btrfs-progs: corrupt-block: Fix -D option
[btrfs-progs-unstable/devel.git] / qgroup-verify.c
blobe2332be2975add7545a2b54bffd0cb43a04cf7ab
1 /*
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>
21 #include <stdio.h>
22 #include <stdlib.h>
23 #include <uuid/uuid.h>
24 #include "kerncompat.h"
25 #include "radix-tree.h"
26 #include "ctree.h"
27 #include "disk-io.h"
28 #include "print-tree.h"
29 #include "utils.h"
30 #include "kernel-shared/ulist.h"
31 #include "rbtree-utils.h"
32 #include "transaction.h"
33 #include "repair.h"
35 #include "qgroup-verify.h"
37 /*#define QGROUP_VERIFY_DEBUG*/
38 static unsigned long tot_extents_scanned = 0;
40 struct qgroup_count;
41 static struct qgroup_count *find_count(u64 qgroupid);
43 struct qgroup_info {
44 u64 referenced;
45 u64 referenced_compressed;
46 u64 exclusive;
47 u64 exclusive_compressed;
50 struct qgroup_count {
51 u64 qgroupid;
52 int subvol_exists;
54 struct btrfs_disk_key key;
55 struct qgroup_info diskinfo;
57 struct qgroup_info info;
59 struct rb_node rb_node;
61 /* Parents when we are a child group */
62 struct list_head groups;
65 * Children when we are a parent group (not currently used but
66 * maintained to mirror kernel handling of qgroups)
68 struct list_head members;
70 u64 cur_refcnt;
72 struct list_head bad_list;
75 static struct counts_tree {
76 struct rb_root root;
77 unsigned int num_groups;
78 unsigned int rescan_running:1;
79 unsigned int qgroup_inconsist:1;
80 } counts = { .root = RB_ROOT };
82 static LIST_HEAD(bad_qgroups);
84 static struct rb_root by_bytenr = RB_ROOT;
87 * Glue structure to represent the relations between qgroups. Mirrored
88 * from kernel.
90 struct btrfs_qgroup_list {
91 struct list_head next_group;
92 struct list_head next_member;
93 struct qgroup_count *group; /* Parent group */
94 struct qgroup_count *member;
97 /* Allow us to reset ref counts during accounting without zeroing each group. */
98 static u64 qgroup_seq = 1ULL;
100 static inline void update_cur_refcnt(struct qgroup_count *c)
102 if (c->cur_refcnt < qgroup_seq)
103 c->cur_refcnt = qgroup_seq;
104 c->cur_refcnt++;
107 static inline u64 group_get_cur_refcnt(struct qgroup_count *c)
109 if (c->cur_refcnt < qgroup_seq)
110 return 0;
111 return c->cur_refcnt - qgroup_seq;
114 static void inc_qgroup_seq(int root_count)
116 qgroup_seq += root_count + 1;
120 * List of interior tree blocks. We walk this list after loading the
121 * extent tree to resolve implied refs. For each interior node we'll
122 * place a shared ref in the ref tree against each child object. This
123 * allows the shared ref resolving code to do the actual work later of
124 * finding roots to account against.
126 * An implied ref is when a tree block has refs on it that may not
127 * exist in any of its child nodes. Even though the refs might not
128 * exist further down the tree, the fact that our interior node has a
129 * ref means we need to account anything below it to all its roots.
131 static struct ulist *tree_blocks = NULL; /* unode->val = bytenr, ->aux
132 * = tree_block pointer */
133 struct tree_block {
134 int level;
135 u64 num_bytes;
138 struct ref {
139 u64 bytenr;
140 u64 num_bytes;
141 u64 parent;
142 u64 root;
144 struct rb_node bytenr_node;
147 #ifdef QGROUP_VERIFY_DEBUG
148 static void print_ref(struct ref *ref)
150 printf("bytenr: %llu\t\tnum_bytes: %llu\t\t parent: %llu\t\t"
151 "root: %llu\n", ref->bytenr, ref->num_bytes,
152 ref->parent, ref->root);
155 static void print_all_refs(void)
157 unsigned long count = 0;
158 struct ref *ref;
159 struct rb_node *node;
161 node = rb_first(&by_bytenr);
162 while (node) {
163 ref = rb_entry(node, struct ref, bytenr_node);
165 print_ref(ref);
167 count++;
168 node = rb_next(node);
171 printf("%lu extents scanned with %lu refs in total.\n",
172 tot_extents_scanned, count);
174 #endif
177 * Store by bytenr in rbtree
179 * The tree is sorted in ascending order by bytenr, then parent, then
180 * root. Since full refs have a parent == 0, those will come before
181 * shared refs.
183 static int compare_ref(struct ref *orig, u64 bytenr, u64 root, u64 parent)
185 if (bytenr < orig->bytenr)
186 return -1;
187 if (bytenr > orig->bytenr)
188 return 1;
190 if (parent < orig->parent)
191 return -1;
192 if (parent > orig->parent)
193 return 1;
195 if (root < orig->root)
196 return -1;
197 if (root > orig->root)
198 return 1;
200 return 0;
204 * insert a new ref into the tree. returns the existing ref entry
205 * if one is already there.
207 static struct ref *insert_ref(struct ref *ref)
209 int ret;
210 struct rb_node **p = &by_bytenr.rb_node;
211 struct rb_node *parent = NULL;
212 struct ref *curr;
214 while (*p) {
215 parent = *p;
216 curr = rb_entry(parent, struct ref, bytenr_node);
218 ret = compare_ref(curr, ref->bytenr, ref->root, ref->parent);
219 if (ret < 0)
220 p = &(*p)->rb_left;
221 else if (ret > 0)
222 p = &(*p)->rb_right;
223 else
224 return curr;
227 rb_link_node(&ref->bytenr_node, parent, p);
228 rb_insert_color(&ref->bytenr_node, &by_bytenr);
229 return ref;
233 * Partial search, returns the first ref with matching bytenr. Caller
234 * can walk forward from there.
236 * Leftmost refs will be full refs - this is used to our advantage
237 * when resolving roots.
