btrfs-progs: check/original: detect missing orphan items correctly

[BUG]
If we have a filesystem with a subvolume that has 0 refs but without an
orphan item, btrfs check won't report any error on it.

  $ btrfs ins dump-tree -t root /dev/test/scratch1
  btrfs-progs v6.15
  root tree
  node 5242880 level 1 items 2 free space 119 generation 11 owner ROOT_TREE
  node 5242880 flags 0x1(WRITTEN) backref revision 1
  fs uuid ff32309e-4e90-4402-b1ef-0a1f9f28bfff
  chunk uuid 6c373b6d-c866-4c8c-81fa-608bf5ef25e3
  	key (EXTENT_TREE ROOT_ITEM 0) block 5267456 gen 11
  	key (ROOT_TREE_DIR DIR_ITEM 2378154706) block 5246976 gen 11
  leaf 5267456 items 6 free space 2339 generation 11 owner ROOT_TREE
  leaf 5267456 flags 0x1(WRITTEN) backref revision 1
  fs uuid ff32309e-4e90-4402-b1ef-0a1f9f28bfff
  	[...]
  leaf 5246976 items 6 free space 1613 generation 11 owner ROOT_TREE
  leaf 5246976 flags 0x1(WRITTEN) backref revision 1
  checksum stored 47620783
  checksum calced 47620783
  fs uuid ff32309e-4e90-4402-b1ef-0a1f9f28bfff
  chunk uuid 6c373b6d-c866-4c8c-81fa-608bf5ef25e3
  	[...]
  	item 4 key (256 ROOT_ITEM 0) itemoff 2202 itemsize 439
  		generation 9 root_dirid 256 bytenr 5898240 byte_limit 0 bytes_used 581632
  		last_snapshot 0 flags 0x1000000000000(none) refs 0 <<<
  		drop_progress key (0 UNKNOWN.0 0) drop_level 0
  		level 2 generation_v2 9
  	item 5 key (DATA_RELOC_TREE ROOT_ITEM 0) itemoff 1763 itemsize 439
  		generation 5 root_dirid 256 bytenr 5287936 byte_limit 0 bytes_used 4096
  		last_snapshot 0 flags 0x0(none) refs 1
  		drop_progress key (0 UNKNOWN.0 0) drop_level 0
  		level 0 generation_v2 5
  	^^^ No orphan item for subvolume 256.

Then "btrfs check" will not report anything wrong with it:

  Opening filesystem to check...
  Checking filesystem on /dev/test/scratch1
  UUID: ff32309e-4e90-4402-b1ef-0a1f9f28bfff
  [1/8] checking log skipped (none written)
  [2/8] checking root items
  [3/8] checking extents
  [4/8] checking free space tree
  [5/8] checking fs roots
  [6/8] checking only csums items (without verifying data)
  [7/8] checking root refs
  [8/8] checking quota groups skipped (not enabled on this FS)
  found 638976 bytes used, no error found <<<
  total csum bytes: 0
  total tree bytes: 638976
  total fs tree bytes: 589824
  total extent tree bytes: 16384
  btree space waste bytes: 289501
  file data blocks allocated: 0
   referenced 0

[CAUSE]
Although we have check_orphan_item() call inside check_root_refs(), it
relies the root record to have its 'found_root_item' member set.
Otherwise it will not report this as an error.

But that 'found_root_item' is always set to 0, if the subvolume has zero
ref. This means any subvolume with 0 refs will not have its orphan item
checked.

[FIX]
- Introduce root_record::expected_ref to record the refs in the root
  item
  So that we can properly compare the number of refs for each subvolume
  tree.

- Set root_record::found_root_item to one unconditionally in
  check_fs_root()
  Since we're accessing the fs root through the root item, the root item
  must exist, no need to bother if the root_refs is zero or not.

- Enhance check_root_refs()
  * Always check if btrfs_root_item::refs match root_record::found_ref
  * If the fs tree has no ref, check for the orphan item

Signed-off-by: Qu Wenruo <[email protected]>

Author: adam900710

check/main.c

@@ -3491,22 +3491,29 @@ static int check_root_refs(struct btrfs_root *root,
 		rec = container_of(cache, struct root_record, cache);
 		cache = next_cache_extent(cache);
 
-		if (rec->found_ref == 0 &&
-		    rec->objectid >= BTRFS_FIRST_FREE_OBJECTID &&
+		/* A subvolume tree, should check if its refs match. */
+		if (rec->objectid >= BTRFS_FIRST_CHUNK_TREE_OBJECTID &&
 		    rec->objectid <= BTRFS_LAST_FREE_OBJECTID) {
-			ret = check_orphan_item(gfs_info->tree_root, rec->objectid);
-			if (ret == 0)
-				continue;
-
-			/*
-			 * If we don't have a root item then we likely just have
-			 * a dir item in a snapshot for this root but no actual
-			 * ref key or anything so it's meaningless.
-			 */
-			if (!rec->found_root_item)
+			if (rec->found_ref != rec->expected_ref) {
+				errors++;
+				fprintf(stderr, "fs tree %llu refs mismatch found %u expect %u\n",
+					rec->objectid, rec->found_ref, rec->expected_ref);
 				continue;
-			errors++;
-			fprintf(stderr, "fs tree %llu not referenced\n", rec->objectid);
+			}
+			if (rec->expected_ref == 0) {
+				ret = check_orphan_item(gfs_info->tree_root, rec->objectid);
+				if (ret == 0)
+					continue;
+				/*
+				 * If we don't have a root item then we likely just have
+				 * a dir item in a snapshot for this root but no actual
+				 * ref key or anything so it's meaningless.
+				 */
+				if (!rec->found_root_item)
+					continue;
+				errors++;
+				fprintf(stderr, "fs tree %llu missing orphan item\n", rec->objectid);
+			}
 		}
 
 		error = 0;
@@ -3728,8 +3735,8 @@ static int check_fs_root(struct btrfs_root *root,
 	if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
 		rec = get_root_rec(root_cache, root->root_key.objectid);
 		BUG_ON(IS_ERR(rec));
-		if (btrfs_root_refs(root_item) > 0)
-			rec->found_root_item = 1;
+		rec->found_root_item = 1;
+		rec->expected_ref = btrfs_root_refs(&root->root_item);
 	}
 
 	memset(&root_node, 0, sizeof(root_node));

check/mode-original.h

@@ -263,7 +263,10 @@ struct root_record {
 	struct cache_extent cache;
 	unsigned int found_root_item:1;
 	u64 objectid;
+	/* The found number of refs in tree root. */
 	u32 found_ref;
+	/* The expected number of refs in root item. */
+	u32 expected_ref;
 };
 
 struct ptr_node {