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;
439 eb = path->nodes[level];
440 ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
447 * 1. We normally enter this function with the path already pointing to
448 * the first item to check. But sometimes, we may enter it with
450 * 2. We are searching for normal backref but bytenr of this leaf
451 * matches shared data backref
452 * 3. The leaf owner is not equal to the root we are searching
454 * For these cases, go to the next leaf before we continue.
457 if (path->slots[0] >= btrfs_header_nritems(eb) ||
458 is_shared_data_backref(preftrees, eb->start) ||
459 ref->root_id != btrfs_header_owner(eb)) {
460 if (time_seq == BTRFS_SEQ_LAST)
461 ret = btrfs_next_leaf(root, path);
463 ret = btrfs_next_old_leaf(root, path, time_seq);
466 while (!ret && count < ref->count) {
468 slot = path->slots[0];
470 btrfs_item_key_to_cpu(eb, &key, slot);
472 if (key.objectid != key_for_search->objectid ||
473 key.type != BTRFS_EXTENT_DATA_KEY)
477 * We are searching for normal backref but bytenr of this leaf
478 * matches shared data backref, OR
479 * the leaf owner is not equal to the root we are searching for
482 (is_shared_data_backref(preftrees, eb->start) ||
483 ref->root_id != btrfs_header_owner(eb))) {
484 if (time_seq == BTRFS_SEQ_LAST)
485 ret = btrfs_next_leaf(root, path);
487 ret = btrfs_next_old_leaf(root, path, time_seq);
490 fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
491 type = btrfs_file_extent_type(eb, fi);
492 if (type == BTRFS_FILE_EXTENT_INLINE)
494 disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
495 data_offset = btrfs_file_extent_offset(eb, fi);
497 if (disk_byte == wanted_disk_byte) {
500 if (ref->key_for_search.offset == key.offset - data_offset)
504 if (extent_item_pos) {
505 ret = check_extent_in_eb(&key, eb, fi,
507 &eie, ignore_offset);
513 ret = ulist_add_merge_ptr(parents, eb->start,
514 eie, (void **)&old, GFP_NOFS);
517 if (!ret && extent_item_pos) {
525 if (time_seq == BTRFS_SEQ_LAST)
526 ret = btrfs_next_item(root, path);
528 ret = btrfs_next_old_item(root, path, time_seq);
534 free_inode_elem_list(eie);
539 * resolve an indirect backref in the form (root_id, key, level)
540 * to a logical address
542 static int resolve_indirect_ref(struct btrfs_fs_info *fs_info,
543 struct btrfs_path *path, u64 time_seq,
544 struct preftrees *preftrees,
545 struct prelim_ref *ref, struct ulist *parents,
546 const u64 *extent_item_pos, bool ignore_offset)
548 struct btrfs_root *root;
549 struct extent_buffer *eb;
552 int level = ref->level;
553 struct btrfs_key search_key = ref->key_for_search;
556 * If we're search_commit_root we could possibly be holding locks on
557 * other tree nodes. This happens when qgroups does backref walks when
558 * adding new delayed refs. To deal with this we need to look in cache
559 * for the root, and if we don't find it then we need to search the
560 * tree_root's commit root, thus the btrfs_get_fs_root_commit_root usage
563 if (path->search_commit_root)
564 root = btrfs_get_fs_root_commit_root(fs_info, path, ref->root_id);
566 root = btrfs_get_fs_root(fs_info, ref->root_id, false);
572 if (!path->search_commit_root &&
573 test_bit(BTRFS_ROOT_DELETING, &root->state)) {
578 if (btrfs_is_testing(fs_info)) {
583 if (path->search_commit_root)
584 root_level = btrfs_header_level(root->commit_root);
585 else if (time_seq == BTRFS_SEQ_LAST)
586 root_level = btrfs_header_level(root->node);
588 root_level = btrfs_old_root_level(root, time_seq);
590 if (root_level + 1 == level)
594 * We can often find data backrefs with an offset that is too large
595 * (>= LLONG_MAX, maximum allowed file offset) due to underflows when
596 * subtracting a file's offset with the data offset of its
597 * corresponding extent data item. This can happen for example in the
600 * So if we detect such case we set the search key's offset to zero to
601 * make sure we will find the matching file extent item at
602 * add_all_parents(), otherwise we will miss it because the offset
603 * taken form the backref is much larger then the offset of the file
604 * extent item. This can make us scan a very large number of file
605 * extent items, but at least it will not make us miss any.
607 * This is an ugly workaround for a behaviour that should have never
608 * existed, but it does and a fix for the clone ioctl would touch a lot
609 * of places, cause backwards incompatibility and would not fix the
610 * problem for extents cloned with older kernels.
612 if (search_key.type == BTRFS_EXTENT_DATA_KEY &&
613 search_key.offset >= LLONG_MAX)
614 search_key.offset = 0;
615 path->lowest_level = level;
616 if (time_seq == BTRFS_SEQ_LAST)
617 ret = btrfs_search_slot(NULL, root, &search_key, path, 0, 0);
619 ret = btrfs_search_old_slot(root, &search_key, path, time_seq);
622 "search slot in root %llu (level %d, ref count %d) returned %d for key (%llu %u %llu)",
623 ref->root_id, level, ref->count, ret,
624 ref->key_for_search.objectid, ref->key_for_search.type,
625 ref->key_for_search.offset);
629 eb = path->nodes[level];
631 if (WARN_ON(!level)) {
636 eb = path->nodes[level];
639 ret = add_all_parents(root, path, parents, preftrees, ref, level,
640 time_seq, extent_item_pos, ignore_offset);
642 btrfs_put_root(root);
644 path->lowest_level = 0;
645 btrfs_release_path(path);
649 static struct extent_inode_elem *
650 unode_aux_to_inode_list(struct ulist_node *node)
654 return (struct extent_inode_elem *)(uintptr_t)node->aux;
657 static void free_leaf_list(struct ulist *ulist)
659 struct ulist_node *node;
660 struct ulist_iterator uiter;
662 ULIST_ITER_INIT(&uiter);
663 while ((node = ulist_next(ulist, &uiter)))
664 free_inode_elem_list(unode_aux_to_inode_list(node));
670 * We maintain three separate rbtrees: one for direct refs, one for
671 * indirect refs which have a key, and one for indirect refs which do not
672 * have a key. Each tree does merge on insertion.
674 * Once all of the references are located, we iterate over the tree of
675 * indirect refs with missing keys. An appropriate key is located and
676 * the ref is moved onto the tree for indirect refs. After all missing
677 * keys are thus located, we iterate over the indirect ref tree, resolve
678 * each reference, and then insert the resolved reference onto the
679 * direct tree (merging there too).
681 * New backrefs (i.e., for parent nodes) are added to the appropriate
682 * rbtree as they are encountered. The new backrefs are subsequently
685 static int resolve_indirect_refs(struct btrfs_fs_info *fs_info,
686 struct btrfs_path *path, u64 time_seq,
687 struct preftrees *preftrees,
688 const u64 *extent_item_pos,
689 struct share_check *sc, bool ignore_offset)
693 struct ulist *parents;
694 struct ulist_node *node;
695 struct ulist_iterator uiter;
696 struct rb_node *rnode;
698 parents = ulist_alloc(GFP_NOFS);
703 * We could trade memory usage for performance here by iterating
704 * the tree, allocating new refs for each insertion, and then
705 * freeing the entire indirect tree when we're done. In some test
706 * cases, the tree can grow quite large (~200k objects).
708 while ((rnode = rb_first_cached(&preftrees->indirect.root))) {
709 struct prelim_ref *ref;
711 ref = rb_entry(rnode, struct prelim_ref, rbnode);
712 if (WARN(ref->parent,
713 "BUG: direct ref found in indirect tree")) {
718 rb_erase_cached(&ref->rbnode, &preftrees->indirect.root);
719 preftrees->indirect.count--;
721 if (ref->count == 0) {
726 if (sc && sc->root_objectid &&
727 ref->root_id != sc->root_objectid) {
729 ret = BACKREF_FOUND_SHARED;
732 err = resolve_indirect_ref(fs_info, path, time_seq, preftrees,
733 ref, parents, extent_item_pos,
736 * we can only tolerate ENOENT,otherwise,we should catch error
737 * and return directly.
739 if (err == -ENOENT) {
740 prelim_ref_insert(fs_info, &preftrees->direct, ref,
749 /* we put the first parent into the ref at hand */
750 ULIST_ITER_INIT(&uiter);
751 node = ulist_next(parents, &uiter);
752 ref->parent = node ? node->val : 0;
753 ref->inode_list = unode_aux_to_inode_list(node);
755 /* Add a prelim_ref(s) for any other parent(s). */
756 while ((node = ulist_next(parents, &uiter))) {
757 struct prelim_ref *new_ref;
759 new_ref = kmem_cache_alloc(btrfs_prelim_ref_cache,
766 memcpy(new_ref, ref, sizeof(*ref));
767 new_ref->parent = node->val;
768 new_ref->inode_list = unode_aux_to_inode_list(node);
769 prelim_ref_insert(fs_info, &preftrees->direct,
774 * Now it's a direct ref, put it in the direct tree. We must
775 * do this last because the ref could be merged/freed here.
777 prelim_ref_insert(fs_info, &preftrees->direct, ref, NULL);
779 ulist_reinit(parents);
784 * We may have inode lists attached to refs in the parents ulist, so we
785 * must free them before freeing the ulist and its refs.
787 free_leaf_list(parents);
792 * read tree blocks and add keys where required.
