xref: /linux/fs/btrfs/backref.c (revision f884ab15afdc5514e88105c92a4e2e1e6539869a)
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 data_offset;
40 	u64 data_len;
41 	struct extent_inode_elem *e;
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 
50 	e = kmalloc(sizeof(*e), GFP_NOFS);
51 	if (!e)
52 		return -ENOMEM;
53 
54 	e->next = *eie;
55 	e->inum = key->objectid;
56 	e->offset = key->offset + (extent_item_pos - data_offset);
57 	*eie = e;
58 
59 	return 0;
60 }
61 
62 static int find_extent_in_eb(struct extent_buffer *eb, u64 wanted_disk_byte,
63 				u64 extent_item_pos,
64 				struct extent_inode_elem **eie)
65 {
66 	u64 disk_byte;
67 	struct btrfs_key key;
68 	struct btrfs_file_extent_item *fi;
69 	int slot;
70 	int nritems;
71 	int extent_type;
72 	int ret;
73 
74 	/*
75 	 * from the shared data ref, we only have the leaf but we need
76 	 * the key. thus, we must look into all items and see that we
77 	 * find one (some) with a reference to our extent item.
78 	 */
79 	nritems = btrfs_header_nritems(eb);
80 	for (slot = 0; slot < nritems; ++slot) {
81 		btrfs_item_key_to_cpu(eb, &key, slot);
82 		if (key.type != BTRFS_EXTENT_DATA_KEY)
83 			continue;
84 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
85 		extent_type = btrfs_file_extent_type(eb, fi);
86 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
87 			continue;
88 		/* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
89 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
90 		if (disk_byte != wanted_disk_byte)
91 			continue;
92 
93 		ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie);
94 		if (ret < 0)
95 			return ret;
96 	}
97 
98 	return 0;
99 }
100 
101 /*
102  * this structure records all encountered refs on the way up to the root
103  */
104 struct __prelim_ref {
105 	struct list_head list;
106 	u64 root_id;
107 	struct btrfs_key key_for_search;
108 	int level;
109 	int count;
110 	struct extent_inode_elem *inode_list;
111 	u64 parent;
112 	u64 wanted_disk_byte;
113 };
114 
115 /*
116  * the rules for all callers of this function are:
117  * - obtaining the parent is the goal
118  * - if you add a key, you must know that it is a correct key
119  * - if you cannot add the parent or a correct key, then we will look into the
120  *   block later to set a correct key
121  *
122  * delayed refs
123  * ============
124  *        backref type | shared | indirect | shared | indirect
125  * information         |   tree |     tree |   data |     data
126  * --------------------+--------+----------+--------+----------
127  *      parent logical |    y   |     -    |    -   |     -
128  *      key to resolve |    -   |     y    |    y   |     y
129  *  tree block logical |    -   |     -    |    -   |     -
130  *  root for resolving |    y   |     y    |    y   |     y
131  *
132  * - column 1:       we've the parent -> done
133  * - column 2, 3, 4: we use the key to find the parent
134  *
135  * on disk refs (inline or keyed)
136  * ==============================
137  *        backref type | shared | indirect | shared | indirect
138  * information         |   tree |     tree |   data |     data
139  * --------------------+--------+----------+--------+----------
140  *      parent logical |    y   |     -    |    y   |     -
141  *      key to resolve |    -   |     -    |    -   |     y
142  *  tree block logical |    y   |     y    |    y   |     y
143  *  root for resolving |    -   |     y    |    y   |     y
144  *
145  * - column 1, 3: we've the parent -> done
146  * - column 2:    we take the first key from the block to find the parent
147  *                (see __add_missing_keys)
148  * - column 4:    we use the key to find the parent
149  *
150  * additional information that's available but not required to find the parent
151  * block might help in merging entries to gain some speed.
152  */
153 
154 static int __add_prelim_ref(struct list_head *head, u64 root_id,
155 			    struct btrfs_key *key, int level,
156 			    u64 parent, u64 wanted_disk_byte, int count)
157 {
158 	struct __prelim_ref *ref;
159 
160 	/* in case we're adding delayed refs, we're holding the refs spinlock */
161 	ref = kmalloc(sizeof(*ref), GFP_ATOMIC);
162 	if (!ref)
163 		return -ENOMEM;
164 
165 	ref->root_id = root_id;
166 	if (key)
167 		ref->key_for_search = *key;
168 	else
169 		memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
170 
171 	ref->inode_list = NULL;
172 	ref->level = level;
173 	ref->count = count;
174 	ref->parent = parent;
175 	ref->wanted_disk_byte = wanted_disk_byte;
176 	list_add_tail(&ref->list, head);
177 
178 	return 0;
179 }
180 
181 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
182 				struct ulist *parents, int level,
183 				struct btrfs_key *key_for_search, u64 time_seq,
184 				u64 wanted_disk_byte,
185 				const u64 *extent_item_pos)
186 {
187 	int ret = 0;
188 	int slot;
189 	struct extent_buffer *eb;
190 	struct btrfs_key key;
191 	struct btrfs_file_extent_item *fi;
192 	struct extent_inode_elem *eie = NULL;
193 	u64 disk_byte;
194 
195 	if (level != 0) {
196 		eb = path->nodes[level];
197 		ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
198 		if (ret < 0)
199 			return ret;
200 		return 0;
201 	}
202 
203 	/*
204 	 * We normally enter this function with the path already pointing to
205 	 * the first item to check. But sometimes, we may enter it with
206 	 * slot==nritems. In that case, go to the next leaf before we continue.
