Make ISO acceptance probes fail on forbidden state

dev/iso-smoke.py

@@ -260,19 +260,43 @@ findmnt -n -o FSTYPE / | grep -qx overlay
 test -f /run/initramfs/live/LiveOS/squashfs.img
 sudo -n true
 for tool in hydra hydra-install qemu-system-x86_64 qemu-img virt-viewer foot labwc; do command -v "$tool"; done
-systemctl is-active NetworkManager hydra-network hydra-dnsmasq
+for service in NetworkManager hydra-network hydra-dnsmasq; do systemctl is-active "$service"; done
 '''
 
 INSTALLED_CHECK = r'''set -eu
+assert_no_match() {
+    if grep "$@"; then
+        printf 'Forbidden live configuration remains\n' >&2
+        exit 1
+    else
+        status=$?
+        [ "$status" -eq 1 ] || exit "$status"
+    fi
+}
 [ "$(findmnt -n -o FSTYPE /)" = ext4 ]
 [ "$(readlink -f "$(findmnt -n -o SOURCE /)")" = /dev/vda2 ]
-! grep -Eq 'root=live:|rd\.live' /proc/cmdline
+assert_no_match -Eq 'root=live:|rd\.live' /proc/cmdline
-! findmnt -n -o FSTYPE / | grep -Eq 'overlay|squashfs'
 test ! -e /etc/hydra/live-mode
-! getent passwd hydra-live
+test ! -L /etc/hydra/live-mode
-! find /etc/systemd/system /etc/sudoers.d -name 90-hydra-live.conf -o -name 90-hydra-live | grep .
+if getent passwd hydra-live; then
+    printf 'Live account remains\n' >&2
+    exit 1
+else
+    status=$?
+    [ "$status" -eq 2 ] || exit "$status"
+fi
+remaining=$(find /etc/systemd/system /etc/sudoers.d \( -name 90-hydra-live.conf -o -name 90-hydra-live \) -print)
+[ -z "$remaining" ]
 test ! -e /etc/greetd/live.toml
-! grep -R -E -- '--autologin|NOPASSWD' /etc/systemd/system/getty@*.service.d /etc/systemd/system/serial-getty@*.service.d /etc/sudoers.d 2>/dev/null
+test ! -L /etc/greetd/live.toml
+for directory in /etc/systemd/system/getty@*.service.d /etc/systemd/system/serial-getty@*.service.d /etc/sudoers.d; do
+    if [ -d "$directory" ]; then
+        assert_no_match -R -E -- '--autologin|NOPASSWD' "$directory"
+    elif [ -e "$directory" ] || [ -L "$directory" ]; then
+        printf 'Unexpected non-directory configuration path\n' >&2
+        exit 1
+    fi
+done
 getent shadow root | cut -d: -f2 | grep -Eq '^[!*]'
 getent shadow hydratest | cut -d: -f2 | grep -Eq '^\$'
 for service in NetworkManager greetd hydra-network hydra-dnsmasq; do systemctl is-active "$service"; done
@@ -329,7 +353,7 @@ try:
     port = 43117
     server_code = "import socket; s=socket.socket(); s.settimeout(15); s.bind(('10.203.0.250',43117)); s.listen(1); c,a=s.accept(); c.sendall(b'HYDRA_LAN_TO_VM_OK'); c.close(); s.close()"
     server = subprocess.Popen(['ip', 'netns', 'exec', guest, 'python3', '-c', server_code], stdout=subprocess.DEVNULL, stderr=subprocess.PIPE)
-    client_code = "import socket,time; deadline=time.monotonic()+10\nwhile True:\n try:\n  s=socket.create_connection(('10.203.0.250',43117),1); break\n except OSError:\n  if time.monotonic()>deadline: raise\n  time.sleep(.1)\nassert s.recv(100)==b'HYDRA_LAN_TO_VM_OK'; s.close(); print('PASS: routed new LAN-to-VM TCP connection on arbitrary port 43117')"
+    client_code = "import socket,time; deadline=time.monotonic()+10\nwhile True:\n try:\n  s=socket.create_connection(('10.203.0.250',43117),1); break\n except OSError:\n  if time.monotonic()>deadline: raise\n  time.sleep(.1)\nexpected=b'HYDRA_LAN_TO_VM_OK'; received=b''\nwhile len(received)<len(expected):\n remaining=deadline-time.monotonic()\n if remaining<=0: raise TimeoutError('TCP marker deadline exceeded')\n s.settimeout(remaining)\n part=s.recv(len(expected)-len(received))\n if not part: raise RuntimeError('Truncated TCP marker')\n received+=part\nassert received==expected; s.close(); print('PASS: routed new LAN-to-VM TCP connection on arbitrary port 43117')"
     print(run('ip', 'netns', 'exec', lan, 'python3', '-c', client_code), end='')
     assert server.wait(timeout=10) == 0
 finally:

