xref: /openbmc/linux/fs/btrfs/backref.c (revision f3539c12d8196ce0a1993364d30b3a18908470d1)
1 /*
2  * Copyright (C) 2011 STRATO.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/vmalloc.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "backref.h"
23 #include "ulist.h"
24 #include "transaction.h"
25 #include "delayed-ref.h"
26 #include "locking.h"
27 
28 /* Just an arbitrary number so we can be sure this happened */
29 #define BACKREF_FOUND_SHARED 6
30 
31 struct extent_inode_elem {
32 	u64 inum;
33 	u64 offset;
34 	struct extent_inode_elem *next;
35 };
36 
37 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
38 				struct btrfs_file_extent_item *fi,
39 				u64 extent_item_pos,
40 				struct extent_inode_elem **eie)
41 {
42 	u64 offset = 0;
43 	struct extent_inode_elem *e;
44 
45 	if (!btrfs_file_extent_compression(eb, fi) &&
46 	    !btrfs_file_extent_encryption(eb, fi) &&
47 	    !btrfs_file_extent_other_encoding(eb, fi)) {
48 		u64 data_offset;
49 		u64 data_len;
50 
51 		data_offset = btrfs_file_extent_offset(eb, fi);
52 		data_len = btrfs_file_extent_num_bytes(eb, fi);
53 
54 		if (extent_item_pos < data_offset ||
55 		    extent_item_pos >= data_offset + data_len)
56 			return 1;
57 		offset = extent_item_pos - data_offset;
58 	}
59 
60 	e = kmalloc(sizeof(*e), GFP_NOFS);
61 	if (!e)
62 		return -ENOMEM;
63 
64 	e->next = *eie;
65 	e->inum = key->objectid;
66 	e->offset = key->offset + offset;
67 	*eie = e;
68 
69 	return 0;
70 }
71 
72 static void free_inode_elem_list(struct extent_inode_elem *eie)
73 {
74 	struct extent_inode_elem *eie_next;
75 
76 	for (; eie; eie = eie_next) {
77 		eie_next = eie->next;
78 		kfree(eie);
79 	}
80 }
81 
82 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
83 				u64 extent_item_pos,
84 				struct extent_inode_elem **eie)
85 {
86 	u64 disk_byte;
87 	struct btrfs_key key;
88 	struct btrfs_file_extent_item *fi;
89 	int slot;
90 	int nritems;
91 	int extent_type;
92 	int ret;
93 
94 	/*
95 	 * from the shared data ref, we only have the leaf but we need
96 	 * the key. thus, we must look into all items and see that we
97 	 * find one (some) with a reference to our extent item.
98 	 */
99 	nritems = btrfs_header_nritems(eb);
100 	for (slot = 0; slot < nritems; ++slot) {
101 		btrfs_item_key_to_cpu(eb, &key, slot);
102 		if (key.type != BTRFS_EXTENT_DATA_KEY)
103 			continue;
104 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
105 		extent_type = btrfs_file_extent_type(eb, fi);
106 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
107 			continue;
108 		/* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
109 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
110 		if (disk_byte != wanted_disk_byte)
111 			continue;
112 
113 		ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
114 		if (ret < 0)
115 			return ret;
116 	}
117 
118 	return 0;
119 }
120 
121 /*
122  * this structure records all encountered refs on the way up to the root
123  */
124 struct __prelim_ref {
125 	struct list_head list;
126 	u64 root_id;
127 	struct btrfs_key key_for_search;
128 	int level;
129 	int count;
130 	struct extent_inode_elem *inode_list;
131 	u64 parent;
132 	u64 wanted_disk_byte;
133 };
134 
135 static struct kmem_cache *btrfs_prelim_ref_cache;
136 
137 int __init btrfs_prelim_ref_init(void)
138 {
139 	btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
140 					sizeof(struct __prelim_ref),
141 					0,
142 					SLAB_MEM_SPREAD,
143 					NULL);
144 	if (!btrfs_prelim_ref_cache)
145 		return -ENOMEM;
146 	return 0;
147 }
148 
149 void btrfs_prelim_ref_exit(void)
150 {
151 	kmem_cache_destroy(btrfs_prelim_ref_cache);
152 }
153 
154 /*
155  * the rules for all callers of this function are:
156  * - obtaining the parent is the goal
157  * - if you add a key, you must know that it is a correct key
158  * - if you cannot add the parent or a correct key, then we will look into the
159  *   block later to set a correct key
160  *
161  * delayed refs
162  * ============
163  *        backref type | shared | indirect | shared | indirect
164  * information         |   tree |     tree |   data |     data
165  * --------------------+--------+----------+--------+----------
166  *      parent logical |    y   |     -    |    -   |     -
167  *      key to resolve |    -   |     y    |    y   |     y
168  *  tree block logical |    -   |     -    |    -   |     -
169  *  root for resolving |    y   |     y    |    y   |     y
170  *
171  * - column 1:       we've the parent -> done
172  * - column 2, 3, 4: we use the key to find the parent
173  *
174  * on disk refs (inline or keyed)
175  * ==============================
176  *        backref type | shared | indirect | shared | indirect
177  * information         |   tree |     tree |   data |     data
178  * --------------------+--------+----------+--------+----------
179  *      parent logical |    y   |     -    |    y   |     -
180  *      key to resolve |    -   |     -    |    -   |     y
181  *  tree block logical |    y   |     y    |    y   |     y
182  *  root for resolving |    -   |     y    |    y   |     y
183  *
184  * - column 1, 3: we've the parent -> done
185  * - column 2:    we take the first key from the block to find the parent
186  *                (see __add_missing_keys)
187  * - column 4:    we use the key to find the parent
188  *
189  * additional information that's available but not required to find the parent
190  * block might help in merging entries to gain some speed.
191  */
192 
193 static int __add_prelim_ref(struct list_head *head, u64 root_id,
194 			    struct btrfs_key *key, int level,
195 			    u64 parent, u64 wanted_disk_byte, int count,
196 			    gfp_t gfp_mask)
197 {
198 	struct __prelim_ref *ref;
199 
200 	if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
201 		return 0;
202 
203 	ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
204 	if (!ref)
205 		return -ENOMEM;
206 
207 	ref->root_id = root_id;
208 	if (key) {
209 		ref->key_for_search = *key;
210 		/*
211 		 * We can often find data backrefs with an offset that is too
212 		 * large (>= LLONG_MAX, maximum allowed file offset) due to
213 		 * underflows when subtracting a file's offset with the data
214 		 * offset of its corresponding extent data item. This can
215 		 * happen for example in the clone ioctl.
216 		 * So if we detect such case we set the search key's offset to
217 		 * zero to make sure we will find the matching file extent item
218 		 * at add_all_parents(), otherwise we will miss it because the
219 		 * offset taken form the backref is much larger then the offset
220 		 * of the file extent item. This can make us scan a very large
221 		 * number of file extent items, but at least it will not make
222 		 * us miss any.
223 		 * This is an ugly workaround for a behaviour that should have
224 		 * never existed, but it does and a fix for the clone ioctl
225 		 * would touch a lot of places, cause backwards incompatibility
226 		 * and would not fix the problem for extents cloned with older
227 		 * kernels.
228 		 */
229 		if (ref->key_for_search.type == BTRFS_EXTENT_DATA_KEY &&
230 		    ref->key_for_search.offset >= LLONG_MAX)
231 			ref->key_for_search.offset = 0;
232 	} else {
233 		memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
234 	}
235 
236 	ref->inode_list = NULL;
237 	ref->level = level;
238 	ref->count = count;
239 	ref->parent = parent;
240 	ref->wanted_disk_byte = wanted_disk_byte;
241 	list_add_tail(&ref->list, head);
242 
243 	return 0;
244 }
245 
246 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
247 			   struct ulist *parents, struct __prelim_ref *ref,
248 			   int level, u64 time_seq, const u64 *extent_item_pos,
249 			   u64 total_refs)
250 {
251 	int ret = 0;
252 	int slot;
253 	struct extent_buffer *eb;
254 	struct btrfs_key key;
255 	struct btrfs_key *key_for_search = &ref->key_for_search;
256 	struct btrfs_file_extent_item *fi;
257 	struct extent_inode_elem *eie = NULL, *old = NULL;
258 	u64 disk_byte;
259 	u64 wanted_disk_byte = ref->wanted_disk_byte;
260 	u64 count = 0;
261 
262 	if (level != 0) {
263 		eb = path->nodes[level];
264 		ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
265 		if (ret < 0)
266 			return ret;
267 		return 0;
268 	}
269 
270 	/*
271 	 * We normally enter this function with the path already pointing to
272 	 * the first item to check. But sometimes, we may enter it with
273 	 * slot==nritems. In that case, go to the next leaf before we continue.
