xref: /openbmc/linux/fs/btrfs/backref.c (revision d2999e1b)
1 /*
2  * Copyright (C) 2011 STRATO.  All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public
14  * License along with this program; if not, write to the
15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16  * Boston, MA 021110-1307, USA.
17  */
18 
19 #include <linux/vmalloc.h>
20 #include "ctree.h"
21 #include "disk-io.h"
22 #include "backref.h"
23 #include "ulist.h"
24 #include "transaction.h"
25 #include "delayed-ref.h"
26 #include "locking.h"
27 
28 struct extent_inode_elem {
29 	u64 inum;
30 	u64 offset;
31 	struct extent_inode_elem *next;
32 };
33 
34 static int check_extent_in_eb(struct btrfs_key *key, struct extent_buffer *eb,
35 				struct btrfs_file_extent_item *fi,
36 				u64 extent_item_pos,
37 				struct extent_inode_elem **eie)
38 {
39 	u64 offset = 0;
40 	struct extent_inode_elem *e;
41 
42 	if (!btrfs_file_extent_compression(eb, fi) &&
43 	    !btrfs_file_extent_encryption(eb, fi) &&
44 	    !btrfs_file_extent_other_encoding(eb, fi)) {
45 		u64 data_offset;
46 		u64 data_len;
47 
48 		data_offset = btrfs_file_extent_offset(eb, fi);
49 		data_len = btrfs_file_extent_num_bytes(eb, fi);
50 
51 		if (extent_item_pos < data_offset ||
52 		    extent_item_pos >= data_offset + data_len)
53 			return 1;
54 		offset = extent_item_pos - data_offset;
55 	}
56 
57 	e = kmalloc(sizeof(*e), GFP_NOFS);
58 	if (!e)
59 		return -ENOMEM;
60 
61 	e->next = *eie;
62 	e->inum = key->objectid;
63 	e->offset = key->offset + offset;
64 	*eie = e;
65 
66 	return 0;
67 }
68 
69 static void free_inode_elem_list(struct extent_inode_elem *eie)
70 {
71 	struct extent_inode_elem *eie_next;
72 
73 	for (; eie; eie = eie_next) {
74 		eie_next = eie->next;
75 		kfree(eie);
76 	}
77 }
78 
79 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
80 				u64 extent_item_pos,
81 				struct extent_inode_elem **eie)
82 {
83 	u64 disk_byte;
84 	struct btrfs_key key;
85 	struct btrfs_file_extent_item *fi;
86 	int slot;
87 	int nritems;
88 	int extent_type;
89 	int ret;
90 
91 	/*
92 	 * from the shared data ref, we only have the leaf but we need
93 	 * the key. thus, we must look into all items and see that we
94 	 * find one (some) with a reference to our extent item.
95 	 */
96 	nritems = btrfs_header_nritems(eb);
97 	for (slot = 0; slot < nritems; ++slot) {
98 		btrfs_item_key_to_cpu(eb, &key, slot);
99 		if (key.type != BTRFS_EXTENT_DATA_KEY)
100 			continue;
101 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
102 		extent_type = btrfs_file_extent_type(eb, fi);
103 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
104 			continue;
105 		/* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
106 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
107 		if (disk_byte != wanted_disk_byte)
108 			continue;
109 
110 		ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
111 		if (ret < 0)
112 			return ret;
113 	}
114 
115 	return 0;
116 }
117 
118 /*
119  * this structure records all encountered refs on the way up to the root
120  */
121 struct __prelim_ref {
122 	struct list_head list;
123 	u64 root_id;
124 	struct btrfs_key key_for_search;
125 	int level;
126 	int count;
127 	struct extent_inode_elem *inode_list;
128 	u64 parent;
129 	u64 wanted_disk_byte;
130 };
131 
132 static struct kmem_cache *btrfs_prelim_ref_cache;
133 
134 int __init btrfs_prelim_ref_init(void)
135 {
136 	btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
137 					sizeof(struct __prelim_ref),
138 					0,
139 					SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
140 					NULL);
141 	if (!btrfs_prelim_ref_cache)
142 		return -ENOMEM;
143 	return 0;
144 }
145 
146 void btrfs_prelim_ref_exit(void)
147 {
148 	if (btrfs_prelim_ref_cache)
149 		kmem_cache_destroy(btrfs_prelim_ref_cache);
150 }
151 
152 /*
153  * the rules for all callers of this function are:
154  * - obtaining the parent is the goal
155  * - if you add a key, you must know that it is a correct key
156  * - if you cannot add the parent or a correct key, then we will look into the
157  *   block later to set a correct key
158  *
159  * delayed refs
160  * ============
161  *        backref type | shared | indirect | shared | indirect
162  * information         |   tree |     tree |   data |     data
163  * --------------------+--------+----------+--------+----------
164  *      parent logical |    y   |     -    |    -   |     -
165  *      key to resolve |    -   |     y    |    y   |     y
166  *  tree block logical |    -   |     -    |    -   |     -
167  *  root for resolving |    y   |     y    |    y   |     y
168  *
169  * - column 1:       we've the parent -> done
170  * - column 2, 3, 4: we use the key to find the parent
171  *
172  * on disk refs (inline or keyed)
173  * ==============================
174  *        backref type | shared | indirect | shared | indirect
175  * information         |   tree |     tree |   data |     data
176  * --------------------+--------+----------+--------+----------
177  *      parent logical |    y   |     -    |    y   |     -
178  *      key to resolve |    -   |     -    |    -   |     y
179  *  tree block logical |    y   |     y    |    y   |     y
180  *  root for resolving |    -   |     y    |    y   |     y
181  *
182  * - column 1, 3: we've the parent -> done
183  * - column 2:    we take the first key from the block to find the parent
184  *                (see __add_missing_keys)
185  * - column 4:    we use the key to find the parent
186  *
187  * additional information that's available but not required to find the parent
188  * block might help in merging entries to gain some speed.
189  */
190 
191 static int __add_prelim_ref(struct list_head *head, u64 root_id,
192 			    struct btrfs_key *key, int level,
193 			    u64 parent, u64 wanted_disk_byte, int count,
194 			    gfp_t gfp_mask)
195 {
196 	struct __prelim_ref *ref;
197 
198 	if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
199 		return 0;
200 
201 	ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
202 	if (!ref)
203 		return -ENOMEM;
204 
205 	ref->root_id = root_id;
206 	if (key)
207 		ref->key_for_search = *key;
208 	else
209 		memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
210 
211 	ref->inode_list = NULL;
212 	ref->level = level;
213 	ref->count = count;
214 	ref->parent = parent;
215 	ref->wanted_disk_byte = wanted_disk_byte;
216 	list_add_tail(&ref->list, head);
217 
218 	return 0;
219 }
220 
221 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
222 			   struct ulist *parents, struct __prelim_ref *ref,
223 			   int level, u64 time_seq, const u64 *extent_item_pos,
224 			   u64 total_refs)
225 {
226 	int ret = 0;
227 	int slot;
228 	struct extent_buffer *eb;
229 	struct btrfs_key key;
230 	struct btrfs_key *key_for_search = &ref->key_for_search;
231 	struct btrfs_file_extent_item *fi;
232 	struct extent_inode_elem *eie = NULL, *old = NULL;
233 	u64 disk_byte;
234 	u64 wanted_disk_byte = ref->wanted_disk_byte;
235 	u64 count = 0;
236 
237 	if (level != 0) {
238 		eb = path->nodes[level];
239 		ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
240 		if (ret < 0)
241 			return ret;
242 		return 0;
243 	}
244 
245 	/*
246 	 * We normally enter this function with the path already pointing to
247 	 * the first item to check. But sometimes, we may enter it with
248 	 * slot==nritems. In that case, go to the next leaf before we continue.