239 static struct ref *find_ref_bytenr(u64 bytenr)
241 struct rb_node *n = by_bytenr.rb_node;
242 struct ref *ref;
244 while (n) {
245 ref = rb_entry(n, struct ref, bytenr_node);
247 if (bytenr < ref->bytenr)
248 n = n->rb_left;
249 else if (bytenr > ref->bytenr)
250 n = n->rb_right;
251 else {
252 /* Walk to the left to find the first item */
253 struct rb_node *node_left = rb_prev(&ref->bytenr_node);
254 struct ref *ref_left;
256 while (node_left) {
257 ref_left = rb_entry(node_left, struct ref,
258 bytenr_node);
259 if (ref_left->bytenr != ref->bytenr)
260 break;
261 ref = ref_left;
262 node_left = rb_prev(node_left);
264 return ref;
267 return NULL;
270 static struct ref *find_ref(u64 bytenr, u64 root, u64 parent)
272 struct rb_node *n = by_bytenr.rb_node;
273 struct ref *ref;
274 int ret;
276 while (n) {
277 ref = rb_entry(n, struct ref, bytenr_node);
279 ret = compare_ref(ref, bytenr, root, parent);
280 if (ret < 0)
281 n = n->rb_left;
282 else if (ret > 0)
283 n = n->rb_right;
284 else
285 return ref;
287 return NULL;
290 static struct ref *alloc_ref(u64 bytenr, u64 root, u64 parent, u64 num_bytes)
292 struct ref *ref = find_ref(bytenr, root, parent);
294 BUG_ON(parent && root);
296 if (ref == NULL) {
297 ref = calloc(1, sizeof(*ref));
298 if (ref) {
299 ref->bytenr = bytenr;
300 ref->root = root;
301 ref->parent = parent;
302 ref->num_bytes = num_bytes;
304 insert_ref(ref);
307 return ref;
310 static void free_ref_node(struct rb_node *node)
312 struct ref *ref = rb_entry(node, struct ref, bytenr_node);
313 free(ref);
316 FREE_RB_BASED_TREE(ref, free_ref_node);
319 * Resolves all the possible roots for the ref at parent.
321 static int find_parent_roots(struct ulist *roots, u64 parent)
323 struct ref *ref;
324 struct rb_node *node;
325 int ret;
328 * Search the rbtree for the first ref with bytenr == parent.
329 * Walk forward so long as bytenr == parent, adding resolved root ids.
330 * For each unresolved root, we recurse
332 ref = find_ref_bytenr(parent);
333 if (!ref) {
334 error("bytenr ref not found for parent %llu",
335 (unsigned long long)parent);
336 return -EIO;
338 node = &ref->bytenr_node;
339 if (ref->bytenr != parent) {
340 error("found bytenr ref does not match parent: %llu != %llu",
341 (unsigned long long)ref->bytenr,
342 (unsigned long long)parent);
343 return -EIO;
348 * Random sanity check, are we actually getting the
349 * leftmost node?
351 struct rb_node *prev_node = rb_prev(&ref->bytenr_node);
352 struct ref *prev;
354 if (prev_node) {
355 prev = rb_entry(prev_node, struct ref, bytenr_node);
356 if (prev->bytenr == parent) {
357 error(
358 "unexpected: prev bytenr same as parent: %llu",
359 (unsigned long long)parent);
360 return -EIO;
365 do {
366 if (ref->root) {
367 if (is_fstree(ref->root)) {
368 ret = ulist_add(roots, ref->root, 0, 0);
369 if (ret < 0)
370 goto out;
372 } else if (ref->parent == ref->bytenr) {
374 * Special loop case for tree reloc tree
376 ref->root = BTRFS_TREE_RELOC_OBJECTID;
377 } else {
378 ret = find_parent_roots(roots, ref->parent);
379 if (ret < 0)
380 goto out;
383 node = rb_next(node);
384 if (node)
385 ref = rb_entry(node, struct ref, bytenr_node);
386 } while (node && ref->bytenr == parent);
388 ret = 0;
389 out:
390 return ret;
393 static int account_one_extent(struct ulist *roots, u64 bytenr, u64 num_bytes)
395 int ret;
396 u64 id, nr_roots, nr_refs;
397 struct qgroup_count *count;
398 struct ulist *counts = ulist_alloc(0);
399 struct ulist *tmp = ulist_alloc(0);
400 struct ulist_iterator uiter;
401 struct ulist_iterator tmp_uiter;
402 struct ulist_node *unode;
403 struct ulist_node *tmp_unode;
404 struct btrfs_qgroup_list *glist;
406 if (!counts || !tmp) {
407 ulist_free(counts);
408 ulist_free(tmp);
409 return ENOMEM;
412 ULIST_ITER_INIT(&uiter);
413 while ((unode = ulist_next(roots, &uiter))) {
414 BUG_ON(unode->val == 0ULL);
417 * For each root, find their corresponding tracking group and
418 * add it to our qgroups list.
420 count = find_count(unode->val);
421 if (!count)
422 continue;
424 BUG_ON(!is_fstree(unode->val));
425 ret = ulist_add(counts, count->qgroupid, ptr_to_u64(count), 0);
426 if (ret < 0)
427 goto out;
430 * Now we look for parents (and parents of those...). Use a tmp
431 * ulist here to avoid re-walking (and re-incrementing) our
432 * already added items on every loop iteration.
434 ulist_reinit(tmp);
435 ret = ulist_add(tmp, count->qgroupid, ptr_to_u64(count), 0);
436 if (ret < 0)
437 goto out;
439 ULIST_ITER_INIT(&tmp_uiter);
440 while ((tmp_unode = ulist_next(tmp, &tmp_uiter))) {
441 /* Bump the refcount on a node every time we see it. */
442 count = u64_to_ptr(tmp_unode->aux);
443 update_cur_refcnt(count);
445 list_for_each_entry(glist, &count->groups, next_group) {
446 struct qgroup_count *parent;
447 parent = glist->group;
448 id = parent->qgroupid;
450 BUG_ON(!count);
452 ret = ulist_add(counts, id, ptr_to_u64(parent),
454 if (ret < 0)
455 goto out;
456 ret = ulist_add(tmp, id, ptr_to_u64(parent),
458 if (ret < 0)
459 goto out;
465 * Now that we have gathered up and counted all the groups, we
466 * can add bytes for this ref.