207 	 */
208 	if (path->slots[0] >= btrfs_header_nritems(path->nodes[0]))
209 		ret = btrfs_next_old_leaf(root, path, time_seq);
210 
211 	while (!ret) {
212 		eb = path->nodes[0];
213 		slot = path->slots[0];
214 
215 		btrfs_item_key_to_cpu(eb, &key, slot);
216 
217 		if (key.objectid != key_for_search->objectid ||
218 		    key.type != BTRFS_EXTENT_DATA_KEY)
219 			break;
220 
221 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
222 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
223 
224 		if (disk_byte == wanted_disk_byte) {
225 			eie = NULL;
226 			if (extent_item_pos) {
227 				ret = check_extent_in_eb(&key, eb, fi,
228 						*extent_item_pos,
229 						&eie);
230 				if (ret < 0)
231 					break;
232 			}
233 			if (!ret) {
234 				ret = ulist_add(parents, eb->start,
235 						(uintptr_t)eie, GFP_NOFS);
236 				if (ret < 0)
237 					break;
238 				if (!extent_item_pos) {
239 					ret = btrfs_next_old_leaf(root, path,
240 							time_seq);
241 					continue;
242 				}
243 			}
244 		}
245 		ret = btrfs_next_old_item(root, path, time_seq);
246 	}
247 
248 	if (ret > 0)
249 		ret = 0;
250 	return ret;
251 }
252 
253 /*
254  * resolve an indirect backref in the form (root_id, key, level)
255  * to a logical address
256  */
257 static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
258 					int search_commit_root,
259 					u64 time_seq,
260 					struct __prelim_ref *ref,
261 					struct ulist *parents,
262 					const u64 *extent_item_pos)
263 {
264 	struct btrfs_path *path;
265 	struct btrfs_root *root;
266 	struct btrfs_key root_key;
267 	struct extent_buffer *eb;
268 	int ret = 0;
269 	int root_level;
270 	int level = ref->level;
271 
272 	path = btrfs_alloc_path();
273 	if (!path)
274 		return -ENOMEM;
275 	path->search_commit_root = !!search_commit_root;
276 
277 	root_key.objectid = ref->root_id;
278 	root_key.type = BTRFS_ROOT_ITEM_KEY;
279 	root_key.offset = (u64)-1;
280 	root = btrfs_read_fs_root_no_name(fs_info, &root_key);
281 	if (IS_ERR(root)) {
282 		ret = PTR_ERR(root);
283 		goto out;
284 	}
285 
286 	root_level = btrfs_old_root_level(root, time_seq);
287 
288 	if (root_level + 1 == level)
289 		goto out;
290 
291 	path->lowest_level = level;
292 	ret = btrfs_search_old_slot(root, &ref->key_for_search, path, time_seq);
293 	pr_debug("search slot in root %llu (level %d, ref count %d) returned "
294 		 "%d for key (%llu %u %llu)\n",
295 		 (unsigned long long)ref->root_id, level, ref->count, ret,
296 		 (unsigned long long)ref->key_for_search.objectid,
297 		 ref->key_for_search.type,
298 		 (unsigned long long)ref->key_for_search.offset);
299 	if (ret < 0)
300 		goto out;
301 
302 	eb = path->nodes[level];
303 	while (!eb) {
304 		if (!level) {
305 			WARN_ON(1);
306 			ret = 1;
307 			goto out;
308 		}
309 		level--;
310 		eb = path->nodes[level];
311 	}
312 
313 	ret = add_all_parents(root, path, parents, level, &ref->key_for_search,
314 				time_seq, ref->wanted_disk_byte,
315 				extent_item_pos);
316 out:
317 	btrfs_free_path(path);
318 	return ret;
319 }
320 
321 /*
322  * resolve all indirect backrefs from the list
323  */
324 static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
325 				   int search_commit_root, u64 time_seq,
326 				   struct list_head *head,
327 				   const u64 *extent_item_pos)
328 {
329 	int err;
330 	int ret = 0;
331 	struct __prelim_ref *ref;
332 	struct __prelim_ref *ref_safe;
333 	struct __prelim_ref *new_ref;
334 	struct ulist *parents;
335 	struct ulist_node *node;
336 	struct ulist_iterator uiter;
337 
338 	parents = ulist_alloc(GFP_NOFS);
339 	if (!parents)
340 		return -ENOMEM;
341 
342 	/*
343 	 * _safe allows us to insert directly after the current item without
344 	 * iterating over the newly inserted items.
345 	 * we're also allowed to re-assign ref during iteration.
346 	 */
347 	list_for_each_entry_safe(ref, ref_safe, head, list) {
348 		if (ref->parent)	/* already direct */
349 			continue;
350 		if (ref->count == 0)
351 			continue;
352 		err = __resolve_indirect_ref(fs_info, search_commit_root,
353 					     time_seq, ref, parents,
354 					     extent_item_pos);
355 		if (err == -ENOMEM)
356 			goto out;
357 		if (err)
358 			continue;
359 
360 		/* we put the first parent into the ref at hand */
361 		ULIST_ITER_INIT(&uiter);
362 		node = ulist_next(parents, &uiter);
363 		ref->parent = node ? node->val : 0;
364 		ref->inode_list = node ?
365 			(struct extent_inode_elem *)(uintptr_t)node->aux : 0;
366 
367 		/* additional parents require new refs being added here */
368 		while ((node = ulist_next(parents, &uiter))) {
369 			new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
370 			if (!new_ref) {
371 				ret = -ENOMEM;
372 				goto out;
373 			}
374 			memcpy(new_ref, ref, sizeof(*ref));
375 			new_ref->parent = node->val;
376 			new_ref->inode_list = (struct extent_inode_elem *)
377 							(uintptr_t)node->aux;
378 			list_add(&new_ref->list, &ref->list);
379 		}
380 		ulist_reinit(parents);
381 	}
382 out:
383 	ulist_free(parents);
384 	return ret;
385 }
386 
387 static inline int ref_for_same_block(struct __prelim_ref *ref1,
388 				     struct __prelim_ref *ref2)
389 {
390 	if (ref1->level != ref2->level)
391 		return 0;
392 	if (ref1->root_id != ref2->root_id)
393 		return 0;
394 	if (ref1->key_for_search.type != ref2->key_for_search.type)
395 		return 0;
396 	if (ref1->key_for_search.objectid != ref2->key_for_search.objectid)
397 		return 0;
398 	if (ref1->key_for_search.offset != ref2->key_for_search.offset)
399 		return 0;
400 	if (ref1->parent != ref2->parent)
401 		return 0;
402 
403 	return 1;
404 }
405 
406 /*
407  * read tree blocks and add keys where required.
408  */
409 static int __add_missing_keys(struct btrfs_fs_info *fs_info,
410 			      struct list_head *head)
411 {
412 	struct list_head *pos;
413 	struct extent_buffer *eb;
414 
415 	list_for_each(pos, head) {
416 		struct __prelim_ref *ref;
417 		ref = list_entry(pos, struct __prelim_ref, list);
418 
419 		if (ref->parent)
420 			continue;
421 		if (ref->key_for_search.type)
422 			continue;
423 		BUG_ON(!ref->wanted_disk_byte);
424 		eb = read_tree_block(fs_info->tree_root, ref->wanted_disk_byte,
425 				     fs_info->tree_root->leafsize, 0);
426 		if (!eb || !extent_buffer_uptodate(eb)) {
427 			free_extent_buffer(eb);
428 			return -EIO;
429 		}
430 		btrfs_tree_read_lock(eb);
431 		if (btrfs_header_level(eb) == 0)
432 			btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
433 		else
434 			btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
435 		btrfs_tree_read_unlock(eb);
436 		free_extent_buffer(eb);
437 	}
438 	return 0;
439 }
440 
441 /*
442  * merge two lists of backrefs and adjust counts accordingly
443  *
444  * mode = 1: merge identical keys, if key is set
445  *    FIXME: if we add more keys in __add_prelim_ref, we can merge more here.
446  *           additionally, we could even add a key range for the blocks we
447  *           looked into to merge even more (-> replace unresolved refs by those
448  *           having a parent).
