deploy: harden kexec-local, key vault items by config, add VM test

kexec-local could never actually jump. nixos-images' kexec-run.sh ends with
`nohup sh -c "sleep 6 && $SCRIPT_DIR/kexec -e" &` and returns immediately, so
the EXIT trap's `rm -rf "$stage"` deleted the kexec binary out from under the
sleeping shell. The box stayed on the old kernel and it looked like a slow boot.
Clear the trap before jumping, verify /sys/kernel/kexec_loaded, then sleep past
the timer.

Preflight everything before the point of no return, since this jumps the machine
you are typing at: CONFIG_KEXEC, kernel lockdown, exec-capable staging dir, free
space, RAM vs image size, and that the tarball holds all five expected files.
Stage on /var/tmp rather than /tmp because kexec-run.sh appends to initrd in
place and execs from that directory. sync before jumping (kexec -e skips
unmount). Confirmation prompt naming the host, since run in the wrong terminal
this kexecs the laptop; --yes skips it.

Drop the ssh-keygen -R added to the remote kexec path: kexec-run.sh copies
/etc/ssh/ssh_host_* into the appended initrd and restore-remote-access.nix
installs them back, so the host key survives the jump.

Proton Pass items are now keyed by <config> instead of <host>, since the address
is incidental and the config name is stable. kexec therefore takes <config>
<host>. Resolve titles among --filter-state active items first: a trashed item
with the same title shadowed the active one and returned an empty password,
which is indistinguishable from "no entry" and silently fell back to prompting
(hit on darman@neptun).

Other fixes: replace `ls glob | head -1` (returns empty with exit 0 on no match)
with a helper that dies; guard against untracked hosts/<config> since flakes
ignore untracked files; feed the sudo password more than once under setsid;
handle empty arrays under set -u; tolerate empty FSTYPE in the SD-card root
partition lookup; preflight zstdcat/dd/lsblk before the destructive dd; list
image and flash in the usage strings.

