xref: /linux/fs/btrfs/backref.c (revision 7a40974fd0efa3698de4c6d1d0ee0436bcc4445d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2011 STRATO.  All rights reserved.
4  */
5 
6 #include <linux/mm.h>
7 #include <linux/rbtree.h>
8 #include <trace/events/btrfs.h>
9 #include "ctree.h"
10 #include "disk-io.h"
11 #include "backref.h"
12 #include "ulist.h"
13 #include "transaction.h"
14 #include "delayed-ref.h"
15 #include "locking.h"
16 #include "misc.h"
17 #include "tree-mod-log.h"
18 #include "fs.h"
19 #include "accessors.h"
20 #include "extent-tree.h"
21 #include "relocation.h"
22 #include "tree-checker.h"
23 
24 /* Just arbitrary numbers so we can be sure one of these happened. */
25 #define BACKREF_FOUND_SHARED     6
26 #define BACKREF_FOUND_NOT_SHARED 7
27 
28 struct extent_inode_elem {
29 	u64 inum;
30 	u64 offset;
31 	u64 num_bytes;
32 	struct extent_inode_elem *next;
33 };
34 
check_extent_in_eb(struct btrfs_backref_walk_ctx * ctx,const struct btrfs_key * key,const struct extent_buffer * eb,const struct btrfs_file_extent_item * fi,struct extent_inode_elem ** eie)35 static int check_extent_in_eb(struct btrfs_backref_walk_ctx *ctx,
36 			      const struct btrfs_key *key,
37 			      const struct extent_buffer *eb,
38 			      const struct btrfs_file_extent_item *fi,
39 			      struct extent_inode_elem **eie)
40 {
41 	const u64 data_len = btrfs_file_extent_num_bytes(eb, fi);
42 	u64 offset = key->offset;
43 	struct extent_inode_elem *e;
44 	const u64 *root_ids;
45 	int root_count;
46 	bool cached;
47 
48 	if (!ctx->ignore_extent_item_pos &&
49 	    !btrfs_file_extent_compression(eb, fi) &&
50 	    !btrfs_file_extent_encryption(eb, fi) &&
51 	    !btrfs_file_extent_other_encoding(eb, fi)) {
52 		u64 data_offset;
53 
54 		data_offset = btrfs_file_extent_offset(eb, fi);
55 
56 		if (ctx->extent_item_pos < data_offset ||
57 		    ctx->extent_item_pos >= data_offset + data_len)
58 			return 1;
59 		offset += ctx->extent_item_pos - data_offset;
60 	}
61 
62 	if (!ctx->indirect_ref_iterator || !ctx->cache_lookup)
63 		goto add_inode_elem;
64 
65 	cached = ctx->cache_lookup(eb->start, ctx->user_ctx, &root_ids,
66 				   &root_count);
67 	if (!cached)
68 		goto add_inode_elem;
69 
70 	for (int i = 0; i < root_count; i++) {
71 		int ret;
72 
73 		ret = ctx->indirect_ref_iterator(key->objectid, offset,
74 						 data_len, root_ids[i],
75 						 ctx->user_ctx);
76 		if (ret)
77 			return ret;
78 	}
79 
80 add_inode_elem:
81 	e = kmalloc(sizeof(*e), GFP_NOFS);
82 	if (!e)
83 		return -ENOMEM;
84 
85 	e->next = *eie;
86 	e->inum = key->objectid;
87 	e->offset = offset;
88 	e->num_bytes = data_len;
89 	*eie = e;
90 
91 	return 0;
92 }
93 
free_inode_elem_list(struct extent_inode_elem * eie)94 static void free_inode_elem_list(struct extent_inode_elem *eie)
95 {
96 	struct extent_inode_elem *eie_next;
97 
98 	for (; eie; eie = eie_next) {
99 		eie_next = eie->next;
100 		kfree(eie);
101 	}
102 }
103 
find_extent_in_eb(struct btrfs_backref_walk_ctx * ctx,const struct extent_buffer * eb,struct extent_inode_elem ** eie)104 static int find_extent_in_eb(struct btrfs_backref_walk_ctx *ctx,
105 			     const struct extent_buffer *eb,
106 			     struct extent_inode_elem **eie)
107 {
108 	u64 disk_byte;
109 	struct btrfs_key key;
110 	struct btrfs_file_extent_item *fi;
111 	int slot;
112 	int nritems;
113 	int extent_type;
114 	int ret;
115 
116 	/*
117 	 * from the shared data ref, we only have the leaf but we need
118 	 * the key. thus, we must look into all items and see that we
119 	 * find one (some) with a reference to our extent item.
120 	 */
121 	nritems = btrfs_header_nritems(eb);
122 	for (slot = 0; slot < nritems; ++slot) {
123 		btrfs_item_key_to_cpu(eb, &key, slot);
124 		if (key.type != BTRFS_EXTENT_DATA_KEY)
125 			continue;
126 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
127 		extent_type = btrfs_file_extent_type(eb, fi);
128 		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
129 			continue;
130 		/* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
131 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
132 		if (disk_byte != ctx->bytenr)
133 			continue;
134 
135 		ret = check_extent_in_eb(ctx, &key, eb, fi, eie);
136 		if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
137 			return ret;
138 	}
139 
140 	return 0;
141 }
142 
143 struct preftree {
144 	struct rb_root_cached root;
145 	unsigned int count;
146 };
147 
148 #define PREFTREE_INIT	{ .root = RB_ROOT_CACHED, .count = 0 }
149 
150 struct preftrees {
151 	struct preftree direct;    /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
152 	struct preftree indirect;  /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
153 	struct preftree indirect_missing_keys;
154 };
155 
156 /*
157  * Checks for a shared extent during backref search.
158  *
159  * The share_count tracks prelim_refs (direct and indirect) having a
160  * ref->count >0:
161  *  - incremented when a ref->count transitions to >0
162  *  - decremented when a ref->count transitions to <1
163  */
164 struct share_check {
165 	struct btrfs_backref_share_check_ctx *ctx;
166 	struct btrfs_root *root;
167 	u64 inum;
168 	u64 data_bytenr;
169 	u64 data_extent_gen;
170 	/*
171 	 * Counts number of inodes that refer to an extent (different inodes in
172 	 * the same root or different roots) that we could find. The sharedness
173 	 * check typically stops once this counter gets greater than 1, so it
174 	 * may not reflect the total number of inodes.
175 	 */
176 	int share_count;
177 	/*
178 	 * The number of times we found our inode refers to the data extent we
179 	 * are determining the sharedness. In other words, how many file extent
180 	 * items we could find for our inode that point to our target data
181 	 * extent. The value we get here after finishing the extent sharedness
182 	 * check may be smaller than reality, but if it ends up being greater
183 	 * than 1, then we know for sure the inode has multiple file extent
184 	 * items that point to our inode, and we can safely assume it's useful
185 	 * to cache the sharedness check result.
186 	 */
187 	int self_ref_count;
188 	bool have_delayed_delete_refs;
189 };
190 
extent_is_shared(struct share_check * sc)191 static inline int extent_is_shared(struct share_check *sc)
192 {
193 	return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
194 }
195 
196 static struct kmem_cache *btrfs_prelim_ref_cache;
197 
btrfs_prelim_ref_init(void)198 int __init btrfs_prelim_ref_init(void)
199 {
200 	btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
201 					sizeof(struct prelim_ref), 0, 0, NULL);
202 	if (!btrfs_prelim_ref_cache)
203 		return -ENOMEM;
204 	return 0;
205 }
206 
btrfs_prelim_ref_exit(void)207 void __cold btrfs_prelim_ref_exit(void)
208 {
209 	kmem_cache_destroy(btrfs_prelim_ref_cache);
210 }
211 
free_pref(struct prelim_ref * ref)212 static void free_pref(struct prelim_ref *ref)
213 {
214 	kmem_cache_free(btrfs_prelim_ref_cache, ref);
215 }
216 
217 /*
218  * Return 0 when both refs are for the same block (and can be merged).
219  * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
220  * indicates a 'higher' block.
221  */
prelim_ref_compare(const struct prelim_ref * ref1,const struct prelim_ref * ref2)222 static int prelim_ref_compare(const struct prelim_ref *ref1,
223 			      const struct prelim_ref *ref2)
224 {
225 	if (ref1->level < ref2->level)
226 		return -1;
227 	if (ref1->level > ref2->level)
228 		return 1;
229 	if (ref1->root_id < ref2->root_id)
230 		return -1;
231 	if (ref1->root_id > ref2->root_id)
232 		return 1;
233 	if (ref1->key_for_search.type < ref2->key_for_search.type)
234 		return -1;
235 	if (ref1->key_for_search.type > ref2->key_for_search.type)
236 		return 1;
237 	if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
238 		return -1;
239 	if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
240 		return 1;
241 	if (ref1->key_for_search.offset < ref2->key_for_search.offset)
242 		return -1;
243 	if (ref1->key_for_search.offset > ref2->key_for_search.offset)
244 		return 1;
245 	if (ref1->parent < ref2->parent)
246 		return -1;
247 	if (ref1->parent > ref2->parent)
248 		return 1;
249 
250 	return 0;
251 }
252 
update_share_count(struct share_check * sc,int oldcount,int newcount,const struct prelim_ref * newref)253 static void update_share_count(struct share_check *sc, int oldcount,
254 			       int newcount, const struct prelim_ref *newref)
255 {
256 	if ((!sc) || (oldcount == 0 && newcount < 1))
257 		return;
258 
259 	if (oldcount > 0 && newcount < 1)
260 		sc->share_count--;
261 	else if (oldcount < 1 && newcount > 0)
262 		sc->share_count++;
263 
264 	if (newref->root_id == btrfs_root_id(sc->root) &&
265 	    newref->wanted_disk_byte == sc->data_bytenr &&
266 	    newref->key_for_search.objectid == sc->inum)
267 		sc->self_ref_count += newref->count;
268 }
269 
270 /*
271  * Add @newref to the @root rbtree, merging identical refs.
272  *
273  * Callers should assume that newref has been freed after calling.
274  */
prelim_ref_insert(const struct btrfs_fs_info * fs_info,struct preftree * preftree,struct prelim_ref * newref,struct share_check * sc)275 static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
276 			      struct preftree *preftree,
277 			      struct prelim_ref *newref,
278 			      struct share_check *sc)
279 {
280 	struct rb_root_cached *root;
281 	struct rb_node **p;
282 	struct rb_node *parent = NULL;
283 	struct prelim_ref *ref;
284 	int result;
285 	bool leftmost = true;
286 
287 	root = &preftree->root;
288 	p = &root->rb_root.rb_node;
289 
290 	while (*p) {
291 		parent = *p;
292 		ref = rb_entry(parent, struct prelim_ref, rbnode);
293 		result = prelim_ref_compare(ref, newref);
294 		if (result < 0) {
295 			p = &(*p)->rb_left;
296 		} else if (result > 0) {
297 			p = &(*p)->rb_right;
298 			leftmost = false;
299 		} else {
300 			/* Identical refs, merge them and free @newref */
301 			struct extent_inode_elem *eie = ref->inode_list;
302 
303 			while (eie && eie->next)
304 				eie = eie->next;
305 
306 			if (!eie)
307 				ref->inode_list = newref->inode_list;
308 			else
309 				eie->next = newref->inode_list;
310 			trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
311 						     preftree->count);
312 			/*
313 			 * A delayed ref can have newref->count < 0.
314 			 * The ref->count is updated to follow any
315 			 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
316 			 */
317 			update_share_count(sc, ref->count,
318 					   ref->count + newref->count, newref);
319 			ref->count += newref->count;
320 			free_pref(newref);
321 			return;
322 		}
323 	}
324 
325 	update_share_count(sc, 0, newref->count, newref);
326 	preftree->count++;
327 	trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
328 	rb_link_node(&newref->rbnode, parent, p);
329 	rb_insert_color_cached(&newref->rbnode, root, leftmost);
330 }
331 
332 /*
333  * Release the entire tree.  We don't care about internal consistency so
334  * just free everything and then reset the tree root.
335  */
prelim_release(struct preftree * preftree)336 static void prelim_release(struct preftree *preftree)
337 {
338 	struct prelim_ref *ref, *next_ref;
339 
340 	rbtree_postorder_for_each_entry_safe(ref, next_ref,
341 					     &preftree->root.rb_root, rbnode) {
342 		free_inode_elem_list(ref->inode_list);
343 		free_pref(ref);
344 	}
345 
346 	preftree->root = RB_ROOT_CACHED;
347 	preftree->count = 0;
348 }
349 
350 /*
351  * the rules for all callers of this function are:
352  * - obtaining the parent is the goal
353  * - if you add a key, you must know that it is a correct key
354  * - if you cannot add the parent or a correct key, then we will look into the
355  *   block later to set a correct key
356  *
357  * delayed refs
358  * ============
359  *        backref type | shared | indirect | shared | indirect
360  * information         |   tree |     tree |   data |     data
361  * --------------------+--------+----------+--------+----------
362  *      parent logical |    y   |     -    |    -   |     -
363  *      key to resolve |    -   |     y    |    y   |     y
364  *  tree block logical |    -   |     -    |    -   |     -
365  *  root for resolving |    y   |     y    |    y   |     y
366  *
367  * - column 1:       we've the parent -> done
368  * - column 2, 3, 4: we use the key to find the parent
369  *
370  * on disk refs (inline or keyed)
371  * ==============================
372  *        backref type | shared | indirect | shared | indirect
373  * information         |   tree |     tree |   data |     data
374  * --------------------+--------+----------+--------+----------
375  *      parent logical |    y   |     -    |    y   |     -
376  *      key to resolve |    -   |     -    |    -   |     y
377  *  tree block logical |    y   |     y    |    y   |     y
378  *  root for resolving |    -   |     y    |    y   |     y
379  *
380  * - column 1, 3: we've the parent -> done
381  * - column 2:    we take the first key from the block to find the parent
382  *                (see add_missing_keys)
383  * - column 4:    we use the key to find the parent
384  *
385  * additional information that's available but not required to find the parent
386  * block might help in merging entries to gain some speed.
387  */
add_prelim_ref(const struct btrfs_fs_info * fs_info,struct preftree * preftree,u64 root_id,const struct btrfs_key * key,int level,u64 parent,u64 wanted_disk_byte,int count,struct share_check * sc,gfp_t gfp_mask)388 static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
389 			  struct preftree *preftree, u64 root_id,
390 			  const struct btrfs_key *key, int level, u64 parent,
391 			  u64 wanted_disk_byte, int count,
392 			  struct share_check *sc, gfp_t gfp_mask)
393 {
394 	struct prelim_ref *ref;
395 
396 	if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
397 		return 0;
398 
399 	ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
400 	if (!ref)
401 		return -ENOMEM;
402 
403 	ref->root_id = root_id;
404 	if (key)
405 		ref->key_for_search = *key;
406 	else
407 		memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
408 
409 	ref->inode_list = NULL;
410 	ref->level = level;
411 	ref->count = count;
412 	ref->parent = parent;
413 	ref->wanted_disk_byte = wanted_disk_byte;
414 	prelim_ref_insert(fs_info, preftree, ref, sc);
415 	return extent_is_shared(sc);
416 }
417 
418 /* direct refs use root == 0, key == NULL */
add_direct_ref(const struct btrfs_fs_info * fs_info,struct preftrees * preftrees,int level,u64 parent,u64 wanted_disk_byte,int count,struct share_check * sc,gfp_t gfp_mask)419 static int add_direct_ref(const struct btrfs_fs_info *fs_info,
420 			  struct preftrees *preftrees, int level, u64 parent,
421 			  u64 wanted_disk_byte, int count,
422 			  struct share_check *sc, gfp_t gfp_mask)
423 {
424 	return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
425 			      parent, wanted_disk_byte, count, sc, gfp_mask);
426 }
427 
428 /* indirect refs use parent == 0 */
add_indirect_ref(const struct btrfs_fs_info * fs_info,struct preftrees * preftrees,u64 root_id,const struct btrfs_key * key,int level,u64 wanted_disk_byte,int count,struct share_check * sc,gfp_t gfp_mask)429 static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
430 			    struct preftrees *preftrees, u64 root_id,
431 			    const struct btrfs_key *key, int level,
432 			    u64 wanted_disk_byte, int count,
433 			    struct share_check *sc, gfp_t gfp_mask)
434 {
435 	struct preftree *tree = &preftrees->indirect;
436 
437 	if (!key)
438 		tree = &preftrees->indirect_missing_keys;
439 	return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
440 			      wanted_disk_byte, count, sc, gfp_mask);
441 }
442 
is_shared_data_backref(struct preftrees * preftrees,u64 bytenr)443 static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
444 {
445 	struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
446 	struct rb_node *parent = NULL;
447 	struct prelim_ref *ref = NULL;
448 	struct prelim_ref target = {};
449 	int result;
450 
451 	target.parent = bytenr;
452 
453 	while (*p) {
454 		parent = *p;
455 		ref = rb_entry(parent, struct prelim_ref, rbnode);
456 		result = prelim_ref_compare(ref, &target);
457 
458 		if (result < 0)
459 			p = &(*p)->rb_left;
460 		else if (result > 0)
461 			p = &(*p)->rb_right;
462 		else
463 			return 1;
464 	}
465 	return 0;
466 }
467 
add_all_parents(struct btrfs_backref_walk_ctx * ctx,struct btrfs_root * root,struct btrfs_path * path,struct ulist * parents,struct preftrees * preftrees,struct prelim_ref * ref,int level)468 static int add_all_parents(struct btrfs_backref_walk_ctx *ctx,
469 			   struct btrfs_root *root, struct btrfs_path *path,
470 			   struct ulist *parents,
471 			   struct preftrees *preftrees, struct prelim_ref *ref,
472 			   int level)
473 {
474 	int ret = 0;
475 	int slot;
476 	struct extent_buffer *eb;
477 	struct btrfs_key key;
478 	struct btrfs_key *key_for_search = &ref->key_for_search;
479 	struct btrfs_file_extent_item *fi;
480 	struct extent_inode_elem *eie = NULL, *old = NULL;
481 	u64 disk_byte;
482 	u64 wanted_disk_byte = ref->wanted_disk_byte;
483 	u64 count = 0;
484 	u64 data_offset;
485 	u8 type;
486 
487 	if (level != 0) {
488 		eb = path->nodes[level];
489 		ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
490 		if (ret < 0)
491 			return ret;
492 		return 0;
493 	}
494 
495 	/*
496 	 * 1. We normally enter this function with the path already pointing to
497 	 *    the first item to check. But sometimes, we may enter it with
498 	 *    slot == nritems.
