Files
Samuel Silva e6ed17b918 nixos/github-runners: GitHub App authentication and multi-org runners
Add three options to `services.github-runners.<name>`:

- `githubApp`: authenticate via a GitHub App installation instead of a
  `tokenFile`. The service derives a short-lived installation token from the
  App's private key on each start, registers the runner, and de-registers it
  on stop. The PEM key is a deployed secret, never copied into the store.

- `count`: fan a single entry out into N identical runner services.
  `github-runner-<name>` is unchanged for `count == 1`, suffixed `-<n>` for
  `count > 1`.

- `orgs`: serve several organisations (or repositories) from one entry,
  fanning out into `github-runner-<name>-<org>-<n>`. The entry-level auth is
  shared across orgs and only the App `login` changes per org; the per-org
  `count` defaults to the entry-level `count`.

This brings the multi-org consumption shape of the soon-to-be-removed srvos
runner role (nix-community/srvos#836, nix-community/srvos#837) into nixpkgs.
An entry without `orgs` and with the default `count = 1` is unchanged, so
existing runners do not re-register.
2026-06-19 15:21:54 -03:00

491 lines
20 KiB
Nix

{
config,
lib,
pkgs,
...
}:
{
config.assertions = lib.flatten (
lib.flip lib.mapAttrsToList config.services.github-runners (
name: cfg:
map (lib.mkIf cfg.enable) [
{
assertion = (cfg.tokenFile == null) != (cfg.githubApp == null);
message = "`services.github-runners.${name}`: Exactly one of `tokenFile` and `githubApp` must be set";
}
{
assertion = cfg.orgs != { } || cfg.url != null;
message = "`services.github-runners.${name}`: `url` must be set unless `orgs` is used";
}
{
assertion = cfg.orgs != { } || cfg.githubApp == null || cfg.githubApp.login != null;
message = "`services.github-runners.${name}`: `githubApp.login` must be set unless `orgs` is used (the login is then derived per org)";
}
{
assertion = cfg.orgs != { } || cfg.count == 1 || cfg.name != null;
message = "`services.github-runners.${name}`: `name` must not be null when `count > 1` (each replica needs a distinct registration name)";
}
{
assertion = !cfg.noDefaultLabels || (cfg.extraLabels != [ ]);
message = "`services.github-runners.${name}`: The `extraLabels` option is mandatory if `noDefaultLabels` is set";
}
{
assertion = cfg.group == null || cfg.user != null;
message = ''`services.github-runners.${name}`: Setting `group` while leaving `user` unset runs the service as `root`. If this is really what you want, set `user = "root"` explicitly'';
}
]
)
);
config.systemd.services =
let
enabledRunners = lib.filterAttrs (_: cfg: cfg.enable) config.services.github-runners;
runnerInstances = lib.concatMapAttrs (
name: cfg:
if cfg.orgs == { } then
# Single org/repo (entry-level `url`), fanned out by `count`. For
# `count == 1` the service keeps the bare `github-runner-<name>` name
# for backwards compatibility; `count > 1` suffixes `-<n>`.
let
suffixes = if cfg.count == 1 then [ "" ] else map (n: "-${toString n}") (lib.range 1 cfg.count);
in
lib.listToAttrs (
map (
suffix:
lib.nameValuePair "${name}${suffix}" (
cfg // { name = if cfg.name == null then null else "${cfg.name}${suffix}"; }
)
) suffixes
)
else
lib.listToAttrs (
lib.concatLists (
lib.flip lib.mapAttrsToList cfg.orgs (
orgName: org:
map (
n:
let
key = "${name}-${orgName}-${toString n}";
in
lib.nameValuePair key (
cfg
// {
url = org.url;
name = key;
extraLabels = cfg.extraLabels ++ org.extraLabels;
githubApp = if cfg.githubApp == null then null else cfg.githubApp // { login = org.login; };
}
)
) (lib.range 1 org.count)
)
)
)
) enabledRunners;
in
(lib.flip lib.mapAttrs' runnerInstances (
name: cfg:
let
svcName = "github-runner-${name}";
systemdDir = "github-runner/${name}";
# %t: Runtime directory root (usually /run); see systemd.unit(5)
runtimeDir = "%t/${systemdDir}";
# %S: State directory root (usually /var/lib); see systemd.unit(5)
stateDir = "%S/${systemdDir}";
# %L: Log directory root (usually /var/log); see systemd.unit(5)
logsDir = "%L/${systemdDir}";
# Name of file stored in service state directory
currentConfigTokenFilename = ".current-token";
workDir = if cfg.workDir == null then runtimeDir else cfg.workDir;
newConfigTokenPath = "$STATE_DIRECTORY/.new-token";
currentConfigTokenPath = "$STATE_DIRECTORY/${currentConfigTokenFilename}";
