Linux Audio

Check our new training course

Loading...
v6.2
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2009 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/sched.h>
   7#include <linux/pagemap.h>
   8#include <linux/writeback.h>
   9#include <linux/blkdev.h>
  10#include <linux/rbtree.h>
  11#include <linux/slab.h>
  12#include <linux/error-injection.h>
  13#include "ctree.h"
  14#include "disk-io.h"
  15#include "transaction.h"
  16#include "volumes.h"
  17#include "locking.h"
  18#include "btrfs_inode.h"
  19#include "async-thread.h"
  20#include "free-space-cache.h"
 
  21#include "qgroup.h"
  22#include "print-tree.h"
  23#include "delalloc-space.h"
  24#include "block-group.h"
  25#include "backref.h"
  26#include "misc.h"
  27#include "subpage.h"
  28#include "zoned.h"
  29#include "inode-item.h"
  30#include "space-info.h"
  31#include "fs.h"
  32#include "accessors.h"
  33#include "extent-tree.h"
  34#include "root-tree.h"
  35#include "file-item.h"
  36#include "relocation.h"
  37#include "super.h"
  38#include "tree-checker.h"
  39
  40/*
  41 * Relocation overview
  42 *
  43 * [What does relocation do]
  44 *
  45 * The objective of relocation is to relocate all extents of the target block
  46 * group to other block groups.
  47 * This is utilized by resize (shrink only), profile converting, compacting
  48 * space, or balance routine to spread chunks over devices.
  49 *
  50 * 		Before		|		After
  51 * ------------------------------------------------------------------
  52 *  BG A: 10 data extents	| BG A: deleted
  53 *  BG B:  2 data extents	| BG B: 10 data extents (2 old + 8 relocated)
  54 *  BG C:  1 extents		| BG C:  3 data extents (1 old + 2 relocated)
  55 *
  56 * [How does relocation work]
  57 *
  58 * 1.   Mark the target block group read-only
  59 *      New extents won't be allocated from the target block group.
  60 *
  61 * 2.1  Record each extent in the target block group
  62 *      To build a proper map of extents to be relocated.
  63 *
  64 * 2.2  Build data reloc tree and reloc trees
  65 *      Data reloc tree will contain an inode, recording all newly relocated
  66 *      data extents.
  67 *      There will be only one data reloc tree for one data block group.
  68 *
  69 *      Reloc tree will be a special snapshot of its source tree, containing
  70 *      relocated tree blocks.
  71 *      Each tree referring to a tree block in target block group will get its
  72 *      reloc tree built.
  73 *
  74 * 2.3  Swap source tree with its corresponding reloc tree
  75 *      Each involved tree only refers to new extents after swap.
  76 *
  77 * 3.   Cleanup reloc trees and data reloc tree.
  78 *      As old extents in the target block group are still referenced by reloc
  79 *      trees, we need to clean them up before really freeing the target block
  80 *      group.
  81 *
  82 * The main complexity is in steps 2.2 and 2.3.
  83 *
  84 * The entry point of relocation is relocate_block_group() function.
 
 
 
 
 
 
 
 
 
  85 */
 
 
 
 
  86
 
 
  87#define RELOCATION_RESERVED_NODES	256
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  88/*
  89 * map address of tree root to tree
  90 */
  91struct mapping_node {
  92	struct {
  93		struct rb_node rb_node;
  94		u64 bytenr;
  95	}; /* Use rb_simle_node for search/insert */
  96	void *data;
  97};
  98
  99struct mapping_tree {
 100	struct rb_root rb_root;
 101	spinlock_t lock;
 102};
 103
 104/*
 105 * present a tree block to process
 106 */
 107struct tree_block {
 108	struct {
 109		struct rb_node rb_node;
 110		u64 bytenr;
 111	}; /* Use rb_simple_node for search/insert */
 112	u64 owner;
 113	struct btrfs_key key;
 114	unsigned int level:8;
 115	unsigned int key_ready:1;
 116};
 117
 118#define MAX_EXTENTS 128
 119
 120struct file_extent_cluster {
 121	u64 start;
 122	u64 end;
 123	u64 boundary[MAX_EXTENTS];
 124	unsigned int nr;
 125};
 126
 127struct reloc_control {
 128	/* block group to relocate */
 129	struct btrfs_block_group *block_group;
 130	/* extent tree */
 131	struct btrfs_root *extent_root;
 132	/* inode for moving data */
 133	struct inode *data_inode;
 134
 135	struct btrfs_block_rsv *block_rsv;
 136
 137	struct btrfs_backref_cache backref_cache;
 138
 139	struct file_extent_cluster cluster;
 140	/* tree blocks have been processed */
 141	struct extent_io_tree processed_blocks;
 142	/* map start of tree root to corresponding reloc tree */
 143	struct mapping_tree reloc_root_tree;
 144	/* list of reloc trees */
 145	struct list_head reloc_roots;
 146	/* list of subvolume trees that get relocated */
 147	struct list_head dirty_subvol_roots;
 148	/* size of metadata reservation for merging reloc trees */
 149	u64 merging_rsv_size;
 150	/* size of relocated tree nodes */
 151	u64 nodes_relocated;
 152	/* reserved size for block group relocation*/
 153	u64 reserved_bytes;
 154
 155	u64 search_start;
 156	u64 extents_found;
 157
 158	unsigned int stage:8;
 159	unsigned int create_reloc_tree:1;
 160	unsigned int merge_reloc_tree:1;
 161	unsigned int found_file_extent:1;
 162};
 163
 164/* stages of data relocation */
 165#define MOVE_DATA_EXTENTS	0
 166#define UPDATE_DATA_PTRS	1
 167
 168static void mark_block_processed(struct reloc_control *rc,
 169				 struct btrfs_backref_node *node)
 
 
 
 
 170{
 171	u32 blocksize;
 
 
 172
 173	if (node->level == 0 ||
 174	    in_range(node->bytenr, rc->block_group->start,
 175		     rc->block_group->length)) {
 176		blocksize = rc->extent_root->fs_info->nodesize;
 177		set_extent_bits(&rc->processed_blocks, node->bytenr,
 178				node->bytenr + blocksize - 1, EXTENT_DIRTY);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 179	}
 180	node->processed = 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 181}
 182
 
 
 
 
 
 
 
 
 183
 184static void mapping_tree_init(struct mapping_tree *tree)
 185{
 186	tree->rb_root = RB_ROOT;
 187	spin_lock_init(&tree->lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 188}
 189
 190/*
 191 * walk up backref nodes until reach node presents tree root
 192 */
 193static struct btrfs_backref_node *walk_up_backref(
 194		struct btrfs_backref_node *node,
 195		struct btrfs_backref_edge *edges[], int *index)
 196{
 197	struct btrfs_backref_edge *edge;
 198	int idx = *index;
 199
 200	while (!list_empty(&node->upper)) {
 201		edge = list_entry(node->upper.next,
 202				  struct btrfs_backref_edge, list[LOWER]);
 203		edges[idx++] = edge;
 204		node = edge->node[UPPER];
 205	}
 206	BUG_ON(node->detached);
 207	*index = idx;
 208	return node;
 209}
 210
 211/*
 212 * walk down backref nodes to find start of next reference path
 213 */
 214static struct btrfs_backref_node *walk_down_backref(
 215		struct btrfs_backref_edge *edges[], int *index)
 216{
 217	struct btrfs_backref_edge *edge;
 218	struct btrfs_backref_node *lower;
 219	int idx = *index;
 220
 221	while (idx > 0) {
 222		edge = edges[idx - 1];
 223		lower = edge->node[LOWER];
 224		if (list_is_last(&edge->list[LOWER], &lower->upper)) {
 225			idx--;
 226			continue;
 227		}
 228		edge = list_entry(edge->list[LOWER].next,
 229				  struct btrfs_backref_edge, list[LOWER]);
 230		edges[idx - 1] = edge;
 231		*index = idx;
 232		return edge->node[UPPER];
 233	}
 234	*index = 0;
 235	return NULL;
 236}
 237
 238static void update_backref_node(struct btrfs_backref_cache *cache,
 239				struct btrfs_backref_node *node, u64 bytenr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 240{
 241	struct rb_node *rb_node;
 242	rb_erase(&node->rb_node, &cache->rb_root);
 243	node->bytenr = bytenr;
 244	rb_node = rb_simple_insert(&cache->rb_root, node->bytenr, &node->rb_node);
 245	if (rb_node)
 246		btrfs_backref_panic(cache->fs_info, bytenr, -EEXIST);
 247}
 248
 249/*
 250 * update backref cache after a transaction commit
 251 */
 252static int update_backref_cache(struct btrfs_trans_handle *trans,
 253				struct btrfs_backref_cache *cache)
 254{
 255	struct btrfs_backref_node *node;
 256	int level = 0;
 257
 258	if (cache->last_trans == 0) {
 259		cache->last_trans = trans->transid;
 260		return 0;
 261	}
 262
 263	if (cache->last_trans == trans->transid)
 264		return 0;
 265
 266	/*
 267	 * detached nodes are used to avoid unnecessary backref
 268	 * lookup. transaction commit changes the extent tree.
 269	 * so the detached nodes are no longer useful.
 270	 */
 271	while (!list_empty(&cache->detached)) {
 272		node = list_entry(cache->detached.next,
 273				  struct btrfs_backref_node, list);
 274		btrfs_backref_cleanup_node(cache, node);
 275	}
 276
 277	while (!list_empty(&cache->changed)) {
 278		node = list_entry(cache->changed.next,
 279				  struct btrfs_backref_node, list);
 280		list_del_init(&node->list);
 281		BUG_ON(node->pending);
 282		update_backref_node(cache, node, node->new_bytenr);
 283	}
 284
 285	/*
 286	 * some nodes can be left in the pending list if there were
 287	 * errors during processing the pending nodes.
 288	 */
 289	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
 290		list_for_each_entry(node, &cache->pending[level], list) {
 291			BUG_ON(!node->pending);
 292			if (node->bytenr == node->new_bytenr)
 293				continue;
 294			update_backref_node(cache, node, node->new_bytenr);
 295		}
 296	}
 297
 298	cache->last_trans = 0;
 299	return 1;
 300}
 301
 302static bool reloc_root_is_dead(struct btrfs_root *root)
 303{
 304	/*
 305	 * Pair with set_bit/clear_bit in clean_dirty_subvols and
 306	 * btrfs_update_reloc_root. We need to see the updated bit before
 307	 * trying to access reloc_root
 308	 */
 309	smp_rmb();
 310	if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state))
 311		return true;
 312	return false;
 313}
 314
 315/*
 316 * Check if this subvolume tree has valid reloc tree.
 317 *
 318 * Reloc tree after swap is considered dead, thus not considered as valid.
 319 * This is enough for most callers, as they don't distinguish dead reloc root
 320 * from no reloc root.  But btrfs_should_ignore_reloc_root() below is a
 321 * special case.
 322 */
 323static bool have_reloc_root(struct btrfs_root *root)
 324{
 325	if (reloc_root_is_dead(root))
 326		return false;
 327	if (!root->reloc_root)
 328		return false;
 329	return true;
 330}
 331
 332int btrfs_should_ignore_reloc_root(struct btrfs_root *root)
 333{
 334	struct btrfs_root *reloc_root;
 335
 336	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
 337		return 0;
 338
 339	/* This root has been merged with its reloc tree, we can ignore it */
 340	if (reloc_root_is_dead(root))
 341		return 1;
 342
 343	reloc_root = root->reloc_root;
 344	if (!reloc_root)
 345		return 0;
 346
 347	if (btrfs_header_generation(reloc_root->commit_root) ==
 348	    root->fs_info->running_transaction->transid)
 349		return 0;
 350	/*
 351	 * if there is reloc tree and it was created in previous
 352	 * transaction backref lookup can find the reloc tree,
 353	 * so backref node for the fs tree root is useless for
 354	 * relocation.
 355	 */
 356	return 1;
 357}
 358
 359/*
 360 * find reloc tree by address of tree root
 361 */
 362struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr)
 
 363{
 364	struct reloc_control *rc = fs_info->reloc_ctl;
 365	struct rb_node *rb_node;
 366	struct mapping_node *node;
 367	struct btrfs_root *root = NULL;
 368
 369	ASSERT(rc);
 370	spin_lock(&rc->reloc_root_tree.lock);
 371	rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root, bytenr);
 372	if (rb_node) {
 373		node = rb_entry(rb_node, struct mapping_node, rb_node);
 374		root = node->data;
 375	}
 376	spin_unlock(&rc->reloc_root_tree.lock);
 377	return btrfs_grab_root(root);
 378}
 379
 380/*
 381 * For useless nodes, do two major clean ups:
 382 *
 383 * - Cleanup the children edges and nodes
 384 *   If child node is also orphan (no parent) during cleanup, then the child
 385 *   node will also be cleaned up.
 386 *
 387 * - Freeing up leaves (level 0), keeps nodes detached
 388 *   For nodes, the node is still cached as "detached"
 389 *
 390 * Return false if @node is not in the @useless_nodes list.
 391 * Return true if @node is in the @useless_nodes list.
 392 */
 393static bool handle_useless_nodes(struct reloc_control *rc,
 394				 struct btrfs_backref_node *node)
 395{
 396	struct btrfs_backref_cache *cache = &rc->backref_cache;
 397	struct list_head *useless_node = &cache->useless_node;
 398	bool ret = false;
 
 
 
 
 
 
 
 
 
 399
 400	while (!list_empty(useless_node)) {
 401		struct btrfs_backref_node *cur;
 
 
 402
 403		cur = list_first_entry(useless_node, struct btrfs_backref_node,
 404				 list);
 405		list_del_init(&cur->list);
 
 
 
 406
 407		/* Only tree root nodes can be added to @useless_nodes */
 408		ASSERT(list_empty(&cur->upper));
 409
 410		if (cur == node)
 411			ret = true;
 
 
 
 
 
 
 
 412
 413		/* The node is the lowest node */
 414		if (cur->lowest) {
 415			list_del_init(&cur->lower);
 416			cur->lowest = 0;
 417		}
 418
 419		/* Cleanup the lower edges */
 420		while (!list_empty(&cur->lower)) {
 421			struct btrfs_backref_edge *edge;
 422			struct btrfs_backref_node *lower;
 423
 424			edge = list_entry(cur->lower.next,
 425					struct btrfs_backref_edge, list[UPPER]);
 426			list_del(&edge->list[UPPER]);
 427			list_del(&edge->list[LOWER]);
 428			lower = edge->node[LOWER];
 429			btrfs_backref_free_edge(cache, edge);
 430
 431			/* Child node is also orphan, queue for cleanup */
 432			if (list_empty(&lower->upper))
 433				list_add(&lower->list, useless_node);
 434		}
 435		/* Mark this block processed for relocation */
 436		mark_block_processed(rc, cur);
 437
 438		/*
 439		 * Backref nodes for tree leaves are deleted from the cache.
 440		 * Backref nodes for upper level tree blocks are left in the
 441		 * cache to avoid unnecessary backref lookup.
 442		 */
 443		if (cur->level > 0) {
 444			list_add(&cur->list, &cache->detached);
 445			cur->detached = 1;
 446		} else {
 447			rb_erase(&cur->rb_node, &cache->rb_root);
 448			btrfs_backref_free_node(cache, cur);
 449		}
 
 
 
 
 450	}
 451	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 452}
 453
 454/*
 455 * Build backref tree for a given tree block. Root of the backref tree
 456 * corresponds the tree block, leaves of the backref tree correspond roots of
 457 * b-trees that reference the tree block.
 458 *
 459 * The basic idea of this function is check backrefs of a given block to find
 460 * upper level blocks that reference the block, and then check backrefs of
 461 * these upper level blocks recursively. The recursion stops when tree root is
 462 * reached or backrefs for the block is cached.
 463 *
 464 * NOTE: if we find that backrefs for a block are cached, we know backrefs for
 465 * all upper level blocks that directly/indirectly reference the block are also
 466 * cached.
 467 */
 468static noinline_for_stack struct btrfs_backref_node *build_backref_tree(
 469			struct reloc_control *rc, struct btrfs_key *node_key,
 470			int level, u64 bytenr)
 471{
 472	struct btrfs_backref_iter *iter;
 473	struct btrfs_backref_cache *cache = &rc->backref_cache;
 474	/* For searching parent of TREE_BLOCK_REF */
 475	struct btrfs_path *path;
 476	struct btrfs_backref_node *cur;
 477	struct btrfs_backref_node *node = NULL;
 478	struct btrfs_backref_edge *edge;
 
 
 
 
 
 
 
 
 
 
 
 
 479	int ret;
 480	int err = 0;
 
 481
 482	iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info);
 483	if (!iter)
 484		return ERR_PTR(-ENOMEM);
 485	path = btrfs_alloc_path();
 486	if (!path) {
 487		err = -ENOMEM;
 488		goto out;
 489	}
 
 
 490
 491	node = btrfs_backref_alloc_node(cache, bytenr, level);
 492	if (!node) {
 493		err = -ENOMEM;
 494		goto out;
 495	}
 496
 
 
 497	node->lowest = 1;
 498	cur = node;
 
 
 
 
 
 
 499
 500	/* Breadth-first search to build backref cache */
 501	do {
 502		ret = btrfs_backref_add_tree_node(cache, path, iter, node_key,
 503						  cur);
 504		if (ret < 0) {
 505			err = ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 506			goto out;
 507		}
 508		edge = list_first_entry_or_null(&cache->pending_edge,
 509				struct btrfs_backref_edge, list[UPPER]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 510		/*
 511		 * The pending list isn't empty, take the first block to
 512		 * process
 513		 */
 514		if (edge) {
 515			list_del_init(&edge->list[UPPER]);
 516			cur = edge->node[UPPER];
 
 
 
 
 
 517		}
 518	} while (edge);
 
 519
 520	/* Finish the upper linkage of newly added edges/nodes */
 521	ret = btrfs_backref_finish_upper_links(cache, node);
 522	if (ret < 0) {
 523		err = ret;
 524		goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 525	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 526
 527	if (handle_useless_nodes(rc, node))
 528		node = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 529out:
 530	btrfs_backref_iter_free(iter);
 531	btrfs_free_path(path);
 532	if (err) {
 533		btrfs_backref_error_cleanup(cache, node);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 534		return ERR_PTR(err);
 535	}
 536	ASSERT(!node || !node->detached);
 537	ASSERT(list_empty(&cache->useless_node) &&
 538	       list_empty(&cache->pending_edge));
 539	return node;
 540}
 541
 542/*
 543 * helper to add backref node for the newly created snapshot.
 544 * the backref node is created by cloning backref node that
 545 * corresponds to root of source tree
 546 */
 547static int clone_backref_node(struct btrfs_trans_handle *trans,
 548			      struct reloc_control *rc,
 549			      struct btrfs_root *src,
 550			      struct btrfs_root *dest)
 551{
 552	struct btrfs_root *reloc_root = src->reloc_root;
 553	struct btrfs_backref_cache *cache = &rc->backref_cache;
 554	struct btrfs_backref_node *node = NULL;
 555	struct btrfs_backref_node *new_node;
 556	struct btrfs_backref_edge *edge;
 557	struct btrfs_backref_edge *new_edge;
 558	struct rb_node *rb_node;
 559
 560	if (cache->last_trans > 0)
 561		update_backref_cache(trans, cache);
 562
 563	rb_node = rb_simple_search(&cache->rb_root, src->commit_root->start);
 564	if (rb_node) {
 565		node = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
 566		if (node->detached)
 567			node = NULL;
 568		else
 569			BUG_ON(node->new_bytenr != reloc_root->node->start);
 570	}
 571
 572	if (!node) {
 573		rb_node = rb_simple_search(&cache->rb_root,
 574					   reloc_root->commit_root->start);
 575		if (rb_node) {
 576			node = rb_entry(rb_node, struct btrfs_backref_node,
 577					rb_node);
 578			BUG_ON(node->detached);
 579		}
 580	}
 581
 582	if (!node)
 583		return 0;
 584
 585	new_node = btrfs_backref_alloc_node(cache, dest->node->start,
 586					    node->level);
 587	if (!new_node)
 588		return -ENOMEM;
 589
 
 
 590	new_node->lowest = node->lowest;
 591	new_node->checked = 1;
 592	new_node->root = btrfs_grab_root(dest);
 593	ASSERT(new_node->root);
 594
 595	if (!node->lowest) {
 596		list_for_each_entry(edge, &node->lower, list[UPPER]) {
 597			new_edge = btrfs_backref_alloc_edge(cache);
 598			if (!new_edge)
 599				goto fail;
 600
 601			btrfs_backref_link_edge(new_edge, edge->node[LOWER],
 602						new_node, LINK_UPPER);
 
 
 603		}
 604	} else {
 605		list_add_tail(&new_node->lower, &cache->leaves);
 606	}
 607
 608	rb_node = rb_simple_insert(&cache->rb_root, new_node->bytenr,
 609				   &new_node->rb_node);
 610	if (rb_node)
 611		btrfs_backref_panic(trans->fs_info, new_node->bytenr, -EEXIST);
 612
 613	if (!new_node->lowest) {
 614		list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
 615			list_add_tail(&new_edge->list[LOWER],
 616				      &new_edge->node[LOWER]->upper);
 617		}
 618	}
 619	return 0;
 620fail:
 621	while (!list_empty(&new_node->lower)) {
 622		new_edge = list_entry(new_node->lower.next,
 623				      struct btrfs_backref_edge, list[UPPER]);
 624		list_del(&new_edge->list[UPPER]);
 625		btrfs_backref_free_edge(cache, new_edge);
 626	}
 627	btrfs_backref_free_node(cache, new_node);
 628	return -ENOMEM;
 629}
 630
 631/*
 632 * helper to add 'address of tree root -> reloc tree' mapping
 633 */
 634static int __must_check __add_reloc_root(struct btrfs_root *root)
 635{
 636	struct btrfs_fs_info *fs_info = root->fs_info;
 637	struct rb_node *rb_node;
 638	struct mapping_node *node;
 639	struct reloc_control *rc = fs_info->reloc_ctl;
 640
 641	node = kmalloc(sizeof(*node), GFP_NOFS);
 642	if (!node)
 643		return -ENOMEM;
 644
 645	node->bytenr = root->commit_root->start;
 646	node->data = root;
 647
 648	spin_lock(&rc->reloc_root_tree.lock);
 649	rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
 650				   node->bytenr, &node->rb_node);
 651	spin_unlock(&rc->reloc_root_tree.lock);
 652	if (rb_node) {
 653		btrfs_err(fs_info,
 654			    "Duplicate root found for start=%llu while inserting into relocation tree",
 655			    node->bytenr);
 656		return -EEXIST;
 657	}
 658
 659	list_add_tail(&root->root_list, &rc->reloc_roots);
 660	return 0;
 661}
 662
 663/*
 664 * helper to delete the 'address of tree root -> reloc tree'
 665 * mapping
 666 */
 667static void __del_reloc_root(struct btrfs_root *root)
 668{
 669	struct btrfs_fs_info *fs_info = root->fs_info;
 670	struct rb_node *rb_node;
 671	struct mapping_node *node = NULL;
 672	struct reloc_control *rc = fs_info->reloc_ctl;
 673	bool put_ref = false;
 674
 675	if (rc && root->node) {
 676		spin_lock(&rc->reloc_root_tree.lock);
 677		rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
 678					   root->commit_root->start);
 679		if (rb_node) {
 680			node = rb_entry(rb_node, struct mapping_node, rb_node);
 681			rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 682			RB_CLEAR_NODE(&node->rb_node);
 683		}
 684		spin_unlock(&rc->reloc_root_tree.lock);
 685		ASSERT(!node || (struct btrfs_root *)node->data == root);
 686	}
 
 
 
 
 
 687
 688	/*
 689	 * We only put the reloc root here if it's on the list.  There's a lot
 690	 * of places where the pattern is to splice the rc->reloc_roots, process
 691	 * the reloc roots, and then add the reloc root back onto
 692	 * rc->reloc_roots.  If we call __del_reloc_root while it's off of the
 693	 * list we don't want the reference being dropped, because the guy
 694	 * messing with the list is in charge of the reference.
 695	 */
 696	spin_lock(&fs_info->trans_lock);
 697	if (!list_empty(&root->root_list)) {
 698		put_ref = true;
 699		list_del_init(&root->root_list);
 700	}
 701	spin_unlock(&fs_info->trans_lock);
 702	if (put_ref)
 703		btrfs_put_root(root);
 704	kfree(node);
 705}
 706
 707/*
 708 * helper to update the 'address of tree root -> reloc tree'
 709 * mapping
 710 */
 711static int __update_reloc_root(struct btrfs_root *root)
 712{
 713	struct btrfs_fs_info *fs_info = root->fs_info;
 714	struct rb_node *rb_node;
 715	struct mapping_node *node = NULL;
 716	struct reloc_control *rc = fs_info->reloc_ctl;
 717
 718	spin_lock(&rc->reloc_root_tree.lock);
 719	rb_node = rb_simple_search(&rc->reloc_root_tree.rb_root,
 720				   root->commit_root->start);
 721	if (rb_node) {
 722		node = rb_entry(rb_node, struct mapping_node, rb_node);
 723		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 724	}
 725	spin_unlock(&rc->reloc_root_tree.lock);
 726
 727	if (!node)
 728		return 0;
 729	BUG_ON((struct btrfs_root *)node->data != root);
 730
 731	spin_lock(&rc->reloc_root_tree.lock);
 732	node->bytenr = root->node->start;
 733	rb_node = rb_simple_insert(&rc->reloc_root_tree.rb_root,
 734				   node->bytenr, &node->rb_node);
 735	spin_unlock(&rc->reloc_root_tree.lock);
 736	if (rb_node)
 737		btrfs_backref_panic(fs_info, node->bytenr, -EEXIST);
 738	return 0;
 739}
 740
 741static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
 742					struct btrfs_root *root, u64 objectid)
 743{
 744	struct btrfs_fs_info *fs_info = root->fs_info;
 745	struct btrfs_root *reloc_root;
 746	struct extent_buffer *eb;
 747	struct btrfs_root_item *root_item;
 748	struct btrfs_key root_key;
 749	int ret = 0;
 750	bool must_abort = false;
 751
 752	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
 753	if (!root_item)
 754		return ERR_PTR(-ENOMEM);
 755
 756	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
 757	root_key.type = BTRFS_ROOT_ITEM_KEY;
 758	root_key.offset = objectid;
 759
 760	if (root->root_key.objectid == objectid) {
 761		u64 commit_root_gen;
 762
 763		/* called by btrfs_init_reloc_root */
 764		ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
 765				      BTRFS_TREE_RELOC_OBJECTID);
 766		if (ret)
 767			goto fail;
 768
 769		/*
 770		 * Set the last_snapshot field to the generation of the commit
 771		 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
 772		 * correctly (returns true) when the relocation root is created
 773		 * either inside the critical section of a transaction commit
 774		 * (through transaction.c:qgroup_account_snapshot()) and when
 775		 * it's created before the transaction commit is started.
 776		 */
 777		commit_root_gen = btrfs_header_generation(root->commit_root);
 778		btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
 779	} else {
 780		/*
 781		 * called by btrfs_reloc_post_snapshot_hook.
 782		 * the source tree is a reloc tree, all tree blocks
 783		 * modified after it was created have RELOC flag
 784		 * set in their headers. so it's OK to not update
 785		 * the 'last_snapshot'.
 786		 */
 787		ret = btrfs_copy_root(trans, root, root->node, &eb,
 788				      BTRFS_TREE_RELOC_OBJECTID);
 789		if (ret)
 790			goto fail;
 791	}
 792
 793	/*
 794	 * We have changed references at this point, we must abort the
 795	 * transaction if anything fails.
 796	 */
 797	must_abort = true;
 798
 799	memcpy(root_item, &root->root_item, sizeof(*root_item));
 800	btrfs_set_root_bytenr(root_item, eb->start);
 801	btrfs_set_root_level(root_item, btrfs_header_level(eb));
 802	btrfs_set_root_generation(root_item, trans->transid);
 803
 804	if (root->root_key.objectid == objectid) {
 805		btrfs_set_root_refs(root_item, 0);
 806		memset(&root_item->drop_progress, 0,
 807		       sizeof(struct btrfs_disk_key));
 808		btrfs_set_root_drop_level(root_item, 0);
 809	}
 810
 811	btrfs_tree_unlock(eb);
 812	free_extent_buffer(eb);
 813
 814	ret = btrfs_insert_root(trans, fs_info->tree_root,
 815				&root_key, root_item);
 816	if (ret)
 817		goto fail;
 818
 819	kfree(root_item);
 820
 821	reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key);
 822	if (IS_ERR(reloc_root)) {
 823		ret = PTR_ERR(reloc_root);
 824		goto abort;
 825	}
 826	set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
 827	reloc_root->last_trans = trans->transid;
 828	return reloc_root;
 829fail:
 830	kfree(root_item);
 831abort:
 832	if (must_abort)
 833		btrfs_abort_transaction(trans, ret);
 834	return ERR_PTR(ret);
 835}
 836
 837/*
 838 * create reloc tree for a given fs tree. reloc tree is just a
 839 * snapshot of the fs tree with special root objectid.
 840 *
 841 * The reloc_root comes out of here with two references, one for
 842 * root->reloc_root, and another for being on the rc->reloc_roots list.
 843 */
 844int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
 845			  struct btrfs_root *root)
 846{
 847	struct btrfs_fs_info *fs_info = root->fs_info;
 848	struct btrfs_root *reloc_root;
 849	struct reloc_control *rc = fs_info->reloc_ctl;
 850	struct btrfs_block_rsv *rsv;
 851	int clear_rsv = 0;
 852	int ret;
 853
 854	if (!rc)
 855		return 0;
 856
 857	/*
 858	 * The subvolume has reloc tree but the swap is finished, no need to
 859	 * create/update the dead reloc tree
 860	 */
 861	if (reloc_root_is_dead(root))
 862		return 0;
 863
 864	/*
 865	 * This is subtle but important.  We do not do
 866	 * record_root_in_transaction for reloc roots, instead we record their
 867	 * corresponding fs root, and then here we update the last trans for the
 868	 * reloc root.  This means that we have to do this for the entire life
 869	 * of the reloc root, regardless of which stage of the relocation we are
 870	 * in.
 871	 */
 872	if (root->reloc_root) {
 873		reloc_root = root->reloc_root;
 874		reloc_root->last_trans = trans->transid;
 875		return 0;
 876	}
 877
 878	/*
 879	 * We are merging reloc roots, we do not need new reloc trees.  Also
 880	 * reloc trees never need their own reloc tree.
 881	 */
 882	if (!rc->create_reloc_tree ||
 883	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
 884		return 0;
 885
 886	if (!trans->reloc_reserved) {
 887		rsv = trans->block_rsv;
 888		trans->block_rsv = rc->block_rsv;
 889		clear_rsv = 1;
 890	}
 891	reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
 892	if (clear_rsv)
 893		trans->block_rsv = rsv;
 894	if (IS_ERR(reloc_root))
 895		return PTR_ERR(reloc_root);
 896
 897	ret = __add_reloc_root(reloc_root);
 898	ASSERT(ret != -EEXIST);
 899	if (ret) {
 900		/* Pairs with create_reloc_root */
 901		btrfs_put_root(reloc_root);
 902		return ret;
 903	}
 904	root->reloc_root = btrfs_grab_root(reloc_root);
 905	return 0;
 906}
 907
 908/*
 909 * update root item of reloc tree
 910 */
 911int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
 912			    struct btrfs_root *root)
 913{
 914	struct btrfs_fs_info *fs_info = root->fs_info;
 915	struct btrfs_root *reloc_root;
 916	struct btrfs_root_item *root_item;
 917	int ret;
 918
 919	if (!have_reloc_root(root))
 920		return 0;
 921
 922	reloc_root = root->reloc_root;
 923	root_item = &reloc_root->root_item;
 924
 925	/*
 926	 * We are probably ok here, but __del_reloc_root() will drop its ref of
 927	 * the root.  We have the ref for root->reloc_root, but just in case
 928	 * hold it while we update the reloc root.
 929	 */
 930	btrfs_grab_root(reloc_root);
 931
 932	/* root->reloc_root will stay until current relocation finished */
 933	if (fs_info->reloc_ctl->merge_reloc_tree &&
 934	    btrfs_root_refs(root_item) == 0) {
 935		set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
 936		/*
 937		 * Mark the tree as dead before we change reloc_root so
 938		 * have_reloc_root will not touch it from now on.
 939		 */
 940		smp_wmb();
 941		__del_reloc_root(reloc_root);
 942	}
 943
 944	if (reloc_root->commit_root != reloc_root->node) {
 945		__update_reloc_root(reloc_root);
 946		btrfs_set_root_node(root_item, reloc_root->node);
 947		free_extent_buffer(reloc_root->commit_root);
 948		reloc_root->commit_root = btrfs_root_node(reloc_root);
 949	}
 950
 951	ret = btrfs_update_root(trans, fs_info->tree_root,
 952				&reloc_root->root_key, root_item);
 953	btrfs_put_root(reloc_root);
 954	return ret;
 
 
 955}
 956
 957/*
 958 * helper to find first cached inode with inode number >= objectid
 959 * in a subvolume
 960 */
 961static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
 962{
 963	struct rb_node *node;
 964	struct rb_node *prev;
 965	struct btrfs_inode *entry;
 966	struct inode *inode;
 967
 968	spin_lock(&root->inode_lock);
 969again:
 970	node = root->inode_tree.rb_node;
 971	prev = NULL;
 972	while (node) {
 973		prev = node;
 974		entry = rb_entry(node, struct btrfs_inode, rb_node);
 975
 976		if (objectid < btrfs_ino(entry))
 977			node = node->rb_left;
 978		else if (objectid > btrfs_ino(entry))
 979			node = node->rb_right;
 980		else
 981			break;
 982	}
 983	if (!node) {
 984		while (prev) {
 985			entry = rb_entry(prev, struct btrfs_inode, rb_node);
 986			if (objectid <= btrfs_ino(entry)) {
 987				node = prev;
 988				break;
 989			}
 990			prev = rb_next(prev);
 991		}
 992	}
 993	while (node) {
 994		entry = rb_entry(node, struct btrfs_inode, rb_node);
 995		inode = igrab(&entry->vfs_inode);
 996		if (inode) {
 997			spin_unlock(&root->inode_lock);
 998			return inode;
 999		}
1000
1001		objectid = btrfs_ino(entry) + 1;
1002		if (cond_resched_lock(&root->inode_lock))
1003			goto again;
1004
1005		node = rb_next(node);
1006	}
1007	spin_unlock(&root->inode_lock);
1008	return NULL;
1009}
1010
 
 
 
 
 
 
 
 
 