274 	 */
275 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
276 		if (time_seq == (u64)-1)
277 			ret = btrfs_next_leaf(root, path);
278 		else
279 			ret = btrfs_next_old_leaf(root, path, time_seq);
280 	}
281 
282 	while (!ret && count < total_refs) {
283 		eb = path->nodes[0];
284 		slot = path->slots[0];
285 
286 		btrfs_item_key_to_cpu(eb, &key, slot);
287 
288 		if (key.objectid != key_for_search->objectid ||
289 		    key.type != BTRFS_EXTENT_DATA_KEY)
290 			break;
291 
292 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
293 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
294 
295 		if (disk_byte == wanted_disk_byte) {
296 			eie = NULL;
297 			old = NULL;
298 			count++;
299 			if (extent_item_pos) {
300 				ret = check_extent_in_eb(&key, eb, fi,
301 						*extent_item_pos,
302 						&eie);
303 				if (ret < 0)
304 					break;
305 			}
306 			if (ret > 0)
307 				goto next;
308 			ret = ulist_add_merge_ptr(parents, eb->start,
309 						  eie, (void **)&old, GFP_NOFS);
310 			if (ret < 0)
311 				break;
312 			if (!ret && extent_item_pos) {
313 				while (old->next)
314 					old = old->next;
315 				old->next = eie;
316 			}
317 			eie = NULL;
318 		}
319 next:
320 		if (time_seq == (u64)-1)
321 			ret = btrfs_next_item(root, path);
322 		else
323 			ret = btrfs_next_old_item(root, path, time_seq);
324 	}
325 
326 	if (ret > 0)
327 		ret = 0;
328 	else if (ret < 0)
329 		free_inode_elem_list(eie);
330 	return ret;
331 }
332 
333 /*
334  * resolve an indirect backref in the form (root_id, key, level)
335  * to a logical address
336  */
337 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
338 				  struct btrfs_path *path, u64 time_seq,
339 				  struct __prelim_ref *ref,
340 				  struct ulist *parents,
341 				  const u64 *extent_item_pos, u64 total_refs)
342 {
343 	struct btrfs_root *root;
344 	struct btrfs_key root_key;
345 	struct extent_buffer *eb;
346 	int ret = 0;
347 	int root_level;
348 	int level = ref->level;
349 	int index;
350 
351 	root_key.objectid = ref->root_id;
352 	root_key.type = BTRFS_ROOT_ITEM_KEY;
353 	root_key.offset = (u64)-1;
354 
355 	index = srcu_read_lock(&fs_info->subvol_srcu);
356 
357 	root = btrfs_get_fs_root(fs_info, &root_key, false);
358 	if (IS_ERR(root)) {
359 		srcu_read_unlock(&fs_info->subvol_srcu, index);
360 		ret = PTR_ERR(root);
361 		goto out;
362 	}
363 
364 	if (btrfs_is_testing(fs_info)) {
365 		srcu_read_unlock(&fs_info->subvol_srcu, index);
366 		ret = -ENOENT;
367 		goto out;
368 	}
369 
370 	if (path->search_commit_root)
371 		root_level = btrfs_header_level(root->commit_root);
372 	else if (time_seq == (u64)-1)
373 		root_level = btrfs_header_level(root->node);
374 	else
375 		root_level = btrfs_old_root_level(root, time_seq);
376 
377 	if (root_level + 1 == level) {
378 		srcu_read_unlock(&fs_info->subvol_srcu, index);
379 		goto out;
380 	}
381 
382 	path->lowest_level = level;
383 	if (time_seq == (u64)-1)
384 		ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path,
385 					0, 0);
386 	else
387 		ret = btrfs_search_old_slot(root, &ref->key_for_search, path,
388 					    time_seq);
389 
390 	/* root node has been locked, we can release @subvol_srcu safely here */
391 	srcu_read_unlock(&fs_info->subvol_srcu, index);
392 
393 	pr_debug("search slot in root %llu (level %d, ref count %d) returned "
394 		 "%d for key (%llu %u %llu)\n",
395 		 ref->root_id, level, ref->count, ret,
396 		 ref->key_for_search.objectid, ref->key_for_search.type,
397 		 ref->key_for_search.offset);
398 	if (ret < 0)
399 		goto out;
400 
401 	eb = path->nodes[level];
402 	while (!eb) {
403 		if (WARN_ON(!level)) {
404 			ret = 1;
405 			goto out;
406 		}
407 		level--;
408 		eb = path->nodes[level];
409 	}
410 
411 	ret = add_all_parents(root, path, parents, ref, level, time_seq,
412 			      extent_item_pos, total_refs);
413 out:
414 	path->lowest_level = 0;
415 	btrfs_release_path(path);
416 	return ret;
417 }
418 
419 /*
420  * resolve all indirect backrefs from the list
421  */
422 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
423 				   struct btrfs_path *path, u64 time_seq,
424 				   struct list_head *head,
425 				   const u64 *extent_item_pos, u64 total_refs,
426 				   u64 root_objectid)
427 {
428 	int err;
429 	int ret = 0;
430 	struct __prelim_ref *ref;
431 	struct __prelim_ref *ref_safe;
432 	struct __prelim_ref *new_ref;
433 	struct ulist *parents;
434 	struct ulist_node *node;
435 	struct ulist_iterator uiter;
436 
437 	parents = ulist_alloc(GFP_NOFS);
438 	if (!parents)
439 		return -ENOMEM;
440 
441 	/*
442 	 * _safe allows us to insert directly after the current item without
443 	 * iterating over the newly inserted items.
444 	 * we're also allowed to re-assign ref during iteration.
445 	 */
446 	list_for_each_entry_safe(ref, ref_safe, head, list) {
447 		if (ref->parent)	/* already direct */
448 			continue;
449 		if (ref->count == 0)
450 			continue;
451 		if (root_objectid && ref->root_id != root_objectid) {
452 			ret = BACKREF_FOUND_SHARED;
453 			goto out;
454 		}
455 		err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
456 					     parents, extent_item_pos,
457 					     total_refs);
458 		/*
459 		 * we can only tolerate ENOENT,otherwise,we should catch error
460 		 * and return directly.
461 		 */
462 		if (err == -ENOENT) {
463 			continue;
464 		} else if (err) {
465 			ret = err;
466 			goto out;
467 		}
468 
469 		/* we put the first parent into the ref at hand */
470 		ULIST_ITER_INIT(&uiter);
471 		node = ulist_next(parents, &uiter);
472 		ref->parent = node ? node->val : 0;
473 		ref->inode_list = node ?
474 			(struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
475 
476 		/* additional parents require new refs being added here */
477 		while ((node = ulist_next(parents, &uiter))) {
478 			new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
479 						   GFP_NOFS);
480 			if (!new_ref) {
481 				ret = -ENOMEM;
482 				goto out;
483 			}
484 			memcpy(new_ref, ref, sizeof(*ref));
485 			new_ref->parent = node->val;
486 			new_ref->inode_list = (struct extent_inode_elem *)
487 							(uintptr_t)node->aux;
488 			list_add(&new_ref->list, &ref->list);
489 		}
490 		ulist_reinit(parents);
491 	}
492 out:
493 	ulist_free(parents);
494 	return ret;
495 }
496 
497 static inline int ref_for_same_block(struct __prelim_ref *ref1,
498 				     struct __prelim_ref *ref2)
499 {
500 	if (ref1->level != ref2->level)
501 		return 0;
502 	if (ref1->root_id != ref2->root_id)
503 		return 0;
504 	if (ref1->key_for_search.type != ref2->key_for_search.type)
505 		return 0;
506 	if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
507 		return 0;
508 	if (ref1->key_for_search.offset != ref2->key_for_search.offset)
509 		return 0;
510 	if (ref1->parent != ref2->parent)
511 		return 0;
512 
513 	return 1;
514 }
515 
516 /*
517  * read tree blocks and add keys where required.
