How far should BLKROSET/set_device_ro() go?

While looking at another bug a while ago, I noticed that in 2.6 at least, set_device_ro() sets the policy on the hd_struct to mark it readonly, but it appears that IO is only really blocked from userspace, via generic_write_checks(). There are bdev_read_only() checks in other places, but

How far should BLKROSET/set_device_ro() go?

While looking at another bug a while ago, I noticed that in 2.6 at least, set_device_ro() sets the policy on the hd_struct to mark it readonly, but it appears that IO is only really blocked from userspace, via generic_write_checks(). There are bdev_read_only() checks in other places, but