xref: /openbmc/linux/fs/btrfs/backref.c (revision 3e30a927)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 STRATO.  All rights reserved.
4  */
5 
6 #include <linux/mm.h>
7 #include <linux/rbtree.h>
8 #include <trace/events/btrfs.h>
9 #include "ctree.h"
10 #include "disk-io.h"
11 #include "backref.h"
12 #include "ulist.h"
13 #include "transaction.h"
14 #include "delayed-ref.h"
15 #include "locking.h"
16 
17 /* Just an arbitrary number so we can be sure this happened */
18 #define BACKREF_FOUND_SHARED 6
19 
20 struct extent_inode_elem {
21 	u64 inum;
22 	u64 offset;
23 	struct extent_inode_elem *next;
24 };
25 
26 static int check_extent_in_eb(const struct btrfs_key *key,
27 			      const struct extent_buffer *eb,
28 			      const struct btrfs_file_extent_item *fi,
29 			      u64 extent_item_pos,
30 			      struct extent_inode_elem **eie,
31 			      bool ignore_offset)
32 {
33 	u64 offset = 0;
34 	struct extent_inode_elem *e;
35 
36 	if (!ignore_offset &&
37 	    !btrfs_file_extent_compression(eb, fi) &&
38 	    !btrfs_file_extent_encryption(eb, fi) &&
39 	    !btrfs_file_extent_other_encoding(eb, fi)) {
40 		u64 data_offset;
41 		u64 data_len;
42 
43 		data_offset = btrfs_file_extent_offset(eb, fi);
44 		data_len = btrfs_file_extent_num_bytes(eb, fi);
45 
46 		if (extent_item_pos < data_offset ||
47 		    extent_item_pos >= data_offset + data_len)
48 			return 1;
49 		offset = extent_item_pos - data_offset;
50 	}
51 
52 	e = kmalloc(sizeof(*e), GFP_NOFS);
53 	if (!e)
54 		return -ENOMEM;
55 
56 	e->next = *eie;
57 	e->inum = key->objectid;
58 	e->offset = key->offset + offset;
59 	*eie = e;
60 
61 	return 0;
62 }
63 
64 static void free_inode_elem_list(struct extent_inode_elem *eie)
65 {
66 	struct extent_inode_elem *eie_next;
67 
68 	for (; eie; eie = eie_next) {
69 		eie_next = eie->next;
70 		kfree(eie);
71 	}
72 }
73 
74 static int find_extent_in_eb(const struct extent_buffer *eb,
75 			     u64 wanted_disk_byte, u64 extent_item_pos,
76 			     struct extent_inode_elem **eie,
77 			     bool ignore_offset)
78 {
79 	u64 disk_byte;
80 	struct btrfs_key key;
81 	struct btrfs_file_extent_item *fi;
82 	int slot;
83 	int nritems;
84 	int extent_type;
85 	int ret;
86 
87 	/*
88 	 * from the shared data ref, we only have the leaf but we need
89 	 * the key. thus, we must look into all items and see that we
90 	 * find one (some) with a reference to our extent item.
91 	 */
92 	nritems = btrfs_header_nritems(eb);
93 	for (slot = 0; slot < nritems; ++slot) {
94 		btrfs_item_key_to_cpu(eb, &key, slot);
95 		if (key.type != BTRFS_EXTENT_DATA_KEY)
96 			continue;
97 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
98 		extent_type = btrfs_file_extent_type(eb, fi);
99 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
100 			continue;
101 		/* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
102 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
103 		if (disk_byte != wanted_disk_byte)
104 			continue;
105 
106 		ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset);
107 		if (ret < 0)
108 			return ret;
109 	}
110 
111 	return 0;
112 }
113 
114 struct preftree {
115 	struct rb_root_cached root;
116 	unsigned int count;
117 };
118 
119 #define PREFTREE_INIT	{ .root = RB_ROOT_CACHED, .count = 0 }
120 
121 struct preftrees {
122 	struct preftree direct;    /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
123 	struct preftree indirect;  /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
124 	struct preftree indirect_missing_keys;
125 };
126 
127 /*
128  * Checks for a shared extent during backref search.
129  *
130  * The share_count tracks prelim_refs (direct and indirect) having a
131  * ref->count >0:
132  *  - incremented when a ref->count transitions to >0
133  *  - decremented when a ref->count transitions to <1
134  */
135 struct share_check {
136 	u64 root_objectid;
137 	u64 inum;
138 	int share_count;
139 };
140 
141 static inline int extent_is_shared(struct share_check *sc)
142 {
143 	return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
144 }
145 
146 static struct kmem_cache *btrfs_prelim_ref_cache;
147 
148 int __init btrfs_prelim_ref_init(void)
149 {
150 	btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
151 					sizeof(struct prelim_ref),
152 					0,
153 					SLAB_MEM_SPREAD,
154 					NULL);
155 	if (!btrfs_prelim_ref_cache)
156 		return -ENOMEM;
157 	return 0;
158 }
159 
160 void __cold btrfs_prelim_ref_exit(void)
161 {
162 	kmem_cache_destroy(btrfs_prelim_ref_cache);
163 }
164 
165 static void free_pref(struct prelim_ref *ref)
166 {
167 	kmem_cache_free(btrfs_prelim_ref_cache, ref);
168 }
169 
170 /*
171  * Return 0 when both refs are for the same block (and can be merged).
172  * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
173  * indicates a 'higher' block.
174  */
175 static int prelim_ref_compare(struct prelim_ref *ref1,
176 			      struct prelim_ref *ref2)
177 {
178 	if (ref1->level < ref2->level)
179 		return -1;
180 	if (ref1->level > ref2->level)
181 		return 1;
182 	if (ref1->root_id < ref2->root_id)
183 		return -1;
184 	if (ref1->root_id > ref2->root_id)
185 		return 1;
186 	if (ref1->key_for_search.type < ref2->key_for_search.type)
187 		return -1;
188 	if (ref1->key_for_search.type > ref2->key_for_search.type)
189 		return 1;
190 	if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
191 		return -1;
192 	if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
193 		return 1;
194 	if (ref1->key_for_search.offset < ref2->key_for_search.offset)
195 		return -1;
196 	if (ref1->key_for_search.offset > ref2->key_for_search.offset)
197 		return 1;
198 	if (ref1->parent < ref2->parent)
199 		return -1;
200 	if (ref1->parent > ref2->parent)
201 		return 1;
202 
203 	return 0;
204 }
205 
206 static void update_share_count(struct share_check *sc, int oldcount,
207 			       int newcount)
208 {
209 	if ((!sc) || (oldcount == 0 && newcount < 1))
210 		return;
211 
212 	if (oldcount > 0 && newcount < 1)
213 		sc->share_count--;
214 	else if (oldcount < 1 && newcount > 0)
215 		sc->share_count++;
216 }
217 
218 /*
219  * Add @newref to the @root rbtree, merging identical refs.
220  *
221  * Callers should assume that newref has been freed after calling.
222  */
223 static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
224 			      struct preftree *preftree,
225 			      struct prelim_ref *newref,
226 			      struct share_check *sc)
227 {
228 	struct rb_root_cached *root;
229 	struct rb_node **p;
230 	struct rb_node *parent = NULL;
231 	struct prelim_ref *ref;
232 	int result;
233 	bool leftmost = true;
234 
235 	root = &preftree->root;
236 	p = &root->rb_root.rb_node;
237 
238 	while (*p) {
239 		parent = *p;
240 		ref = rb_entry(parent, struct prelim_ref, rbnode);
241 		result = prelim_ref_compare(ref, newref);
242 		if (result < 0) {
243 			p = &(*p)->rb_left;
244 		} else if (result > 0) {
245 			p = &(*p)->rb_right;
246 			leftmost = false;
247 		} else {
248 			/* Identical refs, merge them and free @newref */
249 			struct extent_inode_elem *eie = ref->inode_list;
250 
251 			while (eie && eie->next)
252 				eie = eie->next;
253 
254 			if (!eie)
255 				ref->inode_list = newref->inode_list;
256 			else
257 				eie->next = newref->inode_list;
258 			trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
259 						     preftree->count);
260 			/*
261 			 * A delayed ref can have newref->count < 0.
262 			 * The ref->count is updated to follow any
263 			 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
264 			 */
265 			update_share_count(sc, ref->count,
266 					   ref->count + newref->count);
267 			ref->count += newref->count;
268 			free_pref(newref);
269 			return;
270 		}
271 	}
272 
273 	update_share_count(sc, 0, newref->count);
274 	preftree->count++;
275 	trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
276 	rb_link_node(&newref->rbnode, parent, p);
277 	rb_insert_color_cached(&newref->rbnode, root, leftmost);
278 }
279 
280 /*
281  * Release the entire tree.  We don't care about internal consistency so
282  * just free everything and then reset the tree root.
