1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2011 STRATO. All rights reserved.
7 #include <linux/rbtree.h>
8 #include <trace/events/btrfs.h>
13 #include "transaction.h"
14 #include "delayed-ref.h"
17 #include "tree-mod-log.h"
19 #include "accessors.h"
20 #include "extent-tree.h"
21 #include "relocation.h"
22 #include "tree-checker.h"
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
28 struct extent_inode_elem {
32 struct extent_inode_elem *next;
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)
41 const u64 data_len = btrfs_file_extent_num_bytes(eb, fi);
42 u64 offset = key->offset;
43 struct extent_inode_elem *e;
48 if (!btrfs_file_extent_compression(eb, fi) &&
49 !btrfs_file_extent_encryption(eb, fi) &&
50 !btrfs_file_extent_other_encoding(eb, fi)) {
53 data_offset = btrfs_file_extent_offset(eb, fi);
55 if (ctx->extent_item_pos < data_offset ||
56 ctx->extent_item_pos >= data_offset + data_len)
58 offset += ctx->extent_item_pos - data_offset;
61 if (!ctx->indirect_ref_iterator || !ctx->cache_lookup)
64 cached = ctx->cache_lookup(eb->start, ctx->user_ctx, &root_ids,
69 for (int i = 0; i < root_count; i++) {
72 ret = ctx->indirect_ref_iterator(key->objectid, offset,
73 data_len, root_ids[i],
80 e = kmalloc(sizeof(*e), GFP_NOFS);
85 e->inum = key->objectid;
87 e->num_bytes = data_len;
93 static void free_inode_elem_list(struct extent_inode_elem *eie)
95 struct extent_inode_elem *eie_next;
97 for (; eie; eie = eie_next) {
103 static int find_extent_in_eb(struct btrfs_backref_walk_ctx *ctx,
104 const struct extent_buffer *eb,
105 struct extent_inode_elem **eie)
108 struct btrfs_key key;
109 struct btrfs_file_extent_item *fi;
116 * from the shared data ref, we only have the leaf but we need
117 * the key. thus, we must look into all items and see that we
118 * find one (some) with a reference to our extent item.
120 nritems = btrfs_header_nritems(eb);
121 for (slot = 0; slot < nritems; ++slot) {
122 btrfs_item_key_to_cpu(eb, &key, slot);
123 if (key.type != BTRFS_EXTENT_DATA_KEY)
125 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
126 extent_type = btrfs_file_extent_type(eb, fi);
127 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
129 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
130 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
131 if (disk_byte != ctx->bytenr)
134 ret = check_extent_in_eb(ctx, &key, eb, fi, eie);
135 if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
143 struct rb_root_cached root;
147 #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 }
150 struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
151 struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
152 struct preftree indirect_missing_keys;
156 * Checks for a shared extent during backref search.
158 * The share_count tracks prelim_refs (direct and indirect) having a
160 * - incremented when a ref->count transitions to >0
161 * - decremented when a ref->count transitions to <1
164 struct btrfs_backref_share_check_ctx *ctx;
165 struct btrfs_root *root;
170 * Counts number of inodes that refer to an extent (different inodes in
171 * the same root or different roots) that we could find. The sharedness
172 * check typically stops once this counter gets greater than 1, so it
173 * may not reflect the total number of inodes.
177 * The number of times we found our inode refers to the data extent we
178 * are determining the sharedness. In other words, how many file extent
179 * items we could find for our inode that point to our target data
180 * extent. The value we get here after finishing the extent sharedness
181 * check may be smaller than reality, but if it ends up being greater
182 * than 1, then we know for sure the inode has multiple file extent
183 * items that point to our inode, and we can safely assume it's useful
184 * to cache the sharedness check result.
187 bool have_delayed_delete_refs;
190 static inline int extent_is_shared(struct share_check *sc)
192 return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
195 static struct kmem_cache *btrfs_prelim_ref_cache;
197 int __init btrfs_prelim_ref_init(void)
199 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
200 sizeof(struct prelim_ref),
204 if (!btrfs_prelim_ref_cache)
209 void __cold btrfs_prelim_ref_exit(void)
211 kmem_cache_destroy(btrfs_prelim_ref_cache);
214 static void free_pref(struct prelim_ref *ref)
216 kmem_cache_free(btrfs_prelim_ref_cache, ref);
220 * Return 0 when both refs are for the same block (and can be merged).
221 * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
222 * indicates a 'higher' block.
224 static int prelim_ref_compare(struct prelim_ref *ref1,
225 struct prelim_ref *ref2)
227 if (ref1->level < ref2->level)
229 if (ref1->level > ref2->level)
231 if (ref1->root_id < ref2->root_id)
233 if (ref1->root_id > ref2->root_id)
235 if (ref1->key_for_search.type < ref2->key_for_search.type)
237 if (ref1->key_for_search.type > ref2->key_for_search.type)
239 if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
241 if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
243 if (ref1->key_for_search.offset < ref2->key_for_search.offset)
245 if (ref1->key_for_search.offset > ref2->key_for_search.offset)
247 if (ref1->parent < ref2->parent)
249 if (ref1->parent > ref2->parent)
255 static void update_share_count(struct share_check *sc, int oldcount,
256 int newcount, struct prelim_ref *newref)
258 if ((!sc) || (oldcount == 0 && newcount < 1))
261 if (oldcount > 0 && newcount < 1)
263 else if (oldcount < 1 && newcount > 0)
266 if (newref->root_id == sc->root->root_key.objectid &&
267 newref->wanted_disk_byte == sc->data_bytenr &&
268 newref->key_for_search.objectid == sc->inum)
269 sc->self_ref_count += newref->count;
273 * Add @newref to the @root rbtree, merging identical refs.
275 * Callers should assume that newref has been freed after calling.
277 static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
278 struct preftree *preftree,
279 struct prelim_ref *newref,
280 struct share_check *sc)
282 struct rb_root_cached *root;
284 struct rb_node *parent = NULL;
285 struct prelim_ref *ref;
287 bool leftmost = true;
289 root = &preftree->root;
290 p = &root->rb_root.rb_node;
294 ref = rb_entry(parent, struct prelim_ref, rbnode);
295 result = prelim_ref_compare(ref, newref);
298 } else if (result > 0) {
302 /* Identical refs, merge them and free @newref */
303 struct extent_inode_elem *eie = ref->inode_list;
305 while (eie && eie->next)
309 ref->inode_list = newref->inode_list;
311 eie->next = newref->inode_list;
312 trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
315 * A delayed ref can have newref->count < 0.
316 * The ref->count is updated to follow any
317 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
319 update_share_count(sc, ref->count,
320 ref->count + newref->count, newref);
321 ref->count += newref->count;
327 update_share_count(sc, 0, newref->count, newref);
329 trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
330 rb_link_node(&newref->rbnode, parent, p);
331 rb_insert_color_cached(&newref->rbnode, root, leftmost);
335 * Release the entire tree. We don't care about internal consistency so
336 * just free everything and then reset the tree root.
338 static void prelim_release(struct preftree *preftree)
340 struct prelim_ref *ref, *next_ref;
342 rbtree_postorder_for_each_entry_safe(ref, next_ref,
343 &preftree->root.rb_root, rbnode) {
344 free_inode_elem_list(ref->inode_list);
348 preftree->root = RB_ROOT_CACHED;
353 * the rules for all callers of this function are:
354 * - obtaining the parent is the goal
355 * - if you add a key, you must know that it is a correct key
356 * - if you cannot add the parent or a correct key, then we will look into the
357 * block later to set a correct key
361 * backref type | shared | indirect | shared | indirect
362 * information | tree | tree | data | data
363 * --------------------+--------+----------+--------+----------
364 * parent logical | y | - | - | -
365 * key to resolve | - | y | y | y
366 * tree block logical | - | - | - | -
367 * root for resolving | y | y | y | y
369 * - column 1: we've the parent -> done
370 * - column 2, 3, 4: we use the key to find the parent
372 * on disk refs (inline or keyed)
373 * ==============================
374 * backref type | shared | indirect | shared | indirect
375 * information | tree | tree | data | data
376 * --------------------+--------+----------+--------+----------
377 * parent logical | y | - | y | -
378 * key to resolve | - | - | - | y
379 * tree block logical | y | y | y | y
380 * root for resolving | - | y | y | y
382 * - column 1, 3: we've the parent -> done
383 * - column 2: we take the first key from the block to find the parent
384 * (see add_missing_keys)
385 * - column 4: we use the key to find the parent
387 * additional information that's available but not required to find the parent
388 * block might help in merging entries to gain some speed.
390 static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
391 struct preftree *preftree, u64 root_id,
392 const struct btrfs_key *key, int level, u64 parent,
393 u64 wanted_disk_byte, int count,
394 struct share_check *sc, gfp_t gfp_mask)
396 struct prelim_ref *ref;
398 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
401 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
405 ref->root_id = root_id;
407 ref->key_for_search = *key;
409 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
411 ref->inode_list = NULL;
414 ref->parent = parent;
415 ref->wanted_disk_byte = wanted_disk_byte;
416 prelim_ref_insert(fs_info, preftree, ref, sc);
417 return extent_is_shared(sc);
420 /* direct refs use root == 0, key == NULL */
421 static int add_direct_ref(const struct btrfs_fs_info *fs_info,
422 struct preftrees *preftrees, int level, u64 parent,
423 u64 wanted_disk_byte, int count,
424 struct share_check *sc, gfp_t gfp_mask)
426 return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
427 parent, wanted_disk_byte, count, sc, gfp_mask);
430 /* indirect refs use parent == 0 */
431 static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
432 struct preftrees *preftrees, u64 root_id,
433 const struct btrfs_key *key, int level,
434 u64 wanted_disk_byte, int count,
435 struct share_check *sc, gfp_t gfp_mask)
437 struct preftree *tree = &preftrees->indirect;
440 tree = &preftrees->indirect_missing_keys;
441 return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
442 wanted_disk_byte, count, sc, gfp_mask);
445 static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
447 struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
448 struct rb_node *parent = NULL;
449 struct prelim_ref *ref = NULL;
450 struct prelim_ref target = {};
453 target.parent = bytenr;
457 ref = rb_entry(parent, struct prelim_ref, rbnode);
458 result = prelim_ref_compare(ref, &target);
470 static int add_all_parents(struct btrfs_backref_walk_ctx *ctx,
471 struct btrfs_root *root, struct btrfs_path *path,
472 struct ulist *parents,
473 struct preftrees *preftrees, struct prelim_ref *ref,
478 struct extent_buffer *eb;
479 struct btrfs_key key;
480 struct btrfs_key *key_for_search = &ref->key_for_search;
481 struct btrfs_file_extent_item *fi;
482 struct extent_inode_elem *eie = NULL, *old = NULL;
484 u64 wanted_disk_byte = ref->wanted_disk_byte;
489 eb = path->nodes[level];
490 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
497 * 1. We normally enter this function with the path already pointing to
498 * the first item to check. But sometimes, we may enter it with
500 * 2. We are searching for normal backref but bytenr of this leaf
501 * matches shared data backref
502 * 3. The leaf owner is not equal to the root we are searching
504 * For these cases, go to the next leaf before we continue.
