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 /* Just an arbitrary number so we can be sure this happened */
20 #define BACKREF_FOUND_SHARED 6
22 struct extent_inode_elem {
25 struct extent_inode_elem *next;
28 static int check_extent_in_eb(const struct btrfs_key *key,
29 const struct extent_buffer *eb,
30 const struct btrfs_file_extent_item *fi,
32 struct extent_inode_elem **eie,
36 struct extent_inode_elem *e;
39 !btrfs_file_extent_compression(eb, fi) &&
40 !btrfs_file_extent_encryption(eb, fi) &&
41 !btrfs_file_extent_other_encoding(eb, fi)) {
45 data_offset = btrfs_file_extent_offset(eb, fi);
46 data_len = btrfs_file_extent_num_bytes(eb, fi);
48 if (extent_item_pos < data_offset ||
49 extent_item_pos >= data_offset + data_len)
51 offset = extent_item_pos - data_offset;
54 e = kmalloc(sizeof(*e), GFP_NOFS);
59 e->inum = key->objectid;
60 e->offset = key->offset + offset;
66 static void free_inode_elem_list(struct extent_inode_elem *eie)
68 struct extent_inode_elem *eie_next;
70 for (; eie; eie = eie_next) {
76 static int find_extent_in_eb(const struct extent_buffer *eb,
77 u64 wanted_disk_byte, u64 extent_item_pos,
78 struct extent_inode_elem **eie,
83 struct btrfs_file_extent_item *fi;
90 * from the shared data ref, we only have the leaf but we need
91 * the key. thus, we must look into all items and see that we
92 * find one (some) with a reference to our extent item.
94 nritems = btrfs_header_nritems(eb);
95 for (slot = 0; slot < nritems; ++slot) {
96 btrfs_item_key_to_cpu(eb, &key, slot);
97 if (key.type != BTRFS_EXTENT_DATA_KEY)
99 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
100 extent_type = btrfs_file_extent_type(eb, fi);
101 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
103 /* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
104 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
105 if (disk_byte != wanted_disk_byte)
108 ret = check_extent_in_eb(&key, eb, fi, extent_item_pos, eie, ignore_offset);
117 struct rb_root_cached root;
121 #define PREFTREE_INIT { .root = RB_ROOT_CACHED, .count = 0 }
124 struct preftree direct; /* BTRFS_SHARED_[DATA|BLOCK]_REF_KEY */
125 struct preftree indirect; /* BTRFS_[TREE_BLOCK|EXTENT_DATA]_REF_KEY */
126 struct preftree indirect_missing_keys;
130 * Checks for a shared extent during backref search.
132 * The share_count tracks prelim_refs (direct and indirect) having a
134 * - incremented when a ref->count transitions to >0
135 * - decremented when a ref->count transitions to <1
141 bool have_delayed_delete_refs;
144 static inline int extent_is_shared(struct share_check *sc)
146 return (sc && sc->share_count > 1) ? BACKREF_FOUND_SHARED : 0;
149 static struct kmem_cache *btrfs_prelim_ref_cache;
151 int __init btrfs_prelim_ref_init(void)
153 btrfs_prelim_ref_cache = kmem_cache_create("btrfs_prelim_ref",
154 sizeof(struct prelim_ref),
158 if (!btrfs_prelim_ref_cache)
163 void __cold btrfs_prelim_ref_exit(void)
165 kmem_cache_destroy(btrfs_prelim_ref_cache);
168 static void free_pref(struct prelim_ref *ref)
170 kmem_cache_free(btrfs_prelim_ref_cache, ref);
174 * Return 0 when both refs are for the same block (and can be merged).
175 * A -1 return indicates ref1 is a 'lower' block than ref2, while 1
176 * indicates a 'higher' block.
178 static int prelim_ref_compare(struct prelim_ref *ref1,
179 struct prelim_ref *ref2)
181 if (ref1->level < ref2->level)
183 if (ref1->level > ref2->level)
185 if (ref1->root_id < ref2->root_id)
187 if (ref1->root_id > ref2->root_id)
189 if (ref1->key_for_search.type < ref2->key_for_search.type)
191 if (ref1->key_for_search.type > ref2->key_for_search.type)
193 if (ref1->key_for_search.objectid < ref2->key_for_search.objectid)
195 if (ref1->key_for_search.objectid > ref2->key_for_search.objectid)
197 if (ref1->key_for_search.offset < ref2->key_for_search.offset)
199 if (ref1->key_for_search.offset > ref2->key_for_search.offset)
201 if (ref1->parent < ref2->parent)
203 if (ref1->parent > ref2->parent)
209 static void update_share_count(struct share_check *sc, int oldcount,
212 if ((!sc) || (oldcount == 0 && newcount < 1))
215 if (oldcount > 0 && newcount < 1)
217 else if (oldcount < 1 && newcount > 0)
222 * Add @newref to the @root rbtree, merging identical refs.
224 * Callers should assume that newref has been freed after calling.
226 static void prelim_ref_insert(const struct btrfs_fs_info *fs_info,
227 struct preftree *preftree,
228 struct prelim_ref *newref,
229 struct share_check *sc)
231 struct rb_root_cached *root;
233 struct rb_node *parent = NULL;
234 struct prelim_ref *ref;
236 bool leftmost = true;
238 root = &preftree->root;
239 p = &root->rb_root.rb_node;
243 ref = rb_entry(parent, struct prelim_ref, rbnode);
244 result = prelim_ref_compare(ref, newref);
247 } else if (result > 0) {
251 /* Identical refs, merge them and free @newref */
252 struct extent_inode_elem *eie = ref->inode_list;
254 while (eie && eie->next)
258 ref->inode_list = newref->inode_list;
260 eie->next = newref->inode_list;
261 trace_btrfs_prelim_ref_merge(fs_info, ref, newref,
264 * A delayed ref can have newref->count < 0.
265 * The ref->count is updated to follow any
266 * BTRFS_[ADD|DROP]_DELAYED_REF actions.
268 update_share_count(sc, ref->count,
269 ref->count + newref->count);
270 ref->count += newref->count;
276 update_share_count(sc, 0, newref->count);
278 trace_btrfs_prelim_ref_insert(fs_info, newref, NULL, preftree->count);
279 rb_link_node(&newref->rbnode, parent, p);
280 rb_insert_color_cached(&newref->rbnode, root, leftmost);
284 * Release the entire tree. We don't care about internal consistency so
285 * just free everything and then reset the tree root.
287 static void prelim_release(struct preftree *preftree)
289 struct prelim_ref *ref, *next_ref;
291 rbtree_postorder_for_each_entry_safe(ref, next_ref,
292 &preftree->root.rb_root, rbnode) {
293 free_inode_elem_list(ref->inode_list);
297 preftree->root = RB_ROOT_CACHED;
302 * the rules for all callers of this function are:
303 * - obtaining the parent is the goal
304 * - if you add a key, you must know that it is a correct key
305 * - if you cannot add the parent or a correct key, then we will look into the
306 * block later to set a correct key
310 * backref type | shared | indirect | shared | indirect
311 * information | tree | tree | data | data
312 * --------------------+--------+----------+--------+----------
313 * parent logical | y | - | - | -
314 * key to resolve | - | y | y | y
315 * tree block logical | - | - | - | -
316 * root for resolving | y | y | y | y
318 * - column 1: we've the parent -> done
319 * - column 2, 3, 4: we use the key to find the parent
321 * on disk refs (inline or keyed)
322 * ==============================
323 * backref type | shared | indirect | shared | indirect
324 * information | tree | tree | data | data
325 * --------------------+--------+----------+--------+----------
326 * parent logical | y | - | y | -
327 * key to resolve | - | - | - | y
328 * tree block logical | y | y | y | y
329 * root for resolving | - | y | y | y
331 * - column 1, 3: we've the parent -> done
332 * - column 2: we take the first key from the block to find the parent
333 * (see add_missing_keys)
334 * - column 4: we use the key to find the parent
336 * additional information that's available but not required to find the parent
337 * block might help in merging entries to gain some speed.
339 static int add_prelim_ref(const struct btrfs_fs_info *fs_info,
340 struct preftree *preftree, u64 root_id,
341 const struct btrfs_key *key, int level, u64 parent,
342 u64 wanted_disk_byte, int count,
343 struct share_check *sc, gfp_t gfp_mask)
345 struct prelim_ref *ref;
347 if (root_id == BTRFS_DATA_RELOC_TREE_OBJECTID)
350 ref = kmem_cache_alloc(btrfs_prelim_ref_cache, gfp_mask);
354 ref->root_id = root_id;
356 ref->key_for_search = *key;
358 memset(&ref->key_for_search, 0, sizeof(ref->key_for_search));
360 ref->inode_list = NULL;
363 ref->parent = parent;
364 ref->wanted_disk_byte = wanted_disk_byte;
365 prelim_ref_insert(fs_info, preftree, ref, sc);
366 return extent_is_shared(sc);
369 /* direct refs use root == 0, key == NULL */
370 static int add_direct_ref(const struct btrfs_fs_info *fs_info,
371 struct preftrees *preftrees, int level, u64 parent,
372 u64 wanted_disk_byte, int count,
373 struct share_check *sc, gfp_t gfp_mask)
375 return add_prelim_ref(fs_info, &preftrees->direct, 0, NULL, level,
376 parent, wanted_disk_byte, count, sc, gfp_mask);
379 /* indirect refs use parent == 0 */
380 static int add_indirect_ref(const struct btrfs_fs_info *fs_info,
381 struct preftrees *preftrees, u64 root_id,
382 const struct btrfs_key *key, int level,
383 u64 wanted_disk_byte, int count,
384 struct share_check *sc, gfp_t gfp_mask)
386 struct preftree *tree = &preftrees->indirect;
389 tree = &preftrees->indirect_missing_keys;
390 return add_prelim_ref(fs_info, tree, root_id, key, level, 0,
391 wanted_disk_byte, count, sc, gfp_mask);
394 static int is_shared_data_backref(struct preftrees *preftrees, u64 bytenr)
396 struct rb_node **p = &preftrees->direct.root.rb_root.rb_node;
397 struct rb_node *parent = NULL;
398 struct prelim_ref *ref = NULL;
399 struct prelim_ref target = {};
402 target.parent = bytenr;
406 ref = rb_entry(parent, struct prelim_ref, rbnode);
407 result = prelim_ref_compare(ref, &target);
419 static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
420 struct ulist *parents,
421 struct preftrees *preftrees, struct prelim_ref *ref,
422 int level, u64 time_seq, const u64 *extent_item_pos,
427 struct extent_buffer *eb;
428 struct btrfs_key key;
429 struct btrfs_key *key_for_search = &ref->key_for_search;
430 struct btrfs_file_extent_item *fi;
431 struct extent_inode_elem *eie = NULL, *old = NULL;
433 u64 wanted_disk_byte = ref->wanted_disk_byte;
438 eb = path->nodes[level];
439 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
446 * 1. We normally enter this function with the path already pointing to
447 * the first item to check. But sometimes, we may enter it with
449 * 2. We are searching for normal backref but bytenr of this leaf
450 * matches shared data backref
451 * 3. The leaf owner is not equal to the root we are searching
453 * For these cases, go to the next leaf before we continue.