449  * mode = 2: merge identical parents
450  */
451 static void __merge_refs(struct list_head *head, int mode)
452 {
453 	struct list_head *pos1;
454 
455 	list_for_each(pos1, head) {
456 		struct list_head *n2;
457 		struct list_head *pos2;
458 		struct __prelim_ref *ref1;
459 
460 		ref1 = list_entry(pos1, struct __prelim_ref, list);
461 
462 		for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
463 		     pos2 = n2, n2 = pos2->next) {
464 			struct __prelim_ref *ref2;
465 			struct __prelim_ref *xchg;
466 			struct extent_inode_elem *eie;
467 
468 			ref2 = list_entry(pos2, struct __prelim_ref, list);
469 
470 			if (mode == 1) {
471 				if (!ref_for_same_block(ref1, ref2))
472 					continue;
473 				if (!ref1->parent && ref2->parent) {
474 					xchg = ref1;
475 					ref1 = ref2;
476 					ref2 = xchg;
477 				}
478 			} else {
479 				if (ref1->parent != ref2->parent)
480 					continue;
481 			}
482 
483 			eie = ref1->inode_list;
484 			while (eie && eie->next)
485 				eie = eie->next;
486 			if (eie)
487 				eie->next = ref2->inode_list;
488 			else
489 				ref1->inode_list = ref2->inode_list;
490 			ref1->count += ref2->count;
491 
492 			list_del(&ref2->list);
493 			kfree(ref2);
494 		}
495 
496 	}
497 }
498 
499 /*
500  * add all currently queued delayed refs from this head whose seq nr is
501  * smaller or equal that seq to the list
502  */
503 static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
504 			      struct list_head *prefs)
505 {
506 	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
507 	struct rb_node *n = &head->node.rb_node;
508 	struct btrfs_key key;
509 	struct btrfs_key op_key = {0};
510 	int sgn;
511 	int ret = 0;
512 
513 	if (extent_op && extent_op->update_key)
514 		btrfs_disk_key_to_cpu(&op_key, &extent_op->key);
515 
516 	while ((n = rb_prev(n))) {
517 		struct btrfs_delayed_ref_node *node;
518 		node = rb_entry(n, struct btrfs_delayed_ref_node,
519 				rb_node);
520 		if (node->bytenr != head->node.bytenr)
521 			break;
522 		WARN_ON(node->is_head);
523 
524 		if (node->seq > seq)
525 			continue;
526 
527 		switch (node->action) {
528 		case BTRFS_ADD_DELAYED_EXTENT:
529 		case BTRFS_UPDATE_DELAYED_HEAD:
530 			WARN_ON(1);
531 			continue;
532 		case BTRFS_ADD_DELAYED_REF:
533 			sgn = 1;
534 			break;
535 		case BTRFS_DROP_DELAYED_REF:
536 			sgn = -1;
537 			break;
538 		default:
539 			BUG_ON(1);
540 		}
541 		switch (node->type) {
542 		case BTRFS_TREE_BLOCK_REF_KEY: {
543 			struct btrfs_delayed_tree_ref *ref;
544 
545 			ref = btrfs_delayed_node_to_tree_ref(node);
546 			ret = __add_prelim_ref(prefs, ref->root, &op_key,
547 					       ref->level + 1, 0, node->bytenr,
548 					       node->ref_mod * sgn);
549 			break;
550 		}
551 		case BTRFS_SHARED_BLOCK_REF_KEY: {
552 			struct btrfs_delayed_tree_ref *ref;
553 
554 			ref = btrfs_delayed_node_to_tree_ref(node);
555 			ret = __add_prelim_ref(prefs, ref->root, NULL,
556 					       ref->level + 1, ref->parent,
557 					       node->bytenr,
558 					       node->ref_mod * sgn);
559 			break;
560 		}
561 		case BTRFS_EXTENT_DATA_REF_KEY: {
562 			struct btrfs_delayed_data_ref *ref;
563 			ref = btrfs_delayed_node_to_data_ref(node);
564 
565 			key.objectid = ref->objectid;
566 			key.type = BTRFS_EXTENT_DATA_KEY;
567 			key.offset = ref->offset;
568 			ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
569 					       node->bytenr,
570 					       node->ref_mod * sgn);
571 			break;
572 		}
573 		case BTRFS_SHARED_DATA_REF_KEY: {
574 			struct btrfs_delayed_data_ref *ref;
575 
576 			ref = btrfs_delayed_node_to_data_ref(node);
577 
578 			key.objectid = ref->objectid;
579 			key.type = BTRFS_EXTENT_DATA_KEY;
580 			key.offset = ref->offset;
581 			ret = __add_prelim_ref(prefs, ref->root, &key, 0,
582 					       ref->parent, node->bytenr,
583 					       node->ref_mod * sgn);
584 			break;
585 		}
586 		default:
587 			WARN_ON(1);
588 		}
589 		if (ret)
590 			return ret;
591 	}
592 
593 	return 0;
594 }
595 
596 /*
597  * add all inline backrefs for bytenr to the list
598  */
599 static int __add_inline_refs(struct btrfs_fs_info *fs_info,
600 			     struct btrfs_path *path, u64 bytenr,
601 			     int *info_level, struct list_head *prefs)
602 {
603 	int ret = 0;
604 	int slot;
605 	struct extent_buffer *leaf;
606 	struct btrfs_key key;
607 	unsigned long ptr;
608 	unsigned long end;
609 	struct btrfs_extent_item *ei;
610 	u64 flags;
611 	u64 item_size;
612 
613 	/*
614 	 * enumerate all inline refs
615 	 */
616 	leaf = path->nodes[0];
617 	slot = path->slots[0];
618 
619 	item_size = btrfs_item_size_nr(leaf, slot);
620 	BUG_ON(item_size < sizeof(*ei));
621 
622 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
623 	flags = btrfs_extent_flags(leaf, ei);
624 
625 	ptr = (unsigned long)(ei + 1);
626 	end = (unsigned long)ei + item_size;
627 
628 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
629 		struct btrfs_tree_block_info *info;
630 
631 		info = (struct btrfs_tree_block_info *)ptr;
632 		*info_level = btrfs_tree_block_level(leaf, info);
633 		ptr += sizeof(struct btrfs_tree_block_info);
634 		BUG_ON(ptr > end);
635 	} else {
636 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
637 	}
638 
639 	while (ptr < end) {
640 		struct btrfs_extent_inline_ref *iref;
641 		u64 offset;
642 		int type;
643 
644 		iref = (struct btrfs_extent_inline_ref *)ptr;
645 		type = btrfs_extent_inline_ref_type(leaf, iref);
646 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
647 
648 		switch (type) {
649 		case BTRFS_SHARED_BLOCK_REF_KEY:
650 			ret = __add_prelim_ref(prefs, 0, NULL,
651 						*info_level + 1, offset,
652 						bytenr, 1);
653 			break;
654 		case BTRFS_SHARED_DATA_REF_KEY: {
655 			struct btrfs_shared_data_ref *sdref;
656 			int count;
657 
658 			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
659 			count = btrfs_shared_data_ref_count(leaf, sdref);
660 			ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
661 					       bytenr, count);
662 			break;
663 		}
664 		case BTRFS_TREE_BLOCK_REF_KEY:
665 			ret = __add_prelim_ref(prefs, offset, NULL,
666 					       *info_level + 1, 0,
667 					       bytenr, 1);
668 			break;
669 		case BTRFS_EXTENT_DATA_REF_KEY: {
670 			struct btrfs_extent_data_ref *dref;
671 			int count;
672 			u64 root;
673 
674 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
675 			count = btrfs_extent_data_ref_count(leaf, dref);
676 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
677 								      dref);
678 			key.type = BTRFS_EXTENT_DATA_KEY;
679 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
680 			root = btrfs_extent_data_ref_root(leaf, dref);
681 			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
682 					       bytenr, count);
683 			break;
684 		}
685 		default:
686 			WARN_ON(1);
687 		}
688 		if (ret)
689 			return ret;
690 		ptr += btrfs_extent_inline_ref_size(type);
691 	}
692 
693 	return 0;
694 }
695 
696 /*
697  * add all non-inline backrefs for bytenr to the list
698  */
699 static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
700 			    struct btrfs_path *path, u64 bytenr,
701 			    int info_level, struct list_head *prefs)
702 {
703 	struct btrfs_root *extent_root = fs_info->extent_root;
704 	int ret;
705 	int slot;
706 	struct extent_buffer *leaf;
707 	struct btrfs_key key;
708 
709 	while (1) {
710 		ret = btrfs_next_item(extent_root, path);
711 		if (ret < 0)
712 			break;
713 		if (ret) {
714 			ret = 0;
715 			break;
716 		}
717 
718 		slot = path->slots[0];
719 		leaf = path->nodes[0];
720 		btrfs_item_key_to_cpu(leaf, &key, slot);
721 
722 		if (key.objectid != bytenr)
723 			break;
724 		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
725 			continue;
726 		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
727 			break;
728 
729 		switch (key.type) {
730 		case BTRFS_SHARED_BLOCK_REF_KEY:
731 			ret = __add_prelim_ref(prefs, 0, NULL,
732 						info_level + 1, key.offset,
733 						bytenr, 1);
734 			break;
735 		case BTRFS_SHARED_DATA_REF_KEY: {
736 			struct btrfs_shared_data_ref *sdref;
737 			int count;
738 
739 			sdref = btrfs_item_ptr(leaf, slot,
740 					      struct btrfs_shared_data_ref);
741 			count = btrfs_shared_data_ref_count(leaf, sdref);
742 			ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
743 						bytenr, count);
744 			break;
745 		}
746 		case BTRFS_TREE_BLOCK_REF_KEY:
747 			ret = __add_prelim_ref(prefs, key.offset, NULL,
748 					       info_level + 1, 0,
749 					       bytenr, 1);
750 			break;
751 		case BTRFS_EXTENT_DATA_REF_KEY: {
752 			struct btrfs_extent_data_ref *dref;
753 			int count;
754 			u64 root;
755 
756 			dref = btrfs_item_ptr(leaf, slot,
757 					      struct btrfs_extent_data_ref);
758 			count = btrfs_extent_data_ref_count(leaf, dref);
759 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
760 								      dref);
761 			key.type = BTRFS_EXTENT_DATA_KEY;
762 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
763 			root = btrfs_extent_data_ref_root(leaf, dref);
764 			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
765 					       bytenr, count);
766 			break;
767 		}
768 		default:
769 			WARN_ON(1);
770 		}
771 		if (ret)
772 			return ret;
773 
774 	}
775 
776 	return ret;
777 }
778 
779 /*
780  * this adds all existing backrefs (inline backrefs, backrefs and delayed
781  * refs) for the given bytenr to the refs list, merges duplicates and resolves
782  * indirect refs to their parent bytenr.
