Linux Audio

Check our new training course

Loading...
v4.10.11
 
   1/*
   2 * Copyright (C) 2009 Oracle.  All rights reserved.
   3 *
   4 * This program is free software; you can redistribute it and/or
   5 * modify it under the terms of the GNU General Public
   6 * License v2 as published by the Free Software Foundation.
   7 *
   8 * This program is distributed in the hope that it will be useful,
   9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11 * General Public License for more details.
  12 *
  13 * You should have received a copy of the GNU General Public
  14 * License along with this program; if not, write to the
  15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16 * Boston, MA 021110-1307, USA.
  17 */
  18
  19#include <linux/sched.h>
  20#include <linux/pagemap.h>
  21#include <linux/writeback.h>
  22#include <linux/blkdev.h>
  23#include <linux/rbtree.h>
  24#include <linux/slab.h>
 
  25#include "ctree.h"
  26#include "disk-io.h"
  27#include "transaction.h"
  28#include "volumes.h"
  29#include "locking.h"
  30#include "btrfs_inode.h"
  31#include "async-thread.h"
  32#include "free-space-cache.h"
  33#include "inode-map.h"
  34#include "qgroup.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  35
  36/*
  37 * backref_node, mapping_node and tree_block start with this
  38 */
  39struct tree_entry {
  40	struct rb_node rb_node;
  41	u64 bytenr;
  42};
  43
  44/*
  45 * present a tree block in the backref cache
  46 */
  47struct backref_node {
  48	struct rb_node rb_node;
  49	u64 bytenr;
  50
  51	u64 new_bytenr;
  52	/* objectid of tree block owner, can be not uptodate */
  53	u64 owner;
  54	/* link to pending, changed or detached list */
  55	struct list_head list;
  56	/* list of upper level blocks reference this block */
  57	struct list_head upper;
  58	/* list of child blocks in the cache */
  59	struct list_head lower;
  60	/* NULL if this node is not tree root */
  61	struct btrfs_root *root;
  62	/* extent buffer got by COW the block */
  63	struct extent_buffer *eb;
  64	/* level of tree block */
  65	unsigned int level:8;
  66	/* is the block in non-reference counted tree */
  67	unsigned int cowonly:1;
  68	/* 1 if no child node in the cache */
  69	unsigned int lowest:1;
  70	/* is the extent buffer locked */
  71	unsigned int locked:1;
  72	/* has the block been processed */
  73	unsigned int processed:1;
  74	/* have backrefs of this block been checked */
  75	unsigned int checked:1;
  76	/*
  77	 * 1 if corresponding block has been cowed but some upper
  78	 * level block pointers may not point to the new location
  79	 */
  80	unsigned int pending:1;
  81	/*
  82	 * 1 if the backref node isn't connected to any other
  83	 * backref node.
  84	 */
  85	unsigned int detached:1;
  86};
  87
  88/*
  89 * present a block pointer in the backref cache
  90 */
  91struct backref_edge {
  92	struct list_head list[2];
  93	struct backref_node *node[2];
  94};
  95
  96#define LOWER	0
  97#define UPPER	1
  98#define RELOCATION_RESERVED_NODES	256
  99
 100struct backref_cache {
 101	/* red black tree of all backref nodes in the cache */
 102	struct rb_root rb_root;
 103	/* for passing backref nodes to btrfs_reloc_cow_block */
 104	struct backref_node *path[BTRFS_MAX_LEVEL];
 105	/*
 106	 * list of blocks that have been cowed but some block
 107	 * pointers in upper level blocks may not reflect the
 108	 * new location
 109	 */
 110	struct list_head pending[BTRFS_MAX_LEVEL];
 111	/* list of backref nodes with no child node */
 112	struct list_head leaves;
 113	/* list of blocks that have been cowed in current transaction */
 114	struct list_head changed;
 115	/* list of detached backref node. */
 116	struct list_head detached;
 117
 118	u64 last_trans;
 119
 120	int nr_nodes;
 121	int nr_edges;
 122};
 123
 124/*
 125 * map address of tree root to tree
 126 */
 127struct mapping_node {
 128	struct rb_node rb_node;
 129	u64 bytenr;
 
 
 130	void *data;
 131};
 132
 133struct mapping_tree {
 134	struct rb_root rb_root;
 135	spinlock_t lock;
 136};
 137
 138/*
 139 * present a tree block to process
 140 */
 141struct tree_block {
 142	struct rb_node rb_node;
 143	u64 bytenr;
 
 
 
 144	struct btrfs_key key;
 145	unsigned int level:8;
 146	unsigned int key_ready:1;
 147};
 148
 149#define MAX_EXTENTS 128
 150
 151struct file_extent_cluster {
 152	u64 start;
 153	u64 end;
 154	u64 boundary[MAX_EXTENTS];
 155	unsigned int nr;
 
 
 
 
 
 
 
 156};
 157
 158struct reloc_control {
 159	/* block group to relocate */
 160	struct btrfs_block_group_cache *block_group;
 161	/* extent tree */
 162	struct btrfs_root *extent_root;
 163	/* inode for moving data */
 164	struct inode *data_inode;
 165
 166	struct btrfs_block_rsv *block_rsv;
 167
 168	struct backref_cache backref_cache;
 169
 170	struct file_extent_cluster cluster;
 171	/* tree blocks have been processed */
 172	struct extent_io_tree processed_blocks;
 173	/* map start of tree root to corresponding reloc tree */
 174	struct mapping_tree reloc_root_tree;
 175	/* list of reloc trees */
 176	struct list_head reloc_roots;
 
 
 177	/* size of metadata reservation for merging reloc trees */
 178	u64 merging_rsv_size;
 179	/* size of relocated tree nodes */
 180	u64 nodes_relocated;
 181	/* reserved size for block group relocation*/
 182	u64 reserved_bytes;
 183
 184	u64 search_start;
 185	u64 extents_found;
 186
 187	unsigned int stage:8;
 188	unsigned int create_reloc_tree:1;
 189	unsigned int merge_reloc_tree:1;
 190	unsigned int found_file_extent:1;
 191};
 192
 193/* stages of data relocation */
 194#define MOVE_DATA_EXTENTS	0
 195#define UPDATE_DATA_PTRS	1
 196
 197static void remove_backref_node(struct backref_cache *cache,
 198				struct backref_node *node);
 199static void __mark_block_processed(struct reloc_control *rc,
 200				   struct backref_node *node);
 201
 202static void mapping_tree_init(struct mapping_tree *tree)
 203{
 204	tree->rb_root = RB_ROOT;
 205	spin_lock_init(&tree->lock);
 206}
 207
 208static void backref_cache_init(struct backref_cache *cache)
 209{
 210	int i;
 211	cache->rb_root = RB_ROOT;
 212	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
 213		INIT_LIST_HEAD(&cache->pending[i]);
 214	INIT_LIST_HEAD(&cache->changed);
 215	INIT_LIST_HEAD(&cache->detached);
 216	INIT_LIST_HEAD(&cache->leaves);
 217}
 218
 219static void backref_cache_cleanup(struct backref_cache *cache)
 220{
 221	struct backref_node *node;
 222	int i;
 223
 224	while (!list_empty(&cache->detached)) {
 225		node = list_entry(cache->detached.next,
 226				  struct backref_node, list);
 227		remove_backref_node(cache, node);
 228	}
 229
 230	while (!list_empty(&cache->leaves)) {
 231		node = list_entry(cache->leaves.next,
 232				  struct backref_node, lower);
 233		remove_backref_node(cache, node);
 234	}
 235
 236	cache->last_trans = 0;
 237
 238	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
 239		ASSERT(list_empty(&cache->pending[i]));
 240	ASSERT(list_empty(&cache->changed));
 241	ASSERT(list_empty(&cache->detached));
 242	ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
 243	ASSERT(!cache->nr_nodes);
 244	ASSERT(!cache->nr_edges);
 245}
 246
 247static struct backref_node *alloc_backref_node(struct backref_cache *cache)
 248{
 249	struct backref_node *node;
 250
 251	node = kzalloc(sizeof(*node), GFP_NOFS);
 252	if (node) {
 253		INIT_LIST_HEAD(&node->list);
 254		INIT_LIST_HEAD(&node->upper);
 255		INIT_LIST_HEAD(&node->lower);
 256		RB_CLEAR_NODE(&node->rb_node);
 257		cache->nr_nodes++;
 258	}
 259	return node;
 260}
 261
 262static void free_backref_node(struct backref_cache *cache,
 263			      struct backref_node *node)
 264{
 265	if (node) {
 266		cache->nr_nodes--;
 267		kfree(node);
 268	}
 269}
 270
 271static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
 272{
 273	struct backref_edge *edge;
 274
 275	edge = kzalloc(sizeof(*edge), GFP_NOFS);
 276	if (edge)
 277		cache->nr_edges++;
 278	return edge;
 279}
 280
 281static void free_backref_edge(struct backref_cache *cache,
 282			      struct backref_edge *edge)
 283{
 284	if (edge) {
 285		cache->nr_edges--;
 286		kfree(edge);
 287	}
 288}
 289
 290static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
 291				   struct rb_node *node)
 292{
 293	struct rb_node **p = &root->rb_node;
 294	struct rb_node *parent = NULL;
 295	struct tree_entry *entry;
 296
 297	while (*p) {
 298		parent = *p;
 299		entry = rb_entry(parent, struct tree_entry, rb_node);
 300
 301		if (bytenr < entry->bytenr)
 302			p = &(*p)->rb_left;
 303		else if (bytenr > entry->bytenr)
 304			p = &(*p)->rb_right;
 305		else
 306			return parent;
 307	}
 308
 309	rb_link_node(node, parent, p);
 310	rb_insert_color(node, root);
 311	return NULL;
 312}
 313
 314static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
 315{
 316	struct rb_node *n = root->rb_node;
 317	struct tree_entry *entry;
 318
 319	while (n) {
 320		entry = rb_entry(n, struct tree_entry, rb_node);
 321
 322		if (bytenr < entry->bytenr)
 323			n = n->rb_left;
 324		else if (bytenr > entry->bytenr)
 325			n = n->rb_right;
 326		else
 327			return n;
 328	}
 329	return NULL;
 330}
 331
 332static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
 333{
 334
 335	struct btrfs_fs_info *fs_info = NULL;
 336	struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
 337					      rb_node);
 338	if (bnode->root)
 339		fs_info = bnode->root->fs_info;
 340	btrfs_panic(fs_info, errno,
 341		    "Inconsistency in backref cache found at offset %llu",
 342		    bytenr);
 343}
 344
 345/*
 346 * walk up backref nodes until reach node presents tree root
 347 */
 348static struct backref_node *walk_up_backref(struct backref_node *node,
 349					    struct backref_edge *edges[],
 350					    int *index)
 351{
 352	struct backref_edge *edge;
 353	int idx = *index;
 354
 355	while (!list_empty(&node->upper)) {
 356		edge = list_entry(node->upper.next,
 357				  struct backref_edge, list[LOWER]);
 358		edges[idx++] = edge;
 359		node = edge->node[UPPER];
 360	}
 361	BUG_ON(node->detached);
 362	*index = idx;
 363	return node;
 364}
 365
 366/*
 367 * walk down backref nodes to find start of next reference path
 368 */
 369static struct backref_node *walk_down_backref(struct backref_edge *edges[],
 370					      int *index)
 371{
 372	struct backref_edge *edge;
 373	struct backref_node *lower;
 374	int idx = *index;
 375
 376	while (idx > 0) {
 377		edge = edges[idx - 1];
 378		lower = edge->node[LOWER];
 379		if (list_is_last(&edge->list[LOWER], &lower->upper)) {
 380			idx--;
 381			continue;
 382		}
 383		edge = list_entry(edge->list[LOWER].next,
 384				  struct backref_edge, list[LOWER]);
 385		edges[idx - 1] = edge;
 386		*index = idx;
 387		return edge->node[UPPER];
 388	}
 389	*index = 0;
 390	return NULL;
 391}
 392
 393static void unlock_node_buffer(struct backref_node *node)
 394{
 395	if (node->locked) {
 396		btrfs_tree_unlock(node->eb);
 397		node->locked = 0;
 398	}
 399}
 400
 401static void drop_node_buffer(struct backref_node *node)
 402{
 403	if (node->eb) {
 404		unlock_node_buffer(node);
 405		free_extent_buffer(node->eb);
 406		node->eb = NULL;
 407	}
 408}
 409
 410static void drop_backref_node(struct backref_cache *tree,
 411			      struct backref_node *node)
 412{
 413	BUG_ON(!list_empty(&node->upper));
 414
 415	drop_node_buffer(node);
 416	list_del(&node->list);
 417	list_del(&node->lower);
 418	if (!RB_EMPTY_NODE(&node->rb_node))
 419		rb_erase(&node->rb_node, &tree->rb_root);
 420	free_backref_node(tree, node);
 421}
 422
 423/*
 424 * remove a backref node from the backref cache
 425 */
 426static void remove_backref_node(struct backref_cache *cache,
 427				struct backref_node *node)
 428{
 429	struct backref_node *upper;
 430	struct backref_edge *edge;
 431
 432	if (!node)
 433		return;
 434
 435	BUG_ON(!node->lowest && !node->detached);
 436	while (!list_empty(&node->upper)) {
 437		edge = list_entry(node->upper.next, struct backref_edge,
 438				  list[LOWER]);
 439		upper = edge->node[UPPER];
 440		list_del(&edge->list[LOWER]);
 441		list_del(&edge->list[UPPER]);
 442		free_backref_edge(cache, edge);
 443
 444		if (RB_EMPTY_NODE(&upper->rb_node)) {
 445			BUG_ON(!list_empty(&node->upper));
 446			drop_backref_node(cache, node);
 447			node = upper;
 448			node->lowest = 1;
 449			continue;
 450		}
 451		/*
 452		 * add the node to leaf node list if no other
 453		 * child block cached.
 454		 */
 455		if (list_empty(&upper->lower)) {
 456			list_add_tail(&upper->lower, &cache->leaves);
 457			upper->lowest = 1;
 458		}
 459	}
 460
 461	drop_backref_node(cache, node);
 462}
 463
 464static void update_backref_node(struct backref_cache *cache,
 465				struct backref_node *node, u64 bytenr)
 466{
 467	struct rb_node *rb_node;
 468	rb_erase(&node->rb_node, &cache->rb_root);
 469	node->bytenr = bytenr;
 470	rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
 471	if (rb_node)
 472		backref_tree_panic(rb_node, -EEXIST, bytenr);
 473}
 474
 475/*
 476 * update backref cache after a transaction commit
 
 
 
 
 
 477 */
 478static int update_backref_cache(struct btrfs_trans_handle *trans,
 479				struct backref_cache *cache)
 480{
 481	struct backref_node *node;
 482	int level = 0;
 483
 484	if (cache->last_trans == 0) {
 485		cache->last_trans = trans->transid;
 486		return 0;
 487	}
 488
 489	if (cache->last_trans == trans->transid)
 490		return 0;
 491
 492	/*
 493	 * detached nodes are used to avoid unnecessary backref
 494	 * lookup. transaction commit changes the extent tree.
 495	 * so the detached nodes are no longer useful.
 496	 */
 497	while (!list_empty(&cache->detached)) {
 498		node = list_entry(cache->detached.next,
 499				  struct backref_node, list);
 500		remove_backref_node(cache, node);
 501	}
 502
 503	while (!list_empty(&cache->changed)) {
 504		node = list_entry(cache->changed.next,
 505				  struct backref_node, list);
 506		list_del_init(&node->list);
 507		BUG_ON(node->pending);
 508		update_backref_node(cache, node, node->new_bytenr);
 509	}
 510
 511	/*
 512	 * some nodes can be left in the pending list if there were
 513	 * errors during processing the pending nodes.
 514	 */
 515	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
 516		list_for_each_entry(node, &cache->pending[level], list) {
 517			BUG_ON(!node->pending);
 518			if (node->bytenr == node->new_bytenr)
 519				continue;
 520			update_backref_node(cache, node, node->new_bytenr);
 521		}
 522	}
 523
 524	cache->last_trans = 0;
 525	return 1;
 526}
 527
 528
 529static int should_ignore_root(struct btrfs_root *root)
 530{
 531	struct btrfs_root *reloc_root;
 532
 533	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 534		return 0;
 
 
 
 
 535
 536	reloc_root = root->reloc_root;
 537	if (!reloc_root)
 538		return 0;
 539
 540	if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
 541	    root->fs_info->running_transaction->transid - 1)
 542		return 0;
 543	/*
 544	 * if there is reloc tree and it was created in previous
 545	 * transaction backref lookup can find the reloc tree,
 546	 * so backref node for the fs tree root is useless for
 547	 * relocation.
 548	 */
 549	return 1;
 550}
 
 551/*
 552 * find reloc tree by address of tree root
 553 */
 554static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
 555					  u64 bytenr)
 556{
 
 557	struct rb_node *rb_node;
 558	struct mapping_node *node;
 559	struct btrfs_root *root = NULL;
 560
 
 561	spin_lock(&rc->reloc_root_tree.lock);
 562	rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
 563	if (rb_node) {
 564		node = rb_entry(rb_node, struct mapping_node, rb_node);
 565		root = (struct btrfs_root *)node->data;
 566	}
 567	spin_unlock(&rc->reloc_root_tree.lock);
 568	return root;
 569}
 570
 571static int is_cowonly_root(u64 root_objectid)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 572{
 573	if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
 574	    root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
 575	    root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
 576	    root_objectid == BTRFS_DEV_TREE_OBJECTID ||
 577	    root_objectid == BTRFS_TREE_LOG_OBJECTID ||
 578	    root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
 579	    root_objectid == BTRFS_UUID_TREE_OBJECTID ||
 580	    root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
 581	    root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
 582		return 1;
 583	return 0;
 584}
 585
 586static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
 587					u64 root_objectid)
 588{
 589	struct btrfs_key key;
 590
 591	key.objectid = root_objectid;
 592	key.type = BTRFS_ROOT_ITEM_KEY;
 593	if (is_cowonly_root(root_objectid))
 594		key.offset = 0;
 595	else
 596		key.offset = (u64)-1;
 597
 598	return btrfs_get_fs_root(fs_info, &key, false);
 599}
 600
 601#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 602static noinline_for_stack
 603struct btrfs_root *find_tree_root(struct reloc_control *rc,
 604				  struct extent_buffer *leaf,
 605				  struct btrfs_extent_ref_v0 *ref0)
 606{
 607	struct btrfs_root *root;
 608	u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
 609	u64 generation = btrfs_ref_generation_v0(leaf, ref0);
 610
 611	BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
 
 612
 613	root = read_fs_root(rc->extent_root->fs_info, root_objectid);
 614	BUG_ON(IS_ERR(root));
 615
 616	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
 617	    generation != btrfs_root_generation(&root->root_item))
 618		return NULL;
 
 
 619
 620	return root;
 621}
 622#endif
 
 623
 624static noinline_for_stack
 625int find_inline_backref(struct extent_buffer *leaf, int slot,
 626			unsigned long *ptr, unsigned long *end)
 627{
 628	struct btrfs_key key;
 629	struct btrfs_extent_item *ei;
 630	struct btrfs_tree_block_info *bi;
 631	u32 item_size;
 632
 633	btrfs_item_key_to_cpu(leaf, &key, slot);
 634
 635	item_size = btrfs_item_size_nr(leaf, slot);
 636#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 637	if (item_size < sizeof(*ei)) {
 638		WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
 639		return 1;
 640	}
 641#endif
 642	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 643	WARN_ON(!(btrfs_extent_flags(leaf, ei) &
 644		  BTRFS_EXTENT_FLAG_TREE_BLOCK));
 645
 646	if (key.type == BTRFS_EXTENT_ITEM_KEY &&
 647	    item_size <= sizeof(*ei) + sizeof(*bi)) {
 648		WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
 649		return 1;
 650	}
 651	if (key.type == BTRFS_METADATA_ITEM_KEY &&
 652	    item_size <= sizeof(*ei)) {
 653		WARN_ON(item_size < sizeof(*ei));
 654		return 1;
 655	}
 656
 657	if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 658		bi = (struct btrfs_tree_block_info *)(ei + 1);
 659		*ptr = (unsigned long)(bi + 1);
 660	} else {
 661		*ptr = (unsigned long)(ei + 1);
 
 
 
 
 
 
 
 662	}
 663	*end = (unsigned long)ei + item_size;
 664	return 0;
 665}
 666
 667/*
 668 * build backref tree for a given tree block. root of the backref tree
 669 * corresponds the tree block, leaves of the backref tree correspond
 670 * roots of b-trees that reference the tree block.
 671 *
 672 * the basic idea of this function is check backrefs of a given block
 673 * to find upper level blocks that reference the block, and then check
 674 * backrefs of these upper level blocks recursively. the recursion stop
 675 * when tree root is reached or backrefs for the block is cached.
 676 *
 677 * NOTE: if we find backrefs for a block are cached, we know backrefs
 678 * for all upper level blocks that directly/indirectly reference the
 679 * block are also cached.
 680 */
 681static noinline_for_stack
 682struct backref_node *build_backref_tree(struct reloc_control *rc,
 683					struct btrfs_key *node_key,
 684					int level, u64 bytenr)
 685{
 686	struct backref_cache *cache = &rc->backref_cache;
 687	struct btrfs_path *path1;
 688	struct btrfs_path *path2;
 689	struct extent_buffer *eb;
 690	struct btrfs_root *root;
 691	struct backref_node *cur;
 692	struct backref_node *upper;
 693	struct backref_node *lower;
 694	struct backref_node *node = NULL;
 695	struct backref_node *exist = NULL;
 696	struct backref_edge *edge;
 697	struct rb_node *rb_node;
 698	struct btrfs_key key;
 699	unsigned long end;
 700	unsigned long ptr;
 701	LIST_HEAD(list);
 702	LIST_HEAD(useless);
 703	int cowonly;
 704	int ret;
 705	int err = 0;
 706	bool need_check = true;
 707
 708	path1 = btrfs_alloc_path();
 709	path2 = btrfs_alloc_path();
 710	if (!path1 || !path2) {
 711		err = -ENOMEM;
 
 
 712		goto out;
 713	}
 714	path1->reada = READA_FORWARD;
 715	path2->reada = READA_FORWARD;
 716
 717	node = alloc_backref_node(cache);
 718	if (!node) {
 719		err = -ENOMEM;
 720		goto out;
 721	}
 722
 723	node->bytenr = bytenr;
 724	node->level = level;
 725	node->lowest = 1;
 726	cur = node;
 727again:
 728	end = 0;
 729	ptr = 0;
 730	key.objectid = cur->bytenr;
 731	key.type = BTRFS_METADATA_ITEM_KEY;
 732	key.offset = (u64)-1;
 733
 734	path1->search_commit_root = 1;
 735	path1->skip_locking = 1;
 736	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
 737				0, 0);
 738	if (ret < 0) {
 739		err = ret;
 740		goto out;
 741	}
 742	ASSERT(ret);
 743	ASSERT(path1->slots[0]);
 744
 745	path1->slots[0]--;
 746
 747	WARN_ON(cur->checked);
 748	if (!list_empty(&cur->upper)) {
 749		/*
 750		 * the backref was added previously when processing
 751		 * backref of type BTRFS_TREE_BLOCK_REF_KEY
 752		 */
 753		ASSERT(list_is_singular(&cur->upper));
 754		edge = list_entry(cur->upper.next, struct backref_edge,
 755				  list[LOWER]);
 756		ASSERT(list_empty(&edge->list[UPPER]));
 757		exist = edge->node[UPPER];
 758		/*
 759		 * add the upper level block to pending list if we need
 760		 * check its backrefs
 761		 */
 762		if (!exist->checked)
 763			list_add_tail(&edge->list[UPPER], &list);
 764	} else {
 765		exist = NULL;
 766	}
 767
 768	while (1) {
 769		cond_resched();
 770		eb = path1->nodes[0];
 771
 772		if (ptr >= end) {
 773			if (path1->slots[0] >= btrfs_header_nritems(eb)) {
 774				ret = btrfs_next_leaf(rc->extent_root, path1);
 775				if (ret < 0) {
 776					err = ret;
 777					goto out;
 778				}
 779				if (ret > 0)
 780					break;
 781				eb = path1->nodes[0];
 782			}
 783
 784			btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
 785			if (key.objectid != cur->bytenr) {
 786				WARN_ON(exist);
 787				break;
 788			}
 789
 790			if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 791			    key.type == BTRFS_METADATA_ITEM_KEY) {
 792				ret = find_inline_backref(eb, path1->slots[0],
 793							  &ptr, &end);
 794				if (ret)
 795					goto next;
 796			}
 797		}
 798
 799		if (ptr < end) {
 800			/* update key for inline back ref */
 801			struct btrfs_extent_inline_ref *iref;
 802			iref = (struct btrfs_extent_inline_ref *)ptr;
 803			key.type = btrfs_extent_inline_ref_type(eb, iref);
 804			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
 805			WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
 806				key.type != BTRFS_SHARED_BLOCK_REF_KEY);
 807		}
 808
 809		if (exist &&
 810		    ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
 811		      exist->owner == key.offset) ||
 812		     (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
 813		      exist->bytenr == key.offset))) {
 814			exist = NULL;
 815			goto next;
 816		}
 817
 818#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 819		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
 820		    key.type == BTRFS_EXTENT_REF_V0_KEY) {
 821			if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 822				struct btrfs_extent_ref_v0 *ref0;
 823				ref0 = btrfs_item_ptr(eb, path1->slots[0],
 824						struct btrfs_extent_ref_v0);
 825				if (key.objectid == key.offset) {
 826					root = find_tree_root(rc, eb, ref0);
 827					if (root && !should_ignore_root(root))
 828						cur->root = root;
 829					else
 830						list_add(&cur->list, &useless);
 831					break;
 832				}
 833				if (is_cowonly_root(btrfs_ref_root_v0(eb,
 834								      ref0)))
 835					cur->cowonly = 1;
 836			}
 837#else
 838		ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY);
 839		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
 840#endif
 841			if (key.objectid == key.offset) {
 842				/*
 843				 * only root blocks of reloc trees use
 844				 * backref of this type.
 845				 */
 846				root = find_reloc_root(rc, cur->bytenr);
 847				ASSERT(root);
 848				cur->root = root;
 849				break;
 850			}
 851
 852			edge = alloc_backref_edge(cache);
 853			if (!edge) {
 854				err = -ENOMEM;
 855				goto out;
 856			}
 857			rb_node = tree_search(&cache->rb_root, key.offset);
 858			if (!rb_node) {
 859				upper = alloc_backref_node(cache);
 860				if (!upper) {
 861					free_backref_edge(cache, edge);
 862					err = -ENOMEM;
 863					goto out;
 864				}
 865				upper->bytenr = key.offset;
 866				upper->level = cur->level + 1;
 867				/*
 868				 *  backrefs for the upper level block isn't
 869				 *  cached, add the block to pending list
 870				 */
 871				list_add_tail(&edge->list[UPPER], &list);
 872			} else {
 873				upper = rb_entry(rb_node, struct backref_node,
 874						 rb_node);
 875				ASSERT(upper->checked);
 876				INIT_LIST_HEAD(&edge->list[UPPER]);
 877			}
 878			list_add_tail(&edge->list[LOWER], &cur->upper);
 879			edge->node[LOWER] = cur;
 880			edge->node[UPPER] = upper;
 881
 882			goto next;
 883		} else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
 884			goto next;
 885		}
 886
 887		/* key.type == BTRFS_TREE_BLOCK_REF_KEY */
 888		root = read_fs_root(rc->extent_root->fs_info, key.offset);
 889		if (IS_ERR(root)) {
 890			err = PTR_ERR(root);
 891			goto out;
 892		}
 893
 894		if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 895			cur->cowonly = 1;
 896
 897		if (btrfs_root_level(&root->root_item) == cur->level) {
 898			/* tree root */
 899			ASSERT(btrfs_root_bytenr(&root->root_item) ==
 900			       cur->bytenr);
 901			if (should_ignore_root(root))
 902				list_add(&cur->list, &useless);
 903			else
 904				cur->root = root;
 905			break;
 906		}
 907
 908		level = cur->level + 1;
 909
 
 
 910		/*
 911		 * searching the tree to find upper level blocks
 912		 * reference the block.
 913		 */
 914		path2->search_commit_root = 1;
 915		path2->skip_locking = 1;
 916		path2->lowest_level = level;
 917		ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
 918		path2->lowest_level = 0;
 919		if (ret < 0) {
 920			err = ret;
 921			goto out;
 922		}
 923		if (ret > 0 && path2->slots[level] > 0)
 924			path2->slots[level]--;
 925
 926		eb = path2->nodes[level];
 927		if (btrfs_node_blockptr(eb, path2->slots[level]) !=
 928		    cur->bytenr) {
 929			btrfs_err(root->fs_info,
 930	"couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
 931				  cur->bytenr, level - 1, root->objectid,
 932				  node_key->objectid, node_key->type,
 933				  node_key->offset);
 934			err = -ENOENT;
 935			goto out;
 936		}
 937		lower = cur;
 938		need_check = true;
 939		for (; level < BTRFS_MAX_LEVEL; level++) {
 940			if (!path2->nodes[level]) {
 941				ASSERT(btrfs_root_bytenr(&root->root_item) ==
 942				       lower->bytenr);
 943				if (should_ignore_root(root))
 944					list_add(&lower->list, &useless);
 945				else
 946					lower->root = root;
 947				break;
 948			}
 949
 950			edge = alloc_backref_edge(cache);
 951			if (!edge) {
 952				err = -ENOMEM;
 953				goto out;
 954			}
 955
 956			eb = path2->nodes[level];
 957			rb_node = tree_search(&cache->rb_root, eb->start);
 958			if (!rb_node) {
 959				upper = alloc_backref_node(cache);
 960				if (!upper) {
 961					free_backref_edge(cache, edge);
 962					err = -ENOMEM;
 963					goto out;
 964				}
 965				upper->bytenr = eb->start;
 966				upper->owner = btrfs_header_owner(eb);
 967				upper->level = lower->level + 1;
 968				if (!test_bit(BTRFS_ROOT_REF_COWS,
 969					      &root->state))
 970					upper->cowonly = 1;
 971
 972				/*
 973				 * if we know the block isn't shared
 974				 * we can void checking its backrefs.
 975				 */
 976				if (btrfs_block_can_be_shared(root, eb))
 977					upper->checked = 0;
 978				else
 979					upper->checked = 1;
 980
 981				/*
 982				 * add the block to pending list if we
 983				 * need check its backrefs, we only do this once
 984				 * while walking up a tree as we will catch
 985				 * anything else later on.
 986				 */
 987				if (!upper->checked && need_check) {
 988					need_check = false;
 989					list_add_tail(&edge->list[UPPER],
 990						      &list);
 991				} else {
 992					if (upper->checked)
 993						need_check = true;
 994					INIT_LIST_HEAD(&edge->list[UPPER]);
 995				}
 996			} else {
 997				upper = rb_entry(rb_node, struct backref_node,
 998						 rb_node);
 999				ASSERT(upper->checked);
1000				INIT_LIST_HEAD(&edge->list[UPPER]);
1001				if (!upper->owner)
1002					upper->owner = btrfs_header_owner(eb);
1003			}
1004			list_add_tail(&edge->list[LOWER], &lower->upper);
1005			edge->node[LOWER] = lower;
1006			edge->node[UPPER] = upper;
1007
1008			if (rb_node)
1009				break;
1010			lower = upper;
1011			upper = NULL;
1012		}
1013		btrfs_release_path(path2);
1014next:
1015		if (ptr < end) {
1016			ptr += btrfs_extent_inline_ref_size(key.type);
1017			if (ptr >= end) {
1018				WARN_ON(ptr > end);
1019				ptr = 0;
1020				end = 0;
1021			}
1022		}
1023		if (ptr >= end)
1024			path1->slots[0]++;
1025	}
1026	btrfs_release_path(path1);
1027
1028	cur->checked = 1;
1029	WARN_ON(exist);
1030
1031	/* the pending list isn't empty, take the first block to process */
1032	if (!list_empty(&list)) {
1033		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1034		list_del_init(&edge->list[UPPER]);
1035		cur = edge->node[UPPER];
1036		goto again;
1037	}
1038
1039	/*
1040	 * everything goes well, connect backref nodes and insert backref nodes
1041	 * into the cache.
1042	 */
1043	ASSERT(node->checked);
1044	cowonly = node->cowonly;
1045	if (!cowonly) {
1046		rb_node = tree_insert(&cache->rb_root, node->bytenr,
1047				      &node->rb_node);
1048		if (rb_node)
1049			backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1050		list_add_tail(&node->lower, &cache->leaves);
1051	}
1052
1053	list_for_each_entry(edge, &node->upper, list[LOWER])
1054		list_add_tail(&edge->list[UPPER], &list);
1055
1056	while (!list_empty(&list)) {
1057		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1058		list_del_init(&edge->list[UPPER]);
1059		upper = edge->node[UPPER];
1060		if (upper->detached) {
1061			list_del(&edge->list[LOWER]);
1062			lower = edge->node[LOWER];
1063			free_backref_edge(cache, edge);
1064			if (list_empty(&lower->upper))
1065				list_add(&lower->list, &useless);
1066			continue;
1067		}
1068
1069		if (!RB_EMPTY_NODE(&upper->rb_node)) {
1070			if (upper->lowest) {
1071				list_del_init(&upper->lower);
1072				upper->lowest = 0;
1073			}
1074
1075			list_add_tail(&edge->list[UPPER], &upper->lower);
1076			continue;
1077		}
1078
1079		if (!upper->checked) {
1080			/*
1081			 * Still want to blow up for developers since this is a
1082			 * logic bug.
1083			 */
1084			ASSERT(0);
1085			err = -EINVAL;
1086			goto out;
1087		}
1088		if (cowonly != upper->cowonly) {
1089			ASSERT(0);
1090			err = -EINVAL;
1091			goto out;
1092		}
 
1093
1094		if (!cowonly) {
1095			rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1096					      &upper->rb_node);
1097			if (rb_node)
1098				backref_tree_panic(rb_node, -EEXIST,
1099						   upper->bytenr);
1100		}
1101
1102		list_add_tail(&edge->list[UPPER], &upper->lower);
1103
1104		list_for_each_entry(edge, &upper->upper, list[LOWER])
1105			list_add_tail(&edge->list[UPPER], &list);
1106	}
1107	/*
1108	 * process useless backref nodes. backref nodes for tree leaves
1109	 * are deleted from the cache. backref nodes for upper level
1110	 * tree blocks are left in the cache to avoid unnecessary backref
1111	 * lookup.
1112	 */
1113	while (!list_empty(&useless)) {
1114		upper = list_entry(useless.next, struct backref_node, list);
1115		list_del_init(&upper->list);
1116		ASSERT(list_empty(&upper->upper));
1117		if (upper == node)
1118			node = NULL;
1119		if (upper->lowest) {
1120			list_del_init(&upper->lower);
1121			upper->lowest = 0;
1122		}
1123		while (!list_empty(&upper->lower)) {
1124			edge = list_entry(upper->lower.next,
1125					  struct backref_edge, list[UPPER]);
1126			list_del(&edge->list[UPPER]);
1127			list_del(&edge->list[LOWER]);
1128			lower = edge->node[LOWER];
1129			free_backref_edge(cache, edge);
1130
1131			if (list_empty(&lower->upper))
1132				list_add(&lower->list, &useless);
1133		}
1134		__mark_block_processed(rc, upper);
1135		if (upper->level > 0) {
1136			list_add(&upper->list, &cache->detached);
1137			upper->detached = 1;
1138		} else {
1139			rb_erase(&upper->rb_node, &cache->rb_root);
1140			free_backref_node(cache, upper);
1141		}
1142	}
1143out:
1144	btrfs_free_path(path1);
1145	btrfs_free_path(path2);
1146	if (err) {
1147		while (!list_empty(&useless)) {
1148			lower = list_entry(useless.next,
1149					   struct backref_node, list);
1150			list_del_init(&lower->list);
1151		}
1152		while (!list_empty(&list)) {
1153			edge = list_first_entry(&list, struct backref_edge,
1154						list[UPPER]);
1155			list_del(&edge->list[UPPER]);
1156			list_del(&edge->list[LOWER]);
1157			lower = edge->node[LOWER];
1158			upper = edge->node[UPPER];
1159			free_backref_edge(cache, edge);
1160
1161			/*
1162			 * Lower is no longer linked to any upper backref nodes
1163			 * and isn't in the cache, we can free it ourselves.
1164			 */
1165			if (list_empty(&lower->upper) &&
1166			    RB_EMPTY_NODE(&lower->rb_node))
1167				list_add(&lower->list, &useless);
1168
1169			if (!RB_EMPTY_NODE(&upper->rb_node))
1170				continue;
1171
1172			/* Add this guy's upper edges to the list to process */
1173			list_for_each_entry(edge, &upper->upper, list[LOWER])
1174				list_add_tail(&edge->list[UPPER], &list);
1175			if (list_empty(&upper->upper))
1176				list_add(&upper->list, &useless);
1177		}
1178
1179		while (!list_empty(&useless)) {
1180			lower = list_entry(useless.next,
1181					   struct backref_node, list);
1182			list_del_init(&lower->list);
1183			if (lower == node)
1184				node = NULL;
1185			free_backref_node(cache, lower);
1186		}
1187
1188		free_backref_node(cache, node);
1189		return ERR_PTR(err);
1190	}
1191	ASSERT(!node || !node->detached);
 
