1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2011 STRATO.  All rights reserved.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   4 */
   5
   6#include <linux/mm.h>
   7#include <linux/rbtree.h>
   8#include <trace/events/btrfs.h>
   9#include "ctree.h"
  10#include "disk-io.h"
  11#include "backref.h"
  12#include "ulist.h"
  13#include "transaction.h"
  14#include "delayed-ref.h"
  15#include "locking.h"
  16
  17/* Just an arbitrary number so we can be sure this happened */
  18#define BACKREF_FOUND_SHARED 6
  19
  20struct extent_inode_elem {
  21	u64 inum;
  22	u64 offset;
  23	struct extent_inode_elem *next;
  24};
  25
  26static int check_extent_in_eb(const struct btrfs_key *key,
  27			      const struct extent_buffer *eb,
  28			      const struct btrfs_file_extent_item *fi,
  29			      u64 extent_item_pos,
  30			      struct extent_inode_elem **eie,
  31			      bool ignore_offset)
  32{
  33	u64 offset = 0;
  34	struct extent_inode_elem *e;
  35
  36	if (!ignore_offset &&
  37	    !btrfs_file_extent_compression(eb, fi) &&
  38	    !btrfs_file_extent_encryption(eb, fi) &&
  39	    !btrfs_file_extent_other_encoding(eb, fi)) {
  40		u64 data_offset;
  41		u64 data_len;
  42
  43		data_offset = btrfs_file_extent_offset(eb, fi);
  44		data_len = btrfs_file_extent_num_bytes(eb, fi);
  45
  46		if (extent_item_pos < data_offset ||
  47		    extent_item_pos >= data_offset + data_len)
  48			return 1;
  49		offset = extent_item_pos - data_offset;
  50	}
  51
  52	e = kmalloc(sizeof(*e), GFP_NOFS);
  53	if (!e)
  54		return -ENOMEM;
  55
  56	e->next = *eie;
  57	e->inum = key->objectid;
  58	e->offset = key->offset + offset;
  59	*eie = e;
  60
  61	return 0;
  62}
  63
  64static void free_inode_elem_list(struct extent_inode_elem *eie)
  65{
  66	struct extent_inode_elem *eie_next;
  67
  68	for (; eie; eie = eie_next) {
  69		eie_next = eie->next;
  70		kfree(eie);
  71	}
  72}
  73
  74static int find_extent_in_eb(const struct extent_buffer *eb,
  75			     u64 wanted_disk_byte, u64 extent_item_pos,
  76			     struct extent_inode_elem **eie,
  77			     bool ignore_offset)
  78{
  79	u64 disk_byte;
  80	struct btrfs_key key;
  81	struct btrfs_file_extent_item *fi;
  82	int slot;
  83	int nritems;
  84	int extent_type;
  85	int ret;
  86
  87	/*
  88	 * from the shared data ref, we only have the leaf but we need
  89	 * the key. thus, we must look into all items and see that we
  90	 * find one (some) with a reference to our extent item.
  91	 */
  92	nritems = btrfs_header_nritems(eb);
  93	for (slot = 0; slot < nritems; ++slot) {
  94		btrfs_item_key_to_cpu(eb, &key, slot);
  95		if (key.type != BTRFS_EXTENT_DATA_KEY)
  96			continue;
  97		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
  98		extent_type = btrfs_file_extent_type(eb, fi);
  99		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
 100			continue;
 101		/* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
 102		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 103		if (disk_byte != wanted_disk_byte)
 104			continue;
 105
 106		ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset);
 107		if (ret < 0)
 108			return ret;
 109	}
 110
 111	return 0;
 112}
 113
 114struct preftree {
 115	struct rb_root root;
 116	unsigned int count;
 117};
 118
 119#define PREFTREE_INIT	{ .root = RB_ROOT, .count = 0 }
 120
 121struct preftrees {
 122	struct preftree direct;    /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
 123	struct preftree indirect;  /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
 124	struct preftree indirect_missing_keys;
 125};
 126
 127/*
 128 * Checks for a shared extent during backref search.
 129 *
 130 * The share_count tracks prelim_refs (direct and indirect) having a
 131 * ref->count >0:
 132 *  - incremented when a ref->count transitions to >0
 133 *  - decremented when a ref->count transitions to <1
 134 */
 135struct share_check {
 136	u64 root_objectid;
 137	u64 inum;
 138	int share_count;
 
 
 
 
 