456 if (path->slots[0] >= btrfs_header_nritems(eb) ||
457 is_shared_data_backref(preftrees, eb->start) ||
458 ref->root_id != btrfs_header_owner(eb)) {
459 if (time_seq == BTRFS_SEQ_LAST)
460 ret = btrfs_next_leaf(root, path);
462 ret = btrfs_next_old_leaf(root, path, time_seq);
465 while (!ret && count < ref->count) {
467 slot = path->slots[0];
469 btrfs_item_key_to_cpu(eb, &key, slot);
471 if (key.objectid != key_for_search->objectid ||
472 key.type != BTRFS_EXTENT_DATA_KEY)
476 * We are searching for normal backref but bytenr of this leaf
477 * matches shared data backref, OR
478 * the leaf owner is not equal to the root we are searching for
481 (is_shared_data_backref(preftrees, eb->start) ||
482 ref->root_id != btrfs_header_owner(eb))) {
483 if (time_seq == BTRFS_SEQ_LAST)
484 ret = btrfs_next_leaf(root, path);
486 ret = btrfs_next_old_leaf(root, path, time_seq);
489 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
490 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
491 data_offset = btrfs_file_extent_offset(eb, fi);
493 if (disk_byte == wanted_disk_byte) {
496 if (ref->key_for_search.offset == key.offset - data_offset)
500 if (extent_item_pos) {
501 ret = check_extent_in_eb(&key, eb, fi,
503 &eie, ignore_offset);
509 ret = ulist_add_merge_ptr(parents, eb->start,
510 eie, (void **)&old, GFP_NOFS);
513 if (!ret && extent_item_pos) {
521 if (time_seq == BTRFS_SEQ_LAST)
522 ret = btrfs_next_item(root, path);
524 ret = btrfs_next_old_item(root, path, time_seq);
530 free_inode_elem_list(eie);
535 * resolve an indirect backref in the form (root_id, key, level)
536 * to a logical address
538 static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
539 struct btrfs_path *path, u64 time_seq,
540 struct preftrees *preftrees,
541 struct prelim_ref *ref, struct ulist *parents,
542 const u64 *extent_item_pos, bool ignore_offset)
544 struct btrfs_root *root;
545 struct extent_buffer *eb;
548 int level = ref->level;
549 struct btrfs_key search_key = ref->key_for_search;
552 * If we're search_commit_root we could possibly be holding locks on
553 * other tree nodes. This happens when qgroups does backref walks when
554 * adding new delayed refs. To deal with this we need to look in cache
555 * for the root, and if we don't find it then we need to search the
556 * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage
559 if (path->search_commit_root)
560 root = btrfs_get_fs_root_commit_root(fs_info, path, ref->root_id);
562 root = btrfs_get_fs_root(fs_info, ref->root_id, false);
568 if (!path->search_commit_root &&
569 test_bit(BTRFS_ROOT_DELETING, &root->state)) {
574 if (btrfs_is_testing(fs_info)) {
579 if (path->search_commit_root)
580 root_level = btrfs_header_level(root->commit_root);
581 else if (time_seq == BTRFS_SEQ_LAST)
582 root_level = btrfs_header_level(root->node);
584 root_level = btrfs_old_root_level(root, time_seq);
586 if (root_level + 1 == level)
590 * We can often find data backrefs with an offset that is too large
591 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
592 * subtracting a file's offset with the data offset of its
593 * corresponding extent data item. This can happen for example in the
596 * So if we detect such case we set the search key's offset to zero to
597 * make sure we will find the matching file extent item at
598 * add_all_parents(), otherwise we will miss it because the offset
599 * taken form the backref is much larger then the offset of the file
600 * extent item. This can make us scan a very large number of file
601 * extent items, but at least it will not make us miss any.
603 * This is an ugly workaround for a behaviour that should have never
604 * existed, but it does and a fix for the clone ioctl would touch a lot
605 * of places, cause backwards incompatibility and would not fix the
606 * problem for extents cloned with older kernels.
608 if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
609 search_key.offset >= LLONG_MAX)
610 search_key.offset = 0;
611 path->lowest_level = level;
612 if (time_seq == BTRFS_SEQ_LAST)
613 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
615 ret = btrfs_search_old_slot(root, &search_key, path, time_seq);
618 "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
619 ref->root_id, level, ref->count, ret,
620 ref->key_for_search.objectid, ref->key_for_search.type,
621 ref->key_for_search.offset);
625 eb = path->nodes[level];
627 if (WARN_ON(!level)) {
632 eb = path->nodes[level];
635 ret = add_all_parents(root, path, parents, preftrees, ref, level,
636 time_seq, extent_item_pos, ignore_offset);
638 btrfs_put_root(root);
640 path->lowest_level = 0;
641 btrfs_release_path(path);
645 static struct extent_inode_elem *
646 unode_aux_to_inode_list(struct ulist_node *node)
650 return (struct extent_inode_elem *)(uintptr_t)node->aux;
653 static void free_leaf_list(struct ulist *ulist)
655 struct ulist_node *node;
656 struct ulist_iterator uiter;
658 ULIST_ITER_INIT(&uiter);
659 while ((node = ulist_next(ulist, &uiter)))
660 free_inode_elem_list(unode_aux_to_inode_list(node));
666 * We maintain three separate rbtrees: one for direct refs, one for
667 * indirect refs which have a key, and one for indirect refs which do not
668 * have a key. Each tree does merge on insertion.
670 * Once all of the references are located, we iterate over the tree of
671 * indirect refs with missing keys. An appropriate key is located and
672 * the ref is moved onto the tree for indirect refs. After all missing
673 * keys are thus located, we iterate over the indirect ref tree, resolve
674 * each reference, and then insert the resolved reference onto the
675 * direct tree (merging there too).
677 * New backrefs (i.e., for parent nodes) are added to the appropriate
678 * rbtree as they are encountered. The new backrefs are subsequently
681 static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
682 struct btrfs_path *path, u64 time_seq,
683 struct preftrees *preftrees,
684 const u64 *extent_item_pos,
685 struct share_check *sc, bool ignore_offset)
689 struct ulist *parents;
690 struct ulist_node *node;
691 struct ulist_iterator uiter;
692 struct rb_node *rnode;
694 parents = ulist_alloc(GFP_NOFS);
699 * We could trade memory usage for performance here by iterating
700 * the tree, allocating new refs for each insertion, and then
701 * freeing the entire indirect tree when we're done. In some test
702 * cases, the tree can grow quite large (~200k objects).
704 while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
705 struct prelim_ref *ref;
707 ref = rb_entry(rnode, struct prelim_ref, rbnode);
708 if (WARN(ref->parent,
709 "BUG: direct ref found in indirect tree")) {
714 rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
715 preftrees->indirect.count--;
717 if (ref->count == 0) {
722 if (sc && sc->root_objectid &&
723 ref->root_id != sc->root_objectid) {
725 ret = BACKREF_FOUND_SHARED;
728 err = resolve_indirect_ref(fs_info, path, time_seq, preftrees,
729 ref, parents, extent_item_pos,
732 * we can only tolerate ENOENT,otherwise,we should catch error
733 * and return directly.
735 if (err == -ENOENT) {
736 prelim_ref_insert(fs_info, &preftrees->direct, ref,
745 /* we put the first parent into the ref at hand */
746 ULIST_ITER_INIT(&uiter);
747 node = ulist_next(parents, &uiter);
748 ref->parent = node ? node->val : 0;
749 ref->inode_list = unode_aux_to_inode_list(node);
751 /* Add a prelim_ref(s) for any other parent(s). */
752 while ((node = ulist_next(parents, &uiter))) {
753 struct prelim_ref *new_ref;
755 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
762 memcpy(new_ref, ref, sizeof(*ref));
763 new_ref->parent = node->val;
764 new_ref->inode_list = unode_aux_to_inode_list(node);
765 prelim_ref_insert(fs_info, &preftrees->direct,
770 * Now it's a direct ref, put it in the direct tree. We must
771 * do this last because the ref could be merged/freed here.
773 prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
775 ulist_reinit(parents);
780 * We may have inode lists attached to refs in the parents ulist, so we
781 * must free them before freeing the ulist and its refs.
783 free_leaf_list(parents);
788 * read tree blocks and add keys where required.
