)
1645		return 0;
1646
1647	/* reloc trees always use full backref */
1648	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1649		parent = leaf->start;
1650	else
1651		parent = 0;
1652
1653	nritems = btrfs_header_nritems(leaf);
1654	for (i = 0; i < nritems; i++) {
1655		struct btrfs_ref ref = { 0 };
1656
1657		cond_resched();
1658		btrfs_item_key_to_cpu(leaf, &key, i);
1659		if (key.type != BTRFS_EXTENT_DATA_KEY)
1660			continue;
1661		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1662		if (btrfs_file_extent_type(leaf, fi) ==
1663		    BTRFS_FILE_EXTENT_INLINE)
1664			continue;
1665		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1666		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1667		if (bytenr == 0)
1668			continue;
1669		if (!in_block_group(bytenr, rc->block_group))
 
1670			continue;
1671
1672		/*
1673		 * if we are modifying block in fs tree, wait for readpage
1674		 * to complete and drop the extent cache
1675		 */
1676		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1677			if (first) {
1678				inode = find_next_inode(root, key.objectid);
1679				first = 0;
1680			} else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1681				btrfs_add_delayed_iput(inode);
1682				inode = find_next_inode(root, key.objectid);
1683			}
1684			if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
 
 
1685				end = key.offset +
1686				      btrfs_file_extent_num_bytes(leaf, fi);
1687				WARN_ON(!IS_ALIGNED(key.offset,
1688						    fs_info->sectorsize));
1689				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1690				end--;
1691				ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1692						      key.offset, end);
 
1693				if (!ret)
1694					continue;
1695
1696				btrfs_drop_extent_cache(BTRFS_I(inode),
1697						key.offset,	end, 1);
1698				unlock_extent(&BTRFS_I(inode)->io_tree,
1699					      key.offset, end);
1700			}
1701		}
1702
1703		ret = get_new_location(rc->data_inode, &new_bytenr,
1704				       bytenr, num_bytes);
1705		if (ret) {
1706			/*
1707			 * Don't have to abort since we've not changed anything
1708			 * in the file extent yet.
1709			 */
1710			break;
1711		}
1712
1713		btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1714		dirty = 1;
1715
1716		key.offset -= btrfs_file_extent_offset(leaf, fi);
1717		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1718				       num_bytes, parent);
1719		ref.real_root = root->root_key.objectid;
1720		btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1721				    key.objectid, key.offset);
 
1722		ret = btrfs_inc_extent_ref(trans, &ref);
1723		if (ret) {
1724			btrfs_abort_transaction(trans, ret);
1725			break;
1726		}
1727
1728		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1729				       num_bytes, parent);
1730		ref.real_root = root->root_key.objectid;
1731		btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1732				    key.objectid, key.offset);
 
1733		ret = btrfs_free_extent(trans, &ref);
1734		if (ret) {
1735			btrfs_abort_transaction(trans, ret);
1736			break;
1737		}
1738	}
1739	if (dirty)
1740		btrfs_mark_buffer_dirty(leaf);
1741	if (inode)
1742		btrfs_add_delayed_iput(inode);
1743	return ret;
1744}
1745
1746static noinline_for_stack
1747int memcmp_node_keys(struct extent_buffer *eb, int slot,
1748		     struct btrfs_path *path, int level)
1749{
1750	struct btrfs_disk_key key1;
1751	struct btrfs_disk_key key2;
1752	btrfs_node_key(eb, &key1, slot);
1753	btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1754	return memcmp(&key1, &key2, sizeof(key1));
1755}
1756
1757/*
1758 * try to replace tree blocks in fs tree with the new blocks
1759 * in reloc tree. tree blocks haven't been modified since the
1760 * reloc tree was create can be replaced.
1761 *
1762 * if a block was replaced, level of the block + 1 is returned.
1763 * if no block got replaced, 0 is returned. if there are other
1764 * errors, a negative error number is returned.
1765 */
1766static noinline_for_stack
1767int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1768		 struct btrfs_root *dest, struct btrfs_root *src,
1769		 struct btrfs_path *path, struct btrfs_key *next_key,
1770		 int lowest_level, int max_level)
1771{
1772	struct btrfs_fs_info *fs_info = dest->fs_info;
1773	struct extent_buffer *eb;
1774	struct extent_buffer *parent;
1775	struct btrfs_ref ref = { 0 };
1776	struct btrfs_key key;
1777	u64 old_bytenr;
1778	u64 new_bytenr;
1779	u64 old_ptr_gen;
1780	u64 new_ptr_gen;
1781	u64 last_snapshot;
1782	u32 blocksize;
1783	int cow = 0;
1784	int level;
1785	int ret;
1786	int slot;
1787
1788	BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1789	BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1790
1791	last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1792again:
1793	slot = path->slots[lowest_level];
1794	btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1795
1796	eb = btrfs_lock_root_node(dest);
1797	btrfs_set_lock_blocking_write(eb);
1798	level = btrfs_header_level(eb);
1799
1800	if (level < lowest_level) {
1801		btrfs_tree_unlock(eb);
1802		free_extent_buffer(eb);
1803		return 0;
1804	}
1805
1806	if (cow) {
1807		ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1808		BUG_ON(ret);
 
 
 
 
 
1809	}
1810	btrfs_set_lock_blocking_write(eb);
1811
1812	if (next_key) {
1813		next_key->objectid = (u64)-1;
1814		next_key->type = (u8)-1;
1815		next_key->offset = (u64)-1;
1816	}
1817
1818	parent = eb;
1819	while (1) {
1820		struct btrfs_key first_key;
1821
1822		level = btrfs_header_level(parent);
1823		BUG_ON(level < lowest_level);
1824
1825		ret = btrfs_bin_search(parent, &key, level, &slot);
1826		if (ret < 0)
1827			break;
1828		if (ret && slot > 0)
1829			slot--;
1830
1831		if (next_key && slot + 1 < btrfs_header_nritems(parent))
1832			btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1833
1834		old_bytenr = btrfs_node_blockptr(parent, slot);
1835		blocksize = fs_info->nodesize;
1836		old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1837		btrfs_node_key_to_cpu(parent, &first_key, slot);
1838
1839		if (level <= max_level) {
1840			eb = path->nodes[level];
1841			new_bytenr = btrfs_node_blockptr(eb,
1842							path->slots[level]);
1843			new_ptr_gen = btrfs_node_ptr_generation(eb,
1844							path->slots[level]);
1845		} else {
1846			new_bytenr = 0;
1847			new_ptr_gen = 0;
1848		}
1849
1850		if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1851			ret = level;
1852			break;
1853		}
1854
1855		if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1856		    memcmp_node_keys(parent, slot, path, level)) {
1857			if (level <= lowest_level) {
1858				ret = 0;
1859				break;
1860			}
1861
1862			eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1863					     level - 1, &first_key);
1864			if (IS_ERR(eb)) {
1865				ret = PTR_ERR(eb);
1866				break;
1867			} else if (!extent_buffer_uptodate(eb)) {
1868				ret = -EIO;
1869				free_extent_buffer(eb);
1870				break;
1871			}
1872			btrfs_tree_lock(eb);
1873			if (cow) {
1874				ret = btrfs_cow_block(trans, dest, eb, parent,
1875						      slot, &eb);
1876				BUG_ON(ret);
 
 
 
 
 
1877			}
1878			btrfs_set_lock_blocking_write(eb);
1879
1880			btrfs_tree_unlock(parent);
1881			free_extent_buffer(parent);
1882
1883			parent = eb;
1884			continue;
1885		}
1886
1887		if (!cow) {
1888			btrfs_tree_unlock(parent);
1889			free_extent_buffer(parent);
1890			cow = 1;
1891			goto again;
1892		}
1893
1894		btrfs_node_key_to_cpu(path->nodes[level], &key,
1895				      path->slots[level]);
1896		btrfs_release_path(path);
1897
1898		path->lowest_level = level;
 
1899		ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
 
1900		path->lowest_level = 0;
1901		BUG_ON(ret);
 
 
 
 
1902
1903		/*
1904		 * Info qgroup to trace both subtrees.
1905		 *
1906		 * We must trace both trees.
1907		 * 1) Tree reloc subtree
1908		 *    If not traced, we will leak data numbers
1909		 * 2) Fs subtree
1910		 *    If not traced, we will double count old data
1911		 *
1912		 * We don't scan the subtree right now, but only record
1913		 * the swapped tree blocks.
1914		 * The real subtree rescan is delayed until we have new
1915		 * CoW on the subtree root node before transaction commit.
1916		 */
1917		ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1918				rc->block_group, parent, slot,
1919				path->nodes[level], path->slots[level],
1920				last_snapshot);
1921		if (ret < 0)
1922			break;
1923		/*
1924		 * swap blocks in fs tree and reloc tree.
1925		 */
1926		btrfs_set_node_blockptr(parent, slot, new_bytenr);
1927		btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1928		btrfs_mark_buffer_dirty(parent);
1929
1930		btrfs_set_node_blockptr(path->nodes[level],
1931					path->slots[level], old_bytenr);
1932		btrfs_set_node_ptr_generation(path->nodes[level],
1933					      path->slots[level], old_ptr_gen);
1934		btrfs_mark_buffer_dirty(path->nodes[level]);
1935
1936		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1937				       blocksize, path->nodes[level]->start);
1938		ref.skip_qgroup = true;
1939		btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
 
1940		ret = btrfs_inc_extent_ref(trans, &ref);
1941		BUG_ON(ret);
 
 
 
1942		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1943				       blocksize, 0);
1944		ref.skip_qgroup = true;
1945		btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1946		ret = btrfs_inc_extent_ref(trans, &ref);
1947		BUG_ON(ret);
 
 
 
1948
 
1949		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1950				       blocksize, path->nodes[level]->start);
1951		btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid);
1952		ref.skip_qgroup = true;
1953		ret = btrfs_free_extent(trans, &ref);
1954		BUG_ON(ret);
 
 
 
1955
 
1956		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1957				       blocksize, 0);
1958		btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid);
1959		ref.skip_qgroup = true;
1960		ret = btrfs_free_extent(trans, &ref);
1961		BUG_ON(ret);
 
 
 
1962
1963		btrfs_unlock_up_safe(path, 0);
1964
1965		ret = level;
1966		break;
1967	}
1968	btrfs_tree_unlock(parent);
1969	free_extent_buffer(parent);
1970	return ret;
1971}
1972
1973/*
1974 * helper to find next relocated block in reloc tree
1975 */
1976static noinline_for_stack
1977int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1978		       int *level)
1979{
1980	struct extent_buffer *eb;
1981	int i;
1982	u64 last_snapshot;
1983	u32 nritems;
1984
1985	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1986
1987	for (i = 0; i < *level; i++) {
1988		free_extent_buffer(path->nodes[i]);
1989		path->nodes[i] = NULL;
1990	}
1991
1992	for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1993		eb = path->nodes[i];
1994		nritems = btrfs_header_nritems(eb);
1995		while (path->slots[i] + 1 < nritems) {
1996			path->slots[i]++;
1997			if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1998			    last_snapshot)
1999				continue;
2000
2001			*level = i;
2002			return 0;
2003		}
2004		free_extent_buffer(path->nodes[i]);
2005		path->nodes[i] = NULL;
2006	}
2007	return 1;
2008}
2009
2010/*
2011 * walk down reloc tree to find relocated block of lowest level
2012 */
2013static noinline_for_stack
2014int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
2015			 int *level)
2016{
2017	struct btrfs_fs_info *fs_info = root->fs_info;
2018	struct extent_buffer *eb = NULL;
2019	int i;
2020	u64 bytenr;
2021	u64 ptr_gen = 0;
2022	u64 last_snapshot;
2023	u32 nritems;
2024
2025	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2026
2027	for (i = *level; i > 0; i--) {
2028		struct btrfs_key first_key;
2029
2030		eb = path->nodes[i];
2031		nritems = btrfs_header_nritems(eb);
2032		while (path->slots[i] < nritems) {
2033			ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
2034			if (ptr_gen > last_snapshot)
2035				break;
2036			path->slots[i]++;
2037		}
2038		if (path->slots[i] >= nritems) {
2039			if (i == *level)
2040				break;
2041			*level = i + 1;
2042			return 0;
2043		}
2044		if (i == 1) {
2045			*level = i;
2046			return 0;
2047		}
2048
2049		bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2050		btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
2051		eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
2052				     &first_key);
2053		if (IS_ERR(eb)) {
2054			return PTR_ERR(eb);
2055		} else if (!extent_buffer_uptodate(eb)) {
2056			free_extent_buffer(eb);
2057			return -EIO;
2058		}
2059		BUG_ON(btrfs_header_level(eb) != i - 1);
2060		path->nodes[i - 1] = eb;
2061		path->slots[i - 1] = 0;
2062	}
2063	return 1;
2064}
2065
2066/*
2067 * invalidate extent cache for file extents whose key in range of
2068 * [min_key, max_key)
2069 */
2070static int invalidate_extent_cache(struct btrfs_root *root,
2071				   struct btrfs_key *min_key,
2072				   struct btrfs_key *max_key)
2073{
2074	struct btrfs_fs_info *fs_info = root->fs_info;
2075	struct inode *inode = NULL;
2076	u64 objectid;
2077	u64 start, end;
2078	u64 ino;
2079
2080	objectid = min_key->objectid;
2081	while (1) {
 
 
2082		cond_resched();
2083		iput(inode);
2084
2085		if (objectid > max_key->objectid)
2086			break;
2087
2088		inode = find_next_inode(root, objectid);
2089		if (!inode)
2090			break;
2091		ino = btrfs_ino(BTRFS_I(inode));
2092
2093		if (ino > max_key->objectid) {
2094			iput(inode);
2095			break;
2096		}
2097
2098		objectid = ino + 1;
2099		if (!S_ISREG(inode->i_mode))
2100			continue;
2101
2102		if (unlikely(min_key->objectid == ino)) {
2103			if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2104				continue;
2105			if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2106				start = 0;
2107			else {
2108				start = min_key->offset;
2109				WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2110			}
2111		} else {
2112			start = 0;
2113		}
2114
2115		if (unlikely(max_key->objectid == ino)) {
2116			if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2117				continue;
2118			if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2119				end = (u64)-1;
2120			} else {
2121				if (max_key->offset == 0)
2122					continue;
2123				end = max_key->offset;
2124				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2125				end--;
2126			}
2127		} else {
2128			end = (u64)-1;
2129		}
2130
2131		/* the lock_extent waits for readpage to complete */
2132		lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2133		btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2134		unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2135	}
2136	return 0;
2137}
2138
2139static int find_next_key(struct btrfs_path *path, int level,
2140			 struct btrfs_key *key)
2141
2142{
2143	while (level < BTRFS_MAX_LEVEL) {
2144		if (!path->nodes[level])
2145			break;
2146		if (path->slots[level] + 1 <
2147		    btrfs_header_nritems(path->nodes[level])) {
2148			btrfs_node_key_to_cpu(path->nodes[level], key,
2149					      path->slots[level] + 1);
2150			return 0;
2151		}
2152		level++;
2153	}
2154	return 1;
2155}
2156
2157/*
2158 * Insert current subvolume into reloc_control::dirty_subvol_roots
2159 */
2160static void insert_dirty_subvol(struct btrfs_trans_handle *trans,
2161				struct reloc_control *rc,
2162				struct btrfs_root *root)
2163{
2164	struct btrfs_root *reloc_root = root->reloc_root;
2165	struct btrfs_root_item *reloc_root_item;
 
2166
2167	/* @root must be a subvolume tree root with a valid reloc tree */
2168	ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
2169	ASSERT(reloc_root);
2170
2171	reloc_root_item = &reloc_root->root_item;
2172	memset(&reloc_root_item->drop_progress, 0,
2173		sizeof(reloc_root_item->drop_progress));
2174	reloc_root_item->drop_level = 0;
2175	btrfs_set_root_refs(reloc_root_item, 0);
2176	btrfs_update_reloc_root(trans, root);
 
 
2177
2178	if (list_empty(&root->reloc_dirty_list)) {
2179		btrfs_grab_fs_root(root);
2180		list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
2181	}
 
 
2182}
2183
2184static int clean_dirty_subvols(struct reloc_control *rc)
2185{
2186	struct btrfs_root *root;
2187	struct btrfs_root *next;
2188	int ret = 0;
2189	int ret2;
2190
2191	list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
2192				 reloc_dirty_list) {
2193		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
2194			/* Merged subvolume, cleanup its reloc root */
2195			struct btrfs_root *reloc_root = root->reloc_root;
2196
2197			list_del_init(&root->reloc_dirty_list);
2198			root->reloc_root = NULL;
 
 
 
 
 
 
2199			if (reloc_root) {
2200
2201				ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1);
2202				if (ret2 < 0 && !ret)
2203					ret = ret2;
 
 
 
 
 
 
 
2204			}
2205			clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
2206			btrfs_put_fs_root(root);
2207		} else {
2208			/* Orphan reloc tree, just clean it up */
2209			ret2 = btrfs_drop_snapshot(root, NULL, 0, 1);
2210			if (ret2 < 0 && !ret)
2211				ret = ret2;
 
 
 
2212		}
2213	}
2214	return ret;
2215}
2216
2217/*
2218 * merge the relocated tree blocks in reloc tree with corresponding
2219 * fs tree.
2220 */
2221static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2222					       struct btrfs_root *root)
2223{
2224	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2225	struct btrfs_key key;
2226	struct btrfs_key next_key;
2227	struct btrfs_trans_handle *trans = NULL;
2228	struct btrfs_root *reloc_root;
2229	struct btrfs_root_item *root_item;
2230	struct btrfs_path *path;
2231	struct extent_buffer *leaf;
 
2232	int level;
2233	int max_level;
2234	int replaced = 0;
2235	int ret;
2236	int err = 0;
2237	u32 min_reserved;
2238
2239	path = btrfs_alloc_path();
2240	if (!path)
2241		return -ENOMEM;
2242	path->reada = READA_FORWARD;
2243
2244	reloc_root = root->reloc_root;
2245	root_item = &reloc_root->root_item;
2246
2247	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2248		level = btrfs_root_level(root_item);
2249		extent_buffer_get(reloc_root->node);
2250		path->nodes[level] = reloc_root->node;
2251		path->slots[level] = 0;
2252	} else {
2253		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2254
2255		level = root_item->drop_level;
2256		BUG_ON(level == 0);
2257		path->lowest_level = level;
2258		ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2259		path->lowest_level = 0;
2260		if (ret < 0) {
2261			btrfs_free_path(path);
2262			return ret;
2263		}
2264
2265		btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2266				      path->slots[level]);
2267		WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2268
2269		btrfs_unlock_up_safe(path, 0);
2270	}
2271
2272	min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
 
 
 
 
 
 
 
 
 
2273	memset(&next_key, 0, sizeof(next_key));
2274
2275	while (1) {
2276		ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2277					     BTRFS_RESERVE_FLUSH_ALL);
2278		if (ret) {
2279			err = ret;
2280			goto out;
2281		}
2282		trans = btrfs_start_transaction(root, 0);
2283		if (IS_ERR(trans)) {
2284			err = PTR_ERR(trans);
2285			trans = NULL;
2286			goto out;
2287		}
 
 
 
 
 
 
 
 
 
 
 
 
2288		trans->block_rsv = rc->block_rsv;
2289
2290		replaced = 0;
2291		max_level = level;
2292
2293		ret = walk_down_reloc_tree(reloc_root, path, &level);
2294		if (ret < 0) {
2295			err = ret;
2296			goto out;
2297		}
2298		if (ret > 0)
2299			break;
2300
2301		if (!find_next_key(path, level, &key) &&
2302		    btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2303			ret = 0;
2304		} else {
2305			ret = replace_path(trans, rc, root, reloc_root, path,
2306					   &next_key, level, max_level);
2307		}
2308		if (ret < 0) {
2309			err = ret;
2310			goto out;
2311		}
2312
2313		if (ret > 0) {
2314			level = ret;
2315			btrfs_node_key_to_cpu(path->nodes[level], &key,
2316					      path->slots[level]);
2317			replaced = 1;
2318		}
2319
2320		ret = walk_up_reloc_tree(reloc_root, path, &level);
2321		if (ret > 0)
2322			break;
2323
2324		BUG_ON(level == 0);
2325		/*
2326		 * save the merging progress in the drop_progress.
2327		 * this is OK since root refs == 1 in this case.
2328		 */
2329		btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2330			       path->slots[level]);
2331		root_item->drop_level = level;
2332
2333		btrfs_end_transaction_throttle(trans);
2334		trans = NULL;
2335
2336		btrfs_btree_balance_dirty(fs_info);
2337
2338		if (replaced && rc->stage == UPDATE_DATA_PTRS)
2339			invalidate_extent_cache(root, &key, &next_key);
2340	}
2341
2342	/*
2343	 * handle the case only one block in the fs tree need to be
2344	 * relocated and the block is tree root.
2345	 */
2346	leaf = btrfs_lock_root_node(root);
2347	ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
 
2348	btrfs_tree_unlock(leaf);
2349	free_extent_buffer(leaf);
2350	if (ret < 0)
2351		err = ret;
2352out:
2353	btrfs_free_path(path);
2354
2355	if (err == 0)
2356		insert_dirty_subvol(trans, rc, root);
 
 
 
2357
2358	if (trans)
2359		btrfs_end_transaction_throttle(trans);
2360
2361	btrfs_btree_balance_dirty(fs_info);
2362
2363	if (replaced && rc->stage == UPDATE_DATA_PTRS)
2364		invalidate_extent_cache(root, &key, &next_key);
2365
2366	return err;
2367}
2368
2369static noinline_for_stack
2370int prepare_to_merge(struct reloc_control *rc, int err)
2371{
2372	struct btrfs_root *root = rc->extent_root;
2373	struct btrfs_fs_info *fs_info = root->fs_info;
2374	struct btrfs_root *reloc_root;
2375	struct btrfs_trans_handle *trans;
2376	LIST_HEAD(reloc_roots);
2377	u64 num_bytes = 0;
2378	int ret;
2379
2380	mutex_lock(&fs_info->reloc_mutex);
2381	rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2382	rc->merging_rsv_size += rc->nodes_relocated * 2;
2383	mutex_unlock(&fs_info->reloc_mutex);
2384
2385again:
2386	if (!err) {
2387		num_bytes = rc->merging_rsv_size;
2388		ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2389					  BTRFS_RESERVE_FLUSH_ALL);
2390		if (ret)
2391			err = ret;
2392	}
2393
2394	trans = btrfs_join_transaction(rc->extent_root);
2395	if (IS_ERR(trans)) {
2396		if (!err)
2397			btrfs_block_rsv_release(fs_info, rc->block_rsv,
2398						num_bytes);
2399		return PTR_ERR(trans);
2400	}
2401
2402	if (!err) {
2403		if (num_bytes != rc->merging_rsv_size) {
2404			btrfs_end_transaction(trans);
2405			btrfs_block_rsv_release(fs_info, rc->block_rsv,
2406						num_bytes);
2407			goto again;
2408		}
2409	}
2410
2411	rc->merge_reloc_tree = 1;
2412
2413	while (!list_empty(&rc->reloc_roots)) {
2414		reloc_root = list_entry(rc->reloc_roots.next,
2415					struct btrfs_root, root_list);
2416		list_del_init(&reloc_root->root_list);
2417
2418		root = read_fs_root(fs_info, reloc_root->root_key.offset);
2419		BUG_ON(IS_ERR(root));
2420		BUG_ON(root->reloc_root != reloc_root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2421
2422		/*
2423		 * set reference count to 1, so btrfs_recover_relocation
2424		 * knows it should resumes merging
2425		 */
2426		if (!err)
2427			btrfs_set_root_refs(&reloc_root->root_item, 1);
2428		btrfs_update_reloc_root(trans, root);
2429
 
 
 
 
2430		list_add(&reloc_root->root_list, &reloc_roots);
 
 
 
 
 
 
 
 
2431	}
2432
2433	list_splice(&reloc_roots, &rc->reloc_roots);
2434
2435	if (!err)
2436		btrfs_commit_transaction(trans);
2437	else
2438		btrfs_end_transaction(trans);
2439	return err;
2440}
2441
2442static noinline_for_stack
2443void free_reloc_roots(struct list_head *list)
2444{
2445	struct btrfs_root *reloc_root;
2446
2447	while (!list_empty(list)) {
2448		reloc_root = list_entry(list->next, struct btrfs_root,
2449					root_list);
2450		__del_reloc_root(reloc_root);
2451		free_extent_buffer(reloc_root->node);
2452		free_extent_buffer(reloc_root->commit_root);
2453		reloc_root->node = NULL;
2454		reloc_root->commit_root = NULL;
2455	}
2456}
2457
2458static noinline_for_stack
2459void merge_reloc_roots(struct reloc_control *rc)
2460{
2461	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2462	struct btrfs_root *root;
2463	struct btrfs_root *reloc_root;
2464	LIST_HEAD(reloc_roots);
2465	int found = 0;
2466	int ret = 0;
2467again:
2468	root = rc->extent_root;
2469
2470	/*
2471	 * this serializes us with btrfs_record_root_in_transaction,
2472	 * we have to make sure nobody is in the middle of
2473	 * adding their roots to the list while we are
2474	 * doing this splice
2475	 */
2476	mutex_lock(&fs_info->reloc_mutex);
2477	list_splice_init(&rc->reloc_roots, &reloc_roots);
2478	mutex_unlock(&fs_info->reloc_mutex);
2479
2480	while (!list_empty(&reloc_roots)) {
2481		found = 1;
2482		reloc_root = list_entry(reloc_roots.next,
2483					struct btrfs_root, root_list);
2484
 
 
2485		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2486			root = read_fs_root(fs_info,
2487					    reloc_root->root_key.offset);
2488			BUG_ON(IS_ERR(root));
2489			BUG_ON(root->reloc_root != reloc_root);
2490
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2491			ret = merge_reloc_root(rc, root);
 
2492			if (ret) {
2493				if (list_empty(&reloc_root->root_list))
2494					list_add_tail(&reloc_root->root_list,
2495						      &reloc_roots);
2496				goto out;
2497			}
2498		} else {
 
 
 
 
 
 
 
 
 
 
2499			list_del_init(&reloc_root->root_list);
2500			/* Don't forget to queue this reloc root for cleanup */
2501			list_add_tail(&reloc_root->reloc_dirty_list,
2502				      &rc->dirty_subvol_roots);
2503		}
2504	}
2505
2506	if (found) {
2507		found = 0;
2508		goto again;
2509	}
2510out:
2511	if (ret) {
2512		btrfs_handle_fs_error(fs_info, ret, NULL);
2513		if (!list_empty(&reloc_roots))
2514			free_reloc_roots(&reloc_roots);
2515
2516		/* new reloc root may be added */
2517		mutex_lock(&fs_info->reloc_mutex);
2518		list_splice_init(&rc->reloc_roots, &reloc_roots);
2519		mutex_unlock(&fs_info->reloc_mutex);
2520		if (!list_empty(&reloc_roots))
2521			free_reloc_roots(&reloc_roots);
2522	}
2523
2524	BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2525}
2526
2527static void free_block_list(struct rb_root *blocks)
2528{
2529	struct tree_block *block;
2530	struct rb_node *rb_node;
2531	while ((rb_node = rb_first(blocks))) {
2532		block = rb_entry(rb_node, struct tree_block, rb_node);
2533		rb_erase(rb_node, blocks);
2534		kfree(block);
2535	}
2536}
2537
2538static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2539				      struct btrfs_root *reloc_root)
2540{
2541	struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2542	struct btrfs_root *root;
 
2543
2544	if (reloc_root->last_trans == trans->transid)
2545		return 0;
2546
2547	root = read_fs_root(fs_info, reloc_root->root_key.offset);
2548	BUG_ON(IS_ERR(root));
2549	BUG_ON(root->reloc_root != reloc_root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2550
2551	return btrfs_record_root_in_trans(trans, root);
2552}
2553
2554static noinline_for_stack
2555struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2556				     struct reloc_control *rc,
2557				     struct backref_node *node,
2558				     struct backref_edge *edges[])
2559{
2560	struct backref_node *next;
2561	struct btrfs_root *root;
2562	int index = 0;
 
2563
2564	next = node;
2565	while (1) {
2566		cond_resched();
2567		next = walk_up_backref(next, edges, &index);
2568		root = next->root;
2569		BUG_ON(!root);
2570		BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2571
2572		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2573			record_reloc_root_in_trans(trans, root);
 
 
2574			break;
2575		}
2576
2577		btrfs_record_root_in_trans(trans, root);
 
 
2578		root = root->reloc_root;
2579
 
 
 
 
 
 
 
2580		if (next->new_bytenr != root->node->start) {
2581			BUG_ON(next->new_bytenr);
2582			BUG_ON(!list_empty(&next->list));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2583			next->new_bytenr = root->node->start;
2584			next->root = root;
 
 
2585			list_add_tail(&next->list,
2586				      &rc->backref_cache.changed);
2587			__mark_block_processed(rc, next);
2588			break;
2589		}
2590
2591		WARN_ON(1);
2592		root = NULL;
2593		next = walk_down_backref(edges, &index);
2594		if (!next || next->level <= node->level)
2595			break;
2596	}
2597	if (!root)
2598		return NULL;
 
 
 
 
 
 
2599
2600	next = node;
2601	/* setup backref node path for btrfs_reloc_cow_block */
2602	while (1) {
2603		rc->backref_cache.path[next->level] = next;
2604		if (--index < 0)
2605			break;
2606		next = edges[index]->node[UPPER];
2607	}
2608	return root;
2609}
2610
2611/*
2612 * select a tree root for relocation. return NULL if the block
2613 * is reference counted. we should use do_relocation() in this
2614 * case. return a tree root pointer if the block isn't reference
2615 * counted. return -ENOENT if the block is root of reloc tree.
 