283  */
284 static void prelim_release(struct preftree *preftree)
285 {
286 	struct prelim_ref *ref, *next_ref;
287 
288 	rbtree_postorder_for_each_entry_safe(ref, next_ref,
289 					     &preftree->root.rb_root, rbnode)
290 		free_pref(ref);
291 
292 	preftree->root = RB_ROOT_CACHED;
293 	preftree->count = 0;
294 }
295 
296 /*
297  * the rules for all callers of this function are:
298  * - obtaining the parent is the goal
299  * - if you add a key, you must know that it is a correct key
300  * - if you cannot add the parent or a correct key, then we will look into the
301  *   block later to set a correct key
302  *
303  * delayed refs
304  * ============
305  *        backref type | shared | indirect | shared | indirect
306  * information         |   tree |     tree |   data |     data
307  * --------------------+--------+----------+--------+----------
308  *      parent logical |    y   |     -    |    -   |     -
309  *      key to resolve |    -   |     y    |    y   |     y
310  *  tree block logical |    -   |     -    |    -   |     -
311  *  root for resolving |    y   |     y    |    y   |     y
312  *
313  * - column 1:       we've the parent -> done
314  * - column 2, 3, 4: we use the key to find the parent
315  *
316  * on disk refs (inline or keyed)
317  * ==============================
318  *        backref type | shared | indirect | shared | indirect
319  * information         |   tree |     tree |   data |     data
320  * --------------------+--------+----------+--------+----------
321  *      parent logical |    y   |     -    |    y   |     -
322  *      key to resolve |    -   |     -    |    -   |     y
323  *  tree block logical |    y   |     y    |    y   |     y
324  *  root for resolving |    -   |     y    |    y   |     y
325  *
326  * - column 1, 3: we've the parent -> done
327  * - column 2:    we take the first key from the block to find the parent
328  *                (see add_missing_keys)
329  * - column 4:    we use the key to find the parent
330  *
331  * additional information that's available but not required to find the parent
332  * block might help in merging entries to gain some speed.
333  */
334 static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
335 			  struct preftree *preftree, u64 root_id,
336 			  const struct btrfs_key *key, int level, u64 parent,
337 			  u64 wanted_disk_byte, int count,
338 			  struct share_check *sc, gfp_t gfp_mask)
339 {
340 	struct prelim_ref *ref;
341 
342 	if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
343 		return 0;
344 
345 	ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
346 	if (!ref)
347 		return -ENOMEM;
348 
349 	ref->root_id = root_id;
350 	if (key) {
351 		ref->key_for_search = *key;
352 		/*
353 		 * We can often find data backrefs with an offset that is too
354 		 * large (>= LLONG_MAX, maximum allowed file offset) due to
355 		 * underflows when subtracting a file's offset with the data
356 		 * offset of its corresponding extent data item. This can
357 		 * happen for example in the clone ioctl.
358 		 * So if we detect such case we set the search key's offset to
359 		 * zero to make sure we will find the matching file extent item
360 		 * at add_all_parents(), otherwise we will miss it because the
361 		 * offset taken form the backref is much larger then the offset
362 		 * of the file extent item. This can make us scan a very large
363 		 * number of file extent items, but at least it will not make
364 		 * us miss any.
365 		 * This is an ugly workaround for a behaviour that should have
366 		 * never existed, but it does and a fix for the clone ioctl
367 		 * would touch a lot of places, cause backwards incompatibility
368 		 * and would not fix the problem for extents cloned with older
369 		 * kernels.
370 		 */
371 		if (ref->key_for_search.type == BTRFS_EXTENT_DATA_KEY &&
372 		    ref->key_for_search.offset >= LLONG_MAX)
373 			ref->key_for_search.offset = 0;
374 	} else {
375 		memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
376 	}
377 
378 	ref->inode_list = NULL;
379 	ref->level = level;
380 	ref->count = count;
381 	ref->parent = parent;
382 	ref->wanted_disk_byte = wanted_disk_byte;
383 	prelim_ref_insert(fs_info, preftree, ref, sc);
384 	return extent_is_shared(sc);
385 }
386 
387 /* direct refs use root == 0, key == NULL */
388 static int add_direct_ref(const struct btrfs_fs_info *fs_info,
389 			  struct preftrees *preftrees, int level, u64 parent,
390 			  u64 wanted_disk_byte, int count,
391 			  struct share_check *sc, gfp_t gfp_mask)
392 {
393 	return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
394 			      parent, wanted_disk_byte, count, sc, gfp_mask);
395 }
396 
397 /* indirect refs use parent == 0 */
398 static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
399 			    struct preftrees *preftrees, u64 root_id,
400 			    const struct btrfs_key *key, int level,
401 			    u64 wanted_disk_byte, int count,
402 			    struct share_check *sc, gfp_t gfp_mask)
403 {
404 	struct preftree *tree = &preftrees->indirect;
405 
406 	if (!key)
407 		tree = &preftrees->indirect_missing_keys;
408 	return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
409 			      wanted_disk_byte, count, sc, gfp_mask);
410 }
411 
412 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
413 			   struct ulist *parents, struct prelim_ref *ref,
414 			   int level, u64 time_seq, const u64 *extent_item_pos,
415 			   u64 total_refs, bool ignore_offset)
416 {
417 	int ret = 0;
418 	int slot;
419 	struct extent_buffer *eb;
420 	struct btrfs_key key;
421 	struct btrfs_key *key_for_search = &ref->key_for_search;
422 	struct btrfs_file_extent_item *fi;
423 	struct extent_inode_elem *eie = NULL, *old = NULL;
424 	u64 disk_byte;
425 	u64 wanted_disk_byte = ref->wanted_disk_byte;
426 	u64 count = 0;
427 
428 	if (level != 0) {
429 		eb = path->nodes[level];
430 		ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
431 		if (ret < 0)
432 			return ret;
433 		return 0;
434 	}
435 
436 	/*
437 	 * We normally enter this function with the path already pointing to
438 	 * the first item to check. But sometimes, we may enter it with
439 	 * slot==nritems. In that case, go to the next leaf before we continue.
440 	 */
441 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
442 		if (time_seq == SEQ_LAST)
443 			ret = btrfs_next_leaf(root, path);
444 		else
445 			ret = btrfs_next_old_leaf(root, path, time_seq);
446 	}
447 
448 	while (!ret && count < total_refs) {
449 		eb = path->nodes[0];
450 		slot = path->slots[0];
451 
452 		btrfs_item_key_to_cpu(eb, &key, slot);
453 
454 		if (key.objectid != key_for_search->objectid ||
455 		    key.type != BTRFS_EXTENT_DATA_KEY)
456 			break;
457 
458 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
459 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
460 
461 		if (disk_byte == wanted_disk_byte) {
462 			eie = NULL;
463 			old = NULL;
464 			count++;
465 			if (extent_item_pos) {
466 				ret = check_extent_in_eb(&key, eb, fi,
467 						*extent_item_pos,
468 						&eie, ignore_offset);
469 				if (ret < 0)
470 					break;
471 			}
472 			if (ret > 0)
473 				goto next;
474 			ret = ulist_add_merge_ptr(parents, eb->start,
475 						  eie, (void **)&old, GFP_NOFS);
476 			if (ret < 0)
477 				break;
478 			if (!ret && extent_item_pos) {
479 				while (old->next)
480 					old = old->next;
481 				old->next = eie;
482 			}
483 			eie = NULL;
484 		}
485 next:
486 		if (time_seq == SEQ_LAST)
487 			ret = btrfs_next_item(root, path);
488 		else
489 			ret = btrfs_next_old_item(root, path, time_seq);
490 	}
491 
492 	if (ret > 0)
493 		ret = 0;
494 	else if (ret < 0)
495 		free_inode_elem_list(eie);
496 	return ret;
497 }
498 
499 /*
500  * resolve an indirect backref in the form (root_id, key, level)
501  * to a logical address
502  */
503 static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
504 				struct btrfs_path *path, u64 time_seq,
505 				struct prelim_ref *ref, struct ulist *parents,
506 				const u64 *extent_item_pos, u64 total_refs,
507 				bool ignore_offset)
508 {
509 	struct btrfs_root *root;
510 	struct btrfs_key root_key;
511 	struct extent_buffer *eb;
512 	int ret = 0;
513 	int root_level;
514 	int level = ref->level;
515 	int index;
516 
517 	root_key.objectid = ref->root_id;
518 	root_key.type = BTRFS_ROOT_ITEM_KEY;
519 	root_key.offset = (u64)-1;
520 
521 	index = srcu_read_lock(&fs_info->subvol_srcu);
522 
523 	root = btrfs_get_fs_root(fs_info, &root_key, false);
524 	if (IS_ERR(root)) {
525 		srcu_read_unlock(&fs_info->subvol_srcu, index);
526 		ret = PTR_ERR(root);
527 		goto out;
528 	}
529 
530 	if (btrfs_is_testing(fs_info)) {
531 		srcu_read_unlock(&fs_info->subvol_srcu, index);
532 		ret = -ENOENT;
533 		goto out;
534 	}
535 
536 	if (path->search_commit_root)
537 		root_level = btrfs_header_level(root->commit_root);
538 	else if (time_seq == SEQ_LAST)
539 		root_level = btrfs_header_level(root->node);
540 	else
541 		root_level = btrfs_old_root_level(root, time_seq);
542 
543 	if (root_level + 1 == level) {
544 		srcu_read_unlock(&fs_info->subvol_srcu, index);
545 		goto out;
546 	}
547 
548 	path->lowest_level = level;
549 	if (time_seq == SEQ_LAST)
550 		ret = btrfs_search_slot(NULL, root, &ref->key_for_search, path,
551 					0, 0);
552 	else
553 		ret = btrfs_search_old_slot(root, &ref->key_for_search, path,
554 					    time_seq);
555 
556 	/* root node has been locked, we can release @subvol_srcu safely here */
557 	srcu_read_unlock(&fs_info->subvol_srcu, index);
558 
559 	btrfs_debug(fs_info,
560 		"search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
561 		 ref->root_id, level, ref->count, ret,
562 		 ref->key_for_search.objectid, ref->key_for_search.type,
563 		 ref->key_for_search.offset);
564 	if (ret < 0)
565 		goto out;
566 
567 	eb = path->nodes[level];
568 	while (!eb) {
569 		if (WARN_ON(!level)) {
570 			ret = 1;
571 			goto out;
572 		}
573 		level--;
574 		eb = path->nodes[level];
575 	}
576 
577 	ret = add_all_parents(root, path, parents, ref, level, time_seq,
578 			      extent_item_pos, total_refs, ignore_offset);
579 out:
580 	path->lowest_level = 0;
581 	btrfs_release_path(path);
582 	return ret;
583 }
584 
585 static struct extent_inode_elem *
586 unode_aux_to_inode_list(struct ulist_node *node)
587 {
588 	if (!node)
589 		return NULL;
590 	return (struct extent_inode_elem *)(uintptr_t)node->aux;
591 }
592 
593 /*
594  * We maintain three separate rbtrees: one for direct refs, one for
595  * indirect refs which have a key, and one for indirect refs which do not
596  * have a key. Each tree does merge on insertion.
