On Mon, Jul 13, 2020 at 04:45:19PM -0700, Eric Sandeen wrote:

I wandered back into this thread for some reason ... ;)

Since iversion/noiversion is /already/ advertised as a vfs-level mount option,
wouldn't exposing it in /proc/mounts solve the original problem here?

("i_version" is wrong, because it's ext4-specific, but "iversion" is handled
by the vfs, so it's meaningful for any filesystems, and it will also trivially
allow mount(2) to preserve it across remounts for all filesystems that set it by
default.)

Seems like that's the fastest path to fixing the current problems, even if a
long-term goal may be to deprecate it altogether.

But they should not be exposed as a mount option.  E.g. for XFS we
decide internally if we have a useful i_version or not, totally
independent of the mount option that leaked into the VFS.  So we'll
need to fix how the flag is used before doing any new work in this area.