xref: /openbmc/linux/fs/btrfs/backref.c (revision ac73d4bf)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 STRATO.  All rights reserved.
4  */
5 
6 #include <linux/mm.h>
7 #include <linux/rbtree.h>
8 #include <trace/events/btrfs.h>
9 #include "ctree.h"
10 #include "disk-io.h"
11 #include "backref.h"
12 #include "ulist.h"
13 #include "transaction.h"
14 #include "delayed-ref.h"
15 #include "locking.h"
16 #include "misc.h"
17 #include "tree-mod-log.h"
18 
19 /* Just an arbitrary number so we can be sure this happened */
20 #define BACKREF_FOUND_SHARED 6
21 
22 struct extent_inode_elem {
23 	u64 inum;
24 	u64 offset;
25 	struct extent_inode_elem *next;
26 };
27 
28 static int check_extent_in_eb(const struct btrfs_key *key,
29 			      const struct extent_buffer *eb,
30 			      const struct btrfs_file_extent_item *fi,
31 			      u64 extent_item_pos,
32 			      struct extent_inode_elem **eie,
33 			      bool ignore_offset)
34 {
35 	u64 offset = 0;
36 	struct extent_inode_elem *e;
37 
38 	if (!ignore_offset &&
39 	    !btrfs_file_extent_compression(eb, fi) &&
40 	    !btrfs_file_extent_encryption(eb, fi) &&
41 	    !btrfs_file_extent_other_encoding(eb, fi)) {
42 		u64 data_offset;
43 		u64 data_len;
44 
45 		data_offset = btrfs_file_extent_offset(eb, fi);
46 		data_len = btrfs_file_extent_num_bytes(eb, fi);
47 
48 		if (extent_item_pos < data_offset ||
49 		    extent_item_pos >= data_offset + data_len)
50 			return 1;
51 		offset = extent_item_pos - data_offset;
52 	}
53 
54 	e = kmalloc(sizeof(*e), GFP_NOFS);
55 	if (!e)
56 		return -ENOMEM;
57 
58 	e->next = *eie;
59 	e->inum = key->objectid;
60 	e->offset = key->offset + offset;
61 	*eie = e;
62 
63 	return 0;
64 }
65 
66 static void free_inode_elem_list(struct extent_inode_elem *eie)
67 {
68 	struct extent_inode_elem *eie_next;
69 
70 	for (; eie; eie = eie_next) {
71 		eie_next = eie->next;
72 		kfree(eie);
73 	}
74 }
75 
76 static int find_extent_in_eb(const struct extent_buffer *eb,
77 			     u64 wanted_disk_byte, u64 extent_item_pos,
78 			     struct extent_inode_elem **eie,
79 			     bool ignore_offset)
80 {
81 	u64 disk_byte;
82 	struct btrfs_key key;
83 	struct btrfs_file_extent_item *fi;
84 	int slot;
85 	int nritems;
86 	int extent_type;
87 	int ret;
88 
89 	/*
90 	 * from the shared data ref, we only have the leaf but we need
91 	 * the key. thus, we must look into all items and see that we
92 	 * find one (some) with a reference to our extent item.
93 	 */
94 	nritems = btrfs_header_nritems(eb);
95 	for (slot = 0; slot < nritems; ++slot) {
96 		btrfs_item_key_to_cpu(eb, &key, slot);
97 		if (key.type != BTRFS_EXTENT_DATA_KEY)
98 			continue;
99 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
100 		extent_type = btrfs_file_extent_type(eb, fi);
101 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
102 			continue;
103 		/* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
104 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
105 		if (disk_byte != wanted_disk_byte)
106 			continue;
107 
108 		ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset);
109 		if (ret < 0)
110 			return ret;
111 	}
112 
113 	return 0;
114 }
115 
116 struct preftree {
117 	struct rb_root_cached root;
118 	unsigned int count;
119 };
120 
121 #define PREFTREE_INIT	{ .root = RB_ROOT_CACHED, .count = 0 }
122 
123 struct preftrees {
124 	struct preftree direct;    /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
125 	struct preftree indirect;  /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
126 	struct preftree indirect_missing_keys;
127 };
128 
129 /*
130  * Checks for a shared extent during backref search.
131  *
132  * The share_count tracks prelim_refs (direct and indirect) having a
133  * ref->count >0:
134  *  - incremented when a ref->count transitions to >0
135  *  - decremented when a ref->count transitions to <1
136  */
137 struct share_check {
138 	u64 root_objectid;
139 	u64 inum;
140 	int share_count;
141 	bool have_delayed_delete_refs;
142 };
143 
144 static inline int extent_is_shared(struct share_check *sc)
145 {
146 	return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
147 }
148 
149 static struct kmem_cache *btrfs_prelim_ref_cache;
150 
151 int __init btrfs_prelim_ref_init(void)
152 {
153 	btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
154 					sizeof(struct prelim_ref),
155 					0,
156 					SLAB_MEM_SPREAD,
157 					NULL);
158 	if (!btrfs_prelim_ref_cache)
159 		return -ENOMEM;
160 	return 0;
161 }
162 
163 void __cold btrfs_prelim_ref_exit(void)
164 {
165 	kmem_cache_destroy(btrfs_prelim_ref_cache);
166 }
167 
168 static void free_pref(struct prelim_ref *ref)
169 {
170 	kmem_cache_free(btrfs_prelim_ref_cache, ref);
171 }
172 
173 /*
174  * Return 0 when both refs are for the same block (and can be merged).
175  * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
176  * indicates a 'higher' block.
177  */
178 static int prelim_ref_compare(struct prelim_ref *ref1,
179 			      struct prelim_ref *ref2)
180 {
181 	if (ref1->level < ref2->level)
182 		return -1;
183 	if (ref1->level > ref2->level)
184 		return 1;
185 	if (ref1->root_id < ref2->root_id)
186 		return -1;
187 	if (ref1->root_id > ref2->root_id)
188 		return 1;
189 	if (ref1->key_for_search.type < ref2->key_for_search.type)
190 		return -1;
191 	if (ref1->key_for_search.type > ref2->key_for_search.type)
192 		return 1;
193 	if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
194 		return -1;
195 	if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
196 		return 1;
197 	if (ref1->key_for_search.offset < ref2->key_for_search.offset)
198 		return -1;
199 	if (ref1->key_for_search.offset > ref2->key_for_search.offset)
200 		return 1;
201 	if (ref1->parent < ref2->parent)
202 		return -1;
203 	if (ref1->parent > ref2->parent)
204 		return 1;
205 
206 	return 0;
207 }
208 
209 static void update_share_count(struct share_check *sc, int oldcount,
210 			       int newcount)
211 {
212 	if ((!sc) || (oldcount == 0 && newcount < 1))
213 		return;
214 
215 	if (oldcount > 0 && newcount < 1)
216 		sc->share_count--;
217 	else if (oldcount < 1 && newcount > 0)
218 		sc->share_count++;
219 }
220 
221 /*
222  * Add @newref to the @root rbtree, merging identical refs.
223  *
224  * Callers should assume that newref has been freed after calling.
225  */
226 static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
227 			      struct preftree *preftree,
228 			      struct prelim_ref *newref,
229 			      struct share_check *sc)
230 {
231 	struct rb_root_cached *root;
232 	struct rb_node **p;
233 	struct rb_node *parent = NULL;
234 	struct prelim_ref *ref;
235 	int result;
236 	bool leftmost = true;
237 
238 	root = &preftree->root;
239 	p = &root->rb_root.rb_node;
240 
241 	while (*p) {
242 		parent = *p;
243 		ref = rb_entry(parent, struct prelim_ref, rbnode);
244 		result = prelim_ref_compare(ref, newref);
245 		if (result < 0) {
246 			p = &(*p)->rb_left;
247 		} else if (result > 0) {
248 			p = &(*p)->rb_right;
249 			leftmost = false;
250 		} else {
251 			/* Identical refs, merge them and free @newref */
252 			struct extent_inode_elem *eie = ref->inode_list;
253 
254 			while (eie && eie->next)
255 				eie = eie->next;
256 
257 			if (!eie)
258 				ref->inode_list = newref->inode_list;
259 			else
260 				eie->next = newref->inode_list;
261 			trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
262 						     preftree->count);
263 			/*
264 			 * A delayed ref can have newref->count < 0.
265 			 * The ref->count is updated to follow any
266 			 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
267 			 */
268 			update_share_count(sc, ref->count,
269 					   ref->count + newref->count);
270 			ref->count += newref->count;
271 			free_pref(newref);
272 			return;
273 		}
274 	}
275 
276 	update_share_count(sc, 0, newref->count);
277 	preftree->count++;
278 	trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
279 	rb_link_node(&newref->rbnode, parent, p);
280 	rb_insert_color_cached(&newref->rbnode, root, leftmost);
281 }
282 
283 /*
284  * Release the entire tree.  We don't care about internal consistency so
285  * just free everything and then reset the tree root.
286  */
287 static void prelim_release(struct preftree *preftree)
288 {
289 	struct prelim_ref *ref, *next_ref;
290 
291 	rbtree_postorder_for_each_entry_safe(ref, next_ref,
292 					     &preftree->root.rb_root, rbnode)
293 		free_pref(ref);
294 
295 	preftree->root = RB_ROOT_CACHED;
296 	preftree->count = 0;
297 }
298 
299 /*
300  * the rules for all callers of this function are:
301  * - obtaining the parent is the goal
302  * - if you add a key, you must know that it is a correct key
303  * - if you cannot add the parent or a correct key, then we will look into the
304  *   block later to set a correct key
305  *
306  * delayed refs
307  * ============
308  *        backref type | shared | indirect | shared | indirect
309  * information         |   tree |     tree |   data |     data
310  * --------------------+--------+----------+--------+----------
311  *      parent logical |    y   |     -    |    -   |     -
312  *      key to resolve |    -   |     y    |    y   |     y
313  *  tree block logical |    -   |     -    |    -   |     -
314  *  root for resolving |    y   |     y    |    y   |     y
315  *
316  * - column 1:       we've the parent -> done
317  * - column 2, 3, 4: we use the key to find the parent
318  *
319  * on disk refs (inline or keyed)
320  * ==============================
321  *        backref type | shared | indirect | shared | indirect
322  * information         |   tree |     tree |   data |     data
323  * --------------------+--------+----------+--------+----------
324  *      parent logical |    y   |     -    |    y   |     -
325  *      key to resolve |    -   |     -    |    -   |     y
326  *  tree block logical |    y   |     y    |    y   |     y
327  *  root for resolving |    -   |     y    |    y   |     y
328  *
329  * - column 1, 3: we've the parent -> done
330  * - column 2:    we take the first key from the block to find the parent
331  *                (see add_missing_keys)
332  * - column 4:    we use the key to find the parent
333  *
334  * additional information that's available but not required to find the parent
335  * block might help in merging entries to gain some speed.
336  */
337 static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
338 			  struct preftree *preftree, u64 root_id,
339 			  const struct btrfs_key *key, int level, u64 parent,
340 			  u64 wanted_disk_byte, int count,
341 			  struct share_check *sc, gfp_t gfp_mask)
342 {
343 	struct prelim_ref *ref;
344 
345 	if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
346 		return 0;
347 
348 	ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
349 	if (!ref)
350 		return -ENOMEM;
351 
352 	ref->root_id = root_id;
353 	if (key)
354 		ref->key_for_search = *key;
355 	else
356 		memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
357 
358 	ref->inode_list = NULL;
359 	ref->level = level;
360 	ref->count = count;
361 	ref->parent = parent;
362 	ref->wanted_disk_byte = wanted_disk_byte;
363 	prelim_ref_insert(fs_info, preftree, ref, sc);
364 	return extent_is_shared(sc);
365 }
366 
367 /* direct refs use root == 0, key == NULL */
368 static int add_direct_ref(const struct btrfs_fs_info *fs_info,
369 			  struct preftrees *preftrees, int level, u64 parent,
370 			  u64 wanted_disk_byte, int count,
371 			  struct share_check *sc, gfp_t gfp_mask)
372 {
373 	return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
374 			      parent, wanted_disk_byte, count, sc, gfp_mask);
375 }
376 
377 /* indirect refs use parent == 0 */
378 static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
379 			    struct preftrees *preftrees, u64 root_id,
380 			    const struct btrfs_key *key, int level,
381 			    u64 wanted_disk_byte, int count,
382 			    struct share_check *sc, gfp_t gfp_mask)
383 {
384 	struct preftree *tree = &preftrees->indirect;
385 
386 	if (!key)
387 		tree = &preftrees->indirect_missing_keys;
388 	return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
389 			      wanted_disk_byte, count, sc, gfp_mask);
390 }
391 
392 static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
393 {
394 	struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
395 	struct rb_node *parent = NULL;
396 	struct prelim_ref *ref = NULL;
397 	struct prelim_ref target = {};
398 	int result;
399 
400 	target.parent = bytenr;
401 
402 	while (*p) {
403 		parent = *p;
404 		ref = rb_entry(parent, struct prelim_ref, rbnode);
405 		result = prelim_ref_compare(ref, &target);
406 
407 		if (result < 0)
408 			p = &(*p)->rb_left;
409 		else if (result > 0)
410 			p = &(*p)->rb_right;
411 		else
412 			return 1;
413 	}
414 	return 0;
415 }
416 
417 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
418 			   struct ulist *parents,
419 			   struct preftrees *preftrees, struct prelim_ref *ref,
420 			   int level, u64 time_seq, const u64 *extent_item_pos,
421 			   bool ignore_offset)
422 {
423 	int ret = 0;
424 	int slot;
425 	struct extent_buffer *eb;
426 	struct btrfs_key key;
427 	struct btrfs_key *key_for_search = &ref->key_for_search;
428 	struct btrfs_file_extent_item *fi;
429 	struct extent_inode_elem *eie = NULL, *old = NULL;
430 	u64 disk_byte;
431 	u64 wanted_disk_byte = ref->wanted_disk_byte;
432 	u64 count = 0;
433 	u64 data_offset;
434 
435 	if (level != 0) {
436 		eb = path->nodes[level];
437 		ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
438 		if (ret < 0)
439 			return ret;
440 		return 0;
441 	}
442 
443 	/*
444 	 * 1. We normally enter this function with the path already pointing to
445 	 *    the first item to check. But sometimes, we may enter it with
446 	 *    slot == nritems.