468 nr_roots = roots->nnodes;
469 ULIST_ITER_INIT(&uiter);
470 while ((unode = ulist_next(counts, &uiter))) {
471 count = u64_to_ptr(unode->aux);
473 nr_refs = group_get_cur_refcnt(count);
474 if (nr_refs) {
475 count->info.referenced += num_bytes;
476 count->info.referenced_compressed += num_bytes;
478 if (nr_refs == nr_roots) {
479 count->info.exclusive += num_bytes;
480 count->info.exclusive_compressed += num_bytes;
483 #ifdef QGROUP_VERIFY_DEBUG
484 printf("account (%llu, %llu), qgroup %llu/%llu, rfer %llu,"
485 " excl %llu, refs %llu, roots %llu\n", bytenr, num_bytes,
486 btrfs_qgroup_level(count->qgroupid),
487 btrfs_qgroup_subvid(count->qgroupid),
488 count->info.referenced, count->info.exclusive, nr_refs,
489 nr_roots);
490 #endif
493 inc_qgroup_seq(roots->nnodes);
494 ret = 0;
495 out:
496 ulist_free(counts);
497 ulist_free(tmp);
498 return ret;
501 static void print_subvol_info(u64 subvolid, u64 bytenr, u64 num_bytes,
502 struct ulist *roots);
504 * Account each ref. Walk the refs, for each set of refs in a
505 * given bytenr:
507 * - add the roots for direct refs to the ref roots ulist
509 * - resolve all possible roots for shared refs, insert each
510 * of those into ref_roots ulist (this is a recursive process)
512 * - With all roots resolved we can account the ref - this is done in
513 * account_one_extent().
515 static int account_all_refs(int do_qgroups, u64 search_subvol)
517 struct ref *ref;
518 struct rb_node *node;
519 u64 bytenr, num_bytes;
520 struct ulist *roots = ulist_alloc(0);
521 int ret;
523 node = rb_first(&by_bytenr);
524 while (node) {
525 ulist_reinit(roots);
527 ref = rb_entry(node, struct ref, bytenr_node);
529 * Walk forward through the list of refs for this
530 * bytenr, adding roots to our ulist. If it's a full
531 * ref, then we have the easy case. Otherwise we need
532 * to search for roots.
534 bytenr = ref->bytenr;
535 num_bytes = ref->num_bytes;
536 do {
537 BUG_ON(ref->bytenr != bytenr);
538 BUG_ON(ref->num_bytes != num_bytes);
539 if (ref->root) {
540 if (is_fstree(ref->root)) {
541 if (ulist_add(roots, ref->root, 0, 0) < 0)
542 goto enomem;
544 } else {
545 ret = find_parent_roots(roots, ref->parent);
546 if (ret < 0)
547 goto enomem;
551 * When we leave this inner loop, node is set
552 * to next in our tree and will be turned into
553 * a ref object up top
555 node = rb_next(node);
556 if (node)
557 ref = rb_entry(node, struct ref, bytenr_node);
558 } while (node && ref->bytenr == bytenr);
560 if (search_subvol)
561 print_subvol_info(search_subvol, bytenr, num_bytes,
562 roots);
564 if (!do_qgroups)
565 continue;
567 if (account_one_extent(roots, bytenr, num_bytes))
568 goto enomem;
571 ulist_free(roots);
572 return 0;
573 enomem:
574 error("Out of memory while accounting refs for qgroups");
575 return -ENOMEM;
578 static u64 resolve_one_root(u64 bytenr)
580 struct ref *ref = find_ref_bytenr(bytenr);
582 BUG_ON(ref == NULL);
584 if (ref->root)
585 return ref->root;
586 if (ref->parent == bytenr)
587 return BTRFS_TREE_RELOC_OBJECTID;
588 return resolve_one_root(ref->parent);
591 static inline struct tree_block *unode_tree_block(struct ulist_node *unode)
593 return u64_to_ptr(unode->aux);
595 static inline u64 unode_bytenr(struct ulist_node *unode)
597 return unode->val;
600 static int alloc_tree_block(u64 bytenr, u64 num_bytes, int level)
602 struct tree_block *block = calloc(1, sizeof(*block));
604 if (block) {
605 block->num_bytes = num_bytes;
606 block->level = level;
607 if (ulist_add(tree_blocks, bytenr, ptr_to_u64(block), 0) >= 0)
608 return 0;
609 free(block);
611 return -ENOMEM;
614 static void free_tree_blocks(void)
616 struct ulist_iterator uiter;
617 struct ulist_node *unode;
619 if (!tree_blocks)
620 return;
622 ULIST_ITER_INIT(&uiter);
623 while ((unode = ulist_next(tree_blocks, &uiter)))
624 free(unode_tree_block(unode));
625 ulist_free(tree_blocks);
626 tree_blocks = NULL;
629 #ifdef QGROUP_VERIFY_DEBUG
630 static void print_tree_block(u64 bytenr, struct tree_block *block)
632 struct ref *ref;
633 struct rb_node *node;
635 printf("tree block: %llu\t\tlevel: %d\n", (unsigned long long)bytenr,
636 block->level);
638 ref = find_ref_bytenr(bytenr);
639 node = &ref->bytenr_node;
640 do {
641 print_ref(ref);
642 node = rb_next(node);
643 if (node)
644 ref = rb_entry(node, struct ref, bytenr_node);
645 } while (node && ref->bytenr == bytenr);
647 printf("\n");
650 static void print_all_tree_blocks(void)
652 struct ulist_iterator uiter;
653 struct ulist_node *unode;
655 if (!tree_blocks)
656 return;
658 printf("Listing all found interior tree nodes:\n");
660 ULIST_ITER_INIT(&uiter);
661 while ((unode = ulist_next(tree_blocks, &uiter)))
662 print_tree_block(unode_bytenr(unode), unode_tree_block(unode));
664 #endif
666 static int add_refs_for_leaf_items(struct extent_buffer *eb, u64 ref_parent)
668 int nr, i;
669 int extent_type;
670 u64 bytenr, num_bytes;
671 struct btrfs_key key;
672 struct btrfs_disk_key disk_key;
673 struct btrfs_file_extent_item *fi;
675 nr = btrfs_header_nritems(eb);
676 for (i = 0; i < nr; i++) {
677 btrfs_item_key(eb, &disk_key, i);
678 btrfs_disk_key_to_cpu(&key, &disk_key);
680 if (key.type != BTRFS_EXTENT_DATA_KEY)
681 continue;
683 fi = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item);
684 /* filter out: inline, disk_bytenr == 0, compressed?