507 if (path->slots[0] >= btrfs_header_nritems(eb) ||
508 is_shared_data_backref(preftrees, eb->start) ||
509 ref->root_id != btrfs_header_owner(eb)) {
510 if (ctx->time_seq == BTRFS_SEQ_LAST)
511 ret = btrfs_next_leaf(root, path);
513 ret = btrfs_next_old_leaf(root, path, ctx->time_seq);
516 while (!ret && count < ref->count) {
518 slot = path->slots[0];
520 btrfs_item_key_to_cpu(eb, &key, slot);
522 if (key.objectid != key_for_search->objectid ||
523 key.type != BTRFS_EXTENT_DATA_KEY)
527 * We are searching for normal backref but bytenr of this leaf
528 * matches shared data backref, OR
529 * the leaf owner is not equal to the root we are searching for
532 (is_shared_data_backref(preftrees, eb->start) ||
533 ref->root_id != btrfs_header_owner(eb))) {
534 if (ctx->time_seq == BTRFS_SEQ_LAST)
535 ret = btrfs_next_leaf(root, path);
537 ret = btrfs_next_old_leaf(root, path, ctx->time_seq);
540 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
541 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
542 data_offset = btrfs_file_extent_offset(eb, fi);
544 if (disk_byte == wanted_disk_byte) {
547 if (ref->key_for_search.offset == key.offset - data_offset)
551 if (!ctx->ignore_extent_item_pos) {
552 ret = check_extent_in_eb(ctx, &key, eb, fi, &eie);
553 if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
559 ret = ulist_add_merge_ptr(parents, eb->start,
560 eie, (void **)&old, GFP_NOFS);
563 if (!ret && !ctx->ignore_extent_item_pos) {
571 if (ctx->time_seq == BTRFS_SEQ_LAST)
572 ret = btrfs_next_item(root, path);
574 ret = btrfs_next_old_item(root, path, ctx->time_seq);
577 if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
578 free_inode_elem_list(eie);
586 * resolve an indirect backref in the form (root_id, key, level)
587 * to a logical address
589 static int resolve_indirect_ref(struct btrfs_backref_walk_ctx *ctx,
590 struct btrfs_path *path,
591 struct preftrees *preftrees,
592 struct prelim_ref *ref, struct ulist *parents)
594 struct btrfs_root *root;
595 struct extent_buffer *eb;
598 int level = ref->level;
599 struct btrfs_key search_key = ref->key_for_search;
602 * If we're search_commit_root we could possibly be holding locks on
603 * other tree nodes. This happens when qgroups does backref walks when
604 * adding new delayed refs. To deal with this we need to look in cache
605 * for the root, and if we don't find it then we need to search the
606 * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage
609 if (path->search_commit_root)
610 root = btrfs_get_fs_root_commit_root(ctx->fs_info, path, ref->root_id);
612 root = btrfs_get_fs_root(ctx->fs_info, ref->root_id, false);
618 if (!path->search_commit_root &&
619 test_bit(BTRFS_ROOT_DELETING, &root->state)) {
624 if (btrfs_is_testing(ctx->fs_info)) {
629 if (path->search_commit_root)
630 root_level = btrfs_header_level(root->commit_root);
631 else if (ctx->time_seq == BTRFS_SEQ_LAST)
632 root_level = btrfs_header_level(root->node);
634 root_level = btrfs_old_root_level(root, ctx->time_seq);
636 if (root_level + 1 == level)
640 * We can often find data backrefs with an offset that is too large
641 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
642 * subtracting a file's offset with the data offset of its
643 * corresponding extent data item. This can happen for example in the
646 * So if we detect such case we set the search key's offset to zero to
647 * make sure we will find the matching file extent item at
648 * add_all_parents(), otherwise we will miss it because the offset
649 * taken form the backref is much larger then the offset of the file
650 * extent item. This can make us scan a very large number of file
651 * extent items, but at least it will not make us miss any.
653 * This is an ugly workaround for a behaviour that should have never
654 * existed, but it does and a fix for the clone ioctl would touch a lot
655 * of places, cause backwards incompatibility and would not fix the
656 * problem for extents cloned with older kernels.
658 if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
659 search_key.offset >= LLONG_MAX)
660 search_key.offset = 0;
661 path->lowest_level = level;
662 if (ctx->time_seq == BTRFS_SEQ_LAST)
663 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
665 ret = btrfs_search_old_slot(root, &search_key, path, ctx->time_seq);
667 btrfs_debug(ctx->fs_info,
668 "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
669 ref->root_id, level, ref->count, ret,
670 ref->key_for_search.objectid, ref->key_for_search.type,
671 ref->key_for_search.offset);
675 eb = path->nodes[level];
677 if (WARN_ON(!level)) {
682 eb = path->nodes[level];
685 ret = add_all_parents(ctx, root, path, parents, preftrees, ref, level);
687 btrfs_put_root(root);
689 path->lowest_level = 0;
690 btrfs_release_path(path);
694 static struct extent_inode_elem *
695 unode_aux_to_inode_list(struct ulist_node *node)
699 return (struct extent_inode_elem *)(uintptr_t)node->aux;
702 static void free_leaf_list(struct ulist *ulist)
704 struct ulist_node *node;
705 struct ulist_iterator uiter;
707 ULIST_ITER_INIT(&uiter);
708 while ((node = ulist_next(ulist, &uiter)))
709 free_inode_elem_list(unode_aux_to_inode_list(node));
715 * We maintain three separate rbtrees: one for direct refs, one for
716 * indirect refs which have a key, and one for indirect refs which do not
717 * have a key. Each tree does merge on insertion.
719 * Once all of the references are located, we iterate over the tree of
720 * indirect refs with missing keys. An appropriate key is located and
721 * the ref is moved onto the tree for indirect refs. After all missing
722 * keys are thus located, we iterate over the indirect ref tree, resolve
723 * each reference, and then insert the resolved reference onto the
724 * direct tree (merging there too).
726 * New backrefs (i.e., for parent nodes) are added to the appropriate
727 * rbtree as they are encountered. The new backrefs are subsequently
730 static int resolve_indirect_refs(struct btrfs_backref_walk_ctx *ctx,
731 struct btrfs_path *path,
732 struct preftrees *preftrees,
733 struct share_check *sc)
737 struct ulist *parents;
738 struct ulist_node *node;
739 struct ulist_iterator uiter;
740 struct rb_node *rnode;
742 parents = ulist_alloc(GFP_NOFS);
747 * We could trade memory usage for performance here by iterating
748 * the tree, allocating new refs for each insertion, and then
749 * freeing the entire indirect tree when we're done. In some test
750 * cases, the tree can grow quite large (~200k objects).
752 while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
753 struct prelim_ref *ref;
755 ref = rb_entry(rnode, struct prelim_ref, rbnode);
756 if (WARN(ref->parent,
757 "BUG: direct ref found in indirect tree")) {
762 rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
763 preftrees->indirect.count--;
765 if (ref->count == 0) {
770 if (sc && ref->root_id != sc->root->root_key.objectid) {
772 ret = BACKREF_FOUND_SHARED;
775 err = resolve_indirect_ref(ctx, path, preftrees, ref, parents);
777 * we can only tolerate ENOENT,otherwise,we should catch error
778 * and return directly.
780 if (err == -ENOENT) {
781 prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref,
790 /* we put the first parent into the ref at hand */
791 ULIST_ITER_INIT(&uiter);
792 node = ulist_next(parents, &uiter);
793 ref->parent = node ? node->val : 0;
794 ref->inode_list = unode_aux_to_inode_list(node);
796 /* Add a prelim_ref(s) for any other parent(s). */
797 while ((node = ulist_next(parents, &uiter))) {
798 struct prelim_ref *new_ref;
800 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
807 memcpy(new_ref, ref, sizeof(*ref));
808 new_ref->parent = node->val;
809 new_ref->inode_list = unode_aux_to_inode_list(node);
810 prelim_ref_insert(ctx->fs_info, &preftrees->direct,
815 * Now it's a direct ref, put it in the direct tree. We must
816 * do this last because the ref could be merged/freed here.
818 prelim_ref_insert(ctx->fs_info, &preftrees->direct, ref, NULL);
820 ulist_reinit(parents);
825 * We may have inode lists attached to refs in the parents ulist, so we
826 * must free them before freeing the ulist and its refs.
828 free_leaf_list(parents);
833 * read tree blocks and add keys where required.
835 static int add_missing_keys(struct btrfs_fs_info *fs_info,
836 struct preftrees *preftrees, bool lock)
838 struct prelim_ref *ref;
839 struct extent_buffer *eb;
840 struct preftree *tree = &preftrees->indirect_missing_keys;
841 struct rb_node *node;
843 while ((node = rb_first_cached(&tree->root))) {
844 struct btrfs_tree_parent_check check = { 0 };
846 ref = rb_entry(node, struct prelim_ref, rbnode);
847 rb_erase_cached(node, &tree->root);
849 BUG_ON(ref->parent); /* should not be a direct ref */
850 BUG_ON(ref->key_for_search.type);
851 BUG_ON(!ref->wanted_disk_byte);
853 check.level = ref->level - 1;
854 check.owner_root = ref->root_id;
856 eb = read_tree_block(fs_info, ref->wanted_disk_byte, &check);
861 if (!extent_buffer_uptodate(eb)) {
863 free_extent_buffer(eb);
868 btrfs_tree_read_lock(eb);
869 if (btrfs_header_level(eb) == 0)
870 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
872 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
874 btrfs_tree_read_unlock(eb);
875 free_extent_buffer(eb);
876 prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
883 * add all currently queued delayed refs from this head whose seq nr is
884 * smaller or equal that seq to the list
886 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
887 struct btrfs_delayed_ref_head *head, u64 seq,
888 struct preftrees *preftrees, struct share_check *sc)
890 struct btrfs_delayed_ref_node *node;
891 struct btrfs_key key;
896 spin_lock(&head->lock);
897 for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
898 node = rb_entry(n, struct btrfs_delayed_ref_node,
903 switch (node->action) {
904 case BTRFS_ADD_DELAYED_EXTENT:
905 case BTRFS_UPDATE_DELAYED_HEAD:
908 case BTRFS_ADD_DELAYED_REF:
909 count = node->ref_mod;
911 case BTRFS_DROP_DELAYED_REF:
912 count = node->ref_mod * -1;
917 switch (node->type) {
918 case BTRFS_TREE_BLOCK_REF_KEY: {
919 /* NORMAL INDIRECT METADATA backref */
920 struct btrfs_delayed_tree_ref *ref;
921 struct btrfs_key *key_ptr = NULL;
923 if (head->extent_op && head->extent_op->update_key) {
924 btrfs_disk_key_to_cpu(&key, &head->extent_op->key);
928 ref = btrfs_delayed_node_to_tree_ref(node);
929 ret = add_indirect_ref(fs_info, preftrees, ref->root,
930 key_ptr, ref->level + 1,
931 node->bytenr, count, sc,
935 case BTRFS_SHARED_BLOCK_REF_KEY: {
936 /* SHARED DIRECT METADATA backref */
937 struct btrfs_delayed_tree_ref *ref;
939 ref = btrfs_delayed_node_to_tree_ref(node);
941 ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
942 ref->parent, node->bytenr, count,
946 case BTRFS_EXTENT_DATA_REF_KEY: {
947 /* NORMAL INDIRECT DATA backref */
948 struct btrfs_delayed_data_ref *ref;
949 ref = btrfs_delayed_node_to_data_ref(node);
951 key.objectid = ref->objectid;
952 key.type = BTRFS_EXTENT_DATA_KEY;
953 key.offset = ref->offset;
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.