 139};
 140
 141static inline int extent_is_shared(struct share_check *sc)
 142{
 143	return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
 144}
 145
 146static struct kmem_cache *btrfs_prelim_ref_cache;
 147
 148int __init btrfs_prelim_ref_init(void)
 149{
 150	btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
 151					sizeof(struct prelim_ref),
 152					0,
 153					SLAB_MEM_SPREAD,
 154					NULL);
 155	if (!btrfs_prelim_ref_cache)
 156		return -ENOMEM;
 157	return 0;
 158}
 159
 160void __cold btrfs_prelim_ref_exit(void)
 161{
 162	kmem_cache_destroy(btrfs_prelim_ref_cache);
 163}
 164
 165static void free_pref(struct prelim_ref *ref)
 166{
 167	kmem_cache_free(btrfs_prelim_ref_cache, ref);
 168}
 169
 170/*
 171 * Return 0 when both refs are for the same block (and can be merged).
 172 * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
 173 * indicates a 'higher' block.
 174 */
 175static int prelim_ref_compare(struct prelim_ref *ref1,
 176			      struct prelim_ref *ref2)
 177{
 178	if (ref1->level < ref2->level)
 179		return -1;
 180	if (ref1->level > ref2->level)
 181		return 1;
 182	if (ref1->root_id < ref2->root_id)
 183		return -1;
 184	if (ref1->root_id > ref2->root_id)
 185		return 1;
 186	if (ref1->key_for_search.type < ref2->key_for_search.type)
 187		return -1;
 188	if (ref1->key_for_search.type > ref2->key_for_search.type)
 189		return 1;
 190	if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
 191		return -1;
 192	if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
 193		return 1;
 194	if (ref1->key_for_search.offset < ref2->key_for_search.offset)
 195		return -1;
 196	if (ref1->key_for_search.offset > ref2->key_for_search.offset)
 197		return 1;
 198	if (ref1->parent < ref2->parent)
 199		return -1;
 200	if (ref1->parent > ref2->parent)
 201		return 1;
 202
 203	return 0;
 204}
 205
 206static void update_share_count(struct share_check *sc, int oldcount,
 207			       int newcount)
 208{
 209	if ((!sc) || (oldcount == 0 && newcount < 1))
 210		return;
 211
 212	if (oldcount > 0 && newcount < 1)
 213		sc->share_count--;
 214	else if (oldcount < 1 && newcount > 0)
 215		sc->share_count++;
 216}
 217
 218/*
 219 * Add @newref to the @root rbtree, merging identical refs.
 220 *
 221 * Callers should assume that newref has been freed after calling.
 222 */
 223static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
 224			      struct preftree *preftree,
 225			      struct prelim_ref *newref,
 226			      struct share_check *sc)
 227{
 228	struct rb_root *root;
 229	struct rb_node **p;
 230	struct rb_node *parent = NULL;
 231	struct prelim_ref *ref;
 232	int result;
 233
 234	root = &preftree->root;
 235	p = &root->rb_node;
 236
 237	while (*p) {
 238		parent = *p;
 239		ref = rb_entry(parent, struct prelim_ref, rbnode);
 240		result = prelim_ref_compare(ref, newref);
 241		if (result < 0) {
 242			p = &(*p)->rb_left;
 243		} else if (result > 0) {
 244			p = &(*p)->rb_right;
 245		} else {
 246			/* Identical refs, merge them and free @newref */
 247			struct extent_inode_elem *eie = ref->inode_list;
 248
 249			while (eie && eie->next)
 250				eie = eie->next;
 251
 252			if (!eie)
 253				ref->inode_list = newref->inode_list;
 254			else
 255				eie->next = newref->inode_list;
 256			trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
 257						     preftree->count);
 258			/*
 259			 * A delayed ref can have newref->count < 0.
 260			 * The ref->count is updated to follow any
 261			 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
 262			 */
 263			update_share_count(sc, ref->count,
 264					   ref->count + newref->count);
 265			ref->count += newref->count;
 266			free_pref(newref);
 267			return;
 268		}
 269	}
 270
 271	update_share_count(sc, 0, newref->count);
 272	preftree->count++;
 273	trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
 274	rb_link_node(&newref->rbnode, parent, p);
 275	rb_insert_color(&newref->rbnode, root);
 276}
 277
 278/*
 279 * Release the entire tree.  We don't care about internal consistency so
 280 * just free everything and then reset the tree root.
 281 */
 282static void prelim_release(struct preftree *preftree)
 283{
 284	struct prelim_ref *ref, *next_ref;
 285
 286	rbtree_postorder_for_each_entry_safe(ref, next_ref, &preftree->root,
 287					     rbnode)
 288		free_pref(ref);
 289
 290	preftree->root = RB_ROOT;
 291	preftree->count = 0;
 292}
 293
 294/*
 295 * the rules for all callers of this function are:
 296 * - obtaining the parent is the goal
 297 * - if you add a key, you must know that it is a correct key
 298 * - if you cannot add the parent or a correct key, then we will look into the
 299 *   block later to set a correct key
 300 *
 301 * delayed refs
 302 * ============
 303 *        backref type | shared | indirect | shared | indirect
 304 * information         |   tree |     tree |   data |     data
 305 * --------------------+--------+----------+--------+----------
 306 *      parent logical |    y   |     -    |    -   |     -
 307 *      key to resolve |    -   |     y    |    y   |     y
 308 *  tree block logical |    -   |     -    |    -   |     -
 309 *  root for resolving |    y   |     y    |    y   |     y
 310 *
 311 * - column 1:       we've the parent -> done
 312 * - column 2, 3, 4: we use the key to find the parent
 313 *
 314 * on disk refs (inline or keyed)
 315 * ==============================
 316 *        backref type | shared | indirect | shared | indirect
 317 * information         |   tree |     tree |   data |     data
 318 * --------------------+--------+----------+--------+----------
 319 *      parent logical |    y   |     -    |    y   |     -
 320 *      key to resolve |    -   |     -    |    -   |     y
 321 *  tree block logical |    y   |     y    |    y   |     y
 322 *  root for resolving |    -   |     y    |    y   |     y
 323 *
 324 * - column 1, 3: we've the parent -> done
 325 * - column 2:    we take the first key from the block to find the parent
 326 *                (see add_missing_keys)
 327 * - column 4:    we use the key to find the parent
 328 *
 329 * additional information that's available but not required to find the parent
 330 * block might help in merging entries to gain some speed.
 331 */
 332static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
 333			  struct preftree *preftree, u64 root_id,
 334			  const struct btrfs_key *key, int level, u64 parent,
 335			  u64 wanted_disk_byte, int count,
 336			  struct share_check *sc, gfp_t gfp_mask)
 337{
 338	struct prelim_ref *ref;
 339
 340	if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
 341		return 0;
 342
 343	ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
 344	if (!ref)
 345		return -ENOMEM;
 346
 347	ref->root_id = root_id;
 348	if (key) {
 349		ref->key_for_search = *key;
 350		/*
 351		 * We can often find data backrefs with an offset that is too
 352		 * large (>= LLONG_MAX, maximum allowed file offset) due to
 353		 * underflows when subtracting a file's offset with the data
 354		 * offset of its corresponding extent data item. This can
 355		 * happen for example in the clone ioctl.
 356		 * So if we detect such case we set the search key's offset to
 357		 * zero to make sure we will find the matching file extent item
 358		 * at add_all_parents(), otherwise we will miss it because the
 359		 * offset taken form the backref is much larger then the offset
 360		 * of the file extent item. This can make us scan a very large
 361		 * number of file extent items, but at least it will not make
 362		 * us miss any.
 363		 * This is an ugly workaround for a behaviour that should have
 364		 * never existed, but it does and a fix for the clone ioctl
 365		 * would touch a lot of places, cause backwards incompatibility
 366		 * and would not fix the problem for extents cloned with older
 367		 * kernels.
 368		 */
 369		if (ref->key_for_search.type == BTRFS_EXTENT_DATA_KEY &&
 370		    ref->key_for_search.offset >= LLONG_MAX)
 371			ref->key_for_search.offset = 0;
 372	} else {
 373		memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
 374	}
 375
 376	ref->inode_list = NULL;
 377	ref->level = level;
 378	ref->count = count;
 379	ref->parent = parent;
 380	ref->wanted_disk_byte = wanted_disk_byte;
 381	prelim_ref_insert(fs_info, preftree, ref, sc);
 382	return extent_is_shared(sc);
 383}
 384
 385/* direct refs use root == 0, key == NULL */
 386static int add_direct_ref(const struct btrfs_fs_info *fs_info,
 387			  struct preftrees *preftrees, int level, u64 parent,
 388			  u64 wanted_disk_byte, int count,
 389			  struct share_check *sc, gfp_t gfp_mask)
 390{
 391	return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
 392			      parent, wanted_disk_byte, count, sc, gfp_mask);
 393}
 394
 395/* indirect refs use parent == 0 */
 396static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
 397			    struct preftrees *preftrees, u64 root_id,
 398			    const struct btrfs_key *key, int level,
 399			    u64 wanted_disk_byte, int count,
 400			    struct share_check *sc, gfp_t gfp_mask)
 401{
 402	struct preftree *tree = &preftrees->indirect;
 403
 404	if (!key)
 405		tree = &preftrees->indirect_missing_keys;
 406	return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
 407			      wanted_disk_byte, count, sc, gfp_mask);
 408}
 409
 410static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
 411			   struct ulist *parents, struct prelim_ref *ref,
 412			   int level, u64 time_seq, const u64 *extent_item_pos,
 413			   u64 total_refs, bool ignore_offset)
 414{
 415	int ret = 0;
 416	int slot;
 417	struct extent_buffer *eb;
 418	struct btrfs_key key;
 419	struct btrfs_key *key_for_search = &ref->key_for_search;
 420	struct btrfs_file_extent_item *fi;
 421	struct extent_inode_elem *eie = NULL, *old = NULL;
 422	u64 disk_byte;
 423	u64 wanted_disk_byte = ref->wanted_disk_byte;
 424	u64 count = 0;
 425
 426	if (level != 0) {
 427		eb = path->nodes[level];
 428		ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
 429		if (ret < 0)
 430			return ret;
 431		return 0;
 432	}
 433
 434	/*
 435	 * We normally enter this function with the path already pointing to
 436	 * the first item to check. But sometimes, we may enter it with
 437	 * slot==nritems. In that case, go to the next leaf before we continue.
 438	 */
 439	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
 440		if (time_seq == SEQ_LAST)
 441			ret = btrfs_next_leaf(root, path);
 442		else
 443			ret = btrfs_next_old_leaf(root, path, time_seq);
 444	}
 445
 446	while (!ret && count < total_refs) {
 447		eb = path->nodes[0];
 448		slot = path->slots[0];
 449
 450		btrfs_item_key_to_cpu(eb, &key, slot);
 451
 452		if (key.objectid != key_for_search->objectid ||
 453		    key.type != BTRFS_EXTENT_DATA_KEY)
 454			break;
 455
 456		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
 457		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
 458
 459		if (disk_byte == wanted_disk_byte) {
 460			eie = NULL;
 461			old = NULL;
 462			count++;
 463			if (extent_item_pos) {
 464				ret = check_extent_in_eb(&key, eb, fi,
 465						*extent_item_pos,
 466						&eie, ignore_offset);
 467				if (ret < 0)
 468					break;
 469			}
 470			if (ret > 0)
 471				goto next;
 472			ret = ulist_add_merge_ptr(parents, eb->start,
 473						  eie, (void **)&old, GFP_NOFS);
 474			if (ret < 0)
 475				break;
 476			if (!ret && extent_item_pos) {
 477				while (old->next)
 478					old = old->next;
 479				old->next = eie;
 480			}
 481			eie = NULL;
 482		}
 483next:
 484		if (time_seq == SEQ_LAST)
 485			ret = btrfs_next_item(root, path);
 486		else
 487			ret = btrfs_next_old_item(root, path, time_seq);
 488	}
 489
 490	if (ret > 0)
 491		ret = 0;
 492	else if (ret < 0)
 493		free_inode_elem_list(eie);
 494	return ret;
 495}
 496
 497/*
 498 * resolve an indirect backref in the form (root_id, key, level)
 499 * to a logical address
 500 */
 501static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
 502				struct btrfs_path *path, u64 time_seq,
 503				struct prelim_ref *ref, struct ulist *parents,
 504				const u64 *extent_item_pos, u64 total_refs,
 505				bool ignore_offset)
 506{
 507	struct btrfs_root *root;
 508	struct btrfs_key root_key;
 509	struct extent_buffer *eb;
 510	int ret = 0;
 511	int root_level;
 512	int level = ref->level;
 513	int index;
 514
 515	root_key.objectid = ref->root_id;
 516	root_key.type = BTRFS_ROOT_ITEM_KEY;
 517	root_key.offset = (u64)-1;
 518
 519	index = srcu_read_lock(&fs_info->subvol_srcu);
 520
 521	root = btrfs_get_fs_root(fs_info, &root_key, false);
 522	if (IS_ERR(root)) {
 523		srcu_read_unlock(&fs_info->subvol_srcu, index);
 524		ret = PTR_ERR(root);
 525		goto out;
 526	}
 527
 528	if (btrfs_is_testing(fs_info)) {
 529		srcu_read_unlock(&fs_info->subvol_srcu, index);
 530		ret = -ENOENT;
 531		goto out;
 532	}
 533
 534	if (path->search_commit_root)
 535		root_level = btrfs_header_level(root->commit_root);
 536	else if (time_seq == SEQ_LAST)
 537		root_level = btrfs_header_level(root->node);
 538	else
 539		root_level = btrfs_old_root_level(root, time_seq);
 540
 541	if (root_level + 1 == level) {
 542		srcu_read_unlock(&fs_info->subvol_srcu, index);
 543		goto out;
 544	}
 545
 546	path->lowest_level = level;
 547	if (time_seq == SEQ_LAST)
 548		ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path,
 549					0, 0);
 550	else
 551		ret = btrfs_search_old_slot(root, &ref->key_for_search, path,
 552					    time_seq);
 553
 554	/* root node has been locked, we can release @subvol_srcu safely here */
 555	srcu_read_unlock(&fs_info->subvol_srcu, index);
 556
 557	btrfs_debug(fs_info,
 558		"search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
 559		 ref->root_id, level, ref->count, ret,
 560		 ref->key_for_search.objectid, ref->key_for_search.type,
 561		 ref->key_for_search.offset);
 562	if (ret < 0)
 563		goto out;
 564
 565	eb = path->nodes[level];
 566	while (!eb) {
 567		if (WARN_ON(!level)) {
 568			ret = 1;
 569			goto out;
 570		}
 571		level--;
 572		eb = path->nodes[level];
 573	}
 574
 575	ret = add_all_parents(root, path, parents, ref, level, time_seq,
 576			      extent_item_pos, total_refs, ignore_offset);
 577out:
 578	path->lowest_level = 0;
 579	btrfs_release_path(path);
 580	return ret;
 581}
 582
 583static struct extent_inode_elem *
 584unode_aux_to_inode_list(struct ulist_node *node)
 585{
 586	if (!node)
 587		return NULL;
 588	return (struct extent_inode_elem *)(uintptr_t)node->aux;
 589}
 590
 591/*
 592 * We maintain three seperate rbtrees: one for direct refs, one for
 593 * indirect refs which have a key, and one for indirect refs which do not
 594 * have a key. Each tree does merge on insertion.
 595 *
 596 * Once all of the references are located, we iterate over the tree of
 597 * indirect refs with missing keys. An appropriate key is located and
 598 * the ref is moved onto the tree for indirect refs. After all missing
 599 * keys are thus located, we iterate over the indirect ref tree, resolve
 600 * each reference, and then insert the resolved reference onto the
 601 * direct tree (merging there too).
 602 *
 603 * New backrefs (i.e., for parent nodes) are added to the appropriate
 604 * rbtree as they are encountered. The new backrefs are subsequently
 605 * resolved as above.
 606 */
 607static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
 608				 struct btrfs_path *path, u64 time_seq,
 609				 struct preftrees *preftrees,
 610				 const u64 *extent_item_pos, u64 total_refs,
 611				 struct share_check *sc, bool ignore_offset)
 612{
 613	int err;
 614	int ret = 0;
 
