Compare commits

..
Author SHA1 Message Date
beatzaplenty 9bb626327f Unmount nixos-install's leftover chroot bind mounts before zpool export
Check NixOS configurations / eval-hosts (pull_request) Successful in 10m20s
nixos-install bind-mounts /dev, /proc, /sys (and usually /run) into
/mnt to run the target's activation script in a chroot, and doesn't
unmount them again afterward. Left in place, those nested mounts made
ZFS refuse to unmount its own root dataset at /mnt: zpool export
failed with "cannot unmount '/mnt': pool or dataset busy", and because
of this script's set -e, that killed the script before it ever reached
reboot -- silently defeating the export-before-reboot fix from #40 on
every real run, which is why the ZFS-import stall kept recurring.
2026-07-22 04:07:32 +00:00
beatzaplenty 5d7a6327b7 Allow initrd emergency shell access on baremetal-gui
Check NixOS configurations / eval-hosts (pull_request) Successful in 10m45s
The systemd-based initrd (default here, since this host has a ZFS
root) locks the root account by default, so sulogin refuses a shell
if something in the initrd fails and it drops to emergency mode --
confirmed live: it just loops re-entering the target instead of
prompting, making an initrd-level ZFS import failure impossible to
diagnose from the console. Only affects the pre-switch-root initrd
shell, not the installed system's own login.
2026-07-22 03:47:54 +00:00
beatzaplenty 0b9f124713 Merge pull request 'Export ZFS root pool before rebooting from the auto-installer' (#40) from worktree-fix-zfs-install-export into main
Check NixOS configurations / eval-hosts (push) Successful in 10m25s
Reviewed-on: #40
2026-07-22 03:22:32 +00:00
beatzaplentyandClaude Sonnet 5 9479d56e11 Export ZFS root pool before rebooting from the auto-installer
Check NixOS configurations / eval-hosts (pull_request) Successful in 10m19s
disko's --mode ...,mount leaves the pool imported (needed for
nixos-install to write into /mnt), and the script rebooted straight
into the newly-installed system without exporting it. That pool is
still stamped with the live installer's own hostid, which never
matches the target host's declared networking.hostId, and since
boot.zfs.forceImportRoot is false (the recommended setting, not a bug),
the first real boot refuses to force-import an unexported pool from a
different hostid -- which is exactly the ZFS-import stall baremetal-gui
was hitting after install. Exporting all pools right before reboot (a
no-op for non-ZFS hosts) clears the in-use state so import succeeds
regardless of hostid.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-22 03:10:25 +00:00
beatzaplenty c53c1940d6 Merge pull request 'Prompt for a host-key path interactively as a third fallback' (#39) from worktree-gui-wifi-module into main
Check NixOS configurations / eval-hosts (push) Successful in 10m18s
Reviewed-on: #39
2026-07-22 02:57:36 +00:00
beatzaplenty 79e8f9f2ce Prompt for a host-key path interactively as a third fallback
Check NixOS configurations / eval-hosts (pull_request) Successful in 10m19s
If neither /etc/host-keys nor /root/host-keys has the target's SSH
host key, auto-install.sh previously went straight to "continue
without one anyway?". Added a third option in between, gated on
[[ -t 0 ]] (only offered when there's an actual operator at stdin, never
in an unattended/non-interactive run): prompt for an arbitrary
directory (USB stick, other mount, etc.), and if the key pair is
there, copy it into /root/host-keys and install it to /mnt same as the
existing pre-seeded-key path. Falls through to the original
warning+confirm if the prompt is skipped, the path doesn't have the
key, or the run isn't interactive at all.

docs/auto-installer.md updated to mention the new fallback. Quick
bash -n + shellcheck pass only, per request.
2026-07-22 02:54:06 +00:00
3 changed files with 91 additions and 18 deletions

No files matched your search

+4
View File
@@ -195,6 +195,10 @@ default.