794 static int add_missing_keys(struct btrfs_fs_info *fs_info,
795 struct preftrees *preftrees, bool lock)
797 struct prelim_ref *ref;
798 struct extent_buffer *eb;
799 struct preftree *tree = &preftrees->indirect_missing_keys;
800 struct rb_node *node;
802 while ((node = rb_first_cached(&tree->root))) {
803 ref = rb_entry(node, struct prelim_ref, rbnode);
804 rb_erase_cached(node, &tree->root);
806 BUG_ON(ref->parent); /* should not be a direct ref */
807 BUG_ON(ref->key_for_search.type);
808 BUG_ON(!ref->wanted_disk_byte);
810 eb = read_tree_block(fs_info, ref->wanted_disk_byte,
811 ref->root_id, 0, ref->level - 1, NULL);
816 if (!extent_buffer_uptodate(eb)) {
818 free_extent_buffer(eb);
823 btrfs_tree_read_lock(eb);
824 if (btrfs_header_level(eb) == 0)
825 btrfs_item_key_to_cpu(eb, &ref->key_for_search, 0);
827 btrfs_node_key_to_cpu(eb, &ref->key_for_search, 0);
829 btrfs_tree_read_unlock(eb);
830 free_extent_buffer(eb);
831 prelim_ref_insert(fs_info, &preftrees->indirect, ref, NULL);
838 * add all currently queued delayed refs from this head whose seq nr is
839 * smaller or equal that seq to the list
841 static int add_delayed_refs(const struct btrfs_fs_info *fs_info,
842 struct btrfs_delayed_ref_head *head, u64 seq,
843 struct preftrees *preftrees, struct share_check *sc)
845 struct btrfs_delayed_ref_node *node;
846 struct btrfs_key key;
851 spin_lock(&head->lock);
852 for (n = rb_first_cached(&head->ref_tree); n; n = rb_next(n)) {
853 node = rb_entry(n, struct btrfs_delayed_ref_node,
858 switch (node->action) {
859 case BTRFS_ADD_DELAYED_EXTENT:
860 case BTRFS_UPDATE_DELAYED_HEAD:
863 case BTRFS_ADD_DELAYED_REF:
864 count = node->ref_mod;
866 case BTRFS_DROP_DELAYED_REF:
867 count = node->ref_mod * -1;
872 switch (node->type) {
873 case BTRFS_TREE_BLOCK_REF_KEY: {
874 /* NORMAL INDIRECT METADATA backref */
875 struct btrfs_delayed_tree_ref *ref;
876 struct btrfs_key *key_ptr = NULL;
878 if (head->extent_op && head->extent_op->update_key) {
879 btrfs_disk_key_to_cpu(&key, &head->extent_op->key);
883 ref = btrfs_delayed_node_to_tree_ref(node);
884 ret = add_indirect_ref(fs_info, preftrees, ref->root,
885 key_ptr, ref->level + 1,
886 node->bytenr, count, sc,
890 case BTRFS_SHARED_BLOCK_REF_KEY: {
891 /* SHARED DIRECT METADATA backref */
892 struct btrfs_delayed_tree_ref *ref;
894 ref = btrfs_delayed_node_to_tree_ref(node);
896 ret = add_direct_ref(fs_info, preftrees, ref->level + 1,
897 ref->parent, node->bytenr, count,
901 case BTRFS_EXTENT_DATA_REF_KEY: {
902 /* NORMAL INDIRECT DATA backref */
903 struct btrfs_delayed_data_ref *ref;
904 ref = btrfs_delayed_node_to_data_ref(node);
906 key.objectid = ref->objectid;
907 key.type = BTRFS_EXTENT_DATA_KEY;
908 key.offset = ref->offset;
911 * If we have a share check context and a reference for
912 * another inode, we can't exit immediately. This is
913 * because even if this is a BTRFS_ADD_DELAYED_REF
914 * reference we may find next a BTRFS_DROP_DELAYED_REF
915 * which cancels out this ADD reference.
917 * If this is a DROP reference and there was no previous
918 * ADD reference, then we need to signal that when we
919 * process references from the extent tree (through
920 * add_inline_refs() and add_keyed_refs()), we should
921 * not exit early if we find a reference for another
922 * inode, because one of the delayed DROP references
923 * may cancel that reference in the extent tree.
926 sc->have_delayed_delete_refs = true;
928 ret = add_indirect_ref(fs_info, preftrees, ref->root,
929 &key, 0, node->bytenr, count, sc,
933 case BTRFS_SHARED_DATA_REF_KEY: {
934 /* SHARED DIRECT FULL backref */
935 struct btrfs_delayed_data_ref *ref;
937 ref = btrfs_delayed_node_to_data_ref(node);
939 ret = add_direct_ref(fs_info, preftrees, 0, ref->parent,
940 node->bytenr, count, sc,
948 * We must ignore BACKREF_FOUND_SHARED until all delayed
949 * refs have been checked.
951 if (ret && (ret != BACKREF_FOUND_SHARED))
955 ret = extent_is_shared(sc);
957 spin_unlock(&head->lock);
962 * add all inline backrefs for bytenr to the list
964 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
966 static int add_inline_refs(const struct btrfs_fs_info *fs_info,
967 struct btrfs_path *path, u64 bytenr,
968 int *info_level, struct preftrees *preftrees,
969 struct share_check *sc)
973 struct extent_buffer *leaf;
974 struct btrfs_key key;
975 struct btrfs_key found_key;
978 struct btrfs_extent_item *ei;
983 * enumerate all inline refs
985 leaf = path->nodes[0];
986 slot = path->slots[0];
988 item_size = btrfs_item_size(leaf, slot);
989 BUG_ON(item_size < sizeof(*ei));
991 ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
992 flags = btrfs_extent_flags(leaf, ei);
993 btrfs_item_key_to_cpu(leaf, &found_key, slot);
995 ptr = (unsigned long)(ei + 1);
996 end = (unsigned long)ei + item_size;
998 if (found_key.type == BTRFS_EXTENT_ITEM_KEY &&
999 flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1000 struct btrfs_tree_block_info *info;
1002 info = (struct btrfs_tree_block_info *)ptr;
1003 *info_level = btrfs_tree_block_level(leaf, info);
1004 ptr += sizeof(struct btrfs_tree_block_info);
1006 } else if (found_key.type == BTRFS_METADATA_ITEM_KEY) {
1007 *info_level = found_key.offset;
1009 BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
1013 struct btrfs_extent_inline_ref *iref;
1017 iref = (struct btrfs_extent_inline_ref *)ptr;
1018 type = btrfs_get_extent_inline_ref_type(leaf, iref,
1019 BTRFS_REF_TYPE_ANY);
1020 if (type == BTRFS_REF_TYPE_INVALID)
1023 offset = btrfs_extent_inline_ref_offset(leaf, iref);
1026 case BTRFS_SHARED_BLOCK_REF_KEY:
1027 ret = add_direct_ref(fs_info, preftrees,
1028 *info_level + 1, offset,
1029 bytenr, 1, NULL, GFP_NOFS);
1031 case BTRFS_SHARED_DATA_REF_KEY: {
1032 struct btrfs_shared_data_ref *sdref;
1035 sdref = (struct btrfs_shared_data_ref *)(iref + 1);
1036 count = btrfs_shared_data_ref_count(leaf, sdref);
1038 ret = add_direct_ref(fs_info, preftrees, 0, offset,
1039 bytenr, count, sc, GFP_NOFS);
1042 case BTRFS_TREE_BLOCK_REF_KEY:
1043 ret = add_indirect_ref(fs_info, preftrees, offset,
1044 NULL, *info_level + 1,
1045 bytenr, 1, NULL, GFP_NOFS);
1047 case BTRFS_EXTENT_DATA_REF_KEY: {
1048 struct btrfs_extent_data_ref *dref;
1052 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1053 count = btrfs_extent_data_ref_count(leaf, dref);
1054 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1056 key.type = BTRFS_EXTENT_DATA_KEY;
1057 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1059 if (sc && sc->inum && key.objectid != sc->inum &&
1060 !sc->have_delayed_delete_refs) {
1061 ret = BACKREF_FOUND_SHARED;
1065 root = btrfs_extent_data_ref_root(leaf, dref);
1067 ret = add_indirect_ref(fs_info, preftrees, root,
1068 &key, 0, bytenr, count,
1078 ptr += btrfs_extent_inline_ref_size(type);
1085 * add all non-inline backrefs for bytenr to the list
1087 * Returns 0 on success, <0 on error, or BACKREF_FOUND_SHARED.
1089 static int add_keyed_refs(struct btrfs_root *extent_root,
1090 struct btrfs_path *path, u64 bytenr,
1091 int info_level, struct preftrees *preftrees,
1092 struct share_check *sc)
1094 struct btrfs_fs_info *fs_info = extent_root->fs_info;
1097 struct extent_buffer *leaf;
1098 struct btrfs_key key;
1101 ret = btrfs_next_item(extent_root, path);
1109 slot = path->slots[0];
1110 leaf = path->nodes[0];
1111 btrfs_item_key_to_cpu(leaf, &key, slot);
1113 if (key.objectid != bytenr)
1115 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1117 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1121 case BTRFS_SHARED_BLOCK_REF_KEY:
1122 /* SHARED DIRECT METADATA backref */
1123 ret = add_direct_ref(fs_info, preftrees,
1124 info_level + 1, key.offset,
1125 bytenr, 1, NULL, GFP_NOFS);
1127 case BTRFS_SHARED_DATA_REF_KEY: {
1128 /* SHARED DIRECT FULL backref */
1129 struct btrfs_shared_data_ref *sdref;
1132 sdref = btrfs_item_ptr(leaf, slot,
1133 struct btrfs_shared_data_ref);
1134 count = btrfs_shared_data_ref_count(leaf, sdref);
1135 ret = add_direct_ref(fs_info, preftrees, 0,
1136 key.offset, bytenr, count,
1140 case BTRFS_TREE_BLOCK_REF_KEY:
1141 /* NORMAL INDIRECT METADATA backref */
1142 ret = add_indirect_ref(fs_info, preftrees, key.offset,
1143 NULL, info_level + 1, bytenr,
1146 case BTRFS_EXTENT_DATA_REF_KEY: {
1147 /* NORMAL INDIRECT DATA backref */
1148 struct btrfs_extent_data_ref *dref;
1152 dref = btrfs_item_ptr(leaf, slot,
1153 struct btrfs_extent_data_ref);
1154 count = btrfs_extent_data_ref_count(leaf, dref);
1155 key.objectid = btrfs_extent_data_ref_objectid(leaf,
1157 key.type = BTRFS_EXTENT_DATA_KEY;
1158 key.offset = btrfs_extent_data_ref_offset(leaf, dref);
1160 if (sc && sc->inum && key.objectid != sc->inum &&
1161 !sc->have_delayed_delete_refs) {
1162 ret = BACKREF_FOUND_SHARED;
1166 root = btrfs_extent_data_ref_root(leaf, dref);
1167 ret = add_indirect_ref(fs_info, preftrees, root,
1168 &key, 0, bytenr, count,
1184 * this adds all existing backrefs (inline backrefs, backrefs and delayed
1185 * refs) for the given bytenr to the refs list, merges duplicates and resolves
1186 * indirect refs to their parent bytenr.