447 	 * 2. We are searching for normal backref but bytenr of this leaf
448 	 *    matches shared data backref
449 	 * 3. The leaf owner is not equal to the root we are searching
450 	 *
451 	 * For these cases, go to the next leaf before we continue.
452 	 */
453 	eb = path->nodes[0];
454 	if (path->slots[0] >= btrfs_header_nritems(eb) ||
455 	    is_shared_data_backref(preftrees, eb->start) ||
456 	    ref->root_id != btrfs_header_owner(eb)) {
457 		if (time_seq == BTRFS_SEQ_LAST)
458 			ret = btrfs_next_leaf(root, path);
459 		else
460 			ret = btrfs_next_old_leaf(root, path, time_seq);
461 	}
462 
463 	while (!ret && count < ref->count) {
464 		eb = path->nodes[0];
465 		slot = path->slots[0];
466 
467 		btrfs_item_key_to_cpu(eb, &key, slot);
468 
469 		if (key.objectid != key_for_search->objectid ||
470 		    key.type != BTRFS_EXTENT_DATA_KEY)
471 			break;
472 
473 		/*
474 		 * We are searching for normal backref but bytenr of this leaf
475 		 * matches shared data backref, OR
476 		 * the leaf owner is not equal to the root we are searching for
477 		 */
478 		if (slot == 0 &&
479 		    (is_shared_data_backref(preftrees, eb->start) ||
480 		     ref->root_id != btrfs_header_owner(eb))) {
481 			if (time_seq == BTRFS_SEQ_LAST)
482 				ret = btrfs_next_leaf(root, path);
483 			else
484 				ret = btrfs_next_old_leaf(root, path, time_seq);
485 			continue;
486 		}
487 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
488 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
489 		data_offset = btrfs_file_extent_offset(eb, fi);
490 
491 		if (disk_byte == wanted_disk_byte) {
492 			eie = NULL;
493 			old = NULL;
494 			if (ref->key_for_search.offset == key.offset - data_offset)
495 				count++;
496 			else
497 				goto next;
498 			if (extent_item_pos) {
499 				ret = check_extent_in_eb(&key, eb, fi,
500 						*extent_item_pos,
501 						&eie, ignore_offset);
502 				if (ret < 0)
503 					break;
504 			}
505 			if (ret > 0)
506 				goto next;
507 			ret = ulist_add_merge_ptr(parents, eb->start,
508 						  eie, (void **)&old, GFP_NOFS);
509 			if (ret < 0)
510 				break;
511 			if (!ret && extent_item_pos) {
512 				while (old->next)
513 					old = old->next;
514 				old->next = eie;
515 			}
516 			eie = NULL;
517 		}
518 next:
519 		if (time_seq == BTRFS_SEQ_LAST)
520 			ret = btrfs_next_item(root, path);
521 		else
522 			ret = btrfs_next_old_item(root, path, time_seq);
523 	}
524 
525 	if (ret > 0)
526 		ret = 0;
527 	else if (ret < 0)
528 		free_inode_elem_list(eie);
529 	return ret;
530 }
531 
532 /*
533  * resolve an indirect backref in the form (root_id, key, level)
534  * to a logical address
535  */
536 static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
537 				struct btrfs_path *path, u64 time_seq,
538 				struct preftrees *preftrees,
539 				struct prelim_ref *ref, struct ulist *parents,
540 				const u64 *extent_item_pos, bool ignore_offset)
541 {
542 	struct btrfs_root *root;
543 	struct extent_buffer *eb;
544 	int ret = 0;
545 	int root_level;
546 	int level = ref->level;
547 	struct btrfs_key search_key = ref->key_for_search;
548 
549 	/*
550 	 * If we're search_commit_root we could possibly be holding locks on
551 	 * other tree nodes.  This happens when qgroups does backref walks when
552 	 * adding new delayed refs.  To deal with this we need to look in cache
553 	 * for the root, and if we don't find it then we need to search the
554 	 * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage
555 	 * here.
556 	 */
557 	if (path->search_commit_root)
558 		root = btrfs_get_fs_root_commit_root(fs_info, path, ref->root_id);
559 	else
560 		root = btrfs_get_fs_root(fs_info, ref->root_id, false);
561 	if (IS_ERR(root)) {
562 		ret = PTR_ERR(root);
563 		goto out_free;
564 	}
565 
566 	if (!path->search_commit_root &&
567 	    test_bit(BTRFS_ROOT_DELETING, &root->state)) {
568 		ret = -ENOENT;
569 		goto out;
570 	}
571 
572 	if (btrfs_is_testing(fs_info)) {
573 		ret = -ENOENT;
574 		goto out;
575 	}
576 
577 	if (path->search_commit_root)
578 		root_level = btrfs_header_level(root->commit_root);
579 	else if (time_seq == BTRFS_SEQ_LAST)
580 		root_level = btrfs_header_level(root->node);
581 	else
582 		root_level = btrfs_old_root_level(root, time_seq);
583 
584 	if (root_level + 1 == level)
585 		goto out;
586 
587 	/*
588 	 * We can often find data backrefs with an offset that is too large
589 	 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
590 	 * subtracting a file's offset with the data offset of its
591 	 * corresponding extent data item. This can happen for example in the
592 	 * clone ioctl.
593 	 *
594 	 * So if we detect such case we set the search key's offset to zero to
595 	 * make sure we will find the matching file extent item at
596 	 * add_all_parents(), otherwise we will miss it because the offset
597 	 * taken form the backref is much larger then the offset of the file
598 	 * extent item. This can make us scan a very large number of file
599 	 * extent items, but at least it will not make us miss any.
600 	 *
601 	 * This is an ugly workaround for a behaviour that should have never
602 	 * existed, but it does and a fix for the clone ioctl would touch a lot
603 	 * of places, cause backwards incompatibility and would not fix the
604 	 * problem for extents cloned with older kernels.
605 	 */
606 	if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
607 	    search_key.offset >= LLONG_MAX)
608 		search_key.offset = 0;
609 	path->lowest_level = level;
610 	if (time_seq == BTRFS_SEQ_LAST)
611 		ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
612 	else
613 		ret = btrfs_search_old_slot(root, &search_key, path, time_seq);
614 
615 	btrfs_debug(fs_info,
616 		"search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
617 		 ref->root_id, level, ref->count, ret,
618 		 ref->key_for_search.objectid, ref->key_for_search.type,
619 		 ref->key_for_search.offset);
620 	if (ret < 0)
621 		goto out;
622 
623 	eb = path->nodes[level];
624 	while (!eb) {
625 		if (WARN_ON(!level)) {
626 			ret = 1;
627 			goto out;
628 		}
629 		level--;
630 		eb = path->nodes[level];
631 	}
632 
633 	ret = add_all_parents(root, path, parents, preftrees, ref, level,
634 			      time_seq, extent_item_pos, ignore_offset);
635 out:
636 	btrfs_put_root(root);
637 out_free:
638 	path->lowest_level = 0;
639 	btrfs_release_path(path);
640 	return ret;
641 }
642 
643 static struct extent_inode_elem *
644 unode_aux_to_inode_list(struct ulist_node *node)
645 {
646 	if (!node)
647 		return NULL;
648 	return (struct extent_inode_elem *)(uintptr_t)node->aux;
649 }
650 
651 /*
652  * We maintain three separate rbtrees: one for direct refs, one for
653  * indirect refs which have a key, and one for indirect refs which do not
654  * have a key. Each tree does merge on insertion.
655  *
656  * Once all of the references are located, we iterate over the tree of
657  * indirect refs with missing keys. An appropriate key is located and
658  * the ref is moved onto the tree for indirect refs. After all missing
659  * keys are thus located, we iterate over the indirect ref tree, resolve
660  * each reference, and then insert the resolved reference onto the
661  * direct tree (merging there too).
662  *
663  * New backrefs (i.e., for parent nodes) are added to the appropriate
664  * rbtree as they are encountered. The new backrefs are subsequently
665  * resolved as above.
666  */
667 static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
668 				 struct btrfs_path *path, u64 time_seq,
669 				 struct preftrees *preftrees,
670 				 const u64 *extent_item_pos,
671 				 struct share_check *sc, bool ignore_offset)
672 {
673 	int err;
674 	int ret = 0;
675 	struct ulist *parents;
676 	struct ulist_node *node;
677 	struct ulist_iterator uiter;
678 	struct rb_node *rnode;
679 
680 	parents = ulist_alloc(GFP_NOFS);
681 	if (!parents)
682 		return -ENOMEM;
683 
684 	/*
685 	 * We could trade memory usage for performance here by iterating
686 	 * the tree, allocating new refs for each insertion, and then
687 	 * freeing the entire indirect tree when we're done.  In some test
688 	 * cases, the tree can grow quite large (~200k objects).
689 	 */
690 	while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
691 		struct prelim_ref *ref;
692 
693 		ref = rb_entry(rnode, struct prelim_ref, rbnode);
694 		if (WARN(ref->parent,
695 			 "BUG: direct ref found in indirect tree")) {
696 			ret = -EINVAL;
697 			goto out;
698 		}
699 
700 		rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
701 		preftrees->indirect.count--;
702 
703 		if (ref->count == 0) {
704 			free_pref(ref);
705 			continue;
706 		}
707 
708 		if (sc && sc->root_objectid &&
709 		    ref->root_id != sc->root_objectid) {
710 			free_pref(ref);
711 			ret = BACKREF_FOUND_SHARED;
712 			goto out;
713 		}
714 		err = resolve_indirect_ref(fs_info, path, time_seq, preftrees,
715 					   ref, parents, extent_item_pos,
716 					   ignore_offset);
717 		/*
718 		 * we can only tolerate ENOENT,otherwise,we should catch error
719 		 * and return directly.
720 		 */
721 		if (err == -ENOENT) {
722 			prelim_ref_insert(fs_info, &preftrees->direct, ref,
723 					  NULL);
724 			continue;
725 		} else if (err) {
726 			free_pref(ref);
727 			ret = err;
728 			goto out;
729 		}
730 
731 		/* we put the first parent into the ref at hand */
732 		ULIST_ITER_INIT(&uiter);
733 		node = ulist_next(parents, &uiter);
734 		ref->parent = node ? node->val : 0;
735 		ref->inode_list = unode_aux_to_inode_list(node);
736 
737 		/* Add a prelim_ref(s) for any other parent(s). */
738 		while ((node = ulist_next(parents, &uiter))) {
739 			struct prelim_ref *new_ref;
740 
741 			new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
742 						   GFP_NOFS);
743 			if (!new_ref) {
744 				free_pref(ref);
745 				ret = -ENOMEM;
746 				goto out;
747 			}
748 			memcpy(new_ref, ref, sizeof(*ref));
749 			new_ref->parent = node->val;
750 			new_ref->inode_list = unode_aux_to_inode_list(node);
751 			prelim_ref_insert(fs_info, &preftrees->direct,
752 					  new_ref, NULL);
753 		}
754 
755 		/*
756 		 * Now it's a direct ref, put it in the direct tree. We must
757 		 * do this last because the ref could be merged/freed here.
758 		 */
759 		prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
760 
761 		ulist_reinit(parents);
762 		cond_resched();
763 	}
764 out:
765 	ulist_free(parents);
766 	return ret;
767 }
768 
769 /*
770  * read tree blocks and add keys where required.
771  */
772 static int add_missing_keys(struct btrfs_fs_info *fs_info,
773 			    struct preftrees *preftrees, bool lock)
774 {
775 	struct prelim_ref *ref;
776 	struct extent_buffer *eb;
777 	struct preftree *tree = &preftrees->indirect_missing_keys;
778 	struct rb_node *node;
779 
780 	while ((node = rb_first_cached(&tree->root))) {
781 		ref = rb_entry(node, struct prelim_ref, rbnode);
782 		rb_erase_cached(node, &tree->root);
783 
784 		BUG_ON(ref->parent);	/* should not be a direct ref */
785 		BUG_ON(ref->key_for_search.type);
786 		BUG_ON(!ref->wanted_disk_byte);
787 
788 		eb = read_tree_block(fs_info, ref->wanted_disk_byte,
789 				     ref->root_id, 0, ref->level - 1, NULL);
790 		if (IS_ERR(eb)) {
791 			free_pref(ref);
792 			return PTR_ERR(eb);
793 		}
794 		if (!extent_buffer_uptodate(eb)) {
795 			free_pref(ref);
796 			free_extent_buffer(eb);
797 			return -EIO;
798 		}
799 
800 		if (lock)
801 			btrfs_tree_read_lock(eb);
802 		if (btrfs_header_level(eb) == 0)
803 			btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
804 		else
805 			btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
806 		if (lock)
807 			btrfs_tree_read_unlock(eb);
808 		free_extent_buffer(eb);
809 		prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
810 		cond_resched();
811 	}
812 	return 0;
813 }
814 
815 /*
816  * add all currently queued delayed refs from this head whose seq nr is
817  * smaller or equal that seq to the list
818  */
819 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
820 			    struct btrfs_delayed_ref_head *head, u64 seq,
821 			    struct preftrees *preftrees, struct share_check *sc)
822 {
823 	struct btrfs_delayed_ref_node *node;
824 	struct btrfs_key key;
825 	struct rb_node *n;
826 	int count;
827 	int ret = 0;
828 
829 	spin_lock(&head->lock);
830 	for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
831 		node = rb_entry(n, struct btrfs_delayed_ref_node,
832 				ref_node);
833 		if (node->seq > seq)
834 			continue;
835 
836 		switch (node->action) {
837 		case BTRFS_ADD_DELAYED_EXTENT:
838 		case BTRFS_UPDATE_DELAYED_HEAD:
839 			WARN_ON(1);
840 			continue;
841 		case BTRFS_ADD_DELAYED_REF:
842 			count = node->ref_mod;
843 			break;
844 		case BTRFS_DROP_DELAYED_REF:
845 			count = node->ref_mod * -1;
846 			break;
847 		default:
848 			BUG();
849 		}
850 		switch (node->type) {
851 		case BTRFS_TREE_BLOCK_REF_KEY: {
852 			/* NORMAL INDIRECT METADATA backref */
853 			struct btrfs_delayed_tree_ref *ref;
854 			struct btrfs_key *key_ptr = NULL;
855 
856 			if (head->extent_op && head->extent_op->update_key) {
857 				btrfs_disk_key_to_cpu(&key, &head->extent_op->key);
858 				key_ptr = &key;
859 			}
860 
861 			ref = btrfs_delayed_node_to_tree_ref(node);
862 			ret = add_indirect_ref(fs_info, preftrees, ref->root,
863 					       key_ptr, ref->level + 1,
864 					       node->bytenr, count, sc,
865 					       GFP_ATOMIC);
866 			break;
867 		}
868 		case BTRFS_SHARED_BLOCK_REF_KEY: {
869 			/* SHARED DIRECT METADATA backref */
870 			struct btrfs_delayed_tree_ref *ref;
871 
872 			ref = btrfs_delayed_node_to_tree_ref(node);
873 
874 			ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
875 					     ref->parent, node->bytenr, count,
876 					     sc, GFP_ATOMIC);
877 			break;
878 		}
879 		case BTRFS_EXTENT_DATA_REF_KEY: {
880 			/* NORMAL INDIRECT DATA backref */
881 			struct btrfs_delayed_data_ref *ref;
882 			ref = btrfs_delayed_node_to_data_ref(node);
883 
884 			key.objectid = ref->objectid;
885 			key.type = BTRFS_EXTENT_DATA_KEY;
886 			key.offset = ref->offset;
887 
888 			/*
889 			 * If we have a share check context and a reference for
890 			 * another inode, we can't exit immediately. This is
891 			 * because even if this is a BTRFS_ADD_DELAYED_REF
892 			 * reference we may find next a BTRFS_DROP_DELAYED_REF
893 			 * which cancels out this ADD reference.