`auto-install.sh` still supports the older manual path as a fallback: if a
host's key isn't baked in (`/etc/host-keys`), it checks `/root/host-keys`
next, where you can `scp` a key in after boot, same as before this migration.
If neither has it and the script is running interactively (an actual
operator at the other end of stdin, not an unattended run), it prompts for
an arbitrary directory to check (a mounted USB stick, another filesystem,
etc.) and copies the key pair into `/root/host-keys` from there if found.
## Storage
+10
View File
@@ -25,4 +25,14 @@
enable = true;
enable32Bit = true;
};
# The systemd-based initrd (default here since this host has a ZFS root --
# see modules/disko/baremetal.nix) locks the root account by default, so
# sulogin refuses to hand over a shell if something in the initrd (e.g.
# the ZFS pool import) fails and it drops to emergency mode -- confirmed
# live: it just loops re-entering the target instead of prompting. This
# only affects the pre-switch-root initrd shell, not the installed
# system's own login, and is worth the tradeoff on a box already reachable
# at the physical console.
boot.initrd.systemd.emergencyAccess = true;
}
+77 -18
View File
@@ -1,17 +1,20 @@
#!/usr/bin/env nix-shell
#!nix-shell -i bash -p jq disko nixos-install-tools
#!nix-shell -i bash -p jq disko nixos-install-tools zfs
# shellcheck shell=bash
# The only genuinely external tools this script calls directly: `jq`
# (parsing the `nix eval` host list) and `disko`/`nixos-install` (the
# install itself). Everything disko shells out to internally
# (parted/sgdisk/mkfs.*/zfs/...) is self-contained -- disko's own
# generated scripts hardcode absolute Nix store paths for those, they
# (parsing the `nix eval` host list), `disko`/`nixos-install` (the
# install itself), and `zpool` (exporting a ZFS root pool before reboot,
# see the comment above that call below). Everything disko shells out to
# internally (parted/sgdisk/mkfs.*/zfs/...) is self-contained -- disko's
# own generated scripts hardcode absolute Nix store paths for those, they
# don't rely on this script's PATH at all (confirmed by inspecting a
# generated system.build.formatScript). The built installer image
# (modules/installer/common.nix) already has all three in
# environment.systemPackages, so this nix-shell wrapper is a fast no-op
# there; it's what makes the script also work standalone (e.g. run
# directly from a checkout on a stock ISO), where they aren't.
# (modules/installer/common.nix, plus the upstream
# installation-cd-minimal.nix it imports via iso.nix) already has all
# four in environment.systemPackages, so this nix-shell wrapper is a
# fast no-op there; it's what makes the script also work standalone
# (e.g. run directly from a checkout on a stock ISO), where they aren't
# guaranteed.
set -eux
set -euo pipefail
@@ -114,15 +117,37 @@ elif [[ -f "/root/host-keys/${choice}_ssh_host_ed25519_key" ]]; then
install -D -m 0600 "/root/host-keys/${choice}_ssh_host_ed25519_key" /mnt/etc/ssh/ssh_host_ed25519_key
install -D -m 0644 "/root/host-keys/${choice}_ssh_host_ed25519_key.pub" /mnt/etc/ssh/ssh_host_ed25519_key.pub
else
echo "WARNING: no SSH host key found for ${choice} (checked /etc/host-keys and /root/host-keys)"
echo "sops-nix secrets (including the login password) will NOT decrypt on first boot."
echo "Run scripts/secrets/prepare-host-key.sh for host ${choice} on your admin workstation first,"
echo "then either rebuild this image with NIXOS_HOST_KEYS_DIR set, or scp the result to"
echo "/root/host-keys/ on this machine."
read -rp "Continue without a pre-seeded key anyway? (y/N): " skip_key
if [[ ! "$skip_key" =~ ^[Yy]$ ]]; then
echo "Aborted."
exit 1
# Third place a key can come from: an arbitrary path the operator
# points at interactively (e.g. a USB stick, a mount from another
# machine) -- only offered when there's an actual human at the other
# end of stdin to ask, never in a non-interactive run.
key_copied=0
if [[ -t 0 ]]; then
echo "No SSH host key found for ${choice} (checked /etc/host-keys and /root/host-keys)."