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.
971 sc->have_delayed_delete_refs = true;
973 ret = add_indirect_ref(fs_info, preftrees, ref->root,
974 &key, 0, node->bytenr, count, sc,
978 case BTRFS_SHARED_DATA_REF_KEY: {
979 /* SHARED DIRECT FULL backref */
980 struct btrfs_delayed_data_ref *ref;
982 ref = btrfs_delayed_node_to_data_ref(node);
984 ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
985 node->bytenr, count, sc,
993 * We must ignore BACKREF_FOUND_SHARED until all delayed
994 * refs have been checked.
996 if (ret && (ret != BACKREF_FOUND_SHARED))
1000 ret = extent_is_shared(sc);
1002 spin_unlock(&head->lock);
1007 * add all inline backrefs for bytenr to the list
1009 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1011 static int add_inline_refs(struct btrfs_backref_walk_ctx *ctx,
1012 struct btrfs_path *path,
1013 int *info_level, struct preftrees *preftrees,
1014 struct share_check *sc)
1018 struct extent_buffer *leaf;
1019 struct btrfs_key key;
1020 struct btrfs_key found_key;
1023 struct btrfs_extent_item *ei;
1028 * enumerate all inline refs
1030 leaf = path->nodes[0];
1031 slot = path->slots[0];
1033 item_size = btrfs_item_size(leaf, slot);
1034 BUG_ON(item_size < sizeof(*ei));
1036 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
1038 if (ctx->check_extent_item) {
1039 ret = ctx->check_extent_item(ctx->bytenr, ei, leaf, ctx->user_ctx);
1044 flags = btrfs_extent_flags(leaf, ei);
1045 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1047 ptr = (unsigned long)(ei + 1);
1048 end = (unsigned long)ei + item_size;
1050 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
1051 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1052 struct btrfs_tree_block_info *info;
1054 info = (struct btrfs_tree_block_info *)ptr;
1055 *info_level = btrfs_tree_block_level(leaf, info);
1056 ptr += sizeof(struct btrfs_tree_block_info);
1058 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
1059 *info_level = found_key.offset;
1061 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1065 struct btrfs_extent_inline_ref *iref;
1069 iref = (struct btrfs_extent_inline_ref *)ptr;
1070 type = btrfs_get_extent_inline_ref_type(leaf, iref,
1071 BTRFS_REF_TYPE_ANY);
1072 if (type == BTRFS_REF_TYPE_INVALID)
1075 offset = btrfs_extent_inline_ref_offset(leaf, iref);
1078 case BTRFS_SHARED_BLOCK_REF_KEY:
1079 ret = add_direct_ref(ctx->fs_info, preftrees,
1080 *info_level + 1, offset,
1081 ctx->bytenr, 1, NULL, GFP_NOFS);
1083 case BTRFS_SHARED_DATA_REF_KEY: {
1084 struct btrfs_shared_data_ref *sdref;
1087 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
1088 count = btrfs_shared_data_ref_count(leaf, sdref);
1090 ret = add_direct_ref(ctx->fs_info, preftrees, 0, offset,
1091 ctx->bytenr, count, sc, GFP_NOFS);
1094 case BTRFS_TREE_BLOCK_REF_KEY:
1095 ret = add_indirect_ref(ctx->fs_info, preftrees, offset,
1096 NULL, *info_level + 1,
1097 ctx->bytenr, 1, NULL, GFP_NOFS);
1099 case BTRFS_EXTENT_DATA_REF_KEY: {
1100 struct btrfs_extent_data_ref *dref;
1104 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1105 count = btrfs_extent_data_ref_count(leaf, dref);
1106 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1108 key.type = BTRFS_EXTENT_DATA_KEY;
1109 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1111 if (sc && key.objectid != sc->inum &&
1112 !sc->have_delayed_delete_refs) {
1113 ret = BACKREF_FOUND_SHARED;
1117 root = btrfs_extent_data_ref_root(leaf, dref);
1119 if (!ctx->skip_data_ref ||
1120 !ctx->skip_data_ref(root, key.objectid, key.offset,
1122 ret = add_indirect_ref(ctx->fs_info, preftrees,
1123 root, &key, 0, ctx->bytenr,
1124 count, sc, GFP_NOFS);
1132 ptr += btrfs_extent_inline_ref_size(type);
1139 * add all non-inline backrefs for bytenr to the list
1141 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1143 static int add_keyed_refs(struct btrfs_backref_walk_ctx *ctx,
1144 struct btrfs_root *extent_root,
1145 struct btrfs_path *path,
1146 int info_level, struct preftrees *preftrees,
1147 struct share_check *sc)
1149 struct btrfs_fs_info *fs_info = extent_root->fs_info;
1152 struct extent_buffer *leaf;
1153 struct btrfs_key key;
1156 ret = btrfs_next_item(extent_root, path);
1164 slot = path->slots[0];
1165 leaf = path->nodes[0];
1166 btrfs_item_key_to_cpu(leaf, &key, slot);
1168 if (key.objectid != ctx->bytenr)
1170 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1172 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1176 case BTRFS_SHARED_BLOCK_REF_KEY:
1177 /* SHARED DIRECT METADATA backref */
1178 ret = add_direct_ref(fs_info, preftrees,
1179 info_level + 1, key.offset,
1180 ctx->bytenr, 1, NULL, GFP_NOFS);
1182 case BTRFS_SHARED_DATA_REF_KEY: {
1183 /* SHARED DIRECT FULL backref */
1184 struct btrfs_shared_data_ref *sdref;
1187 sdref = btrfs_item_ptr(leaf, slot,
1188 struct btrfs_shared_data_ref);
1189 count = btrfs_shared_data_ref_count(leaf, sdref);
1190 ret = add_direct_ref(fs_info, preftrees, 0,
1191 key.offset, ctx->bytenr, count,
1195 case BTRFS_TREE_BLOCK_REF_KEY:
1196 /* NORMAL INDIRECT METADATA backref */
1197 ret = add_indirect_ref(fs_info, preftrees, key.offset,
1198 NULL, info_level + 1, ctx->bytenr,
1201 case BTRFS_EXTENT_DATA_REF_KEY: {
1202 /* NORMAL INDIRECT DATA backref */
1203 struct btrfs_extent_data_ref *dref;
1207 dref = btrfs_item_ptr(leaf, slot,
1208 struct btrfs_extent_data_ref);
1209 count = btrfs_extent_data_ref_count(leaf, dref);
1210 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1212 key.type = BTRFS_EXTENT_DATA_KEY;
1213 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1215 if (sc && key.objectid != sc->inum &&
1216 !sc->have_delayed_delete_refs) {
1217 ret = BACKREF_FOUND_SHARED;
1221 root = btrfs_extent_data_ref_root(leaf, dref);
1223 if (!ctx->skip_data_ref ||
1224 !ctx->skip_data_ref(root, key.objectid, key.offset,
1226 ret = add_indirect_ref(fs_info, preftrees, root,
1227 &key, 0, ctx->bytenr,
1228 count, sc, GFP_NOFS);
1243 * The caller has joined a transaction or is holding a read lock on the
1244 * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1245 * snapshot field changing while updating or checking the cache.
1247 static bool lookup_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx,
1248 struct btrfs_root *root,
1249 u64 bytenr, int level, bool *is_shared)
1251 struct btrfs_backref_shared_cache_entry *entry;
1253 if (!ctx->use_path_cache)
1256 if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1260 * Level -1 is used for the data extent, which is not reliable to cache
1261 * because its reference count can increase or decrease without us
1262 * realizing. We cache results only for extent buffers that lead from
1263 * the root node down to the leaf with the file extent item.
1267 entry = &ctx->path_cache_entries[level];
1269 /* Unused cache entry or being used for some other extent buffer. */
1270 if (entry->bytenr != bytenr)
1274 * We cached a false result, but the last snapshot generation of the
1275 * root changed, so we now have a snapshot. Don't trust the result.
1277 if (!entry->is_shared &&
1278 entry->gen != btrfs_root_last_snapshot(&root->root_item))
1282 * If we cached a true result and the last generation used for dropping
1283 * a root changed, we can not trust the result, because the dropped root
1284 * could be a snapshot sharing this extent buffer.
1286 if (entry->is_shared &&
1287 entry->gen != btrfs_get_last_root_drop_gen(root->fs_info))
1290 *is_shared = entry->is_shared;
1292 * If the node at this level is shared, than all nodes below are also
1293 * shared. Currently some of the nodes below may be marked as not shared
1294 * because we have just switched from one leaf to another, and switched
1295 * also other nodes above the leaf and below the current level, so mark
1299 for (int i = 0; i < level; i++) {
1300 ctx->path_cache_entries[i].is_shared = true;
1301 ctx->path_cache_entries[i].gen = entry->gen;
1309 * The caller has joined a transaction or is holding a read lock on the
1310 * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1311 * snapshot field changing while updating or checking the cache.
1313 static void store_backref_shared_cache(struct btrfs_backref_share_check_ctx *ctx,
1314 struct btrfs_root *root,
1315 u64 bytenr, int level, bool is_shared)
1317 struct btrfs_backref_shared_cache_entry *entry;
1320 if (!ctx->use_path_cache)
1323 if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1327 * Level -1 is used for the data extent, which is not reliable to cache
1328 * because its reference count can increase or decrease without us
1329 * realizing. We cache results only for extent buffers that lead from
1330 * the root node down to the leaf with the file extent item.
1335 gen = btrfs_get_last_root_drop_gen(root->fs_info);
1337 gen = btrfs_root_last_snapshot(&root->root_item);
1339 entry = &ctx->path_cache_entries[level];
1340 entry->bytenr = bytenr;
1341 entry->is_shared = is_shared;
1345 * If we found an extent buffer is shared, set the cache result for all
1346 * extent buffers below it to true. As nodes in the path are COWed,
1347 * their sharedness is moved to their children, and if a leaf is COWed,
1348 * then the sharedness of a data extent becomes direct, the refcount of
1349 * data extent is increased in the extent item at the extent tree.
1352 for (int i = 0; i < level; i++) {
1353 entry = &ctx->path_cache_entries[i];
1354 entry->is_shared = is_shared;
1361 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1362 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1363 * indirect refs to their parent bytenr.
1364 * When roots are found, they're added to the roots list
1366 * @ctx: Backref walking context object, must be not NULL.
1367 * @sc: If !NULL, then immediately return BACKREF_FOUND_SHARED when a
1368 * shared extent is detected.
1370 * Otherwise this returns 0 for success and <0 for an error.