790 static int add_missing_keys(struct btrfs_fs_info *fs_info,
791 struct preftrees *preftrees, bool lock)
793 struct prelim_ref *ref;
794 struct extent_buffer *eb;
795 struct preftree *tree = &preftrees->indirect_missing_keys;
796 struct rb_node *node;
798 while ((node = rb_first_cached(&tree->root))) {
799 ref = rb_entry(node, struct prelim_ref, rbnode);
800 rb_erase_cached(node, &tree->root);
802 BUG_ON(ref->parent); /* should not be a direct ref */
803 BUG_ON(ref->key_for_search.type);
804 BUG_ON(!ref->wanted_disk_byte);
806 eb = read_tree_block(fs_info, ref->wanted_disk_byte,
807 ref->root_id, 0, ref->level - 1, NULL);
812 if (!extent_buffer_uptodate(eb)) {
814 free_extent_buffer(eb);
819 btrfs_tree_read_lock(eb);
820 if (btrfs_header_level(eb) == 0)
821 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
823 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
825 btrfs_tree_read_unlock(eb);
826 free_extent_buffer(eb);
827 prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
834 * add all currently queued delayed refs from this head whose seq nr is
835 * smaller or equal that seq to the list
837 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
838 struct btrfs_delayed_ref_head *head, u64 seq,
839 struct preftrees *preftrees, struct share_check *sc)
841 struct btrfs_delayed_ref_node *node;
842 struct btrfs_key key;
847 spin_lock(&head->lock);
848 for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
849 node = rb_entry(n, struct btrfs_delayed_ref_node,
854 switch (node->action) {
855 case BTRFS_ADD_DELAYED_EXTENT:
856 case BTRFS_UPDATE_DELAYED_HEAD:
859 case BTRFS_ADD_DELAYED_REF:
860 count = node->ref_mod;
862 case BTRFS_DROP_DELAYED_REF:
863 count = node->ref_mod * -1;
868 switch (node->type) {
869 case BTRFS_TREE_BLOCK_REF_KEY: {
870 /* NORMAL INDIRECT METADATA backref */
871 struct btrfs_delayed_tree_ref *ref;
872 struct btrfs_key *key_ptr = NULL;
874 if (head->extent_op && head->extent_op->update_key) {
875 btrfs_disk_key_to_cpu(&key, &head->extent_op->key);
879 ref = btrfs_delayed_node_to_tree_ref(node);
880 ret = add_indirect_ref(fs_info, preftrees, ref->root,
881 key_ptr, ref->level + 1,
882 node->bytenr, count, sc,
886 case BTRFS_SHARED_BLOCK_REF_KEY: {
887 /* SHARED DIRECT METADATA backref */
888 struct btrfs_delayed_tree_ref *ref;
890 ref = btrfs_delayed_node_to_tree_ref(node);
892 ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
893 ref->parent, node->bytenr, count,
897 case BTRFS_EXTENT_DATA_REF_KEY: {
898 /* NORMAL INDIRECT DATA backref */
899 struct btrfs_delayed_data_ref *ref;
900 ref = btrfs_delayed_node_to_data_ref(node);
902 key.objectid = ref->objectid;
903 key.type = BTRFS_EXTENT_DATA_KEY;
904 key.offset = ref->offset;
907 * If we have a share check context and a reference for
908 * another inode, we can't exit immediately. This is
909 * because even if this is a BTRFS_ADD_DELAYED_REF
910 * reference we may find next a BTRFS_DROP_DELAYED_REF
911 * which cancels out this ADD reference.
913 * If this is a DROP reference and there was no previous
914 * ADD reference, then we need to signal that when we
915 * process references from the extent tree (through
916 * add_inline_refs() and add_keyed_refs()), we should
917 * not exit early if we find a reference for another
918 * inode, because one of the delayed DROP references
919 * may cancel that reference in the extent tree.
922 sc->have_delayed_delete_refs = true;
924 ret = add_indirect_ref(fs_info, preftrees, ref->root,
925 &key, 0, node->bytenr, count, sc,
929 case BTRFS_SHARED_DATA_REF_KEY: {
930 /* SHARED DIRECT FULL backref */
931 struct btrfs_delayed_data_ref *ref;
933 ref = btrfs_delayed_node_to_data_ref(node);
935 ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
936 node->bytenr, count, sc,
944 * We must ignore BACKREF_FOUND_SHARED until all delayed
945 * refs have been checked.
947 if (ret && (ret != BACKREF_FOUND_SHARED))
951 ret = extent_is_shared(sc);
953 spin_unlock(&head->lock);
958 * add all inline backrefs for bytenr to the list
960 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
962 static int add_inline_refs(const struct btrfs_fs_info *fs_info,
963 struct btrfs_path *path, u64 bytenr,
964 int *info_level, struct preftrees *preftrees,
965 struct share_check *sc)
969 struct extent_buffer *leaf;
970 struct btrfs_key key;
971 struct btrfs_key found_key;
974 struct btrfs_extent_item *ei;
979 * enumerate all inline refs
981 leaf = path->nodes[0];
982 slot = path->slots[0];
984 item_size = btrfs_item_size(leaf, slot);
985 BUG_ON(item_size < sizeof(*ei));
987 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
988 flags = btrfs_extent_flags(leaf, ei);
989 btrfs_item_key_to_cpu(leaf, &found_key, slot);
991 ptr = (unsigned long)(ei + 1);
992 end = (unsigned long)ei + item_size;
994 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
995 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
996 struct btrfs_tree_block_info *info;
998 info = (struct btrfs_tree_block_info *)ptr;
999 *info_level = btrfs_tree_block_level(leaf, info);
1000 ptr += sizeof(struct btrfs_tree_block_info);
1002 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
1003 *info_level = found_key.offset;
1005 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1009 struct btrfs_extent_inline_ref *iref;
1013 iref = (struct btrfs_extent_inline_ref *)ptr;
1014 type = btrfs_get_extent_inline_ref_type(leaf, iref,
1015 BTRFS_REF_TYPE_ANY);
1016 if (type == BTRFS_REF_TYPE_INVALID)
1019 offset = btrfs_extent_inline_ref_offset(leaf, iref);
1022 case BTRFS_SHARED_BLOCK_REF_KEY:
1023 ret = add_direct_ref(fs_info, preftrees,
1024 *info_level + 1, offset,
1025 bytenr, 1, NULL, GFP_NOFS);
1027 case BTRFS_SHARED_DATA_REF_KEY: {
1028 struct btrfs_shared_data_ref *sdref;
1031 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
1032 count = btrfs_shared_data_ref_count(leaf, sdref);
1034 ret = add_direct_ref(fs_info, preftrees, 0, offset,
1035 bytenr, count, sc, GFP_NOFS);
1038 case BTRFS_TREE_BLOCK_REF_KEY:
1039 ret = add_indirect_ref(fs_info, preftrees, offset,
1040 NULL, *info_level + 1,
1041 bytenr, 1, NULL, GFP_NOFS);
1043 case BTRFS_EXTENT_DATA_REF_KEY: {
1044 struct btrfs_extent_data_ref *dref;
1048 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1049 count = btrfs_extent_data_ref_count(leaf, dref);
1050 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1052 key.type = BTRFS_EXTENT_DATA_KEY;
1053 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1055 if (sc && sc->inum && key.objectid != sc->inum &&
1056 !sc->have_delayed_delete_refs) {
1057 ret = BACKREF_FOUND_SHARED;
1061 root = btrfs_extent_data_ref_root(leaf, dref);
1063 ret = add_indirect_ref(fs_info, preftrees, root,
1064 &key, 0, bytenr, count,
1074 ptr += btrfs_extent_inline_ref_size(type);
1081 * add all non-inline backrefs for bytenr to the list
1083 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1085 static int add_keyed_refs(struct btrfs_root *extent_root,
1086 struct btrfs_path *path, u64 bytenr,
1087 int info_level, struct preftrees *preftrees,
1088 struct share_check *sc)
1090 struct btrfs_fs_info *fs_info = extent_root->fs_info;
1093 struct extent_buffer *leaf;
1094 struct btrfs_key key;
1097 ret = btrfs_next_item(extent_root, path);
1105 slot = path->slots[0];
1106 leaf = path->nodes[0];
1107 btrfs_item_key_to_cpu(leaf, &key, slot);
1109 if (key.objectid != bytenr)
1111 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1113 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1117 case BTRFS_SHARED_BLOCK_REF_KEY:
1118 /* SHARED DIRECT METADATA backref */
1119 ret = add_direct_ref(fs_info, preftrees,
1120 info_level + 1, key.offset,
1121 bytenr, 1, NULL, GFP_NOFS);
1123 case BTRFS_SHARED_DATA_REF_KEY: {
1124 /* SHARED DIRECT FULL backref */
1125 struct btrfs_shared_data_ref *sdref;
1128 sdref = btrfs_item_ptr(leaf, slot,
1129 struct btrfs_shared_data_ref);
1130 count = btrfs_shared_data_ref_count(leaf, sdref);
1131 ret = add_direct_ref(fs_info, preftrees, 0,
1132 key.offset, bytenr, count,
1136 case BTRFS_TREE_BLOCK_REF_KEY:
1137 /* NORMAL INDIRECT METADATA backref */
1138 ret = add_indirect_ref(fs_info, preftrees, key.offset,
1139 NULL, info_level + 1, bytenr,
1142 case BTRFS_EXTENT_DATA_REF_KEY: {
1143 /* NORMAL INDIRECT DATA backref */
1144 struct btrfs_extent_data_ref *dref;
1148 dref = btrfs_item_ptr(leaf, slot,
1149 struct btrfs_extent_data_ref);
1150 count = btrfs_extent_data_ref_count(leaf, dref);
1151 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1153 key.type = BTRFS_EXTENT_DATA_KEY;
1154 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1156 if (sc && sc->inum && key.objectid != sc->inum &&
1157 !sc->have_delayed_delete_refs) {
1158 ret = BACKREF_FOUND_SHARED;
1162 root = btrfs_extent_data_ref_root(leaf, dref);
1163 ret = add_indirect_ref(fs_info, preftrees, root,
1164 &key, 0, bytenr, count,
1180 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1181 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1182 * indirect refs to their parent bytenr.
1183 * When roots are found, they're added to the roots list
1185 * If time_seq is set to BTRFS_SEQ_LAST, it will not search delayed_refs, and
1186 * behave much like trans == NULL case, the difference only lies in it will not
1188 * The special case is for qgroup to search roots in commit_transaction().
1190 * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1191 * shared extent is detected.
1193 * Otherwise this returns 0 for success and <0 for an error.
1195 * If ignore_offset is set to false, only extent refs whose offsets match
1196 * extent_item_pos are returned. If true, every extent ref is returned
1197 * and extent_item_pos is ignored.