 
 
 615	struct ulist *parents;
 616	struct ulist_node *node;
 617	struct ulist_iterator uiter;
 618	struct rb_node *rnode;
 619
 620	parents = ulist_alloc(GFP_NOFS);
 621	if (!parents)
 622		return -ENOMEM;
 623
 624	/*
 625	 * We could trade memory usage for performance here by iterating
 626	 * the tree, allocating new refs for each insertion, and then
 627	 * freeing the entire indirect tree when we're done.  In some test
 628	 * cases, the tree can grow quite large (~200k objects).
 629	 */
 630	while ((rnode = rb_first(&preftrees->indirect.root))) {
 631		struct prelim_ref *ref;
 632
 633		ref = rb_entry(rnode, struct prelim_ref, rbnode);
 634		if (WARN(ref->parent,
 635			 "BUG: direct ref found in indirect tree")) {
 636			ret = -EINVAL;
 637			goto out;
 638		}
 639
 640		rb_erase(&ref->rbnode, &preftrees->indirect.root);
 641		preftrees->indirect.count--;
 642
 643		if (ref->count == 0) {
 644			free_pref(ref);
 645			continue;
 646		}
 647
 648		if (sc && sc->root_objectid &&
 649		    ref->root_id != sc->root_objectid) {
 650			free_pref(ref);
 651			ret = BACKREF_FOUND_SHARED;
 652			goto out;
 653		}
 654		err = resolve_indirect_ref(fs_info, path, time_seq, ref,
 655					   parents, extent_item_pos,
 656					   total_refs, ignore_offset);
 657		/*
 658		 * we can only tolerate ENOENT,otherwise,we should catch error
 659		 * and return directly.
 660		 */
 661		if (err == -ENOENT) {
 662			prelim_ref_insert(fs_info, &preftrees->direct, ref,
 663					  NULL);
 664			continue;
 665		} else if (err) {
 666			free_pref(ref);
 667			ret = err;
 668			goto out;
 669		}
 670
 671		/* we put the first parent into the ref at hand */
 672		ULIST_ITER_INIT(&uiter);
 673		node = ulist_next(parents, &uiter);
 674		ref->parent = node ? node->val : 0;
 675		ref->inode_list = unode_aux_to_inode_list(node);
 
