{ pkgs, ... }: { name = "boot-stage2"; nodes.machine = { config, pkgs, lib, ... }: { virtualisation = { emptyDiskImages = [ 256 ]; # Mount an ext4 as the upper layer of the Nix store. fileSystems = { "/nix/store" = lib.mkForce { device = "/dev/vdb"; # the above disk image fsType = "ext4"; # data=journal always displays after errors=remount-ro; this is only needed because of the overlay # and #375257 will trigger with `errors=remount-ro` on a non-overlaid store: # see ordering in https://github.com/torvalds/linux/blob/v6.12/fs/ext4/super.c#L2974 options = [ "defaults" "errors=remount-ro" "data=journal" ]; }; }; }; boot = { initrd = { # Format the upper Nix store. postDeviceCommands = '' ${pkgs.e2fsprogs}/bin/mkfs.ext4 /dev/vdb ''; # Overlay the RO store onto it. # Note that bug #375257 can be triggered without an overlay, # using the errors=remount-ro option (or similar) or with an overlay where any of the # paths ends in 'ro'. The offending mountpoint also has to be the last (top) one # if an option ending in 'ro' is the last in the list, so test both cases here. postMountCommands = '' mkdir -p /mnt-root/nix/store/ro /mnt-root/nix/store/rw /mnt-root/nix/store/work mount --bind /mnt-root/nix/.ro-store /mnt-root/nix/store/ro mount -t overlay overlay \ -o lowerdir=/mnt-root/nix/store/ro,upperdir=/mnt-root/nix/store/rw,workdir=/mnt-root/nix/store/work \ /mnt-root/nix/store ''; kernelModules = [ "overlay" ]; }; postBootCommands = '' touch /etc/post-boot-ran mount ''; }; }; testScript = '' machine.wait_for_unit("multi-user.target") machine.succeed("test /etc/post-boot-ran") machine.fail("touch /nix/store/should-not-work"); ''; meta.maintainers = with pkgs.lib.maintainers; [ numinit ]; }