1199 * FIXME some caching might speed things up
1201 static int find_parent_nodes(struct btrfs_trans_handle *trans,
1202 struct btrfs_fs_info *fs_info, u64 bytenr,
1203 u64 time_seq, struct ulist *refs,
1204 struct ulist *roots, const u64 *extent_item_pos,
1205 struct share_check *sc, bool ignore_offset)
1207 struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
1208 struct btrfs_key key;
1209 struct btrfs_path *path;
1210 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1211 struct btrfs_delayed_ref_head *head;
1214 struct prelim_ref *ref;
1215 struct rb_node *node;
1216 struct extent_inode_elem *eie = NULL;
1217 struct preftrees preftrees = {
1218 .direct = PREFTREE_INIT,
1219 .indirect = PREFTREE_INIT,
1220 .indirect_missing_keys = PREFTREE_INIT
1223 key.objectid = bytenr;
1224 key.offset = (u64)-1;
1225 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1226 key.type = BTRFS_METADATA_ITEM_KEY;
1228 key.type = BTRFS_EXTENT_ITEM_KEY;
1230 path = btrfs_alloc_path();
1234 path->search_commit_root = 1;
1235 path->skip_locking = 1;
1238 if (time_seq == BTRFS_SEQ_LAST)
1239 path->skip_locking = 1;
1244 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1248 /* This shouldn't happen, indicates a bug or fs corruption. */
1254 if (trans && likely(trans->type != __TRANS_DUMMY) &&
1255 time_seq != BTRFS_SEQ_LAST) {
1257 * We have a specific time_seq we care about and trans which
1258 * means we have the path lock, we need to grab the ref head and
1259 * lock it so we have a consistent view of the refs at the given
1262 delayed_refs = &trans->transaction->delayed_refs;
1263 spin_lock(&delayed_refs->lock);
1264 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1266 if (!mutex_trylock(&head->mutex)) {
1267 refcount_inc(&head->refs);
1268 spin_unlock(&delayed_refs->lock);
1270 btrfs_release_path(path);
1273 * Mutex was contended, block until it's
1274 * released and try again
1276 mutex_lock(&head->mutex);
1277 mutex_unlock(&head->mutex);
1278 btrfs_put_delayed_ref_head(head);
1281 spin_unlock(&delayed_refs->lock);
1282 ret = add_delayed_refs(fs_info, head, time_seq,
1284 mutex_unlock(&head->mutex);
1288 spin_unlock(&delayed_refs->lock);
1292 if (path->slots[0]) {
1293 struct extent_buffer *leaf;
1297 leaf = path->nodes[0];
1298 slot = path->slots[0];
1299 btrfs_item_key_to_cpu(leaf, &key, slot);
1300 if (key.objectid == bytenr &&
1301 (key.type == BTRFS_EXTENT_ITEM_KEY ||
1302 key.type == BTRFS_METADATA_ITEM_KEY)) {
1303 ret = add_inline_refs(fs_info, path, bytenr,
1304 &info_level, &preftrees, sc);
1307 ret = add_keyed_refs(root, path, bytenr, info_level,
1314 btrfs_release_path(path);
1316 ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
1320 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1322 ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1323 extent_item_pos, sc, ignore_offset);
1327 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1330 * This walks the tree of merged and resolved refs. Tree blocks are
1331 * read in as needed. Unique entries are added to the ulist, and
1332 * the list of found roots is updated.
1334 * We release the entire tree in one go before returning.
1336 node = rb_first_cached(&preftrees.direct.root);
1338 ref = rb_entry(node, struct prelim_ref, rbnode);
1339 node = rb_next(&ref->rbnode);
1341 * ref->count < 0 can happen here if there are delayed
1342 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1343 * prelim_ref_insert() relies on this when merging
1344 * identical refs to keep the overall count correct.
1345 * prelim_ref_insert() will merge only those refs
1346 * which compare identically. Any refs having
1347 * e.g. different offsets would not be merged,
1348 * and would retain their original ref->count < 0.
1350 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1351 if (sc && sc->root_objectid &&
1352 ref->root_id != sc->root_objectid) {
1353 ret = BACKREF_FOUND_SHARED;
1357 /* no parent == root of tree */
1358 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1362 if (ref->count && ref->parent) {
1363 if (extent_item_pos && !ref->inode_list &&
1365 struct extent_buffer *eb;
1367 eb = read_tree_block(fs_info, ref->parent, 0,
1368 0, ref->level, NULL);
1373 if (!extent_buffer_uptodate(eb)) {
1374 free_extent_buffer(eb);
1379 if (!path->skip_locking)
1380 btrfs_tree_read_lock(eb);
1381 ret = find_extent_in_eb(eb, bytenr,
1382 *extent_item_pos, &eie, ignore_offset);
1383 if (!path->skip_locking)
1384 btrfs_tree_read_unlock(eb);
1385 free_extent_buffer(eb);
1388 ref->inode_list = eie;
1390 * We transferred the list ownership to the ref,
1391 * so set to NULL to avoid a double free in case
1392 * an error happens after this.
1396 ret = ulist_add_merge_ptr(refs, ref->parent,
1398 (void **)&eie, GFP_NOFS);
1401 if (!ret && extent_item_pos) {
1403 * We've recorded that parent, so we must extend
1404 * its inode list here.
1406 * However if there was corruption we may not
1407 * have found an eie, return an error in this
1417 eie->next = ref->inode_list;
1421 * We have transferred the inode list ownership from
1422 * this ref to the ref we added to the 'refs' ulist.
1423 * So set this ref's inode list to NULL to avoid
1424 * use-after-free when our caller uses it or double
1425 * frees in case an error happens before we return.
1427 ref->inode_list = NULL;
1433 btrfs_free_path(path);
1435 prelim_release(&preftrees.direct);
1436 prelim_release(&preftrees.indirect);
1437 prelim_release(&preftrees.indirect_missing_keys);
1440 free_inode_elem_list(eie);
1445 * Finds all leafs with a reference to the specified combination of bytenr and
1446 * offset. key_list_head will point to a list of corresponding keys (caller must
1447 * free each list element). The leafs will be stored in the leafs ulist, which
1448 * must be freed with ulist_free.
1450 * returns 0 on success, <0 on error
1452 int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1453 struct btrfs_fs_info *fs_info, u64 bytenr,
1454 u64 time_seq, struct ulist **leafs,
1455 const u64 *extent_item_pos, bool ignore_offset)
1459 *leafs = ulist_alloc(GFP_NOFS);
1463 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1464 *leafs, NULL, extent_item_pos, NULL, ignore_offset);
1465 if (ret < 0 && ret != -ENOENT) {
1466 free_leaf_list(*leafs);
1474 * walk all backrefs for a given extent to find all roots that reference this
1475 * extent. Walking a backref means finding all extents that reference this
1476 * extent and in turn walk the backrefs of those, too. Naturally this is a
1477 * recursive process, but here it is implemented in an iterative fashion: We
1478 * find all referencing extents for the extent in question and put them on a
1479 * list. In turn, we find all referencing extents for those, further appending
1480 * to the list. The way we iterate the list allows adding more elements after
1481 * the current while iterating. The process stops when we reach the end of the
1482 * list. Found roots are added to the roots list.
1484 * returns 0 on success, < 0 on error.
1486 static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1487 struct btrfs_fs_info *fs_info, u64 bytenr,
1488 u64 time_seq, struct ulist **roots,
1492 struct ulist_node *node = NULL;
1493 struct ulist_iterator uiter;
1496 tmp = ulist_alloc(GFP_NOFS);
1499 *roots = ulist_alloc(GFP_NOFS);
1505 ULIST_ITER_INIT(&uiter);
1507 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1508 tmp, *roots, NULL, NULL, ignore_offset);
1509 if (ret < 0 && ret != -ENOENT) {
1515 node = ulist_next(tmp, &uiter);
1526 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1527 struct btrfs_fs_info *fs_info, u64 bytenr,
1528 u64 time_seq, struct ulist **roots,
1529 bool skip_commit_root_sem)
1533 if (!trans && !skip_commit_root_sem)
1534 down_read(&fs_info->commit_root_sem);
1535 ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1536 time_seq, roots, false);
1537 if (!trans && !skip_commit_root_sem)
1538 up_read(&fs_info->commit_root_sem);
1543 * The caller has joined a transaction or is holding a read lock on the
1544 * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1545 * snapshot field changing while updating or checking the cache.
1547 static bool lookup_backref_shared_cache(struct btrfs_backref_shared_cache *cache,
1548 struct btrfs_root *root,
1549 u64 bytenr, int level, bool *is_shared)
1551 struct btrfs_backref_shared_cache_entry *entry;
1553 if (!cache->use_cache)
1556 if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1560 * Level -1 is used for the data extent, which is not reliable to cache
1561 * because its reference count can increase or decrease without us
1562 * realizing. We cache results only for extent buffers that lead from
1563 * the root node down to the leaf with the file extent item.
1567 entry = &cache->entries[level];
1569 /* Unused cache entry or being used for some other extent buffer. */
1570 if (entry->bytenr != bytenr)
1574 * We cached a false result, but the last snapshot generation of the
1575 * root changed, so we now have a snapshot. Don't trust the result.
1577 if (!entry->is_shared &&
1578 entry->gen != btrfs_root_last_snapshot(&root->root_item))
1582 * If we cached a true result and the last generation used for dropping
1583 * a root changed, we can not trust the result, because the dropped root
1584 * could be a snapshot sharing this extent buffer.
1586 if (entry->is_shared &&
1587 entry->gen != btrfs_get_last_root_drop_gen(root->fs_info))
1590 *is_shared = entry->is_shared;
1592 * If the node at this level is shared, than all nodes below are also
1593 * shared. Currently some of the nodes below may be marked as not shared
1594 * because we have just switched from one leaf to another, and switched
1595 * also other nodes above the leaf and below the current level, so mark
1599 for (int i = 0; i < level; i++) {
1600 cache->entries[i].is_shared = true;
1601 cache->entries[i].gen = entry->gen;
1609 * The caller has joined a transaction or is holding a read lock on the
1610 * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1611 * snapshot field changing while updating or checking the cache.
1613 static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache,
1614 struct btrfs_root *root,
1615 u64 bytenr, int level, bool is_shared)
1617 struct btrfs_backref_shared_cache_entry *entry;
1620 if (!cache->use_cache)
1623 if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1627 * Level -1 is used for the data extent, which is not reliable to cache
1628 * because its reference count can increase or decrease without us
1629 * realizing. We cache results only for extent buffers that lead from
1630 * the root node down to the leaf with the file extent item.
