On Thu, May 08, 2025 at 06:56:10AM +0100, Al Viro wrote: > I'll cook something along those lines (on top of "do_move_mount(): don't > leak MNTNS_PROPAGATING on failures") and if it survives overnight tests > post it tomorrow^Win the morning... Got it. Pushed out to viro/vfs.git#fixes; just 4 commits there at the moment - 1) __legitimize_mnt(): check for MNT_SYNC_UMOUNT should be under mount_lock 2) do_umount(): add missing barrier before refcount checks in sync case 3) do_move_mount(): don't leak MNTNS_PROPAGATING on failures 4) fix IS_MNT_PROPAGATING uses I've got reliable reproducers for #3 and #4, but they'll probably need to be massaged for kselftests - any help with that would be very welcome. Patches and those two reproducers in followups; please review and test.
