xref: /openbmc/linux/fs/btrfs/backref.c (revision 8c0b9ee8)
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_RECLAIM_ACCOUNT | 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 	if (btrfs_prelim_ref_cache)
152 		kmem_cache_destroy(btrfs_prelim_ref_cache);
153 }
154 
155 /*
156  * the rules for all callers of this function are:
157  * - obtaining the parent is the goal
158  * - if you add a key, you must know that it is a correct key
159  * - if you cannot add the parent or a correct key, then we will look into the
160  *   block later to set a correct key
161  *
162  * delayed refs
163  * ============
164  *        backref type | shared | indirect | shared | indirect
165  * information         |   tree |     tree |   data |     data
166  * --------------------+--------+----------+--------+----------
167  *      parent logical |    y   |     -    |    -   |     -
168  *      key to resolve |    -   |     y    |    y   |     y
169  *  tree block logical |    -   |     -    |    -   |     -
170  *  root for resolving |    y   |     y    |    y   |     y
171  *
172  * - column 1:       we've the parent -> done
173  * - column 2, 3, 4: we use the key to find the parent
174  *
175  * on disk refs (inline or keyed)
176  * ==============================
177  *        backref type | shared | indirect | shared | indirect
178  * information         |   tree |     tree |   data |     data
179  * --------------------+--------+----------+--------+----------
180  *      parent logical |    y   |     -    |    y   |     -
181  *      key to resolve |    -   |     -    |    -   |     y
182  *  tree block logical |    y   |     y    |    y   |     y
183  *  root for resolving |    -   |     y    |    y   |     y
184  *
185  * - column 1, 3: we've the parent -> done
186  * - column 2:    we take the first key from the block to find the parent
187  *                (see __add_missing_keys)
188  * - column 4:    we use the key to find the parent
189  *
190  * additional information that's available but not required to find the parent
191  * block might help in merging entries to gain some speed.
192  */
193 
194 static int __add_prelim_ref(struct list_head *head, u64 root_id,
195 			    struct btrfs_key *key, int level,
196 			    u64 parent, u64 wanted_disk_byte, int count,
197 			    gfp_t gfp_mask)
198 {
199 	struct __prelim_ref *ref;
200 
201 	if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
202 		return 0;
203 
204 	ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
205 	if (!ref)
206 		return -ENOMEM;
207 
208 	ref->root_id = root_id;
209 	if (key)
210 		ref->key_for_search = *key;
211 	else
212 		memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
213 
214 	ref->inode_list = NULL;
215 	ref->level = level;
216 	ref->count = count;
217 	ref->parent = parent;
218 	ref->wanted_disk_byte = wanted_disk_byte;
219 	list_add_tail(&ref->list, head);
220 
221 	return 0;
222 }
223 
224 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
225 			   struct ulist *parents, struct __prelim_ref *ref,
226 			   int level, u64 time_seq, const u64 *extent_item_pos,
227 			   u64 total_refs)
228 {
229 	int ret = 0;
230 	int slot;
231 	struct extent_buffer *eb;
232 	struct btrfs_key key;
233 	struct btrfs_key *key_for_search = &ref->key_for_search;
234 	struct btrfs_file_extent_item *fi;
235 	struct extent_inode_elem *eie = NULL, *old = NULL;
236 	u64 disk_byte;
237 	u64 wanted_disk_byte = ref->wanted_disk_byte;
238 	u64 count = 0;
239 
240 	if (level != 0) {
241 		eb = path->nodes[level];
242 		ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
243 		if (ret < 0)
244 			return ret;
245 		return 0;
246 	}
247 
248 	/*
249 	 * We normally enter this function with the path already pointing to
250 	 * the first item to check. But sometimes, we may enter it with
251 	 * slot==nritems. In that case, go to the next leaf before we continue.
252 	 */
253 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
254 		ret = btrfs_next_old_leaf(root, path, time_seq);
255 
256 	while (!ret && count < total_refs) {
257 		eb = path->nodes[0];
258 		slot = path->slots[0];
259 
260 		btrfs_item_key_to_cpu(eb, &key, slot);
261 
262 		if (key.objectid != key_for_search->objectid ||
263 		    key.type != BTRFS_EXTENT_DATA_KEY)
264 			break;
265 
266 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
267 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
268 
269 		if (disk_byte == wanted_disk_byte) {
270 			eie = NULL;
271 			old = NULL;
272 			count++;
273 			if (extent_item_pos) {
274 				ret = check_extent_in_eb(&key, eb, fi,
275 						*extent_item_pos,
276 						&eie);
277 				if (ret < 0)
278 					break;
279 			}
280 			if (ret > 0)
281 				goto next;
282 			ret = ulist_add_merge_ptr(parents, eb->start,
283 						  eie, (void **)&old, GFP_NOFS);
284 			if (ret < 0)
285 				break;
286 			if (!ret && extent_item_pos) {
287 				while (old->next)
288 					old = old->next;
289 				old->next = eie;
290 			}
291 			eie = NULL;
292 		}
293 next:
294 		ret = btrfs_next_old_item(root, path, time_seq);
295 	}
296 
297 	if (ret > 0)
298 		ret = 0;
299 	else if (ret < 0)
300 		free_inode_elem_list(eie);
301 	return ret;
302 }
303 
304 /*
305  * resolve an indirect backref in the form (root_id, key, level)
306  * to a logical address
307  */
308 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
309 				  struct btrfs_path *path, u64 time_seq,
310 				  struct __prelim_ref *ref,
311 				  struct ulist *parents,
312 				  const u64 *extent_item_pos, u64 total_refs)
313 {
314 	struct btrfs_root *root;
315 	struct btrfs_key root_key;
316 	struct extent_buffer *eb;
317 	int ret = 0;
318 	int root_level;
319 	int level = ref->level;
320 	int index;
321 
322 	root_key.objectid = ref->root_id;
323 	root_key.type = BTRFS_ROOT_ITEM_KEY;
324 	root_key.offset = (u64)-1;
325 
326 	index = srcu_read_lock(&fs_info->subvol_srcu);
327 
328 	root = btrfs_read_fs_root_no_name(fs_info, &root_key);
329 	if (IS_ERR(root)) {
330 		srcu_read_unlock(&fs_info->subvol_srcu, index);
331 		ret = PTR_ERR(root);
332 		goto out;
333 	}
334 
335 	if (path->search_commit_root)
336 		root_level = btrfs_header_level(root->commit_root);
337 	else
338 		root_level = btrfs_old_root_level(root, time_seq);
339 
340 	if (root_level + 1 == level) {
341 		srcu_read_unlock(&fs_info->subvol_srcu, index);
342 		goto out;
343 	}
344 
345 	path->lowest_level = level;
346 	ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
347 
348 	/* root node has been locked, we can release @subvol_srcu safely here */
349 	srcu_read_unlock(&fs_info->subvol_srcu, index);
350 
351 	pr_debug("search slot in root %llu (level %d, ref count %d) returned "
352 		 "%d for key (%llu %u %llu)\n",
353 		 ref->root_id, level, ref->count, ret,
354 		 ref->key_for_search.objectid, ref->key_for_search.type,
355 		 ref->key_for_search.offset);
356 	if (ret < 0)
357 		goto out;
358 
359 	eb = path->nodes[level];
360 	while (!eb) {
361 		if (WARN_ON(!level)) {
362 			ret = 1;
363 			goto out;
364 		}
365 		level--;
366 		eb = path->nodes[level];
367 	}
368 
369 	ret = add_all_parents(root, path, parents, ref, level, time_seq,
370 			      extent_item_pos, total_refs);
371 out:
372 	path->lowest_level = 0;
373 	btrfs_release_path(path);
374 	return ret;
375 }
376 
377 /*
378  * resolve all indirect backrefs from the list
379  */
380 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
381 				   struct btrfs_path *path, u64 time_seq,
382 				   struct list_head *head,
383 				   const u64 *extent_item_pos, u64 total_refs,
384 				   u64 root_objectid)
385 {
386 	int err;
387 	int ret = 0;
388 	struct __prelim_ref *ref;
389 	struct __prelim_ref *ref_safe;
390 	struct __prelim_ref *new_ref;
391 	struct ulist *parents;
392 	struct ulist_node *node;
393 	struct ulist_iterator uiter;
394 
395 	parents = ulist_alloc(GFP_NOFS);
396 	if (!parents)
397 		return -ENOMEM;
398 
399 	/*
400 	 * _safe allows us to insert directly after the current item without
401 	 * iterating over the newly inserted items.
