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
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
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
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
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
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
207 void __cold btrfs_prelim_ref_exit(void)
208 {
209 kmem_cache_destroy(btrfs_prelim_ref_cache);
210 }
211
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 */
222 static int prelim_ref_compare(struct prelim_ref *ref1,
223 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
253 static void update_share_count(struct share_check *sc, int oldcount,
254 int newcount, 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 */
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 */
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 */
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 */
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 */
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
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
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 */
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 *
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
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 */
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 */
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 */
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 */
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 */
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 */
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 */
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 */
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 */
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 */
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
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
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
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 */
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
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 */
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 */
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 */
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 */
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
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 */
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
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
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
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
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 */
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 */
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
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 */
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
2814 void free_ipath(struct inode_fs_paths *ipath)
2815 {
2816 if (!ipath)
2817 return;
2818 kvfree(ipath->fspath);
2819 kfree(ipath);
2820 }
2821
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
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
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
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 */
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
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
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
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
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
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
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
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 */
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 */
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 */
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 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
3183 while (!list_empty(&cache->pending[i])) {
3184 node = list_first_entry(&cache->pending[i],
3185 struct btrfs_backref_node,
3186 list);
3187 btrfs_backref_cleanup_node(cache, node);
3188 }
3189 }
3190 ASSERT(list_empty(&cache->pending_edge));
3191 ASSERT(list_empty(&cache->useless_node));
3192 ASSERT(list_empty(&cache->changed));
3193 ASSERT(list_empty(&cache->detached));
3194 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
3195 ASSERT(!cache->nr_nodes);
3196 ASSERT(!cache->nr_edges);
3197 }
3198
3199 void btrfs_backref_link_edge(struct btrfs_backref_edge *edge,
3200 struct btrfs_backref_node *lower,
3201 struct btrfs_backref_node *upper,
3202 int link_which)
3203 {
3204 ASSERT(upper && lower && upper->level == lower->level + 1);
3205 edge->node[LOWER] = lower;
3206 edge->node[UPPER] = upper;
3207 if (link_which & LINK_LOWER)
3208 list_add_tail(&edge->list[LOWER], &lower->upper);
3209 if (link_which & LINK_UPPER)
3210 list_add_tail(&edge->list[UPPER], &upper->lower);
3211 }
3212 /*
3213 * Handle direct tree backref
3214 *
3215 * Direct tree backref means, the backref item shows its parent bytenr
3216 * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
3217 *
3218 * @ref_key: The converted backref key.
3219 * For keyed backref, it's the item key.
3220 * For inlined backref, objectid is the bytenr,
3221 * type is btrfs_inline_ref_type, offset is
3222 * btrfs_inline_ref_offset.
3223 */
3224 static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
3225 struct btrfs_key *ref_key,
3226 struct btrfs_backref_node *cur)
3227 {
3228 struct btrfs_backref_edge *edge;
3229 struct btrfs_backref_node *upper;
3230 struct rb_node *rb_node;
3231
3232 ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
3233
3234 /* Only reloc root uses backref pointing to itself */
3235 if (ref_key->objectid == ref_key->offset) {
3236 struct btrfs_root *root;
3237
3238 cur->is_reloc_root = 1;
3239 /* Only reloc backref cache cares about a specific root */
3240 if (cache->is_reloc) {
3241 root = find_reloc_root(cache->fs_info, cur->bytenr);
3242 if (!root)
3243 return -ENOENT;
3244 cur->root = root;
3245 } else {
3246 /*
3247 * For generic purpose backref cache, reloc root node
3248 * is useless.
3249 */
3250 list_add(&cur->list, &cache->useless_node);
3251 }
3252 return 0;
3253 }
3254
3255 edge = btrfs_backref_alloc_edge(cache);
3256 if (!edge)
3257 return -ENOMEM;
3258
3259 rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
3260 if (!rb_node) {
3261 /* Parent node not yet cached */
3262 upper = btrfs_backref_alloc_node(cache, ref_key->offset,
3263 cur->level + 1);
3264 if (!upper) {
3265 btrfs_backref_free_edge(cache, edge);
3266 return -ENOMEM;
3267 }
3268
3269 /*
3270 * Backrefs for the upper level block isn't cached, add the
3271 * block to pending list
3272 */
3273 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3274 } else {
3275 /* Parent node already cached */
3276 upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
3277 ASSERT(upper->checked);
3278 INIT_LIST_HEAD(&edge->list[UPPER]);
3279 }
3280 btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
3281 return 0;
3282 }
3283
3284 /*
3285 * Handle indirect tree backref
3286 *
3287 * Indirect tree backref means, we only know which tree the node belongs to.