685 * not if we can avoid it */
686 extent_type = btrfs_file_extent_type(eb, fi);
688 if (extent_type == BTRFS_FILE_EXTENT_INLINE)
689 continue;
691 bytenr = btrfs_file_extent_disk_bytenr(eb, fi);
692 if (!bytenr)
693 continue;
695 num_bytes = btrfs_file_extent_disk_num_bytes(eb, fi);
696 if (alloc_ref(bytenr, 0, ref_parent, num_bytes) == NULL)
697 return ENOMEM;
700 return 0;
703 static int travel_tree(struct btrfs_fs_info *info, struct btrfs_root *root,
704 u64 bytenr, u64 num_bytes, u64 ref_parent)
706 int ret, nr, i;
707 struct extent_buffer *eb;
708 u64 new_bytenr;
709 u64 new_num_bytes;
711 // printf("travel_tree: bytenr: %llu\tnum_bytes: %llu\tref_parent: %llu\n",
712 // bytenr, num_bytes, ref_parent);
714 eb = read_tree_block(info, bytenr, 0);
715 if (!extent_buffer_uptodate(eb))
716 return -EIO;
718 ret = 0;
719 /* Don't add a ref for our starting tree block to itself */
720 if (bytenr != ref_parent) {
721 if (alloc_ref(bytenr, 0, ref_parent, num_bytes) == NULL)
722 return ENOMEM;
725 if (btrfs_is_leaf(eb)) {
726 ret = add_refs_for_leaf_items(eb, ref_parent);
727 goto out;
731 * Interior nodes are tuples of (key, bytenr) where key is the
732 * leftmost key in the tree block pointed to by bytenr. We
733 * don't have to care about key here, just follow the bytenr
734 * pointer.
736 nr = btrfs_header_nritems(eb);
737 for (i = 0; i < nr; i++) {
738 new_bytenr = btrfs_node_blockptr(eb, i);
739 new_num_bytes = info->nodesize;
741 ret = travel_tree(info, root, new_bytenr, new_num_bytes,
742 ref_parent);
745 out:
746 free_extent_buffer(eb);
747 return ret;
750 static int add_refs_for_implied(struct btrfs_fs_info *info, u64 bytenr,
751 struct tree_block *block)
753 int ret;
754 u64 root_id = resolve_one_root(bytenr);
755 struct btrfs_root *root;
756 struct btrfs_key key;
758 /* Tree reloc tree doesn't contribute qgroup, skip it */
759 if (root_id == BTRFS_TREE_RELOC_OBJECTID)
760 return 0;
761 key.objectid = root_id;
762 key.type = BTRFS_ROOT_ITEM_KEY;
763 key.offset = (u64)-1;
766 * XXX: Don't free the root object as we don't know whether it
767 * came off our fs_info struct or not.
769 root = btrfs_read_fs_root(info, &key);
770 if (!root || IS_ERR(root))
771 return ENOENT;
773 ret = travel_tree(info, root, bytenr, block->num_bytes, bytenr);
774 if (ret)
775 return ret;
777 return 0;
781 * Place shared refs in the ref tree for each child of an interior tree node.
783 static int map_implied_refs(struct btrfs_fs_info *info)
785 int ret = 0;
786 struct ulist_iterator uiter;
787 struct ulist_node *unode;
789 ULIST_ITER_INIT(&uiter);
790 while ((unode = ulist_next(tree_blocks, &uiter))) {
791 ret = add_refs_for_implied(info, unode_bytenr(unode),
792 unode_tree_block(unode));
793 if (ret)
794 goto out;
796 out:
797 return ret;
801 * insert a new root into the tree. returns the existing root entry
802 * if one is already there. qgroupid is used
803 * as the key
805 static int insert_count(struct qgroup_count *qc)
807 struct rb_node **p = &counts.root.rb_node;
808 struct rb_node *parent = NULL;
809 struct qgroup_count *curr;
811 while (*p) {
812 parent = *p;
813 curr = rb_entry(parent, struct qgroup_count, rb_node);
815 if (qc->qgroupid < curr->qgroupid)
816 p = &(*p)->rb_left;
817 else if (qc->qgroupid > curr->qgroupid)
818 p = &(*p)->rb_right;
819 else
820 return EEXIST;
822 counts.num_groups++;
823 rb_link_node(&qc->rb_node, parent, p);
824 rb_insert_color(&qc->rb_node, &counts.root);
825 return 0;
828 static struct qgroup_count *find_count(u64 qgroupid)
830 struct rb_node *n = counts.root.rb_node;
831 struct qgroup_count *count;
833 while (n) {
834 count = rb_entry(n, struct qgroup_count, rb_node);
836 if (qgroupid < count->qgroupid)
837 n = n->rb_left;
838 else if (qgroupid > count->qgroupid)
839 n = n->rb_right;
840 else
841 return count;
843 return NULL;
846 static struct qgroup_count *alloc_count(struct btrfs_disk_key *key,
847 struct extent_buffer *leaf,
848 struct btrfs_qgroup_info_item *disk)
850 struct qgroup_count *c = calloc(1, sizeof(*c));
851 struct qgroup_info *item;
853 if (c) {
854 c->qgroupid = btrfs_disk_key_offset(key);
855 c->key = *key;
857 item = &c->diskinfo;
858 item->referenced = btrfs_qgroup_info_referenced(leaf, disk);
859 item->referenced_compressed =
860 btrfs_qgroup_info_referenced_compressed(leaf, disk);
861 item->exclusive = btrfs_qgroup_info_exclusive(leaf, disk);
862 item->exclusive_compressed =
863 btrfs_qgroup_info_exclusive_compressed(leaf, disk);
864 INIT_LIST_HEAD(&c->groups);
865 INIT_LIST_HEAD(&c->members);
866 INIT_LIST_HEAD(&c->bad_list);
868 if (insert_count(c)) {
869 free(c);
870 c = NULL;
873 return c;
876 static int add_qgroup_relation(u64 memberid, u64 parentid)
878 struct qgroup_count *member;
879 struct qgroup_count *parent;
880 struct btrfs_qgroup_list *list;
882 if (memberid > parentid)
883 return 0;
885 member = find_count(memberid);
886 parent = find_count(parentid);
887 if (!member || !parent)
888 return -ENOENT;
890 list = calloc(1, sizeof(*list));
891 if (!list)
892 return -ENOMEM;
894 list->group = parent;
895 list->member = member;
896 list_add_tail(&list->next_group, &member->groups);
897 list_add_tail(&list->next_member, &parent->members);
899 return 0;
902 static void read_qgroup_status(struct extent_buffer *eb, int slot,
903 struct counts_tree *counts)
905 struct btrfs_qgroup_status_item *status_item;
906 u64 flags;
908 status_item = btrfs_item_ptr(eb, slot, struct btrfs_qgroup_status_item);
909 flags = btrfs_qgroup_status_flags(eb, status_item);
911 * Since qgroup_inconsist/rescan_running is just one bit,
912 * assign value directly won't work.