894 			 *
895 			 * If this is a DROP reference and there was no previous
896 			 * ADD reference, then we need to signal that when we
897 			 * process references from the extent tree (through
898 			 * add_inline_refs() and add_keyed_refs()), we should
899 			 * not exit early if we find a reference for another
900 			 * inode, because one of the delayed DROP references
901 			 * may cancel that reference in the extent tree.
902 			 */
903 			if (sc && count < 0)
904 				sc->have_delayed_delete_refs = true;
905 
906 			ret = add_indirect_ref(fs_info, preftrees, ref->root,
907 					       &key, 0, node->bytenr, count, sc,
908 					       GFP_ATOMIC);
909 			break;
910 		}
911 		case BTRFS_SHARED_DATA_REF_KEY: {
912 			/* SHARED DIRECT FULL backref */
913 			struct btrfs_delayed_data_ref *ref;
914 
915 			ref = btrfs_delayed_node_to_data_ref(node);
916 
917 			ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
918 					     node->bytenr, count, sc,
919 					     GFP_ATOMIC);
920 			break;
921 		}
922 		default:
923 			WARN_ON(1);
924 		}
925 		/*
926 		 * We must ignore BACKREF_FOUND_SHARED until all delayed
927 		 * refs have been checked.
928 		 */
929 		if (ret && (ret != BACKREF_FOUND_SHARED))
930 			break;
931 	}
932 	if (!ret)
933 		ret = extent_is_shared(sc);
934 
935 	spin_unlock(&head->lock);
936 	return ret;
937 }
938 
939 /*
940  * add all inline backrefs for bytenr to the list
941  *
942  * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
943  */
944 static int add_inline_refs(const struct btrfs_fs_info *fs_info,
945 			   struct btrfs_path *path, u64 bytenr,
946 			   int *info_level, struct preftrees *preftrees,
947 			   struct share_check *sc)
948 {
949 	int ret = 0;
950 	int slot;
951 	struct extent_buffer *leaf;
952 	struct btrfs_key key;
953 	struct btrfs_key found_key;
954 	unsigned long ptr;
955 	unsigned long end;
956 	struct btrfs_extent_item *ei;
957 	u64 flags;
958 	u64 item_size;
959 
960 	/*
961 	 * enumerate all inline refs
962 	 */
963 	leaf = path->nodes[0];
964 	slot = path->slots[0];
965 
966 	item_size = btrfs_item_size(leaf, slot);
967 	BUG_ON(item_size < sizeof(*ei));
968 
969 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
970 	flags = btrfs_extent_flags(leaf, ei);
971 	btrfs_item_key_to_cpu(leaf, &found_key, slot);
972 
973 	ptr = (unsigned long)(ei + 1);
974 	end = (unsigned long)ei + item_size;
975 
976 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
977 	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
978 		struct btrfs_tree_block_info *info;
979 
980 		info = (struct btrfs_tree_block_info *)ptr;
981 		*info_level = btrfs_tree_block_level(leaf, info);
982 		ptr += sizeof(struct btrfs_tree_block_info);
983 		BUG_ON(ptr > end);
984 	} else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
985 		*info_level = found_key.offset;
986 	} else {
987 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
988 	}
989 
990 	while (ptr < end) {
991 		struct btrfs_extent_inline_ref *iref;
992 		u64 offset;
993 		int type;
994 
995 		iref = (struct btrfs_extent_inline_ref *)ptr;
996 		type = btrfs_get_extent_inline_ref_type(leaf, iref,
997 							BTRFS_REF_TYPE_ANY);
998 		if (type == BTRFS_REF_TYPE_INVALID)
999 			return -EUCLEAN;
1000 
1001 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
1002 
1003 		switch (type) {
1004 		case BTRFS_SHARED_BLOCK_REF_KEY:
1005 			ret = add_direct_ref(fs_info, preftrees,
1006 					     *info_level + 1, offset,
1007 					     bytenr, 1, NULL, GFP_NOFS);
1008 			break;
1009 		case BTRFS_SHARED_DATA_REF_KEY: {
1010 			struct btrfs_shared_data_ref *sdref;
1011 			int count;
1012 
1013 			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
1014 			count = btrfs_shared_data_ref_count(leaf, sdref);
1015 
1016 			ret = add_direct_ref(fs_info, preftrees, 0, offset,
1017 					     bytenr, count, sc, GFP_NOFS);
1018 			break;
1019 		}
1020 		case BTRFS_TREE_BLOCK_REF_KEY:
1021 			ret = add_indirect_ref(fs_info, preftrees, offset,
1022 					       NULL, *info_level + 1,
1023 					       bytenr, 1, NULL, GFP_NOFS);
1024 			break;
1025 		case BTRFS_EXTENT_DATA_REF_KEY: {
1026 			struct btrfs_extent_data_ref *dref;
1027 			int count;
1028 			u64 root;
1029 
1030 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1031 			count = btrfs_extent_data_ref_count(leaf, dref);
1032 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
1033 								      dref);
1034 			key.type = BTRFS_EXTENT_DATA_KEY;
1035 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1036 
1037 			if (sc && sc->inum && key.objectid != sc->inum &&
1038 			    !sc->have_delayed_delete_refs) {
1039 				ret = BACKREF_FOUND_SHARED;
1040 				break;
1041 			}
1042 
1043 			root = btrfs_extent_data_ref_root(leaf, dref);
1044 
1045 			ret = add_indirect_ref(fs_info, preftrees, root,
1046 					       &key, 0, bytenr, count,
1047 					       sc, GFP_NOFS);
1048 
1049 			break;
1050 		}
1051 		default:
1052 			WARN_ON(1);
1053 		}
1054 		if (ret)
1055 			return ret;
1056 		ptr += btrfs_extent_inline_ref_size(type);
1057 	}
1058 
1059 	return 0;
1060 }
1061 
1062 /*
1063  * add all non-inline backrefs for bytenr to the list
1064  *
1065  * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1066  */
1067 static int add_keyed_refs(struct btrfs_root *extent_root,
1068 			  struct btrfs_path *path, u64 bytenr,
1069 			  int info_level, struct preftrees *preftrees,
1070 			  struct share_check *sc)
1071 {
1072 	struct btrfs_fs_info *fs_info = extent_root->fs_info;
1073 	int ret;
1074 	int slot;
1075 	struct extent_buffer *leaf;
1076 	struct btrfs_key key;
1077 
1078 	while (1) {
1079 		ret = btrfs_next_item(extent_root, path);
1080 		if (ret < 0)
1081 			break;
1082 		if (ret) {
1083 			ret = 0;
1084 			break;
1085 		}
1086 
1087 		slot = path->slots[0];
1088 		leaf = path->nodes[0];
1089 		btrfs_item_key_to_cpu(leaf, &key, slot);
1090 
1091 		if (key.objectid != bytenr)
1092 			break;
1093 		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1094 			continue;
1095 		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1096 			break;
1097 
1098 		switch (key.type) {
1099 		case BTRFS_SHARED_BLOCK_REF_KEY:
1100 			/* SHARED DIRECT METADATA backref */
1101 			ret = add_direct_ref(fs_info, preftrees,
1102 					     info_level + 1, key.offset,
1103 					     bytenr, 1, NULL, GFP_NOFS);
1104 			break;
1105 		case BTRFS_SHARED_DATA_REF_KEY: {
1106 			/* SHARED DIRECT FULL backref */
1107 			struct btrfs_shared_data_ref *sdref;
1108 			int count;
1109 
1110 			sdref = btrfs_item_ptr(leaf, slot,
1111 					      struct btrfs_shared_data_ref);
1112 			count = btrfs_shared_data_ref_count(leaf, sdref);
1113 			ret = add_direct_ref(fs_info, preftrees, 0,
1114 					     key.offset, bytenr, count,
1115 					     sc, GFP_NOFS);
1116 			break;
1117 		}
1118 		case BTRFS_TREE_BLOCK_REF_KEY:
1119 			/* NORMAL INDIRECT METADATA backref */
1120 			ret = add_indirect_ref(fs_info, preftrees, key.offset,
1121 					       NULL, info_level + 1, bytenr,
1122 					       1, NULL, GFP_NOFS);
1123 			break;
1124 		case BTRFS_EXTENT_DATA_REF_KEY: {
1125 			/* NORMAL INDIRECT DATA backref */
1126 			struct btrfs_extent_data_ref *dref;
1127 			int count;
1128 			u64 root;
1129 
1130 			dref = btrfs_item_ptr(leaf, slot,
1131 					      struct btrfs_extent_data_ref);
1132 			count = btrfs_extent_data_ref_count(leaf, dref);
1133 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
1134 								      dref);
1135 			key.type = BTRFS_EXTENT_DATA_KEY;
1136 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1137 
1138 			if (sc && sc->inum && key.objectid != sc->inum &&
1139 			    !sc->have_delayed_delete_refs) {
1140 				ret = BACKREF_FOUND_SHARED;
1141 				break;
1142 			}
1143 
1144 			root = btrfs_extent_data_ref_root(leaf, dref);
1145 			ret = add_indirect_ref(fs_info, preftrees, root,
1146 					       &key, 0, bytenr, count,
1147 					       sc, GFP_NOFS);
1148 			break;
1149 		}
1150 		default:
1151 			WARN_ON(1);
1152 		}
1153 		if (ret)
1154 			return ret;
1155 
1156 	}
1157 
1158 	return ret;
1159 }
1160 
1161 /*
1162  * this adds all existing backrefs (inline backrefs, backrefs and delayed
1163  * refs) for the given bytenr to the refs list, merges duplicates and resolves
1164  * indirect refs to their parent bytenr.
1165  * When roots are found, they're added to the roots list
1166  *
1167  * If time_seq is set to BTRFS_SEQ_LAST, it will not search delayed_refs, and
1168  * behave much like trans == NULL case, the difference only lies in it will not
1169  * commit root.
1170  * The special case is for qgroup to search roots in commit_transaction().
1171  *
1172  * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1173  * shared extent is detected.
1174  *
1175  * Otherwise this returns 0 for success and <0 for an error.
1176  *
1177  * If ignore_offset is set to false, only extent refs whose offsets match
1178  * extent_item_pos are returned.  If true, every extent ref is returned
1179  * and extent_item_pos is ignored.
1180  *
1181  * FIXME some caching might speed things up
1182  */
1183 static int find_parent_nodes(struct btrfs_trans_handle *trans,
1184 			     struct btrfs_fs_info *fs_info, u64 bytenr,
1185 			     u64 time_seq, struct ulist *refs,
1186 			     struct ulist *roots, const u64 *extent_item_pos,
1187 			     struct share_check *sc, bool ignore_offset)
1188 {
1189 	struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
1190 	struct btrfs_key key;
1191 	struct btrfs_path *path;
1192 	struct btrfs_delayed_ref_root *delayed_refs = NULL;
1193 	struct btrfs_delayed_ref_head *head;
1194 	int info_level = 0;
1195 	int ret;
1196 	struct prelim_ref *ref;
1197 	struct rb_node *node;
1198 	struct extent_inode_elem *eie = NULL;
1199 	struct preftrees preftrees = {
1200 		.direct = PREFTREE_INIT,
1201 		.indirect = PREFTREE_INIT,
1202 		.indirect_missing_keys = PREFTREE_INIT
1203 	};
1204 
1205 	key.objectid = bytenr;
1206 	key.offset = (u64)-1;
1207 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1208 		key.type = BTRFS_METADATA_ITEM_KEY;
1209 	else
1210 		key.type = BTRFS_EXTENT_ITEM_KEY;
1211 
1212 	path = btrfs_alloc_path();
1213 	if (!path)
1214 		return -ENOMEM;
1215 	if (!trans) {
1216 		path->search_commit_root = 1;
1217 		path->skip_locking = 1;
1218 	}
1219 
1220 	if (time_seq == BTRFS_SEQ_LAST)
1221 		path->skip_locking = 1;
1222 
1223 again:
1224 	head = NULL;
1225 
1226 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1227 	if (ret < 0)
1228 		goto out;
1229 	if (ret == 0) {
1230 		/* This shouldn't happen, indicates a bug or fs corruption. */
1231 		ASSERT(ret != 0);
1232 		ret = -EUCLEAN;
1233 		goto out;
1234 	}
1235 
1236 	if (trans && likely(trans->type != __TRANS_DUMMY) &&
1237 	    time_seq != BTRFS_SEQ_LAST) {
1238 		/*
1239 		 * We have a specific time_seq we care about and trans which
1240 		 * means we have the path lock, we need to grab the ref head and
1241 		 * lock it so we have a consistent view of the refs at the given
1242 		 * time.