402 	 * we're also allowed to re-assign ref during iteration.
403 	 */
404 	list_for_each_entry_safe(ref, ref_safe, head, list) {
405 		if (ref->parent)	/* already direct */
406 			continue;
407 		if (ref->count == 0)
408 			continue;
409 		if (root_objectid && ref->root_id != root_objectid) {
410 			ret = BACKREF_FOUND_SHARED;
411 			goto out;
412 		}
413 		err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
414 					     parents, extent_item_pos,
415 					     total_refs);
416 		/*
417 		 * we can only tolerate ENOENT,otherwise,we should catch error
418 		 * and return directly.
419 		 */
420 		if (err == -ENOENT) {
421 			continue;
422 		} else if (err) {
423 			ret = err;
424 			goto out;
425 		}
426 
427 		/* we put the first parent into the ref at hand */
428 		ULIST_ITER_INIT(&uiter);
429 		node = ulist_next(parents, &uiter);
430 		ref->parent = node ? node->val : 0;
431 		ref->inode_list = node ?
432 			(struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
433 
434 		/* additional parents require new refs being added here */
435 		while ((node = ulist_next(parents, &uiter))) {
436 			new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
437 						   GFP_NOFS);
438 			if (!new_ref) {
439 				ret = -ENOMEM;
440 				goto out;
441 			}
442 			memcpy(new_ref, ref, sizeof(*ref));
443 			new_ref->parent = node->val;
444 			new_ref->inode_list = (struct extent_inode_elem *)
445 							(uintptr_t)node->aux;
446 			list_add(&new_ref->list, &ref->list);
447 		}
448 		ulist_reinit(parents);
449 	}
450 out:
451 	ulist_free(parents);
452 	return ret;
453 }
454 
455 static inline int ref_for_same_block(struct __prelim_ref *ref1,
456 				     struct __prelim_ref *ref2)
457 {
458 	if (ref1->level != ref2->level)
459 		return 0;
460 	if (ref1->root_id != ref2->root_id)
461 		return 0;
462 	if (ref1->key_for_search.type != ref2->key_for_search.type)
463 		return 0;
464 	if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
465 		return 0;
466 	if (ref1->key_for_search.offset != ref2->key_for_search.offset)
467 		return 0;
468 	if (ref1->parent != ref2->parent)
469 		return 0;
470 
471 	return 1;
472 }
473 
474 /*
475  * read tree blocks and add keys where required.
476  */
477 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
478 			      struct list_head *head)
479 {
480 	struct list_head *pos;
481 	struct extent_buffer *eb;
482 
483 	list_for_each(pos, head) {
484 		struct __prelim_ref *ref;
485 		ref = list_entry(pos, struct __prelim_ref, list);
486 
487 		if (ref->parent)
488 			continue;
489 		if (ref->key_for_search.type)
490 			continue;
491 		BUG_ON(!ref->wanted_disk_byte);
492 		eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
493 				     0);
494 		if (!eb || !extent_buffer_uptodate(eb)) {
495 			free_extent_buffer(eb);
496 			return -EIO;
497 		}
498 		btrfs_tree_read_lock(eb);
499 		if (btrfs_header_level(eb) == 0)
500 			btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
501 		else
502 			btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
503 		btrfs_tree_read_unlock(eb);
504 		free_extent_buffer(eb);
505 	}
506 	return 0;
507 }
508 
509 /*
510  * merge two lists of backrefs and adjust counts accordingly
511  *
512  * mode = 1: merge identical keys, if key is set
513  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
514  *           additionally, we could even add a key range for the blocks we
515  *           looked into to merge even more (-> replace unresolved refs by those
516  *           having a parent).
517  * mode = 2: merge identical parents
518  */
519 static void __merge_refs(struct list_head *head, int mode)
520 {
521 	struct list_head *pos1;
522 
523 	list_for_each(pos1, head) {
524 		struct list_head *n2;
525 		struct list_head *pos2;
526 		struct __prelim_ref *ref1;
527 
528 		ref1 = list_entry(pos1, struct __prelim_ref, list);
529 
530 		for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
531 		     pos2 = n2, n2 = pos2->next) {
532 			struct __prelim_ref *ref2;
533 			struct __prelim_ref *xchg;
534 			struct extent_inode_elem *eie;
535 
536 			ref2 = list_entry(pos2, struct __prelim_ref, list);
537 
538 			if (mode == 1) {
539 				if (!ref_for_same_block(ref1, ref2))
540 					continue;
541 				if (!ref1->parent && ref2->parent) {
542 					xchg = ref1;
543 					ref1 = ref2;
544 					ref2 = xchg;
545 				}
546 			} else {
547 				if (ref1->parent != ref2->parent)
548 					continue;
549 			}
550 
551 			eie = ref1->inode_list;
552 			while (eie && eie->next)
553 				eie = eie->next;
554 			if (eie)
555 				eie->next = ref2->inode_list;
556 			else
557 				ref1->inode_list = ref2->inode_list;
558 			ref1->count += ref2->count;
559 
560 			list_del(&ref2->list);
561 			kmem_cache_free(btrfs_prelim_ref_cache, ref2);
562 		}
563 
564 	}
565 }
566 
567 /*
568  * add all currently queued delayed refs from this head whose seq nr is
569  * smaller or equal that seq to the list
570  */
571 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
572 			      struct list_head *prefs, u64 *total_refs,
573 			      u64 inum)
574 {
575 	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
576 	struct rb_node *n = &head->node.rb_node;
577 	struct btrfs_key key;
578 	struct btrfs_key op_key = {0};
579 	int sgn;
580 	int ret = 0;
581 
582 	if (extent_op && extent_op->update_key)
583 		btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
584 
585 	spin_lock(&head->lock);
586 	n = rb_first(&head->ref_root);
587 	while (n) {
588 		struct btrfs_delayed_ref_node *node;
589 		node = rb_entry(n, struct btrfs_delayed_ref_node,
590 				rb_node);
591 		n = rb_next(n);
592 		if (node->seq > seq)
593 			continue;
594 
595 		switch (node->action) {
596 		case BTRFS_ADD_DELAYED_EXTENT:
597 		case BTRFS_UPDATE_DELAYED_HEAD:
598 			WARN_ON(1);
599 			continue;
600 		case BTRFS_ADD_DELAYED_REF:
601 			sgn = 1;
602 			break;
603 		case BTRFS_DROP_DELAYED_REF:
604 			sgn = -1;
605 			break;
606 		default:
607 			BUG_ON(1);
608 		}
609 		*total_refs += (node->ref_mod * sgn);
610 		switch (node->type) {
611 		case BTRFS_TREE_BLOCK_REF_KEY: {
612 			struct btrfs_delayed_tree_ref *ref;
613 
614 			ref = btrfs_delayed_node_to_tree_ref(node);
615 			ret = __add_prelim_ref(prefs, ref->root, &op_key,
616 					       ref->level + 1, 0, node->bytenr,
617 					       node->ref_mod * sgn, GFP_ATOMIC);
618 			break;
619 		}
620 		case BTRFS_SHARED_BLOCK_REF_KEY: {
621 			struct btrfs_delayed_tree_ref *ref;
622 
623 			ref = btrfs_delayed_node_to_tree_ref(node);
624 			ret = __add_prelim_ref(prefs, ref->root, NULL,
625 					       ref->level + 1, ref->parent,
626 					       node->bytenr,
627 					       node->ref_mod * sgn, GFP_ATOMIC);
628 			break;
629 		}
630 		case BTRFS_EXTENT_DATA_REF_KEY: {
631 			struct btrfs_delayed_data_ref *ref;
632 			ref = btrfs_delayed_node_to_data_ref(node);
633 
634 			key.objectid = ref->objectid;
635 			key.type = BTRFS_EXTENT_DATA_KEY;
636 			key.offset = ref->offset;
637 
638 			/*
639 			 * Found a inum that doesn't match our known inum, we
640 			 * know it's shared.