1011/*
1012 * get new location of data
1013 */
1014static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1015			    u64 bytenr, u64 num_bytes)
1016{
1017	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1018	struct btrfs_path *path;
1019	struct btrfs_file_extent_item *fi;
1020	struct extent_buffer *leaf;
1021	int ret;
1022
1023	path = btrfs_alloc_path();
1024	if (!path)
1025		return -ENOMEM;
1026
1027	bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1028	ret = btrfs_lookup_file_extent(NULL, root, path,
1029			btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1030	if (ret < 0)
1031		goto out;
1032	if (ret > 0) {
1033		ret = -ENOENT;
1034		goto out;
1035	}
1036
1037	leaf = path->nodes[0];
1038	fi = btrfs_item_ptr(leaf, path->slots[0],
1039			    struct btrfs_file_extent_item);
1040
1041	BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1042	       btrfs_file_extent_compression(leaf, fi) ||
1043	       btrfs_file_extent_encryption(leaf, fi) ||
1044	       btrfs_file_extent_other_encoding(leaf, fi));
1045
1046	if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1047		ret = -EINVAL;
1048		goto out;
1049	}
1050
1051	*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1052	ret = 0;
1053out:
1054	btrfs_free_path(path);
1055	return ret;
1056}
1057
1058/*
1059 * update file extent items in the tree leaf to point to
1060 * the new locations.
1061 */
1062static noinline_for_stack
1063int replace_file_extents(struct btrfs_trans_handle *trans,
1064			 struct reloc_control *rc,
1065			 struct btrfs_root *root,
1066			 struct extent_buffer *leaf)
1067{
1068	struct btrfs_fs_info *fs_info = root->fs_info;
1069	struct btrfs_key key;
1070	struct btrfs_file_extent_item *fi;
1071	struct inode *inode = NULL;
1072	u64 parent;
1073	u64 bytenr;
1074	u64 new_bytenr = 0;
1075	u64 num_bytes;
1076	u64 end;
1077	u32 nritems;
1078	u32 i;
1079	int ret = 0;
1080	int first = 1;
1081	int dirty = 0;
1082
1083	if (rc->stage != UPDATE_DATA_PTRS)
1084		return 0;
1085
1086	/* reloc trees always use full backref */
1087	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1088		parent = leaf->start;
1089	else
1090		parent = 0;
1091
1092	nritems = btrfs_header_nritems(leaf);
1093	for (i = 0; i < nritems; i++) {
1094		struct btrfs_ref ref = { 0 };
1095
1096		cond_resched();
1097		btrfs_item_key_to_cpu(leaf, &key, i);
1098		if (key.type != BTRFS_EXTENT_DATA_KEY)
1099			continue;
1100		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1101		if (btrfs_file_extent_type(leaf, fi) ==
1102		    BTRFS_FILE_EXTENT_INLINE)
1103			continue;
1104		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1105		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1106		if (bytenr == 0)
1107			continue;
1108		if (!in_range(bytenr, rc->block_group->start,
1109			      rc->block_group->length))
1110			continue;
1111
1112		/*
1113		 * if we are modifying block in fs tree, wait for read_folio
1114		 * to complete and drop the extent cache
1115		 */
1116		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1117			if (first) {
1118				inode = find_next_inode(root, key.objectid);
1119				first = 0;
1120			} else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1121				btrfs_add_delayed_iput(BTRFS_I(inode));
1122				inode = find_next_inode(root, key.objectid);
1123			}
1124			if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
1125				struct extent_state *cached_state = NULL;
1126
1127				end = key.offset +
1128				      btrfs_file_extent_num_bytes(leaf, fi);
1129				WARN_ON(!IS_ALIGNED(key.offset,
1130						    fs_info->sectorsize));
1131				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1132				end--;
1133				ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1134						      key.offset, end,
1135						      &cached_state);
1136				if (!ret)
1137					continue;
1138
1139				btrfs_drop_extent_map_range(BTRFS_I(inode),
1140							    key.offset, end, true);
1141				unlock_extent(&BTRFS_I(inode)->io_tree,
1142					      key.offset, end, &cached_state);
1143			}
1144		}
1145
1146		ret = get_new_location(rc->data_inode, &new_bytenr,
1147				       bytenr, num_bytes);
1148		if (ret) {
1149			/*
1150			 * Don't have to abort since we've not changed anything
1151			 * in the file extent yet.
1152			 */
1153			break;
1154		}
1155
1156		btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1157		dirty = 1;
1158
1159		key.offset -= btrfs_file_extent_offset(leaf, fi);
1160		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1161				       num_bytes, parent);
1162		btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1163				    key.objectid, key.offset,
1164				    root->root_key.objectid, false);
1165		ret = btrfs_inc_extent_ref(trans, &ref);
1166		if (ret) {
1167			btrfs_abort_transaction(trans, ret);
1168			break;
1169		}
1170
1171		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr,
1172				       num_bytes, parent);
1173		btrfs_init_data_ref(&ref, btrfs_header_owner(leaf),
1174				    key.objectid, key.offset,
1175				    root->root_key.objectid, false);
1176		ret = btrfs_free_extent(trans, &ref);
1177		if (ret) {
1178			btrfs_abort_transaction(trans, ret);
1179			break;
1180		}
1181	}
1182	if (dirty)
1183		btrfs_mark_buffer_dirty(leaf);
1184	if (inode)
1185		btrfs_add_delayed_iput(BTRFS_I(inode));
1186	return ret;
1187}
1188
1189static noinline_for_stack
1190int memcmp_node_keys(struct extent_buffer *eb, int slot,
1191		     struct btrfs_path *path, int level)
1192{
1193	struct btrfs_disk_key key1;
1194	struct btrfs_disk_key key2;
1195	btrfs_node_key(eb, &key1, slot);
1196	btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1197	return memcmp(&key1, &key2, sizeof(key1));
1198}
1199
1200/*
1201 * try to replace tree blocks in fs tree with the new blocks
1202 * in reloc tree. tree blocks haven't been modified since the
1203 * reloc tree was create can be replaced.
1204 *
1205 * if a block was replaced, level of the block + 1 is returned.
1206 * if no block got replaced, 0 is returned. if there are other
1207 * errors, a negative error number is returned.
1208 */
1209static noinline_for_stack
1210int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc,
1211		 struct btrfs_root *dest, struct btrfs_root *src,
1212		 struct btrfs_path *path, struct btrfs_key *next_key,
1213		 int lowest_level, int max_level)
1214{
1215	struct btrfs_fs_info *fs_info = dest->fs_info;
1216	struct extent_buffer *eb;
1217	struct extent_buffer *parent;
1218	struct btrfs_ref ref = { 0 };
1219	struct btrfs_key key;
1220	u64 old_bytenr;
1221	u64 new_bytenr;
1222	u64 old_ptr_gen;
1223	u64 new_ptr_gen;
1224	u64 last_snapshot;
1225	u32 blocksize;
1226	int cow = 0;
1227	int level;
1228	int ret;
1229	int slot;
1230
1231	ASSERT(src->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1232	ASSERT(dest->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1233
1234	last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1235again:
1236	slot = path->slots[lowest_level];
1237	btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1238
1239	eb = btrfs_lock_root_node(dest);
 
1240	level = btrfs_header_level(eb);
1241
1242	if (level < lowest_level) {
1243		btrfs_tree_unlock(eb);
1244		free_extent_buffer(eb);
1245		return 0;
1246	}
1247
1248	if (cow) {
1249		ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb,
1250				      BTRFS_NESTING_COW);
1251		if (ret) {
1252			btrfs_tree_unlock(eb);
1253			free_extent_buffer(eb);
1254			return ret;
1255		}
1256	}
 
1257
1258	if (next_key) {
1259		next_key->objectid = (u64)-1;
1260		next_key->type = (u8)-1;
1261		next_key->offset = (u64)-1;
1262	}
1263
1264	parent = eb;
1265	while (1) {
 
 
1266		level = btrfs_header_level(parent);
1267		ASSERT(level >= lowest_level);
1268
1269		ret = btrfs_bin_search(parent, &key, &slot);
1270		if (ret < 0)
1271			break;
1272		if (ret && slot > 0)
1273			slot--;
1274
1275		if (next_key && slot + 1 < btrfs_header_nritems(parent))
1276			btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1277
1278		old_bytenr = btrfs_node_blockptr(parent, slot);
1279		blocksize = fs_info->nodesize;
1280		old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
 
1281
1282		if (level <= max_level) {
1283			eb = path->nodes[level];
1284			new_bytenr = btrfs_node_blockptr(eb,
1285							path->slots[level]);
1286			new_ptr_gen = btrfs_node_ptr_generation(eb,
1287							path->slots[level]);
1288		} else {
1289			new_bytenr = 0;
1290			new_ptr_gen = 0;
1291		}
1292
1293		if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1294			ret = level;
1295			break;
1296		}
1297
1298		if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1299		    memcmp_node_keys(parent, slot, path, level)) {
1300			if (level <= lowest_level) {
1301				ret = 0;
1302				break;
1303			}
1304
1305			eb = btrfs_read_node_slot(parent, slot);
 
1306			if (IS_ERR(eb)) {
1307				ret = PTR_ERR(eb);
1308				break;
 
 
 
 
1309			}
1310			btrfs_tree_lock(eb);
1311			if (cow) {
1312				ret = btrfs_cow_block(trans, dest, eb, parent,
1313						      slot, &eb,
1314						      BTRFS_NESTING_COW);
1315				if (ret) {
1316					btrfs_tree_unlock(eb);
1317					free_extent_buffer(eb);
1318					break;
1319				}
1320			}
 
1321
1322			btrfs_tree_unlock(parent);
1323			free_extent_buffer(parent);
1324
1325			parent = eb;
1326			continue;
1327		}
1328
1329		if (!cow) {
1330			btrfs_tree_unlock(parent);
1331			free_extent_buffer(parent);
1332			cow = 1;
1333			goto again;
1334		}
1335
1336		btrfs_node_key_to_cpu(path->nodes[level], &key,
1337				      path->slots[level]);
1338		btrfs_release_path(path);
1339
1340		path->lowest_level = level;
1341		set_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1342		ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
1343		clear_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &src->state);
1344		path->lowest_level = 0;
1345		if (ret) {
1346			if (ret > 0)
1347				ret = -ENOENT;
1348			break;
1349		}
1350
1351		/*
1352		 * Info qgroup to trace both subtrees.
1353		 *
1354		 * We must trace both trees.
1355		 * 1) Tree reloc subtree
1356		 *    If not traced, we will leak data numbers
1357		 * 2) Fs subtree
1358		 *    If not traced, we will double count old data
1359		 *
1360		 * We don't scan the subtree right now, but only record
1361		 * the swapped tree blocks.
1362		 * The real subtree rescan is delayed until we have new
1363		 * CoW on the subtree root node before transaction commit.
1364		 */
1365		ret = btrfs_qgroup_add_swapped_blocks(trans, dest,
1366				rc->block_group, parent, slot,
1367				path->nodes[level], path->slots[level],
1368				last_snapshot);
 
 
 
 
 
1369		if (ret < 0)
1370			break;
 
1371		/*
1372		 * swap blocks in fs tree and reloc tree.
1373		 */
1374		btrfs_set_node_blockptr(parent, slot, new_bytenr);
1375		btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1376		btrfs_mark_buffer_dirty(parent);
1377
1378		btrfs_set_node_blockptr(path->nodes[level],
1379					path->slots[level], old_bytenr);
1380		btrfs_set_node_ptr_generation(path->nodes[level],
1381					      path->slots[level], old_ptr_gen);
1382		btrfs_mark_buffer_dirty(path->nodes[level]);
1383
1384		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr,
1385				       blocksize, path->nodes[level]->start);
1386		btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid,
1387				    0, true);
1388		ret = btrfs_inc_extent_ref(trans, &ref);
1389		if (ret) {
1390			btrfs_abort_transaction(trans, ret);
1391			break;
1392		}
1393		btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr,
1394				       blocksize, 0);
1395		btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid, 0,
1396				    true);
1397		ret = btrfs_inc_extent_ref(trans, &ref);
1398		if (ret) {
1399			btrfs_abort_transaction(trans, ret);
1400			break;
1401		}
1402
1403		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr,
1404				       blocksize, path->nodes[level]->start);
1405		btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid,
1406				    0, true);
1407		ret = btrfs_free_extent(trans, &ref);
1408		if (ret) {
1409			btrfs_abort_transaction(trans, ret);
1410			break;
1411		}
1412
1413		btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr,
1414				       blocksize, 0);
1415		btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid,
1416				    0, true);
1417		ret = btrfs_free_extent(trans, &ref);
1418		if (ret) {
1419			btrfs_abort_transaction(trans, ret);
1420			break;
1421		}
1422
1423		btrfs_unlock_up_safe(path, 0);
1424
1425		ret = level;
1426		break;
1427	}
1428	btrfs_tree_unlock(parent);
1429	free_extent_buffer(parent);
1430	return ret;
1431}
1432
1433/*
1434 * helper to find next relocated block in reloc tree
1435 */
1436static noinline_for_stack
1437int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1438		       int *level)
1439{
1440	struct extent_buffer *eb;
1441	int i;
1442	u64 last_snapshot;
1443	u32 nritems;
1444
1445	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1446
1447	for (i = 0; i < *level; i++) {
1448		free_extent_buffer(path->nodes[i]);
1449		path->nodes[i] = NULL;
1450	}
1451
1452	for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1453		eb = path->nodes[i];
1454		nritems = btrfs_header_nritems(eb);
1455		while (path->slots[i] + 1 < nritems) {
1456			path->slots[i]++;
1457			if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
1458			    last_snapshot)
1459				continue;
1460
1461			*level = i;
1462			return 0;
1463		}
1464		free_extent_buffer(path->nodes[i]);
1465		path->nodes[i] = NULL;
1466	}
1467	return 1;
1468}
1469
1470/*
1471 * walk down reloc tree to find relocated block of lowest level
1472 */
1473static noinline_for_stack
1474int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1475			 int *level)
1476{
 
1477	struct extent_buffer *eb = NULL;
1478	int i;
 
1479	u64 ptr_gen = 0;
1480	u64 last_snapshot;
1481	u32 nritems;
1482
1483	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1484
1485	for (i = *level; i > 0; i--) {
 
 
1486		eb = path->nodes[i];
1487		nritems = btrfs_header_nritems(eb);
1488		while (path->slots[i] < nritems) {
1489			ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]);
1490			if (ptr_gen > last_snapshot)
1491				break;
1492			path->slots[i]++;
1493		}
1494		if (path->slots[i] >= nritems) {
1495			if (i == *level)
1496				break;
1497			*level = i + 1;
1498			return 0;
1499		}
1500		if (i == 1) {
1501			*level = i;
1502			return 0;
1503		}
1504
1505		eb = btrfs_read_node_slot(eb, path->slots[i]);
1506		if (IS_ERR(eb))
 
 
 
1507			return PTR_ERR(eb);
 
 
 
 
1508		BUG_ON(btrfs_header_level(eb) != i - 1);
1509		path->nodes[i - 1] = eb;
1510		path->slots[i - 1] = 0;
1511	}
1512	return 1;
1513}
1514
1515/*
1516 * invalidate extent cache for file extents whose key in range of
1517 * [min_key, max_key)
1518 */
1519static int invalidate_extent_cache(struct btrfs_root *root,
1520				   struct btrfs_key *min_key,
1521				   struct btrfs_key *max_key)
1522{
1523	struct btrfs_fs_info *fs_info = root->fs_info;
1524	struct inode *inode = NULL;
1525	u64 objectid;
1526	u64 start, end;
1527	u64 ino;
1528
1529	objectid = min_key->objectid;
1530	while (1) {
1531		struct extent_state *cached_state = NULL;
1532
1533		cond_resched();
1534		iput(inode);
1535
1536		if (objectid > max_key->objectid)
1537			break;
1538
1539		inode = find_next_inode(root, objectid);
1540		if (!inode)
1541			break;
1542		ino = btrfs_ino(BTRFS_I(inode));
1543
1544		if (ino > max_key->objectid) {
1545			iput(inode);
1546			break;
1547		}
1548
1549		objectid = ino + 1;
1550		if (!S_ISREG(inode->i_mode))
1551			continue;
1552
1553		if (unlikely(min_key->objectid == ino)) {
1554			if (min_key->type > BTRFS_EXTENT_DATA_KEY)
1555				continue;
1556			if (min_key->type < BTRFS_EXTENT_DATA_KEY)
1557				start = 0;
1558			else {
1559				start = min_key->offset;
1560				WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
1561			}
1562		} else {
1563			start = 0;
1564		}
1565
1566		if (unlikely(max_key->objectid == ino)) {
1567			if (max_key->type < BTRFS_EXTENT_DATA_KEY)
1568				continue;
1569			if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
1570				end = (u64)-1;
1571			} else {
1572				if (max_key->offset == 0)
1573					continue;
1574				end = max_key->offset;
1575				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1576				end--;
1577			}
1578		} else {
1579			end = (u64)-1;
1580		}
1581
1582		/* the lock_extent waits for read_folio to complete */
1583		lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
1584		btrfs_drop_extent_map_range(BTRFS_I(inode), start, end, true);
1585		unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
1586	}
1587	return 0;
1588}
1589
1590static int find_next_key(struct btrfs_path *path, int level,
1591			 struct btrfs_key *key)
1592
1593{
1594	while (level < BTRFS_MAX_LEVEL) {
1595		if (!path->nodes[level])
1596			break;
1597		if (path->slots[level] + 1 <
1598		    btrfs_header_nritems(path->nodes[level])) {
1599			btrfs_node_key_to_cpu(path->nodes[level], key,
1600					      path->slots[level] + 1);
1601			return 0;
1602		}
1603		level++;
1604	}
1605	return 1;
1606}
1607
1608/*
1609 * Insert current subvolume into reloc_control::dirty_subvol_roots
1610 */
1611static int insert_dirty_subvol(struct btrfs_trans_handle *trans,
1612			       struct reloc_control *rc,
1613			       struct btrfs_root *root)
1614{
1615	struct btrfs_root *reloc_root = root->reloc_root;
1616	struct btrfs_root_item *reloc_root_item;
1617	int ret;
1618
1619	/* @root must be a subvolume tree root with a valid reloc tree */
1620	ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1621	ASSERT(reloc_root);
1622
1623	reloc_root_item = &reloc_root->root_item;
1624	memset(&reloc_root_item->drop_progress, 0,
1625		sizeof(reloc_root_item->drop_progress));
1626	btrfs_set_root_drop_level(reloc_root_item, 0);
1627	btrfs_set_root_refs(reloc_root_item, 0);
1628	ret = btrfs_update_reloc_root(trans, root);
1629	if (ret)
1630		return ret;
1631
1632	if (list_empty(&root->reloc_dirty_list)) {
1633		btrfs_grab_root(root);
1634		list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots);
1635	}
1636
1637	return 0;
1638}
1639
1640static int clean_dirty_subvols(struct reloc_control *rc)
1641{
1642	struct btrfs_root *root;
1643	struct btrfs_root *next;
1644	int ret = 0;
1645	int ret2;
1646
1647	list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots,
1648				 reloc_dirty_list) {
1649		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1650			/* Merged subvolume, cleanup its reloc root */
1651			struct btrfs_root *reloc_root = root->reloc_root;
1652
1653			list_del_init(&root->reloc_dirty_list);
1654			root->reloc_root = NULL;
1655			/*
1656			 * Need barrier to ensure clear_bit() only happens after
1657			 * root->reloc_root = NULL. Pairs with have_reloc_root.
1658			 */
1659			smp_wmb();
1660			clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
1661			if (reloc_root) {
1662				/*
1663				 * btrfs_drop_snapshot drops our ref we hold for
1664				 * ->reloc_root.  If it fails however we must
1665				 * drop the ref ourselves.
1666				 */
1667				ret2 = btrfs_drop_snapshot(reloc_root, 0, 1);
1668				if (ret2 < 0) {
1669					btrfs_put_root(reloc_root);
1670					if (!ret)
1671						ret = ret2;
1672				}
1673			}
1674			btrfs_put_root(root);
1675		} else {
1676			/* Orphan reloc tree, just clean it up */
1677			ret2 = btrfs_drop_snapshot(root, 0, 1);
1678			if (ret2 < 0) {
1679				btrfs_put_root(root);
1680				if (!ret)
1681					ret = ret2;
1682			}
1683		}
1684	}
1685	return ret;
1686}
1687
1688/*
1689 * merge the relocated tree blocks in reloc tree with corresponding
1690 * fs tree.
1691 */
1692static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
1693					       struct btrfs_root *root)
1694{
1695	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
 
1696	struct btrfs_key key;
1697	struct btrfs_key next_key;
1698	struct btrfs_trans_handle *trans = NULL;
1699	struct btrfs_root *reloc_root;
1700	struct btrfs_root_item *root_item;
1701	struct btrfs_path *path;
1702	struct extent_buffer *leaf;
1703	int reserve_level;
1704	int level;
1705	int max_level;
1706	int replaced = 0;
1707	int ret = 0;
 
1708	u32 min_reserved;
1709
1710	path = btrfs_alloc_path();
1711	if (!path)
1712		return -ENOMEM;
1713	path->reada = READA_FORWARD;
1714
1715	reloc_root = root->reloc_root;
1716	root_item = &reloc_root->root_item;
1717
1718	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
1719		level = btrfs_root_level(root_item);
1720		atomic_inc(&reloc_root->node->refs);
1721		path->nodes[level] = reloc_root->node;
1722		path->slots[level] = 0;
1723	} else {
1724		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
1725
1726		level = btrfs_root_drop_level(root_item);
1727		BUG_ON(level == 0);
1728		path->lowest_level = level;
1729		ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
1730		path->lowest_level = 0;
1731		if (ret < 0) {
1732			btrfs_free_path(path);
1733			return ret;
1734		}
1735
1736		btrfs_node_key_to_cpu(path->nodes[level], &next_key,
1737				      path->slots[level]);
1738		WARN_ON(memcmp(&key, &next_key, sizeof(key)));
1739
1740		btrfs_unlock_up_safe(path, 0);
1741	}
1742
1743	/*
1744	 * In merge_reloc_root(), we modify the upper level pointer to swap the
1745	 * tree blocks between reloc tree and subvolume tree.  Thus for tree
1746	 * block COW, we COW at most from level 1 to root level for each tree.
1747	 *
1748	 * Thus the needed metadata size is at most root_level * nodesize,
1749	 * and * 2 since we have two trees to COW.
1750	 */
1751	reserve_level = max_t(int, 1, btrfs_root_level(root_item));
1752	min_reserved = fs_info->nodesize * reserve_level * 2;
1753	memset(&next_key, 0, sizeof(next_key));
1754
1755	while (1) {
1756		ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
1757					     min_reserved,
1758					     BTRFS_RESERVE_FLUSH_LIMIT);
1759		if (ret)
1760			goto out;
 
1761		trans = btrfs_start_transaction(root, 0);
1762		if (IS_ERR(trans)) {
1763			ret = PTR_ERR(trans);
1764			trans = NULL;
1765			goto out;
1766		}
1767
1768		/*
1769		 * At this point we no longer have a reloc_control, so we can't
1770		 * depend on btrfs_init_reloc_root to update our last_trans.
1771		 *
1772		 * But that's ok, we started the trans handle on our
1773		 * corresponding fs_root, which means it's been added to the
1774		 * dirty list.  At commit time we'll still call
1775		 * btrfs_update_reloc_root() and update our root item
1776		 * appropriately.
1777		 */
1778		reloc_root->last_trans = trans->transid;
1779		trans->block_rsv = rc->block_rsv;
1780
1781		replaced = 0;
1782		max_level = level;
1783
1784		ret = walk_down_reloc_tree(reloc_root, path, &level);
1785		if (ret < 0)
 
1786			goto out;
 
1787		if (ret > 0)
1788			break;
1789
1790		if (!find_next_key(path, level, &key) &&
1791		    btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
1792			ret = 0;
1793		} else {
1794			ret = replace_path(trans, rc, root, reloc_root, path,
1795					   &next_key, level, max_level);
1796		}
1797		if (ret < 0)
 
1798			goto out;
 
 
1799		if (ret > 0) {
1800			level = ret;
1801			btrfs_node_key_to_cpu(path->nodes[level], &key,
1802					      path->slots[level]);
1803			replaced = 1;
1804		}
1805
1806		ret = walk_up_reloc_tree(reloc_root, path, &level);
1807		if (ret > 0)
1808			break;
1809
1810		BUG_ON(level == 0);
1811		/*
1812		 * save the merging progress in the drop_progress.
1813		 * this is OK since root refs == 1 in this case.
1814		 */
1815		btrfs_node_key(path->nodes[level], &root_item->drop_progress,
1816			       path->slots[level]);
1817		btrfs_set_root_drop_level(root_item, level);
1818
1819		btrfs_end_transaction_throttle(trans);
1820		trans = NULL;
1821
1822		btrfs_btree_balance_dirty(fs_info);
1823
1824		if (replaced && rc->stage == UPDATE_DATA_PTRS)
1825			invalidate_extent_cache(root, &key, &next_key);
1826	}
1827
1828	/*
1829	 * handle the case only one block in the fs tree need to be
1830	 * relocated and the block is tree root.
1831	 */
1832	leaf = btrfs_lock_root_node(root);
1833	ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf,
1834			      BTRFS_NESTING_COW);
1835	btrfs_tree_unlock(leaf);
1836	free_extent_buffer(leaf);
 
 
1837out:
1838	btrfs_free_path(path);
1839
1840	if (ret == 0) {
1841		ret = insert_dirty_subvol(trans, rc, root);
1842		if (ret)
1843			btrfs_abort_transaction(trans, ret);
 
 
1844	}
1845
1846	if (trans)
1847		btrfs_end_transaction_throttle(trans);
1848
1849	btrfs_btree_balance_dirty(fs_info);
1850
1851	if (replaced && rc->stage == UPDATE_DATA_PTRS)
1852		invalidate_extent_cache(root, &key, &next_key);
1853
1854	return ret;
1855}
1856
1857static noinline_for_stack
1858int prepare_to_merge(struct reloc_control *rc, int err)
1859{
1860	struct btrfs_root *root = rc->extent_root;
1861	struct btrfs_fs_info *fs_info = root->fs_info;
1862	struct btrfs_root *reloc_root;
1863	struct btrfs_trans_handle *trans;
1864	LIST_HEAD(reloc_roots);
1865	u64 num_bytes = 0;
1866	int ret;
1867
1868	mutex_lock(&fs_info->reloc_mutex);
1869	rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
1870	rc->merging_rsv_size += rc->nodes_relocated * 2;
1871	mutex_unlock(&fs_info->reloc_mutex);
1872
1873again:
1874	if (!err) {
1875		num_bytes = rc->merging_rsv_size;
1876		ret = btrfs_block_rsv_add(fs_info, rc->block_rsv, num_bytes,
1877					  BTRFS_RESERVE_FLUSH_ALL);
1878		if (ret)
1879			err = ret;
1880	}
1881
1882	trans = btrfs_join_transaction(rc->extent_root);
1883	if (IS_ERR(trans)) {
1884		if (!err)
1885			btrfs_block_rsv_release(fs_info, rc->block_rsv,
1886						num_bytes, NULL);
1887		return PTR_ERR(trans);
1888	}
1889
1890	if (!err) {
1891		if (num_bytes != rc->merging_rsv_size) {
1892			btrfs_end_transaction(trans);
1893			btrfs_block_rsv_release(fs_info, rc->block_rsv,
1894						num_bytes, NULL);
1895			goto again;
1896		}
1897	}
1898
1899	rc->merge_reloc_tree = 1;
1900
1901	while (!list_empty(&rc->reloc_roots)) {
1902		reloc_root = list_entry(rc->reloc_roots.next,
1903					struct btrfs_root, root_list);
1904		list_del_init(&reloc_root->root_list);
1905
1906		root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1907				false);
1908		if (IS_ERR(root)) {
1909			/*
1910			 * Even if we have an error we need this reloc root
1911			 * back on our list so we can clean up properly.
1912			 */
1913			list_add(&reloc_root->root_list, &reloc_roots);
1914			btrfs_abort_transaction(trans, (int)PTR_ERR(root));
1915			if (!err)
1916				err = PTR_ERR(root);
1917			break;
1918		}
1919		ASSERT(root->reloc_root == reloc_root);
1920
1921		/*
1922		 * set reference count to 1, so btrfs_recover_relocation
1923		 * knows it should resumes merging
1924		 */
1925		if (!err)
1926			btrfs_set_root_refs(&reloc_root->root_item, 1);
1927		ret = btrfs_update_reloc_root(trans, root);
1928
1929		/*
1930		 * Even if we have an error we need this reloc root back on our
1931		 * list so we can clean up properly.
1932		 */
1933		list_add(&reloc_root->root_list, &reloc_roots);
1934		btrfs_put_root(root);
1935
1936		if (ret) {
1937			btrfs_abort_transaction(trans, ret);
1938			if (!err)
1939				err = ret;
1940			break;
1941		}
1942	}
1943
1944	list_splice(&reloc_roots, &rc->reloc_roots);
1945
1946	if (!err)
1947		err = btrfs_commit_transaction(trans);
1948	else
1949		btrfs_end_transaction(trans);
1950	return err;
1951}
1952
1953static noinline_for_stack
1954void free_reloc_roots(struct list_head *list)
1955{
1956	struct btrfs_root *reloc_root, *tmp;
1957
1958	list_for_each_entry_safe(reloc_root, tmp, list, root_list)
 
 
1959		__del_reloc_root(reloc_root);
 
 
 
 
 
1960}
1961
1962static noinline_for_stack
1963void merge_reloc_roots(struct reloc_control *rc)
1964{
1965	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
1966	struct btrfs_root *root;
1967	struct btrfs_root *reloc_root;
1968	LIST_HEAD(reloc_roots);
1969	int found = 0;
1970	int ret = 0;
1971again:
1972	root = rc->extent_root;
1973
1974	/*
1975	 * this serializes us with btrfs_record_root_in_transaction,
1976	 * we have to make sure nobody is in the middle of
1977	 * adding their roots to the list while we are
1978	 * doing this splice
1979	 */
1980	mutex_lock(&fs_info->reloc_mutex);
1981	list_splice_init(&rc->reloc_roots, &reloc_roots);
1982	mutex_unlock(&fs_info->reloc_mutex);
1983
1984	while (!list_empty(&reloc_roots)) {
1985		found = 1;
1986		reloc_root = list_entry(reloc_roots.next,
1987					struct btrfs_root, root_list);
1988
1989		root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
1990					 false);
1991		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
1992			if (IS_ERR(root)) {
1993				/*
1994				 * For recovery we read the fs roots on mount,
1995				 * and if we didn't find the root then we marked
1996				 * the reloc root as a garbage root.  For normal
1997				 * relocation obviously the root should exist in
1998				 * memory.  However there's no reason we can't
1999				 * handle the error properly here just in case.
2000				 */
2001				ASSERT(0);
2002				ret = PTR_ERR(root);
2003				goto out;
2004			}
2005			if (root->reloc_root != reloc_root) {
2006				/*
2007				 * This is actually impossible without something
2008				 * going really wrong (like weird race condition
2009				 * or cosmic rays).
2010				 */
2011				ASSERT(0);
2012				ret = -EINVAL;
2013				goto out;
2014			}
2015			ret = merge_reloc_root(rc, root);
2016			btrfs_put_root(root);
2017			if (ret) {
2018				if (list_empty(&reloc_root->root_list))
2019					list_add_tail(&reloc_root->root_list,
2020						      &reloc_roots);
2021				goto out;
2022			}
2023		} else {
2024			if (!IS_ERR(root)) {
2025				if (root->reloc_root == reloc_root) {
2026					root->reloc_root = NULL;
2027					btrfs_put_root(reloc_root);
2028				}
2029				clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE,
2030					  &root->state);
2031				btrfs_put_root(root);
2032			}
2033
2034			list_del_init(&reloc_root->root_list);
2035			/* Don't forget to queue this reloc root for cleanup */
2036			list_add_tail(&reloc_root->reloc_dirty_list,
2037				      &rc->dirty_subvol_roots);
 
 
 
 
 
2038		}
2039	}
2040
2041	if (found) {
2042		found = 0;
2043		goto again;
2044	}
2045out:
2046	if (ret) {
2047		btrfs_handle_fs_error(fs_info, ret, NULL);
2048		free_reloc_roots(&reloc_roots);
 
2049
2050		/* new reloc root may be added */
2051		mutex_lock(&fs_info->reloc_mutex);
2052		list_splice_init(&rc->reloc_roots, &reloc_roots);
2053		mutex_unlock(&fs_info->reloc_mutex);
2054		free_reloc_roots(&reloc_roots);
 
2055	}
2056
2057	/*
2058	 * We used to have
2059	 *
2060	 * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
2061	 *
2062	 * here, but it's wrong.  If we fail to start the transaction in
2063	 * prepare_to_merge() we will have only 0 ref reloc roots, none of which
2064	 * have actually been removed from the reloc_root_tree rb tree.  This is
2065	 * fine because we're bailing here, and we hold a reference on the root
2066	 * for the list that holds it, so these roots will be cleaned up when we
2067	 * do the reloc_dirty_list afterwards.  Meanwhile the root->reloc_root
2068	 * will be cleaned up on unmount.
2069	 *
2070	 * The remaining nodes will be cleaned up by free_reloc_control.
2071	 */
2072}
2073
2074static void free_block_list(struct rb_root *blocks)
2075{
2076	struct tree_block *block;
2077	struct rb_node *rb_node;
2078	while ((rb_node = rb_first(blocks))) {
2079		block = rb_entry(rb_node, struct tree_block, rb_node);
2080		rb_erase(rb_node, blocks);
2081		kfree(block);
2082	}
2083}
2084
2085static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2086				      struct btrfs_root *reloc_root)
2087{
2088	struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2089	struct btrfs_root *root;
2090	int ret;
2091
2092	if (reloc_root->last_trans == trans->transid)
2093		return 0;
2094
2095	root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset, false);
 
 
2096
2097	/*
2098	 * This should succeed, since we can't have a reloc root without having
2099	 * already looked up the actual root and created the reloc root for this
2100	 * root.
2101	 *
2102	 * However if there's some sort of corruption where we have a ref to a
2103	 * reloc root without a corresponding root this could return ENOENT.
2104	 */
2105	if (IS_ERR(root)) {
2106		ASSERT(0);
2107		return PTR_ERR(root);
2108	}
2109	if (root->reloc_root != reloc_root) {
2110		ASSERT(0);
2111		btrfs_err(fs_info,
2112			  "root %llu has two reloc roots associated with it",
2113			  reloc_root->root_key.offset);
2114		btrfs_put_root(root);
2115		return -EUCLEAN;
2116	}
2117	ret = btrfs_record_root_in_trans(trans, root);
2118	btrfs_put_root(root);
2119
2120	return ret;
2121}
2122
2123static noinline_for_stack
2124struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2125				     struct reloc_control *rc,
2126				     struct btrfs_backref_node *node,
2127				     struct btrfs_backref_edge *edges[])
2128{
2129	struct btrfs_backref_node *next;
2130	struct btrfs_root *root;
2131	int index = 0;
2132	int ret;
2133
2134	next = node;
2135	while (1) {
2136		cond_resched();
2137		next = walk_up_backref(next, edges, &index);
2138		root = next->root;
2139
2140		/*
2141		 * If there is no root, then our references for this block are
2142		 * incomplete, as we should be able to walk all the way up to a
2143		 * block that is owned by a root.
2144		 *
2145		 * This path is only for SHAREABLE roots, so if we come upon a
2146		 * non-SHAREABLE root then we have backrefs that resolve
2147		 * improperly.
2148		 *
2149		 * Both of these cases indicate file system corruption, or a bug
2150		 * in the backref walking code.
2151		 */
2152		if (!root) {
2153			ASSERT(0);
2154			btrfs_err(trans->fs_info,
2155		"bytenr %llu doesn't have a backref path ending in a root",
2156				  node->bytenr);
2157			return ERR_PTR(-EUCLEAN);
2158		}
2159		if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2160			ASSERT(0);
2161			btrfs_err(trans->fs_info,
2162	"bytenr %llu has multiple refs with one ending in a non-shareable root",
2163				  node->bytenr);
2164			return ERR_PTR(-EUCLEAN);
2165		}
2166
2167		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2168			ret = record_reloc_root_in_trans(trans, root);
2169			if (ret)
2170				return ERR_PTR(ret);
2171			break;
2172		}
2173
2174		ret = btrfs_record_root_in_trans(trans, root);
2175		if (ret)
2176			return ERR_PTR(ret);
2177		root = root->reloc_root;
2178
2179		/*
2180		 * We could have raced with another thread which failed, so
2181		 * root->reloc_root may not be set, return ENOENT in this case.
2182		 */
2183		if (!root)
2184			return ERR_PTR(-ENOENT);
2185
2186		if (next->new_bytenr != root->node->start) {
2187			/*
2188			 * We just created the reloc root, so we shouldn't have
2189			 * ->new_bytenr set and this shouldn't be in the changed
2190			 *  list.  If it is then we have multiple roots pointing
2191			 *  at the same bytenr which indicates corruption, or
2192			 *  we've made a mistake in the backref walking code.
2193			 */
2194			ASSERT(next->new_bytenr == 0);
2195			ASSERT(list_empty(&next->list));
2196			if (next->new_bytenr || !list_empty(&next->list)) {
2197				btrfs_err(trans->fs_info,
2198	"bytenr %llu possibly has multiple roots pointing at the same bytenr %llu",
2199					  node->bytenr, next->bytenr);
2200				return ERR_PTR(-EUCLEAN);
2201			}
2202
2203			next->new_bytenr = root->node->start;
2204			btrfs_put_root(next->root);
2205			next->root = btrfs_grab_root(root);
2206			ASSERT(next->root);
2207			list_add_tail(&next->list,
2208				      &rc->backref_cache.changed);
2209			mark_block_processed(rc, next);
2210			break;
2211		}
2212
2213		WARN_ON(1);
2214		root = NULL;
2215		next = walk_down_backref(edges, &index);
2216		if (!next || next->level <= node->level)
2217			break;
2218	}
2219	if (!root) {
2220		/*
2221		 * This can happen if there's fs corruption or if there's a bug
2222		 * in the backref lookup code.
2223		 */
2224		ASSERT(0);
2225		return ERR_PTR(-ENOENT);
2226	}
2227
2228	next = node;
2229	/* setup backref node path for btrfs_reloc_cow_block */
2230	while (1) {
2231		rc->backref_cache.path[next->level] = next;
2232		if (--index < 0)
2233			break;
2234		next = edges[index]->node[UPPER];
2235	}
2236	return root;
2237}
2238
2239/*
2240 * Select a tree root for relocation.
2241 *
2242 * Return NULL if the block is not shareable. We should use do_relocation() in
2243 * this case.
2244 *
2245 * Return a tree root pointer if the block is shareable.
2246 * Return -ENOENT if the block is root of reloc tree.
2247 */
2248static noinline_for_stack
2249struct btrfs_root *select_one_root(struct btrfs_backref_node *node)
2250{
2251	struct btrfs_backref_node *next;
2252	struct btrfs_root *root;
2253	struct btrfs_root *fs_root = NULL;
2254	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2255	int index = 0;
2256
2257	next = node;
2258	while (1) {
2259		cond_resched();
2260		next = walk_up_backref(next, edges, &index);
2261		root = next->root;
 