1372 * FIXME some caching might speed things up
1374 static int find_parent_nodes(struct btrfs_backref_walk_ctx *ctx,
1375 struct share_check *sc)
1377 struct btrfs_root *root = btrfs_extent_root(ctx->fs_info, ctx->bytenr);
1378 struct btrfs_key key;
1379 struct btrfs_path *path;
1380 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1381 struct btrfs_delayed_ref_head *head;
1384 struct prelim_ref *ref;
1385 struct rb_node *node;
1386 struct extent_inode_elem *eie = NULL;
1387 struct preftrees preftrees = {
1388 .direct = PREFTREE_INIT,
1389 .indirect = PREFTREE_INIT,
1390 .indirect_missing_keys = PREFTREE_INIT
1393 /* Roots ulist is not needed when using a sharedness check context. */
1395 ASSERT(ctx->roots == NULL);
1397 key.objectid = ctx->bytenr;
1398 key.offset = (u64)-1;
1399 if (btrfs_fs_incompat(ctx->fs_info, SKINNY_METADATA))
1400 key.type = BTRFS_METADATA_ITEM_KEY;
1402 key.type = BTRFS_EXTENT_ITEM_KEY;
1404 path = btrfs_alloc_path();
1408 path->search_commit_root = 1;
1409 path->skip_locking = 1;
1412 if (ctx->time_seq == BTRFS_SEQ_LAST)
1413 path->skip_locking = 1;
1418 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1422 /* This shouldn't happen, indicates a bug or fs corruption. */
1428 if (ctx->trans && likely(ctx->trans->type != __TRANS_DUMMY) &&
1429 ctx->time_seq != BTRFS_SEQ_LAST) {
1431 * We have a specific time_seq we care about and trans which
1432 * means we have the path lock, we need to grab the ref head and
1433 * lock it so we have a consistent view of the refs at the given
1436 delayed_refs = &ctx->trans->transaction->delayed_refs;
1437 spin_lock(&delayed_refs->lock);
1438 head = btrfs_find_delayed_ref_head(delayed_refs, ctx->bytenr);
1440 if (!mutex_trylock(&head->mutex)) {
1441 refcount_inc(&head->refs);
1442 spin_unlock(&delayed_refs->lock);
1444 btrfs_release_path(path);
1447 * Mutex was contended, block until it's
1448 * released and try again
1450 mutex_lock(&head->mutex);
1451 mutex_unlock(&head->mutex);
1452 btrfs_put_delayed_ref_head(head);
1455 spin_unlock(&delayed_refs->lock);
1456 ret = add_delayed_refs(ctx->fs_info, head, ctx->time_seq,
1458 mutex_unlock(&head->mutex);
1462 spin_unlock(&delayed_refs->lock);
1466 if (path->slots[0]) {
1467 struct extent_buffer *leaf;
1471 leaf = path->nodes[0];
1472 slot = path->slots[0];
1473 btrfs_item_key_to_cpu(leaf, &key, slot);
1474 if (key.objectid == ctx->bytenr &&
1475 (key.type == BTRFS_EXTENT_ITEM_KEY ||
1476 key.type == BTRFS_METADATA_ITEM_KEY)) {
1477 ret = add_inline_refs(ctx, path, &info_level,
1481 ret = add_keyed_refs(ctx, root, path, info_level,
1489 * If we have a share context and we reached here, it means the extent
1490 * is not directly shared (no multiple reference items for it),
1491 * otherwise we would have exited earlier with a return value of
1492 * BACKREF_FOUND_SHARED after processing delayed references or while
1493 * processing inline or keyed references from the extent tree.
1494 * The extent may however be indirectly shared through shared subtrees
1495 * as a result from creating snapshots, so we determine below what is
1496 * its parent node, in case we are dealing with a metadata extent, or
1497 * what's the leaf (or leaves), from a fs tree, that has a file extent
1498 * item pointing to it in case we are dealing with a data extent.
1500 ASSERT(extent_is_shared(sc) == 0);
1503 * If we are here for a data extent and we have a share_check structure
1504 * it means the data extent is not directly shared (does not have
1505 * multiple reference items), so we have to check if a path in the fs
1506 * tree (going from the root node down to the leaf that has the file
1507 * extent item pointing to the data extent) is shared, that is, if any
1508 * of the extent buffers in the path is referenced by other trees.
1510 if (sc && ctx->bytenr == sc->data_bytenr) {
1512 * If our data extent is from a generation more recent than the
1513 * last generation used to snapshot the root, then we know that
1514 * it can not be shared through subtrees, so we can skip
1515 * resolving indirect references, there's no point in
1516 * determining the extent buffers for the path from the fs tree
1517 * root node down to the leaf that has the file extent item that
1518 * points to the data extent.
1520 if (sc->data_extent_gen >
1521 btrfs_root_last_snapshot(&sc->root->root_item)) {
1522 ret = BACKREF_FOUND_NOT_SHARED;
1527 * If we are only determining if a data extent is shared or not
1528 * and the corresponding file extent item is located in the same
1529 * leaf as the previous file extent item, we can skip resolving
1530 * indirect references for a data extent, since the fs tree path
1531 * is the same (same leaf, so same path). We skip as long as the
1532 * cached result for the leaf is valid and only if there's only
1533 * one file extent item pointing to the data extent, because in
1534 * the case of multiple file extent items, they may be located
1535 * in different leaves and therefore we have multiple paths.
1537 if (sc->ctx->curr_leaf_bytenr == sc->ctx->prev_leaf_bytenr &&
1538 sc->self_ref_count == 1) {
1542 cached = lookup_backref_shared_cache(sc->ctx, sc->root,
1543 sc->ctx->curr_leaf_bytenr,
1547 ret = BACKREF_FOUND_SHARED;
1549 ret = BACKREF_FOUND_NOT_SHARED;
1555 btrfs_release_path(path);
1557 ret = add_missing_keys(ctx->fs_info, &preftrees, path->skip_locking == 0);
1561 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1563 ret = resolve_indirect_refs(ctx, path, &preftrees, sc);
1567 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1570 * This walks the tree of merged and resolved refs. Tree blocks are
1571 * read in as needed. Unique entries are added to the ulist, and
1572 * the list of found roots is updated.
1574 * We release the entire tree in one go before returning.
1576 node = rb_first_cached(&preftrees.direct.root);
1578 ref = rb_entry(node, struct prelim_ref, rbnode);
1579 node = rb_next(&ref->rbnode);
1581 * ref->count < 0 can happen here if there are delayed
1582 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1583 * prelim_ref_insert() relies on this when merging
1584 * identical refs to keep the overall count correct.
1585 * prelim_ref_insert() will merge only those refs
1586 * which compare identically. Any refs having
1587 * e.g. different offsets would not be merged,
1588 * and would retain their original ref->count < 0.
1590 if (ctx->roots && ref->count && ref->root_id && ref->parent == 0) {
1591 /* no parent == root of tree */
1592 ret = ulist_add(ctx->roots, ref->root_id, 0, GFP_NOFS);
1596 if (ref->count && ref->parent) {
1597 if (!ctx->ignore_extent_item_pos && !ref->inode_list &&
1599 struct btrfs_tree_parent_check check = { 0 };
1600 struct extent_buffer *eb;
1602 check.level = ref->level;
1604 eb = read_tree_block(ctx->fs_info, ref->parent,
1610 if (!extent_buffer_uptodate(eb)) {
1611 free_extent_buffer(eb);
1616 if (!path->skip_locking)
1617 btrfs_tree_read_lock(eb);
1618 ret = find_extent_in_eb(ctx, eb, &eie);
1619 if (!path->skip_locking)
1620 btrfs_tree_read_unlock(eb);
1621 free_extent_buffer(eb);
1622 if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
1625 ref->inode_list = eie;
1627 * We transferred the list ownership to the ref,
1628 * so set to NULL to avoid a double free in case
1629 * an error happens after this.
1633 ret = ulist_add_merge_ptr(ctx->refs, ref->parent,
1635 (void **)&eie, GFP_NOFS);
1638 if (!ret && !ctx->ignore_extent_item_pos) {
1640 * We've recorded that parent, so we must extend
1641 * its inode list here.
1643 * However if there was corruption we may not
1644 * have found an eie, return an error in this
1654 eie->next = ref->inode_list;
1658 * We have transferred the inode list ownership from
1659 * this ref to the ref we added to the 'refs' ulist.
1660 * So set this ref's inode list to NULL to avoid
1661 * use-after-free when our caller uses it or double
1662 * frees in case an error happens before we return.
1664 ref->inode_list = NULL;
1670 btrfs_free_path(path);
1672 prelim_release(&preftrees.direct);
1673 prelim_release(&preftrees.indirect);
1674 prelim_release(&preftrees.indirect_missing_keys);
1676 if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP || ret < 0)
1677 free_inode_elem_list(eie);
1682 * Finds all leaves with a reference to the specified combination of
1683 * @ctx->bytenr and @ctx->extent_item_pos. The bytenr of the found leaves are
1684 * added to the ulist at @ctx->refs, and that ulist is allocated by this
1685 * function. The caller should free the ulist with free_leaf_list() if
1686 * @ctx->ignore_extent_item_pos is false, otherwise a fimple ulist_free() is
1689 * Returns 0 on success and < 0 on error. On error @ctx->refs is not allocated.
1691 int btrfs_find_all_leafs(struct btrfs_backref_walk_ctx *ctx)
1695 ASSERT(ctx->refs == NULL);
1697 ctx->refs = ulist_alloc(GFP_NOFS);
1701 ret = find_parent_nodes(ctx, NULL);
1702 if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP ||
1703 (ret < 0 && ret != -ENOENT)) {
1704 free_leaf_list(ctx->refs);
1713 * Walk all backrefs for a given extent to find all roots that reference this
1714 * extent. Walking a backref means finding all extents that reference this
1715 * extent and in turn walk the backrefs of those, too. Naturally this is a
1716 * recursive process, but here it is implemented in an iterative fashion: We
1717 * find all referencing extents for the extent in question and put them on a
1718 * list. In turn, we find all referencing extents for those, further appending
1719 * to the list. The way we iterate the list allows adding more elements after
1720 * the current while iterating. The process stops when we reach the end of the
1723 * Found roots are added to @ctx->roots, which is allocated by this function if
1724 * it points to NULL, in which case the caller is responsible for freeing it
1725 * after it's not needed anymore.
1726 * This function requires @ctx->refs to be NULL, as it uses it for allocating a
1727 * ulist to do temporary work, and frees it before returning.
1729 * Returns 0 on success, < 0 on error.
