banger/images/golden/Dockerfile
Thales Maciel 49c5c862b2
golden image: fix systemd boot + sshd startup
Three fixes discovered during end-to-end boot testing on Firecracker:

- Install udev + dbus alongside systemd. Both are Recommends of the
  systemd package, skipped by --no-install-recommends. Without udev,
  systemd never activates device units (dev-vdb.device stays inactive
  even after the kernel enumerates /dev/vdb) and the work-disk mount
  hangs forever. dbus is required by a growing set of services
  (logind, systemd-resolved shim, etc.).

- Ship /usr/lib/tmpfiles.d/sshd.conf creating /run/sshd. Debian's
  openssh-server package doesn't ship one, and ssh.service's own
  RuntimeDirectory=sshd fires too late for the ExecStartPre config
  check, which blows up with 'Missing privilege separation directory'.
  The tmpfiles entry runs in systemd-tmpfiles-setup.service well
  before ssh.service starts.

- Rewrite the ssh.service drop-in to reset the main unit's
  ExecStartPre list. Debian ships `sshd -t` as ExecStartPre #1; that
  fails without host keys and terminates the service before our
  `ssh-keygen -A` fires. Reset + re-add in the correct order: mkdir,
  keygen, then the test.

StandardOutput/Error=journal+console on ssh.service so future sshd
failures surface in the firecracker console log too, not only in the
(unreachable) guest journal.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-04-18 14:59:01 -03:00

115 lines
5.1 KiB
Docker

# banger golden image — Debian bookworm sandbox for development + testing.
#
# Two sections:
# 1. ESSENTIAL — what banger's lifecycle requires to boot the guest.
# 2. OPINION — developer conveniences curated for banger sandboxes.
#
# Banger's guest agents (vsock agent, network bootstrap, first-boot unit)
# are injected at `banger image pull` time, not baked here. Keeping them
# out means this image stays portable enough to run in other contexts.
FROM debian:bookworm-slim
ENV DEBIAN_FRONTEND=noninteractive \
LANG=C.UTF-8 \
LC_ALL=C.UTF-8
# -------- 1. ESSENTIAL --------
# Banger needs: an init (systemd + udev + dbus), sshd (the only
# control channel), TLS roots + curl (first-boot installs + mise
# installer), iproute2 (debugging; `ip` is still useful even when
# the kernel sets IP via cmdline).
#
# udev is a Recommends of the systemd package on Debian. With
# --no-install-recommends it's skipped — and without it systemd never
# activates device units, so fstab mounts of /dev/vdb (banger's work
# disk) hang forever waiting for a device that is already enumerated
# by the kernel but never "seen" by systemd. dbus gets the same
# treatment for the same reason (system-bus-ness services wedge
# without it).
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
systemd systemd-sysv udev dbus \
openssh-server \
ca-certificates \
curl \
iproute2 \
&& rm -rf /var/lib/apt/lists/*
# -------- 2. OPINION --------
# Developer sandbox conveniences. Language runtimes are deliberately
# absent — `mise` (below) handles per-repo `.mise.toml`/`.tool-versions`
# on first `vm run`.
# Core CLI + search/nav + build toolchain + lint/debug + editor/session.
RUN apt-get update \
&& apt-get install -y --no-install-recommends \
git jq less tree file unzip zip rsync \
ripgrep fd-find \
build-essential pkg-config make \
shellcheck sqlite3 \
iputils-ping dnsutils \
vim-tiny tmux htop \
&& rm -rf /var/lib/apt/lists/*
# Docker CE (with Compose v2 + buildx) from the official apt repo.
# Nested-VM docker gives Compose workflows hostname/port isolation
# per banger VM, which is a big part of the sandbox story.
RUN install -m 0755 -d /etc/apt/keyrings \
&& curl -fsSL https://download.docker.com/linux/debian/gpg -o /etc/apt/keyrings/docker.asc \
&& chmod a+r /etc/apt/keyrings/docker.asc \
&& printf 'deb [arch=%s signed-by=/etc/apt/keyrings/docker.asc] https://download.docker.com/linux/debian bookworm stable\n' \
"$(dpkg --print-architecture)" > /etc/apt/sources.list.d/docker.list \
&& apt-get update \
&& apt-get install -y --no-install-recommends \
docker-ce docker-ce-cli containerd.io \
docker-buildx-plugin docker-compose-plugin \
&& rm -rf /var/lib/apt/lists/*
# mise — per-repo version manager. Installed system-wide so the
# bashrc activation reaches every shell.
RUN curl -fsSL https://mise.run | MISE_INSTALL_PATH=/usr/local/bin/mise sh \
&& chmod 0755 /usr/local/bin/mise \
&& install -d /etc/profile.d \
&& printf '%s\n' 'if [ -x /usr/local/bin/mise ]; then eval "$(/usr/local/bin/mise activate bash)"; fi' \
> /etc/profile.d/mise.sh \
&& chmod 0644 /etc/profile.d/mise.sh
# Git default branch — matches the old customize.sh opinion.
RUN git config --system init.defaultBranch main
# `fd-find` installs as `fdfind` on Debian to avoid a long-standing name
# clash. Expose the ergonomic name for interactive use.
RUN ln -s /usr/bin/fdfind /usr/local/bin/fd
# Strip per-image identity so every banger VM gets its own.
# - /etc/machine-id: systemd-firstboot regenerates at boot when empty.
# - SSH host keys: removed here; a ssh.service drop-in (below) runs
# `ssh-keygen -A` before sshd so the VM's first boot generates a
# unique set.
# - /run/sshd tmpfiles entry: Debian's openssh-server package doesn't
# ship one, and ssh.service's own `RuntimeDirectory=sshd` fires too
# late for the ExecStartPre config test, so sshd -t blows up with
# "Missing privilege separation directory: /run/sshd" before the
# daemon ever starts. Creating the dir via tmpfiles.d runs early in
# systemd-tmpfiles-setup, well before ssh.service kicks off.
RUN : > /etc/machine-id \
&& rm -f /etc/ssh/ssh_host_*_key /etc/ssh/ssh_host_*_key.pub \
&& install -d /etc/systemd/system/ssh.service.d \
&& printf '%s\n' \
'[Service]' \
'# Reset main unit ExecStartPre list: Debian ships `sshd -t` as' \
'# the first ExecStartPre, which fails on missing host keys and' \
'# short-circuits the service before ours gets a chance to run.' \
'ExecStartPre=' \
'ExecStartPre=/usr/bin/mkdir -p /run/sshd' \
'ExecStartPre=/usr/bin/ssh-keygen -A' \
'ExecStartPre=/usr/sbin/sshd -t' \
'StandardOutput=journal+console' \
'StandardError=journal+console' \
> /etc/systemd/system/ssh.service.d/banger.conf \
&& rm -f /etc/systemd/system/ssh.service.d/regen-host-keys.conf \
&& printf 'd /run/sshd 0755 root root -\n' > /usr/lib/tmpfiles.d/sshd.conf
# No CMD / ENTRYPOINT: banger boots this via systemd as PID 1 after
# first-boot, not via `docker run`.