1635 gen = btrfs_get_last_root_drop_gen(root->fs_info);
1637 gen = btrfs_root_last_snapshot(&root->root_item);
1639 entry = &cache->entries[level];
1640 entry->bytenr = bytenr;
1641 entry->is_shared = is_shared;
1645 * If we found an extent buffer is shared, set the cache result for all
1646 * extent buffers below it to true. As nodes in the path are COWed,
1647 * their sharedness is moved to their children, and if a leaf is COWed,
1648 * then the sharedness of a data extent becomes direct, the refcount of
1649 * data extent is increased in the extent item at the extent tree.
1652 for (int i = 0; i < level; i++) {
1653 entry = &cache->entries[i];
1654 entry->is_shared = is_shared;
1661 * Check if a data extent is shared or not.
1663 * @root: The root the inode belongs to.
1664 * @inum: Number of the inode whose extent we are checking.
1665 * @bytenr: Logical bytenr of the extent we are checking.
1666 * @extent_gen: Generation of the extent (file extent item) or 0 if it is
1668 * @roots: List of roots this extent is shared among.
1669 * @tmp: Temporary list used for iteration.
1670 * @cache: A backref lookup result cache.
1672 * btrfs_is_data_extent_shared uses the backref walking code but will short
1673 * circuit as soon as it finds a root or inode that doesn't match the
1674 * one passed in. This provides a significant performance benefit for
1675 * callers (such as fiemap) which want to know whether the extent is
1676 * shared but do not need a ref count.
1678 * This attempts to attach to the running transaction in order to account for
1679 * delayed refs, but continues on even when no running transaction exists.
1681 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1683 int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
1685 struct ulist *roots, struct ulist *tmp,
1686 struct btrfs_backref_shared_cache *cache)
1688 struct btrfs_fs_info *fs_info = root->fs_info;
1689 struct btrfs_trans_handle *trans;
1690 struct ulist_iterator uiter;
1691 struct ulist_node *node;
1692 struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem);
1694 struct share_check shared = {
1695 .root_objectid = root->root_key.objectid,
1698 .have_delayed_delete_refs = false,
1705 trans = btrfs_join_transaction_nostart(root);
1706 if (IS_ERR(trans)) {
1707 if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1708 ret = PTR_ERR(trans);
1712 down_read(&fs_info->commit_root_sem);
1714 btrfs_get_tree_mod_seq(fs_info, &elem);
1717 /* -1 means we are in the bytenr of the data extent. */
1719 ULIST_ITER_INIT(&uiter);
1720 cache->use_cache = true;
1725 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1726 roots, NULL, &shared, false);
1727 if (ret == BACKREF_FOUND_SHARED) {
1728 /* this is the only condition under which we return 1 */
1731 store_backref_shared_cache(cache, root, bytenr,
1735 if (ret < 0 && ret != -ENOENT)
1739 * If our data extent is not shared through reflinks and it was
1740 * created in a generation after the last one used to create a
1741 * snapshot of the inode's root, then it can not be shared
1742 * indirectly through subtrees, as that can only happen with
1743 * snapshots. In this case bail out, no need to check for the
1744 * sharedness of extent buffers.
1747 extent_gen > btrfs_root_last_snapshot(&root->root_item))
1751 * If our data extent was not directly shared (without multiple
1752 * reference items), than it might have a single reference item
1753 * with a count > 1 for the same offset, which means there are 2
1754 * (or more) file extent items that point to the data extent -
1755 * this happens when a file extent item needs to be split and
1756 * then one item gets moved to another leaf due to a b+tree leaf
1757 * split when inserting some item. In this case the file extent
1758 * items may be located in different leaves and therefore some
1759 * of the leaves may be referenced through shared subtrees while
1760 * others are not. Since our extent buffer cache only works for
1761 * a single path (by far the most common case and simpler to
1762 * deal with), we can not use it if we have multiple leaves
1763 * (which implies multiple paths).
1765 if (level == -1 && tmp->nnodes > 1)
1766 cache->use_cache = false;
1769 store_backref_shared_cache(cache, root, bytenr,
1771 node = ulist_next(tmp, &uiter);
1776 cached = lookup_backref_shared_cache(cache, root, bytenr, level,
1779 ret = (is_shared ? 1 : 0);
1782 shared.share_count = 0;
1783 shared.have_delayed_delete_refs = false;
1788 btrfs_put_tree_mod_seq(fs_info, &elem);
1789 btrfs_end_transaction(trans);
1791 up_read(&fs_info->commit_root_sem);
1794 ulist_release(roots);
1799 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1800 u64 start_off, struct btrfs_path *path,
1801 struct btrfs_inode_extref **ret_extref,
1805 struct btrfs_key key;
1806 struct btrfs_key found_key;
1807 struct btrfs_inode_extref *extref;
1808 const struct extent_buffer *leaf;
1811 key.objectid = inode_objectid;
1812 key.type = BTRFS_INODE_EXTREF_KEY;
1813 key.offset = start_off;
1815 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1820 leaf = path->nodes[0];
1821 slot = path->slots[0];
1822 if (slot >= btrfs_header_nritems(leaf)) {
1824 * If the item at offset is not found,
1825 * btrfs_search_slot will point us to the slot
1826 * where it should be inserted. In our case
1827 * that will be the slot directly before the
1828 * next INODE_REF_KEY_V2 item. In the case
1829 * that we're pointing to the last slot in a
1830 * leaf, we must move one leaf over.
1832 ret = btrfs_next_leaf(root, path);
1841 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1844 * Check that we're still looking at an extended ref key for
1845 * this particular objectid. If we have different
1846 * objectid or type then there are no more to be found
1847 * in the tree and we can exit.
1850 if (found_key.objectid != inode_objectid)
1852 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1856 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1857 extref = (struct btrfs_inode_extref *)ptr;
1858 *ret_extref = extref;
1860 *found_off = found_key.offset;
1868 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1869 * Elements of the path are separated by '/' and the path is guaranteed to be
1870 * 0-terminated. the path is only given within the current file system.
1871 * Therefore, it never starts with a '/'. the caller is responsible to provide
1872 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1873 * the start point of the resulting string is returned. this pointer is within
1875 * in case the path buffer would overflow, the pointer is decremented further
1876 * as if output was written to the buffer, though no more output is actually
1877 * generated. that way, the caller can determine how much space would be
1878 * required for the path to fit into the buffer. in that case, the returned
1879 * value will be smaller than dest. callers must check this!
1881 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1882 u32 name_len, unsigned long name_off,
1883 struct extent_buffer *eb_in, u64 parent,
1884 char *dest, u32 size)
1889 s64 bytes_left = ((s64)size) - 1;
1890 struct extent_buffer *eb = eb_in;
1891 struct btrfs_key found_key;
1892 struct btrfs_inode_ref *iref;
1894 if (bytes_left >= 0)
1895 dest[bytes_left] = '\0';
1898 bytes_left -= name_len;
1899 if (bytes_left >= 0)
1900 read_extent_buffer(eb, dest + bytes_left,
1901 name_off, name_len);
1903 if (!path->skip_locking)
1904 btrfs_tree_read_unlock(eb);
1905 free_extent_buffer(eb);
1907 ret = btrfs_find_item(fs_root, path, parent, 0,
1908 BTRFS_INODE_REF_KEY, &found_key);
1914 next_inum = found_key.offset;
1916 /* regular exit ahead */
1917 if (parent == next_inum)
1920 slot = path->slots[0];
1921 eb = path->nodes[0];
1922 /* make sure we can use eb after releasing the path */
1924 path->nodes[0] = NULL;
1927 btrfs_release_path(path);
1928 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1930 name_len = btrfs_inode_ref_name_len(eb, iref);
1931 name_off = (unsigned long)(iref + 1);
1935 if (bytes_left >= 0)
1936 dest[bytes_left] = '/';
1939 btrfs_release_path(path);
1942 return ERR_PTR(ret);
1944 return dest + bytes_left;
1948 * this makes the path point to (logical EXTENT_ITEM *)
1949 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1950 * tree blocks and <0 on error.
1952 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1953 struct btrfs_path *path, struct btrfs_key *found_key,
1956 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical);
1961 const struct extent_buffer *eb;
1962 struct btrfs_extent_item *ei;
1963 struct btrfs_key key;
1965 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1966 key.type = BTRFS_METADATA_ITEM_KEY;
1968 key.type = BTRFS_EXTENT_ITEM_KEY;
1969 key.objectid = logical;
1970 key.offset = (u64)-1;
1972 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1976 ret = btrfs_previous_extent_item(extent_root, path, 0);
1982 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1983 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1984 size = fs_info->nodesize;
1985 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1986 size = found_key->offset;
1988 if (found_key->objectid > logical ||
1989 found_key->objectid + size <= logical) {
1990 btrfs_debug(fs_info,
1991 "logical %llu is not within any extent", logical);
1995 eb = path->nodes[0];
1996 item_size = btrfs_item_size(eb, path->slots[0]);
1997 BUG_ON(item_size < sizeof(*ei));
1999 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
2000 flags = btrfs_extent_flags(eb, ei);
2002 btrfs_debug(fs_info,
2003 "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
2004 logical, logical - found_key->objectid, found_key->objectid,
2005 found_key->offset, flags, item_size);
2007 WARN_ON(!flags_ret);
2009 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2010 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
2011 else if (flags & BTRFS_EXTENT_FLAG_DATA)
2012 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
2022 * helper function to iterate extent inline refs. ptr must point to a 0 value
2023 * for the first call and may be modified. it is used to track state.
2024 * if more refs exist, 0 is returned and the next call to
2025 * get_extent_inline_ref must pass the modified ptr parameter to get the
2026 * next ref. after the last ref was processed, 1 is returned.
2027 * returns <0 on error
2029 static int get_extent_inline_ref(unsigned long *ptr,
2030 const struct extent_buffer *eb,
2031 const struct btrfs_key *key,
2032 const struct btrfs_extent_item *ei,
2034 struct btrfs_extent_inline_ref **out_eiref,
2039 struct btrfs_tree_block_info *info;
2043 flags = btrfs_extent_flags(eb, ei);
2044 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2045 if (key->type == BTRFS_METADATA_ITEM_KEY) {
2046 /* a skinny metadata extent */
2048 (struct btrfs_extent_inline_ref *)(ei + 1);
2050 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
2051 info = (struct btrfs_tree_block_info *)(ei + 1);
2053 (struct btrfs_extent_inline_ref *)(info + 1);
2056 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
2058 *ptr = (unsigned long)*out_eiref;
2059 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
2063 end = (unsigned long)ei + item_size;
2064 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
2065 *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
2066 BTRFS_REF_TYPE_ANY);
2067 if (*out_type == BTRFS_REF_TYPE_INVALID)
2070 *ptr += btrfs_extent_inline_ref_size(*out_type);
2071 WARN_ON(*ptr > end);
2073 return 1; /* last */
2079 * reads the tree block backref for an extent. tree level and root are returned
2080 * through out_level and out_root. ptr must point to a 0 value for the first
2081 * call and may be modified (see get_extent_inline_ref comment).