597  *
598  * Once all of the references are located, we iterate over the tree of
599  * indirect refs with missing keys. An appropriate key is located and
600  * the ref is moved onto the tree for indirect refs. After all missing
601  * keys are thus located, we iterate over the indirect ref tree, resolve
602  * each reference, and then insert the resolved reference onto the
603  * direct tree (merging there too).
604  *
605  * New backrefs (i.e., for parent nodes) are added to the appropriate
606  * rbtree as they are encountered. The new backrefs are subsequently
607  * resolved as above.
608  */
609 static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
610 				 struct btrfs_path *path, u64 time_seq,
611 				 struct preftrees *preftrees,
612 				 const u64 *extent_item_pos, u64 total_refs,
613 				 struct share_check *sc, bool ignore_offset)
614 {
615 	int err;
616 	int ret = 0;
617 	struct ulist *parents;
618 	struct ulist_node *node;
619 	struct ulist_iterator uiter;
620 	struct rb_node *rnode;
621 
622 	parents = ulist_alloc(GFP_NOFS);
623 	if (!parents)
624 		return -ENOMEM;
625 
626 	/*
627 	 * We could trade memory usage for performance here by iterating
628 	 * the tree, allocating new refs for each insertion, and then
629 	 * freeing the entire indirect tree when we're done.  In some test
630 	 * cases, the tree can grow quite large (~200k objects).
631 	 */
632 	while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
633 		struct prelim_ref *ref;
634 
635 		ref = rb_entry(rnode, struct prelim_ref, rbnode);
636 		if (WARN(ref->parent,
637 			 "BUG: direct ref found in indirect tree")) {
638 			ret = -EINVAL;
639 			goto out;
640 		}
641 
642 		rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
643 		preftrees->indirect.count--;
644 
645 		if (ref->count == 0) {
646 			free_pref(ref);
647 			continue;
648 		}
649 
650 		if (sc && sc->root_objectid &&
651 		    ref->root_id != sc->root_objectid) {
652 			free_pref(ref);
653 			ret = BACKREF_FOUND_SHARED;
654 			goto out;
655 		}
656 		err = resolve_indirect_ref(fs_info, path, time_seq, ref,
657 					   parents, extent_item_pos,
658 					   total_refs, ignore_offset);
659 		/*
660 		 * we can only tolerate ENOENT,otherwise,we should catch error
661 		 * and return directly.
662 		 */
663 		if (err == -ENOENT) {
664 			prelim_ref_insert(fs_info, &preftrees->direct, ref,
665 					  NULL);
666 			continue;
667 		} else if (err) {
668 			free_pref(ref);
669 			ret = err;
670 			goto out;
671 		}
672 
673 		/* we put the first parent into the ref at hand */
674 		ULIST_ITER_INIT(&uiter);
675 		node = ulist_next(parents, &uiter);
676 		ref->parent = node ? node->val : 0;
677 		ref->inode_list = unode_aux_to_inode_list(node);
678 
679 		/* Add a prelim_ref(s) for any other parent(s). */
680 		while ((node = ulist_next(parents, &uiter))) {
681 			struct prelim_ref *new_ref;
682 
683 			new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
684 						   GFP_NOFS);
685 			if (!new_ref) {
686 				free_pref(ref);
687 				ret = -ENOMEM;
688 				goto out;
689 			}
690 			memcpy(new_ref, ref, sizeof(*ref));
691 			new_ref->parent = node->val;
692 			new_ref->inode_list = unode_aux_to_inode_list(node);
693 			prelim_ref_insert(fs_info, &preftrees->direct,
694 					  new_ref, NULL);
695 		}
696 
697 		/*
698 		 * Now it's a direct ref, put it in the direct tree. We must
699 		 * do this last because the ref could be merged/freed here.
700 		 */
701 		prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
702 
703 		ulist_reinit(parents);
704 		cond_resched();
705 	}
706 out:
707 	ulist_free(parents);
708 	return ret;
709 }
710 
711 /*
712  * read tree blocks and add keys where required.
713  */
714 static int add_missing_keys(struct btrfs_fs_info *fs_info,
715 			    struct preftrees *preftrees, bool lock)
716 {
717 	struct prelim_ref *ref;
718 	struct extent_buffer *eb;
719 	struct preftree *tree = &preftrees->indirect_missing_keys;
720 	struct rb_node *node;
721 
722 	while ((node = rb_first_cached(&tree->root))) {
723 		ref = rb_entry(node, struct prelim_ref, rbnode);
724 		rb_erase_cached(node, &tree->root);
725 
726 		BUG_ON(ref->parent);	/* should not be a direct ref */
727 		BUG_ON(ref->key_for_search.type);
728 		BUG_ON(!ref->wanted_disk_byte);
729 
730 		eb = read_tree_block(fs_info, ref->wanted_disk_byte, 0,
731 				     ref->level - 1, NULL);
732 		if (IS_ERR(eb)) {
733 			free_pref(ref);
734 			return PTR_ERR(eb);
735 		} else if (!extent_buffer_uptodate(eb)) {
736 			free_pref(ref);
737 			free_extent_buffer(eb);
738 			return -EIO;
739 		}
740 		if (lock)
741 			btrfs_tree_read_lock(eb);
742 		if (btrfs_header_level(eb) == 0)
743 			btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
744 		else
745 			btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
746 		if (lock)
747 			btrfs_tree_read_unlock(eb);
748 		free_extent_buffer(eb);
749 		prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
750 		cond_resched();
751 	}
752 	return 0;
753 }
754 
755 /*
756  * add all currently queued delayed refs from this head whose seq nr is
757  * smaller or equal that seq to the list
758  */
759 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
760 			    struct btrfs_delayed_ref_head *head, u64 seq,
761 			    struct preftrees *preftrees, u64 *total_refs,
762 			    struct share_check *sc)
763 {
764 	struct btrfs_delayed_ref_node *node;
765 	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
766 	struct btrfs_key key;
767 	struct btrfs_key tmp_op_key;
768 	struct rb_node *n;
769 	int count;
770 	int ret = 0;
771 
772 	if (extent_op && extent_op->update_key)
773 		btrfs_disk_key_to_cpu(&tmp_op_key, &extent_op->key);
774 
775 	spin_lock(&head->lock);
776 	for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
777 		node = rb_entry(n, struct btrfs_delayed_ref_node,
778 				ref_node);
779 		if (node->seq > seq)
780 			continue;
781 
782 		switch (node->action) {
783 		case BTRFS_ADD_DELAYED_EXTENT:
784 		case BTRFS_UPDATE_DELAYED_HEAD:
785 			WARN_ON(1);
786 			continue;
787 		case BTRFS_ADD_DELAYED_REF:
788 			count = node->ref_mod;
789 			break;
790 		case BTRFS_DROP_DELAYED_REF:
791 			count = node->ref_mod * -1;
792 			break;
793 		default:
794 			BUG();
795 		}
796 		*total_refs += count;
797 		switch (node->type) {
798 		case BTRFS_TREE_BLOCK_REF_KEY: {
799 			/* NORMAL INDIRECT METADATA backref */
800 			struct btrfs_delayed_tree_ref *ref;
801 
802 			ref = btrfs_delayed_node_to_tree_ref(node);
803 			ret = add_indirect_ref(fs_info, preftrees, ref->root,
804 					       &tmp_op_key, ref->level + 1,
805 					       node->bytenr, count, sc,
806 					       GFP_ATOMIC);
807 			break;
808 		}
809 		case BTRFS_SHARED_BLOCK_REF_KEY: {
810 			/* SHARED DIRECT METADATA backref */
811 			struct btrfs_delayed_tree_ref *ref;
812 
813 			ref = btrfs_delayed_node_to_tree_ref(node);
814 
815 			ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
816 					     ref->parent, node->bytenr, count,
817 					     sc, GFP_ATOMIC);
818 			break;
819 		}
820 		case BTRFS_EXTENT_DATA_REF_KEY: {
821 			/* NORMAL INDIRECT DATA backref */
822 			struct btrfs_delayed_data_ref *ref;
823 			ref = btrfs_delayed_node_to_data_ref(node);
824 
825 			key.objectid = ref->objectid;
826 			key.type = BTRFS_EXTENT_DATA_KEY;
827 			key.offset = ref->offset;
828 
829 			/*
830 			 * Found a inum that doesn't match our known inum, we
831 			 * know it's shared.
832 			 */
833 			if (sc && sc->inum && ref->objectid != sc->inum) {
834 				ret = BACKREF_FOUND_SHARED;
835 				goto out;
836 			}
837 
838 			ret = add_indirect_ref(fs_info, preftrees, ref->root,
839 					       &key, 0, node->bytenr, count, sc,
840 					       GFP_ATOMIC);
841 			break;
842 		}
843 		case BTRFS_SHARED_DATA_REF_KEY: {
844 			/* SHARED DIRECT FULL backref */
845 			struct btrfs_delayed_data_ref *ref;
846 
847 			ref = btrfs_delayed_node_to_data_ref(node);
848 
849 			ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
850 					     node->bytenr, count, sc,
851 					     GFP_ATOMIC);
852 			break;
853 		}
854 		default:
855 			WARN_ON(1);
856 		}
857 		/*
858 		 * We must ignore BACKREF_FOUND_SHARED until all delayed
859 		 * refs have been checked.