tests/test_iso_smoke.py

@@ -141,3 +141,159 @@ class SmokeTests(unittest.TestCase):
         with self.assertRaises(self.mod.SmokeError):
             self.mod.qemu_argv('qemu', self.root, comma, self.code,
                                self.variables, None, 2048)
+
+    def probe_fixture(self, case, live=False):
+        # Run the production shell against ordinary scratch files. Commands that
+        # query guest state are boundary fixtures; grep/find/test are real tools.
+        import os
+        root = self.root / 'guest'
+        for directory in ('etc/hydra', 'etc/greetd', 'etc/sudoers.d',
+                          'etc/systemd/system/getty@tty2.service.d',
+                          'etc/systemd/system/serial-getty@ttyS0.service.d',
+                          'usr/share/firmware', 'proc', 'sys/firmware/efi',
+                          'run/initramfs/live/LiveOS'):
+            (root / directory).mkdir(parents=True, exist_ok=True)
+        (root / 'etc/greetd/config.toml').write_text('normal authentication')
+        (root / 'usr/share/firmware/OVMF_CODE.fd').write_text('firmware')
+        (root / 'proc/cmdline').write_text('root=UUID=disk ro' if case != 'live-cmdline' else 'root=live:CDLABEL=HYDRA_OS rd.live.image')
+        if live or case == 'live-marker':
+            (root / 'etc/hydra/live-mode').mkdir(exist_ok=True)
+            (root / 'etc/hydra/live-mode/active').touch()
+        if live:
+            (root / 'run/initramfs/live/LiveOS/squashfs.img').touch()
+        files = {
+            'live-dropin': ('etc/systemd/system/getty@tty2.service.d/90-hydra-live.conf', 'live'),
+            'live-sudoers': ('etc/sudoers.d/90-hydra-live', 'live'),
+            'autologin': ('etc/systemd/system/serial-getty@ttyS0.service.d/other.conf', 'ExecStart=agetty --autologin hydratest'),
+            'nopasswd': ('etc/sudoers.d/other', 'hydratest ALL=(ALL) NOPASSWD: ALL'),
+            'live-greetd': ('etc/greetd/live.toml', 'live'),
+        }
+        if case in files:
+            path, text = files[case]
+            (root / path).write_text(text)
+        if case == 'config-read-error':
+            unreadable = root / 'etc/sudoers.d/unreadable'
+            unreadable.write_text('configuration')
+            unreadable.chmod(0)
+        if case == 'cmdline-read-error':
+            (root / 'proc/cmdline').unlink()
+        commands = self.root / 'fixture-bin'
+        commands.mkdir(exist_ok=True)
+        dispatcher = commands / 'fixture'
+        dispatcher.write_text('''#!/usr/bin/env python3
+import os, pathlib, sys
+name = pathlib.Path(sys.argv[0]).name
+args = sys.argv[1:]
+case = os.environ['SMOKE_CASE']
+if name == 'id': print('hydra-live')
+elif name == 'findmnt': print('/dev/vda2' if 'SOURCE' in args else ('overlay' if os.environ['SMOKE_LIVE']=='1' else 'ext4'))
+elif name == 'systemctl':
+    if args[0] == 'is-active':
+        states = [unit != case for unit in args[1:]]
+        print('\\n'.join('active' if state else 'failed' for state in states))
+        sys.exit(0 if any(states) else 3)
+    print('success' if 'Result' in args else '0')
+elif name == 'getent':
+    if args[:2] == ['passwd','hydra-live']:
+        if case == 'account-lookup-error': sys.exit(3)
+        if case != 'live-account': sys.exit(2)
+        print('hydra-live:x:1000:1000::/home/hydra-live:/bin/bash')
+    elif args[0] == 'shadow': print(args[1] + ':' + ('!' if args[1]=='root' else '$6$hash') + ':1::::::')
+    else: print('127.0.0.1 STREAM example')
+elif name == 'ip': print('10.203.0.1/24' if 'addr' in args else 'default via 10.0.2.2')
+elif name == 'nmcli': print('connected')
+else: print('fixture tool')
+''')
+        dispatcher.chmod(0o755)
+        for command in ('id', 'findmnt', 'systemctl', 'getent', 'ip', 'nft',
+                        'nmcli', 'sudo', 'hydra', 'hydra-install',
+                        'qemu-system-x86_64', 'qemu-img', 'virt-viewer', 'foot', 'labwc'):
+            path = commands / command
+            if not path.exists():
+                path.symlink_to(dispatcher)
+        probe = self.mod.LIVE_CHECK if live else self.mod.INSTALLED_CHECK
+        for path in ('/etc/', '/usr/share', '/proc/cmdline', '/sys/firmware/efi', '/run/initramfs/live/LiveOS/squashfs.img'):
+            probe = probe.replace(path, str(root) + path)
+        env = dict(os.environ, PATH=str(commands) + ':' + os.environ['PATH'],
