On Mon, Jul 24, 2023 at 10:34:20AM -0700, Christoph Hellwig wrote:

On Wed, Jul 12, 2023 at 05:11:00PM -0400, Kent Overstreet wrote:

quoted

bio_iov_iter_get_pages() trims the IO based on the block size of the
block device the IO will be issued to.

However, bcachefs is a multi device filesystem; when we're creating the
bio we don't yet know which block device the bio will be submitted to -
we have to handle the alignment checks elsewhere.

So, we've been trying really hard to always make sure to pass a bdev
to anything that allocates a bio, mostly due due the fact that we
actually derive information like the blk-cgroup associations from it.

The whole blk-cgroup stuff is actually a problem for non-trivial
multi-device setups.  XFS gets away fine because each file just
sits on either the main or RT device and no user I/O goes to the
log device, and btrfs papers over it in a weird way by always
associating with the last added device, which is in many ways gross
and wrong, but at least satisfies the assumptions made in blk-cgroup.

How do you plan to deal with this?  Because I really don't want folks
just to go ahead and ignore the issues, we need to actually sort this
out.

Doing the blk-cgroup association at bio alloc time sounds broken to me,
because of stacking block devices - why was the association not done at
generic_make_request() time?