 
 
2616 */
2617static noinline_for_stack
2618struct btrfs_root *select_one_root(struct backref_node *node)
2619{
2620	struct backref_node *next;
2621	struct btrfs_root *root;
2622	struct btrfs_root *fs_root = NULL;
2623	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2624	int index = 0;
2625
2626	next = node;
2627	while (1) {
2628		cond_resched();
2629		next = walk_up_backref(next, edges, &index);
2630		root = next->root;
2631		BUG_ON(!root);
2632
2633		/* no other choice for non-references counted tree */
2634		if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 
 
 
 
 
 
 
2635			return root;
2636
2637		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2638			fs_root = root;
2639
2640		if (next != node)
2641			return NULL;
2642
2643		next = walk_down_backref(edges, &index);
2644		if (!next || next->level <= node->level)
2645			break;
2646	}
2647
2648	if (!fs_root)
2649		return ERR_PTR(-ENOENT);
2650	return fs_root;
2651}
2652
2653static noinline_for_stack
2654u64 calcu_metadata_size(struct reloc_control *rc,
2655			struct backref_node *node, int reserve)
2656{
2657	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2658	struct backref_node *next = node;
2659	struct backref_edge *edge;
2660	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2661	u64 num_bytes = 0;
2662	int index = 0;
2663
2664	BUG_ON(reserve && node->processed);
2665
2666	while (next) {
2667		cond_resched();
2668		while (1) {
2669			if (next->processed && (reserve || next != node))
2670				break;
2671
2672			num_bytes += fs_info->nodesize;
2673
2674			if (list_empty(&next->upper))
2675				break;
2676
2677			edge = list_entry(next->upper.next,
2678					  struct backref_edge, list[LOWER]);
2679			edges[index++] = edge;
2680			next = edge->node[UPPER];
2681		}
2682		next = walk_down_backref(edges, &index);
2683	}
2684	return num_bytes;
2685}
2686
2687static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2688				  struct reloc_control *rc,
2689				  struct backref_node *node)
2690{
2691	struct btrfs_root *root = rc->extent_root;
2692	struct btrfs_fs_info *fs_info = root->fs_info;
2693	u64 num_bytes;
2694	int ret;
2695	u64 tmp;
2696
2697	num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2698
2699	trans->block_rsv = rc->block_rsv;
2700	rc->reserved_bytes += num_bytes;
2701
2702	/*
2703	 * We are under a transaction here so we can only do limited flushing.
2704	 * If we get an enospc just kick back -EAGAIN so we know to drop the
2705	 * transaction and try to refill when we can flush all the things.
2706	 */
2707	ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2708				BTRFS_RESERVE_FLUSH_LIMIT);
2709	if (ret) {
2710		tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2711		while (tmp <= rc->reserved_bytes)
2712			tmp <<= 1;
2713		/*
2714		 * only one thread can access block_rsv at this point,
2715		 * so we don't need hold lock to protect block_rsv.
2716		 * we expand more reservation size here to allow enough
2717		 * space for relocation and we will return earlier in
2718		 * enospc case.
2719		 */
2720		rc->block_rsv->size = tmp + fs_info->nodesize *
2721				      RELOCATION_RESERVED_NODES;
2722		return -EAGAIN;
2723	}
2724
2725	return 0;
2726}
2727
2728/*
2729 * relocate a block tree, and then update pointers in upper level
2730 * blocks that reference the block to point to the new location.
2731 *
2732 * if called by link_to_upper, the block has already been relocated.
2733 * in that case this function just updates pointers.
2734 */
2735static int do_relocation(struct btrfs_trans_handle *trans,
2736			 struct reloc_control *rc,
2737			 struct backref_node *node,
2738			 struct btrfs_key *key,
2739			 struct btrfs_path *path, int lowest)
2740{
2741	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2742	struct backref_node *upper;
2743	struct backref_edge *edge;
2744	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2745	struct btrfs_root *root;
2746	struct extent_buffer *eb;
2747	u32 blocksize;
2748	u64 bytenr;
2749	u64 generation;
2750	int slot;
2751	int ret;
2752	int err = 0;
2753
2754	BUG_ON(lowest && node->eb);
 
 
 
 
2755
2756	path->lowest_level = node->level + 1;
2757	rc->backref_cache.path[node->level] = node;
2758	list_for_each_entry(edge, &node->upper, list[LOWER]) {
2759		struct btrfs_key first_key;
2760		struct btrfs_ref ref = { 0 };
2761
2762		cond_resched();
2763
2764		upper = edge->node[UPPER];
2765		root = select_reloc_root(trans, rc, upper, edges);
2766		BUG_ON(!root);
 
 
 
2767
2768		if (upper->eb && !upper->locked) {
2769			if (!lowest) {
2770				ret = btrfs_bin_search(upper->eb, key,
2771						       upper->level, &slot);
2772				if (ret < 0) {
2773					err = ret;
2774					goto next;
2775				}
2776				BUG_ON(ret);
2777				bytenr = btrfs_node_blockptr(upper->eb, slot);
2778				if (node->eb->start == bytenr)
2779					goto next;
2780			}
2781			drop_node_buffer(upper);
2782		}
2783
2784		if (!upper->eb) {
2785			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2786			if (ret) {
2787				if (ret < 0)
2788					err = ret;
2789				else
2790					err = -ENOENT;
2791
2792				btrfs_release_path(path);
2793				break;
2794			}
2795
2796			if (!upper->eb) {
2797				upper->eb = path->nodes[upper->level];
2798				path->nodes[upper->level] = NULL;
2799			} else {
2800				BUG_ON(upper->eb != path->nodes[upper->level]);
2801			}
2802
2803			upper->locked = 1;
2804			path->locks[upper->level] = 0;
2805
2806			slot = path->slots[upper->level];
2807			btrfs_release_path(path);
2808		} else {
2809			ret = btrfs_bin_search(upper->eb, key, upper->level,
2810					       &slot);
2811			if (ret < 0) {
2812				err = ret;
2813				goto next;
2814			}
2815			BUG_ON(ret);
2816		}
2817
2818		bytenr = btrfs_node_blockptr(upper->eb, slot);
2819		if (lowest) {
2820			if (bytenr != node->bytenr) {
2821				btrfs_err(root->fs_info,
2822		"lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2823					  bytenr, node->bytenr, slot,
2824					  upper->eb->start);
2825				err = -EIO;
2826				goto next;
2827			}
2828		} else {
2829			if (node->eb->start == bytenr)
2830				goto next;
2831		}
2832
2833		blocksize = root->fs_info->nodesize;
2834		generation = btrfs_node_ptr_generation(upper->eb, slot);
2835		btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2836		eb = read_tree_block(fs_info, bytenr, generation,
2837				     upper->level - 1, &first_key);
2838		if (IS_ERR(eb)) {
2839			err = PTR_ERR(eb);
2840			goto next;
2841		} else if (!extent_buffer_uptodate(eb)) {
2842			free_extent_buffer(eb);
2843			err = -EIO;
2844			goto next;
2845		}
2846		btrfs_tree_lock(eb);
2847		btrfs_set_lock_blocking_write(eb);
2848
2849		if (!node->eb) {
2850			ret = btrfs_cow_block(trans, root, eb, upper->eb,
2851					      slot, &eb);
2852			btrfs_tree_unlock(eb);
2853			free_extent_buffer(eb);
2854			if (ret < 0) {
2855				err = ret;
2856				goto next;
2857			}
2858			BUG_ON(node->eb != eb);
 
 
 
2859		} else {
2860			btrfs_set_node_blockptr(upper->eb, slot,
2861						node->eb->start);
2862			btrfs_set_node_ptr_generation(upper->eb, slot,
2863						      trans->transid);
2864			btrfs_mark_buffer_dirty(upper->eb);
2865
2866			btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2867					       node->eb->start, blocksize,
2868					       upper->eb->start);
2869			ref.real_root = root->root_key.objectid;
2870			btrfs_init_tree_ref(&ref, node->level,
2871					    btrfs_header_owner(upper->eb));
 
2872			ret = btrfs_inc_extent_ref(trans, &ref);
2873			BUG_ON(ret);
2874
2875			ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2876			BUG_ON(ret);
 
2877		}
2878next:
2879		if (!upper->pending)
2880			drop_node_buffer(upper);
2881		else
2882			unlock_node_buffer(upper);
2883		if (err)
2884			break;
2885	}
2886
2887	if (!err && node->pending) {
2888		drop_node_buffer(node);
2889		list_move_tail(&node->list, &rc->backref_cache.changed);
2890		node->pending = 0;
2891	}
2892
2893	path->lowest_level = 0;
2894	BUG_ON(err == -ENOSPC);
2895	return err;
 
 
 
 
 
2896}
2897
2898static int link_to_upper(struct btrfs_trans_handle *trans,
2899			 struct reloc_control *rc,
2900			 struct backref_node *node,
2901			 struct btrfs_path *path)
2902{
2903	struct btrfs_key key;
2904
2905	btrfs_node_key_to_cpu(node->eb, &key, 0);
2906	return do_relocation(trans, rc, node, &key, path, 0);
2907}
2908
2909static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2910				struct reloc_control *rc,
2911				struct btrfs_path *path, int err)
2912{
2913	LIST_HEAD(list);
2914	struct backref_cache *cache = &rc->backref_cache;
2915	struct backref_node *node;
2916	int level;
2917	int ret;
2918
2919	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2920		while (!list_empty(&cache->pending[level])) {
2921			node = list_entry(cache->pending[level].next,
2922					  struct backref_node, list);
2923			list_move_tail(&node->list, &list);
2924			BUG_ON(!node->pending);
2925
2926			if (!err) {
2927				ret = link_to_upper(trans, rc, node, path);
2928				if (ret < 0)
2929					err = ret;
2930			}
2931		}
2932		list_splice_init(&list, &cache->pending[level]);
2933	}
2934	return err;
2935}
2936
2937static void mark_block_processed(struct reloc_control *rc,
2938				 u64 bytenr, u32 blocksize)
2939{
2940	set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2941			EXTENT_DIRTY);
2942}
2943
2944static void __mark_block_processed(struct reloc_control *rc,
2945				   struct backref_node *node)
2946{
2947	u32 blocksize;
2948	if (node->level == 0 ||
2949	    in_block_group(node->bytenr, rc->block_group)) {
2950		blocksize = rc->extent_root->fs_info->nodesize;
2951		mark_block_processed(rc, node->bytenr, blocksize);
2952	}
2953	node->processed = 1;
2954}
2955
2956/*
2957 * mark a block and all blocks directly/indirectly reference the block
2958 * as processed.
2959 */
2960static void update_processed_blocks(struct reloc_control *rc,
2961				    struct backref_node *node)
2962{
2963	struct backref_node *next = node;
2964	struct backref_edge *edge;
2965	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2966	int index = 0;
2967
2968	while (next) {
2969		cond_resched();
2970		while (1) {
2971			if (next->processed)
2972				break;
2973
2974			__mark_block_processed(rc, next);
2975
2976			if (list_empty(&next->upper))
2977				break;
2978
2979			edge = list_entry(next->upper.next,
2980					  struct backref_edge, list[LOWER]);
2981			edges[index++] = edge;
2982			next = edge->node[UPPER];
2983		}
2984		next = walk_down_backref(edges, &index);
2985	}
2986}
2987
2988static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2989{
2990	u32 blocksize = rc->extent_root->fs_info->nodesize;
2991
2992	if (test_range_bit(&rc->processed_blocks, bytenr,
2993			   bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2994		return 1;
2995	return 0;
2996}
2997
2998static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2999			      struct tree_block *block)
3000{
 
 
 
 
 
3001	struct extent_buffer *eb;
3002
3003	BUG_ON(block->key_ready);
3004	eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
3005			     block->level, NULL);
3006	if (IS_ERR(eb)) {
3007		return PTR_ERR(eb);
3008	} else if (!extent_buffer_uptodate(eb)) {
3009		free_extent_buffer(eb);
3010		return -EIO;
3011	}
3012	if (block->level == 0)
3013		btrfs_item_key_to_cpu(eb, &block->key, 0);
3014	else
3015		btrfs_node_key_to_cpu(eb, &block->key, 0);
3016	free_extent_buffer(eb);
3017	block->key_ready = 1;
3018	return 0;
3019}
3020
3021/*
3022 * helper function to relocate a tree block
3023 */
3024static int relocate_tree_block(struct btrfs_trans_handle *trans,
3025				struct reloc_control *rc,
3026				struct backref_node *node,
3027				struct btrfs_key *key,
3028				struct btrfs_path *path)
3029{
3030	struct btrfs_root *root;
3031	int ret = 0;
3032
3033	if (!node)
3034		return 0;
3035
 
 
 
 
 
 
 
 
3036	BUG_ON(node->processed);
3037	root = select_one_root(node);
3038	if (root == ERR_PTR(-ENOENT)) {
3039		update_processed_blocks(rc, node);
3040		goto out;
3041	}
3042
3043	if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
3044		ret = reserve_metadata_space(trans, rc, node);
3045		if (ret)
3046			goto out;
 
 
 
3047	}
3048
3049	if (root) {
3050		if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
3051			BUG_ON(node->new_bytenr);
3052			BUG_ON(!list_empty(&node->list));
3053			btrfs_record_root_in_trans(trans, root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3054			root = root->reloc_root;
3055			node->new_bytenr = root->node->start;
3056			node->root = root;
 
 
3057			list_add_tail(&node->list, &rc->backref_cache.changed);
3058		} else {
3059			path->lowest_level = node->level;
 
 
3060			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
3061			btrfs_release_path(path);
 
 
3062			if (ret > 0)
3063				ret = 0;
3064		}
3065		if (!ret)
3066			update_processed_blocks(rc, node);
3067	} else {
3068		ret = do_relocation(trans, rc, node, key, path, 1);
3069	}
3070out:
3071	if (ret || node->level == 0 || node->cowonly)
3072		remove_backref_node(&rc->backref_cache, node);
3073	return ret;
3074}
3075
3076/*
3077 * relocate a list of blocks
3078 */
3079static noinline_for_stack
3080int relocate_tree_blocks(struct btrfs_trans_handle *trans,
3081			 struct reloc_control *rc, struct rb_root *blocks)
3082{
3083	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3084	struct backref_node *node;
3085	struct btrfs_path *path;
3086	struct tree_block *block;
3087	struct tree_block *next;
3088	int ret;
3089	int err = 0;
3090
3091	path = btrfs_alloc_path();
3092	if (!path) {
3093		err = -ENOMEM;
3094		goto out_free_blocks;
3095	}
3096
3097	/* Kick in readahead for tree blocks with missing keys */
3098	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3099		if (!block->key_ready)
3100			readahead_tree_block(fs_info, block->bytenr);
 
 
3101	}
3102
3103	/* Get first keys */
3104	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3105		if (!block->key_ready) {
3106			err = get_tree_block_key(fs_info, block);
3107			if (err)
3108				goto out_free_path;
3109		}
3110	}
3111
3112	/* Do tree relocation */
3113	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
3114		node = build_backref_tree(rc, &block->key,
3115					  block->level, block->bytenr);
3116		if (IS_ERR(node)) {
3117			err = PTR_ERR(node);
3118			goto out;
3119		}
3120
3121		ret = relocate_tree_block(trans, rc, node, &block->key,
3122					  path);
3123		if (ret < 0) {
3124			if (ret != -EAGAIN || &block->rb_node == rb_first(blocks))
3125				err = ret;
3126			goto out;
3127		}
3128	}
3129out:
3130	err = finish_pending_nodes(trans, rc, path, err);
3131
3132out_free_path:
3133	btrfs_free_path(path);
3134out_free_blocks:
3135	free_block_list(blocks);
3136	return err;
3137}
3138
3139static noinline_for_stack
3140int prealloc_file_extent_cluster(struct inode *inode,
3141				 struct file_extent_cluster *cluster)
3142{
3143	u64 alloc_hint = 0;
3144	u64 start;
3145	u64 end;
3146	u64 offset = BTRFS_I(inode)->index_cnt;
3147	u64 num_bytes;
3148	int nr = 0;
3149	int ret = 0;
 
3150	u64 prealloc_start = cluster->start - offset;
3151	u64 prealloc_end = cluster->end - offset;
3152	u64 cur_offset;
3153	struct extent_changeset *data_reserved = NULL;
3154
3155	BUG_ON(cluster->start != cluster->boundary[0]);
3156	inode_lock(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3157
3158	ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3159					  prealloc_end + 1 - prealloc_start);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3160	if (ret)
3161		goto out;
 
 
 
 
3162
3163	cur_offset = prealloc_start;
3164	while (nr < cluster->nr) {
3165		start = cluster->boundary[nr] - offset;
3166		if (nr + 1 < cluster->nr)
3167			end = cluster->boundary[nr + 1] - 1 - offset;
3168		else
3169			end = cluster->end - offset;
3170
3171		lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3172		num_bytes = end + 1 - start;
3173		if (cur_offset < start)
3174			btrfs_free_reserved_data_space(inode, data_reserved,
3175					cur_offset, start - cur_offset);
3176		ret = btrfs_prealloc_file_range(inode, 0, start,
3177						num_bytes, num_bytes,
3178						end + 1, &alloc_hint);
3179		cur_offset = end + 1;
3180		unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3181		if (ret)
3182			break;
3183		nr++;
3184	}
 
 
3185	if (cur_offset < prealloc_end)
3186		btrfs_free_reserved_data_space(inode, data_reserved,
3187				cur_offset, prealloc_end + 1 - cur_offset);
3188out:
3189	inode_unlock(inode);
3190	extent_changeset_free(data_reserved);
3191	return ret;
3192}
3193
3194static noinline_for_stack
3195int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3196			 u64 block_start)
3197{
3198	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3199	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3200	struct extent_map *em;
 
3201	int ret = 0;
3202
3203	em = alloc_extent_map();
3204	if (!em)
3205		return -ENOMEM;
3206
3207	em->start = start;
3208	em->len = end + 1 - start;
3209	em->block_len = em->len;
3210	em->block_start = block_start;
3211	em->bdev = fs_info->fs_devices->latest_bdev;
3212	set_bit(EXTENT_FLAG_PINNED, &em->flags);
3213
3214	lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3215	while (1) {
3216		write_lock(&em_tree->lock);
3217		ret = add_extent_mapping(em_tree, em, 0);
3218		write_unlock(&em_tree->lock);
3219		if (ret != -EEXIST) {
3220			free_extent_map(em);
3221			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3222		}
3223		btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3224	}
3225	unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
 
 
 
 
 
 
 
 
 
 
 
3226	return ret;
3227}
3228
3229static int relocate_file_extent_cluster(struct inode *inode,
3230					struct file_extent_cluster *cluster)
3231{
3232	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3233	u64 page_start;
3234	u64 page_end;
3235	u64 offset = BTRFS_I(inode)->index_cnt;
3236	unsigned long index;
3237	unsigned long last_index;
3238	struct page *page;
3239	struct file_ra_state *ra;
3240	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3241	int nr = 0;
3242	int ret = 0;
3243
3244	if (!cluster->nr)
3245		return 0;
3246
3247	ra = kzalloc(sizeof(*ra), GFP_NOFS);
3248	if (!ra)
3249		return -ENOMEM;
3250
3251	ret = prealloc_file_extent_cluster(inode, cluster);
3252	if (ret)
3253		goto out;
3254
3255	file_ra_state_init(ra, inode->i_mapping);
3256
3257	ret = setup_extent_mapping(inode, cluster->start - offset,
3258				   cluster->end - offset, cluster->start);
3259	if (ret)
3260		goto out;
3261
3262	index = (cluster->start - offset) >> PAGE_SHIFT;
3263	last_index = (cluster->end - offset) >> PAGE_SHIFT;
3264	while (index <= last_index) {
3265		ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3266				PAGE_SIZE);
3267		if (ret)
3268			goto out;
3269
3270		page = find_lock_page(inode->i_mapping, index);
3271		if (!page) {
3272			page_cache_sync_readahead(inode->i_mapping,
3273						  ra, NULL, index,
3274						  last_index + 1 - index);
3275			page = find_or_create_page(inode->i_mapping, index,
3276						   mask);
3277			if (!page) {
3278				btrfs_delalloc_release_metadata(BTRFS_I(inode),
3279							PAGE_SIZE, true);
3280				btrfs_delalloc_release_extents(BTRFS_I(inode),
3281							PAGE_SIZE);
3282				ret = -ENOMEM;
3283				goto out;
3284			}
3285		}
3286
3287		if (PageReadahead(page)) {
3288			page_cache_async_readahead(inode->i_mapping,
3289						   ra, NULL, page, index,
3290						   last_index + 1 - index);
3291		}
3292
3293		if (!PageUptodate(page)) {
3294			btrfs_readpage(NULL, page);
3295			lock_page(page);
3296			if (!PageUptodate(page)) {
3297				unlock_page(page);
3298				put_page(page);
3299				btrfs_delalloc_release_metadata(BTRFS_I(inode),
3300							PAGE_SIZE, true);
3301				btrfs_delalloc_release_extents(BTRFS_I(inode),
3302							       PAGE_SIZE);
3303				ret = -EIO;
3304				goto out;
3305			}
3306		}
3307
3308		page_start = page_offset(page);
3309		page_end = page_start + PAGE_SIZE - 1;
3310
3311		lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3312
3313		set_page_extent_mapped(page);
3314
3315		if (nr < cluster->nr &&
3316		    page_start + offset == cluster->boundary[nr]) {
3317			set_extent_bits(&BTRFS_I(inode)->io_tree,
3318					page_start, page_end,
3319					EXTENT_BOUNDARY);
3320			nr++;
3321		}
3322
3323		ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
3324						NULL);
3325		if (ret) {
3326			unlock_page(page);
3327			put_page(page);
3328			btrfs_delalloc_release_metadata(BTRFS_I(inode),
3329							 PAGE_SIZE, true);
3330			btrfs_delalloc_release_extents(BTRFS_I(inode),
3331			                               PAGE_SIZE);
3332
3333			clear_extent_bits(&BTRFS_I(inode)->io_tree,
3334					  page_start, page_end,
3335					  EXTENT_LOCKED | EXTENT_BOUNDARY);
3336			goto out;
3337
3338		}
3339		set_page_dirty(page);
3340
3341		unlock_extent(&BTRFS_I(inode)->io_tree,
3342			      page_start, page_end);
3343		unlock_page(page);
3344		put_page(page);
3345
3346		index++;
3347		btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE);
3348		balance_dirty_pages_ratelimited(inode->i_mapping);
3349		btrfs_throttle(fs_info);
3350	}
3351	WARN_ON(nr != cluster->nr);
3352out:
3353	kfree(ra);
3354	return ret;
3355}
3356
3357static noinline_for_stack
3358int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3359			 struct file_extent_cluster *cluster)
3360{
3361	int ret;
 
3362
3363	if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3364		ret = relocate_file_extent_cluster(inode, cluster);
3365		if (ret)
3366			return ret;
3367		cluster->nr = 0;
3368	}
3369
3370	if (!cluster->nr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3371		cluster->start = extent_key->objectid;
 
 
3372	else
3373		BUG_ON(cluster->nr >= MAX_EXTENTS);
3374	cluster->end = extent_key->objectid + extent_key->offset - 1;
3375	cluster->boundary[cluster->nr] = extent_key->objectid;
3376	cluster->nr++;
3377
3378	if (cluster->nr >= MAX_EXTENTS) {
3379		ret = relocate_file_extent_cluster(inode, cluster);
3380		if (ret)
3381			return ret;
3382		cluster->nr = 0;
3383	}
3384	return 0;
3385}
3386
3387/*
3388 * helper to add a tree block to the list.
3389 * the major work is getting the generation and level of the block
3390 */
3391static int add_tree_block(struct reloc_control *rc,
3392			  struct btrfs_key *extent_key,
3393			  struct btrfs_path *path,
3394			  struct rb_root *blocks)
3395{
3396	struct extent_buffer *eb;
3397	struct btrfs_extent_item *ei;
3398	struct btrfs_tree_block_info *bi;
3399	struct tree_block *block;
3400	struct rb_node *rb_node;
3401	u32 item_size;
3402	int level = -1;
3403	u64 generation;
 
3404
3405	eb =  path->nodes[0];
3406	item_size = btrfs_item_size_nr(eb, path->slots[0]);
3407
3408	if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3409	    item_size >= sizeof(*ei) + sizeof(*bi)) {
 
 
3410		ei = btrfs_item_ptr(eb, path->slots[0],
3411				struct btrfs_extent_item);
 
3412		if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3413			bi = (struct btrfs_tree_block_info *)(ei + 1);
3414			level = btrfs_tree_block_level(eb, bi);
 
3415		} else {
3416			level = (int)extent_key->offset;
 
3417		}
3418		generation = btrfs_extent_generation(eb, ei);
3419	} else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3420		btrfs_print_v0_err(eb->fs_info);
3421		btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3422		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3423	} else {
3424		BUG();
 
 
 
 
 
3425	}
3426
3427	btrfs_release_path(path);
3428
3429	BUG_ON(level == -1);
3430
3431	block = kmalloc(sizeof(*block), GFP_NOFS);
3432	if (!block)
3433		return -ENOMEM;
3434
3435	block->bytenr = extent_key->objectid;
3436	block->key.objectid = rc->extent_root->fs_info->nodesize;
3437	block->key.offset = generation;
3438	block->level = level;
3439	block->key_ready = 0;
 
3440
3441	rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3442	if (rb_node)
3443		backref_tree_panic(rb_node, -EEXIST, block->bytenr);
 
3444
3445	return 0;
3446}
3447
3448/*
3449 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3450 */
3451static int __add_tree_block(struct reloc_control *rc,
3452			    u64 bytenr, u32 blocksize,
3453			    struct rb_root *blocks)
3454{
3455	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3456	struct btrfs_path *path;
3457	struct btrfs_key key;
3458	int ret;
3459	bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3460
3461	if (tree_block_processed(bytenr, rc))
3462		return 0;
3463
3464	if (tree_search(blocks, bytenr))
3465		return 0;
3466
3467	path = btrfs_alloc_path();
3468	if (!path)
3469		return -ENOMEM;
3470again:
3471	key.objectid = bytenr;
3472	if (skinny) {
3473		key.type = BTRFS_METADATA_ITEM_KEY;
3474		key.offset = (u64)-1;
3475	} else {
3476		key.type = BTRFS_EXTENT_ITEM_KEY;
3477		key.offset = blocksize;
3478	}
3479
3480	path->search_commit_root = 1;
3481	path->skip_locking = 1;
3482	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3483	if (ret < 0)
3484		goto out;
3485
3486	if (ret > 0 && skinny) {
3487		if (path->slots[0]) {
3488			path->slots[0]--;
3489			btrfs_item_key_to_cpu(path->nodes[0], &key,
3490					      path->slots[0]);
3491			if (key.objectid == bytenr &&
3492			    (key.type == BTRFS_METADATA_ITEM_KEY ||
3493			     (key.type == BTRFS_EXTENT_ITEM_KEY &&
3494			      key.offset == blocksize)))
3495				ret = 0;
3496		}
3497
3498		if (ret) {
3499			skinny = false;
3500			btrfs_release_path(path);
3501			goto again;
3502		}
3503	}
3504	if (ret) {
3505		ASSERT(ret == 1);
3506		btrfs_print_leaf(path->nodes[0]);
3507		btrfs_err(fs_info,
3508	     "tree block extent item (%llu) is not found in extent tree",
3509		     bytenr);
3510		WARN_ON(1);
3511		ret = -EINVAL;
3512		goto out;
3513	}
3514
3515	ret = add_tree_block(rc, &key, path, blocks);
3516out:
3517	btrfs_free_path(path);
3518	return ret;
3519}
3520
3521/*
3522 * helper to check if the block use full backrefs for pointers in it
3523 */
3524static int block_use_full_backref(struct reloc_control *rc,
3525				  struct extent_buffer *eb)
3526{
3527	u64 flags;
3528	int ret;
3529
3530	if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3531	    btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3532		return 1;
3533
3534	ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3535				       eb->start, btrfs_header_level(eb), 1,
3536				       NULL, &flags);
3537	BUG_ON(ret);
3538
3539	if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3540		ret = 1;
3541	else
3542		ret = 0;
3543	return ret;
3544}
3545
3546static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3547				    struct btrfs_block_group_cache *block_group,
3548				    struct inode *inode,
3549				    u64 ino)
3550{
3551	struct btrfs_key key;
3552	struct btrfs_root *root = fs_info->tree_root;
3553	struct btrfs_trans_handle *trans;
3554	int ret = 0;
3555
3556	if (inode)
3557		goto truncate;
3558
3559	key.objectid = ino;
3560	key.type = BTRFS_INODE_ITEM_KEY;
3561	key.offset = 0;
3562
3563	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3564	if (IS_ERR(inode))
3565		return -ENOENT;
3566
3567truncate:
3568	ret = btrfs_check_trunc_cache_free_space(fs_info,
3569						 &fs_info->global_block_rsv);
3570	if (ret)
3571		goto out;
3572
3573	trans = btrfs_join_transaction(root);
3574	if (IS_ERR(trans)) {
3575		ret = PTR_ERR(trans);
3576		goto out;
3577	}
3578
3579	ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3580
3581	btrfs_end_transaction(trans);
3582	btrfs_btree_balance_dirty(fs_info);
3583out:
3584	iput(inode);
3585	return ret;
3586}
3587
3588/*
3589 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3590 * this function scans fs tree to find blocks reference the data extent
3591 */
3592static int find_data_references(struct reloc_control *rc,
3593				struct btrfs_key *extent_key,
3594				struct extent_buffer *leaf,
3595				struct btrfs_extent_data_ref *ref,
3596				struct rb_root *blocks)
3597{
3598	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3599	struct btrfs_path *path;
3600	struct tree_block *block;
3601	struct btrfs_root *root;
3602	struct btrfs_file_extent_item *fi;
3603	struct rb_node *rb_node;
3604	struct btrfs_key key;
3605	u64 ref_root;
3606	u64 ref_objectid;
3607	u64 ref_offset;
3608	u32 ref_count;
3609	u32 nritems;
3610	int err = 0;
3611	int added = 0;
3612	int counted;
3613	int ret;
3614
3615	ref_root = btrfs_extent_data_ref_root(leaf, ref);
3616	ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3617	ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3618	ref_count = btrfs_extent_data_ref_count(leaf, ref);
3619
3620	/*
3621	 * This is an extent belonging to the free space cache, lets just delete
3622	 * it and redo the search.
3623	 */
3624	if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3625		ret = delete_block_group_cache(fs_info, rc->block_group,
3626					       NULL, ref_objectid);
3627		if (ret != -ENOENT)
3628			return ret;
3629		ret = 0;
3630	}
3631
3632	path = btrfs_alloc_path();
3633	if (!path)
3634		return -ENOMEM;
3635	path->reada = READA_FORWARD;
3636
3637	root = read_fs_root(fs_info, ref_root);
3638	if (IS_ERR(root)) {
3639		err = PTR_ERR(root);
3640		goto out;
3641	}
3642
3643	key.objectid = ref_objectid;
3644	key.type = BTRFS_EXTENT_DATA_KEY;
3645	if (ref_offset > ((u64)-1 << 32))
3646		key.offset = 0;
3647	else
3648		key.offset = ref_offset;
3649
3650	path->search_commit_root = 1;
3651	path->skip_locking = 1;
3652	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3653	if (ret < 0) {
3654		err = ret;
3655		goto out;
3656	}
3657
3658	leaf = path->nodes[0];
3659	nritems = btrfs_header_nritems(leaf);
3660	/*
3661	 * the references in tree blocks that use full backrefs
3662	 * are not counted in
3663	 */
3664	if (block_use_full_backref(rc, leaf))
3665		counted = 0;
3666	else
3667		counted = 1;
3668	rb_node = tree_search(blocks, leaf->start);
3669	if (rb_node) {
3670		if (counted)
3671			added = 1;
3672		else
3673			path->slots[0] = nritems;
3674	}
3675
3676	while (ref_count > 0) {
3677		while (path->slots[0] >= nritems) {
3678			ret = btrfs_next_leaf(root, path);
3679			if (ret < 0) {
3680				err = ret;
3681				goto out;
3682			}
3683			if (WARN_ON(ret > 0))
3684				goto out;
3685
3686			leaf = path->nodes[0];
3687			nritems = btrfs_header_nritems(leaf);
3688			added = 0;
3689
3690			if (block_use_full_backref(rc, leaf))
3691				counted = 0;
3692			else
3693				counted = 1;
3694			rb_node = tree_search(blocks, leaf->start);
3695			if (rb_node) {
3696				if (counted)
3697					added = 1;
3698				else
3699					path->slots[0] = nritems;
3700			}
3701		}
3702
3703		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3704		if (WARN_ON(key.objectid != ref_objectid ||
3705		    key.type != BTRFS_EXTENT_DATA_KEY))
 