249 	 */
250 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
251 		ret = btrfs_next_old_leaf(root, path, time_seq);
252 
253 	while (!ret && count < total_refs) {
254 		eb = path->nodes[0];
255 		slot = path->slots[0];
256 
257 		btrfs_item_key_to_cpu(eb, &key, slot);
258 
259 		if (key.objectid != key_for_search->objectid ||
260 		    key.type != BTRFS_EXTENT_DATA_KEY)
261 			break;
262 
263 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
264 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
265 
266 		if (disk_byte == wanted_disk_byte) {
267 			eie = NULL;
268 			old = NULL;
269 			count++;
270 			if (extent_item_pos) {
271 				ret = check_extent_in_eb(&key, eb, fi,
272 						*extent_item_pos,
273 						&eie);
274 				if (ret < 0)
275 					break;
276 			}
277 			if (ret > 0)
278 				goto next;
279 			ret = ulist_add_merge(parents, eb->start,
280 					      (uintptr_t)eie,
281 					      (u64 *)&old, GFP_NOFS);
282 			if (ret < 0)
283 				break;
284 			if (!ret && extent_item_pos) {
285 				while (old->next)
286 					old = old->next;
287 				old->next = eie;
288 			}
289 			eie = NULL;
290 		}
291 next:
292 		ret = btrfs_next_old_item(root, path, time_seq);
293 	}
294 
295 	if (ret > 0)
296 		ret = 0;
297 	else if (ret < 0)
298 		free_inode_elem_list(eie);
299 	return ret;
300 }
301 
302 /*
303  * resolve an indirect backref in the form (root_id, key, level)
304  * to a logical address
305  */
306 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
307 				  struct btrfs_path *path, u64 time_seq,
308 				  struct __prelim_ref *ref,
309 				  struct ulist *parents,
310 				  const u64 *extent_item_pos, u64 total_refs)
311 {
312 	struct btrfs_root *root;
313 	struct btrfs_key root_key;
314 	struct extent_buffer *eb;
315 	int ret = 0;
316 	int root_level;
317 	int level = ref->level;
318 	int index;
319 
320 	root_key.objectid = ref->root_id;
321 	root_key.type = BTRFS_ROOT_ITEM_KEY;
322 	root_key.offset = (u64)-1;
323 
324 	index = srcu_read_lock(&fs_info->subvol_srcu);
325 
326 	root = btrfs_read_fs_root_no_name(fs_info, &root_key);
327 	if (IS_ERR(root)) {
328 		srcu_read_unlock(&fs_info->subvol_srcu, index);
329 		ret = PTR_ERR(root);
330 		goto out;
331 	}
332 
333 	if (path->search_commit_root)
334 		root_level = btrfs_header_level(root->commit_root);
335 	else
336 		root_level = btrfs_old_root_level(root, time_seq);
337 
338 	if (root_level + 1 == level) {
339 		srcu_read_unlock(&fs_info->subvol_srcu, index);
340 		goto out;
341 	}
342 
343 	path->lowest_level = level;
344 	ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
345 
346 	/* root node has been locked, we can release @subvol_srcu safely here */
347 	srcu_read_unlock(&fs_info->subvol_srcu, index);
348 
349 	pr_debug("search slot in root %llu (level %d, ref count %d) returned "
350 		 "%d for key (%llu %u %llu)\n",
351 		 ref->root_id, level, ref->count, ret,
352 		 ref->key_for_search.objectid, ref->key_for_search.type,
353 		 ref->key_for_search.offset);
354 	if (ret < 0)
355 		goto out;
356 
357 	eb = path->nodes[level];
358 	while (!eb) {
359 		if (WARN_ON(!level)) {
360 			ret = 1;
361 			goto out;
362 		}
363 		level--;
364 		eb = path->nodes[level];
365 	}
366 
367 	ret = add_all_parents(root, path, parents, ref, level, time_seq,
368 			      extent_item_pos, total_refs);
369 out:
370 	path->lowest_level = 0;
371 	btrfs_release_path(path);
372 	return ret;
373 }
374 
375 /*
376  * resolve all indirect backrefs from the list
377  */
378 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
379 				   struct btrfs_path *path, u64 time_seq,
380 				   struct list_head *head,
381 				   const u64 *extent_item_pos, u64 total_refs)
382 {
383 	int err;
384 	int ret = 0;
385 	struct __prelim_ref *ref;
386 	struct __prelim_ref *ref_safe;
387 	struct __prelim_ref *new_ref;
388 	struct ulist *parents;
389 	struct ulist_node *node;
390 	struct ulist_iterator uiter;
391 
392 	parents = ulist_alloc(GFP_NOFS);
393 	if (!parents)
394 		return -ENOMEM;
395 
396 	/*
397 	 * _safe allows us to insert directly after the current item without
398 	 * iterating over the newly inserted items.
399 	 * we're also allowed to re-assign ref during iteration.
400 	 */
401 	list_for_each_entry_safe(ref, ref_safe, head, list) {
402 		if (ref->parent)	/* already direct */
403 			continue;
404 		if (ref->count == 0)
405 			continue;
406 		err = __resolve_indirect_ref(fs_info, path, time_seq, ref,
407 					     parents, extent_item_pos,
408 					     total_refs);
409 		/*
410 		 * we can only tolerate ENOENT,otherwise,we should catch error
411 		 * and return directly.
412 		 */
413 		if (err == -ENOENT) {
414 			continue;
415 		} else if (err) {
416 			ret = err;
417 			goto out;
418 		}
419 
420 		/* we put the first parent into the ref at hand */
421 		ULIST_ITER_INIT(&uiter);
422 		node = ulist_next(parents, &uiter);
423 		ref->parent = node ? node->val : 0;
424 		ref->inode_list = node ?
425 			(struct extent_inode_elem *)(uintptr_t)node->aux : NULL;
426 
427 		/* additional parents require new refs being added here */
428 		while ((node = ulist_next(parents, &uiter))) {
429 			new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
430 						   GFP_NOFS);
431 			if (!new_ref) {
432 				ret = -ENOMEM;
433 				goto out;
434 			}
435 			memcpy(new_ref, ref, sizeof(*ref));
436 			new_ref->parent = node->val;
437 			new_ref->inode_list = (struct extent_inode_elem *)
438 							(uintptr_t)node->aux;
439 			list_add(&new_ref->list, &ref->list);
440 		}
441 		ulist_reinit(parents);
442 	}
443 out:
444 	ulist_free(parents);
445 	return ret;
446 }
447 
448 static inline int ref_for_same_block(struct __prelim_ref *ref1,
449 				     struct __prelim_ref *ref2)
450 {
451 	if (ref1->level != ref2->level)
452 		return 0;
453 	if (ref1->root_id != ref2->root_id)
454 		return 0;
455 	if (ref1->key_for_search.type != ref2->key_for_search.type)
456 		return 0;
457 	if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
458 		return 0;
459 	if (ref1->key_for_search.offset != ref2->key_for_search.offset)
460 		return 0;
461 	if (ref1->parent != ref2->parent)
462 		return 0;
463 
464 	return 1;
465 }
466 
467 /*
468  * read tree blocks and add keys where required.
469  */
470 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
471 			      struct list_head *head)
472 {
473 	struct list_head *pos;
474 	struct extent_buffer *eb;
475 
476 	list_for_each(pos, head) {
477 		struct __prelim_ref *ref;
478 		ref = list_entry(pos, struct __prelim_ref, list);
479 
480 		if (ref->parent)
481 			continue;
482 		if (ref->key_for_search.type)
483 			continue;
484 		BUG_ON(!ref->wanted_disk_byte);
485 		eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
486 				     fs_info->tree_root->leafsize, 0);
487 		if (!eb || !extent_buffer_uptodate(eb)) {
488 			free_extent_buffer(eb);
489 			return -EIO;
490 		}
491 		btrfs_tree_read_lock(eb);
492 		if (btrfs_header_level(eb) == 0)
493 			btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
494 		else
495 			btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
496 		btrfs_tree_read_unlock(eb);
497 		free_extent_buffer(eb);
498 	}
499 	return 0;
500 }
501 
502 /*
503  * merge two lists of backrefs and adjust counts accordingly
504  *
505  * mode = 1: merge identical keys, if key is set
506  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
507  *           additionally, we could even add a key range for the blocks we
508  *           looked into to merge even more (-> replace unresolved refs by those
509  *           having a parent).