 676
 677		/* Add a prelim_ref(s) for any other parent(s). */
 678		while ((node = ulist_next(parents, &uiter))) {
 679			struct prelim_ref *new_ref;
 680
 681			new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
 682						   GFP_NOFS);
 683			if (!new_ref) {
 684				free_pref(ref);
 685				ret = -ENOMEM;
 686				goto out;
 687			}
 688			memcpy(new_ref, ref, sizeof(*ref));
 689			new_ref->parent = node->val;
 690			new_ref->inode_list = unode_aux_to_inode_list(node);
 691			prelim_ref_insert(fs_info, &preftrees->direct,
 692					  new_ref, NULL);
 693		}
 694
 695		/*
 696		 * Now it's a direct ref, put it in the the direct tree. We must
 697		 * do this last because the ref could be merged/freed here.
 698		 */
 699		prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
 700
 701		ulist_reinit(parents);
 702		cond_resched();
 703	}
 704out:
 705	ulist_free(parents);
 706	return ret;
 707}
 708
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 709/*
 710 * read tree blocks and add keys where required.
 711 */
 712static int add_missing_keys(struct btrfs_fs_info *fs_info,
 713			    struct preftrees *preftrees)
 714{
 715	struct prelim_ref *ref;
 716	struct extent_buffer *eb;
 717	struct preftree *tree = &preftrees->indirect_missing_keys;
 718	struct rb_node *node;
 719
 720	while ((node = rb_first(&tree->root))) {
 721		ref = rb_entry(node, struct prelim_ref, rbnode);
 722		rb_erase(node, &tree->root);
 723
 724		BUG_ON(ref->parent);	/* should not be a direct ref */
 725		BUG_ON(ref->key_for_search.type);
 726		BUG_ON(!ref->wanted_disk_byte);
 727
 728		eb = read_tree_block(fs_info, ref->wanted_disk_byte, 0,
 729				     ref->level - 1, NULL);
 730		if (IS_ERR(eb)) {
 731			free_pref(ref);
 732			return PTR_ERR(eb);
 733		} else if (!extent_buffer_uptodate(eb)) {
 734			free_pref(ref);
 735			free_extent_buffer(eb);
 736			return -EIO;
 737		}
 738		btrfs_tree_read_lock(eb);
 739		if (btrfs_header_level(eb) == 0)
 740			btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
 741		else
 742			btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
 743		btrfs_tree_read_unlock(eb);
 744		free_extent_buffer(eb);
 745		prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
 746		cond_resched();
 747	}
 748	return 0;
 749}
 750
 751/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 752 * add all currently queued delayed refs from this head whose seq nr is
 753 * smaller or equal that seq to the list
 754 */
 755static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
 756			    struct btrfs_delayed_ref_head *head, u64 seq,
 757			    struct preftrees *preftrees, u64 *total_refs,
 758			    struct share_check *sc)
 759{
 760	struct btrfs_delayed_ref_node *node;
 761	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
 762	struct btrfs_key key;
 763	struct btrfs_key tmp_op_key;
 764	struct rb_node *n;
 765	int count;
 766	int ret = 0;
 767
 768	if (extent_op && extent_op->update_key)
 769		btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key);
 770
 771	spin_lock(&head->lock);
 772	for (n = rb_first(&head->ref_tree); n; n = rb_next(n)) {
 773		node = rb_entry(n, struct btrfs_delayed_ref_node,
 774				ref_node);
 775		if (node->seq > seq)
 776			continue;
 777
 778		switch (node->action) {
 779		case BTRFS_ADD_DELAYED_EXTENT:
 780		case BTRFS_UPDATE_DELAYED_HEAD:
 781			WARN_ON(1);
 782			continue;
 783		case BTRFS_ADD_DELAYED_REF:
 784			count = node->ref_mod;
 785			break;
 786		case BTRFS_DROP_DELAYED_REF:
 787			count = node->ref_mod * -1;
 788			break;
 789		default:
 790			BUG_ON(1);
 791		}
 792		*total_refs += count;
 793		switch (node->type) {
 794		case BTRFS_TREE_BLOCK_REF_KEY: {
 795			/* NORMAL INDIRECT METADATA backref */
 796			struct btrfs_delayed_tree_ref *ref;
 797
 798			ref = btrfs_delayed_node_to_tree_ref(node);
 799			ret = add_indirect_ref(fs_info, preftrees, ref->root,
 800					       &tmp_op_key, ref->level + 1,
 801					       node->bytenr, count, sc,
 802					       GFP_ATOMIC);
 803			break;
 804		}
 805		case BTRFS_SHARED_BLOCK_REF_KEY: {
 806			/* SHARED DIRECT METADATA backref */
 807			struct btrfs_delayed_tree_ref *ref;
 808
 809			ref = btrfs_delayed_node_to_tree_ref(node);
 810
 811			ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
 812					     ref->parent, node->bytenr, count,
 813					     sc, GFP_ATOMIC);
 814			break;
 815		}
 816		case BTRFS_EXTENT_DATA_REF_KEY: {
 817			/* NORMAL INDIRECT DATA backref */
 818			struct btrfs_delayed_data_ref *ref;
 819			ref = btrfs_delayed_node_to_data_ref(node);
 820
 821			key.objectid = ref->objectid;
 822			key.type = BTRFS_EXTENT_DATA_KEY;
 823			key.offset = ref->offset;
 824
 825			/*
 826			 * Found a inum that doesn't match our known inum, we
 827			 * know it's shared.
 828			 */
 829			if (sc && sc->inum && ref->objectid != sc->inum) {
 830				ret = BACKREF_FOUND_SHARED;
 831				goto out;
 832			}
 833
 834			ret = add_indirect_ref(fs_info, preftrees, ref->root,
 835					       &key, 0, node->bytenr, count, sc,
 836					       GFP_ATOMIC);
 837			break;
 838		}
 839		case BTRFS_SHARED_DATA_REF_KEY: {
 840			/* SHARED DIRECT FULL backref */
 841			struct btrfs_delayed_data_ref *ref;
 842
 843			ref = btrfs_delayed_node_to_data_ref(node);
 844
 845			ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
 846					     node->bytenr, count, sc,
 847					     GFP_ATOMIC);
 848			break;
 849		}
 850		default:
 851			WARN_ON(1);
 852		}
 853		/*
 854		 * We must ignore BACKREF_FOUND_SHARED until all delayed
 855		 * refs have been checked.
 856		 */
 857		if (ret && (ret != BACKREF_FOUND_SHARED))
 858			break;
 859	}
 860	if (!ret)
 861		ret = extent_is_shared(sc);
 862out:
 863	spin_unlock(&head->lock);
 864	return ret;
 865}
 866
 867/*
 868 * add all inline backrefs for bytenr to the list
 869 *
 870 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
 871 */
 872static int add_inline_refs(const struct btrfs_fs_info *fs_info,
 873			   struct btrfs_path *path, u64 bytenr,
 874			   int *info_level, struct preftrees *preftrees,
 875			   u64 *total_refs, struct share_check *sc)
 876{
 877	int ret = 0;
 878	int slot;
 879	struct extent_buffer *leaf;
 880	struct btrfs_key key;
 881	struct btrfs_key found_key;
 882	unsigned long ptr;
 883	unsigned long end;
 884	struct btrfs_extent_item *ei;
 885	u64 flags;
 886	u64 item_size;
 887
 888	/*
 889	 * enumerate all inline refs
 890	 */
 891	leaf = path->nodes[0];
 892	slot = path->slots[0];
 893
 894	item_size = btrfs_item_size_nr(leaf, slot);
 895	BUG_ON(item_size < sizeof(*ei));
 896
 897	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
 898	flags = btrfs_extent_flags(leaf, ei);
 899	*total_refs += btrfs_extent_refs(leaf, ei);
 900	btrfs_item_key_to_cpu(leaf, &found_key, slot);
 901
 902	ptr = (unsigned long)(ei + 1);
 903	end = (unsigned long)ei + item_size;
 904
 905	if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
 906	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 907		struct btrfs_tree_block_info *info;
 908
 909		info = (struct btrfs_tree_block_info *)ptr;
 910		*info_level = btrfs_tree_block_level(leaf, info);
 911		ptr += sizeof(struct btrfs_tree_block_info);
 912		BUG_ON(ptr > end);
 913	} else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
 914		*info_level = found_key.offset;
 915	} else {
 916		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
 917	}
 918
 919	while (ptr < end) {
 920		struct btrfs_extent_inline_ref *iref;
 921		u64 offset;
 922		int type;
 923
 924		iref = (struct btrfs_extent_inline_ref *)ptr;
 925		type = btrfs_get_extent_inline_ref_type(leaf, iref,
 926							BTRFS_REF_TYPE_ANY);
 927		if (type == BTRFS_REF_TYPE_INVALID)
 928			return -EINVAL;
 929
 930		offset = btrfs_extent_inline_ref_offset(leaf, iref);
 931
 932		switch (type) {
 933		case BTRFS_SHARED_BLOCK_REF_KEY:
 934			ret = add_direct_ref(fs_info, preftrees,
 935					     *info_level + 1, offset,
 936					     bytenr, 1, NULL, GFP_NOFS);
 937			break;
 938		case BTRFS_SHARED_DATA_REF_KEY: {
 939			struct btrfs_shared_data_ref *sdref;
 940			int count;
 941
 942			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
 943			count = btrfs_shared_data_ref_count(leaf, sdref);
 944
 945			ret = add_direct_ref(fs_info, preftrees, 0, offset,
 946					     bytenr, count, sc, GFP_NOFS);
 947			break;
 948		}
 949		case BTRFS_TREE_BLOCK_REF_KEY:
 950			ret = add_indirect_ref(fs_info, preftrees, offset,
 951					       NULL, *info_level + 1,
 952					       bytenr, 1, NULL, GFP_NOFS);
 953			break;
 954		case BTRFS_EXTENT_DATA_REF_KEY: {
 955			struct btrfs_extent_data_ref *dref;
 956			int count;
 957			u64 root;
 958
 959			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
 960			count = btrfs_extent_data_ref_count(leaf, dref);
 961			key.objectid = btrfs_extent_data_ref_objectid(leaf,
 962								      dref);
 963			key.type = BTRFS_EXTENT_DATA_KEY;
 964			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
 965
 966			if (sc && sc->inum && key.objectid != sc->inum) {
 967				ret = BACKREF_FOUND_SHARED;
 968				break;
 969			}
 970
 971			root = btrfs_extent_data_ref_root(leaf, dref);
 972
 973			ret = add_indirect_ref(fs_info, preftrees, root,
 974					       &key, 0, bytenr, count,
 975					       sc, GFP_NOFS);
 976			break;
 977		}
 978		default:
 979			WARN_ON(1);
 980		}
 981		if (ret)
 982			return ret;
 983		ptr += btrfs_extent_inline_ref_size(type);
 984	}
 985
 986	return 0;
 987}
 988
 989/*
 990 * add all non-inline backrefs for bytenr to the list
 991 *
 992 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
 993 */
 994static int add_keyed_refs(struct btrfs_fs_info *fs_info,
 995			  struct btrfs_path *path, u64 bytenr,
 996			  int info_level, struct preftrees *preftrees,
 997			  struct share_check *sc)
 998{
 999	struct btrfs_root *extent_root = fs_info->extent_root;
1000	int ret;
1001	int slot;
1002	struct extent_buffer *leaf;
1003	struct btrfs_key key;
1004
1005	while (1) {
1006		ret = btrfs_next_item(extent_root, path);
1007		if (ret < 0)
1008			break;
1009		if (ret) {
1010			ret = 0;
1011			break;
1012		}
1013
1014		slot = path->slots[0];
1015		leaf = path->nodes[0];
1016		btrfs_item_key_to_cpu(leaf, &key, slot);
1017
1018		if (key.objectid != bytenr)
1019			break;
1020		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1021			continue;
1022		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1023			break;
1024
1025		switch (key.type) {
1026		case BTRFS_SHARED_BLOCK_REF_KEY:
1027			/* SHARED DIRECT METADATA backref */
1028			ret = add_direct_ref(fs_info, preftrees,
1029					     info_level + 1, key.offset,
1030					     bytenr, 1, NULL, GFP_NOFS);
1031			break;
1032		case BTRFS_SHARED_DATA_REF_KEY: {
1033			/* SHARED DIRECT FULL backref */
1034			struct btrfs_shared_data_ref *sdref;
1035			int count;
1036
1037			sdref = btrfs_item_ptr(leaf, slot,
1038					      struct btrfs_shared_data_ref);
1039			count = btrfs_shared_data_ref_count(leaf, sdref);
1040			ret = add_direct_ref(fs_info, preftrees, 0,
1041					     key.offset, bytenr, count,
1042					     sc, GFP_NOFS);
1043			break;
1044		}
1045		case BTRFS_TREE_BLOCK_REF_KEY:
1046			/* NORMAL INDIRECT METADATA backref */
1047			ret = add_indirect_ref(fs_info, preftrees, key.offset,
1048					       NULL, info_level + 1, bytenr,
1049					       1, NULL, GFP_NOFS);
1050			break;
1051		case BTRFS_EXTENT_DATA_REF_KEY: {
1052			/* NORMAL INDIRECT DATA backref */
1053			struct btrfs_extent_data_ref *dref;
1054			int count;
1055			u64 root;
1056
1057			dref = btrfs_item_ptr(leaf, slot,
1058					      struct btrfs_extent_data_ref);
1059			count = btrfs_extent_data_ref_count(leaf, dref);
1060			key.objectid = btrfs_extent_data_ref_objectid(leaf,
1061								      dref);
1062			key.type = BTRFS_EXTENT_DATA_KEY;
1063			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1064
1065			if (sc && sc->inum && key.objectid != sc->inum) {
1066				ret = BACKREF_FOUND_SHARED;
1067				break;
1068			}
1069
1070			root = btrfs_extent_data_ref_root(leaf, dref);
1071			ret = add_indirect_ref(fs_info, preftrees, root,
1072					       &key, 0, bytenr, count,
1073					       sc, GFP_NOFS);
1074			break;
1075		}
1076		default:
1077			WARN_ON(1);
1078		}
1079		if (ret)
1080			return ret;
1081
1082	}
1083
1084	return ret;
1085}
1086
1087/*
1088 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1089 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1090 * indirect refs to their parent bytenr.
1091 * When roots are found, they're added to the roots list
1092 *
1093 * If time_seq is set to SEQ_LAST, it will not search delayed_refs, and behave
 