518  */
519 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
520 			      struct list_head *head)
521 {
522 	struct __prelim_ref *ref;
523 	struct extent_buffer *eb;
524 
525 	list_for_each_entry(ref, head, list) {
526 		if (ref->parent)
527 			continue;
528 		if (ref->key_for_search.type)
529 			continue;
530 		BUG_ON(!ref->wanted_disk_byte);
531 		eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
532 				     0);
533 		if (IS_ERR(eb)) {
534 			return PTR_ERR(eb);
535 		} else if (!extent_buffer_uptodate(eb)) {
536 			free_extent_buffer(eb);
537 			return -EIO;
538 		}
539 		btrfs_tree_read_lock(eb);
540 		if (btrfs_header_level(eb) == 0)
541 			btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
542 		else
543 			btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
544 		btrfs_tree_read_unlock(eb);
545 		free_extent_buffer(eb);
546 	}
547 	return 0;
548 }
549 
550 /*
551  * merge backrefs and adjust counts accordingly
552  *
553  * mode = 1: merge identical keys, if key is set
554  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
555  *           additionally, we could even add a key range for the blocks we
556  *           looked into to merge even more (-> replace unresolved refs by those
557  *           having a parent).
558  * mode = 2: merge identical parents
559  */
560 static void __merge_refs(struct list_head *head, int mode)
561 {
562 	struct __prelim_ref *pos1;
563 
564 	list_for_each_entry(pos1, head, list) {
565 		struct __prelim_ref *pos2 = pos1, *tmp;
566 
567 		list_for_each_entry_safe_continue(pos2, tmp, head, list) {
568 			struct __prelim_ref *ref1 = pos1, *ref2 = pos2;
569 			struct extent_inode_elem *eie;
570 
571 			if (!ref_for_same_block(ref1, ref2))
572 				continue;
573 			if (mode == 1) {
574 				if (!ref1->parent && ref2->parent)
575 					swap(ref1, ref2);
576 			} else {
577 				if (ref1->parent != ref2->parent)
578 					continue;
579 			}
580 
581 			eie = ref1->inode_list;
582 			while (eie && eie->next)
583 				eie = eie->next;
584 			if (eie)
585 				eie->next = ref2->inode_list;
586 			else
587 				ref1->inode_list = ref2->inode_list;
588 			ref1->count += ref2->count;
589 
590 			list_del(&ref2->list);
591 			kmem_cache_free(btrfs_prelim_ref_cache, ref2);
592 			cond_resched();
593 		}
594 
595 	}
596 }
597 
598 /*
599  * add all currently queued delayed refs from this head whose seq nr is
600  * smaller or equal that seq to the list
601  */
602 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
603 			      struct list_head *prefs, u64 *total_refs,
604 			      u64 inum)
605 {
606 	struct btrfs_delayed_ref_node *node;
607 	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
608 	struct btrfs_key key;
609 	struct btrfs_key op_key = {0};
610 	int sgn;
611 	int ret = 0;
612 
613 	if (extent_op && extent_op->update_key)
614 		btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
615 
616 	spin_lock(&head->lock);
617 	list_for_each_entry(node, &head->ref_list, list) {
618 		if (node->seq > seq)
619 			continue;
620 
621 		switch (node->action) {
622 		case BTRFS_ADD_DELAYED_EXTENT:
623 		case BTRFS_UPDATE_DELAYED_HEAD:
624 			WARN_ON(1);
625 			continue;
626 		case BTRFS_ADD_DELAYED_REF:
627 			sgn = 1;
628 			break;
629 		case BTRFS_DROP_DELAYED_REF:
630 			sgn = -1;
631 			break;
632 		default:
633 			BUG_ON(1);
634 		}
635 		*total_refs += (node->ref_mod * sgn);
636 		switch (node->type) {
637 		case BTRFS_TREE_BLOCK_REF_KEY: {
638 			struct btrfs_delayed_tree_ref *ref;
639 
640 			ref = btrfs_delayed_node_to_tree_ref(node);
641 			ret = __add_prelim_ref(prefs, ref->root, &op_key,
642 					       ref->level + 1, 0, node->bytenr,
643 					       node->ref_mod * sgn, GFP_ATOMIC);
644 			break;
645 		}
646 		case BTRFS_SHARED_BLOCK_REF_KEY: {
647 			struct btrfs_delayed_tree_ref *ref;
648 
649 			ref = btrfs_delayed_node_to_tree_ref(node);
650 			ret = __add_prelim_ref(prefs, 0, NULL,
651 					       ref->level + 1, ref->parent,
652 					       node->bytenr,
653 					       node->ref_mod * sgn, GFP_ATOMIC);
654 			break;
655 		}
656 		case BTRFS_EXTENT_DATA_REF_KEY: {
657 			struct btrfs_delayed_data_ref *ref;
658 			ref = btrfs_delayed_node_to_data_ref(node);
659 
660 			key.objectid = ref->objectid;
661 			key.type = BTRFS_EXTENT_DATA_KEY;
662 			key.offset = ref->offset;
663 
664 			/*
665 			 * Found a inum that doesn't match our known inum, we
666 			 * know it's shared.
667 			 */
668 			if (inum && ref->objectid != inum) {
669 				ret = BACKREF_FOUND_SHARED;
670 				break;
671 			}
672 
673 			ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
674 					       node->bytenr,
675 					       node->ref_mod * sgn, GFP_ATOMIC);
676 			break;
677 		}
678 		case BTRFS_SHARED_DATA_REF_KEY: {
679 			struct btrfs_delayed_data_ref *ref;
680 
681 			ref = btrfs_delayed_node_to_data_ref(node);
682 			ret = __add_prelim_ref(prefs, 0, NULL, 0,
683 					       ref->parent, node->bytenr,
684 					       node->ref_mod * sgn, GFP_ATOMIC);
685 			break;
686 		}
687 		default:
688 			WARN_ON(1);
689 		}
690 		if (ret)
691 			break;
692 	}
693 	spin_unlock(&head->lock);
694 	return ret;
695 }
696 
697 /*
698  * add all inline backrefs for bytenr to the list
699  */
700 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
701 			     struct btrfs_path *path, u64 bytenr,
702 			     int *info_level, struct list_head *prefs,
703 			     u64 *total_refs, u64 inum)
704 {
705 	int ret = 0;
706 	int slot;
707 	struct extent_buffer *leaf;
708 	struct btrfs_key key;
709 	struct btrfs_key found_key;
710 	unsigned long ptr;
711 	unsigned long end;
712 	struct btrfs_extent_item *ei;
713 	u64 flags;
714 	u64 item_size;
715 
716 	/*
717 	 * enumerate all inline refs
718 	 */
719 	leaf = path->nodes[0];
720 	slot = path->slots[0];
721 
722 	item_size = btrfs_item_size_nr(leaf, slot);
723 	BUG_ON(item_size < sizeof(*ei));
724 
725 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
726 	flags = btrfs_extent_flags(leaf, ei);
727 	*total_refs += btrfs_extent_refs(leaf, ei);
728 	btrfs_item_key_to_cpu(leaf, &found_key, slot);
729 
730 	ptr = (unsigned long)(ei + 1);
731 	end = (unsigned long)ei + item_size;
732 
733 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
734 	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
735 		struct btrfs_tree_block_info *info;
736 
737 		info = (struct btrfs_tree_block_info *)ptr;
738 		*info_level = btrfs_tree_block_level(leaf, info);
739 		ptr += sizeof(struct btrfs_tree_block_info);