1187 * When roots are found, they're added to the roots list
1189 * If time_seq is set to BTRFS_SEQ_LAST, it will not search delayed_refs, and
1190 * behave much like trans == NULL case, the difference only lies in it will not
1192 * The special case is for qgroup to search roots in commit_transaction().
1194 * @sc - if !NULL, then immediately return BACKREF_FOUND_SHARED when a
1195 * shared extent is detected.
1197 * Otherwise this returns 0 for success and <0 for an error.
1199 * If ignore_offset is set to false, only extent refs whose offsets match
1200 * extent_item_pos are returned. If true, every extent ref is returned
1201 * and extent_item_pos is ignored.
1203 * FIXME some caching might speed things up
1205 static int find_parent_nodes(struct btrfs_trans_handle *trans,
1206 struct btrfs_fs_info *fs_info, u64 bytenr,
1207 u64 time_seq, struct ulist *refs,
1208 struct ulist *roots, const u64 *extent_item_pos,
1209 struct share_check *sc, bool ignore_offset)
1211 struct btrfs_root *root = btrfs_extent_root(fs_info, bytenr);
1212 struct btrfs_key key;
1213 struct btrfs_path *path;
1214 struct btrfs_delayed_ref_root *delayed_refs = NULL;
1215 struct btrfs_delayed_ref_head *head;
1218 struct prelim_ref *ref;
1219 struct rb_node *node;
1220 struct extent_inode_elem *eie = NULL;
1221 struct preftrees preftrees = {
1222 .direct = PREFTREE_INIT,
1223 .indirect = PREFTREE_INIT,
1224 .indirect_missing_keys = PREFTREE_INIT
1227 key.objectid = bytenr;
1228 key.offset = (u64)-1;
1229 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1230 key.type = BTRFS_METADATA_ITEM_KEY;
1232 key.type = BTRFS_EXTENT_ITEM_KEY;
1234 path = btrfs_alloc_path();
1238 path->search_commit_root = 1;
1239 path->skip_locking = 1;
1242 if (time_seq == BTRFS_SEQ_LAST)
1243 path->skip_locking = 1;
1248 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1252 /* This shouldn't happen, indicates a bug or fs corruption. */
1258 if (trans && likely(trans->type != __TRANS_DUMMY) &&
1259 time_seq != BTRFS_SEQ_LAST) {
1261 * We have a specific time_seq we care about and trans which
1262 * means we have the path lock, we need to grab the ref head and
1263 * lock it so we have a consistent view of the refs at the given
1266 delayed_refs = &trans->transaction->delayed_refs;
1267 spin_lock(&delayed_refs->lock);
1268 head = btrfs_find_delayed_ref_head(delayed_refs, bytenr);
1270 if (!mutex_trylock(&head->mutex)) {
1271 refcount_inc(&head->refs);
1272 spin_unlock(&delayed_refs->lock);
1274 btrfs_release_path(path);
1277 * Mutex was contended, block until it's
1278 * released and try again
1280 mutex_lock(&head->mutex);
1281 mutex_unlock(&head->mutex);
1282 btrfs_put_delayed_ref_head(head);
1285 spin_unlock(&delayed_refs->lock);
1286 ret = add_delayed_refs(fs_info, head, time_seq,
1288 mutex_unlock(&head->mutex);
1292 spin_unlock(&delayed_refs->lock);
1296 if (path->slots[0]) {
1297 struct extent_buffer *leaf;
1301 leaf = path->nodes[0];
1302 slot = path->slots[0];
1303 btrfs_item_key_to_cpu(leaf, &key, slot);
1304 if (key.objectid == bytenr &&
1305 (key.type == BTRFS_EXTENT_ITEM_KEY ||
1306 key.type == BTRFS_METADATA_ITEM_KEY)) {
1307 ret = add_inline_refs(fs_info, path, bytenr,
1308 &info_level, &preftrees, sc);
1311 ret = add_keyed_refs(root, path, bytenr, info_level,
1318 btrfs_release_path(path);
1320 ret = add_missing_keys(fs_info, &preftrees, path->skip_locking == 0);
1324 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect_missing_keys.root.rb_root));
1326 ret = resolve_indirect_refs(fs_info, path, time_seq, &preftrees,
1327 extent_item_pos, sc, ignore_offset);
1331 WARN_ON(!RB_EMPTY_ROOT(&preftrees.indirect.root.rb_root));
1334 * This walks the tree of merged and resolved refs. Tree blocks are
1335 * read in as needed. Unique entries are added to the ulist, and
1336 * the list of found roots is updated.
1338 * We release the entire tree in one go before returning.
1340 node = rb_first_cached(&preftrees.direct.root);
1342 ref = rb_entry(node, struct prelim_ref, rbnode);
1343 node = rb_next(&ref->rbnode);
1345 * ref->count < 0 can happen here if there are delayed
1346 * refs with a node->action of BTRFS_DROP_DELAYED_REF.
1347 * prelim_ref_insert() relies on this when merging
1348 * identical refs to keep the overall count correct.
1349 * prelim_ref_insert() will merge only those refs
1350 * which compare identically. Any refs having
1351 * e.g. different offsets would not be merged,
1352 * and would retain their original ref->count < 0.
1354 if (roots && ref->count && ref->root_id && ref->parent == 0) {
1355 if (sc && sc->root_objectid &&
1356 ref->root_id != sc->root_objectid) {
1357 ret = BACKREF_FOUND_SHARED;
1361 /* no parent == root of tree */
1362 ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
1366 if (ref->count && ref->parent) {
1367 if (extent_item_pos && !ref->inode_list &&
1369 struct extent_buffer *eb;
1371 eb = read_tree_block(fs_info, ref->parent, 0,
1372 0, ref->level, NULL);
1377 if (!extent_buffer_uptodate(eb)) {
1378 free_extent_buffer(eb);
1383 if (!path->skip_locking)
1384 btrfs_tree_read_lock(eb);
1385 ret = find_extent_in_eb(eb, bytenr,
1386 *extent_item_pos, &eie, ignore_offset);
1387 if (!path->skip_locking)
1388 btrfs_tree_read_unlock(eb);
1389 free_extent_buffer(eb);
1392 ref->inode_list = eie;
1394 * We transferred the list ownership to the ref,
1395 * so set to NULL to avoid a double free in case
1396 * an error happens after this.
1400 ret = ulist_add_merge_ptr(refs, ref->parent,
1402 (void **)&eie, GFP_NOFS);
1405 if (!ret && extent_item_pos) {
1407 * We've recorded that parent, so we must extend
1408 * its inode list here.
1410 * However if there was corruption we may not
1411 * have found an eie, return an error in this
1421 eie->next = ref->inode_list;
1425 * We have transferred the inode list ownership from
1426 * this ref to the ref we added to the 'refs' ulist.
1427 * So set this ref's inode list to NULL to avoid
1428 * use-after-free when our caller uses it or double
1429 * frees in case an error happens before we return.
1431 ref->inode_list = NULL;
1437 btrfs_free_path(path);
1439 prelim_release(&preftrees.direct);
1440 prelim_release(&preftrees.indirect);
1441 prelim_release(&preftrees.indirect_missing_keys);
1444 free_inode_elem_list(eie);
1449 * Finds all leafs with a reference to the specified combination of bytenr and
1450 * offset. key_list_head will point to a list of corresponding keys (caller must
1451 * free each list element). The leafs will be stored in the leafs ulist, which
1452 * must be freed with ulist_free.
1454 * returns 0 on success, <0 on error
1456 int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
1457 struct btrfs_fs_info *fs_info, u64 bytenr,
1458 u64 time_seq, struct ulist **leafs,
1459 const u64 *extent_item_pos, bool ignore_offset)
1463 *leafs = ulist_alloc(GFP_NOFS);
1467 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1468 *leafs, NULL, extent_item_pos, NULL, ignore_offset);
1469 if (ret < 0 && ret != -ENOENT) {
1470 free_leaf_list(*leafs);
1478 * walk all backrefs for a given extent to find all roots that reference this
1479 * extent. Walking a backref means finding all extents that reference this
1480 * extent and in turn walk the backrefs of those, too. Naturally this is a
1481 * recursive process, but here it is implemented in an iterative fashion: We
1482 * find all referencing extents for the extent in question and put them on a
1483 * list. In turn, we find all referencing extents for those, further appending
1484 * to the list. The way we iterate the list allows adding more elements after
1485 * the current while iterating. The process stops when we reach the end of the
1486 * list. Found roots are added to the roots list.
1488 * returns 0 on success, < 0 on error.
1490 static int btrfs_find_all_roots_safe(struct btrfs_trans_handle *trans,
1491 struct btrfs_fs_info *fs_info, u64 bytenr,
1492 u64 time_seq, struct ulist **roots,
1496 struct ulist_node *node = NULL;
1497 struct ulist_iterator uiter;
1500 tmp = ulist_alloc(GFP_NOFS);
1503 *roots = ulist_alloc(GFP_NOFS);
1509 ULIST_ITER_INIT(&uiter);
1511 ret = find_parent_nodes(trans, fs_info, bytenr, time_seq,
1512 tmp, *roots, NULL, NULL, ignore_offset);
1513 if (ret < 0 && ret != -ENOENT) {
1519 node = ulist_next(tmp, &uiter);
1530 int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
1531 struct btrfs_fs_info *fs_info, u64 bytenr,
1532 u64 time_seq, struct ulist **roots,
1533 bool skip_commit_root_sem)
1537 if (!trans && !skip_commit_root_sem)
1538 down_read(&fs_info->commit_root_sem);
1539 ret = btrfs_find_all_roots_safe(trans, fs_info, bytenr,
1540 time_seq, roots, false);
1541 if (!trans && !skip_commit_root_sem)
1542 up_read(&fs_info->commit_root_sem);
1547 * The caller has joined a transaction or is holding a read lock on the
1548 * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1549 * snapshot field changing while updating or checking the cache.