3288 * We still need to do a tree search to find out the parents. This is for
3289 * TREE_BLOCK_REF backref (keyed or inlined).
3290 *
3291 * @trans: Transaction handle.
3292 * @ref_key: The same as @ref_key in handle_direct_tree_backref()
3293 * @tree_key: The first key of this tree block.
3294 * @path: A clean (released) path, to avoid allocating path every time
3295 * the function get called.
3296 */
3297 static int handle_indirect_tree_backref(struct btrfs_trans_handle *trans,
3298 struct btrfs_backref_cache *cache,
3299 struct btrfs_path *path,
3300 struct btrfs_key *ref_key,
3301 struct btrfs_key *tree_key,
3302 struct btrfs_backref_node *cur)
3303 {
3304 struct btrfs_fs_info *fs_info = cache->fs_info;
3305 struct btrfs_backref_node *upper;
3306 struct btrfs_backref_node *lower;
3307 struct btrfs_backref_edge *edge;
3308 struct extent_buffer *eb;
3309 struct btrfs_root *root;
3310 struct rb_node *rb_node;
3311 int level;
3312 bool need_check = true;
3313 int ret;
3314
3315 root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
3316 if (IS_ERR(root))
3317 return PTR_ERR(root);
3318 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3319 cur->cowonly = 1;
3320
3321 if (btrfs_root_level(&root->root_item) == cur->level) {
3322 /* Tree root */
3323 ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
3324 /*
3325 * For reloc backref cache, we may ignore reloc root. But for
3326 * general purpose backref cache, we can't rely on
3327 * btrfs_should_ignore_reloc_root() as it may conflict with
3328 * current running relocation and lead to missing root.
3329 *
3330 * For general purpose backref cache, reloc root detection is
3331 * completely relying on direct backref (key->offset is parent
3332 * bytenr), thus only do such check for reloc cache.
3333 */
3334 if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
3335 btrfs_put_root(root);
3336 list_add(&cur->list, &cache->useless_node);
3337 } else {
3338 cur->root = root;
3339 }
3340 return 0;
3341 }
3342
3343 level = cur->level + 1;
3344
3345 /* Search the tree to find parent blocks referring to the block */
3346 path->search_commit_root = 1;
3347 path->skip_locking = 1;
3348 path->lowest_level = level;
3349 ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
3350 path->lowest_level = 0;
3351 if (ret < 0) {
3352 btrfs_put_root(root);
3353 return ret;
3354 }
3355 if (ret > 0 && path->slots[level] > 0)
3356 path->slots[level]--;
3357
3358 eb = path->nodes[level];
3359 if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
3360 btrfs_err(fs_info,
3361 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
3362 cur->bytenr, level - 1, btrfs_root_id(root),
3363 tree_key->objectid, tree_key->type, tree_key->offset);
3364 btrfs_put_root(root);
3365 ret = -ENOENT;
3366 goto out;
3367 }
3368 lower = cur;
3369
3370 /* Add all nodes and edges in the path */
3371 for (; level < BTRFS_MAX_LEVEL; level++) {
3372 if (!path->nodes[level]) {
3373 ASSERT(btrfs_root_bytenr(&root->root_item) ==
3374 lower->bytenr);
3375 /* Same as previous should_ignore_reloc_root() call */
3376 if (btrfs_should_ignore_reloc_root(root) &&
3377 cache->is_reloc) {
3378 btrfs_put_root(root);
3379 list_add(&lower->list, &cache->useless_node);
3380 } else {
3381 lower->root = root;
3382 }
3383 break;
3384 }
3385
3386 edge = btrfs_backref_alloc_edge(cache);
3387 if (!edge) {
3388 btrfs_put_root(root);
3389 ret = -ENOMEM;
3390 goto out;
3391 }
3392
3393 eb = path->nodes[level];
3394 rb_node = rb_simple_search(&cache->rb_root, eb->start);
3395 if (!rb_node) {
3396 upper = btrfs_backref_alloc_node(cache, eb->start,
3397 lower->level + 1);
3398 if (!upper) {
3399 btrfs_put_root(root);
3400 btrfs_backref_free_edge(cache, edge);
3401 ret = -ENOMEM;
3402 goto out;
3403 }
3404 upper->owner = btrfs_header_owner(eb);
3405 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3406 upper->cowonly = 1;
3407
3408 /*
3409 * If we know the block isn't shared we can avoid
3410 * checking its backrefs.
