2 * Copyright (C) 2014 SUSE. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
18 * Authors: Mark Fasheh <mfasheh@suse.de>
23 #include <uuid/uuid.h>
24 #include "kerncompat.h"
25 #include "radix-tree.h"
28 #include "print-tree.h"
31 #include "rbtree-utils.h"
33 #include "qgroup-verify.h"
35 /*#define QGROUP_VERIFY_DEBUG*/
36 static unsigned long tot_extents_scanned = 0;
39 static struct qgroup_count *find_count(u64 qgroupid);
43 u64 referenced_compressed;
45 u64 exclusive_compressed;
52 struct btrfs_disk_key key;
53 struct qgroup_info diskinfo;
55 struct qgroup_info info;
57 struct rb_node rb_node;
59 /* Parents when we are a child group */
60 struct list_head groups;
63 * Children when we are a parent group (not currently used but
64 * maintained to mirror kernel handling of qgroups)
66 struct list_head members;
71 static struct counts_tree {
73 unsigned int num_groups;
74 unsigned int rescan_running:1;
75 unsigned int qgroup_inconsist:1;
76 } counts = { .root = RB_ROOT };
78 static struct rb_root by_bytenr = RB_ROOT;
81 * Glue structure to represent the relations between qgroups. Mirrored
84 struct btrfs_qgroup_list {
85 struct list_head next_group;
86 struct list_head next_member;
87 struct qgroup_count *group; /* Parent group */
88 struct qgroup_count *member;
91 /* Allow us to reset ref counts during accounting without zeroing each group. */
92 static u64 qgroup_seq = 1ULL;
94 static inline void update_cur_refcnt(struct qgroup_count *c)
96 if (c->cur_refcnt < qgroup_seq)
97 c->cur_refcnt = qgroup_seq;
101 static inline u64 group_get_cur_refcnt(struct qgroup_count *c)
103 if (c->cur_refcnt < qgroup_seq)
105 return c->cur_refcnt - qgroup_seq;
108 static void inc_qgroup_seq(int root_count)
110 qgroup_seq += root_count + 1;
114 * List of interior tree blocks. We walk this list after loading the
115 * extent tree to resolve implied refs. For each interior node we'll
116 * place a shared ref in the ref tree against each child object. This
117 * allows the shared ref resolving code to do the actual work later of
118 * finding roots to account against.
120 * An implied ref is when a tree block has refs on it that may not
121 * exist in any of its child nodes. Even though the refs might not
122 * exist further down the tree, the fact that our interior node has a
123 * ref means we need to account anything below it to all its roots.
125 static struct ulist *tree_blocks = NULL; /* unode->val = bytenr, ->aux
126 * = tree_block pointer */
138 struct rb_node bytenr_node;
141 #ifdef QGROUP_VERIFY_DEBUG
142 static void print_ref(struct ref *ref)
144 printf("bytenr: %llu\t\tnum_bytes: %llu\t\t parent: %llu\t\t"
145 "root: %llu\n", ref->bytenr, ref->num_bytes,
146 ref->parent, ref->root);
149 static void print_all_refs(void)
151 unsigned long count = 0;
153 struct rb_node *node;
155 node = rb_first(&by_bytenr);
157 ref = rb_entry(node, struct ref, bytenr_node);
162 node = rb_next(node);
165 printf("%lu extents scanned with %lu refs in total.\n",
166 tot_extents_scanned, count);
171 * Store by bytenr in rbtree
173 * The tree is sorted in ascending order by bytenr, then parent, then
174 * root. Since full refs have a parent == 0, those will come before
177 static int compare_ref(struct ref *orig, u64 bytenr, u64 root, u64 parent)
179 if (bytenr < orig->bytenr)
181 if (bytenr > orig->bytenr)
184 if (parent < orig->parent)
186 if (parent > orig->parent)
189 if (root < orig->root)
191 if (root > orig->root)
198 * insert a new ref into the tree. returns the existing ref entry
199 * if one is already there.