860 		 */
861 		if (ret && (ret != BACKREF_FOUND_SHARED))
862 			break;
863 	}
864 	if (!ret)
865 		ret = extent_is_shared(sc);
866 out:
867 	spin_unlock(&head->lock);
868 	return ret;
869 }
870 
871 /*
872  * add all inline backrefs for bytenr to the list
873  *
874  * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
875  */
876 static int add_inline_refs(const struct btrfs_fs_info *fs_info,
877 			   struct btrfs_path *path, u64 bytenr,
878 			   int *info_level, struct preftrees *preftrees,
879 			   u64 *total_refs, struct share_check *sc)
880 {
881 	int ret = 0;
882 	int slot;
883 	struct extent_buffer *leaf;
884 	struct btrfs_key key;
885 	struct btrfs_key found_key;
886 	unsigned long ptr;
887 	unsigned long end;
888 	struct btrfs_extent_item *ei;
889 	u64 flags;
890 	u64 item_size;
891 
892 	/*
893 	 * enumerate all inline refs
894 	 */
895 	leaf = path->nodes[0];
896 	slot = path->slots[0];
897 
898 	item_size = btrfs_item_size_nr(leaf, slot);
899 	BUG_ON(item_size < sizeof(*ei));
900 
901 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
902 	flags = btrfs_extent_flags(leaf, ei);
903 	*total_refs += btrfs_extent_refs(leaf, ei);
904 	btrfs_item_key_to_cpu(leaf, &found_key, slot);
905 
906 	ptr = (unsigned long)(ei + 1);
907 	end = (unsigned long)ei + item_size;
908 
909 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
910 	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
911 		struct btrfs_tree_block_info *info;
912 
913 		info = (struct btrfs_tree_block_info *)ptr;
914 		*info_level = btrfs_tree_block_level(leaf, info);
915 		ptr += sizeof(struct btrfs_tree_block_info);
916 		BUG_ON(ptr > end);
917 	} else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
918 		*info_level = found_key.offset;
919 	} else {
920 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
921 	}
922 
923 	while (ptr < end) {
924 		struct btrfs_extent_inline_ref *iref;
925 		u64 offset;
926 		int type;
927 
928 		iref = (struct btrfs_extent_inline_ref *)ptr;
929 		type = btrfs_get_extent_inline_ref_type(leaf, iref,
930 							BTRFS_REF_TYPE_ANY);
931 		if (type == BTRFS_REF_TYPE_INVALID)
932 			return -EUCLEAN;
933 
934 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
935 
936 		switch (type) {
937 		case BTRFS_SHARED_BLOCK_REF_KEY:
938 			ret = add_direct_ref(fs_info, preftrees,
939 					     *info_level + 1, offset,
940 					     bytenr, 1, NULL, GFP_NOFS);
941 			break;
942 		case BTRFS_SHARED_DATA_REF_KEY: {
943 			struct btrfs_shared_data_ref *sdref;
944 			int count;
945 
946 			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
947 			count = btrfs_shared_data_ref_count(leaf, sdref);
948 
949 			ret = add_direct_ref(fs_info, preftrees, 0, offset,
950 					     bytenr, count, sc, GFP_NOFS);
951 			break;
952 		}
953 		case BTRFS_TREE_BLOCK_REF_KEY:
954 			ret = add_indirect_ref(fs_info, preftrees, offset,
955 					       NULL, *info_level + 1,
956 					       bytenr, 1, NULL, GFP_NOFS);
957 			break;
958 		case BTRFS_EXTENT_DATA_REF_KEY: {
959 			struct btrfs_extent_data_ref *dref;
960 			int count;
961 			u64 root;
962 
963 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
964 			count = btrfs_extent_data_ref_count(leaf, dref);
965 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
966 								      dref);
967 			key.type = BTRFS_EXTENT_DATA_KEY;
968 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
969 
970 			if (sc && sc->inum && key.objectid != sc->inum) {
971 				ret = BACKREF_FOUND_SHARED;
972 				break;
973 			}
974 
975 			root = btrfs_extent_data_ref_root(leaf, dref);
976 
977 			ret = add_indirect_ref(fs_info, preftrees, root,
978 					       &key, 0, bytenr, count,
979 					       sc, GFP_NOFS);
980 			break;
981 		}
982 		default:
983 			WARN_ON(1);
984 		}
985 		if (ret)
986 			return ret;
987 		ptr += btrfs_extent_inline_ref_size(type);
988 	}
989 
990 	return 0;
991 }
992 
993 /*
994  * add all non-inline backrefs for bytenr to the list
995  *
996  * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
997  */
998 static int add_keyed_refs(struct btrfs_fs_info *fs_info,
999 			  struct btrfs_path *path, u64 bytenr,
1000 			  int info_level, struct preftrees *preftrees,
1001 			  struct share_check *sc)
1002 {
1003 	struct btrfs_root *extent_root = fs_info->extent_root;
1004 	int ret;
1005 	int slot;
1006 	struct extent_buffer *leaf;
1007 	struct btrfs_key key;
1008 
1009 	while (1) {
1010 		ret = btrfs_next_item(extent_root, path);
1011 		if (ret < 0)
1012 			break;
1013 		if (ret) {
1014 			ret = 0;
1015 			break;
1016 		}
1017 
1018 		slot = path->slots[0];
1019 		leaf = path->nodes[0];
1020 		btrfs_item_key_to_cpu(leaf, &key, slot);
1021 
1022 		if (key.objectid != bytenr)
1023 			break;
1024 		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1025 			continue;
1026 		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1027 			break;
1028 
1029 		switch (key.type) {
1030 		case BTRFS_SHARED_BLOCK_REF_KEY:
1031 			/* SHARED DIRECT METADATA backref */
1032 			ret = add_direct_ref(fs_info, preftrees,
1033 					     info_level + 1, key.offset,
1034 					     bytenr, 1, NULL, GFP_NOFS);
1035 			break;
1036 		case BTRFS_SHARED_DATA_REF_KEY: {
1037 			/* SHARED DIRECT FULL backref */
1038 			struct btrfs_shared_data_ref *sdref;
1039 			int count;
1040 
1041 			sdref = btrfs_item_ptr(leaf, slot,
1042 					      struct btrfs_shared_data_ref);
1043 			count = btrfs_shared_data_ref_count(leaf, sdref);
1044 			ret = add_direct_ref(fs_info, preftrees, 0,
1045 					     key.offset, bytenr, count,
1046 					     sc, GFP_NOFS);
1047 			break;
1048 		}
1049 		case BTRFS_TREE_BLOCK_REF_KEY:
1050 			/* NORMAL INDIRECT METADATA backref */
1051 			ret = add_indirect_ref(fs_info, preftrees, key.offset,
1052 					       NULL, info_level + 1, bytenr,
1053 					       1, NULL, GFP_NOFS);
1054 			break;
1055 		case BTRFS_EXTENT_DATA_REF_KEY: {
1056 			/* NORMAL INDIRECT DATA backref */
1057 			struct btrfs_extent_data_ref *dref;
1058 			int count;
1059 			u64 root;
1060 
1061 			dref = btrfs_item_ptr(leaf, slot,
1062 					      struct btrfs_extent_data_ref);
1063 			count = btrfs_extent_data_ref_count(leaf, dref);
1064 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
1065 								      dref);
1066 			key.type = BTRFS_EXTENT_DATA_KEY;
1067 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1068 
1069 			if (sc && sc->inum && key.objectid != sc->inum) {
1070 				ret = BACKREF_FOUND_SHARED;
1071 				break;
1072 			}
1073 
1074 			root = btrfs_extent_data_ref_root(leaf, dref);
1075 			ret = add_indirect_ref(fs_info, preftrees, root,
1076 					       &key, 0, bytenr, count,
1077 					       sc, GFP_NOFS);
1078 			break;
1079 		}
1080 		default:
1081 			WARN_ON(1);
1082 		}
1083 		if (ret)
1084 			return ret;
1085 
1086 	}
1087 
1088 	return ret;
1089 }
1090 
1091 /*
1092  * this adds all existing backrefs (inline backrefs, backrefs and delayed
1093  * refs) for the given bytenr to the refs list, merges duplicates and resolves
1094  * indirect refs to their parent bytenr.
1095  * When roots are found, they're added to the roots list
1096  *
1097  * If time_seq is set to SEQ_LAST, it will not search delayed_refs, and behave
1098  * much like trans == NULL case, the difference only lies in it will not
1099  * commit root.
1100  * The special case is for qgroup to search roots in commit_transaction().
1101  *
1102  * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1103  * shared extent is detected.
1104  *
1105  * Otherwise this returns 0 for success and <0 for an error.
1106  *
1107  * If ignore_offset is set to false, only extent refs whose offsets match
1108  * extent_item_pos are returned.  If true, every extent ref is returned
1109  * and extent_item_pos is ignored.