499 	 * 2. We are searching for normal backref but bytenr of this leaf
500 	 *    matches shared data backref
501 	 * 3. The leaf owner is not equal to the root we are searching
502 	 *
503 	 * For these cases, go to the next leaf before we continue.
504 	 */
505 	eb = path->nodes[0];
506 	if (path->slots[0] >= btrfs_header_nritems(eb) ||
507 	    is_shared_data_backref(preftrees, eb->start) ||
508 	    ref->root_id != btrfs_header_owner(eb)) {
509 		if (ctx->time_seq == BTRFS_SEQ_LAST)
510 			ret = btrfs_next_leaf(root, path);
511 		else
512 			ret = btrfs_next_old_leaf(root, path, ctx->time_seq);
513 	}
514 
515 	while (!ret && count < ref->count) {
516 		eb = path->nodes[0];
517 		slot = path->slots[0];
518 
519 		btrfs_item_key_to_cpu(eb, &key, slot);
520 
521 		if (key.objectid != key_for_search->objectid ||
522 		    key.type != BTRFS_EXTENT_DATA_KEY)
523 			break;
524 
525 		/*
526 		 * We are searching for normal backref but bytenr of this leaf
527 		 * matches shared data backref, OR
528 		 * the leaf owner is not equal to the root we are searching for
529 		 */
530 		if (slot == 0 &&
531 		    (is_shared_data_backref(preftrees, eb->start) ||
532 		     ref->root_id != btrfs_header_owner(eb))) {
533 			if (ctx->time_seq == BTRFS_SEQ_LAST)
534 				ret = btrfs_next_leaf(root, path);
535 			else
536 				ret = btrfs_next_old_leaf(root, path, ctx->time_seq);
537 			continue;
538 		}
539 		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
540 		type = btrfs_file_extent_type(eb, fi);
541 		if (type == BTRFS_FILE_EXTENT_INLINE)
542 			goto next;
543 		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
544 		data_offset = btrfs_file_extent_offset(eb, fi);
545 
546 		if (disk_byte == wanted_disk_byte) {
547 			eie = NULL;
548 			old = NULL;
549 			if (ref->key_for_search.offset == key.offset - data_offset)
550 				count++;
551 			else
552 				goto next;
553 			if (!ctx->skip_inode_ref_list) {
554 				ret = check_extent_in_eb(ctx, &key, eb, fi, &eie);
555 				if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
556 				    ret < 0)
557 					break;
558 			}
559 			if (ret > 0)
560 				goto next;
561 			ret = ulist_add_merge_ptr(parents, eb->start,
562 						  eie, (void **)&old, GFP_NOFS);
563 			if (ret < 0)
564 				break;
565 			if (!ret && !ctx->skip_inode_ref_list) {
566 				while (old->next)
567 					old = old->next;
568 				old->next = eie;
569 			}
570 			eie = NULL;
571 		}
572 next:
573 		if (ctx->time_seq == BTRFS_SEQ_LAST)
574 			ret = btrfs_next_item(root, path);
575 		else
576 			ret = btrfs_next_old_item(root, path, ctx->time_seq);
577 	}
578 
579 	if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
580 		free_inode_elem_list(eie);
581 	else if (ret > 0)
582 		ret = 0;
583 
584 	return ret;
585 }
586 
587 /*
588  * resolve an indirect backref in the form (root_id, key, level)
589  * to a logical address
590  */
resolve_indirect_ref(struct btrfs_backref_walk_ctx * ctx,struct btrfs_path * path,struct preftrees * preftrees,struct prelim_ref * ref,struct ulist * parents)591 static int resolve_indirect_ref(struct btrfs_backref_walk_ctx *ctx,
592 				struct btrfs_path *path,
593 				struct preftrees *preftrees,
594 				struct prelim_ref *ref, struct ulist *parents)
595 {
596 	struct btrfs_root *root;
597 	struct extent_buffer *eb;
598 	int ret = 0;
599 	int root_level;
600 	int level = ref->level;
601 	struct btrfs_key search_key = ref->key_for_search;
602 
603 	/*
604 	 * If we're search_commit_root we could possibly be holding locks on
605 	 * other tree nodes.  This happens when qgroups does backref walks when
606 	 * adding new delayed refs.  To deal with this we need to look in cache
607 	 * for the root, and if we don't find it then we need to search the
608 	 * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage
609 	 * here.
610 	 */
611 	if (path->search_commit_root)
612 		root = btrfs_get_fs_root_commit_root(ctx->fs_info, path, ref->root_id);
613 	else
614 		root = btrfs_get_fs_root(ctx->fs_info, ref->root_id, false);
615 	if (IS_ERR(root)) {
616 		ret = PTR_ERR(root);
617 		goto out_free;
618 	}
619 
620 	if (!path->search_commit_root &&
621 	    test_bit(BTRFS_ROOT_DELETING, &root->state)) {
622 		ret = -ENOENT;
623 		goto out;
624 	}
625 
626 	if (btrfs_is_testing(ctx->fs_info)) {
627 		ret = -ENOENT;
628 		goto out;
629 	}
630 
631 	if (path->search_commit_root)
632 		root_level = btrfs_header_level(root->commit_root);
633 	else if (ctx->time_seq == BTRFS_SEQ_LAST)
634 		root_level = btrfs_header_level(root->node);
635 	else
636 		root_level = btrfs_old_root_level(root, ctx->time_seq);
637 
638 	if (root_level + 1 == level)
639 		goto out;
640 
641 	/*
642 	 * We can often find data backrefs with an offset that is too large
643 	 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
644 	 * subtracting a file's offset with the data offset of its
645 	 * corresponding extent data item. This can happen for example in the
646 	 * clone ioctl.
647 	 *
648 	 * So if we detect such case we set the search key's offset to zero to
649 	 * make sure we will find the matching file extent item at
650 	 * add_all_parents(), otherwise we will miss it because the offset
651 	 * taken form the backref is much larger then the offset of the file
652 	 * extent item. This can make us scan a very large number of file
653 	 * extent items, but at least it will not make us miss any.
654 	 *
655 	 * This is an ugly workaround for a behaviour that should have never
656 	 * existed, but it does and a fix for the clone ioctl would touch a lot
657 	 * of places, cause backwards incompatibility and would not fix the
658 	 * problem for extents cloned with older kernels.
659 	 */
660 	if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
661 	    search_key.offset >= LLONG_MAX)
662 		search_key.offset = 0;
663 	path->lowest_level = level;
664 	if (ctx->time_seq == BTRFS_SEQ_LAST)
665 		ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
666 	else
667 		ret = btrfs_search_old_slot(root, &search_key, path, ctx->time_seq);
668 
669 	btrfs_debug(ctx->fs_info,
670 		"search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
671 		 ref->root_id, level, ref->count, ret,
672 		 ref->key_for_search.objectid, ref->key_for_search.type,
673 		 ref->key_for_search.offset);
674 	if (ret < 0)
675 		goto out;
676 
677 	eb = path->nodes[level];
678 	while (!eb) {
679 		if (WARN_ON(!level)) {
680 			ret = 1;
681 			goto out;
682 		}
683 		level--;
684 		eb = path->nodes[level];
685 	}
686 
687 	ret = add_all_parents(ctx, root, path, parents, preftrees, ref, level);
688 out:
689 	btrfs_put_root(root);
690 out_free:
691 	path->lowest_level = 0;
692 	btrfs_release_path(path);
693 	return ret;
694 }
695 
696 static struct extent_inode_elem *
unode_aux_to_inode_list(struct ulist_node * node)697 unode_aux_to_inode_list(struct ulist_node *node)
698 {
699 	if (!node)
700 		return NULL;
701 	return (struct extent_inode_elem *)(uintptr_t)node->aux;
702 }
703 
free_leaf_list(struct ulist * ulist)704 static void free_leaf_list(struct ulist *ulist)
705 {
706 	struct ulist_node *node;
707 	struct ulist_iterator uiter;
708 
709 	ULIST_ITER_INIT(&uiter);
710 	while ((node = ulist_next(ulist, &uiter)))
711 		free_inode_elem_list(unode_aux_to_inode_list(node));
712 
713 	ulist_free(ulist);
714 }
715 
716 /*
717  * We maintain three separate rbtrees: one for direct refs, one for
718  * indirect refs which have a key, and one for indirect refs which do not
719  * have a key. Each tree does merge on insertion.
720  *
721  * Once all of the references are located, we iterate over the tree of
722  * indirect refs with missing keys. An appropriate key is located and
723  * the ref is moved onto the tree for indirect refs. After all missing
724  * keys are thus located, we iterate over the indirect ref tree, resolve
725  * each reference, and then insert the resolved reference onto the
726  * direct tree (merging there too).
727  *
728  * New backrefs (i.e., for parent nodes) are added to the appropriate
729  * rbtree as they are encountered. The new backrefs are subsequently
730  * resolved as above.
731  */
resolve_indirect_refs(struct btrfs_backref_walk_ctx * ctx,struct btrfs_path * path,struct preftrees * preftrees,struct share_check * sc)732 static int resolve_indirect_refs(struct btrfs_backref_walk_ctx *ctx,
733 				 struct btrfs_path *path,
734 				 struct preftrees *preftrees,
735 				 struct share_check *sc)
736 {
737 	int err;
738 	int ret = 0;
739 	struct ulist *parents;
740 	struct ulist_node *node;
741 	struct ulist_iterator uiter;
742 	struct rb_node *rnode;
743 
744 	parents = ulist_alloc(GFP_NOFS);
745 	if (!parents)
746 		return -ENOMEM;
747 
748 	/*
749 	 * We could trade memory usage for performance here by iterating
750 	 * the tree, allocating new refs for each insertion, and then
751 	 * freeing the entire indirect tree when we're done.  In some test
752 	 * cases, the tree can grow quite large (~200k objects).
753 	 */
754 	while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
755 		struct prelim_ref *ref;
756 
757 		ref = rb_entry(rnode, struct prelim_ref, rbnode);
758 		if (WARN(ref->parent,
759 			 "BUG: direct ref found in indirect tree")) {
760 			ret = -EINVAL;
761 			goto out;
762 		}
763 
764 		rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
765 		preftrees->indirect.count--;
766 
767 		if (ref->count == 0) {
768 			free_pref(ref);
769 			continue;
770 		}
771 
772 		if (sc && ref->root_id != btrfs_root_id(sc->root)) {
773 			free_pref(ref);
774 			ret = BACKREF_FOUND_SHARED;
775 			goto out;
776 		}
777 		err = resolve_indirect_ref(ctx, path, preftrees, ref, parents);
778 		/*
779 		 * we can only tolerate ENOENT,otherwise,we should catch error
780 		 * and return directly.
781 		 */
782 		if (err == -ENOENT) {
783 			prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref,
784 					  NULL);
785 			continue;
786 		} else if (err) {
787 			free_pref(ref);
788 			ret = err;
789 			goto out;
790 		}
791 
792 		/* we put the first parent into the ref at hand */
793 		ULIST_ITER_INIT(&uiter);
794 		node = ulist_next(parents, &uiter);
795 		ref->parent = node ? node->val : 0;
796 		ref->inode_list = unode_aux_to_inode_list(node);
797 
798 		/* Add a prelim_ref(s) for any other parent(s). */
799 		while ((node = ulist_next(parents, &uiter))) {
800 			struct prelim_ref *new_ref;
801 
802 			new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
803 						   GFP_NOFS);
804 			if (!new_ref) {
805 				free_pref(ref);
806 				ret = -ENOMEM;
807 				goto out;
808 			}
809 			memcpy(new_ref, ref, sizeof(*ref));
810 			new_ref->parent = node->val;
811 			new_ref->inode_list = unode_aux_to_inode_list(node);
812 			prelim_ref_insert(ctx->fs_info, &preftrees->direct,
813 					  new_ref, NULL);
814 		}
815 
816 		/*
817 		 * Now it's a direct ref, put it in the direct tree. We must
818 		 * do this last because the ref could be merged/freed here.
819 		 */
820 		prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref, NULL);
821 
822 		ulist_reinit(parents);
823 		cond_resched();
824 	}
825 out:
826 	/*
827 	 * We may have inode lists attached to refs in the parents ulist, so we
828 	 * must free them before freeing the ulist and its refs.
829 	 */
830 	free_leaf_list(parents);
831 	return ret;
832 }
833 
834 /*
835  * read tree blocks and add keys where required.
836  */
add_missing_keys(struct btrfs_fs_info * fs_info,struct preftrees * preftrees,bool lock)837 static int add_missing_keys(struct btrfs_fs_info *fs_info,
838 			    struct preftrees *preftrees, bool lock)
839 {
840 	struct prelim_ref *ref;
841 	struct extent_buffer *eb;
842 	struct preftree *tree = &preftrees->indirect_missing_keys;
843 	struct rb_node *node;
844 
845 	while ((node = rb_first_cached(&tree->root))) {
846 		struct btrfs_tree_parent_check check = { 0 };
847 
848 		ref = rb_entry(node, struct prelim_ref, rbnode);
849 		rb_erase_cached(node, &tree->root);
850 
851 		BUG_ON(ref->parent);	/* should not be a direct ref */
852 		BUG_ON(ref->key_for_search.type);
853 		BUG_ON(!ref->wanted_disk_byte);
854 
855 		check.level = ref->level - 1;
856 		check.owner_root = ref->root_id;
857 
858 		eb = read_tree_block(fs_info, ref->wanted_disk_byte, &check);
859 		if (IS_ERR(eb)) {
860 			free_pref(ref);
861 			return PTR_ERR(eb);
862 		}
863 		if (!extent_buffer_uptodate(eb)) {
864 			free_pref(ref);
865 			free_extent_buffer(eb);
866 			return -EIO;
867 		}
868 
869 		if (lock)
870 			btrfs_tree_read_lock(eb);
871 		if (btrfs_header_level(eb) == 0)
872 			btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
873 		else
874 			btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
875 		if (lock)
876 			btrfs_tree_read_unlock(eb);
877 		free_extent_buffer(eb);
878 		prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
879 		cond_resched();
880 	}
881 	return 0;
882 }
883 
884 /*
885  * add all currently queued delayed refs from this head whose seq nr is
886  * smaller or equal that seq to the list
887  */
add_delayed_refs(const struct btrfs_fs_info * fs_info,struct btrfs_delayed_ref_head * head,u64 seq,struct preftrees * preftrees,struct share_check * sc)888 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
889 			    struct btrfs_delayed_ref_head *head, u64 seq,
890 			    struct preftrees *preftrees, struct share_check *sc)
891 {
892 	struct btrfs_delayed_ref_node *node;
893 	struct btrfs_key key;
894 	struct rb_node *n;
895 	int count;
896 	int ret = 0;
897 
898 	spin_lock(&head->lock);
899 	for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
900 		node = rb_entry(n, struct btrfs_delayed_ref_node,
901 				ref_node);
902 		if (node->seq > seq)
903 			continue;
904 
905 		switch (node->action) {
906 		case BTRFS_ADD_DELAYED_EXTENT:
907 		case BTRFS_UPDATE_DELAYED_HEAD:
908 			WARN_ON(1);
909 			continue;
910 		case BTRFS_ADD_DELAYED_REF:
911 			count = node->ref_mod;
912 			break;
913 		case BTRFS_DROP_DELAYED_REF:
914 			count = node->ref_mod * -1;
915 			break;
916 		default:
917 			BUG();
918 		}
919 		switch (node->type) {
920 		case BTRFS_TREE_BLOCK_REF_KEY: {
921 			/* NORMAL INDIRECT METADATA backref */
922 			struct btrfs_key *key_ptr = NULL;
923 			/* The owner of a tree block ref is the level. */
924 			int level = btrfs_delayed_ref_owner(node);
925 
926 			if (head->extent_op && head->extent_op->update_key) {
927 				btrfs_disk_key_to_cpu(&key, &head->extent_op->key);
928 				key_ptr = &key;
929 			}
930 
931 			ret = add_indirect_ref(fs_info, preftrees, node->ref_root,
932 					       key_ptr, level + 1, node->bytenr,
933 					       count, sc, GFP_ATOMIC);
934 			break;
935 		}
936 		case BTRFS_SHARED_BLOCK_REF_KEY: {
937 			/*
938 			 * SHARED DIRECT METADATA backref
939 			 *
940 			 * The owner of a tree block ref is the level.
941 			 */
942 			int level = btrfs_delayed_ref_owner(node);
943 
944 			ret = add_direct_ref(fs_info, preftrees, level + 1,
945 					     node->parent, node->bytenr, count,
946 					     sc, GFP_ATOMIC);
947 			break;
948 		}
949 		case BTRFS_EXTENT_DATA_REF_KEY: {
950 			/* NORMAL INDIRECT DATA backref */
951 			key.objectid = btrfs_delayed_ref_owner(node);
952 			key.type = BTRFS_EXTENT_DATA_KEY;
953 			key.offset = btrfs_delayed_ref_offset(node);
954 
955 			/*
956 			 * If we have a share check context and a reference for
957 			 * another inode, we can't exit immediately. This is
958 			 * because even if this is a BTRFS_ADD_DELAYED_REF
959 			 * reference we may find next a BTRFS_DROP_DELAYED_REF
960 			 * which cancels out this ADD reference.
