{ config, lib, pkgs, ... }:
with lib;
let
nixos-container = pkgs.substituteAll {
name = "nixos-container";
dir = "bin";
isExecutable = true;
src = ./nixos-container.pl;
perl = "${pkgs.perl}/bin/perl -I${pkgs.perlPackages.FileSlurp}/lib/perl5/site_perl";
inherit (pkgs) utillinux;
};
# The container's init script, a small wrapper around the regular
# NixOS stage-2 init script.
containerInit = pkgs.writeScript "container-init"
''
#! ${pkgs.stdenv.shell} -e
# Initialise the container side of the veth pair.
if [ "$PRIVATE_NETWORK" = 1 ]; then
ip link set host0 name eth0
ip link set dev eth0 up
if [ -n "$HOST_ADDRESS" ]; then
ip route add $HOST_ADDRESS dev eth0
ip route add default via $HOST_ADDRESS
fi
if [ -n "$LOCAL_ADDRESS" ]; then
ip addr add $LOCAL_ADDRESS dev eth0
fi
fi
# Start the regular stage 1 script, passing the bind-mounted
# notification socket from the host to allow the container
# systemd to signal readiness to the host systemd.
NOTIFY_SOCKET=/var/lib/private/host-notify exec "$1"
'';
system = config.nixpkgs.system;
in
{
options = {
boot.isContainer = mkOption {
type = types.bool;
default = false;
description = ''
Whether this NixOS machine is a lightweight container running
in another NixOS system.
'';
};
containers = mkOption {
type = types.attrsOf (types.submodule (
{ config, options, name, ... }:
{
options = {
config = mkOption {
description = ''
A specification of the desired configuration of this
container, as a NixOS module.
'';
};
path = mkOption {
type = types.path;
example = "/nix/var/nix/profiles/containers/webserver";
description = ''
As an alternative to specifying
, you can specify the path to
the evaluated NixOS system configuration, typically a
symlink to a system profile.
'';
};
privateNetwork = mkOption {
type = types.bool;
default = false;
description = ''
Whether to give the container its own private virtual
Ethernet interface. The interface is called
eth0, and is hooked up to the interface
ve-container-name
on the host. If this option is not set, then the
container shares the network interfaces of the host,
and can bind to any port on any interface.
'';
};
hostAddress = mkOption {
type = types.nullOr types.string;
default = null;
example = "10.231.136.1";
description = ''
The IPv4 address assigned to the host interface.
'';
};
localAddress = mkOption {
type = types.nullOr types.string;
default = null;
example = "10.231.136.2";
description = ''
The IPv4 address assigned to eth0
in the container.
'';
};
};
config = mkMerge
[ (mkIf options.config.isDefined {
path = (import ../../lib/eval-config.nix {
inherit system;
modules =
let extraConfig =
{ boot.isContainer = true;
networking.hostName = mkDefault name;
networking.useDHCP = false;
};
in [ extraConfig config.config ];
prefix = [ "containers" name ];
}).config.system.build.toplevel;
})
];
}));
default = {};
example = literalExample
''
{ webserver =
{ path = "/nix/var/nix/profiles/webserver";
};
database =
{ config =
{ config, pkgs, ... }:
{ services.postgresql.enable = true;
services.postgresql.package = pkgs.postgresql92;
};
};
}
'';
description = ''
A set of NixOS system configurations to be run as lightweight
containers. Each container appears as a service
container-name
on the host system, allowing it to be started and stopped via
systemctl .
'';
};
};
config = mkIf (!config.boot.isContainer) {
systemd.services."container@" =
{ description = "Container '%i'";
unitConfig.RequiresMountsFor = [ "/var/lib/containers/%i" ];
path = [ pkgs.iproute ];
environment.INSTANCE = "%i";
environment.root = "/var/lib/containers/%i";
preStart =
''
# Clean up existing machined registration and interfaces.
machinectl terminate "$INSTANCE" 2> /dev/null || true
if [ "$PRIVATE_NETWORK" = 1 ]; then
ip link del dev "ve-$INSTANCE" 2> /dev/null || true
fi
if [ "$PRIVATE_NETWORK" = 1 ]; then
ip link del dev "ve-$INSTANCE" 2> /dev/null || true
fi
'';
script =
''
mkdir -p -m 0755 "$root/etc" "$root/var/lib"
mkdir -p -m 0700 "$root/var/lib/private"
if ! [ -e "$root/etc/os-release" ]; then
touch "$root/etc/os-release"
fi
mkdir -p -m 0755 \
"/nix/var/nix/profiles/per-container/$INSTANCE" \
"/nix/var/nix/gcroots/per-container/$INSTANCE"
cp -f /etc/resolv.conf "$root/etc/resolv.conf"
if [ "$PRIVATE_NETWORK" = 1 ]; then
extraFlags+=" --network-veth"
fi
for iface in $MACVLANS; do
extraFlags+=" --network-macvlan=$iface"