 
3706			break;
3707
3708		fi = btrfs_item_ptr(leaf, path->slots[0],
3709				    struct btrfs_file_extent_item);
3710
3711		if (btrfs_file_extent_type(leaf, fi) ==
3712		    BTRFS_FILE_EXTENT_INLINE)
3713			goto next;
3714
3715		if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3716		    extent_key->objectid)
3717			goto next;
3718
3719		key.offset -= btrfs_file_extent_offset(leaf, fi);
3720		if (key.offset != ref_offset)
3721			goto next;
3722
3723		if (counted)
3724			ref_count--;
3725		if (added)
3726			goto next;
3727
3728		if (!tree_block_processed(leaf->start, rc)) {
3729			block = kmalloc(sizeof(*block), GFP_NOFS);
3730			if (!block) {
3731				err = -ENOMEM;
3732				break;
3733			}
3734			block->bytenr = leaf->start;
3735			btrfs_item_key_to_cpu(leaf, &block->key, 0);
3736			block->level = 0;
3737			block->key_ready = 1;
3738			rb_node = tree_insert(blocks, block->bytenr,
3739					      &block->rb_node);
3740			if (rb_node)
3741				backref_tree_panic(rb_node, -EEXIST,
3742						   block->bytenr);
3743		}
3744		if (counted)
3745			added = 1;
3746		else
3747			path->slots[0] = nritems;
3748next:
3749		path->slots[0]++;
3750
3751	}
3752out:
3753	btrfs_free_path(path);
3754	return err;
 
 
3755}
3756
3757/*
3758 * helper to find all tree blocks that reference a given data extent
3759 */
3760static noinline_for_stack
3761int add_data_references(struct reloc_control *rc,
3762			struct btrfs_key *extent_key,
3763			struct btrfs_path *path,
3764			struct rb_root *blocks)
3765{
3766	struct btrfs_key key;
3767	struct extent_buffer *eb;
3768	struct btrfs_extent_data_ref *dref;
3769	struct btrfs_extent_inline_ref *iref;
3770	unsigned long ptr;
3771	unsigned long end;
3772	u32 blocksize = rc->extent_root->fs_info->nodesize;
3773	int ret = 0;
3774	int err = 0;
3775
3776	eb = path->nodes[0];
3777	ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3778	end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3779	ptr += sizeof(struct btrfs_extent_item);
3780
3781	while (ptr < end) {
3782		iref = (struct btrfs_extent_inline_ref *)ptr;
3783		key.type = btrfs_get_extent_inline_ref_type(eb, iref,
3784							BTRFS_REF_TYPE_DATA);
3785		if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3786			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3787			ret = __add_tree_block(rc, key.offset, blocksize,
3788					       blocks);
3789		} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3790			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3791			ret = find_data_references(rc, extent_key,
3792						   eb, dref, blocks);
3793		} else {
3794			ret = -EUCLEAN;
3795			btrfs_err(rc->extent_root->fs_info,
3796		     "extent %llu slot %d has an invalid inline ref type",
3797			     eb->start, path->slots[0]);
3798		}
3799		if (ret) {
3800			err = ret;
3801			goto out;
3802		}
3803		ptr += btrfs_extent_inline_ref_size(key.type);
3804	}
3805	WARN_ON(ptr > end);
3806
3807	while (1) {
3808		cond_resched();
3809		eb = path->nodes[0];
3810		if (path->slots[0] >= btrfs_header_nritems(eb)) {
3811			ret = btrfs_next_leaf(rc->extent_root, path);
3812			if (ret < 0) {
3813				err = ret;
3814				break;
3815			}
3816			if (ret > 0)
3817				break;
3818			eb = path->nodes[0];
3819		}
3820
3821		btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3822		if (key.objectid != extent_key->objectid)
3823			break;
3824
3825		if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3826			ret = __add_tree_block(rc, key.offset, blocksize,
3827					       blocks);
3828		} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3829			dref = btrfs_item_ptr(eb, path->slots[0],
3830					      struct btrfs_extent_data_ref);
3831			ret = find_data_references(rc, extent_key,
3832						   eb, dref, blocks);
3833		} else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3834			btrfs_print_v0_err(eb->fs_info);
3835			btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3836			ret = -EINVAL;
3837		} else {
3838			ret = 0;
3839		}
3840		if (ret) {
3841			err = ret;
3842			break;
3843		}
3844		path->slots[0]++;
 
 
 
 
 
 
 
3845	}
3846out:
3847	btrfs_release_path(path);
3848	if (err)
3849		free_block_list(blocks);
3850	return err;
 
3851}
3852
3853/*
3854 * helper to find next unprocessed extent
3855 */
3856static noinline_for_stack
3857int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3858		     struct btrfs_key *extent_key)
3859{
3860	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3861	struct btrfs_key key;
3862	struct extent_buffer *leaf;
3863	u64 start, end, last;
3864	int ret;
3865
3866	last = rc->block_group->key.objectid + rc->block_group->key.offset;
3867	while (1) {
 
 
3868		cond_resched();
3869		if (rc->search_start >= last) {
3870			ret = 1;
3871			break;
3872		}
3873
3874		key.objectid = rc->search_start;
3875		key.type = BTRFS_EXTENT_ITEM_KEY;
3876		key.offset = 0;
3877
3878		path->search_commit_root = 1;
3879		path->skip_locking = 1;
3880		ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3881					0, 0);
3882		if (ret < 0)
3883			break;
3884next:
3885		leaf = path->nodes[0];
3886		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3887			ret = btrfs_next_leaf(rc->extent_root, path);
3888			if (ret != 0)
3889				break;
3890			leaf = path->nodes[0];
3891		}
3892
3893		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3894		if (key.objectid >= last) {
3895			ret = 1;
3896			break;
3897		}
3898
3899		if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3900		    key.type != BTRFS_METADATA_ITEM_KEY) {
3901			path->slots[0]++;
3902			goto next;
3903		}
3904
3905		if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3906		    key.objectid + key.offset <= rc->search_start) {
3907			path->slots[0]++;
3908			goto next;
3909		}
3910
3911		if (key.type == BTRFS_METADATA_ITEM_KEY &&
3912		    key.objectid + fs_info->nodesize <=
3913		    rc->search_start) {
3914			path->slots[0]++;
3915			goto next;
3916		}
3917
3918		ret = find_first_extent_bit(&rc->processed_blocks,
3919					    key.objectid, &start, &end,
3920					    EXTENT_DIRTY, NULL);
3921
3922		if (ret == 0 && start <= key.objectid) {
3923			btrfs_release_path(path);
3924			rc->search_start = end + 1;
3925		} else {
3926			if (key.type == BTRFS_EXTENT_ITEM_KEY)
3927				rc->search_start = key.objectid + key.offset;
3928			else
3929				rc->search_start = key.objectid +
3930					fs_info->nodesize;
3931			memcpy(extent_key, &key, sizeof(key));
3932			return 0;
3933		}
3934	}
3935	btrfs_release_path(path);
3936	return ret;
3937}
3938
3939static void set_reloc_control(struct reloc_control *rc)
3940{
3941	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3942
3943	mutex_lock(&fs_info->reloc_mutex);
3944	fs_info->reloc_ctl = rc;
3945	mutex_unlock(&fs_info->reloc_mutex);
3946}
3947
3948static void unset_reloc_control(struct reloc_control *rc)
3949{
3950	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3951
3952	mutex_lock(&fs_info->reloc_mutex);
3953	fs_info->reloc_ctl = NULL;
3954	mutex_unlock(&fs_info->reloc_mutex);
3955}
3956
3957static int check_extent_flags(u64 flags)
3958{
3959	if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3960	    (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3961		return 1;
3962	if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3963	    !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3964		return 1;
3965	if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3966	    (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3967		return 1;
3968	return 0;
3969}
3970
3971static noinline_for_stack
3972int prepare_to_relocate(struct reloc_control *rc)
3973{
3974	struct btrfs_trans_handle *trans;
3975	int ret;
3976
3977	rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3978					      BTRFS_BLOCK_RSV_TEMP);
3979	if (!rc->block_rsv)
3980		return -ENOMEM;
3981
3982	memset(&rc->cluster, 0, sizeof(rc->cluster));
3983	rc->search_start = rc->block_group->key.objectid;
3984	rc->extents_found = 0;
3985	rc->nodes_relocated = 0;
3986	rc->merging_rsv_size = 0;
3987	rc->reserved_bytes = 0;
3988	rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3989			      RELOCATION_RESERVED_NODES;
3990	ret = btrfs_block_rsv_refill(rc->extent_root,
3991				     rc->block_rsv, rc->block_rsv->size,
3992				     BTRFS_RESERVE_FLUSH_ALL);
3993	if (ret)
3994		return ret;
3995
3996	rc->create_reloc_tree = 1;
3997	set_reloc_control(rc);
3998
3999	trans = btrfs_join_transaction(rc->extent_root);
4000	if (IS_ERR(trans)) {
4001		unset_reloc_control(rc);
4002		/*
4003		 * extent tree is not a ref_cow tree and has no reloc_root to
4004		 * cleanup.  And callers are responsible to free the above
4005		 * block rsv.
4006		 */
4007		return PTR_ERR(trans);
4008	}
4009	btrfs_commit_transaction(trans);
4010	return 0;
 
 
 
 
4011}
4012
4013static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
4014{
4015	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
4016	struct rb_root blocks = RB_ROOT;
4017	struct btrfs_key key;
4018	struct btrfs_trans_handle *trans = NULL;
4019	struct btrfs_path *path;
4020	struct btrfs_extent_item *ei;
4021	u64 flags;
4022	u32 item_size;
4023	int ret;
4024	int err = 0;
4025	int progress = 0;
4026
4027	path = btrfs_alloc_path();
4028	if (!path)
4029		return -ENOMEM;
4030	path->reada = READA_FORWARD;
4031
4032	ret = prepare_to_relocate(rc);
4033	if (ret) {
4034		err = ret;
4035		goto out_free;
4036	}
4037
4038	while (1) {
4039		rc->reserved_bytes = 0;
4040		ret = btrfs_block_rsv_refill(rc->extent_root,
4041					rc->block_rsv, rc->block_rsv->size,
4042					BTRFS_RESERVE_FLUSH_ALL);
4043		if (ret) {
4044			err = ret;
4045			break;
4046		}
4047		progress++;
4048		trans = btrfs_start_transaction(rc->extent_root, 0);
4049		if (IS_ERR(trans)) {
4050			err = PTR_ERR(trans);
4051			trans = NULL;
4052			break;
4053		}
4054restart:
4055		if (update_backref_cache(trans, &rc->backref_cache)) {
4056			btrfs_end_transaction(trans);
4057			trans = NULL;
4058			continue;
4059		}
4060
4061		ret = find_next_extent(rc, path, &key);
4062		if (ret < 0)
4063			err = ret;
4064		if (ret != 0)
4065			break;
4066
4067		rc->extents_found++;
4068
4069		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4070				    struct btrfs_extent_item);
4071		item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
4072		if (item_size >= sizeof(*ei)) {
4073			flags = btrfs_extent_flags(path->nodes[0], ei);
4074			ret = check_extent_flags(flags);
4075			BUG_ON(ret);
4076		} else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
4077			err = -EINVAL;
4078			btrfs_print_v0_err(trans->fs_info);
4079			btrfs_abort_transaction(trans, err);
4080			break;
4081		} else {
4082			BUG();
 
 
 
4083		}
4084
4085		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4086			ret = add_tree_block(rc, &key, path, &blocks);
4087		} else if (rc->stage == UPDATE_DATA_PTRS &&
4088			   (flags & BTRFS_EXTENT_FLAG_DATA)) {
4089			ret = add_data_references(rc, &key, path, &blocks);
4090		} else {
4091			btrfs_release_path(path);
4092			ret = 0;
4093		}
4094		if (ret < 0) {
4095			err = ret;
4096			break;
4097		}
4098
4099		if (!RB_EMPTY_ROOT(&blocks)) {
4100			ret = relocate_tree_blocks(trans, rc, &blocks);
4101			if (ret < 0) {
4102				/*
4103				 * if we fail to relocate tree blocks, force to update
4104				 * backref cache when committing transaction.
4105				 */
4106				rc->backref_cache.last_trans = trans->transid - 1;
4107
4108				if (ret != -EAGAIN) {
4109					err = ret;
4110					break;
4111				}
4112				rc->extents_found--;
4113				rc->search_start = key.objectid;
4114			}
4115		}
4116
4117		btrfs_end_transaction_throttle(trans);
4118		btrfs_btree_balance_dirty(fs_info);
4119		trans = NULL;
4120
4121		if (rc->stage == MOVE_DATA_EXTENTS &&
4122		    (flags & BTRFS_EXTENT_FLAG_DATA)) {
4123			rc->found_file_extent = 1;
4124			ret = relocate_data_extent(rc->data_inode,
4125						   &key, &rc->cluster);
4126			if (ret < 0) {
4127				err = ret;
4128				break;
4129			}
4130		}
 
 
 
 
4131	}
4132	if (trans && progress && err == -ENOSPC) {
4133		ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
4134		if (ret == 1) {
4135			err = 0;
4136			progress = 0;
4137			goto restart;
4138		}
4139	}
4140
4141	btrfs_release_path(path);
4142	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4143
4144	if (trans) {
4145		btrfs_end_transaction_throttle(trans);
4146		btrfs_btree_balance_dirty(fs_info);
4147	}
4148
4149	if (!err) {
4150		ret = relocate_file_extent_cluster(rc->data_inode,
4151						   &rc->cluster);
4152		if (ret < 0)
4153			err = ret;
4154	}
4155
4156	rc->create_reloc_tree = 0;
4157	set_reloc_control(rc);
4158
4159	backref_cache_cleanup(&rc->backref_cache);
4160	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4161
 
 
 
 
 
 
 
 
4162	err = prepare_to_merge(rc, err);
4163
4164	merge_reloc_roots(rc);
4165
4166	rc->merge_reloc_tree = 0;
4167	unset_reloc_control(rc);
4168	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4169
4170	/* get rid of pinned extents */
4171	trans = btrfs_join_transaction(rc->extent_root);
4172	if (IS_ERR(trans)) {
4173		err = PTR_ERR(trans);
4174		goto out_free;
4175	}
4176	btrfs_commit_transaction(trans);
 
 
 
4177	ret = clean_dirty_subvols(rc);
4178	if (ret < 0 && !err)
4179		err = ret;
4180out_free:
4181	btrfs_free_block_rsv(fs_info, rc->block_rsv);
4182	btrfs_free_path(path);
4183	return err;
4184}
4185
4186static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4187				 struct btrfs_root *root, u64 objectid)
4188{
4189	struct btrfs_path *path;
4190	struct btrfs_inode_item *item;
4191	struct extent_buffer *leaf;
4192	int ret;
4193
4194	path = btrfs_alloc_path();
4195	if (!path)
4196		return -ENOMEM;
4197
4198	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4199	if (ret)
4200		goto out;
4201
4202	leaf = path->nodes[0];
4203	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4204	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4205	btrfs_set_inode_generation(leaf, item, 1);
4206	btrfs_set_inode_size(leaf, item, 0);
4207	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4208	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4209					  BTRFS_INODE_PREALLOC);
4210	btrfs_mark_buffer_dirty(leaf);
4211out:
4212	btrfs_free_path(path);
4213	return ret;
4214}
4215
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4216/*
4217 * helper to create inode for data relocation.
4218 * the inode is in data relocation tree and its link count is 0
4219 */
4220static noinline_for_stack
4221struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4222				 struct btrfs_block_group_cache *group)
4223{
4224	struct inode *inode = NULL;
4225	struct btrfs_trans_handle *trans;
4226	struct btrfs_root *root;
4227	struct btrfs_key key;
4228	u64 objectid;
4229	int err = 0;
4230
4231	root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4232	if (IS_ERR(root))
4233		return ERR_CAST(root);
4234
4235	trans = btrfs_start_transaction(root, 6);
4236	if (IS_ERR(trans))
 
4237		return ERR_CAST(trans);
 
4238
4239	err = btrfs_find_free_objectid(root, &objectid);
4240	if (err)
4241		goto out;
4242
4243	err = __insert_orphan_inode(trans, root, objectid);
4244	BUG_ON(err);
 
4245
4246	key.objectid = objectid;
4247	key.type = BTRFS_INODE_ITEM_KEY;
4248	key.offset = 0;
4249	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4250	BUG_ON(IS_ERR(inode));
4251	BTRFS_I(inode)->index_cnt = group->key.objectid;
 
 
4252
4253	err = btrfs_orphan_add(trans, BTRFS_I(inode));
4254out:
 
4255	btrfs_end_transaction(trans);
4256	btrfs_btree_balance_dirty(fs_info);
4257	if (err) {
4258		if (inode)
4259			iput(inode);
4260		inode = ERR_PTR(err);
4261	}
4262	return inode;
4263}
4264
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4265static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4266{
4267	struct reloc_control *rc;
4268
4269	rc = kzalloc(sizeof(*rc), GFP_NOFS);
4270	if (!rc)
4271		return NULL;
4272
4273	INIT_LIST_HEAD(&rc->reloc_roots);
4274	INIT_LIST_HEAD(&rc->dirty_subvol_roots);
4275	backref_cache_init(&rc->backref_cache);
4276	mapping_tree_init(&rc->reloc_root_tree);
4277	extent_io_tree_init(fs_info, &rc->processed_blocks,
4278			    IO_TREE_RELOC_BLOCKS, NULL);
4279	return rc;
4280}
4281
 
 
 
 
 
 
 
 
 
 
 
 
4282/*
4283 * Print the block group being relocated
4284 */
4285static void describe_relocation(struct btrfs_fs_info *fs_info,
4286				struct btrfs_block_group_cache *block_group)
4287{
4288	char buf[128] = {'\0'};
4289
4290	btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
4291
4292	btrfs_info(fs_info,
4293		   "relocating block group %llu flags %s",
4294		   block_group->key.objectid, buf);
 
 
 
 
 
 
 
 
 
4295}
4296
4297/*
4298 * function to relocate all extents in a block group.
4299 */
4300int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4301{
4302	struct btrfs_block_group_cache *bg;
4303	struct btrfs_root *extent_root = fs_info->extent_root;
4304	struct reloc_control *rc;
4305	struct inode *inode;
4306	struct btrfs_path *path;
4307	int ret;
4308	int rw = 0;
4309	int err = 0;
4310
 
 
 
 
 
 
 
 
 
 
 
 
 
4311	bg = btrfs_lookup_block_group(fs_info, group_start);
4312	if (!bg)
4313		return -ENOENT;
4314
 
 
 
 
 
 
 
 
 
 
 
4315	if (btrfs_pinned_by_swapfile(fs_info, bg)) {
4316		btrfs_put_block_group(bg);
4317		return -ETXTBSY;
4318	}
4319
4320	rc = alloc_reloc_control(fs_info);
4321	if (!rc) {
4322		btrfs_put_block_group(bg);
4323		return -ENOMEM;
4324	}
4325
 
 
 
 
 
 
4326	rc->extent_root = extent_root;
4327	rc->block_group = bg;
4328
4329	ret = btrfs_inc_block_group_ro(rc->block_group);
4330	if (ret) {
4331		err = ret;
4332		goto out;
4333	}
4334	rw = 1;
4335
4336	path = btrfs_alloc_path();
4337	if (!path) {
4338		err = -ENOMEM;
4339		goto out;
4340	}
4341
4342	inode = lookup_free_space_inode(rc->block_group, path);
4343	btrfs_free_path(path);
4344
4345	if (!IS_ERR(inode))
4346		ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4347	else
4348		ret = PTR_ERR(inode);
4349
4350	if (ret && ret != -ENOENT) {
4351		err = ret;
4352		goto out;
4353	}
4354
4355	rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4356	if (IS_ERR(rc->data_inode)) {
4357		err = PTR_ERR(rc->data_inode);
4358		rc->data_inode = NULL;
4359		goto out;
4360	}
4361
4362	describe_relocation(fs_info, rc->block_group);
4363
4364	btrfs_wait_block_group_reservations(rc->block_group);
4365	btrfs_wait_nocow_writers(rc->block_group);
4366	btrfs_wait_ordered_roots(fs_info, U64_MAX,
4367				 rc->block_group->key.objectid,
4368				 rc->block_group->key.offset);
 
 
 
4369
4370	while (1) {
 
 
4371		mutex_lock(&fs_info->cleaner_mutex);
4372		ret = relocate_block_group(rc);
4373		mutex_unlock(&fs_info->cleaner_mutex);
4374		if (ret < 0)
4375			err = ret;
4376
 
4377		/*
4378		 * We may have gotten ENOSPC after we already dirtied some
4379		 * extents.  If writeout happens while we're relocating a
4380		 * different block group we could end up hitting the
4381		 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
4382		 * btrfs_reloc_cow_block.  Make sure we write everything out
4383		 * properly so we don't trip over this problem, and then break
4384		 * out of the loop if we hit an error.
4385		 */
4386		if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4387			ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4388						       (u64)-1);
4389			if (ret)
4390				err = ret;
4391			invalidate_mapping_pages(rc->data_inode->i_mapping,
4392						 0, -1);
4393			rc->stage = UPDATE_DATA_PTRS;
4394		}
4395
4396		if (err < 0)
4397			goto out;
4398
4399		if (rc->extents_found == 0)
4400			break;
4401
4402		btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4403
4404	}
4405
4406	WARN_ON(rc->block_group->pinned > 0);
4407	WARN_ON(rc->block_group->reserved > 0);
4408	WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4409out:
4410	if (err && rw)
4411		btrfs_dec_block_group_ro(rc->block_group);
4412	iput(rc->data_inode);
4413	btrfs_put_block_group(rc->block_group);
4414	kfree(rc);
 
 
4415	return err;
4416}
4417
4418static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4419{
4420	struct btrfs_fs_info *fs_info = root->fs_info;
4421	struct btrfs_trans_handle *trans;
4422	int ret, err;
4423
4424	trans = btrfs_start_transaction(fs_info->tree_root, 0);
4425	if (IS_ERR(trans))
4426		return PTR_ERR(trans);
4427
4428	memset(&root->root_item.drop_progress, 0,
4429		sizeof(root->root_item.drop_progress));
4430	root->root_item.drop_level = 0;
4431	btrfs_set_root_refs(&root->root_item, 0);
4432	ret = btrfs_update_root(trans, fs_info->tree_root,
4433				&root->root_key, &root->root_item);
4434
4435	err = btrfs_end_transaction(trans);
4436	if (err)
4437		return err;
4438	return ret;
4439}
4440
4441/*
4442 * recover relocation interrupted by system crash.
4443 *
4444 * this function resumes merging reloc trees with corresponding fs trees.
4445 * this is important for keeping the sharing of tree blocks
4446 */
4447int btrfs_recover_relocation(struct btrfs_root *root)
4448{
4449	struct btrfs_fs_info *fs_info = root->fs_info;
4450	LIST_HEAD(reloc_roots);
4451	struct btrfs_key key;
4452	struct btrfs_root *fs_root;
4453	struct btrfs_root *reloc_root;
4454	struct btrfs_path *path;
4455	struct extent_buffer *leaf;
4456	struct reloc_control *rc = NULL;
4457	struct btrfs_trans_handle *trans;
4458	int ret;
4459	int err = 0;
4460
4461	path = btrfs_alloc_path();
4462	if (!path)
4463		return -ENOMEM;
4464	path->reada = READA_BACK;
4465
4466	key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4467	key.type = BTRFS_ROOT_ITEM_KEY;
4468	key.offset = (u64)-1;
4469
4470	while (1) {
4471		ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4472					path, 0, 0);
4473		if (ret < 0) {
4474			err = ret;
4475			goto out;
4476		}
4477		if (ret > 0) {
4478			if (path->slots[0] == 0)
4479				break;
4480			path->slots[0]--;
4481		}
4482		leaf = path->nodes[0];
4483		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4484		btrfs_release_path(path);
4485
4486		if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4487		    key.type != BTRFS_ROOT_ITEM_KEY)
4488			break;
4489
4490		reloc_root = btrfs_read_fs_root(root, &key);
4491		if (IS_ERR(reloc_root)) {
4492			err = PTR_ERR(reloc_root);
4493			goto out;
4494		}
4495
 
4496		list_add(&reloc_root->root_list, &reloc_roots);
4497
4498		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4499			fs_root = read_fs_root(fs_info,
4500					       reloc_root->root_key.offset);
4501			if (IS_ERR(fs_root)) {
4502				ret = PTR_ERR(fs_root);
4503				if (ret != -ENOENT) {
4504					err = ret;
4505					goto out;
4506				}
4507				ret = mark_garbage_root(reloc_root);
4508				if (ret < 0) {
4509					err = ret;
4510					goto out;
4511				}
 
 
4512			}
4513		}
4514
4515		if (key.offset == 0)
4516			break;
4517
4518		key.offset--;
4519	}
4520	btrfs_release_path(path);
4521
4522	if (list_empty(&reloc_roots))
4523		goto out;
4524
4525	rc = alloc_reloc_control(fs_info);
4526	if (!rc) {
4527		err = -ENOMEM;
4528		goto out;
4529	}
4530
4531	rc->extent_root = fs_info->extent_root;
 
 
 
 
 
 
4532
4533	set_reloc_control(rc);
4534
4535	trans = btrfs_join_transaction(rc->extent_root);
4536	if (IS_ERR(trans)) {
4537		unset_reloc_control(rc);
4538		err = PTR_ERR(trans);
4539		goto out_free;
4540	}
4541
4542	rc->merge_reloc_tree = 1;
4543
4544	while (!list_empty(&reloc_roots)) {
4545		reloc_root = list_entry(reloc_roots.next,
4546					struct btrfs_root, root_list);
4547		list_del(&reloc_root->root_list);
4548
4549		if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4550			list_add_tail(&reloc_root->root_list,
4551				      &rc->reloc_roots);
4552			continue;
4553		}
4554
4555		fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
 
4556		if (IS_ERR(fs_root)) {
4557			err = PTR_ERR(fs_root);
4558			goto out_free;
 
 
4559		}
4560
4561		err = __add_reloc_root(reloc_root);
4562		BUG_ON(err < 0); /* -ENOMEM or logic error */
4563		fs_root->reloc_root = reloc_root;
 
 
 
 
 
 
 
4564	}
4565
4566	err = btrfs_commit_transaction(trans);
4567	if (err)
4568		goto out_free;
4569
4570	merge_reloc_roots(rc);
4571
4572	unset_reloc_control(rc);
4573
4574	trans = btrfs_join_transaction(rc->extent_root);
4575	if (IS_ERR(trans)) {
4576		err = PTR_ERR(trans);
4577		goto out_free;
4578	}
4579	err = btrfs_commit_transaction(trans);
4580
4581	ret = clean_dirty_subvols(rc);
4582	if (ret < 0 && !err)
4583		err = ret;
4584out_free:
4585	kfree(rc);
 
 
 
4586out:
4587	if (!list_empty(&reloc_roots))
4588		free_reloc_roots(&reloc_roots);
4589
4590	btrfs_free_path(path);
4591
4592	if (err == 0) {
4593		/* cleanup orphan inode in data relocation tree */
4594		fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4595		if (IS_ERR(fs_root))
4596			err = PTR_ERR(fs_root);
4597		else
4598			err = btrfs_orphan_cleanup(fs_root);
4599	}
4600	return err;
4601}
4602
4603/*
4604 * helper to add ordered checksum for data relocation.
4605 *
4606 * cloning checksum properly handles the nodatasum extents.
4607 * it also saves CPU time to re-calculate the checksum.
4608 */
4609int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4610{
4611	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4612	struct btrfs_ordered_sum *sums;
4613	struct btrfs_ordered_extent *ordered;
4614	int ret;
4615	u64 disk_bytenr;
4616	u64 new_bytenr;
4617	LIST_HEAD(list);
 
4618
4619	ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4620	BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4621
4622	disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4623	ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4624				       disk_bytenr + len - 1, &list, 0);
4625	if (ret)
4626		goto out;
4627
4628	while (!list_empty(&list)) {
4629		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
 
 
4630		list_del_init(&sums->list);
4631
4632		/*
4633		 * We need to offset the new_bytenr based on where the csum is.
4634		 * We need to do this because we will read in entire prealloc
4635		 * extents but we may have written to say the middle of the
4636		 * prealloc extent, so we need to make sure the csum goes with
4637		 * the right disk offset.
4638		 *
4639		 * We can do this because the data reloc inode refers strictly
4640		 * to the on disk bytes, so we don't have to worry about
4641		 * disk_len vs real len like with real inodes since it's all
4642		 * disk length.
4643		 */
4644		new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4645		sums->bytenr = new_bytenr;
4646
4647		btrfs_add_ordered_sum(ordered, sums);
4648	}
4649out:
4650	btrfs_put_ordered_extent(ordered);
4651	return ret;
4652}
4653
4654int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4655			  struct btrfs_root *root, struct extent_buffer *buf,
 
