b_dev vs. b_rdev confusion

These are probably dumb questions but I haven't been able to find definitive answers...

Is there a set rule on when/where one should use a buffer head's b_dev and when/where one should use b_rdev?

I'm been tracing through the kernel source quite a bit and it seems like above ll_rw_blk.c's submit_bh() b_dev is being used. Below there as the request heads off into whichever block device b_rdev is used. Is this true? Things read like the b_dev/b_rdev distinction was introduced to facilitate block drivers intercepting and redirecting requests (lvm, raid, multipathing, &tc.) and I'm trying to figure out if there's a clean delineation of which is used where and if I can safely twiddle b_rdev and pass the buffer head on without copying it. (Are these what are somtimes called "stacking drivers" here?)

All of the md code looks like it copies the buffer head, setting b_dev=b_rdev="real device" in the new bh and leaving b_dev==b_rdev="logical device" in the original bh. I'm assuming they do this for a reason, but it would be nice from a performance standpoint to just touch b_rdev and b_end_io and be done. Is there something I'm missing which necessitates the copy?

And how do the inode->i_dev, i_rdev and i_bdev fit into this? When are those set and does anything get confused by i_rdev moving?

Thanks in advance for any suggestions/info...

Tim

Oops...I take that back. In the md code for raid1, raid5 and multipath I see copies, but not in lvm or linear.

Tim