feat(control-plane): role-scoped signed tokens so the gateway can't drive operator routes
test / integration-docker (pull_request) Successful in 17s
tracker-policy-pr / check-pr (pull_request) Successful in 16s
test / unit (pull_request) Successful in 39s
lint / lint (push) Successful in 56s
test / integration-firecracker (pull_request) Successful in 3m21s
test / coverage (pull_request) Successful in 19s
test / publish-infra (pull_request) Has been skipped

Review follow-up on #469: the data plane held the same control-plane secret
that authorizes every route, so a compromised egress/git-gate could queue a
supervise proposal AND approve it (or rewrite policy, read injected tokens) —
the (source_ip, identity_token) checks attribute the *bottle*, not the caller.

Replace the single shared bearer secret with role-scoped, HMAC-signed tokens
(compact HS256 JWTs, stdlib-only — no new dependency):

  * new `control_auth` mints/verifies `{role}` tokens; roles are `gateway`
    (data plane) and `cli` (host operator/launcher).
  * the orchestrator holds only the signing *key* and verifies; `dispatch`
    gates each route by role — `gateway` reaches /resolve + /supervise/
    {propose,poll}, everything else is `cli`-only (401 unauthenticated,
    403 wrong role).
  * the gateway is handed a pre-minted `gateway` token it cannot rewrite into
    `cli`; the host CLI mints its own `cli` token from the host key.
  * `gateway_init` scopes the signing key to the orchestrator process and the
    gateway token to the data-plane daemons, so even in the combined infra
    container a compromised data-plane daemon never sees the key.

Launchers (docker gateway + infra, macOS infra) inject the minted token(s);
Firecracker stays open behind its nft boundary. Open mode (no key) still grants
full `cli` access — the fail-visible fallback for tests.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
2026-07-24 05:18:39 +00:00
parent 72fdb1d14b
commit 1db2a9eb67
15 changed files with 485 additions and 111 deletions
+17 -7
View File
@@ -48,7 +48,9 @@ from ...orchestrator.lifecycle import (
OrchestratorStartError,
source_hash,
)
from ...control_auth import ROLE_GATEWAY, mint
from ...paths import (
CONTROL_AUTH_JWT_ENV,
CONTROL_PLANE_TOKEN_ENV,
HOST_DB_FILENAME,
host_control_plane_token,
@@ -237,18 +239,26 @@ class MacosInfraService:
# Baked onto the container so `_source_current` can detect a real
# control-plane code change and recreate.
"--env", f"BOT_BOTTLE_SOURCE_HASH={current_hash}",
# The control-plane secret, for BOTH the control plane (to require
# it) and the gateway's PolicyResolver (to present it) — they share
# this one container. Bare `--env NAME` inherits the value from the
# run process below, so the secret never lands on argv or in
# `container inspect`'s command line. The agent runs in a SEPARATE
# container that is never given this var, which is the whole point.
# The control-plane signing key (control plane: verifies tokens) and
# the pre-minted `gateway` JWT (the gateway's PolicyResolver: presents
# it) — they share this one container, and gateway_init scopes each to
# its process so a compromised data-plane daemon never sees the key
# (issue #469 review). Bare `--env NAME` inherits the value from the
# run process below, so neither lands on argv or in `container
# inspect`'s command line. The agent runs in a SEPARATE container that
# is never given these vars, which is the whole point.
"--env", CONTROL_PLANE_TOKEN_ENV,
"--env", CONTROL_AUTH_JWT_ENV,
"--entrypoint", "sh",
self.image,
"-c", _init_script(self.port),
]
run_env = {**os.environ, CONTROL_PLANE_TOKEN_ENV: host_control_plane_token()}
_signing_key = host_control_plane_token()
run_env = {
**os.environ,
CONTROL_PLANE_TOKEN_ENV: _signing_key,
CONTROL_AUTH_JWT_ENV: mint(ROLE_GATEWAY, _signing_key),
}
result = container_mod.run_container_argv(argv, env=run_env)
if result.returncode != 0:
raise OrchestratorStartError(