3411 */
3412 if (btrfs_block_can_be_shared(trans, root, eb))
3413 upper->checked = 0;
3414 else
3415 upper->checked = 1;
3416
3417 /*
3418 * Add the block to pending list if we need to check its
3419 * backrefs, we only do this once while walking up a
3420 * tree as we will catch anything else later on.
3421 */
3422 if (!upper->checked && need_check) {
3423 need_check = false;
3424 list_add_tail(&edge->list[UPPER],
3425 &cache->pending_edge);
3426 } else {
3427 if (upper->checked)
3428 need_check = true;
3429 INIT_LIST_HEAD(&edge->list[UPPER]);
3430 }
3431 } else {
3432 upper = rb_entry(rb_node, struct btrfs_backref_node,
3433 rb_node);
3434 ASSERT(upper->checked);
3435 INIT_LIST_HEAD(&edge->list[UPPER]);
3436 if (!upper->owner)
3437 upper->owner = btrfs_header_owner(eb);
3438 }
3439 btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
3440
3441 if (rb_node) {
3442 btrfs_put_root(root);
3443 break;
3444 }
3445 lower = upper;
3446 upper = NULL;
3447 }
3448 out:
3449 btrfs_release_path(path);
3450 return ret;
3451 }
3452
3453 /*
3454 * Add backref node @cur into @cache.
3455 *
3456 * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
3457 * links aren't yet bi-directional. Needs to finish such links.
3458 * Use btrfs_backref_finish_upper_links() to finish such linkage.
3459 *
3460 * @trans: Transaction handle.
3461 * @path: Released path for indirect tree backref lookup
3462 * @iter: Released backref iter for extent tree search
3463 * @node_key: The first key of the tree block
3464 */
3465 int btrfs_backref_add_tree_node(struct btrfs_trans_handle *trans,
3466 struct btrfs_backref_cache *cache,
3467 struct btrfs_path *path,
3468 struct btrfs_backref_iter *iter,
3469 struct btrfs_key *node_key,
3470 struct btrfs_backref_node *cur)
3471 {
3472 struct btrfs_backref_edge *edge;
3473 struct btrfs_backref_node *exist;
3474 int ret;
3475
3476 ret = btrfs_backref_iter_start(iter, cur->bytenr);
3477 if (ret < 0)
3478 return ret;
3479 /*
3480 * We skip the first btrfs_tree_block_info, as we don't use the key
3481 * stored in it, but fetch it from the tree block
3482 */
3483 if (btrfs_backref_has_tree_block_info(iter)) {
3484 ret = btrfs_backref_iter_next(iter);
3485 if (ret < 0)
3486 goto out;
3487 /* No extra backref? This means the tree block is corrupted */
3488 if (ret > 0) {
3489 ret = -EUCLEAN;
3490 goto out;
3491 }
3492 }
3493 WARN_ON(cur->checked);
3494 if (!list_empty(&cur->upper)) {
3495 /*
3496 * The backref was added previously when processing backref of
3497 * type BTRFS_TREE_BLOCK_REF_KEY
3498 */
3499 ASSERT(list_is_singular(&cur->upper));
3500 edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
3501 list[LOWER]);
3502 ASSERT(list_empty(&edge->list[UPPER]));
3503 exist = edge->node[UPPER];
3504 /*
3505 * Add the upper level block to pending list if we need check
3506 * its backrefs
3507 */
3508 if (!exist->checked)
3509 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3510 } else {
3511 exist = NULL;
3512 }
3513
3514 for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
3515 struct extent_buffer *eb;
3516 struct btrfs_key key;
3517 int type;
3518
3519 cond_resched();
3520 eb = iter->path->nodes[0];
3521
3522 key.objectid = iter->bytenr;
3523 if (btrfs_backref_iter_is_inline_ref(iter)) {
3524 struct btrfs_extent_inline_ref *iref;
3525
3526 /* Update key for inline backref */