 
1094 * much like trans == NULL case, the difference only lies in it will not
1095 * commit root.
1096 * The special case is for qgroup to search roots in commit_transaction().
1097 *
1098 * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1099 * shared extent is detected.
1100 *
1101 * Otherwise this returns 0 for success and <0 for an error.
1102 *
1103 * If ignore_offset is set to false, only extent refs whose offsets match
1104 * extent_item_pos are returned.  If true, every extent ref is returned
1105 * and extent_item_pos is ignored.
1106 *
1107 * FIXME some caching might speed things up
1108 */
1109static int find_parent_nodes(struct btrfs_trans_handle *trans,
1110			     struct btrfs_fs_info *fs_info, u64 bytenr,
1111			     u64 time_seq, struct ulist *refs,
1112			     struct ulist *roots, const u64 *extent_item_pos,
1113			     struct share_check *sc, bool ignore_offset)
1114{
1115	struct btrfs_key key;
1116	struct btrfs_path *path;
1117	struct btrfs_delayed_ref_root *delayed_refs = NULL;
1118	struct btrfs_delayed_ref_head *head;
1119	int info_level = 0;
1120	int ret;
1121	struct prelim_ref *ref;
1122	struct rb_node *node;
 
1123	struct extent_inode_elem *eie = NULL;
1124	/* total of both direct AND indirect refs! */
1125	u64 total_refs = 0;
1126	struct preftrees preftrees = {
1127		.direct = PREFTREE_INIT,
1128		.indirect = PREFTREE_INIT,
1129		.indirect_missing_keys = PREFTREE_INIT
1130	};
1131
1132	key.objectid = bytenr;
1133	key.offset = (u64)-1;
1134	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1135		key.type = BTRFS_METADATA_ITEM_KEY;
1136	else
1137		key.type = BTRFS_EXTENT_ITEM_KEY;
1138
1139	path = btrfs_alloc_path();
1140	if (!path)
1141		return -ENOMEM;
1142	if (!trans) {
1143		path->search_commit_root = 1;
1144		path->skip_locking = 1;
1145	}
1146
1147	if (time_seq == SEQ_LAST)
1148		path->skip_locking = 1;
1149
1150	/*
1151	 * grab both a lock on the path and a lock on the delayed ref head.
1152	 * We need both to get a consistent picture of how the refs look
1153	 * at a specified point in time
1154	 */
1155again:
1156	head = NULL;
1157
1158	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
1159	if (ret < 0)
1160		goto out;
1161	BUG_ON(ret == 0);
1162
1163#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1164	if (trans && likely(trans->type != __TRANS_DUMMY) &&
1165	    time_seq != SEQ_LAST) {
1166#else
1167	if (trans && time_seq != SEQ_LAST) {
1168#endif
1169		/*
1170		 * look if there are updates for this ref queued and lock the
1171		 * head
1172		 */
1173		delayed_refs = &trans->transaction->delayed_refs;
1174		spin_lock(&delayed_refs->lock);
1175		head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1176		if (head) {
1177			if (!mutex_trylock(&head->mutex)) {
1178				refcount_inc(&head->refs);
1179				spin_unlock(&delayed_refs->lock);
1180
1181				btrfs_release_path(path);
1182
1183				/*
1184				 * Mutex was contended, block until it's
1185				 * released and try again
1186				 */
1187				mutex_lock(&head->mutex);
1188				mutex_unlock(&head->mutex);
1189				btrfs_put_delayed_ref_head(head);
1190				goto again;
1191			}
1192			spin_unlock(&delayed_refs->lock);
1193			ret = add_delayed_refs(fs_info, head, time_seq,
1194					       &preftrees, &total_refs, sc);
 
1195			mutex_unlock(&head->mutex);
1196			if (ret)
1197				goto out;
1198		} else {
1199			spin_unlock(&delayed_refs->lock);
1200		}
1201	}
1202
1203	if (path->slots[0]) {
1204		struct extent_buffer *leaf;
1205		int slot;
1206
1207		path->slots[0]--;
1208		leaf = path->nodes[0];
1209		slot = path->slots[0];
1210		btrfs_item_key_to_cpu(leaf, &key, slot);
1211		if (key.objectid == bytenr &&
1212		    (key.type == BTRFS_EXTENT_ITEM_KEY ||
1213		     key.type == BTRFS_METADATA_ITEM_KEY)) {
1214			ret = add_inline_refs(fs_info, path, bytenr,
1215					      &info_level, &preftrees,
1216					      &total_refs, sc);
1217			if (ret)
1218				goto out;
1219			ret = add_keyed_refs(fs_info, path, bytenr, info_level,
1220					     &preftrees, sc);
1221			if (ret)
1222				goto out;
1223		}
1224	}
1225
1226	btrfs_release_path(path);
1227
1228	ret = add_missing_keys(fs_info, &preftrees);
 
 
1229	if (ret)
1230		goto out;
1231
1232	WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root));
1233
1234	ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1235				    extent_item_pos, total_refs, sc, ignore_offset);
 
1236	if (ret)
1237		goto out;
1238
1239	WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root));
1240
1241	/*
1242	 * This walks the tree of merged and resolved refs. Tree blocks are
1243	 * read in as needed. Unique entries are added to the ulist, and
1244	 * the list of found roots is updated.
1245	 *
1246	 * We release the entire tree in one go before returning.
1247	 */
1248	node = rb_first(&preftrees.direct.root);
1249	while (node) {
1250		ref = rb_entry(node, struct prelim_ref, rbnode);
1251		node = rb_next(&ref->rbnode);
1252		/*
1253		 * ref->count < 0 can happen here if there are delayed
1254		 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1255		 * prelim_ref_insert() relies on this when merging
1256		 * identical refs to keep the overall count correct.
1257		 * prelim_ref_insert() will merge only those refs
1258		 * which compare identically.  Any refs having
1259		 * e.g. different offsets would not be merged,
1260		 * and would retain their original ref->count < 0.
1261		 */
1262		if (roots && ref->count && ref->root_id && ref->parent == 0) {
1263			if (sc && sc->root_objectid &&
1264			    ref->root_id != sc->root_objectid) {
1265				ret = BACKREF_FOUND_SHARED;
1266				goto out;
1267			}
1268
1269			/* no parent == root of tree */
1270			ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1271			if (ret < 0)
1272				goto out;
1273		}
1274		if (ref->count && ref->parent) {
1275			if (extent_item_pos && !ref->inode_list &&
1276			    ref->level == 0) {
1277				struct extent_buffer *eb;
1278
1279				eb = read_tree_block(fs_info, ref->parent, 0,
1280						     ref->level, NULL);
1281				if (IS_ERR(eb)) {
1282					ret = PTR_ERR(eb);
1283					goto out;
1284				} else if (!extent_buffer_uptodate(eb)) {
1285					free_extent_buffer(eb);
1286					ret = -EIO;
1287					goto out;
1288				}
1289				btrfs_tree_read_lock(eb);
1290				btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1291				ret = find_extent_in_eb(eb, bytenr,
1292							*extent_item_pos, &eie, ignore_offset);
1293				btrfs_tree_read_unlock_blocking(eb);
1294				free_extent_buffer(eb);
1295				if (ret < 0)
1296					goto out;
1297				ref->inode_list = eie;
1298			}
1299			ret = ulist_add_merge_ptr(refs, ref->parent,
1300						  ref->inode_list,
1301						  (void **)&eie, GFP_NOFS);
1302			if (ret < 0)
1303				goto out;
1304			if (!ret && extent_item_pos) {
1305				/*
1306				 * we've recorded that parent, so we must extend
1307				 * its inode list here
1308				 */
1309				BUG_ON(!eie);
1310				while (eie->next)
1311					eie = eie->next;
1312				eie->next = ref->inode_list;
1313			}
1314			eie = NULL;
1315		}
1316		cond_resched();
 
1317	}
1318
1319out:
1320	btrfs_free_path(path);
1321
1322	prelim_release(&preftrees.direct);
1323	prelim_release(&preftrees.indirect);
1324	prelim_release(&preftrees.indirect_missing_keys);
1325
 
 
 
 
 