 
1192	return node;
1193}
1194
1195/*
1196 * helper to add backref node for the newly created snapshot.
1197 * the backref node is created by cloning backref node that
1198 * corresponds to root of source tree
1199 */
1200static int clone_backref_node(struct btrfs_trans_handle *trans,
1201			      struct reloc_control *rc,
1202			      struct btrfs_root *src,
1203			      struct btrfs_root *dest)
1204{
1205	struct btrfs_root *reloc_root = src->reloc_root;
1206	struct backref_cache *cache = &rc->backref_cache;
1207	struct backref_node *node = NULL;
1208	struct backref_node *new_node;
1209	struct backref_edge *edge;
1210	struct backref_edge *new_edge;
1211	struct rb_node *rb_node;
1212
1213	if (cache->last_trans > 0)
1214		update_backref_cache(trans, cache);
1215
1216	rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1217	if (rb_node) {
1218		node = rb_entry(rb_node, struct backref_node, rb_node);
1219		if (node->detached)
1220			node = NULL;
1221		else
1222			BUG_ON(node->new_bytenr != reloc_root->node->start);
1223	}
1224
1225	if (!node) {
1226		rb_node = tree_search(&cache->rb_root,
1227				      reloc_root->commit_root->start);
1228		if (rb_node) {
1229			node = rb_entry(rb_node, struct backref_node,
1230					rb_node);
1231			BUG_ON(node->detached);
1232		}
1233	}
1234
1235	if (!node)
1236		return 0;
1237
1238	new_node = alloc_backref_node(cache);
 
1239	if (!new_node)
1240		return -ENOMEM;
1241
1242	new_node->bytenr = dest->node->start;
1243	new_node->level = node->level;
1244	new_node->lowest = node->lowest;
1245	new_node->checked = 1;
1246	new_node->root = dest;
 
1247
1248	if (!node->lowest) {
1249		list_for_each_entry(edge, &node->lower, list[UPPER]) {
1250			new_edge = alloc_backref_edge(cache);
1251			if (!new_edge)
1252				goto fail;
1253
1254			new_edge->node[UPPER] = new_node;
1255			new_edge->node[LOWER] = edge->node[LOWER];
1256			list_add_tail(&new_edge->list[UPPER],
1257				      &new_node->lower);
1258		}
1259	} else {
1260		list_add_tail(&new_node->lower, &cache->leaves);
1261	}
1262
1263	rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1264			      &new_node->rb_node);
1265	if (rb_node)
1266		backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1267
1268	if (!new_node->lowest) {
1269		list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1270			list_add_tail(&new_edge->list[LOWER],
1271				      &new_edge->node[LOWER]->upper);
1272		}
1273	}
1274	return 0;
1275fail:
1276	while (!list_empty(&new_node->lower)) {
1277		new_edge = list_entry(new_node->lower.next,
1278				      struct backref_edge, list[UPPER]);
1279		list_del(&new_edge->list[UPPER]);
1280		free_backref_edge(cache, new_edge);
1281	}
1282	free_backref_node(cache, new_node);
1283	return -ENOMEM;
1284}
1285
1286/*
1287 * helper to add 'address of tree root -> reloc tree' mapping
1288 */
1289static int __must_check __add_reloc_root(struct btrfs_root *root)
1290{
1291	struct btrfs_fs_info *fs_info = root->fs_info;
1292	struct rb_node *rb_node;
1293	struct mapping_node *node;
1294	struct reloc_control *rc = fs_info->reloc_ctl;
1295
1296	node = kmalloc(sizeof(*node), GFP_NOFS);
1297	if (!node)
1298		return -ENOMEM;
1299
1300	node->bytenr = root->node->start;
1301	node->data = root;
1302
1303	spin_lock(&rc->reloc_root_tree.lock);
1304	rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1305			      node->bytenr, &node->rb_node);
1306	spin_unlock(&rc->reloc_root_tree.lock);
1307	if (rb_node) {
1308		btrfs_panic(fs_info, -EEXIST,
1309			    "Duplicate root found for start=%llu while inserting into relocation tree",
1310			    node->bytenr);
1311		kfree(node);
1312		return -EEXIST;
1313	}
1314
1315	list_add_tail(&root->root_list, &rc->reloc_roots);
1316	return 0;
1317}
1318
1319/*
1320 * helper to delete the 'address of tree root -> reloc tree'
1321 * mapping
1322 */
1323static void __del_reloc_root(struct btrfs_root *root)
1324{
1325	struct btrfs_fs_info *fs_info = root->fs_info;
1326	struct rb_node *rb_node;
1327	struct mapping_node *node = NULL;
1328	struct reloc_control *rc = fs_info->reloc_ctl;
 
1329
1330	spin_lock(&rc->reloc_root_tree.lock);
1331	rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1332			      root->node->start);
1333	if (rb_node) {
1334		node = rb_entry(rb_node, struct mapping_node, rb_node);
1335		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 
 
 
 
 
1336	}
1337	spin_unlock(&rc->reloc_root_tree.lock);
1338
1339	if (!node)
1340		return;
1341	BUG_ON((struct btrfs_root *)node->data != root);
1342
 
 
 
 
 
 
 
 
1343	spin_lock(&fs_info->trans_lock);
1344	list_del_init(&root->root_list);
 
 
 
1345	spin_unlock(&fs_info->trans_lock);
 
 
1346	kfree(node);
1347}
1348
1349/*
1350 * helper to update the 'address of tree root -> reloc tree'
1351 * mapping
1352 */
1353static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1354{
1355	struct btrfs_fs_info *fs_info = root->fs_info;
1356	struct rb_node *rb_node;
1357	struct mapping_node *node = NULL;
1358	struct reloc_control *rc = fs_info->reloc_ctl;
1359
1360	spin_lock(&rc->reloc_root_tree.lock);
1361	rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1362			      root->node->start);
1363	if (rb_node) {
1364		node = rb_entry(rb_node, struct mapping_node, rb_node);
1365		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1366	}
1367	spin_unlock(&rc->reloc_root_tree.lock);
1368
1369	if (!node)
1370		return 0;
1371	BUG_ON((struct btrfs_root *)node->data != root);
1372
1373	spin_lock(&rc->reloc_root_tree.lock);
1374	node->bytenr = new_bytenr;
1375	rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1376			      node->bytenr, &node->rb_node);
1377	spin_unlock(&rc->reloc_root_tree.lock);
1378	if (rb_node)
1379		backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1380	return 0;
1381}
1382
1383static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1384					struct btrfs_root *root, u64 objectid)
1385{
1386	struct btrfs_fs_info *fs_info = root->fs_info;
1387	struct btrfs_root *reloc_root;
1388	struct extent_buffer *eb;
1389	struct btrfs_root_item *root_item;
1390	struct btrfs_key root_key;
1391	int ret;
 
1392
1393	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1394	BUG_ON(!root_item);
 
1395
1396	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1397	root_key.type = BTRFS_ROOT_ITEM_KEY;
1398	root_key.offset = objectid;
1399
1400	if (root->root_key.objectid == objectid) {
1401		u64 commit_root_gen;
1402
1403		/* called by btrfs_init_reloc_root */
1404		ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1405				      BTRFS_TREE_RELOC_OBJECTID);
1406		BUG_ON(ret);
 
 
1407		/*
1408		 * Set the last_snapshot field to the generation of the commit
1409		 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1410		 * correctly (returns true) when the relocation root is created
1411		 * either inside the critical section of a transaction commit
1412		 * (through transaction.c:qgroup_account_snapshot()) and when
1413		 * it's created before the transaction commit is started.
1414		 */
1415		commit_root_gen = btrfs_header_generation(root->commit_root);
1416		btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1417	} else {
1418		/*
1419		 * called by btrfs_reloc_post_snapshot_hook.
1420		 * the source tree is a reloc tree, all tree blocks
1421		 * modified after it was created have RELOC flag
1422		 * set in their headers. so it's OK to not update
1423		 * the 'last_snapshot'.
1424		 */
1425		ret = btrfs_copy_root(trans, root, root->node, &eb,
1426				      BTRFS_TREE_RELOC_OBJECTID);
1427		BUG_ON(ret);
 
1428	}
1429
 
 
 
 
 
 
1430	memcpy(root_item, &root->root_item, sizeof(*root_item));
1431	btrfs_set_root_bytenr(root_item, eb->start);
1432	btrfs_set_root_level(root_item, btrfs_header_level(eb));
1433	btrfs_set_root_generation(root_item, trans->transid);
1434
1435	if (root->root_key.objectid == objectid) {
1436		btrfs_set_root_refs(root_item, 0);
1437		memset(&root_item->drop_progress, 0,
1438		       sizeof(struct btrfs_disk_key));
1439		root_item->drop_level = 0;
1440	}
1441
1442	btrfs_tree_unlock(eb);
1443	free_extent_buffer(eb);
1444
1445	ret = btrfs_insert_root(trans, fs_info->tree_root,
1446				&root_key, root_item);
1447	BUG_ON(ret);
 
 
1448	kfree(root_item);
1449
1450	reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1451	BUG_ON(IS_ERR(reloc_root));
1452	reloc_root->last_trans = trans->transid;
 
 
 
 
1453	return reloc_root;
 
 
 
 
 
 
1454}
1455
1456/*
1457 * create reloc tree for a given fs tree. reloc tree is just a
1458 * snapshot of the fs tree with special root objectid.
 
 
 
1459 */
1460int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1461			  struct btrfs_root *root)
1462{
1463	struct btrfs_fs_info *fs_info = root->fs_info;
1464	struct btrfs_root *reloc_root;
1465	struct reloc_control *rc = fs_info->reloc_ctl;
1466	struct btrfs_block_rsv *rsv;
1467	int clear_rsv = 0;
1468	int ret;
1469
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1470	if (root->reloc_root) {
1471		reloc_root = root->reloc_root;
1472		reloc_root->last_trans = trans->transid;
1473		return 0;
1474	}
1475
1476	if (!rc || !rc->create_reloc_tree ||
1477	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
 
 
 
1478		return 0;
1479
1480	if (!trans->reloc_reserved) {
1481		rsv = trans->block_rsv;
1482		trans->block_rsv = rc->block_rsv;
1483		clear_rsv = 1;
1484	}
1485	reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1486	if (clear_rsv)
1487		trans->block_rsv = rsv;
 
 
1488
1489	ret = __add_reloc_root(reloc_root);
1490	BUG_ON(ret < 0);
1491	root->reloc_root = reloc_root;
 
 
 
 
 
1492	return 0;
1493}
1494
1495/*
1496 * update root item of reloc tree
1497 */
1498int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1499			    struct btrfs_root *root)
1500{
1501	struct btrfs_fs_info *fs_info = root->fs_info;
1502	struct btrfs_root *reloc_root;
1503	struct btrfs_root_item *root_item;
1504	int ret;
1505
1506	if (!root->reloc_root)
1507		goto out;
1508
1509	reloc_root = root->reloc_root;
1510	root_item = &reloc_root->root_item;
1511
1512	if (fs_info->reloc_ctl->merge_reloc_tree &&
 
 
 
 
 
 
 
 
1513	    btrfs_root_refs(root_item) == 0) {
1514		root->reloc_root = NULL;
 
 
 
 
 
1515		__del_reloc_root(reloc_root);
1516	}
1517
1518	if (reloc_root->commit_root != reloc_root->node) {
 
1519		btrfs_set_root_node(root_item, reloc_root->node);
1520		free_extent_buffer(reloc_root->commit_root);
1521		reloc_root->commit_root = btrfs_root_node(reloc_root);
1522	}
1523
1524	ret = btrfs_update_root(trans, fs_info->tree_root,
1525				&reloc_root->root_key, root_item);
1526	BUG_ON(ret);
1527
1528out:
1529	return 0;
1530}
1531
1532/*
1533 * helper to find first cached inode with inode number >= objectid
1534 * in a subvolume
1535 */
1536static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1537{
1538	struct rb_node *node;
1539	struct rb_node *prev;
1540	struct btrfs_inode *entry;
1541	struct inode *inode;
1542
1543	spin_lock(&root->inode_lock);
1544again:
1545	node = root->inode_tree.rb_node;
1546	prev = NULL;
1547	while (node) {
1548		prev = node;
1549		entry = rb_entry(node, struct btrfs_inode, rb_node);
1550
1551		if (objectid < btrfs_ino(&entry->vfs_inode))
1552			node = node->rb_left;
1553		else if (objectid > btrfs_ino(&entry->vfs_inode))
1554			node = node->rb_right;
1555		else
1556			break;
1557	}
1558	if (!node) {
1559		while (prev) {
1560			entry = rb_entry(prev, struct btrfs_inode, rb_node);
1561			if (objectid <= btrfs_ino(&entry->vfs_inode)) {
1562				node = prev;
1563				break;
1564			}
1565			prev = rb_next(prev);
1566		}
1567	}
1568	while (node) {
1569		entry = rb_entry(node, struct btrfs_inode, rb_node);
1570		inode = igrab(&entry->vfs_inode);
1571		if (inode) {
1572			spin_unlock(&root->inode_lock);
1573			return inode;
1574		}
1575
1576		objectid = btrfs_ino(&entry->vfs_inode) + 1;
1577		if (cond_resched_lock(&root->inode_lock))
1578			goto again;
1579
1580		node = rb_next(node);
1581	}
1582	spin_unlock(&root->inode_lock);
1583	return NULL;
1584}
1585
1586static int in_block_group(u64 bytenr,
1587			  struct btrfs_block_group_cache *block_group)
1588{
1589	if (bytenr >= block_group->key.objectid &&
1590	    bytenr < block_group->key.objectid + block_group->key.offset)
1591		return 1;
1592	return 0;
1593}
1594
1595/*
1596 * get new location of data
1597 */
1598static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1599			    u64 bytenr, u64 num_bytes)
1600{
1601	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1602	struct btrfs_path *path;
1603	struct btrfs_file_extent_item *fi;
1604	struct extent_buffer *leaf;
1605	int ret;
1606
1607	path = btrfs_alloc_path();
1608	if (!path)
1609		return -ENOMEM;
1610
1611	bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1612	ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(reloc_inode),
1613				       bytenr, 0);
1614	if (ret < 0)
1615		goto out;
1616	if (ret > 0) {
1617		ret = -ENOENT;
1618		goto out;
1619	}
1620
1621	leaf = path->nodes[0];
1622	fi = btrfs_item_ptr(leaf, path->slots[0],
1623			    struct btrfs_file_extent_item);
1624
1625	BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1626	       btrfs_file_extent_compression(leaf, fi) ||
1627	       btrfs_file_extent_encryption(leaf, fi) ||
1628	       btrfs_file_extent_other_encoding(leaf, fi));
1629
1630	if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1631		ret = -EINVAL;
1632		goto out;
1633	}
1634
1635	*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1636	ret = 0;
1637out:
1638	btrfs_free_path(path);
1639	return ret;
1640}
1641
1642/*
1643 * update file extent items in the tree leaf to point to
1644 * the new locations.
1645 */
1646static noinline_for_stack
1647int replace_file_extents(struct btrfs_trans_handle *trans,
1648			 struct reloc_control *rc,
1649			 struct btrfs_root *root,
1650			 struct extent_buffer *leaf)
1651{
1652	struct btrfs_fs_info *fs_info = root->fs_info;
1653	struct btrfs_key key;
1654	struct btrfs_file_extent_item *fi;
1655	struct inode *inode = NULL;
1656	u64 parent;
1657	u64 bytenr;
1658	u64 new_bytenr = 0;
1659	u64 num_bytes;
1660	u64 end;
1661	u32 nritems;
1662	u32 i;
1663	int ret = 0;
1664	int first = 1;
1665	int dirty = 0;
1666
1667	if (rc->stage != UPDATE_DATA_PTRS)
1668		return 0;
1669
1670	/* reloc trees always use full backref */
1671	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1672		parent = leaf->start;
1673	else
1674		parent = 0;
1675
1676	nritems = btrfs_header_nritems(leaf);
1677	for (i = 0; i < nritems; i++) {
 
 
1678		cond_resched();
1679		btrfs_item_key_to_cpu(leaf, &key, i);
1680		if (key.type != BTRFS_EXTENT_DATA_KEY)
1681			continue;
1682		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1683		if (btrfs_file_extent_type(leaf, fi) ==
1684		    BTRFS_FILE_EXTENT_INLINE)
1685			continue;
1686		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1687		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1688		if (bytenr == 0)
1689			continue;
1690		if (!in_block_group(bytenr, rc->block_group))
 
1691			continue;
1692
1693		/*
1694		 * if we are modifying block in fs tree, wait for readpage
1695		 * to complete and drop the extent cache
1696		 */
1697		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1698			if (first) {
1699				inode = find_next_inode(root, key.objectid);
1700				first = 0;
1701			} else if (inode && btrfs_ino(inode) < key.objectid) {
1702				btrfs_add_delayed_iput(inode);
1703				inode = find_next_inode(root, key.objectid);
1704			}
1705			if (inode && btrfs_ino(inode) == key.objectid) {
 
 
1706				end = key.offset +
1707				      btrfs_file_extent_num_bytes(leaf, fi);
1708				WARN_ON(!IS_ALIGNED(key.offset,
1709						    fs_info->sectorsize));
1710				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1711				end--;
1712				ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1713						      key.offset, end);
1714				if (!ret)
1715					continue;
 
 
 
 
 
 
1716
1717				btrfs_drop_extent_cache(inode, key.offset, end,
1718							1);
1719				unlock_extent(&BTRFS_I(inode)->io_tree,
1720					      key.offset, end);
1721			}
1722		}
1723
1724		ret = get_new_location(rc->data_inode, &new_bytenr,
1725				       bytenr, num_bytes);
1726		if (ret) {
1727			/*
1728			 * Don't have to abort since we've not changed anything
1729			 * in the file extent yet.
1730			 */
1731			break;
1732		}
1733
1734		btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1735		dirty = 1;
1736
1737		key.offset -= btrfs_file_extent_offset(leaf, fi);
1738		ret = btrfs_inc_extent_ref(trans, fs_info, new_bytenr,
1739					   num_bytes, parent,
1740					   btrfs_header_owner(leaf),
1741					   key.objectid, key.offset);
 
 
 
 
 
1742		if (ret) {
1743			btrfs_abort_transaction(trans, ret);
1744			break;
1745		}
1746
1747		ret = btrfs_free_extent(trans, fs_info, bytenr, num_bytes,
1748					parent, btrfs_header_owner(leaf),
1749					key.objectid, key.offset);
 
 
 
 
 
 
1750		if (ret) {
1751			btrfs_abort_transaction(trans, ret);
1752			break;
1753		}
1754	}
1755	if (dirty)
1756		btrfs_mark_buffer_dirty(leaf);
1757	if (inode)
1758		btrfs_add_delayed_iput(inode);
1759	return ret;
1760}
1761
1762static noinline_for_stack
1763int memcmp_node_keys(struct extent_buffer *eb, int slot,
1764		     struct btrfs_path *path, int level)
1765{
1766	struct btrfs_disk_key key1;
1767	struct btrfs_disk_key key2;
1768	btrfs_node_key(eb, &key1, slot);
1769	btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1770	return memcmp(&key1, &key2, sizeof(key1));
1771}
1772
1773/*
1774 * try to replace tree blocks in fs tree with the new blocks
1775 * in reloc tree. tree blocks haven't been modified since the
1776 * reloc tree was create can be replaced.
1777 *
1778 * if a block was replaced, level of the block + 1 is returned.
1779 * if no block got replaced, 0 is returned. if there are other
1780 * errors, a negative error number is returned.
1781 */
1782static noinline_for_stack
1783int replace_path(struct btrfs_trans_handle *trans,
1784		 struct btrfs_root *dest, struct btrfs_root *src,
1785		 struct btrfs_path *path, struct btrfs_key *next_key,
1786		 int lowest_level, int max_level)
1787{
1788	struct btrfs_fs_info *fs_info = dest->fs_info;
1789	struct extent_buffer *eb;
1790	struct extent_buffer *parent;
 
1791	struct btrfs_key key;
1792	u64 old_bytenr;
1793	u64 new_bytenr;
1794	u64 old_ptr_gen;
1795	u64 new_ptr_gen;
1796	u64 last_snapshot;
1797	u32 blocksize;
1798	int cow = 0;
1799	int level;
1800	int ret;
1801	int slot;
1802
1803	BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1804	BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1805
1806	last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1807again:
1808	slot = path->slots[lowest_level];
1809	btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1810
1811	eb = btrfs_lock_root_node(dest);
1812	btrfs_set_lock_blocking(eb);
1813	level = btrfs_header_level(eb);
1814
1815	if (level < lowest_level) {
1816		btrfs_tree_unlock(eb);
1817		free_extent_buffer(eb);
1818		return 0;
1819	}
1820
1821	if (cow) {
1822		ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1823		BUG_ON(ret);
 
 
 
 
 
1824	}
1825	btrfs_set_lock_blocking(eb);
1826
1827	if (next_key) {
1828		next_key->objectid = (u64)-1;
1829		next_key->type = (u8)-1;
1830		next_key->offset = (u64)-1;
1831	}
1832
1833	parent = eb;
1834	while (1) {
1835		level = btrfs_header_level(parent);
1836		BUG_ON(level < lowest_level);
1837
1838		ret = btrfs_bin_search(parent, &key, level, &slot);
 
 
1839		if (ret && slot > 0)
1840			slot--;
1841
1842		if (next_key && slot + 1 < btrfs_header_nritems(parent))
1843			btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1844
1845		old_bytenr = btrfs_node_blockptr(parent, slot);
1846		blocksize = fs_info->nodesize;
1847		old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1848
1849		if (level <= max_level) {
1850			eb = path->nodes[level];
1851			new_bytenr = btrfs_node_blockptr(eb,
1852							path->slots[level]);
1853			new_ptr_gen = btrfs_node_ptr_generation(eb,
1854							path->slots[level]);
1855		} else {
1856			new_bytenr = 0;
1857			new_ptr_gen = 0;
1858		}
1859
1860		if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1861			ret = level;
1862			break;
1863		}
1864
1865		if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1866		    memcmp_node_keys(parent, slot, path, level)) {
1867			if (level <= lowest_level) {
1868				ret = 0;
1869				break;
1870			}
1871
1872			eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen);
1873			if (IS_ERR(eb)) {
1874				ret = PTR_ERR(eb);
1875				break;
1876			} else if (!extent_buffer_uptodate(eb)) {
1877				ret = -EIO;
1878				free_extent_buffer(eb);
1879				break;
1880			}
1881			btrfs_tree_lock(eb);
1882			if (cow) {
1883				ret = btrfs_cow_block(trans, dest, eb, parent,
1884						      slot, &eb);
1885				BUG_ON(ret);
 
 
 
 
 
1886			}
1887			btrfs_set_lock_blocking(eb);
1888
1889			btrfs_tree_unlock(parent);
1890			free_extent_buffer(parent);
1891
1892			parent = eb;
1893			continue;
1894		}
1895
1896		if (!cow) {
1897			btrfs_tree_unlock(parent);
1898			free_extent_buffer(parent);
1899			cow = 1;
1900			goto again;
1901		}
1902
1903		btrfs_node_key_to_cpu(path->nodes[level], &key,
1904				      path->slots[level]);
1905		btrfs_release_path(path);
1906
1907		path->lowest_level = level;
 
1908		ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
 
1909		path->lowest_level = 0;
1910		BUG_ON(ret);
 
 
 
 
1911
1912		/*
1913		 * Info qgroup to trace both subtrees.
1914		 *
1915		 * We must trace both trees.
1916		 * 1) Tree reloc subtree
1917		 *    If not traced, we will leak data numbers
1918		 * 2) Fs subtree
1919		 *    If not traced, we will double count old data
1920		 *    and tree block numbers, if current trans doesn't free
1921		 *    data reloc tree inode.
 
 
 
1922		 */
1923		ret = btrfs_qgroup_trace_subtree(trans, src, parent,
1924				btrfs_header_generation(parent),
1925				btrfs_header_level(parent));
 
1926		if (ret < 0)
1927			break;
1928		ret = btrfs_qgroup_trace_subtree(trans, dest,
1929				path->nodes[level],
1930				btrfs_header_generation(path->nodes[level]),
1931				btrfs_header_level(path->nodes[level]));
1932		if (ret < 0)
1933			break;
1934
1935		/*
1936		 * swap blocks in fs tree and reloc tree.
1937		 */
1938		btrfs_set_node_blockptr(parent, slot, new_bytenr);
1939		btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1940		btrfs_mark_buffer_dirty(parent);
1941
1942		btrfs_set_node_blockptr(path->nodes[level],
1943					path->slots[level], old_bytenr);
1944		btrfs_set_node_ptr_generation(path->nodes[level],
1945					      path->slots[level], old_ptr_gen);
1946		btrfs_mark_buffer_dirty(path->nodes[level]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1947
1948		ret = btrfs_inc_extent_ref(trans, fs_info, old_bytenr,
1949					blocksize, path->nodes[level]->start,
1950					src->root_key.objectid, level - 1, 0);
1951		BUG_ON(ret);
1952		ret = btrfs_inc_extent_ref(trans, fs_info, new_bytenr,
1953					blocksize, 0, dest->root_key.objectid,
1954					level - 1, 0);
1955		BUG_ON(ret);
1956
1957		ret = btrfs_free_extent(trans, fs_info, new_bytenr, blocksize,
1958					path->nodes[level]->start,
1959					src->root_key.objectid, level - 1, 0);
1960		BUG_ON(ret);
1961
1962		ret = btrfs_free_extent(trans, fs_info, old_bytenr, blocksize,
1963					0, dest->root_key.objectid, level - 1,
1964					0);
1965		BUG_ON(ret);
 
 
 
 
 
 
 
 
 
1966
1967		btrfs_unlock_up_safe(path, 0);
1968
1969		ret = level;
1970		break;
1971	}
1972	btrfs_tree_unlock(parent);
1973	free_extent_buffer(parent);
1974	return ret;
1975}
1976
1977/*
1978 * helper to find next relocated block in reloc tree
1979 */
1980static noinline_for_stack
1981int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1982		       int *level)
1983{
1984	struct extent_buffer *eb;
1985	int i;
1986	u64 last_snapshot;
1987	u32 nritems;
1988
1989	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1990
1991	for (i = 0; i < *level; i++) {
1992		free_extent_buffer(path->nodes[i]);
1993		path->nodes[i] = NULL;
1994	}
1995
1996	for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1997		eb = path->nodes[i];
1998		nritems = btrfs_header_nritems(eb);
1999		while (path->slots[i] + 1 < nritems) {
2000			path->slots[i]++;
2001			if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
2002			    last_snapshot)
2003				continue;
2004
2005			*level = i;
2006			return 0;
2007		}
2008		free_extent_buffer(path->nodes[i]);
2009		path->nodes[i] = NULL;
2010	}
2011	return 1;
2012}
2013
2014/*
2015 * walk down reloc tree to find relocated block of lowest level
2016 */
2017static noinline_for_stack
2018int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
2019			 int *level)
2020{
2021	struct btrfs_fs_info *fs_info = root->fs_info;
2022	struct extent_buffer *eb = NULL;
2023	int i;
2024	u64 bytenr;
2025	u64 ptr_gen = 0;
2026	u64 last_snapshot;
2027	u32 nritems;
2028
2029	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2030
2031	for (i = *level; i > 0; i--) {
2032		eb = path->nodes[i];
2033		nritems = btrfs_header_nritems(eb);
2034		while (path->slots[i] < nritems) {
2035			ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
2036			if (ptr_gen > last_snapshot)
2037				break;
2038			path->slots[i]++;
2039		}
2040		if (path->slots[i] >= nritems) {
2041			if (i == *level)
2042				break;
2043			*level = i + 1;
2044			return 0;
2045		}
2046		if (i == 1) {
2047			*level = i;
2048			return 0;
2049		}
2050
2051		bytenr = btrfs_node_blockptr(eb, path->slots[i]);
2052		eb = read_tree_block(fs_info, bytenr, ptr_gen);
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(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(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 * merge the relocated tree blocks in reloc tree with corresponding
2159 * fs tree.
2160 */
2161static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2162					       struct btrfs_root *root)
2163{
2164	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2165	LIST_HEAD(inode_list);
2166	struct btrfs_key key;
2167	struct btrfs_key next_key;
2168	struct btrfs_trans_handle *trans = NULL;
2169	struct btrfs_root *reloc_root;
2170	struct btrfs_root_item *root_item;
2171	struct btrfs_path *path;
2172	struct extent_buffer *leaf;
 
2173	int level;
2174	int max_level;
2175	int replaced = 0;
2176	int ret;
2177	int err = 0;
2178	u32 min_reserved;
2179
2180	path = btrfs_alloc_path();
2181	if (!path)
2182		return -ENOMEM;
2183	path->reada = READA_FORWARD;
2184
2185	reloc_root = root->reloc_root;
2186	root_item = &reloc_root->root_item;
2187
2188	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2189		level = btrfs_root_level(root_item);
2190		extent_buffer_get(reloc_root->node);
2191		path->nodes[level] = reloc_root->node;
2192		path->slots[level] = 0;
2193	} else {
2194		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2195
2196		level = root_item->drop_level;
2197		BUG_ON(level == 0);
2198		path->lowest_level = level;
2199		ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2200		path->lowest_level = 0;
2201		if (ret < 0) {
2202			btrfs_free_path(path);
2203			return ret;
2204		}
2205
2206		btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2207				      path->slots[level]);
2208		WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2209
2210		btrfs_unlock_up_safe(path, 0);
2211	}
2212
2213	min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
 
 
 
 
 
 
 
 
 
2214	memset(&next_key, 0, sizeof(next_key));
2215
2216	while (1) {
2217		ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2218					     BTRFS_RESERVE_FLUSH_ALL);
2219		if (ret) {
2220			err = ret;
2221			goto out;
2222		}
2223		trans = btrfs_start_transaction(root, 0);
2224		if (IS_ERR(trans)) {
2225			err = PTR_ERR(trans);
2226			trans = NULL;
2227			goto out;
2228		}
 
 
 
 
 
 
 
 
 
 
 
 
2229		trans->block_rsv = rc->block_rsv;
2230
2231		replaced = 0;
2232		max_level = level;
2233
2234		ret = walk_down_reloc_tree(reloc_root, path, &level);
2235		if (ret < 0) {
2236			err = ret;
2237			goto out;
2238		}
2239		if (ret > 0)
2240			break;
2241
2242		if (!find_next_key(path, level, &key) &&
2243		    btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2244			ret = 0;
2245		} else {
2246			ret = replace_path(trans, root, reloc_root, path,
2247					   &next_key, level, max_level);
2248		}
2249		if (ret < 0) {
2250			err = ret;
2251			goto out;
2252		}
2253
2254		if (ret > 0) {
2255			level = ret;
2256			btrfs_node_key_to_cpu(path->nodes[level], &key,
2257					      path->slots[level]);
2258			replaced = 1;
2259		}
2260
2261		ret = walk_up_reloc_tree(reloc_root, path, &level);
2262		if (ret > 0)
2263			break;
2264
2265		BUG_ON(level == 0);
2266		/*
2267		 * save the merging progress in the drop_progress.
2268		 * this is OK since root refs == 1 in this case.
2269		 */
2270		btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2271			       path->slots[level]);
2272		root_item->drop_level = level;
2273
2274		btrfs_end_transaction_throttle(trans);
2275		trans = NULL;
2276
2277		btrfs_btree_balance_dirty(fs_info);
2278
2279		if (replaced && rc->stage == UPDATE_DATA_PTRS)
2280			invalidate_extent_cache(root, &key, &next_key);
2281	}
2282
2283	/*
2284	 * handle the case only one block in the fs tree need to be
2285	 * relocated and the block is tree root.
2286	 */
2287	leaf = btrfs_lock_root_node(root);
2288	ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
 
2289	btrfs_tree_unlock(leaf);
2290	free_extent_buffer(leaf);
2291	if (ret < 0)
2292		err = ret;
2293out:
2294	btrfs_free_path(path);
2295
2296	if (err == 0) {
2297		memset(&root_item->drop_progress, 0,
2298		       sizeof(root_item->drop_progress));
2299		root_item->drop_level = 0;
2300		btrfs_set_root_refs(root_item, 0);
2301		btrfs_update_reloc_root(trans, root);
2302	}
2303
2304	if (trans)
2305		btrfs_end_transaction_throttle(trans);
2306
2307	btrfs_btree_balance_dirty(fs_info);
2308
2309	if (replaced && rc->stage == UPDATE_DATA_PTRS)
2310		invalidate_extent_cache(root, &key, &next_key);
2311
2312	return err;
2313}
2314
2315static noinline_for_stack
2316int prepare_to_merge(struct reloc_control *rc, int err)
2317{
2318	struct btrfs_root *root = rc->extent_root;
2319	struct btrfs_fs_info *fs_info = root->fs_info;
2320	struct btrfs_root *reloc_root;
2321	struct btrfs_trans_handle *trans;
2322	LIST_HEAD(reloc_roots);
2323	u64 num_bytes = 0;
2324	int ret;
2325
2326	mutex_lock(&fs_info->reloc_mutex);
2327	rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2328	rc->merging_rsv_size += rc->nodes_relocated * 2;
2329	mutex_unlock(&fs_info->reloc_mutex);
2330
2331again:
2332	if (!err) {
2333		num_bytes = rc->merging_rsv_size;
2334		ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2335					  BTRFS_RESERVE_FLUSH_ALL);
2336		if (ret)
2337			err = ret;
2338	}
2339
2340	trans = btrfs_join_transaction(rc->extent_root);
2341	if (IS_ERR(trans)) {
2342		if (!err)
2343			btrfs_block_rsv_release(fs_info, rc->block_rsv,
2344						num_bytes);
2345		return PTR_ERR(trans);
2346	}
2347
2348	if (!err) {
2349		if (num_bytes != rc->merging_rsv_size) {
2350			btrfs_end_transaction(trans);
2351			btrfs_block_rsv_release(fs_info, rc->block_rsv,
2352						num_bytes);
2353			goto again;
2354		}
2355	}
2356
2357	rc->merge_reloc_tree = 1;
2358
2359	while (!list_empty(&rc->reloc_roots)) {
2360		reloc_root = list_entry(rc->reloc_roots.next,
2361					struct btrfs_root, root_list);
2362		list_del_init(&reloc_root->root_list);
2363
2364		root = read_fs_root(fs_info, reloc_root->root_key.offset);
2365		BUG_ON(IS_ERR(root));
2366		BUG_ON(root->reloc_root != reloc_root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2367
2368		/*
2369		 * set reference count to 1, so btrfs_recover_relocation
2370		 * knows it should resumes merging
2371		 */
2372		if (!err)
2373			btrfs_set_root_refs(&reloc_root->root_item, 1);
2374		btrfs_update_reloc_root(trans, root);
2375
 
 
 
 
2376		list_add(&reloc_root->root_list, &reloc_roots);
 
 
 
 
 
 
 
