stage-topo-a topology-export: sanitized public docker snapshot module + tests

src/psyc/lines/topology_export.py

@@ -0,0 +1,228 @@
+"""Topology export — sanitized public docker snapshot.
+
+The cockpit's `docker_view.topology()` returns a rich daemon view useful to
+the local operator: container env vars, volume mounts, internal IPs, labels,
+gateways. None of that may leave the node. This module wraps `docker_view`
+with a strict whitelist: only container names, images, states, network names
+and high-level driver/health metadata are exposed. Anything not listed in
+the Pydantic schemas below is dropped before serialization.
+
+Used by `/federation/topology` so peer admin pages can render every node's
+container topology side-by-side with their own.
+"""
+
+from __future__ import annotations
+
+import re
+from datetime import datetime, timezone
+from typing import Any, Dict, List, Optional
+
+from pydantic import BaseModel, Field
+
+from psyc import log
+from psyc.cockpit import docker_view
+from psyc.lines import federation
+
+
+_log = log.get(__name__)
+
+
+# Caps keep the response bounded — a runaway node with thousands of
+# containers shouldn't blow up the peer's panel.
+MAX_CONTAINERS = 200
+MAX_NETWORKS = 50
+
+
+# ---------- data model --------------------------------------------------
+
+class TopologyContainer(BaseModel):
+    """One container — sanitized.
+
+    Strict whitelist: name, short_id, image (tag-only), state, health,
+    network names, compose service label, started_at. No env vars, no
+    volumes, no IPs, no MACs, no port mappings, no full labels dict.
+    """
+    name: str
+    short_id: str
+    image: str
+    state: str
+    health: str
+    networks: List[str] = Field(default_factory=list)
+    service: Optional[str] = None
+    started_at: Optional[str] = None
+
+
+class TopologyNetwork(BaseModel):
+    """One docker network — sanitized.
+
+    Whitelist: name, driver, internal flag, container_count. No subnet,
+    no gateway, no labels, no attached-container details (those are
+    surfaced via the container.networks list).
+    """
+    name: str
+    driver: str
+    internal: bool
+    container_count: int
+
+
+class TopologyExport(BaseModel):
+    """Whole-node container snapshot, public-safe."""
+    node_fingerprint: str
+    generated_at: str
+    host_name: str
+    container_count: int
+    network_count: int
+    containers: List[TopologyContainer] = Field(default_factory=list)
+    networks: List[TopologyNetwork] = Field(default_factory=list)
+
+
+# ---------- sanitizers --------------------------------------------------
+
+_BASIC_AUTH_RE = re.compile(r"^[^/@]+@")
+
+
+def _filter_image_name(s: str) -> str:
+    """Strip credentials from an image reference and drop digests.
+
+    Docker accepts `user:pass@registry/image:tag` for registries with HTTP
+    basic auth — we strip everything up to and including the `@` so leaked
+    creds never reach a peer. We also cut content-addressable digests
+    (`...@sha256:...`) to a clean tag-only form.
+
+    Returns the cleaned `repo/image:tag` string. Empty input → "".
+    """
+    if not s:
+        return ""
+    raw = str(s).strip()
+    if not raw:
+        return ""
+    # Drop digest suffix, e.g. "nginx:1.25@sha256:abcd…" → "nginx:1.25".
+    if "@sha256:" in raw:
+        raw = raw.split("@sha256:", 1)[0]
+    # Strip basic-auth prefix on the registry component.
+    # "user:pass@host/repo:tag" → "host/repo:tag" (we never want creds out).
+    if _BASIC_AUTH_RE.match(raw):
+        raw = raw.split("@", 1)[1]
+    # Cap length defensively.
+    return raw[:160]
+
+
+def _short_id(raw: Any) -> str:
+    s = str(raw or "")
+    return s[:12]
+
+
+def _parse_health(status: str) -> str:
+    """Extract a healthcheck word from the docker "Status" line if present.
+
+    docker's container-list "Status" string includes "(healthy)" or
+    "(unhealthy)" when a healthcheck is configured. We surface just that
+    one-word state and fall back to "—" otherwise — no other free-form
+    text from the daemon leaks out.
+    """
+    if not status:
+        return "—"
+    low = status.lower()
+    if "(healthy)" in low:
+        return "healthy"
+    if "(unhealthy)" in low:
+        return "unhealthy"
+    if "(starting)" in low or "(health: starting)" in low:
+        return "starting"
+    return "—"
+
+
+def _now_iso() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def _empty_export(node_fp: str) -> TopologyExport:
+    return TopologyExport(
+        node_fingerprint=node_fp,