510  * mode = 2: merge identical parents
511  */
512 static void __merge_refs(struct list_head *head, int mode)
513 {
514 	struct list_head *pos1;
515 
516 	list_for_each(pos1, head) {
517 		struct list_head *n2;
518 		struct list_head *pos2;
519 		struct __prelim_ref *ref1;
520 
521 		ref1 = list_entry(pos1, struct __prelim_ref, list);
522 
523 		for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
524 		     pos2 = n2, n2 = pos2->next) {
525 			struct __prelim_ref *ref2;
526 			struct __prelim_ref *xchg;
527 			struct extent_inode_elem *eie;
528 
529 			ref2 = list_entry(pos2, struct __prelim_ref, list);
530 
531 			if (mode == 1) {
532 				if (!ref_for_same_block(ref1, ref2))
533 					continue;
534 				if (!ref1->parent && ref2->parent) {
535 					xchg = ref1;
536 					ref1 = ref2;
537 					ref2 = xchg;
538 				}
539 			} else {
540 				if (ref1->parent != ref2->parent)
541 					continue;
542 			}
543 
544 			eie = ref1->inode_list;
545 			while (eie && eie->next)
546 				eie = eie->next;
547 			if (eie)
548 				eie->next = ref2->inode_list;
549 			else
550 				ref1->inode_list = ref2->inode_list;
551 			ref1->count += ref2->count;
552 
553 			list_del(&ref2->list);
554 			kmem_cache_free(btrfs_prelim_ref_cache, ref2);
555 		}
556 
557 	}
558 }
559 
560 /*
561  * add all currently queued delayed refs from this head whose seq nr is
562  * smaller or equal that seq to the list
563  */
564 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
565 			      struct list_head *prefs, u64 *total_refs)
566 {
567 	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
568 	struct rb_node *n = &head->node.rb_node;
569 	struct btrfs_key key;
570 	struct btrfs_key op_key = {0};
571 	int sgn;
572 	int ret = 0;
573 
574 	if (extent_op && extent_op->update_key)
575 		btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
576 
577 	spin_lock(&head->lock);
578 	n = rb_first(&head->ref_root);
579 	while (n) {
580 		struct btrfs_delayed_ref_node *node;
581 		node = rb_entry(n, struct btrfs_delayed_ref_node,
582 				rb_node);
583 		n = rb_next(n);
584 		if (node->seq > seq)
585 			continue;
586 
587 		switch (node->action) {
588 		case BTRFS_ADD_DELAYED_EXTENT:
589 		case BTRFS_UPDATE_DELAYED_HEAD:
590 			WARN_ON(1);
591 			continue;
592 		case BTRFS_ADD_DELAYED_REF:
593 			sgn = 1;
594 			break;
595 		case BTRFS_DROP_DELAYED_REF:
596 			sgn = -1;
597 			break;
598 		default:
599 			BUG_ON(1);
600 		}
601 		*total_refs += (node->ref_mod * sgn);
602 		switch (node->type) {
603 		case BTRFS_TREE_BLOCK_REF_KEY: {
604 			struct btrfs_delayed_tree_ref *ref;
605 
606 			ref = btrfs_delayed_node_to_tree_ref(node);
607 			ret = __add_prelim_ref(prefs, ref->root, &op_key,
608 					       ref->level + 1, 0, node->bytenr,
609 					       node->ref_mod * sgn, GFP_ATOMIC);
610 			break;
611 		}
612 		case BTRFS_SHARED_BLOCK_REF_KEY: {
613 			struct btrfs_delayed_tree_ref *ref;
614 
615 			ref = btrfs_delayed_node_to_tree_ref(node);
616 			ret = __add_prelim_ref(prefs, ref->root, NULL,
617 					       ref->level + 1, ref->parent,
618 					       node->bytenr,
619 					       node->ref_mod * sgn, GFP_ATOMIC);
620 			break;
621 		}
622 		case BTRFS_EXTENT_DATA_REF_KEY: {
623 			struct btrfs_delayed_data_ref *ref;
624 			ref = btrfs_delayed_node_to_data_ref(node);
625 
626 			key.objectid = ref->objectid;
627 			key.type = BTRFS_EXTENT_DATA_KEY;
628 			key.offset = ref->offset;
629 			ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
630 					       node->bytenr,
631 					       node->ref_mod * sgn, GFP_ATOMIC);
632 			break;
633 		}
634 		case BTRFS_SHARED_DATA_REF_KEY: {
635 			struct btrfs_delayed_data_ref *ref;
636 
637 			ref = btrfs_delayed_node_to_data_ref(node);
638 
639 			key.objectid = ref->objectid;
640 			key.type = BTRFS_EXTENT_DATA_KEY;
641 			key.offset = ref->offset;
642 			ret = __add_prelim_ref(prefs, ref->root, &key, 0,
643 					       ref->parent, node->bytenr,
644 					       node->ref_mod * sgn, GFP_ATOMIC);
645 			break;
646 		}
647 		default:
648 			WARN_ON(1);
649 		}
650 		if (ret)
651 			break;
652 	}
653 	spin_unlock(&head->lock);
654 	return ret;
655 }
656 
657 /*
658  * add all inline backrefs for bytenr to the list
659  */
660 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
661 			     struct btrfs_path *path, u64 bytenr,
662 			     int *info_level, struct list_head *prefs,
663 			     u64 *total_refs)
664 {
665 	int ret = 0;
666 	int slot;
667 	struct extent_buffer *leaf;
668 	struct btrfs_key key;
669 	struct btrfs_key found_key;
670 	unsigned long ptr;
671 	unsigned long end;
672 	struct btrfs_extent_item *ei;
673 	u64 flags;
674 	u64 item_size;
675 
676 	/*
677 	 * enumerate all inline refs
678 	 */
679 	leaf = path->nodes[0];
680 	slot = path->slots[0];
681 
682 	item_size = btrfs_item_size_nr(leaf, slot);
683 	BUG_ON(item_size < sizeof(*ei));
684 
685 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
686 	flags = btrfs_extent_flags(leaf, ei);
687 	*total_refs += btrfs_extent_refs(leaf, ei);
688 	btrfs_item_key_to_cpu(leaf, &found_key, slot);
689 
690 	ptr = (unsigned long)(ei + 1);
691 	end = (unsigned long)ei + item_size;
692 
693 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
694 	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
695 		struct btrfs_tree_block_info *info;
696 
697 		info = (struct btrfs_tree_block_info *)ptr;
698 		*info_level = btrfs_tree_block_level(leaf, info);
699 		ptr += sizeof(struct btrfs_tree_block_info);
700 		BUG_ON(ptr > end);
701 	} else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
702 		*info_level = found_key.offset;
703 	} else {
