Verify owned TTY cleanup before reporting stop

This commit is contained in:
FrameBridge contributors committed 2026-10-27 16:54:03 -04:00
1 parent ba17412c60
commit f6b14c5210
8 files changed
+759 -34

No files matched your search

+290
View File
@@ -0,0 +1,290 @@
"""Owned real-console cleanup, daemon outcome and fail-closed evidence regressions."""
import json
import os
from pathlib import Path
import signal
import subprocess
import sys
import tempfile
import time
import unittest
from unittest import mock
import uuid
ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(ROOT / "expanded"))
os.environ["FRAMEBRIDGE_EXPANDED_PREFIX"] = str(ROOT)
from framebridge_expanded import tty, tty_cleanup, tty_sudo_control as control
class CleanupTests(unittest.TestCase):
def setUp(self):
self.temp = tempfile.TemporaryDirectory(prefix="fb-cleanup-qa-", dir=Path("/run/user") / str(os.getuid()))
self.state = Path(self.temp.name)
self.addCleanup(self.temp.cleanup)
self.addCleanup(tty.stop, self.state)
def dead_daemon(self):
child = subprocess.Popen([sys.executable, "-c", "import time;time.sleep(10)"])
identity = tty._identity(child.pid)
child.terminate()
child.wait()
return tty_cleanup.daemon_identity(identity)
def pidfile(self, daemon):
tty._write_json(self.state / "tty-process.json", {key: daemon[key] for key in ("pid", "birth", "uid", "boot")})
def assert_gone(self, record):
for pid, birth in ((record["pid"], record["birth"]), (record["server_pid"], record["server_birth"])):
process = control._proc(pid)
self.assertFalse(process and process["birth"] == birth and process["state"] not in {"Z", "X"})
controller = control.SudoConsoleController(tty.runtime_path(self.state), timeout=5)
self.assertEqual(controller._scope_members_existing(record), set())
def test_real_close_and_stop_verify_owned_process_death(self):
self.assertTrue(tty.start(self.state)["ok"])
for close_first in (True, False):
opened = tty.request(self.state, {"authorized": True, "op": "open"})
self.assertTrue(opened["ok"], opened)
receipt = tty_cleanup.read(self.state)
self.assertEqual(len(receipt["sessions"]), 1)
record = receipt["sessions"][0]
if close_first:
closed = tty.request(self.state, {"authorized": True, "op": "close", "session_id": record["session_id"]})
self.assertTrue(closed["ok"], closed)
self.assertTrue(closed["result"]["cleanup_verified"])
stopped = tty.stop(self.state)
self.assertTrue(stopped["ok"], stopped)
self.assertTrue(stopped["cleanup_verified"])
self.assert_gone(record)
self.assertTrue(tty.stop(self.state)["cleanup_verified"])
if close_first:
self.assertTrue(tty.start(self.state)["ok"])
def test_failed_shutdown_cannot_become_success_by_stop_or_restart(self):
limits = tty.Limits()
home = self.state / "home"; home.mkdir()
manager = tty.PortableManager(self.state, limits, home)
manager.cleanup_daemon = self.dead_daemon()
manager._record_cleanup()
self.addCleanup(manager.shutdown)
manager.open()
self.pidfile(manager.cleanup_daemon)
with mock.patch.object(manager._sudo, "close", side_effect=control.SudoControlError("TTY_SUDO_SCOPE_STILL_RUNNING")):
result = manager.shutdown()
self.assertFalse(result["cleanup_verified"])
before = (self.state / tty_cleanup.NAME).read_bytes()
for _ in range(2):
result = tty.stop(self.state)
self.assertFalse(result["ok"], result)
self.assertFalse(result["cleanup_verified"])
self.assertEqual(result["error"], "TTY_CLEANUP_UNVERIFIED")
self.assertEqual(tty.start(self.state)["error"], "TTY_CLEANUP_PENDING")
self.assertEqual((self.state / tty_cleanup.NAME).read_bytes(), before)
self.assertTrue(control._proc(tty_cleanup.read(self.state)["sessions"][0]["pid"]))
self.assertTrue(manager.shutdown()["cleanup_verified"])
self.assertTrue(tty.stop(self.state)["cleanup_verified"])
def test_missing_or_other_daemon_receipt_does_not_verify_stop(self):
expected = self.dead_daemon(); self.pidfile(expected)
self.assertFalse(tty.stop(self.state)["cleanup_verified"])
other = self.dead_daemon()
tty_cleanup.write(self.state, other, [], verified=True)
self.assertFalse(tty.stop(self.state)["cleanup_verified"])
def test_untrusted_or_tampered_receipt_fails_closed(self):
daemon = self.dead_daemon(); self.pidfile(daemon)
path = self.state / tty_cleanup.NAME
tty_cleanup.write(self.state, daemon, [], verified=True)
path.chmod(0o644)
self.assertFalse(tty.stop(self.state)["cleanup_verified"])
path.chmod(0o600)
value = json.loads(path.read_text()); value["daemon"]["uid"] += 1
tty._write_json(path, value)
self.assertFalse(tty.stop(self.state)["cleanup_verified"])
value["daemon"] = daemon; value["state_dir"] += "-other"
tty._write_json(path, value)
self.assertFalse(tty.stop(self.state)["cleanup_verified"])
path.unlink(); path.symlink_to(self.state / "tty-process.json")
self.assertFalse(tty.stop(self.state)["cleanup_verified"])