740 		BUG_ON(ptr > end);
741 	} else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
742 		*info_level = found_key.offset;
743 	} else {
744 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
745 	}
746 
747 	while (ptr < end) {
748 		struct btrfs_extent_inline_ref *iref;
749 		u64 offset;
750 		int type;
751 
752 		iref = (struct btrfs_extent_inline_ref *)ptr;
753 		type = btrfs_extent_inline_ref_type(leaf, iref);
754 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
755 
756 		switch (type) {
757 		case BTRFS_SHARED_BLOCK_REF_KEY:
758 			ret = __add_prelim_ref(prefs, 0, NULL,
759 						*info_level + 1, offset,
760 						bytenr, 1, GFP_NOFS);
761 			break;
762 		case BTRFS_SHARED_DATA_REF_KEY: {
763 			struct btrfs_shared_data_ref *sdref;
764 			int count;
765 
766 			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
767 			count = btrfs_shared_data_ref_count(leaf, sdref);
768 			ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
769 					       bytenr, count, GFP_NOFS);
770 			break;
771 		}
772 		case BTRFS_TREE_BLOCK_REF_KEY:
773 			ret = __add_prelim_ref(prefs, offset, NULL,
774 					       *info_level + 1, 0,
775 					       bytenr, 1, GFP_NOFS);
776 			break;
777 		case BTRFS_EXTENT_DATA_REF_KEY: {
778 			struct btrfs_extent_data_ref *dref;
779 			int count;
780 			u64 root;
781 
782 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
783 			count = btrfs_extent_data_ref_count(leaf, dref);
784 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
785 								      dref);
786 			key.type = BTRFS_EXTENT_DATA_KEY;
787 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
788 
789 			if (inum && key.objectid != inum) {
790 				ret = BACKREF_FOUND_SHARED;
791 				break;
792 			}
793 
794 			root = btrfs_extent_data_ref_root(leaf, dref);
795 			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
796 					       bytenr, count, GFP_NOFS);
797 			break;
798 		}
799 		default:
800 			WARN_ON(1);
801 		}
802 		if (ret)
803 			return ret;
804 		ptr += btrfs_extent_inline_ref_size(type);
805 	}
806 
807 	return 0;
808 }
809 
810 /*
811  * add all non-inline backrefs for bytenr to the list
812  */
813 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
814 			    struct btrfs_path *path, u64 bytenr,
815 			    int info_level, struct list_head *prefs, u64 inum)
816 {
817 	struct btrfs_root *extent_root = fs_info->extent_root;
818 	int ret;
819 	int slot;
820 	struct extent_buffer *leaf;
821 	struct btrfs_key key;
822 
823 	while (1) {
824 		ret = btrfs_next_item(extent_root, path);
825 		if (ret < 0)
826 			break;
827 		if (ret) {
828 			ret = 0;
829 			break;
830 		}
831 
832 		slot = path->slots[0];
833 		leaf = path->nodes[0];
834 		btrfs_item_key_to_cpu(leaf, &key, slot);
835 
836 		if (key.objectid != bytenr)
837 			break;
838 		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
839 			continue;
840 		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
841 			break;
842 
843 		switch (key.type) {
844 		case BTRFS_SHARED_BLOCK_REF_KEY:
845 			ret = __add_prelim_ref(prefs, 0, NULL,
846 						info_level + 1, key.offset,
847 						bytenr, 1, GFP_NOFS);
848 			break;
849 		case BTRFS_SHARED_DATA_REF_KEY: {
850 			struct btrfs_shared_data_ref *sdref;
851 			int count;
852 
853 			sdref = btrfs_item_ptr(leaf, slot,
854 					      struct btrfs_shared_data_ref);
855 			count = btrfs_shared_data_ref_count(leaf, sdref);
856 			ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
857 						bytenr, count, GFP_NOFS);
858 			break;
859 		}
860 		case BTRFS_TREE_BLOCK_REF_KEY:
861 			ret = __add_prelim_ref(prefs, key.offset, NULL,
862 					       info_level + 1, 0,
863 					       bytenr, 1, GFP_NOFS);
864 			break;
865 		case BTRFS_EXTENT_DATA_REF_KEY: {
866 			struct btrfs_extent_data_ref *dref;
867 			int count;
868 			u64 root;
869 
870 			dref = btrfs_item_ptr(leaf, slot,
871 					      struct btrfs_extent_data_ref);
872 			count = btrfs_extent_data_ref_count(leaf, dref);
873 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
874 								      dref);
875 			key.type = BTRFS_EXTENT_DATA_KEY;
876 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
877 
878 			if (inum && key.objectid != inum) {
879 				ret = BACKREF_FOUND_SHARED;
880 				break;
881 			}
882 
883 			root = btrfs_extent_data_ref_root(leaf, dref);
884 			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
885 					       bytenr, count, GFP_NOFS);
886 			break;
887 		}
888 		default:
889 			WARN_ON(1);
890 		}
891 		if (ret)
892 			return ret;
893 
894 	}
895 
896 	return ret;
897 }
898 
899 /*
900  * this adds all existing backrefs (inline backrefs, backrefs and delayed
901  * refs) for the given bytenr to the refs list, merges duplicates and resolves
902  * indirect refs to their parent bytenr.
903  * When roots are found, they're added to the roots list
904  *
905  * NOTE: This can return values > 0
906  *
907  * If time_seq is set to (u64)-1, it will not search delayed_refs, and behave
908  * much like trans == NULL case, the difference only lies in it will not
909  * commit root.
910  * The special case is for qgroup to search roots in commit_transaction().
911  *
912  * FIXME some caching might speed things up
913  */
914 static int find_parent_nodes(struct btrfs_trans_handle *trans,
915 			     struct btrfs_fs_info *fs_info, u64 bytenr,
916 			     u64 time_seq, struct ulist *refs,
917 			     struct ulist *roots, const u64 *extent_item_pos,
918 			     u64 root_objectid, u64 inum)
919 {
920 	struct btrfs_key key;
921 	struct btrfs_path *path;
922 	struct btrfs_delayed_ref_root *delayed_refs = NULL;
923 	struct btrfs_delayed_ref_head *head;
924 	int info_level = 0;
925 	int ret;
926 	struct list_head prefs_delayed;
927 	struct list_head prefs;
928 	struct __prelim_ref *ref;
929 	struct extent_inode_elem *eie = NULL;
930 	u64 total_refs = 0;
931 
932 	INIT_LIST_HEAD(&prefs);
933 	INIT_LIST_HEAD(&prefs_delayed);
934 
935 	key.objectid = bytenr;
936 	key.offset = (u64)-1;
937 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
938 		key.type = BTRFS_METADATA_ITEM_KEY;
939 	else
940 		key.type = BTRFS_EXTENT_ITEM_KEY;
941 
942 	path = btrfs_alloc_path();
943 	if (!path)
944 		return -ENOMEM;
945 	if (!trans) {
946 		path->search_commit_root = 1;
947 		path->skip_locking = 1;
948 	}
949 
950 	if (time_seq == (u64)-1)
951 		path->skip_locking = 1;
952 
953 	/*
954 	 * grab both a lock on the path and a lock on the delayed ref head.