1243 		 */
1244 		delayed_refs = &trans->transaction->delayed_refs;
1245 		spin_lock(&delayed_refs->lock);
1246 		head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1247 		if (head) {
1248 			if (!mutex_trylock(&head->mutex)) {
1249 				refcount_inc(&head->refs);
1250 				spin_unlock(&delayed_refs->lock);
1251 
1252 				btrfs_release_path(path);
1253 
1254 				/*
1255 				 * Mutex was contended, block until it's
1256 				 * released and try again
1257 				 */
1258 				mutex_lock(&head->mutex);
1259 				mutex_unlock(&head->mutex);
1260 				btrfs_put_delayed_ref_head(head);
1261 				goto again;
1262 			}
1263 			spin_unlock(&delayed_refs->lock);
1264 			ret = add_delayed_refs(fs_info, head, time_seq,
1265 					       &preftrees, sc);
1266 			mutex_unlock(&head->mutex);
1267 			if (ret)
1268 				goto out;
1269 		} else {
1270 			spin_unlock(&delayed_refs->lock);
1271 		}
1272 	}
1273 
1274 	if (path->slots[0]) {
1275 		struct extent_buffer *leaf;
1276 		int slot;
1277 
1278 		path->slots[0]--;
1279 		leaf = path->nodes[0];
1280 		slot = path->slots[0];
1281 		btrfs_item_key_to_cpu(leaf, &key, slot);
1282 		if (key.objectid == bytenr &&
1283 		    (key.type == BTRFS_EXTENT_ITEM_KEY ||
1284 		     key.type == BTRFS_METADATA_ITEM_KEY)) {
1285 			ret = add_inline_refs(fs_info, path, bytenr,
1286 					      &info_level, &preftrees, sc);
1287 			if (ret)
1288 				goto out;
1289 			ret = add_keyed_refs(root, path, bytenr, info_level,
1290 					     &preftrees, sc);
1291 			if (ret)
1292 				goto out;
1293 		}
1294 	}
1295 
1296 	btrfs_release_path(path);
1297 
1298 	ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
1299 	if (ret)
1300 		goto out;
1301 
1302 	WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1303 
1304 	ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1305 				    extent_item_pos, sc, ignore_offset);
1306 	if (ret)
1307 		goto out;
1308 
1309 	WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1310 
1311 	/*
1312 	 * This walks the tree of merged and resolved refs. Tree blocks are
1313 	 * read in as needed. Unique entries are added to the ulist, and
1314 	 * the list of found roots is updated.
1315 	 *
1316 	 * We release the entire tree in one go before returning.
1317 	 */
1318 	node = rb_first_cached(&preftrees.direct.root);
1319 	while (node) {
1320 		ref = rb_entry(node, struct prelim_ref, rbnode);
1321 		node = rb_next(&ref->rbnode);
1322 		/*
1323 		 * ref->count < 0 can happen here if there are delayed
1324 		 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1325 		 * prelim_ref_insert() relies on this when merging
1326 		 * identical refs to keep the overall count correct.
1327 		 * prelim_ref_insert() will merge only those refs
1328 		 * which compare identically.  Any refs having
1329 		 * e.g. different offsets would not be merged,
1330 		 * and would retain their original ref->count < 0.
1331 		 */
1332 		if (roots && ref->count && ref->root_id && ref->parent == 0) {
1333 			if (sc && sc->root_objectid &&
1334 			    ref->root_id != sc->root_objectid) {
1335 				ret = BACKREF_FOUND_SHARED;
1336 				goto out;
1337 			}
1338 
1339 			/* no parent == root of tree */
1340 			ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1341 			if (ret < 0)
1342 				goto out;
1343 		}
1344 		if (ref->count && ref->parent) {
1345 			if (extent_item_pos && !ref->inode_list &&
1346 			    ref->level == 0) {
1347 				struct extent_buffer *eb;
1348 
1349 				eb = read_tree_block(fs_info, ref->parent, 0,
1350 						     0, ref->level, NULL);
1351 				if (IS_ERR(eb)) {
1352 					ret = PTR_ERR(eb);
1353 					goto out;
1354 				}
1355 				if (!extent_buffer_uptodate(eb)) {
1356 					free_extent_buffer(eb);
1357 					ret = -EIO;
1358 					goto out;
1359 				}
1360 
1361 				if (!path->skip_locking)
1362 					btrfs_tree_read_lock(eb);
1363 				ret = find_extent_in_eb(eb, bytenr,
1364 							*extent_item_pos, &eie, ignore_offset);
1365 				if (!path->skip_locking)
1366 					btrfs_tree_read_unlock(eb);
1367 				free_extent_buffer(eb);
1368 				if (ret < 0)
1369 					goto out;
1370 				ref->inode_list = eie;
1371 			}
1372 			ret = ulist_add_merge_ptr(refs, ref->parent,
1373 						  ref->inode_list,
1374 						  (void **)&eie, GFP_NOFS);
1375 			if (ret < 0)
1376 				goto out;
1377 			if (!ret && extent_item_pos) {
1378 				/*
1379 				 * We've recorded that parent, so we must extend
1380 				 * its inode list here.
1381 				 *
1382 				 * However if there was corruption we may not
1383 				 * have found an eie, return an error in this
1384 				 * case.
1385 				 */
1386 				ASSERT(eie);
1387 				if (!eie) {
1388 					ret = -EUCLEAN;
1389 					goto out;
1390 				}
1391 				while (eie->next)
1392 					eie = eie->next;
1393 				eie->next = ref->inode_list;
1394 			}
1395 			eie = NULL;
1396 		}
1397 		cond_resched();
1398 	}
1399 
1400 out:
1401 	btrfs_free_path(path);
1402 
1403 	prelim_release(&preftrees.direct);
1404 	prelim_release(&preftrees.indirect);
1405 	prelim_release(&preftrees.indirect_missing_keys);
1406 
1407 	if (ret < 0)
1408 		free_inode_elem_list(eie);
1409 	return ret;
1410 }
1411 
1412 static void free_leaf_list(struct ulist *blocks)
1413 {
1414 	struct ulist_node *node = NULL;
1415 	struct extent_inode_elem *eie;
1416 	struct ulist_iterator uiter;
1417 
1418 	ULIST_ITER_INIT(&uiter);
1419 	while ((node = ulist_next(blocks, &uiter))) {
1420 		if (!node->aux)
1421 			continue;
1422 		eie = unode_aux_to_inode_list(node);
1423 		free_inode_elem_list(eie);
1424 		node->aux = 0;
1425 	}
1426 
1427 	ulist_free(blocks);
1428 }
1429 
1430 /*
1431  * Finds all leafs with a reference to the specified combination of bytenr and
1432  * offset. key_list_head will point to a list of corresponding keys (caller must
1433  * free each list element). The leafs will be stored in the leafs ulist, which
1434  * must be freed with ulist_free.
1435  *
1436  * returns 0 on success, <0 on error
1437  */
1438 int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1439 			 struct btrfs_fs_info *fs_info, u64 bytenr,
1440 			 u64 time_seq, struct ulist **leafs,
1441 			 const u64 *extent_item_pos, bool ignore_offset)
1442 {
1443 	int ret;
1444 
1445 	*leafs = ulist_alloc(GFP_NOFS);
1446 	if (!*leafs)
1447 		return -ENOMEM;
1448 
1449 	ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1450 				*leafs, NULL, extent_item_pos, NULL, ignore_offset);
1451 	if (ret < 0 && ret != -ENOENT) {
1452 		free_leaf_list(*leafs);
1453 		return ret;
1454 	}
1455 
1456 	return 0;
1457 }
1458 
1459 /*
1460  * walk all backrefs for a given extent to find all roots that reference this
1461  * extent. Walking a backref means finding all extents that reference this
1462  * extent and in turn walk the backrefs of those, too. Naturally this is a
1463  * recursive process, but here it is implemented in an iterative fashion: We
1464  * find all referencing extents for the extent in question and put them on a
1465  * list. In turn, we find all referencing extents for those, further appending
1466  * to the list. The way we iterate the list allows adding more elements after
1467  * the current while iterating. The process stops when we reach the end of the
1468  * list. Found roots are added to the roots list.
1469  *
1470  * returns 0 on success, < 0 on error.
1471  */
1472 static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1473 				     struct btrfs_fs_info *fs_info, u64 bytenr,
1474 				     u64 time_seq, struct ulist **roots,
1475 				     bool ignore_offset)
1476 {
1477 	struct ulist *tmp;
1478 	struct ulist_node *node = NULL;
1479 	struct ulist_iterator uiter;
1480 	int ret;
1481 
1482 	tmp = ulist_alloc(GFP_NOFS);
1483 	if (!tmp)
1484 		return -ENOMEM;
1485 	*roots = ulist_alloc(GFP_NOFS);
1486 	if (!*roots) {
1487 		ulist_free(tmp);
1488 		return -ENOMEM;
1489 	}
1490 
1491 	ULIST_ITER_INIT(&uiter);
1492 	while (1) {
1493 		ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1494 					tmp, *roots, NULL, NULL, ignore_offset);
1495 		if (ret < 0 && ret != -ENOENT) {
1496 			ulist_free(tmp);
1497 			ulist_free(*roots);
1498 			*roots = NULL;
1499 			return ret;
1500 		}
1501 		node = ulist_next(tmp, &uiter);
1502 		if (!node)
1503 			break;
1504 		bytenr = node->val;
1505 		cond_resched();
1506 	}
1507 
1508 	ulist_free(tmp);
1509 	return 0;
1510 }
1511 
1512 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1513 			 struct btrfs_fs_info *fs_info, u64 bytenr,
1514 			 u64 time_seq, struct ulist **roots,
1515 			 bool skip_commit_root_sem)
1516 {
1517 	int ret;
1518 
1519 	if (!trans && !skip_commit_root_sem)
1520 		down_read(&fs_info->commit_root_sem);
1521 	ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1522 					time_seq, roots, false);
1523 	if (!trans && !skip_commit_root_sem)
1524 		up_read(&fs_info->commit_root_sem);
1525 	return ret;
1526 }
1527 
1528 /*
1529  * The caller has joined a transaction or is holding a read lock on the
1530  * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1531  * snapshot field changing while updating or checking the cache.
1532  */
1533 static bool lookup_backref_shared_cache(struct btrfs_backref_shared_cache *cache,
1534 					struct btrfs_root *root,
1535 					u64 bytenr, int level, bool *is_shared)
1536 {
1537 	struct btrfs_backref_shared_cache_entry *entry;
1538 
1539 	if (!cache->use_cache)
1540 		return false;
1541 
1542 	if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1543 		return false;
1544 
1545 	/*
1546 	 * Level -1 is used for the data extent, which is not reliable to cache
1547 	 * because its reference count can increase or decrease without us
1548 	 * realizing. We cache results only for extent buffers that lead from
1549 	 * the root node down to the leaf with the file extent item.
1550 	 */
1551 	ASSERT(level >= 0);
1552 
1553 	entry = &cache->entries[level];
1554 
1555 	/* Unused cache entry or being used for some other extent buffer. */
1556 	if (entry->bytenr != bytenr)
1557 		return false;
1558 
1559 	/*
1560 	 * We cached a false result, but the last snapshot generation of the
1561 	 * root changed, so we now have a snapshot. Don't trust the result.
1562 	 */
1563 	if (!entry->is_shared &&
1564 	    entry->gen != btrfs_root_last_snapshot(&root->root_item))
1565 		return false;
1566 
1567 	/*
1568 	 * If we cached a true result and the last generation used for dropping
1569 	 * a root changed, we can not trust the result, because the dropped root
1570 	 * could be a snapshot sharing this extent buffer.
1571 	 */
1572 	if (entry->is_shared &&
1573 	    entry->gen != btrfs_get_last_root_drop_gen(root->fs_info))
1574 		return false;
1575 
1576 	*is_shared = entry->is_shared;
1577 	/*
1578 	 * If the node at this level is shared, than all nodes below are also
1579 	 * shared. Currently some of the nodes below may be marked as not shared
1580 	 * because we have just switched from one leaf to another, and switched
1581 	 * also other nodes above the leaf and below the current level, so mark
1582 	 * them as shared.
1583 	 */
1584 	if (*is_shared) {
1585 		for (int i = 0; i < level; i++) {
1586 			cache->entries[i].is_shared = true;
1587 			cache->entries[i].gen = entry->gen;
1588 		}
1589 	}
1590 
1591 	return true;
1592 }
1593 
1594 /*
1595  * The caller has joined a transaction or is holding a read lock on the
1596  * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1597  * snapshot field changing while updating or checking the cache.
1598  */
1599 static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache,
1600 				       struct btrfs_root *root,
1601 				       u64 bytenr, int level, bool is_shared)
1602 {
1603 	struct btrfs_backref_shared_cache_entry *entry;
1604 	u64 gen;
1605 
1606 	if (!cache->use_cache)
1607 		return;
1608 
1609 	if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1610 		return;
1611 
1612 	/*
1613 	 * Level -1 is used for the data extent, which is not reliable to cache
1614 	 * because its reference count can increase or decrease without us
1615 	 * realizing. We cache results only for extent buffers that lead from
1616 	 * the root node down to the leaf with the file extent item.
1617 	 */
1618 	ASSERT(level >= 0);
1619 
1620 	if (is_shared)
1621 		gen = btrfs_get_last_root_drop_gen(root->fs_info);
1622 	else
1623 		gen = btrfs_root_last_snapshot(&root->root_item);
1624 
1625 	entry = &cache->entries[level];
1626 	entry->bytenr = bytenr;
1627 	entry->is_shared = is_shared;
1628 	entry->gen = gen;
1629 
1630 	/*
1631 	 * If we found an extent buffer is shared, set the cache result for all
1632 	 * extent buffers below it to true. As nodes in the path are COWed,
1633 	 * their sharedness is moved to their children, and if a leaf is COWed,
1634 	 * then the sharedness of a data extent becomes direct, the refcount of
1635 	 * data extent is increased in the extent item at the extent tree.
1636 	 */
1637 	if (is_shared) {
1638 		for (int i = 0; i < level; i++) {
1639 			entry = &cache->entries[i];
1640 			entry->is_shared = is_shared;
1641 			entry->gen = gen;
1642 		}
1643 	}
1644 }
1645 
1646 /*
1647  * Check if a data extent is shared or not.
1648  *
1649  * @root:        The root the inode belongs to.
1650  * @inum:        Number of the inode whose extent we are checking.
1651  * @bytenr:      Logical bytenr of the extent we are checking.
1652  * @extent_gen:  Generation of the extent (file extent item) or 0 if it is
1653  *               not known.
1654  * @roots:       List of roots this extent is shared among.
1655  * @tmp:         Temporary list used for iteration.
1656  * @cache:       A backref lookup result cache.
1657  *
1658  * btrfs_is_data_extent_shared uses the backref walking code but will short
1659  * circuit as soon as it finds a root or inode that doesn't match the
1660  * one passed in. This provides a significant performance benefit for
1661  * callers (such as fiemap) which want to know whether the extent is
1662  * shared but do not need a ref count.