read -rp "Path to a directory containing ${choice}_ssh_host_ed25519_key(.pub) (blank to skip): " key_src_dir
if [[ -n "$key_src_dir" && -f "${key_src_dir}/${choice}_ssh_host_ed25519_key" && -f "${key_src_dir}/${choice}_ssh_host_ed25519_key.pub" ]]; then
cp "${key_src_dir}/${choice}_ssh_host_ed25519_key" "${key_src_dir}/${choice}_ssh_host_ed25519_key.pub" /root/host-keys/
key_copied=1
elif [[ -n "$key_src_dir" ]]; then
echo "WARNING: ${choice}_ssh_host_ed25519_key(.pub) not found in ${key_src_dir}."
fi
fi
if [[ "$key_copied" -eq 1 ]]; then
echo "Copied SSH host key for ${choice} from ${key_src_dir}, installing to target..."
install -D -m 0600 "/root/host-keys/${choice}_ssh_host_ed25519_key" /mnt/etc/ssh/ssh_host_ed25519_key
install -D -m 0644 "/root/host-keys/${choice}_ssh_host_ed25519_key.pub" /mnt/etc/ssh/ssh_host_ed25519_key.pub
else
echo "WARNING: no SSH host key found for ${choice} (checked /etc/host-keys and /root/host-keys)"
echo "sops-nix secrets (including the login password) will NOT decrypt on first boot."
echo "Run scripts/secrets/prepare-host-key.sh for host ${choice} on your admin workstation first,"
echo "then either rebuild this image with NIXOS_HOST_KEYS_DIR set, scp the result to"
echo "/root/host-keys/ on this machine, or point at it when prompted above."
read -rp "Continue without a pre-seeded key anyway? (y/N): " skip_key
if [[ ! "$skip_key" =~ ^[Yy]$ ]]; then
echo "Aborted."
exit 1
fi
fi
fi
@@ -140,5 +165,39 @@ nixos-install \
# /etc/ssh/ssh_host_ed25519_key. Nothing NixOS-managed ever cleans this
# up on its own since it was written imperatively, not declaratively.
rm -rf /root/host-keys
# disko's --mode ...,mount left any ZFS root pool imported (that's what
# let nixos-install write into /mnt). If we reboot with it still
# imported, it isn't just "not exported" -- it's stamped with *this*
# live installer environment's hostid, which almost never matches the
# target's own networking.hostId (see hosts/*/host.nix; the installer
# itself sets none). modules/services/zfs/enable-service.nix and
# modules/common/configuration.nix both set boot.zfs.forceImportRoot =
# false deliberately (the safe option per that setting's own docs), so
# the freshly-installed system's first real boot sees a pool "in use by
# another system" and refuses to import it without -f -- which is what
# makes boot stall waiting on the ZFS import. Exporting here (a no-op
# if the chosen host has no ZFS root, e.g. proxmox-*/linode-*) clears
# that in-use state so the next import, from any hostid, succeeds.
#
# nixos-install bind-mounts /dev, /proc, /sys (and usually /run) into
# /mnt to run the target's activation script (switch-to-configuration
# boot) in a chroot, and doesn't unmount them again once it's done.
# Left in place, those nested mounts make ZFS refuse to unmount its own
# root dataset at /mnt -- confirmed live: zpool export failed with
# "cannot unmount '/mnt': pool or dataset busy", and because of this
# script's `set -e`, that killed the script before it ever reached
# reboot, silently defeating the whole point of exporting first.
for chroot_mount in dev proc sys run; do
if mountpoint -q "/mnt/${chroot_mount}"; then
umount -R "/mnt/${chroot_mount}"
fi
done
if [[ -n "$(zpool list -H -o name 2>/dev/null)" ]]; then
echo "Exporting ZFS pool(s) before reboot..."
zpool export -a
fi
sleep 10
reboot