pijul nest
guest › [sign in] [explore]
{/} code [~] changes [>] discussions [*] jobs

Fork channel

Create a new channel as a copy of main.

Rename channel

Rename main to:

Delete channel

Delete main? This cannot be undone.

mode name
-rw-r--r-- README.md
-rw-r--r-- bootstrap.ml
-rw-r--r-- build.ml
-rw-r--r-- configuration.elpe
-rw-r--r-- install-vm.ml
-rw-r--r-- install.ml
README

examples/vm — bootable qcow2 + real-hardware installer

A minimal demo of Elpe.Image, elpe’s derivation for building UEFI-bootable disk images out of Ubuntu packages. No dpkg, no chroot, no root privileges: the whole thing is a bash script running inside elpe’s regular build container.

Build the qcow2

cd examples/vm
elpe build build.ml   # -> <store-path>/disk.qcow2 + run.sh

disk.qcow2 is a GPT image with two partitions:

  • /dev/vda1 — 128 MiB FAT32 ESP holding a Unified Kernel Image at /EFI/BOOT/BOOTX64.EFI (kernel + baked-in cmdline).
  • /dev/vda2 — 4 GiB ext4 rootfs populated from the Ubuntu packages listed in Image.base_packages plus the config’s extra_packages.

Boot it in QEMU

The store path ships with run.sh. It expects OVMF firmware:

sudo apt install ovmf qemu-system-x86
$(elpe build build.ml)/run.sh

Install on real hardware

elpe build install.ml       # -> <store-path>/{install.sh,rootfs.tar.zst,BOOTX64.EFI}

Copy the store path to a USB stick booted from any live Linux with sgdisk, mkfs.fat, mkfs.ext4, tar, and zstd available, then:

sudo ./install.sh /dev/sdX

The script wipes the target, lays down GPT+ESP+ext4, extracts the rootfs, and drops the UKI into the ESP.

Install from inside a VM, onto a simulated disk

This is elpe’s answer to booting a NixOS installer ISO and running nixos-install. Instead of needing a separate live Linux, the built image is the installer: it boots, carries the installer payload and tooling, and lays itself down onto a blank second disk.

elpe build install-vm.ml            # -> disk.qcow2 + target.qcow2 + run.sh
$(elpe build install-vm.ml)/run.sh  # boots the live VM with /dev/vdb attached

run.sh wires two virtio disks: disk.qcow2 (the live system) as /dev/vda and a blank target.qcow2 as /dev/vdb. Inside the VM, log in as root (password elpe) and run:

elpe-install /dev/vdb
systemctl poweroff

elpe-install partitions /dev/vdb (GPT + ESP + ext4), extracts the pristine rootfs from /opt/elpe-install/rootfs.tar.zst, and copies the systemd-boot ESP tree from /opt/elpe-install/esp. Then boot what you installed:

$(elpe build install-vm.ml)/boot-installed.sh

The live image and the installed system share one configuration (install-vm.ml); the base modules — init (systemd, from base_packages), DHCP networking (systemd_networkd_dhcp), and SSH (openssh) — are all it takes to reach a usable login on either.

Notes and current limits

  • No initrd. The Ubuntu kernel has virtio-blk and ext4 built in, so QEMU boots without one; on real hardware add an initrd package (e.g. initramfs-tools) once you need extra module loading.
  • /etc/passwd, /etc/shadow, etc. are synthesised from the OCaml configuration — dpkg post-install scripts don’t run.
  • The default kernel_cmdline hard-codes root=/dev/vda2; override it in your configuration subclass if you use a different disk controller or want to switch to root=LABEL=ELPEROOT.