1551 static bool lookup_backref_shared_cache(struct btrfs_backref_shared_cache *cache,
1552 struct btrfs_root *root,
1553 u64 bytenr, int level, bool *is_shared)
1555 struct btrfs_backref_shared_cache_entry *entry;
1557 if (!cache->use_cache)
1560 if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1564 * Level -1 is used for the data extent, which is not reliable to cache
1565 * because its reference count can increase or decrease without us
1566 * realizing. We cache results only for extent buffers that lead from
1567 * the root node down to the leaf with the file extent item.
1571 entry = &cache->entries[level];
1573 /* Unused cache entry or being used for some other extent buffer. */
1574 if (entry->bytenr != bytenr)
1578 * We cached a false result, but the last snapshot generation of the
1579 * root changed, so we now have a snapshot. Don't trust the result.
1581 if (!entry->is_shared &&
1582 entry->gen != btrfs_root_last_snapshot(&root->root_item))
1586 * If we cached a true result and the last generation used for dropping
1587 * a root changed, we can not trust the result, because the dropped root
1588 * could be a snapshot sharing this extent buffer.
1590 if (entry->is_shared &&
1591 entry->gen != btrfs_get_last_root_drop_gen(root->fs_info))
1594 *is_shared = entry->is_shared;
1596 * If the node at this level is shared, than all nodes below are also
1597 * shared. Currently some of the nodes below may be marked as not shared
1598 * because we have just switched from one leaf to another, and switched
1599 * also other nodes above the leaf and below the current level, so mark
1603 for (int i = 0; i < level; i++) {
1604 cache->entries[i].is_shared = true;
1605 cache->entries[i].gen = entry->gen;
1613 * The caller has joined a transaction or is holding a read lock on the
1614 * fs_info->commit_root_sem semaphore, so no need to worry about the root's last
1615 * snapshot field changing while updating or checking the cache.
1617 static void store_backref_shared_cache(struct btrfs_backref_shared_cache *cache,
1618 struct btrfs_root *root,
1619 u64 bytenr, int level, bool is_shared)
1621 struct btrfs_backref_shared_cache_entry *entry;
1624 if (!cache->use_cache)
1627 if (WARN_ON_ONCE(level >= BTRFS_MAX_LEVEL))
1631 * Level -1 is used for the data extent, which is not reliable to cache
1632 * because its reference count can increase or decrease without us
1633 * realizing. We cache results only for extent buffers that lead from
1634 * the root node down to the leaf with the file extent item.
1639 gen = btrfs_get_last_root_drop_gen(root->fs_info);
1641 gen = btrfs_root_last_snapshot(&root->root_item);
1643 entry = &cache->entries[level];
1644 entry->bytenr = bytenr;
1645 entry->is_shared = is_shared;
1649 * If we found an extent buffer is shared, set the cache result for all
1650 * extent buffers below it to true. As nodes in the path are COWed,
1651 * their sharedness is moved to their children, and if a leaf is COWed,
1652 * then the sharedness of a data extent becomes direct, the refcount of
1653 * data extent is increased in the extent item at the extent tree.
1656 for (int i = 0; i < level; i++) {
1657 entry = &cache->entries[i];
1658 entry->is_shared = is_shared;
1665 * Check if a data extent is shared or not.
1667 * @root: The root the inode belongs to.
1668 * @inum: Number of the inode whose extent we are checking.
1669 * @bytenr: Logical bytenr of the extent we are checking.
1670 * @extent_gen: Generation of the extent (file extent item) or 0 if it is
1672 * @roots: List of roots this extent is shared among.
1673 * @tmp: Temporary list used for iteration.
1674 * @cache: A backref lookup result cache.
1676 * btrfs_is_data_extent_shared uses the backref walking code but will short
1677 * circuit as soon as it finds a root or inode that doesn't match the
1678 * one passed in. This provides a significant performance benefit for
1679 * callers (such as fiemap) which want to know whether the extent is
1680 * shared but do not need a ref count.
1682 * This attempts to attach to the running transaction in order to account for
1683 * delayed refs, but continues on even when no running transaction exists.
1685 * Return: 0 if extent is not shared, 1 if it is shared, < 0 on error.
1687 int btrfs_is_data_extent_shared(struct btrfs_root *root, u64 inum, u64 bytenr,
1689 struct ulist *roots, struct ulist *tmp,
1690 struct btrfs_backref_shared_cache *cache)
1692 struct btrfs_fs_info *fs_info = root->fs_info;
1693 struct btrfs_trans_handle *trans;
1694 struct ulist_iterator uiter;
1695 struct ulist_node *node;
1696 struct btrfs_seq_list elem = BTRFS_SEQ_LIST_INIT(elem);
1698 struct share_check shared = {
1699 .root_objectid = root->root_key.objectid,
1702 .have_delayed_delete_refs = false,
1709 trans = btrfs_join_transaction_nostart(root);
1710 if (IS_ERR(trans)) {
1711 if (PTR_ERR(trans) != -ENOENT && PTR_ERR(trans) != -EROFS) {
1712 ret = PTR_ERR(trans);
1716 down_read(&fs_info->commit_root_sem);
1718 btrfs_get_tree_mod_seq(fs_info, &elem);
1721 /* -1 means we are in the bytenr of the data extent. */
1723 ULIST_ITER_INIT(&uiter);
1724 cache->use_cache = true;
1729 ret = find_parent_nodes(trans, fs_info, bytenr, elem.seq, tmp,
1730 roots, NULL, &shared, false);
1731 if (ret == BACKREF_FOUND_SHARED) {
1732 /* this is the only condition under which we return 1 */
1735 store_backref_shared_cache(cache, root, bytenr,
1739 if (ret < 0 && ret != -ENOENT)
1743 * If our data extent is not shared through reflinks and it was
1744 * created in a generation after the last one used to create a
1745 * snapshot of the inode's root, then it can not be shared
1746 * indirectly through subtrees, as that can only happen with
1747 * snapshots. In this case bail out, no need to check for the
1748 * sharedness of extent buffers.
1751 extent_gen > btrfs_root_last_snapshot(&root->root_item))
1755 * If our data extent was not directly shared (without multiple
1756 * reference items), than it might have a single reference item
1757 * with a count > 1 for the same offset, which means there are 2
1758 * (or more) file extent items that point to the data extent -
1759 * this happens when a file extent item needs to be split and
1760 * then one item gets moved to another leaf due to a b+tree leaf
1761 * split when inserting some item. In this case the file extent
1762 * items may be located in different leaves and therefore some
1763 * of the leaves may be referenced through shared subtrees while
1764 * others are not. Since our extent buffer cache only works for
1765 * a single path (by far the most common case and simpler to
1766 * deal with), we can not use it if we have multiple leaves
1767 * (which implies multiple paths).
1769 if (level == -1 && tmp->nnodes > 1)
1770 cache->use_cache = false;
1773 store_backref_shared_cache(cache, root, bytenr,
1775 node = ulist_next(tmp, &uiter);
1780 cached = lookup_backref_shared_cache(cache, root, bytenr, level,
1783 ret = (is_shared ? 1 : 0);
1786 shared.share_count = 0;
1787 shared.have_delayed_delete_refs = false;
1792 btrfs_put_tree_mod_seq(fs_info, &elem);
1793 btrfs_end_transaction(trans);
1795 up_read(&fs_info->commit_root_sem);
1798 ulist_release(roots);
1803 int btrfs_find_one_extref(struct btrfs_root *root, u64 inode_objectid,
1804 u64 start_off, struct btrfs_path *path,
1805 struct btrfs_inode_extref **ret_extref,
1809 struct btrfs_key key;
1810 struct btrfs_key found_key;
1811 struct btrfs_inode_extref *extref;
1812 const struct extent_buffer *leaf;
1815 key.objectid = inode_objectid;
1816 key.type = BTRFS_INODE_EXTREF_KEY;
1817 key.offset = start_off;
1819 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1824 leaf = path->nodes[0];
1825 slot = path->slots[0];
1826 if (slot >= btrfs_header_nritems(leaf)) {
1828 * If the item at offset is not found,
1829 * btrfs_search_slot will point us to the slot
1830 * where it should be inserted. In our case
1831 * that will be the slot directly before the
1832 * next INODE_REF_KEY_V2 item. In the case
1833 * that we're pointing to the last slot in a
1834 * leaf, we must move one leaf over.
1836 ret = btrfs_next_leaf(root, path);
1845 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1848 * Check that we're still looking at an extended ref key for
1849 * this particular objectid. If we have different
1850 * objectid or type then there are no more to be found
1851 * in the tree and we can exit.
1854 if (found_key.objectid != inode_objectid)
1856 if (found_key.type != BTRFS_INODE_EXTREF_KEY)
1860 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
1861 extref = (struct btrfs_inode_extref *)ptr;
1862 *ret_extref = extref;
1864 *found_off = found_key.offset;
1872 * this iterates to turn a name (from iref/extref) into a full filesystem path.
1873 * Elements of the path are separated by '/' and the path is guaranteed to be
1874 * 0-terminated. the path is only given within the current file system.
1875 * Therefore, it never starts with a '/'. the caller is responsible to provide
1876 * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
1877 * the start point of the resulting string is returned. this pointer is within
1879 * in case the path buffer would overflow, the pointer is decremented further
1880 * as if output was written to the buffer, though no more output is actually
1881 * generated. that way, the caller can determine how much space would be
1882 * required for the path to fit into the buffer. in that case, the returned
1883 * value will be smaller than dest. callers must check this!