1731 static int btrfs_find_all_roots_safe(struct btrfs_backref_walk_ctx *ctx)
1733 const u64 orig_bytenr = ctx->bytenr;
1734 const bool orig_ignore_extent_item_pos = ctx->ignore_extent_item_pos;
1735 bool roots_ulist_allocated = false;
1736 struct ulist_iterator uiter;
1739 ASSERT(ctx->refs == NULL);
1741 ctx->refs = ulist_alloc(GFP_NOFS);
1746 ctx->roots = ulist_alloc(GFP_NOFS);
1748 ulist_free(ctx->refs);
1752 roots_ulist_allocated = true;
1755 ctx->ignore_extent_item_pos = true;
1757 ULIST_ITER_INIT(&uiter);
1759 struct ulist_node *node;
1761 ret = find_parent_nodes(ctx, NULL);
1762 if (ret < 0 && ret != -ENOENT) {
1763 if (roots_ulist_allocated) {
1764 ulist_free(ctx->roots);
1770 node = ulist_next(ctx->refs, &uiter);
1773 ctx->bytenr = node->val;
1777 ulist_free(ctx->refs);
1779 ctx->bytenr = orig_bytenr;
1780 ctx->ignore_extent_item_pos = orig_ignore_extent_item_pos;
1785 int btrfs_find_all_roots(struct btrfs_backref_walk_ctx *ctx,
1786 bool skip_commit_root_sem)
1790 if (!ctx->trans && !skip_commit_root_sem)
1791 down_read(&ctx->fs_info->commit_root_sem);
1792 ret = btrfs_find_all_roots_safe(ctx);
1793 if (!ctx->trans && !skip_commit_root_sem)
1794 up_read(&ctx->fs_info->commit_root_sem);
1798 struct btrfs_backref_share_check_ctx *btrfs_alloc_backref_share_check_ctx(void)
1800 struct btrfs_backref_share_check_ctx *ctx;
1802 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1806 ulist_init(&ctx->refs);
1811 void btrfs_free_backref_share_ctx(struct btrfs_backref_share_check_ctx *ctx)
1816 ulist_release(&ctx->refs);
1821 * Check if a data extent is shared or not.
1823 * @inode: The inode whose extent we are checking.
1824 * @bytenr: Logical bytenr of the extent we are checking.
1825 * @extent_gen: Generation of the extent (file extent item) or 0 if it is
1827 * @ctx: A backref sharedness check context.
1829 * btrfs_is_data_extent_shared uses the backref walking code but will short
1830 * circuit as soon as it finds a root or inode that doesn't match the
1831 * one passed in. This provides a significant performance benefit for
1832 * callers (such as fiemap) which want to know whether the extent is
1833 * shared but do not need a ref count.
1835 * This attempts to attach to the running transaction in order to account for
1836 * delayed refs, but continues on even when no running transaction exists.
1838 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1840 int btrfs_is_data_extent_shared(struct btrfs_inode *inode, u64 bytenr,
1842 struct btrfs_backref_share_check_ctx *ctx)
1844 struct btrfs_backref_walk_ctx walk_ctx = { 0 };
1845 struct btrfs_root *root = inode->root;
1846 struct btrfs_fs_info *fs_info = root->fs_info;
1847 struct btrfs_trans_handle *trans;
1848 struct ulist_iterator uiter;
1849 struct ulist_node *node;
1850 struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem);
1852 struct share_check shared = {
1855 .inum = btrfs_ino(inode),
1856 .data_bytenr = bytenr,
1857 .data_extent_gen = extent_gen,
1859 .self_ref_count = 0,
1860 .have_delayed_delete_refs = false,
1864 for (int i = 0; i < BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE; i++) {
1865 if (ctx->prev_extents_cache[i].bytenr == bytenr)
1866 return ctx->prev_extents_cache[i].is_shared;
1869 ulist_init(&ctx->refs);
1871 trans = btrfs_join_transaction_nostart(root);
1872 if (IS_ERR(trans)) {
1873 if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1874 ret = PTR_ERR(trans);
1878 down_read(&fs_info->commit_root_sem);
1880 btrfs_get_tree_mod_seq(fs_info, &elem);
1881 walk_ctx.time_seq = elem.seq;
1884 walk_ctx.ignore_extent_item_pos = true;
1885 walk_ctx.trans = trans;
1886 walk_ctx.fs_info = fs_info;
1887 walk_ctx.refs = &ctx->refs;
1889 /* -1 means we are in the bytenr of the data extent. */
1891 ULIST_ITER_INIT(&uiter);
1892 ctx->use_path_cache = true;
1897 walk_ctx.bytenr = bytenr;
1898 ret = find_parent_nodes(&walk_ctx, &shared);
1899 if (ret == BACKREF_FOUND_SHARED ||
1900 ret == BACKREF_FOUND_NOT_SHARED) {
1901 /* If shared must return 1, otherwise return 0. */
1902 ret = (ret == BACKREF_FOUND_SHARED) ? 1 : 0;
1904 store_backref_shared_cache(ctx, root, bytenr,
1908 if (ret < 0 && ret != -ENOENT)
1913 * If our data extent was not directly shared (without multiple
1914 * reference items), than it might have a single reference item
1915 * with a count > 1 for the same offset, which means there are 2
1916 * (or more) file extent items that point to the data extent -
1917 * this happens when a file extent item needs to be split and
1918 * then one item gets moved to another leaf due to a b+tree leaf
1919 * split when inserting some item. In this case the file extent
1920 * items may be located in different leaves and therefore some
1921 * of the leaves may be referenced through shared subtrees while
1922 * others are not. Since our extent buffer cache only works for
1923 * a single path (by far the most common case and simpler to
1924 * deal with), we can not use it if we have multiple leaves
1925 * (which implies multiple paths).
1927 if (level == -1 && ctx->refs.nnodes > 1)
1928 ctx->use_path_cache = false;
1931 store_backref_shared_cache(ctx, root, bytenr,
1933 node = ulist_next(&ctx->refs, &uiter);
1938 cached = lookup_backref_shared_cache(ctx, root, bytenr, level,
1941 ret = (is_shared ? 1 : 0);
1944 shared.share_count = 0;
1945 shared.have_delayed_delete_refs = false;
1950 * Cache the sharedness result for the data extent if we know our inode
1951 * has more than 1 file extent item that refers to the data extent.
1953 if (ret >= 0 && shared.self_ref_count > 1) {
1954 int slot = ctx->prev_extents_cache_slot;
1956 ctx->prev_extents_cache[slot].bytenr = shared.data_bytenr;
1957 ctx->prev_extents_cache[slot].is_shared = (ret == 1);
1959 slot = (slot + 1) % BTRFS_BACKREF_CTX_PREV_EXTENTS_SIZE;
1960 ctx->prev_extents_cache_slot = slot;
1964 btrfs_put_tree_mod_seq(fs_info, &elem);
1965 btrfs_end_transaction(trans);
1967 up_read(&fs_info->commit_root_sem);
1970 ulist_release(&ctx->refs);
1971 ctx->prev_leaf_bytenr = ctx->curr_leaf_bytenr;
1976 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1977 u64 start_off, struct btrfs_path *path,
1978 struct btrfs_inode_extref **ret_extref,
1982 struct btrfs_key key;
1983 struct btrfs_key found_key;
1984 struct btrfs_inode_extref *extref;
1985 const struct extent_buffer *leaf;
1988 key.objectid = inode_objectid;
1989 key.type = BTRFS_INODE_EXTREF_KEY;
1990 key.offset = start_off;
1992 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1997 leaf = path->nodes[0];
1998 slot = path->slots[0];
1999 if (slot >= btrfs_header_nritems(leaf)) {
2001 * If the item at offset is not found,
2002 * btrfs_search_slot will point us to the slot
2003 * where it should be inserted. In our case
2004 * that will be the slot directly before the
2005 * next INODE_REF_KEY_V2 item. In the case
2006 * that we're pointing to the last slot in a
2007 * leaf, we must move one leaf over.
2009 ret = btrfs_next_leaf(root, path);
2018 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2021 * Check that we're still looking at an extended ref key for
2022 * this particular objectid. If we have different
2023 * objectid or type then there are no more to be found
2024 * in the tree and we can exit.
2027 if (found_key.objectid != inode_objectid)
2029 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
2033 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
2034 extref = (struct btrfs_inode_extref *)ptr;
2035 *ret_extref = extref;
2037 *found_off = found_key.offset;
2045 * this iterates to turn a name (from iref/extref) into a full filesystem path.
2046 * Elements of the path are separated by '/' and the path is guaranteed to be
2047 * 0-terminated. the path is only given within the current file system.
2048 * Therefore, it never starts with a '/'. the caller is responsible to provide
2049 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
2050 * the start point of the resulting string is returned. this pointer is within
2052 * in case the path buffer would overflow, the pointer is decremented further
2053 * as if output was written to the buffer, though no more output is actually
2054 * generated. that way, the caller can determine how much space would be
2055 * required for the path to fit into the buffer. in that case, the returned
2056 * value will be smaller than dest. callers must check this!
2058 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
2059 u32 name_len, unsigned long name_off,
2060 struct extent_buffer *eb_in, u64 parent,
2061 char *dest, u32 size)
2066 s64 bytes_left = ((s64)size) - 1;
2067 struct extent_buffer *eb = eb_in;
2068 struct btrfs_key found_key;
2069 struct btrfs_inode_ref *iref;
2071 if (bytes_left >= 0)
2072 dest[bytes_left] = '\0';
2075 bytes_left -= name_len;
2076 if (bytes_left >= 0)
2077 read_extent_buffer(eb, dest + bytes_left,
2078 name_off, name_len);
2080 if (!path->skip_locking)
2081 btrfs_tree_read_unlock(eb);
2082 free_extent_buffer(eb);
2084 ret = btrfs_find_item(fs_root, path, parent, 0,
2085 BTRFS_INODE_REF_KEY, &found_key);
2091 next_inum = found_key.offset;
2093 /* regular exit ahead */
2094 if (parent == next_inum)
2097 slot = path->slots[0];
2098 eb = path->nodes[0];
2099 /* make sure we can use eb after releasing the path */
2101 path->nodes[0] = NULL;
2104 btrfs_release_path(path);
2105 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2107 name_len = btrfs_inode_ref_name_len(eb, iref);
2108 name_off = (unsigned long)(iref + 1);
2112 if (bytes_left >= 0)
2113 dest[bytes_left] = '/';
2116 btrfs_release_path(path);
2119 return ERR_PTR(ret);
2121 return dest + bytes_left;
2125 * this makes the path point to (logical EXTENT_ITEM *)
2126 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
2127 * tree blocks and <0 on error.
2129 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
2130 struct btrfs_path *path, struct btrfs_key *found_key,
2133 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical);
2138 const struct extent_buffer *eb;
2139 struct btrfs_extent_item *ei;
2140 struct btrfs_key key;
2142 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
2143 key.type = BTRFS_METADATA_ITEM_KEY;
2145 key.type = BTRFS_EXTENT_ITEM_KEY;
2146 key.objectid = logical;
2147 key.offset = (u64)-1;
2149 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2153 ret = btrfs_previous_extent_item(extent_root, path, 0);
2159 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
2160 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
2161 size = fs_info->nodesize;
2162 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
2163 size = found_key->offset;
2165 if (found_key->objectid > logical ||
2166 found_key->objectid + size <= logical) {
2167 btrfs_debug(fs_info,
2168 "logical %llu is not within any extent", logical);
2172 eb = path->nodes[0];
2173 item_size = btrfs_item_size(eb, path->slots[0]);
2174 BUG_ON(item_size < sizeof(*ei));
2176 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
2177 flags = btrfs_extent_flags(eb, ei);
2179 btrfs_debug(fs_info,
2180 "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
2181 logical, logical - found_key->objectid, found_key->objectid,
2182 found_key->offset, flags, item_size);
2184 WARN_ON(!flags_ret);
2186 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2187 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
2188 else if (flags & BTRFS_EXTENT_FLAG_DATA)
2189 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
2199 * helper function to iterate extent inline refs. ptr must point to a 0 value
2200 * for the first call and may be modified. it is used to track state.