# Wrapper script which expects the full path of the state, working and logs
# directory as arguments. Overrides the respective systemd variables to provide
# unambiguous directory names. This becomes relevant, for example, if the
# caller overrides any of the StateDirectory=, RuntimeDirectory= or LogDirectory=
# to contain more than one directory. This causes systemd to set the respective
# environment variables with the path of all of the given directories, separated
# by a colon.
writeScript =
scriptName: lines:
pkgs.writeShellScript "${svcName}-${scriptName}.sh" ''
set -euo pipefail
STATE_DIRECTORY="$1"
WORK_DIRECTORY="$2"
LOGS_DIRECTORY="$3"
${lines}
'';
ghUrlPath = lib.removePrefix "https://github.com/" cfg.url;
ghUrlSegments = lib.filter (s: s != "") (lib.splitString "/" ghUrlPath);
runnerScope = if lib.length ghUrlSegments >= 2 then "repo" else "org";
appHelper =
helperName: scriptFile:
pkgs.writeShellApplication {
name = helperName;
runtimeInputs = with pkgs; [
jq
curl
openssl
coreutils
];
excludeShellChecks = [
"SC2154"
"SC2116"
];
text = builtins.readFile scriptFile;
};
fetchAccessToken = appHelper "github-app-access-token" ./app-token.sh;
fetchRegistrationToken = appHelper "github-runner-registration-token" ./registration-token.sh;
removeRunner = appHelper "github-runner-remove" ./remove-runner.sh;
appEnv = lib.optionalString (cfg.githubApp != null) ''
export APP_ID=${toString cfg.githubApp.id}
export APP_LOGIN=${lib.escapeShellArg cfg.githubApp.login}
APP_PRIVATE_KEY="$(cat ${lib.escapeShellArg cfg.githubApp.privateKeyFile})"
export APP_PRIVATE_KEY
export RUNNER_SCOPE=${runnerScope}
export ORG_NAME=${lib.escapeShellArg cfg.githubApp.login}
export REPO_URL=${lib.escapeShellArg cfg.url}
${
if cfg.name != null then
"export RUNNER_NAME=${lib.escapeShellArg cfg.name}"
else
''export RUNNER_NAME="$(uname -n)"''
}
'';
in
lib.nameValuePair svcName {
description = "GitHub Actions runner";
wantedBy = [ "multi-user.target" ];
wants = [ "network-online.target" ];
after = [
"network.target"
"network-online.target"
];
environment = {
HOME = workDir;
RUNNER_ROOT = stateDir;
}
// cfg.extraEnvironment;
path =
(with pkgs; [
bashInteractive
coreutils
git
gnutar
gzip
])
++ [
config.nix.package
]
++ cfg.extraPackages;
serviceConfig = lib.mkMerge [
{
ExecStart = "${cfg.package}/bin/Runner.Listener run --startuptype service";
# Does the following, sequentially:
# - If the module configuration or the token has changed, purge the state directory,
# and create the current and the new token file with the contents of the configured
# token. While both files have the same content, only the later is accessible by
# the service user.
# - Configure the runner using the new token file. When finished, delete it.
# - Set up the directory structure by creating the necessary symlinks.
ExecStartPre =
let
runnerRegistrationConfig = lib.getAttrs [
"ephemeral"
"extraLabels"
"name"
"noDefaultLabels"
"runnerGroup"
"tokenFile"
"url"
"workDir"
] cfg;
newConfigPath = builtins.toFile "${svcName}-config.json" (builtins.toJSON runnerRegistrationConfig);
currentConfigPath = "$STATE_DIRECTORY/.nixos-current-config.json";
runnerCredFiles = [
".credentials"
".credentials_rsaparams"
".runner"
];
unconfigureRunner = writeScript "unconfigure" ''
copy_tokens() {
# Copy the configured token file to the state dir and allow the service user to read the file
install --mode=666 ${lib.escapeShellArg cfg.tokenFile} "${newConfigTokenPath}"
# Also copy current file to allow for a diff on the next start
install --mode=600 ${lib.escapeShellArg cfg.tokenFile} "${currentConfigTokenPath}"
}
clean_state() {
find "$STATE_DIRECTORY/" -mindepth 1 -delete
copy_tokens
}
diff_config() {
changed=0
# Check for module config changes
[[ -f "${currentConfigPath}" ]] \
&& ${pkgs.diffutils}/bin/diff -q '${newConfigPath}' "${currentConfigPath}" >/dev/null 2>&1 \
|| changed=1
# Also check the content of the token file
[[ -f "${currentConfigTokenPath}" ]] \
&& ${pkgs.diffutils}/bin/diff -q "${currentConfigTokenPath}" ${lib.escapeShellArg cfg.tokenFile} >/dev/null 2>&1 \
|| changed=1
# If the config has changed, remove old state and copy tokens
if [[ "$changed" -eq 1 ]]; then
echo "Config has changed, removing old runner state."