done
# If the host is 64-bit and the container is 32-bit, add a
# --personality flag.
${optionalString (config.nixpkgs.system == "x86_64-linux") ''
if [ "$(< ''${SYSTEM_PATH:-/nix/var/nix/profiles/per-container/$INSTANCE/system}/system)" = i686-linux ]; then
extraFlags+=" --personality=x86"
fi
''}
# Run systemd-nspawn without startup notification (we'll
# wait for the container systemd to signal readiness).
EXIT_ON_REBOOT=1 NOTIFY_SOCKET= \
exec ${config.systemd.package}/bin/systemd-nspawn \
--keep-unit \
-M "$INSTANCE" -D "$root" $extraFlags \
--bind-ro=/nix/store \
--bind-ro=/nix/var/nix/db \
--bind-ro=/nix/var/nix/daemon-socket \
--bind=/run/systemd/notify:/var/lib/private/host-notify \
--bind="/nix/var/nix/profiles/per-container/$INSTANCE:/nix/var/nix/profiles" \
--bind="/nix/var/nix/gcroots/per-container/$INSTANCE:/nix/var/nix/gcroots" \
--setenv PRIVATE_NETWORK="$PRIVATE_NETWORK" \
--setenv HOST_ADDRESS="$HOST_ADDRESS" \
--setenv LOCAL_ADDRESS="$LOCAL_ADDRESS" \
--setenv PATH="$PATH" \
${containerInit} "''${SYSTEM_PATH:-/nix/var/nix/profiles/system}/init"
'';
postStart =
''
if [ "$PRIVATE_NETWORK" = 1 ]; then
ifaceHost=ve-$INSTANCE
ip link set dev $ifaceHost up
if [ -n "$HOST_ADDRESS" ]; then
ip addr add $HOST_ADDRESS dev $ifaceHost
fi
if [ -n "$LOCAL_ADDRESS" ]; then
ip route add $LOCAL_ADDRESS dev $ifaceHost
fi
fi
'';
preStop =
''
machinectl poweroff "$INSTANCE" || true
'';
restartIfChanged = false;
#reloadIfChanged = true; # FIXME
serviceConfig = {
ExecReload = pkgs.writeScript "reload-container"
''
#! ${pkgs.stdenv.shell} -e
${nixos-container}/bin/nixos-container run "$INSTANCE" -- \
bash --login -c "/nix/var/nix/profiles/system/bin/switch-to-configuration test"
'';
SyslogIdentifier = "container %i";
EnvironmentFile = "-/etc/containers/%i.conf";
Type = "notify";
NotifyAccess = "all";
# Note that on reboot, systemd-nspawn returns 10, so this
# unit will be restarted. On poweroff, it returns 0, so the
# unit won't be restarted.
Restart = "on-failure";
# Hack: we don't want to kill systemd-nspawn, since we call
# "machinectl poweroff" in preStop to shut down the
# container cleanly. But systemd requires sending a signal
# (at least if we want remaining processes to be killed
# after the timeout). So send an ignored signal.
KillMode = "mixed";
KillSignal = "WINCH";
};
};
# Generate a configuration file in /etc/containers for each
# container so that container@.target can get the container
# configuration.
environment.etc = mapAttrs' (name: cfg: nameValuePair "containers/${name}.conf"
{ text =
''
SYSTEM_PATH=${cfg.path}
${optionalString cfg.privateNetwork ''
PRIVATE_NETWORK=1
${optionalString (cfg.hostAddress != null) ''
HOST_ADDRESS=${cfg.hostAddress}
''}
${optionalString (cfg.localAddress != null) ''
LOCAL_ADDRESS=${cfg.localAddress}
''}
''}
'';
}) config.containers;
# FIXME: auto-start containers.
# Generate /etc/hosts entries for the containers.
networking.extraHosts = concatStrings (mapAttrsToList (name: cfg: optionalString (cfg.localAddress != null)
''
${cfg.localAddress} ${name}.containers
'') config.containers);
networking.dhcpcd.denyInterfaces = [ "ve-*" ];
environment.systemPackages = [ nixos-container ];
# Start containers at boot time.
systemd.services.all-containers =
{ description = "All Containers";
wantedBy = [ "multi-user.target" ];
unitConfig.ConditionDirectoryNotEmpty = "/etc/containers";
serviceConfig.Type = "oneshot";
script =
''
res=0
shopt -s nullglob
for i in /etc/containers/*.conf; do
AUTO_START=
source "$i"
if [ "$AUTO_START" = 1 ]; then
systemctl start "container@$(basename "$i" .conf).service" || res=1
fi
done
exit $res
''; # */
};
};
}