961 			 *
962 			 * If this is a DROP reference and there was no previous
963 			 * ADD reference, then we need to signal that when we
964 			 * process references from the extent tree (through
965 			 * add_inline_refs() and add_keyed_refs()), we should
966 			 * not exit early if we find a reference for another
967 			 * inode, because one of the delayed DROP references
968 			 * may cancel that reference in the extent tree.
969 			 */
970 			if (sc && count < 0)
971 				sc->have_delayed_delete_refs = true;
972 
973 			ret = add_indirect_ref(fs_info, preftrees, node->ref_root,
974 					       &key, 0, node->bytenr, count, sc,
975 					       GFP_ATOMIC);
976 			break;
977 		}
978 		case BTRFS_SHARED_DATA_REF_KEY: {
979 			/* SHARED DIRECT FULL backref */
980 			ret = add_direct_ref(fs_info, preftrees, 0, node->parent,
981 					     node->bytenr, count, sc,
982 					     GFP_ATOMIC);
983 			break;
984 		}
985 		default:
986 			WARN_ON(1);
987 		}
988 		/*
989 		 * We must ignore BACKREF_FOUND_SHARED until all delayed
990 		 * refs have been checked.
991 		 */
992 		if (ret && (ret != BACKREF_FOUND_SHARED))
993 			break;
994 	}
995 	if (!ret)
996 		ret = extent_is_shared(sc);
997 
998 	spin_unlock(&head->lock);
999 	return ret;
1000 }
1001 
1002 /*
1003  * add all inline backrefs for bytenr to the list
1004  *
1005  * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1006  */
add_inline_refs(struct btrfs_backref_walk_ctx * ctx,struct btrfs_path * path,int * info_level,struct preftrees * preftrees,struct share_check * sc)1007 static int add_inline_refs(struct btrfs_backref_walk_ctx *ctx,
1008 			   struct btrfs_path *path,
1009 			   int *info_level, struct preftrees *preftrees,
1010 			   struct share_check *sc)
1011 {
1012 	int ret = 0;
1013 	int slot;
1014 	struct extent_buffer *leaf;
1015 	struct btrfs_key key;
1016 	struct btrfs_key found_key;
1017 	unsigned long ptr;
1018 	unsigned long end;
1019 	struct btrfs_extent_item *ei;
1020 	u64 flags;
1021 	u64 item_size;
1022 
1023 	/*
1024 	 * enumerate all inline refs
1025 	 */
1026 	leaf = path->nodes[0];
1027 	slot = path->slots[0];
1028 
1029 	item_size = btrfs_item_size(leaf, slot);
1030 	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1031 
1032 	if (ctx->check_extent_item) {
1033 		ret = ctx->check_extent_item(ctx->bytenr, ei, leaf, ctx->user_ctx);
1034 		if (ret)
1035 			return ret;
1036 	}
1037 
1038 	flags = btrfs_extent_flags(leaf, ei);
1039 	btrfs_item_key_to_cpu(leaf, &found_key, slot);
1040 
1041 	ptr = (unsigned long)(ei + 1);
1042 	end = (unsigned long)ei + item_size;
1043 
1044 	if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1045 	    flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1046 		struct btrfs_tree_block_info *info;
1047 
1048 		info = (struct btrfs_tree_block_info *)ptr;
1049 		*info_level = btrfs_tree_block_level(leaf, info);
1050 		ptr += sizeof(struct btrfs_tree_block_info);
1051 		BUG_ON(ptr > end);
1052 	} else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
1053 		*info_level = found_key.offset;
1054 	} else {
1055 		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1056 	}
1057 
1058 	while (ptr < end) {
1059 		struct btrfs_extent_inline_ref *iref;
1060 		u64 offset;
1061 		int type;
1062 
1063 		iref = (struct btrfs_extent_inline_ref *)ptr;
1064 		type = btrfs_get_extent_inline_ref_type(leaf, iref,
1065 							BTRFS_REF_TYPE_ANY);
1066 		if (type == BTRFS_REF_TYPE_INVALID)
1067 			return -EUCLEAN;
1068 
1069 		offset = btrfs_extent_inline_ref_offset(leaf, iref);
1070 
1071 		switch (type) {
1072 		case BTRFS_SHARED_BLOCK_REF_KEY:
1073 			ret = add_direct_ref(ctx->fs_info, preftrees,
1074 					     *info_level + 1, offset,
1075 					     ctx->bytenr, 1, NULL, GFP_NOFS);
1076 			break;
1077 		case BTRFS_SHARED_DATA_REF_KEY: {
1078 			struct btrfs_shared_data_ref *sdref;
1079 			int count;
1080 
1081 			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
1082 			count = btrfs_shared_data_ref_count(leaf, sdref);
1083 
1084 			ret = add_direct_ref(ctx->fs_info, preftrees, 0, offset,
1085 					     ctx->bytenr, count, sc, GFP_NOFS);
1086 			break;
1087 		}
1088 		case BTRFS_TREE_BLOCK_REF_KEY:
1089 			ret = add_indirect_ref(ctx->fs_info, preftrees, offset,
1090 					       NULL, *info_level + 1,
1091 					       ctx->bytenr, 1, NULL, GFP_NOFS);
1092 			break;
1093 		case BTRFS_EXTENT_DATA_REF_KEY: {
1094 			struct btrfs_extent_data_ref *dref;
1095 			int count;
1096 			u64 root;
1097 
1098 			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1099 			count = btrfs_extent_data_ref_count(leaf, dref);
1100 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
1101 								      dref);
1102 			key.type = BTRFS_EXTENT_DATA_KEY;
1103 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1104 
1105 			if (sc && key.objectid != sc->inum &&
1106 			    !sc->have_delayed_delete_refs) {
1107 				ret = BACKREF_FOUND_SHARED;
1108 				break;
1109 			}
1110 
1111 			root = btrfs_extent_data_ref_root(leaf, dref);
1112 
1113 			if (!ctx->skip_data_ref ||
1114 			    !ctx->skip_data_ref(root, key.objectid, key.offset,
1115 						ctx->user_ctx))
1116 				ret = add_indirect_ref(ctx->fs_info, preftrees,
1117 						       root, &key, 0, ctx->bytenr,
1118 						       count, sc, GFP_NOFS);
1119 			break;
1120 		}
1121 		case BTRFS_EXTENT_OWNER_REF_KEY:
1122 			ASSERT(btrfs_fs_incompat(ctx->fs_info, SIMPLE_QUOTA));
1123 			break;
1124 		default:
1125 			WARN_ON(1);
1126 		}
1127 		if (ret)
1128 			return ret;
1129 		ptr += btrfs_extent_inline_ref_size(type);
1130 	}
1131 
1132 	return 0;
1133 }
1134 
1135 /*
1136  * add all non-inline backrefs for bytenr to the list
1137  *
1138  * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1139  */
add_keyed_refs(struct btrfs_backref_walk_ctx * ctx,struct btrfs_root * extent_root,struct btrfs_path * path,int info_level,struct preftrees * preftrees,struct share_check * sc)1140 static int add_keyed_refs(struct btrfs_backref_walk_ctx *ctx,
1141 			  struct btrfs_root *extent_root,
1142 			  struct btrfs_path *path,
1143 			  int info_level, struct preftrees *preftrees,
1144 			  struct share_check *sc)
1145 {
1146 	struct btrfs_fs_info *fs_info = extent_root->fs_info;
1147 	int ret;
1148 	int slot;
1149 	struct extent_buffer *leaf;
1150 	struct btrfs_key key;
1151 
1152 	while (1) {
1153 		ret = btrfs_next_item(extent_root, path);
1154 		if (ret < 0)
1155 			break;
1156 		if (ret) {
1157 			ret = 0;
1158 			break;
1159 		}
1160 
1161 		slot = path->slots[0];
1162 		leaf = path->nodes[0];
1163 		btrfs_item_key_to_cpu(leaf, &key, slot);
1164 
1165 		if (key.objectid != ctx->bytenr)
1166 			break;
1167 		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1168 			continue;
1169 		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1170 			break;
1171 
1172 		switch (key.type) {
1173 		case BTRFS_SHARED_BLOCK_REF_KEY:
1174 			/* SHARED DIRECT METADATA backref */
1175 			ret = add_direct_ref(fs_info, preftrees,
1176 					     info_level + 1, key.offset,
1177 					     ctx->bytenr, 1, NULL, GFP_NOFS);
1178 			break;
1179 		case BTRFS_SHARED_DATA_REF_KEY: {
1180 			/* SHARED DIRECT FULL backref */
1181 			struct btrfs_shared_data_ref *sdref;
1182 			int count;
1183 
1184 			sdref = btrfs_item_ptr(leaf, slot,
1185 					      struct btrfs_shared_data_ref);
1186 			count = btrfs_shared_data_ref_count(leaf, sdref);
1187 			ret = add_direct_ref(fs_info, preftrees, 0,
1188 					     key.offset, ctx->bytenr, count,
1189 					     sc, GFP_NOFS);
1190 			break;
1191 		}
1192 		case BTRFS_TREE_BLOCK_REF_KEY:
1193 			/* NORMAL INDIRECT METADATA backref */
1194 			ret = add_indirect_ref(fs_info, preftrees, key.offset,
1195 					       NULL, info_level + 1, ctx->bytenr,
1196 					       1, NULL, GFP_NOFS);
1197 			break;
1198 		case BTRFS_EXTENT_DATA_REF_KEY: {
1199 			/* NORMAL INDIRECT DATA backref */
1200 			struct btrfs_extent_data_ref *dref;
1201 			int count;
1202 			u64 root;
1203 
1204 			dref = btrfs_item_ptr(leaf, slot,
1205 					      struct btrfs_extent_data_ref);
1206 			count = btrfs_extent_data_ref_count(leaf, dref);
1207 			key.objectid = btrfs_extent_data_ref_objectid(leaf,
1208 								      dref);
1209 			key.type = BTRFS_EXTENT_DATA_KEY;
1210 			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1211 
1212 			if (sc && key.objectid != sc->inum &&
1213 			    !sc->have_delayed_delete_refs) {
1214 				ret = BACKREF_FOUND_SHARED;
1215 				break;
1216 			}
1217 
1218 			root = btrfs_extent_data_ref_root(leaf, dref);
1219 
1220 			if (!ctx->skip_data_ref ||
1221 			    !ctx->skip_data_ref(root, key.objectid, key.offset,
1222 						ctx->user_ctx))
1223 				ret = add_indirect_ref(fs_info, preftrees, root,
1224 						       &key, 0, ctx->bytenr,
1225 						       count, sc, GFP_NOFS);
1226 			break;
1227 		}
1228 		default:
1229 			WARN_ON(1);
1230 		}
1231 		if (ret)
1232 			return ret;
1233 
1234 	}
1235 
1236 	return ret;
1237 }
1238 
1239 /*
1240  * The caller has joined a transaction or is holding a read lock on the
1241  * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1242  * snapshot field changing while updating or checking the cache.
1243  */
lookup_backref_shared_cache(struct btrfs_backref_share_check_ctx * ctx,struct btrfs_root * root,u64 bytenr,int level,bool * is_shared)1244 static bool lookup_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx,
1245 					struct btrfs_root *root,
1246 					u64 bytenr, int level, bool *is_shared)
1247 {
1248 	const struct btrfs_fs_info *fs_info = root->fs_info;
1249 	struct btrfs_backref_shared_cache_entry *entry;
1250 
1251 	if (!current->journal_info)
1252 		lockdep_assert_held(&fs_info->commit_root_sem);
1253 
1254 	if (!ctx->use_path_cache)
1255 		return false;
1256 
1257 	if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1258 		return false;
1259 
1260 	/*
1261 	 * Level -1 is used for the data extent, which is not reliable to cache
1262 	 * because its reference count can increase or decrease without us
1263 	 * realizing. We cache results only for extent buffers that lead from
1264 	 * the root node down to the leaf with the file extent item.
1265 	 */
1266 	ASSERT(level >= 0);
1267 
1268 	entry = &ctx->path_cache_entries[level];
1269 
1270 	/* Unused cache entry or being used for some other extent buffer. */
1271 	if (entry->bytenr != bytenr)
1272 		return false;
1273 
1274 	/*
1275 	 * We cached a false result, but the last snapshot generation of the
1276 	 * root changed, so we now have a snapshot. Don't trust the result.
1277 	 */
1278 	if (!entry->is_shared &&
1279 	    entry->gen != btrfs_root_last_snapshot(&root->root_item))
1280 		return false;
1281 
1282 	/*
1283 	 * If we cached a true result and the last generation used for dropping
1284 	 * a root changed, we can not trust the result, because the dropped root
1285 	 * could be a snapshot sharing this extent buffer.
1286 	 */
1287 	if (entry->is_shared &&
1288 	    entry->gen != btrfs_get_last_root_drop_gen(fs_info))
1289 		return false;
1290 
1291 	*is_shared = entry->is_shared;
1292 	/*
1293 	 * If the node at this level is shared, than all nodes below are also
1294 	 * shared. Currently some of the nodes below may be marked as not shared
1295 	 * because we have just switched from one leaf to another, and switched
1296 	 * also other nodes above the leaf and below the current level, so mark
1297 	 * them as shared.
1298 	 */
1299 	if (*is_shared) {
1300 		for (int i = 0; i < level; i++) {
1301 			ctx->path_cache_entries[i].is_shared = true;
1302 			ctx->path_cache_entries[i].gen = entry->gen;
1303 		}
1304 	}
1305 
1306 	return true;
1307 }
1308 
1309 /*
1310  * The caller has joined a transaction or is holding a read lock on the
1311  * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1312  * snapshot field changing while updating or checking the cache.
1313  */
store_backref_shared_cache(struct btrfs_backref_share_check_ctx * ctx,struct btrfs_root * root,u64 bytenr,int level,bool is_shared)1314 static void store_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx,
1315 				       struct btrfs_root *root,
1316 				       u64 bytenr, int level, bool is_shared)
1317 {
1318 	const struct btrfs_fs_info *fs_info = root->fs_info;
1319 	struct btrfs_backref_shared_cache_entry *entry;
1320 	u64 gen;
1321 
1322 	if (!current->journal_info)
1323 		lockdep_assert_held(&fs_info->commit_root_sem);
1324 
1325 	if (!ctx->use_path_cache)
1326 		return;
1327 
1328 	if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1329 		return;
1330 
1331 	/*
1332 	 * Level -1 is used for the data extent, which is not reliable to cache
1333 	 * because its reference count can increase or decrease without us
1334 	 * realizing. We cache results only for extent buffers that lead from
1335 	 * the root node down to the leaf with the file extent item.
1336 	 */
1337 	ASSERT(level >= 0);
1338 
1339 	if (is_shared)
1340 		gen = btrfs_get_last_root_drop_gen(fs_info);
1341 	else
1342 		gen = btrfs_root_last_snapshot(&root->root_item);
1343 
1344 	entry = &ctx->path_cache_entries[level];
1345 	entry->bytenr = bytenr;
1346 	entry->is_shared = is_shared;
1347 	entry->gen = gen;
1348 
1349 	/*
1350 	 * If we found an extent buffer is shared, set the cache result for all
1351 	 * extent buffers below it to true. As nodes in the path are COWed,
1352 	 * their sharedness is moved to their children, and if a leaf is COWed,
1353 	 * then the sharedness of a data extent becomes direct, the refcount of
1354 	 * data extent is increased in the extent item at the extent tree.
1355 	 */
1356 	if (is_shared) {
1357 		for (int i = 0; i < level; i++) {
1358 			entry = &ctx->path_cache_entries[i];
1359 			entry->is_shared = is_shared;
1360 			entry->gen = gen;
1361 		}
1362 	}
1363 }
1364 
1365 /*
1366  * this adds all existing backrefs (inline backrefs, backrefs and delayed
1367  * refs) for the given bytenr to the refs list, merges duplicates and resolves
1368  * indirect refs to their parent bytenr.
1369  * When roots are found, they're added to the roots list
1370  *
1371  * @ctx:     Backref walking context object, must be not NULL.
1372  * @sc:      If !NULL, then immediately return BACKREF_FOUND_SHARED when a
1373  *           shared extent is detected.
1374  *
1375  * Otherwise this returns 0 for success and <0 for an error.
1376  *
1377  * FIXME some caching might speed things up
1378  */
find_parent_nodes(struct btrfs_backref_walk_ctx * ctx,struct share_check * sc)1379 static int find_parent_nodes(struct btrfs_backref_walk_ctx *ctx,
1380 			     struct share_check *sc)
1381 {
1382 	struct btrfs_root *root = btrfs_extent_root(ctx->fs_info, ctx->bytenr);
1383 	struct btrfs_key key;
1384 	struct btrfs_path *path;
1385 	struct btrfs_delayed_ref_root *delayed_refs = NULL;
1386 	struct btrfs_delayed_ref_head *head;
1387 	int info_level = 0;
1388 	int ret;
1389 	struct prelim_ref *ref;
1390 	struct rb_node *node;
1391 	struct extent_inode_elem *eie = NULL;
1392 	struct preftrees preftrees = {
1393 		.direct = PREFTREE_INIT,
1394 		.indirect = PREFTREE_INIT,
1395 		.indirect_missing_keys = PREFTREE_INIT
1396 	};
1397 
1398 	/* Roots ulist is not needed when using a sharedness check context. */
1399 	if (sc)
1400 		ASSERT(ctx->roots == NULL);
1401 
1402 	key.objectid = ctx->bytenr;
1403 	key.offset = (u64)-1;
1404 	if (btrfs_fs_incompat(ctx->fs_info, SKINNY_METADATA))
1405 		key.type = BTRFS_METADATA_ITEM_KEY;
1406 	else
1407 		key.type = BTRFS_EXTENT_ITEM_KEY;
1408 
1409 	path = btrfs_alloc_path();
1410 	if (!path)
1411 		return -ENOMEM;
1412 	if (!ctx->trans) {
1413 		path->search_commit_root = 1;
1414 		path->skip_locking = 1;
1415 	}
1416 
1417 	if (ctx->time_seq == BTRFS_SEQ_LAST)
1418 		path->skip_locking = 1;
1419 
1420 again:
1421 	head = NULL;
1422 
1423 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1424 	if (ret < 0)
1425 		goto out;
1426 	if (ret == 0) {
1427 		/*
1428 		 * Key with offset -1 found, there would have to exist an extent
1429 		 * item with such offset, but this is out of the valid range.