2262
2263		/*
2264		 * This can occur if we have incomplete extent refs leading all
2265		 * the way up a particular path, in this case return -EUCLEAN.
2266		 */
2267		if (!root)
2268			return ERR_PTR(-EUCLEAN);
2269
2270		/* No other choice for non-shareable tree */
2271		if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2272			return root;
2273
2274		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2275			fs_root = root;
2276
2277		if (next != node)
2278			return NULL;
2279
2280		next = walk_down_backref(edges, &index);
2281		if (!next || next->level <= node->level)
2282			break;
2283	}
2284
2285	if (!fs_root)
2286		return ERR_PTR(-ENOENT);
2287	return fs_root;
2288}
2289
2290static noinline_for_stack
2291u64 calcu_metadata_size(struct reloc_control *rc,
2292			struct btrfs_backref_node *node, int reserve)
2293{
2294	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2295	struct btrfs_backref_node *next = node;
2296	struct btrfs_backref_edge *edge;
2297	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2298	u64 num_bytes = 0;
2299	int index = 0;
2300
2301	BUG_ON(reserve && node->processed);
2302
2303	while (next) {
2304		cond_resched();
2305		while (1) {
2306			if (next->processed && (reserve || next != node))
2307				break;
2308
2309			num_bytes += fs_info->nodesize;
2310
2311			if (list_empty(&next->upper))
2312				break;
2313
2314			edge = list_entry(next->upper.next,
2315					struct btrfs_backref_edge, list[LOWER]);
2316			edges[index++] = edge;
2317			next = edge->node[UPPER];
2318		}
2319		next = walk_down_backref(edges, &index);
2320	}
2321	return num_bytes;
2322}
2323
2324static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2325				  struct reloc_control *rc,
2326				  struct btrfs_backref_node *node)
2327{
2328	struct btrfs_root *root = rc->extent_root;
2329	struct btrfs_fs_info *fs_info = root->fs_info;
2330	u64 num_bytes;
2331	int ret;
2332	u64 tmp;
2333
2334	num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2335
2336	trans->block_rsv = rc->block_rsv;
2337	rc->reserved_bytes += num_bytes;
2338
2339	/*
2340	 * We are under a transaction here so we can only do limited flushing.
2341	 * If we get an enospc just kick back -EAGAIN so we know to drop the
2342	 * transaction and try to refill when we can flush all the things.
2343	 */
2344	ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv, num_bytes,
2345				     BTRFS_RESERVE_FLUSH_LIMIT);
2346	if (ret) {
2347		tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2348		while (tmp <= rc->reserved_bytes)
2349			tmp <<= 1;
2350		/*
2351		 * only one thread can access block_rsv at this point,
2352		 * so we don't need hold lock to protect block_rsv.
2353		 * we expand more reservation size here to allow enough
2354		 * space for relocation and we will return earlier in
2355		 * enospc case.
2356		 */
2357		rc->block_rsv->size = tmp + fs_info->nodesize *
2358				      RELOCATION_RESERVED_NODES;
2359		return -EAGAIN;
2360	}
2361
2362	return 0;
2363}
2364
2365/*
2366 * relocate a block tree, and then update pointers in upper level
2367 * blocks that reference the block to point to the new location.
2368 *
2369 * if called by link_to_upper, the block has already been relocated.
2370 * in that case this function just updates pointers.
2371 */
2372static int do_relocation(struct btrfs_trans_handle *trans,
2373			 struct reloc_control *rc,
2374			 struct btrfs_backref_node *node,
2375			 struct btrfs_key *key,
2376			 struct btrfs_path *path, int lowest)
2377{
2378	struct btrfs_backref_node *upper;
2379	struct btrfs_backref_edge *edge;
2380	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
 
2381	struct btrfs_root *root;
2382	struct extent_buffer *eb;
2383	u32 blocksize;
2384	u64 bytenr;
 
2385	int slot;
2386	int ret = 0;
 
2387
2388	/*
2389	 * If we are lowest then this is the first time we're processing this
2390	 * block, and thus shouldn't have an eb associated with it yet.
2391	 */
2392	ASSERT(!lowest || !node->eb);
2393
2394	path->lowest_level = node->level + 1;
2395	rc->backref_cache.path[node->level] = node;
2396	list_for_each_entry(edge, &node->upper, list[LOWER]) {
2397		struct btrfs_ref ref = { 0 };
2398
2399		cond_resched();
2400
2401		upper = edge->node[UPPER];
2402		root = select_reloc_root(trans, rc, upper, edges);
2403		if (IS_ERR(root)) {
2404			ret = PTR_ERR(root);
2405			goto next;
2406		}
2407
2408		if (upper->eb && !upper->locked) {
2409			if (!lowest) {
2410				ret = btrfs_bin_search(upper->eb, key, &slot);
2411				if (ret < 0)
2412					goto next;
2413				BUG_ON(ret);
2414				bytenr = btrfs_node_blockptr(upper->eb, slot);
2415				if (node->eb->start == bytenr)
2416					goto next;
2417			}
2418			btrfs_backref_drop_node_buffer(upper);
2419		}
2420
2421		if (!upper->eb) {
2422			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2423			if (ret) {
2424				if (ret > 0)
2425					ret = -ENOENT;
 
 
2426
2427				btrfs_release_path(path);
2428				break;
2429			}
2430
2431			if (!upper->eb) {
2432				upper->eb = path->nodes[upper->level];
2433				path->nodes[upper->level] = NULL;
2434			} else {
2435				BUG_ON(upper->eb != path->nodes[upper->level]);
2436			}
2437
2438			upper->locked = 1;
2439			path->locks[upper->level] = 0;
2440
2441			slot = path->slots[upper->level];
2442			btrfs_release_path(path);
2443		} else {
2444			ret = btrfs_bin_search(upper->eb, key, &slot);
2445			if (ret < 0)
2446				goto next;
2447			BUG_ON(ret);
2448		}
2449
2450		bytenr = btrfs_node_blockptr(upper->eb, slot);
2451		if (lowest) {
2452			if (bytenr != node->bytenr) {
2453				btrfs_err(root->fs_info,
2454		"lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2455					  bytenr, node->bytenr, slot,
2456					  upper->eb->start);
2457				ret = -EIO;
2458				goto next;
2459			}
2460		} else {
2461			if (node->eb->start == bytenr)
2462				goto next;
2463		}
2464
2465		blocksize = root->fs_info->nodesize;
2466		eb = btrfs_read_node_slot(upper->eb, slot);
 
 
 
2467		if (IS_ERR(eb)) {
2468			ret = PTR_ERR(eb);
 
 
 
 
2469			goto next;
2470		}
2471		btrfs_tree_lock(eb);
 
2472
2473		if (!node->eb) {
2474			ret = btrfs_cow_block(trans, root, eb, upper->eb,
2475					      slot, &eb, BTRFS_NESTING_COW);
2476			btrfs_tree_unlock(eb);
2477			free_extent_buffer(eb);
2478			if (ret < 0)
 
2479				goto next;
2480			/*
2481			 * We've just COWed this block, it should have updated
2482			 * the correct backref node entry.
2483			 */
2484			ASSERT(node->eb == eb);
2485		} else {
2486			btrfs_set_node_blockptr(upper->eb, slot,
2487						node->eb->start);
2488			btrfs_set_node_ptr_generation(upper->eb, slot,
2489						      trans->transid);
2490			btrfs_mark_buffer_dirty(upper->eb);
2491
2492			btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF,
2493					       node->eb->start, blocksize,
2494					       upper->eb->start);
2495			btrfs_init_tree_ref(&ref, node->level,
2496					    btrfs_header_owner(upper->eb),
2497					    root->root_key.objectid, false);
2498			ret = btrfs_inc_extent_ref(trans, &ref);
2499			if (!ret)
2500				ret = btrfs_drop_subtree(trans, root, eb,
2501							 upper->eb);
2502			if (ret)
2503				btrfs_abort_transaction(trans, ret);
2504		}
2505next:
2506		if (!upper->pending)
2507			btrfs_backref_drop_node_buffer(upper);
2508		else
2509			btrfs_backref_unlock_node_buffer(upper);
2510		if (ret)
2511			break;
2512	}
2513
2514	if (!ret && node->pending) {
2515		btrfs_backref_drop_node_buffer(node);
2516		list_move_tail(&node->list, &rc->backref_cache.changed);
2517		node->pending = 0;
2518	}
2519
2520	path->lowest_level = 0;
2521
2522	/*
2523	 * We should have allocated all of our space in the block rsv and thus
2524	 * shouldn't ENOSPC.
2525	 */
2526	ASSERT(ret != -ENOSPC);
2527	return ret;
2528}
2529
2530static int link_to_upper(struct btrfs_trans_handle *trans,
2531			 struct reloc_control *rc,
2532			 struct btrfs_backref_node *node,
2533			 struct btrfs_path *path)
2534{
2535	struct btrfs_key key;
2536
2537	btrfs_node_key_to_cpu(node->eb, &key, 0);
2538	return do_relocation(trans, rc, node, &key, path, 0);
2539}
2540
2541static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2542				struct reloc_control *rc,
2543				struct btrfs_path *path, int err)
2544{
2545	LIST_HEAD(list);
2546	struct btrfs_backref_cache *cache = &rc->backref_cache;
2547	struct btrfs_backref_node *node;
2548	int level;
2549	int ret;
2550
2551	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2552		while (!list_empty(&cache->pending[level])) {
2553			node = list_entry(cache->pending[level].next,
2554					  struct btrfs_backref_node, list);
2555			list_move_tail(&node->list, &list);
2556			BUG_ON(!node->pending);
2557
2558			if (!err) {
2559				ret = link_to_upper(trans, rc, node, path);
2560				if (ret < 0)
2561					err = ret;
2562			}
2563		}
2564		list_splice_init(&list, &cache->pending[level]);
2565	}
2566	return err;
2567}
2568
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2569/*
2570 * mark a block and all blocks directly/indirectly reference the block
2571 * as processed.
2572 */
2573static void update_processed_blocks(struct reloc_control *rc,
2574				    struct btrfs_backref_node *node)
2575{
2576	struct btrfs_backref_node *next = node;
2577	struct btrfs_backref_edge *edge;
2578	struct btrfs_backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2579	int index = 0;
2580
2581	while (next) {
2582		cond_resched();
2583		while (1) {
2584			if (next->processed)
2585				break;
2586
2587			mark_block_processed(rc, next);
2588
2589			if (list_empty(&next->upper))
2590				break;
2591
2592			edge = list_entry(next->upper.next,
2593					struct btrfs_backref_edge, list[LOWER]);
2594			edges[index++] = edge;
2595			next = edge->node[UPPER];
2596		}
2597		next = walk_down_backref(edges, &index);
2598	}
2599}
2600
2601static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2602{
2603	u32 blocksize = rc->extent_root->fs_info->nodesize;
2604
2605	if (test_range_bit(&rc->processed_blocks, bytenr,
2606			   bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2607		return 1;
2608	return 0;
2609}
2610
2611static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2612			      struct tree_block *block)
2613{
2614	struct btrfs_tree_parent_check check = {
2615		.level = block->level,
2616		.owner_root = block->owner,
2617		.transid = block->key.offset
2618	};
2619	struct extent_buffer *eb;
2620
2621	eb = read_tree_block(fs_info, block->bytenr, &check);
2622	if (IS_ERR(eb))
 
 
2623		return PTR_ERR(eb);
2624	if (!extent_buffer_uptodate(eb)) {
2625		free_extent_buffer(eb);
2626		return -EIO;
2627	}
 
2628	if (block->level == 0)
2629		btrfs_item_key_to_cpu(eb, &block->key, 0);
2630	else
2631		btrfs_node_key_to_cpu(eb, &block->key, 0);
2632	free_extent_buffer(eb);
2633	block->key_ready = 1;
2634	return 0;
2635}
2636
2637/*
2638 * helper function to relocate a tree block
2639 */
2640static int relocate_tree_block(struct btrfs_trans_handle *trans,
2641				struct reloc_control *rc,
2642				struct btrfs_backref_node *node,
2643				struct btrfs_key *key,
2644				struct btrfs_path *path)
2645{
2646	struct btrfs_root *root;
2647	int ret = 0;
2648
2649	if (!node)
2650		return 0;
2651
2652	/*
2653	 * If we fail here we want to drop our backref_node because we are going
2654	 * to start over and regenerate the tree for it.
2655	 */
2656	ret = reserve_metadata_space(trans, rc, node);
2657	if (ret)
2658		goto out;
2659
2660	BUG_ON(node->processed);
2661	root = select_one_root(node);
2662	if (IS_ERR(root)) {
2663		ret = PTR_ERR(root);
2664
2665		/* See explanation in select_one_root for the -EUCLEAN case. */
2666		ASSERT(ret == -ENOENT);
2667		if (ret == -ENOENT) {
2668			ret = 0;
2669			update_processed_blocks(rc, node);
2670		}
2671		goto out;
2672	}
2673
 
 
 
 
 
 
2674	if (root) {
2675		if (test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) {
2676			/*
2677			 * This block was the root block of a root, and this is
2678			 * the first time we're processing the block and thus it
2679			 * should not have had the ->new_bytenr modified and
2680			 * should have not been included on the changed list.
2681			 *
2682			 * However in the case of corruption we could have
2683			 * multiple refs pointing to the same block improperly,
2684			 * and thus we would trip over these checks.  ASSERT()
2685			 * for the developer case, because it could indicate a
2686			 * bug in the backref code, however error out for a
2687			 * normal user in the case of corruption.
2688			 */
2689			ASSERT(node->new_bytenr == 0);
2690			ASSERT(list_empty(&node->list));
2691			if (node->new_bytenr || !list_empty(&node->list)) {
2692				btrfs_err(root->fs_info,
2693				  "bytenr %llu has improper references to it",
2694					  node->bytenr);
2695				ret = -EUCLEAN;
2696				goto out;
2697			}
2698			ret = btrfs_record_root_in_trans(trans, root);
2699			if (ret)
2700				goto out;
2701			/*
2702			 * Another thread could have failed, need to check if we
2703			 * have reloc_root actually set.
2704			 */
2705			if (!root->reloc_root) {
2706				ret = -ENOENT;
2707				goto out;
2708			}
2709			root = root->reloc_root;
2710			node->new_bytenr = root->node->start;
2711			btrfs_put_root(node->root);
2712			node->root = btrfs_grab_root(root);
2713			ASSERT(node->root);
2714			list_add_tail(&node->list, &rc->backref_cache.changed);
2715		} else {
2716			path->lowest_level = node->level;
2717			if (root == root->fs_info->chunk_root)
2718				btrfs_reserve_chunk_metadata(trans, false);
2719			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2720			btrfs_release_path(path);
2721			if (root == root->fs_info->chunk_root)
2722				btrfs_trans_release_chunk_metadata(trans);
2723			if (ret > 0)
2724				ret = 0;
2725		}
2726		if (!ret)
2727			update_processed_blocks(rc, node);
2728	} else {
2729		ret = do_relocation(trans, rc, node, key, path, 1);
2730	}
2731out:
2732	if (ret || node->level == 0 || node->cowonly)
2733		btrfs_backref_cleanup_node(&rc->backref_cache, node);
2734	return ret;
2735}
2736
2737/*
2738 * relocate a list of blocks
2739 */
2740static noinline_for_stack
2741int relocate_tree_blocks(struct btrfs_trans_handle *trans,
2742			 struct reloc_control *rc, struct rb_root *blocks)
2743{
2744	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2745	struct btrfs_backref_node *node;
2746	struct btrfs_path *path;
2747	struct tree_block *block;
2748	struct tree_block *next;
2749	int ret;
2750	int err = 0;
2751
2752	path = btrfs_alloc_path();
2753	if (!path) {
2754		err = -ENOMEM;
2755		goto out_free_blocks;
2756	}
2757
2758	/* Kick in readahead for tree blocks with missing keys */
2759	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
 
2760		if (!block->key_ready)
2761			btrfs_readahead_tree_block(fs_info, block->bytenr,
2762						   block->owner, 0,
2763						   block->level);
2764	}
2765
2766	/* Get first keys */
2767	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
 
2768		if (!block->key_ready) {
2769			err = get_tree_block_key(fs_info, block);
2770			if (err)
2771				goto out_free_path;
2772		}
 
2773	}
2774
2775	/* Do tree relocation */
2776	rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) {
 
 
2777		node = build_backref_tree(rc, &block->key,
2778					  block->level, block->bytenr);
2779		if (IS_ERR(node)) {
2780			err = PTR_ERR(node);
2781			goto out;
2782		}
2783
2784		ret = relocate_tree_block(trans, rc, node, &block->key,
2785					  path);
2786		if (ret < 0) {
2787			err = ret;
2788			break;
 
2789		}
 
2790	}
2791out:
2792	err = finish_pending_nodes(trans, rc, path, err);
2793
2794out_free_path:
2795	btrfs_free_path(path);
2796out_free_blocks:
2797	free_block_list(blocks);
2798	return err;
2799}
2800
2801static noinline_for_stack int prealloc_file_extent_cluster(
2802				struct btrfs_inode *inode,
2803				struct file_extent_cluster *cluster)
2804{
2805	u64 alloc_hint = 0;
2806	u64 start;
2807	u64 end;
2808	u64 offset = inode->index_cnt;
2809	u64 num_bytes;
2810	int nr;
2811	int ret = 0;
2812	u64 i_size = i_size_read(&inode->vfs_inode);
2813	u64 prealloc_start = cluster->start - offset;
2814	u64 prealloc_end = cluster->end - offset;
2815	u64 cur_offset = prealloc_start;
2816
2817	/*
2818	 * For subpage case, previous i_size may not be aligned to PAGE_SIZE.
2819	 * This means the range [i_size, PAGE_END + 1) is filled with zeros by
2820	 * btrfs_do_readpage() call of previously relocated file cluster.
2821	 *
2822	 * If the current cluster starts in the above range, btrfs_do_readpage()
2823	 * will skip the read, and relocate_one_page() will later writeback
2824	 * the padding zeros as new data, causing data corruption.
2825	 *
2826	 * Here we have to manually invalidate the range (i_size, PAGE_END + 1).
2827	 */
2828	if (!IS_ALIGNED(i_size, PAGE_SIZE)) {
2829		struct address_space *mapping = inode->vfs_inode.i_mapping;
2830		struct btrfs_fs_info *fs_info = inode->root->fs_info;
2831		const u32 sectorsize = fs_info->sectorsize;
2832		struct page *page;
2833
2834		ASSERT(sectorsize < PAGE_SIZE);
2835		ASSERT(IS_ALIGNED(i_size, sectorsize));
2836
2837		/*
2838		 * Subpage can't handle page with DIRTY but without UPTODATE
2839		 * bit as it can lead to the following deadlock:
2840		 *
2841		 * btrfs_read_folio()
2842		 * | Page already *locked*
2843		 * |- btrfs_lock_and_flush_ordered_range()
2844		 *    |- btrfs_start_ordered_extent()
2845		 *       |- extent_write_cache_pages()
2846		 *          |- lock_page()
2847		 *             We try to lock the page we already hold.
2848		 *
2849		 * Here we just writeback the whole data reloc inode, so that
2850		 * we will be ensured to have no dirty range in the page, and
2851		 * are safe to clear the uptodate bits.
2852		 *
2853		 * This shouldn't cause too much overhead, as we need to write
2854		 * the data back anyway.
2855		 */
2856		ret = filemap_write_and_wait(mapping);
2857		if (ret < 0)
2858			return ret;
2859
2860		clear_extent_bits(&inode->io_tree, i_size,
2861				  round_up(i_size, PAGE_SIZE) - 1,
2862				  EXTENT_UPTODATE);
2863		page = find_lock_page(mapping, i_size >> PAGE_SHIFT);
2864		/*
2865		 * If page is freed we don't need to do anything then, as we
2866		 * will re-read the whole page anyway.
2867		 */
2868		if (page) {
2869			btrfs_subpage_clear_uptodate(fs_info, page, i_size,
2870					round_up(i_size, PAGE_SIZE) - i_size);
2871			unlock_page(page);
2872			put_page(page);
2873		}
2874	}
2875
2876	BUG_ON(cluster->start != cluster->boundary[0]);
2877	ret = btrfs_alloc_data_chunk_ondemand(inode,
2878					      prealloc_end + 1 - prealloc_start);
2879	if (ret)
2880		return ret;
2881
2882	btrfs_inode_lock(inode, 0);
2883	for (nr = 0; nr < cluster->nr; nr++) {
2884		struct extent_state *cached_state = NULL;
 
2885
 
 
2886		start = cluster->boundary[nr] - offset;
2887		if (nr + 1 < cluster->nr)
2888			end = cluster->boundary[nr + 1] - 1 - offset;
2889		else
2890			end = cluster->end - offset;
2891
2892		lock_extent(&inode->io_tree, start, end, &cached_state);
2893		num_bytes = end + 1 - start;
2894		ret = btrfs_prealloc_file_range(&inode->vfs_inode, 0, start,
 
 
 
2895						num_bytes, num_bytes,
2896						end + 1, &alloc_hint);
2897		cur_offset = end + 1;
2898		unlock_extent(&inode->io_tree, start, end, &cached_state);
2899		if (ret)
2900			break;
 
2901	}
2902	btrfs_inode_unlock(inode, 0);
2903
2904	if (cur_offset < prealloc_end)
2905		btrfs_free_reserved_data_space_noquota(inode->root->fs_info,
2906					       prealloc_end + 1 - cur_offset);
 
 
 
2907	return ret;
2908}
2909
2910static noinline_for_stack int setup_relocation_extent_mapping(struct inode *inode,
2911				u64 start, u64 end, u64 block_start)
 
2912{
 
 
2913	struct extent_map *em;
2914	struct extent_state *cached_state = NULL;
2915	int ret = 0;
2916
2917	em = alloc_extent_map();
2918	if (!em)
2919		return -ENOMEM;
2920
2921	em->start = start;
2922	em->len = end + 1 - start;
2923	em->block_len = em->len;
2924	em->block_start = block_start;
 
2925	set_bit(EXTENT_FLAG_PINNED, &em->flags);
2926
2927	lock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
2928	ret = btrfs_replace_extent_map_range(BTRFS_I(inode), em, false);
2929	unlock_extent(&BTRFS_I(inode)->io_tree, start, end, &cached_state);
2930	free_extent_map(em);
2931
2932	return ret;
2933}
2934
2935/*
2936 * Allow error injection to test balance/relocation cancellation
2937 */
2938noinline int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info)
2939{
2940	return atomic_read(&fs_info->balance_cancel_req) ||
2941		atomic_read(&fs_info->reloc_cancel_req) ||
2942		fatal_signal_pending(current);
2943}
2944ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE);
2945
2946static u64 get_cluster_boundary_end(struct file_extent_cluster *cluster,
2947				    int cluster_nr)
2948{
2949	/* Last extent, use cluster end directly */
2950	if (cluster_nr >= cluster->nr - 1)
2951		return cluster->end;
2952
2953	/* Use next boundary start*/
2954	return cluster->boundary[cluster_nr + 1] - 1;
2955}
2956
2957static int relocate_one_page(struct inode *inode, struct file_ra_state *ra,
2958			     struct file_extent_cluster *cluster,
2959			     int *cluster_nr, unsigned long page_index)
2960{
2961	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
2962	u64 offset = BTRFS_I(inode)->index_cnt;
2963	const unsigned long last_index = (cluster->end - offset) >> PAGE_SHIFT;
2964	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
2965	struct page *page;
2966	u64 page_start;
2967	u64 page_end;
2968	u64 cur;
2969	int ret;
2970
2971	ASSERT(page_index <= last_index);
2972	page = find_lock_page(inode->i_mapping, page_index);
2973	if (!page) {
2974		page_cache_sync_readahead(inode->i_mapping, ra, NULL,
2975				page_index, last_index + 1 - page_index);
2976		page = find_or_create_page(inode->i_mapping, page_index, mask);
2977		if (!page)
2978			return -ENOMEM;
2979	}
2980	ret = set_page_extent_mapped(page);
2981	if (ret < 0)
2982		goto release_page;
2983
2984	if (PageReadahead(page))
2985		page_cache_async_readahead(inode->i_mapping, ra, NULL,
2986				page_folio(page), page_index,
2987				last_index + 1 - page_index);
2988
2989	if (!PageUptodate(page)) {
2990		btrfs_read_folio(NULL, page_folio(page));
2991		lock_page(page);
2992		if (!PageUptodate(page)) {
2993			ret = -EIO;
2994			goto release_page;
2995		}
2996	}
2997
2998	page_start = page_offset(page);
2999	page_end = page_start + PAGE_SIZE - 1;
3000
3001	/*
3002	 * Start from the cluster, as for subpage case, the cluster can start
3003	 * inside the page.
3004	 */
3005	cur = max(page_start, cluster->boundary[*cluster_nr] - offset);
3006	while (cur <= page_end) {
3007		struct extent_state *cached_state = NULL;
3008		u64 extent_start = cluster->boundary[*cluster_nr] - offset;
3009		u64 extent_end = get_cluster_boundary_end(cluster,
3010						*cluster_nr) - offset;
3011		u64 clamped_start = max(page_start, extent_start);
3012		u64 clamped_end = min(page_end, extent_end);
3013		u32 clamped_len = clamped_end + 1 - clamped_start;
3014
3015		/* Reserve metadata for this range */
3016		ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3017						      clamped_len, clamped_len,
3018						      false);
3019		if (ret)
3020			goto release_page;
3021
3022		/* Mark the range delalloc and dirty for later writeback */
3023		lock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
3024			    &cached_state);
3025		ret = btrfs_set_extent_delalloc(BTRFS_I(inode), clamped_start,
3026						clamped_end, 0, &cached_state);
3027		if (ret) {
3028			clear_extent_bit(&BTRFS_I(inode)->io_tree,
3029					 clamped_start, clamped_end,
3030					 EXTENT_LOCKED | EXTENT_BOUNDARY,
3031					 &cached_state);
3032			btrfs_delalloc_release_metadata(BTRFS_I(inode),
3033							clamped_len, true);
3034			btrfs_delalloc_release_extents(BTRFS_I(inode),
3035						       clamped_len);
3036			goto release_page;
3037		}
3038		btrfs_page_set_dirty(fs_info, page, clamped_start, clamped_len);
3039
3040		/*
3041		 * Set the boundary if it's inside the page.
3042		 * Data relocation requires the destination extents to have the
3043		 * same size as the source.
3044		 * EXTENT_BOUNDARY bit prevents current extent from being merged
3045		 * with previous extent.
3046		 */
3047		if (in_range(cluster->boundary[*cluster_nr] - offset,
3048			     page_start, PAGE_SIZE)) {
3049			u64 boundary_start = cluster->boundary[*cluster_nr] -
3050						offset;
3051			u64 boundary_end = boundary_start +
3052					   fs_info->sectorsize - 1;
3053
3054			set_extent_bits(&BTRFS_I(inode)->io_tree,
3055					boundary_start, boundary_end,
3056					EXTENT_BOUNDARY);
3057		}
3058		unlock_extent(&BTRFS_I(inode)->io_tree, clamped_start, clamped_end,
3059			      &cached_state);
3060		btrfs_delalloc_release_extents(BTRFS_I(inode), clamped_len);
3061		cur += clamped_len;
3062
3063		/* Crossed extent end, go to next extent */
3064		if (cur >= extent_end) {
3065			(*cluster_nr)++;
3066			/* Just finished the last extent of the cluster, exit. */
3067			if (*cluster_nr >= cluster->nr)
3068				break;
3069		}
 
3070	}
3071	unlock_page(page);
3072	put_page(page);
3073
3074	balance_dirty_pages_ratelimited(inode->i_mapping);
3075	btrfs_throttle(fs_info);
3076	if (btrfs_should_cancel_balance(fs_info))
3077		ret = -ECANCELED;
3078	return ret;
3079
3080release_page:
3081	unlock_page(page);
3082	put_page(page);
3083	return ret;
3084}
3085
3086static int relocate_file_extent_cluster(struct inode *inode,
3087					struct file_extent_cluster *cluster)
3088{
 
 
 
3089	u64 offset = BTRFS_I(inode)->index_cnt;
3090	unsigned long index;
3091	unsigned long last_index;
 
3092	struct file_ra_state *ra;
3093	int cluster_nr = 0;
 
3094	int ret = 0;
3095
3096	if (!cluster->nr)
3097		return 0;
3098
3099	ra = kzalloc(sizeof(*ra), GFP_NOFS);
3100	if (!ra)
3101		return -ENOMEM;
3102
3103	ret = prealloc_file_extent_cluster(BTRFS_I(inode), cluster);
3104	if (ret)
3105		goto out;
3106
3107	file_ra_state_init(ra, inode->i_mapping);
3108
3109	ret = setup_relocation_extent_mapping(inode, cluster->start - offset,
3110				   cluster->end - offset, cluster->start);
3111	if (ret)
3112		goto out;
3113
 
3114	last_index = (cluster->end - offset) >> PAGE_SHIFT;
3115	for (index = (cluster->start - offset) >> PAGE_SHIFT;
3116	     index <= last_index && !ret; index++)
3117		ret = relocate_one_page(inode, ra, cluster, &cluster_nr, index);
3118	if (ret == 0)
3119		WARN_ON(cluster_nr != cluster->nr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3120out:
3121	kfree(ra);
3122	return ret;
3123}
3124
3125static noinline_for_stack
3126int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3127			 struct file_extent_cluster *cluster)
3128{
3129	int ret;
3130
3131	if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3132		ret = relocate_file_extent_cluster(inode, cluster);
3133		if (ret)
3134			return ret;
3135		cluster->nr = 0;
3136	}
3137
3138	if (!cluster->nr)
3139		cluster->start = extent_key->objectid;
3140	else
3141		BUG_ON(cluster->nr >= MAX_EXTENTS);
3142	cluster->end = extent_key->objectid + extent_key->offset - 1;
3143	cluster->boundary[cluster->nr] = extent_key->objectid;
3144	cluster->nr++;
3145
3146	if (cluster->nr >= MAX_EXTENTS) {
3147		ret = relocate_file_extent_cluster(inode, cluster);
3148		if (ret)
3149			return ret;
3150		cluster->nr = 0;
3151	}
3152	return 0;
3153}
3154
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3155/*
3156 * helper to add a tree block to the list.
3157 * the major work is getting the generation and level of the block
3158 */
3159static int add_tree_block(struct reloc_control *rc,
3160			  struct btrfs_key *extent_key,
3161			  struct btrfs_path *path,
3162			  struct rb_root *blocks)
3163{
3164	struct extent_buffer *eb;
3165	struct btrfs_extent_item *ei;
3166	struct btrfs_tree_block_info *bi;
3167	struct tree_block *block;
3168	struct rb_node *rb_node;
3169	u32 item_size;
3170	int level = -1;
3171	u64 generation;
3172	u64 owner = 0;
3173
3174	eb =  path->nodes[0];
3175	item_size = btrfs_item_size(eb, path->slots[0]);
3176
3177	if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3178	    item_size >= sizeof(*ei) + sizeof(*bi)) {
3179		unsigned long ptr = 0, end;
3180
3181		ei = btrfs_item_ptr(eb, path->slots[0],
3182				struct btrfs_extent_item);
3183		end = (unsigned long)ei + item_size;
3184		if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3185			bi = (struct btrfs_tree_block_info *)(ei + 1);
3186			level = btrfs_tree_block_level(eb, bi);
3187			ptr = (unsigned long)(bi + 1);
3188		} else {
3189			level = (int)extent_key->offset;
3190			ptr = (unsigned long)(ei + 1);
3191		}
3192		generation = btrfs_extent_generation(eb, ei);
3193
3194		/*
3195		 * We're reading random blocks without knowing their owner ahead
3196		 * of time.  This is ok most of the time, as all reloc roots and
3197		 * fs roots have the same lock type.  However normal trees do
3198		 * not, and the only way to know ahead of time is to read the
3199		 * inline ref offset.  We know it's an fs root if
3200		 *
3201		 * 1. There's more than one ref.
3202		 * 2. There's a SHARED_DATA_REF_KEY set.
3203		 * 3. FULL_BACKREF is set on the flags.
3204		 *
3205		 * Otherwise it's safe to assume that the ref offset == the
3206		 * owner of this block, so we can use that when calling
3207		 * read_tree_block.
3208		 */
3209		if (btrfs_extent_refs(eb, ei) == 1 &&
3210		    !(btrfs_extent_flags(eb, ei) &
3211		      BTRFS_BLOCK_FLAG_FULL_BACKREF) &&
3212		    ptr < end) {
3213			struct btrfs_extent_inline_ref *iref;
3214			int type;
3215
3216			iref = (struct btrfs_extent_inline_ref *)ptr;
3217			type = btrfs_get_extent_inline_ref_type(eb, iref,
3218							BTRFS_REF_TYPE_BLOCK);
3219			if (type == BTRFS_REF_TYPE_INVALID)
3220				return -EINVAL;
3221			if (type == BTRFS_TREE_BLOCK_REF_KEY)
3222				owner = btrfs_extent_inline_ref_offset(eb, iref);
3223		}
3224	} else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) {
3225		btrfs_print_v0_err(eb->fs_info);
3226		btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL);
3227		return -EINVAL;
3228	} else {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3229		BUG();
 
3230	}
3231
3232	btrfs_release_path(path);
3233
3234	BUG_ON(level == -1);
3235
3236	block = kmalloc(sizeof(*block), GFP_NOFS);
3237	if (!block)
3238		return -ENOMEM;
3239
3240	block->bytenr = extent_key->objectid;
3241	block->key.objectid = rc->extent_root->fs_info->nodesize;
3242	block->key.offset = generation;
3243	block->level = level;
3244	block->key_ready = 0;
3245	block->owner = owner;
3246
3247	rb_node = rb_simple_insert(blocks, block->bytenr, &block->rb_node);
3248	if (rb_node)
3249		btrfs_backref_panic(rc->extent_root->fs_info, block->bytenr,
3250				    -EEXIST);
3251
3252	return 0;
3253}
3254
3255/*
3256 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3257 */
3258static int __add_tree_block(struct reloc_control *rc,
3259			    u64 bytenr, u32 blocksize,
3260			    struct rb_root *blocks)
3261{
3262	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3263	struct btrfs_path *path;
3264	struct btrfs_key key;
3265	int ret;
3266	bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3267
3268	if (tree_block_processed(bytenr, rc))
3269		return 0;
3270
3271	if (rb_simple_search(blocks, bytenr))
3272		return 0;
3273
3274	path = btrfs_alloc_path();
3275	if (!path)
3276		return -ENOMEM;
3277again:
3278	key.objectid = bytenr;
3279	if (skinny) {
3280		key.type = BTRFS_METADATA_ITEM_KEY;
3281		key.offset = (u64)-1;
3282	} else {
3283		key.type = BTRFS_EXTENT_ITEM_KEY;
3284		key.offset = blocksize;
3285	}
3286
3287	path->search_commit_root = 1;
3288	path->skip_locking = 1;
3289	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3290	if (ret < 0)
3291		goto out;
3292
3293	if (ret > 0 && skinny) {
3294		if (path->slots[0]) {
3295			path->slots[0]--;
3296			btrfs_item_key_to_cpu(path->nodes[0], &key,
3297					      path->slots[0]);
3298			if (key.objectid == bytenr &&
3299			    (key.type == BTRFS_METADATA_ITEM_KEY ||
3300			     (key.type == BTRFS_EXTENT_ITEM_KEY &&
3301			      key.offset == blocksize)))
3302				ret = 0;
3303		}
3304
3305		if (ret) {
3306			skinny = false;
3307			btrfs_release_path(path);
3308			goto again;
3309		}
3310	}
3311	if (ret) {
3312		ASSERT(ret == 1);
3313		btrfs_print_leaf(path->nodes[0]);
3314		btrfs_err(fs_info,
3315	     "tree block extent item (%llu) is not found in extent tree",
3316		     bytenr);
3317		WARN_ON(1);
3318		ret = -EINVAL;
3319		goto out;
3320	}
3321
3322	ret = add_tree_block(rc, &key, path, blocks);
3323out:
3324	btrfs_free_path(path);
3325	return ret;
3326}
3327
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3328static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3329				    struct btrfs_block_group *block_group,
3330				    struct inode *inode,
3331				    u64 ino)
3332{
 
3333	struct btrfs_root *root = fs_info->tree_root;
3334	struct btrfs_trans_handle *trans;
3335	int ret = 0;
3336
3337	if (inode)
3338		goto truncate;
3339
3340	inode = btrfs_iget(fs_info->sb, ino, root);
3341	if (IS_ERR(inode))
 
 
 
 
 
 
3342		return -ENOENT;
 
3343
3344truncate:
3345	ret = btrfs_check_trunc_cache_free_space(fs_info,
3346						 &fs_info->global_block_rsv);
3347	if (ret)
3348		goto out;
3349
3350	trans = btrfs_join_transaction(root);
3351	if (IS_ERR(trans)) {
3352		ret = PTR_ERR(trans);
3353		goto out;
3354	}
3355
3356	ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3357
3358	btrfs_end_transaction(trans);
3359	btrfs_btree_balance_dirty(fs_info);
3360out:
3361	iput(inode);
3362	return ret;
3363}
3364
3365/*
3366 * Locate the free space cache EXTENT_DATA in root tree leaf and delete the
3367 * cache inode, to avoid free space cache data extent blocking data relocation.
3368 */
3369static int delete_v1_space_cache(struct extent_buffer *leaf,
3370				 struct btrfs_block_group *block_group,
3371				 u64 data_bytenr)
 
 
3372{
3373	u64 space_cache_ino;
3374	struct btrfs_file_extent_item *ei;
 
 
 
 
3375	struct btrfs_key key;
3376	bool found = false;
3377	int i;
 
 
 
 
 
 
3378	int ret;
3379
3380	if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID)
3381		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3382
3383	for (i = 0; i < btrfs_header_nritems(leaf); i++) {
3384		u8 type;
 
 
3385
3386		btrfs_item_key_to_cpu(leaf, &key, i);
3387		if (key.type != BTRFS_EXTENT_DATA_KEY)
3388			continue;
3389		ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
3390		type = btrfs_file_extent_type(leaf, ei);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3391
3392		if ((type == BTRFS_FILE_EXTENT_REG ||
3393		     type == BTRFS_FILE_EXTENT_PREALLOC) &&
3394		    btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) {
3395			found = true;
3396			space_cache_ino = key.objectid;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3397			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3398		}
 
 
 
 
 
 
 
3399	}
3400	if (!found)
3401		return -ENOENT;
3402	ret = delete_block_group_cache(leaf->fs_info, block_group, NULL,
3403					space_cache_ino);
3404	return ret;
3405}
3406
3407/*
3408 * helper to find all tree blocks that reference a given data extent
3409 */
3410static noinline_for_stack
3411int add_data_references(struct reloc_control *rc,
3412			struct btrfs_key *extent_key,
3413			struct btrfs_path *path,
3414			struct rb_root *blocks)
3415{
3416	struct btrfs_backref_walk_ctx ctx = { 0 };
3417	struct ulist_iterator leaf_uiter;
3418	struct ulist_node *ref_node = NULL;
3419	const u32 blocksize = rc->extent_root->fs_info->nodesize;
 
