[core] Sensitive redaction via yaml_util representer

cv.sensitive(...) now returns a SensitiveStr (thin str subclass) so the
tag travels with the validated value. yaml_util.dump constructs a
per-call ESPHomeDumper subclass with a class-attribute redaction flag;
the PyYAML representer for SensitiveStr renders values wrapped in
literal \\033[8m...\\033[28m text when show_secrets is False and raw
when True. The post-dump regex in command_config is deleted.

Also tags wifi.ssid sites with cv.sensitive so SSID coverage isn't lost
when the regex (which matched 'ssid:' via substring) goes away.

No module-level mutable state; the per-call subclass keeps each dump
invocation self-contained and thread-safe by construction.
This commit is contained in:
J. Nick Koston
2026-05-26 21:49:31 -05:00
parent 8d19c55be2
commit 49b9d9313e
6 changed files with 148 additions and 11 deletions

View File

@@ -1412,11 +1412,6 @@ def command_config(args: ArgsProtocol, config: ConfigType) -> int | None:
if not CORE.verbose:
config = strip_default_ids(config)
output = yaml_util.dump(config, args.show_secrets)
# add the console decoration so the front-end can hide the secrets
if not args.show_secrets:
output = re.sub(
r"(password|key|psk|ssid)\: (.+)", r"\1: \\033[8m\2\\033[28m", output
)
if not CORE.quiet:
safe_print(output)
_LOGGER.info("Configuration is valid!")

View File

@@ -271,7 +271,7 @@ EAP_AUTH_SCHEMA = cv.All(
WIFI_NETWORK_BASE = cv.Schema(
{
cv.GenerateID(): cv.declare_id(WiFiAP),
cv.Optional(CONF_SSID): cv.ssid,
cv.Optional(CONF_SSID): cv.sensitive(cv.ssid),
cv.Optional(CONF_PASSWORD): cv.sensitive(validate_password),
cv.Optional(CONF_CHANNEL): validate_channel,
cv.Optional(CONF_MANUAL_IP): STA_MANUAL_IP_SCHEMA,
@@ -434,7 +434,7 @@ CONFIG_SCHEMA = cv.All(
cv.Optional(CONF_NETWORKS): cv.All(
cv.ensure_list(WIFI_NETWORK_STA), cv.Length(max=MAX_WIFI_NETWORKS)
),
cv.Optional(CONF_SSID): cv.ssid,
cv.Optional(CONF_SSID): cv.sensitive(cv.ssid),
cv.Optional(CONF_PASSWORD): cv.sensitive(validate_password),
cv.Optional(CONF_MANUAL_IP): STA_MANUAL_IP_SCHEMA,
cv.Optional(CONF_EAP): EAP_AUTH_SCHEMA,
@@ -850,7 +850,7 @@ async def final_step():
WiFiConfigureAction,
cv.Schema(
{
cv.Required(CONF_SSID): cv.templatable(cv.ssid),
cv.Required(CONF_SSID): cv.sensitive(cv.templatable(cv.ssid)),
cv.Required(CONF_PASSWORD): cv.sensitive(cv.templatable(validate_password)),
cv.Optional(CONF_SAVE, default=True): cv.templatable(cv.boolean),
cv.Optional(CONF_TIMEOUT, default="30000ms"): cv.templatable(

View File

@@ -101,7 +101,7 @@ from esphome.schema_extractors import (
)
from esphome.util import parse_esphome_version
from esphome.voluptuous_schema import _Schema
from esphome.yaml_util import make_data_base
from esphome.yaml_util import SensitiveStr, make_data_base
_LOGGER = logging.getLogger(__name__)
@@ -514,7 +514,13 @@ class SensitiveValidator:
self.inner = inner
def __call__(self, value: typing.Any) -> typing.Any:
return self.inner(value)
validated = self.inner(value)
# Tag string results so yaml_util.dump can mask them. Non-string
# results pass through unchanged; already-tagged values are not
# re-wrapped to keep nested cv.sensitive applications idempotent.
if isinstance(validated, str) and not isinstance(validated, SensitiveStr):
return SensitiveStr(validated)
return validated
def __repr__(self) -> str:
# Mirror the inner validator's repr so ``build_language_schema``'s