1110  *
1111  * FIXME some caching might speed things up
1112  */
1113 static int find_parent_nodes(struct btrfs_trans_handle *trans,
1114 			     struct btrfs_fs_info *fs_info, u64 bytenr,
1115 			     u64 time_seq, struct ulist *refs,
1116 			     struct ulist *roots, const u64 *extent_item_pos,
1117 			     struct share_check *sc, bool ignore_offset)
1118 {
1119 	struct btrfs_key key;
1120 	struct btrfs_path *path;
1121 	struct btrfs_delayed_ref_root *delayed_refs = NULL;
1122 	struct btrfs_delayed_ref_head *head;
1123 	int info_level = 0;
1124 	int ret;
1125 	struct prelim_ref *ref;
1126 	struct rb_node *node;
1127 	struct extent_inode_elem *eie = NULL;
1128 	/* total of both direct AND indirect refs! */
1129 	u64 total_refs = 0;
1130 	struct preftrees preftrees = {
1131 		.direct = PREFTREE_INIT,
1132 		.indirect = PREFTREE_INIT,
1133 		.indirect_missing_keys = PREFTREE_INIT
1134 	};
1135 
1136 	key.objectid = bytenr;
1137 	key.offset = (u64)-1;
1138 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1139 		key.type = BTRFS_METADATA_ITEM_KEY;
1140 	else
1141 		key.type = BTRFS_EXTENT_ITEM_KEY;
1142 
1143 	path = btrfs_alloc_path();
1144 	if (!path)
1145 		return -ENOMEM;
1146 	if (!trans) {
1147 		path->search_commit_root = 1;
1148 		path->skip_locking = 1;
1149 	}
1150 
1151 	if (time_seq == SEQ_LAST)
1152 		path->skip_locking = 1;
1153 
1154 	/*
1155 	 * grab both a lock on the path and a lock on the delayed ref head.
1156 	 * We need both to get a consistent picture of how the refs look
1157 	 * at a specified point in time
1158 	 */
1159 again:
1160 	head = NULL;
1161 
1162 	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
1163 	if (ret < 0)
1164 		goto out;
1165 	BUG_ON(ret == 0);
1166 
1167 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
1168 	if (trans && likely(trans->type != __TRANS_DUMMY) &&
1169 	    time_seq != SEQ_LAST) {
1170 #else
1171 	if (trans && time_seq != SEQ_LAST) {
1172 #endif
1173 		/*
1174 		 * look if there are updates for this ref queued and lock the
1175 		 * head
1176 		 */
1177 		delayed_refs = &trans->transaction->delayed_refs;
1178 		spin_lock(&delayed_refs->lock);
1179 		head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1180 		if (head) {
1181 			if (!mutex_trylock(&head->mutex)) {
1182 				refcount_inc(&head->refs);
1183 				spin_unlock(&delayed_refs->lock);
1184 
1185 				btrfs_release_path(path);
1186 
1187 				/*
1188 				 * Mutex was contended, block until it's
1189 				 * released and try again
1190 				 */
1191 				mutex_lock(&head->mutex);
1192 				mutex_unlock(&head->mutex);
1193 				btrfs_put_delayed_ref_head(head);
1194 				goto again;
1195 			}
1196 			spin_unlock(&delayed_refs->lock);
1197 			ret = add_delayed_refs(fs_info, head, time_seq,
1198 					       &preftrees, &total_refs, sc);
1199 			mutex_unlock(&head->mutex);
1200 			if (ret)
1201 				goto out;
1202 		} else {
1203 			spin_unlock(&delayed_refs->lock);
1204 		}
1205 	}
1206 
1207 	if (path->slots[0]) {
1208 		struct extent_buffer *leaf;
1209 		int slot;
1210 
1211 		path->slots[0]--;
1212 		leaf = path->nodes[0];
1213 		slot = path->slots[0];
1214 		btrfs_item_key_to_cpu(leaf, &key, slot);
1215 		if (key.objectid == bytenr &&
1216 		    (key.type == BTRFS_EXTENT_ITEM_KEY ||
1217 		     key.type == BTRFS_METADATA_ITEM_KEY)) {
1218 			ret = add_inline_refs(fs_info, path, bytenr,
1219 					      &info_level, &preftrees,
1220 					      &total_refs, sc);
1221 			if (ret)
1222 				goto out;
1223 			ret = add_keyed_refs(fs_info, path, bytenr, info_level,
1224 					     &preftrees, sc);
1225 			if (ret)
1226 				goto out;
1227 		}
1228 	}
1229 
1230 	btrfs_release_path(path);
1231 
1232 	ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
1233 	if (ret)
1234 		goto out;
1235 
1236 	WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1237 
1238 	ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1239 				    extent_item_pos, total_refs, sc, ignore_offset);
1240 	if (ret)
1241 		goto out;
1242 
1243 	WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1244 
1245 	/*
1246 	 * This walks the tree of merged and resolved refs. Tree blocks are
1247 	 * read in as needed. Unique entries are added to the ulist, and
1248 	 * the list of found roots is updated.
1249 	 *
1250 	 * We release the entire tree in one go before returning.
1251 	 */
1252 	node = rb_first_cached(&preftrees.direct.root);
1253 	while (node) {
1254 		ref = rb_entry(node, struct prelim_ref, rbnode);
1255 		node = rb_next(&ref->rbnode);
1256 		/*
1257 		 * ref->count < 0 can happen here if there are delayed
1258 		 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1259 		 * prelim_ref_insert() relies on this when merging
1260 		 * identical refs to keep the overall count correct.
1261 		 * prelim_ref_insert() will merge only those refs
1262 		 * which compare identically.  Any refs having
1263 		 * e.g. different offsets would not be merged,
1264 		 * and would retain their original ref->count < 0.
1265 		 */
1266 		if (roots && ref->count && ref->root_id && ref->parent == 0) {
1267 			if (sc && sc->root_objectid &&
1268 			    ref->root_id != sc->root_objectid) {
1269 				ret = BACKREF_FOUND_SHARED;
1270 				goto out;
1271 			}
1272 
1273 			/* no parent == root of tree */
1274 			ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1275 			if (ret < 0)
1276 				goto out;
1277 		}
1278 		if (ref->count && ref->parent) {
1279 			if (extent_item_pos && !ref->inode_list &&
1280 			    ref->level == 0) {
1281 				struct extent_buffer *eb;
1282 
1283 				eb = read_tree_block(fs_info, ref->parent, 0,
1284 						     ref->level, NULL);
1285 				if (IS_ERR(eb)) {
1286 					ret = PTR_ERR(eb);
1287 					goto out;
1288 				} else if (!extent_buffer_uptodate(eb)) {
1289 					free_extent_buffer(eb);
1290 					ret = -EIO;
1291 					goto out;
1292 				}
1293 
1294 				if (!path->skip_locking) {
1295 					btrfs_tree_read_lock(eb);
1296 					btrfs_set_lock_blocking_read(eb);
1297 				}
1298 				ret = find_extent_in_eb(eb, bytenr,
1299 							*extent_item_pos, &eie, ignore_offset);
1300 				if (!path->skip_locking)
1301 					btrfs_tree_read_unlock_blocking(eb);
1302 				free_extent_buffer(eb);
1303 				if (ret < 0)
1304 					goto out;
1305 				ref->inode_list = eie;
1306 			}
1307 			ret = ulist_add_merge_ptr(refs, ref->parent,
1308 						  ref->inode_list,
1309 						  (void **)&eie, GFP_NOFS);
1310 			if (ret < 0)
1311 				goto out;
1312 			if (!ret && extent_item_pos) {
1313 				/*
1314 				 * we've recorded that parent, so we must extend
1315 				 * its inode list here
1316 				 */
1317 				BUG_ON(!eie);
1318 				while (eie->next)
1319 					eie = eie->next;
1320 				eie->next = ref->inode_list;
1321 			}
1322 			eie = NULL;
1323 		}
1324 		cond_resched();
1325 	}
1326 
1327 out:
1328 	btrfs_free_path(path);
1329 
1330 	prelim_release(&preftrees.direct);
1331 	prelim_release(&preftrees.indirect);
1332 	prelim_release(&preftrees.indirect_missing_keys);
1333 
1334 	if (ret < 0)
1335 		free_inode_elem_list(eie);
1336 	return ret;
1337 }
1338 
1339 static void free_leaf_list(struct ulist *blocks)
1340 {
1341 	struct ulist_node *node = NULL;
1342 	struct extent_inode_elem *eie;
1343 	struct ulist_iterator uiter;
1344 
1345 	ULIST_ITER_INIT(&uiter);
1346 	while ((node = ulist_next(blocks, &uiter))) {
1347 		if (!node->aux)
1348 			continue;
1349 		eie = unode_aux_to_inode_list(node);
1350 		free_inode_elem_list(eie);
1351 		node->aux = 0;
1352 	}
1353 
1354 	ulist_free(blocks);
1355 }
1356 
1357 /*
1358  * Finds all leafs with a reference to the specified combination of bytenr and
1359  * offset. key_list_head will point to a list of corresponding keys (caller must
1360  * free each list element). The leafs will be stored in the leafs ulist, which
1361  * must be freed with ulist_free.
1362  *
1363  * returns 0 on success, <0 on error
1364  */
1365 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1366 				struct btrfs_fs_info *fs_info, u64 bytenr,
1367 				u64 time_seq, struct ulist **leafs,
1368 				const u64 *extent_item_pos, bool ignore_offset)
1369 {
1370 	int ret;
1371 
1372 	*leafs = ulist_alloc(GFP_NOFS);
1373 	if (!*leafs)
1374 		return -ENOMEM;
1375 
1376 	ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1377 				*leafs, NULL, extent_item_pos, NULL, ignore_offset);
1378 	if (ret < 0 && ret != -ENOENT) {
1379 		free_leaf_list(*leafs);
1380 		return ret;
1381 	}
1382 
1383 	return 0;
1384 }
1385 
1386 /*
1387  * walk all backrefs for a given extent to find all roots that reference this
1388  * extent. Walking a backref means finding all extents that reference this
1389  * extent and in turn walk the backrefs of those, too. Naturally this is a
1390  * recursive process, but here it is implemented in an iterative fashion: We
1391  * find all referencing extents for the extent in question and put them on a
1392  * list. In turn, we find all referencing extents for those, further appending
1393  * to the list. The way we iterate the list allows adding more elements after
1394  * the current while iterating. The process stops when we reach the end of the
1395  * list. Found roots are added to the roots list.