 
1326	if (ret < 0)
1327		free_inode_elem_list(eie);
1328	return ret;
1329}
1330
1331static void free_leaf_list(struct ulist *blocks)
1332{
1333	struct ulist_node *node = NULL;
1334	struct extent_inode_elem *eie;
1335	struct ulist_iterator uiter;
1336
1337	ULIST_ITER_INIT(&uiter);
1338	while ((node = ulist_next(blocks, &uiter))) {
1339		if (!node->aux)
1340			continue;
1341		eie = unode_aux_to_inode_list(node);
1342		free_inode_elem_list(eie);
1343		node->aux = 0;
1344	}
1345
1346	ulist_free(blocks);
1347}
1348
1349/*
1350 * Finds all leafs with a reference to the specified combination of bytenr and
1351 * offset. key_list_head will point to a list of corresponding keys (caller must
1352 * free each list element). The leafs will be stored in the leafs ulist, which
1353 * must be freed with ulist_free.
1354 *
1355 * returns 0 on success, <0 on error
1356 */
1357static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1358				struct btrfs_fs_info *fs_info, u64 bytenr,
1359				u64 time_seq, struct ulist **leafs,
1360				const u64 *extent_item_pos, bool ignore_offset)
1361{
1362	int ret;
1363
1364	*leafs = ulist_alloc(GFP_NOFS);
1365	if (!*leafs)
1366		return -ENOMEM;
1367
1368	ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1369				*leafs, NULL, extent_item_pos, NULL, ignore_offset);
1370	if (ret < 0 && ret != -ENOENT) {
1371		free_leaf_list(*leafs);
1372		return ret;
1373	}
1374
1375	return 0;
1376}
1377
1378/*
1379 * walk all backrefs for a given extent to find all roots that reference this
1380 * extent. Walking a backref means finding all extents that reference this
1381 * extent and in turn walk the backrefs of those, too. Naturally this is a
1382 * recursive process, but here it is implemented in an iterative fashion: We
1383 * find all referencing extents for the extent in question and put them on a
1384 * list. In turn, we find all referencing extents for those, further appending
1385 * to the list. The way we iterate the list allows adding more elements after
1386 * the current while iterating. The process stops when we reach the end of the
1387 * list. Found roots are added to the roots list.
1388 *
1389 * returns 0 on success, < 0 on error.
1390 */
1391static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1392				     struct btrfs_fs_info *fs_info, u64 bytenr,
1393				     u64 time_seq, struct ulist **roots,
1394				     bool ignore_offset)
1395{
1396	struct ulist *tmp;
1397	struct ulist_node *node = NULL;
1398	struct ulist_iterator uiter;
1399	int ret;
1400
1401	tmp = ulist_alloc(GFP_NOFS);
1402	if (!tmp)
1403		return -ENOMEM;
1404	*roots = ulist_alloc(GFP_NOFS);
1405	if (!*roots) {
1406		ulist_free(tmp);
1407		return -ENOMEM;
1408	}
1409
1410	ULIST_ITER_INIT(&uiter);
1411	while (1) {
1412		ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1413					tmp, *roots, NULL, NULL, ignore_offset);
1414		if (ret < 0 && ret != -ENOENT) {
1415			ulist_free(tmp);
1416			ulist_free(*roots);
1417			return ret;
1418		}
1419		node = ulist_next(tmp, &uiter);
1420		if (!node)
1421			break;
1422		bytenr = node->val;
1423		cond_resched();
1424	}
1425
1426	ulist_free(tmp);
1427	return 0;
1428}
1429
1430int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1431			 struct btrfs_fs_info *fs_info, u64 bytenr,
1432			 u64 time_seq, struct ulist **roots,
1433			 bool ignore_offset)
1434{
1435	int ret;
1436
1437	if (!trans)
1438		down_read(&fs_info->commit_root_sem);
1439	ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1440					time_seq, roots, ignore_offset);
1441	if (!trans)
1442		up_read(&fs_info->commit_root_sem);
1443	return ret;
1444}
1445
1446/**
1447 * btrfs_check_shared - tell us whether an extent is shared
1448 *
 
 
1449 * btrfs_check_shared uses the backref walking code but will short
1450 * circuit as soon as it finds a root or inode that doesn't match the
1451 * one passed in. This provides a significant performance benefit for
1452 * callers (such as fiemap) which want to know whether the extent is
1453 * shared but do not need a ref count.
1454 *
1455 * This attempts to allocate a transaction in order to account for
1456 * delayed refs, but continues on even when the alloc fails.
1457 *
1458 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1459 */
1460int btrfs_check_shared(struct btrfs_root *root, u64 inum, u64 bytenr)
 