704 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
705 	}
706 
707 	while (ptr < end) {
708 		struct btrfs_extent_inline_ref *iref;
709 		u64 offset;
710 		int type;
711 
712 		iref = (struct btrfs_extent_inline_ref *)ptr;
713 		type = btrfs_extent_inline_ref_type(leaf, iref);
714 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
715 
716 		switch (type) {
717 		case BTRFS_SHARED_BLOCK_REF_KEY:
718 			ret = __add_prelim_ref(prefs, 0, NULL,
719 						*info_level + 1, offset,
720 						bytenr, 1, GFP_NOFS);
721 			break;
722 		case BTRFS_SHARED_DATA_REF_KEY: {
723 			struct btrfs_shared_data_ref *sdref;
724 			int count;
725 
726 			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
727 			count = btrfs_shared_data_ref_count(leaf, sdref);
728 			ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
729 					       bytenr, count, GFP_NOFS);
730 			break;
731 		}
732 		case BTRFS_TREE_BLOCK_REF_KEY:
733 			ret = __add_prelim_ref(prefs, offset, NULL,
734 					       *info_level + 1, 0,
735 					       bytenr, 1, GFP_NOFS);
736 			break;
737 		case BTRFS_EXTENT_DATA_REF_KEY: {
738 			struct btrfs_extent_data_ref *dref;
739 			int count;
740 			u64 root;
741 
742 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
743 			count = btrfs_extent_data_ref_count(leaf, dref);
744 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
745 								      dref);
746 			key.type = BTRFS_EXTENT_DATA_KEY;
747 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
748 			root = btrfs_extent_data_ref_root(leaf, dref);
749 			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
750 					       bytenr, count, GFP_NOFS);
751 			break;
752 		}
753 		default:
754 			WARN_ON(1);
755 		}
756 		if (ret)
757 			return ret;
758 		ptr += btrfs_extent_inline_ref_size(type);
759 	}
760 
761 	return 0;
762 }
763 
764 /*
765  * add all non-inline backrefs for bytenr to the list
766  */
767 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
768 			    struct btrfs_path *path, u64 bytenr,
769 			    int info_level, struct list_head *prefs)
770 {
771 	struct btrfs_root *extent_root = fs_info->extent_root;
772 	int ret;
773 	int slot;
774 	struct extent_buffer *leaf;
775 	struct btrfs_key key;
776 
777 	while (1) {
778 		ret = btrfs_next_item(extent_root, path);
779 		if (ret < 0)
780 			break;
781 		if (ret) {
782 			ret = 0;
783 			break;
784 		}
785 
786 		slot = path->slots[0];
787 		leaf = path->nodes[0];
788 		btrfs_item_key_to_cpu(leaf, &key, slot);
789 
790 		if (key.objectid != bytenr)
791 			break;
792 		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
793 			continue;
794 		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
795 			break;
796 
797 		switch (key.type) {
798 		case BTRFS_SHARED_BLOCK_REF_KEY:
799 			ret = __add_prelim_ref(prefs, 0, NULL,
800 						info_level + 1, key.offset,
801 						bytenr, 1, GFP_NOFS);
802 			break;
803 		case BTRFS_SHARED_DATA_REF_KEY: {
804 			struct btrfs_shared_data_ref *sdref;
805 			int count;
806 
807 			sdref = btrfs_item_ptr(leaf, slot,
808 					      struct btrfs_shared_data_ref);
809 			count = btrfs_shared_data_ref_count(leaf, sdref);
810 			ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
811 						bytenr, count, GFP_NOFS);
812 			break;
813 		}
814 		case BTRFS_TREE_BLOCK_REF_KEY:
815 			ret = __add_prelim_ref(prefs, key.offset, NULL,
816 					       info_level + 1, 0,
817 					       bytenr, 1, GFP_NOFS);
818 			break;
819 		case BTRFS_EXTENT_DATA_REF_KEY: {
820 			struct btrfs_extent_data_ref *dref;
821 			int count;
822 			u64 root;
823 
824 			dref = btrfs_item_ptr(leaf, slot,
825 					      struct btrfs_extent_data_ref);
826 			count = btrfs_extent_data_ref_count(leaf, dref);
827 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
828 								      dref);
829 			key.type = BTRFS_EXTENT_DATA_KEY;
830 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
831 			root = btrfs_extent_data_ref_root(leaf, dref);
832 			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
833 					       bytenr, count, GFP_NOFS);
834 			break;
835 		}
836 		default:
837 			WARN_ON(1);
838 		}
839 		if (ret)
840 			return ret;
841 
842 	}
843 
844 	return ret;
845 }
846 
847 /*
848  * this adds all existing backrefs (inline backrefs, backrefs and delayed
849  * refs) for the given bytenr to the refs list, merges duplicates and resolves
850  * indirect refs to their parent bytenr.
851  * When roots are found, they're added to the roots list
852  *
853  * FIXME some caching might speed things up
854  */
855 static int find_parent_nodes(struct btrfs_trans_handle *trans,
856 			     struct btrfs_fs_info *fs_info, u64 bytenr,
857 			     u64 time_seq, struct ulist *refs,
858 			     struct ulist *roots, const u64 *extent_item_pos)
859 {
860 	struct btrfs_key key;
861 	struct btrfs_path *path;
862 	struct btrfs_delayed_ref_root *delayed_refs = NULL;
863 	struct btrfs_delayed_ref_head *head;
864 	int info_level = 0;
865 	int ret;
866 	struct list_head prefs_delayed;
867 	struct list_head prefs;
868 	struct __prelim_ref *ref;
869 	struct extent_inode_elem *eie = NULL;
870 	u64 total_refs = 0;
871 
872 	INIT_LIST_HEAD(&prefs);
873 	INIT_LIST_HEAD(&prefs_delayed);
874 
875 	key.objectid = bytenr;
876 	key.offset = (u64)-1;
877 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
878 		key.type = BTRFS_METADATA_ITEM_KEY;
879 	else
880 		key.type = BTRFS_EXTENT_ITEM_KEY;
881 
882 	path = btrfs_alloc_path();
883 	if (!path)
884 		return -ENOMEM;
885 	if (!trans) {
886 		path->search_commit_root = 1;
887 		path->skip_locking = 1;
888 	}
889 
890 	/*
891 	 * grab both a lock on the path and a lock on the delayed ref head.
