On Fri, Aug 11, 2023 at 12:08:11PM +0200, Christoph Hellwig wrote:

Hi all,

this series against the VFS vfs.super branch finishes off the work to remove
get_super and move (almost) all upcalls to use the holder ops.

The first part is the missing btrfs bits so that all file systems use the
super_block as holder.

The second part is various block driver cleanups so that we use proper
interfaces instead of raw calls to __invalidate_device and fsync_bdev.

The last part than replaces __invalidate_device and fsync_bdev with upcalls
to the file system through the holder ops, and finally removes get_super.

It leaves user_get_super and get_active_super around.  The former is not
used for upcalls in the traditional sense, but for legacy UAPI that for
some weird reason take a dev_t argument (ustat) or a block device path
(quotactl).  get_active_super is only used for calling into the file system
on freeze and should get a similar treatment, but given that Darrick has
changes to that code queued up already this will be handled in the next
merge window.

A git tree is available here:

    git://git.infradead.org/users/hch/misc.git remove-get_super

Gitweb:

    http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/remove-get_super

The CI testing didn't show any regressions or lockdep errors related to moving
the device opening around (which is what I was worried about).  If you look at
the status you'll see the subpage compress timed out, that's not you, that's an
existing bug I haven't devoted time to figuring out yet.  You can add

Reviewed-by: Josef Bacik <josef@toxicpanda.com>

to the series, thanks,

Josef