hermes-agent/tests/tools/test_container_cwd_sanitize.py
Ben Barclay c15945655f
fix(terminal): sanitize host/relative cwd OVERRIDE before it reaches docker run -w (#50636)
terminal_tool() resolves a per-task cwd override that WINS over config["cwd"]:

    cwd = overrides.get("cwd") or config["cwd"]

config["cwd"] is sanitized for container backends in _get_env_config() (host
prefixes /Users//home//C:\\/C:/ and relative paths are replaced with the
backend default /root). But the override was applied RAW — it was never run
through that guard. The gateway/TUI registers the host launch dir as a cwd
override for workspace tracking (tui_gateway/server.py _register_session_cwd
-> _terminal_task_cwd -> _session_cwd -> os.getcwd()), so on a container
backend a host path leaked straight to `docker run -w <host-path>`:

  - Windows desktop: -w C:\Users\<user>  -> container fails to start (exit 125)
  - POSIX:           -w /home/<user>      -> same

The ACP adapter translates its override cwd (acp_adapter/session.py
_translate_acp_cwd), but the gateway path did neither translation nor
sanitization, so the override bypassed the one guard that would have caught it.

Fix: extract the host/relative-path predicate into a shared
_is_unusable_container_cwd() helper (so the existing _get_env_config()
sanitizer and the new guard can't drift), and re-apply it to the *resolved*
cwd at the override-resolution site. Valid in-container override paths
(RL/benchmark sandboxes that set cwd to /workspace, /root, ...) are absolute
non-host paths and pass through untouched.

Tests: unit-pin the predicate (Windows backslash/forwardslash, POSIX home,
macOS /Users, relative, valid container paths) AND an E2E call-site pin that
drives terminal_tool() with a host-path override registered and asserts the
cwd reaching _create_environment is sanitized. Mutation-verified: reverting
the call-site guard makes the two host-path E2E tests fail (showing the raw
host path leaking) while the valid-/workspace-override test stays green.
2026-06-25 02:33:40 +00:00

