secrets-engine/src/secrets_engine/approval_consume.py

363 lines
14 KiB
Python
Raw Normal View History

"""PEP consume-before-side-effect client (GH-DEC-2026-003).
The PEP that is about to cause a protected OpenBao write MUST obtain a
successful approval-engine CAS consume first. Holding a claim or an ALLOW is
not authority to act. Conflict, unavailability, or a missing binding means
do not call OpenBao.
This module does not render an authorization decision. The consume response
is mutation evidence, never a permission.
"""
from __future__ import annotations
import json
import re
from dataclasses import dataclass
from pathlib import Path
from typing import Any, Callable
from urllib.error import HTTPError, URLError
from urllib.request import Request, urlopen
feat: implement the PIP claim + validate authorization join resolve_consume_binding was a `return None` stub, so protocol step 1 of docs/approval-consumption.md (GET /v1/approvals/{id}/claim) and the validation join never existed. validate_action_authorization had no caller in src/ at all - it was reachable only from tests. Production fail-closed was correct, but for an undocumented second reason, and WP-0007-T04's "what remains is not local engine work" was wrong. The join now reproduces the exact CheckRequest via build_action_request, fetches the durable ActionAuthorization, and validates request binding, digest, validity, authority, policy pin, and distinct-approver threshold before offering a consume binding. _require_lane_approval threads the exact field set for provision/rotate/verify/exec so the digest covers the real proposed action. Deliberate choices: - The approval-engine object id is never inferred from a State Hub decision UUID; flex-auth stated GET /decisions/{uuid} is not the durable object. - No default policy pin. flex-auth stated secrets-engine.lifecycle/v1 is example vocabulary, not a published package. - A half-configured join raises rather than returning None, so a partial deployment cannot be mistaken for an unconfigured one. Behavior is unchanged today: every new input is absent by default, so production still fails closed and plan/--dry-run still work. 234 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M65ovP3eiiPHubibvWs9mD Assistant: claude-code Assistant-Model: opus Assistant-Process: 393550@bnt-lap001 Assistant-Session: 4bb359f9-1f12-4410-9e76-079cf23c82e4
2026-09-06 01:01:55 +02:00
from secrets_engine.authorization import (
build_action_request,
request_digest,
validate_action_authorization,
)
from secrets_engine.errors import DecisionError
from secrets_engine.openbao import read_strict_token_file
from secrets_engine.pep_stance import demo_exception_enabled
DIGEST_RE = re.compile(r"^sha256:[0-9a-f]{64}$")
_MAX_BODY = 256 * 1024
@dataclass(frozen=True)
class ConsumeBinding:
"""Inputs the PEP presents to approval-engine consume."""
approval_id: str
request_digest: str
decision_id: str = ""
@dataclass(frozen=True)
class ConsumedApproval:
"""Non-secret confirmation that consume succeeded for this request."""
approval_id: str
request_digest: str
decision_id: str = ""
idempotent: bool = False
consumed_at: str = ""
def as_evidence(self) -> dict[str, object]:
payload: dict[str, object] = {
"approval_consumed": True,
"approval_id": self.approval_id,
"request_digest": self.request_digest,
"approval_consume_idempotent": self.idempotent,
}
if self.decision_id:
payload["decision_id"] = self.decision_id
if self.consumed_at:
payload["approval_consumed_at"] = self.consumed_at
return payload
feat: implement the PIP claim + validate authorization join resolve_consume_binding was a `return None` stub, so protocol step 1 of docs/approval-consumption.md (GET /v1/approvals/{id}/claim) and the validation join never existed. validate_action_authorization had no caller in src/ at all - it was reachable only from tests. Production fail-closed was correct, but for an undocumented second reason, and WP-0007-T04's "what remains is not local engine work" was wrong. The join now reproduces the exact CheckRequest via build_action_request, fetches the durable ActionAuthorization, and validates request binding, digest, validity, authority, policy pin, and distinct-approver threshold before offering a consume binding. _require_lane_approval threads the exact field set for provision/rotate/verify/exec so the digest covers the real proposed action. Deliberate choices: - The approval-engine object id is never inferred from a State Hub decision UUID; flex-auth stated GET /decisions/{uuid} is not the durable object. - No default policy pin. flex-auth stated secrets-engine.lifecycle/v1 is example vocabulary, not a published package. - A half-configured join raises rather than returning None, so a partial deployment cannot be mistaken for an unconfigured one. Behavior is unchanged today: every new input is absent by default, so production still fails closed and plan/--dry-run still work. 234 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M65ovP3eiiPHubibvWs9mD Assistant: claude-code Assistant-Model: opus Assistant-Process: 393550@bnt-lap001 Assistant-Session: 4bb359f9-1f12-4410-9e76-079cf23c82e4
2026-09-06 01:01:55 +02:00
def _authorization_id(entry: Any, decision: Any) -> str:
"""Non-secret approval-engine object id for this lane, or "" if unbound.