 
2377	}
2378
2379	list_splice(&reloc_roots, &rc->reloc_roots);
2380
2381	if (!err)
2382		btrfs_commit_transaction(trans);
2383	else
2384		btrfs_end_transaction(trans);
2385	return err;
2386}
2387
2388static noinline_for_stack
2389void free_reloc_roots(struct list_head *list)
2390{
2391	struct btrfs_root *reloc_root;
2392
2393	while (!list_empty(list)) {
2394		reloc_root = list_entry(list->next, struct btrfs_root,
2395					root_list);
2396		free_extent_buffer(reloc_root->node);
2397		free_extent_buffer(reloc_root->commit_root);
2398		reloc_root->node = NULL;
2399		reloc_root->commit_root = NULL;
2400		__del_reloc_root(reloc_root);
2401	}
2402}
2403
2404static noinline_for_stack
2405void merge_reloc_roots(struct reloc_control *rc)
2406{
2407	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2408	struct btrfs_root *root;
2409	struct btrfs_root *reloc_root;
2410	LIST_HEAD(reloc_roots);
2411	int found = 0;
2412	int ret = 0;
2413again:
2414	root = rc->extent_root;
2415
2416	/*
2417	 * this serializes us with btrfs_record_root_in_transaction,
2418	 * we have to make sure nobody is in the middle of
2419	 * adding their roots to the list while we are
2420	 * doing this splice
2421	 */
2422	mutex_lock(&fs_info->reloc_mutex);
2423	list_splice_init(&rc->reloc_roots, &reloc_roots);
2424	mutex_unlock(&fs_info->reloc_mutex);
2425
2426	while (!list_empty(&reloc_roots)) {
2427		found = 1;
2428		reloc_root = list_entry(reloc_roots.next,
2429					struct btrfs_root, root_list);
2430
 
 
2431		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2432			root = read_fs_root(fs_info,
2433					    reloc_root->root_key.offset);
2434			BUG_ON(IS_ERR(root));
2435			BUG_ON(root->reloc_root != reloc_root);
2436
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2437			ret = merge_reloc_root(rc, root);
 
2438			if (ret) {
2439				if (list_empty(&reloc_root->root_list))
2440					list_add_tail(&reloc_root->root_list,
2441						      &reloc_roots);
2442				goto out;
2443			}
2444		} else {
2445			list_del_init(&reloc_root->root_list);
2446		}
 
 
 
 
 
 
 
2447
2448		ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2449		if (ret < 0) {
2450			if (list_empty(&reloc_root->root_list))
2451				list_add_tail(&reloc_root->root_list,
2452					      &reloc_roots);
2453			goto out;
2454		}
2455	}
2456
2457	if (found) {
2458		found = 0;
2459		goto again;
2460	}
2461out:
2462	if (ret) {
2463		btrfs_handle_fs_error(fs_info, ret, NULL);
2464		if (!list_empty(&reloc_roots))
2465			free_reloc_roots(&reloc_roots);
2466
2467		/* new reloc root may be added */
2468		mutex_lock(&fs_info->reloc_mutex);
2469		list_splice_init(&rc->reloc_roots, &reloc_roots);
2470		mutex_unlock(&fs_info->reloc_mutex);
2471		if (!list_empty(&reloc_roots))
2472			free_reloc_roots(&reloc_roots);
2473	}
2474
2475	BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2476}
2477
2478static void free_block_list(struct rb_root *blocks)
2479{
2480	struct tree_block *block;
2481	struct rb_node *rb_node;
2482	while ((rb_node = rb_first(blocks))) {
2483		block = rb_entry(rb_node, struct tree_block, rb_node);
2484		rb_erase(rb_node, blocks);
2485		kfree(block);
2486	}
2487}
2488
2489static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2490				      struct btrfs_root *reloc_root)
2491{
2492	struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2493	struct btrfs_root *root;
 
2494
2495	if (reloc_root->last_trans == trans->transid)
2496		return 0;
2497
2498	root = read_fs_root(fs_info, reloc_root->root_key.offset);
2499	BUG_ON(IS_ERR(root));
2500	BUG_ON(root->reloc_root != reloc_root);
2501
2502	return btrfs_record_root_in_trans(trans, root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2503}
2504
2505static noinline_for_stack
2506struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2507				     struct reloc_control *rc,
2508				     struct backref_node *node,
2509				     struct backref_edge *edges[])
2510{
2511	struct backref_node *next;
2512	struct btrfs_root *root;
2513	int index = 0;
 
2514
2515	next = node;
2516	while (1) {
2517		cond_resched();
2518		next = walk_up_backref(next, edges, &index);
2519		root = next->root;
2520		BUG_ON(!root);
2521		BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
2522
2523		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2524			record_reloc_root_in_trans(trans, root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2525			break;
2526		}
2527
2528		btrfs_record_root_in_trans(trans, root);
 
 
2529		root = root->reloc_root;
2530
 
 
 
 
 
 
 
2531		if (next->new_bytenr != root->node->start) {
2532			BUG_ON(next->new_bytenr);
2533			BUG_ON(!list_empty(&next->list));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2534			next->new_bytenr = root->node->start;
2535			next->root = root;
 
 
2536			list_add_tail(&next->list,
2537				      &rc->backref_cache.changed);
2538			__mark_block_processed(rc, next);
2539			break;
2540		}
2541
2542		WARN_ON(1);
2543		root = NULL;
2544		next = walk_down_backref(edges, &index);
2545		if (!next || next->level <= node->level)
2546			break;
2547	}
2548	if (!root)
2549		return NULL;
 
 
 
 
 
 
2550
2551	next = node;
2552	/* setup backref node path for btrfs_reloc_cow_block */
2553	while (1) {
2554		rc->backref_cache.path[next->level] = next;
2555		if (--index < 0)
2556			break;
2557		next = edges[index]->node[UPPER];
2558	}
2559	return root;
2560}
2561
2562/*
2563 * select a tree root for relocation. return NULL if the block
2564 * is reference counted. we should use do_relocation() in this
2565 * case. return a tree root pointer if the block isn't reference
2566 * counted. return -ENOENT if the block is root of reloc tree.
 
 
 
2567 */
2568static noinline_for_stack
2569struct btrfs_root *select_one_root(struct backref_node *node)
2570{
2571	struct backref_node *next;
2572	struct btrfs_root *root;
2573	struct btrfs_root *fs_root = NULL;
2574	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2575	int index = 0;
2576
2577	next = node;
2578	while (1) {
2579		cond_resched();
2580		next = walk_up_backref(next, edges, &index);
2581		root = next->root;
2582		BUG_ON(!root);
2583
2584		/* no other choice for non-references counted tree */
2585		if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 
 
 
 
 
 
 
2586			return root;
2587
2588		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2589			fs_root = root;
2590
2591		if (next != node)
2592			return NULL;
2593
2594		next = walk_down_backref(edges, &index);
2595		if (!next || next->level <= node->level)
2596			break;
2597	}
2598
2599	if (!fs_root)
2600		return ERR_PTR(-ENOENT);
2601	return fs_root;
2602}
2603
2604static noinline_for_stack
2605u64 calcu_metadata_size(struct reloc_control *rc,
2606			struct backref_node *node, int reserve)
2607{
2608	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2609	struct backref_node *next = node;
2610	struct backref_edge *edge;
2611	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2612	u64 num_bytes = 0;
2613	int index = 0;
2614
2615	BUG_ON(reserve && node->processed);
2616
2617	while (next) {
2618		cond_resched();
2619		while (1) {
2620			if (next->processed && (reserve || next != node))
2621				break;
2622
2623			num_bytes += fs_info->nodesize;
2624
2625			if (list_empty(&next->upper))
2626				break;
2627
2628			edge = list_entry(next->upper.next,
2629					  struct backref_edge, list[LOWER]);
2630			edges[index++] = edge;
2631			next = edge->node[UPPER];
2632		}
2633		next = walk_down_backref(edges, &index);
2634	}
2635	return num_bytes;
2636}
2637
2638static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2639				  struct reloc_control *rc,
2640				  struct backref_node *node)
2641{
2642	struct btrfs_root *root = rc->extent_root;
2643	struct btrfs_fs_info *fs_info = root->fs_info;
2644	u64 num_bytes;
2645	int ret;
2646	u64 tmp;
2647
2648	num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2649
2650	trans->block_rsv = rc->block_rsv;
2651	rc->reserved_bytes += num_bytes;
2652
2653	/*
2654	 * We are under a transaction here so we can only do limited flushing.
2655	 * If we get an enospc just kick back -EAGAIN so we know to drop the
2656	 * transaction and try to refill when we can flush all the things.
2657	 */
2658	ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2659				BTRFS_RESERVE_FLUSH_LIMIT);
2660	if (ret) {
2661		tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2662		while (tmp <= rc->reserved_bytes)
2663			tmp <<= 1;
2664		/*
2665		 * only one thread can access block_rsv at this point,
2666		 * so we don't need hold lock to protect block_rsv.
2667		 * we expand more reservation size here to allow enough
2668		 * space for relocation and we will return eailer in
2669		 * enospc case.
2670		 */
2671		rc->block_rsv->size = tmp + fs_info->nodesize *
2672				      RELOCATION_RESERVED_NODES;
2673		return -EAGAIN;
2674	}
2675
2676	return 0;
2677}
2678
2679/*
2680 * relocate a block tree, and then update pointers in upper level
2681 * blocks that reference the block to point to the new location.
2682 *
2683 * if called by link_to_upper, the block has already been relocated.
2684 * in that case this function just updates pointers.
2685 */
2686static int do_relocation(struct btrfs_trans_handle *trans,
2687			 struct reloc_control *rc,
2688			 struct backref_node *node,
2689			 struct btrfs_key *key,
2690			 struct btrfs_path *path, int lowest)
2691{
2692	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2693	struct backref_node *upper;
2694	struct backref_edge *edge;
2695	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2696	struct btrfs_root *root;
2697	struct extent_buffer *eb;
2698	u32 blocksize;
2699	u64 bytenr;
2700	u64 generation;
2701	int slot;
2702	int ret;
2703	int err = 0;
2704
2705	BUG_ON(lowest && node->eb);
 
 
 
 
2706
2707	path->lowest_level = node->level + 1;
2708	rc->backref_cache.path[node->level] = node;
2709	list_for_each_entry(edge, &node->upper, list[LOWER]) {
2710		cond_resched();
2711
2712		upper = edge->node[UPPER];
2713		root = select_reloc_root(trans, rc, upper, edges);
2714		BUG_ON(!root);
 
 
 
2715
2716		if (upper->eb && !upper->locked) {
2717			if (!lowest) {
2718				ret = btrfs_bin_search(upper->eb, key,
2719						       upper->level, &slot);
 
2720				BUG_ON(ret);
2721				bytenr = btrfs_node_blockptr(upper->eb, slot);
2722				if (node->eb->start == bytenr)
2723					goto next;
2724			}
2725			drop_node_buffer(upper);
2726		}
2727
2728		if (!upper->eb) {
2729			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2730			if (ret) {
2731				if (ret < 0)
2732					err = ret;
2733				else
2734					err = -ENOENT;
2735
2736				btrfs_release_path(path);
2737				break;
2738			}
2739
2740			if (!upper->eb) {
2741				upper->eb = path->nodes[upper->level];
2742				path->nodes[upper->level] = NULL;
2743			} else {
2744				BUG_ON(upper->eb != path->nodes[upper->level]);
2745			}
2746
2747			upper->locked = 1;
2748			path->locks[upper->level] = 0;
2749
2750			slot = path->slots[upper->level];
2751			btrfs_release_path(path);
2752		} else {
2753			ret = btrfs_bin_search(upper->eb, key, upper->level,
2754					       &slot);
 
2755			BUG_ON(ret);
2756		}
2757
2758		bytenr = btrfs_node_blockptr(upper->eb, slot);
2759		if (lowest) {
2760			if (bytenr != node->bytenr) {
2761				btrfs_err(root->fs_info,
2762		"lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2763					  bytenr, node->bytenr, slot,
2764					  upper->eb->start);
2765				err = -EIO;
2766				goto next;
2767			}
2768		} else {
2769			if (node->eb->start == bytenr)
2770				goto next;
2771		}
2772
2773		blocksize = root->fs_info->nodesize;
2774		generation = btrfs_node_ptr_generation(upper->eb, slot);
2775		eb = read_tree_block(fs_info, bytenr, generation);
2776		if (IS_ERR(eb)) {
2777			err = PTR_ERR(eb);
2778			goto next;
2779		} else if (!extent_buffer_uptodate(eb)) {
2780			free_extent_buffer(eb);
2781			err = -EIO;
2782			goto next;
2783		}
2784		btrfs_tree_lock(eb);
2785		btrfs_set_lock_blocking(eb);
2786
2787		if (!node->eb) {
2788			ret = btrfs_cow_block(trans, root, eb, upper->eb,
2789					      slot, &eb);
2790			btrfs_tree_unlock(eb);
2791			free_extent_buffer(eb);
2792			if (ret < 0) {
2793				err = ret;
2794				goto next;
2795			}
2796			BUG_ON(node->eb != eb);
 
 
 
2797		} else {
 
 
 
 
 
 
 
 
 
2798			btrfs_set_node_blockptr(upper->eb, slot,
2799						node->eb->start);
2800			btrfs_set_node_ptr_generation(upper->eb, slot,
2801						      trans->transid);
2802			btrfs_mark_buffer_dirty(upper->eb);
2803
2804			ret = btrfs_inc_extent_ref(trans, root->fs_info,
2805						node->eb->start, blocksize,
2806						upper->eb->start,
2807						btrfs_header_owner(upper->eb),
2808						node->level, 0);
2809			BUG_ON(ret);
2810
2811			ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2812			BUG_ON(ret);
 
 
 
 
 
 
2813		}
2814next:
2815		if (!upper->pending)
2816			drop_node_buffer(upper);
2817		else
2818			unlock_node_buffer(upper);
2819		if (err)
2820			break;
2821	}
2822
2823	if (!err && node->pending) {
2824		drop_node_buffer(node);
2825		list_move_tail(&node->list, &rc->backref_cache.changed);
2826		node->pending = 0;
2827	}
2828
2829	path->lowest_level = 0;
2830	BUG_ON(err == -ENOSPC);
2831	return err;
 
 
 
 
 
2832}
2833
2834static int link_to_upper(struct btrfs_trans_handle *trans,
2835			 struct reloc_control *rc,
2836			 struct backref_node *node,
2837			 struct btrfs_path *path)
2838{
2839	struct btrfs_key key;
2840
2841	btrfs_node_key_to_cpu(node->eb, &key, 0);
2842	return do_relocation(trans, rc, node, &key, path, 0);
2843}
2844
2845static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2846				struct reloc_control *rc,
2847				struct btrfs_path *path, int err)
2848{
2849	LIST_HEAD(list);
2850	struct backref_cache *cache = &rc->backref_cache;
2851	struct backref_node *node;
2852	int level;
2853	int ret;
2854
2855	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2856		while (!list_empty(&cache->pending[level])) {
2857			node = list_entry(cache->pending[level].next,
2858					  struct backref_node, list);
2859			list_move_tail(&node->list, &list);
2860			BUG_ON(!node->pending);
2861
2862			if (!err) {
2863				ret = link_to_upper(trans, rc, node, path);
2864				if (ret < 0)
2865					err = ret;
2866			}
2867		}
2868		list_splice_init(&list, &cache->pending[level]);
2869	}
2870	return err;
2871}
2872
2873static void mark_block_processed(struct reloc_control *rc,
2874				 u64 bytenr, u32 blocksize)
2875{
2876	set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2877			EXTENT_DIRTY);
2878}
2879
2880static void __mark_block_processed(struct reloc_control *rc,
2881				   struct backref_node *node)
2882{
2883	u32 blocksize;
2884	if (node->level == 0 ||
2885	    in_block_group(node->bytenr, rc->block_group)) {
2886		blocksize = rc->extent_root->fs_info->nodesize;
2887		mark_block_processed(rc, node->bytenr, blocksize);
2888	}
2889	node->processed = 1;
2890}
2891
2892/*
2893 * mark a block and all blocks directly/indirectly reference the block
2894 * as processed.
2895 */
2896static void update_processed_blocks(struct reloc_control *rc,
2897				    struct backref_node *node)
2898{
2899	struct backref_node *next = node;
2900	struct backref_edge *edge;
2901	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2902	int index = 0;
2903
2904	while (next) {
2905		cond_resched();
2906		while (1) {
2907			if (next->processed)
2908				break;
2909
2910			__mark_block_processed(rc, next);
2911
2912			if (list_empty(&next->upper))
2913				break;
2914
2915			edge = list_entry(next->upper.next,
2916					  struct backref_edge, list[LOWER]);
2917			edges[index++] = edge;
2918			next = edge->node[UPPER];
2919		}
2920		next = walk_down_backref(edges, &index);
2921	}
2922}
2923
2924static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2925{
2926	u32 blocksize = rc->extent_root->fs_info->nodesize;
2927
2928	if (test_range_bit(&rc->processed_blocks, bytenr,
2929			   bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2930		return 1;
2931	return 0;
2932}
2933
2934static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2935			      struct tree_block *block)
2936{
 
 
 
 
 
2937	struct extent_buffer *eb;
2938
2939	BUG_ON(block->key_ready);
2940	eb = read_tree_block(fs_info, block->bytenr, block->key.offset);
2941	if (IS_ERR(eb)) {
2942		return PTR_ERR(eb);
2943	} else if (!extent_buffer_uptodate(eb)) {
2944		free_extent_buffer(eb);
2945		return -EIO;
2946	}
2947	WARN_ON(btrfs_header_level(eb) != block->level);
2948	if (block->level == 0)
2949		btrfs_item_key_to_cpu(eb, &block->key, 0);
2950	else
2951		btrfs_node_key_to_cpu(eb, &block->key, 0);
2952	free_extent_buffer(eb);
2953	block->key_ready = 1;
2954	return 0;
2955}
2956
2957/*
2958 * helper function to relocate a tree block
2959 */
2960static int relocate_tree_block(struct btrfs_trans_handle *trans,
2961				struct reloc_control *rc,
2962				struct backref_node *node,
2963				struct btrfs_key *key,
2964				struct btrfs_path *path)
2965{
2966	struct btrfs_root *root;
2967	int ret = 0;
2968
2969	if (!node)
2970		return 0;
2971
 
 
 
 
 
 
 
 
2972	BUG_ON(node->processed);
2973	root = select_one_root(node);
2974	if (root == ERR_PTR(-ENOENT)) {
2975		update_processed_blocks(rc, node);
2976		goto out;
2977	}
2978
2979	if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2980		ret = reserve_metadata_space(trans, rc, node);
2981		if (ret)
2982			goto out;
 
 
 
2983	}
2984
2985	if (root) {
2986		if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2987			BUG_ON(node->new_bytenr);
2988			BUG_ON(!list_empty(&node->list));
2989			btrfs_record_root_in_trans(trans, root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2990			root = root->reloc_root;
2991			node->new_bytenr = root->node->start;
2992			node->root = root;
 
 
2993			list_add_tail(&node->list, &rc->backref_cache.changed);
2994		} else {
2995			path->lowest_level = node->level;
 
 
2996			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2997			btrfs_release_path(path);
 
 
2998			if (ret > 0)
2999				ret = 0;
3000		}
3001		if (!ret)
3002			update_processed_blocks(rc, node);
3003	} else {
3004		ret = do_relocation(trans, rc, node, key, path, 1);
3005	}
3006out:
3007	if (ret || node->level == 0 || node->cowonly)
3008		remove_backref_node(&rc->backref_cache, node);
3009	return ret;
3010}
3011
3012/*
3013 * relocate a list of blocks
3014 */
3015static noinline_for_stack
3016int relocate_tree_blocks(struct btrfs_trans_handle *trans,
3017			 struct reloc_control *rc, struct rb_root *blocks)
3018{
3019	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3020	struct backref_node *node;
3021	struct btrfs_path *path;
3022	struct tree_block *block;
3023	struct rb_node *rb_node;
3024	int ret;
3025	int err = 0;
3026
3027	path = btrfs_alloc_path();
3028	if (!path) {
3029		err = -ENOMEM;
3030		goto out_free_blocks;
3031	}
3032
3033	rb_node = rb_first(blocks);
3034	while (rb_node) {
3035		block = rb_entry(rb_node, struct tree_block, rb_node);
3036		if (!block->key_ready)
3037			readahead_tree_block(fs_info, block->bytenr);
3038		rb_node = rb_next(rb_node);
 
3039	}
3040
3041	rb_node = rb_first(blocks);
3042	while (rb_node) {
3043		block = rb_entry(rb_node, struct tree_block, rb_node);
3044		if (!block->key_ready) {
3045			err = get_tree_block_key(fs_info, block);
3046			if (err)
3047				goto out_free_path;
3048		}
3049		rb_node = rb_next(rb_node);
3050	}
3051
3052	rb_node = rb_first(blocks);
3053	while (rb_node) {
3054		block = rb_entry(rb_node, struct tree_block, rb_node);
3055
3056		node = build_backref_tree(rc, &block->key,
3057					  block->level, block->bytenr);
3058		if (IS_ERR(node)) {
3059			err = PTR_ERR(node);
3060			goto out;
3061		}
3062
3063		ret = relocate_tree_block(trans, rc, node, &block->key,
3064					  path);
3065		if (ret < 0) {
3066			if (ret != -EAGAIN || rb_node == rb_first(blocks))
3067				err = ret;
3068			goto out;
3069		}
3070		rb_node = rb_next(rb_node);
3071	}
3072out:
3073	err = finish_pending_nodes(trans, rc, path, err);
3074
3075out_free_path:
3076	btrfs_free_path(path);
3077out_free_blocks:
3078	free_block_list(blocks);
3079	return err;
3080}
3081
3082static noinline_for_stack
3083int prealloc_file_extent_cluster(struct inode *inode,
3084				 struct file_extent_cluster *cluster)
3085{
 
 
3086	u64 alloc_hint = 0;
3087	u64 start;
3088	u64 end;
3089	u64 offset = BTRFS_I(inode)->index_cnt;
3090	u64 num_bytes;
3091	int nr = 0;
3092	int ret = 0;
 
3093	u64 prealloc_start = cluster->start - offset;
3094	u64 prealloc_end = cluster->end - offset;
3095	u64 cur_offset;
3096
3097	BUG_ON(cluster->start != cluster->boundary[0]);
3098	inode_lock(inode);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3099
3100	ret = btrfs_check_data_free_space(inode, prealloc_start,
3101					  prealloc_end + 1 - prealloc_start);
 
3102	if (ret)
3103		goto out;
 
 
 
 
3104
3105	cur_offset = prealloc_start;
3106	while (nr < cluster->nr) {
3107		start = cluster->boundary[nr] - offset;
3108		if (nr + 1 < cluster->nr)
3109			end = cluster->boundary[nr + 1] - 1 - offset;
3110		else
3111			end = cluster->end - offset;
3112
3113		lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3114		num_bytes = end + 1 - start;
3115		if (cur_offset < start)
3116			btrfs_free_reserved_data_space(inode, cur_offset,
3117					start - cur_offset);
3118		ret = btrfs_prealloc_file_range(inode, 0, start,
3119						num_bytes, num_bytes,
3120						end + 1, &alloc_hint);
3121		cur_offset = end + 1;
3122		unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3123		if (ret)
3124			break;
3125		nr++;
3126	}
 
 
3127	if (cur_offset < prealloc_end)
3128		btrfs_free_reserved_data_space(inode, cur_offset,
3129				       prealloc_end + 1 - cur_offset);
3130out:
3131	inode_unlock(inode);
3132	return ret;
3133}
3134
3135static noinline_for_stack
3136int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3137			 u64 block_start)
3138{
3139	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3140	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3141	struct extent_map *em;
 
 
 
 
3142	int ret = 0;
3143
3144	em = alloc_extent_map();
3145	if (!em)
3146		return -ENOMEM;
3147
3148	em->start = start;
3149	em->len = end + 1 - start;
3150	em->block_len = em->len;
3151	em->block_start = block_start;
3152	em->bdev = fs_info->fs_devices->latest_bdev;
3153	set_bit(EXTENT_FLAG_PINNED, &em->flags);
 
 
 
 
 
3154
3155	lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3156	while (1) {
3157		write_lock(&em_tree->lock);
3158		ret = add_extent_mapping(em_tree, em, 0);
3159		write_unlock(&em_tree->lock);
3160		if (ret != -EEXIST) {
3161			free_extent_map(em);
3162			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3163		}
3164		btrfs_drop_extent_cache(inode, start, end, 0);
3165	}
3166	unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3167	return ret;
3168}
3169
3170static int relocate_file_extent_cluster(struct inode *inode,
3171					struct file_extent_cluster *cluster)
3172{
3173	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3174	u64 page_start;
3175	u64 page_end;
3176	u64 offset = BTRFS_I(inode)->index_cnt;
3177	unsigned long index;
3178	unsigned long last_index;
3179	struct page *page;
3180	struct file_ra_state *ra;
3181	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3182	int nr = 0;
3183	int ret = 0;
3184
3185	if (!cluster->nr)
3186		return 0;
3187
3188	ra = kzalloc(sizeof(*ra), GFP_NOFS);
3189	if (!ra)
3190		return -ENOMEM;
3191
3192	ret = prealloc_file_extent_cluster(inode, cluster);
3193	if (ret)
3194		goto out;
3195
3196	file_ra_state_init(ra, inode->i_mapping);
3197
3198	ret = setup_extent_mapping(inode, cluster->start - offset,
3199				   cluster->end - offset, cluster->start);
3200	if (ret)
3201		goto out;
3202
3203	index = (cluster->start - offset) >> PAGE_SHIFT;
3204	last_index = (cluster->end - offset) >> PAGE_SHIFT;
3205	while (index <= last_index) {
3206		ret = btrfs_delalloc_reserve_metadata(inode, PAGE_SIZE);
3207		if (ret)
3208			goto out;
3209
3210		page = find_lock_page(inode->i_mapping, index);
3211		if (!page) {
3212			page_cache_sync_readahead(inode->i_mapping,
3213						  ra, NULL, index,
3214						  last_index + 1 - index);
3215			page = find_or_create_page(inode->i_mapping, index,
3216						   mask);
3217			if (!page) {
3218				btrfs_delalloc_release_metadata(inode,
3219							PAGE_SIZE);
3220				ret = -ENOMEM;
3221				goto out;
3222			}
3223		}
3224
3225		if (PageReadahead(page)) {
3226			page_cache_async_readahead(inode->i_mapping,
3227						   ra, NULL, page, index,
3228						   last_index + 1 - index);
3229		}
3230
3231		if (!PageUptodate(page)) {
3232			btrfs_readpage(NULL, page);
3233			lock_page(page);
3234			if (!PageUptodate(page)) {
3235				unlock_page(page);
3236				put_page(page);
3237				btrfs_delalloc_release_metadata(inode,
3238							PAGE_SIZE);
3239				ret = -EIO;
3240				goto out;
3241			}
3242		}
3243
3244		page_start = page_offset(page);
3245		page_end = page_start + PAGE_SIZE - 1;
3246
3247		lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3248
3249		set_page_extent_mapped(page);
3250
3251		if (nr < cluster->nr &&
3252		    page_start + offset == cluster->boundary[nr]) {
3253			set_extent_bits(&BTRFS_I(inode)->io_tree,
3254					page_start, page_end,
3255					EXTENT_BOUNDARY);
3256			nr++;
3257		}
3258
3259		btrfs_set_extent_delalloc(inode, page_start, page_end, NULL, 0);
3260		set_page_dirty(page);
3261
3262		unlock_extent(&BTRFS_I(inode)->io_tree,
3263			      page_start, page_end);
3264		unlock_page(page);
3265		put_page(page);
3266
3267		index++;
3268		balance_dirty_pages_ratelimited(inode->i_mapping);
3269		btrfs_throttle(fs_info);
3270	}
3271	WARN_ON(nr != cluster->nr);
3272out:
3273	kfree(ra);
3274	return ret;
3275}
3276
3277static noinline_for_stack
3278int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3279			 struct file_extent_cluster *cluster)
3280{
 
 
3281	int ret;
 
3282
3283	if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3284		ret = relocate_file_extent_cluster(inode, cluster);
3285		if (ret)
3286			return ret;
3287		cluster->nr = 0;
3288	}
3289
3290	if (!cluster->nr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3291		cluster->start = extent_key->objectid;
 
 
3292	else
3293		BUG_ON(cluster->nr >= MAX_EXTENTS);
3294	cluster->end = extent_key->objectid + extent_key->offset - 1;
3295	cluster->boundary[cluster->nr] = extent_key->objectid;
3296	cluster->nr++;
3297
3298	if (cluster->nr >= MAX_EXTENTS) {
3299		ret = relocate_file_extent_cluster(inode, cluster);
3300		if (ret)
3301			return ret;
3302		cluster->nr = 0;
3303	}
3304	return 0;
3305}
3306
3307#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3308static int get_ref_objectid_v0(struct reloc_control *rc,
3309			       struct btrfs_path *path,
3310			       struct btrfs_key *extent_key,
3311			       u64 *ref_objectid, int *path_change)
3312{
3313	struct btrfs_key key;
3314	struct extent_buffer *leaf;
3315	struct btrfs_extent_ref_v0 *ref0;
3316	int ret;
3317	int slot;
3318
3319	leaf = path->nodes[0];
3320	slot = path->slots[0];
3321	while (1) {
3322		if (slot >= btrfs_header_nritems(leaf)) {
3323			ret = btrfs_next_leaf(rc->extent_root, path);
3324			if (ret < 0)
3325				return ret;
3326			BUG_ON(ret > 0);
3327			leaf = path->nodes[0];
3328			slot = path->slots[0];
3329			if (path_change)
3330				*path_change = 1;
3331		}
3332		btrfs_item_key_to_cpu(leaf, &key, slot);
3333		if (key.objectid != extent_key->objectid)
3334			return -ENOENT;
3335
3336		if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3337			slot++;
3338			continue;
3339		}
3340		ref0 = btrfs_item_ptr(leaf, slot,
3341				struct btrfs_extent_ref_v0);
3342		*ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3343		break;
3344	}
3345	return 0;
3346}
3347#endif
3348
3349/*
3350 * helper to add a tree block to the list.
3351 * the major work is getting the generation and level of the block
3352 */
3353static int add_tree_block(struct reloc_control *rc,
3354			  struct btrfs_key *extent_key,
3355			  struct btrfs_path *path,
3356			  struct rb_root *blocks)
3357{
3358	struct extent_buffer *eb;
3359	struct btrfs_extent_item *ei;
3360	struct btrfs_tree_block_info *bi;
3361	struct tree_block *block;
3362	struct rb_node *rb_node;
3363	u32 item_size;
3364	int level = -1;
3365	u64 generation;
 
3366
3367	eb =  path->nodes[0];
3368	item_size = btrfs_item_size_nr(eb, path->slots[0]);
3369
3370	if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3371	    item_size >= sizeof(*ei) + sizeof(*bi)) {
 
 
3372		ei = btrfs_item_ptr(eb, path->slots[0],
3373				struct btrfs_extent_item);
 
3374		if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3375			bi = (struct btrfs_tree_block_info *)(ei + 1);
3376			level = btrfs_tree_block_level(eb, bi);
 
3377		} else {
3378			level = (int)extent_key->offset;
 
3379		}
3380		generation = btrfs_extent_generation(eb, ei);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3381	} else {
3382#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3383		u64 ref_owner;
3384		int ret;
3385
3386		BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3387		ret = get_ref_objectid_v0(rc, path, extent_key,
3388					  &ref_owner, NULL);
3389		if (ret < 0)
3390			return ret;
3391		BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3392		level = (int)ref_owner;
3393		/* FIXME: get real generation */
3394		generation = 0;
3395#else
3396		BUG();
3397#endif
3398	}
3399
3400	btrfs_release_path(path);
3401
3402	BUG_ON(level == -1);
3403
3404	block = kmalloc(sizeof(*block), GFP_NOFS);
3405	if (!block)
3406		return -ENOMEM;
3407
3408	block->bytenr = extent_key->objectid;
3409	block->key.objectid = rc->extent_root->fs_info->nodesize;
3410	block->key.offset = generation;
3411	block->level = level;
3412	block->key_ready = 0;
 
3413
3414	rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3415	if (rb_node)
3416		backref_tree_panic(rb_node, -EEXIST, block->bytenr);
 
3417
3418	return 0;
3419}
3420
3421/*
3422 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3423 */
3424static int __add_tree_block(struct reloc_control *rc,
3425			    u64 bytenr, u32 blocksize,
3426			    struct rb_root *blocks)
3427{
3428	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3429	struct btrfs_path *path;
3430	struct btrfs_key key;
3431	int ret;
3432	bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3433
3434	if (tree_block_processed(bytenr, rc))
3435		return 0;
3436
3437	if (tree_search(blocks, bytenr))
3438		return 0;
3439
3440	path = btrfs_alloc_path();
3441	if (!path)
3442		return -ENOMEM;
3443again:
3444	key.objectid = bytenr;
3445	if (skinny) {
3446		key.type = BTRFS_METADATA_ITEM_KEY;
3447		key.offset = (u64)-1;
3448	} else {
3449		key.type = BTRFS_EXTENT_ITEM_KEY;
3450		key.offset = blocksize;
3451	}
3452
3453	path->search_commit_root = 1;
3454	path->skip_locking = 1;
3455	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3456	if (ret < 0)
3457		goto out;
3458
3459	if (ret > 0 && skinny) {
3460		if (path->slots[0]) {
3461			path->slots[0]--;
3462			btrfs_item_key_to_cpu(path->nodes[0], &key,
3463					      path->slots[0]);
3464			if (key.objectid == bytenr &&
3465			    (key.type == BTRFS_METADATA_ITEM_KEY ||
3466			     (key.type == BTRFS_EXTENT_ITEM_KEY &&
3467			      key.offset == blocksize)))
3468				ret = 0;
3469		}
3470
3471		if (ret) {
3472			skinny = false;
3473			btrfs_release_path(path);
3474			goto again;
3475		}
3476	}
3477	BUG_ON(ret);
 
 
 
 
 
 
 
 
 