201 static struct ref *insert_ref(struct ref *ref)
204 struct rb_node **p = &by_bytenr.rb_node;
205 struct rb_node *parent = NULL;
210 curr = rb_entry(parent, struct ref, bytenr_node);
212 ret = compare_ref(curr, ref->bytenr, ref->root, ref->parent);
221 rb_link_node(&ref->bytenr_node, parent, p);
222 rb_insert_color(&ref->bytenr_node, &by_bytenr);
227 * Partial search, returns the first ref with matching bytenr. Caller
228 * can walk forward from there.
230 * Leftmost refs will be full refs - this is used to our advantage
231 * when resolving roots.
233 static struct ref *find_ref_bytenr(u64 bytenr)
235 struct rb_node *n = by_bytenr.rb_node;
239 ref = rb_entry(n, struct ref, bytenr_node);
241 if (bytenr < ref->bytenr)
243 else if (bytenr > ref->bytenr)
246 /* Walk to the left to find the first item */
247 struct rb_node *node_left = rb_prev(&ref->bytenr_node);
248 struct ref *ref_left;
251 ref_left = rb_entry(node_left, struct ref,
253 if (ref_left->bytenr != ref->bytenr)
256 node_left = rb_prev(node_left);
264 static struct ref *find_ref(u64 bytenr, u64 root, u64 parent)
266 struct rb_node *n = by_bytenr.rb_node;
271 ref = rb_entry(n, struct ref, bytenr_node);
273 ret = compare_ref(ref, bytenr, root, parent);
284 static struct ref *alloc_ref(u64 bytenr, u64 root, u64 parent, u64 num_bytes)
286 struct ref *ref = find_ref(bytenr, root, parent);
288 BUG_ON(parent && root);
291 ref = calloc(1, sizeof(*ref));
293 ref->bytenr = bytenr;
295 ref->parent = parent;
296 ref->num_bytes = num_bytes;
304 static void free_ref_node(struct rb_node *node)
306 struct ref *ref = rb_entry(node, struct ref, bytenr_node);
310 FREE_RB_BASED_TREE(ref, free_ref_node);
313 * Resolves all the possible roots for the ref at parent.
315 static void find_parent_roots(struct ulist *roots, u64 parent)
318 struct rb_node *node;
321 * Search the rbtree for the first ref with bytenr == parent.
322 * Walk forward so long as bytenr == parent, adding resolved root ids.
323 * For each unresolved root, we recurse
325 ref = find_ref_bytenr(parent);
326 node = &ref->bytenr_node;
328 BUG_ON(ref->bytenr != parent);
332 * Random sanity check, are we actually getting the
335 struct rb_node *prev_node = rb_prev(&ref->bytenr_node);
338 prev = rb_entry(prev_node, struct ref, bytenr_node);
339 BUG_ON(prev->bytenr == parent);
345 if (is_fstree(ref->root))
346 ulist_add(roots, ref->root, 0, 0);
348 find_parent_roots(roots, ref->parent);
351 node = rb_next(node);
353 ref = rb_entry(node, struct ref, bytenr_node);
354 } while (node && ref->bytenr == parent);
357 static int account_one_extent(struct ulist *roots, u64 bytenr, u64 num_bytes)
360 u64 id, nr_roots, nr_refs;
361 struct qgroup_count *count;
362 struct ulist *counts = ulist_alloc(0);
363 struct ulist *tmp = ulist_alloc(0);
364 struct ulist_iterator uiter;
365 struct ulist_iterator tmp_uiter;
366 struct ulist_node *unode;
367 struct ulist_node *tmp_unode;
368 struct btrfs_qgroup_list *glist;
370 if (!counts || !tmp) {
376 ULIST_ITER_INIT(&uiter);
377 while ((unode = ulist_next(roots, &uiter))) {
378 BUG_ON(unode->val == 0ULL);
381 * For each root, find their corresponding tracking group and
382 * add it to our qgroups list.
384 count = find_count(unode->val);
388 BUG_ON(!is_fstree(unode->val));
389 ret = ulist_add(counts, count->qgroupid, ptr_to_u64(count), 0);
394 * Now we look for parents (and parents of those...). Use a tmp
395 * ulist here to avoid re-walking (and re-incrementing) our
396 * already added items on every loop iteration.