641 			 */
642 			if (inum && ref->objectid != inum) {
643 				ret = BACKREF_FOUND_SHARED;
644 				break;
645 			}
646 
647 			ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
648 					       node->bytenr,
649 					       node->ref_mod * sgn, GFP_ATOMIC);
650 			break;
651 		}
652 		case BTRFS_SHARED_DATA_REF_KEY: {
653 			struct btrfs_delayed_data_ref *ref;
654 
655 			ref = btrfs_delayed_node_to_data_ref(node);
656 
657 			key.objectid = ref->objectid;
658 			key.type = BTRFS_EXTENT_DATA_KEY;
659 			key.offset = ref->offset;
660 			ret = __add_prelim_ref(prefs, ref->root, &key, 0,
661 					       ref->parent, node->bytenr,
662 					       node->ref_mod * sgn, GFP_ATOMIC);
663 			break;
664 		}
665 		default:
666 			WARN_ON(1);
667 		}
668 		if (ret)
669 			break;
670 	}
671 	spin_unlock(&head->lock);
672 	return ret;
673 }
674 
675 /*
676  * add all inline backrefs for bytenr to the list
677  */
678 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
679 			     struct btrfs_path *path, u64 bytenr,
680 			     int *info_level, struct list_head *prefs,
681 			     u64 *total_refs, u64 inum)
682 {
683 	int ret = 0;
684 	int slot;
685 	struct extent_buffer *leaf;
686 	struct btrfs_key key;
687 	struct btrfs_key found_key;
688 	unsigned long ptr;
689 	unsigned long end;
690 	struct btrfs_extent_item *ei;
691 	u64 flags;
692 	u64 item_size;
693 
694 	/*
695 	 * enumerate all inline refs
696 	 */
697 	leaf = path->nodes[0];
698 	slot = path->slots[0];
699 
700 	item_size = btrfs_item_size_nr(leaf, slot);
701 	BUG_ON(item_size < sizeof(*ei));
702 
703 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
704 	flags = btrfs_extent_flags(leaf, ei);
705 	*total_refs += btrfs_extent_refs(leaf, ei);
706 	btrfs_item_key_to_cpu(leaf, &found_key, slot);
707 
708 	ptr = (unsigned long)(ei + 1);
709 	end = (unsigned long)ei + item_size;
710 
711 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
712 	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
713 		struct btrfs_tree_block_info *info;
714 
715 		info = (struct btrfs_tree_block_info *)ptr;
716 		*info_level = btrfs_tree_block_level(leaf, info);
717 		ptr += sizeof(struct btrfs_tree_block_info);
718 		BUG_ON(ptr > end);
719 	} else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
720 		*info_level = found_key.offset;
721 	} else {
722 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
723 	}
724 
725 	while (ptr < end) {
726 		struct btrfs_extent_inline_ref *iref;
727 		u64 offset;
728 		int type;
729 
730 		iref = (struct btrfs_extent_inline_ref *)ptr;
731 		type = btrfs_extent_inline_ref_type(leaf, iref);
732 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
733 
734 		switch (type) {
735 		case BTRFS_SHARED_BLOCK_REF_KEY:
736 			ret = __add_prelim_ref(prefs, 0, NULL,
737 						*info_level + 1, offset,
738 						bytenr, 1, GFP_NOFS);
739 			break;
740 		case BTRFS_SHARED_DATA_REF_KEY: {
741 			struct btrfs_shared_data_ref *sdref;
742 			int count;
743 
744 			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
745 			count = btrfs_shared_data_ref_count(leaf, sdref);
746 			ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
747 					       bytenr, count, GFP_NOFS);
748 			break;
749 		}
750 		case BTRFS_TREE_BLOCK_REF_KEY:
751 			ret = __add_prelim_ref(prefs, offset, NULL,
752 					       *info_level + 1, 0,
753 					       bytenr, 1, GFP_NOFS);
754 			break;
755 		case BTRFS_EXTENT_DATA_REF_KEY: {
756 			struct btrfs_extent_data_ref *dref;
757 			int count;
758 			u64 root;
759 
760 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
761 			count = btrfs_extent_data_ref_count(leaf, dref);
762 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
763 								      dref);
764 			key.type = BTRFS_EXTENT_DATA_KEY;
765 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
766 
767 			if (inum && key.objectid != inum) {
768 				ret = BACKREF_FOUND_SHARED;
769 				break;
770 			}
771 
772 			root = btrfs_extent_data_ref_root(leaf, dref);
773 			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
774 					       bytenr, count, GFP_NOFS);
775 			break;
776 		}
777 		default:
778 			WARN_ON(1);
779 		}
780 		if (ret)
781 			return ret;
782 		ptr += btrfs_extent_inline_ref_size(type);
783 	}
784 
785 	return 0;
786 }
787 
788 /*
789  * add all non-inline backrefs for bytenr to the list
790  */
791 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
792 			    struct btrfs_path *path, u64 bytenr,
793 			    int info_level, struct list_head *prefs, u64 inum)
794 {
795 	struct btrfs_root *extent_root = fs_info->extent_root;
796 	int ret;
797 	int slot;
798 	struct extent_buffer *leaf;
799 	struct btrfs_key key;
800 
801 	while (1) {
802 		ret = btrfs_next_item(extent_root, path);
803 		if (ret < 0)
804 			break;
805 		if (ret) {
806 			ret = 0;
807 			break;
808 		}
809 
810 		slot = path->slots[0];
811 		leaf = path->nodes[0];
812 		btrfs_item_key_to_cpu(leaf, &key, slot);
813 
814 		if (key.objectid != bytenr)
815 			break;
816 		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
817 			continue;
818 		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
819 			break;
820 
821 		switch (key.type) {
822 		case BTRFS_SHARED_BLOCK_REF_KEY:
823 			ret = __add_prelim_ref(prefs, 0, NULL,
824 						info_level + 1, key.offset,
825 						bytenr, 1, GFP_NOFS);
826 			break;
827 		case BTRFS_SHARED_DATA_REF_KEY: {
828 			struct btrfs_shared_data_ref *sdref;
829 			int count;
830 
831 			sdref = btrfs_item_ptr(leaf, slot,
832 					      struct btrfs_shared_data_ref);
833 			count = btrfs_shared_data_ref_count(leaf, sdref);
834 			ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
835 						bytenr, count, GFP_NOFS);
836 			break;
837 		}
838 		case BTRFS_TREE_BLOCK_REF_KEY:
839 			ret = __add_prelim_ref(prefs, key.offset, NULL,
840 					       info_level + 1, 0,
841 					       bytenr, 1, GFP_NOFS);
842 			break;
843 		case BTRFS_EXTENT_DATA_REF_KEY: {
844 			struct btrfs_extent_data_ref *dref;
845 			int count;
846 			u64 root;
847 
848 			dref = btrfs_item_ptr(leaf, slot,
849 					      struct btrfs_extent_data_ref);
850 			count = btrfs_extent_data_ref_count(leaf, dref);
851 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
852 								      dref);
853 			key.type = BTRFS_EXTENT_DATA_KEY;
854 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
855 
856 			if (inum && key.objectid != inum) {
857 				ret = BACKREF_FOUND_SHARED;
858 				break;
859 			}
860 
861 			root = btrfs_extent_data_ref_root(leaf, dref);
862 			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
863 					       bytenr, count, GFP_NOFS);
864 			break;
865 		}
866 		default:
867 			WARN_ON(1);
868 		}
869 		if (ret)
870 			return ret;
871 
872 	}
873 
874 	return ret;
875 }
876 
877 /*
878  * this adds all existing backrefs (inline backrefs, backrefs and delayed
879  * refs) for the given bytenr to the refs list, merges duplicates and resolves
880  * indirect refs to their parent bytenr.