2201 * if more refs exist, 0 is returned and the next call to
2202 * get_extent_inline_ref must pass the modified ptr parameter to get the
2203 * next ref. after the last ref was processed, 1 is returned.
2204 * returns <0 on error
2206 static int get_extent_inline_ref(unsigned long *ptr,
2207 const struct extent_buffer *eb,
2208 const struct btrfs_key *key,
2209 const struct btrfs_extent_item *ei,
2211 struct btrfs_extent_inline_ref **out_eiref,
2216 struct btrfs_tree_block_info *info;
2220 flags = btrfs_extent_flags(eb, ei);
2221 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2222 if (key->type == BTRFS_METADATA_ITEM_KEY) {
2223 /* a skinny metadata extent */
2225 (struct btrfs_extent_inline_ref *)(ei + 1);
2227 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
2228 info = (struct btrfs_tree_block_info *)(ei + 1);
2230 (struct btrfs_extent_inline_ref *)(info + 1);
2233 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
2235 *ptr = (unsigned long)*out_eiref;
2236 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
2240 end = (unsigned long)ei + item_size;
2241 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
2242 *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
2243 BTRFS_REF_TYPE_ANY);
2244 if (*out_type == BTRFS_REF_TYPE_INVALID)
2247 *ptr += btrfs_extent_inline_ref_size(*out_type);
2248 WARN_ON(*ptr > end);
2250 return 1; /* last */
2256 * reads the tree block backref for an extent. tree level and root are returned
2257 * through out_level and out_root. ptr must point to a 0 value for the first
2258 * call and may be modified (see get_extent_inline_ref comment).
2259 * returns 0 if data was provided, 1 if there was no more data to provide or
2262 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
2263 struct btrfs_key *key, struct btrfs_extent_item *ei,
2264 u32 item_size, u64 *out_root, u8 *out_level)
2268 struct btrfs_extent_inline_ref *eiref;
2270 if (*ptr == (unsigned long)-1)
2274 ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
2279 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
2280 type == BTRFS_SHARED_BLOCK_REF_KEY)
2287 /* we can treat both ref types equally here */
2288 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
2290 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
2291 struct btrfs_tree_block_info *info;
2293 info = (struct btrfs_tree_block_info *)(ei + 1);
2294 *out_level = btrfs_tree_block_level(eb, info);
2296 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
2297 *out_level = (u8)key->offset;
2301 *ptr = (unsigned long)-1;
2306 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
2307 struct extent_inode_elem *inode_list,
2308 u64 root, u64 extent_item_objectid,
2309 iterate_extent_inodes_t *iterate, void *ctx)
2311 struct extent_inode_elem *eie;
2314 for (eie = inode_list; eie; eie = eie->next) {
2315 btrfs_debug(fs_info,
2316 "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
2317 extent_item_objectid, eie->inum,
2319 ret = iterate(eie->inum, eie->offset, eie->num_bytes, root, ctx);
2321 btrfs_debug(fs_info,
2322 "stopping iteration for %llu due to ret=%d",
2323 extent_item_objectid, ret);
2332 * calls iterate() for every inode that references the extent identified by
2333 * the given parameters.
2334 * when the iterator function returns a non-zero value, iteration stops.
2336 int iterate_extent_inodes(struct btrfs_backref_walk_ctx *ctx,
2337 bool search_commit_root,
2338 iterate_extent_inodes_t *iterate, void *user_ctx)
2342 struct ulist_node *ref_node;
2343 struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem);
2344 struct ulist_iterator ref_uiter;
2346 btrfs_debug(ctx->fs_info, "resolving all inodes for extent %llu",
2349 ASSERT(ctx->trans == NULL);
2350 ASSERT(ctx->roots == NULL);
2352 if (!search_commit_root) {
2353 struct btrfs_trans_handle *trans;
2355 trans = btrfs_attach_transaction(ctx->fs_info->tree_root);
2356 if (IS_ERR(trans)) {
2357 if (PTR_ERR(trans) != -ENOENT &&
2358 PTR_ERR(trans) != -EROFS)
2359 return PTR_ERR(trans);
2366 btrfs_get_tree_mod_seq(ctx->fs_info, &seq_elem);
2367 ctx->time_seq = seq_elem.seq;
2369 down_read(&ctx->fs_info->commit_root_sem);
2372 ret = btrfs_find_all_leafs(ctx);
2378 ULIST_ITER_INIT(&ref_uiter);
2379 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
2380 const u64 leaf_bytenr = ref_node->val;
2381 struct ulist_node *root_node;
2382 struct ulist_iterator root_uiter;
2383 struct extent_inode_elem *inode_list;
2385 inode_list = (struct extent_inode_elem *)(uintptr_t)ref_node->aux;
2387 if (ctx->cache_lookup) {
2388 const u64 *root_ids;
2392 cached = ctx->cache_lookup(leaf_bytenr, ctx->user_ctx,
2393 &root_ids, &root_count);
2395 for (int i = 0; i < root_count; i++) {
2396 ret = iterate_leaf_refs(ctx->fs_info,
2410 ctx->roots = ulist_alloc(GFP_NOFS);
2417 ctx->bytenr = leaf_bytenr;
2418 ret = btrfs_find_all_roots_safe(ctx);
2422 if (ctx->cache_store)
2423 ctx->cache_store(leaf_bytenr, ctx->roots, ctx->user_ctx);
2425 ULIST_ITER_INIT(&root_uiter);
2426 while (!ret && (root_node = ulist_next(ctx->roots, &root_uiter))) {
2427 btrfs_debug(ctx->fs_info,
2428 "root %llu references leaf %llu, data list %#llx",
2429 root_node->val, ref_node->val,
2431 ret = iterate_leaf_refs(ctx->fs_info, inode_list,
2432 root_node->val, ctx->bytenr,
2435 ulist_reinit(ctx->roots);
2438 free_leaf_list(refs);
2441 btrfs_put_tree_mod_seq(ctx->fs_info, &seq_elem);
2442 btrfs_end_transaction(ctx->trans);
2445 up_read(&ctx->fs_info->commit_root_sem);
2448 ulist_free(ctx->roots);
2451 if (ret == BTRFS_ITERATE_EXTENT_INODES_STOP)
2457 static int build_ino_list(u64 inum, u64 offset, u64 num_bytes, u64 root, void *ctx)
2459 struct btrfs_data_container *inodes = ctx;
2460 const size_t c = 3 * sizeof(u64);
2462 if (inodes->bytes_left >= c) {
2463 inodes->bytes_left -= c;
2464 inodes->val[inodes->elem_cnt] = inum;
2465 inodes->val[inodes->elem_cnt + 1] = offset;
2466 inodes->val[inodes->elem_cnt + 2] = root;
2467 inodes->elem_cnt += 3;
2469 inodes->bytes_missing += c - inodes->bytes_left;
2470 inodes->bytes_left = 0;
2471 inodes->elem_missed += 3;
2477 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2478 struct btrfs_path *path,
2479 void *ctx, bool ignore_offset)
2481 struct btrfs_backref_walk_ctx walk_ctx = { 0 };
2484 struct btrfs_key found_key;
2485 int search_commit_root = path->search_commit_root;
2487 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2488 btrfs_release_path(path);
2491 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2494 walk_ctx.bytenr = found_key.objectid;
2496 walk_ctx.ignore_extent_item_pos = true;
2498 walk_ctx.extent_item_pos = logical - found_key.objectid;
2499 walk_ctx.fs_info = fs_info;
2501 return iterate_extent_inodes(&walk_ctx, search_commit_root,
2502 build_ino_list, ctx);
2505 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2506 struct extent_buffer *eb, struct inode_fs_paths *ipath);
2508 static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath)
2517 struct btrfs_root *fs_root = ipath->fs_root;
2518 struct btrfs_path *path = ipath->btrfs_path;
2519 struct extent_buffer *eb;
2520 struct btrfs_inode_ref *iref;
2521 struct btrfs_key found_key;
2524 ret = btrfs_find_item(fs_root, path, inum,
2525 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2531 ret = found ? 0 : -ENOENT;
2536 parent = found_key.offset;
2537 slot = path->slots[0];
2538 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2543 btrfs_release_path(path);
2545 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2547 for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) {
2548 name_len = btrfs_inode_ref_name_len(eb, iref);
2549 /* path must be released before calling iterate()! */
2550 btrfs_debug(fs_root->fs_info,
2551 "following ref at offset %u for inode %llu in tree %llu",
2552 cur, found_key.objectid,
2553 fs_root->root_key.objectid);
2554 ret = inode_to_path(parent, name_len,
2555 (unsigned long)(iref + 1), eb, ipath);
2558 len = sizeof(*iref) + name_len;
2559 iref = (struct btrfs_inode_ref *)((char *)iref + len);
2561 free_extent_buffer(eb);
2564 btrfs_release_path(path);
2569 static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath)
2576 struct btrfs_root *fs_root = ipath->fs_root;
2577 struct btrfs_path *path = ipath->btrfs_path;
2578 struct extent_buffer *eb;
2579 struct btrfs_inode_extref *extref;
2585 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2590 ret = found ? 0 : -ENOENT;
2595 slot = path->slots[0];
2596 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2601 btrfs_release_path(path);
2603 item_size = btrfs_item_size(eb, slot);
2604 ptr = btrfs_item_ptr_offset(eb, slot);
2607 while (cur_offset < item_size) {
2610 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2611 parent = btrfs_inode_extref_parent(eb, extref);
2612 name_len = btrfs_inode_extref_name_len(eb, extref);
2613 ret = inode_to_path(parent, name_len,
2614 (unsigned long)&extref->name, eb, ipath);
2618 cur_offset += btrfs_inode_extref_name_len(eb, extref);
2619 cur_offset += sizeof(*extref);
2621 free_extent_buffer(eb);
2626 btrfs_release_path(path);
2632 * returns 0 if the path could be dumped (probably truncated)
2633 * returns <0 in case of an error
2635 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2636 struct extent_buffer *eb, struct inode_fs_paths *ipath)
2640 int i = ipath->fspath->elem_cnt;
2641 const int s_ptr = sizeof(char *);
2644 bytes_left = ipath->fspath->bytes_left > s_ptr ?
2645 ipath->fspath->bytes_left - s_ptr : 0;
2647 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2648 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2649 name_off, eb, inum, fspath_min, bytes_left);
2651 return PTR_ERR(fspath);
2653 if (fspath > fspath_min) {
2654 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2655 ++ipath->fspath->elem_cnt;
2656 ipath->fspath->bytes_left = fspath - fspath_min;
2658 ++ipath->fspath->elem_missed;
2659 ipath->fspath->bytes_missing += fspath_min - fspath;
2660 ipath->fspath->bytes_left = 0;
2667 * this dumps all file system paths to the inode into the ipath struct, provided
2668 * is has been created large enough. each path is zero-terminated and accessed
2669 * from ipath->fspath->val[i].