399 ret = ulist_add(tmp, count->qgroupid, ptr_to_u64(count), 0);
403 ULIST_ITER_INIT(&tmp_uiter);
404 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
405 /* Bump the refcount on a node every time we see it. */
406 count = u64_to_ptr(tmp_unode->aux);
407 update_cur_refcnt(count);
409 list_for_each_entry(glist, &count->groups, next_group) {
410 struct qgroup_count *parent;
411 parent = glist->group;
412 id = parent->qgroupid;
416 ret = ulist_add(counts, id, ptr_to_u64(parent),
420 ret = ulist_add(tmp, id, ptr_to_u64(parent),
429 * Now that we have gathered up and counted all the groups, we
430 * can add bytes for this ref.
432 nr_roots = roots->nnodes;
433 ULIST_ITER_INIT(&uiter);
434 while ((unode = ulist_next(counts, &uiter))) {
435 count = u64_to_ptr(unode->aux);
437 nr_refs = group_get_cur_refcnt(count);
439 count->info.referenced += num_bytes;
440 count->info.referenced_compressed += num_bytes;
442 if (nr_refs == nr_roots) {
443 count->info.exclusive += num_bytes;
444 count->info.exclusive_compressed += num_bytes;
447 #ifdef QGROUP_VERIFY_DEBUG
448 printf("account (%llu, %llu), qgroup %llu/%llu, rfer %llu,"
449 " excl %llu, refs %llu, roots %llu\n", bytenr, num_bytes,
450 btrfs_qgroup_level(count->qgroupid),
451 btrfs_qgroup_subvid(count->qgroupid),
452 count->info.referenced, count->info.exclusive, nr_refs,
457 inc_qgroup_seq(roots->nnodes);
465 static void print_subvol_info(u64 subvolid, u64 bytenr, u64 num_bytes,
466 struct ulist *roots);
468 * Account each ref. Walk the refs, for each set of refs in a
471 * - add the roots for direct refs to the ref roots ulist
473 * - resolve all possible roots for shared refs, insert each
474 * of those into ref_roots ulist (this is a recursive process)
476 * - With all roots resolved we can account the ref - this is done in
477 * account_one_extent().
479 static int account_all_refs(int do_qgroups, u64 search_subvol)
482 struct rb_node *node;
483 u64 bytenr, num_bytes;
484 struct ulist *roots = ulist_alloc(0);
486 node = rb_first(&by_bytenr);
490 ref = rb_entry(node, struct ref, bytenr_node);
492 * Walk forward through the list of refs for this
493 * bytenr, adding roots to our ulist. If it's a full
494 * ref, then we have the easy case. Otherwise we need
495 * to search for roots.
497 bytenr = ref->bytenr;
498 num_bytes = ref->num_bytes;
500 BUG_ON(ref->bytenr != bytenr);
501 BUG_ON(ref->num_bytes != num_bytes);
503 if (is_fstree(ref->root)) {
504 if (ulist_add(roots, ref->root, 0, 0) < 0)
508 find_parent_roots(roots, ref->parent);
512 * When we leave this inner loop, node is set
513 * to next in our tree and will be turned into
514 * a ref object up top
516 node = rb_next(node);
518 ref = rb_entry(node, struct ref, bytenr_node);
519 } while (node && ref->bytenr == bytenr);
522 print_subvol_info(search_subvol, bytenr, num_bytes,
528 if (account_one_extent(roots, bytenr, num_bytes))
535 error("Out of memory while accounting refs for qgroups");
539 static u64 resolve_one_root(u64 bytenr)
541 struct ref *ref = find_ref_bytenr(bytenr);
547 return resolve_one_root(ref->parent);
550 static inline struct tree_block *unode_tree_block(struct ulist_node *unode)
552 return u64_to_ptr(unode->aux);
554 static inline u64 unode_bytenr(struct ulist_node *unode)
559 static int alloc_tree_block(u64 bytenr, u64 num_bytes, int level)
561 struct tree_block *block = calloc(1, sizeof(*block));
564 block->num_bytes = num_bytes;
565 block->level = level;