1663  *
1664  * This attempts to attach to the running transaction in order to account for
1665  * delayed refs, but continues on even when no running transaction exists.
1666  *
1667  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1668  */
1669 int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
1670 				u64 extent_gen,
1671 				struct ulist *roots, struct ulist *tmp,
1672 				struct btrfs_backref_shared_cache *cache)
1673 {
1674 	struct btrfs_fs_info *fs_info = root->fs_info;
1675 	struct btrfs_trans_handle *trans;
1676 	struct ulist_iterator uiter;
1677 	struct ulist_node *node;
1678 	struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem);
1679 	int ret = 0;
1680 	struct share_check shared = {
1681 		.root_objectid = root->root_key.objectid,
1682 		.inum = inum,
1683 		.share_count = 0,
1684 		.have_delayed_delete_refs = false,
1685 	};
1686 	int level;
1687 
1688 	ulist_init(roots);
1689 	ulist_init(tmp);
1690 
1691 	trans = btrfs_join_transaction_nostart(root);
1692 	if (IS_ERR(trans)) {
1693 		if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1694 			ret = PTR_ERR(trans);
1695 			goto out;
1696 		}
1697 		trans = NULL;
1698 		down_read(&fs_info->commit_root_sem);
1699 	} else {
1700 		btrfs_get_tree_mod_seq(fs_info, &elem);
1701 	}
1702 
1703 	/* -1 means we are in the bytenr of the data extent. */
1704 	level = -1;
1705 	ULIST_ITER_INIT(&uiter);
1706 	cache->use_cache = true;
1707 	while (1) {
1708 		bool is_shared;
1709 		bool cached;
1710 
1711 		ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1712 					roots, NULL, &shared, false);
1713 		if (ret == BACKREF_FOUND_SHARED) {
1714 			/* this is the only condition under which we return 1 */
1715 			ret = 1;
1716 			if (level >= 0)
1717 				store_backref_shared_cache(cache, root, bytenr,
1718 							   level, true);
1719 			break;
1720 		}
1721 		if (ret < 0 && ret != -ENOENT)
1722 			break;
1723 		ret = 0;
1724 		/*
1725 		 * If our data extent is not shared through reflinks and it was
1726 		 * created in a generation after the last one used to create a
1727 		 * snapshot of the inode's root, then it can not be shared
1728 		 * indirectly through subtrees, as that can only happen with
1729 		 * snapshots. In this case bail out, no need to check for the
1730 		 * sharedness of extent buffers.
1731 		 */
1732 		if (level == -1 &&
1733 		    extent_gen > btrfs_root_last_snapshot(&root->root_item))
1734 			break;
1735 
1736 		/*
1737 		 * If our data extent was not directly shared (without multiple
1738 		 * reference items), than it might have a single reference item
1739 		 * with a count > 1 for the same offset, which means there are 2
1740 		 * (or more) file extent items that point to the data extent -
1741 		 * this happens when a file extent item needs to be split and
1742 		 * then one item gets moved to another leaf due to a b+tree leaf
1743 		 * split when inserting some item. In this case the file extent
1744 		 * items may be located in different leaves and therefore some
1745 		 * of the leaves may be referenced through shared subtrees while
1746 		 * others are not. Since our extent buffer cache only works for
1747 		 * a single path (by far the most common case and simpler to
1748 		 * deal with), we can not use it if we have multiple leaves
1749 		 * (which implies multiple paths).
1750 		 */
1751 		if (level == -1 && tmp->nnodes > 1)
1752 			cache->use_cache = false;
1753 
1754 		if (level >= 0)
1755 			store_backref_shared_cache(cache, root, bytenr,
1756 						   level, false);
1757 		node = ulist_next(tmp, &uiter);
1758 		if (!node)
1759 			break;
1760 		bytenr = node->val;
1761 		level++;
1762 		cached = lookup_backref_shared_cache(cache, root, bytenr, level,
1763 						     &is_shared);
1764 		if (cached) {
1765 			ret = (is_shared ? 1 : 0);
1766 			break;
1767 		}
1768 		shared.share_count = 0;
1769 		shared.have_delayed_delete_refs = false;
1770 		cond_resched();
1771 	}
1772 
1773 	if (trans) {
1774 		btrfs_put_tree_mod_seq(fs_info, &elem);
1775 		btrfs_end_transaction(trans);
1776 	} else {
1777 		up_read(&fs_info->commit_root_sem);
1778 	}
1779 out:
1780 	ulist_release(roots);
1781 	ulist_release(tmp);
1782 	return ret;
1783 }
1784 
1785 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1786 			  u64 start_off, struct btrfs_path *path,
1787 			  struct btrfs_inode_extref **ret_extref,
1788 			  u64 *found_off)
1789 {
1790 	int ret, slot;
1791 	struct btrfs_key key;
1792 	struct btrfs_key found_key;
1793 	struct btrfs_inode_extref *extref;
1794 	const struct extent_buffer *leaf;
1795 	unsigned long ptr;
1796 
1797 	key.objectid = inode_objectid;
1798 	key.type = BTRFS_INODE_EXTREF_KEY;
1799 	key.offset = start_off;
1800 
1801 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1802 	if (ret < 0)
1803 		return ret;
1804 
1805 	while (1) {
1806 		leaf = path->nodes[0];
1807 		slot = path->slots[0];
1808 		if (slot >= btrfs_header_nritems(leaf)) {
1809 			/*
1810 			 * If the item at offset is not found,
1811 			 * btrfs_search_slot will point us to the slot
1812 			 * where it should be inserted. In our case
1813 			 * that will be the slot directly before the
1814 			 * next INODE_REF_KEY_V2 item. In the case
1815 			 * that we're pointing to the last slot in a
1816 			 * leaf, we must move one leaf over.
1817 			 */
1818 			ret = btrfs_next_leaf(root, path);
1819 			if (ret) {
1820 				if (ret >= 1)
1821 					ret = -ENOENT;
1822 				break;
1823 			}
1824 			continue;
1825 		}
1826 
1827 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1828 
1829 		/*
1830 		 * Check that we're still looking at an extended ref key for
1831 		 * this particular objectid. If we have different
1832 		 * objectid or type then there are no more to be found
1833 		 * in the tree and we can exit.
1834 		 */
1835 		ret = -ENOENT;
1836 		if (found_key.objectid != inode_objectid)
1837 			break;
1838 		if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1839 			break;
1840 
1841 		ret = 0;
1842 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1843 		extref = (struct btrfs_inode_extref *)ptr;
1844 		*ret_extref = extref;
1845 		if (found_off)
1846 			*found_off = found_key.offset;
1847 		break;
1848 	}
1849 
1850 	return ret;
1851 }
1852 
1853 /*
1854  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1855  * Elements of the path are separated by '/' and the path is guaranteed to be
1856  * 0-terminated. the path is only given within the current file system.
1857  * Therefore, it never starts with a '/'. the caller is responsible to provide
1858  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1859  * the start point of the resulting string is returned. this pointer is within
1860  * dest, normally.
1861  * in case the path buffer would overflow, the pointer is decremented further
1862  * as if output was written to the buffer, though no more output is actually
1863  * generated. that way, the caller can determine how much space would be
1864  * required for the path to fit into the buffer. in that case, the returned
1865  * value will be smaller than dest. callers must check this!
1866  */
1867 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1868 			u32 name_len, unsigned long name_off,
1869 			struct extent_buffer *eb_in, u64 parent,
1870 			char *dest, u32 size)
1871 {
1872 	int slot;
1873 	u64 next_inum;
1874 	int ret;
1875 	s64 bytes_left = ((s64)size) - 1;
1876 	struct extent_buffer *eb = eb_in;
1877 	struct btrfs_key found_key;
1878 	struct btrfs_inode_ref *iref;
1879 
1880 	if (bytes_left >= 0)
1881 		dest[bytes_left] = '\0';
1882 
1883 	while (1) {
1884 		bytes_left -= name_len;
1885 		if (bytes_left >= 0)
1886 			read_extent_buffer(eb, dest + bytes_left,
1887 					   name_off, name_len);
1888 		if (eb != eb_in) {
1889 			if (!path->skip_locking)
1890 				btrfs_tree_read_unlock(eb);
1891 			free_extent_buffer(eb);
1892 		}
1893 		ret = btrfs_find_item(fs_root, path, parent, 0,
1894 				BTRFS_INODE_REF_KEY, &found_key);
1895 		if (ret > 0)
1896 			ret = -ENOENT;
1897 		if (ret)
1898 			break;
1899 
1900 		next_inum = found_key.offset;
1901 
1902 		/* regular exit ahead */
1903 		if (parent == next_inum)
1904 			break;
1905 
1906 		slot = path->slots[0];
1907 		eb = path->nodes[0];
1908 		/* make sure we can use eb after releasing the path */
1909 		if (eb != eb_in) {
1910 			path->nodes[0] = NULL;
1911 			path->locks[0] = 0;
1912 		}
1913 		btrfs_release_path(path);
1914 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1915 
1916 		name_len = btrfs_inode_ref_name_len(eb, iref);
1917 		name_off = (unsigned long)(iref + 1);
1918 
1919 		parent = next_inum;
1920 		--bytes_left;
1921 		if (bytes_left >= 0)
1922 			dest[bytes_left] = '/';
1923 	}
1924 
1925 	btrfs_release_path(path);
1926 
1927 	if (ret)
1928 		return ERR_PTR(ret);
1929 
1930 	return dest + bytes_left;
1931 }
1932 
1933 /*
1934  * this makes the path point to (logical EXTENT_ITEM *)
1935  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1936  * tree blocks and <0 on error.
1937  */
1938 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1939 			struct btrfs_path *path, struct btrfs_key *found_key,
1940 			u64 *flags_ret)
1941 {
1942 	struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical);
1943 	int ret;
1944 	u64 flags;
1945 	u64 size = 0;
1946 	u32 item_size;
1947 	const struct extent_buffer *eb;
1948 	struct btrfs_extent_item *ei;
1949 	struct btrfs_key key;
1950 
1951 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1952 		key.type = BTRFS_METADATA_ITEM_KEY;
1953 	else
1954 		key.type = BTRFS_EXTENT_ITEM_KEY;
1955 	key.objectid = logical;
1956 	key.offset = (u64)-1;
1957 
1958 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1959 	if (ret < 0)
1960 		return ret;
1961 
1962 	ret = btrfs_previous_extent_item(extent_root, path, 0);
1963 	if (ret) {
1964 		if (ret > 0)
1965 			ret = -ENOENT;
1966 		return ret;
1967 	}
1968 	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1969 	if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1970 		size = fs_info->nodesize;
1971 	else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1972 		size = found_key->offset;
1973 
1974 	if (found_key->objectid > logical ||
1975 	    found_key->objectid + size <= logical) {
1976 		btrfs_debug(fs_info,
1977 			"logical %llu is not within any extent", logical);
1978 		return -ENOENT;
1979 	}
1980 
1981 	eb = path->nodes[0];
1982 	item_size = btrfs_item_size(eb, path->slots[0]);
1983 	BUG_ON(item_size < sizeof(*ei));
1984 
1985 	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1986 	flags = btrfs_extent_flags(eb, ei);
1987 
1988 	btrfs_debug(fs_info,
1989 		"logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
1990 		 logical, logical - found_key->objectid, found_key->objectid,
1991 		 found_key->offset, flags, item_size);
1992 
1993 	WARN_ON(!flags_ret);
1994 	if (flags_ret) {
1995 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1996 			*flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1997 		else if (flags & BTRFS_EXTENT_FLAG_DATA)
1998 			*flags_ret = BTRFS_EXTENT_FLAG_DATA;
1999 		else
2000 			BUG();
2001 		return 0;
2002 	}
2003 
2004 	return -EIO;
2005 }
2006 
2007 /*
2008  * helper function to iterate extent inline refs. ptr must point to a 0 value
2009  * for the first call and may be modified. it is used to track state.
2010  * if more refs exist, 0 is returned and the next call to
2011  * get_extent_inline_ref must pass the modified ptr parameter to get the
2012  * next ref. after the last ref was processed, 1 is returned.
2013  * returns <0 on error
2014  */
2015 static int get_extent_inline_ref(unsigned long *ptr,
2016 				 const struct extent_buffer *eb,
2017 				 const struct btrfs_key *key,
2018 				 const struct btrfs_extent_item *ei,
2019 				 u32 item_size,
2020 				 struct btrfs_extent_inline_ref **out_eiref,
2021 				 int *out_type)
2022 {
2023 	unsigned long end;
2024 	u64 flags;
2025 	struct btrfs_tree_block_info *info;
2026 
2027 	if (!*ptr) {
2028 		/* first call */
2029 		flags = btrfs_extent_flags(eb, ei);
2030 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2031 			if (key->type == BTRFS_METADATA_ITEM_KEY) {
2032 				/* a skinny metadata extent */
2033 				*out_eiref =
2034 				     (struct btrfs_extent_inline_ref *)(ei + 1);
2035 			} else {
2036 				WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
2037 				info = (struct btrfs_tree_block_info *)(ei + 1);
2038 				*out_eiref =
2039 				   (struct btrfs_extent_inline_ref *)(info + 1);
2040 			}
2041 		} else {
2042 			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
2043 		}
2044 		*ptr = (unsigned long)*out_eiref;
2045 		if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
2046 			return -ENOENT;
2047 	}
2048 
2049 	end = (unsigned long)ei + item_size;
2050 	*out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
2051 	*out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
2052 						     BTRFS_REF_TYPE_ANY);
2053 	if (*out_type == BTRFS_REF_TYPE_INVALID)
2054 		return -EUCLEAN;
2055 
2056 	*ptr += btrfs_extent_inline_ref_size(*out_type);
2057 	WARN_ON(*ptr > end);
2058 	if (*ptr == end)
2059 		return 1; /* last */
2060 
2061 	return 0;
2062 }
2063 
2064 /*
2065  * reads the tree block backref for an extent. tree level and root are returned
2066  * through out_level and out_root. ptr must point to a 0 value for the first
2067  * call and may be modified (see get_extent_inline_ref comment).
2068  * returns 0 if data was provided, 1 if there was no more data to provide or
2069  * <0 on error.