1885 char *btrfs_ref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
1886 u32 name_len, unsigned long name_off,
1887 struct extent_buffer *eb_in, u64 parent,
1888 char *dest, u32 size)
1893 s64 bytes_left = ((s64)size) - 1;
1894 struct extent_buffer *eb = eb_in;
1895 struct btrfs_key found_key;
1896 struct btrfs_inode_ref *iref;
1898 if (bytes_left >= 0)
1899 dest[bytes_left] = '\0';
1902 bytes_left -= name_len;
1903 if (bytes_left >= 0)
1904 read_extent_buffer(eb, dest + bytes_left,
1905 name_off, name_len);
1907 if (!path->skip_locking)
1908 btrfs_tree_read_unlock(eb);
1909 free_extent_buffer(eb);
1911 ret = btrfs_find_item(fs_root, path, parent, 0,
1912 BTRFS_INODE_REF_KEY, &found_key);
1918 next_inum = found_key.offset;
1920 /* regular exit ahead */
1921 if (parent == next_inum)
1924 slot = path->slots[0];
1925 eb = path->nodes[0];
1926 /* make sure we can use eb after releasing the path */
1928 path->nodes[0] = NULL;
1931 btrfs_release_path(path);
1932 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
1934 name_len = btrfs_inode_ref_name_len(eb, iref);
1935 name_off = (unsigned long)(iref + 1);
1939 if (bytes_left >= 0)
1940 dest[bytes_left] = '/';
1943 btrfs_release_path(path);
1946 return ERR_PTR(ret);
1948 return dest + bytes_left;
1952 * this makes the path point to (logical EXTENT_ITEM *)
1953 * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
1954 * tree blocks and <0 on error.
1956 int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
1957 struct btrfs_path *path, struct btrfs_key *found_key,
1960 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, logical);
1965 const struct extent_buffer *eb;
1966 struct btrfs_extent_item *ei;
1967 struct btrfs_key key;
1969 if (btrfs_fs_incompat(fs_info, SKINNY_METADATA))
1970 key.type = BTRFS_METADATA_ITEM_KEY;
1972 key.type = BTRFS_EXTENT_ITEM_KEY;
1973 key.objectid = logical;
1974 key.offset = (u64)-1;
1976 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1980 ret = btrfs_previous_extent_item(extent_root, path, 0);
1986 btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
1987 if (found_key->type == BTRFS_METADATA_ITEM_KEY)
1988 size = fs_info->nodesize;
1989 else if (found_key->type == BTRFS_EXTENT_ITEM_KEY)
1990 size = found_key->offset;
1992 if (found_key->objectid > logical ||
1993 found_key->objectid + size <= logical) {
1994 btrfs_debug(fs_info,
1995 "logical %llu is not within any extent", logical);
1999 eb = path->nodes[0];
2000 item_size = btrfs_item_size(eb, path->slots[0]);
2001 BUG_ON(item_size < sizeof(*ei));
2003 ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
2004 flags = btrfs_extent_flags(eb, ei);
2006 btrfs_debug(fs_info,
2007 "logical %llu is at position %llu within the extent (%llu EXTENT_ITEM %llu) flags %#llx size %u",
2008 logical, logical - found_key->objectid, found_key->objectid,
2009 found_key->offset, flags, item_size);
2011 WARN_ON(!flags_ret);
2013 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2014 *flags_ret = BTRFS_EXTENT_FLAG_TREE_BLOCK;
2015 else if (flags & BTRFS_EXTENT_FLAG_DATA)
2016 *flags_ret = BTRFS_EXTENT_FLAG_DATA;
2026 * helper function to iterate extent inline refs. ptr must point to a 0 value
2027 * for the first call and may be modified. it is used to track state.
2028 * if more refs exist, 0 is returned and the next call to
2029 * get_extent_inline_ref must pass the modified ptr parameter to get the
2030 * next ref. after the last ref was processed, 1 is returned.
2031 * returns <0 on error
2033 static int get_extent_inline_ref(unsigned long *ptr,
2034 const struct extent_buffer *eb,
2035 const struct btrfs_key *key,
2036 const struct btrfs_extent_item *ei,
2038 struct btrfs_extent_inline_ref **out_eiref,
2043 struct btrfs_tree_block_info *info;
2047 flags = btrfs_extent_flags(eb, ei);
2048 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
2049 if (key->type == BTRFS_METADATA_ITEM_KEY) {
2050 /* a skinny metadata extent */
2052 (struct btrfs_extent_inline_ref *)(ei + 1);
2054 WARN_ON(key->type != BTRFS_EXTENT_ITEM_KEY);
2055 info = (struct btrfs_tree_block_info *)(ei + 1);
2057 (struct btrfs_extent_inline_ref *)(info + 1);
2060 *out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
2062 *ptr = (unsigned long)*out_eiref;
2063 if ((unsigned long)(*ptr) >= (unsigned long)ei + item_size)
2067 end = (unsigned long)ei + item_size;
2068 *out_eiref = (struct btrfs_extent_inline_ref *)(*ptr);
2069 *out_type = btrfs_get_extent_inline_ref_type(eb, *out_eiref,
2070 BTRFS_REF_TYPE_ANY);
2071 if (*out_type == BTRFS_REF_TYPE_INVALID)
2074 *ptr += btrfs_extent_inline_ref_size(*out_type);
2075 WARN_ON(*ptr > end);
2077 return 1; /* last */
2083 * reads the tree block backref for an extent. tree level and root are returned
2084 * through out_level and out_root. ptr must point to a 0 value for the first
2085 * call and may be modified (see get_extent_inline_ref comment).
2086 * returns 0 if data was provided, 1 if there was no more data to provide or
2089 int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
2090 struct btrfs_key *key, struct btrfs_extent_item *ei,
2091 u32 item_size, u64 *out_root, u8 *out_level)
2095 struct btrfs_extent_inline_ref *eiref;
2097 if (*ptr == (unsigned long)-1)
2101 ret = get_extent_inline_ref(ptr, eb, key, ei, item_size,
2106 if (type == BTRFS_TREE_BLOCK_REF_KEY ||
2107 type == BTRFS_SHARED_BLOCK_REF_KEY)
2114 /* we can treat both ref types equally here */
2115 *out_root = btrfs_extent_inline_ref_offset(eb, eiref);
2117 if (key->type == BTRFS_EXTENT_ITEM_KEY) {
2118 struct btrfs_tree_block_info *info;
2120 info = (struct btrfs_tree_block_info *)(ei + 1);
2121 *out_level = btrfs_tree_block_level(eb, info);
2123 ASSERT(key->type == BTRFS_METADATA_ITEM_KEY);
2124 *out_level = (u8)key->offset;
2128 *ptr = (unsigned long)-1;
2133 static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
2134 struct extent_inode_elem *inode_list,
2135 u64 root, u64 extent_item_objectid,
2136 iterate_extent_inodes_t *iterate, void *ctx)
2138 struct extent_inode_elem *eie;
2141 for (eie = inode_list; eie; eie = eie->next) {
2142 btrfs_debug(fs_info,
2143 "ref for %llu resolved, key (%llu EXTEND_DATA %llu), root %llu",
2144 extent_item_objectid, eie->inum,
2146 ret = iterate(eie->inum, eie->offset, root, ctx);
2148 btrfs_debug(fs_info,
2149 "stopping iteration for %llu due to ret=%d",
2150 extent_item_objectid, ret);
2159 * calls iterate() for every inode that references the extent identified by
2160 * the given parameters.
2161 * when the iterator function returns a non-zero value, iteration stops.
2163 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
2164 u64 extent_item_objectid, u64 extent_item_pos,
2165 int search_commit_root,
2166 iterate_extent_inodes_t *iterate, void *ctx,
2170 struct btrfs_trans_handle *trans = NULL;
2171 struct ulist *refs = NULL;
2172 struct ulist *roots = NULL;
2173 struct ulist_node *ref_node = NULL;
2174 struct ulist_node *root_node = NULL;
2175 struct btrfs_seq_list seq_elem = BTRFS_SEQ_LIST_INIT(seq_elem);
2176 struct ulist_iterator ref_uiter;
2177 struct ulist_iterator root_uiter;
2179 btrfs_debug(fs_info, "resolving all inodes for extent %llu",
2180 extent_item_objectid);
2182 if (!search_commit_root) {
2183 trans = btrfs_attach_transaction(fs_info->tree_root);
2184 if (IS_ERR(trans)) {
2185 if (PTR_ERR(trans) != -ENOENT &&
2186 PTR_ERR(trans) != -EROFS)
2187 return PTR_ERR(trans);
2193 btrfs_get_tree_mod_seq(fs_info, &seq_elem);
2195 down_read(&fs_info->commit_root_sem);
2197 ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
2198 seq_elem.seq, &refs,
2199 &extent_item_pos, ignore_offset);
2203 ULIST_ITER_INIT(&ref_uiter);
2204 while (!ret && (ref_node = ulist_next(refs, &ref_uiter))) {
2205 ret = btrfs_find_all_roots_safe(trans, fs_info, ref_node->val,
2206 seq_elem.seq, &roots,
2210 ULIST_ITER_INIT(&root_uiter);
2211 while (!ret && (root_node = ulist_next(roots, &root_uiter))) {
2212 btrfs_debug(fs_info,
2213 "root %llu references leaf %llu, data list %#llx",
2214 root_node->val, ref_node->val,
2216 ret = iterate_leaf_refs(fs_info,
2217 (struct extent_inode_elem *)
2218 (uintptr_t)ref_node->aux,
2220 extent_item_objectid,
2226 free_leaf_list(refs);
2229 btrfs_put_tree_mod_seq(fs_info, &seq_elem);
2230 btrfs_end_transaction(trans);
2232 up_read(&fs_info->commit_root_sem);
2238 static int build_ino_list(u64 inum, u64 offset, u64 root, void *ctx)
2240 struct btrfs_data_container *inodes = ctx;
2241 const size_t c = 3 * sizeof(u64);
2243 if (inodes->bytes_left >= c) {
2244 inodes->bytes_left -= c;
2245 inodes->val[inodes->elem_cnt] = inum;
2246 inodes->val[inodes->elem_cnt + 1] = offset;
2247 inodes->val[inodes->elem_cnt + 2] = root;
2248 inodes->elem_cnt += 3;