1430 		 */
1431 		ret = -EUCLEAN;
1432 		goto out;
1433 	}
1434 
1435 	if (ctx->trans && likely(ctx->trans->type != __TRANS_DUMMY) &&
1436 	    ctx->time_seq != BTRFS_SEQ_LAST) {
1437 		/*
1438 		 * We have a specific time_seq we care about and trans which
1439 		 * means we have the path lock, we need to grab the ref head and
1440 		 * lock it so we have a consistent view of the refs at the given
1441 		 * time.
1442 		 */
1443 		delayed_refs = &ctx->trans->transaction->delayed_refs;
1444 		spin_lock(&delayed_refs->lock);
1445 		head = btrfs_find_delayed_ref_head(delayed_refs, ctx->bytenr);
1446 		if (head) {
1447 			if (!mutex_trylock(&head->mutex)) {
1448 				refcount_inc(&head->refs);
1449 				spin_unlock(&delayed_refs->lock);
1450 
1451 				btrfs_release_path(path);
1452 
1453 				/*
1454 				 * Mutex was contended, block until it's
1455 				 * released and try again
1456 				 */
1457 				mutex_lock(&head->mutex);
1458 				mutex_unlock(&head->mutex);
1459 				btrfs_put_delayed_ref_head(head);
1460 				goto again;
1461 			}
1462 			spin_unlock(&delayed_refs->lock);
1463 			ret = add_delayed_refs(ctx->fs_info, head, ctx->time_seq,
1464 					       &preftrees, sc);
1465 			mutex_unlock(&head->mutex);
1466 			if (ret)
1467 				goto out;
1468 		} else {
1469 			spin_unlock(&delayed_refs->lock);
1470 		}
1471 	}
1472 
1473 	if (path->slots[0]) {
1474 		struct extent_buffer *leaf;
1475 		int slot;
1476 
1477 		path->slots[0]--;
1478 		leaf = path->nodes[0];
1479 		slot = path->slots[0];
1480 		btrfs_item_key_to_cpu(leaf, &key, slot);
1481 		if (key.objectid == ctx->bytenr &&
1482 		    (key.type == BTRFS_EXTENT_ITEM_KEY ||
1483 		     key.type == BTRFS_METADATA_ITEM_KEY)) {
1484 			ret = add_inline_refs(ctx, path, &info_level,
1485 					      &preftrees, sc);
1486 			if (ret)
1487 				goto out;
1488 			ret = add_keyed_refs(ctx, root, path, info_level,
1489 					     &preftrees, sc);
1490 			if (ret)
1491 				goto out;
1492 		}
1493 	}
1494 
1495 	/*
1496 	 * If we have a share context and we reached here, it means the extent
1497 	 * is not directly shared (no multiple reference items for it),
1498 	 * otherwise we would have exited earlier with a return value of
1499 	 * BACKREF_FOUND_SHARED after processing delayed references or while
1500 	 * processing inline or keyed references from the extent tree.
1501 	 * The extent may however be indirectly shared through shared subtrees
1502 	 * as a result from creating snapshots, so we determine below what is
1503 	 * its parent node, in case we are dealing with a metadata extent, or
1504 	 * what's the leaf (or leaves), from a fs tree, that has a file extent
1505 	 * item pointing to it in case we are dealing with a data extent.
1506 	 */
1507 	ASSERT(extent_is_shared(sc) == 0);
1508 
1509 	/*
1510 	 * If we are here for a data extent and we have a share_check structure
1511 	 * it means the data extent is not directly shared (does not have
1512 	 * multiple reference items), so we have to check if a path in the fs
1513 	 * tree (going from the root node down to the leaf that has the file
1514 	 * extent item pointing to the data extent) is shared, that is, if any
1515 	 * of the extent buffers in the path is referenced by other trees.
1516 	 */
1517 	if (sc && ctx->bytenr == sc->data_bytenr) {
1518 		/*
1519 		 * If our data extent is from a generation more recent than the
1520 		 * last generation used to snapshot the root, then we know that
1521 		 * it can not be shared through subtrees, so we can skip
1522 		 * resolving indirect references, there's no point in
1523 		 * determining the extent buffers for the path from the fs tree
1524 		 * root node down to the leaf that has the file extent item that
1525 		 * points to the data extent.
1526 		 */
1527 		if (sc->data_extent_gen >
1528 		    btrfs_root_last_snapshot(&sc->root->root_item)) {
1529 			ret = BACKREF_FOUND_NOT_SHARED;
1530 			goto out;
1531 		}
1532 
1533 		/*
1534 		 * If we are only determining if a data extent is shared or not
1535 		 * and the corresponding file extent item is located in the same
1536 		 * leaf as the previous file extent item, we can skip resolving
1537 		 * indirect references for a data extent, since the fs tree path
1538 		 * is the same (same leaf, so same path). We skip as long as the
1539 		 * cached result for the leaf is valid and only if there's only
1540 		 * one file extent item pointing to the data extent, because in
1541 		 * the case of multiple file extent items, they may be located
1542 		 * in different leaves and therefore we have multiple paths.
1543 		 */
1544 		if (sc->ctx->curr_leaf_bytenr == sc->ctx->prev_leaf_bytenr &&
1545 		    sc->self_ref_count == 1) {
1546 			bool cached;
1547 			bool is_shared;
1548 
1549 			cached = lookup_backref_shared_cache(sc->ctx, sc->root,
1550 						     sc->ctx->curr_leaf_bytenr,
1551 						     0, &is_shared);
1552 			if (cached) {
1553 				if (is_shared)
1554 					ret = BACKREF_FOUND_SHARED;
1555 				else
1556 					ret = BACKREF_FOUND_NOT_SHARED;
1557 				goto out;
1558 			}
1559 		}
1560 	}
1561 
1562 	btrfs_release_path(path);
1563 
1564 	ret = add_missing_keys(ctx->fs_info, &preftrees, path->skip_locking == 0);
1565 	if (ret)
1566 		goto out;
1567 
1568 	WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1569 
1570 	ret = resolve_indirect_refs(ctx, path, &preftrees, sc);
1571 	if (ret)
1572 		goto out;
1573 
1574 	WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1575 
1576 	/*
1577 	 * This walks the tree of merged and resolved refs. Tree blocks are
1578 	 * read in as needed. Unique entries are added to the ulist, and
1579 	 * the list of found roots is updated.
1580 	 *
1581 	 * We release the entire tree in one go before returning.
1582 	 */
1583 	node = rb_first_cached(&preftrees.direct.root);
1584 	while (node) {
1585 		ref = rb_entry(node, struct prelim_ref, rbnode);
1586 		node = rb_next(&ref->rbnode);
1587 		/*
1588 		 * ref->count < 0 can happen here if there are delayed
1589 		 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1590 		 * prelim_ref_insert() relies on this when merging
1591 		 * identical refs to keep the overall count correct.
1592 		 * prelim_ref_insert() will merge only those refs
1593 		 * which compare identically.  Any refs having
1594 		 * e.g. different offsets would not be merged,
1595 		 * and would retain their original ref->count < 0.
1596 		 */
1597 		if (ctx->roots && ref->count && ref->root_id && ref->parent == 0) {
1598 			/* no parent == root of tree */
1599 			ret = ulist_add(ctx->roots, ref->root_id, 0, GFP_NOFS);
1600 			if (ret < 0)
1601 				goto out;
1602 		}
1603 		if (ref->count && ref->parent) {
1604 			if (!ctx->skip_inode_ref_list && !ref->inode_list &&
1605 			    ref->level == 0) {
1606 				struct btrfs_tree_parent_check check = { 0 };
1607 				struct extent_buffer *eb;
1608 
1609 				check.level = ref->level;
1610 
1611 				eb = read_tree_block(ctx->fs_info, ref->parent,
1612 						     &check);
1613 				if (IS_ERR(eb)) {
1614 					ret = PTR_ERR(eb);
1615 					goto out;
1616 				}
1617 				if (!extent_buffer_uptodate(eb)) {
1618 					free_extent_buffer(eb);
1619 					ret = -EIO;
1620 					goto out;
1621 				}
1622 
1623 				if (!path->skip_locking)
1624 					btrfs_tree_read_lock(eb);
1625 				ret = find_extent_in_eb(ctx, eb, &eie);
1626 				if (!path->skip_locking)
1627 					btrfs_tree_read_unlock(eb);
1628 				free_extent_buffer(eb);
1629 				if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
1630 				    ret < 0)
1631 					goto out;
1632 				ref->inode_list = eie;
1633 				/*
1634 				 * We transferred the list ownership to the ref,
1635 				 * so set to NULL to avoid a double free in case
1636 				 * an error happens after this.
1637 				 */
1638 				eie = NULL;
1639 			}
1640 			ret = ulist_add_merge_ptr(ctx->refs, ref->parent,
1641 						  ref->inode_list,
1642 						  (void **)&eie, GFP_NOFS);
1643 			if (ret < 0)
1644 				goto out;
1645 			if (!ret && !ctx->skip_inode_ref_list) {
1646 				/*
1647 				 * We've recorded that parent, so we must extend
1648 				 * its inode list here.
1649 				 *
1650 				 * However if there was corruption we may not
1651 				 * have found an eie, return an error in this
1652 				 * case.
1653 				 */
1654 				ASSERT(eie);
1655 				if (!eie) {
1656 					ret = -EUCLEAN;
1657 					goto out;
1658 				}
1659 				while (eie->next)
1660 					eie = eie->next;
1661 				eie->next = ref->inode_list;
1662 			}
1663 			eie = NULL;
1664 			/*
1665 			 * We have transferred the inode list ownership from
1666 			 * this ref to the ref we added to the 'refs' ulist.
1667 			 * So set this ref's inode list to NULL to avoid
1668 			 * use-after-free when our caller uses it or double
1669 			 * frees in case an error happens before we return.
1670 			 */
1671 			ref->inode_list = NULL;
1672 		}
1673 		cond_resched();
1674 	}
1675 
1676 out:
1677 	btrfs_free_path(path);
1678 
1679 	prelim_release(&preftrees.direct);
1680 	prelim_release(&preftrees.indirect);
1681 	prelim_release(&preftrees.indirect_missing_keys);
1682 
1683 	if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
1684 		free_inode_elem_list(eie);
1685 	return ret;
1686 }
1687 
1688 /*
1689  * Finds all leaves with a reference to the specified combination of
1690  * @ctx->bytenr and @ctx->extent_item_pos. The bytenr of the found leaves are
1691  * added to the ulist at @ctx->refs, and that ulist is allocated by this
1692  * function. The caller should free the ulist with free_leaf_list() if
1693  * @ctx->ignore_extent_item_pos is false, otherwise a fimple ulist_free() is
1694  * enough.
1695  *
1696  * Returns 0 on success and < 0 on error. On error @ctx->refs is not allocated.
1697  */
btrfs_find_all_leafs(struct btrfs_backref_walk_ctx * ctx)1698 int btrfs_find_all_leafs(struct btrfs_backref_walk_ctx *ctx)
1699 {
1700 	int ret;
1701 
1702 	ASSERT(ctx->refs == NULL);
1703 
1704 	ctx->refs = ulist_alloc(GFP_NOFS);
1705 	if (!ctx->refs)
1706 		return -ENOMEM;
1707 
1708 	ret = find_parent_nodes(ctx, NULL);
1709 	if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
1710 	    (ret < 0 && ret != -ENOENT)) {
1711 		free_leaf_list(ctx->refs);
1712 		ctx->refs = NULL;
1713 		return ret;
1714 	}
1715 
1716 	return 0;
1717 }
1718 
1719 /*
1720  * Walk all backrefs for a given extent to find all roots that reference this
1721  * extent. Walking a backref means finding all extents that reference this
1722  * extent and in turn walk the backrefs of those, too. Naturally this is a
1723  * recursive process, but here it is implemented in an iterative fashion: We
1724  * find all referencing extents for the extent in question and put them on a
1725  * list. In turn, we find all referencing extents for those, further appending
1726  * to the list. The way we iterate the list allows adding more elements after
1727  * the current while iterating. The process stops when we reach the end of the
1728  * list.
1729  *
1730  * Found roots are added to @ctx->roots, which is allocated by this function if
1731  * it points to NULL, in which case the caller is responsible for freeing it
1732  * after it's not needed anymore.
1733  * This function requires @ctx->refs to be NULL, as it uses it for allocating a
1734  * ulist to do temporary work, and frees it before returning.
1735  *
1736  * Returns 0 on success, < 0 on error.
1737  */
btrfs_find_all_roots_safe(struct btrfs_backref_walk_ctx * ctx)1738 static int btrfs_find_all_roots_safe(struct btrfs_backref_walk_ctx *ctx)
1739 {
1740 	const u64 orig_bytenr = ctx->bytenr;
1741 	const bool orig_skip_inode_ref_list = ctx->skip_inode_ref_list;
1742 	bool roots_ulist_allocated = false;
1743 	struct ulist_iterator uiter;
1744 	int ret = 0;
1745 
1746 	ASSERT(ctx->refs == NULL);
1747 
1748 	ctx->refs = ulist_alloc(GFP_NOFS);
1749 	if (!ctx->refs)
1750 		return -ENOMEM;
1751 
1752 	if (!ctx->roots) {
1753 		ctx->roots = ulist_alloc(GFP_NOFS);
1754 		if (!ctx->roots) {
1755 			ulist_free(ctx->refs);
1756 			ctx->refs = NULL;
1757 			return -ENOMEM;
1758 		}
1759 		roots_ulist_allocated = true;
1760 	}
1761 
1762 	ctx->skip_inode_ref_list = true;
1763 
1764 	ULIST_ITER_INIT(&uiter);
1765 	while (1) {
1766 		struct ulist_node *node;
1767 
1768 		ret = find_parent_nodes(ctx, NULL);
1769 		if (ret < 0 && ret != -ENOENT) {
1770 			if (roots_ulist_allocated) {
1771 				ulist_free(ctx->roots);
1772 				ctx->roots = NULL;
1773 			}
1774 			break;
1775 		}
1776 		ret = 0;
1777 		node = ulist_next(ctx->refs, &uiter);
1778 		if (!node)
1779 			break;
1780 		ctx->bytenr = node->val;
1781 		cond_resched();
1782 	}
1783 
1784 	ulist_free(ctx->refs);
1785 	ctx->refs = NULL;
1786 	ctx->bytenr = orig_bytenr;
1787 	ctx->skip_inode_ref_list = orig_skip_inode_ref_list;
1788 
1789 	return ret;
1790 }
1791 
btrfs_find_all_roots(struct btrfs_backref_walk_ctx * ctx,bool skip_commit_root_sem)1792 int btrfs_find_all_roots(struct btrfs_backref_walk_ctx *ctx,
1793 			 bool skip_commit_root_sem)
1794 {
1795 	int ret;
1796 
1797 	if (!ctx->trans && !skip_commit_root_sem)
1798 		down_read(&ctx->fs_info->commit_root_sem);
1799 	ret = btrfs_find_all_roots_safe(ctx);
1800 	if (!ctx->trans && !skip_commit_root_sem)
1801 		up_read(&ctx->fs_info->commit_root_sem);
1802 	return ret;
1803 }
1804 
btrfs_alloc_backref_share_check_ctx(void)1805 struct btrfs_backref_share_check_ctx *btrfs_alloc_backref_share_check_ctx(void)
1806 {
1807 	struct btrfs_backref_share_check_ctx *ctx;
1808 
1809 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1810 	if (!ctx)
1811 		return NULL;
1812 
1813 	ulist_init(&ctx->refs);
1814 
1815 	return ctx;
1816 }
1817 
btrfs_free_backref_share_ctx(struct btrfs_backref_share_check_ctx * ctx)1818 void btrfs_free_backref_share_ctx(struct btrfs_backref_share_check_ctx *ctx)
1819 {
1820 	if (!ctx)
1821 		return;
1822 
1823 	ulist_release(&ctx->refs);
1824 	kfree(ctx);
1825 }
1826 
1827 /*
1828  * Check if a data extent is shared or not.
1829  *
1830  * @inode:       The inode whose extent we are checking.
1831  * @bytenr:      Logical bytenr of the extent we are checking.
1832  * @extent_gen:  Generation of the extent (file extent item) or 0 if it is
1833  *               not known.
1834  * @ctx:         A backref sharedness check context.
1835  *
1836  * btrfs_is_data_extent_shared uses the backref walking code but will short
1837  * circuit as soon as it finds a root or inode that doesn't match the
1838  * one passed in. This provides a significant performance benefit for
1839  * callers (such as fiemap) which want to know whether the extent is
1840  * shared but do not need a ref count.
1841  *
1842  * This attempts to attach to the running transaction in order to account for
1843  * delayed refs, but continues on even when no running transaction exists.