self.assertEqual(tty.start(self.state)["error"], "TTY_CLEANUP_PENDING")
def test_inflight_operation_receipt_is_preserved_on_restart(self):
daemon = self.dead_daemon()
tty_cleanup.write(self.state, daemon, [], operation_pending=True)
before = (self.state / tty_cleanup.NAME).read_bytes()
self.assertEqual(tty.start(self.state)["error"], "TTY_CLEANUP_PENDING")
self.assertEqual((self.state / tty_cleanup.NAME).read_bytes(), before)
def test_pending_receipt_allows_restart_only_after_natural_identity_death(self):
daemon = self.dead_daemon()
child = subprocess.Popen([sys.executable, "-c", "import time;time.sleep(10)"])
identity = tty._identity(child.pid)
sid = uuid.uuid4().hex
unit = "framebridge-tty-" + "a" * 12 + "-" + sid + "-" + "b" * 8 + ".service"
record = {"session_id": sid, "unit_name": unit, "pid": child.pid, "birth": identity["birth"],
"server_pid": daemon["pid"], "server_birth": daemon["birth"],
"control_group": "/absent-fb-cleanup-qa/" + unit,
"pane_control_group": "/absent-fb-cleanup-qa/" + unit}
tty_cleanup.write(self.state, daemon, [record], errors=[{"error": "TTY_SUDO_SCOPE_STILL_RUNNING"}])
try:
self.assertEqual(tty.start(self.state)["error"], "TTY_CLEANUP_PENDING")
finally:
child.terminate(); child.wait()
self.assertTrue(tty.start(self.state)["ok"])
self.assertTrue(tty.stop(self.state)["cleanup_verified"])
def test_prelaunch_invalid_arguments_do_not_poison_profile(self):
home = self.state / "home"; home.mkdir()
manager = tty.PortableManager(self.state, tty.Limits(), home)
manager.cleanup_daemon = self.dead_daemon()
manager._record_cleanup()
self.addCleanup(manager.shutdown)
with mock.patch.object(manager._sudo, "_run") as run:
with self.assertRaises(control.SudoControlError):
manager._sudo.launch(uuid.uuid4().hex, cwd=self.state / "absent")
with self.assertRaises(tty.TerminalDenied):
manager.open(rows=0)
run.assert_not_called()
self.assertFalse(manager._cleanup_uncertain())
self.assertTrue(manager.shutdown()["cleanup_verified"])
self.assertTrue(tty.start(self.state)["ok"])
self.assertTrue(tty.stop(self.state)["cleanup_verified"])
def test_previous_boot_receipts_recover_without_signaling_old_pid_numbers(self):
daemon = tty_cleanup.daemon_identity(tty._identity(os.getpid()))
daemon["boot"] = str(uuid.uuid4())
for verified, pending in ((True, False), (False, False), (False, True)):
with self.subTest(verified=verified, pending=pending):
tty_cleanup.write(self.state, daemon, [], verified=verified, operation_pending=pending)
self.pidfile(daemon)
with mock.patch.object(os, "pidfd_open") as opening:
self.assertTrue(tty.stop(self.state)["cleanup_verified"])
opening.assert_not_called()
self.assertTrue(tty.start(self.state)["ok"])
self.assertTrue(tty.stop(self.state)["cleanup_verified"])
def test_verified_launch_rollback_does_not_poison_profile(self):
home = self.state / "home"; home.mkdir()
manager = tty.PortableManager(self.state, tty.Limits(), home)
manager.cleanup_daemon = self.dead_daemon(); manager._record_cleanup()
self.addCleanup(manager.shutdown)
original = manager._sudo._inspect
injected = False
def mismatch(record):
nonlocal injected
current = original(record)
if not injected:
injected = True
return {**current, "no_new_privs": 0}
return current
with mock.patch.object(manager._sudo, "_inspect", side_effect=mismatch):
with self.assertRaises(control.SudoControlError) as error:
manager.open()
self.assertEqual(error.exception.code, "TTY_SUDO_PRIVILEGE_STATE_MISMATCH")
self.assertFalse(manager._cleanup_uncertain())
self.assertTrue(manager.shutdown()["cleanup_verified"])
self.assertTrue(manager.open()["session_id"])
self.assertTrue(manager.shutdown()["cleanup_verified"])
def test_partial_launch_and_failed_rollback_remain_unverified(self):
home = self.state / "home"; home.mkdir()
manager = tty.PortableManager(self.state, tty.Limits(), home)
manager.cleanup_daemon = self.dead_daemon()
manager._record_cleanup(); self.pidfile(manager.cleanup_daemon)
original = manager._sudo._run
def failed(command, **kwargs):
if command[0] == manager._sudo.systemd_run:
original(command, **kwargs)
raise control.SudoControlError("TTY_SUDO_BACKEND_TIMEOUT")
if command[0] == manager._sudo.systemctl and "stop" in command:
raise control.SudoControlError("TTY_SUDO_BACKEND_TIMEOUT")
return original(command, **kwargs)
try:
with mock.patch.object(manager._sudo, "_run", side_effect=failed):
with self.assertRaises(control.SudoControlError):
manager.open()
self.assertTrue(manager._sudo._uncertain_launches)
self.assertFalse(manager.shutdown()["cleanup_verified"])
self.assertTrue(tty_cleanup.read(self.state)["operation_pending"])
self.assertFalse(tty.stop(self.state)["cleanup_verified"])
self.assertEqual(tty.start(self.state)["error"], "TTY_CLEANUP_PENDING")
finally:
for unit in manager._sudo._uncertain_launches:
manager._sudo._unit(unit) # Narrow ownership validation first.
