Sidenote: This in no way implies that the patch to libblkid1 is incorrect. It might very well be a strangeness elsewhere * The kernel not clearing the medium change flag when opening a floppy with O_NONBLOCK * qemu passing a changed media in the first place * mount in cloud-init depending someone (via libblkid1) having accessed fd0 before and cleared the medium change flag this way On the other hand it's a regression and it might be the safest choice for maintenance to just add a special case for floppies in libblkid1 and avoid the O_NONBLOCK in that case.