783  * When roots are found, they're added to the roots list
784  *
785  * FIXME some caching might speed things up
786  */
787 static int find_parent_nodes(struct btrfs_trans_handle *trans,
788 			     struct btrfs_fs_info *fs_info, u64 bytenr,
789 			     u64 time_seq, struct ulist *refs,
790 			     struct ulist *roots, const u64 *extent_item_pos)
791 {
792 	struct btrfs_key key;
793 	struct btrfs_path *path;
794 	struct btrfs_delayed_ref_root *delayed_refs = NULL;
795 	struct btrfs_delayed_ref_head *head;
796 	int info_level = 0;
797 	int ret;
798 	int search_commit_root = (trans == BTRFS_BACKREF_SEARCH_COMMIT_ROOT);
799 	struct list_head prefs_delayed;
800 	struct list_head prefs;
801 	struct __prelim_ref *ref;
802 
803 	INIT_LIST_HEAD(&prefs);
804 	INIT_LIST_HEAD(&prefs_delayed);
805 
806 	key.objectid = bytenr;
807 	key.type = BTRFS_EXTENT_ITEM_KEY;
808 	key.offset = (u64)-1;
809 
810 	path = btrfs_alloc_path();
811 	if (!path)
812 		return -ENOMEM;
813 	path->search_commit_root = !!search_commit_root;
814 
815 	/*
816 	 * grab both a lock on the path and a lock on the delayed ref head.
817 	 * We need both to get a consistent picture of how the refs look
818 	 * at a specified point in time
819 	 */
820 again:
821 	head = NULL;
822 
823 	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
824 	if (ret < 0)
825 		goto out;
826 	BUG_ON(ret == 0);
827 
828 	if (trans != BTRFS_BACKREF_SEARCH_COMMIT_ROOT) {
829 		/*
830 		 * look if there are updates for this ref queued and lock the
831 		 * head
832 		 */
833 		delayed_refs = &trans->transaction->delayed_refs;
834 		spin_lock(&delayed_refs->lock);
835 		head = btrfs_find_delayed_ref_head(trans, bytenr);
836 		if (head) {
837 			if (!mutex_trylock(&head->mutex)) {
838 				atomic_inc(&head->node.refs);
839 				spin_unlock(&delayed_refs->lock);
840 
841 				btrfs_release_path(path);
842 
843 				/*
844 				 * Mutex was contended, block until it's
845 				 * released and try again
846 				 */
847 				mutex_lock(&head->mutex);
848 				mutex_unlock(&head->mutex);
849 				btrfs_put_delayed_ref(&head->node);
850 				goto again;
851 			}
852 			ret = __add_delayed_refs(head, time_seq,
853 						 &prefs_delayed);
854 			mutex_unlock(&head->mutex);
855 			if (ret) {
856 				spin_unlock(&delayed_refs->lock);
857 				goto out;
858 			}
859 		}
860 		spin_unlock(&delayed_refs->lock);
861 	}
862 
863 	if (path->slots[0]) {
864 		struct extent_buffer *leaf;
865 		int slot;
866 
867 		path->slots[0]--;
868 		leaf = path->nodes[0];
869 		slot = path->slots[0];
870 		btrfs_item_key_to_cpu(leaf, &key, slot);
871 		if (key.objectid == bytenr &&
872 		    key.type == BTRFS_EXTENT_ITEM_KEY) {
873 			ret = __add_inline_refs(fs_info, path, bytenr,
874 						&info_level, &prefs);
875 			if (ret)
876 				goto out;
877 			ret = __add_keyed_refs(fs_info, path, bytenr,
878 					       info_level, &prefs);
879 			if (ret)
880 				goto out;
881 		}
882 	}
883 	btrfs_release_path(path);
884 
885 	list_splice_init(&prefs_delayed, &prefs);
886 
887 	ret = __add_missing_keys(fs_info, &prefs);
888 	if (ret)
889 		goto out;
890 
891 	__merge_refs(&prefs, 1);
892 
893 	ret = __resolve_indirect_refs(fs_info, search_commit_root, time_seq,
894 				      &prefs, extent_item_pos);
895 	if (ret)
896 		goto out;
897 
898 	__merge_refs(&prefs, 2);
899 
900 	while (!list_empty(&prefs)) {
901 		ref = list_first_entry(&prefs, struct __prelim_ref, list);
902 		list_del(&ref->list);
903 		WARN_ON(ref->count < 0);
904 		if (ref->count && ref->root_id && ref->parent == 0) {
905 			/* no parent == root of tree */
906 			ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
907 			if (ret < 0)
908 				goto out;
909 		}
910 		if (ref->count && ref->parent) {
911 			struct extent_inode_elem *eie = NULL;
912 			if (extent_item_pos && !ref->inode_list) {
913 				u32 bsz;
914 				struct extent_buffer *eb;
915 				bsz = btrfs_level_size(fs_info->extent_root,
916 							info_level);
917 				eb = read_tree_block(fs_info->extent_root,
918 							   ref->parent, bsz, 0);
919 				if (!eb || !extent_buffer_uptodate(eb)) {
920 					free_extent_buffer(eb);
921 					ret = -EIO;
922 					goto out;
923 				}
924 				ret = find_extent_in_eb(eb, bytenr,
925 							*extent_item_pos, &eie);
926 				ref->inode_list = eie;
927 				free_extent_buffer(eb);
928 			}
929 			ret = ulist_add_merge(refs, ref->parent,
930 					      (uintptr_t)ref->inode_list,
931 					      (u64 *)&eie, GFP_NOFS);
932 			if (ret < 0)
933 				goto out;
934 			if (!ret && extent_item_pos) {
935 				/*
936 				 * we've recorded that parent, so we must extend
937 				 * its inode list here
938 				 */
939 				BUG_ON(!eie);
940 				while (eie->next)
941 					eie = eie->next;
942 				eie->next = ref->inode_list;
943 			}
944 		}
945 		kfree(ref);
946 	}
947 
948 out:
949 	btrfs_free_path(path);
950 	while (!list_empty(&prefs)) {
951 		ref = list_first_entry(&prefs, struct __prelim_ref, list);
952 		list_del(&ref->list);
953 		kfree(ref);
954 	}
955 	while (!list_empty(&prefs_delayed)) {
956 		ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
957 				       list);
958 		list_del(&ref->list);
959 		kfree(ref);
960 	}
961 
962 	return ret;
963 }
964 
965 static void free_leaf_list(struct ulist *blocks)
966 {
967 	struct ulist_node *node = NULL;
968 	struct extent_inode_elem *eie;
969 	struct extent_inode_elem *eie_next;
970 	struct ulist_iterator uiter;
971 
972 	ULIST_ITER_INIT(&uiter);
973 	while ((node = ulist_next(blocks, &uiter))) {
974 		if (!node->aux)
975 			continue;
976 		eie = (struct extent_inode_elem *)(uintptr_t)node->aux;
977 		for (; eie; eie = eie_next) {
978 			eie_next = eie->next;
979 			kfree(eie);
980 		}
981 		node->aux = 0;
982 	}
983 
984 	ulist_free(blocks);
985 }
986 
987 /*
988  * Finds all leafs with a reference to the specified combination of bytenr and
989  * offset. key_list_head will point to a list of corresponding keys (caller must
990  * free each list element). The leafs will be stored in the leafs ulist, which
991  * must be freed with ulist_free.
