On Fri, Sep 23, 2022 at 08:46:02AM +0200, Christoph Hellwig wrote: > The new method looks good to me, but the patch structuring here > where we add just the method without caller, than a bunch of > instances and only then use them seems odd to me. Isn't there > some better way to make sure that the newly wire up instances > actually get used from the start? The problem is that we have stacking filesystems and once we switch the vfs over then both stacking filesystems themselves as well as all filesystems they're stackable upon need to be already converted to the new posix acl api. So we can't just switch the vfs or just the stacking filesystems we need them to switch together after conversion has finished. (And we do have more stacking fses then we often think. Yhe famous one is obivously overlayfs but there's ecryptfs and now also ksmbd, technically cachefiles and I'm probably forgetting a few others. Most of these luckily don't mess with posix acls.) I don't think other ways will necessarily give us anything nicer. At least I haven't come up with a better way.
