Linus Torvalds [off-list ref] wrote:

quoted

What about ITER_BVEC_MC ??

That probably would be the best option. Just make it a proper
ITER_xyz, instead of an odd sub-case for one ITER (but set up in such
a way that it looks like it might happen for other ITER_xyz cases).

I'm not sure that buys us anything.  It would then require every call to
iov_iter_is_bvec()[*] to check for two values instead of one - including in
iterate_and_advance() - and *still* we'd have to have the special-casing in
_copy_from_iter() and copy_page_from_iter_atomic().

The issue is that ITER_xyz changes the iteration function - but we don't
actually want to do that; rather, we need to change the step function.

David

[*] There's a bunch of them outside of iov_iter.c.