1844  *
1845  * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1846  */
btrfs_is_data_extent_shared(struct btrfs_inode * inode,u64 bytenr,u64 extent_gen,struct btrfs_backref_share_check_ctx * ctx)1847 int btrfs_is_data_extent_shared(struct btrfs_inode *inode, u64 bytenr,
1848 				u64 extent_gen,
1849 				struct btrfs_backref_share_check_ctx *ctx)
1850 {
1851 	struct btrfs_backref_walk_ctx walk_ctx = { 0 };
1852 	struct btrfs_root *root = inode->root;
1853 	struct btrfs_fs_info *fs_info = root->fs_info;
1854 	struct btrfs_trans_handle *trans;
1855 	struct ulist_iterator uiter;
1856 	struct ulist_node *node;
1857 	struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem);
1858 	int ret = 0;
1859 	struct share_check shared = {
1860 		.ctx = ctx,
1861 		.root = root,
1862 		.inum = btrfs_ino(inode),
1863 		.data_bytenr = bytenr,
1864 		.data_extent_gen = extent_gen,
1865 		.share_count = 0,
1866 		.self_ref_count = 0,
1867 		.have_delayed_delete_refs = false,
1868 	};
1869 	int level;
1870 	bool leaf_cached;
1871 	bool leaf_is_shared;
1872 
1873 	for (int i = 0; i < BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; i++) {
1874 		if (ctx->prev_extents_cache[i].bytenr == bytenr)
1875 			return ctx->prev_extents_cache[i].is_shared;
1876 	}
1877 
1878 	ulist_init(&ctx->refs);
1879 
1880 	trans = btrfs_join_transaction_nostart(root);
1881 	if (IS_ERR(trans)) {
1882 		if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1883 			ret = PTR_ERR(trans);
1884 			goto out;
1885 		}
1886 		trans = NULL;
1887 		down_read(&fs_info->commit_root_sem);
1888 	} else {
1889 		btrfs_get_tree_mod_seq(fs_info, &elem);
1890 		walk_ctx.time_seq = elem.seq;
1891 	}
1892 
1893 	ctx->use_path_cache = true;
1894 
1895 	/*
1896 	 * We may have previously determined that the current leaf is shared.
1897 	 * If it is, then we have a data extent that is shared due to a shared
1898 	 * subtree (caused by snapshotting) and we don't need to check for data
1899 	 * backrefs. If the leaf is not shared, then we must do backref walking
1900 	 * to determine if the data extent is shared through reflinks.
1901 	 */
1902 	leaf_cached = lookup_backref_shared_cache(ctx, root,
1903 						  ctx->curr_leaf_bytenr, 0,
1904 						  &leaf_is_shared);
1905 	if (leaf_cached && leaf_is_shared) {
1906 		ret = 1;
1907 		goto out_trans;
1908 	}
1909 
1910 	walk_ctx.skip_inode_ref_list = true;
1911 	walk_ctx.trans = trans;
1912 	walk_ctx.fs_info = fs_info;
1913 	walk_ctx.refs = &ctx->refs;
1914 
1915 	/* -1 means we are in the bytenr of the data extent. */
1916 	level = -1;
1917 	ULIST_ITER_INIT(&uiter);
1918 	while (1) {
1919 		const unsigned long prev_ref_count = ctx->refs.nnodes;
1920 
1921 		walk_ctx.bytenr = bytenr;
1922 		ret = find_parent_nodes(&walk_ctx, &shared);
1923 		if (ret == BACKREF_FOUND_SHARED ||
1924 		    ret == BACKREF_FOUND_NOT_SHARED) {
1925 			/* If shared must return 1, otherwise return 0. */
1926 			ret = (ret == BACKREF_FOUND_SHARED) ? 1 : 0;
1927 			if (level >= 0)
1928 				store_backref_shared_cache(ctx, root, bytenr,
1929 							   level, ret == 1);
1930 			break;
1931 		}
1932 		if (ret < 0 && ret != -ENOENT)
1933 			break;
1934 		ret = 0;
1935 
1936 		/*
1937 		 * More than one extent buffer (bytenr) may have been added to
1938 		 * the ctx->refs ulist, in which case we have to check multiple
1939 		 * tree paths in case the first one is not shared, so we can not
1940 		 * use the path cache which is made for a single path. Multiple
1941 		 * extent buffers at the current level happen when:
1942 		 *
1943 		 * 1) level -1, the data extent: If our data extent was not
1944 		 *    directly shared (without multiple reference items), then
1945 		 *    it might have a single reference item with a count > 1 for
1946 		 *    the same offset, which means there are 2 (or more) file
1947 		 *    extent items that point to the data extent - this happens
1948 		 *    when a file extent item needs to be split and then one
1949 		 *    item gets moved to another leaf due to a b+tree leaf split
1950 		 *    when inserting some item. In this case the file extent
1951 		 *    items may be located in different leaves and therefore
1952 		 *    some of the leaves may be referenced through shared
1953 		 *    subtrees while others are not. Since our extent buffer
1954 		 *    cache only works for a single path (by far the most common
1955 		 *    case and simpler to deal with), we can not use it if we
1956 		 *    have multiple leaves (which implies multiple paths).
1957 		 *
1958 		 * 2) level >= 0, a tree node/leaf: We can have a mix of direct
1959 		 *    and indirect references on a b+tree node/leaf, so we have
1960 		 *    to check multiple paths, and the extent buffer (the
1961 		 *    current bytenr) may be shared or not. One example is
1962 		 *    during relocation as we may get a shared tree block ref
1963 		 *    (direct ref) and a non-shared tree block ref (indirect
1964 		 *    ref) for the same node/leaf.
1965 		 */
1966 		if ((ctx->refs.nnodes - prev_ref_count) > 1)
1967 			ctx->use_path_cache = false;
1968 
1969 		if (level >= 0)
1970 			store_backref_shared_cache(ctx, root, bytenr,
1971 						   level, false);
1972 		node = ulist_next(&ctx->refs, &uiter);
1973 		if (!node)
1974 			break;
1975 		bytenr = node->val;
1976 		if (ctx->use_path_cache) {
1977 			bool is_shared;
1978 			bool cached;
1979 
1980 			level++;
1981 			cached = lookup_backref_shared_cache(ctx, root, bytenr,
1982 							     level, &is_shared);
1983 			if (cached) {
1984 				ret = (is_shared ? 1 : 0);
1985 				break;
1986 			}
1987 		}
1988 		shared.share_count = 0;
1989 		shared.have_delayed_delete_refs = false;
1990 		cond_resched();
1991 	}
1992 
1993 	/*
1994 	 * If the path cache is disabled, then it means at some tree level we
1995 	 * got multiple parents due to a mix of direct and indirect backrefs or
1996 	 * multiple leaves with file extent items pointing to the same data
1997 	 * extent. We have to invalidate the cache and cache only the sharedness
1998 	 * result for the levels where we got only one node/reference.
1999 	 */
2000 	if (!ctx->use_path_cache) {
2001 		int i = 0;
2002 
2003 		level--;
2004 		if (ret >= 0 && level >= 0) {
2005 			bytenr = ctx->path_cache_entries[level].bytenr;
2006 			ctx->use_path_cache = true;
2007 			store_backref_shared_cache(ctx, root, bytenr, level, ret);
2008 			i = level + 1;
2009 		}
2010 
2011 		for ( ; i < BTRFS_MAX_LEVEL; i++)
2012 			ctx->path_cache_entries[i].bytenr = 0;
2013 	}
2014 
2015 	/*
2016 	 * Cache the sharedness result for the data extent if we know our inode
2017 	 * has more than 1 file extent item that refers to the data extent.
2018 	 */
2019 	if (ret >= 0 && shared.self_ref_count > 1) {
2020 		int slot = ctx->prev_extents_cache_slot;
2021 
2022 		ctx->prev_extents_cache[slot].bytenr = shared.data_bytenr;
2023 		ctx->prev_extents_cache[slot].is_shared = (ret == 1);
2024 
2025 		slot = (slot + 1) % BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE;
2026 		ctx->prev_extents_cache_slot = slot;
2027 	}
2028 
2029 out_trans:
2030 	if (trans) {
2031 		btrfs_put_tree_mod_seq(fs_info, &elem);
2032 		btrfs_end_transaction(trans);
2033 	} else {
2034 		up_read(&fs_info->commit_root_sem);
2035 	}
2036 out:
2037 	ulist_release(&ctx->refs);
2038 	ctx->prev_leaf_bytenr = ctx->curr_leaf_bytenr;
2039 
2040 	return ret;
2041 }
2042 
btrfs_find_one_extref(struct btrfs_root * root,u64 inode_objectid,u64 start_off,struct btrfs_path * path,struct btrfs_inode_extref ** ret_extref,u64 * found_off)2043 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
2044 			  u64 start_off, struct btrfs_path *path,
2045 			  struct btrfs_inode_extref **ret_extref,
2046 			  u64 *found_off)
2047 {
2048 	int ret, slot;
2049 	struct btrfs_key key;
2050 	struct btrfs_key found_key;
2051 	struct btrfs_inode_extref *extref;
2052 	const struct extent_buffer *leaf;
2053 	unsigned long ptr;
2054 
2055 	key.objectid = inode_objectid;
2056 	key.type = BTRFS_INODE_EXTREF_KEY;
2057 	key.offset = start_off;
2058 
2059 	ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2060 	if (ret < 0)
2061 		return ret;
2062 
2063 	while (1) {
2064 		leaf = path->nodes[0];
2065 		slot = path->slots[0];
2066 		if (slot >= btrfs_header_nritems(leaf)) {
2067 			/*
2068 			 * If the item at offset is not found,
2069 			 * btrfs_search_slot will point us to the slot
2070 			 * where it should be inserted. In our case
2071 			 * that will be the slot directly before the
2072 			 * next INODE_REF_KEY_V2 item. In the case
2073 			 * that we're pointing to the last slot in a
2074 			 * leaf, we must move one leaf over.
2075 			 */
2076 			ret = btrfs_next_leaf(root, path);
2077 			if (ret) {
2078 				if (ret >= 1)
2079 					ret = -ENOENT;
2080 				break;
2081 			}
2082 			continue;
2083 		}
2084 
2085 		btrfs_item_key_to_cpu(leaf, &found_key, slot);
2086 
2087 		/*
2088 		 * Check that we're still looking at an extended ref key for
2089 		 * this particular objectid. If we have different
2090 		 * objectid or type then there are no more to be found
2091 		 * in the tree and we can exit.
2092 		 */
2093 		ret = -ENOENT;
2094 		if (found_key.objectid != inode_objectid)
2095 			break;
2096 		if (found_key.type != BTRFS_INODE_EXTREF_KEY)
2097 			break;
2098 
2099 		ret = 0;
2100 		ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2101 		extref = (struct btrfs_inode_extref *)ptr;
2102 		*ret_extref = extref;
2103 		if (found_off)
2104 			*found_off = found_key.offset;
2105 		break;
2106 	}
2107 
2108 	return ret;
2109 }
2110 
2111 /*
2112  * this iterates to turn a name (from iref/extref) into a full filesystem path.
2113  * Elements of the path are separated by '/' and the path is guaranteed to be
2114  * 0-terminated. the path is only given within the current file system.
2115  * Therefore, it never starts with a '/'. the caller is responsible to provide
2116  * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
2117  * the start point of the resulting string is returned. this pointer is within
2118  * dest, normally.
2119  * in case the path buffer would overflow, the pointer is decremented further
2120  * as if output was written to the buffer, though no more output is actually
2121  * generated. that way, the caller can determine how much space would be
2122  * required for the path to fit into the buffer. in that case, the returned
2123  * value will be smaller than dest. callers must check this!
2124  */
btrfs_ref_to_path(struct btrfs_root * fs_root,struct btrfs_path * path,u32 name_len,unsigned long name_off,struct extent_buffer * eb_in,u64 parent,char * dest,u32 size)2125 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
2126 			u32 name_len, unsigned long name_off,
2127 			struct extent_buffer *eb_in, u64 parent,
2128 			char *dest, u32 size)
2129 {
2130 	int slot;
2131 	u64 next_inum;
2132 	int ret;
2133 	s64 bytes_left = ((s64)size) - 1;
2134 	struct extent_buffer *eb = eb_in;
2135 	struct btrfs_key found_key;
2136 	struct btrfs_inode_ref *iref;
2137 
2138 	if (bytes_left >= 0)
2139 		dest[bytes_left] = '\0';
2140 
2141 	while (1) {
2142 		bytes_left -= name_len;
2143 		if (bytes_left >= 0)
2144 			read_extent_buffer(eb, dest + bytes_left,
2145 					   name_off, name_len);
2146 		if (eb != eb_in) {
2147 			if (!path->skip_locking)
2148 				btrfs_tree_read_unlock(eb);
2149 			free_extent_buffer(eb);
2150 		}
2151 		ret = btrfs_find_item(fs_root, path, parent, 0,
2152 				BTRFS_INODE_REF_KEY, &found_key);
2153 		if (ret > 0)
2154 			ret = -ENOENT;
2155 		if (ret)
2156 			break;
2157 
2158 		next_inum = found_key.offset;
2159 
2160 		/* regular exit ahead */
2161 		if (parent == next_inum)
2162 			break;
2163 
2164 		slot = path->slots[0];
2165 		eb = path->nodes[0];
2166 		/* make sure we can use eb after releasing the path */
2167 		if (eb != eb_in) {
2168 			path->nodes[0] = NULL;
2169 			path->locks[0] = 0;
2170 		}
2171 		btrfs_release_path(path);
2172 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2173 
2174 		name_len = btrfs_inode_ref_name_len(eb, iref);
2175 		name_off = (unsigned long)(iref + 1);
2176 
2177 		parent = next_inum;
2178 		--bytes_left;
2179 		if (bytes_left >= 0)
2180 			dest[bytes_left] = '/';
2181 	}
2182 
2183 	btrfs_release_path(path);
2184 
2185 	if (ret)
2186 		return ERR_PTR(ret);
2187 
2188 	return dest + bytes_left;
2189 }
2190 
2191 /*
2192  * this makes the path point to (logical EXTENT_ITEM *)
2193  * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
2194  * tree blocks and <0 on error.
2195  */
extent_from_logical(struct btrfs_fs_info * fs_info,u64 logical,struct btrfs_path * path,struct btrfs_key * found_key,u64 * flags_ret)2196 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
2197 			struct btrfs_path *path, struct btrfs_key *found_key,
2198 			u64 *flags_ret)
2199 {
2200 	struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical);
2201 	int ret;
2202 	u64 flags;
2203 	u64 size = 0;
2204 	u32 item_size;
2205 	const struct extent_buffer *eb;
2206 	struct btrfs_extent_item *ei;
2207 	struct btrfs_key key;
2208 
2209 	if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2210 		key.type = BTRFS_METADATA_ITEM_KEY;
2211 	else
2212 		key.type = BTRFS_EXTENT_ITEM_KEY;
2213 	key.objectid = logical;
2214 	key.offset = (u64)-1;
2215 
2216 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2217 	if (ret < 0)
2218 		return ret;
2219 	if (ret == 0) {
2220 		/*
2221 		 * Key with offset -1 found, there would have to exist an extent
2222 		 * item with such offset, but this is out of the valid range.
2223 		 */
2224 		return -EUCLEAN;
2225 	}
2226 
2227 	ret = btrfs_previous_extent_item(extent_root, path, 0);
2228 	if (ret) {
2229 		if (ret > 0)
2230 			ret = -ENOENT;
2231 		return ret;
2232 	}
2233 	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
2234 	if (found_key->type == BTRFS_METADATA_ITEM_KEY)
2235 		size = fs_info->nodesize;
2236 	else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
2237 		size = found_key->offset;
2238 
2239 	if (found_key->objectid > logical ||
2240 	    found_key->objectid + size <= logical) {
2241 		btrfs_debug(fs_info,
2242 			"logical %llu is not within any extent", logical);
2243 		return -ENOENT;
2244 	}
2245 
2246 	eb = path->nodes[0];
2247 	item_size = btrfs_item_size(eb, path->slots[0]);
2248 
2249 	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
2250 	flags = btrfs_extent_flags(eb, ei);
2251 
2252 	btrfs_debug(fs_info,
2253 		"logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
2254 		 logical, logical - found_key->objectid, found_key->objectid,
2255 		 found_key->offset, flags, item_size);
2256 
2257 	WARN_ON(!flags_ret);
2258 	if (flags_ret) {
2259 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2260 			*flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
2261 		else if (flags & BTRFS_EXTENT_FLAG_DATA)
2262 			*flags_ret = BTRFS_EXTENT_FLAG_DATA;
2263 		else
2264 			BUG();
2265 		return 0;
2266 	}
2267 
2268 	return -EIO;
2269 }
2270 
2271 /*
2272  * helper function to iterate extent inline refs. ptr must point to a 0 value
2273  * for the first call and may be modified. it is used to track state.
2274  * if more refs exist, 0 is returned and the next call to
2275  * get_extent_inline_ref must pass the modified ptr parameter to get the
2276  * next ref. after the last ref was processed, 1 is returned.
2277  * returns <0 on error
2278  */
get_extent_inline_ref(unsigned long * ptr,const struct extent_buffer * eb,const struct btrfs_key * key,const struct btrfs_extent_item * ei,u32 item_size,struct btrfs_extent_inline_ref ** out_eiref,int * out_type)2279 static int get_extent_inline_ref(unsigned long *ptr,
2280 				 const struct extent_buffer *eb,
2281 				 const struct btrfs_key *key,
2282 				 const struct btrfs_extent_item *ei,
2283 				 u32 item_size,
2284 				 struct btrfs_extent_inline_ref **out_eiref,
2285 				 int *out_type)
2286 {
2287 	unsigned long end;
2288 	u64 flags;
2289 	struct btrfs_tree_block_info *info;
2290 
2291 	if (!*ptr) {
2292 		/* first call */
2293 		flags = btrfs_extent_flags(eb, ei);
2294 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2295 			if (key->type == BTRFS_METADATA_ITEM_KEY) {
2296 				/* a skinny metadata extent */
2297 				*out_eiref =
2298 				     (struct btrfs_extent_inline_ref *)(ei + 1);
2299 			} else {
2300 				WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
2301 				info = (struct btrfs_tree_block_info *)(ei + 1);
2302 				*out_eiref =
2303 				   (struct btrfs_extent_inline_ref *)(info + 1);
2304 			}
2305 		} else {
2306 			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
2307 		}
2308 		*ptr = (unsigned long)*out_eiref;
2309 		if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
2310 			return -ENOENT;
2311 	}
2312 
2313 	end = (unsigned long)ei + item_size;
2314 	*out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
2315 	*out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
2316 						     BTRFS_REF_TYPE_ANY);
2317 	if (*out_type == BTRFS_REF_TYPE_INVALID)
2318 		return -EUCLEAN;
2319 
2320 	*ptr += btrfs_extent_inline_ref_size(*out_type);
2321 	WARN_ON(*ptr > end);
2322 	if (*ptr == end)
2323 		return 1; /* last */
2324 
2325 	return 0;
2326 }
2327 
2328 /*
2329  * reads the tree block backref for an extent. tree level and root are returned
2330  * through out_level and out_root. ptr must point to a 0 value for the first
2331  * call and may be modified (see get_extent_inline_ref comment).