2670 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2671 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2672 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2673 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2674 * have been needed to return all paths.
2676 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2681 ret = iterate_inode_refs(inum, ipath);
2684 else if (ret != -ENOENT)
2687 ret = iterate_inode_extrefs(inum, ipath);
2688 if (ret == -ENOENT && found_refs)
2694 struct btrfs_data_container *init_data_container(u32 total_bytes)
2696 struct btrfs_data_container *data;
2699 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2700 data = kvmalloc(alloc_bytes, GFP_KERNEL);
2702 return ERR_PTR(-ENOMEM);
2704 if (total_bytes >= sizeof(*data)) {
2705 data->bytes_left = total_bytes - sizeof(*data);
2706 data->bytes_missing = 0;
2708 data->bytes_missing = sizeof(*data) - total_bytes;
2709 data->bytes_left = 0;
2713 data->elem_missed = 0;
2719 * allocates space to return multiple file system paths for an inode.
2720 * total_bytes to allocate are passed, note that space usable for actual path
2721 * information will be total_bytes - sizeof(struct inode_fs_paths).
2722 * the returned pointer must be freed with free_ipath() in the end.
2724 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2725 struct btrfs_path *path)
2727 struct inode_fs_paths *ifp;
2728 struct btrfs_data_container *fspath;
2730 fspath = init_data_container(total_bytes);
2732 return ERR_CAST(fspath);
2734 ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2737 return ERR_PTR(-ENOMEM);
2740 ifp->btrfs_path = path;
2741 ifp->fspath = fspath;
2742 ifp->fs_root = fs_root;
2747 void free_ipath(struct inode_fs_paths *ipath)
2751 kvfree(ipath->fspath);
2755 struct btrfs_backref_iter *btrfs_backref_iter_alloc(struct btrfs_fs_info *fs_info)
2757 struct btrfs_backref_iter *ret;
2759 ret = kzalloc(sizeof(*ret), GFP_NOFS);
2763 ret->path = btrfs_alloc_path();
2769 /* Current backref iterator only supports iteration in commit root */
2770 ret->path->search_commit_root = 1;
2771 ret->path->skip_locking = 1;
2772 ret->fs_info = fs_info;
2777 int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2779 struct btrfs_fs_info *fs_info = iter->fs_info;
2780 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2781 struct btrfs_path *path = iter->path;
2782 struct btrfs_extent_item *ei;
2783 struct btrfs_key key;
2786 key.objectid = bytenr;
2787 key.type = BTRFS_METADATA_ITEM_KEY;
2788 key.offset = (u64)-1;
2789 iter->bytenr = bytenr;
2791 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2798 if (path->slots[0] == 0) {
2799 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2805 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2806 if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2807 key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2811 memcpy(&iter->cur_key, &key, sizeof(key));
2812 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2814 iter->end_ptr = (u32)(iter->item_ptr +
2815 btrfs_item_size(path->nodes[0], path->slots[0]));
2816 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2817 struct btrfs_extent_item);
2820 * Only support iteration on tree backref yet.
2822 * This is an extra precaution for non skinny-metadata, where
2823 * EXTENT_ITEM is also used for tree blocks, that we can only use
2824 * extent flags to determine if it's a tree block.
2826 if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2830 iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2832 /* If there is no inline backref, go search for keyed backref */
2833 if (iter->cur_ptr >= iter->end_ptr) {
2834 ret = btrfs_next_item(extent_root, path);
2836 /* No inline nor keyed ref */
2844 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2846 if (iter->cur_key.objectid != bytenr ||
2847 (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2848 iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2852 iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2854 iter->item_ptr = iter->cur_ptr;
2855 iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size(
2856 path->nodes[0], path->slots[0]));
2861 btrfs_backref_iter_release(iter);
2866 * Go to the next backref item of current bytenr, can be either inlined or
2869 * Caller needs to check whether it's inline ref or not by iter->cur_key.
2871 * Return 0 if we get next backref without problem.
2872 * Return >0 if there is no extra backref for this bytenr.
2873 * Return <0 if there is something wrong happened.
2875 int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2877 struct extent_buffer *eb = btrfs_backref_get_eb(iter);
2878 struct btrfs_root *extent_root;
2879 struct btrfs_path *path = iter->path;
2880 struct btrfs_extent_inline_ref *iref;
2884 if (btrfs_backref_iter_is_inline_ref(iter)) {
2885 /* We're still inside the inline refs */
2886 ASSERT(iter->cur_ptr < iter->end_ptr);
2888 if (btrfs_backref_has_tree_block_info(iter)) {
2889 /* First tree block info */
2890 size = sizeof(struct btrfs_tree_block_info);
2892 /* Use inline ref type to determine the size */
2895 iref = (struct btrfs_extent_inline_ref *)
2896 ((unsigned long)iter->cur_ptr);
2897 type = btrfs_extent_inline_ref_type(eb, iref);
2899 size = btrfs_extent_inline_ref_size(type);
2901 iter->cur_ptr += size;
2902 if (iter->cur_ptr < iter->end_ptr)
2905 /* All inline items iterated, fall through */
2908 /* We're at keyed items, there is no inline item, go to the next one */
2909 extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr);
2910 ret = btrfs_next_item(extent_root, iter->path);
2914 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
2915 if (iter->cur_key.objectid != iter->bytenr ||
2916 (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2917 iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
2919 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2921 iter->cur_ptr = iter->item_ptr;
2922 iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0],
2927 void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
2928 struct btrfs_backref_cache *cache, int is_reloc)
2932 cache->rb_root = RB_ROOT;
2933 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2934 INIT_LIST_HEAD(&cache->pending[i]);
2935 INIT_LIST_HEAD(&cache->changed);
2936 INIT_LIST_HEAD(&cache->detached);
2937 INIT_LIST_HEAD(&cache->leaves);
2938 INIT_LIST_HEAD(&cache->pending_edge);
2939 INIT_LIST_HEAD(&cache->useless_node);
2940 cache->fs_info = fs_info;
2941 cache->is_reloc = is_reloc;
2944 struct btrfs_backref_node *btrfs_backref_alloc_node(
2945 struct btrfs_backref_cache *cache, u64 bytenr, int level)
2947 struct btrfs_backref_node *node;
2949 ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
2950 node = kzalloc(sizeof(*node), GFP_NOFS);
2954 INIT_LIST_HEAD(&node->list);
2955 INIT_LIST_HEAD(&node->upper);
2956 INIT_LIST_HEAD(&node->lower);
2957 RB_CLEAR_NODE(&node->rb_node);
2959 node->level = level;
2960 node->bytenr = bytenr;
2965 struct btrfs_backref_edge *btrfs_backref_alloc_edge(
2966 struct btrfs_backref_cache *cache)
2968 struct btrfs_backref_edge *edge;
2970 edge = kzalloc(sizeof(*edge), GFP_NOFS);
2977 * Drop the backref node from cache, also cleaning up all its
2978 * upper edges and any uncached nodes in the path.
2980 * This cleanup happens bottom up, thus the node should either
2981 * be the lowest node in the cache or a detached node.
2983 void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
2984 struct btrfs_backref_node *node)
2986 struct btrfs_backref_node *upper;
2987 struct btrfs_backref_edge *edge;
2992 BUG_ON(!node->lowest && !node->detached);
2993 while (!list_empty(&node->upper)) {
2994 edge = list_entry(node->upper.next, struct btrfs_backref_edge,
2996 upper = edge->node[UPPER];
2997 list_del(&edge->list[LOWER]);
2998 list_del(&edge->list[UPPER]);
2999 btrfs_backref_free_edge(cache, edge);
3002 * Add the node to leaf node list if no other child block
3005 if (list_empty(&upper->lower)) {
3006 list_add_tail(&upper->lower, &cache->leaves);
3011 btrfs_backref_drop_node(cache, node);
3015 * Release all nodes/edges from current cache
3017 void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
3019 struct btrfs_backref_node *node;
3022 while (!list_empty(&cache->detached)) {
3023 node = list_entry(cache->detached.next,
3024 struct btrfs_backref_node, list);
3025 btrfs_backref_cleanup_node(cache, node);
3028 while (!list_empty(&cache->leaves)) {
3029 node = list_entry(cache->leaves.next,
3030 struct btrfs_backref_node, lower);
3031 btrfs_backref_cleanup_node(cache, node);
3034 cache->last_trans = 0;
3036 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
3037 ASSERT(list_empty(&cache->pending[i]));
3038 ASSERT(list_empty(&cache->pending_edge));
3039 ASSERT(list_empty(&cache->useless_node));
3040 ASSERT(list_empty(&cache->changed));
3041 ASSERT(list_empty(&cache->detached));
3042 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
3043 ASSERT(!cache->nr_nodes);
3044 ASSERT(!cache->nr_edges);
3048 * Handle direct tree backref
3050 * Direct tree backref means, the backref item shows its parent bytenr
3051 * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
3053 * @ref_key: The converted backref key.
3054 * For keyed backref, it's the item key.
3055 * For inlined backref, objectid is the bytenr,
3056 * type is btrfs_inline_ref_type, offset is
3057 * btrfs_inline_ref_offset.
3059 static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
3060 struct btrfs_key *ref_key,
3061 struct btrfs_backref_node *cur)
3063 struct btrfs_backref_edge *edge;
3064 struct btrfs_backref_node *upper;
3065 struct rb_node *rb_node;
3067 ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
3069 /* Only reloc root uses backref pointing to itself */
3070 if (ref_key->objectid == ref_key->offset) {
3071 struct btrfs_root *root;
3073 cur->is_reloc_root = 1;
3074 /* Only reloc backref cache cares about a specific root */
3075 if (cache->is_reloc) {
3076 root = find_reloc_root(cache->fs_info, cur->bytenr);
3082 * For generic purpose backref cache, reloc root node
3085 list_add(&cur->list, &cache->useless_node);
3090 edge = btrfs_backref_alloc_edge(cache);
3094 rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
3096 /* Parent node not yet cached */
3097 upper = btrfs_backref_alloc_node(cache, ref_key->offset,
3100 btrfs_backref_free_edge(cache, edge);
3105 * Backrefs for the upper level block isn't cached, add the
3106 * block to pending list
3108 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3110 /* Parent node already cached */
3111 upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
3112 ASSERT(upper->checked);
3113 INIT_LIST_HEAD(&edge->list[UPPER]);
3115 btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
3120 * Handle indirect tree backref
3122 * Indirect tree backref means, we only know which tree the node belongs to.
3123 * We still need to do a tree search to find out the parents. This is for
3124 * TREE_BLOCK_REF backref (keyed or inlined).
3126 * @ref_key: The same as @ref_key in handle_direct_tree_backref()
3127 * @tree_key: The first key of this tree block.
3128 * @path: A clean (released) path, to avoid allocating path every time
3129 * the function get called.