881  * When roots are found, they're added to the roots list
882  *
883  * FIXME some caching might speed things up
884  */
885 static int find_parent_nodes(struct btrfs_trans_handle *trans,
886 			     struct btrfs_fs_info *fs_info, u64 bytenr,
887 			     u64 time_seq, struct ulist *refs,
888 			     struct ulist *roots, const u64 *extent_item_pos,
889 			     u64 root_objectid, u64 inum)
890 {
891 	struct btrfs_key key;
892 	struct btrfs_path *path;
893 	struct btrfs_delayed_ref_root *delayed_refs = NULL;
894 	struct btrfs_delayed_ref_head *head;
895 	int info_level = 0;
896 	int ret;
897 	struct list_head prefs_delayed;
898 	struct list_head prefs;
899 	struct __prelim_ref *ref;
900 	struct extent_inode_elem *eie = NULL;
901 	u64 total_refs = 0;
902 
903 	INIT_LIST_HEAD(&prefs);
904 	INIT_LIST_HEAD(&prefs_delayed);
905 
906 	key.objectid = bytenr;
907 	key.offset = (u64)-1;
908 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
909 		key.type = BTRFS_METADATA_ITEM_KEY;
910 	else
911 		key.type = BTRFS_EXTENT_ITEM_KEY;
912 
913 	path = btrfs_alloc_path();
914 	if (!path)
915 		return -ENOMEM;
916 	if (!trans) {
917 		path->search_commit_root = 1;
918 		path->skip_locking = 1;
919 	}
920 
921 	/*
922 	 * grab both a lock on the path and a lock on the delayed ref head.
923 	 * We need both to get a consistent picture of how the refs look
924 	 * at a specified point in time
925 	 */
926 again:
927 	head = NULL;
928 
929 	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
930 	if (ret < 0)
931 		goto out;
932 	BUG_ON(ret == 0);
933 
934 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
935 	if (trans && likely(trans->type != __TRANS_DUMMY)) {
936 #else
937 	if (trans) {
938 #endif
939 		/*
940 		 * look if there are updates for this ref queued and lock the
941 		 * head
942 		 */
943 		delayed_refs = &trans->transaction->delayed_refs;
944 		spin_lock(&delayed_refs->lock);
945 		head = btrfs_find_delayed_ref_head(trans, bytenr);
946 		if (head) {
947 			if (!mutex_trylock(&head->mutex)) {
948 				atomic_inc(&head->node.refs);
949 				spin_unlock(&delayed_refs->lock);
950 
951 				btrfs_release_path(path);
952 
953 				/*
954 				 * Mutex was contended, block until it's
955 				 * released and try again
956 				 */
957 				mutex_lock(&head->mutex);
958 				mutex_unlock(&head->mutex);
959 				btrfs_put_delayed_ref(&head->node);
960 				goto again;
961 			}
962 			spin_unlock(&delayed_refs->lock);
963 			ret = __add_delayed_refs(head, time_seq,
964 						 &prefs_delayed, &total_refs,
965 						 inum);
966 			mutex_unlock(&head->mutex);
967 			if (ret)
968 				goto out;
969 		} else {
970 			spin_unlock(&delayed_refs->lock);
971 		}
972 	}
973 
974 	if (path->slots[0]) {
975 		struct extent_buffer *leaf;
976 		int slot;
977 
978 		path->slots[0]--;
979 		leaf = path->nodes[0];
980 		slot = path->slots[0];
981 		btrfs_item_key_to_cpu(leaf, &key, slot);
982 		if (key.objectid == bytenr &&
983 		    (key.type == BTRFS_EXTENT_ITEM_KEY ||
984 		     key.type == BTRFS_METADATA_ITEM_KEY)) {
985 			ret = __add_inline_refs(fs_info, path, bytenr,
986 						&info_level, &prefs,
987 						&total_refs, inum);
988 			if (ret)
989 				goto out;
990 			ret = __add_keyed_refs(fs_info, path, bytenr,
991 					       info_level, &prefs, inum);
992 			if (ret)
993 				goto out;
994 		}
995 	}
996 	btrfs_release_path(path);
997 
998 	list_splice_init(&prefs_delayed, &prefs);
999 
1000 	ret = __add_missing_keys(fs_info, &prefs);
1001 	if (ret)
1002 		goto out;
1003 
1004 	__merge_refs(&prefs, 1);
1005 
1006 	ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
1007 				      extent_item_pos, total_refs,
1008 				      root_objectid);
1009 	if (ret)
1010 		goto out;
1011 
1012 	__merge_refs(&prefs, 2);
1013 
1014 	while (!list_empty(&prefs)) {
1015 		ref = list_first_entry(&prefs, struct __prelim_ref, list);
1016 		WARN_ON(ref->count < 0);
1017 		if (roots && ref->count && ref->root_id && ref->parent == 0) {
1018 			if (root_objectid && ref->root_id != root_objectid) {
1019 				ret = BACKREF_FOUND_SHARED;
1020 				goto out;
1021 			}
1022 
1023 			/* no parent == root of tree */
1024 			ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1025 			if (ret < 0)
1026 				goto out;
1027 		}
1028 		if (ref->count && ref->parent) {
1029 			if (extent_item_pos && !ref->inode_list &&
1030 			    ref->level == 0) {
1031 				struct extent_buffer *eb;
1032 
1033 				eb = read_tree_block(fs_info->extent_root,
1034 							   ref->parent, 0);
1035 				if (!eb || !extent_buffer_uptodate(eb)) {
1036 					free_extent_buffer(eb);
1037 					ret = -EIO;
1038 					goto out;
1039 				}
1040 				btrfs_tree_read_lock(eb);
1041 				btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1042 				ret = find_extent_in_eb(eb, bytenr,
1043 							*extent_item_pos, &eie);
1044 				btrfs_tree_read_unlock_blocking(eb);
1045 				free_extent_buffer(eb);
1046 				if (ret < 0)
1047 					goto out;
1048 				ref->inode_list = eie;
1049 			}
1050 			ret = ulist_add_merge_ptr(refs, ref->parent,
1051 						  ref->inode_list,
1052 						  (void **)&eie, GFP_NOFS);
1053 			if (ret < 0)
1054 				goto out;
1055 			if (!ret && extent_item_pos) {
1056 				/*
1057 				 * we've recorded that parent, so we must extend
1058 				 * its inode list here
1059 				 */
1060 				BUG_ON(!eie);
1061 				while (eie->next)
1062 					eie = eie->next;
1063 				eie->next = ref->inode_list;
1064 			}
1065 			eie = NULL;
1066 		}
1067 		list_del(&ref->list);
1068 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1069 	}
1070 
1071 out:
1072 	btrfs_free_path(path);
1073 	while (!list_empty(&prefs)) {
1074 		ref = list_first_entry(&prefs, struct __prelim_ref, list);
1075 		list_del(&ref->list);
1076 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1077 	}
1078 	while (!list_empty(&prefs_delayed)) {
1079 		ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1080 				       list);
1081 		list_del(&ref->list);
1082 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1083 	}
1084 	if (ret < 0)
1085 		free_inode_elem_list(eie);
1086 	return ret;
1087 }
1088 
1089 static void free_leaf_list(struct ulist *blocks)
1090 {
1091 	struct ulist_node *node = NULL;
1092 	struct extent_inode_elem *eie;
1093 	struct ulist_iterator uiter;
1094 
1095 	ULIST_ITER_INIT(&uiter);
1096 	while ((node = ulist_next(blocks, &uiter))) {
1097 		if (!node->aux)
1098 			continue;
1099 		eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1100 		free_inode_elem_list(eie);
1101 		node->aux = 0;
1102 	}
1103 
1104 	ulist_free(blocks);
1105 }
1106 
1107 /*
1108  * Finds all leafs with a reference to the specified combination of bytenr and
1109  * offset. key_list_head will point to a list of corresponding keys (caller must
1110  * free each list element). The leafs will be stored in the leafs ulist, which
1111  * must be freed with ulist_free.