echo "The old runner will still appear in the GitHub Actions UI." \
"You have to remove it manually."
clean_state
fi
}
if [[ "${lib.optionalString cfg.ephemeral "1"}" ]]; then
# In ephemeral mode, we always want to start with a clean state
clean_state
elif [[ "$(ls -A "$STATE_DIRECTORY")" ]]; then
# There are state files from a previous run; diff them to decide if we need a new registration
diff_config
else
# The state directory is entirely empty which indicates a first start
copy_tokens
fi
# Always clean workDir
find -H "$WORK_DIRECTORY" -mindepth 1 -delete
'';
unconfigureRunnerGitHubApp = writeScript "unconfigure-github-app" ''
${appEnv}
ACCESS_TOKEN="$(${lib.getExe' fetchAccessToken "github-app-access-token"})"
export ACCESS_TOKEN
${lib.getExe' removeRunner "github-runner-remove"} || true
find "$STATE_DIRECTORY/" -mindepth 1 -delete
umask 000
${lib.getExe' fetchRegistrationToken "github-runner-registration-token"} \
| ${pkgs.jq}/bin/jq -r '.token' > "${newConfigTokenPath}"
install --mode=600 "${newConfigTokenPath}" "${currentConfigTokenPath}"
# Always clean workDir
find -H "$WORK_DIRECTORY" -mindepth 1 -delete
'';
configureRunner =
writeScript "configure" # bash
''
if [[ -e "${newConfigTokenPath}" ]]; then
echo "Configuring GitHub Actions Runner"
# shellcheck disable=SC2054 # don't complain about commas in --labels
args=(
--unattended
--disableupdate
--work "$WORK_DIRECTORY"
--url ${lib.escapeShellArg cfg.url}
--labels ${lib.escapeShellArg (lib.concatStringsSep "," cfg.extraLabels)}
${lib.optionalString (cfg.name != null) "--name ${lib.escapeShellArg cfg.name}"}
${lib.optionalString cfg.replace "--replace"}
${lib.optionalString (
cfg.runnerGroup != null
) "--runnergroup ${lib.escapeShellArg cfg.runnerGroup}"}
${lib.optionalString cfg.ephemeral "--ephemeral"}
${lib.optionalString cfg.noDefaultLabels "--no-default-labels"}
)
token=$(<"${newConfigTokenPath}")
${
if cfg.githubApp != null then
''
args+=(--token "$token")
''
else
''
case ${cfg.tokenType} in
access)
args+=(--pat "$token")
;;
registration)
args+=(--token "$token")
;;
auto)
# If the token file contains a PAT (i.e., it starts with "ghp_" or "github_pat_"),
# we have to use the --pat option, if it is not a PAT, we assume it contains a
# registration token and use the --token option
if [[ "$token" =~ ^gh[a-z]+_* ]] || [[ "$token" =~ ^github_pat_* ]]; then
args+=(--pat "$token")
else
args+=(--token "$token")
fi
;;
esac
''
}
${cfg.package}/bin/Runner.Listener configure "''${args[@]}"
# Move the automatically created _diag dir to the logs dir
mkdir -p "$STATE_DIRECTORY/_diag"
cp -r "$STATE_DIRECTORY/_diag/." "$LOGS_DIRECTORY/"
rm -rf "$STATE_DIRECTORY/_diag/"
# Cleanup token from config
rm "${newConfigTokenPath}"
# Symlink to new config
ln -s '${newConfigPath}' "${currentConfigPath}"
fi
'';
setupWorkDir = writeScript "setup-work-dirs" ''
# Link _diag dir
ln -s "$LOGS_DIRECTORY" "$WORK_DIRECTORY/_diag"
# Link the runner credentials to the work dir
ln -s "$STATE_DIRECTORY"/{${lib.concatStringsSep "," runnerCredFiles}} "$WORK_DIRECTORY/"
'';
in
map
(
x:
"${x} ${
lib.escapeShellArgs [
stateDir
workDir
logsDir
]
}"
)
[
# runs as root
"+${if cfg.githubApp != null then unconfigureRunnerGitHubApp else unconfigureRunner}"
configureRunner
setupWorkDir
];
ExecStopPost = lib.optionals (cfg.githubApp != null) (
let
unregister = writeScript "unregister-github-app" ''
${appEnv}
ACCESS_TOKEN="$(${lib.getExe' fetchAccessToken "github-app-access-token"})"
export ACCESS_TOKEN
${lib.getExe' removeRunner "github-runner-remove"} || true
'';
in
map (
x:
"${x} ${
lib.escapeShellArgs [
stateDir
workDir
logsDir
]
}"
) [ "-+${unregister}" ] # runs as root
);