 
 
3420	int ret = 0;
 
3421
3422	btrfs_release_path(path);
3423
3424	ctx.bytenr = extent_key->objectid;
3425	ctx.ignore_extent_item_pos = true;
3426	ctx.fs_info = rc->extent_root->fs_info;
 
 
 
 
3427
3428	ret = btrfs_find_all_leafs(&ctx);
3429	if (ret < 0)
3430		return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3431
3432	ULIST_ITER_INIT(&leaf_uiter);
3433	while ((ref_node = ulist_next(ctx.refs, &leaf_uiter))) {
3434		struct btrfs_tree_parent_check check = { 0 };
3435		struct extent_buffer *eb;
 
 
 
 
 
 
 
 
 
3436
3437		eb = read_tree_block(ctx.fs_info, ref_node->val, &check);
3438		if (IS_ERR(eb)) {
3439			ret = PTR_ERR(eb);
3440			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3441		}
3442		ret = delete_v1_space_cache(eb, rc->block_group,
3443					    extent_key->objectid);
3444		free_extent_buffer(eb);
3445		if (ret < 0)
3446			break;
3447		ret = __add_tree_block(rc, ref_node->val, blocksize, blocks);
3448		if (ret < 0)
3449			break;
 
 
3450	}
3451	if (ret < 0)
 
 
3452		free_block_list(blocks);
3453	ulist_free(ctx.refs);
3454	return ret;
3455}
3456
3457/*
3458 * helper to find next unprocessed extent
3459 */
3460static noinline_for_stack
3461int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3462		     struct btrfs_key *extent_key)
3463{
3464	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3465	struct btrfs_key key;
3466	struct extent_buffer *leaf;
3467	u64 start, end, last;
3468	int ret;
3469
3470	last = rc->block_group->start + rc->block_group->length;
3471	while (1) {
3472		cond_resched();
3473		if (rc->search_start >= last) {
3474			ret = 1;
3475			break;
3476		}
3477
3478		key.objectid = rc->search_start;
3479		key.type = BTRFS_EXTENT_ITEM_KEY;
3480		key.offset = 0;
3481
3482		path->search_commit_root = 1;
3483		path->skip_locking = 1;
3484		ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3485					0, 0);
3486		if (ret < 0)
3487			break;
3488next:
3489		leaf = path->nodes[0];
3490		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3491			ret = btrfs_next_leaf(rc->extent_root, path);
3492			if (ret != 0)
3493				break;
3494			leaf = path->nodes[0];
3495		}
3496
3497		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3498		if (key.objectid >= last) {
3499			ret = 1;
3500			break;
3501		}
3502
3503		if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3504		    key.type != BTRFS_METADATA_ITEM_KEY) {
3505			path->slots[0]++;
3506			goto next;
3507		}
3508
3509		if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3510		    key.objectid + key.offset <= rc->search_start) {
3511			path->slots[0]++;
3512			goto next;
3513		}
3514
3515		if (key.type == BTRFS_METADATA_ITEM_KEY &&
3516		    key.objectid + fs_info->nodesize <=
3517		    rc->search_start) {
3518			path->slots[0]++;
3519			goto next;
3520		}
3521
3522		ret = find_first_extent_bit(&rc->processed_blocks,
3523					    key.objectid, &start, &end,
3524					    EXTENT_DIRTY, NULL);
3525
3526		if (ret == 0 && start <= key.objectid) {
3527			btrfs_release_path(path);
3528			rc->search_start = end + 1;
3529		} else {
3530			if (key.type == BTRFS_EXTENT_ITEM_KEY)
3531				rc->search_start = key.objectid + key.offset;
3532			else
3533				rc->search_start = key.objectid +
3534					fs_info->nodesize;
3535			memcpy(extent_key, &key, sizeof(key));
3536			return 0;
3537		}
3538	}
3539	btrfs_release_path(path);
3540	return ret;
3541}
3542
3543static void set_reloc_control(struct reloc_control *rc)
3544{
3545	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3546
3547	mutex_lock(&fs_info->reloc_mutex);
3548	fs_info->reloc_ctl = rc;
3549	mutex_unlock(&fs_info->reloc_mutex);
3550}
3551
3552static void unset_reloc_control(struct reloc_control *rc)
3553{
3554	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3555
3556	mutex_lock(&fs_info->reloc_mutex);
3557	fs_info->reloc_ctl = NULL;
3558	mutex_unlock(&fs_info->reloc_mutex);
3559}
3560
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3561static noinline_for_stack
3562int prepare_to_relocate(struct reloc_control *rc)
3563{
3564	struct btrfs_trans_handle *trans;
3565	int ret;
3566
3567	rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3568					      BTRFS_BLOCK_RSV_TEMP);
3569	if (!rc->block_rsv)
3570		return -ENOMEM;
3571
3572	memset(&rc->cluster, 0, sizeof(rc->cluster));
3573	rc->search_start = rc->block_group->start;
3574	rc->extents_found = 0;
3575	rc->nodes_relocated = 0;
3576	rc->merging_rsv_size = 0;
3577	rc->reserved_bytes = 0;
3578	rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
3579			      RELOCATION_RESERVED_NODES;
3580	ret = btrfs_block_rsv_refill(rc->extent_root->fs_info,
3581				     rc->block_rsv, rc->block_rsv->size,
3582				     BTRFS_RESERVE_FLUSH_ALL);
3583	if (ret)
3584		return ret;
3585
3586	rc->create_reloc_tree = 1;
3587	set_reloc_control(rc);
3588
3589	trans = btrfs_join_transaction(rc->extent_root);
3590	if (IS_ERR(trans)) {
3591		unset_reloc_control(rc);
3592		/*
3593		 * extent tree is not a ref_cow tree and has no reloc_root to
3594		 * cleanup.  And callers are responsible to free the above
3595		 * block rsv.
3596		 */
3597		return PTR_ERR(trans);
3598	}
3599
3600	ret = btrfs_commit_transaction(trans);
3601	if (ret)
3602		unset_reloc_control(rc);
3603
3604	return ret;
3605}
3606
3607static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
3608{
3609	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3610	struct rb_root blocks = RB_ROOT;
3611	struct btrfs_key key;
3612	struct btrfs_trans_handle *trans = NULL;
3613	struct btrfs_path *path;
3614	struct btrfs_extent_item *ei;
3615	u64 flags;
 
3616	int ret;
3617	int err = 0;
3618	int progress = 0;
3619
3620	path = btrfs_alloc_path();
3621	if (!path)
3622		return -ENOMEM;
3623	path->reada = READA_FORWARD;
3624
3625	ret = prepare_to_relocate(rc);
3626	if (ret) {
3627		err = ret;
3628		goto out_free;
3629	}
3630
3631	while (1) {
3632		rc->reserved_bytes = 0;
3633		ret = btrfs_block_rsv_refill(fs_info, rc->block_rsv,
3634					     rc->block_rsv->size,
3635					     BTRFS_RESERVE_FLUSH_ALL);
3636		if (ret) {
3637			err = ret;
3638			break;
3639		}
3640		progress++;
3641		trans = btrfs_start_transaction(rc->extent_root, 0);
3642		if (IS_ERR(trans)) {
3643			err = PTR_ERR(trans);
3644			trans = NULL;
3645			break;
3646		}
3647restart:
3648		if (update_backref_cache(trans, &rc->backref_cache)) {
3649			btrfs_end_transaction(trans);
3650			trans = NULL;
3651			continue;
3652		}
3653
3654		ret = find_next_extent(rc, path, &key);
3655		if (ret < 0)
3656			err = ret;
3657		if (ret != 0)
3658			break;
3659
3660		rc->extents_found++;
3661
3662		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
3663				    struct btrfs_extent_item);
3664		flags = btrfs_extent_flags(path->nodes[0], ei);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3665
3666		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
3667			ret = add_tree_block(rc, &key, path, &blocks);
3668		} else if (rc->stage == UPDATE_DATA_PTRS &&
3669			   (flags & BTRFS_EXTENT_FLAG_DATA)) {
3670			ret = add_data_references(rc, &key, path, &blocks);
3671		} else {
3672			btrfs_release_path(path);
3673			ret = 0;
3674		}
3675		if (ret < 0) {
3676			err = ret;
3677			break;
3678		}
3679
3680		if (!RB_EMPTY_ROOT(&blocks)) {
3681			ret = relocate_tree_blocks(trans, rc, &blocks);
3682			if (ret < 0) {
 
 
 
 
 
 
3683				if (ret != -EAGAIN) {
3684					err = ret;
3685					break;
3686				}
3687				rc->extents_found--;
3688				rc->search_start = key.objectid;
3689			}
3690		}
3691
3692		btrfs_end_transaction_throttle(trans);
3693		btrfs_btree_balance_dirty(fs_info);
3694		trans = NULL;
3695
3696		if (rc->stage == MOVE_DATA_EXTENTS &&
3697		    (flags & BTRFS_EXTENT_FLAG_DATA)) {
3698			rc->found_file_extent = 1;
3699			ret = relocate_data_extent(rc->data_inode,
3700						   &key, &rc->cluster);
3701			if (ret < 0) {
3702				err = ret;
3703				break;
3704			}
3705		}
3706		if (btrfs_should_cancel_balance(fs_info)) {
3707			err = -ECANCELED;
3708			break;
3709		}
3710	}
3711	if (trans && progress && err == -ENOSPC) {
3712		ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags);
 
3713		if (ret == 1) {
3714			err = 0;
3715			progress = 0;
3716			goto restart;
3717		}
3718	}
3719
3720	btrfs_release_path(path);
3721	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
3722
3723	if (trans) {
3724		btrfs_end_transaction_throttle(trans);
3725		btrfs_btree_balance_dirty(fs_info);
3726	}
3727
3728	if (!err) {
3729		ret = relocate_file_extent_cluster(rc->data_inode,
3730						   &rc->cluster);
3731		if (ret < 0)
3732			err = ret;
3733	}
3734
3735	rc->create_reloc_tree = 0;
3736	set_reloc_control(rc);
3737
3738	btrfs_backref_release_cache(&rc->backref_cache);
3739	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3740
3741	/*
3742	 * Even in the case when the relocation is cancelled, we should all go
3743	 * through prepare_to_merge() and merge_reloc_roots().
3744	 *
3745	 * For error (including cancelled balance), prepare_to_merge() will
3746	 * mark all reloc trees orphan, then queue them for cleanup in
3747	 * merge_reloc_roots()
3748	 */
3749	err = prepare_to_merge(rc, err);
3750
3751	merge_reloc_roots(rc);
3752
3753	rc->merge_reloc_tree = 0;
3754	unset_reloc_control(rc);
3755	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL);
3756
3757	/* get rid of pinned extents */
3758	trans = btrfs_join_transaction(rc->extent_root);
3759	if (IS_ERR(trans)) {
3760		err = PTR_ERR(trans);
3761		goto out_free;
3762	}
3763	ret = btrfs_commit_transaction(trans);
3764	if (ret && !err)
3765		err = ret;
3766out_free:
3767	ret = clean_dirty_subvols(rc);
3768	if (ret < 0 && !err)
3769		err = ret;
3770	btrfs_free_block_rsv(fs_info, rc->block_rsv);
3771	btrfs_free_path(path);
3772	return err;
3773}
3774
3775static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
3776				 struct btrfs_root *root, u64 objectid)
3777{
3778	struct btrfs_path *path;
3779	struct btrfs_inode_item *item;
3780	struct extent_buffer *leaf;
3781	int ret;
3782
3783	path = btrfs_alloc_path();
3784	if (!path)
3785		return -ENOMEM;
3786
3787	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
3788	if (ret)
3789		goto out;
3790
3791	leaf = path->nodes[0];
3792	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
3793	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
3794	btrfs_set_inode_generation(leaf, item, 1);
3795	btrfs_set_inode_size(leaf, item, 0);
3796	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
3797	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
3798					  BTRFS_INODE_PREALLOC);
3799	btrfs_mark_buffer_dirty(leaf);
3800out:
3801	btrfs_free_path(path);
3802	return ret;
3803}
3804
3805static void delete_orphan_inode(struct btrfs_trans_handle *trans,
3806				struct btrfs_root *root, u64 objectid)
3807{
3808	struct btrfs_path *path;
3809	struct btrfs_key key;
3810	int ret = 0;
3811
3812	path = btrfs_alloc_path();
3813	if (!path) {
3814		ret = -ENOMEM;
3815		goto out;
3816	}
3817
3818	key.objectid = objectid;
3819	key.type = BTRFS_INODE_ITEM_KEY;
3820	key.offset = 0;
3821	ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3822	if (ret) {
3823		if (ret > 0)
3824			ret = -ENOENT;
3825		goto out;
3826	}
3827	ret = btrfs_del_item(trans, root, path);
3828out:
3829	if (ret)
3830		btrfs_abort_transaction(trans, ret);
3831	btrfs_free_path(path);
3832}
3833
3834/*
3835 * helper to create inode for data relocation.
3836 * the inode is in data relocation tree and its link count is 0
3837 */
3838static noinline_for_stack
3839struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
3840				 struct btrfs_block_group *group)
3841{
3842	struct inode *inode = NULL;
3843	struct btrfs_trans_handle *trans;
3844	struct btrfs_root *root;
 
3845	u64 objectid;
3846	int err = 0;
3847
3848	root = btrfs_grab_root(fs_info->data_reloc_root);
 
 
 
3849	trans = btrfs_start_transaction(root, 6);
3850	if (IS_ERR(trans)) {
3851		btrfs_put_root(root);
3852		return ERR_CAST(trans);
3853	}
3854
3855	err = btrfs_get_free_objectid(root, &objectid);
3856	if (err)
3857		goto out;
3858
3859	err = __insert_orphan_inode(trans, root, objectid);
3860	if (err)
3861		goto out;
3862
3863	inode = btrfs_iget(fs_info->sb, objectid, root);
3864	if (IS_ERR(inode)) {
3865		delete_orphan_inode(trans, root, objectid);
3866		err = PTR_ERR(inode);
3867		inode = NULL;
3868		goto out;
3869	}
3870	BTRFS_I(inode)->index_cnt = group->start;
3871
3872	err = btrfs_orphan_add(trans, BTRFS_I(inode));
3873out:
3874	btrfs_put_root(root);
3875	btrfs_end_transaction(trans);
3876	btrfs_btree_balance_dirty(fs_info);
3877	if (err) {
3878		iput(inode);
 
3879		inode = ERR_PTR(err);
3880	}
3881	return inode;
3882}
3883
3884/*
3885 * Mark start of chunk relocation that is cancellable. Check if the cancellation
3886 * has been requested meanwhile and don't start in that case.
3887 *
3888 * Return:
3889 *   0             success
3890 *   -EINPROGRESS  operation is already in progress, that's probably a bug
3891 *   -ECANCELED    cancellation request was set before the operation started
3892 */
3893static int reloc_chunk_start(struct btrfs_fs_info *fs_info)
3894{
3895	if (test_and_set_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) {
3896		/* This should not happen */
3897		btrfs_err(fs_info, "reloc already running, cannot start");
3898		return -EINPROGRESS;
3899	}
3900
3901	if (atomic_read(&fs_info->reloc_cancel_req) > 0) {
3902		btrfs_info(fs_info, "chunk relocation canceled on start");
3903		/*
3904		 * On cancel, clear all requests but let the caller mark
3905		 * the end after cleanup operations.
3906		 */
3907		atomic_set(&fs_info->reloc_cancel_req, 0);
3908		return -ECANCELED;
3909	}
3910	return 0;
3911}
3912
3913/*
3914 * Mark end of chunk relocation that is cancellable and wake any waiters.
3915 */
3916static void reloc_chunk_end(struct btrfs_fs_info *fs_info)
3917{
3918	/* Requested after start, clear bit first so any waiters can continue */
3919	if (atomic_read(&fs_info->reloc_cancel_req) > 0)
3920		btrfs_info(fs_info, "chunk relocation canceled during operation");
3921	clear_and_wake_up_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags);
3922	atomic_set(&fs_info->reloc_cancel_req, 0);
3923}
3924
3925static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
3926{
3927	struct reloc_control *rc;
3928
3929	rc = kzalloc(sizeof(*rc), GFP_NOFS);
3930	if (!rc)
3931		return NULL;
3932
3933	INIT_LIST_HEAD(&rc->reloc_roots);
3934	INIT_LIST_HEAD(&rc->dirty_subvol_roots);
3935	btrfs_backref_init_cache(fs_info, &rc->backref_cache, 1);
3936	mapping_tree_init(&rc->reloc_root_tree);
3937	extent_io_tree_init(fs_info, &rc->processed_blocks, IO_TREE_RELOC_BLOCKS);
3938	return rc;
3939}
3940
3941static void free_reloc_control(struct reloc_control *rc)
3942{
3943	struct mapping_node *node, *tmp;
3944
3945	free_reloc_roots(&rc->reloc_roots);
3946	rbtree_postorder_for_each_entry_safe(node, tmp,
3947			&rc->reloc_root_tree.rb_root, rb_node)
3948		kfree(node);
3949
3950	kfree(rc);
3951}
3952
3953/*
3954 * Print the block group being relocated
3955 */
3956static void describe_relocation(struct btrfs_fs_info *fs_info,
3957				struct btrfs_block_group *block_group)
3958{
3959	char buf[128] = {'\0'};
 
3960
3961	btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3962
3963	btrfs_info(fs_info,
3964		   "relocating block group %llu flags %s",
3965		   block_group->start, buf);
3966}
3967
3968static const char *stage_to_string(int stage)
3969{
3970	if (stage == MOVE_DATA_EXTENTS)
3971		return "move data extents";
3972	if (stage == UPDATE_DATA_PTRS)
3973		return "update data pointers";
3974	return "unknown";
3975}
3976
3977/*
3978 * function to relocate all extents in a block group.
3979 */
3980int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
3981{
3982	struct btrfs_block_group *bg;
3983	struct btrfs_root *extent_root = btrfs_extent_root(fs_info, group_start);
3984	struct reloc_control *rc;
3985	struct inode *inode;
3986	struct btrfs_path *path;
3987	int ret;
3988	int rw = 0;
3989	int err = 0;
3990
3991	/*
3992	 * This only gets set if we had a half-deleted snapshot on mount.  We
3993	 * cannot allow relocation to start while we're still trying to clean up
3994	 * these pending deletions.
3995	 */
3996	ret = wait_on_bit(&fs_info->flags, BTRFS_FS_UNFINISHED_DROPS, TASK_INTERRUPTIBLE);
3997	if (ret)
3998		return ret;
3999
4000	/* We may have been woken up by close_ctree, so bail if we're closing. */
4001	if (btrfs_fs_closing(fs_info))
4002		return -EINTR;
4003
4004	bg = btrfs_lookup_block_group(fs_info, group_start);
4005	if (!bg)
4006		return -ENOENT;
4007
4008	/*
4009	 * Relocation of a data block group creates ordered extents.  Without
4010	 * sb_start_write(), we can freeze the filesystem while unfinished
4011	 * ordered extents are left. Such ordered extents can cause a deadlock
4012	 * e.g. when syncfs() is waiting for their completion but they can't
4013	 * finish because they block when joining a transaction, due to the
4014	 * fact that the freeze locks are being held in write mode.
4015	 */
4016	if (bg->flags & BTRFS_BLOCK_GROUP_DATA)
4017		ASSERT(sb_write_started(fs_info->sb));
4018
4019	if (btrfs_pinned_by_swapfile(fs_info, bg)) {
4020		btrfs_put_block_group(bg);
4021		return -ETXTBSY;
4022	}
4023
4024	rc = alloc_reloc_control(fs_info);
4025	if (!rc) {
4026		btrfs_put_block_group(bg);
4027		return -ENOMEM;
4028	}
4029
4030	ret = reloc_chunk_start(fs_info);
4031	if (ret < 0) {
4032		err = ret;
4033		goto out_put_bg;
4034	}
4035
4036	rc->extent_root = extent_root;
4037	rc->block_group = bg;
4038
4039	ret = btrfs_inc_block_group_ro(rc->block_group, true);
 
 
 
4040	if (ret) {
4041		err = ret;
4042		goto out;
4043	}
4044	rw = 1;
4045
4046	path = btrfs_alloc_path();
4047	if (!path) {
4048		err = -ENOMEM;
4049		goto out;
4050	}
4051
4052	inode = lookup_free_space_inode(rc->block_group, path);
4053	btrfs_free_path(path);
4054
4055	if (!IS_ERR(inode))
4056		ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4057	else
4058		ret = PTR_ERR(inode);
4059
4060	if (ret && ret != -ENOENT) {
4061		err = ret;
4062		goto out;
4063	}
4064
4065	rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4066	if (IS_ERR(rc->data_inode)) {
4067		err = PTR_ERR(rc->data_inode);
4068		rc->data_inode = NULL;
4069		goto out;
4070	}
4071
4072	describe_relocation(fs_info, rc->block_group);
4073
4074	btrfs_wait_block_group_reservations(rc->block_group);
4075	btrfs_wait_nocow_writers(rc->block_group);
4076	btrfs_wait_ordered_roots(fs_info, U64_MAX,
4077				 rc->block_group->start,
4078				 rc->block_group->length);
4079
4080	ret = btrfs_zone_finish(rc->block_group);
4081	WARN_ON(ret && ret != -EAGAIN);
4082
4083	while (1) {
4084		int 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(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 ret;
4176	int err = 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			err = ret;
4192			goto out;
4193		}
4194		if (ret > 0) {
4195			if (path->slots[0] == 0)
4196				break;
4197			path->slots[0]--;
4198		}
4199		leaf = path->nodes[0];
4200		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4201		btrfs_release_path(path);
4202
4203		if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4204		    key.type != BTRFS_ROOT_ITEM_KEY)
4205			break;
4206
4207		reloc_root = btrfs_read_tree_root(fs_info->tree_root, &key);
4208		if (IS_ERR(reloc_root)) {
4209			err = PTR_ERR(reloc_root);
4210			goto out;
4211		}
4212
4213		set_bit(BTRFS_ROOT_SHAREABLE, &reloc_root->state);
4214		list_add(&reloc_root->root_list, &reloc_roots);
4215
4216		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4217			fs_root = btrfs_get_fs_root(fs_info,
4218					reloc_root->root_key.offset, false);
4219			if (IS_ERR(fs_root)) {
4220				ret = PTR_ERR(fs_root);
4221				if (ret != -ENOENT) {
4222					err = ret;
4223					goto out;
4224				}
4225				ret = mark_garbage_root(reloc_root);
4226				if (ret < 0) {
4227					err = ret;
4228					goto out;
4229				}
4230			} else {
4231				btrfs_put_root(fs_root);
4232			}
4233		}
4234
4235		if (key.offset == 0)
4236			break;
4237
4238		key.offset--;
4239	}
4240	btrfs_release_path(path);
4241
4242	if (list_empty(&reloc_roots))
4243		goto out;
4244
4245	rc = alloc_reloc_control(fs_info);
4246	if (!rc) {
4247		err = -ENOMEM;
4248		goto out;
4249	}
4250
4251	ret = reloc_chunk_start(fs_info);
4252	if (ret < 0) {
4253		err = ret;
4254		goto out_end;
4255	}
4256
4257	rc->extent_root = btrfs_extent_root(fs_info, 0);
4258
4259	set_reloc_control(rc);
4260
4261	trans = btrfs_join_transaction(rc->extent_root);
4262	if (IS_ERR(trans)) {
 
4263		err = PTR_ERR(trans);
4264		goto out_unset;
4265	}
4266
4267	rc->merge_reloc_tree = 1;
4268
4269	while (!list_empty(&reloc_roots)) {
4270		reloc_root = list_entry(reloc_roots.next,
4271					struct btrfs_root, root_list);
4272		list_del(&reloc_root->root_list);
4273
4274		if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4275			list_add_tail(&reloc_root->root_list,
4276				      &rc->reloc_roots);
4277			continue;
4278		}
4279
4280		fs_root = btrfs_get_fs_root(fs_info, reloc_root->root_key.offset,
4281					    false);
4282		if (IS_ERR(fs_root)) {
4283			err = PTR_ERR(fs_root);
4284			list_add_tail(&reloc_root->root_list, &reloc_roots);
4285			btrfs_end_transaction(trans);
4286			goto out_unset;
4287		}
4288
4289		err = __add_reloc_root(reloc_root);
4290		ASSERT(err != -EEXIST);
4291		if (err) {
4292			list_add_tail(&reloc_root->root_list, &reloc_roots);
4293			btrfs_put_root(fs_root);
4294			btrfs_end_transaction(trans);
4295			goto out_unset;
4296		}
4297		fs_root->reloc_root = btrfs_grab_root(reloc_root);
4298		btrfs_put_root(fs_root);
4299	}
4300
4301	err = btrfs_commit_transaction(trans);
4302	if (err)
4303		goto out_unset;
4304
4305	merge_reloc_roots(rc);
4306
4307	unset_reloc_control(rc);
4308
4309	trans = btrfs_join_transaction(rc->extent_root);
4310	if (IS_ERR(trans)) {
4311		err = PTR_ERR(trans);
4312		goto out_clean;
4313	}
4314	err = btrfs_commit_transaction(trans);
4315out_clean:
4316	ret = clean_dirty_subvols(rc);
4317	if (ret < 0 && !err)
4318		err = ret;
4319out_unset:
4320	unset_reloc_control(rc);
4321out_end:
4322	reloc_chunk_end(fs_info);
4323	free_reloc_control(rc);
4324out:
4325	free_reloc_roots(&reloc_roots);
 
4326
4327	btrfs_free_path(path);
4328
4329	if (err == 0) {
4330		/* cleanup orphan inode in data relocation tree */
4331		fs_root = btrfs_grab_root(fs_info->data_reloc_root);
4332		ASSERT(fs_root);
4333		err = btrfs_orphan_cleanup(fs_root);
4334		btrfs_put_root(fs_root);
 
4335	}
4336	return err;
4337}
4338
4339/*
4340 * helper to add ordered checksum for data relocation.
4341 *
4342 * cloning checksum properly handles the nodatasum extents.
4343 * it also saves CPU time to re-calculate the checksum.
4344 */
4345int btrfs_reloc_clone_csums(struct btrfs_inode *inode, u64 file_pos, u64 len)
4346{
4347	struct btrfs_fs_info *fs_info = inode->root->fs_info;
4348	struct btrfs_root *csum_root;
4349	struct btrfs_ordered_sum *sums;
4350	struct btrfs_ordered_extent *ordered;
4351	int ret;
4352	u64 disk_bytenr;
4353	u64 new_bytenr;
4354	LIST_HEAD(list);
4355
4356	ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4357	BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len);
4358
4359	disk_bytenr = file_pos + inode->index_cnt;
4360	csum_root = btrfs_csum_root(fs_info, disk_bytenr);
4361	ret = btrfs_lookup_csums_list(csum_root, disk_bytenr,
4362				      disk_bytenr + len - 1, &list, 0, false);
4363	if (ret)
4364		goto out;
4365
4366	while (!list_empty(&list)) {
4367		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4368		list_del_init(&sums->list);
4369
4370		/*
4371		 * We need to offset the new_bytenr based on where the csum is.
4372		 * We need to do this because we will read in entire prealloc
4373		 * extents but we may have written to say the middle of the
4374		 * prealloc extent, so we need to make sure the csum goes with
4375		 * the right disk offset.
4376		 *
4377		 * We can do this because the data reloc inode refers strictly
4378		 * to the on disk bytes, so we don't have to worry about
4379		 * disk_len vs real len like with real inodes since it's all
4380		 * disk length.
4381		 */
4382		new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr;
4383		sums->bytenr = new_bytenr;
4384
4385		btrfs_add_ordered_sum(ordered, sums);
4386	}
4387out:
4388	btrfs_put_ordered_extent(ordered);
4389	return ret;
4390}
4391
4392int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4393			  struct btrfs_root *root, struct extent_buffer *buf,
4394			  struct extent_buffer *cow)
4395{
4396	struct btrfs_fs_info *fs_info = root->fs_info;
4397	struct reloc_control *rc;
4398	struct btrfs_backref_node *node;
4399	int first_cow = 0;
4400	int level;
4401	int ret = 0;
4402
4403	rc = fs_info->reloc_ctl;
4404	if (!rc)
4405		return 0;
4406
4407	BUG_ON(rc->stage == UPDATE_DATA_PTRS && btrfs_is_data_reloc_root(root));
 
 
 
 
 
 
4408
4409	level = btrfs_header_level(buf);
4410	if (btrfs_header_generation(buf) <=
4411	    btrfs_root_last_snapshot(&root->root_item))
4412		first_cow = 1;
4413
4414	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4415	    rc->create_reloc_tree) {
4416		WARN_ON(!first_cow && level == 0);
4417
4418		node = rc->backref_cache.path[level];
4419		BUG_ON(node->bytenr != buf->start &&
4420		       node->new_bytenr != buf->start);
4421
4422		btrfs_backref_drop_node_buffer(node);
4423		atomic_inc(&cow->refs);
4424		node->eb = cow;
4425		node->new_bytenr = cow->start;
4426
4427		if (!node->pending) {
4428			list_move_tail(&node->list,
4429				       &rc->backref_cache.pending[level]);
4430			node->pending = 1;
4431		}
4432
4433		if (first_cow)
4434			mark_block_processed(rc, node);
4435
4436		if (first_cow && level > 0)
4437			rc->nodes_relocated += buf->len;
4438	}
4439
4440	if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4441		ret = replace_file_extents(trans, rc, root, cow);
4442	return ret;
4443}
4444
4445/*
4446 * called before creating snapshot. it calculates metadata reservation
4447 * required for relocating tree blocks in the snapshot
4448 */
4449void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4450			      u64 *bytes_to_reserve)
4451{
4452	struct btrfs_root *root = pending->root;
4453	struct reloc_control *rc = root->fs_info->reloc_ctl;
4454
4455	if (!rc || !have_reloc_root(root))
 
4456		return;
4457
 
4458	if (!rc->merge_reloc_tree)
4459		return;
4460
4461	root = root->reloc_root;
4462	BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4463	/*
4464	 * relocation is in the stage of merging trees. the space
4465	 * used by merging a reloc tree is twice the size of
4466	 * relocated tree nodes in the worst case. half for cowing
4467	 * the reloc tree, half for cowing the fs tree. the space
4468	 * used by cowing the reloc tree will be freed after the
4469	 * tree is dropped. if we create snapshot, cowing the fs
4470	 * tree may use more space than it frees. so we need
4471	 * reserve extra space.
4472	 */
4473	*bytes_to_reserve += rc->nodes_relocated;
4474}
4475
4476/*
4477 * called after snapshot is created. migrate block reservation
4478 * and create reloc root for the newly created snapshot
4479 *
4480 * This is similar to btrfs_init_reloc_root(), we come out of here with two
4481 * references held on the reloc_root, one for root->reloc_root and one for
4482 * rc->reloc_roots.
4483 */
4484int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4485			       struct btrfs_pending_snapshot *pending)
4486{
4487	struct btrfs_root *root = pending->root;
4488	struct btrfs_root *reloc_root;
4489	struct btrfs_root *new_root;
4490	struct reloc_control *rc = root->fs_info->reloc_ctl;
4491	int ret;
4492
4493	if (!rc || !have_reloc_root(root))
4494		return 0;
4495
4496	rc = root->fs_info->reloc_ctl;
4497	rc->merging_rsv_size += rc->nodes_relocated;
4498
4499	if (rc->merge_reloc_tree) {
4500		ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4501					      rc->block_rsv,
4502					      rc->nodes_relocated, true);
4503		if (ret)
4504			return ret;
4505	}
4506
4507	new_root = pending->snap;
4508	reloc_root = create_reloc_root(trans, root->reloc_root,
4509				       new_root->root_key.objectid);
4510	if (IS_ERR(reloc_root))
4511		return PTR_ERR(reloc_root);
4512
4513	ret = __add_reloc_root(reloc_root);
4514	ASSERT(ret != -EEXIST);
4515	if (ret) {
4516		/* Pairs with create_reloc_root */
4517		btrfs_put_root(reloc_root);
4518		return ret;
4519	}
4520	new_root->reloc_root = btrfs_grab_root(reloc_root);
4521
4522	if (rc->create_reloc_tree)
4523		ret = clone_backref_node(trans, rc, root, reloc_root);
4524	return ret;
4525}
v4.17
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2009 Oracle.  All rights reserved.
   4 */
   5
   6#include <linux/sched.h>
   7#include <linux/pagemap.h>
   8#include <linux/writeback.h>
   9#include <linux/blkdev.h>
  10#include <linux/rbtree.h>
  11#include <linux/slab.h>
 
  12#include "ctree.h"
  13#include "disk-io.h"
  14#include "transaction.h"
  15#include "volumes.h"
  16#include "locking.h"
  17#include "btrfs_inode.h"
  18#include "async-thread.h"
  19#include "free-space-cache.h"
  20#include "inode-map.h"
  21#include "qgroup.h"
  22#include "print-tree.h"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  23
  24/*
  25 * backref_node, mapping_node and tree_block start with this
  26 */
  27struct tree_entry {
  28	struct rb_node rb_node;
  29	u64 bytenr;
  30};
  31
  32/*
  33 * present a tree block in the backref cache
  34 */
  35struct backref_node {
  36	struct rb_node rb_node;
  37	u64 bytenr;
  38
  39	u64 new_bytenr;
  40	/* objectid of tree block owner, can be not uptodate */
  41	u64 owner;
  42	/* link to pending, changed or detached list */
  43	struct list_head list;
  44	/* list of upper level blocks reference this block */
  45	struct list_head upper;
  46	/* list of child blocks in the cache */
  47	struct list_head lower;
  48	/* NULL if this node is not tree root */
  49	struct btrfs_root *root;
  50	/* extent buffer got by COW the block */
  51	struct extent_buffer *eb;
  52	/* level of tree block */
  53	unsigned int level:8;
  54	/* is the block in non-reference counted tree */
  55	unsigned int cowonly:1;
  56	/* 1 if no child node in the cache */
  57	unsigned int lowest:1;
  58	/* is the extent buffer locked */
  59	unsigned int locked:1;
  60	/* has the block been processed */
  61	unsigned int processed:1;
  62	/* have backrefs of this block been checked */
  63	unsigned int checked:1;
  64	/*
  65	 * 1 if corresponding block has been cowed but some upper
  66	 * level block pointers may not point to the new location
  67	 */
  68	unsigned int pending:1;
  69	/*
  70	 * 1 if the backref node isn't connected to any other
  71	 * backref node.
  72	 */
  73	unsigned int detached:1;
  74};
  75
  76/*
  77 * present a block pointer in the backref cache
  78 */
  79struct backref_edge {
  80	struct list_head list[2];
  81	struct backref_node *node[2];
  82};
  83
  84#define LOWER	0
  85#define UPPER	1
  86#define RELOCATION_RESERVED_NODES	256
  87
  88struct backref_cache {
  89	/* red black tree of all backref nodes in the cache */
  90	struct rb_root rb_root;
  91	/* for passing backref nodes to btrfs_reloc_cow_block */
  92	struct backref_node *path[BTRFS_MAX_LEVEL];
  93	/*
  94	 * list of blocks that have been cowed but some block
  95	 * pointers in upper level blocks may not reflect the
  96	 * new location
  97	 */
  98	struct list_head pending[BTRFS_MAX_LEVEL];
  99	/* list of backref nodes with no child node */
 100	struct list_head leaves;
 101	/* list of blocks that have been cowed in current transaction */
 102	struct list_head changed;
 103	/* list of detached backref node. */
 104	struct list_head detached;
 105
 106	u64 last_trans;
 107
 108	int nr_nodes;
 109	int nr_edges;
 110};
 111
 112/*
 113 * map address of tree root to tree
 114 */
 115struct mapping_node {
 116	struct rb_node rb_node;
 117	u64 bytenr;
 
 
 118	void *data;
 119};
 120
 121struct mapping_tree {
 122	struct rb_root rb_root;
 123	spinlock_t lock;
 124};
 125
 126/*
 127 * present a tree block to process
 128 */
 129struct tree_block {
 130	struct rb_node rb_node;
 131	u64 bytenr;
 
 
 
 132	struct btrfs_key key;
 133	unsigned int level:8;
 134	unsigned int key_ready:1;
 135};
 136
 137#define MAX_EXTENTS 128
 138
 139struct file_extent_cluster {
 140	u64 start;
 141	u64 end;
 142	u64 boundary[MAX_EXTENTS];
 143	unsigned int nr;
 144};
 145
 146struct reloc_control {
 147	/* block group to relocate */
 148	struct btrfs_block_group_cache *block_group;
 149	/* extent tree */
 150	struct btrfs_root *extent_root;
 151	/* inode for moving data */
 152	struct inode *data_inode;
 153
 154	struct btrfs_block_rsv *block_rsv;
 155
 156	struct backref_cache backref_cache;
 157
 158	struct file_extent_cluster cluster;
 159	/* tree blocks have been processed */
 160	struct extent_io_tree processed_blocks;
 161	/* map start of tree root to corresponding reloc tree */
 162	struct mapping_tree reloc_root_tree;
 163	/* list of reloc trees */
 164	struct list_head reloc_roots;
 