1112  *
1113  * returns 0 on success, <0 on error
1114  */
1115 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1116 				struct btrfs_fs_info *fs_info, u64 bytenr,
1117 				u64 time_seq, struct ulist **leafs,
1118 				const u64 *extent_item_pos)
1119 {
1120 	int ret;
1121 
1122 	*leafs = ulist_alloc(GFP_NOFS);
1123 	if (!*leafs)
1124 		return -ENOMEM;
1125 
1126 	ret = find_parent_nodes(trans, fs_info, bytenr,
1127 				time_seq, *leafs, NULL, extent_item_pos, 0, 0);
1128 	if (ret < 0 && ret != -ENOENT) {
1129 		free_leaf_list(*leafs);
1130 		return ret;
1131 	}
1132 
1133 	return 0;
1134 }
1135 
1136 /*
1137  * walk all backrefs for a given extent to find all roots that reference this
1138  * extent. Walking a backref means finding all extents that reference this
1139  * extent and in turn walk the backrefs of those, too. Naturally this is a
1140  * recursive process, but here it is implemented in an iterative fashion: We
1141  * find all referencing extents for the extent in question and put them on a
1142  * list. In turn, we find all referencing extents for those, further appending
1143  * to the list. The way we iterate the list allows adding more elements after
1144  * the current while iterating. The process stops when we reach the end of the
1145  * list. Found roots are added to the roots list.
1146  *
1147  * returns 0 on success, < 0 on error.
1148  */
1149 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1150 				  struct btrfs_fs_info *fs_info, u64 bytenr,
1151 				  u64 time_seq, struct ulist **roots)
1152 {
1153 	struct ulist *tmp;
1154 	struct ulist_node *node = NULL;
1155 	struct ulist_iterator uiter;
1156 	int ret;
1157 
1158 	tmp = ulist_alloc(GFP_NOFS);
1159 	if (!tmp)
1160 		return -ENOMEM;
1161 	*roots = ulist_alloc(GFP_NOFS);
1162 	if (!*roots) {
1163 		ulist_free(tmp);
1164 		return -ENOMEM;
1165 	}
1166 
1167 	ULIST_ITER_INIT(&uiter);
1168 	while (1) {
1169 		ret = find_parent_nodes(trans, fs_info, bytenr,
1170 					time_seq, tmp, *roots, NULL, 0, 0);
1171 		if (ret < 0 && ret != -ENOENT) {
1172 			ulist_free(tmp);
1173 			ulist_free(*roots);
1174 			return ret;
1175 		}
1176 		node = ulist_next(tmp, &uiter);
1177 		if (!node)
1178 			break;
1179 		bytenr = node->val;
1180 		cond_resched();
1181 	}
1182 
1183 	ulist_free(tmp);
1184 	return 0;
1185 }
1186 
1187 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1188 			 struct btrfs_fs_info *fs_info, u64 bytenr,
1189 			 u64 time_seq, struct ulist **roots)
1190 {
1191 	int ret;
1192 
1193 	if (!trans)
1194 		down_read(&fs_info->commit_root_sem);
1195 	ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1196 	if (!trans)
1197 		up_read(&fs_info->commit_root_sem);
1198 	return ret;
1199 }
1200 
1201 int btrfs_check_shared(struct btrfs_trans_handle *trans,
1202 		       struct btrfs_fs_info *fs_info, u64 root_objectid,
1203 		       u64 inum, u64 bytenr)
1204 {
1205 	struct ulist *tmp = NULL;
1206 	struct ulist *roots = NULL;
1207 	struct ulist_iterator uiter;
1208 	struct ulist_node *node;
1209 	struct seq_list elem = {};
1210 	int ret = 0;
1211 
1212 	tmp = ulist_alloc(GFP_NOFS);
1213 	roots = ulist_alloc(GFP_NOFS);
1214 	if (!tmp || !roots) {
1215 		ulist_free(tmp);
1216 		ulist_free(roots);
1217 		return -ENOMEM;
1218 	}
1219 
1220 	if (trans)
1221 		btrfs_get_tree_mod_seq(fs_info, &elem);
1222 	else
1223 		down_read(&fs_info->commit_root_sem);
1224 	ULIST_ITER_INIT(&uiter);
1225 	while (1) {
1226 		ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1227 					roots, NULL, root_objectid, inum);
1228 		if (ret == BACKREF_FOUND_SHARED) {
1229 			ret = 1;
1230 			break;
1231 		}
1232 		if (ret < 0 && ret != -ENOENT)
1233 			break;
1234 		node = ulist_next(tmp, &uiter);
1235 		if (!node)
1236 			break;
1237 		bytenr = node->val;
1238 		cond_resched();
1239 	}
1240 	if (trans)
1241 		btrfs_put_tree_mod_seq(fs_info, &elem);
1242 	else
1243 		up_read(&fs_info->commit_root_sem);
1244 	ulist_free(tmp);
1245 	ulist_free(roots);
1246 	return ret;
1247 }
1248 
1249 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1250 			  u64 start_off, struct btrfs_path *path,
1251 			  struct btrfs_inode_extref **ret_extref,
1252 			  u64 *found_off)
1253 {
1254 	int ret, slot;
1255 	struct btrfs_key key;
1256 	struct btrfs_key found_key;
1257 	struct btrfs_inode_extref *extref;
1258 	struct extent_buffer *leaf;
1259 	unsigned long ptr;
1260 
1261 	key.objectid = inode_objectid;
1262 	key.type = BTRFS_INODE_EXTREF_KEY;
1263 	key.offset = start_off;
1264 
1265 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1266 	if (ret < 0)
1267 		return ret;
1268 
1269 	while (1) {
1270 		leaf = path->nodes[0];
1271 		slot = path->slots[0];
1272 		if (slot >= btrfs_header_nritems(leaf)) {
1273 			/*
1274 			 * If the item at offset is not found,
1275 			 * btrfs_search_slot will point us to the slot
1276 			 * where it should be inserted. In our case
1277 			 * that will be the slot directly before the
1278 			 * next INODE_REF_KEY_V2 item. In the case
1279 			 * that we're pointing to the last slot in a
1280 			 * leaf, we must move one leaf over.