566 if (ulist_add(tree_blocks, bytenr, ptr_to_u64(block), 0) >= 0)
573 static void free_tree_blocks(void)
575 struct ulist_iterator uiter;
576 struct ulist_node *unode;
581 ULIST_ITER_INIT(&uiter);
582 while ((unode = ulist_next(tree_blocks, &uiter)))
583 free(unode_tree_block(unode));
584 ulist_free(tree_blocks);
588 #ifdef QGROUP_VERIFY_DEBUG
589 static void print_tree_block(u64 bytenr, struct tree_block *block)
592 struct rb_node *node;
594 printf("tree block: %llu\t\tlevel: %d\n", (unsigned long long)bytenr,
597 ref = find_ref_bytenr(bytenr);
598 node = &ref->bytenr_node;
601 node = rb_next(node);
603 ref = rb_entry(node, struct ref, bytenr_node);
604 } while (node && ref->bytenr == bytenr);
609 static void print_all_tree_blocks(void)
611 struct ulist_iterator uiter;
612 struct ulist_node *unode;
617 printf("Listing all found interior tree nodes:\n");
619 ULIST_ITER_INIT(&uiter);
620 while ((unode = ulist_next(tree_blocks, &uiter)))
621 print_tree_block(unode_bytenr(unode), unode_tree_block(unode));
625 static int add_refs_for_leaf_items(struct extent_buffer *eb, u64 ref_parent)
629 u64 bytenr, num_bytes;
630 struct btrfs_key key;
631 struct btrfs_disk_key disk_key;
632 struct btrfs_file_extent_item *fi;
634 nr = btrfs_header_nritems(eb);
635 for (i = 0; i < nr; i++) {
636 btrfs_item_key(eb, &disk_key, i);
637 btrfs_disk_key_to_cpu(&key, &disk_key);
639 if (key.type != BTRFS_EXTENT_DATA_KEY)
642 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
643 /* filter out: inline, disk_bytenr == 0, compressed?
644 * not if we can avoid it */
645 extent_type = btrfs_file_extent_type(eb, fi);
647 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
650 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
654 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
655 if (alloc_ref(bytenr, 0, ref_parent, num_bytes) == NULL)
662 static int travel_tree(struct btrfs_fs_info *info, struct btrfs_root *root,
663 u64 bytenr, u64 num_bytes, u64 ref_parent)
666 struct extent_buffer *eb;
670 // printf("travel_tree: bytenr: %llu\tnum_bytes: %llu\tref_parent: %llu\n",
671 // bytenr, num_bytes, ref_parent);
673 eb = read_tree_block(root, bytenr, num_bytes, 0);
674 if (!extent_buffer_uptodate(eb))
678 /* Don't add a ref for our starting tree block to itself */
679 if (bytenr != ref_parent) {
680 if (alloc_ref(bytenr, 0, ref_parent, num_bytes) == NULL)
684 if (btrfs_is_leaf(eb)) {
685 ret = add_refs_for_leaf_items(eb, ref_parent);
690 * Interior nodes are tuples of (key, bytenr) where key is the
691 * leftmost key in the tree block pointed to by bytenr. We
692 * don't have to care about key here, just follow the bytenr
695 nr = btrfs_header_nritems(eb);
696 for (i = 0; i < nr; i++) {
697 new_bytenr = btrfs_node_blockptr(eb, i);
698 new_num_bytes = root->nodesize;
700 ret = travel_tree(info, root, new_bytenr, new_num_bytes,
705 free_extent_buffer(eb);
709 static int add_refs_for_implied(struct btrfs_fs_info *info, u64 bytenr,
710 struct tree_block *block)
713 u64 root_id = resolve_one_root(bytenr);
714 struct btrfs_root *root;
715 struct btrfs_key key;
717 key.objectid = root_id;
718 key.type = BTRFS_ROOT_ITEM_KEY;
719 key.offset = (u64)-1;
722 * XXX: Don't free the root object as we don't know whether it
723 * came off our fs_info struct or not.
725 root = btrfs_read_fs_root(info, &key);
726 if (!root || IS_ERR(root))
729 ret = travel_tree(info, root, bytenr, block->num_bytes, bytenr);
737 * Place shared refs in the ref tree for each child of an interior tree node.