Add checks.x86_64-linux.kexec-local, a VM test driving the real script. It is
the only way to exercise kexec-local, which cannot be rehearsed on hardware. It
asserts the box left the old kernel, returned as nixos-installer, lost its old
/run, and kept its ssh host key. HOMELAB_KEXEC_TARBALL lets it reuse a prebuilt
installer instead of building ~500MB inside the guest.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-07-22 23:47:20 +02:00
co-authored by Claude Opus 4.8
parent 7bcea764f6
commit 80c2b4fc7b
4 changed files with 521 additions and 60 deletions
+164 -1
View File
@@ -20,6 +20,10 @@
url = "github:nix-community/nixos-images";
inputs.nixpkgs.follows = "nixpkgs";
};
home-manager = {
url = "github:nix-community/home-manager/release-26.05";
inputs.nixpkgs.follows = "nixpkgs";
};
mediamanager-nix = {
url = "github:strangeglyph/mediamanager-nix";
inputs.nixpkgs.follows = "nixpkgs";
@@ -30,7 +34,7 @@
authentik-nix.url = "github:nix-community/authentik-nix";
};
outputs = { self, nixpkgs, disko, sops-nix, nixos-images, mediamanager-nix, authentik-nix, ... }@inputs:
outputs = { self, nixpkgs, disko, sops-nix, nixos-images, home-manager, mediamanager-nix, authentik-nix, ... }@inputs:
let
system = "x86_64-linux";
in
@@ -59,6 +63,19 @@
];
};
# terra — Ryzen 9 5900X desktop (MSI MS-7A32). Replaces CachyOS on the
# OS SSD; Hyprland desktop + tailnet node. See hosts/terra/*.
terra = nixpkgs.lib.nixosSystem {
inherit system;
specialArgs = { inherit inputs; };
modules = [
disko.nixosModules.disko
sops-nix.nixosModules.sops
home-manager.nixosModules.home-manager
./hosts/terra/configuration.nix
];
};
# mercury — Raspberry Pi 3B+ (aarch64), DNS/DHCP. Boots from an SD image:
# nix build .#nixosConfigurations.mercury.config.system.build.sdImage
# (aarch64 build — needs binfmt/qemu on this x86 host, or a remote/aarch64
@@ -146,5 +163,151 @@
];
};
};
# VM test for `./scripts/deploy kexec-local`. Run:
# nix build .#checks.x86_64-linux.kexec-local -L
#
# Worth having because kexec-local is the one command that cannot be
# rehearsed on real hardware: it jumps the machine you are typing at, and
# a failure looks exactly like a slow boot. It regression-tests the
# subtle one — kexec-run.sh backgrounds `sleep 6 && kexec -e` and returns,
# so anything that cleans up the staging dir on exit deletes the binary
# that performs the jump and the box silently never leaves the old kernel.
#
# After the jump the test driver's backdoor is gone with the old kernel,
# so the installer is driven over a forwarded ssh port instead (the same
# approach nixos-images uses in its own kexec test).
checks.${system}.kexec-local =
let
pkgs = nixpkgs.legacyPackages.${system};
tarball = self.nixosConfigurations.kexec.config.system.build.kexecInstallerTarball;
sshKey = nixos-images + "/nix/kexec-installer/ssh-keys/id_ed25519";
in
pkgs.testers.runNixOSTest {
name = "deploy-kexec-local";
nodes.machine = { modulesPath, ... }: {
imports = [ (modulesPath + "/profiles/minimal.nix") ];
virtualisation.vlans = [ ];
# kexec-local refuses to run if RAM < 3x the installer image, and
# the staging dir needs ~3x the tarball on /var/tmp.
virtualisation.memorySize = 4 * 1024;
virtualisation.diskSize = 12 * 1024;
virtualisation.forwardPorts = [{ host.port = 2222; guest.port = 22; }];
services.openssh.enable = true;
users.users.root.openssh.authorizedKeys.keyFiles = [ "${sshKey}.pub" ];
# Everything the script shells out to, minus nix — the test uses the
# HOMELAB_KEXEC_* hook so no build happens inside the VM.
environment.systemPackages = with pkgs; [
bash gnutar coreutils findutils util-linux cpio gzip
];
system.extraDependencies = [ tarball pkgs.cpio pkgs.gzip ];
environment.etc."deploy".source = ./scripts/deploy;
};
testScript = /* python */ ''
import os, shutil, subprocess, tempfile, time
start_all()
machine.wait_for_unit("sshd.service")
# ssh refuses a private key that is group/world readable, and nix
# store paths are 0444 copy it out and tighten the mode.
keydir = tempfile.mkdtemp()
key = os.path.join(keydir, "id_ed25519")
shutil.copyfile("${sshKey}", key)
os.chmod(key, 0o600)
def ssh(cmd, check=True, stdout=None):
return subprocess.run(
[ "${pkgs.openssh}/bin/ssh",
"-o", "StrictHostKeyChecking=no",
"-o", "UserKnownHostsFile=/dev/null",
"-o", "ConnectTimeout=1",
"-i", key,
"-p", "2222", "root@127.0.0.1", "--" ] + cmd,
text=True, check=check, stdout=stdout)
machine.succeed("install -Dm755 /etc/deploy /root/deploy")
# systemd-run starts units with a bare PATH that lacks
# /run/current-system/sw/bin, so `#!/usr/bin/env bash` cannot even
# resolve bash, let alone tar/findmnt/nohup. Set it explicitly.
env = (
" --setenv=PATH=/run/wrappers/bin:/run/current-system/sw/bin"
" --setenv=HOMELAB_KEXEC_TARBALL=${tarball}/nixos-kexec-installer-${system}.tar.gz"
" --setenv=HOMELAB_KEXEC_CPIO=${pkgs.cpio}/bin/cpio"
" --setenv=HOMELAB_KEXEC_GZIP=${pkgs.gzip}/bin/gzip"
)
# Same values for the foreground (non-systemd-run) invocation below.
envsh = (
"HOMELAB_KEXEC_TARBALL=${tarball}/nixos-kexec-installer-${system}.tar.gz"
" HOMELAB_KEXEC_CPIO=${pkgs.cpio}/bin/cpio"
" HOMELAB_KEXEC_GZIP=${pkgs.gzip}/bin/gzip"
)
# Marker on a tmpfs: it must NOT survive the jump, proving we really
# booted a new kernel rather than just restarting a service.
machine.succeed("touch /run/pre-kexec-marker")
host_key_before = machine.succeed("cat /etc/ssh/ssh_host_ed25519_key.pub").strip()
while ssh(["true"], check=False).returncode != 0:
time.sleep(1)
# Refuses without --yes when stdin is not a tty (read gets EOF).
# Must reach the confirmation prompt, so it needs the same env
# otherwise it just dies early on the nix build and proves nothing.
out = machine.fail(f"{envsh} /root/deploy kexec-local </dev/null 2>&1")
assert "using prebuilt kexec installer" in out, \
f"never reached the prompt, so the refusal proves nothing:\n{out}"
# systemd-run so the call returns immediately: the script stays
# alive ~60s on purpose, outliving kexec-run.sh's `sleep 6`.
machine.succeed(f"systemd-run --collect --unit=kexec-local{env} /root/deploy kexec-local --yes")
print("waiting for the jump...")
deadline = time.time() + 300
while ssh(["true"], check=False).returncode == 0:
# Surface a dead unit immediately instead of stalling until the
# deadline and blaming "never left the old kernel".
st = ssh(["systemctl", "is-active", "kexec-local"],
check=False, stdout=subprocess.PIPE).stdout or ""
if st.strip() in ("failed", "inactive"):
# NB: not `log` the driver already binds that name to its
# AbstractLogger and the type check rejects the shadowing.
unit_log = ssh(["journalctl", "-u", "kexec-local", "--no-pager"],
check=False, stdout=subprocess.PIPE).stdout or ""
raise AssertionError(
f"kexec-local.service ended ({st.strip()}) without jumping:\n{unit_log}")
assert time.time() < deadline, "machine never left the old kernel"
time.sleep(1)
print("waiting for the installer...")
deadline = time.time() + 300
while ssh(["true"], check=False).returncode != 0:
assert time.time() < deadline, "installer never came up"
time.sleep(1)
# It really is the RAM installer, not the old system.
host = ssh(["hostname"], stdout=subprocess.PIPE).stdout.strip()
assert host == "nixos-installer", f"hostname is {host}, not nixos-installer"
assert ssh(["ls", "/run/pre-kexec-marker"], check=False).returncode != 0, \
"old /run survived this was not a fresh kernel"
# The host key is carried across (kexec-run.sh copies /etc/ssh into
# the appended initrd), which is why `kexec` does no ssh-keygen -R.
host_key_after = ssh(
["cat", "/etc/ssh/ssh_host_ed25519_key.pub"], stdout=subprocess.PIPE
).stdout.strip()
assert host_key_before == host_key_after, \
f"host key changed: {host_key_before} != {host_key_after}"
machine.crash()
'';
};
};
}