1281 			 */
1282 			ret = btrfs_next_leaf(root, path);
1283 			if (ret) {
1284 				if (ret >= 1)
1285 					ret = -ENOENT;
1286 				break;
1287 			}
1288 			continue;
1289 		}
1290 
1291 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1292 
1293 		/*
1294 		 * Check that we're still looking at an extended ref key for
1295 		 * this particular objectid. If we have different
1296 		 * objectid or type then there are no more to be found
1297 		 * in the tree and we can exit.
1298 		 */
1299 		ret = -ENOENT;
1300 		if (found_key.objectid != inode_objectid)
1301 			break;
1302 		if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1303 			break;
1304 
1305 		ret = 0;
1306 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1307 		extref = (struct btrfs_inode_extref *)ptr;
1308 		*ret_extref = extref;
1309 		if (found_off)
1310 			*found_off = found_key.offset;
1311 		break;
1312 	}
1313 
1314 	return ret;
1315 }
1316 
1317 /*
1318  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1319  * Elements of the path are separated by '/' and the path is guaranteed to be
1320  * 0-terminated. the path is only given within the current file system.
1321  * Therefore, it never starts with a '/'. the caller is responsible to provide
1322  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1323  * the start point of the resulting string is returned. this pointer is within
1324  * dest, normally.
1325  * in case the path buffer would overflow, the pointer is decremented further
1326  * as if output was written to the buffer, though no more output is actually
1327  * generated. that way, the caller can determine how much space would be
1328  * required for the path to fit into the buffer. in that case, the returned
1329  * value will be smaller than dest. callers must check this!
1330  */
1331 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1332 			u32 name_len, unsigned long name_off,
1333 			struct extent_buffer *eb_in, u64 parent,
1334 			char *dest, u32 size)
1335 {
1336 	int slot;
1337 	u64 next_inum;
1338 	int ret;
1339 	s64 bytes_left = ((s64)size) - 1;
1340 	struct extent_buffer *eb = eb_in;
1341 	struct btrfs_key found_key;
1342 	int leave_spinning = path->leave_spinning;
1343 	struct btrfs_inode_ref *iref;
1344 
1345 	if (bytes_left >= 0)
1346 		dest[bytes_left] = '\0';
1347 
1348 	path->leave_spinning = 1;
1349 	while (1) {
1350 		bytes_left -= name_len;
1351 		if (bytes_left >= 0)
1352 			read_extent_buffer(eb, dest + bytes_left,
1353 					   name_off, name_len);
1354 		if (eb != eb_in) {
1355 			btrfs_tree_read_unlock_blocking(eb);
1356 			free_extent_buffer(eb);
1357 		}
1358 		ret = btrfs_find_item(fs_root, path, parent, 0,
1359 				BTRFS_INODE_REF_KEY, &found_key);
1360 		if (ret > 0)
1361 			ret = -ENOENT;
1362 		if (ret)
1363 			break;
1364 
1365 		next_inum = found_key.offset;
1366 
1367 		/* regular exit ahead */
1368 		if (parent == next_inum)
1369 			break;
1370 
1371 		slot = path->slots[0];
1372 		eb = path->nodes[0];
1373 		/* make sure we can use eb after releasing the path */
1374 		if (eb != eb_in) {
1375 			atomic_inc(&eb->refs);
1376 			btrfs_tree_read_lock(eb);
1377 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1378 		}
1379 		btrfs_release_path(path);
1380 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1381 
1382 		name_len = btrfs_inode_ref_name_len(eb, iref);
1383 		name_off = (unsigned long)(iref + 1);
1384 
1385 		parent = next_inum;
1386 		--bytes_left;
1387 		if (bytes_left >= 0)
1388 			dest[bytes_left] = '/';
1389 	}
1390 
1391 	btrfs_release_path(path);
1392 	path->leave_spinning = leave_spinning;
1393 
1394 	if (ret)
1395 		return ERR_PTR(ret);
1396 
1397 	return dest + bytes_left;
1398 }
1399 
1400 /*
1401  * this makes the path point to (logical EXTENT_ITEM *)
1402  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1403  * tree blocks and <0 on error.
1404  */
1405 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1406 			struct btrfs_path *path, struct btrfs_key *found_key,
1407 			u64 *flags_ret)
1408 {
1409 	int ret;
1410 	u64 flags;
1411 	u64 size = 0;
1412 	u32 item_size;
1413 	struct extent_buffer *eb;
1414 	struct btrfs_extent_item *ei;
1415 	struct btrfs_key key;
1416 
1417 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1418 		key.type = BTRFS_METADATA_ITEM_KEY;
1419 	else
1420 		key.type = BTRFS_EXTENT_ITEM_KEY;
1421 	key.objectid = logical;
1422 	key.offset = (u64)-1;
1423 
1424 	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1425 	if (ret < 0)
1426 		return ret;
1427 
1428 	ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1429 	if (ret) {
1430 		if (ret > 0)
1431 			ret = -ENOENT;
1432 		return ret;
1433 	}
1434 	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1435 	if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1436 		size = fs_info->extent_root->nodesize;
1437 	else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1438 		size = found_key->offset;
1439 
1440 	if (found_key->objectid > logical ||
1441 	    found_key->objectid + size <= logical) {
1442 		pr_debug("logical %llu is not within any extent\n", logical);
1443 		return -ENOENT;
1444 	}
1445 
1446 	eb = path->nodes[0];
1447 	item_size = btrfs_item_size_nr(eb, path->slots[0]);
1448 	BUG_ON(item_size < sizeof(*ei));
1449 
1450 	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1451 	flags = btrfs_extent_flags(eb, ei);
1452 
1453 	pr_debug("logical %llu is at position %llu within the extent (%llu "
1454 		 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1455 		 logical, logical - found_key->objectid, found_key->objectid,
1456 		 found_key->offset, flags, item_size);
1457 
1458 	WARN_ON(!flags_ret);
1459 	if (flags_ret) {
1460 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1461 			*flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1462 		else if (flags & BTRFS_EXTENT_FLAG_DATA)
1463 			*flags_ret = BTRFS_EXTENT_FLAG_DATA;
1464 		else
1465 			BUG_ON(1);
1466 		return 0;
1467 	}
1468 
1469 	return -EIO;
1470 }
1471 
1472 /*
1473  * helper function to iterate extent inline refs. ptr must point to a 0 value
1474  * for the first call and may be modified. it is used to track state.
1475  * if more refs exist, 0 is returned and the next call to
1476  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1477  * next ref. after the last ref was processed, 1 is returned.