2250 inodes->bytes_missing += c - inodes->bytes_left;
2251 inodes->bytes_left = 0;
2252 inodes->elem_missed += 3;
2258 int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
2259 struct btrfs_path *path,
2260 void *ctx, bool ignore_offset)
2263 u64 extent_item_pos;
2265 struct btrfs_key found_key;
2266 int search_commit_root = path->search_commit_root;
2268 ret = extent_from_logical(fs_info, logical, path, &found_key, &flags);
2269 btrfs_release_path(path);
2272 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
2275 extent_item_pos = logical - found_key.objectid;
2276 ret = iterate_extent_inodes(fs_info, found_key.objectid,
2277 extent_item_pos, search_commit_root,
2278 build_ino_list, ctx, ignore_offset);
2283 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2284 struct extent_buffer *eb, struct inode_fs_paths *ipath);
2286 static int iterate_inode_refs(u64 inum, struct inode_fs_paths *ipath)
2295 struct btrfs_root *fs_root = ipath->fs_root;
2296 struct btrfs_path *path = ipath->btrfs_path;
2297 struct extent_buffer *eb;
2298 struct btrfs_inode_ref *iref;
2299 struct btrfs_key found_key;
2302 ret = btrfs_find_item(fs_root, path, inum,
2303 parent ? parent + 1 : 0, BTRFS_INODE_REF_KEY,
2309 ret = found ? 0 : -ENOENT;
2314 parent = found_key.offset;
2315 slot = path->slots[0];
2316 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2321 btrfs_release_path(path);
2323 iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
2325 for (cur = 0; cur < btrfs_item_size(eb, slot); cur += len) {
2326 name_len = btrfs_inode_ref_name_len(eb, iref);
2327 /* path must be released before calling iterate()! */
2328 btrfs_debug(fs_root->fs_info,
2329 "following ref at offset %u for inode %llu in tree %llu",
2330 cur, found_key.objectid,
2331 fs_root->root_key.objectid);
2332 ret = inode_to_path(parent, name_len,
2333 (unsigned long)(iref + 1), eb, ipath);
2336 len = sizeof(*iref) + name_len;
2337 iref = (struct btrfs_inode_ref *)((char *)iref + len);
2339 free_extent_buffer(eb);
2342 btrfs_release_path(path);
2347 static int iterate_inode_extrefs(u64 inum, struct inode_fs_paths *ipath)
2354 struct btrfs_root *fs_root = ipath->fs_root;
2355 struct btrfs_path *path = ipath->btrfs_path;
2356 struct extent_buffer *eb;
2357 struct btrfs_inode_extref *extref;
2363 ret = btrfs_find_one_extref(fs_root, inum, offset, path, &extref,
2368 ret = found ? 0 : -ENOENT;
2373 slot = path->slots[0];
2374 eb = btrfs_clone_extent_buffer(path->nodes[0]);
2379 btrfs_release_path(path);
2381 item_size = btrfs_item_size(eb, slot);
2382 ptr = btrfs_item_ptr_offset(eb, slot);
2385 while (cur_offset < item_size) {
2388 extref = (struct btrfs_inode_extref *)(ptr + cur_offset);
2389 parent = btrfs_inode_extref_parent(eb, extref);
2390 name_len = btrfs_inode_extref_name_len(eb, extref);
2391 ret = inode_to_path(parent, name_len,
2392 (unsigned long)&extref->name, eb, ipath);
2396 cur_offset += btrfs_inode_extref_name_len(eb, extref);
2397 cur_offset += sizeof(*extref);
2399 free_extent_buffer(eb);
2404 btrfs_release_path(path);
2410 * returns 0 if the path could be dumped (probably truncated)
2411 * returns <0 in case of an error
2413 static int inode_to_path(u64 inum, u32 name_len, unsigned long name_off,
2414 struct extent_buffer *eb, struct inode_fs_paths *ipath)
2418 int i = ipath->fspath->elem_cnt;
2419 const int s_ptr = sizeof(char *);
2422 bytes_left = ipath->fspath->bytes_left > s_ptr ?
2423 ipath->fspath->bytes_left - s_ptr : 0;
2425 fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
2426 fspath = btrfs_ref_to_path(ipath->fs_root, ipath->btrfs_path, name_len,
2427 name_off, eb, inum, fspath_min, bytes_left);
2429 return PTR_ERR(fspath);
2431 if (fspath > fspath_min) {
2432 ipath->fspath->val[i] = (u64)(unsigned long)fspath;
2433 ++ipath->fspath->elem_cnt;
2434 ipath->fspath->bytes_left = fspath - fspath_min;
2436 ++ipath->fspath->elem_missed;
2437 ipath->fspath->bytes_missing += fspath_min - fspath;
2438 ipath->fspath->bytes_left = 0;
2445 * this dumps all file system paths to the inode into the ipath struct, provided
2446 * is has been created large enough. each path is zero-terminated and accessed
2447 * from ipath->fspath->val[i].
2448 * when it returns, there are ipath->fspath->elem_cnt number of paths available
2449 * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
2450 * number of missed paths is recorded in ipath->fspath->elem_missed, otherwise,
2451 * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
2452 * have been needed to return all paths.
2454 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
2459 ret = iterate_inode_refs(inum, ipath);
2462 else if (ret != -ENOENT)
2465 ret = iterate_inode_extrefs(inum, ipath);
2466 if (ret == -ENOENT && found_refs)
2472 struct btrfs_data_container *init_data_container(u32 total_bytes)
2474 struct btrfs_data_container *data;
2477 alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
2478 data = kvzalloc(alloc_bytes, GFP_KERNEL);
2480 return ERR_PTR(-ENOMEM);
2482 if (total_bytes >= sizeof(*data))
2483 data->bytes_left = total_bytes - sizeof(*data);
2485 data->bytes_missing = sizeof(*data) - total_bytes;
2491 * allocates space to return multiple file system paths for an inode.
2492 * total_bytes to allocate are passed, note that space usable for actual path
2493 * information will be total_bytes - sizeof(struct inode_fs_paths).
2494 * the returned pointer must be freed with free_ipath() in the end.
2496 struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
2497 struct btrfs_path *path)
2499 struct inode_fs_paths *ifp;
2500 struct btrfs_data_container *fspath;
2502 fspath = init_data_container(total_bytes);
2504 return ERR_CAST(fspath);
2506 ifp = kmalloc(sizeof(*ifp), GFP_KERNEL);
2509 return ERR_PTR(-ENOMEM);
2512 ifp->btrfs_path = path;
2513 ifp->fspath = fspath;
2514 ifp->fs_root = fs_root;
2519 void free_ipath(struct inode_fs_paths *ipath)
2523 kvfree(ipath->fspath);
2527 struct btrfs_backref_iter *btrfs_backref_iter_alloc(
2528 struct btrfs_fs_info *fs_info, gfp_t gfp_flag)
2530 struct btrfs_backref_iter *ret;
2532 ret = kzalloc(sizeof(*ret), gfp_flag);
2536 ret->path = btrfs_alloc_path();
2542 /* Current backref iterator only supports iteration in commit root */
2543 ret->path->search_commit_root = 1;
2544 ret->path->skip_locking = 1;
2545 ret->fs_info = fs_info;
2550 int btrfs_backref_iter_start(struct btrfs_backref_iter *iter, u64 bytenr)
2552 struct btrfs_fs_info *fs_info = iter->fs_info;
2553 struct btrfs_root *extent_root = btrfs_extent_root(fs_info, bytenr);
2554 struct btrfs_path *path = iter->path;
2555 struct btrfs_extent_item *ei;
2556 struct btrfs_key key;
2559 key.objectid = bytenr;
2560 key.type = BTRFS_METADATA_ITEM_KEY;
2561 key.offset = (u64)-1;
2562 iter->bytenr = bytenr;
2564 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
2571 if (path->slots[0] == 0) {
2572 WARN_ON(IS_ENABLED(CONFIG_BTRFS_DEBUG));
2578 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
2579 if ((key.type != BTRFS_EXTENT_ITEM_KEY &&
2580 key.type != BTRFS_METADATA_ITEM_KEY) || key.objectid != bytenr) {
2584 memcpy(&iter->cur_key, &key, sizeof(key));
2585 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2587 iter->end_ptr = (u32)(iter->item_ptr +
2588 btrfs_item_size(path->nodes[0], path->slots[0]));
2589 ei = btrfs_item_ptr(path->nodes[0], path->slots[0],
2590 struct btrfs_extent_item);
2593 * Only support iteration on tree backref yet.
2595 * This is an extra precaution for non skinny-metadata, where
2596 * EXTENT_ITEM is also used for tree blocks, that we can only use
2597 * extent flags to determine if it's a tree block.
2599 if (btrfs_extent_flags(path->nodes[0], ei) & BTRFS_EXTENT_FLAG_DATA) {
2603 iter->cur_ptr = (u32)(iter->item_ptr + sizeof(*ei));
2605 /* If there is no inline backref, go search for keyed backref */
2606 if (iter->cur_ptr >= iter->end_ptr) {
2607 ret = btrfs_next_item(extent_root, path);
2609 /* No inline nor keyed ref */
2617 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key,
2619 if (iter->cur_key.objectid != bytenr ||
2620 (iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY &&
2621 iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY)) {
2625 iter->cur_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2627 iter->item_ptr = iter->cur_ptr;
2628 iter->end_ptr = (u32)(iter->item_ptr + btrfs_item_size(
2629 path->nodes[0], path->slots[0]));
2634 btrfs_backref_iter_release(iter);
2639 * Go to the next backref item of current bytenr, can be either inlined or
2642 * Caller needs to check whether it's inline ref or not by iter->cur_key.
2644 * Return 0 if we get next backref without problem.
2645 * Return >0 if there is no extra backref for this bytenr.
2646 * Return <0 if there is something wrong happened.