914 counts->qgroup_inconsist = !!(flags &
915 BTRFS_QGROUP_STATUS_FLAG_INCONSISTENT);
916 counts->rescan_running = !!(flags & BTRFS_QGROUP_STATUS_FLAG_RESCAN);
919 static int load_quota_info(struct btrfs_fs_info *info)
921 int ret;
922 struct btrfs_root *root = info->quota_root;
923 struct btrfs_root *tmproot;
924 struct btrfs_path path;
925 struct btrfs_key key;
926 struct btrfs_key root_key;
927 struct btrfs_disk_key disk_key;
928 struct extent_buffer *leaf;
929 struct btrfs_qgroup_info_item *item;
930 struct qgroup_count *count;
931 int i, nr;
932 int search_relations = 0;
934 loop:
936 * Do 2 passes, the first allocates group counts and reads status
937 * items. The 2nd pass picks up relation items and glues them to their
938 * respective count structures.
940 btrfs_init_path(&path);
942 key.offset = 0;
943 key.objectid = search_relations ? 0 : BTRFS_QGROUP_RELATION_KEY;
944 key.type = 0;
946 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
947 if (ret < 0) {
948 fprintf(stderr, "ERROR: Couldn't search slot: %d\n", ret);
949 goto out;
952 while (1) {
953 leaf = path.nodes[0];
955 nr = btrfs_header_nritems(leaf);
956 for(i = 0; i < nr; i++) {
957 btrfs_item_key(leaf, &disk_key, i);
958 btrfs_disk_key_to_cpu(&key, &disk_key);
960 if (search_relations) {
961 if (key.type == BTRFS_QGROUP_RELATION_KEY) {
962 ret = add_qgroup_relation(key.objectid,
963 key.offset);
964 if (ret) {
965 error("out of memory");
966 goto out;
969 continue;
972 if (key.type == BTRFS_QGROUP_STATUS_KEY) {
973 read_qgroup_status(leaf, i, &counts);
974 continue;
978 * At this point, we can ignore anything that
979 * isn't a qgroup info.
981 if (key.type != BTRFS_QGROUP_INFO_KEY)
982 continue;
984 item = btrfs_item_ptr(leaf, i,
985 struct btrfs_qgroup_info_item);
987 count = alloc_count(&disk_key, leaf, item);
988 if (!count) {
989 ret = ENOMEM;
990 fprintf(stderr, "ERROR: out of memory\n");
991 goto out;
994 root_key.objectid = key.offset;
995 root_key.type = BTRFS_ROOT_ITEM_KEY;
996 root_key.offset = (u64)-1;
997 tmproot = btrfs_read_fs_root_no_cache(info, &root_key);
998 if (tmproot && !IS_ERR(tmproot)) {
999 count->subvol_exists = 1;
1000 btrfs_free_fs_root(tmproot);
1004 ret = btrfs_next_leaf(root, &path);
1005 if (ret != 0)
1006 break;
1009 ret = 0;
1010 btrfs_release_path(&path);
1012 if (!search_relations) {
1013 search_relations = 1;
1014 goto loop;
1017 out:
1018 return ret;
1021 static int add_inline_refs(struct btrfs_fs_info *info,
1022 struct extent_buffer *ei_leaf, int slot,
1023 u64 bytenr, u64 num_bytes, int meta_item)
1025 struct btrfs_extent_item *ei;
1026 struct btrfs_extent_inline_ref *iref;
1027 struct btrfs_extent_data_ref *dref;
1028 u64 flags, root_obj, offset, parent;
1029 u32 item_size = btrfs_item_size_nr(ei_leaf, slot);
1030 int type;
1031 unsigned long end;
1032 unsigned long ptr;
1034 ei = btrfs_item_ptr(ei_leaf, slot, struct btrfs_extent_item);
1035 flags = btrfs_extent_flags(ei_leaf, ei);
1037 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !meta_item) {
1038 struct btrfs_tree_block_info *tbinfo;
1039 tbinfo = (struct btrfs_tree_block_info *)(ei + 1);
1040 iref = (struct btrfs_extent_inline_ref *)(tbinfo + 1);
1041 } else {
1042 iref = (struct btrfs_extent_inline_ref *)(ei + 1);
1045 ptr = (unsigned long)iref;
1046 end = (unsigned long)ei + item_size;
1047 while (ptr < end) {
1048 iref = (struct btrfs_extent_inline_ref *)ptr;
1050 parent = root_obj = 0;
1051 offset = btrfs_extent_inline_ref_offset(ei_leaf, iref);
1052 type = btrfs_extent_inline_ref_type(ei_leaf, iref);
1053 switch (type) {
1054 case BTRFS_TREE_BLOCK_REF_KEY:
1055 root_obj = offset;
1056 break;
1057 case BTRFS_EXTENT_DATA_REF_KEY:
1058 dref = (struct btrfs_extent_data_ref *)(&iref->offset);
1059 root_obj = btrfs_extent_data_ref_root(ei_leaf, dref);
1060 break;
1061 case BTRFS_SHARED_DATA_REF_KEY:
1062 case BTRFS_SHARED_BLOCK_REF_KEY:
1063 parent = offset;
1064 break;
1065 default:
1066 return 1;
1069 if (alloc_ref(bytenr, root_obj, parent, num_bytes) == NULL)
1070 return ENOMEM;
1072 ptr += btrfs_extent_inline_ref_size(type);
1075 return 0;
1078 static int add_keyed_ref(struct btrfs_fs_info *info,
1079 struct btrfs_key *key,
1080 struct extent_buffer *leaf, int slot,
1081 u64 bytenr, u64 num_bytes)
1083 u64 root_obj = 0, parent = 0;
1084 struct btrfs_extent_data_ref *dref;
1086 switch(key->type) {
1087 case BTRFS_TREE_BLOCK_REF_KEY:
1088 root_obj = key->offset;
1089 break;
1090 case BTRFS_EXTENT_DATA_REF_KEY:
1091 dref = btrfs_item_ptr(leaf, slot, struct btrfs_extent_data_ref);
1092 root_obj = btrfs_extent_data_ref_root(leaf, dref);
1093 break;
1094 case BTRFS_SHARED_DATA_REF_KEY:
1095 case BTRFS_SHARED_BLOCK_REF_KEY:
1096 parent = key->offset;
1097 break;
1098 default:
1099 return 1;
1102 if (alloc_ref(bytenr, root_obj, parent, num_bytes) == NULL)
1103 return ENOMEM;
1105 return 0;
1109 * return value of 0 indicates leaf or not meta data. The code that
1110 * calls this does not need to make a distinction between the two as
1111 * it is only concerned with intermediate blocks which will always
1112 * have level > 0.
1114 static int get_tree_block_level(struct btrfs_key *key,
1115 struct extent_buffer *ei_leaf,
1116 int slot)
1118 int level = 0;
1119 int meta_key = key->type == BTRFS_METADATA_ITEM_KEY;
1120 u64 flags;
1121 struct btrfs_extent_item *ei;
1123 ei = btrfs_item_ptr(ei_leaf, slot, struct btrfs_extent_item);
1124 flags = btrfs_extent_flags(ei_leaf, ei);
1126 if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !meta_key) {
1127 struct btrfs_tree_block_info *tbinfo;
1128 tbinfo = (struct btrfs_tree_block_info *)(ei + 1);
1129 level = btrfs_tree_block_level(ei_leaf, tbinfo);
1130 } else if (meta_key) {
1131 /* skinny metadata */
1132 level = (int)key->offset;
1134 return level;
1138 * Walk the extent tree, allocating a ref item for every ref and
1139 * storing it in the bytenr tree.
1141 static int scan_extents(struct btrfs_fs_info *info,
1142 u64 start, u64 end)
1144 int ret, i, nr, level;
1145 struct btrfs_root *root = info->extent_root;
1146 struct btrfs_key key;
1147 struct btrfs_path path;
1148 struct btrfs_disk_key disk_key;
1149 struct extent_buffer *leaf;
1150 u64 bytenr = 0, num_bytes = 0;
1152 btrfs_init_path(&path);
1154 key.objectid = start;
1155 key.type = 0;
1156 key.offset = 0;
1158 ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
1159 if (ret < 0) {
1160 fprintf(stderr, "ERROR: Couldn't search slot: %d\n", ret);
1161 goto out;
1163 path.reada = READA_BACK;
1165 while (1) {
1166 leaf = path.nodes[0];
1168 nr = btrfs_header_nritems(leaf);
1169 for(i = 0; i < nr; i++) {
1170 btrfs_item_key(leaf, &disk_key, i);
1171 btrfs_disk_key_to_cpu(&key, &disk_key);
1173 if (key.objectid < start)
1174 continue;
1176 if (key.objectid > end)
1177 goto done;
1179 if (key.type == BTRFS_EXTENT_ITEM_KEY ||
1180 key.type == BTRFS_METADATA_ITEM_KEY) {
1181 int meta = 0;
1183 tot_extents_scanned++;
1185 bytenr = key.objectid;
1186 num_bytes = key.offset;
1187 if (key.type == BTRFS_METADATA_ITEM_KEY) {
1188 num_bytes = info->nodesize;
1189 meta = 1;
1192 ret = add_inline_refs(info, leaf, i, bytenr,
1193 num_bytes, meta);
1194 if (ret)
1195 goto out;
1197 level = get_tree_block_level(&key, leaf, i);
1198 if (level) {
1199 if (alloc_tree_block(bytenr, num_bytes,
1200 level))
1201 return ENOMEM;
1204 continue;
1207 if (key.type > BTRFS_SHARED_DATA_REF_KEY)
1208 continue;
1209 if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
1210 continue;
1213 * Keyed refs should come after their extent
1214 * item in the tree. As a result, the value of
1215 * bytenr and num_bytes should be unchanged
1216 * from the above block that catches the
1217 * original extent item.