1396  *
1397  * returns 0 on success, < 0 on error.
1398  */
1399 static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1400 				     struct btrfs_fs_info *fs_info, u64 bytenr,
1401 				     u64 time_seq, struct ulist **roots,
1402 				     bool ignore_offset)
1403 {
1404 	struct ulist *tmp;
1405 	struct ulist_node *node = NULL;
1406 	struct ulist_iterator uiter;
1407 	int ret;
1408 
1409 	tmp = ulist_alloc(GFP_NOFS);
1410 	if (!tmp)
1411 		return -ENOMEM;
1412 	*roots = ulist_alloc(GFP_NOFS);
1413 	if (!*roots) {
1414 		ulist_free(tmp);
1415 		return -ENOMEM;
1416 	}
1417 
1418 	ULIST_ITER_INIT(&uiter);
1419 	while (1) {
1420 		ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1421 					tmp, *roots, NULL, NULL, ignore_offset);
1422 		if (ret < 0 && ret != -ENOENT) {
1423 			ulist_free(tmp);
1424 			ulist_free(*roots);
1425 			return ret;
1426 		}
1427 		node = ulist_next(tmp, &uiter);
1428 		if (!node)
1429 			break;
1430 		bytenr = node->val;
1431 		cond_resched();
1432 	}
1433 
1434 	ulist_free(tmp);
1435 	return 0;
1436 }
1437 
1438 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1439 			 struct btrfs_fs_info *fs_info, u64 bytenr,
1440 			 u64 time_seq, struct ulist **roots,
1441 			 bool ignore_offset)
1442 {
1443 	int ret;
1444 
1445 	if (!trans)
1446 		down_read(&fs_info->commit_root_sem);
1447 	ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1448 					time_seq, roots, ignore_offset);
1449 	if (!trans)
1450 		up_read(&fs_info->commit_root_sem);
1451 	return ret;
1452 }
1453 
1454 /**
1455  * btrfs_check_shared - tell us whether an extent is shared
1456  *
1457  * btrfs_check_shared uses the backref walking code but will short
1458  * circuit as soon as it finds a root or inode that doesn't match the
1459  * one passed in. This provides a significant performance benefit for
1460  * callers (such as fiemap) which want to know whether the extent is
1461  * shared but do not need a ref count.
1462  *
1463  * This attempts to attach to the running transaction in order to account for
1464  * delayed refs, but continues on even when no running transaction exists.
1465  *
1466  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1467  */
1468 int btrfs_check_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
1469 		struct ulist *roots, struct ulist *tmp)
1470 {
1471 	struct btrfs_fs_info *fs_info = root->fs_info;
1472 	struct btrfs_trans_handle *trans;
1473 	struct ulist_iterator uiter;
1474 	struct ulist_node *node;
1475 	struct seq_list elem = SEQ_LIST_INIT(elem);
1476 	int ret = 0;
1477 	struct share_check shared = {
1478 		.root_objectid = root->root_key.objectid,
1479 		.inum = inum,
1480 		.share_count = 0,
1481 	};
1482 
1483 	ulist_init(roots);
1484 	ulist_init(tmp);
1485 
1486 	trans = btrfs_join_transaction_nostart(root);
1487 	if (IS_ERR(trans)) {
1488 		if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1489 			ret = PTR_ERR(trans);
1490 			goto out;
1491 		}
1492 		trans = NULL;
1493 		down_read(&fs_info->commit_root_sem);
1494 	} else {
1495 		btrfs_get_tree_mod_seq(fs_info, &elem);
1496 	}
1497 
1498 	ULIST_ITER_INIT(&uiter);
1499 	while (1) {
1500 		ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1501 					roots, NULL, &shared, false);
1502 		if (ret == BACKREF_FOUND_SHARED) {
1503 			/* this is the only condition under which we return 1 */
1504 			ret = 1;
1505 			break;
1506 		}
1507 		if (ret < 0 && ret != -ENOENT)
1508 			break;
1509 		ret = 0;
1510 		node = ulist_next(tmp, &uiter);
1511 		if (!node)
1512 			break;
1513 		bytenr = node->val;
1514 		shared.share_count = 0;
1515 		cond_resched();
1516 	}
1517 
1518 	if (trans) {
1519 		btrfs_put_tree_mod_seq(fs_info, &elem);
1520 		btrfs_end_transaction(trans);
1521 	} else {
1522 		up_read(&fs_info->commit_root_sem);
1523 	}
1524 out:
1525 	ulist_release(roots);
1526 	ulist_release(tmp);
1527 	return ret;
1528 }
1529 
1530 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1531 			  u64 start_off, struct btrfs_path *path,
1532 			  struct btrfs_inode_extref **ret_extref,
1533 			  u64 *found_off)
1534 {
1535 	int ret, slot;
1536 	struct btrfs_key key;
1537 	struct btrfs_key found_key;
1538 	struct btrfs_inode_extref *extref;
1539 	const struct extent_buffer *leaf;
1540 	unsigned long ptr;
1541 
1542 	key.objectid = inode_objectid;
1543 	key.type = BTRFS_INODE_EXTREF_KEY;
1544 	key.offset = start_off;
1545 
1546 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1547 	if (ret < 0)
1548 		return ret;
1549 
1550 	while (1) {
1551 		leaf = path->nodes[0];
1552 		slot = path->slots[0];
1553 		if (slot >= btrfs_header_nritems(leaf)) {
1554 			/*
1555 			 * If the item at offset is not found,
1556 			 * btrfs_search_slot will point us to the slot
1557 			 * where it should be inserted. In our case
1558 			 * that will be the slot directly before the
1559 			 * next INODE_REF_KEY_V2 item. In the case
1560 			 * that we're pointing to the last slot in a
1561 			 * leaf, we must move one leaf over.
1562 			 */
1563 			ret = btrfs_next_leaf(root, path);
1564 			if (ret) {
1565 				if (ret >= 1)
1566 					ret = -ENOENT;
1567 				break;
1568 			}
1569 			continue;
1570 		}
1571 
1572 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1573 
1574 		/*
1575 		 * Check that we're still looking at an extended ref key for
1576 		 * this particular objectid. If we have different
1577 		 * objectid or type then there are no more to be found
1578 		 * in the tree and we can exit.
1579 		 */
1580 		ret = -ENOENT;
1581 		if (found_key.objectid != inode_objectid)
1582 			break;
1583 		if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1584 			break;
1585 
1586 		ret = 0;
1587 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1588 		extref = (struct btrfs_inode_extref *)ptr;
1589 		*ret_extref = extref;
1590 		if (found_off)
1591 			*found_off = found_key.offset;
1592 		break;
1593 	}
1594 
1595 	return ret;
1596 }
1597 
1598 /*
1599  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1600  * Elements of the path are separated by '/' and the path is guaranteed to be
1601  * 0-terminated. the path is only given within the current file system.
1602  * Therefore, it never starts with a '/'. the caller is responsible to provide
1603  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1604  * the start point of the resulting string is returned. this pointer is within
1605  * dest, normally.
1606  * in case the path buffer would overflow, the pointer is decremented further
1607  * as if output was written to the buffer, though no more output is actually
1608  * generated. that way, the caller can determine how much space would be
1609  * required for the path to fit into the buffer. in that case, the returned
1610  * value will be smaller than dest. callers must check this!
1611  */
1612 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1613 			u32 name_len, unsigned long name_off,
1614 			struct extent_buffer *eb_in, u64 parent,
1615 			char *dest, u32 size)
1616 {
1617 	int slot;
1618 	u64 next_inum;
1619 	int ret;
1620 	s64 bytes_left = ((s64)size) - 1;
1621 	struct extent_buffer *eb = eb_in;
1622 	struct btrfs_key found_key;
1623 	int leave_spinning = path->leave_spinning;
1624 	struct btrfs_inode_ref *iref;
1625 
1626 	if (bytes_left >= 0)
1627 		dest[bytes_left] = '\0';
1628 
1629 	path->leave_spinning = 1;
1630 	while (1) {
1631 		bytes_left -= name_len;
1632 		if (bytes_left >= 0)
1633 			read_extent_buffer(eb, dest + bytes_left,
1634 					   name_off, name_len);
1635 		if (eb != eb_in) {
1636 			if (!path->skip_locking)
1637 				btrfs_tree_read_unlock_blocking(eb);
1638 			free_extent_buffer(eb);
1639 		}
1640 		ret = btrfs_find_item(fs_root, path, parent, 0,
1641 				BTRFS_INODE_REF_KEY, &found_key);
1642 		if (ret > 0)
1643 			ret = -ENOENT;
1644 		if (ret)
1645 			break;
1646 
1647 		next_inum = found_key.offset;
1648 
1649 		/* regular exit ahead */
1650 		if (parent == next_inum)
1651 			break;
1652 
1653 		slot = path->slots[0];
1654 		eb = path->nodes[0];
1655 		/* make sure we can use eb after releasing the path */
1656 		if (eb != eb_in) {
1657 			if (!path->skip_locking)
1658 				btrfs_set_lock_blocking_read(eb);
1659 			path->nodes[0] = NULL;
1660 			path->locks[0] = 0;
1661 		}
1662 		btrfs_release_path(path);
1663 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1664 
1665 		name_len = btrfs_inode_ref_name_len(eb, iref);
1666 		name_off = (unsigned long)(iref + 1);
1667 
1668 		parent = next_inum;
1669 		--bytes_left;
1670 		if (bytes_left >= 0)
1671 			dest[bytes_left] = '/';
1672 	}
1673 
1674 	btrfs_release_path(path);
1675 	path->leave_spinning = leave_spinning;
1676 
1677 	if (ret)
1678 		return ERR_PTR(ret);
1679 
1680 	return dest + bytes_left;
1681 }
1682 
1683 /*
1684  * this makes the path point to (logical EXTENT_ITEM *)
1685  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1686  * tree blocks and <0 on error.
