rewrite sops backend for secret generation and add tests

pkgs/clan-cli/clan_cli/machines/facts.py

@@ -0,0 +1,9 @@
+from .folders import machine_folder
+
+
+def machine_has_fact(machine: str, fact: str) -> bool:
+    return (machine_folder(machine) / "facts" / fact).exists()
+
+
+def machine_get_fact(machine: str, fact: str) -> str:
+    return (machine_folder(machine) / "facts" / fact).read_text()

pkgs/clan-cli/clan_cli/nix.py

@@ -1,5 +1,8 @@
+import json
 import os
+import subprocess
 import tempfile
+from typing import Any
 
 from .dirs import nixpkgs_flake, nixpkgs_source, unfree_nixpkgs
 
@@ -25,6 +28,16 @@ def nix_build(
     )
 
 
+def nix_config() -> dict[str, Any]:
+    cmd = nix_command(["show-config", "--json"])
+    proc = subprocess.run(cmd, check=True, text=True, stdout=subprocess.PIPE)
+    data = json.loads(proc.stdout)
+    config = {}
+    for key, value in data.items():
+        config[key] = value["value"]
+    return config
+
+
 def nix_eval(flags: list[str]) -> list[str]:
     default_flags = nix_command(
         [

pkgs/clan-cli/clan_cli/secrets/generate.py

@@ -1,31 +1,41 @@
 import argparse
 import os
+import shlex
+import shutil
 import subprocess
 import sys
+from pathlib import Path
+from tempfile import TemporaryDirectory
+from typing import Any
 
 from clan_cli.errors import ClanError
 
-from ..dirs import get_clan_flake_toplevel
-from ..nix import nix_build
+from ..dirs import get_clan_flake_toplevel, module_root
+from ..nix import nix_build, nix_config
+from .folders import sops_secrets_folder
+from .machines import add_machine, has_machine
+from .secrets import encrypt_secret, has_secret
+from .sops import generate_private_key
 
 
 def generate_secrets(machine: str) -> None:
     clan_dir = get_clan_flake_toplevel().as_posix().strip()
     env = os.environ.copy()
     env["CLAN_DIR"] = clan_dir
+    env["PYTHONPATH"] = str(module_root().parent)
+    config = nix_config()
+    system = config["system"]
 
-    proc = subprocess.run(
-        nix_build(
-            [
-                f'path:{clan_dir}#nixosConfigurations."{machine}".config.system.clan.generateSecrets'
-            ]
-        ),
-        capture_output=True,
-        text=True,
+    cmd = nix_build(
+        [
+            f'path:{clan_dir}#nixosConfigurations."{machine}".config.system.clan.{system}.generateSecrets'
+        ]
     )
+    proc = subprocess.run(cmd, stdout=subprocess.PIPE, text=True)
     if proc.returncode != 0:
-        print(proc.stderr, file=sys.stderr)
-        raise ClanError(f"failed to generate secrets:\n{proc.stderr}")
+        raise ClanError(
+            f"failed to generate secrets:\n{shlex.join(cmd)}\nexited with {proc.returncode}"
+        )
 
     secret_generator_script = proc.stdout.strip()
     print(secret_generator_script)
@@ -40,6 +50,87 @@ def generate_secrets(machine: str) -> None:
         print("successfully generated secrets")
 