1219 BUG_ON(key.objectid != bytenr);
1221 ret = add_keyed_ref(info, &key, leaf, i, bytenr,
1222 num_bytes);
1223 if (ret)
1224 goto out;
1227 ret = btrfs_next_leaf(root, &path);
1228 if (ret != 0) {
1229 if (ret < 0) {
1230 fprintf(stderr,
1231 "ERROR: Next leaf failed: %d\n", ret);
1232 goto out;
1234 break;
1237 done:
1238 ret = 0;
1239 out:
1240 btrfs_release_path(&path);
1242 return ret;
1245 static void print_fields(u64 bytes, u64 bytes_compressed, char *prefix,
1246 char *type)
1248 printf("%s\t\t%s %llu %s compressed %llu\n",
1249 prefix, type, (unsigned long long)bytes, type,
1250 (unsigned long long)bytes_compressed);
1253 static void print_fields_signed(long long bytes,
1254 long long bytes_compressed,
1255 char *prefix, char *type)
1257 printf("%s\t\t%s %lld %s compressed %lld\n",
1258 prefix, type, bytes, type, bytes_compressed);
1261 static inline int qgroup_printable(struct qgroup_count *c)
1263 return !!(c->subvol_exists || btrfs_qgroup_level(c->qgroupid));
1266 static int report_qgroup_difference(struct qgroup_count *count, int verbose)
1268 int is_different;
1269 struct qgroup_info *info = &count->info;
1270 struct qgroup_info *disk = &count->diskinfo;
1271 long long excl_diff = info->exclusive - disk->exclusive;
1272 long long ref_diff = info->referenced - disk->referenced;
1274 is_different = excl_diff || ref_diff;
1276 if (verbose || (is_different && qgroup_printable(count))) {
1277 printf("Counts for qgroup id: %llu/%llu %s\n",
1278 btrfs_qgroup_level(count->qgroupid),
1279 btrfs_qgroup_subvid(count->qgroupid),
1280 is_different ? "are different" : "");
1282 print_fields(info->referenced, info->referenced_compressed,
1283 "our:", "referenced");
1284 print_fields(disk->referenced, disk->referenced_compressed,
1285 "disk:", "referenced");
1286 if (ref_diff)
1287 print_fields_signed(ref_diff, ref_diff,
1288 "diff:", "referenced");
1289 print_fields(info->exclusive, info->exclusive_compressed,
1290 "our:", "exclusive");
1291 print_fields(disk->exclusive, disk->exclusive_compressed,
1292 "disk:", "exclusive");
1293 if (excl_diff)
1294 print_fields_signed(excl_diff, excl_diff,
1295 "diff:", "exclusive");
1298 return is_different;
1302 * Report qgroups errors
1303 * Return 0 if nothing wrong.
1304 * Return <0 if any qgroup is inconsistent.
1306 * @all: if set, all qgroup will be checked and reported even already
1307 * inconsistent or under rescan.
1309 int report_qgroups(int all)
1311 struct rb_node *node;
1312 struct qgroup_count *c;
1313 bool found_err = false;
1315 if (!repair && counts.rescan_running) {
1316 if (all) {
1317 printf(
1318 "Qgroup rescan is running, a difference in qgroup counts is expected\n");
1319 } else {
1320 printf(
1321 "Qgroup rescan is running, qgroups will not be printed.\n");
1322 return 0;
1325 if (counts.qgroup_inconsist && !counts.rescan_running)
1326 fprintf(stderr, "Qgroup are marked as inconsistent.\n");
1327 node = rb_first(&counts.root);
1328 while (node) {
1329 c = rb_entry(node, struct qgroup_count, rb_node);
1331 if (report_qgroup_difference(c, all)) {
1332 list_add_tail(&c->bad_list, &bad_qgroups);
1333 found_err = true;
1336 node = rb_next(node);
1338 if (found_err)
1339 return -EUCLEAN;
1340 return 0;
1343 void free_qgroup_counts(void)
1345 struct rb_node *node;
1346 struct qgroup_count *c;
1347 struct btrfs_qgroup_list *glist, *tmpglist;
1349 node = rb_first(&counts.root);
1350 while (node) {
1351 c = rb_entry(node, struct qgroup_count, rb_node);
1353 list_del(&c->bad_list);
1355 list_for_each_entry_safe(glist, tmpglist, &c->groups,
1356 next_group) {
1357 list_del(&glist->next_group);
1358 list_del(&glist->next_member);
1359 free(glist);
1361 list_for_each_entry_safe(glist, tmpglist, &c->members,
1362 next_group) {
1363 list_del(&glist->next_group);
1364 list_del(&glist->next_member);
1365 free(glist);
1368 node = rb_next(node);
1370 rb_erase(&c->rb_node, &counts.root);
1371 free(c);
1375 int qgroup_verify_all(struct btrfs_fs_info *info)
1377 int ret;
1379 if (!info->quota_enabled)
1380 return 0;
1382 tree_blocks = ulist_alloc(0);
1383 if (!tree_blocks) {
1384 fprintf(stderr,
1385 "ERROR: Out of memory while allocating ulist.\n");
1386 return ENOMEM;
1389 ret = load_quota_info(info);
1390 if (ret) {
1391 fprintf(stderr, "ERROR: Loading qgroups from disk: %d\n", ret);
1392 goto out;
1396 * Put all extent refs into our rbtree
1398 ret = scan_extents(info, 0, ~0ULL);
1399 if (ret) {
1400 fprintf(stderr, "ERROR: while scanning extent tree: %d\n", ret);
1401 goto out;
1404 ret = map_implied_refs(info);
1405 if (ret) {
1406 fprintf(stderr, "ERROR: while mapping refs: %d\n", ret);
1407 goto out;
1410 ret = account_all_refs(1, 0);
1412 out:
1414 * Don't free the qgroup count records as they will be walked
1415 * later via the print function.