View File

@@ -52,6 +52,16 @@ _load_listeners: list[Callable[[Path], None]] = []
DocumentPath = list[str | int]
class SensitiveStr(str):
"""Marker subclass for validated strings that should be masked in
user-visible YAML output. ``cv.sensitive`` wraps validated values in this
type so ``dump()`` can render them with ANSI conceal codes without
needing a post-process regex.
"""
__slots__ = ()
@contextmanager
def track_yaml_loads() -> Generator[list[Path]]:
"""Context manager that records every file loaded by the YAML loader.
@@ -808,11 +818,18 @@ def dump(dict_, show_secrets=False, sort_keys=False):
if show_secrets:
_SECRET_VALUES.clear()
_SECRET_CACHE.clear()
# Per-call subclass so the redaction flag lives on the dumper class itself,
# not in module-level mutable state. Cheap (one type object per call),
# encapsulated, and thread-safe by construction.
class _Dumper(ESPHomeDumper):
_redact_sensitive = not show_secrets
return yaml.dump(
dict_,
default_flow_style=False,
allow_unicode=True,
Dumper=ESPHomeDumper,
Dumper=_Dumper,
sort_keys=sort_keys,
)
@@ -958,6 +975,10 @@ def format_path(path: DocumentPath, current_obj: Any) -> str:
class ESPHomeDumper(yaml.SafeDumper):
# Default for the base class; per-call subclass in ``dump()`` overrides.
# When True, ``represent_sensitive`` wraps values in ANSI conceal codes.
_redact_sensitive: bool = False
def represent_mapping(self, tag, mapping, flow_style=None):
value = []
node = yaml.MappingNode(tag, value, flow_style=flow_style)
@@ -992,6 +1013,23 @@ class ESPHomeDumper(yaml.SafeDumper):
return self.represent_secret(value)
return self.represent_scalar(tag="tag:yaml.org,2002:str", value=str(value))
def represent_sensitive(self, value):
# ``!secret`` references win: keep the original representation so the
# dumped YAML round-trips back to ``!secret name`` instead of leaking
# the resolved value.
if is_secret(value):
return self.represent_secret(value)
if self._redact_sensitive:
# Emit the conceal sequence as literal ``\033`` text (not actual
# ESC bytes) so the dump matches the previous regex-based output
# and downstream consumers like device-builder, which match
# ``\033[8m...\033[28m`` against the rendered text, keep working.
return self.represent_scalar(
tag="tag:yaml.org,2002:str",
value=f"\\033[8m{value}\\033[28m",
)
return self.represent_scalar(tag="tag:yaml.org,2002:str", value=str(value))
# pylint: disable=arguments-renamed
def represent_bool(self, value):
return self.represent_scalar(
@@ -1063,6 +1101,9 @@ ESPHomeDumper.add_multi_representer(
)
ESPHomeDumper.add_multi_representer(bool, ESPHomeDumper.represent_bool)
ESPHomeDumper.add_multi_representer(str, ESPHomeDumper.represent_stringify)
# Registered after ``str`` so ``add_multi_representer``'s MRO lookup prefers
# the more specific ``SensitiveStr`` representer over the bare-``str`` one.
ESPHomeDumper.add_multi_representer(SensitiveStr, ESPHomeDumper.represent_sensitive)
ESPHomeDumper.add_multi_representer(int, ESPHomeDumper.represent_int)
ESPHomeDumper.add_multi_representer(float, ESPHomeDumper.represent_float)
ESPHomeDumper.add_multi_representer(_BaseAddress, ESPHomeDumper.represent_stringify)