955 	 * We need both to get a consistent picture of how the refs look
956 	 * at a specified point in time
957 	 */
958 again:
959 	head = NULL;
960 
961 	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
962 	if (ret < 0)
963 		goto out;
964 	BUG_ON(ret == 0);
965 
966 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
967 	if (trans && likely(trans->type != __TRANS_DUMMY) &&
968 	    time_seq != (u64)-1) {
969 #else
970 	if (trans && time_seq != (u64)-1) {
971 #endif
972 		/*
973 		 * look if there are updates for this ref queued and lock the
974 		 * head
975 		 */
976 		delayed_refs = &trans->transaction->delayed_refs;
977 		spin_lock(&delayed_refs->lock);
978 		head = btrfs_find_delayed_ref_head(trans, bytenr);
979 		if (head) {
980 			if (!mutex_trylock(&head->mutex)) {
981 				atomic_inc(&head->node.refs);
982 				spin_unlock(&delayed_refs->lock);
983 
984 				btrfs_release_path(path);
985 
986 				/*
987 				 * Mutex was contended, block until it's
988 				 * released and try again
989 				 */
990 				mutex_lock(&head->mutex);
991 				mutex_unlock(&head->mutex);
992 				btrfs_put_delayed_ref(&head->node);
993 				goto again;
994 			}
995 			spin_unlock(&delayed_refs->lock);
996 			ret = __add_delayed_refs(head, time_seq,
997 						 &prefs_delayed, &total_refs,
998 						 inum);
999 			mutex_unlock(&head->mutex);
1000 			if (ret)
1001 				goto out;
1002 		} else {
1003 			spin_unlock(&delayed_refs->lock);
1004 		}
1005 	}
1006 
1007 	if (path->slots[0]) {
1008 		struct extent_buffer *leaf;
1009 		int slot;
1010 
1011 		path->slots[0]--;
1012 		leaf = path->nodes[0];
1013 		slot = path->slots[0];
1014 		btrfs_item_key_to_cpu(leaf, &key, slot);
1015 		if (key.objectid == bytenr &&
1016 		    (key.type == BTRFS_EXTENT_ITEM_KEY ||
1017 		     key.type == BTRFS_METADATA_ITEM_KEY)) {
1018 			ret = __add_inline_refs(fs_info, path, bytenr,
1019 						&info_level, &prefs,
1020 						&total_refs, inum);
1021 			if (ret)
1022 				goto out;
1023 			ret = __add_keyed_refs(fs_info, path, bytenr,
1024 					       info_level, &prefs, inum);
1025 			if (ret)
1026 				goto out;
1027 		}
1028 	}
1029 	btrfs_release_path(path);
1030 
1031 	list_splice_init(&prefs_delayed, &prefs);
1032 
1033 	ret = __add_missing_keys(fs_info, &prefs);
1034 	if (ret)
1035 		goto out;
1036 
1037 	__merge_refs(&prefs, 1);
1038 
1039 	ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1040 				      extent_item_pos, total_refs,
1041 				      root_objectid);
1042 	if (ret)
1043 		goto out;
1044 
1045 	__merge_refs(&prefs, 2);
1046 
1047 	while (!list_empty(&prefs)) {
1048 		ref = list_first_entry(&prefs, struct __prelim_ref, list);
1049 		WARN_ON(ref->count < 0);
1050 		if (roots && ref->count && ref->root_id && ref->parent == 0) {
1051 			if (root_objectid && ref->root_id != root_objectid) {
1052 				ret = BACKREF_FOUND_SHARED;
1053 				goto out;
1054 			}
1055 
1056 			/* no parent == root of tree */
1057 			ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1058 			if (ret < 0)
1059 				goto out;
1060 		}
1061 		if (ref->count && ref->parent) {
1062 			if (extent_item_pos && !ref->inode_list &&
1063 			    ref->level == 0) {
1064 				struct extent_buffer *eb;
1065 
1066 				eb = read_tree_block(fs_info->extent_root,
1067 							   ref->parent, 0);
1068 				if (IS_ERR(eb)) {
1069 					ret = PTR_ERR(eb);
1070 					goto out;
1071 				} else if (!extent_buffer_uptodate(eb)) {
1072 					free_extent_buffer(eb);
1073 					ret = -EIO;
1074 					goto out;
1075 				}
1076 				btrfs_tree_read_lock(eb);
1077 				btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1078 				ret = find_extent_in_eb(eb, bytenr,
1079 							*extent_item_pos, &eie);
1080 				btrfs_tree_read_unlock_blocking(eb);
1081 				free_extent_buffer(eb);
1082 				if (ret < 0)
1083 					goto out;
1084 				ref->inode_list = eie;
1085 			}
1086 			ret = ulist_add_merge_ptr(refs, ref->parent,
1087 						  ref->inode_list,
1088 						  (void **)&eie, GFP_NOFS);
1089 			if (ret < 0)
1090 				goto out;
1091 			if (!ret && extent_item_pos) {
1092 				/*
1093 				 * we've recorded that parent, so we must extend
1094 				 * its inode list here
1095 				 */
1096 				BUG_ON(!eie);
1097 				while (eie->next)
1098 					eie = eie->next;
1099 				eie->next = ref->inode_list;
1100 			}
1101 			eie = NULL;
1102 		}
1103 		list_del(&ref->list);
1104 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1105 	}
1106 
1107 out:
1108 	btrfs_free_path(path);
1109 	while (!list_empty(&prefs)) {
1110 		ref = list_first_entry(&prefs, struct __prelim_ref, list);
1111 		list_del(&ref->list);
1112 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1113 	}
1114 	while (!list_empty(&prefs_delayed)) {
1115 		ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1116 				       list);
1117 		list_del(&ref->list);
1118 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1119 	}
1120 	if (ret < 0)
1121 		free_inode_elem_list(eie);
1122 	return ret;
1123 }
1124 
1125 static void free_leaf_list(struct ulist *blocks)
1126 {
1127 	struct ulist_node *node = NULL;
1128 	struct extent_inode_elem *eie;
1129 	struct ulist_iterator uiter;
1130 
1131 	ULIST_ITER_INIT(&uiter);
1132 	while ((node = ulist_next(blocks, &uiter))) {
1133 		if (!node->aux)
1134 			continue;
1135 		eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1136 		free_inode_elem_list(eie);
1137 		node->aux = 0;
1138 	}
1139 
1140 	ulist_free(blocks);
1141 }
1142 
1143 /*
1144  * Finds all leafs with a reference to the specified combination of bytenr and
1145  * offset. key_list_head will point to a list of corresponding keys (caller must
1146  * free each list element). The leafs will be stored in the leafs ulist, which
1147  * must be freed with ulist_free.
1148  *
1149  * returns 0 on success, <0 on error
1150  */
1151 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1152 				struct btrfs_fs_info *fs_info, u64 bytenr,
1153 				u64 time_seq, struct ulist **leafs,
1154 				const u64 *extent_item_pos)
1155 {
1156 	int ret;
1157 
1158 	*leafs = ulist_alloc(GFP_NOFS);
1159 	if (!*leafs)
1160 		return -ENOMEM;
1161 
1162 	ret = find_parent_nodes(trans, fs_info, bytenr,
1163 				time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1164 	if (ret < 0 && ret != -ENOENT) {
1165 		free_leaf_list(*leafs);
1166 		return ret;
1167 	}
1168 
1169 	return 0;
1170 }
1171 
1172 /*
1173  * walk all backrefs for a given extent to find all roots that reference this
1174  * extent. Walking a backref means finding all extents that reference this
1175  * extent and in turn walk the backrefs of those, too. Naturally this is a
1176  * recursive process, but here it is implemented in an iterative fashion: We
1177  * find all referencing extents for the extent in question and put them on a
1178  * list. In turn, we find all referencing extents for those, further appending
1179  * to the list. The way we iterate the list allows adding more elements after
1180  * the current while iterating. The process stops when we reach the end of the
1181  * list. Found roots are added to the roots list.
1182  *
1183  * returns 0 on success, < 0 on error.
1184  */
1185 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1186 				  struct btrfs_fs_info *fs_info, u64 bytenr,
1187 				  u64 time_seq, struct ulist **roots)
1188 {
1189 	struct ulist *tmp;
1190 	struct ulist_node *node = NULL;
1191 	struct ulist_iterator uiter;
1192 	int ret;
1193 
1194 	tmp = ulist_alloc(GFP_NOFS);
1195 	if (!tmp)
1196 		return -ENOMEM;
1197 	*roots = ulist_alloc(GFP_NOFS);
1198 	if (!*roots) {
1199 		ulist_free(tmp);
1200 		return -ENOMEM;
1201 	}
1202 
1203 	ULIST_ITER_INIT(&uiter);
1204 	while (1) {
1205 		ret = find_parent_nodes(trans, fs_info, bytenr,
1206 					time_seq, tmp, *roots, NULL, 0, 0);
1207 		if (ret < 0 && ret != -ENOENT) {
1208 			ulist_free(tmp);
1209 			ulist_free(*roots);
1210 			return ret;
1211 		}
1212 		node = ulist_next(tmp, &uiter);
1213 		if (!node)
1214 			break;
1215 		bytenr = node->val;
1216 		cond_resched();
1217 	}
1218 
1219 	ulist_free(tmp);
1220 	return 0;
1221 }
1222 
1223 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1224 			 struct btrfs_fs_info *fs_info, u64 bytenr,
1225 			 u64 time_seq, struct ulist **roots)
1226 {
1227 	int ret;
1228 
1229 	if (!trans)
1230 		down_read(&fs_info->commit_root_sem);
1231 	ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1232 	if (!trans)
1233 		up_read(&fs_info->commit_root_sem);
1234 	return ret;
1235 }
1236 
1237 /**
1238  * btrfs_check_shared - tell us whether an extent is shared
1239  *
1240  * @trans: optional trans handle
1241  *
1242  * btrfs_check_shared uses the backref walking code but will short
1243  * circuit as soon as it finds a root or inode that doesn't match the
1244  * one passed in. This provides a significant performance benefit for
1245  * callers (such as fiemap) which want to know whether the extent is
1246  * shared but do not need a ref count.