1417 free_tree_blocks();
1418 free_ref_tree(&by_bytenr);
1419 return ret;
1422 static void __print_subvol_info(u64 bytenr, u64 num_bytes, struct ulist *roots)
1424 int n = roots->nnodes;
1425 struct ulist_iterator uiter;
1426 struct ulist_node *unode;
1428 printf("%llu\t%llu\t%d\t", bytenr, num_bytes, n);
1430 ULIST_ITER_INIT(&uiter);
1431 while ((unode = ulist_next(roots, &uiter))) {
1432 printf("%llu ", unode->val);
1434 printf("\n");
1437 static void print_subvol_info(u64 subvolid, u64 bytenr, u64 num_bytes,
1438 struct ulist *roots)
1440 struct ulist_iterator uiter;
1441 struct ulist_node *unode;
1443 ULIST_ITER_INIT(&uiter);
1444 while ((unode = ulist_next(roots, &uiter))) {
1445 BUG_ON(unode->val == 0ULL);
1446 if (unode->val == subvolid) {
1447 __print_subvol_info(bytenr, num_bytes, roots);
1448 return;
1455 int print_extent_state(struct btrfs_fs_info *info, u64 subvol)
1457 int ret;
1459 tree_blocks = ulist_alloc(0);
1460 if (!tree_blocks) {
1461 fprintf(stderr,
1462 "ERROR: Out of memory while allocating ulist.\n");
1463 return ENOMEM;
1467 * Put all extent refs into our rbtree
1469 ret = scan_extents(info, 0, ~0ULL);
1470 if (ret) {
1471 fprintf(stderr, "ERROR: while scanning extent tree: %d\n", ret);
1472 goto out;
1475 ret = map_implied_refs(info);
1476 if (ret) {
1477 fprintf(stderr, "ERROR: while mapping refs: %d\n", ret);
1478 goto out;
1481 printf("Offset\t\tLen\tRoot Refs\tRoots\n");
1482 ret = account_all_refs(0, subvol);
1484 out:
1485 free_tree_blocks();
1486 free_ref_tree(&by_bytenr);
1487 return ret;
1490 static int repair_qgroup_info(struct btrfs_fs_info *info,
1491 struct qgroup_count *count)
1493 int ret;
1494 struct btrfs_root *root = info->quota_root;
1495 struct btrfs_trans_handle *trans;
1496 struct btrfs_path path;
1497 struct btrfs_qgroup_info_item *info_item;
1498 struct btrfs_key key;
1500 printf("Repair qgroup %llu/%llu\n", btrfs_qgroup_level(count->qgroupid),
1501 btrfs_qgroup_subvid(count->qgroupid));
1503 trans = btrfs_start_transaction(root, 1);
1504 if (IS_ERR(trans))
1505 return PTR_ERR(trans);
1507 btrfs_init_path(&path);
1508 key.objectid = 0;
1509 key.type = BTRFS_QGROUP_INFO_KEY;
1510 key.offset = count->qgroupid;
1511 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
1512 if (ret) {
1513 error("could not find disk item for qgroup %llu/%llu",
1514 btrfs_qgroup_level(count->qgroupid),
1515 btrfs_qgroup_subvid(count->qgroupid));
1516 if (ret > 0)
1517 ret = -ENOENT;
1518 goto out;
1521 info_item = btrfs_item_ptr(path.nodes[0], path.slots[0],
1522 struct btrfs_qgroup_info_item);
1524 btrfs_set_qgroup_info_generation(path.nodes[0], info_item,
1525 trans->transid);
1527 btrfs_set_qgroup_info_referenced(path.nodes[0], info_item,
1528 count->info.referenced);
1529 btrfs_set_qgroup_info_referenced_compressed(path.nodes[0], info_item,
1530 count->info.referenced_compressed);
1532 btrfs_set_qgroup_info_exclusive(path.nodes[0], info_item,
1533 count->info.exclusive);
1534 btrfs_set_qgroup_info_exclusive_compressed(path.nodes[0], info_item,
1535 count->info.exclusive_compressed);
1537 btrfs_mark_buffer_dirty(path.nodes[0]);
1539 out:
1540 btrfs_commit_transaction(trans, root);
1541 btrfs_release_path(&path);
1543 return ret;
1546 static int repair_qgroup_status(struct btrfs_fs_info *info)
1548 int ret;
1549 struct btrfs_root *root = info->quota_root;
1550 struct btrfs_trans_handle *trans;
1551 struct btrfs_path path;
1552 struct btrfs_key key;
1553 struct btrfs_qgroup_status_item *status_item;
1555 printf("Repair qgroup status item\n");
1557 trans = btrfs_start_transaction(root, 1);
1558 if (IS_ERR(trans))
1559 return PTR_ERR(trans);
1561 btrfs_init_path(&path);
1562 key.objectid = 0;
1563 key.type = BTRFS_QGROUP_STATUS_KEY;
1564 key.offset = 0;
1565 ret = btrfs_search_slot(trans, root, &key, &path, 0, 1);
1566 if (ret) {
1567 error("could not find qgroup status item");
1568 if (ret > 0)
1569 ret = -ENOENT;
1570 goto out;
1573 status_item = btrfs_item_ptr(path.nodes[0], path.slots[0],
1574 struct btrfs_qgroup_status_item);
1575 btrfs_set_qgroup_status_flags(path.nodes[0], status_item,
1576 BTRFS_QGROUP_STATUS_FLAG_ON);
1577 btrfs_set_qgroup_status_rescan(path.nodes[0], status_item, 0);
1578 btrfs_set_qgroup_status_generation(path.nodes[0], status_item,
1579 trans->transid);
1581 btrfs_mark_buffer_dirty(path.nodes[0]);
1583 out:
1584 btrfs_commit_transaction(trans, root);
1585 btrfs_release_path(&path);
1587 return ret;
1590 int repair_qgroups(struct btrfs_fs_info *info, int *repaired)
1592 int ret = 0;
1593 struct qgroup_count *count, *tmpcount;
1595 *repaired = 0;
1597 if (!repair)
1598 return 0;
1600 list_for_each_entry_safe(count, tmpcount, &bad_qgroups, bad_list) {
1601 ret = repair_qgroup_info(info, count);
1602 if (ret) {
1603 goto out;
1606 (*repaired)++;
1608 list_del_init(&count->bad_list);
1612 * Do this step last as we want the latest transaction id on
1613 * our qgroup status to avoid a (useless) warning after
1614 * mount.
1616 if (*repaired || counts.qgroup_inconsist || counts.rescan_running) {
1617 ret = repair_qgroup_status(info);
1618 if (ret)
1619 goto out;
1621 (*repaired)++;
1624 out:
1625 return ret;