4656			  struct extent_buffer *cow)
4657{
4658	struct btrfs_fs_info *fs_info = root->fs_info;
4659	struct reloc_control *rc;
4660	struct backref_node *node;
4661	int first_cow = 0;
4662	int level;
4663	int ret = 0;
4664
4665	rc = fs_info->reloc_ctl;
4666	if (!rc)
4667		return 0;
4668
4669	BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4670	       root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4671
4672	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4673		if (buf == root->node)
4674			__update_reloc_root(root, cow->start);
4675	}
4676
4677	level = btrfs_header_level(buf);
4678	if (btrfs_header_generation(buf) <=
4679	    btrfs_root_last_snapshot(&root->root_item))
4680		first_cow = 1;
4681
4682	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4683	    rc->create_reloc_tree) {
4684		WARN_ON(!first_cow && level == 0);
4685
4686		node = rc->backref_cache.path[level];
4687		BUG_ON(node->bytenr != buf->start &&
4688		       node->new_bytenr != buf->start);
4689
4690		drop_node_buffer(node);
4691		extent_buffer_get(cow);
4692		node->eb = cow;
4693		node->new_bytenr = cow->start;
4694
4695		if (!node->pending) {
4696			list_move_tail(&node->list,
4697				       &rc->backref_cache.pending[level]);
4698			node->pending = 1;
4699		}
4700
4701		if (first_cow)
4702			__mark_block_processed(rc, node);
4703
4704		if (first_cow && level > 0)
4705			rc->nodes_relocated += buf->len;
4706	}
4707
4708	if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4709		ret = replace_file_extents(trans, rc, root, cow);
4710	return ret;
4711}
4712
4713/*
4714 * called before creating snapshot. it calculates metadata reservation
4715 * required for relocating tree blocks in the snapshot
4716 */
4717void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4718			      u64 *bytes_to_reserve)
4719{
4720	struct btrfs_root *root = pending->root;
4721	struct reloc_control *rc = root->fs_info->reloc_ctl;
4722
4723	if (!root->reloc_root || !rc)
4724		return;
4725
4726	if (!rc->merge_reloc_tree)
4727		return;
4728
4729	root = root->reloc_root;
4730	BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4731	/*
4732	 * relocation is in the stage of merging trees. the space
4733	 * used by merging a reloc tree is twice the size of
4734	 * relocated tree nodes in the worst case. half for cowing
4735	 * the reloc tree, half for cowing the fs tree. the space
4736	 * used by cowing the reloc tree will be freed after the
4737	 * tree is dropped. if we create snapshot, cowing the fs
4738	 * tree may use more space than it frees. so we need
4739	 * reserve extra space.
4740	 */
4741	*bytes_to_reserve += rc->nodes_relocated;
4742}
4743
4744/*
4745 * called after snapshot is created. migrate block reservation
4746 * and create reloc root for the newly created snapshot
 
 
 
 
4747 */
4748int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4749			       struct btrfs_pending_snapshot *pending)
4750{
4751	struct btrfs_root *root = pending->root;
4752	struct btrfs_root *reloc_root;
4753	struct btrfs_root *new_root;
4754	struct reloc_control *rc = root->fs_info->reloc_ctl;
4755	int ret;
4756
4757	if (!root->reloc_root || !rc)
4758		return 0;
4759
4760	rc = root->fs_info->reloc_ctl;
4761	rc->merging_rsv_size += rc->nodes_relocated;
4762
4763	if (rc->merge_reloc_tree) {
4764		ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4765					      rc->block_rsv,
4766					      rc->nodes_relocated, true);
4767		if (ret)
4768			return ret;
4769	}
4770
4771	new_root = pending->snap;
4772	reloc_root = create_reloc_root(trans, root->reloc_root,
4773				       new_root->root_key.objectid);
4774	if (IS_ERR(reloc_root))
4775		return PTR_ERR(reloc_root);
4776
4777	ret = __add_reloc_root(reloc_root);
4778	BUG_ON(ret < 0);
4779	new_root->reloc_root = reloc_root;
 
 
 
 
 
4780
4781	if (rc->create_reloc_tree)
4782		ret = clone_backref_node(trans, rc, root, reloc_root);
4783	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4784}
v6.9.4
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2009 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/sched.h>
   7#include <linux/pagemap.h>
   8#include <linux/writeback.h>
   9#include <linux/blkdev.h>
  10#include <linux/rbtree.h>
  11#include <linux/slab.h>
  12#include <linux/error-injection.h>
  13#include "ctree.h"
  14#include "disk-io.h"
  15#include "transaction.h"
  16#include "volumes.h"
  17#include "locking.h"
  18#include "btrfs_inode.h"
  19#include "async-thread.h"
  20#include "free-space-cache.h"
 
  21#include "qgroup.h"
  22#include "print-tree.h"
  23#include "delalloc-space.h"
  24#include "block-group.h"
  25#include "backref.h"
  26#include "misc.h"
  27#include "subpage.h"
  28#include "zoned.h"
  29#include "inode-item.h"
  30#include "space-info.h"
  31#include "fs.h"
  32#include "accessors.h"
  33#include "extent-tree.h"
  34#include "root-tree.h"
  35#include "file-item.h"
  36#include "relocation.h"
  37#include "super.h"
  38#include "tree-checker.h"
  39
  40/*
  41 * Relocation overview
  42 *
  43 * [What does relocation do]
  44 *
  45 * The objective of relocation is to relocate all extents of the target block
  46 * group to other block groups.
  47 * This is utilized by resize (shrink only), profile converting, compacting
  48 * space, or balance routine to spread chunks over devices.
  49 *
  50 * 		Before		|		After
  51 * ------------------------------------------------------------------
  52 *  BG A: 10 data extents	| BG A: deleted
  53 *  BG B:  2 data extents	| BG B: 10 data extents (2 old + 8 relocated)
  54 *  BG C:  1 extents		| BG C:  3 data extents (1 old + 2 relocated)
  55 *
  56 * [How does relocation work]
  57 *
  58 * 1.   Mark the target block group read-only
  59 *      New extents won't be allocated from the target block group.
  60 *
  61 * 2.1  Record each extent in the target block group
  62 *      To build a proper map of extents to be relocated.
  63 *
  64 * 2.2  Build data reloc tree and reloc trees
  65 *      Data reloc tree will contain an inode, recording all newly relocated
  66 *      data extents.
  67 *      There will be only one data reloc tree for one data block group.
  68 *
  69 *      Reloc tree will be a special snapshot of its source tree, containing
  70 *      relocated tree blocks.
  71 *      Each tree referring to a tree block in target block group will get its
  72 *      reloc tree built.
  73 *
  74 * 2.3  Swap source tree with its corresponding reloc tree
  75 *      Each involved tree only refers to new extents after swap.
  76 *
  77 * 3.   Cleanup reloc trees and data reloc tree.
  78 *      As old extents in the target block group are still referenced by reloc
  79 *      trees, we need to clean them up before really freeing the target block
  80 *      group.
  81 *
  82 * The main complexity is in steps 2.2 and 2.3.
  83 *
  84 * The entry point of relocation is relocate_block_group() function.
 
 
 
 
 
 
 
 
 
  85 */
 
 
 
 
  86
 
 
  87#define RELOCATION_RESERVED_NODES	256
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  88/*
  89 * map address of tree root to tree
  90 */
  91struct mapping_node {
  92	struct {
  93		struct rb_node rb_node;
  94		u64 bytenr;
  95	}; /* Use rb_simle_node for search/insert */
  96	void *data;
  97};
  98
  99struct mapping_tree {
 100	struct rb_root rb_root;
 101	spinlock_t lock;
 102};
 103
 104/*
 105 * present a tree block to process
 106 */
 107struct tree_block {
 108	struct {
 109		struct rb_node rb_node;
 110		u64 bytenr;
 111	}; /* Use rb_simple_node for search/insert */
 112	u64 owner;
 113	struct btrfs_key key;
 114	u8 level;
 115	bool key_ready;
 116};
 117
 118#define MAX_EXTENTS 128
 119
 120struct file_extent_cluster {
 121	u64 start;
 122	u64 end;
 123	u64 boundary[MAX_EXTENTS];
 124	unsigned int nr;
 125	u64 owning_root;
 126};
 127
 128/* Stages of data relocation. */
 129enum reloc_stage {
 130	MOVE_DATA_EXTENTS,
 131	UPDATE_DATA_PTRS
 132};
 133
 134struct reloc_control {
 135	/* block group to relocate */
 136	struct btrfs_block_group *block_group;
 137	/* extent tree */
 138	struct btrfs_root *extent_root;
 139	/* inode for moving data */
 140	struct inode *data_inode;
 141
 142	struct btrfs_block_rsv *block_rsv;
 143
 144	struct btrfs_backref_cache backref_cache;
 145
 146	struct file_extent_cluster cluster;
 147	/* tree blocks have been processed */
 148	struct extent_io_tree processed_blocks;
 149	/* map start of tree root to corresponding reloc tree */
 150	struct mapping_tree reloc_root_tree;
 151	/* list of reloc trees */
 152	struct list_head reloc_roots;
 153	/* list of subvolume trees that get relocated */
 154	struct list_head dirty_subvol_roots;
 155	/* size of metadata reservation for merging reloc trees */
 156	u64 merging_rsv_size;
 157	/* size of relocated tree nodes */
 158	u64 nodes_relocated;
 159	/* reserved size for block group relocation*/
 160	u64 reserved_bytes;
 161
 162	u64 search_start;
 163	u64 extents_found;
 164
 165	enum reloc_stage stage;
 166	bool create_reloc_tree;
 167	bool merge_reloc_tree;
 168	bool found_file_extent;
 169};
 170
 171static void mark_block_processed(struct reloc_control *rc,
 172				 struct btrfs_backref_node *node)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 173{
 174	u32 blocksize;
 
 
 
 
 175
 176	if (node->level == 0 ||
 177	    in_range(node->bytenr, rc->block_group->start,
 178		     rc->block_group->length)) {
 179		blocksize = rc->extent_root->fs_info->nodesize;
 180		set_extent_bit(&rc->processed_blocks, node->bytenr,
 181			       node->bytenr + blocksize - 1, EXTENT_DIRTY, NULL);
 182	}
 183	node->processed = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 184}
 185
 186/*
 187 * walk up backref nodes until reach node presents tree root
 188 */
 189static struct btrfs_backref_node *walk_up_backref(
 190		struct btrfs_backref_node *node,
 191		struct btrfs_backref_edge *edges[], int *index)
 192{
 193	struct btrfs_backref_edge *edge;
 194	int idx = *index;
 195
 196	while (!list_empty(&node->upper)) {
 197		edge = list_entry(node->upper.next,
 198				  struct btrfs_backref_edge, list[LOWER]);
 199		edges[idx++] = edge;
 200		node = edge->node[UPPER];
 201	}
 202	BUG_ON(node->detached);
 203	*index = idx;
 204	return node;
 205}
 206
 207/*
 208 * walk down backref nodes to find start of next reference path
 209 */
 210static struct btrfs_backref_node *walk_down_backref(
 211		struct btrfs_backref_edge *edges[], int *index)
 212{
 213	struct btrfs_backref_edge *edge;
 214	struct btrfs_backref_node *lower;
 215	int idx = *index;
 216
 217	while (idx > 0) {
 218		edge = edges[idx - 1];
 219		lower = edge->node[LOWER];
 220		if (list_is_last(&edge->list[LOWER], &lower->upper)) {
 221			idx--;
 222			continue;
 223		}
 224		edge = list_entry(edge->list[LOWER].next,
 225				  struct btrfs_backref_edge, list[LOWER]);
 226		edges[idx - 1] = edge;
 227		*index = idx;
 228		return edge->node[UPPER];
 229	}
 230	*index = 0;
 231	return NULL;
 232}
 233
 234static void update_backref_node(struct btrfs_backref_cache *cache,
 235				struct btrfs_backref_node *node, u64 bytenr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 236{
 237	struct rb_node *rb_node;
 238	rb_erase(&node->rb_node, &cache->rb_root);
 239	node->bytenr = bytenr;
 240	rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node);
 241	if (rb_node)
 242		btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST);
 243}
 244
 245/*
 246 * update backref cache after a transaction commit
 247 */
 248static int update_backref_cache(struct btrfs_trans_handle *trans,
 249				struct btrfs_backref_cache *cache)
 250{
 251	struct btrfs_backref_node *node;
 252	int level = 0;
 253
 254	if (cache->last_trans == 0) {
 255		cache->last_trans = trans->transid;
 256		return 0;
 257	}
 258
 259	if (cache->last_trans == trans->transid)
 260		return 0;
 261
 262	/*
 263	 * detached nodes are used to avoid unnecessary backref
 264	 * lookup. transaction commit changes the extent tree.
 265	 * so the detached nodes are no longer useful.
 266	 */
 267	while (!list_empty(&cache->detached)) {
 268		node = list_entry(cache->detached.next,
 269				  struct btrfs_backref_node, list);
 270		btrfs_backref_cleanup_node(cache, node);
 271	}
 272
 273	while (!list_empty(&cache->changed)) {
 274		node = list_entry(cache->changed.next,
 275				  struct btrfs_backref_node, list);
 276		list_del_init(&node->list);
 277		BUG_ON(node->pending);
 278		update_backref_node(cache, node, node->new_bytenr);
 279	}
 280
 281	/*
 282	 * some nodes can be left in the pending list if there were
 283	 * errors during processing the pending nodes.
 284	 */
 285	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
 286		list_for_each_entry(node, &cache->pending[level], list) {
 287			BUG_ON(!node->pending);
 288			if (node->bytenr == node->new_bytenr)
 289				continue;
 290			update_backref_node(cache, node, node->new_bytenr);
 291		}
 292	}
 293
 294	cache->last_trans = 0;
 295	return 1;
 296}
 297
 298static bool reloc_root_is_dead(const struct btrfs_root *root)
 299{
 300	/*
 301	 * Pair with set_bit/clear_bit in clean_dirty_subvols and
 302	 * btrfs_update_reloc_root. We need to see the updated bit before
 303	 * trying to access reloc_root
 304	 */
 305	smp_rmb();
 306	if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
 307		return true;
 308	return false;
 309}
 310
 311/*
 312 * Check if this subvolume tree has valid reloc tree.
 313 *
 314 * Reloc tree after swap is considered dead, thus not considered as valid.
 315 * This is enough for most callers, as they don't distinguish dead reloc root
 316 * from no reloc root.  But btrfs_should_ignore_reloc_root() below is a
 317 * special case.
 318 */
 319static bool have_reloc_root(const struct btrfs_root *root)
 320{
 321	if (reloc_root_is_dead(root))
 322		return false;
 323	if (!root->reloc_root)
 324		return false;
 325	return true;
 326}
 327
 328bool btrfs_should_ignore_reloc_root(const struct btrfs_root *root)
 329{
 330	struct btrfs_root *reloc_root;
 331
 332	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 333		return false;
 334
 335	/* This root has been merged with its reloc tree, we can ignore it */
 336	if (reloc_root_is_dead(root))
 337		return true;
 338
 339	reloc_root = root->reloc_root;
 340	if (!reloc_root)
 341		return false;
 342
 343	if (btrfs_header_generation(reloc_root->commit_root) ==
 344	    root->fs_info->running_transaction->transid)
 345		return false;
 346	/*
 347	 * If there is reloc tree and it was created in previous transaction
 348	 * backref lookup can find the reloc tree, so backref node for the fs
 349	 * tree root is useless for relocation.
 
 350	 */
 351	return true;
 352}
 353
 354/*
 355 * find reloc tree by address of tree root
 356 */
 357struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 
 358{
 359	struct reloc_control *rc = fs_info->reloc_ctl;
 360	struct rb_node *rb_node;
 361	struct mapping_node *node;
 362	struct btrfs_root *root = NULL;
 363
 364	ASSERT(rc);
 365	spin_lock(&rc->reloc_root_tree.lock);
 366	rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
 367	if (rb_node) {
 368		node = rb_entry(rb_node, struct mapping_node, rb_node);
 369		root = node->data;
 370	}
 371	spin_unlock(&rc->reloc_root_tree.lock);
 372	return btrfs_grab_root(root);
 373}
 374
 375/*
 376 * For useless nodes, do two major clean ups:
 377 *
 378 * - Cleanup the children edges and nodes
 379 *   If child node is also orphan (no parent) during cleanup, then the child
 380 *   node will also be cleaned up.
 381 *
 382 * - Freeing up leaves (level 0), keeps nodes detached
 383 *   For nodes, the node is still cached as "detached"
 384 *
 385 * Return false if @node is not in the @useless_nodes list.
 386 * Return true if @node is in the @useless_nodes list.
 387 */
 388static bool handle_useless_nodes(struct reloc_control *rc,
 389				 struct btrfs_backref_node *node)
 390{
 391	struct btrfs_backref_cache *cache = &rc->backref_cache;
 392	struct list_head *useless_node = &cache->useless_node;
 393	bool ret = false;
 
 
 
 
 
 
 
 
 
 394
 395	while (!list_empty(useless_node)) {
 396		struct btrfs_backref_node *cur;
 
 
 397
 398		cur = list_first_entry(useless_node, struct btrfs_backref_node,
 399				 list);
 400		list_del_init(&cur->list);
 
 
 
 401
 402		/* Only tree root nodes can be added to @useless_nodes */
 403		ASSERT(list_empty(&cur->upper));
 404
 405		if (cur == node)
 406			ret = true;
 407
 408		/* The node is the lowest node */
 409		if (cur->lowest) {
 410			list_del_init(&cur->lower);
 411			cur->lowest = 0;
 412		}
 413
 414		/* Cleanup the lower edges */
 415		while (!list_empty(&cur->lower)) {
 416			struct btrfs_backref_edge *edge;
 417			struct btrfs_backref_node *lower;
 418
 419			edge = list_entry(cur->lower.next,
 420					struct btrfs_backref_edge, list[UPPER]);
 421			list_del(&edge->list[UPPER]);
 422			list_del(&edge->list[LOWER]);
 423			lower = edge->node[LOWER];
 424			btrfs_backref_free_edge(cache, edge);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 425
 426			/* Child node is also orphan, queue for cleanup */
 427			if (list_empty(&lower->upper))
 428				list_add(&lower->list, useless_node);
 429		}
 430		/* Mark this block processed for relocation */
 431		mark_block_processed(rc, cur);
 432
 433		/*
 434		 * Backref nodes for tree leaves are deleted from the cache.
 435		 * Backref nodes for upper level tree blocks are left in the
 436		 * cache to avoid unnecessary backref lookup.
 437		 */
 438		if (cur->level > 0) {
 439			list_add(&cur->list, &cache->detached);
 440			cur->detached = 1;
 441		} else {
 442			rb_erase(&cur->rb_node, &cache->rb_root);
 443			btrfs_backref_free_node(cache, cur);
 444		}
 445	}
 446	return ret;
 
 447}
 448
 449/*
 450 * Build backref tree for a given tree block. Root of the backref tree
 451 * corresponds the tree block, leaves of the backref tree correspond roots of
 452 * b-trees that reference the tree block.
 453 *
 454 * The basic idea of this function is check backrefs of a given block to find
 455 * upper level blocks that reference the block, and then check backrefs of
 456 * these upper level blocks recursively. The recursion stops when tree root is
 457 * reached or backrefs for the block is cached.
 458 *
 459 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
 460 * all upper level blocks that directly/indirectly reference the block are also
 461 * cached.
 462 */
 463static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
 464			struct btrfs_trans_handle *trans,
 465			struct reloc_control *rc, struct btrfs_key *node_key,
 466			int level, u64 bytenr)
 467{
 468	struct btrfs_backref_iter *iter;
 469	struct btrfs_backref_cache *cache = &rc->backref_cache;
 470	/* For searching parent of TREE_BLOCK_REF */
 471	struct btrfs_path *path;
 472	struct btrfs_backref_node *cur;
 473	struct btrfs_backref_node *node = NULL;
 474	struct btrfs_backref_edge *edge;
 
 
 
 
 
 
 
 
 
 
 
 475	int ret;
 476	int err = 0;
 
 477
 478	iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info);
 479	if (!iter)
 480		return ERR_PTR(-ENOMEM);
 481	path = btrfs_alloc_path();
 482	if (!path) {
 483		err = -ENOMEM;
 484		goto out;
 485	}
 
 
 486
 487	node = btrfs_backref_alloc_node(cache, bytenr, level);
 488	if (!node) {
 489		err = -ENOMEM;
 490		goto out;
 491	}
 492
 
 
 493	node->lowest = 1;
 494	cur = node;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 495
 496	/* Breadth-first search to build backref cache */
 497	do {
 498		ret = btrfs_backref_add_tree_node(trans, cache, path, iter,
 499						  node_key, cur);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 500		if (ret < 0) {
 501			err = ret;
 502			goto out;
 503		}
 504		edge = list_first_entry_or_null(&cache->pending_edge,
 505				struct btrfs_backref_edge, list[UPPER]);
 506		/*
 507		 * The pending list isn't empty, take the first block to
 508		 * process
 509		 */
 510		if (edge) {
 511			list_del_init(&edge->list[UPPER]);
 512			cur = edge->node[UPPER];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 513		}
 514	} while (edge);
 515
 516	/* Finish the upper linkage of newly added edges/nodes */
 517	ret = btrfs_backref_finish_upper_links(cache, node);
 518	if (ret < 0) {
 519		err = ret;
 520		goto out;
 521	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 522
 523	if (handle_useless_nodes(rc, node))
 524		node = NULL;
 
 
 
 
 
 
 
 
 
 
 525out:
 526	btrfs_free_path(iter->path);
 527	kfree(iter);
 528	btrfs_free_path(path);
 529	if (err) {
 530		btrfs_backref_error_cleanup(cache, node);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 531		return ERR_PTR(err);
 532	}
 533	ASSERT(!node || !node->detached);
 534	ASSERT(list_empty(&cache->useless_node) &&
 535	       list_empty(&cache->pending_edge));
 536	return node;
 537}
 538
 539/*
 540 * helper to add backref node for the newly created snapshot.
 541 * the backref node is created by cloning backref node that
 542 * corresponds to root of source tree
 543 */
 544static int clone_backref_node(struct btrfs_trans_handle *trans,
 545			      struct reloc_control *rc,
 546			      const struct btrfs_root *src,
 547			      struct btrfs_root *dest)
 548{
 549	struct btrfs_root *reloc_root = src->reloc_root;
 550	struct btrfs_backref_cache *cache = &rc->backref_cache;
 551	struct btrfs_backref_node *node = NULL;
 552	struct btrfs_backref_node *new_node;
 553	struct btrfs_backref_edge *edge;
 554	struct btrfs_backref_edge *new_edge;
 555	struct rb_node *rb_node;
 556
 557	if (cache->last_trans > 0)
 558		update_backref_cache(trans, cache);
 559
 560	rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
 561	if (rb_node) {
 562		node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
 563		if (node->detached)
 564			node = NULL;
 565		else
 566			BUG_ON(node->new_bytenr != reloc_root->node->start);
 567	}
 568
 569	if (!node) {
 570		rb_node = rb_simple_search(&cache->rb_root,
 571					   reloc_root->commit_root->start);
 572		if (rb_node) {
 573			node = rb_entry(rb_node, struct btrfs_backref_node,
 574					rb_node);
 575			BUG_ON(node->detached);
 576		}
 577	}
 578
 579	if (!node)
 580		return 0;
 581
 582	new_node = btrfs_backref_alloc_node(cache, dest->node->start,
 583					    node->level);
 584	if (!new_node)
 585		return -ENOMEM;
 586
 
 
 587	new_node->lowest = node->lowest;
 588	new_node->checked = 1;
 589	new_node->root = btrfs_grab_root(dest);
 590	ASSERT(new_node->root);
 591
 592	if (!node->lowest) {
 593		list_for_each_entry(edge, &node->lower, list[UPPER]) {
 594			new_edge = btrfs_backref_alloc_edge(cache);
 595			if (!new_edge)
 596				goto fail;
 597
 598			btrfs_backref_link_edge(new_edge, edge->node[LOWER],
 599						new_node, LINK_UPPER);
 
 
 600		}
 601	} else {
 602		list_add_tail(&new_node->lower, &cache->leaves);
 603	}
 604
 605	rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
 606				   &new_node->rb_node);
 607	if (rb_node)
 608		btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
 609
 610	if (!new_node->lowest) {
 611		list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
 612			list_add_tail(&new_edge->list[LOWER],
 613				      &new_edge->node[LOWER]->upper);
 614		}
 615	}
 616	return 0;
 617fail:
 618	while (!list_empty(&new_node->lower)) {
 619		new_edge = list_entry(new_node->lower.next,
 620				      struct btrfs_backref_edge, list[UPPER]);
 621		list_del(&new_edge->list[UPPER]);
 622		btrfs_backref_free_edge(cache, new_edge);
 623	}
 624	btrfs_backref_free_node(cache, new_node);
 625	return -ENOMEM;
 626}
 627
 628/*
 629 * helper to add 'address of tree root -> reloc tree' mapping
 630 */
 631static int __add_reloc_root(struct btrfs_root *root)
 632{
 633	struct btrfs_fs_info *fs_info = root->fs_info;
 634	struct rb_node *rb_node;
 635	struct mapping_node *node;
 636	struct reloc_control *rc = fs_info->reloc_ctl;
 637
 638	node = kmalloc(sizeof(*node), GFP_NOFS);
 639	if (!node)
 640		return -ENOMEM;
 641
 642	node->bytenr = root->commit_root->start;
 643	node->data = root;
 644
 645	spin_lock(&rc->reloc_root_tree.lock);
 646	rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
 647				   node->bytenr, &node->rb_node);
 648	spin_unlock(&rc->reloc_root_tree.lock);
 649	if (rb_node) {
 650		btrfs_err(fs_info,
 651			    "Duplicate root found for start=%llu while inserting into relocation tree",
 652			    node->bytenr);
 653		return -EEXIST;
 654	}
 655
 656	list_add_tail(&root->root_list, &rc->reloc_roots);
 657	return 0;
 658}
 659
 660/*
 661 * helper to delete the 'address of tree root -> reloc tree'
 662 * mapping
 663 */
 664static void __del_reloc_root(struct btrfs_root *root)
 665{
 666	struct btrfs_fs_info *fs_info = root->fs_info;
 667	struct rb_node *rb_node;
 668	struct mapping_node *node = NULL;
 669	struct reloc_control *rc = fs_info->reloc_ctl;
 670	bool put_ref = false;
 671
 672	if (rc && root->node) {
 673		spin_lock(&rc->reloc_root_tree.lock);
 674		rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
 675					   root->commit_root->start);
 676		if (rb_node) {
 677			node = rb_entry(rb_node, struct mapping_node, rb_node);
 678			rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 679			RB_CLEAR_NODE(&node->rb_node);
 680		}
 681		spin_unlock(&rc->reloc_root_tree.lock);
 682		ASSERT(!node || (struct btrfs_root *)node->data == root);
 
 
 683	}
 684
 685	/*
 686	 * We only put the reloc root here if it's on the list.  There's a lot
 687	 * of places where the pattern is to splice the rc->reloc_roots, process
 688	 * the reloc roots, and then add the reloc root back onto
 689	 * rc->reloc_roots.  If we call __del_reloc_root while it's off of the
 690	 * list we don't want the reference being dropped, because the guy
 691	 * messing with the list is in charge of the reference.
 692	 */
 693	spin_lock(&fs_info->trans_lock);
 694	if (!list_empty(&root->root_list)) {
 695		put_ref = true;
 696		list_del_init(&root->root_list);
 697	}
 698	spin_unlock(&fs_info->trans_lock);
 699	if (put_ref)
 700		btrfs_put_root(root);
 701	kfree(node);
 702}
 703
 704/*
 705 * helper to update the 'address of tree root -> reloc tree'
 706 * mapping
 707 */
 708static int __update_reloc_root(struct btrfs_root *root)
 709{
 710	struct btrfs_fs_info *fs_info = root->fs_info;
 711	struct rb_node *rb_node;
 712	struct mapping_node *node = NULL;
 713	struct reloc_control *rc = fs_info->reloc_ctl;
 714
 715	spin_lock(&rc->reloc_root_tree.lock);
 716	rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
 717				   root->commit_root->start);
 718	if (rb_node) {
 719		node = rb_entry(rb_node, struct mapping_node, rb_node);
 720		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 721	}
 722	spin_unlock(&rc->reloc_root_tree.lock);
 723
 724	if (!node)
 725		return 0;
 726	BUG_ON((struct btrfs_root *)node->data != root);
 727
 728	spin_lock(&rc->reloc_root_tree.lock);
 729	node->bytenr = root->node->start;
 730	rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
 731				   node->bytenr, &node->rb_node);
 732	spin_unlock(&rc->reloc_root_tree.lock);
 733	if (rb_node)
 734		btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
 735	return 0;
 736}
 737
 738static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
 739					struct btrfs_root *root, u64 objectid)
 740{
 741	struct btrfs_fs_info *fs_info = root->fs_info;
 742	struct btrfs_root *reloc_root;
 743	struct extent_buffer *eb;
 744	struct btrfs_root_item *root_item;
 745	struct btrfs_key root_key;
 746	int ret = 0;
 747	bool must_abort = false;
 748
 749	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
 750	if (!root_item)
 751		return ERR_PTR(-ENOMEM);
 752
 753	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
 754	root_key.type = BTRFS_ROOT_ITEM_KEY;
 755	root_key.offset = objectid;
 756
 757	if (root->root_key.objectid == objectid) {
 758		u64 commit_root_gen;
 759
 760		/* called by btrfs_init_reloc_root */
 761		ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
 762				      BTRFS_TREE_RELOC_OBJECTID);
 763		if (ret)
 764			goto fail;
 765
 766		/*
 767		 * Set the last_snapshot field to the generation of the commit
 768		 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
 769		 * correctly (returns true) when the relocation root is created
 770		 * either inside the critical section of a transaction commit
 771		 * (through transaction.c:qgroup_account_snapshot()) and when
 772		 * it's created before the transaction commit is started.
 773		 */
 774		commit_root_gen = btrfs_header_generation(root->commit_root);
 775		btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
 776	} else {
 777		/*
 778		 * called by btrfs_reloc_post_snapshot_hook.
 779		 * the source tree is a reloc tree, all tree blocks
 780		 * modified after it was created have RELOC flag
 781		 * set in their headers. so it's OK to not update
 782		 * the 'last_snapshot'.
 783		 */
 784		ret = btrfs_copy_root(trans, root, root->node, &eb,
 785				      BTRFS_TREE_RELOC_OBJECTID);
 786		if (ret)
 787			goto fail;
 788	}
 789
 790	/*
 791	 * We have changed references at this point, we must abort the
 792	 * transaction if anything fails.
 793	 */
 794	must_abort = true;
 795
 796	memcpy(root_item, &root->root_item, sizeof(*root_item));
 797	btrfs_set_root_bytenr(root_item, eb->start);
 798	btrfs_set_root_level(root_item, btrfs_header_level(eb));
 799	btrfs_set_root_generation(root_item, trans->transid);
 800
 801	if (root->root_key.objectid == objectid) {
 802		btrfs_set_root_refs(root_item, 0);
 803		memset(&root_item->drop_progress, 0,
 804		       sizeof(struct btrfs_disk_key));
 805		btrfs_set_root_drop_level(root_item, 0);
 806	}
 807
 808	btrfs_tree_unlock(eb);
 809	free_extent_buffer(eb);
 810
 811	ret = btrfs_insert_root(trans, fs_info->tree_root,
 812				&root_key, root_item);
 813	if (ret)
 814		goto fail;
 815
 816	kfree(root_item);
 817
 818	reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
 819	if (IS_ERR(reloc_root)) {
 820		ret = PTR_ERR(reloc_root);
 821		goto abort;
 822	}
 823	set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
 824	reloc_root->last_trans = trans->transid;
 825	return reloc_root;
 826fail:
 827	kfree(root_item);
 828abort:
 829	if (must_abort)
 830		btrfs_abort_transaction(trans, ret);
 831	return ERR_PTR(ret);
 832}
 833
 834/*
 835 * create reloc tree for a given fs tree. reloc tree is just a
 836 * snapshot of the fs tree with special root objectid.
 837 *
 838 * The reloc_root comes out of here with two references, one for
 839 * root->reloc_root, and another for being on the rc->reloc_roots list.
 840 */
 841int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
 842			  struct btrfs_root *root)
 843{
 844	struct btrfs_fs_info *fs_info = root->fs_info;
 845	struct btrfs_root *reloc_root;
 846	struct reloc_control *rc = fs_info->reloc_ctl;
 847	struct btrfs_block_rsv *rsv;
 848	int clear_rsv = 0;
 849	int ret;
 850
 851	if (!rc)
 852		return 0;
 853
 854	/*
 855	 * The subvolume has reloc tree but the swap is finished, no need to
 856	 * create/update the dead reloc tree
 857	 */
 858	if (reloc_root_is_dead(root))
 859		return 0;
 860
 861	/*
 862	 * This is subtle but important.  We do not do
 863	 * record_root_in_transaction for reloc roots, instead we record their
 864	 * corresponding fs root, and then here we update the last trans for the
 865	 * reloc root.  This means that we have to do this for the entire life
 866	 * of the reloc root, regardless of which stage of the relocation we are
 867	 * in.
 868	 */
 869	if (root->reloc_root) {
 870		reloc_root = root->reloc_root;
 871		reloc_root->last_trans = trans->transid;
 872		return 0;
 873	}
 874
 875	/*
 876	 * We are merging reloc roots, we do not need new reloc trees.  Also
 877	 * reloc trees never need their own reloc tree.
 878	 */
 879	if (!rc->create_reloc_tree ||
 880	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
 881		return 0;
 882
 883	if (!trans->reloc_reserved) {
 884		rsv = trans->block_rsv;
 885		trans->block_rsv = rc->block_rsv;
 886		clear_rsv = 1;
 887	}
 888	reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
 889	if (clear_rsv)
 890		trans->block_rsv = rsv;
 891	if (IS_ERR(reloc_root))
 892		return PTR_ERR(reloc_root);
 893
 894	ret = __add_reloc_root(reloc_root);
 895	ASSERT(ret != -EEXIST);
 896	if (ret) {
 897		/* Pairs with create_reloc_root */
 898		btrfs_put_root(reloc_root);
 899		return ret;
 900	}
 901	root->reloc_root = btrfs_grab_root(reloc_root);
 902	return 0;
 903}
 904
 905/*
 906 * update root item of reloc tree
 907 */
 908int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
 909			    struct btrfs_root *root)
 910{
 911	struct btrfs_fs_info *fs_info = root->fs_info;
 912	struct btrfs_root *reloc_root;
 913	struct btrfs_root_item *root_item;
 914	int ret;
 915
 916	if (!have_reloc_root(root))
 917		return 0;
 