2070  */
2071 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
2072 			    struct btrfs_key *key, struct btrfs_extent_item *ei,
2073 			    u32 item_size, u64 *out_root, u8 *out_level)
2074 {
2075 	int ret;
2076 	int type;
2077 	struct btrfs_extent_inline_ref *eiref;
2078 
2079 	if (*ptr == (unsigned long)-1)
2080 		return 1;
2081 
2082 	while (1) {
2083 		ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
2084 					      &eiref, &type);
2085 		if (ret < 0)
2086 			return ret;
2087 
2088 		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
2089 		    type == BTRFS_SHARED_BLOCK_REF_KEY)
2090 			break;
2091 
2092 		if (ret == 1)
2093 			return 1;
2094 	}
2095 
2096 	/* we can treat both ref types equally here */
2097 	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
2098 
2099 	if (key->type == BTRFS_EXTENT_ITEM_KEY) {
2100 		struct btrfs_tree_block_info *info;
2101 
2102 		info = (struct btrfs_tree_block_info *)(ei + 1);
2103 		*out_level = btrfs_tree_block_level(eb, info);
2104 	} else {
2105 		ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
2106 		*out_level = (u8)key->offset;
2107 	}
2108 
2109 	if (ret == 1)
2110 		*ptr = (unsigned long)-1;
2111 
2112 	return 0;
2113 }
2114 
2115 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
2116 			     struct extent_inode_elem *inode_list,
2117 			     u64 root, u64 extent_item_objectid,
2118 			     iterate_extent_inodes_t *iterate, void *ctx)
2119 {
2120 	struct extent_inode_elem *eie;
2121 	int ret = 0;
2122 
2123 	for (eie = inode_list; eie; eie = eie->next) {
2124 		btrfs_debug(fs_info,
2125 			    "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
2126 			    extent_item_objectid, eie->inum,
2127 			    eie->offset, root);
2128 		ret = iterate(eie->inum, eie->offset, root, ctx);
2129 		if (ret) {
2130 			btrfs_debug(fs_info,
2131 				    "stopping iteration for %llu due to ret=%d",
2132 				    extent_item_objectid, ret);
2133 			break;
2134 		}
2135 	}
2136 
2137 	return ret;
2138 }
2139 
2140 /*
2141  * calls iterate() for every inode that references the extent identified by
2142  * the given parameters.
2143  * when the iterator function returns a non-zero value, iteration stops.
2144  */
2145 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
2146 				u64 extent_item_objectid, u64 extent_item_pos,
2147 				int search_commit_root,
2148 				iterate_extent_inodes_t *iterate, void *ctx,
2149 				bool ignore_offset)
2150 {
2151 	int ret;
2152 	struct btrfs_trans_handle *trans = NULL;
2153 	struct ulist *refs = NULL;
2154 	struct ulist *roots = NULL;
2155 	struct ulist_node *ref_node = NULL;
2156 	struct ulist_node *root_node = NULL;
2157 	struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem);
2158 	struct ulist_iterator ref_uiter;
2159 	struct ulist_iterator root_uiter;
2160 
2161 	btrfs_debug(fs_info, "resolving all inodes for extent %llu",
2162 			extent_item_objectid);
2163 
2164 	if (!search_commit_root) {
2165 		trans = btrfs_attach_transaction(fs_info->tree_root);
2166 		if (IS_ERR(trans)) {
2167 			if (PTR_ERR(trans) != -ENOENT &&
2168 			    PTR_ERR(trans) != -EROFS)
2169 				return PTR_ERR(trans);
2170 			trans = NULL;
2171 		}
2172 	}
2173 
2174 	if (trans)
2175 		btrfs_get_tree_mod_seq(fs_info, &seq_elem);
2176 	else
2177 		down_read(&fs_info->commit_root_sem);
2178 
2179 	ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
2180 				   seq_elem.seq, &refs,
2181 				   &extent_item_pos, ignore_offset);
2182 	if (ret)
2183 		goto out;
2184 
2185 	ULIST_ITER_INIT(&ref_uiter);
2186 	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
2187 		ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
2188 						seq_elem.seq, &roots,
2189 						ignore_offset);
2190 		if (ret)
2191 			break;
2192 		ULIST_ITER_INIT(&root_uiter);
2193 		while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
2194 			btrfs_debug(fs_info,
2195 				    "root %llu references leaf %llu, data list %#llx",
2196 				    root_node->val, ref_node->val,
2197 				    ref_node->aux);
2198 			ret = iterate_leaf_refs(fs_info,
2199 						(struct extent_inode_elem *)
2200 						(uintptr_t)ref_node->aux,
2201 						root_node->val,
2202 						extent_item_objectid,
2203 						iterate, ctx);
2204 		}
2205 		ulist_free(roots);
2206 	}
2207 
2208 	free_leaf_list(refs);
2209 out:
2210 	if (trans) {
2211 		btrfs_put_tree_mod_seq(fs_info, &seq_elem);
2212 		btrfs_end_transaction(trans);
2213 	} else {
2214 		up_read(&fs_info->commit_root_sem);
2215 	}
2216 
2217 	return ret;
2218 }
2219 
2220 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
2221 {
2222 	struct btrfs_data_container *inodes = ctx;
2223 	const size_t c = 3 * sizeof(u64);
2224 
2225 	if (inodes->bytes_left >= c) {
2226 		inodes->bytes_left -= c;
2227 		inodes->val[inodes->elem_cnt] = inum;
2228 		inodes->val[inodes->elem_cnt + 1] = offset;
2229 		inodes->val[inodes->elem_cnt + 2] = root;
2230 		inodes->elem_cnt += 3;
2231 	} else {
2232 		inodes->bytes_missing += c - inodes->bytes_left;
2233 		inodes->bytes_left = 0;
2234 		inodes->elem_missed += 3;
2235 	}
2236 
2237 	return 0;
2238 }
2239 
2240 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2241 				struct btrfs_path *path,
2242 				void *ctx, bool ignore_offset)
2243 {
2244 	int ret;
2245 	u64 extent_item_pos;
2246 	u64 flags = 0;
2247 	struct btrfs_key found_key;
2248 	int search_commit_root = path->search_commit_root;
2249 
2250 	ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2251 	btrfs_release_path(path);
2252 	if (ret < 0)
2253 		return ret;
2254 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2255 		return -EINVAL;
2256 
2257 	extent_item_pos = logical - found_key.objectid;
2258 	ret = iterate_extent_inodes(fs_info, found_key.objectid,
2259 					extent_item_pos, search_commit_root,
2260 					build_ino_list, ctx, ignore_offset);
2261 
2262 	return ret;
2263 }
2264 
2265 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2266 			 struct extent_buffer *eb, struct inode_fs_paths *ipath);
2267 
2268 static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath)
2269 {
2270 	int ret = 0;
2271 	int slot;
2272 	u32 cur;
2273 	u32 len;
2274 	u32 name_len;
2275 	u64 parent = 0;
2276 	int found = 0;
2277 	struct btrfs_root *fs_root = ipath->fs_root;
2278 	struct btrfs_path *path = ipath->btrfs_path;
2279 	struct extent_buffer *eb;
2280 	struct btrfs_inode_ref *iref;
2281 	struct btrfs_key found_key;
2282 
2283 	while (!ret) {
2284 		ret = btrfs_find_item(fs_root, path, inum,
2285 				parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2286 				&found_key);
2287 
2288 		if (ret < 0)
2289 			break;
2290 		if (ret) {
2291 			ret = found ? 0 : -ENOENT;
2292 			break;
2293 		}
2294 		++found;
2295 
2296 		parent = found_key.offset;
2297 		slot = path->slots[0];
2298 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
2299 		if (!eb) {
2300 			ret = -ENOMEM;
2301 			break;
2302 		}
2303 		btrfs_release_path(path);
2304 
2305 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2306 
2307 		for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) {
2308 			name_len = btrfs_inode_ref_name_len(eb, iref);
2309 			/* path must be released before calling iterate()! */
2310 			btrfs_debug(fs_root->fs_info,
2311 				"following ref at offset %u for inode %llu in tree %llu",
2312 				cur, found_key.objectid,
2313 				fs_root->root_key.objectid);
2314 			ret = inode_to_path(parent, name_len,
2315 				      (unsigned long)(iref + 1), eb, ipath);
2316 			if (ret)
2317 				break;
2318 			len = sizeof(*iref) + name_len;
2319 			iref = (struct btrfs_inode_ref *)((char *)iref + len);
2320 		}
2321 		free_extent_buffer(eb);
2322 	}
2323 
2324 	btrfs_release_path(path);
2325 
2326 	return ret;
2327 }
2328 
2329 static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath)
2330 {
2331 	int ret;
2332 	int slot;
2333 	u64 offset = 0;
2334 	u64 parent;
2335 	int found = 0;
2336 	struct btrfs_root *fs_root = ipath->fs_root;
2337 	struct btrfs_path *path = ipath->btrfs_path;
2338 	struct extent_buffer *eb;
2339 	struct btrfs_inode_extref *extref;
2340 	u32 item_size;
2341 	u32 cur_offset;
2342 	unsigned long ptr;
2343 
2344 	while (1) {
2345 		ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2346 					    &offset);
2347 		if (ret < 0)
2348 			break;
2349 		if (ret) {
2350 			ret = found ? 0 : -ENOENT;
2351 			break;
2352 		}
2353 		++found;
2354 
2355 		slot = path->slots[0];
2356 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
2357 		if (!eb) {
2358 			ret = -ENOMEM;
2359 			break;
2360 		}
2361 		btrfs_release_path(path);
2362 
2363 		item_size = btrfs_item_size(eb, slot);
2364 		ptr = btrfs_item_ptr_offset(eb, slot);
2365 		cur_offset = 0;
2366 
2367 		while (cur_offset < item_size) {
2368 			u32 name_len;
2369 
2370 			extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2371 			parent = btrfs_inode_extref_parent(eb, extref);
2372 			name_len = btrfs_inode_extref_name_len(eb, extref);
2373 			ret = inode_to_path(parent, name_len,
2374 				      (unsigned long)&extref->name, eb, ipath);
2375 			if (ret)
2376 				break;
2377 
2378 			cur_offset += btrfs_inode_extref_name_len(eb, extref);
2379 			cur_offset += sizeof(*extref);
2380 		}
2381 		free_extent_buffer(eb);
2382 
2383 		offset++;
2384 	}
2385 
2386 	btrfs_release_path(path);
2387 
2388 	return ret;
2389 }
2390 
2391 /*
2392  * returns 0 if the path could be dumped (probably truncated)
2393  * returns <0 in case of an error
2394  */
2395 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2396 			 struct extent_buffer *eb, struct inode_fs_paths *ipath)
2397 {
2398 	char *fspath;
2399 	char *fspath_min;
2400 	int i = ipath->fspath->elem_cnt;
2401 	const int s_ptr = sizeof(char *);
2402 	u32 bytes_left;
2403 
2404 	bytes_left = ipath->fspath->bytes_left > s_ptr ?
2405 					ipath->fspath->bytes_left - s_ptr : 0;
2406 
2407 	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2408 	fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2409 				   name_off, eb, inum, fspath_min, bytes_left);
2410 	if (IS_ERR(fspath))
2411 		return PTR_ERR(fspath);
2412 
2413 	if (fspath > fspath_min) {
2414 		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2415 		++ipath->fspath->elem_cnt;
2416 		ipath->fspath->bytes_left = fspath - fspath_min;
2417 	} else {
2418 		++ipath->fspath->elem_missed;
2419 		ipath->fspath->bytes_missing += fspath_min - fspath;
2420 		ipath->fspath->bytes_left = 0;
2421 	}
2422 
2423 	return 0;
2424 }
2425 
2426 /*
2427  * this dumps all file system paths to the inode into the ipath struct, provided
2428  * is has been created large enough. each path is zero-terminated and accessed
2429  * from ipath->fspath->val[i].
2430  * when it returns, there are ipath->fspath->elem_cnt number of paths available
2431  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2432  * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2433  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2434  * have been needed to return all paths.
2435  */
2436 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2437 {
2438 	int ret;
2439 	int found_refs = 0;
2440 
2441 	ret = iterate_inode_refs(inum, ipath);
2442 	if (!ret)
2443 		++found_refs;
2444 	else if (ret != -ENOENT)
2445 		return ret;
2446 
2447 	ret = iterate_inode_extrefs(inum, ipath);
2448 	if (ret == -ENOENT && found_refs)
2449 		return 0;
2450 
2451 	return ret;
2452 }
2453 
2454 struct btrfs_data_container *init_data_container(u32 total_bytes)
2455 {
2456 	struct btrfs_data_container *data;
2457 	size_t alloc_bytes;
2458 
2459 	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2460 	data = kvmalloc(alloc_bytes, GFP_KERNEL);
2461 	if (!data)
2462 		return ERR_PTR(-ENOMEM);
2463 
2464 	if (total_bytes >= sizeof(*data)) {
2465 		data->bytes_left = total_bytes - sizeof(*data);
2466 		data->bytes_missing = 0;
2467 	} else {
2468 		data->bytes_missing = sizeof(*data) - total_bytes;
2469 		data->bytes_left = 0;
2470 	}
2471 
2472 	data->elem_cnt = 0;
2473 	data->elem_missed = 0;
2474 
2475 	return data;
2476 }
2477 
2478 /*
2479  * allocates space to return multiple file system paths for an inode.
2480  * total_bytes to allocate are passed, note that space usable for actual path
2481  * information will be total_bytes - sizeof(struct inode_fs_paths).
2482  * the returned pointer must be freed with free_ipath() in the end.