739 static int map_implied_refs(struct btrfs_fs_info *info)
742 struct ulist_iterator uiter;
743 struct ulist_node *unode;
745 ULIST_ITER_INIT(&uiter);
746 while ((unode = ulist_next(tree_blocks, &uiter))) {
747 ret = add_refs_for_implied(info, unode_bytenr(unode),
748 unode_tree_block(unode));
757 * insert a new root into the tree. returns the existing root entry
758 * if one is already there. qgroupid is used
761 static int insert_count(struct qgroup_count *qc)
763 struct rb_node **p = &counts.root.rb_node;
764 struct rb_node *parent = NULL;
765 struct qgroup_count *curr;
769 curr = rb_entry(parent, struct qgroup_count, rb_node);
771 if (qc->qgroupid < curr->qgroupid)
773 else if (qc->qgroupid > curr->qgroupid)
779 rb_link_node(&qc->rb_node, parent, p);
780 rb_insert_color(&qc->rb_node, &counts.root);
784 static struct qgroup_count *find_count(u64 qgroupid)
786 struct rb_node *n = counts.root.rb_node;
787 struct qgroup_count *count;
790 count = rb_entry(n, struct qgroup_count, rb_node);
792 if (qgroupid < count->qgroupid)
794 else if (qgroupid > count->qgroupid)
802 static struct qgroup_count *alloc_count(struct btrfs_disk_key *key,
803 struct extent_buffer *leaf,
804 struct btrfs_qgroup_info_item *disk)
806 struct qgroup_count *c = calloc(1, sizeof(*c));
807 struct qgroup_info *item;
810 c->qgroupid = btrfs_disk_key_offset(key);
814 item->referenced = btrfs_qgroup_info_referenced(leaf, disk);
815 item->referenced_compressed =
816 btrfs_qgroup_info_referenced_compressed(leaf, disk);
817 item->exclusive = btrfs_qgroup_info_exclusive(leaf, disk);
818 item->exclusive_compressed =
819 btrfs_qgroup_info_exclusive_compressed(leaf, disk);
820 INIT_LIST_HEAD(&c->groups);
821 INIT_LIST_HEAD(&c->members);
823 if (insert_count(c)) {
831 static int add_qgroup_relation(u64 memberid, u64 parentid)
833 struct qgroup_count *member;
834 struct qgroup_count *parent;
835 struct btrfs_qgroup_list *list;
837 if (memberid > parentid)
840 member = find_count(memberid);
841 parent = find_count(parentid);
842 if (!member || !parent)
845 list = calloc(1, sizeof(*list));
849 list->group = parent;
850 list->member = member;
851 list_add_tail(&list->next_group, &member->groups);
852 list_add_tail(&list->next_member, &parent->members);
857 static void read_qgroup_status(struct btrfs_path *path,
858 struct counts_tree *counts)
860 struct btrfs_qgroup_status_item *status_item;
863 status_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
864 struct btrfs_qgroup_status_item);
865 flags = btrfs_qgroup_status_flags(path->nodes[0], status_item);
867 * Since qgroup_inconsist/rescan_running is just one bit,
868 * assign value directly won't work.
870 counts->qgroup_inconsist = !!(flags &
871 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT);
872 counts->rescan_running = !!(flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN);
875 static int load_quota_info(struct btrfs_fs_info *info)
878 struct btrfs_root *root = info->quota_root;
879 struct btrfs_root *tmproot;
880 struct btrfs_path path;
881 struct btrfs_key key;
882 struct btrfs_key root_key;
883 struct btrfs_disk_key disk_key;
884 struct extent_buffer *leaf;
885 struct btrfs_qgroup_info_item *item;
886 struct qgroup_count *count;
888 int search_relations = 0;
892 * Do 2 passes, the first allocates group counts and reads status
893 * items. The 2nd pass picks up relation items and glues them to their
894 * respective count structures.
896 btrfs_init_path(&path);
899 key.objectid = search_relations ? 0 : BTRFS_QGROUP_RELATION_KEY;
902 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
904 fprintf(stderr, "ERROR: Couldn't search slot: %d\n", ret);
909 leaf = path.nodes[0];
911 nr = btrfs_header_nritems(leaf);
912 for(i = 0; i < nr; i++) {
913 btrfs_item_key(leaf, &disk_key, i);
914 btrfs_disk_key_to_cpu(&key, &disk_key);
916 if (search_relations) {
917 if (key.type == BTRFS_QGROUP_RELATION_KEY) {
918 ret = add_qgroup_relation(key.objectid,
921 error("out of memory");
928 if (key.type == BTRFS_QGROUP_STATUS_KEY) {
929 read_qgroup_status(&path, &counts);
934 * At this point, we can ignore anything that
935 * isn't a qgroup info.
