Re: [PATCH RFC] btrfs: verify the endio function at layer boundary

[Christoph Hellwig <hch@infradead.org>]

So based on my staring at this area for a while I think the fundamental
probem is that btrfs passes a struct bio in a lot of places where it
should pass a btrfs_bio.

Basically all the layers that eventually call into btrfs_map_bio should
all be working on a btrfs_bio.  The fact that the btrfs_bio contains
a struct bio is an implementation detail that should be mostly invisible
to the consumers of the btrfs_bio API.  That also means the "high level"
end_io callbacks should take a btrfs_bio and we can just use good old
type safety for this sanity check.  I have started this work, but as
my bio cleanup series already got big enough I've not finished and
posted it.