2483  */
2484 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2485 					struct btrfs_path *path)
2486 {
2487 	struct inode_fs_paths *ifp;
2488 	struct btrfs_data_container *fspath;
2489 
2490 	fspath = init_data_container(total_bytes);
2491 	if (IS_ERR(fspath))
2492 		return ERR_CAST(fspath);
2493 
2494 	ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2495 	if (!ifp) {
2496 		kvfree(fspath);
2497 		return ERR_PTR(-ENOMEM);
2498 	}
2499 
2500 	ifp->btrfs_path = path;
2501 	ifp->fspath = fspath;
2502 	ifp->fs_root = fs_root;
2503 
2504 	return ifp;
2505 }
2506 
2507 void free_ipath(struct inode_fs_paths *ipath)
2508 {
2509 	if (!ipath)
2510 		return;
2511 	kvfree(ipath->fspath);
2512 	kfree(ipath);
2513 }
2514 
2515 struct btrfs_backref_iter *btrfs_backref_iter_alloc(
2516 		struct btrfs_fs_info *fs_info, gfp_t gfp_flag)
2517 {
2518 	struct btrfs_backref_iter *ret;
2519 
2520 	ret = kzalloc(sizeof(*ret), gfp_flag);
2521 	if (!ret)
2522 		return NULL;
2523 
2524 	ret->path = btrfs_alloc_path();
2525 	if (!ret->path) {
2526 		kfree(ret);
2527 		return NULL;
2528 	}
2529 
2530 	/* Current backref iterator only supports iteration in commit root */
2531 	ret->path->search_commit_root = 1;
2532 	ret->path->skip_locking = 1;
2533 	ret->fs_info = fs_info;
2534 
2535 	return ret;
2536 }
2537 
2538 int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2539 {
2540 	struct btrfs_fs_info *fs_info = iter->fs_info;
2541 	struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2542 	struct btrfs_path *path = iter->path;
2543 	struct btrfs_extent_item *ei;
2544 	struct btrfs_key key;
2545 	int ret;
2546 
2547 	key.objectid = bytenr;
2548 	key.type = BTRFS_METADATA_ITEM_KEY;
2549 	key.offset = (u64)-1;
2550 	iter->bytenr = bytenr;
2551 
2552 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2553 	if (ret < 0)
2554 		return ret;
2555 	if (ret == 0) {
2556 		ret = -EUCLEAN;
2557 		goto release;
2558 	}
2559 	if (path->slots[0] == 0) {
2560 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2561 		ret = -EUCLEAN;
2562 		goto release;
2563 	}
2564 	path->slots[0]--;
2565 
2566 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2567 	if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2568 	     key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2569 		ret = -ENOENT;
2570 		goto release;
2571 	}
2572 	memcpy(&iter->cur_key, &key, sizeof(key));
2573 	iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2574 						    path->slots[0]);
2575 	iter->end_ptr = (u32)(iter->item_ptr +
2576 			btrfs_item_size(path->nodes[0], path->slots[0]));
2577 	ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2578 			    struct btrfs_extent_item);
2579 
2580 	/*
2581 	 * Only support iteration on tree backref yet.
2582 	 *
2583 	 * This is an extra precaution for non skinny-metadata, where
2584 	 * EXTENT_ITEM is also used for tree blocks, that we can only use
2585 	 * extent flags to determine if it's a tree block.
2586 	 */
2587 	if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2588 		ret = -ENOTSUPP;
2589 		goto release;
2590 	}
2591 	iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2592 
2593 	/* If there is no inline backref, go search for keyed backref */
2594 	if (iter->cur_ptr >= iter->end_ptr) {
2595 		ret = btrfs_next_item(extent_root, path);
2596 
2597 		/* No inline nor keyed ref */
2598 		if (ret > 0) {
2599 			ret = -ENOENT;
2600 			goto release;
2601 		}
2602 		if (ret < 0)
2603 			goto release;
2604 
2605 		btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2606 				path->slots[0]);
2607 		if (iter->cur_key.objectid != bytenr ||
2608 		    (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2609 		     iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2610 			ret = -ENOENT;
2611 			goto release;
2612 		}
2613 		iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2614 							   path->slots[0]);
2615 		iter->item_ptr = iter->cur_ptr;
2616 		iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size(
2617 				      path->nodes[0], path->slots[0]));
2618 	}
2619 
2620 	return 0;
2621 release:
2622 	btrfs_backref_iter_release(iter);
2623 	return ret;
2624 }
2625 
2626 /*
2627  * Go to the next backref item of current bytenr, can be either inlined or
2628  * keyed.
2629  *
2630  * Caller needs to check whether it's inline ref or not by iter->cur_key.
2631  *
2632  * Return 0 if we get next backref without problem.
2633  * Return >0 if there is no extra backref for this bytenr.
2634  * Return <0 if there is something wrong happened.
2635  */
2636 int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2637 {
2638 	struct extent_buffer *eb = btrfs_backref_get_eb(iter);
2639 	struct btrfs_root *extent_root;
2640 	struct btrfs_path *path = iter->path;
2641 	struct btrfs_extent_inline_ref *iref;
2642 	int ret;
2643 	u32 size;
2644 
2645 	if (btrfs_backref_iter_is_inline_ref(iter)) {
2646 		/* We're still inside the inline refs */
2647 		ASSERT(iter->cur_ptr < iter->end_ptr);
2648 
2649 		if (btrfs_backref_has_tree_block_info(iter)) {
2650 			/* First tree block info */
2651 			size = sizeof(struct btrfs_tree_block_info);
2652 		} else {
2653 			/* Use inline ref type to determine the size */
2654 			int type;
2655 
2656 			iref = (struct btrfs_extent_inline_ref *)
2657 				((unsigned long)iter->cur_ptr);
2658 			type = btrfs_extent_inline_ref_type(eb, iref);
2659 
2660 			size = btrfs_extent_inline_ref_size(type);
2661 		}
2662 		iter->cur_ptr += size;
2663 		if (iter->cur_ptr < iter->end_ptr)
2664 			return 0;
2665 
2666 		/* All inline items iterated, fall through */
2667 	}
2668 
2669 	/* We're at keyed items, there is no inline item, go to the next one */
2670 	extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr);
2671 	ret = btrfs_next_item(extent_root, iter->path);
2672 	if (ret)
2673 		return ret;
2674 
2675 	btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
2676 	if (iter->cur_key.objectid != iter->bytenr ||
2677 	    (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2678 	     iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
2679 		return 1;
2680 	iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2681 					path->slots[0]);
2682 	iter->cur_ptr = iter->item_ptr;
2683 	iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0],
2684 						path->slots[0]);
2685 	return 0;
2686 }
2687 
2688 void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
2689 			      struct btrfs_backref_cache *cache, int is_reloc)
2690 {
2691 	int i;
2692 
2693 	cache->rb_root = RB_ROOT;
2694 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2695 		INIT_LIST_HEAD(&cache->pending[i]);
2696 	INIT_LIST_HEAD(&cache->changed);
2697 	INIT_LIST_HEAD(&cache->detached);
2698 	INIT_LIST_HEAD(&cache->leaves);
2699 	INIT_LIST_HEAD(&cache->pending_edge);
2700 	INIT_LIST_HEAD(&cache->useless_node);
2701 	cache->fs_info = fs_info;
2702 	cache->is_reloc = is_reloc;
2703 }
2704 
2705 struct btrfs_backref_node *btrfs_backref_alloc_node(
2706 		struct btrfs_backref_cache *cache, u64 bytenr, int level)
2707 {
2708 	struct btrfs_backref_node *node;
2709 
2710 	ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
2711 	node = kzalloc(sizeof(*node), GFP_NOFS);
2712 	if (!node)
2713 		return node;
2714 
2715 	INIT_LIST_HEAD(&node->list);
2716 	INIT_LIST_HEAD(&node->upper);
2717 	INIT_LIST_HEAD(&node->lower);
2718 	RB_CLEAR_NODE(&node->rb_node);
2719 	cache->nr_nodes++;
2720 	node->level = level;
2721 	node->bytenr = bytenr;
2722 
2723 	return node;
2724 }
2725 
2726 struct btrfs_backref_edge *btrfs_backref_alloc_edge(
2727 		struct btrfs_backref_cache *cache)
2728 {
2729 	struct btrfs_backref_edge *edge;
2730 
2731 	edge = kzalloc(sizeof(*edge), GFP_NOFS);
2732 	if (edge)
2733 		cache->nr_edges++;
2734 	return edge;
2735 }
2736 
2737 /*
2738  * Drop the backref node from cache, also cleaning up all its
2739  * upper edges and any uncached nodes in the path.
2740  *
2741  * This cleanup happens bottom up, thus the node should either
2742  * be the lowest node in the cache or a detached node.
2743  */
2744 void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
2745 				struct btrfs_backref_node *node)
2746 {
2747 	struct btrfs_backref_node *upper;
2748 	struct btrfs_backref_edge *edge;
2749 
2750 	if (!node)
2751 		return;
2752 
2753 	BUG_ON(!node->lowest && !node->detached);
2754 	while (!list_empty(&node->upper)) {
2755 		edge = list_entry(node->upper.next, struct btrfs_backref_edge,
2756 				  list[LOWER]);
2757 		upper = edge->node[UPPER];
2758 		list_del(&edge->list[LOWER]);
2759 		list_del(&edge->list[UPPER]);
2760 		btrfs_backref_free_edge(cache, edge);
2761 
2762 		/*
2763 		 * Add the node to leaf node list if no other child block
2764 		 * cached.
2765 		 */
2766 		if (list_empty(&upper->lower)) {
2767 			list_add_tail(&upper->lower, &cache->leaves);
2768 			upper->lowest = 1;
2769 		}
2770 	}
2771 
2772 	btrfs_backref_drop_node(cache, node);
2773 }
2774 
2775 /*
2776  * Release all nodes/edges from current cache
2777  */
2778 void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
2779 {
2780 	struct btrfs_backref_node *node;
2781 	int i;
2782 
2783 	while (!list_empty(&cache->detached)) {
2784 		node = list_entry(cache->detached.next,
2785 				  struct btrfs_backref_node, list);
2786 		btrfs_backref_cleanup_node(cache, node);
2787 	}
2788 
2789 	while (!list_empty(&cache->leaves)) {
2790 		node = list_entry(cache->leaves.next,
2791 				  struct btrfs_backref_node, lower);
2792 		btrfs_backref_cleanup_node(cache, node);
2793 	}
2794 
2795 	cache->last_trans = 0;
2796 
2797 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2798 		ASSERT(list_empty(&cache->pending[i]));
2799 	ASSERT(list_empty(&cache->pending_edge));
2800 	ASSERT(list_empty(&cache->useless_node));
2801 	ASSERT(list_empty(&cache->changed));
2802 	ASSERT(list_empty(&cache->detached));
2803 	ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
2804 	ASSERT(!cache->nr_nodes);
2805 	ASSERT(!cache->nr_edges);
2806 }
2807 
2808 /*
2809  * Handle direct tree backref
2810  *
2811  * Direct tree backref means, the backref item shows its parent bytenr
2812  * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
2813  *
2814  * @ref_key:	The converted backref key.
2815  *		For keyed backref, it's the item key.
2816  *		For inlined backref, objectid is the bytenr,
2817  *		type is btrfs_inline_ref_type, offset is
2818  *		btrfs_inline_ref_offset.
2819  */
2820 static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
2821 				      struct btrfs_key *ref_key,
2822 				      struct btrfs_backref_node *cur)
2823 {
2824 	struct btrfs_backref_edge *edge;
2825 	struct btrfs_backref_node *upper;
2826 	struct rb_node *rb_node;
2827 
2828 	ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
2829 
2830 	/* Only reloc root uses backref pointing to itself */
2831 	if (ref_key->objectid == ref_key->offset) {
2832 		struct btrfs_root *root;
2833 
2834 		cur->is_reloc_root = 1;
2835 		/* Only reloc backref cache cares about a specific root */
2836 		if (cache->is_reloc) {
2837 			root = find_reloc_root(cache->fs_info, cur->bytenr);
2838 			if (!root)
2839 				return -ENOENT;
2840 			cur->root = root;
2841 		} else {
2842 			/*
2843 			 * For generic purpose backref cache, reloc root node
2844 			 * is useless.
2845 			 */
2846 			list_add(&cur->list, &cache->useless_node);
2847 		}
2848 		return 0;
2849 	}
2850 
2851 	edge = btrfs_backref_alloc_edge(cache);
2852 	if (!edge)
2853 		return -ENOMEM;
2854 
2855 	rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
2856 	if (!rb_node) {
2857 		/* Parent node not yet cached */
2858 		upper = btrfs_backref_alloc_node(cache, ref_key->offset,
2859 					   cur->level + 1);
2860 		if (!upper) {
2861 			btrfs_backref_free_edge(cache, edge);
2862 			return -ENOMEM;
2863 		}
2864 
2865 		/*
2866 		 *  Backrefs for the upper level block isn't cached, add the
2867 		 *  block to pending list
2868 		 */
2869 		list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2870 	} else {
2871 		/* Parent node already cached */
2872 		upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
2873 		ASSERT(upper->checked);
2874 		INIT_LIST_HEAD(&edge->list[UPPER]);
2875 	}
2876 	btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
2877 	return 0;
2878 }
2879 
2880 /*
2881  * Handle indirect tree backref
2882  *
2883  * Indirect tree backref means, we only know which tree the node belongs to.
2884  * We still need to do a tree search to find out the parents. This is for
2885  * TREE_BLOCK_REF backref (keyed or inlined).
2886  *
2887  * @ref_key:	The same as @ref_key in  handle_direct_tree_backref()
2888  * @tree_key:	The first key of this tree block.
2889  * @path:	A clean (released) path, to avoid allocating path every time
2890  *		the function get called.
2891  */
2892 static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache,
2893 					struct btrfs_path *path,
2894 					struct btrfs_key *ref_key,
2895 					struct btrfs_key *tree_key,
2896 					struct btrfs_backref_node *cur)
2897 {
2898 	struct btrfs_fs_info *fs_info = cache->fs_info;
2899 	struct btrfs_backref_node *upper;
2900 	struct btrfs_backref_node *lower;
2901 	struct btrfs_backref_edge *edge;
2902 	struct extent_buffer *eb;
2903 	struct btrfs_root *root;
2904 	struct rb_node *rb_node;
2905 	int level;
2906 	bool need_check = true;
2907 	int ret;
2908 
2909 	root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
2910 	if (IS_ERR(root))
2911 		return PTR_ERR(root);
2912 	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2913 		cur->cowonly = 1;
2914 
2915 	if (btrfs_root_level(&root->root_item) == cur->level) {
2916 		/* Tree root */
2917 		ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
2918 		/*
2919 		 * For reloc backref cache, we may ignore reloc root.  But for
2920 		 * general purpose backref cache, we can't rely on
2921 		 * btrfs_should_ignore_reloc_root() as it may conflict with
2922 		 * current running relocation and lead to missing root.
2923 		 *
2924 		 * For general purpose backref cache, reloc root detection is
2925 		 * completely relying on direct backref (key->offset is parent
2926 		 * bytenr), thus only do such check for reloc cache.