+                   SMOKE_CASE=case, SMOKE_LIVE='1' if live else '0')
+        result = subprocess.run(['/bin/sh', '-c', self.mod.frame_command(probe, 'probecheck')],
+                                env=env, capture_output=True, timeout=10)
+        parsed = self.mod.parse_frame(result.stdout, 'probecheck')
+        self.assertIsNotNone(parsed, result.stderr)
+        return parsed[0]
+
+    def test_installed_probe_rejects_each_retained_live_artifact(self):
+        self.assertEqual(self.probe_fixture('clean'), 0)
+        for case in ('live-cmdline', 'live-account', 'live-dropin', 'live-sudoers',
+                     'autologin', 'nopasswd', 'live-marker', 'live-greetd'):
+            with self.subTest(case=case):
+                # Give each negative case an independent clean filesystem.
+                import shutil
+                shutil.rmtree(self.root / 'guest')
+                self.assertNotEqual(self.probe_fixture(case), 0)
+
+    def test_installed_probe_does_not_treat_account_query_error_as_absence(self):
+        self.assertNotEqual(self.probe_fixture('account-lookup-error'), 0)
+
+    def test_installed_probe_rejects_unreadable_configuration(self):
+        self.assertNotEqual(self.probe_fixture('config-read-error'), 0)
+
+    def test_installed_probe_rejects_unreadable_boot_arguments(self):
+        self.assertNotEqual(self.probe_fixture('cmdline-read-error'), 0)
+
+    def test_installed_probe_rejects_each_failed_required_service(self):
+        self.assertEqual(self.probe_fixture('clean'), 0)
+        for service in ('NetworkManager', 'greetd', 'hydra-network', 'hydra-dnsmasq'):
+            with self.subTest(service=service):
+                self.assertNotEqual(self.probe_fixture(service), 0)
+
+    def test_live_probe_rejects_each_failed_required_service(self):
+        self.assertEqual(self.probe_fixture('clean', live=True), 0)
+        for service in ('NetworkManager', 'hydra-network', 'hydra-dnsmasq'):
+            with self.subTest(service=service):
+                self.assertNotEqual(self.probe_fixture(service, live=True), 0)
+
+    def tcp_probe(self, fragments):
+        # Execute the exact Python client embedded in the guest network probe,
+        # redirecting only its address to a fresh loopback socket.
+        import ast
+        import time
+        source = self.mod.NETWORK_CHECK.split("<<'NETWORK'\n", 1)[1].rsplit('\nNETWORK\n', 1)[0]
+        assignments = (node for node in ast.walk(ast.parse(source)) if isinstance(node, ast.Assign))
+        value = next(node.value for node in assignments
+                     if any(isinstance(target, ast.Name) and target.id == 'client_code' for target in node.targets))
+        client = ast.parse(ast.literal_eval(value))
+        listener = socket.socket()
+        self.addCleanup(listener.close)
+        listener.bind(('127.0.0.1', 0))
+        listener.listen(1)
+        listener.settimeout(3)
+        host, port = listener.getsockname()
+        class Redirect(ast.NodeTransformer):
+            def visit_Tuple(self, node):
+                if isinstance(node.elts[0], ast.Constant) and node.elts[0].value == '10.203.0.250':
+                    return ast.Tuple(elts=[ast.Constant(host), ast.Constant(port)], ctx=ast.Load())
+                return self.generic_visit(node)
+        client = ast.fix_missing_locations(Redirect().visit(client))
+        def server():
+            connection, _ = listener.accept()
+            with connection:
+                for fragment in fragments:
+                    try:
+                        connection.sendall(fragment)
+                    except BrokenPipeError:
+                        break
+                    time.sleep(.05)
+        thread = threading.Thread(target=server)
+        thread.start()
+        result = subprocess.run(['python3', '-c', ast.unparse(client)], capture_output=True, timeout=5)
+        thread.join(timeout=3)
+        self.assertFalse(thread.is_alive())
+        return result
+
+    def test_tcp_probe_accepts_complete_marker_across_fragmented_reads(self):
+        result = self.tcp_probe([b'HYDRA_', b'LAN_TO_', b'VM_OK'])
+        self.assertEqual(result.returncode, 0, result.stderr)
+
+    def test_tcp_probe_rejects_truncated_marker(self):
+        self.assertNotEqual(self.tcp_probe([b'HYDRA_LAN_TO_']).returncode, 0)