892 	 * We need both to get a consistent picture of how the refs look
893 	 * at a specified point in time
894 	 */
895 again:
896 	head = NULL;
897 
898 	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
899 	if (ret < 0)
900 		goto out;
901 	BUG_ON(ret == 0);
902 
903 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
904 	if (trans && likely(trans->type != __TRANS_DUMMY)) {
905 #else
906 	if (trans) {
907 #endif
908 		/*
909 		 * look if there are updates for this ref queued and lock the
910 		 * head
911 		 */
912 		delayed_refs = &trans->transaction->delayed_refs;
913 		spin_lock(&delayed_refs->lock);
914 		head = btrfs_find_delayed_ref_head(trans, bytenr);
915 		if (head) {
916 			if (!mutex_trylock(&head->mutex)) {
917 				atomic_inc(&head->node.refs);
918 				spin_unlock(&delayed_refs->lock);
919 
920 				btrfs_release_path(path);
921 
922 				/*
923 				 * Mutex was contended, block until it's
924 				 * released and try again
925 				 */
926 				mutex_lock(&head->mutex);
927 				mutex_unlock(&head->mutex);
928 				btrfs_put_delayed_ref(&head->node);
929 				goto again;
930 			}
931 			spin_unlock(&delayed_refs->lock);
932 			ret = __add_delayed_refs(head, time_seq,
933 						 &prefs_delayed, &total_refs);
934 			mutex_unlock(&head->mutex);
935 			if (ret)
936 				goto out;
937 		} else {
938 			spin_unlock(&delayed_refs->lock);
939 		}
940 	}
941 
942 	if (path->slots[0]) {
943 		struct extent_buffer *leaf;
944 		int slot;
945 
946 		path->slots[0]--;
947 		leaf = path->nodes[0];
948 		slot = path->slots[0];
949 		btrfs_item_key_to_cpu(leaf, &key, slot);
950 		if (key.objectid == bytenr &&
951 		    (key.type == BTRFS_EXTENT_ITEM_KEY ||
952 		     key.type == BTRFS_METADATA_ITEM_KEY)) {
953 			ret = __add_inline_refs(fs_info, path, bytenr,
954 						&info_level, &prefs,
955 						&total_refs);
956 			if (ret)
957 				goto out;
958 			ret = __add_keyed_refs(fs_info, path, bytenr,
959 					       info_level, &prefs);
960 			if (ret)
961 				goto out;
962 		}
963 	}
964 	btrfs_release_path(path);
965 
966 	list_splice_init(&prefs_delayed, &prefs);
967 
968 	ret = __add_missing_keys(fs_info, &prefs);
969 	if (ret)
970 		goto out;
971 
972 	__merge_refs(&prefs, 1);
973 
974 	ret = __resolve_indirect_refs(fs_info, path, time_seq, &prefs,
975 				      extent_item_pos, total_refs);
976 	if (ret)
977 		goto out;
978 
979 	__merge_refs(&prefs, 2);
980 
981 	while (!list_empty(&prefs)) {
982 		ref = list_first_entry(&prefs, struct __prelim_ref, list);
983 		WARN_ON(ref->count < 0);
984 		if (roots && ref->count && ref->root_id && ref->parent == 0) {
985 			/* no parent == root of tree */
986 			ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
987 			if (ret < 0)
988 				goto out;
989 		}
990 		if (ref->count && ref->parent) {
991 			if (extent_item_pos && !ref->inode_list &&
992 			    ref->level == 0) {
993 				u32 bsz;
994 				struct extent_buffer *eb;
995 				bsz = btrfs_level_size(fs_info->extent_root,
996 							ref->level);
997 				eb = read_tree_block(fs_info->extent_root,
998 							   ref->parent, bsz, 0);
999 				if (!eb || !extent_buffer_uptodate(eb)) {
1000 					free_extent_buffer(eb);
1001 					ret = -EIO;
1002 					goto out;
1003 				}
1004 				ret = find_extent_in_eb(eb, bytenr,
1005 							*extent_item_pos, &eie);
1006 				free_extent_buffer(eb);
1007 				if (ret < 0)
1008 					goto out;
1009 				ref->inode_list = eie;
1010 			}
1011 			ret = ulist_add_merge(refs, ref->parent,
1012 					      (uintptr_t)ref->inode_list,
1013 					      (u64 *)&eie, GFP_NOFS);
1014 			if (ret < 0)
1015 				goto out;
1016 			if (!ret && extent_item_pos) {
1017 				/*
1018 				 * we've recorded that parent, so we must extend
1019 				 * its inode list here
1020 				 */
1021 				BUG_ON(!eie);
1022 				while (eie->next)
1023 					eie = eie->next;
1024 				eie->next = ref->inode_list;
1025 			}
1026 			eie = NULL;
1027 		}
1028 		list_del(&ref->list);
1029 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1030 	}
1031 
1032 out:
1033 	btrfs_free_path(path);
1034 	while (!list_empty(&prefs)) {
1035 		ref = list_first_entry(&prefs, struct __prelim_ref, list);
1036 		list_del(&ref->list);
1037 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1038 	}
1039 	while (!list_empty(&prefs_delayed)) {
1040 		ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
1041 				       list);
1042 		list_del(&ref->list);
1043 		kmem_cache_free(btrfs_prelim_ref_cache, ref);
1044 	}
1045 	if (ret < 0)
1046 		free_inode_elem_list(eie);
1047 	return ret;
1048 }
1049 
1050 static void free_leaf_list(struct ulist *blocks)
1051 {
1052 	struct ulist_node *node = NULL;
1053 	struct extent_inode_elem *eie;
1054 	struct ulist_iterator uiter;
1055 
1056 	ULIST_ITER_INIT(&uiter);
1057 	while ((node = ulist_next(blocks, &uiter))) {
1058 		if (!node->aux)
1059 			continue;
1060 		eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
1061 		free_inode_elem_list(eie);
1062 		node->aux = 0;
1063 	}
1064 
1065 	ulist_free(blocks);
1066 }
1067 
1068 /*
1069  * Finds all leafs with a reference to the specified combination of bytenr and
1070  * offset. key_list_head will point to a list of corresponding keys (caller must
1071  * free each list element). The leafs will be stored in the leafs ulist, which
1072  * must be freed with ulist_free.
1073  *
1074  * returns 0 on success, <0 on error
1075  */
1076 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1077 				struct btrfs_fs_info *fs_info, u64 bytenr,
1078 				u64 time_seq, struct ulist **leafs,
1079 				const u64 *extent_item_pos)
1080 {
1081 	int ret;
1082 
1083 	*leafs = ulist_alloc(GFP_NOFS);
1084 	if (!*leafs)
1085 		return -ENOMEM;
1086 
1087 	ret = find_parent_nodes(trans, fs_info, bytenr,
1088 				time_seq, *leafs, NULL, extent_item_pos);
1089 	if (ret < 0 && ret != -ENOENT) {
1090 		free_leaf_list(*leafs);
1091 		return ret;
1092 	}
1093 
1094 	return 0;
1095 }
1096 
1097 /*
1098  * walk all backrefs for a given extent to find all roots that reference this
1099  * extent. Walking a backref means finding all extents that reference this
1100  * extent and in turn walk the backrefs of those, too. Naturally this is a
1101  * recursive process, but here it is implemented in an iterative fashion: We
1102  * find all referencing extents for the extent in question and put them on a
1103  * list. In turn, we find all referencing extents for those, further appending
1104  * to the list. The way we iterate the list allows adding more elements after
1105  * the current while iterating. The process stops when we reach the end of the
1106  * list. Found roots are added to the roots list.
1107  *
1108  * returns 0 on success, < 0 on error.
1109  */
1110 static int __btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1111 				  struct btrfs_fs_info *fs_info, u64 bytenr,
1112 				  u64 time_seq, struct ulist **roots)
1113 {
1114 	struct ulist *tmp;
1115 	struct ulist_node *node = NULL;
1116 	struct ulist_iterator uiter;
1117 	int ret;
1118 
1119 	tmp = ulist_alloc(GFP_NOFS);
1120 	if (!tmp)
1121 		return -ENOMEM;
1122 	*roots = ulist_alloc(GFP_NOFS);
1123 	if (!*roots) {
1124 		ulist_free(tmp);
1125 		return -ENOMEM;
1126 	}
1127 
1128 	ULIST_ITER_INIT(&uiter);
1129 	while (1) {
1130 		ret = find_parent_nodes(trans, fs_info, bytenr,
1131 					time_seq, tmp, *roots, NULL);
1132 		if (ret < 0 && ret != -ENOENT) {
1133 			ulist_free(tmp);
1134 			ulist_free(*roots);
1135 			return ret;
1136 		}
1137 		node = ulist_next(tmp, &uiter);
1138 		if (!node)
1139 			break;
1140 		bytenr = node->val;
1141 		cond_resched();
1142 	}
1143 
1144 	ulist_free(tmp);
1145 	return 0;
1146 }
1147 
1148 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1149 			 struct btrfs_fs_info *fs_info, u64 bytenr,
1150 			 u64 time_seq, struct ulist **roots)
1151 {
1152 	int ret;
1153 
1154 	if (!trans)
1155 		down_read(&fs_info->commit_root_sem);
1156 	ret = __btrfs_find_all_roots(trans, fs_info, bytenr, time_seq, roots);
1157 	if (!trans)
1158 		up_read(&fs_info->commit_root_sem);
1159 	return ret;
1160 }
1161 
1162 /*
1163  * this makes the path point to (inum INODE_ITEM ioff)
1164  */
1165 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1166 			struct btrfs_path *path)
1167 {
1168 	struct btrfs_key key;
1169 	return btrfs_find_item(fs_root, path, inum, ioff,
1170 			BTRFS_INODE_ITEM_KEY, &key);
1171 }
1172 
1173 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1174 				struct btrfs_path *path,
1175 				struct btrfs_key *found_key)
1176 {
1177 	return btrfs_find_item(fs_root, path, inum, ioff,
1178 			BTRFS_INODE_REF_KEY, found_key);
1179 }
1180 
1181 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1182 			  u64 start_off, struct btrfs_path *path,
1183 			  struct btrfs_inode_extref **ret_extref,
1184 			  u64 *found_off)
1185 {
1186 	int ret, slot;
1187 	struct btrfs_key key;
1188 	struct btrfs_key found_key;
1189 	struct btrfs_inode_extref *extref;
1190 	struct extent_buffer *leaf;
1191 	unsigned long ptr;
1192 
1193 	key.objectid = inode_objectid;
1194 	btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1195 	key.offset = start_off;
1196 
1197 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1198 	if (ret < 0)
1199 		return ret;
1200 
1201 	while (1) {
1202 		leaf = path->nodes[0];
1203 		slot = path->slots[0];
1204 		if (slot >= btrfs_header_nritems(leaf)) {
1205 			/*
1206 			 * If the item at offset is not found,
1207 			 * btrfs_search_slot will point us to the slot
1208 			 * where it should be inserted. In our case
1209 			 * that will be the slot directly before the
1210 			 * next INODE_REF_KEY_V2 item. In the case
1211 			 * that we're pointing to the last slot in a
1212 			 * leaf, we must move one leaf over.