2927 		 */
2928 		if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
2929 			btrfs_put_root(root);
2930 			list_add(&cur->list, &cache->useless_node);
2931 		} else {
2932 			cur->root = root;
2933 		}
2934 		return 0;
2935 	}
2936 
2937 	level = cur->level + 1;
2938 
2939 	/* Search the tree to find parent blocks referring to the block */
2940 	path->search_commit_root = 1;
2941 	path->skip_locking = 1;
2942 	path->lowest_level = level;
2943 	ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
2944 	path->lowest_level = 0;
2945 	if (ret < 0) {
2946 		btrfs_put_root(root);
2947 		return ret;
2948 	}
2949 	if (ret > 0 && path->slots[level] > 0)
2950 		path->slots[level]--;
2951 
2952 	eb = path->nodes[level];
2953 	if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
2954 		btrfs_err(fs_info,
2955 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
2956 			  cur->bytenr, level - 1, root->root_key.objectid,
2957 			  tree_key->objectid, tree_key->type, tree_key->offset);
2958 		btrfs_put_root(root);
2959 		ret = -ENOENT;
2960 		goto out;
2961 	}
2962 	lower = cur;
2963 
2964 	/* Add all nodes and edges in the path */
2965 	for (; level < BTRFS_MAX_LEVEL; level++) {
2966 		if (!path->nodes[level]) {
2967 			ASSERT(btrfs_root_bytenr(&root->root_item) ==
2968 			       lower->bytenr);
2969 			/* Same as previous should_ignore_reloc_root() call */
2970 			if (btrfs_should_ignore_reloc_root(root) &&
2971 			    cache->is_reloc) {
2972 				btrfs_put_root(root);
2973 				list_add(&lower->list, &cache->useless_node);
2974 			} else {
2975 				lower->root = root;
2976 			}
2977 			break;
2978 		}
2979 
2980 		edge = btrfs_backref_alloc_edge(cache);
2981 		if (!edge) {
2982 			btrfs_put_root(root);
2983 			ret = -ENOMEM;
2984 			goto out;
2985 		}
2986 
2987 		eb = path->nodes[level];
2988 		rb_node = rb_simple_search(&cache->rb_root, eb->start);
2989 		if (!rb_node) {
2990 			upper = btrfs_backref_alloc_node(cache, eb->start,
2991 							 lower->level + 1);
2992 			if (!upper) {
2993 				btrfs_put_root(root);
2994 				btrfs_backref_free_edge(cache, edge);
2995 				ret = -ENOMEM;
2996 				goto out;
2997 			}
2998 			upper->owner = btrfs_header_owner(eb);
2999 			if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3000 				upper->cowonly = 1;
3001 
3002 			/*
3003 			 * If we know the block isn't shared we can avoid
3004 			 * checking its backrefs.
3005 			 */
3006 			if (btrfs_block_can_be_shared(root, eb))
3007 				upper->checked = 0;
3008 			else
3009 				upper->checked = 1;
3010 
3011 			/*
3012 			 * Add the block to pending list if we need to check its
3013 			 * backrefs, we only do this once while walking up a
3014 			 * tree as we will catch anything else later on.
3015 			 */
3016 			if (!upper->checked && need_check) {
3017 				need_check = false;
3018 				list_add_tail(&edge->list[UPPER],
3019 					      &cache->pending_edge);
3020 			} else {
3021 				if (upper->checked)
3022 					need_check = true;
3023 				INIT_LIST_HEAD(&edge->list[UPPER]);
3024 			}
3025 		} else {
3026 			upper = rb_entry(rb_node, struct btrfs_backref_node,
3027 					 rb_node);
3028 			ASSERT(upper->checked);
3029 			INIT_LIST_HEAD(&edge->list[UPPER]);
3030 			if (!upper->owner)
3031 				upper->owner = btrfs_header_owner(eb);
3032 		}
3033 		btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
3034 
3035 		if (rb_node) {
3036 			btrfs_put_root(root);
3037 			break;
3038 		}
3039 		lower = upper;
3040 		upper = NULL;
3041 	}
3042 out:
3043 	btrfs_release_path(path);
3044 	return ret;
3045 }
3046 
3047 /*
3048  * Add backref node @cur into @cache.
3049  *
3050  * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
3051  *	 links aren't yet bi-directional. Needs to finish such links.
3052  *	 Use btrfs_backref_finish_upper_links() to finish such linkage.
3053  *
3054  * @path:	Released path for indirect tree backref lookup
3055  * @iter:	Released backref iter for extent tree search
3056  * @node_key:	The first key of the tree block
3057  */
3058 int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache,
3059 				struct btrfs_path *path,
3060 				struct btrfs_backref_iter *iter,
3061 				struct btrfs_key *node_key,
3062 				struct btrfs_backref_node *cur)
3063 {
3064 	struct btrfs_fs_info *fs_info = cache->fs_info;
3065 	struct btrfs_backref_edge *edge;
3066 	struct btrfs_backref_node *exist;
3067 	int ret;
3068 
3069 	ret = btrfs_backref_iter_start(iter, cur->bytenr);
3070 	if (ret < 0)
3071 		return ret;
3072 	/*
3073 	 * We skip the first btrfs_tree_block_info, as we don't use the key
3074 	 * stored in it, but fetch it from the tree block
3075 	 */
3076 	if (btrfs_backref_has_tree_block_info(iter)) {
3077 		ret = btrfs_backref_iter_next(iter);
3078 		if (ret < 0)
3079 			goto out;
3080 		/* No extra backref? This means the tree block is corrupted */
3081 		if (ret > 0) {
3082 			ret = -EUCLEAN;
3083 			goto out;
3084 		}
3085 	}
3086 	WARN_ON(cur->checked);
3087 	if (!list_empty(&cur->upper)) {
3088 		/*
3089 		 * The backref was added previously when processing backref of
3090 		 * type BTRFS_TREE_BLOCK_REF_KEY
3091 		 */
3092 		ASSERT(list_is_singular(&cur->upper));
3093 		edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
3094 				  list[LOWER]);
3095 		ASSERT(list_empty(&edge->list[UPPER]));
3096 		exist = edge->node[UPPER];
3097 		/*
3098 		 * Add the upper level block to pending list if we need check
3099 		 * its backrefs
3100 		 */
3101 		if (!exist->checked)
3102 			list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3103 	} else {
3104 		exist = NULL;
3105 	}
3106 
3107 	for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
3108 		struct extent_buffer *eb;
3109 		struct btrfs_key key;
3110 		int type;
3111 
3112 		cond_resched();
3113 		eb = btrfs_backref_get_eb(iter);
3114 
3115 		key.objectid = iter->bytenr;
3116 		if (btrfs_backref_iter_is_inline_ref(iter)) {
3117 			struct btrfs_extent_inline_ref *iref;
3118 
3119 			/* Update key for inline backref */
3120 			iref = (struct btrfs_extent_inline_ref *)
3121 				((unsigned long)iter->cur_ptr);
3122 			type = btrfs_get_extent_inline_ref_type(eb, iref,
3123 							BTRFS_REF_TYPE_BLOCK);
3124 			if (type == BTRFS_REF_TYPE_INVALID) {
3125 				ret = -EUCLEAN;
3126 				goto out;
3127 			}
3128 			key.type = type;
3129 			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3130 		} else {
3131 			key.type = iter->cur_key.type;
3132 			key.offset = iter->cur_key.offset;
3133 		}
3134 
3135 		/*
3136 		 * Parent node found and matches current inline ref, no need to
3137 		 * rebuild this node for this inline ref
3138 		 */
3139 		if (exist &&
3140 		    ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
3141 		      exist->owner == key.offset) ||
3142 		     (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
3143 		      exist->bytenr == key.offset))) {
3144 			exist = NULL;
3145 			continue;
3146 		}
3147 
3148 		/* SHARED_BLOCK_REF means key.offset is the parent bytenr */
3149 		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3150 			ret = handle_direct_tree_backref(cache, &key, cur);
3151 			if (ret < 0)
3152 				goto out;
3153 			continue;
3154 		} else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3155 			ret = -EINVAL;
3156 			btrfs_print_v0_err(fs_info);
3157 			btrfs_handle_fs_error(fs_info, ret, NULL);
3158 			goto out;
3159 		} else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
3160 			continue;
3161 		}
3162 
3163 		/*
3164 		 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
3165 		 * means the root objectid. We need to search the tree to get
3166 		 * its parent bytenr.
3167 		 */
3168 		ret = handle_indirect_tree_backref(cache, path, &key, node_key,
3169 						   cur);
3170 		if (ret < 0)
3171 			goto out;
3172 	}
3173 	ret = 0;
3174 	cur->checked = 1;
3175 	WARN_ON(exist);
3176 out:
3177 	btrfs_backref_iter_release(iter);
3178 	return ret;
3179 }
3180 
3181 /*
3182  * Finish the upwards linkage created by btrfs_backref_add_tree_node()
3183  */
3184 int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
3185 				     struct btrfs_backref_node *start)
3186 {
3187 	struct list_head *useless_node = &cache->useless_node;
3188 	struct btrfs_backref_edge *edge;
3189 	struct rb_node *rb_node;
3190 	LIST_HEAD(pending_edge);
3191 
3192 	ASSERT(start->checked);
3193 
3194 	/* Insert this node to cache if it's not COW-only */
3195 	if (!start->cowonly) {
3196 		rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
3197 					   &start->rb_node);
3198 		if (rb_node)
3199 			btrfs_backref_panic(cache->fs_info, start->bytenr,
3200 					    -EEXIST);
3201 		list_add_tail(&start->lower, &cache->leaves);
3202 	}
3203 
3204 	/*
3205 	 * Use breadth first search to iterate all related edges.
3206 	 *
3207 	 * The starting points are all the edges of this node
3208 	 */
3209 	list_for_each_entry(edge, &start->upper, list[LOWER])
3210 		list_add_tail(&edge->list[UPPER], &pending_edge);
3211 
3212 	while (!list_empty(&pending_edge)) {
3213 		struct btrfs_backref_node *upper;
3214 		struct btrfs_backref_node *lower;
3215 
3216 		edge = list_first_entry(&pending_edge,
3217 				struct btrfs_backref_edge, list[UPPER]);
3218 		list_del_init(&edge->list[UPPER]);
3219 		upper = edge->node[UPPER];
3220 		lower = edge->node[LOWER];
3221 
3222 		/* Parent is detached, no need to keep any edges */
3223 		if (upper->detached) {
3224 			list_del(&edge->list[LOWER]);
3225 			btrfs_backref_free_edge(cache, edge);
3226 
3227 			/* Lower node is orphan, queue for cleanup */
3228 			if (list_empty(&lower->upper))
3229 				list_add(&lower->list, useless_node);
3230 			continue;
3231 		}
3232 
3233 		/*
3234 		 * All new nodes added in current build_backref_tree() haven't
3235 		 * been linked to the cache rb tree.
3236 		 * So if we have upper->rb_node populated, this means a cache
3237 		 * hit. We only need to link the edge, as @upper and all its
3238 		 * parents have already been linked.
3239 		 */
3240 		if (!RB_EMPTY_NODE(&upper->rb_node)) {
3241 			if (upper->lowest) {
3242 				list_del_init(&upper->lower);
3243 				upper->lowest = 0;
3244 			}
3245 
3246 			list_add_tail(&edge->list[UPPER], &upper->lower);
3247 			continue;
3248 		}
3249 
3250 		/* Sanity check, we shouldn't have any unchecked nodes */
3251 		if (!upper->checked) {
3252 			ASSERT(0);
3253 			return -EUCLEAN;
3254 		}
3255 
3256 		/* Sanity check, COW-only node has non-COW-only parent */
3257 		if (start->cowonly != upper->cowonly) {
3258 			ASSERT(0);
3259 			return -EUCLEAN;
3260 		}
3261 
3262 		/* Only cache non-COW-only (subvolume trees) tree blocks */
3263 		if (!upper->cowonly) {
3264 			rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3265 						   &upper->rb_node);
3266 			if (rb_node) {
3267 				btrfs_backref_panic(cache->fs_info,
3268 						upper->bytenr, -EEXIST);
3269 				return -EUCLEAN;
3270 			}
3271 		}
3272 
3273 		list_add_tail(&edge->list[UPPER], &upper->lower);
3274 
3275 		/*
3276 		 * Also queue all the parent edges of this uncached node
3277 		 * to finish the upper linkage
3278 		 */
3279 		list_for_each_entry(edge, &upper->upper, list[LOWER])
3280 			list_add_tail(&edge->list[UPPER], &pending_edge);
3281 	}
3282 	return 0;
3283 }
3284 
3285 void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3286 				 struct btrfs_backref_node *node)
3287 {
3288 	struct btrfs_backref_node *lower;
3289 	struct btrfs_backref_node *upper;
3290 	struct btrfs_backref_edge *edge;
3291 
3292 	while (!list_empty(&cache->useless_node)) {
3293 		lower = list_first_entry(&cache->useless_node,
3294 				   struct btrfs_backref_node, list);
3295 		list_del_init(&lower->list);
3296 	}
3297 	while (!list_empty(&cache->pending_edge)) {
3298 		edge = list_first_entry(&cache->pending_edge,
3299 				struct btrfs_backref_edge, list[UPPER]);
3300 		list_del(&edge->list[UPPER]);
3301 		list_del(&edge->list[LOWER]);
3302 		lower = edge->node[LOWER];
3303 		upper = edge->node[UPPER];
3304 		btrfs_backref_free_edge(cache, edge);
3305 
3306 		/*
3307 		 * Lower is no longer linked to any upper backref nodes and
3308 		 * isn't in the cache, we can free it ourselves.
3309 		 */
3310 		if (list_empty(&lower->upper) &&
3311 		    RB_EMPTY_NODE(&lower->rb_node))
3312 			list_add(&lower->list, &cache->useless_node);
3313 
3314 		if (!RB_EMPTY_NODE(&upper->rb_node))
3315 			continue;
3316 
3317 		/* Add this guy's upper edges to the list to process */
3318 		list_for_each_entry(edge, &upper->upper, list[LOWER])
3319 			list_add_tail(&edge->list[UPPER],
3320 				      &cache->pending_edge);
3321 		if (list_empty(&upper->upper))
3322 			list_add(&upper->list, &cache->useless_node);
3323 	}
3324 
3325 	while (!list_empty(&cache->useless_node)) {
3326 		lower = list_first_entry(&cache->useless_node,
3327 				   struct btrfs_backref_node, list);
3328 		list_del_init(&lower->list);
3329 		if (lower == node)
3330 			node = NULL;
3331 		btrfs_backref_drop_node(cache, lower);
3332 	}
3333 
3334 	btrfs_backref_cleanup_node(cache, node);
3335 	ASSERT(list_empty(&cache->useless_node) &&
3336 	       list_empty(&cache->pending_edge));
3337 }
3338