flex-auth: ``ActionAuthorization.id`` is the approval-engine object UUID.
It is never the State Hub decision UUID, so it is not inferred from one.
"""
approval = getattr(entry, "approval", None) or {}
if isinstance(approval, dict):
declared = str(approval.get("authorization_id", "") or "").strip()
if declared:
return declared
for attr in ("authorization_id", "action_authorization_id"):
served = str(getattr(decision, attr, "") or "").strip()
if served:
return served
return ""
def _request_purpose(entry: Any) -> str:
"""Declared purpose for the request context. Never invented at call time."""
approval = getattr(entry, "approval", None) or {}
if isinstance(approval, dict):
declared = str(approval.get("purpose", "") or "").strip()
if declared:
return declared
for consumer in getattr(entry, "consumers", None) or []:
if isinstance(consumer, dict):
declared = str(consumer.get("purpose", "") or "").strip()
if declared:
return declared
return ""
def fetch_action_authorization(
*,
base_url: str,
token_file: Path,
authorization_id: str,
timeout_seconds: float = 3,
opener: Callable[..., Any] = urlopen,
) -> dict[str, Any]:
"""GET /v1/approvals/{id}/claim (PIP). Any non-200 fails closed."""
if not base_url or not base_url.startswith(("http://", "https://")):
raise DecisionError("approval-engine claim URL is missing or invalid")
ident = authorization_id.strip()
if not ident or "/" in ident or any(ch.isspace() for ch in ident):
raise DecisionError("approval claim requires a concrete authorization id")
token = read_strict_token_file(Path(token_file), purpose="approval claim credential")
request = Request(
base_url.rstrip("/") + f"/v1/approvals/{ident}/claim",
method="GET",
)
request.add_header("Authorization", f"Bearer {token}")
request.add_header("Accept", "application/json")
try:
with opener(request, timeout=timeout_seconds) as response:
if getattr(response, "status", 200) != 200:
raise DecisionError("approval claim did not return the authorization")
payload = json.loads(response.read(_MAX_BODY).decode("utf-8"))
except HTTPError as e:
raise DecisionError(f"approval claim refused: {_status_message(e.code)}") from e
except URLError as e:
raise DecisionError("approval-engine is unreachable for claim") from e
except json.JSONDecodeError as e:
raise DecisionError("approval claim returned a non-JSON body") from e
if not isinstance(payload, dict):
raise DecisionError("approval claim returned a non-object body")
return payload
def resolve_consume_binding(
feat: implement the PIP claim + validate authorization join resolve_consume_binding was a `return None` stub, so protocol step 1 of docs/approval-consumption.md (GET /v1/approvals/{id}/claim) and the validation join never existed. validate_action_authorization had no caller in src/ at all - it was reachable only from tests. Production fail-closed was correct, but for an undocumented second reason, and WP-0007-T04's "what remains is not local engine work" was wrong. The join now reproduces the exact CheckRequest via build_action_request, fetches the durable ActionAuthorization, and validates request binding, digest, validity, authority, policy pin, and distinct-approver threshold before offering a consume binding. _require_lane_approval threads the exact field set for provision/rotate/verify/exec so the digest covers the real proposed action. Deliberate choices: - The approval-engine object id is never inferred from a State Hub decision UUID; flex-auth stated GET /decisions/{uuid} is not the durable object. - No default policy pin. flex-auth stated secrets-engine.lifecycle/v1 is example vocabulary, not a published package. - A half-configured join raises rather than returning None, so a partial deployment cannot be mistaken for an unconfigured one. Behavior is unchanged today: every new input is absent by default, so production still fails closed and plan/--dry-run still work. 234 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M65ovP3eiiPHubibvWs9mD Assistant: claude-code Assistant-Model: opus Assistant-Process: 393550@bnt-lap001 Assistant-Session: 4bb359f9-1f12-4410-9e76-079cf23c82e4