1687  */
1688 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1689 			struct btrfs_path *path, struct btrfs_key *found_key,
1690 			u64 *flags_ret)
1691 {
1692 	int ret;
1693 	u64 flags;
1694 	u64 size = 0;
1695 	u32 item_size;
1696 	const struct extent_buffer *eb;
1697 	struct btrfs_extent_item *ei;
1698 	struct btrfs_key key;
1699 
1700 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1701 		key.type = BTRFS_METADATA_ITEM_KEY;
1702 	else
1703 		key.type = BTRFS_EXTENT_ITEM_KEY;
1704 	key.objectid = logical;
1705 	key.offset = (u64)-1;
1706 
1707 	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1708 	if (ret < 0)
1709 		return ret;
1710 
1711 	ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1712 	if (ret) {
1713 		if (ret > 0)
1714 			ret = -ENOENT;
1715 		return ret;
1716 	}
1717 	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1718 	if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1719 		size = fs_info->nodesize;
1720 	else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1721 		size = found_key->offset;
1722 
1723 	if (found_key->objectid > logical ||
1724 	    found_key->objectid + size <= logical) {
1725 		btrfs_debug(fs_info,
1726 			"logical %llu is not within any extent", logical);
1727 		return -ENOENT;
1728 	}
1729 
1730 	eb = path->nodes[0];
1731 	item_size = btrfs_item_size_nr(eb, path->slots[0]);
1732 	BUG_ON(item_size < sizeof(*ei));
1733 
1734 	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1735 	flags = btrfs_extent_flags(eb, ei);
1736 
1737 	btrfs_debug(fs_info,
1738 		"logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
1739 		 logical, logical - found_key->objectid, found_key->objectid,
1740 		 found_key->offset, flags, item_size);
1741 
1742 	WARN_ON(!flags_ret);
1743 	if (flags_ret) {
1744 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1745 			*flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1746 		else if (flags & BTRFS_EXTENT_FLAG_DATA)
1747 			*flags_ret = BTRFS_EXTENT_FLAG_DATA;
1748 		else
1749 			BUG();
1750 		return 0;
1751 	}
1752 
1753 	return -EIO;
1754 }
1755 
1756 /*
1757  * helper function to iterate extent inline refs. ptr must point to a 0 value
1758  * for the first call and may be modified. it is used to track state.
1759  * if more refs exist, 0 is returned and the next call to
1760  * get_extent_inline_ref must pass the modified ptr parameter to get the
1761  * next ref. after the last ref was processed, 1 is returned.
1762  * returns <0 on error
1763  */
1764 static int get_extent_inline_ref(unsigned long *ptr,
1765 				 const struct extent_buffer *eb,
1766 				 const struct btrfs_key *key,
1767 				 const struct btrfs_extent_item *ei,
1768 				 u32 item_size,
1769 				 struct btrfs_extent_inline_ref **out_eiref,
1770 				 int *out_type)
1771 {
1772 	unsigned long end;
1773 	u64 flags;
1774 	struct btrfs_tree_block_info *info;
1775 
1776 	if (!*ptr) {
1777 		/* first call */
1778 		flags = btrfs_extent_flags(eb, ei);
1779 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1780 			if (key->type == BTRFS_METADATA_ITEM_KEY) {
1781 				/* a skinny metadata extent */
1782 				*out_eiref =
1783 				     (struct btrfs_extent_inline_ref *)(ei + 1);
1784 			} else {
1785 				WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1786 				info = (struct btrfs_tree_block_info *)(ei + 1);
1787 				*out_eiref =
1788 				   (struct btrfs_extent_inline_ref *)(info + 1);
1789 			}
1790 		} else {
1791 			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1792 		}
1793 		*ptr = (unsigned long)*out_eiref;
1794 		if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1795 			return -ENOENT;
1796 	}
1797 
1798 	end = (unsigned long)ei + item_size;
1799 	*out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1800 	*out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
1801 						     BTRFS_REF_TYPE_ANY);
1802 	if (*out_type == BTRFS_REF_TYPE_INVALID)
1803 		return -EUCLEAN;
1804 
1805 	*ptr += btrfs_extent_inline_ref_size(*out_type);
1806 	WARN_ON(*ptr > end);
1807 	if (*ptr == end)
1808 		return 1; /* last */
1809 
1810 	return 0;
1811 }
1812 
1813 /*
1814  * reads the tree block backref for an extent. tree level and root are returned
1815  * through out_level and out_root. ptr must point to a 0 value for the first
1816  * call and may be modified (see get_extent_inline_ref comment).
1817  * returns 0 if data was provided, 1 if there was no more data to provide or
1818  * <0 on error.
1819  */
1820 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1821 			    struct btrfs_key *key, struct btrfs_extent_item *ei,
1822 			    u32 item_size, u64 *out_root, u8 *out_level)
1823 {
1824 	int ret;
1825 	int type;
1826 	struct btrfs_extent_inline_ref *eiref;
1827 
1828 	if (*ptr == (unsigned long)-1)
1829 		return 1;
1830 
1831 	while (1) {
1832 		ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
1833 					      &eiref, &type);
1834 		if (ret < 0)
1835 			return ret;
1836 
1837 		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1838 		    type == BTRFS_SHARED_BLOCK_REF_KEY)
1839 			break;
1840 
1841 		if (ret == 1)
1842 			return 1;
1843 	}
1844 
1845 	/* we can treat both ref types equally here */
1846 	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1847 
1848 	if (key->type == BTRFS_EXTENT_ITEM_KEY) {
1849 		struct btrfs_tree_block_info *info;
1850 
1851 		info = (struct btrfs_tree_block_info *)(ei + 1);
1852 		*out_level = btrfs_tree_block_level(eb, info);
1853 	} else {
1854 		ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
1855 		*out_level = (u8)key->offset;
1856 	}
1857 
1858 	if (ret == 1)
1859 		*ptr = (unsigned long)-1;
1860 
1861 	return 0;
1862 }
1863 
1864 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
1865 			     struct extent_inode_elem *inode_list,
1866 			     u64 root, u64 extent_item_objectid,
1867 			     iterate_extent_inodes_t *iterate, void *ctx)
1868 {
1869 	struct extent_inode_elem *eie;
1870 	int ret = 0;
1871 
1872 	for (eie = inode_list; eie; eie = eie->next) {
1873 		btrfs_debug(fs_info,
1874 			    "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
1875 			    extent_item_objectid, eie->inum,
1876 			    eie->offset, root);
1877 		ret = iterate(eie->inum, eie->offset, root, ctx);
1878 		if (ret) {
1879 			btrfs_debug(fs_info,
1880 				    "stopping iteration for %llu due to ret=%d",
1881 				    extent_item_objectid, ret);
1882 			break;
1883 		}
1884 	}
1885 
1886 	return ret;
1887 }
1888 
1889 /*
1890  * calls iterate() for every inode that references the extent identified by
1891  * the given parameters.
1892  * when the iterator function returns a non-zero value, iteration stops.