3131 static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache,
3132 struct btrfs_path *path,
3133 struct btrfs_key *ref_key,
3134 struct btrfs_key *tree_key,
3135 struct btrfs_backref_node *cur)
3137 struct btrfs_fs_info *fs_info = cache->fs_info;
3138 struct btrfs_backref_node *upper;
3139 struct btrfs_backref_node *lower;
3140 struct btrfs_backref_edge *edge;
3141 struct extent_buffer *eb;
3142 struct btrfs_root *root;
3143 struct rb_node *rb_node;
3145 bool need_check = true;
3148 root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
3150 return PTR_ERR(root);
3151 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3154 if (btrfs_root_level(&root->root_item) == cur->level) {
3156 ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
3158 * For reloc backref cache, we may ignore reloc root. But for
3159 * general purpose backref cache, we can't rely on
3160 * btrfs_should_ignore_reloc_root() as it may conflict with
3161 * current running relocation and lead to missing root.
3163 * For general purpose backref cache, reloc root detection is
3164 * completely relying on direct backref (key->offset is parent
3165 * bytenr), thus only do such check for reloc cache.
3167 if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
3168 btrfs_put_root(root);
3169 list_add(&cur->list, &cache->useless_node);
3176 level = cur->level + 1;
3178 /* Search the tree to find parent blocks referring to the block */
3179 path->search_commit_root = 1;
3180 path->skip_locking = 1;
3181 path->lowest_level = level;
3182 ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
3183 path->lowest_level = 0;
3185 btrfs_put_root(root);
3188 if (ret > 0 && path->slots[level] > 0)
3189 path->slots[level]--;
3191 eb = path->nodes[level];
3192 if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
3194 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
3195 cur->bytenr, level - 1, root->root_key.objectid,
3196 tree_key->objectid, tree_key->type, tree_key->offset);
3197 btrfs_put_root(root);
3203 /* Add all nodes and edges in the path */
3204 for (; level < BTRFS_MAX_LEVEL; level++) {
3205 if (!path->nodes[level]) {
3206 ASSERT(btrfs_root_bytenr(&root->root_item) ==
3208 /* Same as previous should_ignore_reloc_root() call */
3209 if (btrfs_should_ignore_reloc_root(root) &&
3211 btrfs_put_root(root);
3212 list_add(&lower->list, &cache->useless_node);
3219 edge = btrfs_backref_alloc_edge(cache);
3221 btrfs_put_root(root);
3226 eb = path->nodes[level];
3227 rb_node = rb_simple_search(&cache->rb_root, eb->start);
3229 upper = btrfs_backref_alloc_node(cache, eb->start,
3232 btrfs_put_root(root);
3233 btrfs_backref_free_edge(cache, edge);
3237 upper->owner = btrfs_header_owner(eb);
3238 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3242 * If we know the block isn't shared we can avoid
3243 * checking its backrefs.
3245 if (btrfs_block_can_be_shared(root, eb))
3251 * Add the block to pending list if we need to check its
3252 * backrefs, we only do this once while walking up a
3253 * tree as we will catch anything else later on.
3255 if (!upper->checked && need_check) {
3257 list_add_tail(&edge->list[UPPER],
3258 &cache->pending_edge);
3262 INIT_LIST_HEAD(&edge->list[UPPER]);
3265 upper = rb_entry(rb_node, struct btrfs_backref_node,
3267 ASSERT(upper->checked);
3268 INIT_LIST_HEAD(&edge->list[UPPER]);
3270 upper->owner = btrfs_header_owner(eb);
3272 btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
3275 btrfs_put_root(root);
3282 btrfs_release_path(path);
3287 * Add backref node @cur into @cache.
3289 * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
3290 * links aren't yet bi-directional. Needs to finish such links.
3291 * Use btrfs_backref_finish_upper_links() to finish such linkage.
3293 * @path: Released path for indirect tree backref lookup
3294 * @iter: Released backref iter for extent tree search
3295 * @node_key: The first key of the tree block
3297 int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache,
3298 struct btrfs_path *path,
3299 struct btrfs_backref_iter *iter,
3300 struct btrfs_key *node_key,
3301 struct btrfs_backref_node *cur)
3303 struct btrfs_fs_info *fs_info = cache->fs_info;
3304 struct btrfs_backref_edge *edge;
3305 struct btrfs_backref_node *exist;
3308 ret = btrfs_backref_iter_start(iter, cur->bytenr);
3312 * We skip the first btrfs_tree_block_info, as we don't use the key
3313 * stored in it, but fetch it from the tree block
3315 if (btrfs_backref_has_tree_block_info(iter)) {
3316 ret = btrfs_backref_iter_next(iter);
3319 /* No extra backref? This means the tree block is corrupted */
3325 WARN_ON(cur->checked);
3326 if (!list_empty(&cur->upper)) {
3328 * The backref was added previously when processing backref of
3329 * type BTRFS_TREE_BLOCK_REF_KEY
3331 ASSERT(list_is_singular(&cur->upper));
3332 edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
3334 ASSERT(list_empty(&edge->list[UPPER]));
3335 exist = edge->node[UPPER];
3337 * Add the upper level block to pending list if we need check
3340 if (!exist->checked)
3341 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3346 for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
3347 struct extent_buffer *eb;
3348 struct btrfs_key key;
3352 eb = btrfs_backref_get_eb(iter);
3354 key.objectid = iter->bytenr;
3355 if (btrfs_backref_iter_is_inline_ref(iter)) {
3356 struct btrfs_extent_inline_ref *iref;
3358 /* Update key for inline backref */
3359 iref = (struct btrfs_extent_inline_ref *)
3360 ((unsigned long)iter->cur_ptr);
3361 type = btrfs_get_extent_inline_ref_type(eb, iref,
3362 BTRFS_REF_TYPE_BLOCK);
3363 if (type == BTRFS_REF_TYPE_INVALID) {
3368 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3370 key.type = iter->cur_key.type;
3371 key.offset = iter->cur_key.offset;
3375 * Parent node found and matches current inline ref, no need to
3376 * rebuild this node for this inline ref
3379 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
3380 exist->owner == key.offset) ||
3381 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
3382 exist->bytenr == key.offset))) {
3387 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
3388 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3389 ret = handle_direct_tree_backref(cache, &key, cur);
3393 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3395 btrfs_print_v0_err(fs_info);
3396 btrfs_handle_fs_error(fs_info, ret, NULL);
3398 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
3403 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
3404 * means the root objectid. We need to search the tree to get
3405 * its parent bytenr.
3407 ret = handle_indirect_tree_backref(cache, path, &key, node_key,
3416 btrfs_backref_iter_release(iter);
3421 * Finish the upwards linkage created by btrfs_backref_add_tree_node()
3423 int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
3424 struct btrfs_backref_node *start)
3426 struct list_head *useless_node = &cache->useless_node;
3427 struct btrfs_backref_edge *edge;
3428 struct rb_node *rb_node;
3429 LIST_HEAD(pending_edge);
3431 ASSERT(start->checked);
3433 /* Insert this node to cache if it's not COW-only */
3434 if (!start->cowonly) {
3435 rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
3438 btrfs_backref_panic(cache->fs_info, start->bytenr,
3440 list_add_tail(&start->lower, &cache->leaves);
3444 * Use breadth first search to iterate all related edges.
3446 * The starting points are all the edges of this node
3448 list_for_each_entry(edge, &start->upper, list[LOWER])
3449 list_add_tail(&edge->list[UPPER], &pending_edge);
3451 while (!list_empty(&pending_edge)) {
3452 struct btrfs_backref_node *upper;
3453 struct btrfs_backref_node *lower;
3455 edge = list_first_entry(&pending_edge,
3456 struct btrfs_backref_edge, list[UPPER]);
3457 list_del_init(&edge->list[UPPER]);
3458 upper = edge->node[UPPER];
3459 lower = edge->node[LOWER];
3461 /* Parent is detached, no need to keep any edges */
3462 if (upper->detached) {
3463 list_del(&edge->list[LOWER]);
3464 btrfs_backref_free_edge(cache, edge);
3466 /* Lower node is orphan, queue for cleanup */
3467 if (list_empty(&lower->upper))
3468 list_add(&lower->list, useless_node);
3473 * All new nodes added in current build_backref_tree() haven't
3474 * been linked to the cache rb tree.
3475 * So if we have upper->rb_node populated, this means a cache
3476 * hit. We only need to link the edge, as @upper and all its
3477 * parents have already been linked.
3479 if (!RB_EMPTY_NODE(&upper->rb_node)) {
3480 if (upper->lowest) {
3481 list_del_init(&upper->lower);
3485 list_add_tail(&edge->list[UPPER], &upper->lower);
3489 /* Sanity check, we shouldn't have any unchecked nodes */
3490 if (!upper->checked) {
3495 /* Sanity check, COW-only node has non-COW-only parent */
3496 if (start->cowonly != upper->cowonly) {
3501 /* Only cache non-COW-only (subvolume trees) tree blocks */
3502 if (!upper->cowonly) {
3503 rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3506 btrfs_backref_panic(cache->fs_info,
3507 upper->bytenr, -EEXIST);
3512 list_add_tail(&edge->list[UPPER], &upper->lower);
3515 * Also queue all the parent edges of this uncached node
3516 * to finish the upper linkage
3518 list_for_each_entry(edge, &upper->upper, list[LOWER])
3519 list_add_tail(&edge->list[UPPER], &pending_edge);
3524 void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3525 struct btrfs_backref_node *node)
3527 struct btrfs_backref_node *lower;
3528 struct btrfs_backref_node *upper;
3529 struct btrfs_backref_edge *edge;
3531 while (!list_empty(&cache->useless_node)) {
3532 lower = list_first_entry(&cache->useless_node,
3533 struct btrfs_backref_node, list);
3534 list_del_init(&lower->list);
3536 while (!list_empty(&cache->pending_edge)) {
3537 edge = list_first_entry(&cache->pending_edge,
3538 struct btrfs_backref_edge, list[UPPER]);
3539 list_del(&edge->list[UPPER]);
3540 list_del(&edge->list[LOWER]);
3541 lower = edge->node[LOWER];
3542 upper = edge->node[UPPER];
3543 btrfs_backref_free_edge(cache, edge);
3546 * Lower is no longer linked to any upper backref nodes and
3547 * isn't in the cache, we can free it ourselves.
3549 if (list_empty(&lower->upper) &&
3550 RB_EMPTY_NODE(&lower->rb_node))
3551 list_add(&lower->list, &cache->useless_node);
3553 if (!RB_EMPTY_NODE(&upper->rb_node))
3556 /* Add this guy's upper edges to the list to process */
3557 list_for_each_entry(edge, &upper->upper, list[LOWER])
3558 list_add_tail(&edge->list[UPPER],
3559 &cache->pending_edge);
3560 if (list_empty(&upper->upper))
3561 list_add(&upper->list, &cache->useless_node);
3564 while (!list_empty(&cache->useless_node)) {
3565 lower = list_first_entry(&cache->useless_node,
3566 struct btrfs_backref_node, list);
3567 list_del_init(&lower->list);
3570 btrfs_backref_drop_node(cache, lower);
3573 btrfs_backref_cleanup_node(cache, node);
3574 ASSERT(list_empty(&cache->useless_node) &&
3575 list_empty(&cache->pending_edge));