On Wed, Nov 08, 2023 at 06:07:10AM -0800, Christoph Hellwig wrote: > On Wed, Nov 08, 2023 at 09:22:33AM +0100, Christian Brauner wrote: > > > /mnt/1/foo to /mnt/2/bar will get your dev 8 for /mnt/2/bar > > > > > > So the device number changes at the mount point here, bind mount or not. > > > > Yes, I know. But /mnt/2/ will have the device number of the > > superblock/filesystem it belongs to and so will /mnt/1. Creating a > > bind-mount won't suddenly change the device number and decoupling it > > from the superblock it is a bind-mount of. > > It doesn't any more then just changing st_dev. But at least it aligns > to the boundary that such a change always aligned to in not just Linux > but most (if not all?) Unix variants and thus where it is expected. I'm not parsing that sentence fully tbh. But the point stands that this alters how mountinfo works. bind-mounts don't deviate from the device number of the superblock they belong to and there's no reason to tie that st_dev change in stat() that is btrfs specific to vfsmounts. That's not going to happen and has been rejected before. And that is on top of all other problems with making subvolumes automounted vfsmounts that were outlined.