3478
3479	ret = add_tree_block(rc, &key, path, blocks);
3480out:
3481	btrfs_free_path(path);
3482	return ret;
3483}
3484
3485/*
3486 * helper to check if the block use full backrefs for pointers in it
3487 */
3488static int block_use_full_backref(struct reloc_control *rc,
3489				  struct extent_buffer *eb)
3490{
3491	u64 flags;
3492	int ret;
3493
3494	if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3495	    btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3496		return 1;
3497
3498	ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3499				       eb->start, btrfs_header_level(eb), 1,
3500				       NULL, &flags);
3501	BUG_ON(ret);
3502
3503	if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3504		ret = 1;
3505	else
3506		ret = 0;
3507	return ret;
3508}
3509
3510static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3511				    struct btrfs_block_group_cache *block_group,
3512				    struct inode *inode,
3513				    u64 ino)
3514{
3515	struct btrfs_key key;
3516	struct btrfs_root *root = fs_info->tree_root;
3517	struct btrfs_trans_handle *trans;
3518	int ret = 0;
3519
3520	if (inode)
3521		goto truncate;
3522
3523	key.objectid = ino;
3524	key.type = BTRFS_INODE_ITEM_KEY;
3525	key.offset = 0;
3526
3527	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3528	if (IS_ERR(inode) || is_bad_inode(inode)) {
3529		if (!IS_ERR(inode))
3530			iput(inode);
3531		return -ENOENT;
3532	}
3533
3534truncate:
3535	ret = btrfs_check_trunc_cache_free_space(fs_info,
3536						 &fs_info->global_block_rsv);
3537	if (ret)
3538		goto out;
3539
3540	trans = btrfs_join_transaction(root);
3541	if (IS_ERR(trans)) {
3542		ret = PTR_ERR(trans);
3543		goto out;
3544	}
3545
3546	ret = btrfs_truncate_free_space_cache(root, trans, block_group, inode);
3547
3548	btrfs_end_transaction(trans);
3549	btrfs_btree_balance_dirty(fs_info);
3550out:
3551	iput(inode);
3552	return ret;
3553}
3554
3555/*
3556 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3557 * this function scans fs tree to find blocks reference the data extent
3558 */
3559static int find_data_references(struct reloc_control *rc,
3560				struct btrfs_key *extent_key,
3561				struct extent_buffer *leaf,
3562				struct btrfs_extent_data_ref *ref,
3563				struct rb_root *blocks)
3564{
3565	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3566	struct btrfs_path *path;
3567	struct tree_block *block;
3568	struct btrfs_root *root;
3569	struct btrfs_file_extent_item *fi;
3570	struct rb_node *rb_node;
3571	struct btrfs_key key;
3572	u64 ref_root;
3573	u64 ref_objectid;
3574	u64 ref_offset;
3575	u32 ref_count;
3576	u32 nritems;
3577	int err = 0;
3578	int added = 0;
3579	int counted;
3580	int ret;
3581
3582	ref_root = btrfs_extent_data_ref_root(leaf, ref);
3583	ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3584	ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3585	ref_count = btrfs_extent_data_ref_count(leaf, ref);
3586
3587	/*
3588	 * This is an extent belonging to the free space cache, lets just delete
3589	 * it and redo the search.
3590	 */
3591	if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3592		ret = delete_block_group_cache(fs_info, rc->block_group,
3593					       NULL, ref_objectid);
3594		if (ret != -ENOENT)
3595			return ret;
3596		ret = 0;
3597	}
3598
3599	path = btrfs_alloc_path();
3600	if (!path)
3601		return -ENOMEM;
3602	path->reada = READA_FORWARD;
3603
3604	root = read_fs_root(fs_info, ref_root);
3605	if (IS_ERR(root)) {
3606		err = PTR_ERR(root);
3607		goto out;
3608	}
3609
3610	key.objectid = ref_objectid;
3611	key.type = BTRFS_EXTENT_DATA_KEY;
3612	if (ref_offset > ((u64)-1 << 32))
3613		key.offset = 0;
3614	else
3615		key.offset = ref_offset;
3616
3617	path->search_commit_root = 1;
3618	path->skip_locking = 1;
3619	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3620	if (ret < 0) {
3621		err = ret;
3622		goto out;
3623	}
3624
3625	leaf = path->nodes[0];
3626	nritems = btrfs_header_nritems(leaf);
3627	/*
3628	 * the references in tree blocks that use full backrefs
3629	 * are not counted in
3630	 */
3631	if (block_use_full_backref(rc, leaf))
3632		counted = 0;
3633	else
3634		counted = 1;
3635	rb_node = tree_search(blocks, leaf->start);
3636	if (rb_node) {
3637		if (counted)
3638			added = 1;
3639		else
3640			path->slots[0] = nritems;
3641	}
3642
3643	while (ref_count > 0) {
3644		while (path->slots[0] >= nritems) {
3645			ret = btrfs_next_leaf(root, path);
3646			if (ret < 0) {
3647				err = ret;
3648				goto out;
3649			}
3650			if (WARN_ON(ret > 0))
3651				goto out;
3652
3653			leaf = path->nodes[0];
3654			nritems = btrfs_header_nritems(leaf);
3655			added = 0;
3656
3657			if (block_use_full_backref(rc, leaf))
3658				counted = 0;
3659			else
3660				counted = 1;
3661			rb_node = tree_search(blocks, leaf->start);
3662			if (rb_node) {
3663				if (counted)
3664					added = 1;
3665				else
3666					path->slots[0] = nritems;
3667			}
3668		}
3669
3670		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3671		if (WARN_ON(key.objectid != ref_objectid ||
3672		    key.type != BTRFS_EXTENT_DATA_KEY))
 
 
3673			break;
3674
3675		fi = btrfs_item_ptr(leaf, path->slots[0],
3676				    struct btrfs_file_extent_item);
3677
3678		if (btrfs_file_extent_type(leaf, fi) ==
3679		    BTRFS_FILE_EXTENT_INLINE)
3680			goto next;
3681
3682		if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3683		    extent_key->objectid)
3684			goto next;
3685
3686		key.offset -= btrfs_file_extent_offset(leaf, fi);
3687		if (key.offset != ref_offset)
3688			goto next;
3689
3690		if (counted)
3691			ref_count--;
3692		if (added)
3693			goto next;
3694
3695		if (!tree_block_processed(leaf->start, rc)) {
3696			block = kmalloc(sizeof(*block), GFP_NOFS);
3697			if (!block) {
3698				err = -ENOMEM;
3699				break;
3700			}
3701			block->bytenr = leaf->start;
3702			btrfs_item_key_to_cpu(leaf, &block->key, 0);
3703			block->level = 0;
3704			block->key_ready = 1;
3705			rb_node = tree_insert(blocks, block->bytenr,
3706					      &block->rb_node);
3707			if (rb_node)
3708				backref_tree_panic(rb_node, -EEXIST,
3709						   block->bytenr);
3710		}
3711		if (counted)
3712			added = 1;
3713		else
3714			path->slots[0] = nritems;
3715next:
3716		path->slots[0]++;
3717
3718	}
3719out:
3720	btrfs_free_path(path);
3721	return err;
 
 
3722}
3723
3724/*
3725 * helper to find all tree blocks that reference a given data extent
3726 */
3727static noinline_for_stack
3728int add_data_references(struct reloc_control *rc,
3729			struct btrfs_key *extent_key,
3730			struct btrfs_path *path,
3731			struct rb_root *blocks)
3732{
3733	struct btrfs_key key;
3734	struct extent_buffer *eb;
3735	struct btrfs_extent_data_ref *dref;
3736	struct btrfs_extent_inline_ref *iref;
3737	unsigned long ptr;
3738	unsigned long end;
3739	u32 blocksize = rc->extent_root->fs_info->nodesize;
3740	int ret = 0;
3741	int err = 0;
3742
3743	eb = path->nodes[0];
3744	ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3745	end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3746#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3747	if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3748		ptr = end;
3749	else
3750#endif
3751		ptr += sizeof(struct btrfs_extent_item);
3752
3753	while (ptr < end) {
3754		iref = (struct btrfs_extent_inline_ref *)ptr;
3755		key.type = btrfs_extent_inline_ref_type(eb, iref);
3756		if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3757			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3758			ret = __add_tree_block(rc, key.offset, blocksize,
3759					       blocks);
3760		} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3761			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3762			ret = find_data_references(rc, extent_key,
3763						   eb, dref, blocks);
3764		} else {
3765			BUG();
3766		}
3767		if (ret) {
3768			err = ret;
3769			goto out;
3770		}
3771		ptr += btrfs_extent_inline_ref_size(key.type);
3772	}
3773	WARN_ON(ptr > end);
3774
3775	while (1) {
3776		cond_resched();
3777		eb = path->nodes[0];
3778		if (path->slots[0] >= btrfs_header_nritems(eb)) {
3779			ret = btrfs_next_leaf(rc->extent_root, path);
3780			if (ret < 0) {
3781				err = ret;
3782				break;
3783			}
3784			if (ret > 0)
3785				break;
3786			eb = path->nodes[0];
3787		}
3788
3789		btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3790		if (key.objectid != extent_key->objectid)
3791			break;
 
3792
3793#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3794		if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3795		    key.type == BTRFS_EXTENT_REF_V0_KEY) {
3796#else
3797		BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3798		if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3799#endif
3800			ret = __add_tree_block(rc, key.offset, blocksize,
3801					       blocks);
3802		} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3803			dref = btrfs_item_ptr(eb, path->slots[0],
3804					      struct btrfs_extent_data_ref);
3805			ret = find_data_references(rc, extent_key,
3806						   eb, dref, blocks);
3807		} else {
3808			ret = 0;
3809		}
3810		if (ret) {
3811			err = ret;
3812			break;
3813		}
3814		path->slots[0]++;
 
 
 
 
 
 
 
3815	}
3816out:
3817	btrfs_release_path(path);
3818	if (err)
3819		free_block_list(blocks);
3820	return err;
 
3821}
3822
3823/*
3824 * helper to find next unprocessed extent
3825 */
3826static noinline_for_stack
3827int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3828		     struct btrfs_key *extent_key)
3829{
3830	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3831	struct btrfs_key key;
3832	struct extent_buffer *leaf;
3833	u64 start, end, last;
3834	int ret;
3835
3836	last = rc->block_group->key.objectid + rc->block_group->key.offset;
3837	while (1) {
 
 
3838		cond_resched();
3839		if (rc->search_start >= last) {
3840			ret = 1;
3841			break;
3842		}
3843
3844		key.objectid = rc->search_start;
3845		key.type = BTRFS_EXTENT_ITEM_KEY;
3846		key.offset = 0;
3847
3848		path->search_commit_root = 1;
3849		path->skip_locking = 1;
3850		ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3851					0, 0);
3852		if (ret < 0)
3853			break;
3854next:
3855		leaf = path->nodes[0];
3856		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3857			ret = btrfs_next_leaf(rc->extent_root, path);
3858			if (ret != 0)
3859				break;
3860			leaf = path->nodes[0];
3861		}
3862
3863		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3864		if (key.objectid >= last) {
3865			ret = 1;
3866			break;
3867		}
3868
3869		if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3870		    key.type != BTRFS_METADATA_ITEM_KEY) {
3871			path->slots[0]++;
3872			goto next;
3873		}
3874
3875		if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3876		    key.objectid + key.offset <= rc->search_start) {
3877			path->slots[0]++;
3878			goto next;
3879		}
3880
3881		if (key.type == BTRFS_METADATA_ITEM_KEY &&
3882		    key.objectid + fs_info->nodesize <=
3883		    rc->search_start) {
3884			path->slots[0]++;
3885			goto next;
3886		}
3887
3888		ret = find_first_extent_bit(&rc->processed_blocks,
3889					    key.objectid, &start, &end,
3890					    EXTENT_DIRTY, NULL);
3891
3892		if (ret == 0 && start <= key.objectid) {
3893			btrfs_release_path(path);
3894			rc->search_start = end + 1;
3895		} else {
3896			if (key.type == BTRFS_EXTENT_ITEM_KEY)
3897				rc->search_start = key.objectid + key.offset;
3898			else
3899				rc->search_start = key.objectid +
3900					fs_info->nodesize;
3901			memcpy(extent_key, &key, sizeof(key));
3902			return 0;
3903		}
3904	}
3905	btrfs_release_path(path);
3906	return ret;
3907}
3908
3909static void set_reloc_control(struct reloc_control *rc)
3910{
3911	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3912
3913	mutex_lock(&fs_info->reloc_mutex);
3914	fs_info->reloc_ctl = rc;
3915	mutex_unlock(&fs_info->reloc_mutex);
3916}
3917
3918static void unset_reloc_control(struct reloc_control *rc)
3919{
3920	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3921
3922	mutex_lock(&fs_info->reloc_mutex);
3923	fs_info->reloc_ctl = NULL;
3924	mutex_unlock(&fs_info->reloc_mutex);
3925}
3926
3927static int check_extent_flags(u64 flags)
3928{
3929	if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3930	    (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3931		return 1;
3932	if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3933	    !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3934		return 1;
3935	if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3936	    (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3937		return 1;
3938	return 0;
3939}
3940
3941static noinline_for_stack
3942int prepare_to_relocate(struct reloc_control *rc)
3943{
3944	struct btrfs_trans_handle *trans;
3945	int ret;
3946
3947	rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3948					      BTRFS_BLOCK_RSV_TEMP);
3949	if (!rc->block_rsv)
3950		return -ENOMEM;
3951
3952	memset(&rc->cluster, 0, sizeof(rc->cluster));
3953	rc->search_start = rc->block_group->key.objectid;
3954	rc->extents_found = 0;
3955	rc->nodes_relocated = 0;
3956	rc->merging_rsv_size = 0;
3957	rc->reserved_bytes = 0;
3958	rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3959			      RELOCATION_RESERVED_NODES;
3960	ret = btrfs_block_rsv_refill(rc->extent_root,
3961				     rc->block_rsv, rc->block_rsv->size,
3962				     BTRFS_RESERVE_FLUSH_ALL);
3963	if (ret)
3964		return ret;
3965
3966	rc->create_reloc_tree = 1;
3967	set_reloc_control(rc);
3968
3969	trans = btrfs_join_transaction(rc->extent_root);
3970	if (IS_ERR(trans)) {
3971		unset_reloc_control(rc);
3972		/*
3973		 * extent tree is not a ref_cow tree and has no reloc_root to
3974		 * cleanup.  And callers are responsible to free the above
3975		 * block rsv.
3976		 */
3977		return PTR_ERR(trans);
3978	}
3979	btrfs_commit_transaction(trans);
3980	return 0;
 
 
 
 
3981}
3982
3983static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3984{
3985	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3986	struct rb_root blocks = RB_ROOT;
3987	struct btrfs_key key;
3988	struct btrfs_trans_handle *trans = NULL;
3989	struct btrfs_path *path;
3990	struct btrfs_extent_item *ei;
3991	u64 flags;
3992	u32 item_size;
3993	int ret;
3994	int err = 0;
3995	int progress = 0;
3996
3997	path = btrfs_alloc_path();
3998	if (!path)
3999		return -ENOMEM;
4000	path->reada = READA_FORWARD;
4001
4002	ret = prepare_to_relocate(rc);
4003	if (ret) {
4004		err = ret;
4005		goto out_free;
4006	}
4007
4008	while (1) {
4009		rc->reserved_bytes = 0;
4010		ret = btrfs_block_rsv_refill(rc->extent_root,
4011					rc->block_rsv, rc->block_rsv->size,
4012					BTRFS_RESERVE_FLUSH_ALL);
4013		if (ret) {
4014			err = ret;
4015			break;
4016		}
4017		progress++;
4018		trans = btrfs_start_transaction(rc->extent_root, 0);
4019		if (IS_ERR(trans)) {
4020			err = PTR_ERR(trans);
4021			trans = NULL;
4022			break;
4023		}
4024restart:
4025		if (update_backref_cache(trans, &rc->backref_cache)) {
4026			btrfs_end_transaction(trans);
4027			continue;
4028		}
4029
4030		ret = find_next_extent(rc, path, &key);
4031		if (ret < 0)
4032			err = ret;
4033		if (ret != 0)
4034			break;
4035
4036		rc->extents_found++;
4037
4038		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4039				    struct btrfs_extent_item);
4040		item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
4041		if (item_size >= sizeof(*ei)) {
4042			flags = btrfs_extent_flags(path->nodes[0], ei);
4043			ret = check_extent_flags(flags);
4044			BUG_ON(ret);
4045
4046		} else {
4047#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4048			u64 ref_owner;
4049			int path_change = 0;
4050
4051			BUG_ON(item_size !=
4052			       sizeof(struct btrfs_extent_item_v0));
4053			ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
4054						  &path_change);
4055			if (ret < 0) {
4056				err = ret;
4057				break;
4058			}
4059			if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
4060				flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
4061			else
4062				flags = BTRFS_EXTENT_FLAG_DATA;
4063
4064			if (path_change) {
4065				btrfs_release_path(path);
 
 
 
 
 
 
 
 
 
4066
4067				path->search_commit_root = 1;
4068				path->skip_locking = 1;
4069				ret = btrfs_search_slot(NULL, rc->extent_root,
4070							&key, path, 0, 0);
4071				if (ret < 0) {
4072					err = ret;
4073					break;
4074				}
4075				BUG_ON(ret > 0);
4076			}
4077#else
4078			BUG();
4079#endif
4080		}
4081
4082		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4083			ret = add_tree_block(rc, &key, path, &blocks);
4084		} else if (rc->stage == UPDATE_DATA_PTRS &&
4085			   (flags & BTRFS_EXTENT_FLAG_DATA)) {
4086			ret = add_data_references(rc, &key, path, &blocks);
4087		} else {
4088			btrfs_release_path(path);
4089			ret = 0;
4090		}
4091		if (ret < 0) {
4092			err = ret;
4093			break;
4094		}
4095
4096		if (!RB_EMPTY_ROOT(&blocks)) {
4097			ret = relocate_tree_blocks(trans, rc, &blocks);
4098			if (ret < 0) {
4099				/*
4100				 * if we fail to relocate tree blocks, force to update
4101				 * backref cache when committing transaction.
4102				 */
4103				rc->backref_cache.last_trans = trans->transid - 1;
4104
4105				if (ret != -EAGAIN) {
4106					err = ret;
4107					break;
4108				}
4109				rc->extents_found--;
4110				rc->search_start = key.objectid;
4111			}
4112		}
4113
4114		btrfs_end_transaction_throttle(trans);
4115		btrfs_btree_balance_dirty(fs_info);
4116		trans = NULL;
4117
4118		if (rc->stage == MOVE_DATA_EXTENTS &&
4119		    (flags & BTRFS_EXTENT_FLAG_DATA)) {
4120			rc->found_file_extent = 1;
4121			ret = relocate_data_extent(rc->data_inode,
4122						   &key, &rc->cluster);
4123			if (ret < 0) {
4124				err = ret;
4125				break;
4126			}
4127		}
 
 
 
 
4128	}
4129	if (trans && progress && err == -ENOSPC) {
4130		ret = btrfs_force_chunk_alloc(trans, fs_info,
4131					      rc->block_group->flags);
4132		if (ret == 1) {
4133			err = 0;
4134			progress = 0;
4135			goto restart;
4136		}
4137	}
4138
4139	btrfs_release_path(path);
4140	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4141
4142	if (trans) {
4143		btrfs_end_transaction_throttle(trans);
4144		btrfs_btree_balance_dirty(fs_info);
4145	}
4146
4147	if (!err) {
4148		ret = relocate_file_extent_cluster(rc->data_inode,
4149						   &rc->cluster);
4150		if (ret < 0)
4151			err = ret;
4152	}
4153
4154	rc->create_reloc_tree = 0;
4155	set_reloc_control(rc);
4156
4157	backref_cache_cleanup(&rc->backref_cache);
4158	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4159
 
 
 
 
 
 
 
 
4160	err = prepare_to_merge(rc, err);
4161
4162	merge_reloc_roots(rc);
4163
4164	rc->merge_reloc_tree = 0;
4165	unset_reloc_control(rc);
4166	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4167
4168	/* get rid of pinned extents */
4169	trans = btrfs_join_transaction(rc->extent_root);
4170	if (IS_ERR(trans)) {
4171		err = PTR_ERR(trans);
4172		goto out_free;
4173	}
4174	btrfs_commit_transaction(trans);
 
 
4175out_free:
 
 
 
4176	btrfs_free_block_rsv(fs_info, rc->block_rsv);
4177	btrfs_free_path(path);
4178	return err;
4179}
4180
4181static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4182				 struct btrfs_root *root, u64 objectid)
4183{
4184	struct btrfs_path *path;
4185	struct btrfs_inode_item *item;
4186	struct extent_buffer *leaf;
4187	int ret;
4188
4189	path = btrfs_alloc_path();
4190	if (!path)
4191		return -ENOMEM;
4192
4193	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4194	if (ret)
4195		goto out;
4196
4197	leaf = path->nodes[0];
4198	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4199	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4200	btrfs_set_inode_generation(leaf, item, 1);
4201	btrfs_set_inode_size(leaf, item, 0);
4202	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4203	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4204					  BTRFS_INODE_PREALLOC);
4205	btrfs_mark_buffer_dirty(leaf);
4206out:
4207	btrfs_free_path(path);
4208	return ret;
4209}
4210
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4211/*
4212 * helper to create inode for data relocation.
4213 * the inode is in data relocation tree and its link count is 0
4214 */
4215static noinline_for_stack
4216struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4217				 struct btrfs_block_group_cache *group)
4218{
4219	struct inode *inode = NULL;
4220	struct btrfs_trans_handle *trans;
4221	struct btrfs_root *root;
4222	struct btrfs_key key;
4223	u64 objectid;
4224	int err = 0;
4225
4226	root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4227	if (IS_ERR(root))
4228		return ERR_CAST(root);
4229
 
4230	trans = btrfs_start_transaction(root, 6);
4231	if (IS_ERR(trans))
 
4232		return ERR_CAST(trans);
 
4233
4234	err = btrfs_find_free_objectid(root, &objectid);
4235	if (err)
4236		goto out;
4237
4238	err = __insert_orphan_inode(trans, root, objectid);
4239	BUG_ON(err);
 
4240
4241	key.objectid = objectid;
4242	key.type = BTRFS_INODE_ITEM_KEY;
4243	key.offset = 0;
4244	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4245	BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
4246	BTRFS_I(inode)->index_cnt = group->key.objectid;
 
 
4247
4248	err = btrfs_orphan_add(trans, inode);
4249out:
 
4250	btrfs_end_transaction(trans);
4251	btrfs_btree_balance_dirty(fs_info);
4252	if (err) {
4253		if (inode)
4254			iput(inode);
4255		inode = ERR_PTR(err);
4256	}
4257	return inode;
4258}
4259
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4260static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4261{
4262	struct reloc_control *rc;
4263
4264	rc = kzalloc(sizeof(*rc), GFP_NOFS);
4265	if (!rc)
4266		return NULL;
4267
4268	INIT_LIST_HEAD(&rc->reloc_roots);
4269	backref_cache_init(&rc->backref_cache);
4270	mapping_tree_init(&rc->reloc_root_tree);
4271	extent_io_tree_init(&rc->processed_blocks,
4272			    fs_info->btree_inode->i_mapping);
 
4273	return rc;
4274}
4275
 
 
 
 
 
 
 
 
 
 
 
 
4276/*
4277 * Print the block group being relocated
4278 */
4279static void describe_relocation(struct btrfs_fs_info *fs_info,
4280				struct btrfs_block_group_cache *block_group)
4281{
4282	char buf[128];		/* prefixed by a '|' that'll be dropped */
4283	u64 flags = block_group->flags;
4284
4285	/* Shouldn't happen */
4286	if (!flags) {
4287		strcpy(buf, "|NONE");
4288	} else {
4289		char *bp = buf;
4290
4291#define DESCRIBE_FLAG(f, d) \
4292		if (flags & BTRFS_BLOCK_GROUP_##f) { \
4293			bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \
4294			flags &= ~BTRFS_BLOCK_GROUP_##f; \
4295		}
4296		DESCRIBE_FLAG(DATA,     "data");
4297		DESCRIBE_FLAG(SYSTEM,   "system");
4298		DESCRIBE_FLAG(METADATA, "metadata");
4299		DESCRIBE_FLAG(RAID0,    "raid0");
4300		DESCRIBE_FLAG(RAID1,    "raid1");
4301		DESCRIBE_FLAG(DUP,      "dup");
4302		DESCRIBE_FLAG(RAID10,   "raid10");
4303		DESCRIBE_FLAG(RAID5,    "raid5");
4304		DESCRIBE_FLAG(RAID6,    "raid6");
4305		if (flags)
4306			snprintf(buf, buf - bp + sizeof(buf), "|0x%llx", flags);
4307#undef DESCRIBE_FLAG
4308	}
4309
4310	btrfs_info(fs_info,
4311		   "relocating block group %llu flags %s",
4312		   block_group->key.objectid, buf + 1);
4313}
4314
4315/*
4316 * function to relocate all extents in a block group.
4317 */
4318int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4319{
4320	struct btrfs_root *extent_root = fs_info->extent_root;
 
4321	struct reloc_control *rc;
4322	struct inode *inode;
4323	struct btrfs_path *path;
4324	int ret;
4325	int rw = 0;
4326	int err = 0;
4327
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4328	rc = alloc_reloc_control(fs_info);
4329	if (!rc)
 
4330		return -ENOMEM;
 
4331
4332	rc->extent_root = extent_root;
 
 
 
 
4333
4334	rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4335	BUG_ON(!rc->block_group);
4336
4337	ret = btrfs_inc_block_group_ro(extent_root, rc->block_group);
4338	if (ret) {
4339		err = ret;
4340		goto out;
4341	}
4342	rw = 1;
4343
4344	path = btrfs_alloc_path();
4345	if (!path) {
4346		err = -ENOMEM;
4347		goto out;
4348	}
4349
4350	inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
4351					path);
4352	btrfs_free_path(path);
4353
4354	if (!IS_ERR(inode))
4355		ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4356	else
4357		ret = PTR_ERR(inode);
4358
4359	if (ret && ret != -ENOENT) {
4360		err = ret;
4361		goto out;
4362	}
4363
4364	rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4365	if (IS_ERR(rc->data_inode)) {
4366		err = PTR_ERR(rc->data_inode);
4367		rc->data_inode = NULL;
4368		goto out;
4369	}
4370
4371	describe_relocation(fs_info, rc->block_group);
4372
4373	btrfs_wait_block_group_reservations(rc->block_group);
4374	btrfs_wait_nocow_writers(rc->block_group);
4375	btrfs_wait_ordered_roots(fs_info, -1,
4376				 rc->block_group->key.objectid,
4377				 rc->block_group->key.offset);
 
4378
4379	while (1) {
 
 
4380		mutex_lock(&fs_info->cleaner_mutex);
4381		ret = relocate_block_group(rc);
4382		mutex_unlock(&fs_info->cleaner_mutex);
4383		if (ret < 0) {
4384			err = ret;
4385			goto out;
4386		}
4387
4388		if (rc->extents_found == 0)
4389			break;
4390
4391		btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4392
 
 
 
 
 
 
 
 
 
 
4393		if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4394			ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4395						       (u64)-1);
4396			if (ret) {
4397				err = ret;
4398				goto out;
4399			}
4400			invalidate_mapping_pages(rc->data_inode->i_mapping,
4401						 0, -1);
4402			rc->stage = UPDATE_DATA_PTRS;
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);
4580out_free:
4581	kfree(rc);
 
 
 
 
 
 
 
4582out:
4583	if (!list_empty(&reloc_roots))
4584		free_reloc_roots(&reloc_roots);
4585
4586	btrfs_free_path(path);
4587
4588	if (err == 0) {
4589		/* cleanup orphan inode in data relocation tree */
4590		fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4591		if (IS_ERR(fs_root))
4592			err = PTR_ERR(fs_root);
4593		else
4594			err = btrfs_orphan_cleanup(fs_root);
4595	}
4596	return err;
4597}
4598
4599/*
4600 * helper to add ordered checksum for data relocation.
4601 *
4602 * cloning checksum properly handles the nodatasum extents.
4603 * it also saves CPU time to re-calculate the checksum.
4604 */
4605int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4606{
4607	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
4608	struct btrfs_ordered_sum *sums;
4609	struct btrfs_ordered_extent *ordered;
4610	int ret;
4611	u64 disk_bytenr;
4612	u64 new_bytenr;
4613	LIST_HEAD(list);
 
4614
4615	ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4616	BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4617
4618	disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4619	ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4620				       disk_bytenr + len - 1, &list, 0);
4621	if (ret)
4622		goto out;
4623
4624	while (!list_empty(&list)) {
4625		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
 
 
4626		list_del_init(&sums->list);
4627
4628		/*
4629		 * We need to offset the new_bytenr based on where the csum is.
4630		 * We need to do this because we will read in entire prealloc
4631		 * extents but we may have written to say the middle of the
4632		 * prealloc extent, so we need to make sure the csum goes with
4633		 * the right disk offset.
4634		 *
4635		 * We can do this because the data reloc inode refers strictly
4636		 * to the on disk bytes, so we don't have to worry about
4637		 * disk_len vs real len like with real inodes since it's all
4638		 * disk length.
4639		 */
4640		new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4641		sums->bytenr = new_bytenr;
4642
4643		btrfs_add_ordered_sum(inode, ordered, sums);
4644	}
4645out:
4646	btrfs_put_ordered_extent(ordered);
4647	return ret;
4648}
4649
4650int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4651			  struct btrfs_root *root, struct extent_buffer *buf,
 
4652			  struct extent_buffer *cow)
4653{
4654	struct btrfs_fs_info *fs_info = root->fs_info;
4655	struct reloc_control *rc;
4656	struct backref_node *node;
4657	int first_cow = 0;
4658	int level;
4659	int ret = 0;
4660
4661	rc = fs_info->reloc_ctl;
4662	if (!rc)
4663		return 0;
4664
4665	BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4666	       root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4667
4668	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4669		if (buf == root->node)
4670			__update_reloc_root(root, cow->start);
4671	}
4672
4673	level = btrfs_header_level(buf);
4674	if (btrfs_header_generation(buf) <=
4675	    btrfs_root_last_snapshot(&root->root_item))
4676		first_cow = 1;
4677
4678	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4679	    rc->create_reloc_tree) {
4680		WARN_ON(!first_cow && level == 0);
4681
4682		node = rc->backref_cache.path[level];
4683		BUG_ON(node->bytenr != buf->start &&
4684		       node->new_bytenr != buf->start);
4685
4686		drop_node_buffer(node);
4687		extent_buffer_get(cow);
 
 
 
 
 
 
 
 
 
 
 
 
4688		node->eb = cow;
4689		node->new_bytenr = cow->start;
4690
4691		if (!node->pending) {
4692			list_move_tail(&node->list,
4693				       &rc->backref_cache.pending[level]);
4694			node->pending = 1;
4695		}
4696
4697		if (first_cow)
4698			__mark_block_processed(rc, node);
4699
4700		if (first_cow && level > 0)
4701			rc->nodes_relocated += buf->len;
4702	}
4703
4704	if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4705		ret = replace_file_extents(trans, rc, root, cow);
4706	return ret;
4707}
4708
4709/*
4710 * called before creating snapshot. it calculates metadata reservation
4711 * required for relocating tree blocks in the snapshot
4712 */
4713void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4714			      u64 *bytes_to_reserve)
4715{
4716	struct btrfs_root *root;
4717	struct reloc_control *rc;
4718
4719	root = pending->root;
4720	if (!root->reloc_root)
4721		return;
4722
4723	rc = root->fs_info->reloc_ctl;
4724	if (!rc->merge_reloc_tree)
4725		return;
4726
4727	root = root->reloc_root;
4728	BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4729	/*
4730	 * relocation is in the stage of merging trees. the space
4731	 * used by merging a reloc tree is twice the size of
4732	 * relocated tree nodes in the worst case. half for cowing
4733	 * the reloc tree, half for cowing the fs tree. the space
4734	 * used by cowing the reloc tree will be freed after the
4735	 * tree is dropped. if we create snapshot, cowing the fs
4736	 * tree may use more space than it frees. so we need
4737	 * reserve extra space.
4738	 */
4739	*bytes_to_reserve += rc->nodes_relocated;
4740}
4741
4742/*
4743 * called after snapshot is created. migrate block reservation
4744 * and create reloc root for the newly created snapshot
 
 
 
 
4745 */
4746int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4747			       struct btrfs_pending_snapshot *pending)
4748{
4749	struct btrfs_root *root = pending->root;
4750	struct btrfs_root *reloc_root;
4751	struct btrfs_root *new_root;
4752	struct reloc_control *rc;
4753	int ret;
4754
4755	if (!root->reloc_root)
4756		return 0;
4757
4758	rc = root->fs_info->reloc_ctl;
4759	rc->merging_rsv_size += rc->nodes_relocated;
4760
4761	if (rc->merge_reloc_tree) {
4762		ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4763					      rc->block_rsv,
4764					      rc->nodes_relocated, 1);
4765		if (ret)
4766			return ret;
4767	}
4768
4769	new_root = pending->snap;
4770	reloc_root = create_reloc_root(trans, root->reloc_root,
4771				       new_root->root_key.objectid);
4772	if (IS_ERR(reloc_root))
4773		return PTR_ERR(reloc_root);
4774
4775	ret = __add_reloc_root(reloc_root);
4776	BUG_ON(ret < 0);
4777	new_root->reloc_root = reloc_root;
 
 
 
 
 
4778
4779	if (rc->create_reloc_tree)
4780		ret = clone_backref_node(trans, rc, root, reloc_root);
4781	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4782}
v6.13.7
   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#include "raid-stripe-tree.h"
  40
  41/*
  42 * Relocation overview
  43 *
  44 * [What does relocation do]
  45 *
  46 * The objective of relocation is to relocate all extents of the target block
  47 * group to other block groups.
  48 * This is utilized by resize (shrink only), profile converting, compacting
  49 * space, or balance routine to spread chunks over devices.
  50 *
  51 * 		Before		|		After
  52 * ------------------------------------------------------------------
  53 *  BG A: 10 data extents	| BG A: deleted
  54 *  BG B:  2 data extents	| BG B: 10 data extents (2 old + 8 relocated)
  55 *  BG C:  1 extents		| BG C:  3 data extents (1 old + 2 relocated)
  56 *
  57 * [How does relocation work]
  58 *
  59 * 1.   Mark the target block group read-only
  60 *      New extents won't be allocated from the target block group.
  61 *
  62 * 2.1  Record each extent in the target block group
  63 *      To build a proper map of extents to be relocated.
  64 *
  65 * 2.2  Build data reloc tree and reloc trees
  66 *      Data reloc tree will contain an inode, recording all newly relocated
  67 *      data extents.
  68 *      There will be only one data reloc tree for one data block group.
  69 *
  70 *      Reloc tree will be a special snapshot of its source tree, containing
  71 *      relocated tree blocks.
  72 *      Each tree referring to a tree block in target block group will get its
  73 *      reloc tree built.
  74 *
  75 * 2.3  Swap source tree with its corresponding reloc tree
  76 *      Each involved tree only refers to new extents after swap.
  77 *
  78 * 3.   Cleanup reloc trees and data reloc tree.
  79 *      As old extents in the target block group are still referenced by reloc
  80 *      trees, we need to clean them up before really freeing the target block
  81 *      group.
  82 *
  83 * The main complexity is in steps 2.2 and 2.3.
  84 *
  85 * The entry point of relocation is relocate_block_group() function.
 