 918
 919	reloc_root = root->reloc_root;
 920	root_item = &reloc_root->root_item;
 921
 922	/*
 923	 * We are probably ok here, but __del_reloc_root() will drop its ref of
 924	 * the root.  We have the ref for root->reloc_root, but just in case
 925	 * hold it while we update the reloc root.
 926	 */
 927	btrfs_grab_root(reloc_root);
 928
 929	/* root->reloc_root will stay until current relocation finished */
 930	if (fs_info->reloc_ctl->merge_reloc_tree &&
 931	    btrfs_root_refs(root_item) == 0) {
 932		set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
 933		/*
 934		 * Mark the tree as dead before we change reloc_root so
 935		 * have_reloc_root will not touch it from now on.
 936		 */
 937		smp_wmb();
 938		__del_reloc_root(reloc_root);
 939	}
 940
 941	if (reloc_root->commit_root != reloc_root->node) {
 942		__update_reloc_root(reloc_root);
 943		btrfs_set_root_node(root_item, reloc_root->node);
 944		free_extent_buffer(reloc_root->commit_root);
 945		reloc_root->commit_root = btrfs_root_node(reloc_root);
 946	}
 947
 948	ret = btrfs_update_root(trans, fs_info->tree_root,
 949				&reloc_root->root_key, root_item);
 950	btrfs_put_root(reloc_root);
 951	return ret;
 
 
 952}
 953
 954/*
 955 * helper to find first cached inode with inode number >= objectid
 956 * in a subvolume
 957 */
 958static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
 959{
 960	struct rb_node *node;
 961	struct rb_node *prev;
 962	struct btrfs_inode *entry;
 963	struct inode *inode;
 964
 965	spin_lock(&root->inode_lock);
 966again:
 967	node = root->inode_tree.rb_node;
 968	prev = NULL;
 969	while (node) {
 970		prev = node;
 971		entry = rb_entry(node, struct btrfs_inode, rb_node);
 972
 973		if (objectid < btrfs_ino(entry))
 974			node = node->rb_left;
 975		else if (objectid > btrfs_ino(entry))
 976			node = node->rb_right;
 977		else
 978			break;
 979	}
 980	if (!node) {
 981		while (prev) {
 982			entry = rb_entry(prev, struct btrfs_inode, rb_node);
 983			if (objectid <= btrfs_ino(entry)) {
 984				node = prev;
 985				break;
 986			}
 987			prev = rb_next(prev);
 988		}
 989	}
 990	while (node) {
 991		entry = rb_entry(node, struct btrfs_inode, rb_node);
 992		inode = igrab(&entry->vfs_inode);
 993		if (inode) {
 994			spin_unlock(&root->inode_lock);
 995			return inode;
 996		}
 997
 998		objectid = btrfs_ino(entry) + 1;
 999		if (cond_resched_lock(&root->inode_lock))
1000			goto again;
1001
1002		node = rb_next(node);
1003	}
1004	spin_unlock(&root->inode_lock);
1005	return NULL;
1006}
1007
 
 
 
 
 
 
 
 
 
1008/*
1009 * get new location of data
1010 */
1011static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1012			    u64 bytenr, u64 num_bytes)
1013{
1014	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1015	struct btrfs_path *path;
1016	struct btrfs_file_extent_item *fi;
1017	struct extent_buffer *leaf;
1018	int ret;
1019
1020	path = btrfs_alloc_path();
1021	if (!path)
1022		return -ENOMEM;
1023
1024	bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1025	ret = btrfs_lookup_file_extent(NULL, root, path,
1026			btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1027	if (ret < 0)
1028		goto out;
1029	if (ret > 0) {
1030		ret = -ENOENT;
1031		goto out;
1032	}
1033
1034	leaf = path->nodes[0];
1035	fi = btrfs_item_ptr(leaf, path->slots[0],
1036			    struct btrfs_file_extent_item);
1037
1038	BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1039	       btrfs_file_extent_compression(leaf, fi) ||
1040	       btrfs_file_extent_encryption(leaf, fi) ||
1041	       btrfs_file_extent_other_encoding(leaf, fi));
1042
1043	if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1044		ret = -EINVAL;
1045		goto out;
1046	}
1047
1048	*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1049	ret = 0;
1050out:
1051	btrfs_free_path(path);
1052	return ret;
1053}
1054
1055/*
1056 * update file extent items in the tree leaf to point to
1057 * the new locations.
1058 */
1059static noinline_for_stack
1060int replace_file_extents(struct btrfs_trans_handle *trans,
1061			 struct reloc_control *rc,
1062			 struct btrfs_root *root,
1063			 struct extent_buffer *leaf)
1064{
1065	struct btrfs_fs_info *fs_info = root->fs_info;
1066	struct btrfs_key key;
1067	struct btrfs_file_extent_item *fi;
1068	struct inode *inode = NULL;
1069	u64 parent;
1070	u64 bytenr;
1071	u64 new_bytenr = 0;
1072	u64 num_bytes;
1073	u64 end;
1074	u32 nritems;
1075	u32 i;
1076	int ret = 0;
1077	int first = 1;
1078	int dirty = 0;
1079
1080	if (rc->stage != UPDATE_DATA_PTRS)
1081		return 0;
1082
1083	/* reloc trees always use full backref */
1084	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1085		parent = leaf->start;
1086	else
1087		parent = 0;
1088
1089	nritems = btrfs_header_nritems(leaf);
1090	for (i = 0; i < nritems; i++) {
1091		struct btrfs_ref ref = { 0 };
1092
1093		cond_resched();
1094		btrfs_item_key_to_cpu(leaf, &key, i);
1095		if (key.type != BTRFS_EXTENT_DATA_KEY)
1096			continue;
1097		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1098		if (btrfs_file_extent_type(leaf, fi) ==
1099		    BTRFS_FILE_EXTENT_INLINE)
1100			continue;
1101		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1102		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1103		if (bytenr == 0)
1104			continue;
1105		if (!in_range(bytenr, rc->block_group->start,
1106			      rc->block_group->length))
1107			continue;
1108
1109		/*
1110		 * if we are modifying block in fs tree, wait for read_folio
1111		 * to complete and drop the extent cache
1112		 */
1113		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1114			if (first) {
1115				inode = find_next_inode(root, key.objectid);
1116				first = 0;
1117			} else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1118				btrfs_add_delayed_iput(BTRFS_I(inode));
1119				inode = find_next_inode(root, key.objectid);
1120			}
1121			if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1122				struct extent_state *cached_state = NULL;
1123
1124				end = key.offset +
1125				      btrfs_file_extent_num_bytes(leaf, fi);
1126				WARN_ON(!IS_ALIGNED(key.offset,
1127						    fs_info->sectorsize));
1128				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1129				end--;
1130				ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1131						      key.offset, end,
1132						      &cached_state);
1133				if (!ret)
1134					continue;
1135
1136				btrfs_drop_extent_map_range(BTRFS_I(inode),
1137							    key.offset, end, true);
1138				unlock_extent(&BTRFS_I(inode)->io_tree,
1139					      key.offset, end, &cached_state);
1140			}
1141		}
1142
1143		ret = get_new_location(rc->data_inode, &new_bytenr,
1144				       bytenr, num_bytes);
1145		if (ret) {
1146			/*
1147			 * Don't have to abort since we've not changed anything
1148			 * in the file extent yet.
1149			 */
1150			break;
1151		}
1152
1153		btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1154		dirty = 1;
1155
1156		key.offset -= btrfs_file_extent_offset(leaf, fi);
1157		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1158				       num_bytes, parent, root->root_key.objectid);
 
1159		btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1160				    key.objectid, key.offset,
1161				    root->root_key.objectid, false);
1162		ret = btrfs_inc_extent_ref(trans, &ref);
1163		if (ret) {
1164			btrfs_abort_transaction(trans, ret);
1165			break;
1166		}
1167
1168		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1169				       num_bytes, parent, root->root_key.objectid);
 
1170		btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1171				    key.objectid, key.offset,
1172				    root->root_key.objectid, false);
1173		ret = btrfs_free_extent(trans, &ref);
1174		if (ret) {
1175			btrfs_abort_transaction(trans, ret);
1176			break;
1177		}
1178	}
1179	if (dirty)
1180		btrfs_mark_buffer_dirty(trans, leaf);
1181	if (inode)
1182		btrfs_add_delayed_iput(BTRFS_I(inode));
1183	return ret;
1184}
1185
1186static noinline_for_stack int memcmp_node_keys(const struct extent_buffer *eb,
1187					       int slot, const struct btrfs_path *path,
1188					       int level)
1189{
1190	struct btrfs_disk_key key1;
1191	struct btrfs_disk_key key2;
1192	btrfs_node_key(eb, &key1, slot);
1193	btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1194	return memcmp(&key1, &key2, sizeof(key1));
1195}
1196
1197/*
1198 * try to replace tree blocks in fs tree with the new blocks
1199 * in reloc tree. tree blocks haven't been modified since the
1200 * reloc tree was create can be replaced.
1201 *
1202 * if a block was replaced, level of the block + 1 is returned.
1203 * if no block got replaced, 0 is returned. if there are other
1204 * errors, a negative error number is returned.
1205 */
1206static noinline_for_stack
1207int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1208		 struct btrfs_root *dest, struct btrfs_root *src,
1209		 struct btrfs_path *path, struct btrfs_key *next_key,
1210		 int lowest_level, int max_level)
1211{
1212	struct btrfs_fs_info *fs_info = dest->fs_info;
1213	struct extent_buffer *eb;
1214	struct extent_buffer *parent;
1215	struct btrfs_ref ref = { 0 };
1216	struct btrfs_key key;
1217	u64 old_bytenr;
1218	u64 new_bytenr;
1219	u64 old_ptr_gen;
1220	u64 new_ptr_gen;
1221	u64 last_snapshot;
1222	u32 blocksize;
1223	int cow = 0;
1224	int level;
1225	int ret;
1226	int slot;
1227
1228	ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1229	ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1230
1231	last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1232again:
1233	slot = path->slots[lowest_level];
1234	btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1235
1236	eb = btrfs_lock_root_node(dest);
 
1237	level = btrfs_header_level(eb);
1238
1239	if (level < lowest_level) {
1240		btrfs_tree_unlock(eb);
1241		free_extent_buffer(eb);
1242		return 0;
1243	}
1244
1245	if (cow) {
1246		ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1247				      BTRFS_NESTING_COW);
1248		if (ret) {
1249			btrfs_tree_unlock(eb);
1250			free_extent_buffer(eb);
1251			return ret;
1252		}
1253	}
 
1254
1255	if (next_key) {
1256		next_key->objectid = (u64)-1;
1257		next_key->type = (u8)-1;
1258		next_key->offset = (u64)-1;
1259	}
1260
1261	parent = eb;
1262	while (1) {
 
 
1263		level = btrfs_header_level(parent);
1264		ASSERT(level >= lowest_level);
1265
1266		ret = btrfs_bin_search(parent, 0, &key, &slot);
1267		if (ret < 0)
1268			break;
1269		if (ret && slot > 0)
1270			slot--;
1271
1272		if (next_key && slot + 1 < btrfs_header_nritems(parent))
1273			btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1274
1275		old_bytenr = btrfs_node_blockptr(parent, slot);
1276		blocksize = fs_info->nodesize;
1277		old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
 
1278
1279		if (level <= max_level) {
1280			eb = path->nodes[level];
1281			new_bytenr = btrfs_node_blockptr(eb,
1282							path->slots[level]);
1283			new_ptr_gen = btrfs_node_ptr_generation(eb,
1284							path->slots[level]);
1285		} else {
1286			new_bytenr = 0;
1287			new_ptr_gen = 0;
1288		}
1289
1290		if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1291			ret = level;
1292			break;
1293		}
1294
1295		if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1296		    memcmp_node_keys(parent, slot, path, level)) {
1297			if (level <= lowest_level) {
1298				ret = 0;
1299				break;
1300			}
1301
1302			eb = btrfs_read_node_slot(parent, slot);
 
1303			if (IS_ERR(eb)) {
1304				ret = PTR_ERR(eb);
1305				break;
 
 
 
 
1306			}
1307			btrfs_tree_lock(eb);
1308			if (cow) {
1309				ret = btrfs_cow_block(trans, dest, eb, parent,
1310						      slot, &eb,
1311						      BTRFS_NESTING_COW);
1312				if (ret) {
1313					btrfs_tree_unlock(eb);
1314					free_extent_buffer(eb);
1315					break;
1316				}
1317			}
 
1318
1319			btrfs_tree_unlock(parent);
1320			free_extent_buffer(parent);
1321
1322			parent = eb;
1323			continue;
1324		}
1325
1326		if (!cow) {
1327			btrfs_tree_unlock(parent);
1328			free_extent_buffer(parent);
1329			cow = 1;
1330			goto again;
1331		}
1332
1333		btrfs_node_key_to_cpu(path->nodes[level], &key,
1334				      path->slots[level]);
1335		btrfs_release_path(path);
1336
1337		path->lowest_level = level;
1338		set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1339		ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1340		clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1341		path->lowest_level = 0;
1342		if (ret) {
1343			if (ret > 0)
1344				ret = -ENOENT;
1345			break;
1346		}
1347
1348		/*
1349		 * Info qgroup to trace both subtrees.
1350		 *
1351		 * We must trace both trees.
1352		 * 1) Tree reloc subtree
1353		 *    If not traced, we will leak data numbers
1354		 * 2) Fs subtree
1355		 *    If not traced, we will double count old data
1356		 *
1357		 * We don't scan the subtree right now, but only record
1358		 * the swapped tree blocks.
1359		 * The real subtree rescan is delayed until we have new
1360		 * CoW on the subtree root node before transaction commit.
1361		 */
1362		ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1363				rc->block_group, parent, slot,
1364				path->nodes[level], path->slots[level],
1365				last_snapshot);
1366		if (ret < 0)
1367			break;
1368		/*
1369		 * swap blocks in fs tree and reloc tree.
1370		 */
1371		btrfs_set_node_blockptr(parent, slot, new_bytenr);
1372		btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1373		btrfs_mark_buffer_dirty(trans, parent);
1374
1375		btrfs_set_node_blockptr(path->nodes[level],
1376					path->slots[level], old_bytenr);
1377		btrfs_set_node_ptr_generation(path->nodes[level],
1378					      path->slots[level], old_ptr_gen);
1379		btrfs_mark_buffer_dirty(trans, path->nodes[level]);
1380
1381		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1382				       blocksize, path->nodes[level]->start,
1383				       src->root_key.objectid);
1384		btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid,
1385				    0, true);
1386		ret = btrfs_inc_extent_ref(trans, &ref);
1387		if (ret) {
1388			btrfs_abort_transaction(trans, ret);
1389			break;
1390		}
1391		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1392				       blocksize, 0, dest->root_key.objectid);
1393		btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0,
1394				    true);
1395		ret = btrfs_inc_extent_ref(trans, &ref);
1396		if (ret) {
1397			btrfs_abort_transaction(trans, ret);
1398			break;
1399		}
1400
1401		/* We don't know the real owning_root, use 0. */
1402		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1403				       blocksize, path->nodes[level]->start, 0);
1404		btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid,
1405				    0, true);
1406		ret = btrfs_free_extent(trans, &ref);
1407		if (ret) {
1408			btrfs_abort_transaction(trans, ret);
1409			break;
1410		}
1411
1412		/* We don't know the real owning_root, use 0. */
1413		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1414				       blocksize, 0, 0);
1415		btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid,
1416				    0, true);
1417		ret = btrfs_free_extent(trans, &ref);
1418		if (ret) {
1419			btrfs_abort_transaction(trans, ret);
1420			break;
1421		}
1422
1423		btrfs_unlock_up_safe(path, 0);
1424
1425		ret = level;
1426		break;
1427	}
1428	btrfs_tree_unlock(parent);
1429	free_extent_buffer(parent);
1430	return ret;
1431}
1432
1433/*
1434 * helper to find next relocated block in reloc tree
1435 */
1436static noinline_for_stack
1437int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1438		       int *level)
1439{
1440	struct extent_buffer *eb;
1441	int i;
1442	u64 last_snapshot;
1443	u32 nritems;
1444
1445	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1446
1447	for (i = 0; i < *level; i++) {
1448		free_extent_buffer(path->nodes[i]);
1449		path->nodes[i] = NULL;
1450	}
1451
1452	for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1453		eb = path->nodes[i];
1454		nritems = btrfs_header_nritems(eb);
1455		while (path->slots[i] + 1 < nritems) {
1456			path->slots[i]++;
1457			if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1458			    last_snapshot)
1459				continue;
1460
1461			*level = i;
1462			return 0;
1463		}
1464		free_extent_buffer(path->nodes[i]);
1465		path->nodes[i] = NULL;
1466	}
1467	return 1;
1468}
1469
1470/*
1471 * walk down reloc tree to find relocated block of lowest level
1472 */
1473static noinline_for_stack
1474int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1475			 int *level)
1476{
 
1477	struct extent_buffer *eb = NULL;
1478	int i;
 
1479	u64 ptr_gen = 0;
1480	u64 last_snapshot;
1481	u32 nritems;
1482
1483	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1484
1485	for (i = *level; i > 0; i--) {
 
 
1486		eb = path->nodes[i];
1487		nritems = btrfs_header_nritems(eb);
1488		while (path->slots[i] < nritems) {
1489			ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1490			if (ptr_gen > last_snapshot)
1491				break;
1492			path->slots[i]++;
1493		}
1494		if (path->slots[i] >= nritems) {
1495			if (i == *level)
1496				break;
1497			*level = i + 1;
1498			return 0;
1499		}
1500		if (i == 1) {
1501			*level = i;
1502			return 0;
1503		}
1504
1505		eb = btrfs_read_node_slot(eb, path->slots[i]);
1506		if (IS_ERR(eb))
 
 
 
1507			return PTR_ERR(eb);
 
 
 
 
1508		BUG_ON(btrfs_header_level(eb) != i - 1);
1509		path->nodes[i - 1] = eb;
1510		path->slots[i - 1] = 0;
1511	}
1512	return 1;
1513}
1514
1515/*
1516 * invalidate extent cache for file extents whose key in range of
1517 * [min_key, max_key)
1518 */
1519static int invalidate_extent_cache(struct btrfs_root *root,
1520				   const struct btrfs_key *min_key,
1521				   const struct btrfs_key *max_key)
1522{
1523	struct btrfs_fs_info *fs_info = root->fs_info;
1524	struct inode *inode = NULL;
1525	u64 objectid;
1526	u64 start, end;
1527	u64 ino;
1528
1529	objectid = min_key->objectid;
1530	while (1) {
1531		struct extent_state *cached_state = NULL;
1532
1533		cond_resched();
1534		iput(inode);
1535
1536		if (objectid > max_key->objectid)
1537			break;
1538
1539		inode = find_next_inode(root, objectid);
1540		if (!inode)
1541			break;
1542		ino = btrfs_ino(BTRFS_I(inode));
1543
1544		if (ino > max_key->objectid) {
1545			iput(inode);
1546			break;
1547		}
1548
1549		objectid = ino + 1;
1550		if (!S_ISREG(inode->i_mode))
1551			continue;
1552
1553		if (unlikely(min_key->objectid == ino)) {
1554			if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1555				continue;
1556			if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1557				start = 0;
1558			else {
1559				start = min_key->offset;
1560				WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1561			}
1562		} else {
1563			start = 0;
1564		}
1565
1566		if (unlikely(max_key->objectid == ino)) {
1567			if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1568				continue;
1569			if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1570				end = (u64)-1;
1571			} else {
1572				if (max_key->offset == 0)
1573					continue;
1574				end = max_key->offset;
1575				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1576				end--;
1577			}
1578		} else {
1579			end = (u64)-1;
1580		}
1581
1582		/* the lock_extent waits for read_folio to complete */
1583		lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
1584		btrfs_drop_extent_map_range(BTRFS_I(inode), start, end, true);
1585		unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
1586	}
1587	return 0;
1588}
1589
1590static int find_next_key(struct btrfs_path *path, int level,
1591			 struct btrfs_key *key)
1592
1593{
1594	while (level < BTRFS_MAX_LEVEL) {
1595		if (!path->nodes[level])
1596			break;
1597		if (path->slots[level] + 1 <
1598		    btrfs_header_nritems(path->nodes[level])) {
1599			btrfs_node_key_to_cpu(path->nodes[level], key,
1600					      path->slots[level] + 1);
1601			return 0;
1602		}
1603		level++;
1604	}
1605	return 1;
1606}
1607
1608/*
1609 * Insert current subvolume into reloc_control::dirty_subvol_roots
1610 */
1611static int insert_dirty_subvol(struct btrfs_trans_handle *trans,
1612			       struct reloc_control *rc,
1613			       struct btrfs_root *root)
1614{
1615	struct btrfs_root *reloc_root = root->reloc_root;
1616	struct btrfs_root_item *reloc_root_item;
1617	int ret;
1618
1619	/* @root must be a subvolume tree root with a valid reloc tree */
1620	ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1621	ASSERT(reloc_root);
1622
1623	reloc_root_item = &reloc_root->root_item;
1624	memset(&reloc_root_item->drop_progress, 0,
1625		sizeof(reloc_root_item->drop_progress));
1626	btrfs_set_root_drop_level(reloc_root_item, 0);
1627	btrfs_set_root_refs(reloc_root_item, 0);
1628	ret = btrfs_update_reloc_root(trans, root);
1629	if (ret)
1630		return ret;
1631
1632	if (list_empty(&root->reloc_dirty_list)) {
1633		btrfs_grab_root(root);
1634		list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1635	}
1636
1637	return 0;
1638}
1639
1640static int clean_dirty_subvols(struct reloc_control *rc)
1641{
1642	struct btrfs_root *root;
1643	struct btrfs_root *next;
1644	int ret = 0;
1645	int ret2;
1646
1647	list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1648				 reloc_dirty_list) {
1649		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1650			/* Merged subvolume, cleanup its reloc root */
1651			struct btrfs_root *reloc_root = root->reloc_root;
1652
1653			list_del_init(&root->reloc_dirty_list);
1654			root->reloc_root = NULL;
1655			/*
1656			 * Need barrier to ensure clear_bit() only happens after
1657			 * root->reloc_root = NULL. Pairs with have_reloc_root.
1658			 */
1659			smp_wmb();
1660			clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1661			if (reloc_root) {
1662				/*
1663				 * btrfs_drop_snapshot drops our ref we hold for
1664				 * ->reloc_root.  If it fails however we must
1665				 * drop the ref ourselves.
1666				 */
1667				ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1668				if (ret2 < 0) {
1669					btrfs_put_root(reloc_root);
1670					if (!ret)
1671						ret = ret2;
1672				}
1673			}
1674			btrfs_put_root(root);
 
1675		} else {
1676			/* Orphan reloc tree, just clean it up */
1677			ret2 = btrfs_drop_snapshot(root, 0, 1);
1678			if (ret2 < 0) {
1679				btrfs_put_root(root);
1680				if (!ret)
1681					ret = ret2;
1682			}
1683		}
1684	}
1685	return ret;
1686}
1687
1688/*
1689 * merge the relocated tree blocks in reloc tree with corresponding
1690 * fs tree.
1691 */
1692static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1693					       struct btrfs_root *root)
1694{
1695	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1696	struct btrfs_key key;
1697	struct btrfs_key next_key;
1698	struct btrfs_trans_handle *trans = NULL;
1699	struct btrfs_root *reloc_root;
1700	struct btrfs_root_item *root_item;
1701	struct btrfs_path *path;
1702	struct extent_buffer *leaf;
1703	int reserve_level;
1704	int level;
1705	int max_level;
1706	int replaced = 0;
1707	int ret = 0;
 
1708	u32 min_reserved;
1709
1710	path = btrfs_alloc_path();
1711	if (!path)
1712		return -ENOMEM;
1713	path->reada = READA_FORWARD;
1714
1715	reloc_root = root->reloc_root;
1716	root_item = &reloc_root->root_item;
1717
1718	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1719		level = btrfs_root_level(root_item);
1720		atomic_inc(&reloc_root->node->refs);
1721		path->nodes[level] = reloc_root->node;
1722		path->slots[level] = 0;
1723	} else {
1724		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1725
1726		level = btrfs_root_drop_level(root_item);
1727		BUG_ON(level == 0);
1728		path->lowest_level = level;
1729		ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1730		path->lowest_level = 0;
1731		if (ret < 0) {
1732			btrfs_free_path(path);
1733			return ret;
1734		}
1735
1736		btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1737				      path->slots[level]);
1738		WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1739
1740		btrfs_unlock_up_safe(path, 0);
1741	}
1742
1743	/*
1744	 * In merge_reloc_root(), we modify the upper level pointer to swap the
1745	 * tree blocks between reloc tree and subvolume tree.  Thus for tree
1746	 * block COW, we COW at most from level 1 to root level for each tree.
1747	 *
1748	 * Thus the needed metadata size is at most root_level * nodesize,
1749	 * and * 2 since we have two trees to COW.
1750	 */
1751	reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1752	min_reserved = fs_info->nodesize * reserve_level * 2;
1753	memset(&next_key, 0, sizeof(next_key));
1754
1755	while (1) {
1756		ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
1757					     min_reserved,
1758					     BTRFS_RESERVE_FLUSH_LIMIT);
1759		if (ret)
1760			goto out;
 
1761		trans = btrfs_start_transaction(root, 0);
1762		if (IS_ERR(trans)) {
1763			ret = PTR_ERR(trans);
1764			trans = NULL;
1765			goto out;
1766		}
1767
1768		/*
1769		 * At this point we no longer have a reloc_control, so we can't
1770		 * depend on btrfs_init_reloc_root to update our last_trans.
1771		 *
1772		 * But that's ok, we started the trans handle on our
1773		 * corresponding fs_root, which means it's been added to the
1774		 * dirty list.  At commit time we'll still call
1775		 * btrfs_update_reloc_root() and update our root item
1776		 * appropriately.
1777		 */
1778		reloc_root->last_trans = trans->transid;
1779		trans->block_rsv = rc->block_rsv;
1780
1781		replaced = 0;
1782		max_level = level;
1783
1784		ret = walk_down_reloc_tree(reloc_root, path, &level);
1785		if (ret < 0)
 
1786			goto out;
 
1787		if (ret > 0)
1788			break;
1789
1790		if (!find_next_key(path, level, &key) &&
1791		    btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1792			ret = 0;
1793		} else {
1794			ret = replace_path(trans, rc, root, reloc_root, path,
1795					   &next_key, level, max_level);
1796		}
1797		if (ret < 0)
 
1798			goto out;
 