2332  * returns 0 if data was provided, 1 if there was no more data to provide or
2333  * <0 on error.
2334  */
tree_backref_for_extent(unsigned long * ptr,struct extent_buffer * eb,struct btrfs_key * key,struct btrfs_extent_item * ei,u32 item_size,u64 * out_root,u8 * out_level)2335 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
2336 			    struct btrfs_key *key, struct btrfs_extent_item *ei,
2337 			    u32 item_size, u64 *out_root, u8 *out_level)
2338 {
2339 	int ret;
2340 	int type;
2341 	struct btrfs_extent_inline_ref *eiref;
2342 
2343 	if (*ptr == (unsigned long)-1)
2344 		return 1;
2345 
2346 	while (1) {
2347 		ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
2348 					      &eiref, &type);
2349 		if (ret < 0)
2350 			return ret;
2351 
2352 		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
2353 		    type == BTRFS_SHARED_BLOCK_REF_KEY)
2354 			break;
2355 
2356 		if (ret == 1)
2357 			return 1;
2358 	}
2359 
2360 	/* we can treat both ref types equally here */
2361 	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
2362 
2363 	if (key->type == BTRFS_EXTENT_ITEM_KEY) {
2364 		struct btrfs_tree_block_info *info;
2365 
2366 		info = (struct btrfs_tree_block_info *)(ei + 1);
2367 		*out_level = btrfs_tree_block_level(eb, info);
2368 	} else {
2369 		ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
2370 		*out_level = (u8)key->offset;
2371 	}
2372 
2373 	if (ret == 1)
2374 		*ptr = (unsigned long)-1;
2375 
2376 	return 0;
2377 }
2378 
iterate_leaf_refs(struct btrfs_fs_info * fs_info,struct extent_inode_elem * inode_list,u64 root,u64 extent_item_objectid,iterate_extent_inodes_t * iterate,void * ctx)2379 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
2380 			     struct extent_inode_elem *inode_list,
2381 			     u64 root, u64 extent_item_objectid,
2382 			     iterate_extent_inodes_t *iterate, void *ctx)
2383 {
2384 	struct extent_inode_elem *eie;
2385 	int ret = 0;
2386 
2387 	for (eie = inode_list; eie; eie = eie->next) {
2388 		btrfs_debug(fs_info,
2389 			    "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
2390 			    extent_item_objectid, eie->inum,
2391 			    eie->offset, root);
2392 		ret = iterate(eie->inum, eie->offset, eie->num_bytes, root, ctx);
2393 		if (ret) {
2394 			btrfs_debug(fs_info,
2395 				    "stopping iteration for %llu due to ret=%d",
2396 				    extent_item_objectid, ret);
2397 			break;
2398 		}
2399 	}
2400 
2401 	return ret;
2402 }
2403 
2404 /*
2405  * calls iterate() for every inode that references the extent identified by
2406  * the given parameters.
2407  * when the iterator function returns a non-zero value, iteration stops.
2408  */
iterate_extent_inodes(struct btrfs_backref_walk_ctx * ctx,bool search_commit_root,iterate_extent_inodes_t * iterate,void * user_ctx)2409 int iterate_extent_inodes(struct btrfs_backref_walk_ctx *ctx,
2410 			  bool search_commit_root,
2411 			  iterate_extent_inodes_t *iterate, void *user_ctx)
2412 {
2413 	int ret;
2414 	struct ulist *refs;
2415 	struct ulist_node *ref_node;
2416 	struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem);
2417 	struct ulist_iterator ref_uiter;
2418 
2419 	btrfs_debug(ctx->fs_info, "resolving all inodes for extent %llu",
2420 		    ctx->bytenr);
2421 
2422 	ASSERT(ctx->trans == NULL);
2423 	ASSERT(ctx->roots == NULL);
2424 
2425 	if (!search_commit_root) {
2426 		struct btrfs_trans_handle *trans;
2427 
2428 		trans = btrfs_attach_transaction(ctx->fs_info->tree_root);
2429 		if (IS_ERR(trans)) {
2430 			if (PTR_ERR(trans) != -ENOENT &&
2431 			    PTR_ERR(trans) != -EROFS)
2432 				return PTR_ERR(trans);
2433 			trans = NULL;
2434 		}
2435 		ctx->trans = trans;
2436 	}
2437 
2438 	if (ctx->trans) {
2439 		btrfs_get_tree_mod_seq(ctx->fs_info, &seq_elem);
2440 		ctx->time_seq = seq_elem.seq;
2441 	} else {
2442 		down_read(&ctx->fs_info->commit_root_sem);
2443 	}
2444 
2445 	ret = btrfs_find_all_leafs(ctx);
2446 	if (ret)
2447 		goto out;
2448 	refs = ctx->refs;
2449 	ctx->refs = NULL;
2450 
2451 	ULIST_ITER_INIT(&ref_uiter);
2452 	while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
2453 		const u64 leaf_bytenr = ref_node->val;
2454 		struct ulist_node *root_node;
2455 		struct ulist_iterator root_uiter;
2456 		struct extent_inode_elem *inode_list;
2457 
2458 		inode_list = (struct extent_inode_elem *)(uintptr_t)ref_node->aux;
2459 
2460 		if (ctx->cache_lookup) {
2461 			const u64 *root_ids;
2462 			int root_count;
2463 			bool cached;
2464 
2465 			cached = ctx->cache_lookup(leaf_bytenr, ctx->user_ctx,
2466 						   &root_ids, &root_count);
2467 			if (cached) {
2468 				for (int i = 0; i < root_count; i++) {
2469 					ret = iterate_leaf_refs(ctx->fs_info,
2470 								inode_list,
2471 								root_ids[i],
2472 								leaf_bytenr,
2473 								iterate,
2474 								user_ctx);
2475 					if (ret)
2476 						break;
2477 				}
2478 				continue;
2479 			}
2480 		}
2481 
2482 		if (!ctx->roots) {
2483 			ctx->roots = ulist_alloc(GFP_NOFS);
2484 			if (!ctx->roots) {
2485 				ret = -ENOMEM;
2486 				break;
2487 			}
2488 		}
2489 
2490 		ctx->bytenr = leaf_bytenr;
2491 		ret = btrfs_find_all_roots_safe(ctx);
2492 		if (ret)
2493 			break;
2494 
2495 		if (ctx->cache_store)
2496 			ctx->cache_store(leaf_bytenr, ctx->roots, ctx->user_ctx);
2497 
2498 		ULIST_ITER_INIT(&root_uiter);
2499 		while (!ret && (root_node = ulist_next(ctx->roots, &root_uiter))) {
2500 			btrfs_debug(ctx->fs_info,
2501 				    "root %llu references leaf %llu, data list %#llx",
2502 				    root_node->val, ref_node->val,
2503 				    ref_node->aux);
2504 			ret = iterate_leaf_refs(ctx->fs_info, inode_list,
2505 						root_node->val, ctx->bytenr,
2506 						iterate, user_ctx);
2507 		}
2508 		ulist_reinit(ctx->roots);
2509 	}
2510 
2511 	free_leaf_list(refs);
2512 out:
2513 	if (ctx->trans) {
2514 		btrfs_put_tree_mod_seq(ctx->fs_info, &seq_elem);
2515 		btrfs_end_transaction(ctx->trans);
2516 		ctx->trans = NULL;
2517 	} else {
2518 		up_read(&ctx->fs_info->commit_root_sem);
2519 	}
2520 
2521 	ulist_free(ctx->roots);
2522 	ctx->roots = NULL;
2523 
2524 	if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP)
2525 		ret = 0;
2526 
2527 	return ret;
2528 }
2529 
build_ino_list(u64 inum,u64 offset,u64 num_bytes,u64 root,void * ctx)2530 static int build_ino_list(u64 inum, u64 offset, u64 num_bytes, u64 root, void *ctx)
2531 {
2532 	struct btrfs_data_container *inodes = ctx;
2533 	const size_t c = 3 * sizeof(u64);
2534 
2535 	if (inodes->bytes_left >= c) {
2536 		inodes->bytes_left -= c;
2537 		inodes->val[inodes->elem_cnt] = inum;
2538 		inodes->val[inodes->elem_cnt + 1] = offset;
2539 		inodes->val[inodes->elem_cnt + 2] = root;
2540 		inodes->elem_cnt += 3;
2541 	} else {
2542 		inodes->bytes_missing += c - inodes->bytes_left;
2543 		inodes->bytes_left = 0;
2544 		inodes->elem_missed += 3;
2545 	}
2546 
2547 	return 0;
2548 }
2549 
iterate_inodes_from_logical(u64 logical,struct btrfs_fs_info * fs_info,struct btrfs_path * path,void * ctx,bool ignore_offset)2550 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2551 				struct btrfs_path *path,
2552 				void *ctx, bool ignore_offset)
2553 {
2554 	struct btrfs_backref_walk_ctx walk_ctx = { 0 };
2555 	int ret;
2556 	u64 flags = 0;
2557 	struct btrfs_key found_key;
2558 	int search_commit_root = path->search_commit_root;
2559 
2560 	ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2561 	btrfs_release_path(path);
2562 	if (ret < 0)
2563 		return ret;
2564 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2565 		return -EINVAL;
2566 
2567 	walk_ctx.bytenr = found_key.objectid;
2568 	if (ignore_offset)
2569 		walk_ctx.ignore_extent_item_pos = true;
2570 	else
2571 		walk_ctx.extent_item_pos = logical - found_key.objectid;
2572 	walk_ctx.fs_info = fs_info;
2573 
2574 	return iterate_extent_inodes(&walk_ctx, search_commit_root,
2575 				     build_ino_list, ctx);
2576 }
2577 
2578 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2579 			 struct extent_buffer *eb, struct inode_fs_paths *ipath);
2580 
iterate_inode_refs(u64 inum,struct inode_fs_paths * ipath)2581 static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath)
2582 {
2583 	int ret = 0;
2584 	int slot;
2585 	u32 cur;
2586 	u32 len;
2587 	u32 name_len;
2588 	u64 parent = 0;
2589 	int found = 0;
2590 	struct btrfs_root *fs_root = ipath->fs_root;
2591 	struct btrfs_path *path = ipath->btrfs_path;
2592 	struct extent_buffer *eb;
2593 	struct btrfs_inode_ref *iref;
2594 	struct btrfs_key found_key;
2595 
2596 	while (!ret) {
2597 		ret = btrfs_find_item(fs_root, path, inum,
2598 				parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2599 				&found_key);
2600 
2601 		if (ret < 0)
2602 			break;
2603 		if (ret) {
2604 			ret = found ? 0 : -ENOENT;
2605 			break;
2606 		}
2607 		++found;
2608 
2609 		parent = found_key.offset;
2610 		slot = path->slots[0];
2611 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
2612 		if (!eb) {
2613 			ret = -ENOMEM;
2614 			break;
2615 		}
2616 		btrfs_release_path(path);
2617 
2618 		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2619 
2620 		for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) {
2621 			name_len = btrfs_inode_ref_name_len(eb, iref);
2622 			/* path must be released before calling iterate()! */
2623 			btrfs_debug(fs_root->fs_info,
2624 				"following ref at offset %u for inode %llu in tree %llu",
2625 				cur, found_key.objectid,
2626 				btrfs_root_id(fs_root));
2627 			ret = inode_to_path(parent, name_len,
2628 				      (unsigned long)(iref + 1), eb, ipath);
2629 			if (ret)
2630 				break;
2631 			len = sizeof(*iref) + name_len;
2632 			iref = (struct btrfs_inode_ref *)((char *)iref + len);
2633 		}
2634 		free_extent_buffer(eb);
2635 	}
2636 
2637 	btrfs_release_path(path);
2638 
2639 	return ret;
2640 }
2641 
iterate_inode_extrefs(u64 inum,struct inode_fs_paths * ipath)2642 static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath)
2643 {
2644 	int ret;
2645 	int slot;
2646 	u64 offset = 0;
2647 	u64 parent;
2648 	int found = 0;
2649 	struct btrfs_root *fs_root = ipath->fs_root;
2650 	struct btrfs_path *path = ipath->btrfs_path;
2651 	struct extent_buffer *eb;
2652 	struct btrfs_inode_extref *extref;
2653 	u32 item_size;
2654 	u32 cur_offset;
2655 	unsigned long ptr;
2656 
2657 	while (1) {
2658 		ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2659 					    &offset);
2660 		if (ret < 0)
2661 			break;
2662 		if (ret) {
2663 			ret = found ? 0 : -ENOENT;
2664 			break;
2665 		}
2666 		++found;
2667 
2668 		slot = path->slots[0];
2669 		eb = btrfs_clone_extent_buffer(path->nodes[0]);
2670 		if (!eb) {
2671 			ret = -ENOMEM;
2672 			break;
2673 		}
2674 		btrfs_release_path(path);
2675 
2676 		item_size = btrfs_item_size(eb, slot);
2677 		ptr = btrfs_item_ptr_offset(eb, slot);
2678 		cur_offset = 0;
2679 
2680 		while (cur_offset < item_size) {
2681 			u32 name_len;
2682 
2683 			extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2684 			parent = btrfs_inode_extref_parent(eb, extref);
2685 			name_len = btrfs_inode_extref_name_len(eb, extref);
2686 			ret = inode_to_path(parent, name_len,
2687 				      (unsigned long)&extref->name, eb, ipath);
2688 			if (ret)
2689 				break;
2690 
2691 			cur_offset += btrfs_inode_extref_name_len(eb, extref);
2692 			cur_offset += sizeof(*extref);
2693 		}
2694 		free_extent_buffer(eb);
2695 
2696 		offset++;
2697 	}
2698 
2699 	btrfs_release_path(path);
2700 
2701 	return ret;
2702 }
2703 
2704 /*
2705  * returns 0 if the path could be dumped (probably truncated)
2706  * returns <0 in case of an error
2707  */
inode_to_path(u64 inum,u32 name_len,unsigned long name_off,struct extent_buffer * eb,struct inode_fs_paths * ipath)2708 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2709 			 struct extent_buffer *eb, struct inode_fs_paths *ipath)
2710 {
2711 	char *fspath;
2712 	char *fspath_min;
2713 	int i = ipath->fspath->elem_cnt;
2714 	const int s_ptr = sizeof(char *);
2715 	u32 bytes_left;
2716 
2717 	bytes_left = ipath->fspath->bytes_left > s_ptr ?
2718 					ipath->fspath->bytes_left - s_ptr : 0;
2719 
2720 	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2721 	fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2722 				   name_off, eb, inum, fspath_min, bytes_left);
2723 	if (IS_ERR(fspath))
2724 		return PTR_ERR(fspath);
2725 
2726 	if (fspath > fspath_min) {
2727 		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2728 		++ipath->fspath->elem_cnt;
2729 		ipath->fspath->bytes_left = fspath - fspath_min;
2730 	} else {
2731 		++ipath->fspath->elem_missed;
2732 		ipath->fspath->bytes_missing += fspath_min - fspath;
2733 		ipath->fspath->bytes_left = 0;
2734 	}
2735 
2736 	return 0;
2737 }
2738 
2739 /*
2740  * this dumps all file system paths to the inode into the ipath struct, provided
2741  * is has been created large enough. each path is zero-terminated and accessed
2742  * from ipath->fspath->val[i].
2743  * when it returns, there are ipath->fspath->elem_cnt number of paths available
2744  * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2745  * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2746  * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2747  * have been needed to return all paths.
2748  */
paths_from_inode(u64 inum,struct inode_fs_paths * ipath)2749 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2750 {
2751 	int ret;
2752 	int found_refs = 0;
2753 
2754 	ret = iterate_inode_refs(inum, ipath);
2755 	if (!ret)
2756 		++found_refs;
2757 	else if (ret != -ENOENT)
2758 		return ret;
2759 
2760 	ret = iterate_inode_extrefs(inum, ipath);
2761 	if (ret == -ENOENT && found_refs)
2762 		return 0;
2763 
2764 	return ret;
2765 }
2766 
init_data_container(u32 total_bytes)2767 struct btrfs_data_container *init_data_container(u32 total_bytes)
2768 {
2769 	struct btrfs_data_container *data;
2770 	size_t alloc_bytes;
2771 
2772 	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2773 	data = kvzalloc(alloc_bytes, GFP_KERNEL);
2774 	if (!data)
2775 		return ERR_PTR(-ENOMEM);
2776 
2777 	if (total_bytes >= sizeof(*data))
2778 		data->bytes_left = total_bytes - sizeof(*data);
2779 	else
2780 		data->bytes_missing = sizeof(*data) - total_bytes;
2781 
2782 	return data;
2783 }
2784 
2785 /*
2786  * allocates space to return multiple file system paths for an inode.
2787  * total_bytes to allocate are passed, note that space usable for actual path
2788  * information will be total_bytes - sizeof(struct inode_fs_paths).
2789  * the returned pointer must be freed with free_ipath() in the end.