 
 
 
 
 
 
 
 
  86 */
 
 
 
 
  87
 
 
  88#define RELOCATION_RESERVED_NODES	256
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  89/*
  90 * map address of tree root to tree
  91 */
  92struct mapping_node {
  93	struct {
  94		struct rb_node rb_node;
  95		u64 bytenr;
  96	}; /* Use rb_simle_node for search/insert */
  97	void *data;
  98};
  99
 100struct mapping_tree {
 101	struct rb_root rb_root;
 102	spinlock_t lock;
 103};
 104
 105/*
 106 * present a tree block to process
 107 */
 108struct tree_block {
 109	struct {
 110		struct rb_node rb_node;
 111		u64 bytenr;
 112	}; /* Use rb_simple_node for search/insert */
 113	u64 owner;
 114	struct btrfs_key key;
 115	u8 level;
 116	bool key_ready;
 117};
 118
 119#define MAX_EXTENTS 128
 120
 121struct file_extent_cluster {
 122	u64 start;
 123	u64 end;
 124	u64 boundary[MAX_EXTENTS];
 125	unsigned int nr;
 126	u64 owning_root;
 127};
 128
 129/* Stages of data relocation. */
 130enum reloc_stage {
 131	MOVE_DATA_EXTENTS,
 132	UPDATE_DATA_PTRS
 133};
 134
 135struct reloc_control {
 136	/* block group to relocate */
 137	struct btrfs_block_group *block_group;
 138	/* extent tree */
 139	struct btrfs_root *extent_root;
 140	/* inode for moving data */
 141	struct inode *data_inode;
 142
 143	struct btrfs_block_rsv *block_rsv;
 144
 145	struct btrfs_backref_cache backref_cache;
 146
 147	struct file_extent_cluster cluster;
 148	/* tree blocks have been processed */
 149	struct extent_io_tree processed_blocks;
 150	/* map start of tree root to corresponding reloc tree */
 151	struct mapping_tree reloc_root_tree;
 152	/* list of reloc trees */
 153	struct list_head reloc_roots;
 154	/* list of subvolume trees that get relocated */
 155	struct list_head dirty_subvol_roots;
 156	/* size of metadata reservation for merging reloc trees */
 157	u64 merging_rsv_size;
 158	/* size of relocated tree nodes */
 159	u64 nodes_relocated;
 160	/* reserved size for block group relocation*/
 161	u64 reserved_bytes;
 162
 163	u64 search_start;
 164	u64 extents_found;
 165
 166	enum reloc_stage stage;
 167	bool create_reloc_tree;
 168	bool merge_reloc_tree;
 169	bool found_file_extent;
 170};
 171
 172static void mark_block_processed(struct reloc_control *rc,
 173				 struct btrfs_backref_node *node)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 174{
 175	u32 blocksize;
 
 
 
 
 176
 177	if (node->level == 0 ||
 178	    in_range(node->bytenr, rc->block_group->start,
 179		     rc->block_group->length)) {
 180		blocksize = rc->extent_root->fs_info->nodesize;
 181		set_extent_bit(&rc->processed_blocks, node->bytenr,
 182			       node->bytenr + blocksize - 1, EXTENT_DIRTY, NULL);
 183	}
 184	node->processed = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 185}
 186
 187/*
 188 * walk up backref nodes until reach node presents tree root
 189 */
 190static struct btrfs_backref_node *walk_up_backref(
 191		struct btrfs_backref_node *node,
 192		struct btrfs_backref_edge *edges[], int *index)
 193{
 194	struct btrfs_backref_edge *edge;
 195	int idx = *index;
 196
 197	while (!list_empty(&node->upper)) {
 198		edge = list_entry(node->upper.next,
 199				  struct btrfs_backref_edge, list[LOWER]);
 200		edges[idx++] = edge;
 201		node = edge->node[UPPER];
 202	}
 203	BUG_ON(node->detached);
 204	*index = idx;
 205	return node;
 206}
 207
 208/*
 209 * walk down backref nodes to find start of next reference path
 210 */
 211static struct btrfs_backref_node *walk_down_backref(
 212		struct btrfs_backref_edge *edges[], int *index)
 213{
 214	struct btrfs_backref_edge *edge;
 215	struct btrfs_backref_node *lower;
 216	int idx = *index;
 217
 218	while (idx > 0) {
 219		edge = edges[idx - 1];
 220		lower = edge->node[LOWER];
 221		if (list_is_last(&edge->list[LOWER], &lower->upper)) {
 222			idx--;
 223			continue;
 224		}
 225		edge = list_entry(edge->list[LOWER].next,
 226				  struct btrfs_backref_edge, list[LOWER]);
 227		edges[idx - 1] = edge;
 228		*index = idx;
 229		return edge->node[UPPER];
 230	}
 231	*index = 0;
 232	return NULL;
 233}
 234
 235static bool reloc_root_is_dead(const struct btrfs_root *root)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 236{
 237	/*
 238	 * Pair with set_bit/clear_bit in clean_dirty_subvols and
 239	 * btrfs_update_reloc_root. We need to see the updated bit before
 240	 * trying to access reloc_root
 241	 */
 242	smp_rmb();
 243	if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
 244		return true;
 245	return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 246}
 247
 248/*
 249 * Check if this subvolume tree has valid reloc tree.
 250 *
 251 * Reloc tree after swap is considered dead, thus not considered as valid.
 252 * This is enough for most callers, as they don't distinguish dead reloc root
 253 * from no reloc root.  But btrfs_should_ignore_reloc_root() below is a
 254 * special case.
 255 */
 256static bool have_reloc_root(const struct btrfs_root *root)
 
 257{
 258	if (reloc_root_is_dead(root))
 259		return false;
 260	if (!root->reloc_root)
 261		return false;
 262	return true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 263}
 264
 265bool btrfs_should_ignore_reloc_root(const struct btrfs_root *root)
 
 266{
 267	struct btrfs_root *reloc_root;
 268
 269	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 270		return false;
 271
 272	/* This root has been merged with its reloc tree, we can ignore it */
 273	if (reloc_root_is_dead(root))
 274		return true;
 275
 276	reloc_root = root->reloc_root;
 277	if (!reloc_root)
 278		return false;
 279
 280	if (btrfs_header_generation(reloc_root->commit_root) ==
 281	    root->fs_info->running_transaction->transid)
 282		return false;
 283	/*
 284	 * If there is reloc tree and it was created in previous transaction
 285	 * backref lookup can find the reloc tree, so backref node for the fs
 286	 * tree root is useless for relocation.
 
 287	 */
 288	return true;
 289}
 290
 291/*
 292 * find reloc tree by address of tree root
 293 */
 294struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 
 295{
 296	struct reloc_control *rc = fs_info->reloc_ctl;
 297	struct rb_node *rb_node;
 298	struct mapping_node *node;
 299	struct btrfs_root *root = NULL;
 300
 301	ASSERT(rc);
 302	spin_lock(&rc->reloc_root_tree.lock);
 303	rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
 304	if (rb_node) {
 305		node = rb_entry(rb_node, struct mapping_node, rb_node);
 306		root = node->data;
 307	}
 308	spin_unlock(&rc->reloc_root_tree.lock);
 309	return btrfs_grab_root(root);
 310}
 311
 312/*
 313 * For useless nodes, do two major clean ups:
 314 *
 315 * - Cleanup the children edges and nodes
 316 *   If child node is also orphan (no parent) during cleanup, then the child
 317 *   node will also be cleaned up.
 318 *
 319 * - Freeing up leaves (level 0), keeps nodes detached
 320 *   For nodes, the node is still cached as "detached"
 321 *
 322 * Return false if @node is not in the @useless_nodes list.
 323 * Return true if @node is in the @useless_nodes list.
 324 */
 325static bool handle_useless_nodes(struct reloc_control *rc,
 326				 struct btrfs_backref_node *node)
 327{
 328	struct btrfs_backref_cache *cache = &rc->backref_cache;
 329	struct list_head *useless_node = &cache->useless_node;
 330	bool ret = false;
 
 
 
 
 
 
 
 
 
 331
 332	while (!list_empty(useless_node)) {
 333		struct btrfs_backref_node *cur;
 
 
 334
 335		cur = list_first_entry(useless_node, struct btrfs_backref_node,
 336				 list);
 337		list_del_init(&cur->list);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 338
 339		/* Only tree root nodes can be added to @useless_nodes */
 340		ASSERT(list_empty(&cur->upper));
 341
 342		if (cur == node)
 343			ret = true;
 344
 345		/* The node is the lowest node */
 346		if (cur->lowest) {
 347			list_del_init(&cur->lower);
 348			cur->lowest = 0;
 349		}
 350
 351		/* Cleanup the lower edges */
 352		while (!list_empty(&cur->lower)) {
 353			struct btrfs_backref_edge *edge;
 354			struct btrfs_backref_node *lower;
 355
 356			edge = list_entry(cur->lower.next,
 357					struct btrfs_backref_edge, list[UPPER]);
 358			list_del(&edge->list[UPPER]);
 359			list_del(&edge->list[LOWER]);
 360			lower = edge->node[LOWER];
 361			btrfs_backref_free_edge(cache, edge);
 
 
 
 
 362
 363			/* Child node is also orphan, queue for cleanup */
 364			if (list_empty(&lower->upper))
 365				list_add(&lower->list, useless_node);
 366		}
 367		/* Mark this block processed for relocation */
 368		mark_block_processed(rc, cur);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 369
 370		/*
 371		 * Backref nodes for tree leaves are deleted from the cache.
 372		 * Backref nodes for upper level tree blocks are left in the
 373		 * cache to avoid unnecessary backref lookup.
 374		 */
 375		if (cur->level > 0) {
 376			list_add(&cur->list, &cache->detached);
 377			cur->detached = 1;
 378		} else {
 379			rb_erase(&cur->rb_node, &cache->rb_root);
 380			btrfs_backref_free_node(cache, cur);
 381		}
 382	}
 383	return ret;
 
 384}
 385
 386/*
 387 * Build backref tree for a given tree block. Root of the backref tree
 388 * corresponds the tree block, leaves of the backref tree correspond roots of
 389 * b-trees that reference the tree block.
 390 *
 391 * The basic idea of this function is check backrefs of a given block to find
 392 * upper level blocks that reference the block, and then check backrefs of
 393 * these upper level blocks recursively. The recursion stops when tree root is
 394 * reached or backrefs for the block is cached.
 395 *
 396 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
 397 * all upper level blocks that directly/indirectly reference the block are also
 398 * cached.
 399 */
 400static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
 401			struct btrfs_trans_handle *trans,
 402			struct reloc_control *rc, struct btrfs_key *node_key,
 403			int level, u64 bytenr)
 404{
 405	struct btrfs_backref_iter *iter;
 406	struct btrfs_backref_cache *cache = &rc->backref_cache;
 407	/* For searching parent of TREE_BLOCK_REF */
 408	struct btrfs_path *path;
 409	struct btrfs_backref_node *cur;
 410	struct btrfs_backref_node *node = NULL;
 411	struct btrfs_backref_edge *edge;
 
 
 
 
 
 
 
 
 
 
 
 412	int ret;
 
 
 413
 414	iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info);
 415	if (!iter)
 416		return ERR_PTR(-ENOMEM);
 417	path = btrfs_alloc_path();
 418	if (!path) {
 419		ret = -ENOMEM;
 420		goto out;
 421	}
 
 
 422
 423	node = btrfs_backref_alloc_node(cache, bytenr, level);
 424	if (!node) {
 425		ret = -ENOMEM;
 426		goto out;
 427	}
 428
 
 
 429	node->lowest = 1;
 430	cur = node;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 431
 432	/* Breadth-first search to build backref cache */
 433	do {
 434		ret = btrfs_backref_add_tree_node(trans, cache, path, iter,
 435						  node_key, cur);
 436		if (ret < 0)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 437			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 438
 439		edge = list_first_entry_or_null(&cache->pending_edge,
 440				struct btrfs_backref_edge, list[UPPER]);
 441		/*
 442		 * The pending list isn't empty, take the first block to
 443		 * process
 444		 */
 445		if (edge) {
 446			list_del_init(&edge->list[UPPER]);
 447			cur = edge->node[UPPER];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 448		}
 449	} while (edge);
 450
 451	/* Finish the upper linkage of newly added edges/nodes */
 452	ret = btrfs_backref_finish_upper_links(cache, node);
 453	if (ret < 0)
 454		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 455
 456	if (handle_useless_nodes(rc, node))
 457		node = NULL;
 
 
 
 
 
 
 
 
 
 
 458out:
 459	btrfs_free_path(iter->path);
 460	kfree(iter);
 461	btrfs_free_path(path);
 462	if (ret) {
 463		btrfs_backref_error_cleanup(cache, node);
 464		return ERR_PTR(ret);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 465	}
 466	ASSERT(!node || !node->detached);
 467	ASSERT(list_empty(&cache->useless_node) &&
 468	       list_empty(&cache->pending_edge));
 469	return node;
 470}
 471
 472/*
 473 * helper to add backref node for the newly created snapshot.
 474 * the backref node is created by cloning backref node that
 475 * corresponds to root of source tree
 476 */
 477static int clone_backref_node(struct btrfs_trans_handle *trans,
 478			      struct reloc_control *rc,
 479			      const struct btrfs_root *src,
 480			      struct btrfs_root *dest)
 481{
 482	struct btrfs_root *reloc_root = src->reloc_root;
 483	struct btrfs_backref_cache *cache = &rc->backref_cache;
 484	struct btrfs_backref_node *node = NULL;
 485	struct btrfs_backref_node *new_node;
 486	struct btrfs_backref_edge *edge;
 487	struct btrfs_backref_edge *new_edge;
 488	struct rb_node *rb_node;
 489
 490	rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
 
 
 
 491	if (rb_node) {
 492		node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
 493		if (node->detached)
 494			node = NULL;
 495		else
 496			BUG_ON(node->new_bytenr != reloc_root->node->start);
 497	}
 498
 499	if (!node) {
 500		rb_node = rb_simple_search(&cache->rb_root,
 501					   reloc_root->commit_root->start);
 502		if (rb_node) {
 503			node = rb_entry(rb_node, struct btrfs_backref_node,
 504					rb_node);
 505			BUG_ON(node->detached);
 506		}
 507	}
 508
 509	if (!node)
 510		return 0;
 511
 512	new_node = btrfs_backref_alloc_node(cache, dest->node->start,
 513					    node->level);
 514	if (!new_node)
 515		return -ENOMEM;
 516
 
 
 517	new_node->lowest = node->lowest;
 518	new_node->checked = 1;
 519	new_node->root = btrfs_grab_root(dest);
 520	ASSERT(new_node->root);
 521
 522	if (!node->lowest) {
 523		list_for_each_entry(edge, &node->lower, list[UPPER]) {
 524			new_edge = btrfs_backref_alloc_edge(cache);
 525			if (!new_edge)
 526				goto fail;
 527
 528			btrfs_backref_link_edge(new_edge, edge->node[LOWER],
 529						new_node, LINK_UPPER);
 
 
 530		}
 531	} else {
 532		list_add_tail(&new_node->lower, &cache->leaves);
 533	}
 534
 535	rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
 536				   &new_node->rb_node);
 537	if (rb_node)
 538		btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
 539
 540	if (!new_node->lowest) {
 541		list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
 542			list_add_tail(&new_edge->list[LOWER],
 543				      &new_edge->node[LOWER]->upper);
 544		}
 545	}
 546	return 0;
 547fail:
 548	while (!list_empty(&new_node->lower)) {
 549		new_edge = list_entry(new_node->lower.next,
 550				      struct btrfs_backref_edge, list[UPPER]);
 551		list_del(&new_edge->list[UPPER]);
 552		btrfs_backref_free_edge(cache, new_edge);
 553	}
 554	btrfs_backref_free_node(cache, new_node);
 555	return -ENOMEM;
 556}
 557
 558/*
 559 * helper to add 'address of tree root -> reloc tree' mapping
 560 */
 561static int __add_reloc_root(struct btrfs_root *root)
 562{
 563	struct btrfs_fs_info *fs_info = root->fs_info;
 564	struct rb_node *rb_node;
 565	struct mapping_node *node;
 566	struct reloc_control *rc = fs_info->reloc_ctl;
 567
 568	node = kmalloc(sizeof(*node), GFP_NOFS);
 569	if (!node)
 570		return -ENOMEM;
 571
 572	node->bytenr = root->commit_root->start;
 573	node->data = root;
 574
 575	spin_lock(&rc->reloc_root_tree.lock);
 576	rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
 577				   node->bytenr, &node->rb_node);
 578	spin_unlock(&rc->reloc_root_tree.lock);
 579	if (rb_node) {
 580		btrfs_err(fs_info,
 581			    "Duplicate root found for start=%llu while inserting into relocation tree",
 582			    node->bytenr);
 
 583		return -EEXIST;
 584	}
 585
 586	list_add_tail(&root->root_list, &rc->reloc_roots);
 587	return 0;
 588}
 589
 590/*
 591 * helper to delete the 'address of tree root -> reloc tree'
 592 * mapping
 593 */
 594static void __del_reloc_root(struct btrfs_root *root)
 595{
 596	struct btrfs_fs_info *fs_info = root->fs_info;
 597	struct rb_node *rb_node;
 598	struct mapping_node *node = NULL;
 599	struct reloc_control *rc = fs_info->reloc_ctl;
 600	bool put_ref = false;
 601
 602	if (rc && root->node) {
 603		spin_lock(&rc->reloc_root_tree.lock);
 604		rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
 605					   root->commit_root->start);
 606		if (rb_node) {
 607			node = rb_entry(rb_node, struct mapping_node, rb_node);
 608			rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 609			RB_CLEAR_NODE(&node->rb_node);
 610		}
 611		spin_unlock(&rc->reloc_root_tree.lock);
 612		ASSERT(!node || (struct btrfs_root *)node->data == root);
 613	}
 
 
 
 
 
 614
 615	/*
 616	 * We only put the reloc root here if it's on the list.  There's a lot
 617	 * of places where the pattern is to splice the rc->reloc_roots, process
 618	 * the reloc roots, and then add the reloc root back onto
 619	 * rc->reloc_roots.  If we call __del_reloc_root while it's off of the
 620	 * list we don't want the reference being dropped, because the guy
 621	 * messing with the list is in charge of the reference.
 622	 */
 623	spin_lock(&fs_info->trans_lock);
 624	if (!list_empty(&root->root_list)) {
 625		put_ref = true;
 626		list_del_init(&root->root_list);
 627	}
 628	spin_unlock(&fs_info->trans_lock);
 629	if (put_ref)
 630		btrfs_put_root(root);
 631	kfree(node);
 632}
 633
 634/*
 635 * helper to update the 'address of tree root -> reloc tree'
 636 * mapping
 637 */
 638static int __update_reloc_root(struct btrfs_root *root)
 639{
 640	struct btrfs_fs_info *fs_info = root->fs_info;
 641	struct rb_node *rb_node;
 642	struct mapping_node *node = NULL;
 643	struct reloc_control *rc = fs_info->reloc_ctl;
 644
 645	spin_lock(&rc->reloc_root_tree.lock);
 646	rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
 647				   root->commit_root->start);
 648	if (rb_node) {
 649		node = rb_entry(rb_node, struct mapping_node, rb_node);
 650		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 651	}
 652	spin_unlock(&rc->reloc_root_tree.lock);
 653
 654	if (!node)
 655		return 0;
 656	BUG_ON((struct btrfs_root *)node->data != root);
 657
 658	spin_lock(&rc->reloc_root_tree.lock);
 659	node->bytenr = root->node->start;
 660	rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
 661				   node->bytenr, &node->rb_node);
 662	spin_unlock(&rc->reloc_root_tree.lock);
 663	if (rb_node)
 664		btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
 665	return 0;
 666}
 667
 668static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
 669					struct btrfs_root *root, u64 objectid)
 670{
 671	struct btrfs_fs_info *fs_info = root->fs_info;
 672	struct btrfs_root *reloc_root;
 673	struct extent_buffer *eb;
 674	struct btrfs_root_item *root_item;
 675	struct btrfs_key root_key;
 676	int ret = 0;
 677	bool must_abort = false;
 678
 679	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
 680	if (!root_item)
 681		return ERR_PTR(-ENOMEM);
 682
 683	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
 684	root_key.type = BTRFS_ROOT_ITEM_KEY;
 685	root_key.offset = objectid;
 686
 687	if (btrfs_root_id(root) == objectid) {
 688		u64 commit_root_gen;
 689
 690		/* called by btrfs_init_reloc_root */
 691		ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
 692				      BTRFS_TREE_RELOC_OBJECTID);
 693		if (ret)
 694			goto fail;
 695
 696		/*
 697		 * Set the last_snapshot field to the generation of the commit
 698		 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
 699		 * correctly (returns true) when the relocation root is created
 700		 * either inside the critical section of a transaction commit
 701		 * (through transaction.c:qgroup_account_snapshot()) and when
 702		 * it's created before the transaction commit is started.
 703		 */
 704		commit_root_gen = btrfs_header_generation(root->commit_root);
 705		btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
 706	} else {
 707		/*
 708		 * called by btrfs_reloc_post_snapshot_hook.
 709		 * the source tree is a reloc tree, all tree blocks
 710		 * modified after it was created have RELOC flag
 711		 * set in their headers. so it's OK to not update
 712		 * the 'last_snapshot'.
 713		 */
 714		ret = btrfs_copy_root(trans, root, root->node, &eb,
 715				      BTRFS_TREE_RELOC_OBJECTID);
 716		if (ret)
 717			goto fail;
 718	}
 719
 720	/*
 721	 * We have changed references at this point, we must abort the
 722	 * transaction if anything fails.
 723	 */
 724	must_abort = true;
 725
 726	memcpy(root_item, &root->root_item, sizeof(*root_item));
 727	btrfs_set_root_bytenr(root_item, eb->start);
 728	btrfs_set_root_level(root_item, btrfs_header_level(eb));
 729	btrfs_set_root_generation(root_item, trans->transid);
 730
 731	if (btrfs_root_id(root) == objectid) {
 732		btrfs_set_root_refs(root_item, 0);
 733		memset(&root_item->drop_progress, 0,
 734		       sizeof(struct btrfs_disk_key));
 735		btrfs_set_root_drop_level(root_item, 0);
 736	}
 737
 738	btrfs_tree_unlock(eb);
 739	free_extent_buffer(eb);
 740
 741	ret = btrfs_insert_root(trans, fs_info->tree_root,
 742				&root_key, root_item);
 743	if (ret)
 744		goto fail;
 745
 746	kfree(root_item);
 747
 748	reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
 749	if (IS_ERR(reloc_root)) {
 750		ret = PTR_ERR(reloc_root);
 751		goto abort;
 752	}
 753	set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
 754	btrfs_set_root_last_trans(reloc_root, trans->transid);
 755	return reloc_root;
 756fail:
 757	kfree(root_item);
 758abort:
 759	if (must_abort)
 760		btrfs_abort_transaction(trans, ret);
 761	return ERR_PTR(ret);
 762}
 763
 764/*
 765 * create reloc tree for a given fs tree. reloc tree is just a
 766 * snapshot of the fs tree with special root objectid.
 767 *
 768 * The reloc_root comes out of here with two references, one for
 769 * root->reloc_root, and another for being on the rc->reloc_roots list.
 770 */
 771int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
 772			  struct btrfs_root *root)
 773{
 774	struct btrfs_fs_info *fs_info = root->fs_info;
 775	struct btrfs_root *reloc_root;
 776	struct reloc_control *rc = fs_info->reloc_ctl;
 777	struct btrfs_block_rsv *rsv;
 778	int clear_rsv = 0;
 779	int ret;
 780
 781	if (!rc)
 782		return 0;
 783
 784	/*
 785	 * The subvolume has reloc tree but the swap is finished, no need to
 786	 * create/update the dead reloc tree
 787	 */
 788	if (reloc_root_is_dead(root))
 789		return 0;
 790
 791	/*
 792	 * This is subtle but important.  We do not do
 793	 * record_root_in_transaction for reloc roots, instead we record their
 794	 * corresponding fs root, and then here we update the last trans for the
 795	 * reloc root.  This means that we have to do this for the entire life
 796	 * of the reloc root, regardless of which stage of the relocation we are
 797	 * in.
 798	 */
 799	if (root->reloc_root) {
 800		reloc_root = root->reloc_root;
 801		btrfs_set_root_last_trans(reloc_root, trans->transid);
 802		return 0;
 803	}
 804
 805	/*
 806	 * We are merging reloc roots, we do not need new reloc trees.  Also
 807	 * reloc trees never need their own reloc tree.
 808	 */
 809	if (!rc->create_reloc_tree || btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID)
 810		return 0;
 811
 812	if (!trans->reloc_reserved) {
 813		rsv = trans->block_rsv;
 814		trans->block_rsv = rc->block_rsv;
 815		clear_rsv = 1;
 816	}
 817	reloc_root = create_reloc_root(trans, root, btrfs_root_id(root));
 818	if (clear_rsv)
 819		trans->block_rsv = rsv;
 820	if (IS_ERR(reloc_root))
 821		return PTR_ERR(reloc_root);
 822
 823	ret = __add_reloc_root(reloc_root);
 824	ASSERT(ret != -EEXIST);
 825	if (ret) {
 826		/* Pairs with create_reloc_root */
 827		btrfs_put_root(reloc_root);
 828		return ret;
 829	}
 830	root->reloc_root = btrfs_grab_root(reloc_root);
 831	return 0;
 832}
 833
 834/*
 835 * update root item of reloc tree
 836 */
 837int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
 838			    struct btrfs_root *root)
 839{
 840	struct btrfs_fs_info *fs_info = root->fs_info;
 841	struct btrfs_root *reloc_root;
 842	struct btrfs_root_item *root_item;
 843	int ret;
 844
 845	if (!have_reloc_root(root))
 846		return 0;
 847
 848	reloc_root = root->reloc_root;
 849	root_item = &reloc_root->root_item;
 850
 851	/*
 852	 * We are probably ok here, but __del_reloc_root() will drop its ref of
 853	 * the root.  We have the ref for root->reloc_root, but just in case
 854	 * hold it while we update the reloc root.
 855	 */
 856	btrfs_grab_root(reloc_root);
 857
 858	/* root->reloc_root will stay until current relocation finished */
 859	if (fs_info->reloc_ctl && fs_info->reloc_ctl->merge_reloc_tree &&
 860	    btrfs_root_refs(root_item) == 0) {
 861		set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
 862		/*
 863		 * Mark the tree as dead before we change reloc_root so
 864		 * have_reloc_root will not touch it from now on.
 865		 */
 866		smp_wmb();
 867		__del_reloc_root(reloc_root);
 868	}
 869
 870	if (reloc_root->commit_root != reloc_root->node) {
 871		__update_reloc_root(reloc_root);
 872		btrfs_set_root_node(root_item, reloc_root->node);
 873		free_extent_buffer(reloc_root->commit_root);
 874		reloc_root->commit_root = btrfs_root_node(reloc_root);
 875	}
 876
 877	ret = btrfs_update_root(trans, fs_info->tree_root,
 878				&reloc_root->root_key, root_item);
 879	btrfs_put_root(reloc_root);
 880	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 881}
 882
 883/*
 884 * get new location of data
 885 */
 886static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
 887			    u64 bytenr, u64 num_bytes)
 888{
 889	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
 890	struct btrfs_path *path;
 891	struct btrfs_file_extent_item *fi;
 892	struct extent_buffer *leaf;
 893	int ret;
 894
 895	path = btrfs_alloc_path();
 896	if (!path)
 897		return -ENOMEM;
 898
 899	bytenr -= BTRFS_I(reloc_inode)->reloc_block_group_start;
 900	ret = btrfs_lookup_file_extent(NULL, root, path,
 901			btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
 902	if (ret < 0)
 903		goto out;
 904	if (ret > 0) {
 905		ret = -ENOENT;
 906		goto out;
 907	}
 908
 909	leaf = path->nodes[0];
 910	fi = btrfs_item_ptr(leaf, path->slots[0],
 911			    struct btrfs_file_extent_item);
 912
 913	BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
 914	       btrfs_file_extent_compression(leaf, fi) ||
 915	       btrfs_file_extent_encryption(leaf, fi) ||
 916	       btrfs_file_extent_other_encoding(leaf, fi));
 917
 918	if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
 919		ret = -EINVAL;
 920		goto out;
 921	}
 922
 923	*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 924	ret = 0;
 925out:
 926	btrfs_free_path(path);
 927	return ret;
 928}
 929
 930/*
 931 * update file extent items in the tree leaf to point to
 932 * the new locations.
 933 */
 934static noinline_for_stack
 935int replace_file_extents(struct btrfs_trans_handle *trans,
 936			 struct reloc_control *rc,
 937			 struct btrfs_root *root,
 938			 struct extent_buffer *leaf)
 939{
 940	struct btrfs_fs_info *fs_info = root->fs_info;
 941	struct btrfs_key key;
 942	struct btrfs_file_extent_item *fi;
 943	struct btrfs_inode *inode = NULL;
 944	u64 parent;
 945	u64 bytenr;
 946	u64 new_bytenr = 0;
 947	u64 num_bytes;
 948	u64 end;
 949	u32 nritems;
 950	u32 i;
 951	int ret = 0;
 952	int first = 1;
 953	int dirty = 0;
 954
 955	if (rc->stage != UPDATE_DATA_PTRS)
 956		return 0;
 957
 958	/* reloc trees always use full backref */
 959	if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID)
 960		parent = leaf->start;
 961	else
 962		parent = 0;
 963
 964	nritems = btrfs_header_nritems(leaf);
 965	for (i = 0; i < nritems; i++) {
 966		struct btrfs_ref ref = { 0 };
 967
 968		cond_resched();
 969		btrfs_item_key_to_cpu(leaf, &key, i);
 970		if (key.type != BTRFS_EXTENT_DATA_KEY)
 971			continue;
 972		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
 973		if (btrfs_file_extent_type(leaf, fi) ==
 974		    BTRFS_FILE_EXTENT_INLINE)
 975			continue;
 976		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 977		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 978		if (bytenr == 0)
 979			continue;
 980		if (!in_range(bytenr, rc->block_group->start,
 981			      rc->block_group->length))
 982			continue;
 983
 984		/*
 985		 * if we are modifying block in fs tree, wait for read_folio
 986		 * to complete and drop the extent cache
 987		 */
 988		if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID) {
 989			if (first) {
 990				inode = btrfs_find_first_inode(root, key.objectid);
 991				first = 0;
 992			} else if (inode && btrfs_ino(inode) < key.objectid) {
 993				btrfs_add_delayed_iput(inode);
 994				inode = btrfs_find_first_inode(root, key.objectid);
 995			}
 996			if (inode && btrfs_ino(inode) == key.objectid) {
 997				struct extent_state *cached_state = NULL;
 998
 999				end = key.offset +
1000				      btrfs_file_extent_num_bytes(leaf, fi);
1001				WARN_ON(!IS_ALIGNED(key.offset,
1002						    fs_info->sectorsize));
1003				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1004				end--;
1005				/* Take mmap lock to serialize with reflinks. */
1006				if (!down_read_trylock(&inode->i_mmap_lock))
 
1007					continue;
1008				ret = try_lock_extent(&inode->io_tree, key.offset,
1009						      end, &cached_state);
1010				if (!ret) {
1011					up_read(&inode->i_mmap_lock);
1012					continue;
1013				}
1014
1015				btrfs_drop_extent_map_range(inode, key.offset, end, true);
1016				unlock_extent(&inode->io_tree, key.offset, end,
1017					      &cached_state);
1018				up_read(&inode->i_mmap_lock);
1019			}
1020		}
1021
1022		ret = get_new_location(rc->data_inode, &new_bytenr,
1023				       bytenr, num_bytes);
1024		if (ret) {
1025			/*
1026			 * Don't have to abort since we've not changed anything
1027			 * in the file extent yet.
1028			 */
1029			break;
1030		}
1031
1032		btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1033		dirty = 1;
1034
1035		key.offset -= btrfs_file_extent_offset(leaf, fi);
1036		ref.action = BTRFS_ADD_DELAYED_REF;
1037		ref.bytenr = new_bytenr;
1038		ref.num_bytes = num_bytes;
1039		ref.parent = parent;
1040		ref.owning_root = btrfs_root_id(root);
1041		ref.ref_root = btrfs_header_owner(leaf);
1042		btrfs_init_data_ref(&ref, key.objectid, key.offset,
1043				    btrfs_root_id(root), false);
1044		ret = btrfs_inc_extent_ref(trans, &ref);
1045		if (ret) {
1046			btrfs_abort_transaction(trans, ret);
1047			break;
1048		}
1049
1050		ref.action = BTRFS_DROP_DELAYED_REF;
1051		ref.bytenr = bytenr;
1052		ref.num_bytes = num_bytes;
1053		ref.parent = parent;
1054		ref.owning_root = btrfs_root_id(root);
1055		ref.ref_root = btrfs_header_owner(leaf);
1056		btrfs_init_data_ref(&ref, key.objectid, key.offset,
1057				    btrfs_root_id(root), false);
1058		ret = btrfs_free_extent(trans, &ref);
1059		if (ret) {
1060			btrfs_abort_transaction(trans, ret);
1061			break;
1062		}
1063	}
1064	if (dirty)
1065		btrfs_mark_buffer_dirty(trans, leaf);
1066	if (inode)
1067		btrfs_add_delayed_iput(inode);
1068	return ret;
1069}
1070
1071static noinline_for_stack int memcmp_node_keys(const struct extent_buffer *eb,
1072					       int slot, const struct btrfs_path *path,
1073					       int level)
1074{
1075	struct btrfs_disk_key key1;
1076	struct btrfs_disk_key key2;
1077	btrfs_node_key(eb, &key1, slot);
1078	btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1079	return memcmp(&key1, &key2, sizeof(key1));
1080}
1081
1082/*
1083 * try to replace tree blocks in fs tree with the new blocks
1084 * in reloc tree. tree blocks haven't been modified since the
1085 * reloc tree was create can be replaced.
1086 *
1087 * if a block was replaced, level of the block + 1 is returned.
1088 * if no block got replaced, 0 is returned. if there are other
1089 * errors, a negative error number is returned.
1090 */
1091static noinline_for_stack
1092int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1093		 struct btrfs_root *dest, struct btrfs_root *src,
1094		 struct btrfs_path *path, struct btrfs_key *next_key,
1095		 int lowest_level, int max_level)
1096{
1097	struct btrfs_fs_info *fs_info = dest->fs_info;
1098	struct extent_buffer *eb;
1099	struct extent_buffer *parent;
1100	struct btrfs_ref ref = { 0 };
1101	struct btrfs_key key;
1102	u64 old_bytenr;
1103	u64 new_bytenr;
1104	u64 old_ptr_gen;
1105	u64 new_ptr_gen;
1106	u64 last_snapshot;
1107	u32 blocksize;
1108	int cow = 0;
1109	int level;
1110	int ret;
1111	int slot;
1112
1113	ASSERT(btrfs_root_id(src) == BTRFS_TREE_RELOC_OBJECTID);
1114	ASSERT(btrfs_root_id(dest) != BTRFS_TREE_RELOC_OBJECTID);
1115
1116	last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1117again:
1118	slot = path->slots[lowest_level];
1119	btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1120
1121	eb = btrfs_lock_root_node(dest);
 
1122	level = btrfs_header_level(eb);
1123
1124	if (level < lowest_level) {
1125		btrfs_tree_unlock(eb);
1126		free_extent_buffer(eb);
1127		return 0;
1128	}
1129
1130	if (cow) {
1131		ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1132				      BTRFS_NESTING_COW);
1133		if (ret) {
1134			btrfs_tree_unlock(eb);
1135			free_extent_buffer(eb);
1136			return ret;
1137		}
1138	}
 
1139
1140	if (next_key) {
1141		next_key->objectid = (u64)-1;
1142		next_key->type = (u8)-1;
1143		next_key->offset = (u64)-1;
1144	}
1145
1146	parent = eb;
1147	while (1) {
1148		level = btrfs_header_level(parent);
1149		ASSERT(level >= lowest_level);
1150
1151		ret = btrfs_bin_search(parent, 0, &key, &slot);
1152		if (ret < 0)
1153			break;
1154		if (ret && slot > 0)
1155			slot--;
1156
1157		if (next_key && slot + 1 < btrfs_header_nritems(parent))
1158			btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1159
1160		old_bytenr = btrfs_node_blockptr(parent, slot);
1161		blocksize = fs_info->nodesize;
1162		old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1163
1164		if (level <= max_level) {
1165			eb = path->nodes[level];
1166			new_bytenr = btrfs_node_blockptr(eb,
1167							path->slots[level]);
1168			new_ptr_gen = btrfs_node_ptr_generation(eb,
1169							path->slots[level]);
1170		} else {
1171			new_bytenr = 0;
1172			new_ptr_gen = 0;
1173		}
1174
1175		if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1176			ret = level;
1177			break;
1178		}
1179
1180		if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1181		    memcmp_node_keys(parent, slot, path, level)) {
1182			if (level <= lowest_level) {
1183				ret = 0;
1184				break;
1185			}
1186
1187			eb = btrfs_read_node_slot(parent, slot);
1188			if (IS_ERR(eb)) {
1189				ret = PTR_ERR(eb);
1190				break;
 
 
 
 
1191			}
1192			btrfs_tree_lock(eb);
1193			if (cow) {
1194				ret = btrfs_cow_block(trans, dest, eb, parent,
1195						      slot, &eb,
1196						      BTRFS_NESTING_COW);
1197				if (ret) {
1198					btrfs_tree_unlock(eb);
1199					free_extent_buffer(eb);
1200					break;
1201				}
1202			}
 
1203
1204			btrfs_tree_unlock(parent);
1205			free_extent_buffer(parent);
1206
1207			parent = eb;
1208			continue;
1209		}
1210
1211		if (!cow) {
1212			btrfs_tree_unlock(parent);
1213			free_extent_buffer(parent);
1214			cow = 1;
1215			goto again;
1216		}
1217
1218		btrfs_node_key_to_cpu(path->nodes[level], &key,
1219				      path->slots[level]);
1220		btrfs_release_path(path);
1221
1222		path->lowest_level = level;
1223		set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1224		ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1225		clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1226		path->lowest_level = 0;
1227		if (ret) {
1228			if (ret > 0)
1229				ret = -ENOENT;
1230			break;
1231		}
1232
1233		/*
1234		 * Info qgroup to trace both subtrees.
1235		 *
1236		 * We must trace both trees.
1237		 * 1) Tree reloc subtree
1238		 *    If not traced, we will leak data numbers
1239		 * 2) Fs subtree
1240		 *    If not traced, we will double count old data
1241		 *
1242		 * We don't scan the subtree right now, but only record
1243		 * the swapped tree blocks.
1244		 * The real subtree rescan is delayed until we have new
1245		 * CoW on the subtree root node before transaction commit.
1246		 */
1247		ret = btrfs_qgroup_add_swapped_blocks(dest,
1248				rc->block_group, parent, slot,
1249				path->nodes[level], path->slots[level],
1250				last_snapshot);
1251		if (ret < 0)
1252			break;
 
 
 
 
 
 
 