 
 165	/* size of metadata reservation for merging reloc trees */
 166	u64 merging_rsv_size;
 167	/* size of relocated tree nodes */
 168	u64 nodes_relocated;
 169	/* reserved size for block group relocation*/
 170	u64 reserved_bytes;
 171
 172	u64 search_start;
 173	u64 extents_found;
 174
 175	unsigned int stage:8;
 176	unsigned int create_reloc_tree:1;
 177	unsigned int merge_reloc_tree:1;
 178	unsigned int found_file_extent:1;
 179};
 180
 181/* stages of data relocation */
 182#define MOVE_DATA_EXTENTS	0
 183#define UPDATE_DATA_PTRS	1
 184
 185static void remove_backref_node(struct backref_cache *cache,
 186				struct backref_node *node);
 187static void __mark_block_processed(struct reloc_control *rc,
 188				   struct backref_node *node);
 189
 190static void mapping_tree_init(struct mapping_tree *tree)
 191{
 192	tree->rb_root = RB_ROOT;
 193	spin_lock_init(&tree->lock);
 194}
 195
 196static void backref_cache_init(struct backref_cache *cache)
 197{
 198	int i;
 199	cache->rb_root = RB_ROOT;
 200	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
 201		INIT_LIST_HEAD(&cache->pending[i]);
 202	INIT_LIST_HEAD(&cache->changed);
 203	INIT_LIST_HEAD(&cache->detached);
 204	INIT_LIST_HEAD(&cache->leaves);
 205}
 206
 207static void backref_cache_cleanup(struct backref_cache *cache)
 208{
 209	struct backref_node *node;
 210	int i;
 211
 212	while (!list_empty(&cache->detached)) {
 213		node = list_entry(cache->detached.next,
 214				  struct backref_node, list);
 215		remove_backref_node(cache, node);
 216	}
 217
 218	while (!list_empty(&cache->leaves)) {
 219		node = list_entry(cache->leaves.next,
 220				  struct backref_node, lower);
 221		remove_backref_node(cache, node);
 222	}
 223
 224	cache->last_trans = 0;
 225
 226	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
 227		ASSERT(list_empty(&cache->pending[i]));
 228	ASSERT(list_empty(&cache->changed));
 229	ASSERT(list_empty(&cache->detached));
 230	ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
 231	ASSERT(!cache->nr_nodes);
 232	ASSERT(!cache->nr_edges);
 233}
 234
 235static struct backref_node *alloc_backref_node(struct backref_cache *cache)
 236{
 237	struct backref_node *node;
 238
 239	node = kzalloc(sizeof(*node), GFP_NOFS);
 240	if (node) {
 241		INIT_LIST_HEAD(&node->list);
 242		INIT_LIST_HEAD(&node->upper);
 243		INIT_LIST_HEAD(&node->lower);
 244		RB_CLEAR_NODE(&node->rb_node);
 245		cache->nr_nodes++;
 246	}
 247	return node;
 248}
 249
 250static void free_backref_node(struct backref_cache *cache,
 251			      struct backref_node *node)
 252{
 253	if (node) {
 254		cache->nr_nodes--;
 255		kfree(node);
 256	}
 257}
 258
 259static struct backref_edge *alloc_backref_edge(struct backref_cache *cache)
 260{
 261	struct backref_edge *edge;
 262
 263	edge = kzalloc(sizeof(*edge), GFP_NOFS);
 264	if (edge)
 265		cache->nr_edges++;
 266	return edge;
 267}
 268
 269static void free_backref_edge(struct backref_cache *cache,
 270			      struct backref_edge *edge)
 271{
 272	if (edge) {
 273		cache->nr_edges--;
 274		kfree(edge);
 275	}
 276}
 277
 278static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr,
 279				   struct rb_node *node)
 280{
 281	struct rb_node **p = &root->rb_node;
 282	struct rb_node *parent = NULL;
 283	struct tree_entry *entry;
 284
 285	while (*p) {
 286		parent = *p;
 287		entry = rb_entry(parent, struct tree_entry, rb_node);
 288
 289		if (bytenr < entry->bytenr)
 290			p = &(*p)->rb_left;
 291		else if (bytenr > entry->bytenr)
 292			p = &(*p)->rb_right;
 293		else
 294			return parent;
 295	}
 296
 297	rb_link_node(node, parent, p);
 298	rb_insert_color(node, root);
 299	return NULL;
 300}
 301
 302static struct rb_node *tree_search(struct rb_root *root, u64 bytenr)
 303{
 304	struct rb_node *n = root->rb_node;
 305	struct tree_entry *entry;
 306
 307	while (n) {
 308		entry = rb_entry(n, struct tree_entry, rb_node);
 309
 310		if (bytenr < entry->bytenr)
 311			n = n->rb_left;
 312		else if (bytenr > entry->bytenr)
 313			n = n->rb_right;
 314		else
 315			return n;
 316	}
 317	return NULL;
 318}
 319
 320static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr)
 321{
 322
 323	struct btrfs_fs_info *fs_info = NULL;
 324	struct backref_node *bnode = rb_entry(rb_node, struct backref_node,
 325					      rb_node);
 326	if (bnode->root)
 327		fs_info = bnode->root->fs_info;
 328	btrfs_panic(fs_info, errno,
 329		    "Inconsistency in backref cache found at offset %llu",
 330		    bytenr);
 331}
 332
 333/*
 334 * walk up backref nodes until reach node presents tree root
 335 */
 336static struct backref_node *walk_up_backref(struct backref_node *node,
 337					    struct backref_edge *edges[],
 338					    int *index)
 339{
 340	struct backref_edge *edge;
 341	int idx = *index;
 342
 343	while (!list_empty(&node->upper)) {
 344		edge = list_entry(node->upper.next,
 345				  struct backref_edge, list[LOWER]);
 346		edges[idx++] = edge;
 347		node = edge->node[UPPER];
 348	}
 349	BUG_ON(node->detached);
 350	*index = idx;
 351	return node;
 352}
 353
 354/*
 355 * walk down backref nodes to find start of next reference path
 356 */
 357static struct backref_node *walk_down_backref(struct backref_edge *edges[],
 358					      int *index)
 359{
 360	struct backref_edge *edge;
 361	struct backref_node *lower;
 362	int idx = *index;
 363
 364	while (idx > 0) {
 365		edge = edges[idx - 1];
 366		lower = edge->node[LOWER];
 367		if (list_is_last(&edge->list[LOWER], &lower->upper)) {
 368			idx--;
 369			continue;
 370		}
 371		edge = list_entry(edge->list[LOWER].next,
 372				  struct backref_edge, list[LOWER]);
 373		edges[idx - 1] = edge;
 374		*index = idx;
 375		return edge->node[UPPER];
 376	}
 377	*index = 0;
 378	return NULL;
 379}
 380
 381static void unlock_node_buffer(struct backref_node *node)
 382{
 383	if (node->locked) {
 384		btrfs_tree_unlock(node->eb);
 385		node->locked = 0;
 386	}
 387}
 388
 389static void drop_node_buffer(struct backref_node *node)
 390{
 391	if (node->eb) {
 392		unlock_node_buffer(node);
 393		free_extent_buffer(node->eb);
 394		node->eb = NULL;
 395	}
 396}
 397
 398static void drop_backref_node(struct backref_cache *tree,
 399			      struct backref_node *node)
 400{
 401	BUG_ON(!list_empty(&node->upper));
 402
 403	drop_node_buffer(node);
 404	list_del(&node->list);
 405	list_del(&node->lower);
 406	if (!RB_EMPTY_NODE(&node->rb_node))
 407		rb_erase(&node->rb_node, &tree->rb_root);
 408	free_backref_node(tree, node);
 409}
 410
 411/*
 412 * remove a backref node from the backref cache
 413 */
 414static void remove_backref_node(struct backref_cache *cache,
 415				struct backref_node *node)
 416{
 417	struct backref_node *upper;
 418	struct backref_edge *edge;
 419
 420	if (!node)
 421		return;
 422
 423	BUG_ON(!node->lowest && !node->detached);
 424	while (!list_empty(&node->upper)) {
 425		edge = list_entry(node->upper.next, struct backref_edge,
 426				  list[LOWER]);
 427		upper = edge->node[UPPER];
 428		list_del(&edge->list[LOWER]);
 429		list_del(&edge->list[UPPER]);
 430		free_backref_edge(cache, edge);
 431
 432		if (RB_EMPTY_NODE(&upper->rb_node)) {
 433			BUG_ON(!list_empty(&node->upper));
 434			drop_backref_node(cache, node);
 435			node = upper;
 436			node->lowest = 1;
 437			continue;
 438		}
 439		/*
 440		 * add the node to leaf node list if no other
 441		 * child block cached.
 442		 */
 443		if (list_empty(&upper->lower)) {
 444			list_add_tail(&upper->lower, &cache->leaves);
 445			upper->lowest = 1;
 446		}
 447	}
 448
 449	drop_backref_node(cache, node);
 450}
 451
 452static void update_backref_node(struct backref_cache *cache,
 453				struct backref_node *node, u64 bytenr)
 454{
 455	struct rb_node *rb_node;
 456	rb_erase(&node->rb_node, &cache->rb_root);
 457	node->bytenr = bytenr;
 458	rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node);
 459	if (rb_node)
 460		backref_tree_panic(rb_node, -EEXIST, bytenr);
 461}
 462
 463/*
 464 * update backref cache after a transaction commit
 465 */
 466static int update_backref_cache(struct btrfs_trans_handle *trans,
 467				struct backref_cache *cache)
 468{
 469	struct backref_node *node;
 470	int level = 0;
 471
 472	if (cache->last_trans == 0) {
 473		cache->last_trans = trans->transid;
 474		return 0;
 475	}
 476
 477	if (cache->last_trans == trans->transid)
 478		return 0;
 479
 480	/*
 481	 * detached nodes are used to avoid unnecessary backref
 482	 * lookup. transaction commit changes the extent tree.
 483	 * so the detached nodes are no longer useful.
 484	 */
 485	while (!list_empty(&cache->detached)) {
 486		node = list_entry(cache->detached.next,
 487				  struct backref_node, list);
 488		remove_backref_node(cache, node);
 489	}
 490
 491	while (!list_empty(&cache->changed)) {
 492		node = list_entry(cache->changed.next,
 493				  struct backref_node, list);
 494		list_del_init(&node->list);
 495		BUG_ON(node->pending);
 496		update_backref_node(cache, node, node->new_bytenr);
 497	}
 498
 499	/*
 500	 * some nodes can be left in the pending list if there were
 501	 * errors during processing the pending nodes.
 502	 */
 503	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
 504		list_for_each_entry(node, &cache->pending[level], list) {
 505			BUG_ON(!node->pending);
 506			if (node->bytenr == node->new_bytenr)
 507				continue;
 508			update_backref_node(cache, node, node->new_bytenr);
 509		}
 510	}
 511
 512	cache->last_trans = 0;
 513	return 1;
 514}
 515
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 516
 517static int should_ignore_root(struct btrfs_root *root)
 518{
 519	struct btrfs_root *reloc_root;
 520
 521	if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 522		return 0;
 523
 
 
 
 
 524	reloc_root = root->reloc_root;
 525	if (!reloc_root)
 526		return 0;
 527
 528	if (btrfs_root_last_snapshot(&reloc_root->root_item) ==
 529	    root->fs_info->running_transaction->transid - 1)
 530		return 0;
 531	/*
 532	 * if there is reloc tree and it was created in previous
 533	 * transaction backref lookup can find the reloc tree,
 534	 * so backref node for the fs tree root is useless for
 535	 * relocation.
 536	 */
 537	return 1;
 538}
 
 539/*
 540 * find reloc tree by address of tree root
 541 */
 542static struct btrfs_root *find_reloc_root(struct reloc_control *rc,
 543					  u64 bytenr)
 544{
 
 545	struct rb_node *rb_node;
 546	struct mapping_node *node;
 547	struct btrfs_root *root = NULL;
 548
 
 549	spin_lock(&rc->reloc_root_tree.lock);
 550	rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr);
 551	if (rb_node) {
 552		node = rb_entry(rb_node, struct mapping_node, rb_node);
 553		root = (struct btrfs_root *)node->data;
 554	}
 555	spin_unlock(&rc->reloc_root_tree.lock);
 556	return root;
 557}
 558
 559static int is_cowonly_root(u64 root_objectid)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 560{
 561	if (root_objectid == BTRFS_ROOT_TREE_OBJECTID ||
 562	    root_objectid == BTRFS_EXTENT_TREE_OBJECTID ||
 563	    root_objectid == BTRFS_CHUNK_TREE_OBJECTID ||
 564	    root_objectid == BTRFS_DEV_TREE_OBJECTID ||
 565	    root_objectid == BTRFS_TREE_LOG_OBJECTID ||
 566	    root_objectid == BTRFS_CSUM_TREE_OBJECTID ||
 567	    root_objectid == BTRFS_UUID_TREE_OBJECTID ||
 568	    root_objectid == BTRFS_QUOTA_TREE_OBJECTID ||
 569	    root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
 570		return 1;
 571	return 0;
 572}
 573
 574static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info,
 575					u64 root_objectid)
 576{
 577	struct btrfs_key key;
 578
 579	key.objectid = root_objectid;
 580	key.type = BTRFS_ROOT_ITEM_KEY;
 581	if (is_cowonly_root(root_objectid))
 582		key.offset = 0;
 583	else
 584		key.offset = (u64)-1;
 585
 586	return btrfs_get_fs_root(fs_info, &key, false);
 587}
 588
 589#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 590static noinline_for_stack
 591struct btrfs_root *find_tree_root(struct reloc_control *rc,
 592				  struct extent_buffer *leaf,
 593				  struct btrfs_extent_ref_v0 *ref0)
 594{
 595	struct btrfs_root *root;
 596	u64 root_objectid = btrfs_ref_root_v0(leaf, ref0);
 597	u64 generation = btrfs_ref_generation_v0(leaf, ref0);
 598
 599	BUG_ON(root_objectid == BTRFS_TREE_RELOC_OBJECTID);
 
 
 
 
 600
 601	root = read_fs_root(rc->extent_root->fs_info, root_objectid);
 602	BUG_ON(IS_ERR(root));
 
 
 603
 604	if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) &&
 605	    generation != btrfs_root_generation(&root->root_item))
 606		return NULL;
 
 
 
 607
 608	return root;
 609}
 610#endif
 
 
 
 611
 612static noinline_for_stack
 613int find_inline_backref(struct extent_buffer *leaf, int slot,
 614			unsigned long *ptr, unsigned long *end)
 615{
 616	struct btrfs_key key;
 617	struct btrfs_extent_item *ei;
 618	struct btrfs_tree_block_info *bi;
 619	u32 item_size;
 620
 621	btrfs_item_key_to_cpu(leaf, &key, slot);
 622
 623	item_size = btrfs_item_size_nr(leaf, slot);
 624#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 625	if (item_size < sizeof(*ei)) {
 626		WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0));
 627		return 1;
 628	}
 629#endif
 630	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 631	WARN_ON(!(btrfs_extent_flags(leaf, ei) &
 632		  BTRFS_EXTENT_FLAG_TREE_BLOCK));
 633
 634	if (key.type == BTRFS_EXTENT_ITEM_KEY &&
 635	    item_size <= sizeof(*ei) + sizeof(*bi)) {
 636		WARN_ON(item_size < sizeof(*ei) + sizeof(*bi));
 637		return 1;
 638	}
 639	if (key.type == BTRFS_METADATA_ITEM_KEY &&
 640	    item_size <= sizeof(*ei)) {
 641		WARN_ON(item_size < sizeof(*ei));
 642		return 1;
 643	}
 644
 645	if (key.type == BTRFS_EXTENT_ITEM_KEY) {
 646		bi = (struct btrfs_tree_block_info *)(ei + 1);
 647		*ptr = (unsigned long)(bi + 1);
 648	} else {
 649		*ptr = (unsigned long)(ei + 1);
 650	}
 651	*end = (unsigned long)ei + item_size;
 652	return 0;
 653}
 654
 655/*
 656 * build backref tree for a given tree block. root of the backref tree
 657 * corresponds the tree block, leaves of the backref tree correspond
 658 * roots of b-trees that reference the tree block.
 659 *
 660 * the basic idea of this function is check backrefs of a given block
 661 * to find upper level blocks that reference the block, and then check
 662 * backrefs of these upper level blocks recursively. the recursion stop
 663 * when tree root is reached or backrefs for the block is cached.
 664 *
 665 * NOTE: if we find backrefs for a block are cached, we know backrefs
 666 * for all upper level blocks that directly/indirectly reference the
 667 * block are also cached.
 668 */
 669static noinline_for_stack
 670struct backref_node *build_backref_tree(struct reloc_control *rc,
 671					struct btrfs_key *node_key,
 672					int level, u64 bytenr)
 673{
 674	struct backref_cache *cache = &rc->backref_cache;
 675	struct btrfs_path *path1;
 676	struct btrfs_path *path2;
 677	struct extent_buffer *eb;
 678	struct btrfs_root *root;
 679	struct backref_node *cur;
 680	struct backref_node *upper;
 681	struct backref_node *lower;
 682	struct backref_node *node = NULL;
 683	struct backref_node *exist = NULL;
 684	struct backref_edge *edge;
 685	struct rb_node *rb_node;
 686	struct btrfs_key key;
 687	unsigned long end;
 688	unsigned long ptr;
 689	LIST_HEAD(list);
 690	LIST_HEAD(useless);
 691	int cowonly;
 692	int ret;
 693	int err = 0;
 694	bool need_check = true;
 695
 696	path1 = btrfs_alloc_path();
 697	path2 = btrfs_alloc_path();
 698	if (!path1 || !path2) {
 
 
 699		err = -ENOMEM;
 700		goto out;
 701	}
 702	path1->reada = READA_FORWARD;
 703	path2->reada = READA_FORWARD;
 704
 705	node = alloc_backref_node(cache);
 706	if (!node) {
 707		err = -ENOMEM;
 708		goto out;
 709	}
 710
 711	node->bytenr = bytenr;
 712	node->level = level;
 713	node->lowest = 1;
 714	cur = node;
 715again:
 716	end = 0;
 717	ptr = 0;
 718	key.objectid = cur->bytenr;
 719	key.type = BTRFS_METADATA_ITEM_KEY;
 720	key.offset = (u64)-1;
 721
 722	path1->search_commit_root = 1;
 723	path1->skip_locking = 1;
 724	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1,
 725				0, 0);
 726	if (ret < 0) {
 727		err = ret;
 728		goto out;
 729	}
 730	ASSERT(ret);
 731	ASSERT(path1->slots[0]);
 732
 733	path1->slots[0]--;
 734
 735	WARN_ON(cur->checked);
 736	if (!list_empty(&cur->upper)) {
 737		/*
 738		 * the backref was added previously when processing
 739		 * backref of type BTRFS_TREE_BLOCK_REF_KEY
 740		 */
 741		ASSERT(list_is_singular(&cur->upper));
 742		edge = list_entry(cur->upper.next, struct backref_edge,
 743				  list[LOWER]);
 744		ASSERT(list_empty(&edge->list[UPPER]));
 745		exist = edge->node[UPPER];
 746		/*
 747		 * add the upper level block to pending list if we need
 748		 * check its backrefs
 749		 */
 750		if (!exist->checked)
 751			list_add_tail(&edge->list[UPPER], &list);
 752	} else {
 753		exist = NULL;
 754	}
 755
 756	while (1) {
 757		cond_resched();
 758		eb = path1->nodes[0];
 759
 760		if (ptr >= end) {
 761			if (path1->slots[0] >= btrfs_header_nritems(eb)) {
 762				ret = btrfs_next_leaf(rc->extent_root, path1);
 763				if (ret < 0) {
 764					err = ret;
 765					goto out;
 766				}
 767				if (ret > 0)
 768					break;
 769				eb = path1->nodes[0];
 770			}
 771
 772			btrfs_item_key_to_cpu(eb, &key, path1->slots[0]);
 773			if (key.objectid != cur->bytenr) {
 774				WARN_ON(exist);
 775				break;
 776			}
 777
 778			if (key.type == BTRFS_EXTENT_ITEM_KEY ||
 779			    key.type == BTRFS_METADATA_ITEM_KEY) {
 780				ret = find_inline_backref(eb, path1->slots[0],
 781							  &ptr, &end);
 782				if (ret)
 783					goto next;
 784			}
 785		}
 786
 787		if (ptr < end) {
 788			/* update key for inline back ref */
 789			struct btrfs_extent_inline_ref *iref;
 790			int type;
 791			iref = (struct btrfs_extent_inline_ref *)ptr;
 792			type = btrfs_get_extent_inline_ref_type(eb, iref,
 793							BTRFS_REF_TYPE_BLOCK);
 794			if (type == BTRFS_REF_TYPE_INVALID) {
 795				err = -EINVAL;
 796				goto out;
 797			}
 798			key.type = type;
 799			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
 800
 801			WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY &&
 802				key.type != BTRFS_SHARED_BLOCK_REF_KEY);
 803		}
 804
 805		if (exist &&
 806		    ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
 807		      exist->owner == key.offset) ||
 808		     (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
 809		      exist->bytenr == key.offset))) {
 810			exist = NULL;
 811			goto next;
 812		}
 813
 814#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
 815		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY ||
 816		    key.type == BTRFS_EXTENT_REF_V0_KEY) {
 817			if (key.type == BTRFS_EXTENT_REF_V0_KEY) {
 818				struct btrfs_extent_ref_v0 *ref0;
 819				ref0 = btrfs_item_ptr(eb, path1->slots[0],
 820						struct btrfs_extent_ref_v0);
 821				if (key.objectid == key.offset) {
 822					root = find_tree_root(rc, eb, ref0);
 823					if (root && !should_ignore_root(root))
 824						cur->root = root;
 825					else
 826						list_add(&cur->list, &useless);
 827					break;
 828				}
 829				if (is_cowonly_root(btrfs_ref_root_v0(eb,
 830								      ref0)))
 831					cur->cowonly = 1;
 832			}
 833#else
 834		ASSERT(key.type != BTRFS_EXTENT_REF_V0_KEY);
 835		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
 836#endif
 837			if (key.objectid == key.offset) {
 838				/*
 839				 * only root blocks of reloc trees use
 840				 * backref of this type.
 841				 */
 842				root = find_reloc_root(rc, cur->bytenr);
 843				ASSERT(root);
 844				cur->root = root;
 845				break;
 846			}
 847
 848			edge = alloc_backref_edge(cache);
 849			if (!edge) {
 850				err = -ENOMEM;
 851				goto out;
 852			}
 853			rb_node = tree_search(&cache->rb_root, key.offset);
 854			if (!rb_node) {
 855				upper = alloc_backref_node(cache);
 856				if (!upper) {
 857					free_backref_edge(cache, edge);
 858					err = -ENOMEM;
 859					goto out;
 860				}
 861				upper->bytenr = key.offset;
 862				upper->level = cur->level + 1;
 863				/*
 864				 *  backrefs for the upper level block isn't
 865				 *  cached, add the block to pending list
 866				 */
 867				list_add_tail(&edge->list[UPPER], &list);
 868			} else {
 869				upper = rb_entry(rb_node, struct backref_node,
 870						 rb_node);
 871				ASSERT(upper->checked);
 872				INIT_LIST_HEAD(&edge->list[UPPER]);
 873			}
 874			list_add_tail(&edge->list[LOWER], &cur->upper);
 875			edge->node[LOWER] = cur;
 876			edge->node[UPPER] = upper;
 877
 878			goto next;
 879		} else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
 880			goto next;
 881		}
 882
 883		/* key.type == BTRFS_TREE_BLOCK_REF_KEY */
 884		root = read_fs_root(rc->extent_root->fs_info, key.offset);
 885		if (IS_ERR(root)) {
 886			err = PTR_ERR(root);
 887			goto out;
 888		}
 889
 890		if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 891			cur->cowonly = 1;
 892
 893		if (btrfs_root_level(&root->root_item) == cur->level) {
 894			/* tree root */
 895			ASSERT(btrfs_root_bytenr(&root->root_item) ==
 896			       cur->bytenr);
 897			if (should_ignore_root(root))
 898				list_add(&cur->list, &useless);
 899			else
 900				cur->root = root;
 901			break;
 902		}
 903
 904		level = cur->level + 1;
 905
 906		/*
 907		 * searching the tree to find upper level blocks
 908		 * reference the block.
 909		 */
 910		path2->search_commit_root = 1;
 911		path2->skip_locking = 1;
 912		path2->lowest_level = level;
 913		ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0);
 914		path2->lowest_level = 0;
 915		if (ret < 0) {
 916			err = ret;
 917			goto out;
 918		}
 919		if (ret > 0 && path2->slots[level] > 0)
 920			path2->slots[level]--;
 921
 922		eb = path2->nodes[level];
 923		if (btrfs_node_blockptr(eb, path2->slots[level]) !=
 924		    cur->bytenr) {
 925			btrfs_err(root->fs_info,
 926	"couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
 927				  cur->bytenr, level - 1, root->objectid,
 928				  node_key->objectid, node_key->type,
 929				  node_key->offset);
 930			err = -ENOENT;
 931			goto out;
 932		}
 933		lower = cur;
 934		need_check = true;
 935		for (; level < BTRFS_MAX_LEVEL; level++) {
 936			if (!path2->nodes[level]) {
 937				ASSERT(btrfs_root_bytenr(&root->root_item) ==
 938				       lower->bytenr);
 939				if (should_ignore_root(root))
 940					list_add(&lower->list, &useless);
 941				else
 942					lower->root = root;
 943				break;
 944			}
 945
 946			edge = alloc_backref_edge(cache);
 947			if (!edge) {
 948				err = -ENOMEM;
 949				goto out;
 950			}
 951
 952			eb = path2->nodes[level];
 953			rb_node = tree_search(&cache->rb_root, eb->start);
 954			if (!rb_node) {
 955				upper = alloc_backref_node(cache);
 956				if (!upper) {
 957					free_backref_edge(cache, edge);
 958					err = -ENOMEM;
 959					goto out;
 960				}
 961				upper->bytenr = eb->start;
 962				upper->owner = btrfs_header_owner(eb);
 963				upper->level = lower->level + 1;
 964				if (!test_bit(BTRFS_ROOT_REF_COWS,
 965					      &root->state))
 966					upper->cowonly = 1;
 967
 968				/*
 969				 * if we know the block isn't shared
 970				 * we can void checking its backrefs.
 971				 */
 972				if (btrfs_block_can_be_shared(root, eb))
 973					upper->checked = 0;
 974				else
 975					upper->checked = 1;
 976
 977				/*
 978				 * add the block to pending list if we
 979				 * need check its backrefs, we only do this once
 980				 * while walking up a tree as we will catch
 981				 * anything else later on.
 982				 */
 983				if (!upper->checked && need_check) {
 984					need_check = false;
 985					list_add_tail(&edge->list[UPPER],
 986						      &list);
 987				} else {
 988					if (upper->checked)
 989						need_check = true;
 990					INIT_LIST_HEAD(&edge->list[UPPER]);
 991				}
 992			} else {
 993				upper = rb_entry(rb_node, struct backref_node,
 994						 rb_node);
 995				ASSERT(upper->checked);
 996				INIT_LIST_HEAD(&edge->list[UPPER]);
 997				if (!upper->owner)
 998					upper->owner = btrfs_header_owner(eb);
 999			}
1000			list_add_tail(&edge->list[LOWER], &lower->upper);
1001			edge->node[LOWER] = lower;
1002			edge->node[UPPER] = upper;
1003
1004			if (rb_node)
1005				break;
1006			lower = upper;
1007			upper = NULL;
1008		}
1009		btrfs_release_path(path2);
1010next:
1011		if (ptr < end) {
1012			ptr += btrfs_extent_inline_ref_size(key.type);
1013			if (ptr >= end) {
1014				WARN_ON(ptr > end);
1015				ptr = 0;
1016				end = 0;
1017			}
1018		}
1019		if (ptr >= end)
1020			path1->slots[0]++;
1021	}
1022	btrfs_release_path(path1);
1023
1024	cur->checked = 1;
1025	WARN_ON(exist);
1026
1027	/* the pending list isn't empty, take the first block to process */
1028	if (!list_empty(&list)) {
1029		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1030		list_del_init(&edge->list[UPPER]);
1031		cur = edge->node[UPPER];
1032		goto again;
1033	}
1034
1035	/*
1036	 * everything goes well, connect backref nodes and insert backref nodes
1037	 * into the cache.
1038	 */
1039	ASSERT(node->checked);
1040	cowonly = node->cowonly;
1041	if (!cowonly) {
1042		rb_node = tree_insert(&cache->rb_root, node->bytenr,
1043				      &node->rb_node);
1044		if (rb_node)
1045			backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1046		list_add_tail(&node->lower, &cache->leaves);
1047	}
1048
1049	list_for_each_entry(edge, &node->upper, list[LOWER])
1050		list_add_tail(&edge->list[UPPER], &list);
1051
1052	while (!list_empty(&list)) {
1053		edge = list_entry(list.next, struct backref_edge, list[UPPER]);
1054		list_del_init(&edge->list[UPPER]);
1055		upper = edge->node[UPPER];
1056		if (upper->detached) {
1057			list_del(&edge->list[LOWER]);
1058			lower = edge->node[LOWER];
1059			free_backref_edge(cache, edge);
1060			if (list_empty(&lower->upper))
1061				list_add(&lower->list, &useless);
1062			continue;
1063		}
1064
1065		if (!RB_EMPTY_NODE(&upper->rb_node)) {
1066			if (upper->lowest) {
1067				list_del_init(&upper->lower);
1068				upper->lowest = 0;
1069			}
1070
1071			list_add_tail(&edge->list[UPPER], &upper->lower);
1072			continue;
1073		}
1074
1075		if (!upper->checked) {
1076			/*
1077			 * Still want to blow up for developers since this is a
1078			 * logic bug.
1079			 */
1080			ASSERT(0);
1081			err = -EINVAL;
1082			goto out;
1083		}
1084		if (cowonly != upper->cowonly) {
1085			ASSERT(0);
1086			err = -EINVAL;
1087			goto out;
1088		}
1089
1090		if (!cowonly) {
1091			rb_node = tree_insert(&cache->rb_root, upper->bytenr,
1092					      &upper->rb_node);
1093			if (rb_node)
1094				backref_tree_panic(rb_node, -EEXIST,
1095						   upper->bytenr);
1096		}
1097
1098		list_add_tail(&edge->list[UPPER], &upper->lower);
1099
1100		list_for_each_entry(edge, &upper->upper, list[LOWER])
1101			list_add_tail(&edge->list[UPPER], &list);
1102	}
1103	/*
1104	 * process useless backref nodes. backref nodes for tree leaves
1105	 * are deleted from the cache. backref nodes for upper level
1106	 * tree blocks are left in the cache to avoid unnecessary backref
1107	 * lookup.
1108	 */
1109	while (!list_empty(&useless)) {
1110		upper = list_entry(useless.next, struct backref_node, list);
1111		list_del_init(&upper->list);
1112		ASSERT(list_empty(&upper->upper));
1113		if (upper == node)
1114			node = NULL;
1115		if (upper->lowest) {
1116			list_del_init(&upper->lower);
1117			upper->lowest = 0;
1118		}
1119		while (!list_empty(&upper->lower)) {
1120			edge = list_entry(upper->lower.next,
1121					  struct backref_edge, list[UPPER]);
1122			list_del(&edge->list[UPPER]);
1123			list_del(&edge->list[LOWER]);
1124			lower = edge->node[LOWER];
1125			free_backref_edge(cache, edge);
1126
1127			if (list_empty(&lower->upper))
1128				list_add(&lower->list, &useless);
1129		}
1130		__mark_block_processed(rc, upper);
1131		if (upper->level > 0) {
1132			list_add(&upper->list, &cache->detached);
1133			upper->detached = 1;
1134		} else {
1135			rb_erase(&upper->rb_node, &cache->rb_root);
1136			free_backref_node(cache, upper);
1137		}
1138	}
1139out:
1140	btrfs_free_path(path1);
1141	btrfs_free_path(path2);
1142	if (err) {
1143		while (!list_empty(&useless)) {
1144			lower = list_entry(useless.next,
1145					   struct backref_node, list);
1146			list_del_init(&lower->list);
1147		}
1148		while (!list_empty(&list)) {
1149			edge = list_first_entry(&list, struct backref_edge,
1150						list[UPPER]);
1151			list_del(&edge->list[UPPER]);
1152			list_del(&edge->list[LOWER]);
1153			lower = edge->node[LOWER];
1154			upper = edge->node[UPPER];
1155			free_backref_edge(cache, edge);
1156
1157			/*
1158			 * Lower is no longer linked to any upper backref nodes
1159			 * and isn't in the cache, we can free it ourselves.
1160			 */
1161			if (list_empty(&lower->upper) &&
1162			    RB_EMPTY_NODE(&lower->rb_node))
1163				list_add(&lower->list, &useless);
1164
1165			if (!RB_EMPTY_NODE(&upper->rb_node))
1166				continue;
1167
1168			/* Add this guy's upper edges to the list to process */
1169			list_for_each_entry(edge, &upper->upper, list[LOWER])
1170				list_add_tail(&edge->list[UPPER], &list);
1171			if (list_empty(&upper->upper))
1172				list_add(&upper->list, &useless);
1173		}
1174
1175		while (!list_empty(&useless)) {
1176			lower = list_entry(useless.next,
1177					   struct backref_node, list);
1178			list_del_init(&lower->list);
1179			if (lower == node)
1180				node = NULL;
1181			free_backref_node(cache, lower);
1182		}
1183
1184		free_backref_node(cache, node);
1185		return ERR_PTR(err);
1186	}
1187	ASSERT(!node || !node->detached);
 
 
1188	return node;
1189}
1190
1191/*
1192 * helper to add backref node for the newly created snapshot.
1193 * the backref node is created by cloning backref node that
1194 * corresponds to root of source tree
1195 */
1196static int clone_backref_node(struct btrfs_trans_handle *trans,
1197			      struct reloc_control *rc,
1198			      struct btrfs_root *src,
1199			      struct btrfs_root *dest)
1200{
1201	struct btrfs_root *reloc_root = src->reloc_root;
1202	struct backref_cache *cache = &rc->backref_cache;
1203	struct backref_node *node = NULL;
1204	struct backref_node *new_node;
1205	struct backref_edge *edge;
1206	struct backref_edge *new_edge;
1207	struct rb_node *rb_node;
1208
1209	if (cache->last_trans > 0)
1210		update_backref_cache(trans, cache);
1211
1212	rb_node = tree_search(&cache->rb_root, src->commit_root->start);
1213	if (rb_node) {
1214		node = rb_entry(rb_node, struct backref_node, rb_node);
1215		if (node->detached)
1216			node = NULL;
1217		else
1218			BUG_ON(node->new_bytenr != reloc_root->node->start);
1219	}
1220
1221	if (!node) {
1222		rb_node = tree_search(&cache->rb_root,
1223				      reloc_root->commit_root->start);
1224		if (rb_node) {
1225			node = rb_entry(rb_node, struct backref_node,
1226					rb_node);
1227			BUG_ON(node->detached);
1228		}
1229	}
1230
1231	if (!node)
1232		return 0;
1233
1234	new_node = alloc_backref_node(cache);
 
1235	if (!new_node)
1236		return -ENOMEM;
1237
1238	new_node->bytenr = dest->node->start;
1239	new_node->level = node->level;
1240	new_node->lowest = node->lowest;
1241	new_node->checked = 1;
1242	new_node->root = dest;
 
1243
1244	if (!node->lowest) {
1245		list_for_each_entry(edge, &node->lower, list[UPPER]) {
1246			new_edge = alloc_backref_edge(cache);
1247			if (!new_edge)
1248				goto fail;
1249
1250			new_edge->node[UPPER] = new_node;
1251			new_edge->node[LOWER] = edge->node[LOWER];
1252			list_add_tail(&new_edge->list[UPPER],
1253				      &new_node->lower);
1254		}
1255	} else {
1256		list_add_tail(&new_node->lower, &cache->leaves);
1257	}
1258
1259	rb_node = tree_insert(&cache->rb_root, new_node->bytenr,
1260			      &new_node->rb_node);
1261	if (rb_node)
1262		backref_tree_panic(rb_node, -EEXIST, new_node->bytenr);
1263
1264	if (!new_node->lowest) {
1265		list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) {
1266			list_add_tail(&new_edge->list[LOWER],
1267				      &new_edge->node[LOWER]->upper);
1268		}
1269	}
1270	return 0;
1271fail:
1272	while (!list_empty(&new_node->lower)) {
1273		new_edge = list_entry(new_node->lower.next,
1274				      struct backref_edge, list[UPPER]);
1275		list_del(&new_edge->list[UPPER]);
1276		free_backref_edge(cache, new_edge);
1277	}
1278	free_backref_node(cache, new_node);
1279	return -ENOMEM;
1280}
1281
1282/*
1283 * helper to add 'address of tree root -> reloc tree' mapping
1284 */
1285static int __must_check __add_reloc_root(struct btrfs_root *root)
1286{
1287	struct btrfs_fs_info *fs_info = root->fs_info;
1288	struct rb_node *rb_node;
1289	struct mapping_node *node;
1290	struct reloc_control *rc = fs_info->reloc_ctl;
1291
1292	node = kmalloc(sizeof(*node), GFP_NOFS);
1293	if (!node)
1294		return -ENOMEM;
1295
1296	node->bytenr = root->node->start;
1297	node->data = root;
1298
1299	spin_lock(&rc->reloc_root_tree.lock);
1300	rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1301			      node->bytenr, &node->rb_node);
1302	spin_unlock(&rc->reloc_root_tree.lock);
1303	if (rb_node) {
1304		btrfs_panic(fs_info, -EEXIST,
1305			    "Duplicate root found for start=%llu while inserting into relocation tree",
1306			    node->bytenr);
 
1307	}
1308
1309	list_add_tail(&root->root_list, &rc->reloc_roots);
1310	return 0;
1311}
1312
1313/*
1314 * helper to delete the 'address of tree root -> reloc tree'
1315 * mapping
1316 */
1317static void __del_reloc_root(struct btrfs_root *root)
1318{
1319	struct btrfs_fs_info *fs_info = root->fs_info;
1320	struct rb_node *rb_node;
1321	struct mapping_node *node = NULL;
1322	struct reloc_control *rc = fs_info->reloc_ctl;
 
1323
1324	spin_lock(&rc->reloc_root_tree.lock);
1325	rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1326			      root->node->start);
1327	if (rb_node) {
1328		node = rb_entry(rb_node, struct mapping_node, rb_node);
1329		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
 
 
 
 
 
1330	}
1331	spin_unlock(&rc->reloc_root_tree.lock);
1332
1333	if (!node)
1334		return;
1335	BUG_ON((struct btrfs_root *)node->data != root);
1336
 
 
 
 
 
 
 
 
1337	spin_lock(&fs_info->trans_lock);
1338	list_del_init(&root->root_list);
 
 
 
1339	spin_unlock(&fs_info->trans_lock);
 
 
1340	kfree(node);
1341}
1342
1343/*
1344 * helper to update the 'address of tree root -> reloc tree'
1345 * mapping
1346 */
1347static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr)
1348{
1349	struct btrfs_fs_info *fs_info = root->fs_info;
1350	struct rb_node *rb_node;
1351	struct mapping_node *node = NULL;
1352	struct reloc_control *rc = fs_info->reloc_ctl;
1353
1354	spin_lock(&rc->reloc_root_tree.lock);
1355	rb_node = tree_search(&rc->reloc_root_tree.rb_root,
1356			      root->node->start);
1357	if (rb_node) {
1358		node = rb_entry(rb_node, struct mapping_node, rb_node);
1359		rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root);
1360	}
1361	spin_unlock(&rc->reloc_root_tree.lock);
1362
1363	if (!node)
1364		return 0;
1365	BUG_ON((struct btrfs_root *)node->data != root);
1366
1367	spin_lock(&rc->reloc_root_tree.lock);
1368	node->bytenr = new_bytenr;
1369	rb_node = tree_insert(&rc->reloc_root_tree.rb_root,
1370			      node->bytenr, &node->rb_node);
1371	spin_unlock(&rc->reloc_root_tree.lock);
1372	if (rb_node)
1373		backref_tree_panic(rb_node, -EEXIST, node->bytenr);
1374	return 0;
1375}
1376
1377static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans,
1378					struct btrfs_root *root, u64 objectid)
1379{
1380	struct btrfs_fs_info *fs_info = root->fs_info;
1381	struct btrfs_root *reloc_root;
1382	struct extent_buffer *eb;
1383	struct btrfs_root_item *root_item;
1384	struct btrfs_key root_key;
1385	int ret;
 
1386
1387	root_item = kmalloc(sizeof(*root_item), GFP_NOFS);
1388	BUG_ON(!root_item);
 
1389
1390	root_key.objectid = BTRFS_TREE_RELOC_OBJECTID;
1391	root_key.type = BTRFS_ROOT_ITEM_KEY;
1392	root_key.offset = objectid;
1393
1394	if (root->root_key.objectid == objectid) {
1395		u64 commit_root_gen;
1396
1397		/* called by btrfs_init_reloc_root */
1398		ret = btrfs_copy_root(trans, root, root->commit_root, &eb,
1399				      BTRFS_TREE_RELOC_OBJECTID);
1400		BUG_ON(ret);
 
 
1401		/*
1402		 * Set the last_snapshot field to the generation of the commit
1403		 * root - like this ctree.c:btrfs_block_can_be_shared() behaves
1404		 * correctly (returns true) when the relocation root is created
1405		 * either inside the critical section of a transaction commit
1406		 * (through transaction.c:qgroup_account_snapshot()) and when
1407		 * it's created before the transaction commit is started.
1408		 */
1409		commit_root_gen = btrfs_header_generation(root->commit_root);
1410		btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen);
1411	} else {
1412		/*
1413		 * called by btrfs_reloc_post_snapshot_hook.
1414		 * the source tree is a reloc tree, all tree blocks
1415		 * modified after it was created have RELOC flag
1416		 * set in their headers. so it's OK to not update
1417		 * the 'last_snapshot'.
1418		 */
1419		ret = btrfs_copy_root(trans, root, root->node, &eb,
1420				      BTRFS_TREE_RELOC_OBJECTID);
1421		BUG_ON(ret);
 
1422	}
1423
 
 
 
 
 
 
1424	memcpy(root_item, &root->root_item, sizeof(*root_item));
1425	btrfs_set_root_bytenr(root_item, eb->start);
1426	btrfs_set_root_level(root_item, btrfs_header_level(eb));
1427	btrfs_set_root_generation(root_item, trans->transid);
1428
1429	if (root->root_key.objectid == objectid) {
1430		btrfs_set_root_refs(root_item, 0);
1431		memset(&root_item->drop_progress, 0,
1432		       sizeof(struct btrfs_disk_key));
1433		root_item->drop_level = 0;
1434	}
1435
1436	btrfs_tree_unlock(eb);
1437	free_extent_buffer(eb);
1438
1439	ret = btrfs_insert_root(trans, fs_info->tree_root,
1440				&root_key, root_item);
1441	BUG_ON(ret);
 
 
1442	kfree(root_item);
1443
1444	reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key);
1445	BUG_ON(IS_ERR(reloc_root));
 
 
 
 
1446	reloc_root->last_trans = trans->transid;
1447	return reloc_root;
 
 
 
 
 
 
1448}
1449
1450/*
1451 * create reloc tree for a given fs tree. reloc tree is just a
1452 * snapshot of the fs tree with special root objectid.
 