 
+def generate_host_key(machine_name: str) -> None:
+    if has_machine(machine_name):
+        return
+    priv_key, pub_key = generate_private_key()
+    encrypt_secret(sops_secrets_folder() / f"{machine_name}-age.key", priv_key)
+    add_machine(machine_name, pub_key, False)
+
+
+def generate_secrets_group(
+    secret_group: str, machine_name: str, tempdir: Path, secret_options: dict[str, Any]
+) -> None:
+    clan_dir = get_clan_flake_toplevel()
+    secrets = secret_options["secrets"]
+    needs_regeneration = any(
+        not has_secret(f"{machine_name}-{secret['name']}")
+        for secret in secrets.values()
+    )
+    generator = secret_options["generator"]
+    subdir = tempdir / secret_group
+    if needs_regeneration:
+        facts_dir = subdir / "facts"
+        facts_dir.mkdir(parents=True)
+        secrets_dir = subdir / "secrets"
+        secrets_dir.mkdir(parents=True)
+
+        text = f"""\
+set -euo pipefail
+facts={shlex.quote(str(facts_dir))}
+secrets={shlex.quote(str(secrets_dir))}
+{generator}
+        """
+        try:
+            subprocess.run(["bash", "-c", text], check=True)
+        except subprocess.CalledProcessError:
+            msg = "failed to the following command:\n"
+            msg += text
+            raise ClanError(msg)
+        for secret in secrets.values():
+            secret_file = secrets_dir / secret["name"]
+            if not secret_file.is_file():
+                msg = f"did not generate a file for '{secret['name']}' when running the following command:\n"
+                msg += text
+                raise ClanError(msg)
+            encrypt_secret(
+                sops_secrets_folder() / f"{machine_name}-{secret['name']}",
+                secret_file.read_text(),
+            )
+        for fact in secret_options["facts"].values():
+            fact_file = facts_dir / fact["name"]
+            if not fact_file.is_file():
+                msg = f"did not generate a file for '{fact['name']}' when running the following command:\n"
+                msg += text
+                raise ClanError(msg)
+            fact_path = clan_dir.joinpath(fact["path"])
+            fact_path.parent.mkdir(parents=True, exist_ok=True)
+            shutil.copyfile(fact_file, fact_path)
+
+
+# this is called by the sops.nix clan core module
+def generate_secrets_from_nix(
+    machine_name: str,
+    secret_submodules: dict[str, Any],
+) -> None:
+    generate_host_key(machine_name)
+    errors = {}
+    with TemporaryDirectory() as d:
+        # if any of the secrets are missing, we regenerate all connected facts/secrets
+        for secret_group, secret_options in secret_submodules.items():
+            try:
+                generate_secrets_group(
+                    secret_group, machine_name, Path(d), secret_options
+                )
+            except ClanError as e:
+                errors[secret_group] = e
+    for secret_group, error in errors.items():
+        print(f"failed to generate secrets for {machine_name}/{secret_group}:")
+        print(error, file=sys.stderr)
+    if len(errors) > 0:
+        sys.exit(1)
+
+
 def generate_command(args: argparse.Namespace) -> None:
     generate_secrets(args.machine)
 

pkgs/clan-cli/clan_cli/secrets/upload.py

@@ -1,20 +1,25 @@
 import argparse
 import json
 import subprocess
-
-from clan_cli.errors import ClanError
+import sys
+from pathlib import Path
 
 from ..dirs import get_clan_flake_toplevel
-from ..nix import nix_build, nix_eval
+from ..errors import ClanError
+from ..nix import nix_build, nix_config, nix_eval
+from ..ssh import parse_deployment_address
+from .secrets import decrypt_secret, has_secret
 
 
 def upload_secrets(machine: str) -> None:
     clan_dir = get_clan_flake_toplevel().as_posix()
+    config = nix_config()
+    system = config["system"]
 
     proc = subprocess.run(
         nix_build(
             [
-                f'{clan_dir}#nixosConfigurations."{machine}".config.system.clan.uploadSecrets'
+                f'{clan_dir}#nixosConfigurations."{machine}".config.system.clan.{system}.uploadSecrets'
             ]
         ),
         stdout=subprocess.PIPE,
@@ -48,6 +53,34 @@ def upload_secrets(machine: str) -> None:
         print("successfully uploaded secrets")
 
 
+# this is called by the sops.nix clan core module
+def upload_age_key_from_nix(
+    machine_name: str, deployment_address: str, age_key_file: str
+) -> None:
+    secret_name = f"{machine_name}-age.key"
+    if not has_secret(secret_name):  # skip uploading the secret, not managed by us
+        return
+    secret = decrypt_secret(secret_name)
+
+    h = parse_deployment_address(machine_name, deployment_address)
+    path = Path(age_key_file)
+
+    proc = h.run(
+        [
+            "bash",
+            "-c",
+            'mkdir -p "$0" && echo -n "$1" > "$2"',
+            str(path.parent),
+            secret,
+            age_key_file,
+        ],
+        check=False,
+    )
+    if proc.returncode != 0:
+        print(f"failed to upload age key to {deployment_address}")
+        sys.exit(1)
+
+
 def upload_command(args: argparse.Namespace) -> None:
     upload_secrets(args.machine)
 