1253		/*
1254		 * swap blocks in fs tree and reloc tree.
1255		 */
1256		btrfs_set_node_blockptr(parent, slot, new_bytenr);
1257		btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1258		btrfs_mark_buffer_dirty(trans, parent);
1259
1260		btrfs_set_node_blockptr(path->nodes[level],
1261					path->slots[level], old_bytenr);
1262		btrfs_set_node_ptr_generation(path->nodes[level],
1263					      path->slots[level], old_ptr_gen);
1264		btrfs_mark_buffer_dirty(trans, path->nodes[level]);
1265
1266		ref.action = BTRFS_ADD_DELAYED_REF;
1267		ref.bytenr = old_bytenr;
1268		ref.num_bytes = blocksize;
1269		ref.parent = path->nodes[level]->start;
1270		ref.owning_root = btrfs_root_id(src);
1271		ref.ref_root = btrfs_root_id(src);
1272		btrfs_init_tree_ref(&ref, level - 1, 0, true);
1273		ret = btrfs_inc_extent_ref(trans, &ref);
1274		if (ret) {
1275			btrfs_abort_transaction(trans, ret);
1276			break;
1277		}
1278
1279		ref.action = BTRFS_ADD_DELAYED_REF;
1280		ref.bytenr = new_bytenr;
1281		ref.num_bytes = blocksize;
1282		ref.parent = 0;
1283		ref.owning_root = btrfs_root_id(dest);
1284		ref.ref_root = btrfs_root_id(dest);
1285		btrfs_init_tree_ref(&ref, level - 1, 0, true);
1286		ret = btrfs_inc_extent_ref(trans, &ref);
1287		if (ret) {
1288			btrfs_abort_transaction(trans, ret);
1289			break;
1290		}
1291
1292		/* We don't know the real owning_root, use 0. */
1293		ref.action = BTRFS_DROP_DELAYED_REF;
1294		ref.bytenr = new_bytenr;
1295		ref.num_bytes = blocksize;
1296		ref.parent = path->nodes[level]->start;
1297		ref.owning_root = 0;
1298		ref.ref_root = btrfs_root_id(src);
1299		btrfs_init_tree_ref(&ref, level - 1, 0, true);
1300		ret = btrfs_free_extent(trans, &ref);
1301		if (ret) {
1302			btrfs_abort_transaction(trans, ret);
1303			break;
1304		}
1305
1306		/* We don't know the real owning_root, use 0. */
1307		ref.action = BTRFS_DROP_DELAYED_REF;
1308		ref.bytenr = old_bytenr;
1309		ref.num_bytes = blocksize;
1310		ref.parent = 0;
1311		ref.owning_root = 0;
1312		ref.ref_root = btrfs_root_id(dest);
1313		btrfs_init_tree_ref(&ref, level - 1, 0, true);
1314		ret = btrfs_free_extent(trans, &ref);
1315		if (ret) {
1316			btrfs_abort_transaction(trans, ret);
1317			break;
1318		}
1319
1320		btrfs_unlock_up_safe(path, 0);
1321
1322		ret = level;
1323		break;
1324	}
1325	btrfs_tree_unlock(parent);
1326	free_extent_buffer(parent);
1327	return ret;
1328}
1329
1330/*
1331 * helper to find next relocated block in reloc tree
1332 */
1333static noinline_for_stack
1334int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1335		       int *level)
1336{
1337	struct extent_buffer *eb;
1338	int i;
1339	u64 last_snapshot;
1340	u32 nritems;
1341
1342	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1343
1344	for (i = 0; i < *level; i++) {
1345		free_extent_buffer(path->nodes[i]);
1346		path->nodes[i] = NULL;
1347	}
1348
1349	for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1350		eb = path->nodes[i];
1351		nritems = btrfs_header_nritems(eb);
1352		while (path->slots[i] + 1 < nritems) {
1353			path->slots[i]++;
1354			if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1355			    last_snapshot)
1356				continue;
1357
1358			*level = i;
1359			return 0;
1360		}
1361		free_extent_buffer(path->nodes[i]);
1362		path->nodes[i] = NULL;
1363	}
1364	return 1;
1365}
1366
1367/*
1368 * walk down reloc tree to find relocated block of lowest level
1369 */
1370static noinline_for_stack
1371int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1372			 int *level)
1373{
 
1374	struct extent_buffer *eb = NULL;
1375	int i;
 
1376	u64 ptr_gen = 0;
1377	u64 last_snapshot;
1378	u32 nritems;
1379
1380	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1381
1382	for (i = *level; i > 0; i--) {
1383		eb = path->nodes[i];
1384		nritems = btrfs_header_nritems(eb);
1385		while (path->slots[i] < nritems) {
1386			ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1387			if (ptr_gen > last_snapshot)
1388				break;
1389			path->slots[i]++;
1390		}
1391		if (path->slots[i] >= nritems) {
1392			if (i == *level)
1393				break;
1394			*level = i + 1;
1395			return 0;
1396		}
1397		if (i == 1) {
1398			*level = i;
1399			return 0;
1400		}
1401
1402		eb = btrfs_read_node_slot(eb, path->slots[i]);
1403		if (IS_ERR(eb))
 
1404			return PTR_ERR(eb);
 
 
 
 
1405		BUG_ON(btrfs_header_level(eb) != i - 1);
1406		path->nodes[i - 1] = eb;
1407		path->slots[i - 1] = 0;
1408	}
1409	return 1;
1410}
1411
1412/*
1413 * invalidate extent cache for file extents whose key in range of
1414 * [min_key, max_key)
1415 */
1416static int invalidate_extent_cache(struct btrfs_root *root,
1417				   const struct btrfs_key *min_key,
1418				   const struct btrfs_key *max_key)
1419{
1420	struct btrfs_fs_info *fs_info = root->fs_info;
1421	struct btrfs_inode *inode = NULL;
1422	u64 objectid;
1423	u64 start, end;
1424	u64 ino;
1425
1426	objectid = min_key->objectid;
1427	while (1) {
1428		struct extent_state *cached_state = NULL;
1429
1430		cond_resched();
1431		if (inode)
1432			iput(&inode->vfs_inode);
1433
1434		if (objectid > max_key->objectid)
1435			break;
1436
1437		inode = btrfs_find_first_inode(root, objectid);
1438		if (!inode)
1439			break;
1440		ino = btrfs_ino(inode);
1441
1442		if (ino > max_key->objectid) {
1443			iput(&inode->vfs_inode);
1444			break;
1445		}
1446
1447		objectid = ino + 1;
1448		if (!S_ISREG(inode->vfs_inode.i_mode))
1449			continue;
1450
1451		if (unlikely(min_key->objectid == ino)) {
1452			if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1453				continue;
1454			if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1455				start = 0;
1456			else {
1457				start = min_key->offset;
1458				WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1459			}
1460		} else {
1461			start = 0;
1462		}
1463
1464		if (unlikely(max_key->objectid == ino)) {
1465			if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1466				continue;
1467			if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1468				end = (u64)-1;
1469			} else {
1470				if (max_key->offset == 0)
1471					continue;
1472				end = max_key->offset;
1473				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1474				end--;
1475			}
1476		} else {
1477			end = (u64)-1;
1478		}
1479
1480		/* the lock_extent waits for read_folio to complete */
1481		lock_extent(&inode->io_tree, start, end, &cached_state);
1482		btrfs_drop_extent_map_range(inode, start, end, true);
1483		unlock_extent(&inode->io_tree, start, end, &cached_state);
1484	}
1485	return 0;
1486}
1487
1488static int find_next_key(struct btrfs_path *path, int level,
1489			 struct btrfs_key *key)
1490
1491{
1492	while (level < BTRFS_MAX_LEVEL) {
1493		if (!path->nodes[level])
1494			break;
1495		if (path->slots[level] + 1 <
1496		    btrfs_header_nritems(path->nodes[level])) {
1497			btrfs_node_key_to_cpu(path->nodes[level], key,
1498					      path->slots[level] + 1);
1499			return 0;
1500		}
1501		level++;
1502	}
1503	return 1;
1504}
1505
1506/*
1507 * Insert current subvolume into reloc_control::dirty_subvol_roots
1508 */
1509static int insert_dirty_subvol(struct btrfs_trans_handle *trans,
1510			       struct reloc_control *rc,
1511			       struct btrfs_root *root)
1512{
1513	struct btrfs_root *reloc_root = root->reloc_root;
1514	struct btrfs_root_item *reloc_root_item;
1515	int ret;
1516
1517	/* @root must be a subvolume tree root with a valid reloc tree */
1518	ASSERT(btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID);
1519	ASSERT(reloc_root);
1520
1521	reloc_root_item = &reloc_root->root_item;
1522	memset(&reloc_root_item->drop_progress, 0,
1523		sizeof(reloc_root_item->drop_progress));
1524	btrfs_set_root_drop_level(reloc_root_item, 0);
1525	btrfs_set_root_refs(reloc_root_item, 0);
1526	ret = btrfs_update_reloc_root(trans, root);
1527	if (ret)
1528		return ret;
1529
1530	if (list_empty(&root->reloc_dirty_list)) {
1531		btrfs_grab_root(root);
1532		list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1533	}
1534
1535	return 0;
1536}
1537
1538static int clean_dirty_subvols(struct reloc_control *rc)
1539{
1540	struct btrfs_root *root;
1541	struct btrfs_root *next;
1542	int ret = 0;
1543	int ret2;
1544
1545	list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1546				 reloc_dirty_list) {
1547		if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID) {
1548			/* Merged subvolume, cleanup its reloc root */
1549			struct btrfs_root *reloc_root = root->reloc_root;
1550
1551			list_del_init(&root->reloc_dirty_list);
1552			root->reloc_root = NULL;
1553			/*
1554			 * Need barrier to ensure clear_bit() only happens after
1555			 * root->reloc_root = NULL. Pairs with have_reloc_root.
1556			 */
1557			smp_wmb();
1558			clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1559			if (reloc_root) {
1560				/*
1561				 * btrfs_drop_snapshot drops our ref we hold for
1562				 * ->reloc_root.  If it fails however we must
1563				 * drop the ref ourselves.
1564				 */
1565				ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1566				if (ret2 < 0) {
1567					btrfs_put_root(reloc_root);
1568					if (!ret)
1569						ret = ret2;
1570				}
1571			}
1572			btrfs_put_root(root);
1573		} else {
1574			/* Orphan reloc tree, just clean it up */
1575			ret2 = btrfs_drop_snapshot(root, 0, 1);
1576			if (ret2 < 0) {
1577				btrfs_put_root(root);
1578				if (!ret)
1579					ret = ret2;
1580			}
1581		}
1582	}
1583	return ret;
1584}
1585
1586/*
1587 * merge the relocated tree blocks in reloc tree with corresponding
1588 * fs tree.
1589 */
1590static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1591					       struct btrfs_root *root)
1592{
1593	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
 
1594	struct btrfs_key key;
1595	struct btrfs_key next_key;
1596	struct btrfs_trans_handle *trans = NULL;
1597	struct btrfs_root *reloc_root;
1598	struct btrfs_root_item *root_item;
1599	struct btrfs_path *path;
1600	struct extent_buffer *leaf;
1601	int reserve_level;
1602	int level;
1603	int max_level;
1604	int replaced = 0;
1605	int ret = 0;
 
1606	u32 min_reserved;
1607
1608	path = btrfs_alloc_path();
1609	if (!path)
1610		return -ENOMEM;
1611	path->reada = READA_FORWARD;
1612
1613	reloc_root = root->reloc_root;
1614	root_item = &reloc_root->root_item;
1615
1616	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1617		level = btrfs_root_level(root_item);
1618		atomic_inc(&reloc_root->node->refs);
1619		path->nodes[level] = reloc_root->node;
1620		path->slots[level] = 0;
1621	} else {
1622		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1623
1624		level = btrfs_root_drop_level(root_item);
1625		BUG_ON(level == 0);
1626		path->lowest_level = level;
1627		ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1628		path->lowest_level = 0;
1629		if (ret < 0) {
1630			btrfs_free_path(path);
1631			return ret;
1632		}
1633
1634		btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1635				      path->slots[level]);
1636		WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1637
1638		btrfs_unlock_up_safe(path, 0);
1639	}
1640
1641	/*
1642	 * In merge_reloc_root(), we modify the upper level pointer to swap the
1643	 * tree blocks between reloc tree and subvolume tree.  Thus for tree
1644	 * block COW, we COW at most from level 1 to root level for each tree.
1645	 *
1646	 * Thus the needed metadata size is at most root_level * nodesize,
1647	 * and * 2 since we have two trees to COW.
1648	 */
1649	reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1650	min_reserved = fs_info->nodesize * reserve_level * 2;
1651	memset(&next_key, 0, sizeof(next_key));
1652
1653	while (1) {
1654		ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
1655					     min_reserved,
1656					     BTRFS_RESERVE_FLUSH_LIMIT);
1657		if (ret)
1658			goto out;
 
1659		trans = btrfs_start_transaction(root, 0);
1660		if (IS_ERR(trans)) {
1661			ret = PTR_ERR(trans);
1662			trans = NULL;
1663			goto out;
1664		}
1665
1666		/*
1667		 * At this point we no longer have a reloc_control, so we can't
1668		 * depend on btrfs_init_reloc_root to update our last_trans.
1669		 *
1670		 * But that's ok, we started the trans handle on our
1671		 * corresponding fs_root, which means it's been added to the
1672		 * dirty list.  At commit time we'll still call
1673		 * btrfs_update_reloc_root() and update our root item
1674		 * appropriately.
1675		 */
1676		btrfs_set_root_last_trans(reloc_root, trans->transid);
1677		trans->block_rsv = rc->block_rsv;
1678
1679		replaced = 0;
1680		max_level = level;
1681
1682		ret = walk_down_reloc_tree(reloc_root, path, &level);
1683		if (ret < 0)
 
1684			goto out;
 
1685		if (ret > 0)
1686			break;
1687
1688		if (!find_next_key(path, level, &key) &&
1689		    btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1690			ret = 0;
1691		} else {
1692			ret = replace_path(trans, rc, root, reloc_root, path,
1693					   &next_key, level, max_level);
1694		}
1695		if (ret < 0)
 
1696			goto out;
 
 
1697		if (ret > 0) {
1698			level = ret;
1699			btrfs_node_key_to_cpu(path->nodes[level], &key,
1700					      path->slots[level]);
1701			replaced = 1;
1702		}
1703
1704		ret = walk_up_reloc_tree(reloc_root, path, &level);
1705		if (ret > 0)
1706			break;
1707
1708		BUG_ON(level == 0);
1709		/*
1710		 * save the merging progress in the drop_progress.
1711		 * this is OK since root refs == 1 in this case.
1712		 */
1713		btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1714			       path->slots[level]);
1715		btrfs_set_root_drop_level(root_item, level);
1716
1717		btrfs_end_transaction_throttle(trans);
1718		trans = NULL;
1719
1720		btrfs_btree_balance_dirty(fs_info);
1721
1722		if (replaced && rc->stage == UPDATE_DATA_PTRS)
1723			invalidate_extent_cache(root, &key, &next_key);
1724	}
1725
1726	/*
1727	 * handle the case only one block in the fs tree need to be
1728	 * relocated and the block is tree root.
1729	 */
1730	leaf = btrfs_lock_root_node(root);
1731	ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1732			      BTRFS_NESTING_COW);
1733	btrfs_tree_unlock(leaf);
1734	free_extent_buffer(leaf);
 
 
1735out:
1736	btrfs_free_path(path);
1737
1738	if (ret == 0) {
1739		ret = insert_dirty_subvol(trans, rc, root);
1740		if (ret)
1741			btrfs_abort_transaction(trans, ret);
 
 
1742	}
1743
1744	if (trans)
1745		btrfs_end_transaction_throttle(trans);
1746
1747	btrfs_btree_balance_dirty(fs_info);
1748
1749	if (replaced && rc->stage == UPDATE_DATA_PTRS)
1750		invalidate_extent_cache(root, &key, &next_key);
1751
1752	return ret;
1753}
1754
1755static noinline_for_stack
1756int prepare_to_merge(struct reloc_control *rc, int err)
1757{
1758	struct btrfs_root *root = rc->extent_root;
1759	struct btrfs_fs_info *fs_info = root->fs_info;
1760	struct btrfs_root *reloc_root;
1761	struct btrfs_trans_handle *trans;
1762	LIST_HEAD(reloc_roots);
1763	u64 num_bytes = 0;
1764	int ret;
1765
1766	mutex_lock(&fs_info->reloc_mutex);
1767	rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1768	rc->merging_rsv_size += rc->nodes_relocated * 2;
1769	mutex_unlock(&fs_info->reloc_mutex);
1770
1771again:
1772	if (!err) {
1773		num_bytes = rc->merging_rsv_size;
1774		ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes,
1775					  BTRFS_RESERVE_FLUSH_ALL);
1776		if (ret)
1777			err = ret;
1778	}
1779
1780	trans = btrfs_join_transaction(rc->extent_root);
1781	if (IS_ERR(trans)) {
1782		if (!err)
1783			btrfs_block_rsv_release(fs_info, rc->block_rsv,
1784						num_bytes, NULL);
1785		return PTR_ERR(trans);
1786	}
1787
1788	if (!err) {
1789		if (num_bytes != rc->merging_rsv_size) {
1790			btrfs_end_transaction(trans);
1791			btrfs_block_rsv_release(fs_info, rc->block_rsv,
1792						num_bytes, NULL);
1793			goto again;
1794		}
1795	}
1796
1797	rc->merge_reloc_tree = true;
1798
1799	while (!list_empty(&rc->reloc_roots)) {
1800		reloc_root = list_entry(rc->reloc_roots.next,
1801					struct btrfs_root, root_list);
1802		list_del_init(&reloc_root->root_list);
1803
1804		root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1805				false);
1806		if (IS_ERR(root)) {
1807			/*
1808			 * Even if we have an error we need this reloc root
1809			 * back on our list so we can clean up properly.
1810			 */
1811			list_add(&reloc_root->root_list, &reloc_roots);
1812			btrfs_abort_transaction(trans, (int)PTR_ERR(root));
1813			if (!err)
1814				err = PTR_ERR(root);
1815			break;
1816		}
1817
1818		if (unlikely(root->reloc_root != reloc_root)) {
1819			if (root->reloc_root) {
1820				btrfs_err(fs_info,
1821"reloc tree mismatch, root %lld has reloc root key (%lld %u %llu) gen %llu, expect reloc root key (%lld %u %llu) gen %llu",
1822					  btrfs_root_id(root),
1823					  btrfs_root_id(root->reloc_root),
1824					  root->reloc_root->root_key.type,
1825					  root->reloc_root->root_key.offset,
1826					  btrfs_root_generation(
1827						  &root->reloc_root->root_item),
1828					  btrfs_root_id(reloc_root),
1829					  reloc_root->root_key.type,
1830					  reloc_root->root_key.offset,
1831					  btrfs_root_generation(
1832						  &reloc_root->root_item));
1833			} else {
1834				btrfs_err(fs_info,
1835"reloc tree mismatch, root %lld has no reloc root, expect reloc root key (%lld %u %llu) gen %llu",
1836					  btrfs_root_id(root),
1837					  btrfs_root_id(reloc_root),
1838					  reloc_root->root_key.type,
1839					  reloc_root->root_key.offset,
1840					  btrfs_root_generation(
1841						  &reloc_root->root_item));
1842			}
1843			list_add(&reloc_root->root_list, &reloc_roots);
1844			btrfs_put_root(root);
1845			btrfs_abort_transaction(trans, -EUCLEAN);
1846			if (!err)
1847				err = -EUCLEAN;
1848			break;
1849		}
1850
1851		/*
1852		 * set reference count to 1, so btrfs_recover_relocation
1853		 * knows it should resumes merging
1854		 */
1855		if (!err)
1856			btrfs_set_root_refs(&reloc_root->root_item, 1);
1857		ret = btrfs_update_reloc_root(trans, root);
1858
1859		/*
1860		 * Even if we have an error we need this reloc root back on our
1861		 * list so we can clean up properly.
1862		 */
1863		list_add(&reloc_root->root_list, &reloc_roots);
1864		btrfs_put_root(root);
1865
1866		if (ret) {
1867			btrfs_abort_transaction(trans, ret);
1868			if (!err)
1869				err = ret;
1870			break;
1871		}
1872	}
1873
1874	list_splice(&reloc_roots, &rc->reloc_roots);
1875
1876	if (!err)
1877		err = btrfs_commit_transaction(trans);
1878	else
1879		btrfs_end_transaction(trans);
1880	return err;
1881}
1882
1883static noinline_for_stack
1884void free_reloc_roots(struct list_head *list)
1885{
1886	struct btrfs_root *reloc_root, *tmp;
1887
1888	list_for_each_entry_safe(reloc_root, tmp, list, root_list)
 
 
 
 
 
 
1889		__del_reloc_root(reloc_root);
 
1890}
1891
1892static noinline_for_stack
1893void merge_reloc_roots(struct reloc_control *rc)
1894{
1895	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1896	struct btrfs_root *root;
1897	struct btrfs_root *reloc_root;
1898	LIST_HEAD(reloc_roots);
1899	int found = 0;
1900	int ret = 0;
1901again:
1902	root = rc->extent_root;
1903
1904	/*
1905	 * this serializes us with btrfs_record_root_in_transaction,
1906	 * we have to make sure nobody is in the middle of
1907	 * adding their roots to the list while we are
1908	 * doing this splice
1909	 */
1910	mutex_lock(&fs_info->reloc_mutex);
1911	list_splice_init(&rc->reloc_roots, &reloc_roots);
1912	mutex_unlock(&fs_info->reloc_mutex);
1913
1914	while (!list_empty(&reloc_roots)) {
1915		found = 1;
1916		reloc_root = list_entry(reloc_roots.next,
1917					struct btrfs_root, root_list);
1918
1919		root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1920					 false);
1921		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1922			if (WARN_ON(IS_ERR(root))) {
1923				/*
1924				 * For recovery we read the fs roots on mount,
1925				 * and if we didn't find the root then we marked
1926				 * the reloc root as a garbage root.  For normal
1927				 * relocation obviously the root should exist in
1928				 * memory.  However there's no reason we can't
1929				 * handle the error properly here just in case.
1930				 */
1931				ret = PTR_ERR(root);
1932				goto out;
1933			}
1934			if (WARN_ON(root->reloc_root != reloc_root)) {
1935				/*
1936				 * This can happen if on-disk metadata has some
1937				 * corruption, e.g. bad reloc tree key offset.
1938				 */
1939				ret = -EINVAL;
1940				goto out;
1941			}
1942			ret = merge_reloc_root(rc, root);
1943			btrfs_put_root(root);
1944			if (ret) {
1945				if (list_empty(&reloc_root->root_list))
1946					list_add_tail(&reloc_root->root_list,
1947						      &reloc_roots);
1948				goto out;
1949			}
1950		} else {
1951			if (!IS_ERR(root)) {
1952				if (root->reloc_root == reloc_root) {
1953					root->reloc_root = NULL;
1954					btrfs_put_root(reloc_root);
1955				}
1956				clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
1957					  &root->state);
1958				btrfs_put_root(root);
1959			}
1960
1961			list_del_init(&reloc_root->root_list);
1962			/* Don't forget to queue this reloc root for cleanup */
1963			list_add_tail(&reloc_root->reloc_dirty_list,
1964				      &rc->dirty_subvol_roots);
 
 
1965		}
1966	}
1967
1968	if (found) {
1969		found = 0;
1970		goto again;
1971	}
1972out:
1973	if (ret) {
1974		btrfs_handle_fs_error(fs_info, ret, NULL);
1975		free_reloc_roots(&reloc_roots);
 
1976
1977		/* new reloc root may be added */
1978		mutex_lock(&fs_info->reloc_mutex);
1979		list_splice_init(&rc->reloc_roots, &reloc_roots);
1980		mutex_unlock(&fs_info->reloc_mutex);
1981		free_reloc_roots(&reloc_roots);
 
1982	}
1983
1984	/*
1985	 * We used to have
1986	 *
1987	 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
1988	 *
1989	 * here, but it's wrong.  If we fail to start the transaction in
1990	 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
1991	 * have actually been removed from the reloc_root_tree rb tree.  This is
1992	 * fine because we're bailing here, and we hold a reference on the root
1993	 * for the list that holds it, so these roots will be cleaned up when we
1994	 * do the reloc_dirty_list afterwards.  Meanwhile the root->reloc_root
1995	 * will be cleaned up on unmount.
1996	 *
1997	 * The remaining nodes will be cleaned up by free_reloc_control.
1998	 */
1999}
2000
2001static void free_block_list(struct rb_root *blocks)
2002{
2003	struct tree_block *block;
2004	struct rb_node *rb_node;
2005	while ((rb_node = rb_first(blocks))) {
2006		block = rb_entry(rb_node, struct tree_block, rb_node);
2007		rb_erase(rb_node, blocks);
2008		kfree(block);
2009	}
2010}
2011
2012static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2013				      struct btrfs_root *reloc_root)
2014{
2015	struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2016	struct btrfs_root *root;
2017	int ret;
2018
2019	if (btrfs_get_root_last_trans(reloc_root) == trans->transid)
2020		return 0;
2021
2022	root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
 
 
2023
2024	/*
2025	 * This should succeed, since we can't have a reloc root without having
2026	 * already looked up the actual root and created the reloc root for this
2027	 * root.
2028	 *
2029	 * However if there's some sort of corruption where we have a ref to a
2030	 * reloc root without a corresponding root this could return ENOENT.
2031	 */
2032	if (IS_ERR(root)) {
2033		ASSERT(0);
2034		return PTR_ERR(root);
2035	}
2036	if (root->reloc_root != reloc_root) {
2037		ASSERT(0);
2038		btrfs_err(fs_info,
2039			  "root %llu has two reloc roots associated with it",
2040			  reloc_root->root_key.offset);
2041		btrfs_put_root(root);
2042		return -EUCLEAN;
2043	}
2044	ret = btrfs_record_root_in_trans(trans, root);
2045	btrfs_put_root(root);
2046
2047	return ret;
2048}
2049
2050static noinline_for_stack
2051struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2052				     struct reloc_control *rc,
2053				     struct btrfs_backref_node *node,
2054				     struct btrfs_backref_edge *edges[])
2055{
2056	struct btrfs_backref_node *next;
2057	struct btrfs_root *root;
2058	int index = 0;
2059	int ret;
2060
2061	next = node;
2062	while (1) {
2063		cond_resched();
2064		next = walk_up_backref(next, edges, &index);
2065		root = next->root;
 
 
2066
2067		/*
2068		 * If there is no root, then our references for this block are
2069		 * incomplete, as we should be able to walk all the way up to a
2070		 * block that is owned by a root.
2071		 *
2072		 * This path is only for SHAREABLE roots, so if we come upon a
2073		 * non-SHAREABLE root then we have backrefs that resolve
2074		 * improperly.
2075		 *
2076		 * Both of these cases indicate file system corruption, or a bug
2077		 * in the backref walking code.
2078		 */
2079		if (!root) {
2080			ASSERT(0);
2081			btrfs_err(trans->fs_info,
2082		"bytenr %llu doesn't have a backref path ending in a root",
2083				  node->bytenr);
2084			return ERR_PTR(-EUCLEAN);
2085		}
2086		if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2087			ASSERT(0);
2088			btrfs_err(trans->fs_info,
2089	"bytenr %llu has multiple refs with one ending in a non-shareable root",
2090				  node->bytenr);
2091			return ERR_PTR(-EUCLEAN);
2092		}
2093
2094		if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID) {
2095			ret = record_reloc_root_in_trans(trans, root);
2096			if (ret)
2097				return ERR_PTR(ret);
2098			break;
2099		}
2100
2101		ret = btrfs_record_root_in_trans(trans, root);
2102		if (ret)
2103			return ERR_PTR(ret);
2104		root = root->reloc_root;
2105
2106		/*
2107		 * We could have raced with another thread which failed, so
2108		 * root->reloc_root may not be set, return ENOENT in this case.
2109		 */
2110		if (!root)
2111			return ERR_PTR(-ENOENT);
2112
2113		if (next->new_bytenr != root->node->start) {
2114			/*
2115			 * We just created the reloc root, so we shouldn't have
2116			 * ->new_bytenr set and this shouldn't be in the changed
2117			 *  list.  If it is then we have multiple roots pointing
2118			 *  at the same bytenr which indicates corruption, or
2119			 *  we've made a mistake in the backref walking code.
2120			 */
2121			ASSERT(next->new_bytenr == 0);
2122			ASSERT(list_empty(&next->list));
2123			if (next->new_bytenr || !list_empty(&next->list)) {
2124				btrfs_err(trans->fs_info,
2125	"bytenr %llu possibly has multiple roots pointing at the same bytenr %llu",
2126					  node->bytenr, next->bytenr);
2127				return ERR_PTR(-EUCLEAN);
2128			}
2129
2130			next->new_bytenr = root->node->start;
2131			btrfs_put_root(next->root);
2132			next->root = btrfs_grab_root(root);
2133			ASSERT(next->root);
2134			list_add_tail(&next->list,
2135				      &rc->backref_cache.changed);
2136			mark_block_processed(rc, next);
2137			break;
2138		}
2139
2140		WARN_ON(1);
2141		root = NULL;
2142		next = walk_down_backref(edges, &index);
2143		if (!next || next->level <= node->level)
2144			break;
2145	}
2146	if (!root) {
2147		/*
2148		 * This can happen if there's fs corruption or if there's a bug
2149		 * in the backref lookup code.
2150		 */
2151		ASSERT(0);
2152		return ERR_PTR(-ENOENT);
2153	}
2154
2155	next = node;
2156	/* setup backref node path for btrfs_reloc_cow_block */
2157	while (1) {
2158		rc->backref_cache.path[next->level] = next;
2159		if (--index < 0)
2160			break;
2161		next = edges[index]->node[UPPER];
2162	}
2163	return root;
2164}
2165
2166/*
2167 * Select a tree root for relocation.
2168 *
2169 * Return NULL if the block is not shareable. We should use do_relocation() in
2170 * this case.
2171 *
2172 * Return a tree root pointer if the block is shareable.
2173 * Return -ENOENT if the block is root of reloc tree.
2174 */
2175static noinline_for_stack
2176struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2177{
2178	struct btrfs_backref_node *next;
2179	struct btrfs_root *root;
2180	struct btrfs_root *fs_root = NULL;
2181	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2182	int index = 0;
2183
2184	next = node;
2185	while (1) {
2186		cond_resched();
2187		next = walk_up_backref(next, edges, &index);
2188		root = next->root;
 
2189
2190		/*
2191		 * This can occur if we have incomplete extent refs leading all
2192		 * the way up a particular path, in this case return -EUCLEAN.
2193		 */
2194		if (!root)
2195			return ERR_PTR(-EUCLEAN);
2196
2197		/* No other choice for non-shareable tree */
2198		if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2199			return root;
2200
2201		if (btrfs_root_id(root) != BTRFS_TREE_RELOC_OBJECTID)
2202			fs_root = root;
2203
2204		if (next != node)
2205			return NULL;
2206
2207		next = walk_down_backref(edges, &index);
2208		if (!next || next->level <= node->level)
2209			break;
2210	}
2211
2212	if (!fs_root)
2213		return ERR_PTR(-ENOENT);
2214	return fs_root;
2215}
2216
2217static noinline_for_stack u64 calcu_metadata_size(struct reloc_control *rc,
2218						  struct btrfs_backref_node *node)
 
2219{
2220	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2221	struct btrfs_backref_node *next = node;
2222	struct btrfs_backref_edge *edge;
2223	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2224	u64 num_bytes = 0;
2225	int index = 0;
2226
2227	BUG_ON(node->processed);
2228
2229	while (next) {
2230		cond_resched();
2231		while (1) {
2232			if (next->processed)
2233				break;
2234
2235			num_bytes += fs_info->nodesize;
2236
2237			if (list_empty(&next->upper))
2238				break;
2239
2240			edge = list_entry(next->upper.next,
2241					struct btrfs_backref_edge, list[LOWER]);
2242			edges[index++] = edge;
2243			next = edge->node[UPPER];
2244		}
2245		next = walk_down_backref(edges, &index);
2246	}
2247	return num_bytes;
2248}
2249
2250static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2251				  struct reloc_control *rc,
2252				  struct btrfs_backref_node *node)
2253{
2254	struct btrfs_root *root = rc->extent_root;
2255	struct btrfs_fs_info *fs_info = root->fs_info;
2256	u64 num_bytes;
2257	int ret;
2258	u64 tmp;
2259
2260	num_bytes = calcu_metadata_size(rc, node) * 2;
2261
2262	trans->block_rsv = rc->block_rsv;
2263	rc->reserved_bytes += num_bytes;
2264
2265	/*
2266	 * We are under a transaction here so we can only do limited flushing.
2267	 * If we get an enospc just kick back -EAGAIN so we know to drop the
2268	 * transaction and try to refill when we can flush all the things.
2269	 */
2270	ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes,
2271				     BTRFS_RESERVE_FLUSH_LIMIT);
2272	if (ret) {
2273		tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2274		while (tmp <= rc->reserved_bytes)
2275			tmp <<= 1;
2276		/*
2277		 * only one thread can access block_rsv at this point,
2278		 * so we don't need hold lock to protect block_rsv.
2279		 * we expand more reservation size here to allow enough
2280		 * space for relocation and we will return earlier in
2281		 * enospc case.
2282		 */
2283		rc->block_rsv->size = tmp + fs_info->nodesize *
2284				      RELOCATION_RESERVED_NODES;
2285		return -EAGAIN;
2286	}
2287
2288	return 0;
2289}
2290
2291/*
2292 * relocate a block tree, and then update pointers in upper level
2293 * blocks that reference the block to point to the new location.
2294 *
2295 * if called by link_to_upper, the block has already been relocated.
2296 * in that case this function just updates pointers.
2297 */
2298static int do_relocation(struct btrfs_trans_handle *trans,
2299			 struct reloc_control *rc,
2300			 struct btrfs_backref_node *node,
2301			 struct btrfs_key *key,
2302			 struct btrfs_path *path, int lowest)
2303{
2304	struct btrfs_backref_node *upper;
2305	struct btrfs_backref_edge *edge;
2306	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
 
2307	struct btrfs_root *root;
2308	struct extent_buffer *eb;
2309	u32 blocksize;
2310	u64 bytenr;
 
2311	int slot;
2312	int ret = 0;
 
2313
2314	/*
2315	 * If we are lowest then this is the first time we're processing this
2316	 * block, and thus shouldn't have an eb associated with it yet.
2317	 */
2318	ASSERT(!lowest || !node->eb);
2319
2320	path->lowest_level = node->level + 1;
2321	rc->backref_cache.path[node->level] = node;
2322	list_for_each_entry(edge, &node->upper, list[LOWER]) {
2323		cond_resched();
2324
2325		upper = edge->node[UPPER];
2326		root = select_reloc_root(trans, rc, upper, edges);
2327		if (IS_ERR(root)) {
2328			ret = PTR_ERR(root);
2329			goto next;
2330		}
2331
2332		if (upper->eb && !upper->locked) {
2333			if (!lowest) {
2334				ret = btrfs_bin_search(upper->eb, 0, key, &slot);
2335				if (ret < 0)
2336					goto next;
2337				BUG_ON(ret);
2338				bytenr = btrfs_node_blockptr(upper->eb, slot);
2339				if (node->eb->start == bytenr)
2340					goto next;
2341			}
2342			btrfs_backref_drop_node_buffer(upper);
2343		}
2344
2345		if (!upper->eb) {
2346			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2347			if (ret) {
2348				if (ret > 0)
2349					ret = -ENOENT;
 
 
2350
2351				btrfs_release_path(path);
2352				break;
2353			}
2354
2355			if (!upper->eb) {
2356				upper->eb = path->nodes[upper->level];
2357				path->nodes[upper->level] = NULL;
2358			} else {
2359				BUG_ON(upper->eb != path->nodes[upper->level]);
2360			}
2361
2362			upper->locked = 1;
2363			path->locks[upper->level] = 0;
2364
2365			slot = path->slots[upper->level];
2366			btrfs_release_path(path);
2367		} else {
2368			ret = btrfs_bin_search(upper->eb, 0, key, &slot);
2369			if (ret < 0)
2370				goto next;
2371			BUG_ON(ret);
2372		}
2373
2374		bytenr = btrfs_node_blockptr(upper->eb, slot);
2375		if (lowest) {
2376			if (bytenr != node->bytenr) {
2377				btrfs_err(root->fs_info,
2378		"lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2379					  bytenr, node->bytenr, slot,
2380					  upper->eb->start);
2381				ret = -EIO;
2382				goto next;
2383			}
2384		} else {
2385			if (node->eb->start == bytenr)
2386				goto next;
2387		}
2388
2389		blocksize = root->fs_info->nodesize;
2390		eb = btrfs_read_node_slot(upper->eb, slot);
 