1247  *
1248  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1249  */
1250 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1251 		       struct btrfs_fs_info *fs_info, u64 root_objectid,
1252 		       u64 inum, u64 bytenr)
1253 {
1254 	struct ulist *tmp = NULL;
1255 	struct ulist *roots = NULL;
1256 	struct ulist_iterator uiter;
1257 	struct ulist_node *node;
1258 	struct seq_list elem = SEQ_LIST_INIT(elem);
1259 	int ret = 0;
1260 
1261 	tmp = ulist_alloc(GFP_NOFS);
1262 	roots = ulist_alloc(GFP_NOFS);
1263 	if (!tmp || !roots) {
1264 		ulist_free(tmp);
1265 		ulist_free(roots);
1266 		return -ENOMEM;
1267 	}
1268 
1269 	if (trans)
1270 		btrfs_get_tree_mod_seq(fs_info, &elem);
1271 	else
1272 		down_read(&fs_info->commit_root_sem);
1273 	ULIST_ITER_INIT(&uiter);
1274 	while (1) {
1275 		ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1276 					roots, NULL, root_objectid, inum);
1277 		if (ret == BACKREF_FOUND_SHARED) {
1278 			/* this is the only condition under which we return 1 */
1279 			ret = 1;
1280 			break;
1281 		}
1282 		if (ret < 0 && ret != -ENOENT)
1283 			break;
1284 		ret = 0;
1285 		node = ulist_next(tmp, &uiter);
1286 		if (!node)
1287 			break;
1288 		bytenr = node->val;
1289 		cond_resched();
1290 	}
1291 	if (trans)
1292 		btrfs_put_tree_mod_seq(fs_info, &elem);
1293 	else
1294 		up_read(&fs_info->commit_root_sem);
1295 	ulist_free(tmp);
1296 	ulist_free(roots);
1297 	return ret;
1298 }
1299 
1300 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1301 			  u64 start_off, struct btrfs_path *path,
1302 			  struct btrfs_inode_extref **ret_extref,
1303 			  u64 *found_off)
1304 {
1305 	int ret, slot;
1306 	struct btrfs_key key;
1307 	struct btrfs_key found_key;
1308 	struct btrfs_inode_extref *extref;
1309 	struct extent_buffer *leaf;
1310 	unsigned long ptr;
1311 
1312 	key.objectid = inode_objectid;
1313 	key.type = BTRFS_INODE_EXTREF_KEY;
1314 	key.offset = start_off;
1315 
1316 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1317 	if (ret < 0)
1318 		return ret;
1319 
1320 	while (1) {
1321 		leaf = path->nodes[0];
1322 		slot = path->slots[0];
1323 		if (slot >= btrfs_header_nritems(leaf)) {
1324 			/*
1325 			 * If the item at offset is not found,
1326 			 * btrfs_search_slot will point us to the slot
1327 			 * where it should be inserted. In our case
1328 			 * that will be the slot directly before the
1329 			 * next INODE_REF_KEY_V2 item. In the case
1330 			 * that we're pointing to the last slot in a
1331 			 * leaf, we must move one leaf over.
1332 			 */
1333 			ret = btrfs_next_leaf(root, path);
1334 			if (ret) {
1335 				if (ret >= 1)
1336 					ret = -ENOENT;
1337 				break;
1338 			}
1339 			continue;
1340 		}
1341 
1342 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1343 
1344 		/*
1345 		 * Check that we're still looking at an extended ref key for
1346 		 * this particular objectid. If we have different
1347 		 * objectid or type then there are no more to be found
1348 		 * in the tree and we can exit.
1349 		 */
1350 		ret = -ENOENT;
1351 		if (found_key.objectid != inode_objectid)
1352 			break;
1353 		if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1354 			break;
1355 
1356 		ret = 0;
1357 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1358 		extref = (struct btrfs_inode_extref *)ptr;
1359 		*ret_extref = extref;
1360 		if (found_off)
1361 			*found_off = found_key.offset;
1362 		break;
1363 	}
1364 
1365 	return ret;
1366 }
1367 
1368 /*
1369  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1370  * Elements of the path are separated by '/' and the path is guaranteed to be
1371  * 0-terminated. the path is only given within the current file system.
1372  * Therefore, it never starts with a '/'. the caller is responsible to provide
1373  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1374  * the start point of the resulting string is returned. this pointer is within
1375  * dest, normally.
1376  * in case the path buffer would overflow, the pointer is decremented further
1377  * as if output was written to the buffer, though no more output is actually
1378  * generated. that way, the caller can determine how much space would be
1379  * required for the path to fit into the buffer. in that case, the returned
1380  * value will be smaller than dest. callers must check this!
1381  */
1382 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1383 			u32 name_len, unsigned long name_off,
1384 			struct extent_buffer *eb_in, u64 parent,
1385 			char *dest, u32 size)
1386 {
1387 	int slot;
1388 	u64 next_inum;
1389 	int ret;
1390 	s64 bytes_left = ((s64)size) - 1;
1391 	struct extent_buffer *eb = eb_in;
1392 	struct btrfs_key found_key;
1393 	int leave_spinning = path->leave_spinning;
1394 	struct btrfs_inode_ref *iref;
1395 
1396 	if (bytes_left >= 0)
1397 		dest[bytes_left] = '\0';
1398 
1399 	path->leave_spinning = 1;
1400 	while (1) {
1401 		bytes_left -= name_len;
1402 		if (bytes_left >= 0)
1403 			read_extent_buffer(eb, dest + bytes_left,
1404 					   name_off, name_len);
1405 		if (eb != eb_in) {
1406 			if (!path->skip_locking)
1407 				btrfs_tree_read_unlock_blocking(eb);
1408 			free_extent_buffer(eb);
1409 		}
1410 		ret = btrfs_find_item(fs_root, path, parent, 0,
1411 				BTRFS_INODE_REF_KEY, &found_key);
1412 		if (ret > 0)
1413 			ret = -ENOENT;
1414 		if (ret)
1415 			break;
1416 
1417 		next_inum = found_key.offset;
1418 
1419 		/* regular exit ahead */
1420 		if (parent == next_inum)
1421 			break;
1422 
1423 		slot = path->slots[0];
1424 		eb = path->nodes[0];
1425 		/* make sure we can use eb after releasing the path */
1426 		if (eb != eb_in) {
1427 			if (!path->skip_locking)
1428 				btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1429 			path->nodes[0] = NULL;
1430 			path->locks[0] = 0;
1431 		}
1432 		btrfs_release_path(path);
1433 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1434 
1435 		name_len = btrfs_inode_ref_name_len(eb, iref);
1436 		name_off = (unsigned long)(iref + 1);
1437 
1438 		parent = next_inum;
1439 		--bytes_left;
1440 		if (bytes_left >= 0)
1441 			dest[bytes_left] = '/';
1442 	}
1443 
1444 	btrfs_release_path(path);
1445 	path->leave_spinning = leave_spinning;
1446 
1447 	if (ret)
1448 		return ERR_PTR(ret);
1449 
1450 	return dest + bytes_left;
1451 }
1452 
1453 /*
1454  * this makes the path point to (logical EXTENT_ITEM *)
1455  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1456  * tree blocks and <0 on error.
1457  */
1458 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1459 			struct btrfs_path *path, struct btrfs_key *found_key,
1460 			u64 *flags_ret)
1461 {
1462 	int ret;
1463 	u64 flags;
1464 	u64 size = 0;
1465 	u32 item_size;
1466 	struct extent_buffer *eb;
1467 	struct btrfs_extent_item *ei;
1468 	struct btrfs_key key;
1469 
1470 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1471 		key.type = BTRFS_METADATA_ITEM_KEY;
1472 	else
1473 		key.type = BTRFS_EXTENT_ITEM_KEY;
1474 	key.objectid = logical;
1475 	key.offset = (u64)-1;
1476 
1477 	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1478 	if (ret < 0)
1479 		return ret;
1480 
1481 	ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1482 	if (ret) {
1483 		if (ret > 0)
1484 			ret = -ENOENT;
1485 		return ret;
1486 	}
1487 	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1488 	if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1489 		size = fs_info->extent_root->nodesize;
1490 	else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1491 		size = found_key->offset;
1492 
1493 	if (found_key->objectid > logical ||
1494 	    found_key->objectid + size <= logical) {
1495 		pr_debug("logical %llu is not within any extent\n", logical);
1496 		return -ENOENT;
1497 	}
1498 
1499 	eb = path->nodes[0];
1500 	item_size = btrfs_item_size_nr(eb, path->slots[0]);
1501 	BUG_ON(item_size < sizeof(*ei));
1502 
1503 	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1504 	flags = btrfs_extent_flags(eb, ei);
1505 
1506 	pr_debug("logical %llu is at position %llu within the extent (%llu "
1507 		 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1508 		 logical, logical - found_key->objectid, found_key->objectid,
1509 		 found_key->offset, flags, item_size);
1510 
1511 	WARN_ON(!flags_ret);
1512 	if (flags_ret) {
1513 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1514 			*flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1515 		else if (flags & BTRFS_EXTENT_FLAG_DATA)
1516 			*flags_ret = BTRFS_EXTENT_FLAG_DATA;
1517 		else
1518 			BUG_ON(1);
1519 		return 0;
1520 	}
1521 
1522 	return -EIO;
1523 }
1524 
1525 /*
1526  * helper function to iterate extent inline refs. ptr must point to a 0 value
1527  * for the first call and may be modified. it is used to track state.