937 if (key.type != BTRFS_QGROUP_INFO_KEY)
940 item = btrfs_item_ptr(leaf, i,
941 struct btrfs_qgroup_info_item);
943 count = alloc_count(&disk_key, leaf, item);
946 fprintf(stderr, "ERROR: out of memory\n");
950 root_key.objectid = key.offset;
951 root_key.type = BTRFS_ROOT_ITEM_KEY;
952 root_key.offset = (u64)-1;
953 tmproot = btrfs_read_fs_root_no_cache(info, &root_key);
954 if (tmproot && !IS_ERR(tmproot)) {
955 count->subvol_exists = 1;
956 btrfs_free_fs_root(tmproot);
960 ret = btrfs_next_leaf(root, &path);
966 btrfs_release_path(&path);
968 if (!search_relations) {
969 search_relations = 1;
977 static int add_inline_refs(struct btrfs_fs_info *info,
978 struct extent_buffer *ei_leaf, int slot,
979 u64 bytenr, u64 num_bytes, int meta_item)
981 struct btrfs_extent_item *ei;
982 struct btrfs_extent_inline_ref *iref;
983 struct btrfs_extent_data_ref *dref;
984 u64 flags, root_obj, offset, parent;
985 u32 item_size = btrfs_item_size_nr(ei_leaf, slot);
990 ei = btrfs_item_ptr(ei_leaf, slot, struct btrfs_extent_item);
991 flags = btrfs_extent_flags(ei_leaf, ei);
993 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !meta_item) {
994 struct btrfs_tree_block_info *tbinfo;
995 tbinfo = (struct btrfs_tree_block_info *)(ei + 1);
996 iref = (struct btrfs_extent_inline_ref *)(tbinfo + 1);
998 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
1001 ptr = (unsigned long)iref;
1002 end = (unsigned long)ei + item_size;
1004 iref = (struct btrfs_extent_inline_ref *)ptr;
1006 parent = root_obj = 0;
1007 offset = btrfs_extent_inline_ref_offset(ei_leaf, iref);
1008 type = btrfs_extent_inline_ref_type(ei_leaf, iref);
1010 case BTRFS_TREE_BLOCK_REF_KEY:
1013 case BTRFS_EXTENT_DATA_REF_KEY:
1014 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1015 root_obj = btrfs_extent_data_ref_root(ei_leaf, dref);
1017 case BTRFS_SHARED_DATA_REF_KEY:
1018 case BTRFS_SHARED_BLOCK_REF_KEY:
1025 if (alloc_ref(bytenr, root_obj, parent, num_bytes) == NULL)
1028 ptr += btrfs_extent_inline_ref_size(type);
1034 static int add_keyed_ref(struct btrfs_fs_info *info,
1035 struct btrfs_key *key,
1036 struct extent_buffer *leaf, int slot,
1037 u64 bytenr, u64 num_bytes)
1039 u64 root_obj = 0, parent = 0;
1040 struct btrfs_extent_data_ref *dref;
1043 case BTRFS_TREE_BLOCK_REF_KEY:
1044 root_obj = key->offset;
1046 case BTRFS_EXTENT_DATA_REF_KEY:
1047 dref = btrfs_item_ptr(leaf, slot, struct btrfs_extent_data_ref);
1048 root_obj = btrfs_extent_data_ref_root(leaf, dref);
1050 case BTRFS_SHARED_DATA_REF_KEY:
1051 case BTRFS_SHARED_BLOCK_REF_KEY:
1052 parent = key->offset;
1058 if (alloc_ref(bytenr, root_obj, parent, num_bytes) == NULL)
1065 * return value of 0 indicates leaf or not meta data. The code that
1066 * calls this does not need to make a distinction between the two as
1067 * it is only concerned with intermediate blocks which will always
1070 static int get_tree_block_level(struct btrfs_key *key,
1071 struct extent_buffer *ei_leaf,
1075 int meta_key = key->type == BTRFS_METADATA_ITEM_KEY;
1077 struct btrfs_extent_item *ei;
1079 ei = btrfs_item_ptr(ei_leaf, slot, struct btrfs_extent_item);
1080 flags = btrfs_extent_flags(ei_leaf, ei);
1082 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !meta_key) {
1083 struct btrfs_tree_block_info *tbinfo;
1084 tbinfo = (struct btrfs_tree_block_info *)(ei + 1);
1085 level = btrfs_tree_block_level(ei_leaf, tbinfo);
1086 } else if (meta_key) {
1087 /* skinny metadata */
1088 level = (int)key->offset;
1094 * Walk the extent tree, allocating a ref item for every ref and
1095 * storing it in the bytenr tree.