 
1461{
1462	struct btrfs_fs_info *fs_info = root->fs_info;
1463	struct btrfs_trans_handle *trans;
1464	struct ulist *tmp = NULL;
1465	struct ulist *roots = NULL;
1466	struct ulist_iterator uiter;
1467	struct ulist_node *node;
1468	struct seq_list elem = SEQ_LIST_INIT(elem);
1469	int ret = 0;
1470	struct share_check shared = {
1471		.root_objectid = root->objectid,
1472		.inum = inum,
1473		.share_count = 0,
1474	};
1475
1476	tmp = ulist_alloc(GFP_NOFS);
1477	roots = ulist_alloc(GFP_NOFS);
1478	if (!tmp || !roots) {
1479		ulist_free(tmp);
1480		ulist_free(roots);
1481		return -ENOMEM;
1482	}
1483
1484	trans = btrfs_join_transaction(root);
1485	if (IS_ERR(trans)) {
1486		trans = NULL;
1487		down_read(&fs_info->commit_root_sem);
1488	} else {
1489		btrfs_get_tree_mod_seq(fs_info, &elem);
1490	}
1491
1492	ULIST_ITER_INIT(&uiter);
1493	while (1) {
1494		ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1495					roots, NULL, &shared, false);
1496		if (ret == BACKREF_FOUND_SHARED) {
1497			/* this is the only condition under which we return 1 */
1498			ret = 1;
1499			break;
1500		}
1501		if (ret < 0 && ret != -ENOENT)
1502			break;
1503		ret = 0;
1504		node = ulist_next(tmp, &uiter);
1505		if (!node)
1506			break;
1507		bytenr = node->val;
1508		shared.share_count = 0;
1509		cond_resched();
1510	}
1511
1512	if (trans) {
1513		btrfs_put_tree_mod_seq(fs_info, &elem);
1514		btrfs_end_transaction(trans);
1515	} else {
1516		up_read(&fs_info->commit_root_sem);
1517	}
1518	ulist_free(tmp);
1519	ulist_free(roots);
1520	return ret;
1521}
1522
1523int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1524			  u64 start_off, struct btrfs_path *path,
1525			  struct btrfs_inode_extref **ret_extref,
1526			  u64 *found_off)
1527{
1528	int ret, slot;
1529	struct btrfs_key key;
1530	struct btrfs_key found_key;
1531	struct btrfs_inode_extref *extref;
1532	const struct extent_buffer *leaf;
1533	unsigned long ptr;
1534
1535	key.objectid = inode_objectid;
1536	key.type = BTRFS_INODE_EXTREF_KEY;
1537	key.offset = start_off;
1538
1539	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1540	if (ret < 0)
1541		return ret;
1542
1543	while (1) {
1544		leaf = path->nodes[0];
1545		slot = path->slots[0];
1546		if (slot >= btrfs_header_nritems(leaf)) {
1547			/*
1548			 * If the item at offset is not found,
1549			 * btrfs_search_slot will point us to the slot
1550			 * where it should be inserted. In our case
1551			 * that will be the slot directly before the
1552			 * next INODE_REF_KEY_V2 item. In the case
1553			 * that we're pointing to the last slot in a
1554			 * leaf, we must move one leaf over.
1555			 */
1556			ret = btrfs_next_leaf(root, path);
1557			if (ret) {
1558				if (ret >= 1)
1559					ret = -ENOENT;
1560				break;
1561			}
1562			continue;
1563		}
1564
1565		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1566
1567		/*
1568		 * Check that we're still looking at an extended ref key for
1569		 * this particular objectid. If we have different
1570		 * objectid or type then there are no more to be found
1571		 * in the tree and we can exit.
1572		 */
1573		ret = -ENOENT;
1574		if (found_key.objectid != inode_objectid)
1575			break;
1576		if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1577			break;
1578
1579		ret = 0;
1580		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1581		extref = (struct btrfs_inode_extref *)ptr;
1582		*ret_extref = extref;
1583		if (found_off)
1584			*found_off = found_key.offset;
1585		break;
1586	}
1587
1588	return ret;
1589}
1590
1591/*
1592 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1593 * Elements of the path are separated by '/' and the path is guaranteed to be
1594 * 0-terminated. the path is only given within the current file system.
1595 * Therefore, it never starts with a '/'. the caller is responsible to provide
1596 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1597 * the start point of the resulting string is returned. this pointer is within
1598 * dest, normally.
1599 * in case the path buffer would overflow, the pointer is decremented further
1600 * as if output was written to the buffer, though no more output is actually
1601 * generated. that way, the caller can determine how much space would be
1602 * required for the path to fit into the buffer. in that case, the returned
1603 * value will be smaller than dest. callers must check this!
1604 */
1605char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1606			u32 name_len, unsigned long name_off,
1607			struct extent_buffer *eb_in, u64 parent,
1608			char *dest, u32 size)
1609{
1610	int slot;
1611	u64 next_inum;
1612	int ret;
1613	s64 bytes_left = ((s64)size) - 1;
1614	struct extent_buffer *eb = eb_in;
1615	struct btrfs_key found_key;
1616	int leave_spinning = path->leave_spinning;
1617	struct btrfs_inode_ref *iref;
1618
1619	if (bytes_left >= 0)
1620		dest[bytes_left] = '\0';
1621
1622	path->leave_spinning = 1;
1623	while (1) {
1624		bytes_left -= name_len;
1625		if (bytes_left >= 0)
1626			read_extent_buffer(eb, dest + bytes_left,
1627					   name_off, name_len);
1628		if (eb != eb_in) {
1629			if (!path->skip_locking)
1630				btrfs_tree_read_unlock_blocking(eb);
1631			free_extent_buffer(eb);
1632		}
1633		ret = btrfs_find_item(fs_root, path, parent, 0,
1634				BTRFS_INODE_REF_KEY, &found_key);
1635		if (ret > 0)
1636			ret = -ENOENT;
1637		if (ret)
1638			break;
1639
1640		next_inum = found_key.offset;
1641
1642		/* regular exit ahead */
1643		if (parent == next_inum)
1644			break;
1645
1646		slot = path->slots[0];
1647		eb = path->nodes[0];
1648		/* make sure we can use eb after releasing the path */
1649		if (eb != eb_in) {
1650			if (!path->skip_locking)
1651				btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1652			path->nodes[0] = NULL;
1653			path->locks[0] = 0;
1654		}
1655		btrfs_release_path(path);
1656		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1657
1658		name_len = btrfs_inode_ref_name_len(eb, iref);
1659		name_off = (unsigned long)(iref + 1);
1660
1661		parent = next_inum;
1662		--bytes_left;
1663		if (bytes_left >= 0)
1664			dest[bytes_left] = '/';
1665	}
1666
1667	btrfs_release_path(path);
1668	path->leave_spinning = leave_spinning;
1669
1670	if (ret)
1671		return ERR_PTR(ret);
1672
1673	return dest + bytes_left;
1674}
1675
1676/*
1677 * this makes the path point to (logical EXTENT_ITEM *)
1678 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1679 * tree blocks and <0 on error.
1680 */
1681int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1682			struct btrfs_path *path, struct btrfs_key *found_key,
1683			u64 *flags_ret)
1684{
1685	int ret;
1686	u64 flags;
1687	u64 size = 0;
1688	u32 item_size;
1689	const struct extent_buffer *eb;
1690	struct btrfs_extent_item *ei;
1691	struct btrfs_key key;
1692
1693	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1694		key.type = BTRFS_METADATA_ITEM_KEY;
1695	else
1696		key.type = BTRFS_EXTENT_ITEM_KEY;
1697	key.objectid = logical;
1698	key.offset = (u64)-1;
1699
1700	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1701	if (ret < 0)
1702		return ret;
1703
1704	ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1705	if (ret) {
1706		if (ret > 0)
1707			ret = -ENOENT;
1708		return ret;
1709	}
1710	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1711	if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1712		size = fs_info->nodesize;
1713	else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1714		size = found_key->offset;
1715
1716	if (found_key->objectid > logical ||
1717	    found_key->objectid + size <= logical) {
1718		btrfs_debug(fs_info,
1719			"logical %llu is not within any extent", logical);
1720		return -ENOENT;
1721	}
1722
1723	eb = path->nodes[0];
1724	item_size = btrfs_item_size_nr(eb, path->slots[0]);
1725	BUG_ON(item_size < sizeof(*ei));
1726
1727	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1728	flags = btrfs_extent_flags(eb, ei);
1729
1730	btrfs_debug(fs_info,
1731		"logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
1732		 logical, logical - found_key->objectid, found_key->objectid,
1733		 found_key->offset, flags, item_size);
1734
1735	WARN_ON(!flags_ret);
1736	if (flags_ret) {
1737		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1738			*flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1739		else if (flags & BTRFS_EXTENT_FLAG_DATA)
1740			*flags_ret = BTRFS_EXTENT_FLAG_DATA;
1741		else
1742			BUG_ON(1);
1743		return 0;
1744	}
1745
1746	return -EIO;
1747}
1748
1749/*
1750 * helper function to iterate extent inline refs. ptr must point to a 0 value
1751 * for the first call and may be modified. it is used to track state.
1752 * if more refs exist, 0 is returned and the next call to
1753 * get_extent_inline_ref must pass the modified ptr parameter to get the
1754 * next ref. after the last ref was processed, 1 is returned.
1755 * returns <0 on error
1756 */
1757static int get_extent_inline_ref(unsigned long *ptr,
1758				 const struct extent_buffer *eb,
1759				 const struct btrfs_key *key,
1760				 const struct btrfs_extent_item *ei,
1761				 u32 item_size,
1762				 struct btrfs_extent_inline_ref **out_eiref,
1763				 int *out_type)
1764{
1765	unsigned long end;
1766	u64 flags;
1767	struct btrfs_tree_block_info *info;
1768
1769	if (!*ptr) {
1770		/* first call */
1771		flags = btrfs_extent_flags(eb, ei);
1772		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1773			if (key->type == BTRFS_METADATA_ITEM_KEY) {
1774				/* a skinny metadata extent */
1775				*out_eiref =
1776				     (struct btrfs_extent_inline_ref *)(ei + 1);
1777			} else {
1778				WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1779				info = (struct btrfs_tree_block_info *)(ei + 1);
1780				*out_eiref =
1781				   (struct btrfs_extent_inline_ref *)(info + 1);
1782			}
1783		} else {
1784			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1785		}
1786		*ptr = (unsigned long)*out_eiref;
1787		if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1788			return -ENOENT;
1789	}
1790
1791	end = (unsigned long)ei + item_size;
1792	*out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1793	*out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
1794						     BTRFS_REF_TYPE_ANY);
1795	if (*out_type == BTRFS_REF_TYPE_INVALID)
1796		return -EINVAL;
1797
1798	*ptr += btrfs_extent_inline_ref_size(*out_type);
1799	WARN_ON(*ptr > end);
1800	if (*ptr == end)
1801		return 1; /* last */
1802
1803	return 0;
1804}
1805
1806/*
1807 * reads the tree block backref for an extent. tree level and root are returned
1808 * through out_level and out_root. ptr must point to a 0 value for the first
1809 * call and may be modified (see get_extent_inline_ref comment).
1810 * returns 0 if data was provided, 1 if there was no more data to provide or
1811 * <0 on error.
1812 */
1813int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1814			    struct btrfs_key *key, struct btrfs_extent_item *ei,
1815			    u32 item_size, u64 *out_root, u8 *out_level)
1816{
1817	int ret;
1818	int type;
1819	struct btrfs_extent_inline_ref *eiref;
1820
1821	if (*ptr == (unsigned long)-1)
1822		return 1;
1823
1824	while (1) {
1825		ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
1826					      &eiref, &type);
1827		if (ret < 0)
1828			return ret;
1829
1830		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1831		    type == BTRFS_SHARED_BLOCK_REF_KEY)
1832			break;
1833
1834		if (ret == 1)
1835			return 1;
1836	}
1837
1838	/* we can treat both ref types equally here */
1839	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1840
1841	if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1842		struct btrfs_tree_block_info *info;
1843
1844		info = (struct btrfs_tree_block_info *)(ei + 1);