 
 
1453 */
1454int btrfs_init_reloc_root(struct btrfs_trans_handle *trans,
1455			  struct btrfs_root *root)
1456{
1457	struct btrfs_fs_info *fs_info = root->fs_info;
1458	struct btrfs_root *reloc_root;
1459	struct reloc_control *rc = fs_info->reloc_ctl;
1460	struct btrfs_block_rsv *rsv;
1461	int clear_rsv = 0;
1462	int ret;
1463
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1464	if (root->reloc_root) {
1465		reloc_root = root->reloc_root;
1466		reloc_root->last_trans = trans->transid;
1467		return 0;
1468	}
1469
1470	if (!rc || !rc->create_reloc_tree ||
 
 
 
 
1471	    root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1472		return 0;
1473
1474	if (!trans->reloc_reserved) {
1475		rsv = trans->block_rsv;
1476		trans->block_rsv = rc->block_rsv;
1477		clear_rsv = 1;
1478	}
1479	reloc_root = create_reloc_root(trans, root, root->root_key.objectid);
1480	if (clear_rsv)
1481		trans->block_rsv = rsv;
 
 
1482
1483	ret = __add_reloc_root(reloc_root);
1484	BUG_ON(ret < 0);
1485	root->reloc_root = reloc_root;
 
 
 
 
 
1486	return 0;
1487}
1488
1489/*
1490 * update root item of reloc tree
1491 */
1492int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
1493			    struct btrfs_root *root)
1494{
1495	struct btrfs_fs_info *fs_info = root->fs_info;
1496	struct btrfs_root *reloc_root;
1497	struct btrfs_root_item *root_item;
1498	int ret;
1499
1500	if (!root->reloc_root)
1501		goto out;
1502
1503	reloc_root = root->reloc_root;
1504	root_item = &reloc_root->root_item;
1505
 
 
 
 
 
 
 
 
1506	if (fs_info->reloc_ctl->merge_reloc_tree &&
1507	    btrfs_root_refs(root_item) == 0) {
1508		root->reloc_root = NULL;
 
 
 
 
 
1509		__del_reloc_root(reloc_root);
1510	}
1511
1512	if (reloc_root->commit_root != reloc_root->node) {
 
1513		btrfs_set_root_node(root_item, reloc_root->node);
1514		free_extent_buffer(reloc_root->commit_root);
1515		reloc_root->commit_root = btrfs_root_node(reloc_root);
1516	}
1517
1518	ret = btrfs_update_root(trans, fs_info->tree_root,
1519				&reloc_root->root_key, root_item);
1520	BUG_ON(ret);
1521
1522out:
1523	return 0;
1524}
1525
1526/*
1527 * helper to find first cached inode with inode number >= objectid
1528 * in a subvolume
1529 */
1530static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid)
1531{
1532	struct rb_node *node;
1533	struct rb_node *prev;
1534	struct btrfs_inode *entry;
1535	struct inode *inode;
1536
1537	spin_lock(&root->inode_lock);
1538again:
1539	node = root->inode_tree.rb_node;
1540	prev = NULL;
1541	while (node) {
1542		prev = node;
1543		entry = rb_entry(node, struct btrfs_inode, rb_node);
1544
1545		if (objectid < btrfs_ino(entry))
1546			node = node->rb_left;
1547		else if (objectid > btrfs_ino(entry))
1548			node = node->rb_right;
1549		else
1550			break;
1551	}
1552	if (!node) {
1553		while (prev) {
1554			entry = rb_entry(prev, struct btrfs_inode, rb_node);
1555			if (objectid <= btrfs_ino(entry)) {
1556				node = prev;
1557				break;
1558			}
1559			prev = rb_next(prev);
1560		}
1561	}
1562	while (node) {
1563		entry = rb_entry(node, struct btrfs_inode, rb_node);
1564		inode = igrab(&entry->vfs_inode);
1565		if (inode) {
1566			spin_unlock(&root->inode_lock);
1567			return inode;
1568		}
1569
1570		objectid = btrfs_ino(entry) + 1;
1571		if (cond_resched_lock(&root->inode_lock))
1572			goto again;
1573
1574		node = rb_next(node);
1575	}
1576	spin_unlock(&root->inode_lock);
1577	return NULL;
1578}
1579
1580static int in_block_group(u64 bytenr,
1581			  struct btrfs_block_group_cache *block_group)
1582{
1583	if (bytenr >= block_group->key.objectid &&
1584	    bytenr < block_group->key.objectid + block_group->key.offset)
1585		return 1;
1586	return 0;
1587}
1588
1589/*
1590 * get new location of data
1591 */
1592static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr,
1593			    u64 bytenr, u64 num_bytes)
1594{
1595	struct btrfs_root *root = BTRFS_I(reloc_inode)->root;
1596	struct btrfs_path *path;
1597	struct btrfs_file_extent_item *fi;
1598	struct extent_buffer *leaf;
1599	int ret;
1600
1601	path = btrfs_alloc_path();
1602	if (!path)
1603		return -ENOMEM;
1604
1605	bytenr -= BTRFS_I(reloc_inode)->index_cnt;
1606	ret = btrfs_lookup_file_extent(NULL, root, path,
1607			btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0);
1608	if (ret < 0)
1609		goto out;
1610	if (ret > 0) {
1611		ret = -ENOENT;
1612		goto out;
1613	}
1614
1615	leaf = path->nodes[0];
1616	fi = btrfs_item_ptr(leaf, path->slots[0],
1617			    struct btrfs_file_extent_item);
1618
1619	BUG_ON(btrfs_file_extent_offset(leaf, fi) ||
1620	       btrfs_file_extent_compression(leaf, fi) ||
1621	       btrfs_file_extent_encryption(leaf, fi) ||
1622	       btrfs_file_extent_other_encoding(leaf, fi));
1623
1624	if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) {
1625		ret = -EINVAL;
1626		goto out;
1627	}
1628
1629	*new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1630	ret = 0;
1631out:
1632	btrfs_free_path(path);
1633	return ret;
1634}
1635
1636/*
1637 * update file extent items in the tree leaf to point to
1638 * the new locations.
1639 */
1640static noinline_for_stack
1641int replace_file_extents(struct btrfs_trans_handle *trans,
1642			 struct reloc_control *rc,
1643			 struct btrfs_root *root,
1644			 struct extent_buffer *leaf)
1645{
1646	struct btrfs_fs_info *fs_info = root->fs_info;
1647	struct btrfs_key key;
1648	struct btrfs_file_extent_item *fi;
1649	struct inode *inode = NULL;
1650	u64 parent;
1651	u64 bytenr;
1652	u64 new_bytenr = 0;
1653	u64 num_bytes;
1654	u64 end;
1655	u32 nritems;
1656	u32 i;
1657	int ret = 0;
1658	int first = 1;
1659	int dirty = 0;
1660
1661	if (rc->stage != UPDATE_DATA_PTRS)
1662		return 0;
1663
1664	/* reloc trees always use full backref */
1665	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1666		parent = leaf->start;
1667	else
1668		parent = 0;
1669
1670	nritems = btrfs_header_nritems(leaf);
1671	for (i = 0; i < nritems; i++) {
 
 
1672		cond_resched();
1673		btrfs_item_key_to_cpu(leaf, &key, i);
1674		if (key.type != BTRFS_EXTENT_DATA_KEY)
1675			continue;
1676		fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item);
1677		if (btrfs_file_extent_type(leaf, fi) ==
1678		    BTRFS_FILE_EXTENT_INLINE)
1679			continue;
1680		bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1681		num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1682		if (bytenr == 0)
1683			continue;
1684		if (!in_block_group(bytenr, rc->block_group))
 
1685			continue;
1686
1687		/*
1688		 * if we are modifying block in fs tree, wait for readpage
1689		 * to complete and drop the extent cache
1690		 */
1691		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
1692			if (first) {
1693				inode = find_next_inode(root, key.objectid);
1694				first = 0;
1695			} else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) {
1696				btrfs_add_delayed_iput(inode);
1697				inode = find_next_inode(root, key.objectid);
1698			}
1699			if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) {
 
 
1700				end = key.offset +
1701				      btrfs_file_extent_num_bytes(leaf, fi);
1702				WARN_ON(!IS_ALIGNED(key.offset,
1703						    fs_info->sectorsize));
1704				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
1705				end--;
1706				ret = try_lock_extent(&BTRFS_I(inode)->io_tree,
1707						      key.offset, end);
 
1708				if (!ret)
1709					continue;
1710
1711				btrfs_drop_extent_cache(BTRFS_I(inode),
1712						key.offset,	end, 1);
1713				unlock_extent(&BTRFS_I(inode)->io_tree,
1714					      key.offset, end);
1715			}
1716		}
1717
1718		ret = get_new_location(rc->data_inode, &new_bytenr,
1719				       bytenr, num_bytes);
1720		if (ret) {
1721			/*
1722			 * Don't have to abort since we've not changed anything
1723			 * in the file extent yet.
1724			 */
1725			break;
1726		}
1727
1728		btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr);
1729		dirty = 1;
1730
1731		key.offset -= btrfs_file_extent_offset(leaf, fi);
1732		ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
1733					   num_bytes, parent,
1734					   btrfs_header_owner(leaf),
1735					   key.objectid, key.offset);
 
 
1736		if (ret) {
1737			btrfs_abort_transaction(trans, ret);
1738			break;
1739		}
1740
1741		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
1742					parent, btrfs_header_owner(leaf),
1743					key.objectid, key.offset);
 
 
 
1744		if (ret) {
1745			btrfs_abort_transaction(trans, ret);
1746			break;
1747		}
1748	}
1749	if (dirty)
1750		btrfs_mark_buffer_dirty(leaf);
1751	if (inode)
1752		btrfs_add_delayed_iput(inode);
1753	return ret;
1754}
1755
1756static noinline_for_stack
1757int memcmp_node_keys(struct extent_buffer *eb, int slot,
1758		     struct btrfs_path *path, int level)
1759{
1760	struct btrfs_disk_key key1;
1761	struct btrfs_disk_key key2;
1762	btrfs_node_key(eb, &key1, slot);
1763	btrfs_node_key(path->nodes[level], &key2, path->slots[level]);
1764	return memcmp(&key1, &key2, sizeof(key1));
1765}
1766
1767/*
1768 * try to replace tree blocks in fs tree with the new blocks
1769 * in reloc tree. tree blocks haven't been modified since the
1770 * reloc tree was create can be replaced.
1771 *
1772 * if a block was replaced, level of the block + 1 is returned.
1773 * if no block got replaced, 0 is returned. if there are other
1774 * errors, a negative error number is returned.
1775 */
1776static noinline_for_stack
1777int replace_path(struct btrfs_trans_handle *trans,
1778		 struct btrfs_root *dest, struct btrfs_root *src,
1779		 struct btrfs_path *path, struct btrfs_key *next_key,
1780		 int lowest_level, int max_level)
1781{
1782	struct btrfs_fs_info *fs_info = dest->fs_info;
1783	struct extent_buffer *eb;
1784	struct extent_buffer *parent;
 
1785	struct btrfs_key key;
1786	u64 old_bytenr;
1787	u64 new_bytenr;
1788	u64 old_ptr_gen;
1789	u64 new_ptr_gen;
1790	u64 last_snapshot;
1791	u32 blocksize;
1792	int cow = 0;
1793	int level;
1794	int ret;
1795	int slot;
1796
1797	BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
1798	BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID);
1799
1800	last_snapshot = btrfs_root_last_snapshot(&src->root_item);
1801again:
1802	slot = path->slots[lowest_level];
1803	btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot);
1804
1805	eb = btrfs_lock_root_node(dest);
1806	btrfs_set_lock_blocking(eb);
1807	level = btrfs_header_level(eb);
1808
1809	if (level < lowest_level) {
1810		btrfs_tree_unlock(eb);
1811		free_extent_buffer(eb);
1812		return 0;
1813	}
1814
1815	if (cow) {
1816		ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb);
1817		BUG_ON(ret);
 
 
 
 
 
1818	}
1819	btrfs_set_lock_blocking(eb);
1820
1821	if (next_key) {
1822		next_key->objectid = (u64)-1;
1823		next_key->type = (u8)-1;
1824		next_key->offset = (u64)-1;
1825	}
1826
1827	parent = eb;
1828	while (1) {
1829		struct btrfs_key first_key;
1830
1831		level = btrfs_header_level(parent);
1832		BUG_ON(level < lowest_level);
1833
1834		ret = btrfs_bin_search(parent, &key, level, &slot);
 
 
1835		if (ret && slot > 0)
1836			slot--;
1837
1838		if (next_key && slot + 1 < btrfs_header_nritems(parent))
1839			btrfs_node_key_to_cpu(parent, next_key, slot + 1);
1840
1841		old_bytenr = btrfs_node_blockptr(parent, slot);
1842		blocksize = fs_info->nodesize;
1843		old_ptr_gen = btrfs_node_ptr_generation(parent, slot);
1844		btrfs_node_key_to_cpu(parent, &first_key, slot);
1845
1846		if (level <= max_level) {
1847			eb = path->nodes[level];
1848			new_bytenr = btrfs_node_blockptr(eb,
1849							path->slots[level]);
1850			new_ptr_gen = btrfs_node_ptr_generation(eb,
1851							path->slots[level]);
1852		} else {
1853			new_bytenr = 0;
1854			new_ptr_gen = 0;
1855		}
1856
1857		if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) {
1858			ret = level;
1859			break;
1860		}
1861
1862		if (new_bytenr == 0 || old_ptr_gen > last_snapshot ||
1863		    memcmp_node_keys(parent, slot, path, level)) {
1864			if (level <= lowest_level) {
1865				ret = 0;
1866				break;
1867			}
1868
1869			eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen,
1870					     level - 1, &first_key);
1871			if (IS_ERR(eb)) {
1872				ret = PTR_ERR(eb);
1873				break;
1874			} else if (!extent_buffer_uptodate(eb)) {
1875				ret = -EIO;
1876				free_extent_buffer(eb);
1877				break;
1878			}
1879			btrfs_tree_lock(eb);
1880			if (cow) {
1881				ret = btrfs_cow_block(trans, dest, eb, parent,
1882						      slot, &eb);
1883				BUG_ON(ret);
 
 
 
 
 
1884			}
1885			btrfs_set_lock_blocking(eb);
1886
1887			btrfs_tree_unlock(parent);
1888			free_extent_buffer(parent);
1889
1890			parent = eb;
1891			continue;
1892		}
1893
1894		if (!cow) {
1895			btrfs_tree_unlock(parent);
1896			free_extent_buffer(parent);
1897			cow = 1;
1898			goto again;
1899		}
1900
1901		btrfs_node_key_to_cpu(path->nodes[level], &key,
1902				      path->slots[level]);
1903		btrfs_release_path(path);
1904
1905		path->lowest_level = level;
 
1906		ret = btrfs_search_slot(trans, src, &key, path, 0, 1);
 
1907		path->lowest_level = 0;
1908		BUG_ON(ret);
 
 
 
 
1909
1910		/*
1911		 * Info qgroup to trace both subtrees.
1912		 *
1913		 * We must trace both trees.
1914		 * 1) Tree reloc subtree
1915		 *    If not traced, we will leak data numbers
1916		 * 2) Fs subtree
1917		 *    If not traced, we will double count old data
1918		 *    and tree block numbers, if current trans doesn't free
1919		 *    data reloc tree inode.
 
 
 
1920		 */
1921		ret = btrfs_qgroup_trace_subtree(trans, src, parent,
1922				btrfs_header_generation(parent),
1923				btrfs_header_level(parent));
1924		if (ret < 0)
1925			break;
1926		ret = btrfs_qgroup_trace_subtree(trans, dest,
1927				path->nodes[level],
1928				btrfs_header_generation(path->nodes[level]),
1929				btrfs_header_level(path->nodes[level]));
1930		if (ret < 0)
1931			break;
1932
1933		/*
1934		 * swap blocks in fs tree and reloc tree.
1935		 */
1936		btrfs_set_node_blockptr(parent, slot, new_bytenr);
1937		btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen);
1938		btrfs_mark_buffer_dirty(parent);
1939
1940		btrfs_set_node_blockptr(path->nodes[level],
1941					path->slots[level], old_bytenr);
1942		btrfs_set_node_ptr_generation(path->nodes[level],
1943					      path->slots[level], old_ptr_gen);
1944		btrfs_mark_buffer_dirty(path->nodes[level]);
1945
1946		ret = btrfs_inc_extent_ref(trans, src, old_bytenr,
1947					blocksize, path->nodes[level]->start,
1948					src->root_key.objectid, level - 1, 0);
1949		BUG_ON(ret);
1950		ret = btrfs_inc_extent_ref(trans, dest, new_bytenr,
1951					blocksize, 0, dest->root_key.objectid,
1952					level - 1, 0);
1953		BUG_ON(ret);
1954
1955		ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
1956					path->nodes[level]->start,
1957					src->root_key.objectid, level - 1, 0);
1958		BUG_ON(ret);
1959
1960		ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
1961					0, dest->root_key.objectid, level - 1,
1962					0);
1963		BUG_ON(ret);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1964
1965		btrfs_unlock_up_safe(path, 0);
1966
1967		ret = level;
1968		break;
1969	}
1970	btrfs_tree_unlock(parent);
1971	free_extent_buffer(parent);
1972	return ret;
1973}
1974
1975/*
1976 * helper to find next relocated block in reloc tree
1977 */
1978static noinline_for_stack
1979int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
1980		       int *level)
1981{
1982	struct extent_buffer *eb;
1983	int i;
1984	u64 last_snapshot;
1985	u32 nritems;
1986
1987	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
1988
1989	for (i = 0; i < *level; i++) {
1990		free_extent_buffer(path->nodes[i]);
1991		path->nodes[i] = NULL;
1992	}
1993
1994	for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) {
1995		eb = path->nodes[i];
1996		nritems = btrfs_header_nritems(eb);
1997		while (path->slots[i] + 1 < nritems) {
1998			path->slots[i]++;
1999			if (btrfs_node_ptr_generation(eb, path->slots[i]) <=
2000			    last_snapshot)
2001				continue;
2002
2003			*level = i;
2004			return 0;
2005		}
2006		free_extent_buffer(path->nodes[i]);
2007		path->nodes[i] = NULL;
2008	}
2009	return 1;
2010}
2011
2012/*
2013 * walk down reloc tree to find relocated block of lowest level
2014 */
2015static noinline_for_stack
2016int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path,
2017			 int *level)
2018{
2019	struct btrfs_fs_info *fs_info = root->fs_info;
2020	struct extent_buffer *eb = NULL;
2021	int i;
2022	u64 bytenr;
2023	u64 ptr_gen = 0;
2024	u64 last_snapshot;
2025	u32 nritems;
2026
2027	last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2028
2029	for (i = *level; i > 0; i--) {
2030		struct btrfs_key first_key;
2031
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		btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]);
2053		eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1,
2054				     &first_key);
2055		if (IS_ERR(eb)) {
2056			return PTR_ERR(eb);
2057		} else if (!extent_buffer_uptodate(eb)) {
2058			free_extent_buffer(eb);
2059			return -EIO;
2060		}
2061		BUG_ON(btrfs_header_level(eb) != i - 1);
2062		path->nodes[i - 1] = eb;
2063		path->slots[i - 1] = 0;
2064	}
2065	return 1;
2066}
2067
2068/*
2069 * invalidate extent cache for file extents whose key in range of
2070 * [min_key, max_key)
2071 */
2072static int invalidate_extent_cache(struct btrfs_root *root,
2073				   struct btrfs_key *min_key,
2074				   struct btrfs_key *max_key)
2075{
2076	struct btrfs_fs_info *fs_info = root->fs_info;
2077	struct inode *inode = NULL;
2078	u64 objectid;
2079	u64 start, end;
2080	u64 ino;
2081
2082	objectid = min_key->objectid;
2083	while (1) {
 
 
2084		cond_resched();
2085		iput(inode);
2086
2087		if (objectid > max_key->objectid)
2088			break;
2089
2090		inode = find_next_inode(root, objectid);
2091		if (!inode)
2092			break;
2093		ino = btrfs_ino(BTRFS_I(inode));
2094
2095		if (ino > max_key->objectid) {
2096			iput(inode);
2097			break;
2098		}
2099
2100		objectid = ino + 1;
2101		if (!S_ISREG(inode->i_mode))
2102			continue;
2103
2104		if (unlikely(min_key->objectid == ino)) {
2105			if (min_key->type > BTRFS_EXTENT_DATA_KEY)
2106				continue;
2107			if (min_key->type < BTRFS_EXTENT_DATA_KEY)
2108				start = 0;
2109			else {
2110				start = min_key->offset;
2111				WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize));
2112			}
2113		} else {
2114			start = 0;
2115		}
2116
2117		if (unlikely(max_key->objectid == ino)) {
2118			if (max_key->type < BTRFS_EXTENT_DATA_KEY)
2119				continue;
2120			if (max_key->type > BTRFS_EXTENT_DATA_KEY) {
2121				end = (u64)-1;
2122			} else {
2123				if (max_key->offset == 0)
2124					continue;
2125				end = max_key->offset;
2126				WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize));
2127				end--;
2128			}
2129		} else {
2130			end = (u64)-1;
2131		}
2132
2133		/* the lock_extent waits for readpage to complete */
2134		lock_extent(&BTRFS_I(inode)->io_tree, start, end);
2135		btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1);
2136		unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
2137	}
2138	return 0;
2139}
2140
2141static int find_next_key(struct btrfs_path *path, int level,
2142			 struct btrfs_key *key)
2143
2144{
2145	while (level < BTRFS_MAX_LEVEL) {
2146		if (!path->nodes[level])
2147			break;
2148		if (path->slots[level] + 1 <
2149		    btrfs_header_nritems(path->nodes[level])) {
2150			btrfs_node_key_to_cpu(path->nodes[level], key,
2151					      path->slots[level] + 1);
2152			return 0;
2153		}
2154		level++;
2155	}
2156	return 1;
2157}
2158
2159/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2160 * merge the relocated tree blocks in reloc tree with corresponding
2161 * fs tree.
2162 */
2163static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
2164					       struct btrfs_root *root)
2165{
2166	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2167	LIST_HEAD(inode_list);
2168	struct btrfs_key key;
2169	struct btrfs_key next_key;
2170	struct btrfs_trans_handle *trans = NULL;
2171	struct btrfs_root *reloc_root;
2172	struct btrfs_root_item *root_item;
2173	struct btrfs_path *path;
2174	struct extent_buffer *leaf;
 
2175	int level;
2176	int max_level;
2177	int replaced = 0;
2178	int ret;
2179	int err = 0;
2180	u32 min_reserved;
2181
2182	path = btrfs_alloc_path();
2183	if (!path)
2184		return -ENOMEM;
2185	path->reada = READA_FORWARD;
2186
2187	reloc_root = root->reloc_root;
2188	root_item = &reloc_root->root_item;
2189
2190	if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
2191		level = btrfs_root_level(root_item);
2192		extent_buffer_get(reloc_root->node);
2193		path->nodes[level] = reloc_root->node;
2194		path->slots[level] = 0;
2195	} else {
2196		btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
2197
2198		level = root_item->drop_level;
2199		BUG_ON(level == 0);
2200		path->lowest_level = level;
2201		ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
2202		path->lowest_level = 0;
2203		if (ret < 0) {
2204			btrfs_free_path(path);
2205			return ret;
2206		}
2207
2208		btrfs_node_key_to_cpu(path->nodes[level], &next_key,
2209				      path->slots[level]);
2210		WARN_ON(memcmp(&key, &next_key, sizeof(key)));
2211
2212		btrfs_unlock_up_safe(path, 0);
2213	}
2214
2215	min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
 
 
 
 
 
 
 
 
 
2216	memset(&next_key, 0, sizeof(next_key));
2217
2218	while (1) {
2219		ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved,
2220					     BTRFS_RESERVE_FLUSH_ALL);
2221		if (ret) {
2222			err = ret;
2223			goto out;
2224		}
2225		trans = btrfs_start_transaction(root, 0);
2226		if (IS_ERR(trans)) {
2227			err = PTR_ERR(trans);
2228			trans = NULL;
2229			goto out;
2230		}
 
 
 
 
 
 
 
 
 
 
 
 
2231		trans->block_rsv = rc->block_rsv;
2232
2233		replaced = 0;
2234		max_level = level;
2235
2236		ret = walk_down_reloc_tree(reloc_root, path, &level);
2237		if (ret < 0) {
2238			err = ret;
2239			goto out;
2240		}
2241		if (ret > 0)
2242			break;
2243
2244		if (!find_next_key(path, level, &key) &&
2245		    btrfs_comp_cpu_keys(&next_key, &key) >= 0) {
2246			ret = 0;
2247		} else {
2248			ret = replace_path(trans, root, reloc_root, path,
2249					   &next_key, level, max_level);
2250		}
2251		if (ret < 0) {
2252			err = ret;
2253			goto out;
2254		}
2255
2256		if (ret > 0) {
2257			level = ret;
2258			btrfs_node_key_to_cpu(path->nodes[level], &key,
2259					      path->slots[level]);
2260			replaced = 1;
2261		}
2262
2263		ret = walk_up_reloc_tree(reloc_root, path, &level);
2264		if (ret > 0)
2265			break;
2266
2267		BUG_ON(level == 0);
2268		/*
2269		 * save the merging progress in the drop_progress.
2270		 * this is OK since root refs == 1 in this case.
2271		 */
2272		btrfs_node_key(path->nodes[level], &root_item->drop_progress,
2273			       path->slots[level]);
2274		root_item->drop_level = level;
2275
2276		btrfs_end_transaction_throttle(trans);
2277		trans = NULL;
2278
2279		btrfs_btree_balance_dirty(fs_info);
2280
2281		if (replaced && rc->stage == UPDATE_DATA_PTRS)
2282			invalidate_extent_cache(root, &key, &next_key);
2283	}
2284
2285	/*
2286	 * handle the case only one block in the fs tree need to be
2287	 * relocated and the block is tree root.
2288	 */
2289	leaf = btrfs_lock_root_node(root);
2290	ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf);
 
2291	btrfs_tree_unlock(leaf);
2292	free_extent_buffer(leaf);
2293	if (ret < 0)
2294		err = ret;
2295out:
2296	btrfs_free_path(path);
2297
2298	if (err == 0) {
2299		memset(&root_item->drop_progress, 0,
2300		       sizeof(root_item->drop_progress));
2301		root_item->drop_level = 0;
2302		btrfs_set_root_refs(root_item, 0);
2303		btrfs_update_reloc_root(trans, root);
2304	}
2305
2306	if (trans)
2307		btrfs_end_transaction_throttle(trans);
2308
2309	btrfs_btree_balance_dirty(fs_info);
2310
2311	if (replaced && rc->stage == UPDATE_DATA_PTRS)
2312		invalidate_extent_cache(root, &key, &next_key);
2313
2314	return err;
2315}
2316
2317static noinline_for_stack
2318int prepare_to_merge(struct reloc_control *rc, int err)
2319{
2320	struct btrfs_root *root = rc->extent_root;
2321	struct btrfs_fs_info *fs_info = root->fs_info;
2322	struct btrfs_root *reloc_root;
2323	struct btrfs_trans_handle *trans;
2324	LIST_HEAD(reloc_roots);
2325	u64 num_bytes = 0;
2326	int ret;
2327
2328	mutex_lock(&fs_info->reloc_mutex);
2329	rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
2330	rc->merging_rsv_size += rc->nodes_relocated * 2;
2331	mutex_unlock(&fs_info->reloc_mutex);
2332
2333again:
2334	if (!err) {
2335		num_bytes = rc->merging_rsv_size;
2336		ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes,
2337					  BTRFS_RESERVE_FLUSH_ALL);
2338		if (ret)
2339			err = ret;
2340	}
2341
2342	trans = btrfs_join_transaction(rc->extent_root);
2343	if (IS_ERR(trans)) {
2344		if (!err)
2345			btrfs_block_rsv_release(fs_info, rc->block_rsv,
2346						num_bytes);
2347		return PTR_ERR(trans);
2348	}
2349
2350	if (!err) {
2351		if (num_bytes != rc->merging_rsv_size) {
2352			btrfs_end_transaction(trans);
2353			btrfs_block_rsv_release(fs_info, rc->block_rsv,
2354						num_bytes);
2355			goto again;
2356		}
2357	}
2358
2359	rc->merge_reloc_tree = 1;
2360
2361	while (!list_empty(&rc->reloc_roots)) {
2362		reloc_root = list_entry(rc->reloc_roots.next,
2363					struct btrfs_root, root_list);
2364		list_del_init(&reloc_root->root_list);
2365
2366		root = read_fs_root(fs_info, reloc_root->root_key.offset);
2367		BUG_ON(IS_ERR(root));
2368		BUG_ON(root->reloc_root != reloc_root);
 
 
 
 
 
 
 
 
 
 
 
2369
2370		/*
2371		 * set reference count to 1, so btrfs_recover_relocation
2372		 * knows it should resumes merging
2373		 */
2374		if (!err)
2375			btrfs_set_root_refs(&reloc_root->root_item, 1);
2376		btrfs_update_reloc_root(trans, root);
2377
 
 
 
 
2378		list_add(&reloc_root->root_list, &reloc_roots);
 
 
 
 
 
 
 
 
2379	}
2380
2381	list_splice(&reloc_roots, &rc->reloc_roots);
2382
2383	if (!err)
2384		btrfs_commit_transaction(trans);
2385	else
2386		btrfs_end_transaction(trans);
2387	return err;
2388}
2389
2390static noinline_for_stack
2391void free_reloc_roots(struct list_head *list)
2392{
2393	struct btrfs_root *reloc_root;
2394
2395	while (!list_empty(list)) {
2396		reloc_root = list_entry(list->next, struct btrfs_root,
2397					root_list);
2398		__del_reloc_root(reloc_root);
2399		free_extent_buffer(reloc_root->node);
2400		free_extent_buffer(reloc_root->commit_root);
2401		reloc_root->node = NULL;
2402		reloc_root->commit_root = NULL;
2403	}
2404}
2405
2406static noinline_for_stack
2407void merge_reloc_roots(struct reloc_control *rc)
2408{
2409	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2410	struct btrfs_root *root;
2411	struct btrfs_root *reloc_root;
2412	LIST_HEAD(reloc_roots);
2413	int found = 0;
2414	int ret = 0;
2415again:
2416	root = rc->extent_root;
2417
2418	/*
2419	 * this serializes us with btrfs_record_root_in_transaction,
2420	 * we have to make sure nobody is in the middle of
2421	 * adding their roots to the list while we are
2422	 * doing this splice
2423	 */
2424	mutex_lock(&fs_info->reloc_mutex);
2425	list_splice_init(&rc->reloc_roots, &reloc_roots);
2426	mutex_unlock(&fs_info->reloc_mutex);
2427
2428	while (!list_empty(&reloc_roots)) {
2429		found = 1;
2430		reloc_root = list_entry(reloc_roots.next,
2431					struct btrfs_root, root_list);
2432
 
 
2433		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
2434			root = read_fs_root(fs_info,
2435					    reloc_root->root_key.offset);
2436			BUG_ON(IS_ERR(root));
2437			BUG_ON(root->reloc_root != reloc_root);
2438
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2439			ret = merge_reloc_root(rc, root);
 
2440			if (ret) {
2441				if (list_empty(&reloc_root->root_list))
2442					list_add_tail(&reloc_root->root_list,
2443						      &reloc_roots);
2444				goto out;
2445			}
2446		} else {
 
 
 
 
 
 
 
 
 
 
2447			list_del_init(&reloc_root->root_list);
2448		}
2449
2450		ret = btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
2451		if (ret < 0) {
2452			if (list_empty(&reloc_root->root_list))
2453				list_add_tail(&reloc_root->root_list,
2454					      &reloc_roots);
2455			goto out;
2456		}
2457	}
2458
2459	if (found) {
2460		found = 0;
2461		goto again;
2462	}
2463out:
2464	if (ret) {
2465		btrfs_handle_fs_error(fs_info, ret, NULL);
2466		if (!list_empty(&reloc_roots))
2467			free_reloc_roots(&reloc_roots);
2468
2469		/* new reloc root may be added */
2470		mutex_lock(&fs_info->reloc_mutex);
2471		list_splice_init(&rc->reloc_roots, &reloc_roots);
2472		mutex_unlock(&fs_info->reloc_mutex);
2473		if (!list_empty(&reloc_roots))
2474			free_reloc_roots(&reloc_roots);
2475	}
2476
2477	BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2478}
2479
2480static void free_block_list(struct rb_root *blocks)
2481{
2482	struct tree_block *block;
2483	struct rb_node *rb_node;
2484	while ((rb_node = rb_first(blocks))) {
2485		block = rb_entry(rb_node, struct tree_block, rb_node);
2486		rb_erase(rb_node, blocks);
2487		kfree(block);
2488	}
2489}
2490
2491static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans,
2492				      struct btrfs_root *reloc_root)
2493{
2494	struct btrfs_fs_info *fs_info = reloc_root->fs_info;
2495	struct btrfs_root *root;
 
2496
2497	if (reloc_root->last_trans == trans->transid)
2498		return 0;
2499
2500	root = read_fs_root(fs_info, reloc_root->root_key.offset);
2501	BUG_ON(IS_ERR(root));
2502	BUG_ON(root->reloc_root != reloc_root);
2503
2504	return btrfs_record_root_in_trans(trans, root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2505}
2506
2507static noinline_for_stack
2508struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans,
2509				     struct reloc_control *rc,
2510				     struct backref_node *node,
2511				     struct backref_edge *edges[])
2512{
2513	struct backref_node *next;
2514	struct btrfs_root *root;
2515	int index = 0;
 
2516
2517	next = node;
2518	while (1) {
2519		cond_resched();
2520		next = walk_up_backref(next, edges, &index);
2521		root = next->root;
2522		BUG_ON(!root);
2523		BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2524
2525		if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
2526			record_reloc_root_in_trans(trans, root);
 
 
2527			break;
2528		}
2529
2530		btrfs_record_root_in_trans(trans, root);
 
 
2531		root = root->reloc_root;
2532
 
 
 
 
 
 
 
2533		if (next->new_bytenr != root->node->start) {
2534			BUG_ON(next->new_bytenr);
2535			BUG_ON(!list_empty(&next->list));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2536			next->new_bytenr = root->node->start;
2537			next->root = root;
 
 
2538			list_add_tail(&next->list,
2539				      &rc->backref_cache.changed);
2540			__mark_block_processed(rc, next);
2541			break;
2542		}
2543
2544		WARN_ON(1);
2545		root = NULL;
2546		next = walk_down_backref(edges, &index);
2547		if (!next || next->level <= node->level)
2548			break;
2549	}
2550	if (!root)
2551		return NULL;
 
 
 
 
 
 
2552
2553	next = node;
2554	/* setup backref node path for btrfs_reloc_cow_block */
2555	while (1) {
2556		rc->backref_cache.path[next->level] = next;
2557		if (--index < 0)
2558			break;
2559		next = edges[index]->node[UPPER];
2560	}
2561	return root;
2562}
2563
2564/*
2565 * select a tree root for relocation. return NULL if the block
2566 * is reference counted. we should use do_relocation() in this
2567 * case. return a tree root pointer if the block isn't reference
2568 * counted. return -ENOENT if the block is root of reloc tree.
 