1893  */
1894 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1895 				u64 extent_item_objectid, u64 extent_item_pos,
1896 				int search_commit_root,
1897 				iterate_extent_inodes_t *iterate, void *ctx,
1898 				bool ignore_offset)
1899 {
1900 	int ret;
1901 	struct btrfs_trans_handle *trans = NULL;
1902 	struct ulist *refs = NULL;
1903 	struct ulist *roots = NULL;
1904 	struct ulist_node *ref_node = NULL;
1905 	struct ulist_node *root_node = NULL;
1906 	struct seq_list tree_mod_seq_elem = SEQ_LIST_INIT(tree_mod_seq_elem);
1907 	struct ulist_iterator ref_uiter;
1908 	struct ulist_iterator root_uiter;
1909 
1910 	btrfs_debug(fs_info, "resolving all inodes for extent %llu",
1911 			extent_item_objectid);
1912 
1913 	if (!search_commit_root) {
1914 		trans = btrfs_attach_transaction(fs_info->extent_root);
1915 		if (IS_ERR(trans)) {
1916 			if (PTR_ERR(trans) != -ENOENT &&
1917 			    PTR_ERR(trans) != -EROFS)
1918 				return PTR_ERR(trans);
1919 			trans = NULL;
1920 		}
1921 	}
1922 
1923 	if (trans)
1924 		btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1925 	else
1926 		down_read(&fs_info->commit_root_sem);
1927 
1928 	ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1929 				   tree_mod_seq_elem.seq, &refs,
1930 				   &extent_item_pos, ignore_offset);
1931 	if (ret)
1932 		goto out;
1933 
1934 	ULIST_ITER_INIT(&ref_uiter);
1935 	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1936 		ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
1937 						tree_mod_seq_elem.seq, &roots,
1938 						ignore_offset);
1939 		if (ret)
1940 			break;
1941 		ULIST_ITER_INIT(&root_uiter);
1942 		while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1943 			btrfs_debug(fs_info,
1944 				    "root %llu references leaf %llu, data list %#llx",
1945 				    root_node->val, ref_node->val,
1946 				    ref_node->aux);
1947 			ret = iterate_leaf_refs(fs_info,
1948 						(struct extent_inode_elem *)
1949 						(uintptr_t)ref_node->aux,
1950 						root_node->val,
1951 						extent_item_objectid,
1952 						iterate, ctx);
1953 		}
1954 		ulist_free(roots);
1955 	}
1956 
1957 	free_leaf_list(refs);
1958 out:
1959 	if (trans) {
1960 		btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1961 		btrfs_end_transaction(trans);
1962 	} else {
1963 		up_read(&fs_info->commit_root_sem);
1964 	}
1965 
1966 	return ret;
1967 }
1968 
1969 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1970 				struct btrfs_path *path,
1971 				iterate_extent_inodes_t *iterate, void *ctx,
1972 				bool ignore_offset)
1973 {
1974 	int ret;
1975 	u64 extent_item_pos;
1976 	u64 flags = 0;
1977 	struct btrfs_key found_key;
1978 	int search_commit_root = path->search_commit_root;
1979 
1980 	ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1981 	btrfs_release_path(path);
1982 	if (ret < 0)
1983 		return ret;
1984 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1985 		return -EINVAL;
1986 
1987 	extent_item_pos = logical - found_key.objectid;
1988 	ret = iterate_extent_inodes(fs_info, found_key.objectid,
1989 					extent_item_pos, search_commit_root,
1990 					iterate, ctx, ignore_offset);
1991 
1992 	return ret;
1993 }
1994 
1995 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1996 			      struct extent_buffer *eb, void *ctx);
1997 
1998 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1999 			      struct btrfs_path *path,
2000 			      iterate_irefs_t *iterate, void *ctx)
2001 {
2002 	int ret = 0;
2003 	int slot;
2004 	u32 cur;
2005 	u32 len;
2006 	u32 name_len;
2007 	u64 parent = 0;
2008 	int found = 0;
2009 	struct extent_buffer *eb;
2010 	struct btrfs_item *item;
2011 	struct btrfs_inode_ref *iref;
2012 	struct btrfs_key found_key;
2013 
2014 	while (!ret) {
2015 		ret = btrfs_find_item(fs_root, path, inum,
2016 				parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2017 				&found_key);
2018 
2019 		if (ret < 0)
2020 			break;
2021 		if (ret) {
2022 			ret = found ? 0 : -ENOENT;
2023 			break;
2024 		}
2025 		++found;
2026 
2027 		parent = found_key.offset;
2028 		slot = path->slots[0];
2029 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
2030 		if (!eb) {
2031 			ret = -ENOMEM;
2032 			break;
2033 		}
2034 		btrfs_release_path(path);
2035 
2036 		item = btrfs_item_nr(slot);
2037 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2038 
2039 		for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
2040 			name_len = btrfs_inode_ref_name_len(eb, iref);
2041 			/* path must be released before calling iterate()! */
2042 			btrfs_debug(fs_root->fs_info,
2043 				"following ref at offset %u for inode %llu in tree %llu",
2044 				cur, found_key.objectid,
2045 				fs_root->root_key.objectid);
2046 			ret = iterate(parent, name_len,
2047 				      (unsigned long)(iref + 1), eb, ctx);
2048 			if (ret)
2049 				break;
2050 			len = sizeof(*iref) + name_len;
2051 			iref = (struct btrfs_inode_ref *)((char *)iref + len);
2052 		}
2053 		free_extent_buffer(eb);
2054 	}
2055 
2056 	btrfs_release_path(path);
2057 
2058 	return ret;
2059 }
2060 
2061 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
2062 				 struct btrfs_path *path,
2063 				 iterate_irefs_t *iterate, void *ctx)
2064 {
2065 	int ret;
2066 	int slot;
2067 	u64 offset = 0;
2068 	u64 parent;
2069 	int found = 0;
2070 	struct extent_buffer *eb;
2071 	struct btrfs_inode_extref *extref;
2072 	u32 item_size;
2073 	u32 cur_offset;
2074 	unsigned long ptr;
2075 
2076 	while (1) {
2077 		ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2078 					    &offset);
2079 		if (ret < 0)
2080 			break;
2081 		if (ret) {
2082 			ret = found ? 0 : -ENOENT;
2083 			break;
2084 		}
2085 		++found;
2086 
2087 		slot = path->slots[0];
2088 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
2089 		if (!eb) {
2090 			ret = -ENOMEM;
2091 			break;
2092 		}
2093 		btrfs_release_path(path);
2094 
2095 		item_size = btrfs_item_size_nr(eb, slot);
2096 		ptr = btrfs_item_ptr_offset(eb, slot);
2097 		cur_offset = 0;
2098 
2099 		while (cur_offset < item_size) {
2100 			u32 name_len;
2101 
2102 			extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2103 			parent = btrfs_inode_extref_parent(eb, extref);
2104 			name_len = btrfs_inode_extref_name_len(eb, extref);
2105 			ret = iterate(parent, name_len,
2106 				      (unsigned long)&extref->name, eb, ctx);
2107 			if (ret)
2108 				break;
2109 
2110 			cur_offset += btrfs_inode_extref_name_len(eb, extref);
2111 			cur_offset += sizeof(*extref);
2112 		}
2113 		free_extent_buffer(eb);
2114 
2115 		offset++;
2116 	}
2117 
2118 	btrfs_release_path(path);
2119 
2120 	return ret;
2121 }
2122 
2123 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
2124 			 struct btrfs_path *path, iterate_irefs_t *iterate,
2125 			 void *ctx)
2126 {
2127 	int ret;
2128 	int found_refs = 0;
2129 
2130 	ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
2131 	if (!ret)
2132 		++found_refs;
2133 	else if (ret != -ENOENT)
2134 		return ret;
2135 
2136 	ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
2137 	if (ret == -ENOENT && found_refs)
2138 		return 0;
2139 
2140 	return ret;
2141 }
2142 
2143 /*
2144  * returns 0 if the path could be dumped (probably truncated)
2145  * returns <0 in case of an error
2146  */
2147 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2148 			 struct extent_buffer *eb, void *ctx)
2149 {
2150 	struct inode_fs_paths *ipath = ctx;
2151 	char *fspath;
2152 	char *fspath_min;
2153 	int i = ipath->fspath->elem_cnt;
2154 	const int s_ptr = sizeof(char *);
2155 	u32 bytes_left;
2156 
2157 	bytes_left = ipath->fspath->bytes_left > s_ptr ?
2158 					ipath->fspath->bytes_left - s_ptr : 0;
2159 
2160 	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2161 	fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2162 				   name_off, eb, inum, fspath_min, bytes_left);
2163 	if (IS_ERR(fspath))
2164 		return PTR_ERR(fspath);
2165 
2166 	if (fspath > fspath_min) {
2167 		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2168 		++ipath->fspath->elem_cnt;
2169 		ipath->fspath->bytes_left = fspath - fspath_min;
2170 	} else {
2171 		++ipath->fspath->elem_missed;
2172 		ipath->fspath->bytes_missing += fspath_min - fspath;
2173 		ipath->fspath->bytes_left = 0;
2174 	}
2175 
2176 	return 0;
2177 }
2178 
2179 /*
2180  * this dumps all file system paths to the inode into the ipath struct, provided
2181  * is has been created large enough. each path is zero-terminated and accessed
2182  * from ipath->fspath->val[i].
2183  * when it returns, there are ipath->fspath->elem_cnt number of paths available
2184  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2185  * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2186  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2187  * have been needed to return all paths.
2188  */
2189 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2190 {
2191 	return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
2192 			     inode_to_path, ipath);
2193 }
2194 
2195 struct btrfs_data_container *init_data_container(u32 total_bytes)
2196 {
2197 	struct btrfs_data_container *data;
2198 	size_t alloc_bytes;
2199 
2200 	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2201 	data = kvmalloc(alloc_bytes, GFP_KERNEL);
2202 	if (!data)
2203 		return ERR_PTR(-ENOMEM);
2204 
2205 	if (total_bytes >= sizeof(*data)) {
2206 		data->bytes_left = total_bytes - sizeof(*data);
2207 		data->bytes_missing = 0;
2208 	} else {
2209 		data->bytes_missing = sizeof(*data) - total_bytes;
2210 		data->bytes_left = 0;
2211 	}
2212 
2213 	data->elem_cnt = 0;
2214 	data->elem_missed = 0;
2215 
2216 	return data;
2217 }
2218 
2219 /*
2220  * allocates space to return multiple file system paths for an inode.
2221  * total_bytes to allocate are passed, note that space usable for actual path
2222  * information will be total_bytes - sizeof(struct inode_fs_paths).
2223  * the returned pointer must be freed with free_ipath() in the end.
2224  */
2225 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2226 					struct btrfs_path *path)
2227 {
2228 	struct inode_fs_paths *ifp;
2229 	struct btrfs_data_container *fspath;
2230 
2231 	fspath = init_data_container(total_bytes);
2232 	if (IS_ERR(fspath))
2233 		return ERR_CAST(fspath);
2234 
2235 	ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2236 	if (!ifp) {
2237 		kvfree(fspath);
2238 		return ERR_PTR(-ENOMEM);
2239 	}
2240 
2241 	ifp->btrfs_path = path;
2242 	ifp->fspath = fspath;
2243 	ifp->fs_root = fs_root;
2244 
2245 	return ifp;
2246 }
2247 
2248 void free_ipath(struct inode_fs_paths *ipath)
2249 {
2250 	if (!ipath)
2251 		return;
2252 	kvfree(ipath->fspath);
2253 	kfree(ipath);
2254 }
2255