original([manager._sudo.systemctl, "--user", "stop", unit], optional=True)
def test_real_close_kills_hup_and_term_ignoring_owned_job(self):
controller = control.SudoConsoleController(tty.runtime_path(self.state), timeout=5)
record = controller.launch(uuid.uuid4().hex, cwd=self.state)
self.addCleanup(lambda: controller.close(record) if record["session_id"] in controller._records else None)
marker = self.state / "child.pid"
script = "import os,signal,time;signal.signal(signal.SIGHUP,signal.SIG_IGN);signal.signal(signal.SIGTERM,signal.SIG_IGN);open(" + repr(str(marker)) + ",'w').write(str(os.getpid()));time.sleep(30)"
import shlex
command = "nohup " + shlex.quote(sys.executable) + " -c " + shlex.quote(script) + " >/dev/null 2>&1 & disown"
controller._tmux(record, "send-keys", "-t", record["pane"], "-l", command)
controller._tmux(record, "send-keys", "-t", record["pane"], "Enter")
deadline = time.monotonic() + 3
while not marker.exists() and time.monotonic() < deadline:
time.sleep(0.03)
self.assertTrue(marker.exists())
child = control._proc(int(marker.read_text()))
calls = []
original = controller._run
def run(command, **kwargs):
calls.append(command)
return original(command, **kwargs)
started = time.monotonic()
with mock.patch.object(controller, "_run", side_effect=run):
self.assertTrue(controller.close(record)["closed"])
self.assertLess(time.monotonic() - started, 5)
killed = control._proc(child["pid"])
self.assertFalse(killed and killed["birth"] == child["birth"] and killed["state"] not in {"Z", "X"})
self.assert_gone(record)
kill_index = next(i for i, call in enumerate(calls) if "kill-session" in call)
stops = [(i, call) for i, call in enumerate(calls) if "stop" in call]
self.assertTrue(stops)
self.assertTrue(all(i > kill_index and "--no-block" in call for i, call in stops))
class PidfdGuardTests(unittest.TestCase):
def test_reused_pid_after_pidfd_open_is_not_signaled(self):
controller = control.SudoConsoleController.__new__(control.SudoConsoleController)
controller.uid = os.getuid()
controller.timeout = 0.06
controller.systemctl = "systemctl"
record = {"session_id": "a" * 32, "unit_name": "owned.service", "name": "private-session",
"pid": 123456, "birth": "1", "server_pid": 123457, "server_birth": "2",
"control_group": "/private/owned.service", "pane_control_group": "/private/own.scope",
"pane_scope_unit": "own.scope", "busy_reason": None}
member = {"pid": 123458, "birth": "3", "uids": (os.getuid(),) * 4, "state": "S"}
opened = False
def pidfd_open(pid):
nonlocal opened
self.assertEqual(pid, member["pid"])
opened = True
return 987
def process(pid):
if pid == member["pid"]:
return {**member, "birth": "4" if opened else "3"}
return None
with (mock.patch.object(controller, "_inspect", return_value=record),
mock.patch.object(controller, "_tmux"),
mock.patch.object(controller, "_run"),
mock.patch.object(controller, "_close_members", return_value={member["pid"]: member}),
mock.patch.object(controller, "_terminated", return_value=record),
mock.patch.object(controller, "_pane_group") as group,
mock.patch.object(control, "_proc", side_effect=process),
mock.patch.object(os, "pidfd_open", side_effect=pidfd_open),
mock.patch.object(os, "close") as closing,
mock.patch.object(signal, "pidfd_send_signal") as sending):
with self.assertRaises(control.SudoControlError):
controller._close(record)
self.assertTrue(opened)
sending.assert_not_called()
group.assert_not_called()
closing.assert_called_once_with(987)
if __name__ == "__main__":
unittest.main()