2391		if (IS_ERR(eb)) {
2392			ret = PTR_ERR(eb);
 
 
 
 
2393			goto next;
2394		}
2395		btrfs_tree_lock(eb);
 
2396
2397		if (!node->eb) {
2398			ret = btrfs_cow_block(trans, root, eb, upper->eb,
2399					      slot, &eb, BTRFS_NESTING_COW);
2400			btrfs_tree_unlock(eb);
2401			free_extent_buffer(eb);
2402			if (ret < 0)
 
2403				goto next;
2404			/*
2405			 * We've just COWed this block, it should have updated
2406			 * the correct backref node entry.
2407			 */
2408			ASSERT(node->eb == eb);
2409		} else {
2410			struct btrfs_ref ref = {
2411				.action = BTRFS_ADD_DELAYED_REF,
2412				.bytenr = node->eb->start,
2413				.num_bytes = blocksize,
2414				.parent = upper->eb->start,
2415				.owning_root = btrfs_header_owner(upper->eb),
2416				.ref_root = btrfs_header_owner(upper->eb),
2417			};
2418
2419			btrfs_set_node_blockptr(upper->eb, slot,
2420						node->eb->start);
2421			btrfs_set_node_ptr_generation(upper->eb, slot,
2422						      trans->transid);
2423			btrfs_mark_buffer_dirty(trans, upper->eb);
 
 
 
 
 
 
 
2424
2425			btrfs_init_tree_ref(&ref, node->level,
2426					    btrfs_root_id(root), false);
2427			ret = btrfs_inc_extent_ref(trans, &ref);
2428			if (!ret)
2429				ret = btrfs_drop_subtree(trans, root, eb,
2430							 upper->eb);
2431			if (ret)
2432				btrfs_abort_transaction(trans, ret);
2433		}
2434next:
2435		if (!upper->pending)
2436			btrfs_backref_drop_node_buffer(upper);
2437		else
2438			btrfs_backref_unlock_node_buffer(upper);
2439		if (ret)
2440			break;
2441	}
2442
2443	if (!ret && node->pending) {
2444		btrfs_backref_drop_node_buffer(node);
2445		list_move_tail(&node->list, &rc->backref_cache.changed);
2446		node->pending = 0;
2447	}
2448
2449	path->lowest_level = 0;
2450
2451	/*
2452	 * We should have allocated all of our space in the block rsv and thus
2453	 * shouldn't ENOSPC.
2454	 */
2455	ASSERT(ret != -ENOSPC);
2456	return ret;
2457}
2458
2459static int link_to_upper(struct btrfs_trans_handle *trans,
2460			 struct reloc_control *rc,
2461			 struct btrfs_backref_node *node,
2462			 struct btrfs_path *path)
2463{
2464	struct btrfs_key key;
2465
2466	btrfs_node_key_to_cpu(node->eb, &key, 0);
2467	return do_relocation(trans, rc, node, &key, path, 0);
2468}
2469
2470static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2471				struct reloc_control *rc,
2472				struct btrfs_path *path, int err)
2473{
2474	LIST_HEAD(list);
2475	struct btrfs_backref_cache *cache = &rc->backref_cache;
2476	struct btrfs_backref_node *node;
2477	int level;
2478	int ret;
2479
2480	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2481		while (!list_empty(&cache->pending[level])) {
2482			node = list_entry(cache->pending[level].next,
2483					  struct btrfs_backref_node, list);
2484			list_move_tail(&node->list, &list);
2485			BUG_ON(!node->pending);
2486
2487			if (!err) {
2488				ret = link_to_upper(trans, rc, node, path);
2489				if (ret < 0)
2490					err = ret;
2491			}
2492		}
2493		list_splice_init(&list, &cache->pending[level]);
2494	}
2495	return err;
2496}
2497
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2498/*
2499 * mark a block and all blocks directly/indirectly reference the block
2500 * as processed.
2501 */
2502static void update_processed_blocks(struct reloc_control *rc,
2503				    struct btrfs_backref_node *node)
2504{
2505	struct btrfs_backref_node *next = node;
2506	struct btrfs_backref_edge *edge;
2507	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2508	int index = 0;
2509
2510	while (next) {
2511		cond_resched();
2512		while (1) {
2513			if (next->processed)
2514				break;
2515
2516			mark_block_processed(rc, next);
2517
2518			if (list_empty(&next->upper))
2519				break;
2520
2521			edge = list_entry(next->upper.next,
2522					struct btrfs_backref_edge, list[LOWER]);
2523			edges[index++] = edge;
2524			next = edge->node[UPPER];
2525		}
2526		next = walk_down_backref(edges, &index);
2527	}
2528}
2529
2530static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2531{
2532	u32 blocksize = rc->extent_root->fs_info->nodesize;
2533
2534	if (test_range_bit(&rc->processed_blocks, bytenr,
2535			   bytenr + blocksize - 1, EXTENT_DIRTY, NULL))
2536		return 1;
2537	return 0;
2538}
2539
2540static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2541			      struct tree_block *block)
2542{
2543	struct btrfs_tree_parent_check check = {
2544		.level = block->level,
2545		.owner_root = block->owner,
2546		.transid = block->key.offset
2547	};
2548	struct extent_buffer *eb;
2549
2550	eb = read_tree_block(fs_info, block->bytenr, &check);
2551	if (IS_ERR(eb))
 
2552		return PTR_ERR(eb);
2553	if (!extent_buffer_uptodate(eb)) {
2554		free_extent_buffer(eb);
2555		return -EIO;
2556	}
 
2557	if (block->level == 0)
2558		btrfs_item_key_to_cpu(eb, &block->key, 0);
2559	else
2560		btrfs_node_key_to_cpu(eb, &block->key, 0);
2561	free_extent_buffer(eb);
2562	block->key_ready = true;
2563	return 0;
2564}
2565
2566/*
2567 * helper function to relocate a tree block
2568 */
2569static int relocate_tree_block(struct btrfs_trans_handle *trans,
2570				struct reloc_control *rc,
2571				struct btrfs_backref_node *node,
2572				struct btrfs_key *key,
2573				struct btrfs_path *path)
2574{
2575	struct btrfs_root *root;
2576	int ret = 0;
2577
2578	if (!node)
2579		return 0;
2580
2581	/*
2582	 * If we fail here we want to drop our backref_node because we are going
2583	 * to start over and regenerate the tree for it.
2584	 */
2585	ret = reserve_metadata_space(trans, rc, node);
2586	if (ret)
2587		goto out;
2588
2589	BUG_ON(node->processed);
2590	root = select_one_root(node);
2591	if (IS_ERR(root)) {
2592		ret = PTR_ERR(root);
 
 
2593
2594		/* See explanation in select_one_root for the -EUCLEAN case. */
2595		ASSERT(ret == -ENOENT);
2596		if (ret == -ENOENT) {
2597			ret = 0;
2598			update_processed_blocks(rc, node);
2599		}
2600		goto out;
2601	}
2602
2603	if (root) {
2604		if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2605			/*
2606			 * This block was the root block of a root, and this is
2607			 * the first time we're processing the block and thus it
2608			 * should not have had the ->new_bytenr modified and
2609			 * should have not been included on the changed list.
2610			 *
2611			 * However in the case of corruption we could have
2612			 * multiple refs pointing to the same block improperly,
2613			 * and thus we would trip over these checks.  ASSERT()
2614			 * for the developer case, because it could indicate a
2615			 * bug in the backref code, however error out for a
2616			 * normal user in the case of corruption.
2617			 */
2618			ASSERT(node->new_bytenr == 0);
2619			ASSERT(list_empty(&node->list));
2620			if (node->new_bytenr || !list_empty(&node->list)) {
2621				btrfs_err(root->fs_info,
2622				  "bytenr %llu has improper references to it",
2623					  node->bytenr);
2624				ret = -EUCLEAN;
2625				goto out;
2626			}
2627			ret = btrfs_record_root_in_trans(trans, root);
2628			if (ret)
2629				goto out;
2630			/*
2631			 * Another thread could have failed, need to check if we
2632			 * have reloc_root actually set.
2633			 */
2634			if (!root->reloc_root) {
2635				ret = -ENOENT;
2636				goto out;
2637			}
2638			root = root->reloc_root;
2639			node->new_bytenr = root->node->start;
2640			btrfs_put_root(node->root);
2641			node->root = btrfs_grab_root(root);
2642			ASSERT(node->root);
2643			list_add_tail(&node->list, &rc->backref_cache.changed);
2644		} else {
2645			path->lowest_level = node->level;
2646			if (root == root->fs_info->chunk_root)
2647				btrfs_reserve_chunk_metadata(trans, false);
2648			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2649			btrfs_release_path(path);
2650			if (root == root->fs_info->chunk_root)
2651				btrfs_trans_release_chunk_metadata(trans);
2652			if (ret > 0)
2653				ret = 0;
2654		}
2655		if (!ret)
2656			update_processed_blocks(rc, node);
2657	} else {
2658		ret = do_relocation(trans, rc, node, key, path, 1);
2659	}
2660out:
2661	if (ret || node->level == 0 || node->cowonly)
2662		btrfs_backref_cleanup_node(&rc->backref_cache, node);
2663	return ret;
2664}
2665
2666/*
2667 * relocate a list of blocks
2668 */
2669static noinline_for_stack
2670int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2671			 struct reloc_control *rc, struct rb_root *blocks)
2672{
2673	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2674	struct btrfs_backref_node *node;
2675	struct btrfs_path *path;
2676	struct tree_block *block;
2677	struct tree_block *next;
2678	int ret = 0;
 
2679
2680	path = btrfs_alloc_path();
2681	if (!path) {
2682		ret = -ENOMEM;
2683		goto out_free_blocks;
2684	}
2685
2686	/* Kick in readahead for tree blocks with missing keys */
2687	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
 
2688		if (!block->key_ready)
2689			btrfs_readahead_tree_block(fs_info, block->bytenr,
2690						   block->owner, 0,
2691						   block->level);
2692	}
2693
2694	/* Get first keys */
2695	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
 
2696		if (!block->key_ready) {
2697			ret = get_tree_block_key(fs_info, block);
2698			if (ret)
2699				goto out_free_path;
2700		}
 
2701	}
2702
2703	/* Do tree relocation */
2704	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
2705		node = build_backref_tree(trans, rc, &block->key,
 
 
2706					  block->level, block->bytenr);
2707		if (IS_ERR(node)) {
2708			ret = PTR_ERR(node);
2709			goto out;
2710		}
2711
2712		ret = relocate_tree_block(trans, rc, node, &block->key,
2713					  path);
2714		if (ret < 0)
2715			break;
 
 
 
 
2716	}
2717out:
2718	ret = finish_pending_nodes(trans, rc, path, ret);
2719
2720out_free_path:
2721	btrfs_free_path(path);
2722out_free_blocks:
2723	free_block_list(blocks);
2724	return ret;
2725}
2726
2727static noinline_for_stack int prealloc_file_extent_cluster(struct reloc_control *rc)
 
 
2728{
2729	const struct file_extent_cluster *cluster = &rc->cluster;
2730	struct btrfs_inode *inode = BTRFS_I(rc->data_inode);
2731	u64 alloc_hint = 0;
2732	u64 start;
2733	u64 end;
2734	u64 offset = inode->reloc_block_group_start;
2735	u64 num_bytes;
2736	int nr;
2737	int ret = 0;
2738	u64 i_size = i_size_read(&inode->vfs_inode);
2739	u64 prealloc_start = cluster->start - offset;
2740	u64 prealloc_end = cluster->end - offset;
2741	u64 cur_offset = prealloc_start;
2742
2743	/*
2744	 * For subpage case, previous i_size may not be aligned to PAGE_SIZE.
2745	 * This means the range [i_size, PAGE_END + 1) is filled with zeros by
2746	 * btrfs_do_readpage() call of previously relocated file cluster.
2747	 *
2748	 * If the current cluster starts in the above range, btrfs_do_readpage()
2749	 * will skip the read, and relocate_one_folio() will later writeback
2750	 * the padding zeros as new data, causing data corruption.
2751	 *
2752	 * Here we have to manually invalidate the range (i_size, PAGE_END + 1).
2753	 */
2754	if (!PAGE_ALIGNED(i_size)) {
2755		struct address_space *mapping = inode->vfs_inode.i_mapping;
2756		struct btrfs_fs_info *fs_info = inode->root->fs_info;
2757		const u32 sectorsize = fs_info->sectorsize;
2758		struct folio *folio;
2759
2760		ASSERT(sectorsize < PAGE_SIZE);
2761		ASSERT(IS_ALIGNED(i_size, sectorsize));
2762
2763		/*
2764		 * Subpage can't handle page with DIRTY but without UPTODATE
2765		 * bit as it can lead to the following deadlock:
2766		 *
2767		 * btrfs_read_folio()
2768		 * | Page already *locked*
2769		 * |- btrfs_lock_and_flush_ordered_range()
2770		 *    |- btrfs_start_ordered_extent()
2771		 *       |- extent_write_cache_pages()
2772		 *          |- lock_page()
2773		 *             We try to lock the page we already hold.
2774		 *
2775		 * Here we just writeback the whole data reloc inode, so that
2776		 * we will be ensured to have no dirty range in the page, and
2777		 * are safe to clear the uptodate bits.
2778		 *
2779		 * This shouldn't cause too much overhead, as we need to write
2780		 * the data back anyway.
2781		 */
2782		ret = filemap_write_and_wait(mapping);
2783		if (ret < 0)
2784			return ret;
2785
2786		clear_extent_bits(&inode->io_tree, i_size,
2787				  round_up(i_size, PAGE_SIZE) - 1,
2788				  EXTENT_UPTODATE);
2789		folio = filemap_lock_folio(mapping, i_size >> PAGE_SHIFT);
2790		/*
2791		 * If page is freed we don't need to do anything then, as we
2792		 * will re-read the whole page anyway.
2793		 */
2794		if (!IS_ERR(folio)) {
2795			btrfs_subpage_clear_uptodate(fs_info, folio, i_size,
2796					round_up(i_size, PAGE_SIZE) - i_size);
2797			folio_unlock(folio);
2798			folio_put(folio);
2799		}
2800	}
2801
2802	BUG_ON(cluster->start != cluster->boundary[0]);
2803	ret = btrfs_alloc_data_chunk_ondemand(inode,
2804					      prealloc_end + 1 - prealloc_start);
2805	if (ret)
2806		return ret;
2807
2808	btrfs_inode_lock(inode, 0);
2809	for (nr = 0; nr < cluster->nr; nr++) {
2810		struct extent_state *cached_state = NULL;
2811
 
 
2812		start = cluster->boundary[nr] - offset;
2813		if (nr + 1 < cluster->nr)
2814			end = cluster->boundary[nr + 1] - 1 - offset;
2815		else
2816			end = cluster->end - offset;
2817
2818		lock_extent(&inode->io_tree, start, end, &cached_state);
2819		num_bytes = end + 1 - start;
2820		ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
 
 
 
2821						num_bytes, num_bytes,
2822						end + 1, &alloc_hint);
2823		cur_offset = end + 1;
2824		unlock_extent(&inode->io_tree, start, end, &cached_state);
2825		if (ret)
2826			break;
 
2827	}
2828	btrfs_inode_unlock(inode, 0);
2829
2830	if (cur_offset < prealloc_end)
2831		btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2832					       prealloc_end + 1 - cur_offset);
 
 
2833	return ret;
2834}
2835
2836static noinline_for_stack int setup_relocation_extent_mapping(struct reloc_control *rc)
 
 
2837{
2838	struct btrfs_inode *inode = BTRFS_I(rc->data_inode);
 
2839	struct extent_map *em;
2840	struct extent_state *cached_state = NULL;
2841	u64 offset = inode->reloc_block_group_start;
2842	u64 start = rc->cluster.start - offset;
2843	u64 end = rc->cluster.end - offset;
2844	int ret = 0;
2845
2846	em = alloc_extent_map();
2847	if (!em)
2848		return -ENOMEM;
2849
2850	em->start = start;
2851	em->len = end + 1 - start;
2852	em->disk_bytenr = rc->cluster.start;
2853	em->disk_num_bytes = em->len;
2854	em->ram_bytes = em->len;
2855	em->flags |= EXTENT_FLAG_PINNED;
2856
2857	lock_extent(&inode->io_tree, start, end, &cached_state);
2858	ret = btrfs_replace_extent_map_range(inode, em, false);
2859	unlock_extent(&inode->io_tree, start, end, &cached_state);
2860	free_extent_map(em);
2861
2862	return ret;
2863}
2864
2865/*
2866 * Allow error injection to test balance/relocation cancellation
2867 */
2868noinline int btrfs_should_cancel_balance(const struct btrfs_fs_info *fs_info)
2869{
2870	return atomic_read(&fs_info->balance_cancel_req) ||
2871		atomic_read(&fs_info->reloc_cancel_req) ||
2872		fatal_signal_pending(current);
2873}
2874ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2875
2876static u64 get_cluster_boundary_end(const struct file_extent_cluster *cluster,
2877				    int cluster_nr)
2878{
2879	/* Last extent, use cluster end directly */
2880	if (cluster_nr >= cluster->nr - 1)
2881		return cluster->end;
2882
2883	/* Use next boundary start*/
2884	return cluster->boundary[cluster_nr + 1] - 1;
2885}
2886
2887static int relocate_one_folio(struct reloc_control *rc,
2888			      struct file_ra_state *ra,
2889			      int *cluster_nr, unsigned long index)
2890{
2891	const struct file_extent_cluster *cluster = &rc->cluster;
2892	struct inode *inode = rc->data_inode;
2893	struct btrfs_fs_info *fs_info = inode_to_fs_info(inode);
2894	u64 offset = BTRFS_I(inode)->reloc_block_group_start;
2895	const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT;
2896	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2897	struct folio *folio;
2898	u64 folio_start;
2899	u64 folio_end;
2900	u64 cur;
2901	int ret;
2902	const bool use_rst = btrfs_need_stripe_tree_update(fs_info, rc->block_group->flags);
2903
2904	ASSERT(index <= last_index);
2905again:
2906	folio = filemap_lock_folio(inode->i_mapping, index);
2907	if (IS_ERR(folio)) {
2908
2909		/*
2910		 * On relocation we're doing readahead on the relocation inode,
2911		 * but if the filesystem is backed by a RAID stripe tree we can
2912		 * get ENOENT (e.g. due to preallocated extents not being
2913		 * mapped in the RST) from the lookup.
2914		 *
2915		 * But readahead doesn't handle the error and submits invalid
2916		 * reads to the device, causing a assertion failures.
2917		 */
2918		if (!use_rst)
2919			page_cache_sync_readahead(inode->i_mapping, ra, NULL,
2920						  index, last_index + 1 - index);
2921		folio = __filemap_get_folio(inode->i_mapping, index,
2922					    FGP_LOCK | FGP_ACCESSED | FGP_CREAT,
2923					    mask);
2924		if (IS_ERR(folio))
2925			return PTR_ERR(folio);
2926	}
2927
2928	WARN_ON(folio_order(folio));
2929
2930	if (folio_test_readahead(folio) && !use_rst)
2931		page_cache_async_readahead(inode->i_mapping, ra, NULL,
2932					   folio, last_index + 1 - index);
2933
2934	if (!folio_test_uptodate(folio)) {
2935		btrfs_read_folio(NULL, folio);
2936		folio_lock(folio);
2937		if (!folio_test_uptodate(folio)) {
2938			ret = -EIO;
2939			goto release_folio;
2940		}
2941		if (folio->mapping != inode->i_mapping) {
2942			folio_unlock(folio);
2943			folio_put(folio);
2944			goto again;
2945		}
 
2946	}
2947
2948	/*
2949	 * We could have lost folio private when we dropped the lock to read the
2950	 * folio above, make sure we set_page_extent_mapped here so we have any
2951	 * of the subpage blocksize stuff we need in place.
2952	 */
2953	ret = set_folio_extent_mapped(folio);
2954	if (ret < 0)
2955		goto release_folio;
2956
2957	folio_start = folio_pos(folio);
2958	folio_end = folio_start + PAGE_SIZE - 1;
2959
2960	/*
2961	 * Start from the cluster, as for subpage case, the cluster can start
2962	 * inside the folio.
2963	 */
2964	cur = max(folio_start, cluster->boundary[*cluster_nr] - offset);
2965	while (cur <= folio_end) {
2966		struct extent_state *cached_state = NULL;
2967		u64 extent_start = cluster->boundary[*cluster_nr] - offset;
2968		u64 extent_end = get_cluster_boundary_end(cluster,
2969						*cluster_nr) - offset;
2970		u64 clamped_start = max(folio_start, extent_start);
2971		u64 clamped_end = min(folio_end, extent_end);
2972		u32 clamped_len = clamped_end + 1 - clamped_start;
2973
2974		/* Reserve metadata for this range */
2975		ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
2976						      clamped_len, clamped_len,
2977						      false);
2978		if (ret)
2979			goto release_folio;
2980
2981		/* Mark the range delalloc and dirty for later writeback */
2982		lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
2983			    &cached_state);
2984		ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start,
2985						clamped_end, 0, &cached_state);
2986		if (ret) {
2987			clear_extent_bit(&BTRFS_I(inode)->io_tree,
2988					 clamped_start, clamped_end,
2989					 EXTENT_LOCKED | EXTENT_BOUNDARY,
2990					 &cached_state);
2991			btrfs_delalloc_release_metadata(BTRFS_I(inode),
2992							clamped_len, true);
2993			btrfs_delalloc_release_extents(BTRFS_I(inode),
2994						       clamped_len);
2995			goto release_folio;
2996		}
2997		btrfs_folio_set_dirty(fs_info, folio, clamped_start, clamped_len);
2998
2999		/*
3000		 * Set the boundary if it's inside the folio.
3001		 * Data relocation requires the destination extents to have the
3002		 * same size as the source.
3003		 * EXTENT_BOUNDARY bit prevents current extent from being merged
3004		 * with previous extent.
3005		 */
3006		if (in_range(cluster->boundary[*cluster_nr] - offset, folio_start, PAGE_SIZE)) {
3007			u64 boundary_start = cluster->boundary[*cluster_nr] -
3008						offset;
3009			u64 boundary_end = boundary_start +
3010					   fs_info->sectorsize - 1;
3011
3012			set_extent_bit(&BTRFS_I(inode)->io_tree,
3013				       boundary_start, boundary_end,
3014				       EXTENT_BOUNDARY, NULL);
3015		}
3016		unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
3017			      &cached_state);
3018		btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len);
3019		cur += clamped_len;
3020
3021		/* Crossed extent end, go to next extent */
3022		if (cur >= extent_end) {
3023			(*cluster_nr)++;
3024			/* Just finished the last extent of the cluster, exit. */
3025			if (*cluster_nr >= cluster->nr)
3026				break;
3027		}
3028	}
3029	folio_unlock(folio);
3030	folio_put(folio);
3031
3032	balance_dirty_pages_ratelimited(inode->i_mapping);
3033	btrfs_throttle(fs_info);
3034	if (btrfs_should_cancel_balance(fs_info))
3035		ret = -ECANCELED;
3036	return ret;
3037
3038release_folio:
3039	folio_unlock(folio);
3040	folio_put(folio);
3041	return ret;
3042}
3043
3044static int relocate_file_extent_cluster(struct reloc_control *rc)
 
3045{
3046	struct inode *inode = rc->data_inode;
3047	const struct file_extent_cluster *cluster = &rc->cluster;
3048	u64 offset = BTRFS_I(inode)->reloc_block_group_start;
 
3049	unsigned long index;
3050	unsigned long last_index;
 
3051	struct file_ra_state *ra;
3052	int cluster_nr = 0;
 
3053	int ret = 0;
3054
3055	if (!cluster->nr)
3056		return 0;
3057
3058	ra = kzalloc(sizeof(*ra), GFP_NOFS);
3059	if (!ra)
3060		return -ENOMEM;
3061
3062	ret = prealloc_file_extent_cluster(rc);
3063	if (ret)
3064		goto out;
3065
3066	file_ra_state_init(ra, inode->i_mapping);
3067
3068	ret = setup_relocation_extent_mapping(rc);
 
3069	if (ret)
3070		goto out;
3071
 
3072	last_index = (cluster->end - offset) >> PAGE_SHIFT;
3073	for (index = (cluster->start - offset) >> PAGE_SHIFT;
3074	     index <= last_index && !ret; index++)
3075		ret = relocate_one_folio(rc, ra, &cluster_nr, index);
3076	if (ret == 0)
3077		WARN_ON(cluster_nr != cluster->nr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3078out:
3079	kfree(ra);
3080	return ret;
3081}
3082
3083static noinline_for_stack int relocate_data_extent(struct reloc_control *rc,
3084					   const struct btrfs_key *extent_key)
 
3085{
3086	struct inode *inode = rc->data_inode;
3087	struct file_extent_cluster *cluster = &rc->cluster;
3088	int ret;
3089	struct btrfs_root *root = BTRFS_I(inode)->root;
3090
3091	if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3092		ret = relocate_file_extent_cluster(rc);
3093		if (ret)
3094			return ret;
3095		cluster->nr = 0;
3096	}
3097
3098	/*
3099	 * Under simple quotas, we set root->relocation_src_root when we find
3100	 * the extent. If adjacent extents have different owners, we can't merge
3101	 * them while relocating. Handle this by storing the owning root that
3102	 * started a cluster and if we see an extent from a different root break
3103	 * cluster formation (just like the above case of non-adjacent extents).
3104	 *
3105	 * Without simple quotas, relocation_src_root is always 0, so we should
3106	 * never see a mismatch, and it should have no effect on relocation
3107	 * clusters.
3108	 */
3109	if (cluster->nr > 0 && cluster->owning_root != root->relocation_src_root) {
3110		u64 tmp = root->relocation_src_root;
3111
3112		/*
3113		 * root->relocation_src_root is the state that actually affects
3114		 * the preallocation we do here, so set it to the root owning
3115		 * the cluster we need to relocate.
3116		 */
3117		root->relocation_src_root = cluster->owning_root;
3118		ret = relocate_file_extent_cluster(rc);
3119		if (ret)
3120			return ret;
3121		cluster->nr = 0;
3122		/* And reset it back for the current extent's owning root. */
3123		root->relocation_src_root = tmp;
3124	}
3125
3126	if (!cluster->nr) {
3127		cluster->start = extent_key->objectid;
3128		cluster->owning_root = root->relocation_src_root;
3129	}
3130	else
3131		BUG_ON(cluster->nr >= MAX_EXTENTS);
3132	cluster->end = extent_key->objectid + extent_key->offset - 1;
3133	cluster->boundary[cluster->nr] = extent_key->objectid;
3134	cluster->nr++;
3135
3136	if (cluster->nr >= MAX_EXTENTS) {
3137		ret = relocate_file_extent_cluster(rc);
3138		if (ret)
3139			return ret;
3140		cluster->nr = 0;
3141	}
3142	return 0;
3143}
3144
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3145/*
3146 * helper to add a tree block to the list.
3147 * the major work is getting the generation and level of the block
3148 */
3149static int add_tree_block(struct reloc_control *rc,
3150			  const struct btrfs_key *extent_key,
3151			  struct btrfs_path *path,
3152			  struct rb_root *blocks)
3153{
3154	struct extent_buffer *eb;
3155	struct btrfs_extent_item *ei;
3156	struct btrfs_tree_block_info *bi;
3157	struct tree_block *block;
3158	struct rb_node *rb_node;
3159	u32 item_size;
3160	int level = -1;
3161	u64 generation;
3162	u64 owner = 0;
3163
3164	eb =  path->nodes[0];
3165	item_size = btrfs_item_size(eb, path->slots[0]);
3166
3167	if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3168	    item_size >= sizeof(*ei) + sizeof(*bi)) {
3169		unsigned long ptr = 0, end;
3170
3171		ei = btrfs_item_ptr(eb, path->slots[0],
3172				struct btrfs_extent_item);
3173		end = (unsigned long)ei + item_size;
3174		if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3175			bi = (struct btrfs_tree_block_info *)(ei + 1);
3176			level = btrfs_tree_block_level(eb, bi);
3177			ptr = (unsigned long)(bi + 1);
3178		} else {
3179			level = (int)extent_key->offset;
3180			ptr = (unsigned long)(ei + 1);
3181		}
3182		generation = btrfs_extent_generation(eb, ei);
3183
3184		/*
3185		 * We're reading random blocks without knowing their owner ahead
3186		 * of time.  This is ok most of the time, as all reloc roots and
3187		 * fs roots have the same lock type.  However normal trees do
3188		 * not, and the only way to know ahead of time is to read the
3189		 * inline ref offset.  We know it's an fs root if
3190		 *
3191		 * 1. There's more than one ref.
3192		 * 2. There's a SHARED_DATA_REF_KEY set.
3193		 * 3. FULL_BACKREF is set on the flags.
3194		 *
3195		 * Otherwise it's safe to assume that the ref offset == the
3196		 * owner of this block, so we can use that when calling
3197		 * read_tree_block.
3198		 */
3199		if (btrfs_extent_refs(eb, ei) == 1 &&
3200		    !(btrfs_extent_flags(eb, ei) &
3201		      BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
3202		    ptr < end) {
3203			struct btrfs_extent_inline_ref *iref;
3204			int type;
3205
3206			iref = (struct btrfs_extent_inline_ref *)ptr;
3207			type = btrfs_get_extent_inline_ref_type(eb, iref,
3208							BTRFS_REF_TYPE_BLOCK);
3209			if (type == BTRFS_REF_TYPE_INVALID)
3210				return -EINVAL;
3211			if (type == BTRFS_TREE_BLOCK_REF_KEY)
3212				owner = btrfs_extent_inline_ref_offset(eb, iref);
3213		}
3214	} else {
3215		btrfs_print_leaf(eb);
3216		btrfs_err(rc->block_group->fs_info,
3217			  "unrecognized tree backref at tree block %llu slot %u",
3218			  eb->start, path->slots[0]);
3219		btrfs_release_path(path);
3220		return -EUCLEAN;
 
 
 
 
 
 
 
 
 
 
3221	}
3222
3223	btrfs_release_path(path);
3224
3225	BUG_ON(level == -1);
3226
3227	block = kmalloc(sizeof(*block), GFP_NOFS);
3228	if (!block)
3229		return -ENOMEM;
3230
3231	block->bytenr = extent_key->objectid;
3232	block->key.objectid = rc->extent_root->fs_info->nodesize;
3233	block->key.offset = generation;
3234	block->level = level;
3235	block->key_ready = false;
3236	block->owner = owner;
3237
3238	rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
3239	if (rb_node)
3240		btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
3241				    -EEXIST);
3242
3243	return 0;
3244}
3245
3246/*
3247 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3248 */
3249static int __add_tree_block(struct reloc_control *rc,
3250			    u64 bytenr, u32 blocksize,
3251			    struct rb_root *blocks)
3252{
3253	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3254	struct btrfs_path *path;
3255	struct btrfs_key key;
3256	int ret;
3257	bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3258
3259	if (tree_block_processed(bytenr, rc))
3260		return 0;
3261
3262	if (rb_simple_search(blocks, bytenr))
3263		return 0;
3264
3265	path = btrfs_alloc_path();
3266	if (!path)
3267		return -ENOMEM;
3268again:
3269	key.objectid = bytenr;
3270	if (skinny) {
3271		key.type = BTRFS_METADATA_ITEM_KEY;
3272		key.offset = (u64)-1;
3273	} else {
3274		key.type = BTRFS_EXTENT_ITEM_KEY;
3275		key.offset = blocksize;
3276	}
3277
3278	path->search_commit_root = 1;
3279	path->skip_locking = 1;
3280	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3281	if (ret < 0)
3282		goto out;
3283
3284	if (ret > 0 && skinny) {
3285		if (path->slots[0]) {
3286			path->slots[0]--;
3287			btrfs_item_key_to_cpu(path->nodes[0], &key,
3288					      path->slots[0]);
3289			if (key.objectid == bytenr &&
3290			    (key.type == BTRFS_METADATA_ITEM_KEY ||
3291			     (key.type == BTRFS_EXTENT_ITEM_KEY &&
3292			      key.offset == blocksize)))
3293				ret = 0;
3294		}
3295
3296		if (ret) {
3297			skinny = false;
3298			btrfs_release_path(path);
3299			goto again;
3300		}
3301	}
3302	if (ret) {
3303		ASSERT(ret == 1);
3304		btrfs_print_leaf(path->nodes[0]);
3305		btrfs_err(fs_info,
3306	     "tree block extent item (%llu) is not found in extent tree",
3307		     bytenr);
3308		WARN_ON(1);
3309		ret = -EINVAL;
3310		goto out;
3311	}
3312
3313	ret = add_tree_block(rc, &key, path, blocks);
3314out:
3315	btrfs_free_path(path);
3316	return ret;
3317}
3318
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3319static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3320				    struct btrfs_block_group *block_group,
3321				    struct inode *inode,
3322				    u64 ino)
3323{
 
3324	struct btrfs_root *root = fs_info->tree_root;
3325	struct btrfs_trans_handle *trans;
3326	int ret = 0;
3327
3328	if (inode)
3329		goto truncate;
3330
3331	inode = btrfs_iget(ino, root);
3332	if (IS_ERR(inode))
 
 
 
 
 
 
3333		return -ENOENT;
 
3334
3335truncate:
3336	ret = btrfs_check_trunc_cache_free_space(fs_info,
3337						 &fs_info->global_block_rsv);
3338	if (ret)
3339		goto out;
3340
3341	trans = btrfs_join_transaction(root);
3342	if (IS_ERR(trans)) {
3343		ret = PTR_ERR(trans);
3344		goto out;
3345	}
3346
3347	ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3348
3349	btrfs_end_transaction(trans);
3350	btrfs_btree_balance_dirty(fs_info);
3351out:
3352	iput(inode);
3353	return ret;
3354}
3355
3356/*
3357 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3358 * cache inode, to avoid free space cache data extent blocking data relocation.
3359 */
3360static int delete_v1_space_cache(struct extent_buffer *leaf,
3361				 struct btrfs_block_group *block_group,
3362				 u64 data_bytenr)
 
 
3363{
3364	u64 space_cache_ino;
3365	struct btrfs_file_extent_item *ei;
 
 
 
 
3366	struct btrfs_key key;
3367	bool found = false;
3368	int i;
 
 
 
 
 
 
3369	int ret;
3370
3371	if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3372		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3373
3374	for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3375		u8 type;
 
3376
3377		btrfs_item_key_to_cpu(leaf, &key, i);
3378		if (key.type != BTRFS_EXTENT_DATA_KEY)
3379			continue;
3380		ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3381		type = btrfs_file_extent_type(leaf, ei);
 
 
 
 
 
 
 
3382
3383		if ((type == BTRFS_FILE_EXTENT_REG ||
3384		     type == BTRFS_FILE_EXTENT_PREALLOC) &&
3385		    btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3386			found = true;
3387			space_cache_ino = key.objectid;
3388			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3389		}
 
 
 
 
 
 
 
3390	}
3391	if (!found)
3392		return -ENOENT;
3393	ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3394					space_cache_ino);
3395	return ret;
3396}
3397
3398/*
3399 * helper to find all tree blocks that reference a given data extent
3400 */
3401static noinline_for_stack int add_data_references(struct reloc_control *rc,
3402						  const struct btrfs_key *extent_key,
3403						  struct btrfs_path *path,
3404						  struct rb_root *blocks)
3405{
3406	struct btrfs_backref_walk_ctx ctx = { 0 };
3407	struct ulist_iterator leaf_uiter;
3408	struct ulist_node *ref_node = NULL;
3409	const u32 blocksize = rc->extent_root->fs_info->nodesize;
 
 
 
 
3410	int ret = 0;
 
3411
3412	btrfs_release_path(path);
 
 
 
 
 
 
 
 
3413
3414	ctx.bytenr = extent_key->objectid;
3415	ctx.skip_inode_ref_list = true;
3416	ctx.fs_info = rc->extent_root->fs_info;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3417
3418	ret = btrfs_find_all_leafs(&ctx);
3419	if (ret < 0)
3420		return ret;
 