3527 iref = (struct btrfs_extent_inline_ref *)
3528 ((unsigned long)iter->cur_ptr);
3529 type = btrfs_get_extent_inline_ref_type(eb, iref,
3530 BTRFS_REF_TYPE_BLOCK);
3531 if (type == BTRFS_REF_TYPE_INVALID) {
3532 ret = -EUCLEAN;
3533 goto out;
3534 }
3535 key.type = type;
3536 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3537 } else {
3538 key.type = iter->cur_key.type;
3539 key.offset = iter->cur_key.offset;
3540 }
3541
3542 /*
3543 * Parent node found and matches current inline ref, no need to
3544 * rebuild this node for this inline ref
3545 */
3546 if (exist &&
3547 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
3548 exist->owner == key.offset) ||
3549 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
3550 exist->bytenr == key.offset))) {
3551 exist = NULL;
3552 continue;
3553 }
3554
3555 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
3556 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3557 ret = handle_direct_tree_backref(cache, &key, cur);
3558 if (ret < 0)
3559 goto out;
3560 } else if (key.type == BTRFS_TREE_BLOCK_REF_KEY) {
3561 /*
3562 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref
3563 * offset means the root objectid. We need to search
3564 * the tree to get its parent bytenr.
3565 */
3566 ret = handle_indirect_tree_backref(trans, cache, path,
3567 &key, node_key, cur);
3568 if (ret < 0)
3569 goto out;
3570 }
3571 /*
3572 * Unrecognized tree backref items (if it can pass tree-checker)
3573 * would be ignored.
3574 */
3575 }
3576 ret = 0;
3577 cur->checked = 1;
3578 WARN_ON(exist);
3579 out:
3580 btrfs_backref_iter_release(iter);
3581 return ret;
3582 }
3583
3584 /*
3585 * Finish the upwards linkage created by btrfs_backref_add_tree_node()
3586 */
3587 int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
3588 struct btrfs_backref_node *start)
3589 {
3590 struct list_head *useless_node = &cache->useless_node;
3591 struct btrfs_backref_edge *edge;
3592 struct rb_node *rb_node;
3593 LIST_HEAD(pending_edge);
3594
3595 ASSERT(start->checked);
3596
3597 /* Insert this node to cache if it's not COW-only */
3598 if (!start->cowonly) {
3599 rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
3600 &start->rb_node);
3601 if (rb_node)
3602 btrfs_backref_panic(cache->fs_info, start->bytenr,
3603 -EEXIST);
3604 list_add_tail(&start->lower, &cache->leaves);
3605 }
3606
3607 /*
3608 * Use breadth first search to iterate all related edges.
3609 *
3610 * The starting points are all the edges of this node
3611 */
3612 list_for_each_entry(edge, &start->upper, list[LOWER])
3613 list_add_tail(&edge->list[UPPER], &pending_edge);
3614
3615 while (!list_empty(&pending_edge)) {
3616 struct btrfs_backref_node *upper;
3617 struct btrfs_backref_node *lower;
3618
3619 edge = list_first_entry(&pending_edge,
3620 struct btrfs_backref_edge, list[UPPER]);
3621 list_del_init(&edge->list[UPPER]);
3622 upper = edge->node[UPPER];
3623 lower = edge->node[LOWER];
3624
3625 /* Parent is detached, no need to keep any edges */
3626 if (upper->detached) {
3627 list_del(&edge->list[LOWER]);
3628 btrfs_backref_free_edge(cache, edge);
3629
3630 /* Lower node is orphan, queue for cleanup */
3631 if (list_empty(&lower->upper))
3632 list_add(&lower->list, useless_node);
3633 continue;
3634 }
3635
3636 /*
3637 * All new nodes added in current build_backref_tree() haven't
3638 * been linked to the cache rb tree.