View File

@@ -145,6 +145,46 @@ def test_sensitive__custom_inner_delegates_validation() -> None:
validator(123)
def test_sensitive__wraps_string_result_in_sensitive_str() -> None:
from esphome.yaml_util import SensitiveStr
validator = config_validation.sensitive()
result = validator("hunter2")
assert isinstance(result, SensitiveStr)
assert isinstance(result, str)
assert result == "hunter2"
def test_sensitive__does_not_double_tag_already_sensitive() -> None:
from esphome.yaml_util import SensitiveStr
# If the inner validator already returns a SensitiveStr (e.g., nested
# cv.sensitive wrappers), re-tagging is a no-op rather than a new
# SensitiveStr around the same value.
pre_tagged = SensitiveStr("hunter2")
def inner(_value):
return pre_tagged
validator = config_validation.sensitive(inner)
result = validator("anything")
assert result is pre_tagged
def test_sensitive__non_string_result_passes_through() -> None:
# If an inner validator returns something other than a string (e.g., a
# Lambda template), the sensitive wrapper must not coerce it.
sentinel = object()
def inner(_value):
return sentinel
validator = config_validation.sensitive(inner)
assert validator("anything") is sentinel
def test_sensitive__is_detectable_via_isinstance() -> None:
validator = config_validation.sensitive()

View File

@@ -15,6 +15,7 @@ from esphome.yaml_util import (
DiscoveredYamlFiles,
ESPHomeDataBase,
ESPLiteralValue,
SensitiveStr,
discover_user_yaml_files,
force_load_include_files,
format_path,
@@ -1340,3 +1341,57 @@ def test_frontmatter_included_file_stored(tmp_path: Path) -> None:
assert main.resolve() not in core.CORE.frontmatter
# Included file's frontmatter is captured
assert core.CORE.frontmatter[inc.resolve()]["child_meta"] == "hello"
def test_sensitive_str__is_a_str_subclass() -> None:
value = SensitiveStr("hunter2")
assert isinstance(value, str)
assert value == "hunter2"
def test_dump__redacts_sensitive_str_by_default() -> None:
out = yaml_util.dump({"password": SensitiveStr("hunter2")})
assert "\\033[8mhunter2\\033[28m" in out
assert "hunter2" not in out.replace(
"\\033[8mhunter2\\033[28m", ""
) # the raw value is only present inside the wrap
def test_dump__show_secrets_emits_sensitive_str_raw() -> None:
out = yaml_util.dump({"password": SensitiveStr("hunter2")}, show_secrets=True)
assert "hunter2" in out
assert "\\033[8m" not in out
assert "\\033[28m" not in out
def test_dump__plain_str_is_not_redacted() -> None:
out = yaml_util.dump({"hostname": "myserver"})
assert "myserver" in out
assert "\\033[8m" not in out
def test_dump__secret_reference_wins_over_redaction() -> None:
# If the value also has an entry in _SECRET_VALUES (i.e., it was loaded
# via !secret), the dump should render it as !secret <name>, not as a
# redacted scalar. SensitiveStr layered on top must not change that.
value = SensitiveStr("hunter2")
yaml_util._SECRET_VALUES[str(value)] = "my_secret_name"
try:
out = yaml_util.dump({"password": value})
assert "!secret" in out
assert "my_secret_name" in out
assert "\\033[8m" not in out
finally:
yaml_util._SECRET_VALUES.clear()
def test_dump__redaction_flag_does_not_leak_between_calls() -> None:
# Per-call _Dumper subclass means show_secrets in one call doesn't
# affect another. Run them in both orders to catch any leakage.
redacted = yaml_util.dump({"password": SensitiveStr("hunter2")})
raw = yaml_util.dump({"password": SensitiveStr("hunter2")}, show_secrets=True)
redacted_again = yaml_util.dump({"password": SensitiveStr("hunter2")})
assert "\\033[8m" in redacted
assert "\\033[8m" not in raw
assert "\\033[8m" in redacted_again