1528  * if more refs exist, 0 is returned and the next call to
1529  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1530  * next ref. after the last ref was processed, 1 is returned.
1531  * returns <0 on error
1532  */
1533 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1534 				   struct btrfs_key *key,
1535 				   struct btrfs_extent_item *ei, u32 item_size,
1536 				   struct btrfs_extent_inline_ref **out_eiref,
1537 				   int *out_type)
1538 {
1539 	unsigned long end;
1540 	u64 flags;
1541 	struct btrfs_tree_block_info *info;
1542 
1543 	if (!*ptr) {
1544 		/* first call */
1545 		flags = btrfs_extent_flags(eb, ei);
1546 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1547 			if (key->type == BTRFS_METADATA_ITEM_KEY) {
1548 				/* a skinny metadata extent */
1549 				*out_eiref =
1550 				     (struct btrfs_extent_inline_ref *)(ei + 1);
1551 			} else {
1552 				WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1553 				info = (struct btrfs_tree_block_info *)(ei + 1);
1554 				*out_eiref =
1555 				   (struct btrfs_extent_inline_ref *)(info + 1);
1556 			}
1557 		} else {
1558 			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1559 		}
1560 		*ptr = (unsigned long)*out_eiref;
1561 		if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1562 			return -ENOENT;
1563 	}
1564 
1565 	end = (unsigned long)ei + item_size;
1566 	*out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1567 	*out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1568 
1569 	*ptr += btrfs_extent_inline_ref_size(*out_type);
1570 	WARN_ON(*ptr > end);
1571 	if (*ptr == end)
1572 		return 1; /* last */
1573 
1574 	return 0;
1575 }
1576 
1577 /*
1578  * reads the tree block backref for an extent. tree level and root are returned
1579  * through out_level and out_root. ptr must point to a 0 value for the first
1580  * call and may be modified (see __get_extent_inline_ref comment).
1581  * returns 0 if data was provided, 1 if there was no more data to provide or
1582  * <0 on error.
1583  */
1584 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1585 			    struct btrfs_key *key, struct btrfs_extent_item *ei,
1586 			    u32 item_size, u64 *out_root, u8 *out_level)
1587 {
1588 	int ret;
1589 	int type;
1590 	struct btrfs_extent_inline_ref *eiref;
1591 
1592 	if (*ptr == (unsigned long)-1)
1593 		return 1;
1594 
1595 	while (1) {
1596 		ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1597 					      &eiref, &type);
1598 		if (ret < 0)
1599 			return ret;
1600 
1601 		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1602 		    type == BTRFS_SHARED_BLOCK_REF_KEY)
1603 			break;
1604 
1605 		if (ret == 1)
1606 			return 1;
1607 	}
1608 
1609 	/* we can treat both ref types equally here */
1610 	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1611 
1612 	if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1613 		struct btrfs_tree_block_info *info;
1614 
1615 		info = (struct btrfs_tree_block_info *)(ei + 1);
1616 		*out_level = btrfs_tree_block_level(eb, info);
1617 	} else {
1618 		ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1619 		*out_level = (u8)key->offset;
1620 	}
1621 
1622 	if (ret == 1)
1623 		*ptr = (unsigned long)-1;
1624 
1625 	return 0;
1626 }
1627 
1628 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1629 				u64 root, u64 extent_item_objectid,
1630 				iterate_extent_inodes_t *iterate, void *ctx)
1631 {
1632 	struct extent_inode_elem *eie;
1633 	int ret = 0;
1634 
1635 	for (eie = inode_list; eie; eie = eie->next) {
1636 		pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1637 			 "root %llu\n", extent_item_objectid,
1638 			 eie->inum, eie->offset, root);
1639 		ret = iterate(eie->inum, eie->offset, root, ctx);
1640 		if (ret) {
1641 			pr_debug("stopping iteration for %llu due to ret=%d\n",
1642 				 extent_item_objectid, ret);
1643 			break;
1644 		}
1645 	}
1646 
1647 	return ret;
1648 }
1649 
1650 /*
1651  * calls iterate() for every inode that references the extent identified by
1652  * the given parameters.
1653  * when the iterator function returns a non-zero value, iteration stops.
1654  */
1655 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1656 				u64 extent_item_objectid, u64 extent_item_pos,
1657 				int search_commit_root,
1658 				iterate_extent_inodes_t *iterate, void *ctx)
1659 {
1660 	int ret;
1661 	struct btrfs_trans_handle *trans = NULL;
1662 	struct ulist *refs = NULL;
1663 	struct ulist *roots = NULL;
1664 	struct ulist_node *ref_node = NULL;
1665 	struct ulist_node *root_node = NULL;
1666 	struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1667 	struct ulist_iterator ref_uiter;
1668 	struct ulist_iterator root_uiter;
1669 
1670 	pr_debug("resolving all inodes for extent %llu\n",
1671 			extent_item_objectid);
1672 
1673 	if (!search_commit_root) {
1674 		trans = btrfs_join_transaction(fs_info->extent_root);
1675 		if (IS_ERR(trans))
1676 			return PTR_ERR(trans);
1677 		btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1678 	} else {
1679 		down_read(&fs_info->commit_root_sem);
1680 	}
1681 
1682 	ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1683 				   tree_mod_seq_elem.seq, &refs,
1684 				   &extent_item_pos);
1685 	if (ret)
1686 		goto out;
1687 
1688 	ULIST_ITER_INIT(&ref_uiter);
1689 	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1690 		ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1691 					     tree_mod_seq_elem.seq, &roots);
1692 		if (ret)
1693 			break;
1694 		ULIST_ITER_INIT(&root_uiter);
1695 		while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1696 			pr_debug("root %llu references leaf %llu, data list "
1697 				 "%#llx\n", root_node->val, ref_node->val,
1698 				 ref_node->aux);
1699 			ret = iterate_leaf_refs((struct extent_inode_elem *)
1700 						(uintptr_t)ref_node->aux,
1701 						root_node->val,
1702 						extent_item_objectid,
1703 						iterate, ctx);
1704 		}
1705 		ulist_free(roots);
1706 	}
1707 
1708 	free_leaf_list(refs);
1709 out:
1710 	if (!search_commit_root) {
1711 		btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1712 		btrfs_end_transaction(trans, fs_info->extent_root);
1713 	} else {
1714 		up_read(&fs_info->commit_root_sem);
1715 	}
1716 
1717 	return ret;
1718 }
1719 
1720 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1721 				struct btrfs_path *path,
1722 				iterate_extent_inodes_t *iterate, void *ctx)
1723 {
1724 	int ret;
1725 	u64 extent_item_pos;
1726 	u64 flags = 0;
1727 	struct btrfs_key found_key;
1728 	int search_commit_root = path->search_commit_root;
1729 
1730 	ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1731 	btrfs_release_path(path);
1732 	if (ret < 0)
1733 		return ret;
1734 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1735 		return -EINVAL;
1736 
1737 	extent_item_pos = logical - found_key.objectid;
1738 	ret = iterate_extent_inodes(fs_info, found_key.objectid,
1739 					extent_item_pos, search_commit_root,
1740 					iterate, ctx);
1741 
1742 	return ret;
1743 }
1744 
1745 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1746 			      struct extent_buffer *eb, void *ctx);
1747 
1748 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1749 			      struct btrfs_path *path,
1750 			      iterate_irefs_t *iterate, void *ctx)
1751 {
1752 	int ret = 0;
1753 	int slot;
1754 	u32 cur;
1755 	u32 len;
1756 	u32 name_len;
1757 	u64 parent = 0;