992  *
993  * returns 0 on success, <0 on error
994  */
995 static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
996 				struct btrfs_fs_info *fs_info, u64 bytenr,
997 				u64 time_seq, struct ulist **leafs,
998 				const u64 *extent_item_pos)
999 {
1000 	struct ulist *tmp;
1001 	int ret;
1002 
1003 	tmp = ulist_alloc(GFP_NOFS);
1004 	if (!tmp)
1005 		return -ENOMEM;
1006 	*leafs = ulist_alloc(GFP_NOFS);
1007 	if (!*leafs) {
1008 		ulist_free(tmp);
1009 		return -ENOMEM;
1010 	}
1011 
1012 	ret = find_parent_nodes(trans, fs_info, bytenr,
1013 				time_seq, *leafs, tmp, extent_item_pos);
1014 	ulist_free(tmp);
1015 
1016 	if (ret < 0 && ret != -ENOENT) {
1017 		free_leaf_list(*leafs);
1018 		return ret;
1019 	}
1020 
1021 	return 0;
1022 }
1023 
1024 /*
1025  * walk all backrefs for a given extent to find all roots that reference this
1026  * extent. Walking a backref means finding all extents that reference this
1027  * extent and in turn walk the backrefs of those, too. Naturally this is a
1028  * recursive process, but here it is implemented in an iterative fashion: We
1029  * find all referencing extents for the extent in question and put them on a
1030  * list. In turn, we find all referencing extents for those, further appending
1031  * to the list. The way we iterate the list allows adding more elements after
1032  * the current while iterating. The process stops when we reach the end of the
1033  * list. Found roots are added to the roots list.
1034  *
1035  * returns 0 on success, < 0 on error.
1036  */
1037 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1038 				struct btrfs_fs_info *fs_info, u64 bytenr,
1039 				u64 time_seq, struct ulist **roots)
1040 {
1041 	struct ulist *tmp;
1042 	struct ulist_node *node = NULL;
1043 	struct ulist_iterator uiter;
1044 	int ret;
1045 
1046 	tmp = ulist_alloc(GFP_NOFS);
1047 	if (!tmp)
1048 		return -ENOMEM;
1049 	*roots = ulist_alloc(GFP_NOFS);
1050 	if (!*roots) {
1051 		ulist_free(tmp);
1052 		return -ENOMEM;
1053 	}
1054 
1055 	ULIST_ITER_INIT(&uiter);
1056 	while (1) {
1057 		ret = find_parent_nodes(trans, fs_info, bytenr,
1058 					time_seq, tmp, *roots, NULL);
1059 		if (ret < 0 && ret != -ENOENT) {
1060 			ulist_free(tmp);
1061 			ulist_free(*roots);
1062 			return ret;
1063 		}
1064 		node = ulist_next(tmp, &uiter);
1065 		if (!node)
1066 			break;
1067 		bytenr = node->val;
1068 	}
1069 
1070 	ulist_free(tmp);
1071 	return 0;
1072 }
1073 
1074 
1075 static int __inode_info(u64 inum, u64 ioff, u8 key_type,
1076 			struct btrfs_root *fs_root, struct btrfs_path *path,
1077 			struct btrfs_key *found_key)
1078 {
1079 	int ret;
1080 	struct btrfs_key key;
1081 	struct extent_buffer *eb;
1082 
1083 	key.type = key_type;
1084 	key.objectid = inum;
1085 	key.offset = ioff;
1086 
1087 	ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
1088 	if (ret < 0)
1089 		return ret;
1090 
1091 	eb = path->nodes[0];
1092 	if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
1093 		ret = btrfs_next_leaf(fs_root, path);
1094 		if (ret)
1095 			return ret;
1096 		eb = path->nodes[0];
1097 	}
1098 
1099 	btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
1100 	if (found_key->type != key.type || found_key->objectid != key.objectid)
1101 		return 1;
1102 
1103 	return 0;
1104 }
1105 
1106 /*
1107  * this makes the path point to (inum INODE_ITEM ioff)
1108  */
1109 int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1110 			struct btrfs_path *path)
1111 {
1112 	struct btrfs_key key;
1113 	return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
1114 				&key);
1115 }
1116 
1117 static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
1118 				struct btrfs_path *path,
1119 				struct btrfs_key *found_key)
1120 {
1121 	return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
1122 				found_key);
1123 }
1124 
1125 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1126 			  u64 start_off, struct btrfs_path *path,
1127 			  struct btrfs_inode_extref **ret_extref,
1128 			  u64 *found_off)
1129 {
1130 	int ret, slot;
1131 	struct btrfs_key key;
1132 	struct btrfs_key found_key;
1133 	struct btrfs_inode_extref *extref;
1134 	struct extent_buffer *leaf;
1135 	unsigned long ptr;
1136 
1137 	key.objectid = inode_objectid;
1138 	btrfs_set_key_type(&key, BTRFS_INODE_EXTREF_KEY);
1139 	key.offset = start_off;
1140 
1141 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1142 	if (ret < 0)
1143 		return ret;
1144 
1145 	while (1) {
1146 		leaf = path->nodes[0];
1147 		slot = path->slots[0];
1148 		if (slot >= btrfs_header_nritems(leaf)) {
1149 			/*
1150 			 * If the item at offset is not found,
1151 			 * btrfs_search_slot will point us to the slot
1152 			 * where it should be inserted. In our case
1153 			 * that will be the slot directly before the
1154 			 * next INODE_REF_KEY_V2 item. In the case
1155 			 * that we're pointing to the last slot in a
1156 			 * leaf, we must move one leaf over.