 
 
2569 */
2570static noinline_for_stack
2571struct btrfs_root *select_one_root(struct backref_node *node)
2572{
2573	struct backref_node *next;
2574	struct btrfs_root *root;
2575	struct btrfs_root *fs_root = NULL;
2576	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2577	int index = 0;
2578
2579	next = node;
2580	while (1) {
2581		cond_resched();
2582		next = walk_up_backref(next, edges, &index);
2583		root = next->root;
2584		BUG_ON(!root);
2585
2586		/* no other choice for non-references counted tree */
2587		if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state))
 
 
 
 
 
 
 
2588			return root;
2589
2590		if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID)
2591			fs_root = root;
2592
2593		if (next != node)
2594			return NULL;
2595
2596		next = walk_down_backref(edges, &index);
2597		if (!next || next->level <= node->level)
2598			break;
2599	}
2600
2601	if (!fs_root)
2602		return ERR_PTR(-ENOENT);
2603	return fs_root;
2604}
2605
2606static noinline_for_stack
2607u64 calcu_metadata_size(struct reloc_control *rc,
2608			struct backref_node *node, int reserve)
2609{
2610	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2611	struct backref_node *next = node;
2612	struct backref_edge *edge;
2613	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2614	u64 num_bytes = 0;
2615	int index = 0;
2616
2617	BUG_ON(reserve && node->processed);
2618
2619	while (next) {
2620		cond_resched();
2621		while (1) {
2622			if (next->processed && (reserve || next != node))
2623				break;
2624
2625			num_bytes += fs_info->nodesize;
2626
2627			if (list_empty(&next->upper))
2628				break;
2629
2630			edge = list_entry(next->upper.next,
2631					  struct backref_edge, list[LOWER]);
2632			edges[index++] = edge;
2633			next = edge->node[UPPER];
2634		}
2635		next = walk_down_backref(edges, &index);
2636	}
2637	return num_bytes;
2638}
2639
2640static int reserve_metadata_space(struct btrfs_trans_handle *trans,
2641				  struct reloc_control *rc,
2642				  struct backref_node *node)
2643{
2644	struct btrfs_root *root = rc->extent_root;
2645	struct btrfs_fs_info *fs_info = root->fs_info;
2646	u64 num_bytes;
2647	int ret;
2648	u64 tmp;
2649
2650	num_bytes = calcu_metadata_size(rc, node, 1) * 2;
2651
2652	trans->block_rsv = rc->block_rsv;
2653	rc->reserved_bytes += num_bytes;
2654
2655	/*
2656	 * We are under a transaction here so we can only do limited flushing.
2657	 * If we get an enospc just kick back -EAGAIN so we know to drop the
2658	 * transaction and try to refill when we can flush all the things.
2659	 */
2660	ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes,
2661				BTRFS_RESERVE_FLUSH_LIMIT);
2662	if (ret) {
2663		tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES;
2664		while (tmp <= rc->reserved_bytes)
2665			tmp <<= 1;
2666		/*
2667		 * only one thread can access block_rsv at this point,
2668		 * so we don't need hold lock to protect block_rsv.
2669		 * we expand more reservation size here to allow enough
2670		 * space for relocation and we will return eailer in
2671		 * enospc case.
2672		 */
2673		rc->block_rsv->size = tmp + fs_info->nodesize *
2674				      RELOCATION_RESERVED_NODES;
2675		return -EAGAIN;
2676	}
2677
2678	return 0;
2679}
2680
2681/*
2682 * relocate a block tree, and then update pointers in upper level
2683 * blocks that reference the block to point to the new location.
2684 *
2685 * if called by link_to_upper, the block has already been relocated.
2686 * in that case this function just updates pointers.
2687 */
2688static int do_relocation(struct btrfs_trans_handle *trans,
2689			 struct reloc_control *rc,
2690			 struct backref_node *node,
2691			 struct btrfs_key *key,
2692			 struct btrfs_path *path, int lowest)
2693{
2694	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
2695	struct backref_node *upper;
2696	struct backref_edge *edge;
2697	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2698	struct btrfs_root *root;
2699	struct extent_buffer *eb;
2700	u32 blocksize;
2701	u64 bytenr;
2702	u64 generation;
2703	int slot;
2704	int ret;
2705	int err = 0;
2706
2707	BUG_ON(lowest && node->eb);
 
 
 
 
2708
2709	path->lowest_level = node->level + 1;
2710	rc->backref_cache.path[node->level] = node;
2711	list_for_each_entry(edge, &node->upper, list[LOWER]) {
2712		struct btrfs_key first_key;
2713
2714		cond_resched();
2715
2716		upper = edge->node[UPPER];
2717		root = select_reloc_root(trans, rc, upper, edges);
2718		BUG_ON(!root);
 
 
 
2719
2720		if (upper->eb && !upper->locked) {
2721			if (!lowest) {
2722				ret = btrfs_bin_search(upper->eb, key,
2723						       upper->level, &slot);
 
2724				BUG_ON(ret);
2725				bytenr = btrfs_node_blockptr(upper->eb, slot);
2726				if (node->eb->start == bytenr)
2727					goto next;
2728			}
2729			drop_node_buffer(upper);
2730		}
2731
2732		if (!upper->eb) {
2733			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
2734			if (ret) {
2735				if (ret < 0)
2736					err = ret;
2737				else
2738					err = -ENOENT;
2739
2740				btrfs_release_path(path);
2741				break;
2742			}
2743
2744			if (!upper->eb) {
2745				upper->eb = path->nodes[upper->level];
2746				path->nodes[upper->level] = NULL;
2747			} else {
2748				BUG_ON(upper->eb != path->nodes[upper->level]);
2749			}
2750
2751			upper->locked = 1;
2752			path->locks[upper->level] = 0;
2753
2754			slot = path->slots[upper->level];
2755			btrfs_release_path(path);
2756		} else {
2757			ret = btrfs_bin_search(upper->eb, key, upper->level,
2758					       &slot);
 
2759			BUG_ON(ret);
2760		}
2761
2762		bytenr = btrfs_node_blockptr(upper->eb, slot);
2763		if (lowest) {
2764			if (bytenr != node->bytenr) {
2765				btrfs_err(root->fs_info,
2766		"lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu",
2767					  bytenr, node->bytenr, slot,
2768					  upper->eb->start);
2769				err = -EIO;
2770				goto next;
2771			}
2772		} else {
2773			if (node->eb->start == bytenr)
2774				goto next;
2775		}
2776
2777		blocksize = root->fs_info->nodesize;
2778		generation = btrfs_node_ptr_generation(upper->eb, slot);
2779		btrfs_node_key_to_cpu(upper->eb, &first_key, slot);
2780		eb = read_tree_block(fs_info, bytenr, generation,
2781				     upper->level - 1, &first_key);
2782		if (IS_ERR(eb)) {
2783			err = PTR_ERR(eb);
2784			goto next;
2785		} else if (!extent_buffer_uptodate(eb)) {
2786			free_extent_buffer(eb);
2787			err = -EIO;
2788			goto next;
2789		}
2790		btrfs_tree_lock(eb);
2791		btrfs_set_lock_blocking(eb);
2792
2793		if (!node->eb) {
2794			ret = btrfs_cow_block(trans, root, eb, upper->eb,
2795					      slot, &eb);
2796			btrfs_tree_unlock(eb);
2797			free_extent_buffer(eb);
2798			if (ret < 0) {
2799				err = ret;
2800				goto next;
2801			}
2802			BUG_ON(node->eb != eb);
 
 
 
2803		} else {
2804			btrfs_set_node_blockptr(upper->eb, slot,
2805						node->eb->start);
2806			btrfs_set_node_ptr_generation(upper->eb, slot,
2807						      trans->transid);
2808			btrfs_mark_buffer_dirty(upper->eb);
2809
2810			ret = btrfs_inc_extent_ref(trans, root,
2811						node->eb->start, blocksize,
2812						upper->eb->start,
2813						btrfs_header_owner(upper->eb),
2814						node->level, 0);
2815			BUG_ON(ret);
2816
2817			ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
2818			BUG_ON(ret);
 
 
 
2819		}
2820next:
2821		if (!upper->pending)
2822			drop_node_buffer(upper);
2823		else
2824			unlock_node_buffer(upper);
2825		if (err)
2826			break;
2827	}
2828
2829	if (!err && node->pending) {
2830		drop_node_buffer(node);
2831		list_move_tail(&node->list, &rc->backref_cache.changed);
2832		node->pending = 0;
2833	}
2834
2835	path->lowest_level = 0;
2836	BUG_ON(err == -ENOSPC);
2837	return err;
 
 
 
 
 
2838}
2839
2840static int link_to_upper(struct btrfs_trans_handle *trans,
2841			 struct reloc_control *rc,
2842			 struct backref_node *node,
2843			 struct btrfs_path *path)
2844{
2845	struct btrfs_key key;
2846
2847	btrfs_node_key_to_cpu(node->eb, &key, 0);
2848	return do_relocation(trans, rc, node, &key, path, 0);
2849}
2850
2851static int finish_pending_nodes(struct btrfs_trans_handle *trans,
2852				struct reloc_control *rc,
2853				struct btrfs_path *path, int err)
2854{
2855	LIST_HEAD(list);
2856	struct backref_cache *cache = &rc->backref_cache;
2857	struct backref_node *node;
2858	int level;
2859	int ret;
2860
2861	for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2862		while (!list_empty(&cache->pending[level])) {
2863			node = list_entry(cache->pending[level].next,
2864					  struct backref_node, list);
2865			list_move_tail(&node->list, &list);
2866			BUG_ON(!node->pending);
2867
2868			if (!err) {
2869				ret = link_to_upper(trans, rc, node, path);
2870				if (ret < 0)
2871					err = ret;
2872			}
2873		}
2874		list_splice_init(&list, &cache->pending[level]);
2875	}
2876	return err;
2877}
2878
2879static void mark_block_processed(struct reloc_control *rc,
2880				 u64 bytenr, u32 blocksize)
2881{
2882	set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1,
2883			EXTENT_DIRTY);
2884}
2885
2886static void __mark_block_processed(struct reloc_control *rc,
2887				   struct backref_node *node)
2888{
2889	u32 blocksize;
2890	if (node->level == 0 ||
2891	    in_block_group(node->bytenr, rc->block_group)) {
2892		blocksize = rc->extent_root->fs_info->nodesize;
2893		mark_block_processed(rc, node->bytenr, blocksize);
2894	}
2895	node->processed = 1;
2896}
2897
2898/*
2899 * mark a block and all blocks directly/indirectly reference the block
2900 * as processed.
2901 */
2902static void update_processed_blocks(struct reloc_control *rc,
2903				    struct backref_node *node)
2904{
2905	struct backref_node *next = node;
2906	struct backref_edge *edge;
2907	struct backref_edge *edges[BTRFS_MAX_LEVEL - 1];
2908	int index = 0;
2909
2910	while (next) {
2911		cond_resched();
2912		while (1) {
2913			if (next->processed)
2914				break;
2915
2916			__mark_block_processed(rc, next);
2917
2918			if (list_empty(&next->upper))
2919				break;
2920
2921			edge = list_entry(next->upper.next,
2922					  struct backref_edge, list[LOWER]);
2923			edges[index++] = edge;
2924			next = edge->node[UPPER];
2925		}
2926		next = walk_down_backref(edges, &index);
2927	}
2928}
2929
2930static int tree_block_processed(u64 bytenr, struct reloc_control *rc)
2931{
2932	u32 blocksize = rc->extent_root->fs_info->nodesize;
2933
2934	if (test_range_bit(&rc->processed_blocks, bytenr,
2935			   bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL))
2936		return 1;
2937	return 0;
2938}
2939
2940static int get_tree_block_key(struct btrfs_fs_info *fs_info,
2941			      struct tree_block *block)
2942{
 
 
 
 
 
2943	struct extent_buffer *eb;
2944
2945	BUG_ON(block->key_ready);
2946	eb = read_tree_block(fs_info, block->bytenr, block->key.offset,
2947			     block->level, NULL);
2948	if (IS_ERR(eb)) {
2949		return PTR_ERR(eb);
2950	} else if (!extent_buffer_uptodate(eb)) {
2951		free_extent_buffer(eb);
2952		return -EIO;
2953	}
2954	WARN_ON(btrfs_header_level(eb) != block->level);
2955	if (block->level == 0)
2956		btrfs_item_key_to_cpu(eb, &block->key, 0);
2957	else
2958		btrfs_node_key_to_cpu(eb, &block->key, 0);
2959	free_extent_buffer(eb);
2960	block->key_ready = 1;
2961	return 0;
2962}
2963
2964/*
2965 * helper function to relocate a tree block
2966 */
2967static int relocate_tree_block(struct btrfs_trans_handle *trans,
2968				struct reloc_control *rc,
2969				struct backref_node *node,
2970				struct btrfs_key *key,
2971				struct btrfs_path *path)
2972{
2973	struct btrfs_root *root;
2974	int ret = 0;
2975
2976	if (!node)
2977		return 0;
2978
 
 
 
 
 
 
 
 
2979	BUG_ON(node->processed);
2980	root = select_one_root(node);
2981	if (root == ERR_PTR(-ENOENT)) {
2982		update_processed_blocks(rc, node);
 
 
 
 
 
 
 
2983		goto out;
2984	}
2985
2986	if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2987		ret = reserve_metadata_space(trans, rc, node);
2988		if (ret)
2989			goto out;
2990	}
2991
2992	if (root) {
2993		if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) {
2994			BUG_ON(node->new_bytenr);
2995			BUG_ON(!list_empty(&node->list));
2996			btrfs_record_root_in_trans(trans, root);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2997			root = root->reloc_root;
2998			node->new_bytenr = root->node->start;
2999			node->root = root;
 
 
3000			list_add_tail(&node->list, &rc->backref_cache.changed);
3001		} else {
3002			path->lowest_level = node->level;
 
 
3003			ret = btrfs_search_slot(trans, root, key, path, 0, 1);
3004			btrfs_release_path(path);
 
 
3005			if (ret > 0)
3006				ret = 0;
3007		}
3008		if (!ret)
3009			update_processed_blocks(rc, node);
3010	} else {
3011		ret = do_relocation(trans, rc, node, key, path, 1);
3012	}
3013out:
3014	if (ret || node->level == 0 || node->cowonly)
3015		remove_backref_node(&rc->backref_cache, node);
3016	return ret;
3017}
3018
3019/*
3020 * relocate a list of blocks
3021 */
3022static noinline_for_stack
3023int relocate_tree_blocks(struct btrfs_trans_handle *trans,
3024			 struct reloc_control *rc, struct rb_root *blocks)
3025{
3026	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3027	struct backref_node *node;
3028	struct btrfs_path *path;
3029	struct tree_block *block;
3030	struct rb_node *rb_node;
3031	int ret;
3032	int err = 0;
3033
3034	path = btrfs_alloc_path();
3035	if (!path) {
3036		err = -ENOMEM;
3037		goto out_free_blocks;
3038	}
3039
3040	rb_node = rb_first(blocks);
3041	while (rb_node) {
3042		block = rb_entry(rb_node, struct tree_block, rb_node);
3043		if (!block->key_ready)
3044			readahead_tree_block(fs_info, block->bytenr);
3045		rb_node = rb_next(rb_node);
 
3046	}
3047
3048	rb_node = rb_first(blocks);
3049	while (rb_node) {
3050		block = rb_entry(rb_node, struct tree_block, rb_node);
3051		if (!block->key_ready) {
3052			err = get_tree_block_key(fs_info, block);
3053			if (err)
3054				goto out_free_path;
3055		}
3056		rb_node = rb_next(rb_node);
3057	}
3058
3059	rb_node = rb_first(blocks);
3060	while (rb_node) {
3061		block = rb_entry(rb_node, struct tree_block, rb_node);
3062
3063		node = build_backref_tree(rc, &block->key,
3064					  block->level, block->bytenr);
3065		if (IS_ERR(node)) {
3066			err = PTR_ERR(node);
3067			goto out;
3068		}
3069
3070		ret = relocate_tree_block(trans, rc, node, &block->key,
3071					  path);
3072		if (ret < 0) {
3073			if (ret != -EAGAIN || rb_node == rb_first(blocks))
3074				err = ret;
3075			goto out;
3076		}
3077		rb_node = rb_next(rb_node);
3078	}
3079out:
3080	err = finish_pending_nodes(trans, rc, path, err);
3081
3082out_free_path:
3083	btrfs_free_path(path);
3084out_free_blocks:
3085	free_block_list(blocks);
3086	return err;
3087}
3088
3089static noinline_for_stack
3090int prealloc_file_extent_cluster(struct inode *inode,
3091				 struct file_extent_cluster *cluster)
3092{
3093	u64 alloc_hint = 0;
3094	u64 start;
3095	u64 end;
3096	u64 offset = BTRFS_I(inode)->index_cnt;
3097	u64 num_bytes;
3098	int nr = 0;
3099	int ret = 0;
 
3100	u64 prealloc_start = cluster->start - offset;
3101	u64 prealloc_end = cluster->end - offset;
3102	u64 cur_offset;
3103	struct extent_changeset *data_reserved = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3104
3105	BUG_ON(cluster->start != cluster->boundary[0]);
3106	inode_lock(inode);
 
 
 
3107
3108	ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start,
3109					  prealloc_end + 1 - prealloc_start);
3110	if (ret)
3111		goto out;
3112
3113	cur_offset = prealloc_start;
3114	while (nr < cluster->nr) {
3115		start = cluster->boundary[nr] - offset;
3116		if (nr + 1 < cluster->nr)
3117			end = cluster->boundary[nr + 1] - 1 - offset;
3118		else
3119			end = cluster->end - offset;
3120
3121		lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3122		num_bytes = end + 1 - start;
3123		if (cur_offset < start)
3124			btrfs_free_reserved_data_space(inode, data_reserved,
3125					cur_offset, start - cur_offset);
3126		ret = btrfs_prealloc_file_range(inode, 0, start,
3127						num_bytes, num_bytes,
3128						end + 1, &alloc_hint);
3129		cur_offset = end + 1;
3130		unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
3131		if (ret)
3132			break;
3133		nr++;
3134	}
 
 
3135	if (cur_offset < prealloc_end)
3136		btrfs_free_reserved_data_space(inode, data_reserved,
3137				cur_offset, prealloc_end + 1 - cur_offset);
3138out:
3139	inode_unlock(inode);
3140	extent_changeset_free(data_reserved);
3141	return ret;
3142}
3143
3144static noinline_for_stack
3145int setup_extent_mapping(struct inode *inode, u64 start, u64 end,
3146			 u64 block_start)
3147{
3148	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3149	struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3150	struct extent_map *em;
 
3151	int ret = 0;
3152
3153	em = alloc_extent_map();
3154	if (!em)
3155		return -ENOMEM;
3156
3157	em->start = start;
3158	em->len = end + 1 - start;
3159	em->block_len = em->len;
3160	em->block_start = block_start;
3161	em->bdev = fs_info->fs_devices->latest_bdev;
3162	set_bit(EXTENT_FLAG_PINNED, &em->flags);
3163
3164	lock_extent(&BTRFS_I(inode)->io_tree, start, end);
3165	while (1) {
3166		write_lock(&em_tree->lock);
3167		ret = add_extent_mapping(em_tree, em, 0);
3168		write_unlock(&em_tree->lock);
3169		if (ret != -EEXIST) {
3170			free_extent_map(em);
3171			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3172		}
3173		btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0);
3174	}
3175	unlock_extent(&BTRFS_I(inode)->io_tree, start, end);
 
 
 
 
 
 
 
 
 
 
 
3176	return ret;
3177}
3178
3179static int relocate_file_extent_cluster(struct inode *inode,
3180					struct file_extent_cluster *cluster)
3181{
3182	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
3183	u64 page_start;
3184	u64 page_end;
3185	u64 offset = BTRFS_I(inode)->index_cnt;
3186	unsigned long index;
3187	unsigned long last_index;
3188	struct page *page;
3189	struct file_ra_state *ra;
3190	gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping);
3191	int nr = 0;
3192	int ret = 0;
3193
3194	if (!cluster->nr)
3195		return 0;
3196
3197	ra = kzalloc(sizeof(*ra), GFP_NOFS);
3198	if (!ra)
3199		return -ENOMEM;
3200
3201	ret = prealloc_file_extent_cluster(inode, cluster);
3202	if (ret)
3203		goto out;
3204
3205	file_ra_state_init(ra, inode->i_mapping);
3206
3207	ret = setup_extent_mapping(inode, cluster->start - offset,
3208				   cluster->end - offset, cluster->start);
3209	if (ret)
3210		goto out;
3211
3212	index = (cluster->start - offset) >> PAGE_SHIFT;
3213	last_index = (cluster->end - offset) >> PAGE_SHIFT;
3214	while (index <= last_index) {
3215		ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode),
3216				PAGE_SIZE);
3217		if (ret)
3218			goto out;
3219
3220		page = find_lock_page(inode->i_mapping, index);
3221		if (!page) {
3222			page_cache_sync_readahead(inode->i_mapping,
3223						  ra, NULL, index,
3224						  last_index + 1 - index);
3225			page = find_or_create_page(inode->i_mapping, index,
3226						   mask);
3227			if (!page) {
3228				btrfs_delalloc_release_metadata(BTRFS_I(inode),
3229							PAGE_SIZE, true);
3230				ret = -ENOMEM;
3231				goto out;
3232			}
3233		}
3234
3235		if (PageReadahead(page)) {
3236			page_cache_async_readahead(inode->i_mapping,
3237						   ra, NULL, page, index,
3238						   last_index + 1 - index);
3239		}
3240
3241		if (!PageUptodate(page)) {
3242			btrfs_readpage(NULL, page);
3243			lock_page(page);
3244			if (!PageUptodate(page)) {
3245				unlock_page(page);
3246				put_page(page);
3247				btrfs_delalloc_release_metadata(BTRFS_I(inode),
3248							PAGE_SIZE, true);
3249				btrfs_delalloc_release_extents(BTRFS_I(inode),
3250							       PAGE_SIZE, true);
3251				ret = -EIO;
3252				goto out;
3253			}
3254		}
3255
3256		page_start = page_offset(page);
3257		page_end = page_start + PAGE_SIZE - 1;
3258
3259		lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end);
3260
3261		set_page_extent_mapped(page);
3262
3263		if (nr < cluster->nr &&
3264		    page_start + offset == cluster->boundary[nr]) {
3265			set_extent_bits(&BTRFS_I(inode)->io_tree,
3266					page_start, page_end,
3267					EXTENT_BOUNDARY);
3268			nr++;
3269		}
3270
3271		ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0,
3272						NULL, 0);
3273		if (ret) {
3274			unlock_page(page);
3275			put_page(page);
3276			btrfs_delalloc_release_metadata(BTRFS_I(inode),
3277							 PAGE_SIZE, true);
3278			btrfs_delalloc_release_extents(BTRFS_I(inode),
3279			                               PAGE_SIZE, true);
3280
3281			clear_extent_bits(&BTRFS_I(inode)->io_tree,
3282					  page_start, page_end,
3283					  EXTENT_LOCKED | EXTENT_BOUNDARY);
3284			goto out;
3285
3286		}
3287		set_page_dirty(page);
3288
3289		unlock_extent(&BTRFS_I(inode)->io_tree,
3290			      page_start, page_end);
3291		unlock_page(page);
3292		put_page(page);
3293
3294		index++;
3295		btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE,
3296					       false);
3297		balance_dirty_pages_ratelimited(inode->i_mapping);
3298		btrfs_throttle(fs_info);
3299	}
3300	WARN_ON(nr != cluster->nr);
3301out:
3302	kfree(ra);
3303	return ret;
3304}
3305
3306static noinline_for_stack
3307int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
3308			 struct file_extent_cluster *cluster)
3309{
3310	int ret;
3311
3312	if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
3313		ret = relocate_file_extent_cluster(inode, cluster);
3314		if (ret)
3315			return ret;
3316		cluster->nr = 0;
3317	}
3318
3319	if (!cluster->nr)
3320		cluster->start = extent_key->objectid;
3321	else
3322		BUG_ON(cluster->nr >= MAX_EXTENTS);
3323	cluster->end = extent_key->objectid + extent_key->offset - 1;
3324	cluster->boundary[cluster->nr] = extent_key->objectid;
3325	cluster->nr++;
3326
3327	if (cluster->nr >= MAX_EXTENTS) {
3328		ret = relocate_file_extent_cluster(inode, cluster);
3329		if (ret)
3330			return ret;
3331		cluster->nr = 0;
3332	}
3333	return 0;
3334}
3335
3336#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3337static int get_ref_objectid_v0(struct reloc_control *rc,
3338			       struct btrfs_path *path,
3339			       struct btrfs_key *extent_key,
3340			       u64 *ref_objectid, int *path_change)
3341{
3342	struct btrfs_key key;
3343	struct extent_buffer *leaf;
3344	struct btrfs_extent_ref_v0 *ref0;
3345	int ret;
3346	int slot;
3347
3348	leaf = path->nodes[0];
3349	slot = path->slots[0];
3350	while (1) {
3351		if (slot >= btrfs_header_nritems(leaf)) {
3352			ret = btrfs_next_leaf(rc->extent_root, path);
3353			if (ret < 0)
3354				return ret;
3355			BUG_ON(ret > 0);
3356			leaf = path->nodes[0];
3357			slot = path->slots[0];
3358			if (path_change)
3359				*path_change = 1;
3360		}
3361		btrfs_item_key_to_cpu(leaf, &key, slot);
3362		if (key.objectid != extent_key->objectid)
3363			return -ENOENT;
3364
3365		if (key.type != BTRFS_EXTENT_REF_V0_KEY) {
3366			slot++;
3367			continue;
3368		}
3369		ref0 = btrfs_item_ptr(leaf, slot,
3370				struct btrfs_extent_ref_v0);
3371		*ref_objectid = btrfs_ref_objectid_v0(leaf, ref0);
3372		break;
3373	}
3374	return 0;
3375}
3376#endif
3377
3378/*
3379 * helper to add a tree block to the list.
3380 * the major work is getting the generation and level of the block
3381 */
3382static int add_tree_block(struct reloc_control *rc,
3383			  struct btrfs_key *extent_key,
3384			  struct btrfs_path *path,
3385			  struct rb_root *blocks)
3386{
3387	struct extent_buffer *eb;
3388	struct btrfs_extent_item *ei;
3389	struct btrfs_tree_block_info *bi;
3390	struct tree_block *block;
3391	struct rb_node *rb_node;
3392	u32 item_size;
3393	int level = -1;
3394	u64 generation;
 
3395
3396	eb =  path->nodes[0];
3397	item_size = btrfs_item_size_nr(eb, path->slots[0]);
3398
3399	if (extent_key->type == BTRFS_METADATA_ITEM_KEY ||
3400	    item_size >= sizeof(*ei) + sizeof(*bi)) {
 
 
3401		ei = btrfs_item_ptr(eb, path->slots[0],
3402				struct btrfs_extent_item);
 
3403		if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) {
3404			bi = (struct btrfs_tree_block_info *)(ei + 1);
3405			level = btrfs_tree_block_level(eb, bi);
 
3406		} else {
3407			level = (int)extent_key->offset;
 
3408		}
3409		generation = btrfs_extent_generation(eb, ei);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3410	} else {
3411#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3412		u64 ref_owner;
3413		int ret;
3414
3415		BUG_ON(item_size != sizeof(struct btrfs_extent_item_v0));
3416		ret = get_ref_objectid_v0(rc, path, extent_key,
3417					  &ref_owner, NULL);
3418		if (ret < 0)
3419			return ret;
3420		BUG_ON(ref_owner >= BTRFS_MAX_LEVEL);
3421		level = (int)ref_owner;
3422		/* FIXME: get real generation */
3423		generation = 0;
3424#else
3425		BUG();
3426#endif
3427	}
3428
3429	btrfs_release_path(path);
3430
3431	BUG_ON(level == -1);
3432
3433	block = kmalloc(sizeof(*block), GFP_NOFS);
3434	if (!block)
3435		return -ENOMEM;
3436
3437	block->bytenr = extent_key->objectid;
3438	block->key.objectid = rc->extent_root->fs_info->nodesize;
3439	block->key.offset = generation;
3440	block->level = level;
3441	block->key_ready = 0;
 
3442
3443	rb_node = tree_insert(blocks, block->bytenr, &block->rb_node);
3444	if (rb_node)
3445		backref_tree_panic(rb_node, -EEXIST, block->bytenr);
 
3446
3447	return 0;
3448}
3449
3450/*
3451 * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY
3452 */
3453static int __add_tree_block(struct reloc_control *rc,
3454			    u64 bytenr, u32 blocksize,
3455			    struct rb_root *blocks)
3456{
3457	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3458	struct btrfs_path *path;
3459	struct btrfs_key key;
3460	int ret;
3461	bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA);
3462
3463	if (tree_block_processed(bytenr, rc))
3464		return 0;
3465
3466	if (tree_search(blocks, bytenr))
3467		return 0;
3468
3469	path = btrfs_alloc_path();
3470	if (!path)
3471		return -ENOMEM;
3472again:
3473	key.objectid = bytenr;
3474	if (skinny) {
3475		key.type = BTRFS_METADATA_ITEM_KEY;
3476		key.offset = (u64)-1;
3477	} else {
3478		key.type = BTRFS_EXTENT_ITEM_KEY;
3479		key.offset = blocksize;
3480	}
3481
3482	path->search_commit_root = 1;
3483	path->skip_locking = 1;
3484	ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0);
3485	if (ret < 0)
3486		goto out;
3487
3488	if (ret > 0 && skinny) {
3489		if (path->slots[0]) {
3490			path->slots[0]--;
3491			btrfs_item_key_to_cpu(path->nodes[0], &key,
3492					      path->slots[0]);
3493			if (key.objectid == bytenr &&
3494			    (key.type == BTRFS_METADATA_ITEM_KEY ||
3495			     (key.type == BTRFS_EXTENT_ITEM_KEY &&
3496			      key.offset == blocksize)))
3497				ret = 0;
3498		}
3499
3500		if (ret) {
3501			skinny = false;
3502			btrfs_release_path(path);
3503			goto again;
3504		}
3505	}
3506	if (ret) {
3507		ASSERT(ret == 1);
3508		btrfs_print_leaf(path->nodes[0]);
3509		btrfs_err(fs_info,
3510	     "tree block extent item (%llu) is not found in extent tree",
3511		     bytenr);
3512		WARN_ON(1);
3513		ret = -EINVAL;
3514		goto out;
3515	}
3516
3517	ret = add_tree_block(rc, &key, path, blocks);
3518out:
3519	btrfs_free_path(path);
3520	return ret;
3521}
3522
3523/*
3524 * helper to check if the block use full backrefs for pointers in it
3525 */
3526static int block_use_full_backref(struct reloc_control *rc,
3527				  struct extent_buffer *eb)
3528{
3529	u64 flags;
3530	int ret;
3531
3532	if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) ||
3533	    btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV)
3534		return 1;
3535
3536	ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info,
3537				       eb->start, btrfs_header_level(eb), 1,
3538				       NULL, &flags);
3539	BUG_ON(ret);
3540
3541	if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)
3542		ret = 1;
3543	else
3544		ret = 0;
3545	return ret;
3546}
3547
3548static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
3549				    struct btrfs_block_group_cache *block_group,
3550				    struct inode *inode,
3551				    u64 ino)
3552{
3553	struct btrfs_key key;
3554	struct btrfs_root *root = fs_info->tree_root;
3555	struct btrfs_trans_handle *trans;
3556	int ret = 0;
3557
3558	if (inode)
3559		goto truncate;
3560
3561	key.objectid = ino;
3562	key.type = BTRFS_INODE_ITEM_KEY;
3563	key.offset = 0;
3564
3565	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
3566	if (IS_ERR(inode) || is_bad_inode(inode)) {
3567		if (!IS_ERR(inode))
3568			iput(inode);
3569		return -ENOENT;
3570	}
3571
3572truncate:
3573	ret = btrfs_check_trunc_cache_free_space(fs_info,
3574						 &fs_info->global_block_rsv);
3575	if (ret)
3576		goto out;
3577
3578	trans = btrfs_join_transaction(root);
3579	if (IS_ERR(trans)) {
3580		ret = PTR_ERR(trans);
3581		goto out;
3582	}
3583
3584	ret = btrfs_truncate_free_space_cache(trans, block_group, inode);
3585
3586	btrfs_end_transaction(trans);
3587	btrfs_btree_balance_dirty(fs_info);
3588out:
3589	iput(inode);
3590	return ret;
3591}
3592
3593/*
3594 * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
3595 * this function scans fs tree to find blocks reference the data extent
3596 */
3597static int find_data_references(struct reloc_control *rc,
3598				struct btrfs_key *extent_key,
3599				struct extent_buffer *leaf,
3600				struct btrfs_extent_data_ref *ref,
3601				struct rb_root *blocks)
3602{
3603	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3604	struct btrfs_path *path;
3605	struct tree_block *block;
3606	struct btrfs_root *root;
3607	struct btrfs_file_extent_item *fi;
3608	struct rb_node *rb_node;
3609	struct btrfs_key key;
3610	u64 ref_root;
3611	u64 ref_objectid;
3612	u64 ref_offset;
3613	u32 ref_count;
3614	u32 nritems;
3615	int err = 0;
3616	int added = 0;
3617	int counted;
3618	int ret;
3619
3620	ref_root = btrfs_extent_data_ref_root(leaf, ref);
3621	ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
3622	ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
3623	ref_count = btrfs_extent_data_ref_count(leaf, ref);
3624
3625	/*
3626	 * This is an extent belonging to the free space cache, lets just delete
3627	 * it and redo the search.
3628	 */
3629	if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
3630		ret = delete_block_group_cache(fs_info, rc->block_group,
3631					       NULL, ref_objectid);
3632		if (ret != -ENOENT)
3633			return ret;
3634		ret = 0;
3635	}
3636
3637	path = btrfs_alloc_path();
3638	if (!path)
3639		return -ENOMEM;
3640	path->reada = READA_FORWARD;
3641
3642	root = read_fs_root(fs_info, ref_root);
3643	if (IS_ERR(root)) {
3644		err = PTR_ERR(root);
3645		goto out;
3646	}
3647
3648	key.objectid = ref_objectid;
3649	key.type = BTRFS_EXTENT_DATA_KEY;
3650	if (ref_offset > ((u64)-1 << 32))
3651		key.offset = 0;
3652	else
3653		key.offset = ref_offset;
3654
3655	path->search_commit_root = 1;
3656	path->skip_locking = 1;
3657	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3658	if (ret < 0) {
3659		err = ret;
3660		goto out;
3661	}
3662
3663	leaf = path->nodes[0];
3664	nritems = btrfs_header_nritems(leaf);
3665	/*
3666	 * the references in tree blocks that use full backrefs
3667	 * are not counted in
3668	 */
3669	if (block_use_full_backref(rc, leaf))
3670		counted = 0;
3671	else
3672		counted = 1;
3673	rb_node = tree_search(blocks, leaf->start);
3674	if (rb_node) {
3675		if (counted)
3676			added = 1;
3677		else
3678			path->slots[0] = nritems;
3679	}
3680
3681	while (ref_count > 0) {
3682		while (path->slots[0] >= nritems) {
3683			ret = btrfs_next_leaf(root, path);
3684			if (ret < 0) {
3685				err = ret;
3686				goto out;
3687			}
3688			if (WARN_ON(ret > 0))
3689				goto out;
3690
3691			leaf = path->nodes[0];
3692			nritems = btrfs_header_nritems(leaf);
3693			added = 0;
3694
3695			if (block_use_full_backref(rc, leaf))
3696				counted = 0;
3697			else
3698				counted = 1;
3699			rb_node = tree_search(blocks, leaf->start);
3700			if (rb_node) {
3701				if (counted)
3702					added = 1;
3703				else
3704					path->slots[0] = nritems;
3705			}
3706		}
3707
3708		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3709		if (WARN_ON(key.objectid != ref_objectid ||
3710		    key.type != BTRFS_EXTENT_DATA_KEY))
3711			break;
3712
3713		fi = btrfs_item_ptr(leaf, path->slots[0],
3714				    struct btrfs_file_extent_item);
3715
3716		if (btrfs_file_extent_type(leaf, fi) ==
3717		    BTRFS_FILE_EXTENT_INLINE)
3718			goto next;
3719
3720		if (btrfs_file_extent_disk_bytenr(leaf, fi) !=
3721		    extent_key->objectid)
3722			goto next;
3723
3724		key.offset -= btrfs_file_extent_offset(leaf, fi);
3725		if (key.offset != ref_offset)
3726			goto next;
3727
3728		if (counted)
3729			ref_count--;
3730		if (added)
3731			goto next;
3732
3733		if (!tree_block_processed(leaf->start, rc)) {
3734			block = kmalloc(sizeof(*block), GFP_NOFS);
3735			if (!block) {
3736				err = -ENOMEM;
3737				break;
3738			}
3739			block->bytenr = leaf->start;
3740			btrfs_item_key_to_cpu(leaf, &block->key, 0);
3741			block->level = 0;
3742			block->key_ready = 1;
3743			rb_node = tree_insert(blocks, block->bytenr,
3744					      &block->rb_node);
3745			if (rb_node)
3746				backref_tree_panic(rb_node, -EEXIST,
3747						   block->bytenr);
3748		}
3749		if (counted)
3750			added = 1;
3751		else
3752			path->slots[0] = nritems;
3753next:
3754		path->slots[0]++;
3755
3756	}
3757out:
3758	btrfs_free_path(path);
3759	return err;
 