1478  * returns <0 on error
1479  */
1480 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1481 				   struct btrfs_key *key,
1482 				   struct btrfs_extent_item *ei, u32 item_size,
1483 				   struct btrfs_extent_inline_ref **out_eiref,
1484 				   int *out_type)
1485 {
1486 	unsigned long end;
1487 	u64 flags;
1488 	struct btrfs_tree_block_info *info;
1489 
1490 	if (!*ptr) {
1491 		/* first call */
1492 		flags = btrfs_extent_flags(eb, ei);
1493 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1494 			if (key->type == BTRFS_METADATA_ITEM_KEY) {
1495 				/* a skinny metadata extent */
1496 				*out_eiref =
1497 				     (struct btrfs_extent_inline_ref *)(ei + 1);
1498 			} else {
1499 				WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1500 				info = (struct btrfs_tree_block_info *)(ei + 1);
1501 				*out_eiref =
1502 				   (struct btrfs_extent_inline_ref *)(info + 1);
1503 			}
1504 		} else {
1505 			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1506 		}
1507 		*ptr = (unsigned long)*out_eiref;
1508 		if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1509 			return -ENOENT;
1510 	}
1511 
1512 	end = (unsigned long)ei + item_size;
1513 	*out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1514 	*out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1515 
1516 	*ptr += btrfs_extent_inline_ref_size(*out_type);
1517 	WARN_ON(*ptr > end);
1518 	if (*ptr == end)
1519 		return 1; /* last */
1520 
1521 	return 0;
1522 }
1523 
1524 /*
1525  * reads the tree block backref for an extent. tree level and root are returned
1526  * through out_level and out_root. ptr must point to a 0 value for the first
1527  * call and may be modified (see __get_extent_inline_ref comment).
1528  * returns 0 if data was provided, 1 if there was no more data to provide or
1529  * <0 on error.
1530  */
1531 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1532 			    struct btrfs_key *key, struct btrfs_extent_item *ei,
1533 			    u32 item_size, u64 *out_root, u8 *out_level)
1534 {
1535 	int ret;
1536 	int type;
1537 	struct btrfs_extent_inline_ref *eiref;
1538 
1539 	if (*ptr == (unsigned long)-1)
1540 		return 1;
1541 
1542 	while (1) {
1543 		ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1544 					      &eiref, &type);
1545 		if (ret < 0)
1546 			return ret;
1547 
1548 		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1549 		    type == BTRFS_SHARED_BLOCK_REF_KEY)
1550 			break;
1551 
1552 		if (ret == 1)
1553 			return 1;
1554 	}
1555 
1556 	/* we can treat both ref types equally here */
1557 	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1558 
1559 	if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1560 		struct btrfs_tree_block_info *info;
1561 
1562 		info = (struct btrfs_tree_block_info *)(ei + 1);
1563 		*out_level = btrfs_tree_block_level(eb, info);
1564 	} else {
1565 		ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1566 		*out_level = (u8)key->offset;
1567 	}
1568 
1569 	if (ret == 1)
1570 		*ptr = (unsigned long)-1;
1571 
1572 	return 0;
1573 }
1574 
1575 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1576 				u64 root, u64 extent_item_objectid,
1577 				iterate_extent_inodes_t *iterate, void *ctx)
1578 {
1579 	struct extent_inode_elem *eie;
1580 	int ret = 0;
1581 
1582 	for (eie = inode_list; eie; eie = eie->next) {
1583 		pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1584 			 "root %llu\n", extent_item_objectid,
1585 			 eie->inum, eie->offset, root);
1586 		ret = iterate(eie->inum, eie->offset, root, ctx);
1587 		if (ret) {
1588 			pr_debug("stopping iteration for %llu due to ret=%d\n",
1589 				 extent_item_objectid, ret);
1590 			break;
1591 		}
1592 	}
1593 
1594 	return ret;
1595 }
1596 
1597 /*
1598  * calls iterate() for every inode that references the extent identified by
1599  * the given parameters.
1600  * when the iterator function returns a non-zero value, iteration stops.
1601  */
1602 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1603 				u64 extent_item_objectid, u64 extent_item_pos,
1604 				int search_commit_root,
1605 				iterate_extent_inodes_t *iterate, void *ctx)
1606 {
1607 	int ret;
1608 	struct btrfs_trans_handle *trans = NULL;
1609 	struct ulist *refs = NULL;
1610 	struct ulist *roots = NULL;
1611 	struct ulist_node *ref_node = NULL;
1612 	struct ulist_node *root_node = NULL;
1613 	struct seq_list tree_mod_seq_elem = {};
1614 	struct ulist_iterator ref_uiter;
1615 	struct ulist_iterator root_uiter;
1616 
1617 	pr_debug("resolving all inodes for extent %llu\n",
1618 			extent_item_objectid);
1619 
1620 	if (!search_commit_root) {
1621 		trans = btrfs_join_transaction(fs_info->extent_root);
1622 		if (IS_ERR(trans))
1623 			return PTR_ERR(trans);
1624 		btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1625 	} else {
1626 		down_read(&fs_info->commit_root_sem);
1627 	}
1628 
1629 	ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1630 				   tree_mod_seq_elem.seq, &refs,
1631 				   &extent_item_pos);
1632 	if (ret)
1633 		goto out;
1634 
1635 	ULIST_ITER_INIT(&ref_uiter);
1636 	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1637 		ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1638 					     tree_mod_seq_elem.seq, &roots);
1639 		if (ret)
1640 			break;
1641 		ULIST_ITER_INIT(&root_uiter);
1642 		while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1643 			pr_debug("root %llu references leaf %llu, data list "
1644 				 "%#llx\n", root_node->val, ref_node->val,
1645 				 ref_node->aux);
1646 			ret = iterate_leaf_refs((struct extent_inode_elem *)
1647 						(uintptr_t)ref_node->aux,
1648 						root_node->val,
1649 						extent_item_objectid,
1650 						iterate, ctx);
1651 		}
1652 		ulist_free(roots);
1653 	}
1654 
1655 	free_leaf_list(refs);
1656 out:
1657 	if (!search_commit_root) {
1658 		btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1659 		btrfs_end_transaction(trans, fs_info->extent_root);
1660 	} else {
1661 		up_read(&fs_info->commit_root_sem);
1662 	}
1663 
1664 	return ret;
1665 }
1666 
1667 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1668 				struct btrfs_path *path,
1669 				iterate_extent_inodes_t *iterate, void *ctx)
1670 {
1671 	int ret;
1672 	u64 extent_item_pos;
1673 	u64 flags = 0;
1674 	struct btrfs_key found_key;
1675 	int search_commit_root = path->search_commit_root;
1676 
1677 	ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1678 	btrfs_release_path(path);
1679 	if (ret < 0)
1680 		return ret;
1681 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1682 		return -EINVAL;
1683 
1684 	extent_item_pos = logical - found_key.objectid;
1685 	ret = iterate_extent_inodes(fs_info, found_key.objectid,
1686 					extent_item_pos, search_commit_root,
1687 					iterate, ctx);
1688 
1689 	return ret;
1690 }
1691 
1692 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1693 			      struct extent_buffer *eb, void *ctx);
1694 
1695 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1696 			      struct btrfs_path *path,
1697 			      iterate_irefs_t *iterate, void *ctx)
1698 {
1699 	int ret = 0;
1700 	int slot;
1701 	u32 cur;
1702 	u32 len;
1703 	u32 name_len;
1704 	u64 parent = 0;
1705 	int found = 0;
1706 	struct extent_buffer *eb;
1707 	struct btrfs_item *item;
1708 	struct btrfs_inode_ref *iref;