1157 			 */
1158 			ret = btrfs_next_leaf(root, path);
1159 			if (ret) {
1160 				if (ret >= 1)
1161 					ret = -ENOENT;
1162 				break;
1163 			}
1164 			continue;
1165 		}
1166 
1167 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
1168 
1169 		/*
1170 		 * Check that we're still looking at an extended ref key for
1171 		 * this particular objectid. If we have different
1172 		 * objectid or type then there are no more to be found
1173 		 * in the tree and we can exit.
1174 		 */
1175 		ret = -ENOENT;
1176 		if (found_key.objectid != inode_objectid)
1177 			break;
1178 		if (btrfs_key_type(&found_key) != BTRFS_INODE_EXTREF_KEY)
1179 			break;
1180 
1181 		ret = 0;
1182 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1183 		extref = (struct btrfs_inode_extref *)ptr;
1184 		*ret_extref = extref;
1185 		if (found_off)
1186 			*found_off = found_key.offset;
1187 		break;
1188 	}
1189 
1190 	return ret;
1191 }
1192 
1193 /*
1194  * this iterates to turn a name (from iref/extref) into a full filesystem path.
1195  * Elements of the path are separated by '/' and the path is guaranteed to be
1196  * 0-terminated. the path is only given within the current file system.
1197  * Therefore, it never starts with a '/'. the caller is responsible to provide
1198  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1199  * the start point of the resulting string is returned. this pointer is within
1200  * dest, normally.
1201  * in case the path buffer would overflow, the pointer is decremented further
1202  * as if output was written to the buffer, though no more output is actually
1203  * generated. that way, the caller can determine how much space would be
1204  * required for the path to fit into the buffer. in that case, the returned
1205  * value will be smaller than dest. callers must check this!
1206  */
1207 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1208 			u32 name_len, unsigned long name_off,
1209 			struct extent_buffer *eb_in, u64 parent,
1210 			char *dest, u32 size)
1211 {
1212 	int slot;
1213 	u64 next_inum;
1214 	int ret;
1215 	s64 bytes_left = ((s64)size) - 1;
1216 	struct extent_buffer *eb = eb_in;
1217 	struct btrfs_key found_key;
1218 	int leave_spinning = path->leave_spinning;
1219 	struct btrfs_inode_ref *iref;
1220 
1221 	if (bytes_left >= 0)
1222 		dest[bytes_left] = '\0';
1223 
1224 	path->leave_spinning = 1;
1225 	while (1) {
1226 		bytes_left -= name_len;
1227 		if (bytes_left >= 0)
1228 			read_extent_buffer(eb, dest + bytes_left,
1229 					   name_off, name_len);
1230 		if (eb != eb_in) {
1231 			btrfs_tree_read_unlock_blocking(eb);
1232 			free_extent_buffer(eb);
1233 		}
1234 		ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
1235 		if (ret > 0)
1236 			ret = -ENOENT;
1237 		if (ret)
1238 			break;
1239 
1240 		next_inum = found_key.offset;
1241 
1242 		/* regular exit ahead */
1243 		if (parent == next_inum)
1244 			break;
1245 
1246 		slot = path->slots[0];
1247 		eb = path->nodes[0];
1248 		/* make sure we can use eb after releasing the path */
1249 		if (eb != eb_in) {
1250 			atomic_inc(&eb->refs);
1251 			btrfs_tree_read_lock(eb);
1252 			btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1253 		}
1254 		btrfs_release_path(path);
1255 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1256 
1257 		name_len = btrfs_inode_ref_name_len(eb, iref);
1258 		name_off = (unsigned long)(iref + 1);
1259 
1260 		parent = next_inum;
1261 		--bytes_left;
1262 		if (bytes_left >= 0)
1263 			dest[bytes_left] = '/';
1264 	}
1265 
1266 	btrfs_release_path(path);
1267 	path->leave_spinning = leave_spinning;
1268 
1269 	if (ret)
1270 		return ERR_PTR(ret);
1271 
1272 	return dest + bytes_left;
1273 }
1274 
1275 /*
1276  * this makes the path point to (logical EXTENT_ITEM *)
1277  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1278  * tree blocks and <0 on error.
1279  */
1280 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1281 			struct btrfs_path *path, struct btrfs_key *found_key,
1282 			u64 *flags_ret)
1283 {
1284 	int ret;
1285 	u64 flags;
1286 	u32 item_size;
1287 	struct extent_buffer *eb;
1288 	struct btrfs_extent_item *ei;
1289 	struct btrfs_key key;
1290 
1291 	key.type = BTRFS_EXTENT_ITEM_KEY;
1292 	key.objectid = logical;
1293 	key.offset = (u64)-1;
1294 
1295 	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
1296 	if (ret < 0)
1297 		return ret;
1298 	ret = btrfs_previous_item(fs_info->extent_root, path,
1299 					0, BTRFS_EXTENT_ITEM_KEY);
1300 	if (ret < 0)
1301 		return ret;
1302 
1303 	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1304 	if (found_key->type != BTRFS_EXTENT_ITEM_KEY ||
1305 	    found_key->objectid > logical ||
1306 	    found_key->objectid + found_key->offset <= logical) {
1307 		pr_debug("logical %llu is not within any extent\n",
1308 			 (unsigned long long)logical);
1309 		return -ENOENT;
1310 	}
1311 
1312 	eb = path->nodes[0];
1313 	item_size = btrfs_item_size_nr(eb, path->slots[0]);
1314 	BUG_ON(item_size < sizeof(*ei));
1315 
1316 	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
1317 	flags = btrfs_extent_flags(eb, ei);
1318 
1319 	pr_debug("logical %llu is at position %llu within the extent (%llu "
1320 		 "EXTENT_ITEM %llu) flags %#llx size %u\n",
1321 		 (unsigned long long)logical,
1322 		 (unsigned long long)(logical - found_key->objectid),
1323 		 (unsigned long long)found_key->objectid,
1324 		 (unsigned long long)found_key->offset,
1325 		 (unsigned long long)flags, item_size);
1326 
1327 	WARN_ON(!flags_ret);
1328 	if (flags_ret) {
1329 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1330 			*flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
1331 		else if (flags & BTRFS_EXTENT_FLAG_DATA)
1332 			*flags_ret = BTRFS_EXTENT_FLAG_DATA;
1333 		else
1334 			BUG_ON(1);
1335 		return 0;
1336 	}
1337 
1338 	return -EIO;
1339 }
1340 
1341 /*
1342  * helper function to iterate extent inline refs. ptr must point to a 0 value
1343  * for the first call and may be modified. it is used to track state.
1344  * if more refs exist, 0 is returned and the next call to
1345  * __get_extent_inline_ref must pass the modified ptr parameter to get the
1346  * next ref. after the last ref was processed, 1 is returned.
1347  * returns <0 on error
1348  */
1349 static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
1350 				struct btrfs_extent_item *ei, u32 item_size,
1351 				struct btrfs_extent_inline_ref **out_eiref,
1352 				int *out_type)
1353 {
1354 	unsigned long end;
1355 	u64 flags;
1356 	struct btrfs_tree_block_info *info;
1357 
1358 	if (!*ptr) {
1359 		/* first call */
1360 		flags = btrfs_extent_flags(eb, ei);
1361 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1362 			info = (struct btrfs_tree_block_info *)(ei + 1);
1363 			*out_eiref =
1364 				(struct btrfs_extent_inline_ref *)(info + 1);
1365 		} else {
1366 			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
1367 		}
1368 		*ptr = (unsigned long)*out_eiref;
1369 		if ((void *)*ptr >= (void *)ei + item_size)
1370 			return -ENOENT;
1371 	}
1372 
1373 	end = (unsigned long)ei + item_size;
1374 	*out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
1375 	*out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
1376 
1377 	*ptr += btrfs_extent_inline_ref_size(*out_type);
1378 	WARN_ON(*ptr > end);
1379 	if (*ptr == end)
1380 		return 1; /* last */
1381 
1382 	return 0;
1383 }
1384 
1385 /*
1386  * reads the tree block backref for an extent. tree level and root are returned
1387  * through out_level and out_root. ptr must point to a 0 value for the first
1388  * call and may be modified (see __get_extent_inline_ref comment).