1213 			 */
1214 			ret = btrfs_next_leaf(root, path);
1215 			if (ret) {
1216 				if (ret >= 1)
1217 					ret = -ENOENT;
1218 				break;
1219 			}
1220 			continue;
1221 		}
1222 
1223 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1224 
1225 		/*
1226 		 * Check that we're still looking at an extended ref key for
1227 		 * this particular objectid. If we have different
1228 		 * objectid or type then there are no more to be found
1229 		 * in the tree and we can exit.
1230 		 */
1231 		ret = -ENOENT;
1232 		if (found_key.objectid != inode_objectid)
1233 			break;
1234 		if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1235 			break;
1236 
1237 		ret = 0;
1238 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1239 		extref = (struct btrfs_inode_extref *)ptr;
1240 		*ret_extref = extref;
1241 		if (found_off)
1242 			*found_off = found_key.offset;
1243 		break;
1244 	}
1245 
1246 	return ret;
1247 }
1248 
1249 /*
1250  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1251  * Elements of the path are separated by '/' and the path is guaranteed to be
1252  * 0-terminated. the path is only given within the current file system.
1253  * Therefore, it never starts with a '/'. the caller is responsible to provide
1254  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1255  * the start point of the resulting string is returned. this pointer is within
1256  * dest, normally.
1257  * in case the path buffer would overflow, the pointer is decremented further
1258  * as if output was written to the buffer, though no more output is actually
1259  * generated. that way, the caller can determine how much space would be
1260  * required for the path to fit into the buffer. in that case, the returned
1261  * value will be smaller than dest. callers must check this!
1262  */
1263 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1264 			u32 name_len, unsigned long name_off,
1265 			struct extent_buffer *eb_in, u64 parent,
1266 			char *dest, u32 size)
1267 {
1268 	int slot;
1269 	u64 next_inum;
1270 	int ret;
1271 	s64 bytes_left = ((s64)size) - 1;
1272 	struct extent_buffer *eb = eb_in;
1273 	struct btrfs_key found_key;
1274 	int leave_spinning = path->leave_spinning;
1275 	struct btrfs_inode_ref *iref;
1276 
1277 	if (bytes_left >= 0)
1278 		dest[bytes_left] = '\0';
1279 
1280 	path->leave_spinning = 1;
1281 	while (1) {
1282 		bytes_left -= name_len;
1283 		if (bytes_left >= 0)
1284 			read_extent_buffer(eb, dest + bytes_left,
1285 					   name_off, name_len);
1286 		if (eb != eb_in) {
1287 			btrfs_tree_read_unlock_blocking(eb);
1288 			free_extent_buffer(eb);
1289 		}
1290 		ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1291 		if (ret > 0)
1292 			ret = -ENOENT;
1293 		if (ret)
1294 			break;
1295 
1296 		next_inum = found_key.offset;
1297 
1298 		/* regular exit ahead */
1299 		if (parent == next_inum)
1300 			break;
1301 
1302 		slot = path->slots[0];
1303 		eb = path->nodes[0];
1304 		/* make sure we can use eb after releasing the path */
1305 		if (eb != eb_in) {
1306 			atomic_inc(&eb->refs);
1307 			btrfs_tree_read_lock(eb);
1308 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1309 		}
1310 		btrfs_release_path(path);
1311 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1312 
1313 		name_len = btrfs_inode_ref_name_len(eb, iref);
1314 		name_off = (unsigned long)(iref + 1);
1315 
1316 		parent = next_inum;
1317 		--bytes_left;
1318 		if (bytes_left >= 0)
1319 			dest[bytes_left] = '/';
1320 	}
1321 
1322 	btrfs_release_path(path);
1323 	path->leave_spinning = leave_spinning;
1324 
1325 	if (ret)
1326 		return ERR_PTR(ret);
1327 
1328 	return dest + bytes_left;
1329 }
1330 
1331 /*
1332  * this makes the path point to (logical EXTENT_ITEM *)
1333  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1334  * tree blocks and <0 on error.
1335  */
1336 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1337 			struct btrfs_path *path, struct btrfs_key *found_key,
1338 			u64 *flags_ret)
1339 {
1340 	int ret;
1341 	u64 flags;
1342 	u64 size = 0;
1343 	u32 item_size;
1344 	struct extent_buffer *eb;
1345 	struct btrfs_extent_item *ei;
1346 	struct btrfs_key key;
1347 
1348 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1349 		key.type = BTRFS_METADATA_ITEM_KEY;
1350 	else
1351 		key.type = BTRFS_EXTENT_ITEM_KEY;
1352 	key.objectid = logical;
1353 	key.offset = (u64)-1;
1354 
1355 	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1356 	if (ret < 0)
1357 		return ret;
1358 
1359 	ret = btrfs_previous_extent_item(fs_info->extent_root, path, 0);
1360 	if (ret) {
1361 		if (ret > 0)
1362 			ret = -ENOENT;
1363 		return ret;
1364 	}
1365 	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1366 	if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1367 		size = fs_info->extent_root->leafsize;
1368 	else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1369 		size = found_key->offset;
1370 
1371 	if (found_key->objectid > logical ||
1372 	    found_key->objectid + size <= logical) {
1373 		pr_debug("logical %llu is not within any extent\n", logical);
1374 		return -ENOENT;
1375 	}
1376 
1377 	eb = path->nodes[0];
1378 	item_size = btrfs_item_size_nr(eb, path->slots[0]);
1379 	BUG_ON(item_size < sizeof(*ei));
1380 
1381 	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1382 	flags = btrfs_extent_flags(eb, ei);
1383 
1384 	pr_debug("logical %llu is at position %llu within the extent (%llu "
1385 		 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1386 		 logical, logical - found_key->objectid, found_key->objectid,
1387 		 found_key->offset, flags, item_size);
1388 
1389 	WARN_ON(!flags_ret);
1390 	if (flags_ret) {
1391 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1392 			*flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1393 		else if (flags & BTRFS_EXTENT_FLAG_DATA)
1394 			*flags_ret = BTRFS_EXTENT_FLAG_DATA;
1395 		else
1396 			BUG_ON(1);
1397 		return 0;
1398 	}
1399 
1400 	return -EIO;
1401 }
1402 
1403 /*
1404  * helper function to iterate extent inline refs. ptr must point to a 0 value
1405  * for the first call and may be modified. it is used to track state.