146 lines
6.4 KiB
Python

"""Regression tests for host-path cwd sanitization on container backends.
Two code paths in ``tools/terminal_tool.py`` must reject a host (or relative)
working directory before it reaches ``docker run -w``:
1. ``_get_env_config()`` sanitizes the ``TERMINAL_CWD``-derived ``config["cwd"]``.
2. ``terminal_tool()`` resolves a *per-task cwd override* that WINS over
``config["cwd"]`` (registered by the gateway/TUI for workspace tracking,
and by RL/benchmark envs). That override was applied RAW — never sanitized
— so a host cwd (e.g. a Windows desktop session's ``C:\\Users\\<user>``)
leaked straight to ``docker run -w C:\\Users\\<user>``, which fails to start
the container (exit 125). The sanitizer at path #1 lists ``C:\\``/``C:/`` as
host prefixes but only ever ran against ``config["cwd"]``, so the override
bypassed the one guard that would have caught it.
Both paths now share ``_is_unusable_container_cwd()``; these tests pin its
behaviour so neither path can regress.
"""
import tools.terminal_tool as tt
class TestIsUnusableContainerCwd:
def test_windows_backslash_host_path_rejected(self):
# The exact shape from the bug report: a Windows host cwd reaching a
# Linux container's -w flag.
assert tt._is_unusable_container_cwd(r"C:\Users\someuser") is True
def test_windows_forwardslash_host_path_rejected(self):
assert tt._is_unusable_container_cwd("C:/Users/someuser") is True
def test_posix_home_host_path_rejected(self):
assert tt._is_unusable_container_cwd("/home/ben/projects") is True
def test_macos_users_host_path_rejected(self):
assert tt._is_unusable_container_cwd("/Users/ben/projects") is True
def test_relative_path_rejected(self):
assert tt._is_unusable_container_cwd(".") is True
assert tt._is_unusable_container_cwd("src/app") is True
def test_valid_container_workspace_accepted(self):
# In-container paths that RL/benchmark overrides legitimately set must
# pass through untouched.
assert tt._is_unusable_container_cwd("/workspace") is False
assert tt._is_unusable_container_cwd("/root") is False
assert tt._is_unusable_container_cwd("/app") is False
assert tt._is_unusable_container_cwd("/opt/project") is False
def test_empty_is_not_flagged(self):
# Empty/None-ish cwd is handled by the caller's `or config["cwd"]`
# fallback, not by flagging it here.
assert tt._is_unusable_container_cwd("") is False
def test_host_prefixes_include_windows_and_posix(self):
# Guard the constant itself — the Windows entries are the ones that
# were load-bearing for the reported desktop bug.
assert r"C:\\"[:2] in tt._HOST_CWD_PREFIXES or "C:\\" in tt._HOST_CWD_PREFIXES
assert "C:/" in tt._HOST_CWD_PREFIXES
assert "/home/" in tt._HOST_CWD_PREFIXES
assert "/Users/" in tt._HOST_CWD_PREFIXES
def test_container_backends_set(self):
assert tt._CONTAINER_BACKENDS == frozenset(
{"docker", "singularity", "modal", "daytona"}
)
class TestOverrideCwdSanitizedAtCallSite:
"""E2E pin: a per-task cwd OVERRIDE that is a host path must NOT reach the
container builder. This is the actual reported bug — the gateway/TUI
registers the host launch dir as a cwd override, which previously won over
the (sanitized) config["cwd"] and flowed raw into `docker run -w`.
"""
def _run_and_capture_cwd(self, monkeypatch, override_cwd, config_cwd="/root"):
"""Drive terminal_tool() on the docker backend with a host-path cwd
override registered, and return the cwd that reached _create_environment
(i.e. the cwd that would be passed to `docker run -w`).
"""
captured = {}
config = {
"env_type": "docker",
"docker_image": "pytorch/pytorch:latest",
"cwd": config_cwd,
"host_cwd": None,
"timeout": 180,
"lifetime_seconds": 300,
"container_cpu": 1,
"container_memory": 5120,
"container_disk": 51200,
"container_persistent": True,
"docker_volumes": [],
"docker_env": {},
"docker_extra_args": [],
"docker_mount_cwd_to_workspace": False,
"docker_run_as_host_user": False,
"docker_forward_env": [],
"modal_mode": "auto",
}
class _DummyEnv:
cwd = config_cwd
def execute(self, *a, **k):
return {"output": "", "exit_code": 0}
def fake_create_environment(env_type, image, cwd, timeout, **kwargs):
captured["cwd"] = cwd
return _DummyEnv()
monkeypatch.setattr(tt, "_get_env_config", lambda: config)
monkeypatch.setattr(tt, "_start_cleanup_thread", lambda: None)
monkeypatch.setattr(tt, "_check_all_guards", lambda *a, **k: {"approved": True})
monkeypatch.setattr(tt, "_create_environment", fake_create_environment)
# Force a fresh environment build so _create_environment is invoked.
monkeypatch.setattr(tt, "_active_environments", {})
monkeypatch.setattr(tt, "_last_activity", {})
task_id = "sess-host-cwd"
tt.register_task_env_overrides(task_id, {"cwd": override_cwd})
try:
tt.terminal_tool(command="pwd", task_id=task_id)
finally:
tt.clear_task_env_overrides(task_id)
tt._active_environments.pop(task_id, None)
tt._active_environments.pop("default", None)
return captured.get("cwd")
def test_windows_host_override_does_not_reach_container(self, monkeypatch):
# The bug: C:\Users\<user> registered as override → docker run -w C:\Users\<user> → exit 125.
cwd = self._run_and_capture_cwd(monkeypatch, r"C:\Users\someuser")
assert cwd == "/root", (
f"Host-path cwd override leaked to the container builder: {cwd!r}. "
"It must be sanitized back to config['cwd']."
)
def test_posix_host_override_does_not_reach_container(self, monkeypatch):
cwd = self._run_and_capture_cwd(monkeypatch, "/home/someuser/project")
assert cwd == "/root"
def test_valid_container_override_is_preserved(self, monkeypatch):
# RL/benchmark envs set an in-container path; it must pass through.
cwd = self._run_and_capture_cwd(monkeypatch, "/workspace/task42")
assert cwd == "/workspace/task42"