1389  * returns 0 if data was provided, 1 if there was no more data to provide or
1390  * <0 on error.
1391  */
1392 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
1393 				struct btrfs_extent_item *ei, u32 item_size,
1394 				u64 *out_root, u8 *out_level)
1395 {
1396 	int ret;
1397 	int type;
1398 	struct btrfs_tree_block_info *info;
1399 	struct btrfs_extent_inline_ref *eiref;
1400 
1401 	if (*ptr == (unsigned long)-1)
1402 		return 1;
1403 
1404 	while (1) {
1405 		ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
1406 						&eiref, &type);
1407 		if (ret < 0)
1408 			return ret;
1409 
1410 		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
1411 		    type == BTRFS_SHARED_BLOCK_REF_KEY)
1412 			break;
1413 
1414 		if (ret == 1)
1415 			return 1;
1416 	}
1417 
1418 	/* we can treat both ref types equally here */
1419 	info = (struct btrfs_tree_block_info *)(ei + 1);
1420 	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
1421 	*out_level = btrfs_tree_block_level(eb, info);
1422 
1423 	if (ret == 1)
1424 		*ptr = (unsigned long)-1;
1425 
1426 	return 0;
1427 }
1428 
1429 static int iterate_leaf_refs(struct extent_inode_elem *inode_list,
1430 				u64 root, u64 extent_item_objectid,
1431 				iterate_extent_inodes_t *iterate, void *ctx)
1432 {
1433 	struct extent_inode_elem *eie;
1434 	int ret = 0;
1435 
1436 	for (eie = inode_list; eie; eie = eie->next) {
1437 		pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
1438 			 "root %llu\n", extent_item_objectid,
1439 			 eie->inum, eie->offset, root);
1440 		ret = iterate(eie->inum, eie->offset, root, ctx);
1441 		if (ret) {
1442 			pr_debug("stopping iteration for %llu due to ret=%d\n",
1443 				 extent_item_objectid, ret);
1444 			break;
1445 		}
1446 	}
1447 
1448 	return ret;
1449 }
1450 
1451 /*
1452  * calls iterate() for every inode that references the extent identified by
1453  * the given parameters.
1454  * when the iterator function returns a non-zero value, iteration stops.
1455  */
1456 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
1457 				u64 extent_item_objectid, u64 extent_item_pos,
1458 				int search_commit_root,
1459 				iterate_extent_inodes_t *iterate, void *ctx)
1460 {
1461 	int ret;
1462 	struct btrfs_trans_handle *trans;
1463 	struct ulist *refs = NULL;
1464 	struct ulist *roots = NULL;
1465 	struct ulist_node *ref_node = NULL;
1466 	struct ulist_node *root_node = NULL;
1467 	struct seq_list tree_mod_seq_elem = {};
1468 	struct ulist_iterator ref_uiter;
1469 	struct ulist_iterator root_uiter;
1470 
1471 	pr_debug("resolving all inodes for extent %llu\n",
1472 			extent_item_objectid);
1473 
1474 	if (search_commit_root) {
1475 		trans = BTRFS_BACKREF_SEARCH_COMMIT_ROOT;
1476 	} else {
1477 		trans = btrfs_join_transaction(fs_info->extent_root);
1478 		if (IS_ERR(trans))
1479 			return PTR_ERR(trans);
1480 		btrfs_get_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1481 	}
1482 
1483 	ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
1484 				   tree_mod_seq_elem.seq, &refs,
1485 				   &extent_item_pos);
1486 	if (ret)
1487 		goto out;
1488 
1489 	ULIST_ITER_INIT(&ref_uiter);
1490 	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
1491 		ret = btrfs_find_all_roots(trans, fs_info, ref_node->val,
1492 					   tree_mod_seq_elem.seq, &roots);
1493 		if (ret)
1494 			break;
1495 		ULIST_ITER_INIT(&root_uiter);
1496 		while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
1497 			pr_debug("root %llu references leaf %llu, data list "
1498 				 "%#llx\n", root_node->val, ref_node->val,
1499 				 (long long)ref_node->aux);
1500 			ret = iterate_leaf_refs((struct extent_inode_elem *)
1501 						(uintptr_t)ref_node->aux,
1502 						root_node->val,
1503 						extent_item_objectid,
1504 						iterate, ctx);
1505 		}
1506 		ulist_free(roots);
1507 	}
1508 
1509 	free_leaf_list(refs);
1510 out:
1511 	if (!search_commit_root) {
1512 		btrfs_put_tree_mod_seq(fs_info, &tree_mod_seq_elem);
1513 		btrfs_end_transaction(trans, fs_info->extent_root);
1514 	}
1515 
1516 	return ret;
1517 }
1518 
1519 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
1520 				struct btrfs_path *path,
1521 				iterate_extent_inodes_t *iterate, void *ctx)
1522 {
1523 	int ret;
1524 	u64 extent_item_pos;
1525 	u64 flags = 0;
1526 	struct btrfs_key found_key;
1527 	int search_commit_root = path->search_commit_root;
1528 
1529 	ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
1530 	btrfs_release_path(path);
1531 	if (ret < 0)
1532 		return ret;
1533 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
1534 		return -EINVAL;
1535 
1536 	extent_item_pos = logical - found_key.objectid;
1537 	ret = iterate_extent_inodes(fs_info, found_key.objectid,
1538 					extent_item_pos, search_commit_root,
1539 					iterate, ctx);
1540 
1541 	return ret;
1542 }
1543 
1544 typedef int (iterate_irefs_t)(u64 parent, u32 name_len, unsigned long name_off,
1545 			      struct extent_buffer *eb, void *ctx);
1546 
1547 static int iterate_inode_refs(u64 inum, struct btrfs_root *fs_root,
1548 			      struct btrfs_path *path,
1549 			      iterate_irefs_t *iterate, void *ctx)
1550 {
1551 	int ret = 0;
1552 	int slot;
1553 	u32 cur;
1554 	u32 len;
1555 	u32 name_len;
1556 	u64 parent = 0;
1557 	int found = 0;
1558 	struct extent_buffer *eb;
1559 	struct btrfs_item *item;
1560 	struct btrfs_inode_ref *iref;
1561 	struct btrfs_key found_key;
1562 
1563 	while (!ret) {
1564 		path->leave_spinning = 1;
1565 		ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
1566 				     &found_key);
1567 		if (ret < 0)
1568 			break;
1569 		if (ret) {
1570 			ret = found ? 0 : -ENOENT;
1571 			break;
1572 		}
1573 		++found;
1574 
1575 		parent = found_key.offset;
1576 		slot = path->slots[0];
1577 		eb = path->nodes[0];
1578 		/* make sure we can use eb after releasing the path */
1579 		atomic_inc(&eb->refs);
1580 		btrfs_tree_read_lock(eb);
1581 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1582 		btrfs_release_path(path);
1583 
1584 		item = btrfs_item_nr(eb, slot);
1585 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1586 
1587 		for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
1588 			name_len = btrfs_inode_ref_name_len(eb, iref);
1589 			/* path must be released before calling iterate()! */
1590 			pr_debug("following ref at offset %u for inode %llu in "
1591 				 "tree %llu\n", cur,