 
1799		if (ret > 0) {
1800			level = ret;
1801			btrfs_node_key_to_cpu(path->nodes[level], &key,
1802					      path->slots[level]);
1803			replaced = 1;
1804		}
1805
1806		ret = walk_up_reloc_tree(reloc_root, path, &level);
1807		if (ret > 0)
1808			break;
1809
1810		BUG_ON(level == 0);
1811		/*
1812		 * save the merging progress in the drop_progress.
1813		 * this is OK since root refs == 1 in this case.
1814		 */
1815		btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1816			       path->slots[level]);
1817		btrfs_set_root_drop_level(root_item, level);
1818
1819		btrfs_end_transaction_throttle(trans);
1820		trans = NULL;
1821
1822		btrfs_btree_balance_dirty(fs_info);
1823
1824		if (replaced && rc->stage == UPDATE_DATA_PTRS)
1825			invalidate_extent_cache(root, &key, &next_key);
1826	}
1827
1828	/*
1829	 * handle the case only one block in the fs tree need to be
1830	 * relocated and the block is tree root.
1831	 */
1832	leaf = btrfs_lock_root_node(root);
1833	ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1834			      BTRFS_NESTING_COW);
1835	btrfs_tree_unlock(leaf);
1836	free_extent_buffer(leaf);
 
 
1837out:
1838	btrfs_free_path(path);
1839
1840	if (ret == 0) {
1841		ret = insert_dirty_subvol(trans, rc, root);
1842		if (ret)
1843			btrfs_abort_transaction(trans, ret);
1844	}
1845
1846	if (trans)
1847		btrfs_end_transaction_throttle(trans);
1848
1849	btrfs_btree_balance_dirty(fs_info);
1850
1851	if (replaced && rc->stage == UPDATE_DATA_PTRS)
1852		invalidate_extent_cache(root, &key, &next_key);
1853
1854	return ret;
1855}
1856
1857static noinline_for_stack
1858int prepare_to_merge(struct reloc_control *rc, int err)
1859{
1860	struct btrfs_root *root = rc->extent_root;
1861	struct btrfs_fs_info *fs_info = root->fs_info;
1862	struct btrfs_root *reloc_root;
1863	struct btrfs_trans_handle *trans;
1864	LIST_HEAD(reloc_roots);
1865	u64 num_bytes = 0;
1866	int ret;
1867
1868	mutex_lock(&fs_info->reloc_mutex);
1869	rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1870	rc->merging_rsv_size += rc->nodes_relocated * 2;
1871	mutex_unlock(&fs_info->reloc_mutex);
1872
1873again:
1874	if (!err) {
1875		num_bytes = rc->merging_rsv_size;
1876		ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes,
1877					  BTRFS_RESERVE_FLUSH_ALL);
1878		if (ret)
1879			err = ret;
1880	}
1881
1882	trans = btrfs_join_transaction(rc->extent_root);
1883	if (IS_ERR(trans)) {
1884		if (!err)
1885			btrfs_block_rsv_release(fs_info, rc->block_rsv,
1886						num_bytes, NULL);
1887		return PTR_ERR(trans);
1888	}
1889
1890	if (!err) {
1891		if (num_bytes != rc->merging_rsv_size) {
1892			btrfs_end_transaction(trans);
1893			btrfs_block_rsv_release(fs_info, rc->block_rsv,
1894						num_bytes, NULL);
1895			goto again;
1896		}
1897	}
1898
1899	rc->merge_reloc_tree = true;
1900
1901	while (!list_empty(&rc->reloc_roots)) {
1902		reloc_root = list_entry(rc->reloc_roots.next,
1903					struct btrfs_root, root_list);
1904		list_del_init(&reloc_root->root_list);
1905
1906		root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1907				false);
1908		if (IS_ERR(root)) {
1909			/*
1910			 * Even if we have an error we need this reloc root
1911			 * back on our list so we can clean up properly.
1912			 */
1913			list_add(&reloc_root->root_list, &reloc_roots);
1914			btrfs_abort_transaction(trans, (int)PTR_ERR(root));
1915			if (!err)
1916				err = PTR_ERR(root);
1917			break;
1918		}
1919
1920		if (unlikely(root->reloc_root != reloc_root)) {
1921			if (root->reloc_root) {
1922				btrfs_err(fs_info,
1923"reloc tree mismatch, root %lld has reloc root key (%lld %u %llu) gen %llu, expect reloc root key (%lld %u %llu) gen %llu",
1924					  root->root_key.objectid,
1925					  root->reloc_root->root_key.objectid,
1926					  root->reloc_root->root_key.type,
1927					  root->reloc_root->root_key.offset,
1928					  btrfs_root_generation(
1929						  &root->reloc_root->root_item),
1930					  reloc_root->root_key.objectid,
1931					  reloc_root->root_key.type,
1932					  reloc_root->root_key.offset,
1933					  btrfs_root_generation(
1934						  &reloc_root->root_item));
1935			} else {
1936				btrfs_err(fs_info,
1937"reloc tree mismatch, root %lld has no reloc root, expect reloc root key (%lld %u %llu) gen %llu",
1938					  root->root_key.objectid,
1939					  reloc_root->root_key.objectid,
1940					  reloc_root->root_key.type,
1941					  reloc_root->root_key.offset,
1942					  btrfs_root_generation(
1943						  &reloc_root->root_item));
1944			}
1945			list_add(&reloc_root->root_list, &reloc_roots);
1946			btrfs_put_root(root);
1947			btrfs_abort_transaction(trans, -EUCLEAN);
1948			if (!err)
1949				err = -EUCLEAN;
1950			break;
1951		}
1952
1953		/*
1954		 * set reference count to 1, so btrfs_recover_relocation
1955		 * knows it should resumes merging
1956		 */
1957		if (!err)
1958			btrfs_set_root_refs(&reloc_root->root_item, 1);
1959		ret = btrfs_update_reloc_root(trans, root);
1960
1961		/*
1962		 * Even if we have an error we need this reloc root back on our
1963		 * list so we can clean up properly.
1964		 */
1965		list_add(&reloc_root->root_list, &reloc_roots);
1966		btrfs_put_root(root);
1967
1968		if (ret) {
1969			btrfs_abort_transaction(trans, ret);
1970			if (!err)
1971				err = ret;
1972			break;
1973		}
1974	}
1975
1976	list_splice(&reloc_roots, &rc->reloc_roots);
1977
1978	if (!err)
1979		err = btrfs_commit_transaction(trans);
1980	else
1981		btrfs_end_transaction(trans);
1982	return err;
1983}
1984
1985static noinline_for_stack
1986void free_reloc_roots(struct list_head *list)
1987{
1988	struct btrfs_root *reloc_root, *tmp;
1989
1990	list_for_each_entry_safe(reloc_root, tmp, list, root_list)
 
 
1991		__del_reloc_root(reloc_root);
 
 
 
 
 
1992}
1993
1994static noinline_for_stack
1995void merge_reloc_roots(struct reloc_control *rc)
1996{
1997	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1998	struct btrfs_root *root;
1999	struct btrfs_root *reloc_root;
2000	LIST_HEAD(reloc_roots);
2001	int found = 0;
2002	int ret = 0;
2003again:
2004	root = rc->extent_root;
2005
2006	/*
2007	 * this serializes us with btrfs_record_root_in_transaction,
2008	 * we have to make sure nobody is in the middle of
2009	 * adding their roots to the list while we are
2010	 * doing this splice
2011	 */
2012	mutex_lock(&fs_info->reloc_mutex);
2013	list_splice_init(&rc->reloc_roots, &reloc_roots);
2014	mutex_unlock(&fs_info->reloc_mutex);
2015
2016	while (!list_empty(&reloc_roots)) {
2017		found = 1;
2018		reloc_root = list_entry(reloc_roots.next,
2019					struct btrfs_root, root_list);
2020
2021		root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
2022					 false);
2023		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2024			if (WARN_ON(IS_ERR(root))) {
2025				/*
2026				 * For recovery we read the fs roots on mount,
2027				 * and if we didn't find the root then we marked
2028				 * the reloc root as a garbage root.  For normal
2029				 * relocation obviously the root should exist in
2030				 * memory.  However there's no reason we can't
2031				 * handle the error properly here just in case.
2032				 */
2033				ret = PTR_ERR(root);
2034				goto out;
2035			}
2036			if (WARN_ON(root->reloc_root != reloc_root)) {
2037				/*
2038				 * This can happen if on-disk metadata has some
2039				 * corruption, e.g. bad reloc tree key offset.
2040				 */
2041				ret = -EINVAL;
2042				goto out;
2043			}
2044			ret = merge_reloc_root(rc, root);
2045			btrfs_put_root(root);
2046			if (ret) {
2047				if (list_empty(&reloc_root->root_list))
2048					list_add_tail(&reloc_root->root_list,
2049						      &reloc_roots);
2050				goto out;
2051			}
2052		} else {
2053			if (!IS_ERR(root)) {
2054				if (root->reloc_root == reloc_root) {
2055					root->reloc_root = NULL;
2056					btrfs_put_root(reloc_root);
2057				}
2058				clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
2059					  &root->state);
2060				btrfs_put_root(root);
2061			}
2062
2063			list_del_init(&reloc_root->root_list);
2064			/* Don't forget to queue this reloc root for cleanup */
2065			list_add_tail(&reloc_root->reloc_dirty_list,
2066				      &rc->dirty_subvol_roots);
2067		}
2068	}
2069
2070	if (found) {
2071		found = 0;
2072		goto again;
2073	}
2074out:
2075	if (ret) {
2076		btrfs_handle_fs_error(fs_info, ret, NULL);
2077		free_reloc_roots(&reloc_roots);
 
2078
2079		/* new reloc root may be added */
2080		mutex_lock(&fs_info->reloc_mutex);
2081		list_splice_init(&rc->reloc_roots, &reloc_roots);
2082		mutex_unlock(&fs_info->reloc_mutex);
2083		free_reloc_roots(&reloc_roots);
 
2084	}
2085
2086	/*
2087	 * We used to have
2088	 *
2089	 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2090	 *
2091	 * here, but it's wrong.  If we fail to start the transaction in
2092	 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
2093	 * have actually been removed from the reloc_root_tree rb tree.  This is
2094	 * fine because we're bailing here, and we hold a reference on the root
2095	 * for the list that holds it, so these roots will be cleaned up when we
2096	 * do the reloc_dirty_list afterwards.  Meanwhile the root->reloc_root
2097	 * will be cleaned up on unmount.
2098	 *
2099	 * The remaining nodes will be cleaned up by free_reloc_control.
2100	 */
2101}
2102
2103static void free_block_list(struct rb_root *blocks)
2104{
2105	struct tree_block *block;
2106	struct rb_node *rb_node;
2107	while ((rb_node = rb_first(blocks))) {
2108		block = rb_entry(rb_node, struct tree_block, rb_node);
2109		rb_erase(rb_node, blocks);
2110		kfree(block);
2111	}
2112}
2113
2114static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2115				      struct btrfs_root *reloc_root)
2116{
2117	struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2118	struct btrfs_root *root;
2119	int ret;
2120
2121	if (reloc_root->last_trans == trans->transid)
2122		return 0;
2123
2124	root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
2125
2126	/*
2127	 * This should succeed, since we can't have a reloc root without having
2128	 * already looked up the actual root and created the reloc root for this
2129	 * root.
2130	 *
2131	 * However if there's some sort of corruption where we have a ref to a
2132	 * reloc root without a corresponding root this could return ENOENT.
2133	 */
2134	if (IS_ERR(root)) {
2135		ASSERT(0);
2136		return PTR_ERR(root);
2137	}
2138	if (root->reloc_root != reloc_root) {
2139		ASSERT(0);
2140		btrfs_err(fs_info,
2141			  "root %llu has two reloc roots associated with it",
2142			  reloc_root->root_key.offset);
2143		btrfs_put_root(root);
2144		return -EUCLEAN;
2145	}
2146	ret = btrfs_record_root_in_trans(trans, root);
2147	btrfs_put_root(root);
2148
2149	return ret;
2150}
2151
2152static noinline_for_stack
2153struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2154				     struct reloc_control *rc,
2155				     struct btrfs_backref_node *node,
2156				     struct btrfs_backref_edge *edges[])
2157{
2158	struct btrfs_backref_node *next;
2159	struct btrfs_root *root;
2160	int index = 0;
2161	int ret;
2162
2163	next = node;
2164	while (1) {
2165		cond_resched();
2166		next = walk_up_backref(next, edges, &index);
2167		root = next->root;
2168
2169		/*
2170		 * If there is no root, then our references for this block are
2171		 * incomplete, as we should be able to walk all the way up to a
2172		 * block that is owned by a root.
2173		 *
2174		 * This path is only for SHAREABLE roots, so if we come upon a
2175		 * non-SHAREABLE root then we have backrefs that resolve
2176		 * improperly.
2177		 *
2178		 * Both of these cases indicate file system corruption, or a bug
2179		 * in the backref walking code.
2180		 */
2181		if (!root) {
2182			ASSERT(0);
2183			btrfs_err(trans->fs_info,
2184		"bytenr %llu doesn't have a backref path ending in a root",
2185				  node->bytenr);
2186			return ERR_PTR(-EUCLEAN);
2187		}
2188		if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2189			ASSERT(0);
2190			btrfs_err(trans->fs_info,
2191	"bytenr %llu has multiple refs with one ending in a non-shareable root",
2192				  node->bytenr);
2193			return ERR_PTR(-EUCLEAN);
2194		}
2195
2196		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2197			ret = record_reloc_root_in_trans(trans, root);
2198			if (ret)
2199				return ERR_PTR(ret);
2200			break;
2201		}
2202
2203		ret = btrfs_record_root_in_trans(trans, root);
2204		if (ret)
2205			return ERR_PTR(ret);
2206		root = root->reloc_root;
2207
2208		/*
2209		 * We could have raced with another thread which failed, so
2210		 * root->reloc_root may not be set, return ENOENT in this case.
2211		 */
2212		if (!root)
2213			return ERR_PTR(-ENOENT);
2214
2215		if (next->new_bytenr != root->node->start) {
2216			/*
2217			 * We just created the reloc root, so we shouldn't have
2218			 * ->new_bytenr set and this shouldn't be in the changed
2219			 *  list.  If it is then we have multiple roots pointing
2220			 *  at the same bytenr which indicates corruption, or
2221			 *  we've made a mistake in the backref walking code.
2222			 */
2223			ASSERT(next->new_bytenr == 0);
2224			ASSERT(list_empty(&next->list));
2225			if (next->new_bytenr || !list_empty(&next->list)) {
2226				btrfs_err(trans->fs_info,
2227	"bytenr %llu possibly has multiple roots pointing at the same bytenr %llu",
2228					  node->bytenr, next->bytenr);
2229				return ERR_PTR(-EUCLEAN);
2230			}
2231
2232			next->new_bytenr = root->node->start;
2233			btrfs_put_root(next->root);
2234			next->root = btrfs_grab_root(root);
2235			ASSERT(next->root);
2236			list_add_tail(&next->list,
2237				      &rc->backref_cache.changed);
2238			mark_block_processed(rc, next);
2239			break;
2240		}
2241
2242		WARN_ON(1);
2243		root = NULL;
2244		next = walk_down_backref(edges, &index);
2245		if (!next || next->level <= node->level)
2246			break;
2247	}
2248	if (!root) {
2249		/*
2250		 * This can happen if there's fs corruption or if there's a bug
2251		 * in the backref lookup code.
2252		 */
2253		ASSERT(0);
2254		return ERR_PTR(-ENOENT);
2255	}
2256
2257	next = node;
2258	/* setup backref node path for btrfs_reloc_cow_block */
2259	while (1) {
2260		rc->backref_cache.path[next->level] = next;
2261		if (--index < 0)
2262			break;
2263		next = edges[index]->node[UPPER];
2264	}
2265	return root;
2266}
2267
2268/*
2269 * Select a tree root for relocation.
2270 *
2271 * Return NULL if the block is not shareable. We should use do_relocation() in
2272 * this case.
2273 *
2274 * Return a tree root pointer if the block is shareable.
2275 * Return -ENOENT if the block is root of reloc tree.
2276 */
2277static noinline_for_stack
2278struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2279{
2280	struct btrfs_backref_node *next;
2281	struct btrfs_root *root;
2282	struct btrfs_root *fs_root = NULL;
2283	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2284	int index = 0;
2285
2286	next = node;
2287	while (1) {
2288		cond_resched();
2289		next = walk_up_backref(next, edges, &index);
2290		root = next->root;
 
2291
2292		/*
2293		 * This can occur if we have incomplete extent refs leading all
2294		 * the way up a particular path, in this case return -EUCLEAN.
2295		 */
2296		if (!root)
2297			return ERR_PTR(-EUCLEAN);
2298
2299		/* No other choice for non-shareable tree */
2300		if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2301			return root;
2302
2303		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2304			fs_root = root;
2305
2306		if (next != node)
2307			return NULL;
2308
2309		next = walk_down_backref(edges, &index);
2310		if (!next || next->level <= node->level)
2311			break;
2312	}
2313
2314	if (!fs_root)
2315		return ERR_PTR(-ENOENT);
2316	return fs_root;
2317}
2318
2319static noinline_for_stack
2320u64 calcu_metadata_size(struct reloc_control *rc,
2321			struct btrfs_backref_node *node, int reserve)
2322{
2323	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2324	struct btrfs_backref_node *next = node;
2325	struct btrfs_backref_edge *edge;
2326	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2327	u64 num_bytes = 0;
2328	int index = 0;
2329
2330	BUG_ON(reserve && node->processed);
2331
2332	while (next) {
2333		cond_resched();
2334		while (1) {
2335			if (next->processed && (reserve || next != node))
2336				break;
2337
2338			num_bytes += fs_info->nodesize;
2339
2340			if (list_empty(&next->upper))
2341				break;
2342
2343			edge = list_entry(next->upper.next,
2344					struct btrfs_backref_edge, list[LOWER]);
2345			edges[index++] = edge;
2346			next = edge->node[UPPER];
2347		}
2348		next = walk_down_backref(edges, &index);
2349	}
2350	return num_bytes;
2351}
2352
2353static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2354				  struct reloc_control *rc,
2355				  struct btrfs_backref_node *node)
2356{
2357	struct btrfs_root *root = rc->extent_root;
2358	struct btrfs_fs_info *fs_info = root->fs_info;
2359	u64 num_bytes;
2360	int ret;
2361	u64 tmp;
2362
2363	num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2364
2365	trans->block_rsv = rc->block_rsv;
2366	rc->reserved_bytes += num_bytes;
2367
2368	/*
2369	 * We are under a transaction here so we can only do limited flushing.
2370	 * If we get an enospc just kick back -EAGAIN so we know to drop the
2371	 * transaction and try to refill when we can flush all the things.
2372	 */
2373	ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes,
2374				     BTRFS_RESERVE_FLUSH_LIMIT);
2375	if (ret) {
2376		tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2377		while (tmp <= rc->reserved_bytes)
2378			tmp <<= 1;
2379		/*
2380		 * only one thread can access block_rsv at this point,
2381		 * so we don't need hold lock to protect block_rsv.
2382		 * we expand more reservation size here to allow enough
2383		 * space for relocation and we will return earlier in
2384		 * enospc case.
2385		 */
2386		rc->block_rsv->size = tmp + fs_info->nodesize *
2387				      RELOCATION_RESERVED_NODES;
2388		return -EAGAIN;
2389	}
2390
2391	return 0;
2392}
2393
2394/*
2395 * relocate a block tree, and then update pointers in upper level
2396 * blocks that reference the block to point to the new location.
2397 *
2398 * if called by link_to_upper, the block has already been relocated.
2399 * in that case this function just updates pointers.
2400 */
2401static int do_relocation(struct btrfs_trans_handle *trans,
2402			 struct reloc_control *rc,
2403			 struct btrfs_backref_node *node,
2404			 struct btrfs_key *key,
2405			 struct btrfs_path *path, int lowest)
2406{
2407	struct btrfs_backref_node *upper;
2408	struct btrfs_backref_edge *edge;
2409	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
 
2410	struct btrfs_root *root;
2411	struct extent_buffer *eb;
2412	u32 blocksize;
2413	u64 bytenr;
 
2414	int slot;
2415	int ret = 0;
 
2416
2417	/*
2418	 * If we are lowest then this is the first time we're processing this
2419	 * block, and thus shouldn't have an eb associated with it yet.
2420	 */
2421	ASSERT(!lowest || !node->eb);
2422
2423	path->lowest_level = node->level + 1;
2424	rc->backref_cache.path[node->level] = node;
2425	list_for_each_entry(edge, &node->upper, list[LOWER]) {
 
2426		struct btrfs_ref ref = { 0 };
2427
2428		cond_resched();
2429
2430		upper = edge->node[UPPER];
2431		root = select_reloc_root(trans, rc, upper, edges);
2432		if (IS_ERR(root)) {
2433			ret = PTR_ERR(root);
2434			goto next;
2435		}
2436
2437		if (upper->eb && !upper->locked) {
2438			if (!lowest) {
2439				ret = btrfs_bin_search(upper->eb, 0, key, &slot);
2440				if (ret < 0)
 
 
2441					goto next;
 
2442				BUG_ON(ret);
2443				bytenr = btrfs_node_blockptr(upper->eb, slot);
2444				if (node->eb->start == bytenr)
2445					goto next;
2446			}
2447			btrfs_backref_drop_node_buffer(upper);
2448		}
2449
2450		if (!upper->eb) {
2451			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2452			if (ret) {
2453				if (ret > 0)
2454					ret = -ENOENT;
 
 
2455
2456				btrfs_release_path(path);
2457				break;
2458			}
2459
2460			if (!upper->eb) {
2461				upper->eb = path->nodes[upper->level];
2462				path->nodes[upper->level] = NULL;
2463			} else {
2464				BUG_ON(upper->eb != path->nodes[upper->level]);
2465			}
2466
2467			upper->locked = 1;
2468			path->locks[upper->level] = 0;
2469
2470			slot = path->slots[upper->level];
2471			btrfs_release_path(path);
2472		} else {
2473			ret = btrfs_bin_search(upper->eb, 0, key, &slot);
2474			if (ret < 0)
 
 
2475				goto next;
 
2476			BUG_ON(ret);
2477		}
2478
2479		bytenr = btrfs_node_blockptr(upper->eb, slot);
2480		if (lowest) {
2481			if (bytenr != node->bytenr) {
2482				btrfs_err(root->fs_info,
2483		"lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2484					  bytenr, node->bytenr, slot,
2485					  upper->eb->start);
2486				ret = -EIO;
2487				goto next;
2488			}
2489		} else {
2490			if (node->eb->start == bytenr)
2491				goto next;
2492		}
2493
2494		blocksize = root->fs_info->nodesize;
2495		eb = btrfs_read_node_slot(upper->eb, slot);
 
 
 
2496		if (IS_ERR(eb)) {
2497			ret = PTR_ERR(eb);
 
 
 
 
2498			goto next;
2499		}
2500		btrfs_tree_lock(eb);
 
2501
2502		if (!node->eb) {
2503			ret = btrfs_cow_block(trans, root, eb, upper->eb,
2504					      slot, &eb, BTRFS_NESTING_COW);
2505			btrfs_tree_unlock(eb);
2506			free_extent_buffer(eb);
2507			if (ret < 0)
 
2508				goto next;
2509			/*
2510			 * We've just COWed this block, it should have updated
2511			 * the correct backref node entry.
2512			 */
2513			ASSERT(node->eb == eb);
2514		} else {
2515			btrfs_set_node_blockptr(upper->eb, slot,
2516						node->eb->start);
2517			btrfs_set_node_ptr_generation(upper->eb, slot,
2518						      trans->transid);
2519			btrfs_mark_buffer_dirty(trans, upper->eb);
2520
2521			btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2522					       node->eb->start, blocksize,
2523					       upper->eb->start,
2524					       btrfs_header_owner(upper->eb));
2525			btrfs_init_tree_ref(&ref, node->level,
2526					    btrfs_header_owner(upper->eb),
2527					    root->root_key.objectid, false);
2528			ret = btrfs_inc_extent_ref(trans, &ref);
2529			if (!ret)
2530				ret = btrfs_drop_subtree(trans, root, eb,
2531							 upper->eb);
2532			if (ret)
2533				btrfs_abort_transaction(trans, ret);
2534		}
2535next:
2536		if (!upper->pending)
2537			btrfs_backref_drop_node_buffer(upper);
2538		else
2539			btrfs_backref_unlock_node_buffer(upper);
2540		if (ret)
2541			break;
2542	}
2543
2544	if (!ret && node->pending) {
2545		btrfs_backref_drop_node_buffer(node);
2546		list_move_tail(&node->list, &rc->backref_cache.changed);
2547		node->pending = 0;
2548	}
2549
2550	path->lowest_level = 0;
2551
2552	/*
2553	 * We should have allocated all of our space in the block rsv and thus
2554	 * shouldn't ENOSPC.
2555	 */
2556	ASSERT(ret != -ENOSPC);
2557	return ret;
2558}
2559
2560static int link_to_upper(struct btrfs_trans_handle *trans,
2561			 struct reloc_control *rc,
2562			 struct btrfs_backref_node *node,
2563			 struct btrfs_path *path)
2564{
2565	struct btrfs_key key;
2566
2567	btrfs_node_key_to_cpu(node->eb, &key, 0);
2568	return do_relocation(trans, rc, node, &key, path, 0);
2569}
2570
2571static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2572				struct reloc_control *rc,
2573				struct btrfs_path *path, int err)
2574{
2575	LIST_HEAD(list);
2576	struct btrfs_backref_cache *cache = &rc->backref_cache;
2577	struct btrfs_backref_node *node;
2578	int level;
2579	int ret;
2580
2581	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2582		while (!list_empty(&cache->pending[level])) {
2583			node = list_entry(cache->pending[level].next,
2584					  struct btrfs_backref_node, list);
2585			list_move_tail(&node->list, &list);
2586			BUG_ON(!node->pending);
2587
2588			if (!err) {
2589				ret = link_to_upper(trans, rc, node, path);
2590				if (ret < 0)
2591					err = ret;
2592			}
2593		}
2594		list_splice_init(&list, &cache->pending[level]);
2595	}
2596	return err;
2597}
2598
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2599/*
2600 * mark a block and all blocks directly/indirectly reference the block
2601 * as processed.
2602 */
2603static void update_processed_blocks(struct reloc_control *rc,
2604				    struct btrfs_backref_node *node)
2605{
2606	struct btrfs_backref_node *next = node;
2607	struct btrfs_backref_edge *edge;
2608	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2609	int index = 0;
2610
2611	while (next) {
2612		cond_resched();
2613		while (1) {
2614			if (next->processed)
2615				break;
2616
2617			mark_block_processed(rc, next);
2618
2619			if (list_empty(&next->upper))
2620				break;
2621
2622			edge = list_entry(next->upper.next,
2623					struct btrfs_backref_edge, list[LOWER]);
2624			edges[index++] = edge;
2625			next = edge->node[UPPER];
2626		}
2627		next = walk_down_backref(edges, &index);
2628	}
2629}
2630
2631static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2632{
2633	u32 blocksize = rc->extent_root->fs_info->nodesize;
2634
2635	if (test_range_bit(&rc->processed_blocks, bytenr,
2636			   bytenr + blocksize - 1, EXTENT_DIRTY, NULL))
2637		return 1;
2638	return 0;
2639}
2640
2641static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2642			      struct tree_block *block)
2643{
2644	struct btrfs_tree_parent_check check = {
2645		.level = block->level,
2646		.owner_root = block->owner,
2647		.transid = block->key.offset
2648	};
2649	struct extent_buffer *eb;
2650
2651	eb = read_tree_block(fs_info, block->bytenr, &check);
2652	if (IS_ERR(eb))
 
 
2653		return PTR_ERR(eb);
2654	if (!extent_buffer_uptodate(eb)) {
2655		free_extent_buffer(eb);
2656		return -EIO;
2657	}
2658	if (block->level == 0)
2659		btrfs_item_key_to_cpu(eb, &block->key, 0);
2660	else
2661		btrfs_node_key_to_cpu(eb, &block->key, 0);
2662	free_extent_buffer(eb);
2663	block->key_ready = true;
2664	return 0;
2665}
2666
2667/*
2668 * helper function to relocate a tree block
2669 */
2670static int relocate_tree_block(struct btrfs_trans_handle *trans,
2671				struct reloc_control *rc,
2672				struct btrfs_backref_node *node,
2673				struct btrfs_key *key,
2674				struct btrfs_path *path)
2675{
2676	struct btrfs_root *root;
2677	int ret = 0;
2678
2679	if (!node)
2680		return 0;
2681
2682	/*
2683	 * If we fail here we want to drop our backref_node because we are going
2684	 * to start over and regenerate the tree for it.
2685	 */
2686	ret = reserve_metadata_space(trans, rc, node);
2687	if (ret)
2688		goto out;
2689
2690	BUG_ON(node->processed);
2691	root = select_one_root(node);
2692	if (IS_ERR(root)) {
2693		ret = PTR_ERR(root);
 
 
2694
2695		/* See explanation in select_one_root for the -EUCLEAN case. */
2696		ASSERT(ret == -ENOENT);
2697		if (ret == -ENOENT) {
2698			ret = 0;
2699			update_processed_blocks(rc, node);
2700		}
2701		goto out;
2702	}
2703
2704	if (root) {
2705		if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2706			/*
2707			 * This block was the root block of a root, and this is
2708			 * the first time we're processing the block and thus it
2709			 * should not have had the ->new_bytenr modified and
2710			 * should have not been included on the changed list.
2711			 *
2712			 * However in the case of corruption we could have
2713			 * multiple refs pointing to the same block improperly,
2714			 * and thus we would trip over these checks.  ASSERT()
2715			 * for the developer case, because it could indicate a
2716			 * bug in the backref code, however error out for a
2717			 * normal user in the case of corruption.
2718			 */
2719			ASSERT(node->new_bytenr == 0);
2720			ASSERT(list_empty(&node->list));
2721			if (node->new_bytenr || !list_empty(&node->list)) {
2722				btrfs_err(root->fs_info,
2723				  "bytenr %llu has improper references to it",
2724					  node->bytenr);
2725				ret = -EUCLEAN;
2726				goto out;
2727			}
2728			ret = btrfs_record_root_in_trans(trans, root);
2729			if (ret)
2730				goto out;
2731			/*
2732			 * Another thread could have failed, need to check if we
2733			 * have reloc_root actually set.
2734			 */
2735			if (!root->reloc_root) {
2736				ret = -ENOENT;
2737				goto out;
2738			}
2739			root = root->reloc_root;
2740			node->new_bytenr = root->node->start;
2741			btrfs_put_root(node->root);
2742			node->root = btrfs_grab_root(root);
2743			ASSERT(node->root);
2744			list_add_tail(&node->list, &rc->backref_cache.changed);
2745		} else {
2746			path->lowest_level = node->level;
2747			if (root == root->fs_info->chunk_root)
2748				btrfs_reserve_chunk_metadata(trans, false);
2749			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2750			btrfs_release_path(path);
2751			if (root == root->fs_info->chunk_root)
2752				btrfs_trans_release_chunk_metadata(trans);
2753			if (ret > 0)
2754				ret = 0;
2755		}
2756		if (!ret)
2757			update_processed_blocks(rc, node);
2758	} else {
2759		ret = do_relocation(trans, rc, node, key, path, 1);
2760	}
2761out:
2762	if (ret || node->level == 0 || node->cowonly)
2763		btrfs_backref_cleanup_node(&rc->backref_cache, node);
2764	return ret;
2765}
2766
2767/*
2768 * relocate a list of blocks
2769 */
2770static noinline_for_stack
2771int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2772			 struct reloc_control *rc, struct rb_root *blocks)
2773{
2774	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2775	struct btrfs_backref_node *node;
2776	struct btrfs_path *path;
2777	struct tree_block *block;
2778	struct tree_block *next;
2779	int ret;
2780	int err = 0;
2781
2782	path = btrfs_alloc_path();
2783	if (!path) {
2784		err = -ENOMEM;
2785		goto out_free_blocks;
2786	}
2787
2788	/* Kick in readahead for tree blocks with missing keys */
2789	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2790		if (!block->key_ready)
2791			btrfs_readahead_tree_block(fs_info, block->bytenr,
2792						   block->owner, 0,
2793						   block->level);
2794	}
2795
2796	/* Get first keys */
2797	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2798		if (!block->key_ready) {
2799			err = get_tree_block_key(fs_info, block);
2800			if (err)
2801				goto out_free_path;
2802		}
2803	}
2804
2805	/* Do tree relocation */
2806	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2807		node = build_backref_tree(trans, rc, &block->key,
2808					  block->level, block->bytenr);
2809		if (IS_ERR(node)) {
2810			err = PTR_ERR(node);
2811			goto out;
2812		}
2813
2814		ret = relocate_tree_block(trans, rc, node, &block->key,
2815					  path);
2816		if (ret < 0) {
2817			err = ret;
2818			break;
 