2648 int btrfs_backref_iter_next(struct btrfs_backref_iter *iter)
2650 struct extent_buffer *eb = btrfs_backref_get_eb(iter);
2651 struct btrfs_root *extent_root;
2652 struct btrfs_path *path = iter->path;
2653 struct btrfs_extent_inline_ref *iref;
2657 if (btrfs_backref_iter_is_inline_ref(iter)) {
2658 /* We're still inside the inline refs */
2659 ASSERT(iter->cur_ptr < iter->end_ptr);
2661 if (btrfs_backref_has_tree_block_info(iter)) {
2662 /* First tree block info */
2663 size = sizeof(struct btrfs_tree_block_info);
2665 /* Use inline ref type to determine the size */
2668 iref = (struct btrfs_extent_inline_ref *)
2669 ((unsigned long)iter->cur_ptr);
2670 type = btrfs_extent_inline_ref_type(eb, iref);
2672 size = btrfs_extent_inline_ref_size(type);
2674 iter->cur_ptr += size;
2675 if (iter->cur_ptr < iter->end_ptr)
2678 /* All inline items iterated, fall through */
2681 /* We're at keyed items, there is no inline item, go to the next one */
2682 extent_root = btrfs_extent_root(iter->fs_info, iter->bytenr);
2683 ret = btrfs_next_item(extent_root, iter->path);
2687 btrfs_item_key_to_cpu(path->nodes[0], &iter->cur_key, path->slots[0]);
2688 if (iter->cur_key.objectid != iter->bytenr ||
2689 (iter->cur_key.type != BTRFS_TREE_BLOCK_REF_KEY &&
2690 iter->cur_key.type != BTRFS_SHARED_BLOCK_REF_KEY))
2692 iter->item_ptr = (u32)btrfs_item_ptr_offset(path->nodes[0],
2694 iter->cur_ptr = iter->item_ptr;
2695 iter->end_ptr = iter->item_ptr + (u32)btrfs_item_size(path->nodes[0],
2700 void btrfs_backref_init_cache(struct btrfs_fs_info *fs_info,
2701 struct btrfs_backref_cache *cache, int is_reloc)
2705 cache->rb_root = RB_ROOT;
2706 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2707 INIT_LIST_HEAD(&cache->pending[i]);
2708 INIT_LIST_HEAD(&cache->changed);
2709 INIT_LIST_HEAD(&cache->detached);
2710 INIT_LIST_HEAD(&cache->leaves);
2711 INIT_LIST_HEAD(&cache->pending_edge);
2712 INIT_LIST_HEAD(&cache->useless_node);
2713 cache->fs_info = fs_info;
2714 cache->is_reloc = is_reloc;
2717 struct btrfs_backref_node *btrfs_backref_alloc_node(
2718 struct btrfs_backref_cache *cache, u64 bytenr, int level)
2720 struct btrfs_backref_node *node;
2722 ASSERT(level >= 0 && level < BTRFS_MAX_LEVEL);
2723 node = kzalloc(sizeof(*node), GFP_NOFS);
2727 INIT_LIST_HEAD(&node->list);
2728 INIT_LIST_HEAD(&node->upper);
2729 INIT_LIST_HEAD(&node->lower);
2730 RB_CLEAR_NODE(&node->rb_node);
2732 node->level = level;
2733 node->bytenr = bytenr;
2738 struct btrfs_backref_edge *btrfs_backref_alloc_edge(
2739 struct btrfs_backref_cache *cache)
2741 struct btrfs_backref_edge *edge;
2743 edge = kzalloc(sizeof(*edge), GFP_NOFS);
2750 * Drop the backref node from cache, also cleaning up all its
2751 * upper edges and any uncached nodes in the path.
2753 * This cleanup happens bottom up, thus the node should either
2754 * be the lowest node in the cache or a detached node.
2756 void btrfs_backref_cleanup_node(struct btrfs_backref_cache *cache,
2757 struct btrfs_backref_node *node)
2759 struct btrfs_backref_node *upper;
2760 struct btrfs_backref_edge *edge;
2765 BUG_ON(!node->lowest && !node->detached);
2766 while (!list_empty(&node->upper)) {
2767 edge = list_entry(node->upper.next, struct btrfs_backref_edge,
2769 upper = edge->node[UPPER];
2770 list_del(&edge->list[LOWER]);
2771 list_del(&edge->list[UPPER]);
2772 btrfs_backref_free_edge(cache, edge);
2775 * Add the node to leaf node list if no other child block
2778 if (list_empty(&upper->lower)) {
2779 list_add_tail(&upper->lower, &cache->leaves);
2784 btrfs_backref_drop_node(cache, node);
2788 * Release all nodes/edges from current cache
2790 void btrfs_backref_release_cache(struct btrfs_backref_cache *cache)
2792 struct btrfs_backref_node *node;
2795 while (!list_empty(&cache->detached)) {
2796 node = list_entry(cache->detached.next,
2797 struct btrfs_backref_node, list);
2798 btrfs_backref_cleanup_node(cache, node);
2801 while (!list_empty(&cache->leaves)) {
2802 node = list_entry(cache->leaves.next,
2803 struct btrfs_backref_node, lower);
2804 btrfs_backref_cleanup_node(cache, node);
2807 cache->last_trans = 0;
2809 for (i = 0; i < BTRFS_MAX_LEVEL; i++)
2810 ASSERT(list_empty(&cache->pending[i]));
2811 ASSERT(list_empty(&cache->pending_edge));
2812 ASSERT(list_empty(&cache->useless_node));
2813 ASSERT(list_empty(&cache->changed));
2814 ASSERT(list_empty(&cache->detached));
2815 ASSERT(RB_EMPTY_ROOT(&cache->rb_root));
2816 ASSERT(!cache->nr_nodes);
2817 ASSERT(!cache->nr_edges);
2821 * Handle direct tree backref
2823 * Direct tree backref means, the backref item shows its parent bytenr
2824 * directly. This is for SHARED_BLOCK_REF backref (keyed or inlined).
2826 * @ref_key: The converted backref key.
2827 * For keyed backref, it's the item key.
2828 * For inlined backref, objectid is the bytenr,
2829 * type is btrfs_inline_ref_type, offset is
2830 * btrfs_inline_ref_offset.
2832 static int handle_direct_tree_backref(struct btrfs_backref_cache *cache,
2833 struct btrfs_key *ref_key,
2834 struct btrfs_backref_node *cur)
2836 struct btrfs_backref_edge *edge;
2837 struct btrfs_backref_node *upper;
2838 struct rb_node *rb_node;
2840 ASSERT(ref_key->type == BTRFS_SHARED_BLOCK_REF_KEY);
2842 /* Only reloc root uses backref pointing to itself */
2843 if (ref_key->objectid == ref_key->offset) {
2844 struct btrfs_root *root;
2846 cur->is_reloc_root = 1;
2847 /* Only reloc backref cache cares about a specific root */
2848 if (cache->is_reloc) {
2849 root = find_reloc_root(cache->fs_info, cur->bytenr);
2855 * For generic purpose backref cache, reloc root node
2858 list_add(&cur->list, &cache->useless_node);
2863 edge = btrfs_backref_alloc_edge(cache);
2867 rb_node = rb_simple_search(&cache->rb_root, ref_key->offset);
2869 /* Parent node not yet cached */
2870 upper = btrfs_backref_alloc_node(cache, ref_key->offset,
2873 btrfs_backref_free_edge(cache, edge);
2878 * Backrefs for the upper level block isn't cached, add the
2879 * block to pending list
2881 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
2883 /* Parent node already cached */
2884 upper = rb_entry(rb_node, struct btrfs_backref_node, rb_node);
2885 ASSERT(upper->checked);
2886 INIT_LIST_HEAD(&edge->list[UPPER]);
2888 btrfs_backref_link_edge(edge, cur, upper, LINK_LOWER);
2893 * Handle indirect tree backref
2895 * Indirect tree backref means, we only know which tree the node belongs to.
2896 * We still need to do a tree search to find out the parents. This is for
2897 * TREE_BLOCK_REF backref (keyed or inlined).
2899 * @ref_key: The same as @ref_key in handle_direct_tree_backref()
2900 * @tree_key: The first key of this tree block.
2901 * @path: A clean (released) path, to avoid allocating path every time
2902 * the function get called.
2904 static int handle_indirect_tree_backref(struct btrfs_backref_cache *cache,
2905 struct btrfs_path *path,
2906 struct btrfs_key *ref_key,
2907 struct btrfs_key *tree_key,
2908 struct btrfs_backref_node *cur)
2910 struct btrfs_fs_info *fs_info = cache->fs_info;
2911 struct btrfs_backref_node *upper;
2912 struct btrfs_backref_node *lower;
2913 struct btrfs_backref_edge *edge;
2914 struct extent_buffer *eb;
2915 struct btrfs_root *root;
2916 struct rb_node *rb_node;
2918 bool need_check = true;
2921 root = btrfs_get_fs_root(fs_info, ref_key->offset, false);
2923 return PTR_ERR(root);
2924 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
2927 if (btrfs_root_level(&root->root_item) == cur->level) {
2929 ASSERT(btrfs_root_bytenr(&root->root_item) == cur->bytenr);
2931 * For reloc backref cache, we may ignore reloc root. But for
2932 * general purpose backref cache, we can't rely on
2933 * btrfs_should_ignore_reloc_root() as it may conflict with
2934 * current running relocation and lead to missing root.
2936 * For general purpose backref cache, reloc root detection is
2937 * completely relying on direct backref (key->offset is parent
2938 * bytenr), thus only do such check for reloc cache.
2940 if (btrfs_should_ignore_reloc_root(root) && cache->is_reloc) {
2941 btrfs_put_root(root);
2942 list_add(&cur->list, &cache->useless_node);
2949 level = cur->level + 1;
2951 /* Search the tree to find parent blocks referring to the block */
2952 path->search_commit_root = 1;
2953 path->skip_locking = 1;
2954 path->lowest_level = level;
2955 ret = btrfs_search_slot(NULL, root, tree_key, path, 0, 0);
2956 path->lowest_level = 0;
2958 btrfs_put_root(root);
2961 if (ret > 0 && path->slots[level] > 0)
2962 path->slots[level]--;
2964 eb = path->nodes[level];
2965 if (btrfs_node_blockptr(eb, path->slots[level]) != cur->bytenr) {
2967 "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)",
2968 cur->bytenr, level - 1, root->root_key.objectid,
2969 tree_key->objectid, tree_key->type, tree_key->offset);
2970 btrfs_put_root(root);
2976 /* Add all nodes and edges in the path */
2977 for (; level < BTRFS_MAX_LEVEL; level++) {
2978 if (!path->nodes[level]) {
2979 ASSERT(btrfs_root_bytenr(&root->root_item) ==
2981 /* Same as previous should_ignore_reloc_root() call */
2982 if (btrfs_should_ignore_reloc_root(root) &&
2984 btrfs_put_root(root);
2985 list_add(&lower->list, &cache->useless_node);
2992 edge = btrfs_backref_alloc_edge(cache);
2994 btrfs_put_root(root);
2999 eb = path->nodes[level];
3000 rb_node = rb_simple_search(&cache->rb_root, eb->start);
3002 upper = btrfs_backref_alloc_node(cache, eb->start,
3005 btrfs_put_root(root);
3006 btrfs_backref_free_edge(cache, edge);
3010 upper->owner = btrfs_header_owner(eb);
3011 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state))
3015 * If we know the block isn't shared we can avoid
3016 * checking its backrefs.