1406  * if more refs exist, 0 is returned and the next call to
1407  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1408  * next ref. after the last ref was processed, 1 is returned.
1409  * returns <0 on error
1410  */
1411 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1412 				   struct btrfs_key *key,
1413 				   struct btrfs_extent_item *ei, u32 item_size,
1414 				   struct btrfs_extent_inline_ref **out_eiref,
1415 				   int *out_type)
1416 {
1417 	unsigned long end;
1418 	u64 flags;
1419 	struct btrfs_tree_block_info *info;
1420 
1421 	if (!*ptr) {
1422 		/* first call */
1423 		flags = btrfs_extent_flags(eb, ei);
1424 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1425 			if (key->type == BTRFS_METADATA_ITEM_KEY) {
1426 				/* a skinny metadata extent */
1427 				*out_eiref =
1428 				     (struct btrfs_extent_inline_ref *)(ei + 1);
1429 			} else {
1430 				WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
1431 				info = (struct btrfs_tree_block_info *)(ei + 1);
1432 				*out_eiref =
1433 				   (struct btrfs_extent_inline_ref *)(info + 1);
1434 			}
1435 		} else {
1436 			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1437 		}
1438 		*ptr = (unsigned long)*out_eiref;
1439 		if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
1440 			return -ENOENT;
1441 	}
1442 
1443 	end = (unsigned long)ei + item_size;
1444 	*out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
1445 	*out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1446 
1447 	*ptr += btrfs_extent_inline_ref_size(*out_type);
1448 	WARN_ON(*ptr > end);
1449 	if (*ptr == end)
1450 		return 1; /* last */
1451 
1452 	return 0;
1453 }
1454 
1455 /*
1456  * reads the tree block backref for an extent. tree level and root are returned
1457  * through out_level and out_root. ptr must point to a 0 value for the first
1458  * call and may be modified (see __get_extent_inline_ref comment).
1459  * returns 0 if data was provided, 1 if there was no more data to provide or
1460  * <0 on error.
1461  */
1462 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1463 			    struct btrfs_key *key, struct btrfs_extent_item *ei,
1464 			    u32 item_size, u64 *out_root, u8 *out_level)
1465 {
1466 	int ret;
1467 	int type;
1468 	struct btrfs_tree_block_info *info;
1469 	struct btrfs_extent_inline_ref *eiref;
1470 
1471 	if (*ptr == (unsigned long)-1)
1472 		return 1;
1473 
1474 	while (1) {
1475 		ret = __get_extent_inline_ref(ptr, eb, key, ei, item_size,
1476 					      &eiref, &type);
1477 		if (ret < 0)
1478 			return ret;
1479 
1480 		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1481 		    type == BTRFS_SHARED_BLOCK_REF_KEY)
1482 			break;
1483 
1484 		if (ret == 1)
1485 			return 1;
1486 	}
1487 
1488 	/* we can treat both ref types equally here */
1489 	info = (struct btrfs_tree_block_info *)(ei + 1);
1490 	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1491 	*out_level = btrfs_tree_block_level(eb, info);
1492 
1493 	if (ret == 1)
1494 		*ptr = (unsigned long)-1;
1495 
1496 	return 0;
1497 }
1498 
1499 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1500 				u64 root, u64 extent_item_objectid,
1501 				iterate_extent_inodes_t *iterate, void *ctx)
1502 {
1503 	struct extent_inode_elem *eie;
1504 	int ret = 0;
1505 
1506 	for (eie = inode_list; eie; eie = eie->next) {
1507 		pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1508 			 "root %llu\n", extent_item_objectid,
1509 			 eie->inum, eie->offset, root);
1510 		ret = iterate(eie->inum, eie->offset, root, ctx);
1511 		if (ret) {
1512 			pr_debug("stopping iteration for %llu due to ret=%d\n",
1513 				 extent_item_objectid, ret);
1514 			break;
1515 		}
1516 	}
1517 
1518 	return ret;
1519 }
1520 
1521 /*
1522  * calls iterate() for every inode that references the extent identified by
1523  * the given parameters.
1524  * when the iterator function returns a non-zero value, iteration stops.
1525  */
1526 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1527 				u64 extent_item_objectid, u64 extent_item_pos,
1528 				int search_commit_root,
1529 				iterate_extent_inodes_t *iterate, void *ctx)
1530 {
1531 	int ret;
1532 	struct btrfs_trans_handle *trans = NULL;
1533 	struct ulist *refs = NULL;
1534 	struct ulist *roots = NULL;
1535 	struct ulist_node *ref_node = NULL;
1536 	struct ulist_node *root_node = NULL;
1537 	struct seq_list tree_mod_seq_elem = {};
1538 	struct ulist_iterator ref_uiter;
1539 	struct ulist_iterator root_uiter;
1540 
1541 	pr_debug("resolving all inodes for extent %llu\n",
1542 			extent_item_objectid);
1543 
1544 	if (!search_commit_root) {
1545 		trans = btrfs_join_transaction(fs_info->extent_root);
1546 		if (IS_ERR(trans))
1547 			return PTR_ERR(trans);
1548 		btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1549 	} else {
1550 		down_read(&fs_info->commit_root_sem);
1551 	}
1552 
1553 	ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1554 				   tree_mod_seq_elem.seq, &refs,
1555 				   &extent_item_pos);
1556 	if (ret)
1557 		goto out;
1558 
1559 	ULIST_ITER_INIT(&ref_uiter);
1560 	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1561 		ret = __btrfs_find_all_roots(trans, fs_info, ref_node->val,
1562 					     tree_mod_seq_elem.seq, &roots);
1563 		if (ret)
1564 			break;
1565 		ULIST_ITER_INIT(&root_uiter);
1566 		while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1567 			pr_debug("root %llu references leaf %llu, data list "
1568 				 "%#llx\n", root_node->val, ref_node->val,
1569 				 ref_node->aux);
1570 			ret = iterate_leaf_refs((struct extent_inode_elem *)
1571 						(uintptr_t)ref_node->aux,
1572 						root_node->val,
1573 						extent_item_objectid,
1574 						iterate, ctx);
1575 		}
1576 		ulist_free(roots);
1577 	}
1578 
1579 	free_leaf_list(refs);
1580 out:
1581 	if (!search_commit_root) {
1582 		btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1583 		btrfs_end_transaction(trans, fs_info->extent_root);
1584 	} else {
1585 		up_read(&fs_info->commit_root_sem);
1586 	}
1587 
1588 	return ret;
1589 }
1590 
1591 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1592 				struct btrfs_path *path,
1593 				iterate_extent_inodes_t *iterate, void *ctx)
1594 {
1595 	int ret;
1596 	u64 extent_item_pos;
1597 	u64 flags = 0;
1598 	struct btrfs_key found_key;
1599 	int search_commit_root = path->search_commit_root;
1600 
1601 	ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1602 	btrfs_release_path(path);
1603 	if (ret < 0)
1604 		return ret;
1605 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1606 		return -EINVAL;
1607 
1608 	extent_item_pos = logical - found_key.objectid;
1609 	ret = iterate_extent_inodes(fs_info, found_key.objectid,
1610 					extent_item_pos, search_commit_root,
1611 					iterate, ctx);
1612 
1613 	return ret;
1614 }
1615 
1616 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1617 			      struct extent_buffer *eb, void *ctx);
1618 
1619 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1620 			      struct btrfs_path *path,
1621 			      iterate_irefs_t *iterate, void *ctx)
1622 {
1623 	int ret = 0;
1624 	int slot;
1625 	u32 cur;
1626 	u32 len;
1627 	u32 name_len;
1628 	u64 parent = 0;
1629 	int found = 0;
1630 	struct extent_buffer *eb;
1631 	struct btrfs_item *item;