2026-09-06 01:01:55 +02:00
cfg: Any,
entry: Any,
action: str,
decision: Any,
*,
fields: tuple[str, ...] = (),
policy_targets: tuple[str, ...] = (),
auth_targets: tuple[str, ...] = (),
opener: Callable[..., Any] | None = None,
) -> ConsumeBinding | None:
feat: implement the PIP claim + validate authorization join resolve_consume_binding was a `return None` stub, so protocol step 1 of docs/approval-consumption.md (GET /v1/approvals/{id}/claim) and the validation join never existed. validate_action_authorization had no caller in src/ at all - it was reachable only from tests. Production fail-closed was correct, but for an undocumented second reason, and WP-0007-T04's "what remains is not local engine work" was wrong. The join now reproduces the exact CheckRequest via build_action_request, fetches the durable ActionAuthorization, and validates request binding, digest, validity, authority, policy pin, and distinct-approver threshold before offering a consume binding. _require_lane_approval threads the exact field set for provision/rotate/verify/exec so the digest covers the real proposed action. Deliberate choices: - The approval-engine object id is never inferred from a State Hub decision UUID; flex-auth stated GET /decisions/{uuid} is not the durable object. - No default policy pin. flex-auth stated secrets-engine.lifecycle/v1 is example vocabulary, not a published package. - A half-configured join raises rather than returning None, so a partial deployment cannot be mistaken for an unconfigured one. Behavior is unchanged today: every new input is absent by default, so production still fails closed and plan/--dry-run still work. 234 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M65ovP3eiiPHubibvWs9mD Assistant: claude-code Assistant-Model: opus Assistant-Process: 393550@bnt-lap001 Assistant-Session: 4bb359f9-1f12-4410-9e76-079cf23c82e4
2026-09-06 01:01:55 +02:00
"""Join the proposed action to a served ActionAuthorization (PIP + validate).
feat: implement the PIP claim + validate authorization join resolve_consume_binding was a `return None` stub, so protocol step 1 of docs/approval-consumption.md (GET /v1/approvals/{id}/claim) and the validation join never existed. validate_action_authorization had no caller in src/ at all - it was reachable only from tests. Production fail-closed was correct, but for an undocumented second reason, and WP-0007-T04's "what remains is not local engine work" was wrong. The join now reproduces the exact CheckRequest via build_action_request, fetches the durable ActionAuthorization, and validates request binding, digest, validity, authority, policy pin, and distinct-approver threshold before offering a consume binding. _require_lane_approval threads the exact field set for provision/rotate/verify/exec so the digest covers the real proposed action. Deliberate choices: - The approval-engine object id is never inferred from a State Hub decision UUID; flex-auth stated GET /decisions/{uuid} is not the durable object. - No default policy pin. flex-auth stated secrets-engine.lifecycle/v1 is example vocabulary, not a published package. - A half-configured join raises rather than returning None, so a partial deployment cannot be mistaken for an unconfigured one. Behavior is unchanged today: every new input is absent by default, so production still fails closed and plan/--dry-run still work. 234 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M65ovP3eiiPHubibvWs9mD Assistant: claude-code Assistant-Model: opus Assistant-Process: 393550@bnt-lap001 Assistant-Session: 4bb359f9-1f12-4410-9e76-079cf23c82e4
2026-09-06 01:01:55 +02:00
Returns None only when this repo holds no configured serving path, which
keeps production fail-closed exactly as it was before the join existed.
Anything configured-but-wrong raises instead of degrading to None: a
half-configured PEP must not look like an unconfigured one.