2082 * returns 0 if data was provided, 1 if there was no more data to provide or
2085 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
2086 struct btrfs_key *key, struct btrfs_extent_item *ei,
2087 u32 item_size, u64 *out_root, u8 *out_level)
2091 struct btrfs_extent_inline_ref *eiref;
2093 if (*ptr == (unsigned long)-1)
2097 ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
2102 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
2103 type == BTRFS_SHARED_BLOCK_REF_KEY)
2110 /* we can treat both ref types equally here */
2111 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
2113 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
2114 struct btrfs_tree_block_info *info;
2116 info = (struct btrfs_tree_block_info *)(ei + 1);
2117 *out_level = btrfs_tree_block_level(eb, info);
2119 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
2120 *out_level = (u8)key->offset;
2124 *ptr = (unsigned long)-1;
2129 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
2130 struct extent_inode_elem *inode_list,
2131 u64 root, u64 extent_item_objectid,
2132 iterate_extent_inodes_t *iterate, void *ctx)
2134 struct extent_inode_elem *eie;
2137 for (eie = inode_list; eie; eie = eie->next) {
2138 btrfs_debug(fs_info,
2139 "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
2140 extent_item_objectid, eie->inum,
2142 ret = iterate(eie->inum, eie->offset, root, ctx);
2144 btrfs_debug(fs_info,
2145 "stopping iteration for %llu due to ret=%d",
2146 extent_item_objectid, ret);
2155 * calls iterate() for every inode that references the extent identified by
2156 * the given parameters.
2157 * when the iterator function returns a non-zero value, iteration stops.
2159 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
2160 u64 extent_item_objectid, u64 extent_item_pos,
2161 int search_commit_root,
2162 iterate_extent_inodes_t *iterate, void *ctx,
2166 struct btrfs_trans_handle *trans = NULL;
2167 struct ulist *refs = NULL;
2168 struct ulist *roots = NULL;
2169 struct ulist_node *ref_node = NULL;
2170 struct ulist_node *root_node = NULL;
2171 struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem);
2172 struct ulist_iterator ref_uiter;
2173 struct ulist_iterator root_uiter;
2175 btrfs_debug(fs_info, "resolving all inodes for extent %llu",
2176 extent_item_objectid);
2178 if (!search_commit_root) {
2179 trans = btrfs_attach_transaction(fs_info->tree_root);
2180 if (IS_ERR(trans)) {
2181 if (PTR_ERR(trans) != -ENOENT &&
2182 PTR_ERR(trans) != -EROFS)
2183 return PTR_ERR(trans);
2189 btrfs_get_tree_mod_seq(fs_info, &seq_elem);
2191 down_read(&fs_info->commit_root_sem);
2193 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
2194 seq_elem.seq, &refs,
2195 &extent_item_pos, ignore_offset);
2199 ULIST_ITER_INIT(&ref_uiter);
2200 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
2201 ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
2202 seq_elem.seq, &roots,
2206 ULIST_ITER_INIT(&root_uiter);
2207 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
2208 btrfs_debug(fs_info,
2209 "root %llu references leaf %llu, data list %#llx",
2210 root_node->val, ref_node->val,
2212 ret = iterate_leaf_refs(fs_info,
2213 (struct extent_inode_elem *)
2214 (uintptr_t)ref_node->aux,
2216 extent_item_objectid,
2222 free_leaf_list(refs);
2225 btrfs_put_tree_mod_seq(fs_info, &seq_elem);
2226 btrfs_end_transaction(trans);
2228 up_read(&fs_info->commit_root_sem);
2234 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
2236 struct btrfs_data_container *inodes = ctx;
2237 const size_t c = 3 * sizeof(u64);
2239 if (inodes->bytes_left >= c) {
2240 inodes->bytes_left -= c;
2241 inodes->val[inodes->elem_cnt] = inum;
2242 inodes->val[inodes->elem_cnt + 1] = offset;
2243 inodes->val[inodes->elem_cnt + 2] = root;
2244 inodes->elem_cnt += 3;
2246 inodes->bytes_missing += c - inodes->bytes_left;
2247 inodes->bytes_left = 0;
2248 inodes->elem_missed += 3;
2254 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2255 struct btrfs_path *path,
2256 void *ctx, bool ignore_offset)
2259 u64 extent_item_pos;
2261 struct btrfs_key found_key;
2262 int search_commit_root = path->search_commit_root;
2264 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2265 btrfs_release_path(path);
2268 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2271 extent_item_pos = logical - found_key.objectid;
2272 ret = iterate_extent_inodes(fs_info, found_key.objectid,
2273 extent_item_pos, search_commit_root,
2274 build_ino_list, ctx, ignore_offset);
2279 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2280 struct extent_buffer *eb, struct inode_fs_paths *ipath);
2282 static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath)
2291 struct btrfs_root *fs_root = ipath->fs_root;
2292 struct btrfs_path *path = ipath->btrfs_path;
2293 struct extent_buffer *eb;
2294 struct btrfs_inode_ref *iref;
2295 struct btrfs_key found_key;
2298 ret = btrfs_find_item(fs_root, path, inum,
2299 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2305 ret = found ? 0 : -ENOENT;
2310 parent = found_key.offset;
2311 slot = path->slots[0];
2312 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2317 btrfs_release_path(path);
2319 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2321 for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) {
2322 name_len = btrfs_inode_ref_name_len(eb, iref);
2323 /* path must be released before calling iterate()! */
2324 btrfs_debug(fs_root->fs_info,
2325 "following ref at offset %u for inode %llu in tree %llu",
2326 cur, found_key.objectid,
2327 fs_root->root_key.objectid);
2328 ret = inode_to_path(parent, name_len,
2329 (unsigned long)(iref + 1), eb, ipath);
2332 len = sizeof(*iref) + name_len;
2333 iref = (struct btrfs_inode_ref *)((char *)iref + len);
2335 free_extent_buffer(eb);
2338 btrfs_release_path(path);
2343 static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath)
2350 struct btrfs_root *fs_root = ipath->fs_root;
2351 struct btrfs_path *path = ipath->btrfs_path;
2352 struct extent_buffer *eb;
2353 struct btrfs_inode_extref *extref;
2359 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2364 ret = found ? 0 : -ENOENT;
2369 slot = path->slots[0];
2370 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2375 btrfs_release_path(path);
2377 item_size = btrfs_item_size(eb, slot);
2378 ptr = btrfs_item_ptr_offset(eb, slot);
2381 while (cur_offset < item_size) {
2384 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2385 parent = btrfs_inode_extref_parent(eb, extref);
2386 name_len = btrfs_inode_extref_name_len(eb, extref);
2387 ret = inode_to_path(parent, name_len,
2388 (unsigned long)&extref->name, eb, ipath);
2392 cur_offset += btrfs_inode_extref_name_len(eb, extref);
2393 cur_offset += sizeof(*extref);
2395 free_extent_buffer(eb);
2400 btrfs_release_path(path);
2406 * returns 0 if the path could be dumped (probably truncated)
2407 * returns <0 in case of an error
2409 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2410 struct extent_buffer *eb, struct inode_fs_paths *ipath)
2414 int i = ipath->fspath->elem_cnt;
2415 const int s_ptr = sizeof(char *);
2418 bytes_left = ipath->fspath->bytes_left > s_ptr ?
2419 ipath->fspath->bytes_left - s_ptr : 0;
2421 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2422 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2423 name_off, eb, inum, fspath_min, bytes_left);
2425 return PTR_ERR(fspath);
2427 if (fspath > fspath_min) {
2428 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2429 ++ipath->fspath->elem_cnt;
2430 ipath->fspath->bytes_left = fspath - fspath_min;
2432 ++ipath->fspath->elem_missed;
2433 ipath->fspath->bytes_missing += fspath_min - fspath;
2434 ipath->fspath->bytes_left = 0;
2441 * this dumps all file system paths to the inode into the ipath struct, provided
2442 * is has been created large enough. each path is zero-terminated and accessed
2443 * from ipath->fspath->val[i].
2444 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2445 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2446 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2447 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2448 * have been needed to return all paths.
2450 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2455 ret = iterate_inode_refs(inum, ipath);
2458 else if (ret != -ENOENT)
2461 ret = iterate_inode_extrefs(inum, ipath);
2462 if (ret == -ENOENT && found_refs)
2468 struct btrfs_data_container *init_data_container(u32 total_bytes)
2470 struct btrfs_data_container *data;
2473 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2474 data = kvmalloc(alloc_bytes, GFP_KERNEL);
2476 return ERR_PTR(-ENOMEM);
2478 if (total_bytes >= sizeof(*data)) {
2479 data->bytes_left = total_bytes - sizeof(*data);
2480 data->bytes_missing = 0;
2482 data->bytes_missing = sizeof(*data) - total_bytes;
2483 data->bytes_left = 0;
2487 data->elem_missed = 0;
2493 * allocates space to return multiple file system paths for an inode.
2494 * total_bytes to allocate are passed, note that space usable for actual path
2495 * information will be total_bytes - sizeof(struct inode_fs_paths).
2496 * the returned pointer must be freed with free_ipath() in the end.