2819		}
2820	}
2821out:
2822	err = finish_pending_nodes(trans, rc, path, err);
2823
2824out_free_path:
2825	btrfs_free_path(path);
2826out_free_blocks:
2827	free_block_list(blocks);
2828	return err;
2829}
2830
2831static noinline_for_stack int prealloc_file_extent_cluster(
2832				struct btrfs_inode *inode,
2833				const struct file_extent_cluster *cluster)
2834{
2835	u64 alloc_hint = 0;
2836	u64 start;
2837	u64 end;
2838	u64 offset = inode->index_cnt;
2839	u64 num_bytes;
2840	int nr;
2841	int ret = 0;
2842	u64 i_size = i_size_read(&inode->vfs_inode);
2843	u64 prealloc_start = cluster->start - offset;
2844	u64 prealloc_end = cluster->end - offset;
2845	u64 cur_offset = prealloc_start;
 
2846
2847	/*
2848	 * For subpage case, previous i_size may not be aligned to PAGE_SIZE.
2849	 * This means the range [i_size, PAGE_END + 1) is filled with zeros by
2850	 * btrfs_do_readpage() call of previously relocated file cluster.
2851	 *
2852	 * If the current cluster starts in the above range, btrfs_do_readpage()
2853	 * will skip the read, and relocate_one_page() will later writeback
2854	 * the padding zeros as new data, causing data corruption.
2855	 *
2856	 * Here we have to manually invalidate the range (i_size, PAGE_END + 1).
2857	 */
2858	if (!PAGE_ALIGNED(i_size)) {
2859		struct address_space *mapping = inode->vfs_inode.i_mapping;
2860		struct btrfs_fs_info *fs_info = inode->root->fs_info;
2861		const u32 sectorsize = fs_info->sectorsize;
2862		struct page *page;
2863
2864		ASSERT(sectorsize < PAGE_SIZE);
2865		ASSERT(IS_ALIGNED(i_size, sectorsize));
2866
2867		/*
2868		 * Subpage can't handle page with DIRTY but without UPTODATE
2869		 * bit as it can lead to the following deadlock:
2870		 *
2871		 * btrfs_read_folio()
2872		 * | Page already *locked*
2873		 * |- btrfs_lock_and_flush_ordered_range()
2874		 *    |- btrfs_start_ordered_extent()
2875		 *       |- extent_write_cache_pages()
2876		 *          |- lock_page()
2877		 *             We try to lock the page we already hold.
2878		 *
2879		 * Here we just writeback the whole data reloc inode, so that
2880		 * we will be ensured to have no dirty range in the page, and
2881		 * are safe to clear the uptodate bits.
2882		 *
2883		 * This shouldn't cause too much overhead, as we need to write
2884		 * the data back anyway.
2885		 */
2886		ret = filemap_write_and_wait(mapping);
2887		if (ret < 0)
2888			return ret;
2889
2890		clear_extent_bits(&inode->io_tree, i_size,
2891				  round_up(i_size, PAGE_SIZE) - 1,
2892				  EXTENT_UPTODATE);
2893		page = find_lock_page(mapping, i_size >> PAGE_SHIFT);
2894		/*
2895		 * If page is freed we don't need to do anything then, as we
2896		 * will re-read the whole page anyway.
2897		 */
2898		if (page) {
2899			btrfs_subpage_clear_uptodate(fs_info, page_folio(page), i_size,
2900					round_up(i_size, PAGE_SIZE) - i_size);
2901			unlock_page(page);
2902			put_page(page);
2903		}
2904	}
2905
2906	BUG_ON(cluster->start != cluster->boundary[0]);
2907	ret = btrfs_alloc_data_chunk_ondemand(inode,
2908					      prealloc_end + 1 - prealloc_start);
2909	if (ret)
2910		return ret;
2911
2912	btrfs_inode_lock(inode, 0);
2913	for (nr = 0; nr < cluster->nr; nr++) {
2914		struct extent_state *cached_state = NULL;
2915
 
 
2916		start = cluster->boundary[nr] - offset;
2917		if (nr + 1 < cluster->nr)
2918			end = cluster->boundary[nr + 1] - 1 - offset;
2919		else
2920			end = cluster->end - offset;
2921
2922		lock_extent(&inode->io_tree, start, end, &cached_state);
2923		num_bytes = end + 1 - start;
2924		ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
 
 
 
2925						num_bytes, num_bytes,
2926						end + 1, &alloc_hint);
2927		cur_offset = end + 1;
2928		unlock_extent(&inode->io_tree, start, end, &cached_state);
2929		if (ret)
2930			break;
 
2931	}
2932	btrfs_inode_unlock(inode, 0);
2933
2934	if (cur_offset < prealloc_end)
2935		btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2936					       prealloc_end + 1 - cur_offset);
 
 
 
2937	return ret;
2938}
2939
2940static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode,
2941				u64 start, u64 end, u64 block_start)
 
2942{
 
 
2943	struct extent_map *em;
2944	struct extent_state *cached_state = NULL;
2945	int ret = 0;
2946
2947	em = alloc_extent_map();
2948	if (!em)
2949		return -ENOMEM;
2950
2951	em->start = start;
2952	em->len = end + 1 - start;
2953	em->block_len = em->len;
2954	em->block_start = block_start;
2955	em->flags |= EXTENT_FLAG_PINNED;
 
2956
2957	lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
2958	ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false);
2959	unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
2960	free_extent_map(em);
2961
2962	return ret;
2963}
2964
2965/*
2966 * Allow error injection to test balance/relocation cancellation
2967 */
2968noinline int btrfs_should_cancel_balance(const struct btrfs_fs_info *fs_info)
2969{
2970	return atomic_read(&fs_info->balance_cancel_req) ||
2971		atomic_read(&fs_info->reloc_cancel_req) ||
2972		fatal_signal_pending(current);
2973}
2974ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2975
2976static u64 get_cluster_boundary_end(const struct file_extent_cluster *cluster,
2977				    int cluster_nr)
2978{
2979	/* Last extent, use cluster end directly */
2980	if (cluster_nr >= cluster->nr - 1)
2981		return cluster->end;
2982
2983	/* Use next boundary start*/
2984	return cluster->boundary[cluster_nr + 1] - 1;
2985}
2986
2987static int relocate_one_page(struct inode *inode, struct file_ra_state *ra,
2988			     const struct file_extent_cluster *cluster,
2989			     int *cluster_nr, unsigned long page_index)
2990{
2991	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2992	u64 offset = BTRFS_I(inode)->index_cnt;
2993	const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT;
2994	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2995	struct page *page;
2996	u64 page_start;
2997	u64 page_end;
2998	u64 cur;
2999	int ret;
3000
3001	ASSERT(page_index <= last_index);
3002	page = find_lock_page(inode->i_mapping, page_index);
3003	if (!page) {
3004		page_cache_sync_readahead(inode->i_mapping, ra, NULL,
3005				page_index, last_index + 1 - page_index);
3006		page = find_or_create_page(inode->i_mapping, page_index, mask);
3007		if (!page)
3008			return -ENOMEM;
3009	}
3010
3011	if (PageReadahead(page))
3012		page_cache_async_readahead(inode->i_mapping, ra, NULL,
3013				page_folio(page), page_index,
3014				last_index + 1 - page_index);
3015
3016	if (!PageUptodate(page)) {
3017		btrfs_read_folio(NULL, page_folio(page));
3018		lock_page(page);
3019		if (!PageUptodate(page)) {
3020			ret = -EIO;
3021			goto release_page;
3022		}
3023	}
3024
3025	/*
3026	 * We could have lost page private when we dropped the lock to read the
3027	 * page above, make sure we set_page_extent_mapped here so we have any
3028	 * of the subpage blocksize stuff we need in place.
3029	 */
3030	ret = set_page_extent_mapped(page);
3031	if (ret < 0)
3032		goto release_page;
3033
3034	page_start = page_offset(page);
3035	page_end = page_start + PAGE_SIZE - 1;
3036
3037	/*
3038	 * Start from the cluster, as for subpage case, the cluster can start
3039	 * inside the page.
3040	 */
3041	cur = max(page_start, cluster->boundary[*cluster_nr] - offset);
3042	while (cur <= page_end) {
3043		struct extent_state *cached_state = NULL;
3044		u64 extent_start = cluster->boundary[*cluster_nr] - offset;
3045		u64 extent_end = get_cluster_boundary_end(cluster,
3046						*cluster_nr) - offset;
3047		u64 clamped_start = max(page_start, extent_start);
3048		u64 clamped_end = min(page_end, extent_end);
3049		u32 clamped_len = clamped_end + 1 - clamped_start;
3050
3051		/* Reserve metadata for this range */
3052		ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3053						      clamped_len, clamped_len,
3054						      false);
3055		if (ret)
3056			goto release_page;
3057
3058		/* Mark the range delalloc and dirty for later writeback */
3059		lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
3060			    &cached_state);
3061		ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start,
3062						clamped_end, 0, &cached_state);
3063		if (ret) {
3064			clear_extent_bit(&BTRFS_I(inode)->io_tree,
3065					 clamped_start, clamped_end,
3066					 EXTENT_LOCKED | EXTENT_BOUNDARY,
3067					 &cached_state);
3068			btrfs_delalloc_release_metadata(BTRFS_I(inode),
3069							clamped_len, true);
3070			btrfs_delalloc_release_extents(BTRFS_I(inode),
3071						       clamped_len);
3072			goto release_page;
3073		}
3074		btrfs_folio_set_dirty(fs_info, page_folio(page),
3075				      clamped_start, clamped_len);
3076
3077		/*
3078		 * Set the boundary if it's inside the page.
3079		 * Data relocation requires the destination extents to have the
3080		 * same size as the source.
3081		 * EXTENT_BOUNDARY bit prevents current extent from being merged
3082		 * with previous extent.
3083		 */
3084		if (in_range(cluster->boundary[*cluster_nr] - offset,
3085			     page_start, PAGE_SIZE)) {
3086			u64 boundary_start = cluster->boundary[*cluster_nr] -
3087						offset;
3088			u64 boundary_end = boundary_start +
3089					   fs_info->sectorsize - 1;
3090
3091			set_extent_bit(&BTRFS_I(inode)->io_tree,
3092				       boundary_start, boundary_end,
3093				       EXTENT_BOUNDARY, NULL);
3094		}
3095		unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
3096			      &cached_state);
3097		btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len);
3098		cur += clamped_len;
3099
3100		/* Crossed extent end, go to next extent */
3101		if (cur >= extent_end) {
3102			(*cluster_nr)++;
3103			/* Just finished the last extent of the cluster, exit. */
3104			if (*cluster_nr >= cluster->nr)
3105				break;
3106		}
 
3107	}
3108	unlock_page(page);
3109	put_page(page);
3110
3111	balance_dirty_pages_ratelimited(inode->i_mapping);
3112	btrfs_throttle(fs_info);
3113	if (btrfs_should_cancel_balance(fs_info))
3114		ret = -ECANCELED;
3115	return ret;
3116
3117release_page:
3118	unlock_page(page);
3119	put_page(page);
3120	return ret;
3121}
3122
3123static int relocate_file_extent_cluster(struct inode *inode,
3124					const struct file_extent_cluster *cluster)
3125{
 
 
 
3126	u64 offset = BTRFS_I(inode)->index_cnt;
3127	unsigned long index;
3128	unsigned long last_index;
 
3129	struct file_ra_state *ra;
3130	int cluster_nr = 0;
 
3131	int ret = 0;
3132
3133	if (!cluster->nr)
3134		return 0;
3135
3136	ra = kzalloc(sizeof(*ra), GFP_NOFS);
3137	if (!ra)
3138		return -ENOMEM;
3139
3140	ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
3141	if (ret)
3142		goto out;
3143
3144	file_ra_state_init(ra, inode->i_mapping);
3145
3146	ret = setup_relocation_extent_mapping(inode, cluster->start - offset,
3147				   cluster->end - offset, cluster->start);
3148	if (ret)
3149		goto out;
3150
 
3151	last_index = (cluster->end - offset) >> PAGE_SHIFT;
3152	for (index = (cluster->start - offset) >> PAGE_SHIFT;
3153	     index <= last_index && !ret; index++)
3154		ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index);
3155	if (ret == 0)
3156		WARN_ON(cluster_nr != cluster->nr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3157out:
3158	kfree(ra);
3159	return ret;
3160}
3161
3162static noinline_for_stack int relocate_data_extent(struct inode *inode,
3163				const struct btrfs_key *extent_key,
3164				struct file_extent_cluster *cluster)
3165{
3166	int ret;
3167	struct btrfs_root *root = BTRFS_I(inode)->root;
3168
3169	if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3170		ret = relocate_file_extent_cluster(inode, cluster);
3171		if (ret)
3172			return ret;
3173		cluster->nr = 0;
3174	}
3175
3176	/*
3177	 * Under simple quotas, we set root->relocation_src_root when we find
3178	 * the extent. If adjacent extents have different owners, we can't merge
3179	 * them while relocating. Handle this by storing the owning root that
3180	 * started a cluster and if we see an extent from a different root break
3181	 * cluster formation (just like the above case of non-adjacent extents).
3182	 *
3183	 * Without simple quotas, relocation_src_root is always 0, so we should
3184	 * never see a mismatch, and it should have no effect on relocation
3185	 * clusters.
3186	 */
3187	if (cluster->nr > 0 && cluster->owning_root != root->relocation_src_root) {
3188		u64 tmp = root->relocation_src_root;
3189
3190		/*
3191		 * root->relocation_src_root is the state that actually affects
3192		 * the preallocation we do here, so set it to the root owning
3193		 * the cluster we need to relocate.
3194		 */
3195		root->relocation_src_root = cluster->owning_root;
3196		ret = relocate_file_extent_cluster(inode, cluster);
3197		if (ret)
3198			return ret;
3199		cluster->nr = 0;
3200		/* And reset it back for the current extent's owning root. */
3201		root->relocation_src_root = tmp;
3202	}
3203
3204	if (!cluster->nr) {
3205		cluster->start = extent_key->objectid;
3206		cluster->owning_root = root->relocation_src_root;
3207	}
3208	else
3209		BUG_ON(cluster->nr >= MAX_EXTENTS);
3210	cluster->end = extent_key->objectid + extent_key->offset - 1;
3211	cluster->boundary[cluster->nr] = extent_key->objectid;
3212	cluster->nr++;
3213
3214	if (cluster->nr >= MAX_EXTENTS) {
3215		ret = relocate_file_extent_cluster(inode, cluster);
3216		if (ret)
3217			return ret;
3218		cluster->nr = 0;
3219	}
3220	return 0;
3221}
3222
3223/*
3224 * helper to add a tree block to the list.
3225 * the major work is getting the generation and level of the block
3226 */
3227static int add_tree_block(struct reloc_control *rc,
3228			  const struct btrfs_key *extent_key,
3229			  struct btrfs_path *path,
3230			  struct rb_root *blocks)
3231{
3232	struct extent_buffer *eb;
3233	struct btrfs_extent_item *ei;
3234	struct btrfs_tree_block_info *bi;
3235	struct tree_block *block;
3236	struct rb_node *rb_node;
3237	u32 item_size;
3238	int level = -1;
3239	u64 generation;
3240	u64 owner = 0;
3241
3242	eb =  path->nodes[0];
3243	item_size = btrfs_item_size(eb, path->slots[0]);
3244
3245	if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3246	    item_size >= sizeof(*ei) + sizeof(*bi)) {
3247		unsigned long ptr = 0, end;
3248
3249		ei = btrfs_item_ptr(eb, path->slots[0],
3250				struct btrfs_extent_item);
3251		end = (unsigned long)ei + item_size;
3252		if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3253			bi = (struct btrfs_tree_block_info *)(ei + 1);
3254			level = btrfs_tree_block_level(eb, bi);
3255			ptr = (unsigned long)(bi + 1);
3256		} else {
3257			level = (int)extent_key->offset;
3258			ptr = (unsigned long)(ei + 1);
3259		}
3260		generation = btrfs_extent_generation(eb, ei);
3261
3262		/*
3263		 * We're reading random blocks without knowing their owner ahead
3264		 * of time.  This is ok most of the time, as all reloc roots and
3265		 * fs roots have the same lock type.  However normal trees do
3266		 * not, and the only way to know ahead of time is to read the
3267		 * inline ref offset.  We know it's an fs root if
3268		 *
3269		 * 1. There's more than one ref.
3270		 * 2. There's a SHARED_DATA_REF_KEY set.
3271		 * 3. FULL_BACKREF is set on the flags.
3272		 *
3273		 * Otherwise it's safe to assume that the ref offset == the
3274		 * owner of this block, so we can use that when calling
3275		 * read_tree_block.
3276		 */
3277		if (btrfs_extent_refs(eb, ei) == 1 &&
3278		    !(btrfs_extent_flags(eb, ei) &
3279		      BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
3280		    ptr < end) {
3281			struct btrfs_extent_inline_ref *iref;
3282			int type;
3283
3284			iref = (struct btrfs_extent_inline_ref *)ptr;
3285			type = btrfs_get_extent_inline_ref_type(eb, iref,
3286							BTRFS_REF_TYPE_BLOCK);
3287			if (type == BTRFS_REF_TYPE_INVALID)
3288				return -EINVAL;
3289			if (type == BTRFS_TREE_BLOCK_REF_KEY)
3290				owner = btrfs_extent_inline_ref_offset(eb, iref);
3291		}
3292	} else {
3293		btrfs_print_leaf(eb);
3294		btrfs_err(rc->block_group->fs_info,
3295			  "unrecognized tree backref at tree block %llu slot %u",
3296			  eb->start, path->slots[0]);
3297		btrfs_release_path(path);
3298		return -EUCLEAN;
3299	}
3300
3301	btrfs_release_path(path);
3302
3303	BUG_ON(level == -1);
3304
3305	block = kmalloc(sizeof(*block), GFP_NOFS);
3306	if (!block)
3307		return -ENOMEM;
3308
3309	block->bytenr = extent_key->objectid;
3310	block->key.objectid = rc->extent_root->fs_info->nodesize;
3311	block->key.offset = generation;
3312	block->level = level;
3313	block->key_ready = false;
3314	block->owner = owner;
3315
3316	rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
3317	if (rb_node)
3318		btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
3319				    -EEXIST);
3320
3321	return 0;
3322}
3323
3324/*
3325 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3326 */
3327static int __add_tree_block(struct reloc_control *rc,
3328			    u64 bytenr, u32 blocksize,
3329			    struct rb_root *blocks)
3330{
3331	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3332	struct btrfs_path *path;
3333	struct btrfs_key key;
3334	int ret;
3335	bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3336
3337	if (tree_block_processed(bytenr, rc))
3338		return 0;
3339
3340	if (rb_simple_search(blocks, bytenr))
3341		return 0;
3342
3343	path = btrfs_alloc_path();
3344	if (!path)
3345		return -ENOMEM;
3346again:
3347	key.objectid = bytenr;
3348	if (skinny) {
3349		key.type = BTRFS_METADATA_ITEM_KEY;
3350		key.offset = (u64)-1;
3351	} else {
3352		key.type = BTRFS_EXTENT_ITEM_KEY;
3353		key.offset = blocksize;
3354	}
3355
3356	path->search_commit_root = 1;
3357	path->skip_locking = 1;
3358	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3359	if (ret < 0)
3360		goto out;
3361
3362	if (ret > 0 && skinny) {
3363		if (path->slots[0]) {
3364			path->slots[0]--;
3365			btrfs_item_key_to_cpu(path->nodes[0], &key,
3366					      path->slots[0]);
3367			if (key.objectid == bytenr &&
3368			    (key.type == BTRFS_METADATA_ITEM_KEY ||
3369			     (key.type == BTRFS_EXTENT_ITEM_KEY &&
3370			      key.offset == blocksize)))
3371				ret = 0;
3372		}
3373
3374		if (ret) {
3375			skinny = false;
3376			btrfs_release_path(path);
3377			goto again;
3378		}
3379	}
3380	if (ret) {
3381		ASSERT(ret == 1);
3382		btrfs_print_leaf(path->nodes[0]);
3383		btrfs_err(fs_info,
3384	     "tree block extent item (%llu) is not found in extent tree",
3385		     bytenr);
3386		WARN_ON(1);
3387		ret = -EINVAL;
3388		goto out;
3389	}
3390
3391	ret = add_tree_block(rc, &key, path, blocks);
3392out:
3393	btrfs_free_path(path);
3394	return ret;
3395}
3396
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3397static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3398				    struct btrfs_block_group *block_group,
3399				    struct inode *inode,
3400				    u64 ino)
3401{
 
3402	struct btrfs_root *root = fs_info->tree_root;
3403	struct btrfs_trans_handle *trans;
3404	int ret = 0;
3405
3406	if (inode)
3407		goto truncate;
3408
3409	inode = btrfs_iget(fs_info->sb, ino, root);
 
 
 
 
3410	if (IS_ERR(inode))
3411		return -ENOENT;
3412
3413truncate:
3414	ret = btrfs_check_trunc_cache_free_space(fs_info,
3415						 &fs_info->global_block_rsv);
3416	if (ret)
3417		goto out;
3418
3419	trans = btrfs_join_transaction(root);
3420	if (IS_ERR(trans)) {
3421		ret = PTR_ERR(trans);
3422		goto out;
3423	}
3424
3425	ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3426
3427	btrfs_end_transaction(trans);
3428	btrfs_btree_balance_dirty(fs_info);
3429out:
3430	iput(inode);
3431	return ret;
3432}
3433
3434/*
3435 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3436 * cache inode, to avoid free space cache data extent blocking data relocation.
3437 */
3438static int delete_v1_space_cache(struct extent_buffer *leaf,
3439				 struct btrfs_block_group *block_group,
3440				 u64 data_bytenr)
 
 
3441{
3442	u64 space_cache_ino;
3443	struct btrfs_file_extent_item *ei;
 
 
 
 
3444	struct btrfs_key key;
3445	bool found = false;
3446	int i;
 
 
 
 
 
 
3447	int ret;
3448
3449	if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3450		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3451
3452	for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3453		u8 type;
 
3454
3455		btrfs_item_key_to_cpu(leaf, &key, i);
3456		if (key.type != BTRFS_EXTENT_DATA_KEY)
3457			continue;
3458		ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3459		type = btrfs_file_extent_type(leaf, ei);
 
 
 
 
 
 
 
3460
3461		if ((type == BTRFS_FILE_EXTENT_REG ||
3462		     type == BTRFS_FILE_EXTENT_PREALLOC) &&
3463		    btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3464			found = true;
3465			space_cache_ino = key.objectid;
3466			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3467		}
 
 
 
 
 
 
 
3468	}
3469	if (!found)
3470		return -ENOENT;
3471	ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3472					space_cache_ino);
3473	return ret;
3474}
3475
3476/*
3477 * helper to find all tree blocks that reference a given data extent
3478 */
3479static noinline_for_stack int add_data_references(struct reloc_control *rc,
3480						  const struct btrfs_key *extent_key,
3481						  struct btrfs_path *path,
3482						  struct rb_root *blocks)
3483{
3484	struct btrfs_backref_walk_ctx ctx = { 0 };
3485	struct ulist_iterator leaf_uiter;
3486	struct ulist_node *ref_node = NULL;
3487	const u32 blocksize = rc->extent_root->fs_info->nodesize;
 
 
 
 
3488	int ret = 0;
 
3489
3490	btrfs_release_path(path);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3491
3492	ctx.bytenr = extent_key->objectid;
3493	ctx.skip_inode_ref_list = true;
3494	ctx.fs_info = rc->extent_root->fs_info;
 
 
 
 
 
 
 
 
 
 
3495
3496	ret = btrfs_find_all_leafs(&ctx);
3497	if (ret < 0)
3498		return ret;
3499
3500	ULIST_ITER_INIT(&leaf_uiter);
3501	while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) {
3502		struct btrfs_tree_parent_check check = { 0 };
3503		struct extent_buffer *eb;
3504
3505		eb = read_tree_block(ctx.fs_info, ref_node->val, &check);
3506		if (IS_ERR(eb)) {
3507			ret = PTR_ERR(eb);
 
 
 
 
 
 
 
 
 
3508			break;
3509		}
3510		ret = delete_v1_space_cache(eb, rc->block_group,
3511					    extent_key->objectid);
3512		free_extent_buffer(eb);
3513		if (ret < 0)
3514			break;
3515		ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3516		if (ret < 0)
3517			break;
3518	}
3519	if (ret < 0)
 
 
3520		free_block_list(blocks);
3521	ulist_free(ctx.refs);
3522	return ret;
3523}
3524
3525/*
3526 * helper to find next unprocessed extent
3527 */
3528static noinline_for_stack
3529int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3530		     struct btrfs_key *extent_key)
3531{
3532	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3533	struct btrfs_key key;
3534	struct extent_buffer *leaf;
3535	u64 start, end, last;
3536	int ret;
3537
3538	last = rc->block_group->start + rc->block_group->length;
3539	while (1) {
3540		bool block_found;
3541
3542		cond_resched();
3543		if (rc->search_start >= last) {
3544			ret = 1;
3545			break;
3546		}
3547
3548		key.objectid = rc->search_start;
3549		key.type = BTRFS_EXTENT_ITEM_KEY;
3550		key.offset = 0;
3551
3552		path->search_commit_root = 1;
3553		path->skip_locking = 1;
3554		ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3555					0, 0);
3556		if (ret < 0)
3557			break;
3558next:
3559		leaf = path->nodes[0];
3560		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3561			ret = btrfs_next_leaf(rc->extent_root, path);
3562			if (ret != 0)
3563				break;
3564			leaf = path->nodes[0];
3565		}
3566
3567		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3568		if (key.objectid >= last) {
3569			ret = 1;
3570			break;
3571		}
3572
3573		if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3574		    key.type != BTRFS_METADATA_ITEM_KEY) {
3575			path->slots[0]++;
3576			goto next;
3577		}
3578
3579		if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3580		    key.objectid + key.offset <= rc->search_start) {
3581			path->slots[0]++;
3582			goto next;
3583		}
3584
3585		if (key.type == BTRFS_METADATA_ITEM_KEY &&
3586		    key.objectid + fs_info->nodesize <=
3587		    rc->search_start) {
3588			path->slots[0]++;
3589			goto next;
3590		}
3591
3592		block_found = find_first_extent_bit(&rc->processed_blocks,
3593						    key.objectid, &start, &end,
3594						    EXTENT_DIRTY, NULL);
3595
3596		if (block_found && start <= key.objectid) {
3597			btrfs_release_path(path);
3598			rc->search_start = end + 1;
3599		} else {
3600			if (key.type == BTRFS_EXTENT_ITEM_KEY)
3601				rc->search_start = key.objectid + key.offset;
3602			else
3603				rc->search_start = key.objectid +
3604					fs_info->nodesize;
3605			memcpy(extent_key, &key, sizeof(key));
3606			return 0;
3607		}
3608	}
3609	btrfs_release_path(path);
3610	return ret;
3611}
3612
3613static void set_reloc_control(struct reloc_control *rc)
3614{
3615	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3616
3617	mutex_lock(&fs_info->reloc_mutex);
3618	fs_info->reloc_ctl = rc;
3619	mutex_unlock(&fs_info->reloc_mutex);
3620}
3621
3622static void unset_reloc_control(struct reloc_control *rc)
3623{
3624	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3625
3626	mutex_lock(&fs_info->reloc_mutex);
3627	fs_info->reloc_ctl = NULL;
3628	mutex_unlock(&fs_info->reloc_mutex);
3629}
3630
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3631static noinline_for_stack
3632int prepare_to_relocate(struct reloc_control *rc)
3633{
3634	struct btrfs_trans_handle *trans;
3635	int ret;
3636
3637	rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3638					      BTRFS_BLOCK_RSV_TEMP);
3639	if (!rc->block_rsv)
3640		return -ENOMEM;
3641
3642	memset(&rc->cluster, 0, sizeof(rc->cluster));
3643	rc->search_start = rc->block_group->start;
3644	rc->extents_found = 0;
3645	rc->nodes_relocated = 0;
3646	rc->merging_rsv_size = 0;
3647	rc->reserved_bytes = 0;
3648	rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3649			      RELOCATION_RESERVED_NODES;
3650	ret = btrfs_block_rsv_refill(rc->extent_root->fs_info,
3651				     rc->block_rsv, rc->block_rsv->size,
3652				     BTRFS_RESERVE_FLUSH_ALL);
3653	if (ret)
3654		return ret;
3655
3656	rc->create_reloc_tree = true;
3657	set_reloc_control(rc);
3658
3659	trans = btrfs_join_transaction(rc->extent_root);
3660	if (IS_ERR(trans)) {
3661		unset_reloc_control(rc);
3662		/*
3663		 * extent tree is not a ref_cow tree and has no reloc_root to
3664		 * cleanup.  And callers are responsible to free the above
3665		 * block rsv.
3666		 */
3667		return PTR_ERR(trans);
3668	}
3669
3670	ret = btrfs_commit_transaction(trans);
3671	if (ret)
3672		unset_reloc_control(rc);
3673
3674	return ret;
3675}
3676
3677static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3678{
3679	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3680	struct rb_root blocks = RB_ROOT;
3681	struct btrfs_key key;
3682	struct btrfs_trans_handle *trans = NULL;
3683	struct btrfs_path *path;
3684	struct btrfs_extent_item *ei;
3685	u64 flags;
 