"""
feat: implement the PIP claim + validate authorization join resolve_consume_binding was a `return None` stub, so protocol step 1 of docs/approval-consumption.md (GET /v1/approvals/{id}/claim) and the validation join never existed. validate_action_authorization had no caller in src/ at all - it was reachable only from tests. Production fail-closed was correct, but for an undocumented second reason, and WP-0007-T04's "what remains is not local engine work" was wrong. The join now reproduces the exact CheckRequest via build_action_request, fetches the durable ActionAuthorization, and validates request binding, digest, validity, authority, policy pin, and distinct-approver threshold before offering a consume binding. _require_lane_approval threads the exact field set for provision/rotate/verify/exec so the digest covers the real proposed action. Deliberate choices: - The approval-engine object id is never inferred from a State Hub decision UUID; flex-auth stated GET /decisions/{uuid} is not the durable object. - No default policy pin. flex-auth stated secrets-engine.lifecycle/v1 is example vocabulary, not a published package. - A half-configured join raises rather than returning None, so a partial deployment cannot be mistaken for an unconfigured one. Behavior is unchanged today: every new input is absent by default, so production still fails closed and plan/--dry-run still work. 234 tests pass. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M65ovP3eiiPHubibvWs9mD Assistant: claude-code Assistant-Model: opus Assistant-Process: 393550@bnt-lap001 Assistant-Session: 4bb359f9-1f12-4410-9e76-079cf23c82e4
2026-09-06 01:01:55 +02:00
base_url = str(getattr(cfg, "approval_url", "") or "")
token_file = getattr(cfg, "approval_token_file", None)
authorization_id = _authorization_id(entry, decision)
if not base_url or not token_file or not authorization_id:
return None
subject_id = str(getattr(cfg, "authorization_subject_id", "") or "")
subject_type = str(getattr(cfg, "authorization_subject_type", "") or "")
if not subject_id or not subject_type:
raise DecisionError(
"authorization join requires SECRETS_ENGINE_AUTHORIZATION_SUBJECT_ID "
"and _SUBJECT_TYPE; the PEP must not assert an unnamed subject"
)
package = str(getattr(cfg, "authorization_policy_package", "") or "")
version = str(getattr(cfg, "authorization_policy_version", "") or "")
if not package or not version:
raise DecisionError(
"authorization join requires an explicitly configured policy "
"package/version pin; the published example vocabulary "
"(secrets-engine.lifecycle/v1) is not a live pin"
)
purpose = _request_purpose(entry)
if not purpose:
raise DecisionError(
"authorization join requires a declared approval/consumer purpose"
)
envelope = fetch_action_authorization(
base_url=base_url,
token_file=Path(token_file),
authorization_id=authorization_id,
opener=opener or urlopen,
)
# The Check request id is an opaque correlator chosen by the requester, so
# the PEP cannot regenerate it and adopts the served one. Every
# security-relevant field (subject, action, resource, context) is still
# compared exactly by validate_action_authorization, and the served
# binding digest is recomputed against the served request, so adopting the
# id cannot let a mismatched request validate.
served_request = envelope.get("request")
served_id = ""
if isinstance(served_request, dict):
served_id = str(served_request.get("id", "") or "")
expected_request = build_action_request(
entry,
action,
subject_id=subject_id,
subject_type=subject_type,
purpose=purpose,
fields=fields,
policy_targets=policy_targets,
auth_targets=auth_targets,
request_id=served_id,
)
validated = validate_action_authorization(
envelope,
expected_request,
accepted_policy_packages={package},
accepted_policy_versions={version},
minimum_approval_count=int(getattr(cfg, "authorization_min_approvals", 1) or 1),
)
if validated.action != action:
raise DecisionError("action authorization does not bind this action")
return ConsumeBinding(
approval_id=validated.authorization_id,
request_digest=request_digest(expected_request),
decision_id=validated.decision_id,
)
def consume_approval(
*,
base_url: str,
token_file: Path,
binding: ConsumeBinding,
timeout_seconds: float = 3,
opener: Callable[..., Any] = urlopen,
) -> ConsumedApproval:
"""POST /v1/approvals/{id}/consume. Fail closed on anything but confirmed use."""