1592 				 (unsigned long long)found_key.objectid,
1593 				 (unsigned long long)fs_root->objectid);
1594 			ret = iterate(parent, name_len,
1595 				      (unsigned long)(iref + 1), eb, ctx);
1596 			if (ret)
1597 				break;
1598 			len = sizeof(*iref) + name_len;
1599 			iref = (struct btrfs_inode_ref *)((char *)iref + len);
1600 		}
1601 		btrfs_tree_read_unlock_blocking(eb);
1602 		free_extent_buffer(eb);
1603 	}
1604 
1605 	btrfs_release_path(path);
1606 
1607 	return ret;
1608 }
1609 
1610 static int iterate_inode_extrefs(u64 inum, struct btrfs_root *fs_root,
1611 				 struct btrfs_path *path,
1612 				 iterate_irefs_t *iterate, void *ctx)
1613 {
1614 	int ret;
1615 	int slot;
1616 	u64 offset = 0;
1617 	u64 parent;
1618 	int found = 0;
1619 	struct extent_buffer *eb;
1620 	struct btrfs_inode_extref *extref;
1621 	struct extent_buffer *leaf;
1622 	u32 item_size;
1623 	u32 cur_offset;
1624 	unsigned long ptr;
1625 
1626 	while (1) {
1627 		ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
1628 					    &offset);
1629 		if (ret < 0)
1630 			break;
1631 		if (ret) {
1632 			ret = found ? 0 : -ENOENT;
1633 			break;
1634 		}
1635 		++found;
1636 
1637 		slot = path->slots[0];
1638 		eb = path->nodes[0];
1639 		/* make sure we can use eb after releasing the path */
1640 		atomic_inc(&eb->refs);
1641 
1642 		btrfs_tree_read_lock(eb);
1643 		btrfs_set_lock_blocking_rw(eb, BTRFS_READ_LOCK);
1644 		btrfs_release_path(path);
1645 
1646 		leaf = path->nodes[0];
1647 		item_size = btrfs_item_size_nr(leaf, path->slots[0]);
1648 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1649 		cur_offset = 0;
1650 
1651 		while (cur_offset < item_size) {
1652 			u32 name_len;
1653 
1654 			extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
1655 			parent = btrfs_inode_extref_parent(eb, extref);
1656 			name_len = btrfs_inode_extref_name_len(eb, extref);
1657 			ret = iterate(parent, name_len,
1658 				      (unsigned long)&extref->name, eb, ctx);
1659 			if (ret)
1660 				break;
1661 
1662 			cur_offset += btrfs_inode_extref_name_len(leaf, extref);
1663 			cur_offset += sizeof(*extref);
1664 		}
1665 		btrfs_tree_read_unlock_blocking(eb);
1666 		free_extent_buffer(eb);
1667 
1668 		offset++;
1669 	}
1670 
1671 	btrfs_release_path(path);
1672 
1673 	return ret;
1674 }
1675 
1676 static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
1677 			 struct btrfs_path *path, iterate_irefs_t *iterate,
1678 			 void *ctx)
1679 {
1680 	int ret;
1681 	int found_refs = 0;
1682 
1683 	ret = iterate_inode_refs(inum, fs_root, path, iterate, ctx);
1684 	if (!ret)
1685 		++found_refs;
1686 	else if (ret != -ENOENT)
1687 		return ret;
1688 
1689 	ret = iterate_inode_extrefs(inum, fs_root, path, iterate, ctx);
1690 	if (ret == -ENOENT && found_refs)
1691 		return 0;
1692 
1693 	return ret;
1694 }
1695 
1696 /*
1697  * returns 0 if the path could be dumped (probably truncated)
1698  * returns <0 in case of an error
1699  */
1700 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
1701 			 struct extent_buffer *eb, void *ctx)
1702 {
1703 	struct inode_fs_paths *ipath = ctx;
1704 	char *fspath;
1705 	char *fspath_min;
1706 	int i = ipath->fspath->elem_cnt;
1707 	const int s_ptr = sizeof(char *);
1708 	u32 bytes_left;
1709 
1710 	bytes_left = ipath->fspath->bytes_left > s_ptr ?
1711 					ipath->fspath->bytes_left - s_ptr : 0;
1712 
1713 	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
1714 	fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
1715 				   name_off, eb, inum, fspath_min, bytes_left);
1716 	if (IS_ERR(fspath))
1717 		return PTR_ERR(fspath);
1718 
1719 	if (fspath > fspath_min) {
1720 		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
1721 		++ipath->fspath->elem_cnt;
1722 		ipath->fspath->bytes_left = fspath - fspath_min;
1723 	} else {
1724 		++ipath->fspath->elem_missed;
1725 		ipath->fspath->bytes_missing += fspath_min - fspath;
1726 		ipath->fspath->bytes_left = 0;
1727 	}
1728 
1729 	return 0;
1730 }
1731 
1732 /*
1733  * this dumps all file system paths to the inode into the ipath struct, provided
1734  * is has been created large enough. each path is zero-terminated and accessed
1735  * from ipath->fspath->val[i].
1736  * when it returns, there are ipath->fspath->elem_cnt number of paths available
1737  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
1738  * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
1739  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
1740  * have been needed to return all paths.
1741  */
1742 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
1743 {
1744 	return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
1745 			     inode_to_path, ipath);
1746 }
1747 
1748 struct btrfs_data_container *init_data_container(u32 total_bytes)
1749 {
1750 	struct btrfs_data_container *data;
1751 	size_t alloc_bytes;
1752 
1753 	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
1754 	data = vmalloc(alloc_bytes);
1755 	if (!data)
1756 		return ERR_PTR(-ENOMEM);
1757 
1758 	if (total_bytes >= sizeof(*data)) {
1759 		data->bytes_left = total_bytes - sizeof(*data);
1760 		data->bytes_missing = 0;
1761 	} else {
1762 		data->bytes_missing = sizeof(*data) - total_bytes;
1763 		data->bytes_left = 0;
1764 	}
1765 
1766 	data->elem_cnt = 0;
1767 	data->elem_missed = 0;
1768 
1769 	return data;
1770 }
1771 
1772 /*
1773  * allocates space to return multiple file system paths for an inode.
1774  * total_bytes to allocate are passed, note that space usable for actual path
1775  * information will be total_bytes - sizeof(struct inode_fs_paths).
1776  * the returned pointer must be freed with free_ipath() in the end.
1777  */
1778 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
1779 					struct btrfs_path *path)
1780 {
1781 	struct inode_fs_paths *ifp;
1782 	struct btrfs_data_container *fspath;
1783 
1784 	fspath = init_data_container(total_bytes);
1785 	if (IS_ERR(fspath))
1786 		return (void *)fspath;
1787 
1788 	ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
1789 	if (!ifp) {
1790 		kfree(fspath);
1791 		return ERR_PTR(-ENOMEM);
1792 	}
1793 
1794 	ifp->btrfs_path = path;
1795 	ifp->fspath = fspath;
1796 	ifp->fs_root = fs_root;
1797 
1798 	return ifp;
1799 }
1800 
1801 void free_ipath(struct inode_fs_paths *ipath)
1802 {
1803 	if (!ipath)
1804 		return;
1805 	vfree(ipath->fspath);
1806 	kfree(ipath);
1807 }
1808