2790  */
init_ipath(s32 total_bytes,struct btrfs_root * fs_root,struct btrfs_path * path)2791 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2792 					struct btrfs_path *path)
2793 {
2794 	struct inode_fs_paths *ifp;
2795 	struct btrfs_data_container *fspath;
2796 
2797 	fspath = init_data_container(total_bytes);
2798 	if (IS_ERR(fspath))
2799 		return ERR_CAST(fspath);
2800 
2801 	ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2802 	if (!ifp) {
2803 		kvfree(fspath);
2804 		return ERR_PTR(-ENOMEM);
2805 	}
2806 
2807 	ifp->btrfs_path = path;
2808 	ifp->fspath = fspath;
2809 	ifp->fs_root = fs_root;
2810 
2811 	return ifp;
2812 }
2813 
free_ipath(struct inode_fs_paths * ipath)2814 void free_ipath(struct inode_fs_paths *ipath)
2815 {
2816 	if (!ipath)
2817 		return;
2818 	kvfree(ipath->fspath);
2819 	kfree(ipath);
2820 }
2821 
btrfs_backref_iter_alloc(struct btrfs_fs_info * fs_info)2822 struct btrfs_backref_iter *btrfs_backref_iter_alloc(struct btrfs_fs_info *fs_info)
2823 {
2824 	struct btrfs_backref_iter *ret;
2825 
2826 	ret = kzalloc(sizeof(*ret), GFP_NOFS);
2827 	if (!ret)
2828 		return NULL;
2829 
2830 	ret->path = btrfs_alloc_path();
2831 	if (!ret->path) {
2832 		kfree(ret);
2833 		return NULL;
2834 	}
2835 
2836 	/* Current backref iterator only supports iteration in commit root */
2837 	ret->path->search_commit_root = 1;
2838 	ret->path->skip_locking = 1;
2839 	ret->fs_info = fs_info;
2840 
2841 	return ret;
2842 }
2843 
btrfs_backref_iter_release(struct btrfs_backref_iter * iter)2844 static void btrfs_backref_iter_release(struct btrfs_backref_iter *iter)
2845 {
2846 	iter->bytenr = 0;
2847 	iter->item_ptr = 0;
2848 	iter->cur_ptr = 0;
2849 	iter->end_ptr = 0;
2850 	btrfs_release_path(iter->path);
2851 	memset(&iter->cur_key, 0, sizeof(iter->cur_key));
2852 }
2853 
btrfs_backref_iter_start(struct btrfs_backref_iter * iter,u64 bytenr)2854 int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2855 {
2856 	struct btrfs_fs_info *fs_info = iter->fs_info;
2857 	struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2858 	struct btrfs_path *path = iter->path;
2859 	struct btrfs_extent_item *ei;
2860 	struct btrfs_key key;
2861 	int ret;
2862 
2863 	key.objectid = bytenr;
2864 	key.type = BTRFS_METADATA_ITEM_KEY;
2865 	key.offset = (u64)-1;
2866 	iter->bytenr = bytenr;
2867 
2868 	ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2869 	if (ret < 0)
2870 		return ret;
2871 	if (ret == 0) {
2872 		/*
2873 		 * Key with offset -1 found, there would have to exist an extent
2874 		 * item with such offset, but this is out of the valid range.
2875 		 */
2876 		ret = -EUCLEAN;
2877 		goto release;
2878 	}
2879 	if (path->slots[0] == 0) {
2880 		WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2881 		ret = -EUCLEAN;
2882 		goto release;
2883 	}
2884 	path->slots[0]--;
2885 
2886 	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2887 	if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2888 	     key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2889 		ret = -ENOENT;
2890 		goto release;
2891 	}
2892 	memcpy(&iter->cur_key, &key, sizeof(key));
2893 	iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2894 						    path->slots[0]);
2895 	iter->end_ptr = (u32)(iter->item_ptr +
2896 			btrfs_item_size(path->nodes[0], path->slots[0]));
2897 	ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2898 			    struct btrfs_extent_item);
2899 
2900 	/*
2901 	 * Only support iteration on tree backref yet.
2902 	 *
2903 	 * This is an extra precaution for non skinny-metadata, where
2904 	 * EXTENT_ITEM is also used for tree blocks, that we can only use
2905 	 * extent flags to determine if it's a tree block.
2906 	 */
2907 	if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2908 		ret = -ENOTSUPP;
2909 		goto release;
2910 	}
2911 	iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2912 
2913 	/* If there is no inline backref, go search for keyed backref */
2914 	if (iter->cur_ptr >= iter->end_ptr) {
2915 		ret = btrfs_next_item(extent_root, path);
2916 
2917 		/* No inline nor keyed ref */
2918 		if (ret > 0) {
2919 			ret = -ENOENT;
2920 			goto release;
2921 		}
2922 		if (ret < 0)
2923 			goto release;
2924 
2925 		btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2926 				path->slots[0]);
2927 		if (iter->cur_key.objectid != bytenr ||
2928 		    (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2929 		     iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2930 			ret = -ENOENT;
2931 			goto release;
2932 		}
2933 		iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2934 							   path->slots[0]);
2935 		iter->item_ptr = iter->cur_ptr;
2936 		iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size(
2937 				      path->nodes[0], path->slots[0]));
2938 	}
2939 
2940 	return 0;
2941 release:
2942 	btrfs_backref_iter_release(iter);
2943 	return ret;
2944 }
2945 
btrfs_backref_iter_is_inline_ref(struct btrfs_backref_iter * iter)2946 static bool btrfs_backref_iter_is_inline_ref(struct btrfs_backref_iter *iter)
2947 {
2948 	if (iter->cur_key.type == BTRFS_EXTENT_ITEM_KEY ||
2949 	    iter->cur_key.type == BTRFS_METADATA_ITEM_KEY)
2950 		return true;
2951 	return false;
2952 }
2953 
2954 /*
2955  * Go to the next backref item of current bytenr, can be either inlined or
2956  * keyed.
2957  *
2958  * Caller needs to check whether it's inline ref or not by iter->cur_key.
2959  *
2960  * Return 0 if we get next backref without problem.
2961  * Return >0 if there is no extra backref for this bytenr.
2962  * Return <0 if there is something wrong happened.
2963  */
btrfs_backref_iter_next(struct btrfs_backref_iter * iter)2964 int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2965 {
2966 	struct extent_buffer *eb = iter->path->nodes[0];
2967 	struct btrfs_root *extent_root;
2968 	struct btrfs_path *path = iter->path;
2969 	struct btrfs_extent_inline_ref *iref;
2970 	int ret;
2971 	u32 size;
2972 
2973 	if (btrfs_backref_iter_is_inline_ref(iter)) {
2974 		/* We're still inside the inline refs */
2975 		ASSERT(iter->cur_ptr < iter->end_ptr);
2976 
2977 		if (btrfs_backref_has_tree_block_info(iter)) {
2978 			/* First tree block info */
2979 			size = sizeof(struct btrfs_tree_block_info);
2980 		} else {
2981 			/* Use inline ref type to determine the size */
2982 			int type;
2983 
2984 			iref = (struct btrfs_extent_inline_ref *)
2985 				((unsigned long)iter->cur_ptr);
2986 			type = btrfs_extent_inline_ref_type(eb, iref);
2987 
2988 			size = btrfs_extent_inline_ref_size(type);
2989 		}
2990 		iter->cur_ptr += size;
2991 		if (iter->cur_ptr < iter->end_ptr)
2992 			return 0;
2993 
2994 		/* All inline items iterated, fall through */
2995 	}
2996 
2997 	/* We're at keyed items, there is no inline item, go to the next one */
2998 	extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr);
2999 	ret = btrfs_next_item(extent_root, iter->path);
3000 	if (ret)
3001 		return ret;
3002 
3003 	btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
3004 	if (iter->cur_key.objectid != iter->bytenr ||
3005 	    (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
3006 	     iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
3007 		return 1;
3008 	iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
3009 					path->slots[0]);
3010 	iter->cur_ptr = iter->item_ptr;
3011 	iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0],
3012 						path->slots[0]);
3013 	return 0;
3014 }
3015 
btrfs_backref_init_cache(struct btrfs_fs_info * fs_info,struct btrfs_backref_cache * cache,bool is_reloc)3016 void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
3017 			      struct btrfs_backref_cache *cache, bool is_reloc)
3018 {
3019 	int i;
3020 
3021 	cache->rb_root = RB_ROOT;
3022 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3023 		INIT_LIST_HEAD(&cache->pending[i]);
3024 	INIT_LIST_HEAD(&cache->changed);
3025 	INIT_LIST_HEAD(&cache->detached);
3026 	INIT_LIST_HEAD(&cache->leaves);
3027 	INIT_LIST_HEAD(&cache->pending_edge);
3028 	INIT_LIST_HEAD(&cache->useless_node);
3029 	cache->fs_info = fs_info;
3030 	cache->is_reloc = is_reloc;
3031 }
3032 
btrfs_backref_alloc_node(struct btrfs_backref_cache * cache,u64 bytenr,int level)3033 struct btrfs_backref_node *btrfs_backref_alloc_node(
3034 		struct btrfs_backref_cache *cache, u64 bytenr, int level)
3035 {
3036 	struct btrfs_backref_node *node;
3037 
3038 	ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
3039 	node = kzalloc(sizeof(*node), GFP_NOFS);
3040 	if (!node)
3041 		return node;
3042 
3043 	INIT_LIST_HEAD(&node->list);
3044 	INIT_LIST_HEAD(&node->upper);
3045 	INIT_LIST_HEAD(&node->lower);
3046 	RB_CLEAR_NODE(&node->rb_node);
3047 	cache->nr_nodes++;
3048 	node->level = level;
3049 	node->bytenr = bytenr;
3050 
3051 	return node;
3052 }
3053 
btrfs_backref_free_node(struct btrfs_backref_cache * cache,struct btrfs_backref_node * node)3054 void btrfs_backref_free_node(struct btrfs_backref_cache *cache,
3055 			     struct btrfs_backref_node *node)
3056 {
3057 	if (node) {
3058 		ASSERT(list_empty(&node->list));
3059 		ASSERT(list_empty(&node->lower));
3060 		ASSERT(node->eb == NULL);
3061 		cache->nr_nodes--;
3062 		btrfs_put_root(node->root);
3063 		kfree(node);
3064 	}
3065 }
3066 
btrfs_backref_alloc_edge(struct btrfs_backref_cache * cache)3067 struct btrfs_backref_edge *btrfs_backref_alloc_edge(
3068 		struct btrfs_backref_cache *cache)
3069 {
3070 	struct btrfs_backref_edge *edge;
3071 
3072 	edge = kzalloc(sizeof(*edge), GFP_NOFS);
3073 	if (edge)
3074 		cache->nr_edges++;
3075 	return edge;
3076 }
3077 
btrfs_backref_free_edge(struct btrfs_backref_cache * cache,struct btrfs_backref_edge * edge)3078 void btrfs_backref_free_edge(struct btrfs_backref_cache *cache,
3079 			     struct btrfs_backref_edge *edge)
3080 {
3081 	if (edge) {
3082 		cache->nr_edges--;
3083 		kfree(edge);
3084 	}
3085 }
3086 
btrfs_backref_unlock_node_buffer(struct btrfs_backref_node * node)3087 void btrfs_backref_unlock_node_buffer(struct btrfs_backref_node *node)
3088 {
3089 	if (node->locked) {
3090 		btrfs_tree_unlock(node->eb);
3091 		node->locked = 0;
3092 	}
3093 }
3094 
btrfs_backref_drop_node_buffer(struct btrfs_backref_node * node)3095 void btrfs_backref_drop_node_buffer(struct btrfs_backref_node *node)
3096 {
3097 	if (node->eb) {
3098 		btrfs_backref_unlock_node_buffer(node);
3099 		free_extent_buffer(node->eb);
3100 		node->eb = NULL;
3101 	}
3102 }
3103 
3104 /*
3105  * Drop the backref node from cache without cleaning up its children
3106  * edges.
3107  *
3108  * This can only be called on node without parent edges.
3109  * The children edges are still kept as is.
3110  */
btrfs_backref_drop_node(struct btrfs_backref_cache * tree,struct btrfs_backref_node * node)3111 void btrfs_backref_drop_node(struct btrfs_backref_cache *tree,
3112 			     struct btrfs_backref_node *node)
3113 {
3114 	ASSERT(list_empty(&node->upper));
3115 
3116 	btrfs_backref_drop_node_buffer(node);
3117 	list_del_init(&node->list);
3118 	list_del_init(&node->lower);
3119 	if (!RB_EMPTY_NODE(&node->rb_node))
3120 		rb_erase(&node->rb_node, &tree->rb_root);
3121 	btrfs_backref_free_node(tree, node);
3122 }
3123 
3124 /*
3125  * Drop the backref node from cache, also cleaning up all its
3126  * upper edges and any uncached nodes in the path.
3127  *
3128  * This cleanup happens bottom up, thus the node should either
3129  * be the lowest node in the cache or a detached node.
3130  */
btrfs_backref_cleanup_node(struct btrfs_backref_cache * cache,struct btrfs_backref_node * node)3131 void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
3132 				struct btrfs_backref_node *node)
3133 {
3134 	struct btrfs_backref_node *upper;
3135 	struct btrfs_backref_edge *edge;
3136 
3137 	if (!node)
3138 		return;
3139 
3140 	BUG_ON(!node->lowest && !node->detached);
3141 	while (!list_empty(&node->upper)) {
3142 		edge = list_entry(node->upper.next, struct btrfs_backref_edge,
3143 				  list[LOWER]);
3144 		upper = edge->node[UPPER];
3145 		list_del(&edge->list[LOWER]);
3146 		list_del(&edge->list[UPPER]);
3147 		btrfs_backref_free_edge(cache, edge);
3148 
3149 		/*
3150 		 * Add the node to leaf node list if no other child block
3151 		 * cached.
3152 		 */
3153 		if (list_empty(&upper->lower)) {
3154 			list_add_tail(&upper->lower, &cache->leaves);
3155 			upper->lowest = 1;
3156 		}
3157 	}
3158 
3159 	btrfs_backref_drop_node(cache, node);
3160 }
3161 
3162 /*
3163  * Release all nodes/edges from current cache
3164  */
btrfs_backref_release_cache(struct btrfs_backref_cache * cache)3165 void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
3166 {
3167 	struct btrfs_backref_node *node;
3168 	int i;
3169 
3170 	while (!list_empty(&cache->detached)) {
3171 		node = list_entry(cache->detached.next,
3172 				  struct btrfs_backref_node, list);
3173 		btrfs_backref_cleanup_node(cache, node);
3174 	}
3175 
3176 	while (!list_empty(&cache->leaves)) {
3177 		node = list_entry(cache->leaves.next,
3178 				  struct btrfs_backref_node, lower);
3179 		btrfs_backref_cleanup_node(cache, node);
3180 	}
3181 
3182 	cache->last_trans = 0;
3183 
3184 	for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3185 		ASSERT(list_empty(&cache->pending[i]));
3186 	ASSERT(list_empty(&cache->pending_edge));
3187 	ASSERT(list_empty(&cache->useless_node));
3188 	ASSERT(list_empty(&cache->changed));
3189 	ASSERT(list_empty(&cache->detached));
3190 	ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
3191 	ASSERT(!cache->nr_nodes);
3192 	ASSERT(!cache->nr_edges);
3193 }
3194 
btrfs_backref_link_edge(struct btrfs_backref_edge * edge,struct btrfs_backref_node * lower,struct btrfs_backref_node * upper,int link_which)3195 void btrfs_backref_link_edge(struct btrfs_backref_edge *edge,
3196 			     struct btrfs_backref_node *lower,
3197 			     struct btrfs_backref_node *upper,
3198 			     int link_which)
3199 {
3200 	ASSERT(upper && lower && upper->level == lower->level + 1);
3201 	edge->node[LOWER] = lower;
3202 	edge->node[UPPER] = upper;
3203 	if (link_which & LINK_LOWER)
3204 		list_add_tail(&edge->list[LOWER], &lower->upper);
3205 	if (link_which & LINK_UPPER)
3206 		list_add_tail(&edge->list[UPPER], &upper->lower);
3207 }
3208 /*
3209  * Handle direct tree backref
3210  *
3211  * Direct tree backref means, the backref item shows its parent bytenr
3212  * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
3213  *
3214  * @ref_key:	The converted backref key.
3215  *		For keyed backref, it's the item key.
3216  *		For inlined backref, objectid is the bytenr,
3217  *		type is btrfs_inline_ref_type, offset is
3218  *		btrfs_inline_ref_offset.
3219  */
handle_direct_tree_backref(struct btrfs_backref_cache * cache,struct btrfs_key * ref_key,struct btrfs_backref_node * cur)3220 static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
3221 				      struct btrfs_key *ref_key,
3222 				      struct btrfs_backref_node *cur)
3223 {
3224 	struct btrfs_backref_edge *edge;
3225 	struct btrfs_backref_node *upper;
3226 	struct rb_node *rb_node;
3227 
3228 	ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
3229 
3230 	/* Only reloc root uses backref pointing to itself */
3231 	if (ref_key->objectid == ref_key->offset) {
3232 		struct btrfs_root *root;
3233 
3234 		cur->is_reloc_root = 1;
3235 		/* Only reloc backref cache cares about a specific root */
3236 		if (cache->is_reloc) {
3237 			root = find_reloc_root(cache->fs_info, cur->bytenr);
3238 			if (!root)
3239 				return -ENOENT;
3240 			cur->root = root;
3241 		} else {
3242 			/*
3243 			 * For generic purpose backref cache, reloc root node
3244 			 * is useless.
3245 			 */
3246 			list_add(&cur->list, &cache->useless_node);
3247 		}
3248 		return 0;
3249 	}
3250 
3251 	edge = btrfs_backref_alloc_edge(cache);
3252 	if (!edge)
3253 		return -ENOMEM;
3254 
3255 	rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
3256 	if (!rb_node) {
3257 		/* Parent node not yet cached */
3258 		upper = btrfs_backref_alloc_node(cache, ref_key->offset,
3259 					   cur->level + 1);
3260 		if (!upper) {
3261 			btrfs_backref_free_edge(cache, edge);
3262 			return -ENOMEM;
3263 		}
3264 
3265 		/*
3266 		 *  Backrefs for the upper level block isn't cached, add the
3267 		 *  block to pending list
3268 		 */
3269 		list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3270 	} else {
3271 		/* Parent node already cached */
3272 		upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
3273 		ASSERT(upper->checked);
3274 		INIT_LIST_HEAD(&edge->list[UPPER]);
3275 	}
3276 	btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
3277 	return 0;
3278 }
3279 
3280 /*
3281  * Handle indirect tree backref
3282  *
3283  * Indirect tree backref means, we only know which tree the node belongs to.