 
3760}
3761
3762/*
3763 * helper to find all tree blocks that reference a given data extent
3764 */
3765static noinline_for_stack
3766int add_data_references(struct reloc_control *rc,
3767			struct btrfs_key *extent_key,
3768			struct btrfs_path *path,
3769			struct rb_root *blocks)
3770{
3771	struct btrfs_key key;
3772	struct extent_buffer *eb;
3773	struct btrfs_extent_data_ref *dref;
3774	struct btrfs_extent_inline_ref *iref;
3775	unsigned long ptr;
3776	unsigned long end;
3777	u32 blocksize = rc->extent_root->fs_info->nodesize;
3778	int ret = 0;
3779	int err = 0;
3780
3781	eb = path->nodes[0];
3782	ptr = btrfs_item_ptr_offset(eb, path->slots[0]);
3783	end = ptr + btrfs_item_size_nr(eb, path->slots[0]);
3784#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3785	if (ptr + sizeof(struct btrfs_extent_item_v0) == end)
3786		ptr = end;
3787	else
3788#endif
3789		ptr += sizeof(struct btrfs_extent_item);
3790
3791	while (ptr < end) {
3792		iref = (struct btrfs_extent_inline_ref *)ptr;
3793		key.type = btrfs_get_extent_inline_ref_type(eb, iref,
3794							BTRFS_REF_TYPE_DATA);
3795		if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3796			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3797			ret = __add_tree_block(rc, key.offset, blocksize,
3798					       blocks);
3799		} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3800			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
3801			ret = find_data_references(rc, extent_key,
3802						   eb, dref, blocks);
3803		} else {
3804			ret = -EINVAL;
3805			btrfs_err(rc->extent_root->fs_info,
3806		     "extent %llu slot %d has an invalid inline ref type",
3807			     eb->start, path->slots[0]);
3808		}
3809		if (ret) {
3810			err = ret;
3811			goto out;
3812		}
3813		ptr += btrfs_extent_inline_ref_size(key.type);
3814	}
3815	WARN_ON(ptr > end);
3816
3817	while (1) {
3818		cond_resched();
3819		eb = path->nodes[0];
3820		if (path->slots[0] >= btrfs_header_nritems(eb)) {
3821			ret = btrfs_next_leaf(rc->extent_root, path);
3822			if (ret < 0) {
3823				err = ret;
3824				break;
3825			}
3826			if (ret > 0)
3827				break;
3828			eb = path->nodes[0];
3829		}
3830
3831		btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
3832		if (key.objectid != extent_key->objectid)
 
3833			break;
3834
3835#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
3836		if (key.type == BTRFS_SHARED_DATA_REF_KEY ||
3837		    key.type == BTRFS_EXTENT_REF_V0_KEY) {
3838#else
3839		BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY);
3840		if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
3841#endif
3842			ret = __add_tree_block(rc, key.offset, blocksize,
3843					       blocks);
3844		} else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
3845			dref = btrfs_item_ptr(eb, path->slots[0],
3846					      struct btrfs_extent_data_ref);
3847			ret = find_data_references(rc, extent_key,
3848						   eb, dref, blocks);
3849		} else {
3850			ret = 0;
3851		}
3852		if (ret) {
3853			err = ret;
 
 
 
 
 
3854			break;
3855		}
3856		path->slots[0]++;
3857	}
3858out:
3859	btrfs_release_path(path);
3860	if (err)
3861		free_block_list(blocks);
3862	return err;
 
3863}
3864
3865/*
3866 * helper to find next unprocessed extent
3867 */
3868static noinline_for_stack
3869int find_next_extent(struct reloc_control *rc, struct btrfs_path *path,
3870		     struct btrfs_key *extent_key)
3871{
3872	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3873	struct btrfs_key key;
3874	struct extent_buffer *leaf;
3875	u64 start, end, last;
3876	int ret;
3877
3878	last = rc->block_group->key.objectid + rc->block_group->key.offset;
3879	while (1) {
3880		cond_resched();
3881		if (rc->search_start >= last) {
3882			ret = 1;
3883			break;
3884		}
3885
3886		key.objectid = rc->search_start;
3887		key.type = BTRFS_EXTENT_ITEM_KEY;
3888		key.offset = 0;
3889
3890		path->search_commit_root = 1;
3891		path->skip_locking = 1;
3892		ret = btrfs_search_slot(NULL, rc->extent_root, &key, path,
3893					0, 0);
3894		if (ret < 0)
3895			break;
3896next:
3897		leaf = path->nodes[0];
3898		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
3899			ret = btrfs_next_leaf(rc->extent_root, path);
3900			if (ret != 0)
3901				break;
3902			leaf = path->nodes[0];
3903		}
3904
3905		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3906		if (key.objectid >= last) {
3907			ret = 1;
3908			break;
3909		}
3910
3911		if (key.type != BTRFS_EXTENT_ITEM_KEY &&
3912		    key.type != BTRFS_METADATA_ITEM_KEY) {
3913			path->slots[0]++;
3914			goto next;
3915		}
3916
3917		if (key.type == BTRFS_EXTENT_ITEM_KEY &&
3918		    key.objectid + key.offset <= rc->search_start) {
3919			path->slots[0]++;
3920			goto next;
3921		}
3922
3923		if (key.type == BTRFS_METADATA_ITEM_KEY &&
3924		    key.objectid + fs_info->nodesize <=
3925		    rc->search_start) {
3926			path->slots[0]++;
3927			goto next;
3928		}
3929
3930		ret = find_first_extent_bit(&rc->processed_blocks,
3931					    key.objectid, &start, &end,
3932					    EXTENT_DIRTY, NULL);
3933
3934		if (ret == 0 && start <= key.objectid) {
3935			btrfs_release_path(path);
3936			rc->search_start = end + 1;
3937		} else {
3938			if (key.type == BTRFS_EXTENT_ITEM_KEY)
3939				rc->search_start = key.objectid + key.offset;
3940			else
3941				rc->search_start = key.objectid +
3942					fs_info->nodesize;
3943			memcpy(extent_key, &key, sizeof(key));
3944			return 0;
3945		}
3946	}
3947	btrfs_release_path(path);
3948	return ret;
3949}
3950
3951static void set_reloc_control(struct reloc_control *rc)
3952{
3953	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3954
3955	mutex_lock(&fs_info->reloc_mutex);
3956	fs_info->reloc_ctl = rc;
3957	mutex_unlock(&fs_info->reloc_mutex);
3958}
3959
3960static void unset_reloc_control(struct reloc_control *rc)
3961{
3962	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
3963
3964	mutex_lock(&fs_info->reloc_mutex);
3965	fs_info->reloc_ctl = NULL;
3966	mutex_unlock(&fs_info->reloc_mutex);
3967}
3968
3969static int check_extent_flags(u64 flags)
3970{
3971	if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3972	    (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3973		return 1;
3974	if (!(flags & BTRFS_EXTENT_FLAG_DATA) &&
3975	    !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK))
3976		return 1;
3977	if ((flags & BTRFS_EXTENT_FLAG_DATA) &&
3978	    (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF))
3979		return 1;
3980	return 0;
3981}
3982
3983static noinline_for_stack
3984int prepare_to_relocate(struct reloc_control *rc)
3985{
3986	struct btrfs_trans_handle *trans;
3987	int ret;
3988
3989	rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info,
3990					      BTRFS_BLOCK_RSV_TEMP);
3991	if (!rc->block_rsv)
3992		return -ENOMEM;
3993
3994	memset(&rc->cluster, 0, sizeof(rc->cluster));
3995	rc->search_start = rc->block_group->key.objectid;
3996	rc->extents_found = 0;
3997	rc->nodes_relocated = 0;
3998	rc->merging_rsv_size = 0;
3999	rc->reserved_bytes = 0;
4000	rc->block_rsv->size = rc->extent_root->fs_info->nodesize *
4001			      RELOCATION_RESERVED_NODES;
4002	ret = btrfs_block_rsv_refill(rc->extent_root,
4003				     rc->block_rsv, rc->block_rsv->size,
4004				     BTRFS_RESERVE_FLUSH_ALL);
4005	if (ret)
4006		return ret;
4007
4008	rc->create_reloc_tree = 1;
4009	set_reloc_control(rc);
4010
4011	trans = btrfs_join_transaction(rc->extent_root);
4012	if (IS_ERR(trans)) {
4013		unset_reloc_control(rc);
4014		/*
4015		 * extent tree is not a ref_cow tree and has no reloc_root to
4016		 * cleanup.  And callers are responsible to free the above
4017		 * block rsv.
4018		 */
4019		return PTR_ERR(trans);
4020	}
4021	btrfs_commit_transaction(trans);
4022	return 0;
 
 
 
 
4023}
4024
4025static noinline_for_stack int relocate_block_group(struct reloc_control *rc)
4026{
4027	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
4028	struct rb_root blocks = RB_ROOT;
4029	struct btrfs_key key;
4030	struct btrfs_trans_handle *trans = NULL;
4031	struct btrfs_path *path;
4032	struct btrfs_extent_item *ei;
4033	u64 flags;
4034	u32 item_size;
4035	int ret;
4036	int err = 0;
4037	int progress = 0;
4038
4039	path = btrfs_alloc_path();
4040	if (!path)
4041		return -ENOMEM;
4042	path->reada = READA_FORWARD;
4043
4044	ret = prepare_to_relocate(rc);
4045	if (ret) {
4046		err = ret;
4047		goto out_free;
4048	}
4049
4050	while (1) {
4051		rc->reserved_bytes = 0;
4052		ret = btrfs_block_rsv_refill(rc->extent_root,
4053					rc->block_rsv, rc->block_rsv->size,
4054					BTRFS_RESERVE_FLUSH_ALL);
4055		if (ret) {
4056			err = ret;
4057			break;
4058		}
4059		progress++;
4060		trans = btrfs_start_transaction(rc->extent_root, 0);
4061		if (IS_ERR(trans)) {
4062			err = PTR_ERR(trans);
4063			trans = NULL;
4064			break;
4065		}
4066restart:
4067		if (update_backref_cache(trans, &rc->backref_cache)) {
4068			btrfs_end_transaction(trans);
 
4069			continue;
4070		}
4071
4072		ret = find_next_extent(rc, path, &key);
4073		if (ret < 0)
4074			err = ret;
4075		if (ret != 0)
4076			break;
4077
4078		rc->extents_found++;
4079
4080		ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
4081				    struct btrfs_extent_item);
4082		item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]);
4083		if (item_size >= sizeof(*ei)) {
4084			flags = btrfs_extent_flags(path->nodes[0], ei);
4085			ret = check_extent_flags(flags);
4086			BUG_ON(ret);
4087
4088		} else {
4089#ifdef BTRFS_COMPAT_EXTENT_TREE_V0
4090			u64 ref_owner;
4091			int path_change = 0;
4092
4093			BUG_ON(item_size !=
4094			       sizeof(struct btrfs_extent_item_v0));
4095			ret = get_ref_objectid_v0(rc, path, &key, &ref_owner,
4096						  &path_change);
4097			if (ret < 0) {
4098				err = ret;
4099				break;
4100			}
4101			if (ref_owner < BTRFS_FIRST_FREE_OBJECTID)
4102				flags = BTRFS_EXTENT_FLAG_TREE_BLOCK;
4103			else
4104				flags = BTRFS_EXTENT_FLAG_DATA;
4105
4106			if (path_change) {
4107				btrfs_release_path(path);
4108
4109				path->search_commit_root = 1;
4110				path->skip_locking = 1;
4111				ret = btrfs_search_slot(NULL, rc->extent_root,
4112							&key, path, 0, 0);
4113				if (ret < 0) {
4114					err = ret;
4115					break;
4116				}
4117				BUG_ON(ret > 0);
4118			}
4119#else
4120			BUG();
4121#endif
4122		}
4123
4124		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
4125			ret = add_tree_block(rc, &key, path, &blocks);
4126		} else if (rc->stage == UPDATE_DATA_PTRS &&
4127			   (flags & BTRFS_EXTENT_FLAG_DATA)) {
4128			ret = add_data_references(rc, &key, path, &blocks);
4129		} else {
4130			btrfs_release_path(path);
4131			ret = 0;
4132		}
4133		if (ret < 0) {
4134			err = ret;
4135			break;
4136		}
4137
4138		if (!RB_EMPTY_ROOT(&blocks)) {
4139			ret = relocate_tree_blocks(trans, rc, &blocks);
4140			if (ret < 0) {
4141				/*
4142				 * if we fail to relocate tree blocks, force to update
4143				 * backref cache when committing transaction.
4144				 */
4145				rc->backref_cache.last_trans = trans->transid - 1;
4146
4147				if (ret != -EAGAIN) {
4148					err = ret;
4149					break;
4150				}
4151				rc->extents_found--;
4152				rc->search_start = key.objectid;
4153			}
4154		}
4155
4156		btrfs_end_transaction_throttle(trans);
4157		btrfs_btree_balance_dirty(fs_info);
4158		trans = NULL;
4159
4160		if (rc->stage == MOVE_DATA_EXTENTS &&
4161		    (flags & BTRFS_EXTENT_FLAG_DATA)) {
4162			rc->found_file_extent = 1;
4163			ret = relocate_data_extent(rc->data_inode,
4164						   &key, &rc->cluster);
4165			if (ret < 0) {
4166				err = ret;
4167				break;
4168			}
4169		}
 
 
 
 
4170	}
4171	if (trans && progress && err == -ENOSPC) {
4172		ret = btrfs_force_chunk_alloc(trans, fs_info,
4173					      rc->block_group->flags);
4174		if (ret == 1) {
4175			err = 0;
4176			progress = 0;
4177			goto restart;
4178		}
4179	}
4180
4181	btrfs_release_path(path);
4182	clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY);
4183
4184	if (trans) {
4185		btrfs_end_transaction_throttle(trans);
4186		btrfs_btree_balance_dirty(fs_info);
4187	}
4188
4189	if (!err) {
4190		ret = relocate_file_extent_cluster(rc->data_inode,
4191						   &rc->cluster);
4192		if (ret < 0)
4193			err = ret;
4194	}
4195
4196	rc->create_reloc_tree = 0;
4197	set_reloc_control(rc);
4198
4199	backref_cache_cleanup(&rc->backref_cache);
4200	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4201
 
 
 
 
 
 
 
 
4202	err = prepare_to_merge(rc, err);
4203
4204	merge_reloc_roots(rc);
4205
4206	rc->merge_reloc_tree = 0;
4207	unset_reloc_control(rc);
4208	btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1);
4209
4210	/* get rid of pinned extents */
4211	trans = btrfs_join_transaction(rc->extent_root);
4212	if (IS_ERR(trans)) {
4213		err = PTR_ERR(trans);
4214		goto out_free;
4215	}
4216	btrfs_commit_transaction(trans);
 
 
4217out_free:
 
 
 
4218	btrfs_free_block_rsv(fs_info, rc->block_rsv);
4219	btrfs_free_path(path);
4220	return err;
4221}
4222
4223static int __insert_orphan_inode(struct btrfs_trans_handle *trans,
4224				 struct btrfs_root *root, u64 objectid)
4225{
4226	struct btrfs_path *path;
4227	struct btrfs_inode_item *item;
4228	struct extent_buffer *leaf;
4229	int ret;
4230
4231	path = btrfs_alloc_path();
4232	if (!path)
4233		return -ENOMEM;
4234
4235	ret = btrfs_insert_empty_inode(trans, root, path, objectid);
4236	if (ret)
4237		goto out;
4238
4239	leaf = path->nodes[0];
4240	item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item);
4241	memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item));
4242	btrfs_set_inode_generation(leaf, item, 1);
4243	btrfs_set_inode_size(leaf, item, 0);
4244	btrfs_set_inode_mode(leaf, item, S_IFREG | 0600);
4245	btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS |
4246					  BTRFS_INODE_PREALLOC);
4247	btrfs_mark_buffer_dirty(leaf);
4248out:
4249	btrfs_free_path(path);
4250	return ret;
4251}
4252
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4253/*
4254 * helper to create inode for data relocation.
4255 * the inode is in data relocation tree and its link count is 0
4256 */
4257static noinline_for_stack
4258struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info,
4259				 struct btrfs_block_group_cache *group)
4260{
4261	struct inode *inode = NULL;
4262	struct btrfs_trans_handle *trans;
4263	struct btrfs_root *root;
4264	struct btrfs_key key;
4265	u64 objectid;
4266	int err = 0;
4267
4268	root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4269	if (IS_ERR(root))
4270		return ERR_CAST(root);
4271
4272	trans = btrfs_start_transaction(root, 6);
4273	if (IS_ERR(trans))
 
4274		return ERR_CAST(trans);
 
4275
4276	err = btrfs_find_free_objectid(root, &objectid);
4277	if (err)
4278		goto out;
4279
4280	err = __insert_orphan_inode(trans, root, objectid);
4281	BUG_ON(err);
 
4282
4283	key.objectid = objectid;
4284	key.type = BTRFS_INODE_ITEM_KEY;
4285	key.offset = 0;
4286	inode = btrfs_iget(fs_info->sb, &key, root, NULL);
4287	BUG_ON(IS_ERR(inode) || is_bad_inode(inode));
4288	BTRFS_I(inode)->index_cnt = group->key.objectid;
 
 
4289
4290	err = btrfs_orphan_add(trans, BTRFS_I(inode));
4291out:
 
4292	btrfs_end_transaction(trans);
4293	btrfs_btree_balance_dirty(fs_info);
4294	if (err) {
4295		if (inode)
4296			iput(inode);
4297		inode = ERR_PTR(err);
4298	}
4299	return inode;
4300}
4301
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4302static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info)
4303{
4304	struct reloc_control *rc;
4305
4306	rc = kzalloc(sizeof(*rc), GFP_NOFS);
4307	if (!rc)
4308		return NULL;
4309
4310	INIT_LIST_HEAD(&rc->reloc_roots);
4311	backref_cache_init(&rc->backref_cache);
 
4312	mapping_tree_init(&rc->reloc_root_tree);
4313	extent_io_tree_init(&rc->processed_blocks, NULL);
4314	return rc;
4315}
4316
 
 
 
 
 
 
 
 
 
 
 
 
4317/*
4318 * Print the block group being relocated
4319 */
4320static void describe_relocation(struct btrfs_fs_info *fs_info,
4321				struct btrfs_block_group_cache *block_group)
4322{
4323	char buf[128];		/* prefixed by a '|' that'll be dropped */
4324	u64 flags = block_group->flags;
4325
4326	/* Shouldn't happen */
4327	if (!flags) {
4328		strcpy(buf, "|NONE");
4329	} else {
4330		char *bp = buf;
4331
4332#define DESCRIBE_FLAG(f, d) \
4333		if (flags & BTRFS_BLOCK_GROUP_##f) { \
4334			bp += snprintf(bp, buf - bp + sizeof(buf), "|%s", d); \
4335			flags &= ~BTRFS_BLOCK_GROUP_##f; \
4336		}
4337		DESCRIBE_FLAG(DATA,     "data");
4338		DESCRIBE_FLAG(SYSTEM,   "system");
4339		DESCRIBE_FLAG(METADATA, "metadata");
4340		DESCRIBE_FLAG(RAID0,    "raid0");
4341		DESCRIBE_FLAG(RAID1,    "raid1");
4342		DESCRIBE_FLAG(DUP,      "dup");
4343		DESCRIBE_FLAG(RAID10,   "raid10");
4344		DESCRIBE_FLAG(RAID5,    "raid5");
4345		DESCRIBE_FLAG(RAID6,    "raid6");
4346		if (flags)
4347			snprintf(buf, buf - bp + sizeof(buf), "|0x%llx", flags);
4348#undef DESCRIBE_FLAG
4349	}
4350
4351	btrfs_info(fs_info,
4352		   "relocating block group %llu flags %s",
4353		   block_group->key.objectid, buf + 1);
 
 
 
 
 
 
 
 
 
4354}
4355
4356/*
4357 * function to relocate all extents in a block group.
4358 */
4359int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start)
4360{
4361	struct btrfs_root *extent_root = fs_info->extent_root;
 
4362	struct reloc_control *rc;
4363	struct inode *inode;
4364	struct btrfs_path *path;
4365	int ret;
4366	int rw = 0;
4367	int err = 0;
4368
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4369	rc = alloc_reloc_control(fs_info);
4370	if (!rc)
 
4371		return -ENOMEM;
 
 
 
 
 
 
 
4372
4373	rc->extent_root = extent_root;
 
4374
4375	rc->block_group = btrfs_lookup_block_group(fs_info, group_start);
4376	BUG_ON(!rc->block_group);
4377
4378	ret = btrfs_inc_block_group_ro(fs_info, rc->block_group);
4379	if (ret) {
4380		err = ret;
4381		goto out;
4382	}
4383	rw = 1;
4384
4385	path = btrfs_alloc_path();
4386	if (!path) {
4387		err = -ENOMEM;
4388		goto out;
4389	}
4390
4391	inode = lookup_free_space_inode(fs_info, rc->block_group, path);
4392	btrfs_free_path(path);
4393
4394	if (!IS_ERR(inode))
4395		ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0);
4396	else
4397		ret = PTR_ERR(inode);
4398
4399	if (ret && ret != -ENOENT) {
4400		err = ret;
4401		goto out;
4402	}
4403
4404	rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
4405	if (IS_ERR(rc->data_inode)) {
4406		err = PTR_ERR(rc->data_inode);
4407		rc->data_inode = NULL;
4408		goto out;
4409	}
4410
4411	describe_relocation(fs_info, rc->block_group);
4412
4413	btrfs_wait_block_group_reservations(rc->block_group);
4414	btrfs_wait_nocow_writers(rc->block_group);
4415	btrfs_wait_ordered_roots(fs_info, U64_MAX,
4416				 rc->block_group->key.objectid,
4417				 rc->block_group->key.offset);
 
 
 
4418
4419	while (1) {
 
 
4420		mutex_lock(&fs_info->cleaner_mutex);
4421		ret = relocate_block_group(rc);
4422		mutex_unlock(&fs_info->cleaner_mutex);
4423		if (ret < 0) {
4424			err = ret;
4425			goto out;
4426		}
4427
4428		if (rc->extents_found == 0)
4429			break;
4430
4431		btrfs_info(fs_info, "found %llu extents", rc->extents_found);
4432
 
 
 
 
 
 
 
 
 
 
4433		if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) {
4434			ret = btrfs_wait_ordered_range(rc->data_inode, 0,
4435						       (u64)-1);
4436			if (ret) {
4437				err = ret;
4438				goto out;
4439			}
4440			invalidate_mapping_pages(rc->data_inode->i_mapping,
4441						 0, -1);
4442			rc->stage = UPDATE_DATA_PTRS;
4443		}
 
 
 
 
 
 
 
 
 
4444	}
4445
4446	WARN_ON(rc->block_group->pinned > 0);
4447	WARN_ON(rc->block_group->reserved > 0);
4448	WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0);
4449out:
4450	if (err && rw)
4451		btrfs_dec_block_group_ro(rc->block_group);
4452	iput(rc->data_inode);
4453	btrfs_put_block_group(rc->block_group);
4454	kfree(rc);
 
 
4455	return err;
4456}
4457
4458static noinline_for_stack int mark_garbage_root(struct btrfs_root *root)
4459{
4460	struct btrfs_fs_info *fs_info = root->fs_info;
4461	struct btrfs_trans_handle *trans;
4462	int ret, err;
4463
4464	trans = btrfs_start_transaction(fs_info->tree_root, 0);
4465	if (IS_ERR(trans))
4466		return PTR_ERR(trans);
4467
4468	memset(&root->root_item.drop_progress, 0,
4469		sizeof(root->root_item.drop_progress));
4470	root->root_item.drop_level = 0;
4471	btrfs_set_root_refs(&root->root_item, 0);
4472	ret = btrfs_update_root(trans, fs_info->tree_root,
4473				&root->root_key, &root->root_item);
4474
4475	err = btrfs_end_transaction(trans);
4476	if (err)
4477		return err;
4478	return ret;
4479}
4480
4481/*
4482 * recover relocation interrupted by system crash.
4483 *
4484 * this function resumes merging reloc trees with corresponding fs trees.
4485 * this is important for keeping the sharing of tree blocks
4486 */
4487int btrfs_recover_relocation(struct btrfs_root *root)
4488{
4489	struct btrfs_fs_info *fs_info = root->fs_info;
4490	LIST_HEAD(reloc_roots);
4491	struct btrfs_key key;
4492	struct btrfs_root *fs_root;
4493	struct btrfs_root *reloc_root;
4494	struct btrfs_path *path;
4495	struct extent_buffer *leaf;
4496	struct reloc_control *rc = NULL;
4497	struct btrfs_trans_handle *trans;
4498	int ret;
4499	int err = 0;
4500
4501	path = btrfs_alloc_path();
4502	if (!path)
4503		return -ENOMEM;
4504	path->reada = READA_BACK;
4505
4506	key.objectid = BTRFS_TREE_RELOC_OBJECTID;
4507	key.type = BTRFS_ROOT_ITEM_KEY;
4508	key.offset = (u64)-1;
4509
4510	while (1) {
4511		ret = btrfs_search_slot(NULL, fs_info->tree_root, &key,
4512					path, 0, 0);
4513		if (ret < 0) {
4514			err = ret;
4515			goto out;
4516		}
4517		if (ret > 0) {
4518			if (path->slots[0] == 0)
4519				break;
4520			path->slots[0]--;
4521		}
4522		leaf = path->nodes[0];
4523		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4524		btrfs_release_path(path);
4525
4526		if (key.objectid != BTRFS_TREE_RELOC_OBJECTID ||
4527		    key.type != BTRFS_ROOT_ITEM_KEY)
4528			break;
4529
4530		reloc_root = btrfs_read_fs_root(root, &key);
4531		if (IS_ERR(reloc_root)) {
4532			err = PTR_ERR(reloc_root);
4533			goto out;
4534		}
4535
 
4536		list_add(&reloc_root->root_list, &reloc_roots);
4537
4538		if (btrfs_root_refs(&reloc_root->root_item) > 0) {
4539			fs_root = read_fs_root(fs_info,
4540					       reloc_root->root_key.offset);
4541			if (IS_ERR(fs_root)) {
4542				ret = PTR_ERR(fs_root);
4543				if (ret != -ENOENT) {
4544					err = ret;
4545					goto out;
4546				}
4547				ret = mark_garbage_root(reloc_root);
4548				if (ret < 0) {
4549					err = ret;
4550					goto out;
4551				}
 
 
4552			}
4553		}
4554
4555		if (key.offset == 0)
4556			break;
4557
4558		key.offset--;
4559	}
4560	btrfs_release_path(path);
4561
4562	if (list_empty(&reloc_roots))
4563		goto out;
4564
4565	rc = alloc_reloc_control(fs_info);
4566	if (!rc) {
4567		err = -ENOMEM;
4568		goto out;
4569	}
4570
4571	rc->extent_root = fs_info->extent_root;
 
 
 
 
 
 
4572
4573	set_reloc_control(rc);
4574
4575	trans = btrfs_join_transaction(rc->extent_root);
4576	if (IS_ERR(trans)) {
4577		unset_reloc_control(rc);
4578		err = PTR_ERR(trans);
4579		goto out_free;
4580	}
4581
4582	rc->merge_reloc_tree = 1;
4583
4584	while (!list_empty(&reloc_roots)) {
4585		reloc_root = list_entry(reloc_roots.next,
4586					struct btrfs_root, root_list);
4587		list_del(&reloc_root->root_list);
4588
4589		if (btrfs_root_refs(&reloc_root->root_item) == 0) {
4590			list_add_tail(&reloc_root->root_list,
4591				      &rc->reloc_roots);
4592			continue;
4593		}
4594
4595		fs_root = read_fs_root(fs_info, reloc_root->root_key.offset);
 
4596		if (IS_ERR(fs_root)) {
4597			err = PTR_ERR(fs_root);
4598			goto out_free;
 
 
4599		}
4600
4601		err = __add_reloc_root(reloc_root);
4602		BUG_ON(err < 0); /* -ENOMEM or logic error */
4603		fs_root->reloc_root = reloc_root;
 
 
 
 
 
 
 
4604	}
4605
4606	err = btrfs_commit_transaction(trans);
4607	if (err)
4608		goto out_free;
4609
4610	merge_reloc_roots(rc);
4611
4612	unset_reloc_control(rc);
4613
4614	trans = btrfs_join_transaction(rc->extent_root);
4615	if (IS_ERR(trans)) {
4616		err = PTR_ERR(trans);
4617		goto out_free;
4618	}
4619	err = btrfs_commit_transaction(trans);
4620out_free:
4621	kfree(rc);
 
 
 
 
 
 
 
4622out:
4623	if (!list_empty(&reloc_roots))
4624		free_reloc_roots(&reloc_roots);
4625
4626	btrfs_free_path(path);
4627
4628	if (err == 0) {
4629		/* cleanup orphan inode in data relocation tree */
4630		fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
4631		if (IS_ERR(fs_root))
4632			err = PTR_ERR(fs_root);
4633		else
4634			err = btrfs_orphan_cleanup(fs_root);
4635	}
4636	return err;
4637}
4638
4639/*
4640 * helper to add ordered checksum for data relocation.
4641 *
4642 * cloning checksum properly handles the nodatasum extents.
4643 * it also saves CPU time to re-calculate the checksum.
4644 */
4645int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len)
4646{
4647	struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb);
 
4648	struct btrfs_ordered_sum *sums;
4649	struct btrfs_ordered_extent *ordered;
4650	int ret;
4651	u64 disk_bytenr;
4652	u64 new_bytenr;
4653	LIST_HEAD(list);
4654
4655	ordered = btrfs_lookup_ordered_extent(inode, file_pos);
4656	BUG_ON(ordered->file_offset != file_pos || ordered->len != len);
4657
4658	disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt;
4659	ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr,
4660				       disk_bytenr + len - 1, &list, 0);
 
4661	if (ret)
4662		goto out;
4663
4664	while (!list_empty(&list)) {
4665		sums = list_entry(list.next, struct btrfs_ordered_sum, list);
4666		list_del_init(&sums->list);
4667
4668		/*
4669		 * We need to offset the new_bytenr based on where the csum is.
4670		 * We need to do this because we will read in entire prealloc
4671		 * extents but we may have written to say the middle of the
4672		 * prealloc extent, so we need to make sure the csum goes with
4673		 * the right disk offset.
4674		 *
4675		 * We can do this because the data reloc inode refers strictly
4676		 * to the on disk bytes, so we don't have to worry about
4677		 * disk_len vs real len like with real inodes since it's all
4678		 * disk length.
4679		 */
4680		new_bytenr = ordered->start + (sums->bytenr - disk_bytenr);
4681		sums->bytenr = new_bytenr;
4682
4683		btrfs_add_ordered_sum(inode, ordered, sums);
4684	}
4685out:
4686	btrfs_put_ordered_extent(ordered);
4687	return ret;
4688}
4689
4690int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans,
4691			  struct btrfs_root *root, struct extent_buffer *buf,
4692			  struct extent_buffer *cow)
4693{
4694	struct btrfs_fs_info *fs_info = root->fs_info;
4695	struct reloc_control *rc;
4696	struct backref_node *node;
4697	int first_cow = 0;
4698	int level;
4699	int ret = 0;
4700
4701	rc = fs_info->reloc_ctl;
4702	if (!rc)
4703		return 0;
4704
4705	BUG_ON(rc->stage == UPDATE_DATA_PTRS &&
4706	       root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID);
4707
4708	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
4709		if (buf == root->node)
4710			__update_reloc_root(root, cow->start);
4711	}
4712
4713	level = btrfs_header_level(buf);
4714	if (btrfs_header_generation(buf) <=
4715	    btrfs_root_last_snapshot(&root->root_item))
4716		first_cow = 1;
4717
4718	if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID &&
4719	    rc->create_reloc_tree) {
4720		WARN_ON(!first_cow && level == 0);
4721
4722		node = rc->backref_cache.path[level];
4723		BUG_ON(node->bytenr != buf->start &&
4724		       node->new_bytenr != buf->start);
4725
4726		drop_node_buffer(node);
4727		extent_buffer_get(cow);
4728		node->eb = cow;
4729		node->new_bytenr = cow->start;
4730
4731		if (!node->pending) {
4732			list_move_tail(&node->list,
4733				       &rc->backref_cache.pending[level]);
4734			node->pending = 1;
4735		}
4736
4737		if (first_cow)
4738			__mark_block_processed(rc, node);
4739
4740		if (first_cow && level > 0)
4741			rc->nodes_relocated += buf->len;
4742	}
4743
4744	if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS)
4745		ret = replace_file_extents(trans, rc, root, cow);
4746	return ret;
4747}
4748
4749/*
4750 * called before creating snapshot. it calculates metadata reservation
4751 * required for relocating tree blocks in the snapshot
4752 */
4753void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending,
4754			      u64 *bytes_to_reserve)
4755{
4756	struct btrfs_root *root;
4757	struct reloc_control *rc;
4758
4759	root = pending->root;
4760	if (!root->reloc_root)
4761		return;
4762
4763	rc = root->fs_info->reloc_ctl;
4764	if (!rc->merge_reloc_tree)
4765		return;
4766
4767	root = root->reloc_root;
4768	BUG_ON(btrfs_root_refs(&root->root_item) == 0);
4769	/*
4770	 * relocation is in the stage of merging trees. the space
4771	 * used by merging a reloc tree is twice the size of
4772	 * relocated tree nodes in the worst case. half for cowing
4773	 * the reloc tree, half for cowing the fs tree. the space
4774	 * used by cowing the reloc tree will be freed after the
4775	 * tree is dropped. if we create snapshot, cowing the fs
4776	 * tree may use more space than it frees. so we need
4777	 * reserve extra space.
4778	 */
4779	*bytes_to_reserve += rc->nodes_relocated;
4780}
4781
4782/*
4783 * called after snapshot is created. migrate block reservation
4784 * and create reloc root for the newly created snapshot
 
 
 
 
4785 */
4786int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans,
4787			       struct btrfs_pending_snapshot *pending)
4788{
4789	struct btrfs_root *root = pending->root;
4790	struct btrfs_root *reloc_root;
4791	struct btrfs_root *new_root;
4792	struct reloc_control *rc;
4793	int ret;
4794
4795	if (!root->reloc_root)
4796		return 0;
4797
4798	rc = root->fs_info->reloc_ctl;
4799	rc->merging_rsv_size += rc->nodes_relocated;
4800
4801	if (rc->merge_reloc_tree) {
4802		ret = btrfs_block_rsv_migrate(&pending->block_rsv,
4803					      rc->block_rsv,
4804					      rc->nodes_relocated, 1);
4805		if (ret)
4806			return ret;
4807	}
4808
4809	new_root = pending->snap;
4810	reloc_root = create_reloc_root(trans, root->reloc_root,
4811				       new_root->root_key.objectid);
4812	if (IS_ERR(reloc_root))
4813		return PTR_ERR(reloc_root);
4814
4815	ret = __add_reloc_root(reloc_root);
4816	BUG_ON(ret < 0);
4817	new_root->reloc_root = reloc_root;
 
 
 
 
 
4818
4819	if (rc->create_reloc_tree)
4820		ret = clone_backref_node(trans, rc, root, reloc_root);
4821	return ret;
4822}