1097 static int scan_extents(struct btrfs_fs_info *info,
1100 int ret, i, nr, level;
1101 struct btrfs_root *root = info->extent_root;
1102 struct btrfs_key key;
1103 struct btrfs_path path;
1104 struct btrfs_disk_key disk_key;
1105 struct extent_buffer *leaf;
1106 u64 bytenr = 0, num_bytes = 0;
1108 btrfs_init_path(&path);
1110 key.objectid = start;
1114 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1116 fprintf(stderr, "ERROR: Couldn't search slot: %d\n", ret);
1122 leaf = path.nodes[0];
1124 nr = btrfs_header_nritems(leaf);
1125 for(i = 0; i < nr; i++) {
1126 btrfs_item_key(leaf, &disk_key, i);
1127 btrfs_disk_key_to_cpu(&key, &disk_key);
1129 if (key.objectid < start)
1132 if (key.objectid > end)
1135 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1136 key.type == BTRFS_METADATA_ITEM_KEY) {
1139 tot_extents_scanned++;
1141 bytenr = key.objectid;
1142 num_bytes = key.offset;
1143 if (key.type == BTRFS_METADATA_ITEM_KEY) {
1144 num_bytes = info->extent_root->nodesize;
1148 ret = add_inline_refs(info, leaf, i, bytenr,
1153 level = get_tree_block_level(&key, leaf, i);
1155 if (alloc_tree_block(bytenr, num_bytes,
1163 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1165 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1169 * Keyed refs should come after their extent
1170 * item in the tree. As a result, the value of
1171 * bytenr and num_bytes should be unchanged
1172 * from the above block that catches the
1173 * original extent item.
1175 BUG_ON(key.objectid != bytenr);
1177 ret = add_keyed_ref(info, &key, leaf, i, bytenr,
1183 ret = btrfs_next_leaf(root, &path);
1187 "ERROR: Next leaf failed: %d\n", ret);
1196 btrfs_release_path(&path);
1201 static void print_fields(u64 bytes, u64 bytes_compressed, char *prefix,
1204 printf("%s\t\t%s %llu %s compressed %llu\n",
1205 prefix, type, (unsigned long long)bytes, type,
1206 (unsigned long long)bytes_compressed);
1209 static void print_fields_signed(long long bytes,
1210 long long bytes_compressed,
1211 char *prefix, char *type)
1213 printf("%s\t\t%s %lld %s compressed %lld\n",
1214 prefix, type, bytes, type, bytes_compressed);
1217 static inline int qgroup_printable(struct qgroup_count *c)
1219 return !!(c->subvol_exists || btrfs_qgroup_level(c->qgroupid));
1222 static int report_qgroup_difference(struct qgroup_count *count, int verbose)
1225 struct qgroup_info *info = &count->info;
1226 struct qgroup_info *disk = &count->diskinfo;
1227 long long excl_diff = info->exclusive - disk->exclusive;
1228 long long ref_diff = info->referenced - disk->referenced;
1230 is_different = excl_diff || ref_diff;
1232 if (verbose || (is_different && qgroup_printable(count))) {
1233 printf("Counts for qgroup id: %llu/%llu %s\n",
1234 btrfs_qgroup_level(count->qgroupid),
1235 btrfs_qgroup_subvid(count->qgroupid),
1236 is_different ? "are different" : "");
1238 print_fields(info->referenced, info->referenced_compressed,
1239 "our:", "referenced");
1240 print_fields(disk->referenced, disk->referenced_compressed,
1241 "disk:", "referenced");
1243 print_fields_signed(ref_diff, ref_diff,
1244 "diff:", "referenced");
1245 print_fields(info->exclusive, info->exclusive_compressed,
1246 "our:", "exclusive");
1247 print_fields(disk->exclusive, disk->exclusive_compressed,