# If running in ephemeral mode, restart the service on-exit (i.e., successful de-registration of the runner)
# to trigger a fresh registration.
Restart = if cfg.ephemeral then "on-success" else "no";
# If the runner exits with `ReturnCode.RetryableError = 2`, always restart the service:
# https://github.com/actions/runner/blob/40ed7f8/src/Runner.Common/Constants.cs#L146
RestartForceExitStatus = [ 2 ];
# Contains _diag
LogsDirectory = [ systemdDir ];
# Default RUNNER_ROOT which contains ephemeral Runner data
RuntimeDirectory = [ systemdDir ];
# Home of persistent runner data, e.g., credentials
StateDirectory = [ systemdDir ];
StateDirectoryMode = "0700";
WorkingDirectory = workDir;
InaccessiblePaths = [
# Token file in the state directory
"${stateDir}/${currentConfigTokenFilename}"
]
# Token file path given in the configuration, if visible to the service
++ lib.optional (cfg.tokenFile != null) "-${cfg.tokenFile}"
++ lib.optional (cfg.githubApp != null) "-${cfg.githubApp.privateKeyFile}";
KillSignal = "SIGINT";
# Hardening (may overlap with DynamicUser=)
# The following options are only for optimizing:
# systemd-analyze security github-runner
AmbientCapabilities = lib.mkBefore [ "" ];
CapabilityBoundingSet = lib.mkBefore [ "" ];
# ProtectClock= adds DeviceAllow=char-rtc r
DeviceAllow = lib.mkBefore [ "" ];
NoNewPrivileges = lib.mkDefault true;
PrivateDevices = lib.mkDefault true;
PrivateMounts = lib.mkDefault true;
PrivateTmp = lib.mkDefault true;
PrivateUsers = lib.mkDefault true;
ProtectClock = lib.mkDefault true;
ProtectControlGroups = lib.mkDefault true;
ProtectHome = lib.mkDefault true;
ProtectHostname = lib.mkDefault true;
ProtectKernelLogs = lib.mkDefault true;
ProtectKernelModules = lib.mkDefault true;
ProtectKernelTunables = lib.mkDefault true;
ProtectSystem = lib.mkDefault "strict";
RemoveIPC = lib.mkDefault true;
RestrictNamespaces = lib.mkDefault true;
RestrictRealtime = lib.mkDefault true;
RestrictSUIDSGID = lib.mkDefault true;
UMask = lib.mkDefault "0066";
ProtectProc = lib.mkDefault "invisible";
SystemCallFilter = lib.mkBefore [
"~@clock"
"~@cpu-emulation"
"~@module"
"~@mount"
"~@obsolete"
"~@raw-io"
"~@reboot"
"~capset"
"~setdomainname"
"~sethostname"
];
RestrictAddressFamilies = lib.mkBefore [
"AF_INET"
"AF_INET6"
"AF_UNIX"
"AF_NETLINK"
];
BindPaths = lib.optionals (cfg.workDir != null) [ cfg.workDir ];
# Needs network access
PrivateNetwork = lib.mkDefault false;
# Cannot be true due to Node
MemoryDenyWriteExecute = lib.mkDefault false;
# The more restrictive "pid" option makes `nix` commands in CI emit
# "GC Warning: Couldn't read /proc/stat"
# You may want to set this to "pid" if not using `nix` commands
ProcSubset = lib.mkDefault "all";
# Coverage programs for compiled code such as `cargo-tarpaulin` disable
# ASLR (address space layout randomization) which requires the
# `personality` syscall
# You may want to set this to `true` if not using coverage tooling on
# compiled code
LockPersonality = lib.mkDefault false;
DynamicUser = lib.mkDefault true;
}
(lib.mkIf (cfg.user != null) {
DynamicUser = false;
User = cfg.user;
})
(lib.mkIf (cfg.group != null) {
DynamicUser = false;
Group = cfg.group;
})
cfg.serviceOverrides
];
}
));
}