1632 	struct btrfs_inode_ref *iref;
1633 	struct btrfs_key found_key;
1634 
1635 	while (!ret) {
1636 		ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1637 				     &found_key);
1638 		if (ret < 0)
1639 			break;
1640 		if (ret) {
1641 			ret = found ? 0 : -ENOENT;
1642 			break;
1643 		}
1644 		++found;
1645 
1646 		parent = found_key.offset;
1647 		slot = path->slots[0];
1648 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
1649 		if (!eb) {
1650 			ret = -ENOMEM;
1651 			break;
1652 		}
1653 		extent_buffer_get(eb);
1654 		btrfs_tree_read_lock(eb);
1655 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1656 		btrfs_release_path(path);
1657 
1658 		item = btrfs_item_nr(slot);
1659 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1660 
1661 		for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1662 			name_len = btrfs_inode_ref_name_len(eb, iref);
1663 			/* path must be released before calling iterate()! */
1664 			pr_debug("following ref at offset %u for inode %llu in "
1665 				 "tree %llu\n", cur, found_key.objectid,
1666 				 fs_root->objectid);
1667 			ret = iterate(parent, name_len,
1668 				      (unsigned long)(iref + 1), eb, ctx);
1669 			if (ret)
1670 				break;
1671 			len = sizeof(*iref) + name_len;
1672 			iref = (struct btrfs_inode_ref *)((char *)iref + len);
1673 		}
1674 		btrfs_tree_read_unlock_blocking(eb);
1675 		free_extent_buffer(eb);
1676 	}
1677 
1678 	btrfs_release_path(path);
1679 
1680 	return ret;
1681 }
1682 
1683 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1684 				 struct btrfs_path *path,
1685 				 iterate_irefs_t *iterate, void *ctx)
1686 {
1687 	int ret;
1688 	int slot;
1689 	u64 offset = 0;
1690 	u64 parent;
1691 	int found = 0;
1692 	struct extent_buffer *eb;
1693 	struct btrfs_inode_extref *extref;
1694 	struct extent_buffer *leaf;
1695 	u32 item_size;
1696 	u32 cur_offset;
1697 	unsigned long ptr;
1698 
1699 	while (1) {
1700 		ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1701 					    &offset);
1702 		if (ret < 0)
1703 			break;
1704 		if (ret) {
1705 			ret = found ? 0 : -ENOENT;
1706 			break;
1707 		}
1708 		++found;
1709 
1710 		slot = path->slots[0];
1711 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
1712 		if (!eb) {
1713 			ret = -ENOMEM;
1714 			break;
1715 		}
1716 		extent_buffer_get(eb);
1717 
1718 		btrfs_tree_read_lock(eb);
1719 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1720 		btrfs_release_path(path);
1721 
1722 		leaf = path->nodes[0];
1723 		item_size = btrfs_item_size_nr(leaf, slot);
1724 		ptr = btrfs_item_ptr_offset(leaf, slot);
1725 		cur_offset = 0;
1726 
1727 		while (cur_offset < item_size) {
1728 			u32 name_len;
1729 
1730 			extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1731 			parent = btrfs_inode_extref_parent(eb, extref);
1732 			name_len = btrfs_inode_extref_name_len(eb, extref);
1733 			ret = iterate(parent, name_len,
1734 				      (unsigned long)&extref->name, eb, ctx);
1735 			if (ret)
1736 				break;
1737 
1738 			cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1739 			cur_offset += sizeof(*extref);
1740 		}
1741 		btrfs_tree_read_unlock_blocking(eb);
1742 		free_extent_buffer(eb);
1743 
1744 		offset++;
1745 	}
1746 
1747 	btrfs_release_path(path);
1748 
1749 	return ret;
1750 }
1751 
1752 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1753 			 struct btrfs_path *path, iterate_irefs_t *iterate,
1754 			 void *ctx)
1755 {
1756 	int ret;
1757 	int found_refs = 0;
1758 
1759 	ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1760 	if (!ret)
1761 		++found_refs;
1762 	else if (ret != -ENOENT)
1763 		return ret;
1764 
1765 	ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1766 	if (ret == -ENOENT && found_refs)
1767 		return 0;
1768 
1769 	return ret;
1770 }
1771 
1772 /*
1773  * returns 0 if the path could be dumped (probably truncated)
1774  * returns <0 in case of an error
1775  */
1776 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1777 			 struct extent_buffer *eb, void *ctx)
1778 {
1779 	struct inode_fs_paths *ipath = ctx;
1780 	char *fspath;
1781 	char *fspath_min;
1782 	int i = ipath->fspath->elem_cnt;
1783 	const int s_ptr = sizeof(char *);
1784 	u32 bytes_left;
1785 
1786 	bytes_left = ipath->fspath->bytes_left > s_ptr ?
1787 					ipath->fspath->bytes_left - s_ptr : 0;
1788 
1789 	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1790 	fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1791 				   name_off, eb, inum, fspath_min, bytes_left);
1792 	if (IS_ERR(fspath))
1793 		return PTR_ERR(fspath);
1794 
1795 	if (fspath > fspath_min) {
1796 		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1797 		++ipath->fspath->elem_cnt;
1798 		ipath->fspath->bytes_left = fspath - fspath_min;
1799 	} else {
1800 		++ipath->fspath->elem_missed;
1801 		ipath->fspath->bytes_missing += fspath_min - fspath;
1802 		ipath->fspath->bytes_left = 0;
1803 	}
1804 
1805 	return 0;
1806 }
1807 
1808 /*
1809  * this dumps all file system paths to the inode into the ipath struct, provided
1810  * is has been created large enough. each path is zero-terminated and accessed
1811  * from ipath->fspath->val[i].
1812  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1813  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1814  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1815  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1816  * have been needed to return all paths.
1817  */
1818 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1819 {
1820 	return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1821 			     inode_to_path, ipath);
1822 }
1823 
1824 struct btrfs_data_container *init_data_container(u32 total_bytes)
1825 {
1826 	struct btrfs_data_container *data;
1827 	size_t alloc_bytes;
1828 
1829 	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1830 	data = vmalloc(alloc_bytes);
1831 	if (!data)
1832 		return ERR_PTR(-ENOMEM);
1833 
1834 	if (total_bytes >= sizeof(*data)) {
1835 		data->bytes_left = total_bytes - sizeof(*data);
1836 		data->bytes_missing = 0;
1837 	} else {
1838 		data->bytes_missing = sizeof(*data) - total_bytes;
1839 		data->bytes_left = 0;
1840 	}
1841 
1842 	data->elem_cnt = 0;
1843 	data->elem_missed = 0;
1844 
1845 	return data;
1846 }
1847 
1848 /*
1849  * allocates space to return multiple file system paths for an inode.
1850  * total_bytes to allocate are passed, note that space usable for actual path
1851  * information will be total_bytes - sizeof(struct inode_fs_paths).
1852  * the returned pointer must be freed with free_ipath() in the end.
1853  */
1854 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1855 					struct btrfs_path *path)
1856 {
1857 	struct inode_fs_paths *ifp;
1858 	struct btrfs_data_container *fspath;
1859 
1860 	fspath = init_data_container(total_bytes);
1861 	if (IS_ERR(fspath))
1862 		return (void *)fspath;
1863 
1864 	ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1865 	if (!ifp) {
1866 		kfree(fspath);
1867 		return ERR_PTR(-ENOMEM);
1868 	}
1869 
1870 	ifp->btrfs_path = path;
1871 	ifp->fspath = fspath;
1872 	ifp->fs_root = fs_root;
1873 
1874 	return ifp;
1875 }
1876 
1877 void free_ipath(struct inode_fs_paths *ipath)
1878 {
1879 	if (!ipath)
1880 		return;
1881 	vfree(ipath->fspath);
1882 	kfree(ipath);
1883 }
1884