3686	int ret;
3687	int err = 0;
3688	int progress = 0;
3689
3690	path = btrfs_alloc_path();
3691	if (!path)
3692		return -ENOMEM;
3693	path->reada = READA_FORWARD;
3694
3695	ret = prepare_to_relocate(rc);
3696	if (ret) {
3697		err = ret;
3698		goto out_free;
3699	}
3700
3701	while (1) {
3702		rc->reserved_bytes = 0;
3703		ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
3704					     rc->block_rsv->size,
3705					     BTRFS_RESERVE_FLUSH_ALL);
3706		if (ret) {
3707			err = ret;
3708			break;
3709		}
3710		progress++;
3711		trans = btrfs_start_transaction(rc->extent_root, 0);
3712		if (IS_ERR(trans)) {
3713			err = PTR_ERR(trans);
3714			trans = NULL;
3715			break;
3716		}
3717restart:
3718		if (update_backref_cache(trans, &rc->backref_cache)) {
3719			btrfs_end_transaction(trans);
3720			trans = NULL;
3721			continue;
3722		}
3723
3724		ret = find_next_extent(rc, path, &key);
3725		if (ret < 0)
3726			err = ret;
3727		if (ret != 0)
3728			break;
3729
3730		rc->extents_found++;
3731
3732		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3733				    struct btrfs_extent_item);
3734		flags = btrfs_extent_flags(path->nodes[0], ei);
3735
3736		/*
3737		 * If we are relocating a simple quota owned extent item, we
3738		 * need to note the owner on the reloc data root so that when
3739		 * we allocate the replacement item, we can attribute it to the
3740		 * correct eventual owner (rather than the reloc data root).
3741		 */
3742		if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) {
3743			struct btrfs_root *root = BTRFS_I(rc->data_inode)->root;
3744			u64 owning_root_id = btrfs_get_extent_owner_root(fs_info,
3745								 path->nodes[0],
3746								 path->slots[0]);
3747
3748			root->relocation_src_root = owning_root_id;
3749		}
3750
3751		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3752			ret = add_tree_block(rc, &key, path, &blocks);
3753		} else if (rc->stage == UPDATE_DATA_PTRS &&
3754			   (flags & BTRFS_EXTENT_FLAG_DATA)) {
3755			ret = add_data_references(rc, &key, path, &blocks);
3756		} else {
3757			btrfs_release_path(path);
3758			ret = 0;
3759		}
3760		if (ret < 0) {
3761			err = ret;
3762			break;
3763		}
3764
3765		if (!RB_EMPTY_ROOT(&blocks)) {
3766			ret = relocate_tree_blocks(trans, rc, &blocks);
3767			if (ret < 0) {
 
 
 
 
 
 
3768				if (ret != -EAGAIN) {
3769					err = ret;
3770					break;
3771				}
3772				rc->extents_found--;
3773				rc->search_start = key.objectid;
3774			}
3775		}
3776
3777		btrfs_end_transaction_throttle(trans);
3778		btrfs_btree_balance_dirty(fs_info);
3779		trans = NULL;
3780
3781		if (rc->stage == MOVE_DATA_EXTENTS &&
3782		    (flags & BTRFS_EXTENT_FLAG_DATA)) {
3783			rc->found_file_extent = true;
3784			ret = relocate_data_extent(rc->data_inode,
3785						   &key, &rc->cluster);
3786			if (ret < 0) {
3787				err = ret;
3788				break;
3789			}
3790		}
3791		if (btrfs_should_cancel_balance(fs_info)) {
3792			err = -ECANCELED;
3793			break;
3794		}
3795	}
3796	if (trans && progress && err == -ENOSPC) {
3797		ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
3798		if (ret == 1) {
3799			err = 0;
3800			progress = 0;
3801			goto restart;
3802		}
3803	}
3804
3805	btrfs_release_path(path);
3806	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3807
3808	if (trans) {
3809		btrfs_end_transaction_throttle(trans);
3810		btrfs_btree_balance_dirty(fs_info);
3811	}
3812
3813	if (!err) {
3814		ret = relocate_file_extent_cluster(rc->data_inode,
3815						   &rc->cluster);
3816		if (ret < 0)
3817			err = ret;
3818	}
3819
3820	rc->create_reloc_tree = false;
3821	set_reloc_control(rc);
3822
3823	btrfs_backref_release_cache(&rc->backref_cache);
3824	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3825
3826	/*
3827	 * Even in the case when the relocation is cancelled, we should all go
3828	 * through prepare_to_merge() and merge_reloc_roots().
3829	 *
3830	 * For error (including cancelled balance), prepare_to_merge() will
3831	 * mark all reloc trees orphan, then queue them for cleanup in
3832	 * merge_reloc_roots()
3833	 */
3834	err = prepare_to_merge(rc, err);
3835
3836	merge_reloc_roots(rc);
3837
3838	rc->merge_reloc_tree = false;
3839	unset_reloc_control(rc);
3840	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3841
3842	/* get rid of pinned extents */
3843	trans = btrfs_join_transaction(rc->extent_root);
3844	if (IS_ERR(trans)) {
3845		err = PTR_ERR(trans);
3846		goto out_free;
3847	}
3848	ret = btrfs_commit_transaction(trans);
3849	if (ret && !err)
3850		err = ret;
3851out_free:
3852	ret = clean_dirty_subvols(rc);
3853	if (ret < 0 && !err)
3854		err = ret;
 
3855	btrfs_free_block_rsv(fs_info, rc->block_rsv);
3856	btrfs_free_path(path);
3857	return err;
3858}
3859
3860static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3861				 struct btrfs_root *root, u64 objectid)
3862{
3863	struct btrfs_path *path;
3864	struct btrfs_inode_item *item;
3865	struct extent_buffer *leaf;
3866	int ret;
3867
3868	path = btrfs_alloc_path();
3869	if (!path)
3870		return -ENOMEM;
3871
3872	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3873	if (ret)
3874		goto out;
3875
3876	leaf = path->nodes[0];
3877	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3878	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3879	btrfs_set_inode_generation(leaf, item, 1);
3880	btrfs_set_inode_size(leaf, item, 0);
3881	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3882	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3883					  BTRFS_INODE_PREALLOC);
3884	btrfs_mark_buffer_dirty(trans, leaf);
3885out:
3886	btrfs_free_path(path);
3887	return ret;
3888}
3889
3890static void delete_orphan_inode(struct btrfs_trans_handle *trans,
3891				struct btrfs_root *root, u64 objectid)
3892{
3893	struct btrfs_path *path;
3894	struct btrfs_key key;
3895	int ret = 0;
3896
3897	path = btrfs_alloc_path();
3898	if (!path) {
3899		ret = -ENOMEM;
3900		goto out;
3901	}
3902
3903	key.objectid = objectid;
3904	key.type = BTRFS_INODE_ITEM_KEY;
3905	key.offset = 0;
3906	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3907	if (ret) {
3908		if (ret > 0)
3909			ret = -ENOENT;
3910		goto out;
3911	}
3912	ret = btrfs_del_item(trans, root, path);
3913out:
3914	if (ret)
3915		btrfs_abort_transaction(trans, ret);
3916	btrfs_free_path(path);
3917}
3918
3919/*
3920 * helper to create inode for data relocation.
3921 * the inode is in data relocation tree and its link count is 0
3922 */
3923static noinline_for_stack struct inode *create_reloc_inode(
3924					struct btrfs_fs_info *fs_info,
3925					const struct btrfs_block_group *group)
3926{
3927	struct inode *inode = NULL;
3928	struct btrfs_trans_handle *trans;
3929	struct btrfs_root *root;
 
3930	u64 objectid;
3931	int err = 0;
3932
3933	root = btrfs_grab_root(fs_info->data_reloc_root);
 
 
 
3934	trans = btrfs_start_transaction(root, 6);
3935	if (IS_ERR(trans)) {
3936		btrfs_put_root(root);
3937		return ERR_CAST(trans);
3938	}
3939
3940	err = btrfs_get_free_objectid(root, &objectid);
3941	if (err)
3942		goto out;
3943
3944	err = __insert_orphan_inode(trans, root, objectid);
3945	if (err)
3946		goto out;
3947
3948	inode = btrfs_iget(fs_info->sb, objectid, root);
3949	if (IS_ERR(inode)) {
3950		delete_orphan_inode(trans, root, objectid);
3951		err = PTR_ERR(inode);
3952		inode = NULL;
3953		goto out;
3954	}
3955	BTRFS_I(inode)->index_cnt = group->start;
3956
3957	err = btrfs_orphan_add(trans, BTRFS_I(inode));
3958out:
3959	btrfs_put_root(root);
3960	btrfs_end_transaction(trans);
3961	btrfs_btree_balance_dirty(fs_info);
3962	if (err) {
3963		iput(inode);
 
3964		inode = ERR_PTR(err);
3965	}
3966	return inode;
3967}
3968
3969/*
3970 * Mark start of chunk relocation that is cancellable. Check if the cancellation
3971 * has been requested meanwhile and don't start in that case.
3972 *
3973 * Return:
3974 *   0             success
3975 *   -EINPROGRESS  operation is already in progress, that's probably a bug
3976 *   -ECANCELED    cancellation request was set before the operation started
3977 */
3978static int reloc_chunk_start(struct btrfs_fs_info *fs_info)
3979{
3980	if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) {
3981		/* This should not happen */
3982		btrfs_err(fs_info, "reloc already running, cannot start");
3983		return -EINPROGRESS;
3984	}
3985
3986	if (atomic_read(&fs_info->reloc_cancel_req) > 0) {
3987		btrfs_info(fs_info, "chunk relocation canceled on start");
3988		/*
3989		 * On cancel, clear all requests but let the caller mark
3990		 * the end after cleanup operations.
3991		 */
3992		atomic_set(&fs_info->reloc_cancel_req, 0);
3993		return -ECANCELED;
3994	}
3995	return 0;
3996}
3997
3998/*
3999 * Mark end of chunk relocation that is cancellable and wake any waiters.
4000 */
4001static void reloc_chunk_end(struct btrfs_fs_info *fs_info)
4002{
4003	/* Requested after start, clear bit first so any waiters can continue */
4004	if (atomic_read(&fs_info->reloc_cancel_req) > 0)
4005		btrfs_info(fs_info, "chunk relocation canceled during operation");
4006	clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags);
4007	atomic_set(&fs_info->reloc_cancel_req, 0);
4008}
4009
4010static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4011{
4012	struct reloc_control *rc;
4013
4014	rc = kzalloc(sizeof(*rc), GFP_NOFS);
4015	if (!rc)
4016		return NULL;
4017
4018	INIT_LIST_HEAD(&rc->reloc_roots);
4019	INIT_LIST_HEAD(&rc->dirty_subvol_roots);
4020	btrfs_backref_init_cache(fs_info, &rc->backref_cache, true);
4021	rc->reloc_root_tree.rb_root = RB_ROOT;
4022	spin_lock_init(&rc->reloc_root_tree.lock);
4023	extent_io_tree_init(fs_info, &rc->processed_blocks, IO_TREE_RELOC_BLOCKS);
4024	return rc;
4025}
4026
4027static void free_reloc_control(struct reloc_control *rc)
4028{
4029	struct mapping_node *node, *tmp;
4030
4031	free_reloc_roots(&rc->reloc_roots);
4032	rbtree_postorder_for_each_entry_safe(node, tmp,
4033			&rc->reloc_root_tree.rb_root, rb_node)
4034		kfree(node);
4035
4036	kfree(rc);
4037}
4038
4039/*
4040 * Print the block group being relocated
4041 */
4042static void describe_relocation(struct btrfs_fs_info *fs_info,
4043				struct btrfs_block_group *block_group)
4044{
4045	char buf[128] = {'\0'};
4046
4047	btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
4048
4049	btrfs_info(fs_info,
4050		   "relocating block group %llu flags %s",
4051		   block_group->start, buf);
4052}
4053
4054static const char *stage_to_string(enum reloc_stage stage)
4055{
4056	if (stage == MOVE_DATA_EXTENTS)
4057		return "move data extents";
4058	if (stage == UPDATE_DATA_PTRS)
4059		return "update data pointers";
4060	return "unknown";
4061}
4062
4063/*
4064 * function to relocate all extents in a block group.
4065 */
4066int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4067{
4068	struct btrfs_block_group *bg;
4069	struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start);
4070	struct reloc_control *rc;
4071	struct inode *inode;
4072	struct btrfs_path *path;
4073	int ret;
4074	int rw = 0;
4075	int err = 0;
4076
4077	/*
4078	 * This only gets set if we had a half-deleted snapshot on mount.  We
4079	 * cannot allow relocation to start while we're still trying to clean up
4080	 * these pending deletions.
4081	 */
4082	ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE);
4083	if (ret)
4084		return ret;
4085
4086	/* We may have been woken up by close_ctree, so bail if we're closing. */
4087	if (btrfs_fs_closing(fs_info))
4088		return -EINTR;
4089
4090	bg = btrfs_lookup_block_group(fs_info, group_start);
4091	if (!bg)
4092		return -ENOENT;
4093
4094	/*
4095	 * Relocation of a data block group creates ordered extents.  Without
4096	 * sb_start_write(), we can freeze the filesystem while unfinished
4097	 * ordered extents are left. Such ordered extents can cause a deadlock
4098	 * e.g. when syncfs() is waiting for their completion but they can't
4099	 * finish because they block when joining a transaction, due to the
4100	 * fact that the freeze locks are being held in write mode.
4101	 */
4102	if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
4103		ASSERT(sb_write_started(fs_info->sb));
4104
4105	if (btrfs_pinned_by_swapfile(fs_info, bg)) {
4106		btrfs_put_block_group(bg);
4107		return -ETXTBSY;
4108	}
4109
4110	rc = alloc_reloc_control(fs_info);
4111	if (!rc) {
4112		btrfs_put_block_group(bg);
4113		return -ENOMEM;
4114	}
4115
4116	ret = reloc_chunk_start(fs_info);
4117	if (ret < 0) {
4118		err = ret;
4119		goto out_put_bg;
4120	}
4121
4122	rc->extent_root = extent_root;
4123	rc->block_group = bg;
4124
4125	ret = btrfs_inc_block_group_ro(rc->block_group, true);
4126	if (ret) {
4127		err = ret;
4128		goto out;
4129	}
4130	rw = 1;
4131
4132	path = btrfs_alloc_path();
4133	if (!path) {
4134		err = -ENOMEM;
4135		goto out;
4136	}
4137
4138	inode = lookup_free_space_inode(rc->block_group, path);
4139	btrfs_free_path(path);
4140
4141	if (!IS_ERR(inode))
4142		ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4143	else
4144		ret = PTR_ERR(inode);
4145
4146	if (ret && ret != -ENOENT) {
4147		err = ret;
4148		goto out;
4149	}
4150
4151	rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4152	if (IS_ERR(rc->data_inode)) {
4153		err = PTR_ERR(rc->data_inode);
4154		rc->data_inode = NULL;
4155		goto out;
4156	}
4157
4158	describe_relocation(fs_info, rc->block_group);
4159
4160	btrfs_wait_block_group_reservations(rc->block_group);
4161	btrfs_wait_nocow_writers(rc->block_group);
4162	btrfs_wait_ordered_roots(fs_info, U64_MAX,
4163				 rc->block_group->start,
4164				 rc->block_group->length);
4165
4166	ret = btrfs_zone_finish(rc->block_group);
4167	WARN_ON(ret && ret != -EAGAIN);
4168
4169	while (1) {
4170		enum reloc_stage finishes_stage;
4171
4172		mutex_lock(&fs_info->cleaner_mutex);
4173		ret = relocate_block_group(rc);
4174		mutex_unlock(&fs_info->cleaner_mutex);
4175		if (ret < 0)
4176			err = ret;
4177
4178		finishes_stage = rc->stage;
4179		/*
4180		 * We may have gotten ENOSPC after we already dirtied some
4181		 * extents.  If writeout happens while we're relocating a
4182		 * different block group we could end up hitting the
4183		 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
4184		 * btrfs_reloc_cow_block.  Make sure we write everything out
4185		 * properly so we don't trip over this problem, and then break
4186		 * out of the loop if we hit an error.
4187		 */
4188		if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4189			ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4190						       (u64)-1);
4191			if (ret)
4192				err = ret;
4193			invalidate_mapping_pages(rc->data_inode->i_mapping,
4194						 0, -1);
4195			rc->stage = UPDATE_DATA_PTRS;
4196		}
4197
4198		if (err < 0)
4199			goto out;
4200
4201		if (rc->extents_found == 0)
4202			break;
4203
4204		btrfs_info(fs_info, "found %llu extents, stage: %s",
4205			   rc->extents_found, stage_to_string(finishes_stage));
4206	}
4207
4208	WARN_ON(rc->block_group->pinned > 0);
4209	WARN_ON(rc->block_group->reserved > 0);
4210	WARN_ON(rc->block_group->used > 0);
4211out:
4212	if (err && rw)
4213		btrfs_dec_block_group_ro(rc->block_group);
4214	iput(rc->data_inode);
4215out_put_bg:
4216	btrfs_put_block_group(bg);
4217	reloc_chunk_end(fs_info);
4218	free_reloc_control(rc);
4219	return err;
4220}
4221
4222static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4223{
4224	struct btrfs_fs_info *fs_info = root->fs_info;
4225	struct btrfs_trans_handle *trans;
4226	int ret, err;
4227
4228	trans = btrfs_start_transaction(fs_info->tree_root, 0);
4229	if (IS_ERR(trans))
4230		return PTR_ERR(trans);
4231
4232	memset(&root->root_item.drop_progress, 0,
4233		sizeof(root->root_item.drop_progress));
4234	btrfs_set_root_drop_level(&root->root_item, 0);
4235	btrfs_set_root_refs(&root->root_item, 0);
4236	ret = btrfs_update_root(trans, fs_info->tree_root,
4237				&root->root_key, &root->root_item);
4238
4239	err = btrfs_end_transaction(trans);
4240	if (err)
4241		return err;
4242	return ret;
4243}
4244
4245/*
4246 * recover relocation interrupted by system crash.
4247 *
4248 * this function resumes merging reloc trees with corresponding fs trees.
4249 * this is important for keeping the sharing of tree blocks
4250 */
4251int btrfs_recover_relocation(struct btrfs_fs_info *fs_info)
4252{
 
4253	LIST_HEAD(reloc_roots);
4254	struct btrfs_key key;
4255	struct btrfs_root *fs_root;
4256	struct btrfs_root *reloc_root;
4257	struct btrfs_path *path;
4258	struct extent_buffer *leaf;
4259	struct reloc_control *rc = NULL;
4260	struct btrfs_trans_handle *trans;
4261	int ret;
4262	int err = 0;
4263
4264	path = btrfs_alloc_path();
4265	if (!path)
4266		return -ENOMEM;
4267	path->reada = READA_BACK;
4268
4269	key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4270	key.type = BTRFS_ROOT_ITEM_KEY;
4271	key.offset = (u64)-1;
4272
4273	while (1) {
4274		ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4275					path, 0, 0);
4276		if (ret < 0) {
4277			err = ret;
4278			goto out;
4279		}
4280		if (ret > 0) {
4281			if (path->slots[0] == 0)
4282				break;
4283			path->slots[0]--;
4284		}
4285		leaf = path->nodes[0];
4286		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4287		btrfs_release_path(path);
4288
4289		if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4290		    key.type != BTRFS_ROOT_ITEM_KEY)
4291			break;
4292
4293		reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key);
4294		if (IS_ERR(reloc_root)) {
4295			err = PTR_ERR(reloc_root);
4296			goto out;
4297		}
4298
4299		set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
4300		list_add(&reloc_root->root_list, &reloc_roots);
4301
4302		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4303			fs_root = btrfs_get_fs_root(fs_info,
4304					reloc_root->root_key.offset, false);
4305			if (IS_ERR(fs_root)) {
4306				ret = PTR_ERR(fs_root);
4307				if (ret != -ENOENT) {
4308					err = ret;
4309					goto out;
4310				}
4311				ret = mark_garbage_root(reloc_root);
4312				if (ret < 0) {
4313					err = ret;
4314					goto out;
4315				}
4316			} else {
4317				btrfs_put_root(fs_root);
4318			}
4319		}
4320
4321		if (key.offset == 0)
4322			break;
4323
4324		key.offset--;
4325	}
4326	btrfs_release_path(path);
4327
4328	if (list_empty(&reloc_roots))
4329		goto out;
4330
4331	rc = alloc_reloc_control(fs_info);
4332	if (!rc) {
4333		err = -ENOMEM;
4334		goto out;
4335	}
4336
4337	ret = reloc_chunk_start(fs_info);
4338	if (ret < 0) {
4339		err = ret;
4340		goto out_end;
4341	}
4342
4343	rc->extent_root = btrfs_extent_root(fs_info, 0);
4344
4345	set_reloc_control(rc);
4346
4347	trans = btrfs_join_transaction(rc->extent_root);
4348	if (IS_ERR(trans)) {
 
4349		err = PTR_ERR(trans);
4350		goto out_unset;
4351	}
4352
4353	rc->merge_reloc_tree = true;
4354
4355	while (!list_empty(&reloc_roots)) {
4356		reloc_root = list_entry(reloc_roots.next,
4357					struct btrfs_root, root_list);
4358		list_del(&reloc_root->root_list);
4359
4360		if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4361			list_add_tail(&reloc_root->root_list,
4362				      &rc->reloc_roots);
4363			continue;
4364		}
4365
4366		fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
4367					    false);
4368		if (IS_ERR(fs_root)) {
4369			err = PTR_ERR(fs_root);
4370			list_add_tail(&reloc_root->root_list, &reloc_roots);
4371			btrfs_end_transaction(trans);
4372			goto out_unset;
4373		}
4374
4375		err = __add_reloc_root(reloc_root);
4376		ASSERT(err != -EEXIST);
4377		if (err) {
4378			list_add_tail(&reloc_root->root_list, &reloc_roots);
4379			btrfs_put_root(fs_root);
4380			btrfs_end_transaction(trans);
4381			goto out_unset;
4382		}
4383		fs_root->reloc_root = btrfs_grab_root(reloc_root);
4384		btrfs_put_root(fs_root);
4385	}
4386
4387	err = btrfs_commit_transaction(trans);
4388	if (err)
4389		goto out_unset;
4390
4391	merge_reloc_roots(rc);
4392
4393	unset_reloc_control(rc);
4394
4395	trans = btrfs_join_transaction(rc->extent_root);
4396	if (IS_ERR(trans)) {
4397		err = PTR_ERR(trans);
4398		goto out_clean;
4399	}
4400	err = btrfs_commit_transaction(trans);
4401out_clean:
4402	ret = clean_dirty_subvols(rc);
4403	if (ret < 0 && !err)
4404		err = ret;
4405out_unset:
4406	unset_reloc_control(rc);
4407out_end:
4408	reloc_chunk_end(fs_info);
4409	free_reloc_control(rc);
4410out:
4411	free_reloc_roots(&reloc_roots);
 
4412
4413	btrfs_free_path(path);
4414
4415	if (err == 0) {
4416		/* cleanup orphan inode in data relocation tree */
4417		fs_root = btrfs_grab_root(fs_info->data_reloc_root);
4418		ASSERT(fs_root);
4419		err = btrfs_orphan_cleanup(fs_root);
4420		btrfs_put_root(fs_root);
 
4421	}
4422	return err;
4423}
4424
4425/*
4426 * helper to add ordered checksum for data relocation.
4427 *
4428 * cloning checksum properly handles the nodatasum extents.
4429 * it also saves CPU time to re-calculate the checksum.
4430 */
4431int btrfs_reloc_clone_csums(struct btrfs_ordered_extent *ordered)
4432{
4433	struct btrfs_inode *inode = BTRFS_I(ordered->inode);
4434	struct btrfs_fs_info *fs_info = inode->root->fs_info;
4435	u64 disk_bytenr = ordered->file_offset + inode->index_cnt;
4436	struct btrfs_root *csum_root = btrfs_csum_root(fs_info, disk_bytenr);
 
 
4437	LIST_HEAD(list);
4438	int ret;
4439
4440	ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4441				      disk_bytenr + ordered->num_bytes - 1,
4442				      &list, 0, false);
 
 
 
4443	if (ret)
4444		return ret;
4445
4446	while (!list_empty(&list)) {
4447		struct btrfs_ordered_sum *sums =
4448			list_entry(list.next, struct btrfs_ordered_sum, list);
4449
4450		list_del_init(&sums->list);
4451
4452		/*
4453		 * We need to offset the new_bytenr based on where the csum is.
4454		 * We need to do this because we will read in entire prealloc
4455		 * extents but we may have written to say the middle of the
4456		 * prealloc extent, so we need to make sure the csum goes with
4457		 * the right disk offset.
4458		 *
4459		 * We can do this because the data reloc inode refers strictly
4460		 * to the on disk bytes, so we don't have to worry about
4461		 * disk_len vs real len like with real inodes since it's all
4462		 * disk length.
4463		 */
4464		sums->logical = ordered->disk_bytenr + sums->logical - disk_bytenr;
 
 
4465		btrfs_add_ordered_sum(ordered, sums);
4466	}
4467
4468	return 0;
 
4469}
4470
4471int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4472			  struct btrfs_root *root,
4473			  const struct extent_buffer *buf,
4474			  struct extent_buffer *cow)
4475{
4476	struct btrfs_fs_info *fs_info = root->fs_info;
4477	struct reloc_control *rc;
4478	struct btrfs_backref_node *node;
4479	int first_cow = 0;
4480	int level;
4481	int ret = 0;
4482
4483	rc = fs_info->reloc_ctl;
4484	if (!rc)
4485		return 0;
4486
4487	BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root));
 
 
 
 
 
 
4488
4489	level = btrfs_header_level(buf);
4490	if (btrfs_header_generation(buf) <=
4491	    btrfs_root_last_snapshot(&root->root_item))
4492		first_cow = 1;
4493
4494	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4495	    rc->create_reloc_tree) {
4496		WARN_ON(!first_cow && level == 0);
4497
4498		node = rc->backref_cache.path[level];
4499		BUG_ON(node->bytenr != buf->start &&
4500		       node->new_bytenr != buf->start);
4501
4502		btrfs_backref_drop_node_buffer(node);
4503		atomic_inc(&cow->refs);
4504		node->eb = cow;
4505		node->new_bytenr = cow->start;
4506
4507		if (!node->pending) {
4508			list_move_tail(&node->list,
4509				       &rc->backref_cache.pending[level]);
4510			node->pending = 1;
4511		}
4512
4513		if (first_cow)
4514			mark_block_processed(rc, node);
4515
4516		if (first_cow && level > 0)
4517			rc->nodes_relocated += buf->len;
4518	}
4519
4520	if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4521		ret = replace_file_extents(trans, rc, root, cow);
4522	return ret;
4523}
4524
4525/*
4526 * called before creating snapshot. it calculates metadata reservation
4527 * required for relocating tree blocks in the snapshot
4528 */
4529void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4530			      u64 *bytes_to_reserve)
4531{
4532	struct btrfs_root *root = pending->root;
4533	struct reloc_control *rc = root->fs_info->reloc_ctl;
4534
4535	if (!rc || !have_reloc_root(root))
4536		return;
4537
4538	if (!rc->merge_reloc_tree)
4539		return;
4540
4541	root = root->reloc_root;
4542	BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4543	/*
4544	 * relocation is in the stage of merging trees. the space
4545	 * used by merging a reloc tree is twice the size of
4546	 * relocated tree nodes in the worst case. half for cowing
4547	 * the reloc tree, half for cowing the fs tree. the space
4548	 * used by cowing the reloc tree will be freed after the
4549	 * tree is dropped. if we create snapshot, cowing the fs
4550	 * tree may use more space than it frees. so we need
4551	 * reserve extra space.
4552	 */
4553	*bytes_to_reserve += rc->nodes_relocated;
4554}
4555
4556/*
4557 * called after snapshot is created. migrate block reservation
4558 * and create reloc root for the newly created snapshot
4559 *
4560 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4561 * references held on the reloc_root, one for root->reloc_root and one for
4562 * rc->reloc_roots.
4563 */
4564int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4565			       struct btrfs_pending_snapshot *pending)
4566{
4567	struct btrfs_root *root = pending->root;
4568	struct btrfs_root *reloc_root;
4569	struct btrfs_root *new_root;
4570	struct reloc_control *rc = root->fs_info->reloc_ctl;
4571	int ret;
4572
4573	if (!rc || !have_reloc_root(root))
4574		return 0;
4575
4576	rc = root->fs_info->reloc_ctl;
4577	rc->merging_rsv_size += rc->nodes_relocated;
4578
4579	if (rc->merge_reloc_tree) {
4580		ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4581					      rc->block_rsv,
4582					      rc->nodes_relocated, true);
4583		if (ret)
4584			return ret;
4585	}
4586
4587	new_root = pending->snap;
4588	reloc_root = create_reloc_root(trans, root->reloc_root,
4589				       new_root->root_key.objectid);
4590	if (IS_ERR(reloc_root))
4591		return PTR_ERR(reloc_root);
4592
4593	ret = __add_reloc_root(reloc_root);
4594	ASSERT(ret != -EEXIST);
4595	if (ret) {
4596		/* Pairs with create_reloc_root */
4597		btrfs_put_root(reloc_root);
4598		return ret;
4599	}
4600	new_root->reloc_root = btrfs_grab_root(reloc_root);
4601
4602	if (rc->create_reloc_tree)
4603		ret = clone_backref_node(trans, rc, root, reloc_root);
4604	return ret;
4605}
4606
4607/*
4608 * Get the current bytenr for the block group which is being relocated.
4609 *
4610 * Return U64_MAX if no running relocation.
4611 */
4612u64 btrfs_get_reloc_bg_bytenr(const struct btrfs_fs_info *fs_info)
4613{
4614	u64 logical = U64_MAX;
4615
4616	lockdep_assert_held(&fs_info->reloc_mutex);
4617
4618	if (fs_info->reloc_ctl && fs_info->reloc_ctl->block_group)
4619		logical = fs_info->reloc_ctl->block_group->start;
4620	return logical;
4621}