1248 "disk:", "exclusive");
1250 print_fields_signed(excl_diff, excl_diff,
1251 "diff:", "exclusive");
1253 return (is_different && count->subvol_exists);
1256 int report_qgroups(int all)
1258 struct rb_node *node;
1259 struct qgroup_count *c;
1262 if (counts.rescan_running) {
1265 "Qgroup rescan is running, qgroup counts difference is expected\n");
1268 "Qgroup rescan is running, ignore qgroup check\n");
1272 if (counts.qgroup_inconsist && !counts.rescan_running)
1273 fprintf(stderr, "Qgroup is already inconsistent before checking\n");
1274 node = rb_first(&counts.root);
1276 c = rb_entry(node, struct qgroup_count, rb_node);
1277 ret |= report_qgroup_difference(c, all);
1278 node = rb_next(node);
1283 void free_qgroup_counts(void)
1285 struct rb_node *node;
1286 struct qgroup_count *c;
1287 struct btrfs_qgroup_list *glist, *tmpglist;
1289 node = rb_first(&counts.root);
1291 c = rb_entry(node, struct qgroup_count, rb_node);
1293 list_for_each_entry_safe(glist, tmpglist, &c->groups,
1295 list_del(&glist->next_group);
1296 list_del(&glist->next_member);
1299 list_for_each_entry_safe(glist, tmpglist, &c->members,
1301 list_del(&glist->next_group);
1302 list_del(&glist->next_member);
1306 node = rb_next(node);
1308 rb_erase(&c->rb_node, &counts.root);
1313 int qgroup_verify_all(struct btrfs_fs_info *info)
1317 if (!info->quota_enabled)
1320 tree_blocks = ulist_alloc(0);
1323 "ERROR: Out of memory while allocating ulist.\n");
1327 ret = load_quota_info(info);
1329 fprintf(stderr, "ERROR: Loading qgroups from disk: %d\n", ret);
1334 * Put all extent refs into our rbtree
1336 ret = scan_extents(info, 0, ~0ULL);
1338 fprintf(stderr, "ERROR: while scanning extent tree: %d\n", ret);
1342 ret = map_implied_refs(info);
1344 fprintf(stderr, "ERROR: while mapping refs: %d\n", ret);
1348 ret = account_all_refs(1, 0);
1352 * Don't free the qgroup count records as they will be walked
1353 * later via the print function.
1356 free_ref_tree(&by_bytenr);
1360 static void __print_subvol_info(u64 bytenr, u64 num_bytes, struct ulist *roots)
1362 int n = roots->nnodes;
1363 struct ulist_iterator uiter;
1364 struct ulist_node *unode;
1366 printf("%llu\t%llu\t%d\t", bytenr, num_bytes, n);
1368 ULIST_ITER_INIT(&uiter);
1369 while ((unode = ulist_next(roots, &uiter))) {
1370 printf("%llu ", unode->val);
1375 static void print_subvol_info(u64 subvolid, u64 bytenr, u64 num_bytes,
1376 struct ulist *roots)
1378 struct ulist_iterator uiter;
1379 struct ulist_node *unode;
1381 ULIST_ITER_INIT(&uiter);
1382 while ((unode = ulist_next(roots, &uiter))) {
1383 BUG_ON(unode->val == 0ULL);
1384 if (unode->val == subvolid) {
1385 __print_subvol_info(bytenr, num_bytes, roots);
1393 int print_extent_state(struct btrfs_fs_info *info, u64 subvol)
1397 tree_blocks = ulist_alloc(0);
1400 "ERROR: Out of memory while allocating ulist.\n");
1405 * Put all extent refs into our rbtree
1407 ret = scan_extents(info, 0, ~0ULL);
1409 fprintf(stderr, "ERROR: while scanning extent tree: %d\n", ret);
1413 ret = map_implied_refs(info);
1415 fprintf(stderr, "ERROR: while mapping refs: %d\n", ret);
1419 printf("Offset\t\tLen\tRoot Refs\tRoots\n");
1420 ret = account_all_refs(0, subvol);
1424 free_ref_tree(&by_bytenr);