1845		*out_level = btrfs_tree_block_level(eb, info);
1846	} else {
1847		ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1848		*out_level = (u8)key->offset;
1849	}
1850
1851	if (ret == 1)
1852		*ptr = (unsigned long)-1;
1853
1854	return 0;
1855}
1856
1857static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
1858			     struct extent_inode_elem *inode_list,
1859			     u64 root, u64 extent_item_objectid,
1860			     iterate_extent_inodes_t *iterate, void *ctx)
1861{
1862	struct extent_inode_elem *eie;
1863	int ret = 0;
1864
1865	for (eie = inode_list; eie; eie = eie->next) {
1866		btrfs_debug(fs_info,
1867			    "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
1868			    extent_item_objectid, eie->inum,
1869			    eie->offset, root);
1870		ret = iterate(eie->inum, eie->offset, root, ctx);
1871		if (ret) {
1872			btrfs_debug(fs_info,
1873				    "stopping iteration for %llu due to ret=%d",
1874				    extent_item_objectid, ret);
1875			break;
1876		}
1877	}
1878
1879	return ret;
1880}
1881
1882/*
1883 * calls iterate() for every inode that references the extent identified by
1884 * the given parameters.
1885 * when the iterator function returns a non-zero value, iteration stops.
1886 */
1887int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1888				u64 extent_item_objectid, u64 extent_item_pos,
1889				int search_commit_root,
1890				iterate_extent_inodes_t *iterate, void *ctx,
1891				bool ignore_offset)
1892{
1893	int ret;
1894	struct btrfs_trans_handle *trans = NULL;
1895	struct ulist *refs = NULL;
1896	struct ulist *roots = NULL;
1897	struct ulist_node *ref_node = NULL;
1898	struct ulist_node *root_node = NULL;
1899	struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1900	struct ulist_iterator ref_uiter;
1901	struct ulist_iterator root_uiter;
1902
1903	btrfs_debug(fs_info, "resolving all inodes for extent %llu",
1904			extent_item_objectid);
1905
1906	if (!search_commit_root) {
1907		trans = btrfs_join_transaction(fs_info->extent_root);
1908		if (IS_ERR(trans))
1909			return PTR_ERR(trans);
1910		btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1911	} else {
1912		down_read(&fs_info->commit_root_sem);
1913	}
1914
1915	ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1916				   tree_mod_seq_elem.seq, &refs,
1917				   &extent_item_pos, ignore_offset);
1918	if (ret)
1919		goto out;
1920
1921	ULIST_ITER_INIT(&ref_uiter);
1922	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1923		ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
1924						tree_mod_seq_elem.seq, &roots,
1925						ignore_offset);
1926		if (ret)
1927			break;
1928		ULIST_ITER_INIT(&root_uiter);
1929		while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1930			btrfs_debug(fs_info,
1931				    "root %llu references leaf %llu, data list %#llx",
1932				    root_node->val, ref_node->val,
1933				    ref_node->aux);
1934			ret = iterate_leaf_refs(fs_info,
1935						(struct extent_inode_elem *)
1936						(uintptr_t)ref_node->aux,
1937						root_node->val,
1938						extent_item_objectid,
1939						iterate, ctx);
1940		}
1941		ulist_free(roots);
1942	}
1943
1944	free_leaf_list(refs);
1945out:
1946	if (!search_commit_root) {
1947		btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1948		btrfs_end_transaction(trans);
1949	} else {
1950		up_read(&fs_info->commit_root_sem);
1951	}
1952
1953	return ret;
1954}
1955
1956int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1957				struct btrfs_path *path,
1958				iterate_extent_inodes_t *iterate, void *ctx,
1959				bool ignore_offset)
1960{
1961	int ret;
1962	u64 extent_item_pos;
1963	u64 flags = 0;
1964	struct btrfs_key found_key;
1965	int search_commit_root = path->search_commit_root;
1966
1967	ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1968	btrfs_release_path(path);
1969	if (ret < 0)
1970		return ret;
1971	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1972		return -EINVAL;
1973
1974	extent_item_pos = logical - found_key.objectid;
1975	ret = iterate_extent_inodes(fs_info, found_key.objectid,
1976					extent_item_pos, search_commit_root,
1977					iterate, ctx, ignore_offset);
1978
1979	return ret;
1980}
1981
1982typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1983			      struct extent_buffer *eb, void *ctx);
1984
1985static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1986			      struct btrfs_path *path,
1987			      iterate_irefs_t *iterate, void *ctx)
1988{
1989	int ret = 0;
1990	int slot;
1991	u32 cur;
1992	u32 len;
1993	u32 name_len;
1994	u64 parent = 0;
1995	int found = 0;
1996	struct extent_buffer *eb;
1997	struct btrfs_item *item;
1998	struct btrfs_inode_ref *iref;
1999	struct btrfs_key found_key;
2000
2001	while (!ret) {
2002		ret = btrfs_find_item(fs_root, path, inum,
2003				parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2004				&found_key);
2005
2006		if (ret < 0)
2007			break;
2008		if (ret) {
2009			ret = found ? 0 : -ENOENT;
2010			break;
2011		}
2012		++found;
2013
2014		parent = found_key.offset;
2015		slot = path->slots[0];
2016		eb = btrfs_clone_extent_buffer(path->nodes[0]);
2017		if (!eb) {
2018			ret = -ENOMEM;
2019			break;
2020		}
2021		extent_buffer_get(eb);
2022		btrfs_tree_read_lock(eb);
2023		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
2024		btrfs_release_path(path);
2025
2026		item = btrfs_item_nr(slot);
2027		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2028
2029		for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
2030			name_len = btrfs_inode_ref_name_len(eb, iref);
2031			/* path must be released before calling iterate()! */
2032			btrfs_debug(fs_root->fs_info,
2033				"following ref at offset %u for inode %llu in tree %llu",
2034				cur, found_key.objectid, fs_root->objectid);
2035			ret = iterate(parent, name_len,
2036				      (unsigned long)(iref + 1), eb, ctx);
2037			if (ret)
2038				break;
2039			len = sizeof(*iref) + name_len;
2040			iref = (struct btrfs_inode_ref *)((char *)iref + len);
2041		}
2042		btrfs_tree_read_unlock_blocking(eb);
2043		free_extent_buffer(eb);
2044	}
2045
2046	btrfs_release_path(path);
2047
2048	return ret;
2049}
2050
2051static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
2052				 struct btrfs_path *path,
2053				 iterate_irefs_t *iterate, void *ctx)
2054{
2055	int ret;
2056	int slot;
2057	u64 offset = 0;
2058	u64 parent;
2059	int found = 0;
2060	struct extent_buffer *eb;
2061	struct btrfs_inode_extref *extref;
2062	u32 item_size;
2063	u32 cur_offset;
2064	unsigned long ptr;
2065
2066	while (1) {
2067		ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2068					    &offset);
2069		if (ret < 0)
2070			break;
2071		if (ret) {
2072			ret = found ? 0 : -ENOENT;
2073			break;
2074		}
2075		++found;
2076
2077		slot = path->slots[0];
2078		eb = btrfs_clone_extent_buffer(path->nodes[0]);
2079		if (!eb) {
2080			ret = -ENOMEM;
2081			break;
2082		}
2083		extent_buffer_get(eb);
2084
2085		btrfs_tree_read_lock(eb);
2086		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
2087		btrfs_release_path(path);
2088
2089		item_size = btrfs_item_size_nr(eb, slot);
2090		ptr = btrfs_item_ptr_offset(eb, slot);
2091		cur_offset = 0;
2092
2093		while (cur_offset < item_size) {
2094			u32 name_len;
2095
2096			extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2097			parent = btrfs_inode_extref_parent(eb, extref);
2098			name_len = btrfs_inode_extref_name_len(eb, extref);
2099			ret = iterate(parent, name_len,
2100				      (unsigned long)&extref->name, eb, ctx);
2101			if (ret)
2102				break;
2103
2104			cur_offset += btrfs_inode_extref_name_len(eb, extref);
2105			cur_offset += sizeof(*extref);
2106		}
2107		btrfs_tree_read_unlock_blocking(eb);
2108		free_extent_buffer(eb);
2109
2110		offset++;
2111	}
2112
2113	btrfs_release_path(path);
2114
2115	return ret;
2116}
2117
2118static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
2119			 struct btrfs_path *path, iterate_irefs_t *iterate,
2120			 void *ctx)
2121{
2122	int ret;
2123	int found_refs = 0;
2124
2125	ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
2126	if (!ret)
2127		++found_refs;
2128	else if (ret != -ENOENT)
2129		return ret;
2130
2131	ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
2132	if (ret == -ENOENT && found_refs)
2133		return 0;
2134
2135	return ret;
2136}
2137
2138/*
2139 * returns 0 if the path could be dumped (probably truncated)
2140 * returns <0 in case of an error
2141 */
2142static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2143			 struct extent_buffer *eb, void *ctx)
2144{
2145	struct inode_fs_paths *ipath = ctx;
2146	char *fspath;
2147	char *fspath_min;
2148	int i = ipath->fspath->elem_cnt;
2149	const int s_ptr = sizeof(char *);
2150	u32 bytes_left;
2151
2152	bytes_left = ipath->fspath->bytes_left > s_ptr ?
2153					ipath->fspath->bytes_left - s_ptr : 0;
2154
2155	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2156	fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2157				   name_off, eb, inum, fspath_min, bytes_left);
2158	if (IS_ERR(fspath))
2159		return PTR_ERR(fspath);
2160
2161	if (fspath > fspath_min) {
2162		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2163		++ipath->fspath->elem_cnt;
2164		ipath->fspath->bytes_left = fspath - fspath_min;
2165	} else {
2166		++ipath->fspath->elem_missed;
2167		ipath->fspath->bytes_missing += fspath_min - fspath;
2168		ipath->fspath->bytes_left = 0;
2169	}
2170
2171	return 0;
2172}
2173
2174/*
2175 * this dumps all file system paths to the inode into the ipath struct, provided
2176 * is has been created large enough. each path is zero-terminated and accessed
2177 * from ipath->fspath->val[i].
2178 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2179 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2180 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2181 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2182 * have been needed to return all paths.
2183 */
2184int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2185{
2186	return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
2187			     inode_to_path, ipath);
2188}
2189
2190struct btrfs_data_container *init_data_container(u32 total_bytes)
2191{
2192	struct btrfs_data_container *data;
2193	size_t alloc_bytes;
2194
2195	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2196	data = kvmalloc(alloc_bytes, GFP_KERNEL);
2197	if (!data)
2198		return ERR_PTR(-ENOMEM);
2199
2200	if (total_bytes >= sizeof(*data)) {
2201		data->bytes_left = total_bytes - sizeof(*data);
2202		data->bytes_missing = 0;
2203	} else {
2204		data->bytes_missing = sizeof(*data) - total_bytes;
2205		data->bytes_left = 0;
2206	}
2207
2208	data->elem_cnt = 0;
2209	data->elem_missed = 0;
2210
2211	return data;
2212}
2213
2214/*
2215 * allocates space to return multiple file system paths for an inode.
2216 * total_bytes to allocate are passed, note that space usable for actual path
2217 * information will be total_bytes - sizeof(struct inode_fs_paths).
2218 * the returned pointer must be freed with free_ipath() in the end.
2219 */
2220struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2221					struct btrfs_path *path)
2222{
2223	struct inode_fs_paths *ifp;
2224	struct btrfs_data_container *fspath;
2225
2226	fspath = init_data_container(total_bytes);
2227	if (IS_ERR(fspath))
2228		return (void *)fspath;
2229
2230	ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2231	if (!ifp) {
2232		kvfree(fspath);
2233		return ERR_PTR(-ENOMEM);
2234	}
2235
2236	ifp->btrfs_path = path;
2237	ifp->fspath = fspath;
2238	ifp->fs_root = fs_root;
2239
2240	return ifp;
2241}
2242
2243void free_ipath(struct inode_fs_paths *ipath)
2244{
2245	if (!ipath)
2246		return;
2247	kvfree(ipath->fspath);
2248	kfree(ipath);
2249}