1709 	struct btrfs_key found_key;
1710 
1711 	while (!ret) {
1712 		ret = btrfs_find_item(fs_root, path, inum,
1713 				parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
1714 				&found_key);
1715 
1716 		if (ret < 0)
1717 			break;
1718 		if (ret) {
1719 			ret = found ? 0 : -ENOENT;
1720 			break;
1721 		}
1722 		++found;
1723 
1724 		parent = found_key.offset;
1725 		slot = path->slots[0];
1726 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
1727 		if (!eb) {
1728 			ret = -ENOMEM;
1729 			break;
1730 		}
1731 		extent_buffer_get(eb);
1732 		btrfs_tree_read_lock(eb);
1733 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1734 		btrfs_release_path(path);
1735 
1736 		item = btrfs_item_nr(slot);
1737 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1738 
1739 		for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1740 			name_len = btrfs_inode_ref_name_len(eb, iref);
1741 			/* path must be released before calling iterate()! */
1742 			pr_debug("following ref at offset %u for inode %llu in "
1743 				 "tree %llu\n", cur, found_key.objectid,
1744 				 fs_root->objectid);
1745 			ret = iterate(parent, name_len,
1746 				      (unsigned long)(iref + 1), eb, ctx);
1747 			if (ret)
1748 				break;
1749 			len = sizeof(*iref) + name_len;
1750 			iref = (struct btrfs_inode_ref *)((char *)iref + len);
1751 		}
1752 		btrfs_tree_read_unlock_blocking(eb);
1753 		free_extent_buffer(eb);
1754 	}
1755 
1756 	btrfs_release_path(path);
1757 
1758 	return ret;
1759 }
1760 
1761 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1762 				 struct btrfs_path *path,
1763 				 iterate_irefs_t *iterate, void *ctx)
1764 {
1765 	int ret;
1766 	int slot;
1767 	u64 offset = 0;
1768 	u64 parent;
1769 	int found = 0;
1770 	struct extent_buffer *eb;
1771 	struct btrfs_inode_extref *extref;
1772 	struct extent_buffer *leaf;
1773 	u32 item_size;
1774 	u32 cur_offset;
1775 	unsigned long ptr;
1776 
1777 	while (1) {
1778 		ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1779 					    &offset);
1780 		if (ret < 0)
1781 			break;
1782 		if (ret) {
1783 			ret = found ? 0 : -ENOENT;
1784 			break;
1785 		}
1786 		++found;
1787 
1788 		slot = path->slots[0];
1789 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
1790 		if (!eb) {
1791 			ret = -ENOMEM;
1792 			break;
1793 		}
1794 		extent_buffer_get(eb);
1795 
1796 		btrfs_tree_read_lock(eb);
1797 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1798 		btrfs_release_path(path);
1799 
1800 		leaf = path->nodes[0];
1801 		item_size = btrfs_item_size_nr(leaf, slot);
1802 		ptr = btrfs_item_ptr_offset(leaf, slot);
1803 		cur_offset = 0;
1804 
1805 		while (cur_offset < item_size) {
1806 			u32 name_len;
1807 
1808 			extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1809 			parent = btrfs_inode_extref_parent(eb, extref);
1810 			name_len = btrfs_inode_extref_name_len(eb, extref);
1811 			ret = iterate(parent, name_len,
1812 				      (unsigned long)&extref->name, eb, ctx);
1813 			if (ret)
1814 				break;
1815 
1816 			cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1817 			cur_offset += sizeof(*extref);
1818 		}
1819 		btrfs_tree_read_unlock_blocking(eb);
1820 		free_extent_buffer(eb);
1821 
1822 		offset++;
1823 	}
1824 
1825 	btrfs_release_path(path);
1826 
1827 	return ret;
1828 }
1829 
1830 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1831 			 struct btrfs_path *path, iterate_irefs_t *iterate,
1832 			 void *ctx)
1833 {
1834 	int ret;
1835 	int found_refs = 0;
1836 
1837 	ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1838 	if (!ret)
1839 		++found_refs;
1840 	else if (ret != -ENOENT)
1841 		return ret;
1842 
1843 	ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1844 	if (ret == -ENOENT && found_refs)
1845 		return 0;
1846 
1847 	return ret;
1848 }
1849 
1850 /*
1851  * returns 0 if the path could be dumped (probably truncated)
1852  * returns <0 in case of an error
1853  */
1854 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1855 			 struct extent_buffer *eb, void *ctx)
1856 {
1857 	struct inode_fs_paths *ipath = ctx;
1858 	char *fspath;
1859 	char *fspath_min;
1860 	int i = ipath->fspath->elem_cnt;
1861 	const int s_ptr = sizeof(char *);
1862 	u32 bytes_left;
1863 
1864 	bytes_left = ipath->fspath->bytes_left > s_ptr ?
1865 					ipath->fspath->bytes_left - s_ptr : 0;
1866 
1867 	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1868 	fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1869 				   name_off, eb, inum, fspath_min, bytes_left);
1870 	if (IS_ERR(fspath))
1871 		return PTR_ERR(fspath);
1872 
1873 	if (fspath > fspath_min) {
1874 		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1875 		++ipath->fspath->elem_cnt;
1876 		ipath->fspath->bytes_left = fspath - fspath_min;
1877 	} else {
1878 		++ipath->fspath->elem_missed;
1879 		ipath->fspath->bytes_missing += fspath_min - fspath;
1880 		ipath->fspath->bytes_left = 0;
1881 	}
1882 
1883 	return 0;
1884 }
1885 
1886 /*
1887  * this dumps all file system paths to the inode into the ipath struct, provided
1888  * is has been created large enough. each path is zero-terminated and accessed
1889  * from ipath->fspath->val[i].
1890  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1891  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1892  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1893  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1894  * have been needed to return all paths.
1895  */
1896 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1897 {
1898 	return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1899 			     inode_to_path, ipath);
1900 }
1901 
1902 struct btrfs_data_container *init_data_container(u32 total_bytes)
1903 {
1904 	struct btrfs_data_container *data;
1905 	size_t alloc_bytes;
1906 
1907 	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1908 	data = vmalloc(alloc_bytes);
1909 	if (!data)
1910 		return ERR_PTR(-ENOMEM);
1911 
1912 	if (total_bytes >= sizeof(*data)) {
1913 		data->bytes_left = total_bytes - sizeof(*data);
1914 		data->bytes_missing = 0;
1915 	} else {
1916 		data->bytes_missing = sizeof(*data) - total_bytes;
1917 		data->bytes_left = 0;
1918 	}
1919 
1920 	data->elem_cnt = 0;
1921 	data->elem_missed = 0;
1922 
1923 	return data;
1924 }
1925 
1926 /*
1927  * allocates space to return multiple file system paths for an inode.
1928  * total_bytes to allocate are passed, note that space usable for actual path
1929  * information will be total_bytes - sizeof(struct inode_fs_paths).
1930  * the returned pointer must be freed with free_ipath() in the end.
1931  */
1932 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1933 					struct btrfs_path *path)
1934 {
1935 	struct inode_fs_paths *ifp;
1936 	struct btrfs_data_container *fspath;
1937 
1938 	fspath = init_data_container(total_bytes);
1939 	if (IS_ERR(fspath))
1940 		return (void *)fspath;
1941 
1942 	ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1943 	if (!ifp) {
1944 		kfree(fspath);
1945 		return ERR_PTR(-ENOMEM);
1946 	}
1947 
1948 	ifp->btrfs_path = path;
1949 	ifp->fspath = fspath;
1950 	ifp->fs_root = fs_root;
1951 
1952 	return ifp;
1953 }
1954 
1955 void free_ipath(struct inode_fs_paths *ipath)
1956 {
1957 	if (!ipath)
1958 		return;
1959 	vfree(ipath->fspath);
1960 	kfree(ipath);
1961 }
1962