3639 * So if we have upper->rb_node populated, this means a cache
3640 * hit. We only need to link the edge, as @upper and all its
3641 * parents have already been linked.
3642 */
3643 if (!RB_EMPTY_NODE(&upper->rb_node)) {
3644 if (upper->lowest) {
3645 list_del_init(&upper->lower);
3646 upper->lowest = 0;
3647 }
3648
3649 list_add_tail(&edge->list[UPPER], &upper->lower);
3650 continue;
3651 }
3652
3653 /* Sanity check, we shouldn't have any unchecked nodes */
3654 if (!upper->checked) {
3655 ASSERT(0);
3656 return -EUCLEAN;
3657 }
3658
3659 /* Sanity check, COW-only node has non-COW-only parent */
3660 if (start->cowonly != upper->cowonly) {
3661 ASSERT(0);
3662 return -EUCLEAN;
3663 }
3664
3665 /* Only cache non-COW-only (subvolume trees) tree blocks */
3666 if (!upper->cowonly) {
3667 rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3668 &upper->rb_node);
3669 if (rb_node) {
3670 btrfs_backref_panic(cache->fs_info,
3671 upper->bytenr, -EEXIST);
3672 return -EUCLEAN;
3673 }
3674 }
3675
3676 list_add_tail(&edge->list[UPPER], &upper->lower);
3677
3678 /*
3679 * Also queue all the parent edges of this uncached node
3680 * to finish the upper linkage
3681 */
3682 list_for_each_entry(edge, &upper->upper, list[LOWER])
3683 list_add_tail(&edge->list[UPPER], &pending_edge);
3684 }
3685 return 0;
3686 }
3687
3688 void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3689 struct btrfs_backref_node *node)
3690 {
3691 struct btrfs_backref_node *lower;
3692 struct btrfs_backref_node *upper;
3693 struct btrfs_backref_edge *edge;
3694
3695 while (!list_empty(&cache->useless_node)) {
3696 lower = list_first_entry(&cache->useless_node,
3697 struct btrfs_backref_node, list);
3698 list_del_init(&lower->list);
3699 }
3700 while (!list_empty(&cache->pending_edge)) {
3701 edge = list_first_entry(&cache->pending_edge,
3702 struct btrfs_backref_edge, list[UPPER]);
3703 list_del(&edge->list[UPPER]);
3704 list_del(&edge->list[LOWER]);
3705 lower = edge->node[LOWER];
3706 upper = edge->node[UPPER];
3707 btrfs_backref_free_edge(cache, edge);
3708
3709 /*
3710 * Lower is no longer linked to any upper backref nodes and
3711 * isn't in the cache, we can free it ourselves.
3712 */
3713 if (list_empty(&lower->upper) &&
3714 RB_EMPTY_NODE(&lower->rb_node))
3715 list_add(&lower->list, &cache->useless_node);
3716
3717 if (!RB_EMPTY_NODE(&upper->rb_node))
3718 continue;
3719
3720 /* Add this guy's upper edges to the list to process */
3721 list_for_each_entry(edge, &upper->upper, list[LOWER])
3722 list_add_tail(&edge->list[UPPER],
3723 &cache->pending_edge);
3724 if (list_empty(&upper->upper))
3725 list_add(&upper->list, &cache->useless_node);
3726 }
3727
3728 while (!list_empty(&cache->useless_node)) {
3729 lower = list_first_entry(&cache->useless_node,
3730 struct btrfs_backref_node, list);
3731 list_del_init(&lower->list);
3732 if (lower == node)
3733 node = NULL;
3734 btrfs_backref_drop_node(cache, lower);
3735 }
3736
3737 btrfs_backref_cleanup_node(cache, node);
3738 ASSERT(list_empty(&cache->useless_node) &&
3739 list_empty(&cache->pending_edge));
3740 }
3741
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.