if not base_url or not base_url.startswith(("http://", "https://")):
raise DecisionError("approval-engine consume URL is missing or invalid")
approval_id = binding.approval_id.strip()
if not approval_id or "/" in approval_id or any(ch.isspace() for ch in approval_id):
raise DecisionError("approval consume requires a concrete approval id")
if not DIGEST_RE.fullmatch(binding.request_digest):
raise DecisionError("approval consume requires the canonical request digest")
token = read_strict_token_file(Path(token_file), purpose="approval consume credential")
body: dict[str, str] = {"request_digest": binding.request_digest}
if binding.decision_id:
body["decision_id"] = binding.decision_id
encoded = json.dumps(body).encode("utf-8")
request = Request(
base_url.rstrip("/") + f"/v1/approvals/{approval_id}/consume",
data=encoded,
method="POST",
headers={
"Authorization": f"Bearer {token}",
"Content-Type": "application/json",
"Accept": "application/json",
},
)
try:
response = opener(request, timeout=timeout_seconds)
try:
status = int(response.getcode())
raw = response.read(_MAX_BODY + 1)
finally:
response.close()
except HTTPError as exc:
status = int(getattr(exc, "code", 0) or 0)
try:
exc.read(_MAX_BODY)
except Exception:
pass
raise DecisionError(_status_message(status)) from None
except (URLError, TimeoutError, OSError):
raise DecisionError(
"approval-engine unreachable; OpenBao must not be called"
) from None
if status != 200 or len(raw) > _MAX_BODY:
raise DecisionError(_status_message(status if status != 200 else 502))
try:
payload = json.loads(raw.decode("utf-8"))
except (UnicodeDecodeError, json.JSONDecodeError) as exc:
raise DecisionError("approval consume returned invalid JSON") from exc
if not isinstance(payload, dict):
raise DecisionError("approval consume returned invalid payload")
if payload.get("status") != "consumed":
raise DecisionError("approval consumption was not confirmed")
if payload.get("request_digest") != binding.request_digest:
raise DecisionError("approval consume digest does not match the request")
decision_id = payload.get("decision_id")
if decision_id is not None and (
not isinstance(decision_id, str) or not decision_id
):
raise DecisionError("approval consume returned an invalid decision id")
consumed_at = payload.get("consumed_at")
if consumed_at is not None and not isinstance(consumed_at, str):
raise DecisionError("approval consume returned an invalid consumed_at")
return ConsumedApproval(
approval_id=str(payload.get("approval_id") or approval_id),
request_digest=binding.request_digest,
decision_id=decision_id or binding.decision_id,
idempotent=bool(payload.get("idempotent")),
consumed_at=consumed_at or "",
)
def require_production_consume(
cfg: Any,
entry: Any,
*,
binding: ConsumeBinding | None,
evidence: Any = None,
opener: Callable[..., Any] | None = None,
) -> ConsumedApproval | None:
"""CAS-consume before a production OpenBao call. No-op off the prod path.
Build/test remain fail-open relative to approval-engine. The three-factor
unsafe-demo exception is not a consume path. Missing binding, URL, or
credential fail closed so a stance bypass cannot reach OpenBao.
"""
if getattr(entry, "stage", "") != "prod":
return None
if demo_exception_enabled(cfg):
return None
if binding is None:
raise DecisionError(
"production OpenBao call requires CAS consume of an approval "
"after an access-engine ALLOW; no durable consume binding is served"
)
base_url = str(getattr(cfg, "approval_url", "") or "")
token_file = getattr(cfg, "approval_token_file", None)
if not base_url:
raise DecisionError(
"production OpenBao call requires approval-engine consume; "
"SECRETS_ENGINE_APPROVAL_URL is unset"
)
if not token_file:
raise DecisionError(
"production OpenBao call requires approval-engine consume; "
"SECRETS_ENGINE_APPROVAL_TOKEN_FILE is unset"
)
consumed = consume_approval(
base_url=base_url,
token_file=Path(token_file),
binding=binding,
opener=opener or urlopen,
)
if evidence is not None and hasattr(evidence, "mark_consumed"):
evidence.mark_consumed(consumed)
return consumed
def _status_message(status: int) -> str:
if status == 409:
return "approval consume conflict; OpenBao must not be called"
if status == 404:
return "approval not found; OpenBao must not be called"
if status in {401, 403}:
return "approval consume unauthorized; OpenBao must not be called"
if status == 503:
return "approval-engine unavailable; OpenBao must not be called"
if status == 0:
return "approval-engine unreachable; OpenBao must not be called"
return "approval consume failed; OpenBao must not be called"