2498 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2499 struct btrfs_path *path)
2501 struct inode_fs_paths *ifp;
2502 struct btrfs_data_container *fspath;
2504 fspath = init_data_container(total_bytes);
2506 return ERR_CAST(fspath);
2508 ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2511 return ERR_PTR(-ENOMEM);
2514 ifp->btrfs_path = path;
2515 ifp->fspath = fspath;
2516 ifp->fs_root = fs_root;
2521 void free_ipath(struct inode_fs_paths *ipath)
2525 kvfree(ipath->fspath);
2529 struct btrfs_backref_iter *btrfs_backref_iter_alloc(
2530 struct btrfs_fs_info *fs_info, gfp_t gfp_flag)
2532 struct btrfs_backref_iter *ret;
2534 ret = kzalloc(sizeof(*ret), gfp_flag);
2538 ret->path = btrfs_alloc_path();
2544 /* Current backref iterator only supports iteration in commit root */
2545 ret->path->search_commit_root = 1;
2546 ret->path->skip_locking = 1;
2547 ret->fs_info = fs_info;
2552 int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2554 struct btrfs_fs_info *fs_info = iter->fs_info;
2555 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2556 struct btrfs_path *path = iter->path;
2557 struct btrfs_extent_item *ei;
2558 struct btrfs_key key;
2561 key.objectid = bytenr;
2562 key.type = BTRFS_METADATA_ITEM_KEY;
2563 key.offset = (u64)-1;
2564 iter->bytenr = bytenr;
2566 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2573 if (path->slots[0] == 0) {
2574 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2580 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2581 if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2582 key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2586 memcpy(&iter->cur_key, &key, sizeof(key));
2587 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2589 iter->end_ptr = (u32)(iter->item_ptr +
2590 btrfs_item_size(path->nodes[0], path->slots[0]));
2591 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2592 struct btrfs_extent_item);
2595 * Only support iteration on tree backref yet.
2597 * This is an extra precaution for non skinny-metadata, where
2598 * EXTENT_ITEM is also used for tree blocks, that we can only use
2599 * extent flags to determine if it's a tree block.
2601 if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2605 iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2607 /* If there is no inline backref, go search for keyed backref */
2608 if (iter->cur_ptr >= iter->end_ptr) {
2609 ret = btrfs_next_item(extent_root, path);
2611 /* No inline nor keyed ref */
2619 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2621 if (iter->cur_key.objectid != bytenr ||
2622 (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2623 iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2627 iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2629 iter->item_ptr = iter->cur_ptr;
2630 iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size(
2631 path->nodes[0], path->slots[0]));
2636 btrfs_backref_iter_release(iter);
2641 * Go to the next backref item of current bytenr, can be either inlined or
2644 * Caller needs to check whether it's inline ref or not by iter->cur_key.
2646 * Return 0 if we get next backref without problem.
2647 * Return >0 if there is no extra backref for this bytenr.
2648 * Return <0 if there is something wrong happened.
2650 int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2652 struct extent_buffer *eb = btrfs_backref_get_eb(iter);
2653 struct btrfs_root *extent_root;
2654 struct btrfs_path *path = iter->path;
2655 struct btrfs_extent_inline_ref *iref;
2659 if (btrfs_backref_iter_is_inline_ref(iter)) {
2660 /* We're still inside the inline refs */
2661 ASSERT(iter->cur_ptr < iter->end_ptr);
2663 if (btrfs_backref_has_tree_block_info(iter)) {
2664 /* First tree block info */
2665 size = sizeof(struct btrfs_tree_block_info);
2667 /* Use inline ref type to determine the size */
2670 iref = (struct btrfs_extent_inline_ref *)
2671 ((unsigned long)iter->cur_ptr);
2672 type = btrfs_extent_inline_ref_type(eb, iref);
2674 size = btrfs_extent_inline_ref_size(type);
2676 iter->cur_ptr += size;
2677 if (iter->cur_ptr < iter->end_ptr)
2680 /* All inline items iterated, fall through */
2683 /* We're at keyed items, there is no inline item, go to the next one */
2684 extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr);
2685 ret = btrfs_next_item(extent_root, iter->path);
2689 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
2690 if (iter->cur_key.objectid != iter->bytenr ||
2691 (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2692 iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
2694 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2696 iter->cur_ptr = iter->item_ptr;
2697 iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0],
2702 void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
2703 struct btrfs_backref_cache *cache, int is_reloc)
2707 cache->rb_root = RB_ROOT;
2708 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2709 INIT_LIST_HEAD(&cache->pending[i]);
2710 INIT_LIST_HEAD(&cache->changed);
2711 INIT_LIST_HEAD(&cache->detached);
2712 INIT_LIST_HEAD(&cache->leaves);
2713 INIT_LIST_HEAD(&cache->pending_edge);
2714 INIT_LIST_HEAD(&cache->useless_node);
2715 cache->fs_info = fs_info;
2716 cache->is_reloc = is_reloc;
2719 struct btrfs_backref_node *btrfs_backref_alloc_node(
2720 struct btrfs_backref_cache *cache, u64 bytenr, int level)
2722 struct btrfs_backref_node *node;
2724 ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
2725 node = kzalloc(sizeof(*node), GFP_NOFS);
2729 INIT_LIST_HEAD(&node->list);
2730 INIT_LIST_HEAD(&node->upper);
2731 INIT_LIST_HEAD(&node->lower);
2732 RB_CLEAR_NODE(&node->rb_node);
2734 node->level = level;
2735 node->bytenr = bytenr;
2740 struct btrfs_backref_edge *btrfs_backref_alloc_edge(
2741 struct btrfs_backref_cache *cache)
2743 struct btrfs_backref_edge *edge;
2745 edge = kzalloc(sizeof(*edge), GFP_NOFS);
2752 * Drop the backref node from cache, also cleaning up all its
2753 * upper edges and any uncached nodes in the path.
2755 * This cleanup happens bottom up, thus the node should either
2756 * be the lowest node in the cache or a detached node.
2758 void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
2759 struct btrfs_backref_node *node)
2761 struct btrfs_backref_node *upper;
2762 struct btrfs_backref_edge *edge;
2767 BUG_ON(!node->lowest && !node->detached);
2768 while (!list_empty(&node->upper)) {
2769 edge = list_entry(node->upper.next, struct btrfs_backref_edge,
2771 upper = edge->node[UPPER];
2772 list_del(&edge->list[LOWER]);
2773 list_del(&edge->list[UPPER]);
2774 btrfs_backref_free_edge(cache, edge);
2777 * Add the node to leaf node list if no other child block
2780 if (list_empty(&upper->lower)) {
2781 list_add_tail(&upper->lower, &cache->leaves);
2786 btrfs_backref_drop_node(cache, node);
2790 * Release all nodes/edges from current cache
2792 void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
2794 struct btrfs_backref_node *node;
2797 while (!list_empty(&cache->detached)) {
2798 node = list_entry(cache->detached.next,
2799 struct btrfs_backref_node, list);
2800 btrfs_backref_cleanup_node(cache, node);
2803 while (!list_empty(&cache->leaves)) {
2804 node = list_entry(cache->leaves.next,
2805 struct btrfs_backref_node, lower);
2806 btrfs_backref_cleanup_node(cache, node);
2809 cache->last_trans = 0;
2811 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2812 ASSERT(list_empty(&cache->pending[i]));
2813 ASSERT(list_empty(&cache->pending_edge));
2814 ASSERT(list_empty(&cache->useless_node));
2815 ASSERT(list_empty(&cache->changed));
2816 ASSERT(list_empty(&cache->detached));
2817 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
2818 ASSERT(!cache->nr_nodes);
2819 ASSERT(!cache->nr_edges);
2823 * Handle direct tree backref
2825 * Direct tree backref means, the backref item shows its parent bytenr
2826 * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
2828 * @ref_key: The converted backref key.
2829 * For keyed backref, it's the item key.
2830 * For inlined backref, objectid is the bytenr,
2831 * type is btrfs_inline_ref_type, offset is
2832 * btrfs_inline_ref_offset.
2834 static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
2835 struct btrfs_key *ref_key,
2836 struct btrfs_backref_node *cur)
2838 struct btrfs_backref_edge *edge;
2839 struct btrfs_backref_node *upper;
2840 struct rb_node *rb_node;
2842 ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
2844 /* Only reloc root uses backref pointing to itself */
2845 if (ref_key->objectid == ref_key->offset) {
2846 struct btrfs_root *root;
2848 cur->is_reloc_root = 1;
2849 /* Only reloc backref cache cares about a specific root */
2850 if (cache->is_reloc) {
2851 root = find_reloc_root(cache->fs_info, cur->bytenr);
2857 * For generic purpose backref cache, reloc root node
2860 list_add(&cur->list, &cache->useless_node);
2865 edge = btrfs_backref_alloc_edge(cache);
2869 rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
2871 /* Parent node not yet cached */
2872 upper = btrfs_backref_alloc_node(cache, ref_key->offset,
2875 btrfs_backref_free_edge(cache, edge);
2880 * Backrefs for the upper level block isn't cached, add the
2881 * block to pending list
2883 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2885 /* Parent node already cached */
2886 upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
2887 ASSERT(upper->checked);
2888 INIT_LIST_HEAD(&edge->list[UPPER]);
2890 btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
2895 * Handle indirect tree backref
2897 * Indirect tree backref means, we only know which tree the node belongs to.
2898 * We still need to do a tree search to find out the parents. This is for
2899 * TREE_BLOCK_REF backref (keyed or inlined).
2901 * @ref_key: The same as @ref_key in handle_direct_tree_backref()
2902 * @tree_key: The first key of this tree block.
2903 * @path: A clean (released) path, to avoid allocating path every time
2904 * the function get called.
2906 static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache,
2907 struct btrfs_path *path,
2908 struct btrfs_key *ref_key,
2909 struct btrfs_key *tree_key,
2910 struct btrfs_backref_node *cur)
2912 struct btrfs_fs_info *fs_info = cache->fs_info;
2913 struct btrfs_backref_node *upper;
2914 struct btrfs_backref_node *lower;
2915 struct btrfs_backref_edge *edge;
2916 struct extent_buffer *eb;
2917 struct btrfs_root *root;
2918 struct rb_node *rb_node;
2920 bool need_check = true;
2923 root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
2925 return PTR_ERR(root);
2926 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2929 if (btrfs_root_level(&root->root_item) == cur->level) {
2931 ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
2933 * For reloc backref cache, we may ignore reloc root. But for
2934 * general purpose backref cache, we can't rely on
2935 * btrfs_should_ignore_reloc_root() as it may conflict with
2936 * current running relocation and lead to missing root.
2938 * For general purpose backref cache, reloc root detection is
2939 * completely relying on direct backref (key->offset is parent
2940 * bytenr), thus only do such check for reloc cache.
