btrfs: zoned: fix double counting of split ordered extent

commit f79645d upstream.

btrfs_add_ordered_extent_*() add num_bytes to fs_info->ordered_bytes.
Then, splitting an ordered extent will call btrfs_add_ordered_extent_*()
again for split extents, leading to double counting of the region of
a split extent. These leaked bytes are finally reported at unmount time
as follow:

  BTRFS info (device dm-1): at unmount dio bytes count 364544

Fix the double counting by subtracting split extent's size from
fs_info->ordered_bytes.

Fixes: d22002f ("btrfs: zoned: split ordered extent when bio is sent")
CC: stable@vger.kernel.org # 5.12+
Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Signed-off-by: Naohiro Aota <naohiro.aota@wdc.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

fs/btrfs/ordered-data.c

@@ -1052,6 +1052,7 @@ static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
 				u64 len)
 {
 	struct inode *inode = ordered->inode;
+	struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
 	u64 file_offset = ordered->file_offset + pos;
 	u64 disk_bytenr = ordered->disk_bytenr + pos;
 	u64 num_bytes = len;
@@ -1069,6 +1070,13 @@ static int clone_ordered_extent(struct btrfs_ordered_extent *ordered, u64 pos,
 	else
 		type = __ffs(flags_masked);
 
+	/*
+	 * The splitting extent is already counted and will be added again
+	 * in btrfs_add_ordered_extent_*(). Subtract num_bytes to avoid
+	 * double counting.
+	 */
+	percpu_counter_add_batch(&fs_info->ordered_bytes, -num_bytes,
+				 fs_info->delalloc_batch);
 	if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered->flags)) {
 		WARN_ON_ONCE(1);
 		ret = btrfs_add_ordered_extent_compress(BTRFS_I(inode),