1758 	int found = 0;
1759 	struct extent_buffer *eb;
1760 	struct btrfs_item *item;
1761 	struct btrfs_inode_ref *iref;
1762 	struct btrfs_key found_key;
1763 
1764 	while (!ret) {
1765 		ret = btrfs_find_item(fs_root, path, inum,
1766 				parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
1767 				&found_key);
1768 
1769 		if (ret < 0)
1770 			break;
1771 		if (ret) {
1772 			ret = found ? 0 : -ENOENT;
1773 			break;
1774 		}
1775 		++found;
1776 
1777 		parent = found_key.offset;
1778 		slot = path->slots[0];
1779 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
1780 		if (!eb) {
1781 			ret = -ENOMEM;
1782 			break;
1783 		}
1784 		extent_buffer_get(eb);
1785 		btrfs_tree_read_lock(eb);
1786 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1787 		btrfs_release_path(path);
1788 
1789 		item = btrfs_item_nr(slot);
1790 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1791 
1792 		for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1793 			name_len = btrfs_inode_ref_name_len(eb, iref);
1794 			/* path must be released before calling iterate()! */
1795 			pr_debug("following ref at offset %u for inode %llu in "
1796 				 "tree %llu\n", cur, found_key.objectid,
1797 				 fs_root->objectid);
1798 			ret = iterate(parent, name_len,
1799 				      (unsigned long)(iref + 1), eb, ctx);
1800 			if (ret)
1801 				break;
1802 			len = sizeof(*iref) + name_len;
1803 			iref = (struct btrfs_inode_ref *)((char *)iref + len);
1804 		}
1805 		btrfs_tree_read_unlock_blocking(eb);
1806 		free_extent_buffer(eb);
1807 	}
1808 
1809 	btrfs_release_path(path);
1810 
1811 	return ret;
1812 }
1813 
1814 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1815 				 struct btrfs_path *path,
1816 				 iterate_irefs_t *iterate, void *ctx)
1817 {
1818 	int ret;
1819 	int slot;
1820 	u64 offset = 0;
1821 	u64 parent;
1822 	int found = 0;
1823 	struct extent_buffer *eb;
1824 	struct btrfs_inode_extref *extref;
1825 	u32 item_size;
1826 	u32 cur_offset;
1827 	unsigned long ptr;
1828 
1829 	while (1) {
1830 		ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1831 					    &offset);
1832 		if (ret < 0)
1833 			break;
1834 		if (ret) {
1835 			ret = found ? 0 : -ENOENT;
1836 			break;
1837 		}
1838 		++found;
1839 
1840 		slot = path->slots[0];
1841 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
1842 		if (!eb) {
1843 			ret = -ENOMEM;
1844 			break;
1845 		}
1846 		extent_buffer_get(eb);
1847 
1848 		btrfs_tree_read_lock(eb);
1849 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1850 		btrfs_release_path(path);
1851 
1852 		item_size = btrfs_item_size_nr(eb, slot);
1853 		ptr = btrfs_item_ptr_offset(eb, slot);
1854 		cur_offset = 0;
1855 
1856 		while (cur_offset < item_size) {
1857 			u32 name_len;
1858 
1859 			extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1860 			parent = btrfs_inode_extref_parent(eb, extref);
1861 			name_len = btrfs_inode_extref_name_len(eb, extref);
1862 			ret = iterate(parent, name_len,
1863 				      (unsigned long)&extref->name, eb, ctx);
1864 			if (ret)
1865 				break;
1866 
1867 			cur_offset += btrfs_inode_extref_name_len(eb, extref);
1868 			cur_offset += sizeof(*extref);
1869 		}
1870 		btrfs_tree_read_unlock_blocking(eb);
1871 		free_extent_buffer(eb);
1872 
1873 		offset++;
1874 	}
1875 
1876 	btrfs_release_path(path);
1877 
1878 	return ret;
1879 }
1880 
1881 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1882 			 struct btrfs_path *path, iterate_irefs_t *iterate,
1883 			 void *ctx)
1884 {
1885 	int ret;
1886 	int found_refs = 0;
1887 
1888 	ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1889 	if (!ret)
1890 		++found_refs;
1891 	else if (ret != -ENOENT)
1892 		return ret;
1893 
1894 	ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1895 	if (ret == -ENOENT && found_refs)
1896 		return 0;
1897 
1898 	return ret;
1899 }
1900 
1901 /*
1902  * returns 0 if the path could be dumped (probably truncated)
1903  * returns <0 in case of an error
1904  */
1905 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1906 			 struct extent_buffer *eb, void *ctx)
1907 {
1908 	struct inode_fs_paths *ipath = ctx;
1909 	char *fspath;
1910 	char *fspath_min;
1911 	int i = ipath->fspath->elem_cnt;
1912 	const int s_ptr = sizeof(char *);
1913 	u32 bytes_left;
1914 
1915 	bytes_left = ipath->fspath->bytes_left > s_ptr ?
1916 					ipath->fspath->bytes_left - s_ptr : 0;
1917 
1918 	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1919 	fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1920 				   name_off, eb, inum, fspath_min, bytes_left);
1921 	if (IS_ERR(fspath))
1922 		return PTR_ERR(fspath);
1923 
1924 	if (fspath > fspath_min) {
1925 		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1926 		++ipath->fspath->elem_cnt;
1927 		ipath->fspath->bytes_left = fspath - fspath_min;
1928 	} else {
1929 		++ipath->fspath->elem_missed;
1930 		ipath->fspath->bytes_missing += fspath_min - fspath;
1931 		ipath->fspath->bytes_left = 0;
1932 	}
1933 
1934 	return 0;
1935 }
1936 
1937 /*
1938  * this dumps all file system paths to the inode into the ipath struct, provided
1939  * is has been created large enough. each path is zero-terminated and accessed
1940  * from ipath->fspath->val[i].
1941  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1942  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1943  * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
1944  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1945  * have been needed to return all paths.
1946  */
1947 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1948 {
1949 	return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1950 			     inode_to_path, ipath);
1951 }
1952 
1953 struct btrfs_data_container *init_data_container(u32 total_bytes)
1954 {
1955 	struct btrfs_data_container *data;
1956 	size_t alloc_bytes;
1957 
1958 	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1959 	data = vmalloc(alloc_bytes);
1960 	if (!data)
1961 		return ERR_PTR(-ENOMEM);
1962 
1963 	if (total_bytes >= sizeof(*data)) {
1964 		data->bytes_left = total_bytes - sizeof(*data);
1965 		data->bytes_missing = 0;
1966 	} else {
1967 		data->bytes_missing = sizeof(*data) - total_bytes;
1968 		data->bytes_left = 0;
1969 	}
1970 
1971 	data->elem_cnt = 0;
1972 	data->elem_missed = 0;
1973 
1974 	return data;
1975 }
1976 
1977 /*
1978  * allocates space to return multiple file system paths for an inode.
1979  * total_bytes to allocate are passed, note that space usable for actual path
1980  * information will be total_bytes - sizeof(struct inode_fs_paths).
1981  * the returned pointer must be freed with free_ipath() in the end.
1982  */
1983 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1984 					struct btrfs_path *path)
1985 {
1986 	struct inode_fs_paths *ifp;
1987 	struct btrfs_data_container *fspath;
1988 
1989 	fspath = init_data_container(total_bytes);
1990 	if (IS_ERR(fspath))
1991 		return (void *)fspath;
1992 
1993 	ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1994 	if (!ifp) {
1995 		vfree(fspath);
1996 		return ERR_PTR(-ENOMEM);
1997 	}
1998 
1999 	ifp->btrfs_path = path;
2000 	ifp->fspath = fspath;
2001 	ifp->fs_root = fs_root;
2002 
2003 	return ifp;
2004 }
2005 
2006 void free_ipath(struct inode_fs_paths *ipath)
2007 {
2008 	if (!ipath)
2009 		return;
2010 	vfree(ipath->fspath);
2011 	kfree(ipath);
2012 }
2013