3018 if (btrfs_block_can_be_shared(root, eb))
3024 * Add the block to pending list if we need to check its
3025 * backrefs, we only do this once while walking up a
3026 * tree as we will catch anything else later on.
3028 if (!upper->checked && need_check) {
3030 list_add_tail(&edge->list[UPPER],
3031 &cache->pending_edge);
3035 INIT_LIST_HEAD(&edge->list[UPPER]);
3038 upper = rb_entry(rb_node, struct btrfs_backref_node,
3040 ASSERT(upper->checked);
3041 INIT_LIST_HEAD(&edge->list[UPPER]);
3043 upper->owner = btrfs_header_owner(eb);
3045 btrfs_backref_link_edge(edge, lower, upper, LINK_LOWER);
3048 btrfs_put_root(root);
3055 btrfs_release_path(path);
3060 * Add backref node @cur into @cache.
3062 * NOTE: Even if the function returned 0, @cur is not yet cached as its upper
3063 * links aren't yet bi-directional. Needs to finish such links.
3064 * Use btrfs_backref_finish_upper_links() to finish such linkage.
3066 * @path: Released path for indirect tree backref lookup
3067 * @iter: Released backref iter for extent tree search
3068 * @node_key: The first key of the tree block
3070 int btrfs_backref_add_tree_node(struct btrfs_backref_cache *cache,
3071 struct btrfs_path *path,
3072 struct btrfs_backref_iter *iter,
3073 struct btrfs_key *node_key,
3074 struct btrfs_backref_node *cur)
3076 struct btrfs_fs_info *fs_info = cache->fs_info;
3077 struct btrfs_backref_edge *edge;
3078 struct btrfs_backref_node *exist;
3081 ret = btrfs_backref_iter_start(iter, cur->bytenr);
3085 * We skip the first btrfs_tree_block_info, as we don't use the key
3086 * stored in it, but fetch it from the tree block
3088 if (btrfs_backref_has_tree_block_info(iter)) {
3089 ret = btrfs_backref_iter_next(iter);
3092 /* No extra backref? This means the tree block is corrupted */
3098 WARN_ON(cur->checked);
3099 if (!list_empty(&cur->upper)) {
3101 * The backref was added previously when processing backref of
3102 * type BTRFS_TREE_BLOCK_REF_KEY
3104 ASSERT(list_is_singular(&cur->upper));
3105 edge = list_entry(cur->upper.next, struct btrfs_backref_edge,
3107 ASSERT(list_empty(&edge->list[UPPER]));
3108 exist = edge->node[UPPER];
3110 * Add the upper level block to pending list if we need check
3113 if (!exist->checked)
3114 list_add_tail(&edge->list[UPPER], &cache->pending_edge);
3119 for (; ret == 0; ret = btrfs_backref_iter_next(iter)) {
3120 struct extent_buffer *eb;
3121 struct btrfs_key key;
3125 eb = btrfs_backref_get_eb(iter);
3127 key.objectid = iter->bytenr;
3128 if (btrfs_backref_iter_is_inline_ref(iter)) {
3129 struct btrfs_extent_inline_ref *iref;
3131 /* Update key for inline backref */
3132 iref = (struct btrfs_extent_inline_ref *)
3133 ((unsigned long)iter->cur_ptr);
3134 type = btrfs_get_extent_inline_ref_type(eb, iref,
3135 BTRFS_REF_TYPE_BLOCK);
3136 if (type == BTRFS_REF_TYPE_INVALID) {
3141 key.offset = btrfs_extent_inline_ref_offset(eb, iref);
3143 key.type = iter->cur_key.type;
3144 key.offset = iter->cur_key.offset;
3148 * Parent node found and matches current inline ref, no need to
3149 * rebuild this node for this inline ref
3152 ((key.type == BTRFS_TREE_BLOCK_REF_KEY &&
3153 exist->owner == key.offset) ||
3154 (key.type == BTRFS_SHARED_BLOCK_REF_KEY &&
3155 exist->bytenr == key.offset))) {
3160 /* SHARED_BLOCK_REF means key.offset is the parent bytenr */
3161 if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) {
3162 ret = handle_direct_tree_backref(cache, &key, cur);
3166 } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) {
3168 btrfs_print_v0_err(fs_info);
3169 btrfs_handle_fs_error(fs_info, ret, NULL);
3171 } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) {
3176 * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset
3177 * means the root objectid. We need to search the tree to get
3178 * its parent bytenr.
3180 ret = handle_indirect_tree_backref(cache, path, &key, node_key,
3189 btrfs_backref_iter_release(iter);
3194 * Finish the upwards linkage created by btrfs_backref_add_tree_node()
3196 int btrfs_backref_finish_upper_links(struct btrfs_backref_cache *cache,
3197 struct btrfs_backref_node *start)
3199 struct list_head *useless_node = &cache->useless_node;
3200 struct btrfs_backref_edge *edge;
3201 struct rb_node *rb_node;
3202 LIST_HEAD(pending_edge);
3204 ASSERT(start->checked);
3206 /* Insert this node to cache if it's not COW-only */
3207 if (!start->cowonly) {
3208 rb_node = rb_simple_insert(&cache->rb_root, start->bytenr,
3211 btrfs_backref_panic(cache->fs_info, start->bytenr,
3213 list_add_tail(&start->lower, &cache->leaves);
3217 * Use breadth first search to iterate all related edges.
3219 * The starting points are all the edges of this node
3221 list_for_each_entry(edge, &start->upper, list[LOWER])
3222 list_add_tail(&edge->list[UPPER], &pending_edge);
3224 while (!list_empty(&pending_edge)) {
3225 struct btrfs_backref_node *upper;
3226 struct btrfs_backref_node *lower;
3228 edge = list_first_entry(&pending_edge,
3229 struct btrfs_backref_edge, list[UPPER]);
3230 list_del_init(&edge->list[UPPER]);
3231 upper = edge->node[UPPER];
3232 lower = edge->node[LOWER];
3234 /* Parent is detached, no need to keep any edges */
3235 if (upper->detached) {
3236 list_del(&edge->list[LOWER]);
3237 btrfs_backref_free_edge(cache, edge);
3239 /* Lower node is orphan, queue for cleanup */
3240 if (list_empty(&lower->upper))
3241 list_add(&lower->list, useless_node);
3246 * All new nodes added in current build_backref_tree() haven't
3247 * been linked to the cache rb tree.
3248 * So if we have upper->rb_node populated, this means a cache
3249 * hit. We only need to link the edge, as @upper and all its
3250 * parents have already been linked.
3252 if (!RB_EMPTY_NODE(&upper->rb_node)) {
3253 if (upper->lowest) {
3254 list_del_init(&upper->lower);
3258 list_add_tail(&edge->list[UPPER], &upper->lower);
3262 /* Sanity check, we shouldn't have any unchecked nodes */
3263 if (!upper->checked) {
3268 /* Sanity check, COW-only node has non-COW-only parent */
3269 if (start->cowonly != upper->cowonly) {
3274 /* Only cache non-COW-only (subvolume trees) tree blocks */
3275 if (!upper->cowonly) {
3276 rb_node = rb_simple_insert(&cache->rb_root, upper->bytenr,
3279 btrfs_backref_panic(cache->fs_info,
3280 upper->bytenr, -EEXIST);
3285 list_add_tail(&edge->list[UPPER], &upper->lower);
3288 * Also queue all the parent edges of this uncached node
3289 * to finish the upper linkage
3291 list_for_each_entry(edge, &upper->upper, list[LOWER])
3292 list_add_tail(&edge->list[UPPER], &pending_edge);
3297 void btrfs_backref_error_cleanup(struct btrfs_backref_cache *cache,
3298 struct btrfs_backref_node *node)
3300 struct btrfs_backref_node *lower;
3301 struct btrfs_backref_node *upper;
3302 struct btrfs_backref_edge *edge;
3304 while (!list_empty(&cache->useless_node)) {
3305 lower = list_first_entry(&cache->useless_node,
3306 struct btrfs_backref_node, list);
3307 list_del_init(&lower->list);
3309 while (!list_empty(&cache->pending_edge)) {
3310 edge = list_first_entry(&cache->pending_edge,
3311 struct btrfs_backref_edge, list[UPPER]);
3312 list_del(&edge->list[UPPER]);
3313 list_del(&edge->list[LOWER]);
3314 lower = edge->node[LOWER];
3315 upper = edge->node[UPPER];
3316 btrfs_backref_free_edge(cache, edge);
3319 * Lower is no longer linked to any upper backref nodes and
3320 * isn't in the cache, we can free it ourselves.
3322 if (list_empty(&lower->upper) &&
3323 RB_EMPTY_NODE(&lower->rb_node))
3324 list_add(&lower->list, &cache->useless_node);
3326 if (!RB_EMPTY_NODE(&upper->rb_node))
3329 /* Add this guy's upper edges to the list to process */
3330 list_for_each_entry(edge, &upper->upper, list[LOWER])
3331 list_add_tail(&edge->list[UPPER],
3332 &cache->pending_edge);
3333 if (list_empty(&upper->upper))
3334 list_add(&upper->list, &cache->useless_node);
3337 while (!list_empty(&cache->useless_node)) {
3338 lower = list_first_entry(&cache->useless_node,
3339 struct btrfs_backref_node, list);
3340 list_del_init(&lower->list);
3343 btrfs_backref_drop_node(cache, lower);
3346 btrfs_backref_cleanup_node(cache, node);
3347 ASSERT(list_empty(&cache->useless_node) &&
3348 list_empty(&cache->pending_edge));