 
 
 
 
 
 
 
 
 
3421
3422	ULIST_ITER_INIT(&leaf_uiter);
3423	while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) {
3424		struct btrfs_tree_parent_check check = { 0 };
3425		struct extent_buffer *eb;
3426
3427		eb = read_tree_block(ctx.fs_info, ref_node->val, &check);
3428		if (IS_ERR(eb)) {
3429			ret = PTR_ERR(eb);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3430			break;
3431		}
3432		ret = delete_v1_space_cache(eb, rc->block_group,
3433					    extent_key->objectid);
3434		free_extent_buffer(eb);
3435		if (ret < 0)
3436			break;
3437		ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3438		if (ret < 0)
3439			break;
3440	}
3441	if (ret < 0)
 
 
3442		free_block_list(blocks);
3443	ulist_free(ctx.refs);
3444	return ret;
3445}
3446
3447/*
3448 * helper to find next unprocessed extent
3449 */
3450static noinline_for_stack
3451int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3452		     struct btrfs_key *extent_key)
3453{
3454	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3455	struct btrfs_key key;
3456	struct extent_buffer *leaf;
3457	u64 start, end, last;
3458	int ret;
3459
3460	last = rc->block_group->start + rc->block_group->length;
3461	while (1) {
3462		bool block_found;
3463
3464		cond_resched();
3465		if (rc->search_start >= last) {
3466			ret = 1;
3467			break;
3468		}
3469
3470		key.objectid = rc->search_start;
3471		key.type = BTRFS_EXTENT_ITEM_KEY;
3472		key.offset = 0;
3473
3474		path->search_commit_root = 1;
3475		path->skip_locking = 1;
3476		ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3477					0, 0);
3478		if (ret < 0)
3479			break;
3480next:
3481		leaf = path->nodes[0];
3482		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3483			ret = btrfs_next_leaf(rc->extent_root, path);
3484			if (ret != 0)
3485				break;
3486			leaf = path->nodes[0];
3487		}
3488
3489		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3490		if (key.objectid >= last) {
3491			ret = 1;
3492			break;
3493		}
3494
3495		if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3496		    key.type != BTRFS_METADATA_ITEM_KEY) {
3497			path->slots[0]++;
3498			goto next;
3499		}
3500
3501		if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3502		    key.objectid + key.offset <= rc->search_start) {
3503			path->slots[0]++;
3504			goto next;
3505		}
3506
3507		if (key.type == BTRFS_METADATA_ITEM_KEY &&
3508		    key.objectid + fs_info->nodesize <=
3509		    rc->search_start) {
3510			path->slots[0]++;
3511			goto next;
3512		}
3513
3514		block_found = find_first_extent_bit(&rc->processed_blocks,
3515						    key.objectid, &start, &end,
3516						    EXTENT_DIRTY, NULL);
3517
3518		if (block_found && start <= key.objectid) {
3519			btrfs_release_path(path);
3520			rc->search_start = end + 1;
3521		} else {
3522			if (key.type == BTRFS_EXTENT_ITEM_KEY)
3523				rc->search_start = key.objectid + key.offset;
3524			else
3525				rc->search_start = key.objectid +
3526					fs_info->nodesize;
3527			memcpy(extent_key, &key, sizeof(key));
3528			return 0;
3529		}
3530	}
3531	btrfs_release_path(path);
3532	return ret;
3533}
3534
3535static void set_reloc_control(struct reloc_control *rc)
3536{
3537	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3538
3539	mutex_lock(&fs_info->reloc_mutex);
3540	fs_info->reloc_ctl = rc;
3541	mutex_unlock(&fs_info->reloc_mutex);
3542}
3543
3544static void unset_reloc_control(struct reloc_control *rc)
3545{
3546	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3547
3548	mutex_lock(&fs_info->reloc_mutex);
3549	fs_info->reloc_ctl = NULL;
3550	mutex_unlock(&fs_info->reloc_mutex);
3551}
3552
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3553static noinline_for_stack
3554int prepare_to_relocate(struct reloc_control *rc)
3555{
3556	struct btrfs_trans_handle *trans;
3557	int ret;
3558
3559	rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3560					      BTRFS_BLOCK_RSV_TEMP);
3561	if (!rc->block_rsv)
3562		return -ENOMEM;
3563
3564	memset(&rc->cluster, 0, sizeof(rc->cluster));
3565	rc->search_start = rc->block_group->start;
3566	rc->extents_found = 0;
3567	rc->nodes_relocated = 0;
3568	rc->merging_rsv_size = 0;
3569	rc->reserved_bytes = 0;
3570	rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3571			      RELOCATION_RESERVED_NODES;
3572	ret = btrfs_block_rsv_refill(rc->extent_root->fs_info,
3573				     rc->block_rsv, rc->block_rsv->size,
3574				     BTRFS_RESERVE_FLUSH_ALL);
3575	if (ret)
3576		return ret;
3577
3578	rc->create_reloc_tree = true;
3579	set_reloc_control(rc);
3580
3581	trans = btrfs_join_transaction(rc->extent_root);
3582	if (IS_ERR(trans)) {
3583		unset_reloc_control(rc);
3584		/*
3585		 * extent tree is not a ref_cow tree and has no reloc_root to
3586		 * cleanup.  And callers are responsible to free the above
3587		 * block rsv.
3588		 */
3589		return PTR_ERR(trans);
3590	}
3591
3592	ret = btrfs_commit_transaction(trans);
3593	if (ret)
3594		unset_reloc_control(rc);
3595
3596	return ret;
3597}
3598
3599static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3600{
3601	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3602	struct rb_root blocks = RB_ROOT;
3603	struct btrfs_key key;
3604	struct btrfs_trans_handle *trans = NULL;
3605	struct btrfs_path *path;
3606	struct btrfs_extent_item *ei;
3607	u64 flags;
 
3608	int ret;
3609	int err = 0;
3610	int progress = 0;
3611
3612	path = btrfs_alloc_path();
3613	if (!path)
3614		return -ENOMEM;
3615	path->reada = READA_FORWARD;
3616
3617	ret = prepare_to_relocate(rc);
3618	if (ret) {
3619		err = ret;
3620		goto out_free;
3621	}
3622
3623	while (1) {
3624		rc->reserved_bytes = 0;
3625		ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
3626					     rc->block_rsv->size,
3627					     BTRFS_RESERVE_FLUSH_ALL);
3628		if (ret) {
3629			err = ret;
3630			break;
3631		}
3632		progress++;
3633		trans = btrfs_start_transaction(rc->extent_root, 0);
3634		if (IS_ERR(trans)) {
3635			err = PTR_ERR(trans);
3636			trans = NULL;
3637			break;
3638		}
3639restart:
3640		if (rc->backref_cache.last_trans != trans->transid)
3641			btrfs_backref_release_cache(&rc->backref_cache);
3642		rc->backref_cache.last_trans = trans->transid;
 
3643
3644		ret = find_next_extent(rc, path, &key);
3645		if (ret < 0)
3646			err = ret;
3647		if (ret != 0)
3648			break;
3649
3650		rc->extents_found++;
3651
3652		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3653				    struct btrfs_extent_item);
3654		flags = btrfs_extent_flags(path->nodes[0], ei);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3655
3656		/*
3657		 * If we are relocating a simple quota owned extent item, we
3658		 * need to note the owner on the reloc data root so that when
3659		 * we allocate the replacement item, we can attribute it to the
3660		 * correct eventual owner (rather than the reloc data root).
3661		 */
3662		if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) {
3663			struct btrfs_root *root = BTRFS_I(rc->data_inode)->root;
3664			u64 owning_root_id = btrfs_get_extent_owner_root(fs_info,
3665								 path->nodes[0],
3666								 path->slots[0]);
3667
3668			root->relocation_src_root = owning_root_id;
 
 
 
 
 
 
 
 
 
 
 
 
3669		}
3670
3671		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3672			ret = add_tree_block(rc, &key, path, &blocks);
3673		} else if (rc->stage == UPDATE_DATA_PTRS &&
3674			   (flags & BTRFS_EXTENT_FLAG_DATA)) {
3675			ret = add_data_references(rc, &key, path, &blocks);
3676		} else {
3677			btrfs_release_path(path);
3678			ret = 0;
3679		}
3680		if (ret < 0) {
3681			err = ret;
3682			break;
3683		}
3684
3685		if (!RB_EMPTY_ROOT(&blocks)) {
3686			ret = relocate_tree_blocks(trans, rc, &blocks);
3687			if (ret < 0) {
 
 
 
 
 
 
3688				if (ret != -EAGAIN) {
3689					err = ret;
3690					break;
3691				}
3692				rc->extents_found--;
3693				rc->search_start = key.objectid;
3694			}
3695		}
3696
3697		btrfs_end_transaction_throttle(trans);
3698		btrfs_btree_balance_dirty(fs_info);
3699		trans = NULL;
3700
3701		if (rc->stage == MOVE_DATA_EXTENTS &&
3702		    (flags & BTRFS_EXTENT_FLAG_DATA)) {
3703			rc->found_file_extent = true;
3704			ret = relocate_data_extent(rc, &key);
 
3705			if (ret < 0) {
3706				err = ret;
3707				break;
3708			}
3709		}
3710		if (btrfs_should_cancel_balance(fs_info)) {
3711			err = -ECANCELED;
3712			break;
3713		}
3714	}
3715	if (trans && progress && err == -ENOSPC) {
3716		ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
 
3717		if (ret == 1) {
3718			err = 0;
3719			progress = 0;
3720			goto restart;
3721		}
3722	}
3723
3724	btrfs_release_path(path);
3725	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3726
3727	if (trans) {
3728		btrfs_end_transaction_throttle(trans);
3729		btrfs_btree_balance_dirty(fs_info);
3730	}
3731
3732	if (!err) {
3733		ret = relocate_file_extent_cluster(rc);
 
3734		if (ret < 0)
3735			err = ret;
3736	}
3737
3738	rc->create_reloc_tree = false;
3739	set_reloc_control(rc);
3740
3741	btrfs_backref_release_cache(&rc->backref_cache);
3742	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3743
3744	/*
3745	 * Even in the case when the relocation is cancelled, we should all go
3746	 * through prepare_to_merge() and merge_reloc_roots().
3747	 *
3748	 * For error (including cancelled balance), prepare_to_merge() will
3749	 * mark all reloc trees orphan, then queue them for cleanup in
3750	 * merge_reloc_roots()
3751	 */
3752	err = prepare_to_merge(rc, err);
3753
3754	merge_reloc_roots(rc);
3755
3756	rc->merge_reloc_tree = false;
3757	unset_reloc_control(rc);
3758	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3759
3760	/* get rid of pinned extents */
3761	trans = btrfs_join_transaction(rc->extent_root);
3762	if (IS_ERR(trans)) {
3763		err = PTR_ERR(trans);
3764		goto out_free;
3765	}
3766	ret = btrfs_commit_transaction(trans);
3767	if (ret && !err)
3768		err = ret;
3769out_free:
3770	ret = clean_dirty_subvols(rc);
3771	if (ret < 0 && !err)
3772		err = ret;
3773	btrfs_free_block_rsv(fs_info, rc->block_rsv);
3774	btrfs_free_path(path);
3775	return err;
3776}
3777
3778static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3779				 struct btrfs_root *root, u64 objectid)
3780{
3781	struct btrfs_path *path;
3782	struct btrfs_inode_item *item;
3783	struct extent_buffer *leaf;
3784	int ret;
3785
3786	path = btrfs_alloc_path();
3787	if (!path)
3788		return -ENOMEM;
3789
3790	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3791	if (ret)
3792		goto out;
3793
3794	leaf = path->nodes[0];
3795	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3796	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3797	btrfs_set_inode_generation(leaf, item, 1);
3798	btrfs_set_inode_size(leaf, item, 0);
3799	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3800	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3801					  BTRFS_INODE_PREALLOC);
3802	btrfs_mark_buffer_dirty(trans, leaf);
3803out:
3804	btrfs_free_path(path);
3805	return ret;
3806}
3807
3808static void delete_orphan_inode(struct btrfs_trans_handle *trans,
3809				struct btrfs_root *root, u64 objectid)
3810{
3811	struct btrfs_path *path;
3812	struct btrfs_key key;
3813	int ret = 0;
3814
3815	path = btrfs_alloc_path();
3816	if (!path) {
3817		ret = -ENOMEM;
3818		goto out;
3819	}
3820
3821	key.objectid = objectid;
3822	key.type = BTRFS_INODE_ITEM_KEY;
3823	key.offset = 0;
3824	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3825	if (ret) {
3826		if (ret > 0)
3827			ret = -ENOENT;
3828		goto out;
3829	}
3830	ret = btrfs_del_item(trans, root, path);
3831out:
3832	if (ret)
3833		btrfs_abort_transaction(trans, ret);
3834	btrfs_free_path(path);
3835}
3836
3837/*
3838 * helper to create inode for data relocation.
3839 * the inode is in data relocation tree and its link count is 0
3840 */
3841static noinline_for_stack struct inode *create_reloc_inode(
3842					struct btrfs_fs_info *fs_info,
3843					const struct btrfs_block_group *group)
3844{
3845	struct inode *inode = NULL;
3846	struct btrfs_trans_handle *trans;
3847	struct btrfs_root *root;
 
3848	u64 objectid;
3849	int ret = 0;
 
 
 
 
3850
3851	root = btrfs_grab_root(fs_info->data_reloc_root);
3852	trans = btrfs_start_transaction(root, 6);
3853	if (IS_ERR(trans)) {
3854		btrfs_put_root(root);
3855		return ERR_CAST(trans);
3856	}
3857
3858	ret = btrfs_get_free_objectid(root, &objectid);
3859	if (ret)
3860		goto out;
3861
3862	ret = __insert_orphan_inode(trans, root, objectid);
3863	if (ret)
3864		goto out;
3865
3866	inode = btrfs_iget(objectid, root);
3867	if (IS_ERR(inode)) {
3868		delete_orphan_inode(trans, root, objectid);
3869		ret = PTR_ERR(inode);
3870		inode = NULL;
3871		goto out;
3872	}
3873	BTRFS_I(inode)->reloc_block_group_start = group->start;
3874
3875	ret = btrfs_orphan_add(trans, BTRFS_I(inode));
3876out:
3877	btrfs_put_root(root);
3878	btrfs_end_transaction(trans);
3879	btrfs_btree_balance_dirty(fs_info);
3880	if (ret) {
3881		iput(inode);
3882		inode = ERR_PTR(ret);
 
3883	}
3884	return inode;
3885}
3886
3887/*
3888 * Mark start of chunk relocation that is cancellable. Check if the cancellation
3889 * has been requested meanwhile and don't start in that case.
3890 *
3891 * Return:
3892 *   0             success
3893 *   -EINPROGRESS  operation is already in progress, that's probably a bug
3894 *   -ECANCELED    cancellation request was set before the operation started
3895 */
3896static int reloc_chunk_start(struct btrfs_fs_info *fs_info)
3897{
3898	if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) {
3899		/* This should not happen */
3900		btrfs_err(fs_info, "reloc already running, cannot start");
3901		return -EINPROGRESS;
3902	}
3903
3904	if (atomic_read(&fs_info->reloc_cancel_req) > 0) {
3905		btrfs_info(fs_info, "chunk relocation canceled on start");
3906		/*
3907		 * On cancel, clear all requests but let the caller mark
3908		 * the end after cleanup operations.
3909		 */
3910		atomic_set(&fs_info->reloc_cancel_req, 0);
3911		return -ECANCELED;
3912	}
3913	return 0;
3914}
3915
3916/*
3917 * Mark end of chunk relocation that is cancellable and wake any waiters.
3918 */
3919static void reloc_chunk_end(struct btrfs_fs_info *fs_info)
3920{
3921	/* Requested after start, clear bit first so any waiters can continue */
3922	if (atomic_read(&fs_info->reloc_cancel_req) > 0)
3923		btrfs_info(fs_info, "chunk relocation canceled during operation");
3924	clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags);
3925	atomic_set(&fs_info->reloc_cancel_req, 0);
3926}
3927
3928static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3929{
3930	struct reloc_control *rc;
3931
3932	rc = kzalloc(sizeof(*rc), GFP_NOFS);
3933	if (!rc)
3934		return NULL;
3935
3936	INIT_LIST_HEAD(&rc->reloc_roots);
3937	INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3938	btrfs_backref_init_cache(fs_info, &rc->backref_cache, true);
3939	rc->reloc_root_tree.rb_root = RB_ROOT;
3940	spin_lock_init(&rc->reloc_root_tree.lock);
3941	extent_io_tree_init(fs_info, &rc->processed_blocks, IO_TREE_RELOC_BLOCKS);
3942	return rc;
3943}
3944
3945static void free_reloc_control(struct reloc_control *rc)
3946{
3947	struct mapping_node *node, *tmp;
3948
3949	free_reloc_roots(&rc->reloc_roots);
3950	rbtree_postorder_for_each_entry_safe(node, tmp,
3951			&rc->reloc_root_tree.rb_root, rb_node)
3952		kfree(node);
3953
3954	kfree(rc);
3955}
3956
3957/*
3958 * Print the block group being relocated
3959 */
3960static void describe_relocation(struct btrfs_block_group *block_group)
 
3961{
3962	char buf[128] = {'\0'};
 
3963
3964	btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
3965
3966	btrfs_info(block_group->fs_info, "relocating block group %llu flags %s",
3967		   block_group->start, buf);
3968}
3969
3970static const char *stage_to_string(enum reloc_stage stage)
3971{
3972	if (stage == MOVE_DATA_EXTENTS)
3973		return "move data extents";
3974	if (stage == UPDATE_DATA_PTRS)
3975		return "update data pointers";
3976	return "unknown";
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3977}
3978
3979/*
3980 * function to relocate all extents in a block group.
3981 */
3982int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3983{
3984	struct btrfs_block_group *bg;
3985	struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start);
3986	struct reloc_control *rc;
3987	struct inode *inode;
3988	struct btrfs_path *path;
3989	int ret;
3990	int rw = 0;
3991	int err = 0;
3992
3993	/*
3994	 * This only gets set if we had a half-deleted snapshot on mount.  We
3995	 * cannot allow relocation to start while we're still trying to clean up
3996	 * these pending deletions.
3997	 */
3998	ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE);
3999	if (ret)
4000		return ret;
4001
4002	/* We may have been woken up by close_ctree, so bail if we're closing. */
4003	if (btrfs_fs_closing(fs_info))
4004		return -EINTR;
4005
4006	bg = btrfs_lookup_block_group(fs_info, group_start);
4007	if (!bg)
4008		return -ENOENT;
4009
4010	/*
4011	 * Relocation of a data block group creates ordered extents.  Without
4012	 * sb_start_write(), we can freeze the filesystem while unfinished
4013	 * ordered extents are left. Such ordered extents can cause a deadlock
4014	 * e.g. when syncfs() is waiting for their completion but they can't
4015	 * finish because they block when joining a transaction, due to the
4016	 * fact that the freeze locks are being held in write mode.
4017	 */
4018	if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
4019		ASSERT(sb_write_started(fs_info->sb));
4020
4021	if (btrfs_pinned_by_swapfile(fs_info, bg)) {
4022		btrfs_put_block_group(bg);
4023		return -ETXTBSY;
4024	}
4025
4026	rc = alloc_reloc_control(fs_info);
4027	if (!rc) {
4028		btrfs_put_block_group(bg);
4029		return -ENOMEM;
4030	}
4031
4032	ret = reloc_chunk_start(fs_info);
4033	if (ret < 0) {
4034		err = ret;
4035		goto out_put_bg;
4036	}
4037
4038	rc->extent_root = extent_root;
4039	rc->block_group = bg;
4040
4041	ret = btrfs_inc_block_group_ro(rc->block_group, true);
4042	if (ret) {
4043		err = ret;
4044		goto out;
4045	}
4046	rw = 1;
4047
4048	path = btrfs_alloc_path();
4049	if (!path) {
4050		err = -ENOMEM;
4051		goto out;
4052	}
4053
4054	inode = lookup_free_space_inode(rc->block_group, path);
 
4055	btrfs_free_path(path);
4056
4057	if (!IS_ERR(inode))
4058		ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4059	else
4060		ret = PTR_ERR(inode);
4061
4062	if (ret && ret != -ENOENT) {
4063		err = ret;
4064		goto out;
4065	}
4066
4067	rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4068	if (IS_ERR(rc->data_inode)) {
4069		err = PTR_ERR(rc->data_inode);
4070		rc->data_inode = NULL;
4071		goto out;
4072	}
4073
4074	describe_relocation(rc->block_group);
4075
4076	btrfs_wait_block_group_reservations(rc->block_group);
4077	btrfs_wait_nocow_writers(rc->block_group);
4078	btrfs_wait_ordered_roots(fs_info, U64_MAX, rc->block_group);
4079
4080	ret = btrfs_zone_finish(rc->block_group);
4081	WARN_ON(ret && ret != -EAGAIN);
4082
4083	while (1) {
4084		enum reloc_stage finishes_stage;
4085
4086		mutex_lock(&fs_info->cleaner_mutex);
4087		ret = relocate_block_group(rc);
4088		mutex_unlock(&fs_info->cleaner_mutex);
4089		if (ret < 0)
4090			err = ret;
 
 
 
 
 
 
 
4091
4092		finishes_stage = rc->stage;
4093		/*
4094		 * We may have gotten ENOSPC after we already dirtied some
4095		 * extents.  If writeout happens while we're relocating a
4096		 * different block group we could end up hitting the
4097		 * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in
4098		 * btrfs_reloc_cow_block.  Make sure we write everything out
4099		 * properly so we don't trip over this problem, and then break
4100		 * out of the loop if we hit an error.
4101		 */
4102		if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4103			ret = btrfs_wait_ordered_range(BTRFS_I(rc->data_inode), 0,
4104						       (u64)-1);
4105			if (ret)
4106				err = ret;
 
 
4107			invalidate_mapping_pages(rc->data_inode->i_mapping,
4108						 0, -1);
4109			rc->stage = UPDATE_DATA_PTRS;
4110		}
4111
4112		if (err < 0)
4113			goto out;
4114
4115		if (rc->extents_found == 0)
4116			break;
4117
4118		btrfs_info(fs_info, "found %llu extents, stage: %s",
4119			   rc->extents_found, stage_to_string(finishes_stage));
4120	}
4121
4122	WARN_ON(rc->block_group->pinned > 0);
4123	WARN_ON(rc->block_group->reserved > 0);
4124	WARN_ON(rc->block_group->used > 0);
4125out:
4126	if (err && rw)
4127		btrfs_dec_block_group_ro(rc->block_group);
4128	iput(rc->data_inode);
4129out_put_bg:
4130	btrfs_put_block_group(bg);
4131	reloc_chunk_end(fs_info);
4132	free_reloc_control(rc);
4133	return err;
4134}
4135
4136static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4137{
4138	struct btrfs_fs_info *fs_info = root->fs_info;
4139	struct btrfs_trans_handle *trans;
4140	int ret, err;
4141
4142	trans = btrfs_start_transaction(fs_info->tree_root, 0);
4143	if (IS_ERR(trans))
4144		return PTR_ERR(trans);
4145
4146	memset(&root->root_item.drop_progress, 0,
4147		sizeof(root->root_item.drop_progress));
4148	btrfs_set_root_drop_level(&root->root_item, 0);
4149	btrfs_set_root_refs(&root->root_item, 0);
4150	ret = btrfs_update_root(trans, fs_info->tree_root,
4151				&root->root_key, &root->root_item);
4152
4153	err = btrfs_end_transaction(trans);
4154	if (err)
4155		return err;
4156	return ret;
4157}
4158
4159/*
4160 * recover relocation interrupted by system crash.
4161 *
4162 * this function resumes merging reloc trees with corresponding fs trees.
4163 * this is important for keeping the sharing of tree blocks
4164 */
4165int btrfs_recover_relocation(struct btrfs_fs_info *fs_info)
4166{
 
4167	LIST_HEAD(reloc_roots);
4168	struct btrfs_key key;
4169	struct btrfs_root *fs_root;
4170	struct btrfs_root *reloc_root;
4171	struct btrfs_path *path;
4172	struct extent_buffer *leaf;
4173	struct reloc_control *rc = NULL;
4174	struct btrfs_trans_handle *trans;
4175	int ret2;
4176	int ret = 0;
4177
4178	path = btrfs_alloc_path();
4179	if (!path)
4180		return -ENOMEM;
4181	path->reada = READA_BACK;
4182
4183	key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4184	key.type = BTRFS_ROOT_ITEM_KEY;
4185	key.offset = (u64)-1;
4186
4187	while (1) {
4188		ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4189					path, 0, 0);
4190		if (ret < 0)
 
4191			goto out;
 
4192		if (ret > 0) {
4193			if (path->slots[0] == 0)
4194				break;
4195			path->slots[0]--;
4196		}
4197		ret = 0;
4198		leaf = path->nodes[0];
4199		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4200		btrfs_release_path(path);
4201
4202		if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4203		    key.type != BTRFS_ROOT_ITEM_KEY)
4204			break;
4205
4206		reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key);
4207		if (IS_ERR(reloc_root)) {
4208			ret = PTR_ERR(reloc_root);
4209			goto out;
4210		}
4211
4212		set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
4213		list_add(&reloc_root->root_list, &reloc_roots);
4214
4215		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4216			fs_root = btrfs_get_fs_root(fs_info,
4217					reloc_root->root_key.offset, false);
4218			if (IS_ERR(fs_root)) {
4219				ret = PTR_ERR(fs_root);
4220				if (ret != -ENOENT)
 
4221					goto out;
 
4222				ret = mark_garbage_root(reloc_root);
4223				if (ret < 0)
 
4224					goto out;
4225				ret = 0;
4226			} else {
4227				btrfs_put_root(fs_root);
4228			}
4229		}
4230
4231		if (key.offset == 0)
4232			break;
4233
4234		key.offset--;
4235	}
4236	btrfs_release_path(path);
4237
4238	if (list_empty(&reloc_roots))
4239		goto out;
4240
4241	rc = alloc_reloc_control(fs_info);
4242	if (!rc) {
4243		ret = -ENOMEM;
4244		goto out;
4245	}
4246
4247	ret = reloc_chunk_start(fs_info);
4248	if (ret < 0)
4249		goto out_end;
4250
4251	rc->extent_root = btrfs_extent_root(fs_info, 0);
4252
4253	set_reloc_control(rc);
4254
4255	trans = btrfs_join_transaction(rc->extent_root);
4256	if (IS_ERR(trans)) {
4257		ret = PTR_ERR(trans);
4258		goto out_unset;
 
4259	}
4260
4261	rc->merge_reloc_tree = true;
4262
4263	while (!list_empty(&reloc_roots)) {
4264		reloc_root = list_entry(reloc_roots.next,
4265					struct btrfs_root, root_list);
4266		list_del(&reloc_root->root_list);
4267
4268		if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4269			list_add_tail(&reloc_root->root_list,
4270				      &rc->reloc_roots);
4271			continue;
4272		}
4273
4274		fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
4275					    false);
4276		if (IS_ERR(fs_root)) {
4277			ret = PTR_ERR(fs_root);
4278			list_add_tail(&reloc_root->root_list, &reloc_roots);
4279			btrfs_end_transaction(trans);
4280			goto out_unset;
4281		}
4282
4283		ret = __add_reloc_root(reloc_root);
4284		ASSERT(ret != -EEXIST);
4285		if (ret) {
4286			list_add_tail(&reloc_root->root_list, &reloc_roots);
4287			btrfs_put_root(fs_root);
4288			btrfs_end_transaction(trans);
4289			goto out_unset;
4290		}
4291		fs_root->reloc_root = btrfs_grab_root(reloc_root);
4292		btrfs_put_root(fs_root);
4293	}
4294
4295	ret = btrfs_commit_transaction(trans);
4296	if (ret)
4297		goto out_unset;
4298
4299	merge_reloc_roots(rc);
4300
4301	unset_reloc_control(rc);
4302
4303	trans = btrfs_join_transaction(rc->extent_root);
4304	if (IS_ERR(trans)) {
4305		ret = PTR_ERR(trans);
4306		goto out_clean;
4307	}
4308	ret = btrfs_commit_transaction(trans);
4309out_clean:
4310	ret2 = clean_dirty_subvols(rc);
4311	if (ret2 < 0 && !ret)
4312		ret = ret2;
4313out_unset:
4314	unset_reloc_control(rc);
4315out_end:
4316	reloc_chunk_end(fs_info);
4317	free_reloc_control(rc);
4318out:
4319	free_reloc_roots(&reloc_roots);
 
4320
4321	btrfs_free_path(path);
4322
4323	if (ret == 0) {
4324		/* cleanup orphan inode in data relocation tree */
4325		fs_root = btrfs_grab_root(fs_info->data_reloc_root);
4326		ASSERT(fs_root);
4327		ret = btrfs_orphan_cleanup(fs_root);
4328		btrfs_put_root(fs_root);
 
4329	}
4330	return ret;
4331}
4332
4333/*
4334 * helper to add ordered checksum for data relocation.
4335 *
4336 * cloning checksum properly handles the nodatasum extents.
4337 * it also saves CPU time to re-calculate the checksum.
4338 */
4339int btrfs_reloc_clone_csums(struct btrfs_ordered_extent *ordered)
4340{
4341	struct btrfs_inode *inode = ordered->inode;
4342	struct btrfs_fs_info *fs_info = inode->root->fs_info;
4343	u64 disk_bytenr = ordered->file_offset + inode->reloc_block_group_start;
4344	struct btrfs_root *csum_root = btrfs_csum_root(fs_info, disk_bytenr);
 
 
4345	LIST_HEAD(list);
4346	int ret;
4347
4348	ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4349				      disk_bytenr + ordered->num_bytes - 1,
4350				      &list, false);
4351	if (ret < 0) {
4352		btrfs_mark_ordered_extent_error(ordered);
4353		return ret;
4354	}
 
4355
4356	while (!list_empty(&list)) {
4357		struct btrfs_ordered_sum *sums =
4358			list_entry(list.next, struct btrfs_ordered_sum, list);
4359
4360		list_del_init(&sums->list);
4361
4362		/*
4363		 * We need to offset the new_bytenr based on where the csum is.
4364		 * We need to do this because we will read in entire prealloc
4365		 * extents but we may have written to say the middle of the
4366		 * prealloc extent, so we need to make sure the csum goes with
4367		 * the right disk offset.
4368		 *
4369		 * We can do this because the data reloc inode refers strictly
4370		 * to the on disk bytes, so we don't have to worry about
4371		 * disk_len vs real len like with real inodes since it's all
4372		 * disk length.
4373		 */
4374		sums->logical = ordered->disk_bytenr + sums->logical - disk_bytenr;
4375		btrfs_add_ordered_sum(ordered, sums);
 
 
4376	}
4377
4378	return 0;
 
4379}
4380
4381int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4382			  struct btrfs_root *root,
4383			  const struct extent_buffer *buf,
4384			  struct extent_buffer *cow)
4385{
4386	struct btrfs_fs_info *fs_info = root->fs_info;
4387	struct reloc_control *rc;
4388	struct btrfs_backref_node *node;
4389	int first_cow = 0;
4390	int level;
4391	int ret = 0;
4392
4393	rc = fs_info->reloc_ctl;
4394	if (!rc)
4395		return 0;
4396
4397	BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root));
 
 
 
 
 
 
4398
4399	level = btrfs_header_level(buf);
4400	if (btrfs_header_generation(buf) <=
4401	    btrfs_root_last_snapshot(&root->root_item))
4402		first_cow = 1;
4403
4404	if (btrfs_root_id(root) == BTRFS_TREE_RELOC_OBJECTID && rc->create_reloc_tree) {
 
4405		WARN_ON(!first_cow && level == 0);
4406
4407		node = rc->backref_cache.path[level];
 
 
4408
4409		/*
4410		 * If node->bytenr != buf->start and node->new_bytenr !=
4411		 * buf->start then we've got the wrong backref node for what we
4412		 * expected to see here and the cache is incorrect.
4413		 */
4414		if (unlikely(node->bytenr != buf->start && node->new_bytenr != buf->start)) {
4415			btrfs_err(fs_info,
4416"bytenr %llu was found but our backref cache was expecting %llu or %llu",
4417				  buf->start, node->bytenr, node->new_bytenr);
4418			return -EUCLEAN;
4419		}
4420
4421		btrfs_backref_drop_node_buffer(node);
4422		atomic_inc(&cow->refs);
4423		node->eb = cow;
4424		node->new_bytenr = cow->start;
4425
4426		if (!node->pending) {
4427			list_move_tail(&node->list,
4428				       &rc->backref_cache.pending[level]);
4429			node->pending = 1;
4430		}
4431
4432		if (first_cow)
4433			mark_block_processed(rc, node);
4434
4435		if (first_cow && level > 0)
4436			rc->nodes_relocated += buf->len;
4437	}
4438
4439	if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4440		ret = replace_file_extents(trans, rc, root, cow);
4441	return ret;
4442}
4443
4444/*
4445 * called before creating snapshot. it calculates metadata reservation
4446 * required for relocating tree blocks in the snapshot
4447 */
4448void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4449			      u64 *bytes_to_reserve)
4450{
4451	struct btrfs_root *root = pending->root;
4452	struct reloc_control *rc = root->fs_info->reloc_ctl;
4453
4454	if (!rc || !have_reloc_root(root))
 
4455		return;
4456
 
4457	if (!rc->merge_reloc_tree)
4458		return;
4459
4460	root = root->reloc_root;
4461	BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4462	/*
4463	 * relocation is in the stage of merging trees. the space
4464	 * used by merging a reloc tree is twice the size of
4465	 * relocated tree nodes in the worst case. half for cowing
4466	 * the reloc tree, half for cowing the fs tree. the space
4467	 * used by cowing the reloc tree will be freed after the
4468	 * tree is dropped. if we create snapshot, cowing the fs
4469	 * tree may use more space than it frees. so we need
4470	 * reserve extra space.
4471	 */
4472	*bytes_to_reserve += rc->nodes_relocated;
4473}
4474
4475/*
4476 * called after snapshot is created. migrate block reservation
4477 * and create reloc root for the newly created snapshot
4478 *
4479 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4480 * references held on the reloc_root, one for root->reloc_root and one for
4481 * rc->reloc_roots.
4482 */
4483int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4484			       struct btrfs_pending_snapshot *pending)
4485{
4486	struct btrfs_root *root = pending->root;
4487	struct btrfs_root *reloc_root;
4488	struct btrfs_root *new_root;
4489	struct reloc_control *rc = root->fs_info->reloc_ctl;
4490	int ret;
4491
4492	if (!rc || !have_reloc_root(root))
4493		return 0;
4494
4495	rc = root->fs_info->reloc_ctl;
4496	rc->merging_rsv_size += rc->nodes_relocated;
4497
4498	if (rc->merge_reloc_tree) {
4499		ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4500					      rc->block_rsv,
4501					      rc->nodes_relocated, true);
4502		if (ret)
4503			return ret;
4504	}
4505
4506	new_root = pending->snap;
4507	reloc_root = create_reloc_root(trans, root->reloc_root, btrfs_root_id(new_root));
 
4508	if (IS_ERR(reloc_root))
4509		return PTR_ERR(reloc_root);
4510
4511	ret = __add_reloc_root(reloc_root);
4512	ASSERT(ret != -EEXIST);
4513	if (ret) {
4514		/* Pairs with create_reloc_root */
4515		btrfs_put_root(reloc_root);
4516		return ret;
4517	}
4518	new_root->reloc_root = btrfs_grab_root(reloc_root);
4519
4520	if (rc->create_reloc_tree)
4521		ret = clone_backref_node(trans, rc, root, reloc_root);
4522	return ret;
4523}
4524
4525/*
4526 * Get the current bytenr for the block group which is being relocated.
4527 *
4528 * Return U64_MAX if no running relocation.
4529 */
4530u64 btrfs_get_reloc_bg_bytenr(const struct btrfs_fs_info *fs_info)
4531{
4532	u64 logical = U64_MAX;
4533
4534	lockdep_assert_held(&fs_info->reloc_mutex);
4535
4536	if (fs_info->reloc_ctl && fs_info->reloc_ctl->block_group)
4537		logical = fs_info->reloc_ctl->block_group->start;
4538	return logical;
4539}