3284  * We still need to do a tree search to find out the parents. This is for
3285  * TREE_BLOCK_REF backref (keyed or inlined).
3286  *
3287  * @trans:	Transaction handle.
3288  * @ref_key:	The same as @ref_key in  handle_direct_tree_backref()
3289  * @tree_key:	The first key of this tree block.
3290  * @path:	A clean (released) path, to avoid allocating path every time
3291  *		the function get called.
3292  */
handle_indirect_tree_backref(struct btrfs_trans_handle * trans,struct btrfs_backref_cache * cache,struct btrfs_path * path,struct btrfs_key * ref_key,struct btrfs_key * tree_key,struct btrfs_backref_node * cur)3293 static int handle_indirect_tree_backref(struct btrfs_trans_handle *trans,
3294 					struct btrfs_backref_cache *cache,
3295 					struct btrfs_path *path,
3296 					struct btrfs_key *ref_key,
3297 					struct btrfs_key *tree_key,
3298 					struct btrfs_backref_node *cur)
3299 {
3300 	struct btrfs_fs_info *fs_info = cache->fs_info;
3301 	struct btrfs_backref_node *upper;
3302 	struct btrfs_backref_node *lower;
3303 	struct btrfs_backref_edge *edge;
3304 	struct extent_buffer *eb;
3305 	struct btrfs_root *root;
3306 	struct rb_node *rb_node;
3307 	int level;
3308 	bool need_check = true;
3309 	int ret;
3310 
3311 	root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
3312 	if (IS_ERR(root))
3313 		return PTR_ERR(root);
3314 	if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3315 		cur->cowonly = 1;
3316 
3317 	if (btrfs_root_level(&root->root_item) == cur->level) {
3318 		/* Tree root */
3319 		ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
3320 		/*
3321 		 * For reloc backref cache, we may ignore reloc root.  But for
3322 		 * general purpose backref cache, we can't rely on
3323 		 * btrfs_should_ignore_reloc_root() as it may conflict with
3324 		 * current running relocation and lead to missing root.
3325 		 *
3326 		 * For general purpose backref cache, reloc root detection is
3327 		 * completely relying on direct backref (key->offset is parent
3328 		 * bytenr), thus only do such check for reloc cache.
3329 		 */
3330 		if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
3331 			btrfs_put_root(root);
3332 			list_add(&cur->list, &cache->useless_node);
3333 		} else {
3334 			cur->root = root;
3335 		}
3336 		return 0;
3337 	}
3338 
3339 	level = cur->level + 1;
3340 
3341 	/* Search the tree to find parent blocks referring to the block */
3342 	path->search_commit_root = 1;
3343 	path->skip_locking = 1;
3344 	path->lowest_level = level;
3345 	ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
3346 	path->lowest_level = 0;
3347 	if (ret < 0) {
3348 		btrfs_put_root(root);
3349 		return ret;
3350 	}
3351 	if (ret > 0 && path->slots[level] > 0)
3352 		path->slots[level]--;
3353 
3354 	eb = path->nodes[level];
3355 	if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
3356 		btrfs_err(fs_info,
3357 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
3358 			  cur->bytenr, level - 1, btrfs_root_id(root),
3359 			  tree_key->objectid, tree_key->type, tree_key->offset);
3360 		btrfs_put_root(root);
3361 		ret = -ENOENT;
3362 		goto out;
3363 	}
3364 	lower = cur;
3365 
3366 	/* Add all nodes and edges in the path */
3367 	for (; level < BTRFS_MAX_LEVEL; level++) {
3368 		if (!path->nodes[level]) {
3369 			ASSERT(btrfs_root_bytenr(&root->root_item) ==
3370 			       lower->bytenr);
3371 			/* Same as previous should_ignore_reloc_root() call */
3372 			if (btrfs_should_ignore_reloc_root(root) &&
3373 			    cache->is_reloc) {
3374 				btrfs_put_root(root);
3375 				list_add(&lower->list, &cache->useless_node);
3376 			} else {
3377 				lower->root = root;
3378 			}
3379 			break;
3380 		}
3381 
3382 		edge = btrfs_backref_alloc_edge(cache);
3383 		if (!edge) {
3384 			btrfs_put_root(root);
3385 			ret = -ENOMEM;
3386 			goto out;
3387 		}
3388 
3389 		eb = path->nodes[level];
3390 		rb_node = rb_simple_search(&cache->rb_root, eb->start);
3391 		if (!rb_node) {
3392 			upper = btrfs_backref_alloc_node(cache, eb->start,
3393 							 lower->level + 1);
3394 			if (!upper) {
3395 				btrfs_put_root(root);
3396 				btrfs_backref_free_edge(cache, edge);
3397 				ret = -ENOMEM;
3398 				goto out;
3399 			}
3400 			upper->owner = btrfs_header_owner(eb);
3401 			if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3402 				upper->cowonly = 1;
3403 
3404 			/*
3405 			 * If we know the block isn't shared we can avoid
3406 			 * checking its backrefs.
3407 			 */
3408 			if (btrfs_block_can_be_shared(trans, root, eb))
3409 				upper->checked = 0;
3410 			else
3411 				upper->checked = 1;
3412 
3413 			/*
3414 			 * Add the block to pending list if we need to check its
3415 			 * backrefs, we only do this once while walking up a
3416 			 * tree as we will catch anything else later on.
3417 			 */
3418 			if (!upper->checked && need_check) {
3419 				need_check = false;
3420 				list_add_tail(&edge->list[UPPER],
3421 					      &cache->pending_edge);
3422 			} else {
3423 				if (upper->checked)
3424 					need_check = true;
3425 				INIT_LIST_HEAD(&edge->list[UPPER]);
3426 			}
3427 		} else {
3428 			upper = rb_entry(rb_node, struct btrfs_backref_node,
3429 					 rb_node);
3430 			ASSERT(upper->checked);
3431 			INIT_LIST_HEAD(&edge->list[UPPER]);
3432 			if (!upper->owner)
3433 				upper->owner = btrfs_header_owner(eb);
3434 		}
3435 		btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
3436 
3437 		if (rb_node) {
3438 			btrfs_put_root(root);
3439 			break;
3440 		}
3441 		lower = upper;
3442 		upper = NULL;
3443 	}
3444 out:
3445 	btrfs_release_path(path);
3446 	return ret;
3447 }
3448 
3449 /*
3450  * Add backref node @cur into @cache.
3451  *
3452  * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
3453  *	 links aren't yet bi-directional. Needs to finish such links.
3454  *	 Use btrfs_backref_finish_upper_links() to finish such linkage.
3455  *
3456  * @trans:	Transaction handle.
3457  * @path:	Released path for indirect tree backref lookup
3458  * @iter:	Released backref iter for extent tree search
3459  * @node_key:	The first key of the tree block
3460  */
btrfs_backref_add_tree_node(struct btrfs_trans_handle * trans,struct btrfs_backref_cache * cache,struct btrfs_path * path,struct btrfs_backref_iter * iter,struct btrfs_key * node_key,struct btrfs_backref_node * cur)3461 int btrfs_backref_add_tree_node(struct btrfs_trans_handle *trans,
3462 				struct btrfs_backref_cache *cache,
3463 				struct btrfs_path *path,
3464 				struct btrfs_backref_iter *iter,
3465 				struct btrfs_key *node_key,
3466 				struct btrfs_backref_node *cur)
3467 {
3468 	struct btrfs_backref_edge *edge;
3469 	struct btrfs_backref_node *exist;
3470 	int ret;
3471 
3472 	ret = btrfs_backref_iter_start(iter, cur->bytenr);
3473 	if (ret < 0)
3474 		return ret;
3475 	/*
3476 	 * We skip the first btrfs_tree_block_info, as we don't use the key
3477 	 * stored in it, but fetch it from the tree block
3478 	 */
3479 	if (btrfs_backref_has_tree_block_info(iter)) {
3480 		ret = btrfs_backref_iter_next(iter);
3481 		if (ret < 0)
3482 			goto out;
3483 		/* No extra backref? This means the tree block is corrupted */
3484 		if (ret > 0) {
3485 			ret = -EUCLEAN;
3486 			goto out;
3487 		}
3488 	}
3489 	WARN_ON(cur->checked);
3490 	if (!list_empty(&cur->upper)) {
3491 		/*
3492 		 * The backref was added previously when processing backref of
3493 		 * type BTRFS_TREE_BLOCK_REF_KEY
3494 		 */
3495 		ASSERT(list_is_singular(&cur->upper));
3496 		edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
3497 				  list[LOWER]);
3498 		ASSERT(list_empty(&edge->list[UPPER]));
3499 		exist = edge->node[UPPER];
3500 		/*
3501 		 * Add the upper level block to pending list if we need check
3502 		 * its backrefs
3503 		 */
3504 		if (!exist->checked)
3505 			list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3506 	} else {
3507 		exist = NULL;
3508 	}
3509 
3510 	for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
3511 		struct extent_buffer *eb;
3512 		struct btrfs_key key;
3513 		int type;
3514 
3515 		cond_resched();
3516 		eb = iter->path->nodes[0];
3517 
3518 		key.objectid = iter->bytenr;
3519 		if (btrfs_backref_iter_is_inline_ref(iter)) {
3520 			struct btrfs_extent_inline_ref *iref;
3521 
3522 			/* Update key for inline backref */
3523 			iref = (struct btrfs_extent_inline_ref *)
3524 				((unsigned long)iter->cur_ptr);
3525 			type = btrfs_get_extent_inline_ref_type(eb, iref,
3526 							BTRFS_REF_TYPE_BLOCK);
3527 			if (type == BTRFS_REF_TYPE_INVALID) {
3528 				ret = -EUCLEAN;
3529 				goto out;
3530 			}
3531 			key.type = type;
3532 			key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3533 		} else {
3534 			key.type = iter->cur_key.type;
3535 			key.offset = iter->cur_key.offset;
3536 		}
3537 
3538 		/*
3539 		 * Parent node found and matches current inline ref, no need to
3540 		 * rebuild this node for this inline ref
3541 		 */
3542 		if (exist &&
3543 		    ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
3544 		      exist->owner == key.offset) ||
3545 		     (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
3546 		      exist->bytenr == key.offset))) {
3547 			exist = NULL;
3548 			continue;
3549 		}
3550 
3551 		/* SHARED_BLOCK_REF means key.offset is the parent bytenr */
3552 		if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3553 			ret = handle_direct_tree_backref(cache, &key, cur);
3554 			if (ret < 0)
3555 				goto out;
3556 		} else if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
3557 			/*
3558 			 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref
3559 			 * offset means the root objectid. We need to search
3560 			 * the tree to get its parent bytenr.
3561 			 */
3562 			ret = handle_indirect_tree_backref(trans, cache, path,
3563 							   &key, node_key, cur);
3564 			if (ret < 0)
3565 				goto out;
3566 		}
3567 		/*
3568 		 * Unrecognized tree backref items (if it can pass tree-checker)
3569 		 * would be ignored.
3570 		 */
3571 	}
3572 	ret = 0;
3573 	cur->checked = 1;
3574 	WARN_ON(exist);
3575 out:
3576 	btrfs_backref_iter_release(iter);
3577 	return ret;
3578 }
3579 
3580 /*
3581  * Finish the upwards linkage created by btrfs_backref_add_tree_node()
3582  */
btrfs_backref_finish_upper_links(struct btrfs_backref_cache * cache,struct btrfs_backref_node * start)3583 int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
3584 				     struct btrfs_backref_node *start)
3585 {
3586 	struct list_head *useless_node = &cache->useless_node;
3587 	struct btrfs_backref_edge *edge;
3588 	struct rb_node *rb_node;
3589 	LIST_HEAD(pending_edge);
3590 
3591 	ASSERT(start->checked);
3592 
3593 	/* Insert this node to cache if it's not COW-only */
3594 	if (!start->cowonly) {
3595 		rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
3596 					   &start->rb_node);
3597 		if (rb_node)
3598 			btrfs_backref_panic(cache->fs_info, start->bytenr,
3599 					    -EEXIST);
3600 		list_add_tail(&start->lower, &cache->leaves);
3601 	}
3602 
3603 	/*
3604 	 * Use breadth first search to iterate all related edges.
3605 	 *
3606 	 * The starting points are all the edges of this node
3607 	 */
3608 	list_for_each_entry(edge, &start->upper, list[LOWER])
3609 		list_add_tail(&edge->list[UPPER], &pending_edge);
3610 
3611 	while (!list_empty(&pending_edge)) {
3612 		struct btrfs_backref_node *upper;
3613 		struct btrfs_backref_node *lower;
3614 
3615 		edge = list_first_entry(&pending_edge,
3616 				struct btrfs_backref_edge, list[UPPER]);
3617 		list_del_init(&edge->list[UPPER]);
3618 		upper = edge->node[UPPER];
3619 		lower = edge->node[LOWER];
3620 
3621 		/* Parent is detached, no need to keep any edges */
3622 		if (upper->detached) {
3623 			list_del(&edge->list[LOWER]);
3624 			btrfs_backref_free_edge(cache, edge);
3625 
3626 			/* Lower node is orphan, queue for cleanup */
3627 			if (list_empty(&lower->upper))
3628 				list_add(&lower->list, useless_node);
3629 			continue;
3630 		}
3631 
3632 		/*
3633 		 * All new nodes added in current build_backref_tree() haven't
3634 		 * been linked to the cache rb tree.
3635 		 * So if we have upper->rb_node populated, this means a cache
3636 		 * hit. We only need to link the edge, as @upper and all its
3637 		 * parents have already been linked.
3638 		 */
3639 		if (!RB_EMPTY_NODE(&upper->rb_node)) {
3640 			if (upper->lowest) {
3641 				list_del_init(&upper->lower);
3642 				upper->lowest = 0;
3643 			}
3644 
3645 			list_add_tail(&edge->list[UPPER], &upper->lower);
3646 			continue;
3647 		}
3648 
3649 		/* Sanity check, we shouldn't have any unchecked nodes */
3650 		if (!upper->checked) {
3651 			ASSERT(0);
3652 			return -EUCLEAN;
3653 		}
3654 
3655 		/* Sanity check, COW-only node has non-COW-only parent */
3656 		if (start->cowonly != upper->cowonly) {
3657 			ASSERT(0);
3658 			return -EUCLEAN;
3659 		}
3660 
3661 		/* Only cache non-COW-only (subvolume trees) tree blocks */
3662 		if (!upper->cowonly) {
3663 			rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3664 						   &upper->rb_node);
3665 			if (rb_node) {
3666 				btrfs_backref_panic(cache->fs_info,
3667 						upper->bytenr, -EEXIST);
3668 				return -EUCLEAN;
3669 			}
3670 		}
3671 
3672 		list_add_tail(&edge->list[UPPER], &upper->lower);
3673 
3674 		/*
3675 		 * Also queue all the parent edges of this uncached node
3676 		 * to finish the upper linkage
3677 		 */
3678 		list_for_each_entry(edge, &upper->upper, list[LOWER])
3679 			list_add_tail(&edge->list[UPPER], &pending_edge);
3680 	}
3681 	return 0;
3682 }
3683 
btrfs_backref_error_cleanup(struct btrfs_backref_cache * cache,struct btrfs_backref_node * node)3684 void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3685 				 struct btrfs_backref_node *node)
3686 {
3687 	struct btrfs_backref_node *lower;
3688 	struct btrfs_backref_node *upper;
3689 	struct btrfs_backref_edge *edge;
3690 
3691 	while (!list_empty(&cache->useless_node)) {
3692 		lower = list_first_entry(&cache->useless_node,
3693 				   struct btrfs_backref_node, list);
3694 		list_del_init(&lower->list);
3695 	}
3696 	while (!list_empty(&cache->pending_edge)) {
3697 		edge = list_first_entry(&cache->pending_edge,
3698 				struct btrfs_backref_edge, list[UPPER]);
3699 		list_del(&edge->list[UPPER]);
3700 		list_del(&edge->list[LOWER]);
3701 		lower = edge->node[LOWER];
3702 		upper = edge->node[UPPER];
3703 		btrfs_backref_free_edge(cache, edge);
3704 
3705 		/*
3706 		 * Lower is no longer linked to any upper backref nodes and
3707 		 * isn't in the cache, we can free it ourselves.
3708 		 */
3709 		if (list_empty(&lower->upper) &&
3710 		    RB_EMPTY_NODE(&lower->rb_node))
3711 			list_add(&lower->list, &cache->useless_node);
3712 
3713 		if (!RB_EMPTY_NODE(&upper->rb_node))
3714 			continue;
3715 
3716 		/* Add this guy's upper edges to the list to process */
3717 		list_for_each_entry(edge, &upper->upper, list[LOWER])
3718 			list_add_tail(&edge->list[UPPER],
3719 				      &cache->pending_edge);
3720 		if (list_empty(&upper->upper))
3721 			list_add(&upper->list, &cache->useless_node);
3722 	}
3723 
3724 	while (!list_empty(&cache->useless_node)) {
3725 		lower = list_first_entry(&cache->useless_node,
3726 				   struct btrfs_backref_node, list);
3727 		list_del_init(&lower->list);
3728 		if (lower == node)
3729 			node = NULL;
3730 		btrfs_backref_drop_node(cache, lower);
3731 	}
3732 
3733 	btrfs_backref_cleanup_node(cache, node);
3734 	ASSERT(list_empty(&cache->useless_node) &&
3735 	       list_empty(&cache->pending_edge));
3736 }
3737