pkgs/clan-cli/tests/test_flake_with_core/flake.nix

@@ -11,6 +11,17 @@
       machines = {
         vm1 = { modulesPath, ... }: {
           imports = [ "${toString modulesPath}/virtualisation/qemu-vm.nix" ];
+          clan.networking.deploymentAddress = "__CLAN_DEPLOYMENT_ADDRESS__";
+          sops.age.keyFile = "__CLAN_SOPS_KEY_PATH__";
+
+          clanCore.secrets.testpassword = {
+            generator = ''
+              echo "secret1" > "$secrets/secret1"
+              echo "fact1" > "$facts/fact1"
+            '';
+            secrets.secret1 = { };
+            facts.fact1 = { };
+          };
         };
       };
     };

pkgs/clan-cli/tests/test_secrets_generate.py

@@ -0,0 +1,38 @@
+from pathlib import Path
+from typing import TYPE_CHECKING
+
+import pytest
+from cli import Cli
+
+from clan_cli.machines.facts import machine_get_fact
+from clan_cli.secrets.folders import sops_secrets_folder
+from clan_cli.secrets.secrets import has_secret
+
+if TYPE_CHECKING:
+    from age_keys import KeyPair
+
+
+@pytest.mark.impure
+def test_upload_secret(
+    monkeypatch: pytest.MonkeyPatch,
+    test_flake_with_core: Path,
+    age_keys: list["KeyPair"],
+) -> None:
+    monkeypatch.chdir(test_flake_with_core)
+    monkeypatch.setenv("SOPS_AGE_KEY", age_keys[0].privkey)
+    cli = Cli()
+    cli.run(["secrets", "users", "add", "user1", age_keys[0].pubkey])
+    cli.run(["secrets", "generate", "vm1"])
+    has_secret("vm1-age.key")
+    has_secret("vm1-secret1")
+    fact1 = machine_get_fact("vm1", "fact1")
+    assert fact1 == "fact1\n"
+    age_key = sops_secrets_folder().joinpath("vm1-age.key").joinpath("secret")
+    secret1 = sops_secrets_folder().joinpath("vm1-secret1").joinpath("secret")
+    age_key_mtime = age_key.lstat().st_mtime_ns
+    secret1_mtime = secret1.lstat().st_mtime_ns
+
+    # test idempotency
+    cli.run(["secrets", "generate", "vm1"])
+    assert age_key.lstat().st_mtime_ns == age_key_mtime
+    assert secret1.lstat().st_mtime_ns == secret1_mtime

pkgs/clan-cli/tests/test_secrets_upload.py

@@ -0,0 +1,40 @@
+from pathlib import Path
+from typing import TYPE_CHECKING
+
+import pytest
+from cli import Cli
+
+from clan_cli.ssh import HostGroup
+
+if TYPE_CHECKING:
+    from age_keys import KeyPair
+
+
+@pytest.mark.impure
+def test_upload_secret(
+    monkeypatch: pytest.MonkeyPatch,
+    test_flake_with_core: Path,
+    host_group: HostGroup,
+    age_keys: list["KeyPair"],
+) -> None:
+    monkeypatch.chdir(test_flake_with_core)
+    monkeypatch.setenv("SOPS_AGE_KEY", age_keys[0].privkey)
+
+    cli = Cli()
+    cli.run(["secrets", "users", "add", "user1", age_keys[0].pubkey])
+
+    cli.run(["secrets", "machines", "add", "vm1", age_keys[1].pubkey])
+    monkeypatch.setenv("SOPS_NIX_SECRET", age_keys[0].privkey)
+    cli.run(["secrets", "set", "vm1-age.key"])
+
+    flake = test_flake_with_core.joinpath("flake.nix")
+    host = host_group.hosts[0]
+    addr = f"{host.user}@{host.host}:{host.port}?StrictHostKeyChecking=no&UserKnownHostsFile=/dev/null&IdentityFile={host.key}"
+    new_text = flake.read_text().replace("__CLAN_DEPLOYMENT_ADDRESS__", addr)
+    sops_key = test_flake_with_core.joinpath("sops.key")
+    new_text = new_text.replace("__CLAN_SOPS_KEY_PATH__", str(sops_key))
+
+    flake.write_text(new_text)
+    cli.run(["secrets", "upload", "vm1"])
+    assert sops_key.exists()
+    assert sops_key.read_text() == age_keys[0].privkey