Fsck with s3 backend?

How does the FSCK work with S3, does it have to read all of the blocks, or just the object/commits?

Given S3/S3 compatible services charge for requests and bandwidth I’m wondering how expensive it could get if you have to run an FSCK.