2942 if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
2943 btrfs_put_root(root);
2944 list_add(&cur->list, &cache->useless_node);
2951 level = cur->level + 1;
2953 /* Search the tree to find parent blocks referring to the block */
2954 path->search_commit_root = 1;
2955 path->skip_locking = 1;
2956 path->lowest_level = level;
2957 ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
2958 path->lowest_level = 0;
2960 btrfs_put_root(root);
2963 if (ret > 0 && path->slots[level] > 0)
2964 path->slots[level]--;
2966 eb = path->nodes[level];
2967 if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
2969 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
2970 cur->bytenr, level - 1, root->root_key.objectid,
2971 tree_key->objectid, tree_key->type, tree_key->offset);
2972 btrfs_put_root(root);
2978 /* Add all nodes and edges in the path */
2979 for (; level < BTRFS_MAX_LEVEL; level++) {
2980 if (!path->nodes[level]) {
2981 ASSERT(btrfs_root_bytenr(&root->root_item) ==
2983 /* Same as previous should_ignore_reloc_root() call */
2984 if (btrfs_should_ignore_reloc_root(root) &&
2986 btrfs_put_root(root);
2987 list_add(&lower->list, &cache->useless_node);
2994 edge = btrfs_backref_alloc_edge(cache);
2996 btrfs_put_root(root);
3001 eb = path->nodes[level];
3002 rb_node = rb_simple_search(&cache->rb_root, eb->start);
3004 upper = btrfs_backref_alloc_node(cache, eb->start,
3007 btrfs_put_root(root);
3008 btrfs_backref_free_edge(cache, edge);
3012 upper->owner = btrfs_header_owner(eb);
3013 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3017 * If we know the block isn't shared we can avoid
3018 * checking its backrefs.
3020 if (btrfs_block_can_be_shared(root, eb))
3026 * Add the block to pending list if we need to check its
3027 * backrefs, we only do this once while walking up a
3028 * tree as we will catch anything else later on.
3030 if (!upper->checked && need_check) {
3032 list_add_tail(&edge->list[UPPER],
3033 &cache->pending_edge);
3037 INIT_LIST_HEAD(&edge->list[UPPER]);
3040 upper = rb_entry(rb_node, struct btrfs_backref_node,
3042 ASSERT(upper->checked);
3043 INIT_LIST_HEAD(&edge->list[UPPER]);
3045 upper->owner = btrfs_header_owner(eb);
3047 btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
3050 btrfs_put_root(root);
3057 btrfs_release_path(path);
3062 * Add backref node @cur into @cache.
3064 * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
3065 * links aren't yet bi-directional. Needs to finish such links.
3066 * Use btrfs_backref_finish_upper_links() to finish such linkage.
3068 * @path: Released path for indirect tree backref lookup
3069 * @iter: Released backref iter for extent tree search
3070 * @node_key: The first key of the tree block
3072 int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache,
3073 struct btrfs_path *path,
3074 struct btrfs_backref_iter *iter,
3075 struct btrfs_key *node_key,
3076 struct btrfs_backref_node *cur)
3078 struct btrfs_fs_info *fs_info = cache->fs_info;
3079 struct btrfs_backref_edge *edge;
3080 struct btrfs_backref_node *exist;
3083 ret = btrfs_backref_iter_start(iter, cur->bytenr);
3087 * We skip the first btrfs_tree_block_info, as we don't use the key
3088 * stored in it, but fetch it from the tree block
3090 if (btrfs_backref_has_tree_block_info(iter)) {
3091 ret = btrfs_backref_iter_next(iter);
3094 /* No extra backref? This means the tree block is corrupted */
3100 WARN_ON(cur->checked);
3101 if (!list_empty(&cur->upper)) {
3103 * The backref was added previously when processing backref of
3104 * type BTRFS_TREE_BLOCK_REF_KEY
3106 ASSERT(list_is_singular(&cur->upper));
3107 edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
3109 ASSERT(list_empty(&edge->list[UPPER]));
3110 exist = edge->node[UPPER];
3112 * Add the upper level block to pending list if we need check
3115 if (!exist->checked)
3116 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3121 for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
3122 struct extent_buffer *eb;
3123 struct btrfs_key key;
3127 eb = btrfs_backref_get_eb(iter);
3129 key.objectid = iter->bytenr;
3130 if (btrfs_backref_iter_is_inline_ref(iter)) {
3131 struct btrfs_extent_inline_ref *iref;
3133 /* Update key for inline backref */
3134 iref = (struct btrfs_extent_inline_ref *)
3135 ((unsigned long)iter->cur_ptr);
3136 type = btrfs_get_extent_inline_ref_type(eb, iref,
3137 BTRFS_REF_TYPE_BLOCK);
3138 if (type == BTRFS_REF_TYPE_INVALID) {
3143 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3145 key.type = iter->cur_key.type;
3146 key.offset = iter->cur_key.offset;
3150 * Parent node found and matches current inline ref, no need to
3151 * rebuild this node for this inline ref
3154 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
3155 exist->owner == key.offset) ||
3156 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
3157 exist->bytenr == key.offset))) {
3162 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
3163 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3164 ret = handle_direct_tree_backref(cache, &key, cur);
3168 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3170 btrfs_print_v0_err(fs_info);
3171 btrfs_handle_fs_error(fs_info, ret, NULL);
3173 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
3178 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
3179 * means the root objectid. We need to search the tree to get
3180 * its parent bytenr.
3182 ret = handle_indirect_tree_backref(cache, path, &key, node_key,
3191 btrfs_backref_iter_release(iter);
3196 * Finish the upwards linkage created by btrfs_backref_add_tree_node()
3198 int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
3199 struct btrfs_backref_node *start)
3201 struct list_head *useless_node = &cache->useless_node;
3202 struct btrfs_backref_edge *edge;
3203 struct rb_node *rb_node;
3204 LIST_HEAD(pending_edge);
3206 ASSERT(start->checked);
3208 /* Insert this node to cache if it's not COW-only */
3209 if (!start->cowonly) {
3210 rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
3213 btrfs_backref_panic(cache->fs_info, start->bytenr,
3215 list_add_tail(&start->lower, &cache->leaves);
3219 * Use breadth first search to iterate all related edges.
3221 * The starting points are all the edges of this node
3223 list_for_each_entry(edge, &start->upper, list[LOWER])
3224 list_add_tail(&edge->list[UPPER], &pending_edge);
3226 while (!list_empty(&pending_edge)) {
3227 struct btrfs_backref_node *upper;
3228 struct btrfs_backref_node *lower;
3230 edge = list_first_entry(&pending_edge,
3231 struct btrfs_backref_edge, list[UPPER]);
3232 list_del_init(&edge->list[UPPER]);
3233 upper = edge->node[UPPER];
3234 lower = edge->node[LOWER];
3236 /* Parent is detached, no need to keep any edges */
3237 if (upper->detached) {
3238 list_del(&edge->list[LOWER]);
3239 btrfs_backref_free_edge(cache, edge);
3241 /* Lower node is orphan, queue for cleanup */
3242 if (list_empty(&lower->upper))
3243 list_add(&lower->list, useless_node);
3248 * All new nodes added in current build_backref_tree() haven't
3249 * been linked to the cache rb tree.
3250 * So if we have upper->rb_node populated, this means a cache
3251 * hit. We only need to link the edge, as @upper and all its
3252 * parents have already been linked.
3254 if (!RB_EMPTY_NODE(&upper->rb_node)) {
3255 if (upper->lowest) {
3256 list_del_init(&upper->lower);
3260 list_add_tail(&edge->list[UPPER], &upper->lower);
3264 /* Sanity check, we shouldn't have any unchecked nodes */
3265 if (!upper->checked) {
3270 /* Sanity check, COW-only node has non-COW-only parent */
3271 if (start->cowonly != upper->cowonly) {
3276 /* Only cache non-COW-only (subvolume trees) tree blocks */
3277 if (!upper->cowonly) {
3278 rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3281 btrfs_backref_panic(cache->fs_info,
3282 upper->bytenr, -EEXIST);
3287 list_add_tail(&edge->list[UPPER], &upper->lower);
3290 * Also queue all the parent edges of this uncached node
3291 * to finish the upper linkage
3293 list_for_each_entry(edge, &upper->upper, list[LOWER])
3294 list_add_tail(&edge->list[UPPER], &pending_edge);
3299 void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3300 struct btrfs_backref_node *node)
3302 struct btrfs_backref_node *lower;
3303 struct btrfs_backref_node *upper;
3304 struct btrfs_backref_edge *edge;
3306 while (!list_empty(&cache->useless_node)) {
3307 lower = list_first_entry(&cache->useless_node,
3308 struct btrfs_backref_node, list);
3309 list_del_init(&lower->list);
3311 while (!list_empty(&cache->pending_edge)) {
3312 edge = list_first_entry(&cache->pending_edge,
3313 struct btrfs_backref_edge, list[UPPER]);
3314 list_del(&edge->list[UPPER]);
3315 list_del(&edge->list[LOWER]);
3316 lower = edge->node[LOWER];
3317 upper = edge->node[UPPER];
3318 btrfs_backref_free_edge(cache, edge);
3321 * Lower is no longer linked to any upper backref nodes and
3322 * isn't in the cache, we can free it ourselves.
3324 if (list_empty(&lower->upper) &&
3325 RB_EMPTY_NODE(&lower->rb_node))
3326 list_add(&lower->list, &cache->useless_node);
3328 if (!RB_EMPTY_NODE(&upper->rb_node))
3331 /* Add this guy's upper edges to the list to process */
3332 list_for_each_entry(edge, &upper->upper, list[LOWER])
3333 list_add_tail(&edge->list[UPPER],
3334 &cache->pending_edge);
3335 if (list_empty(&upper->upper))
3336 list_add(&upper->list, &cache->useless_node);
3339 while (!list_empty(&cache->useless_node)) {
3340 lower = list_first_entry(&cache->useless_node,
3341 struct btrfs_backref_node, list);
3342 list_del_init(&lower->list);
3345 btrfs_backref_drop_node(cache, lower);
3348 btrfs_backref_cleanup_node(cache, node);
3349 ASSERT(list_empty(&cache->useless_node) &&
3350 list_empty(&cache->pending_edge));