+        generated_at=_now_iso(),
+        host_name="",
+        container_count=0,
+        network_count=0,
+        containers=[],
+        networks=[],
+    )
+
+
+# ---------- builder ------------------------------------------------------
+
+def build_export() -> TopologyExport:
+    """Sanitized snapshot of this node's docker topology.
+
+    Calls `docker_view.topology()` and re-projects every field through the
+    Pydantic whitelist above. If the proxy is unreachable (e.g. dev box
+    without docker-socket-proxy) we return an empty export rather than
+    raising — the public endpoint must never 500.
+    """
+    try:
+        node_fp = federation.node_fingerprint()
+    except Exception as exc:  # noqa: BLE001 — keep endpoint defensive
+        _log.warning("topology_export.fp.error", error=str(exc))
+        node_fp = ""
+
+    try:
+        raw = docker_view.topology()
+    except Exception as exc:  # noqa: BLE001 — docker proxy may be down
+        _log.warning("topology_export.docker.error", error=str(exc))
+        return _empty_export(node_fp)
+
+    # docker_view.topology() returns a dict with `containers`, `networks`,
+    # `host`, `error` fields. We treat any non-None error as "empty export"
+    # rather than partially leaking through whatever did succeed.
+    if raw.get("error"):
+        return _empty_export(node_fp)
+
+    raw_host = raw.get("host") or {}
+    host_name_raw = str(raw_host.get("name") or "")
+    # Truncate the docker host id — it can be the actual machine hostname.
+    # Keep it short, no domain. Defensive even though docker host names are
+    # generally low-sensitivity.
+    host_name = host_name_raw[:24]
+
+    raw_containers = raw.get("containers") or []
+    raw_networks = raw.get("networks") or []
+
+    containers: List[TopologyContainer] = []
+    for c in raw_containers[:MAX_CONTAINERS]:
+        nets_raw = c.get("networks") or []
+        net_names: List[str] = []
+        for nd in nets_raw:
+            nm = nd.get("name") if isinstance(nd, dict) else None
+            if nm:
+                net_names.append(str(nm)[:64])
+        containers.append(TopologyContainer(
+            name=str(c.get("name") or "?")[:64],
+            short_id=_short_id(c.get("id")),
+            image=_filter_image_name(c.get("image") or ""),
+            state=str(c.get("state") or "")[:24],
+            health=_parse_health(str(c.get("status") or "")),
+            networks=net_names[:12],
+            # docker_view doesn't currently surface the compose service label
+            # or started_at; leave them None until that lands.
+            service=None,
+            started_at=None,
+        ))
+
+    networks: List[TopologyNetwork] = []
+    for n in raw_networks[:MAX_NETWORKS]:
+        attached = n.get("containers") or []
+        networks.append(TopologyNetwork(
+            name=str(n.get("name") or "")[:64],
+            driver=str(n.get("driver") or "")[:24],
+            internal=bool(n.get("internal")),
+            container_count=len(attached),
+        ))
+
+    return TopologyExport(
+        node_fingerprint=node_fp,
+        generated_at=_now_iso(),
+        host_name=host_name,
+        container_count=len(containers),
+        network_count=len(networks),
+        containers=containers,
+        networks=networks,
+    )

tests/test_topology_export.py

@@ -0,0 +1,245 @@
+"""Topology export — whitelist sanitization unit tests.
+
+The big invariant: nothing from docker_view.topology() escapes that isn't
+in the Pydantic schema. We assert via model_fields introspection AND via a
+JSON-dump scan over a fixture that contains every dangerous field.
+
+The /federation/topology endpoint contract lives in the sibling tests added
+alongside it; this module covers the builder + the sanitizer in isolation.
+"""
+
+from __future__ import annotations
+
+import json
+from typing import Any, Dict
+from unittest.mock import patch
+
+import pytest
+from sqlalchemy import create_engine
+
+from psyc import db
+from psyc.cockpit import docker_view
+from psyc.lines import federation, topology_export
+from psyc.lines.topology_export import (
+    TopologyContainer,
+    TopologyExport,
+    TopologyNetwork,
+    _filter_image_name,
+    build_export,
+)
+
+
+# ---------- fixtures ----------------------------------------------------
+
+@pytest.fixture
+def fresh_db(tmp_path, monkeypatch):
+    test_db = tmp_path / "test.db"
+    eng = create_engine(f"sqlite:///{test_db}", future=True)
+    db._metadata.create_all(eng, checkfirst=True)
+    monkeypatch.setattr(db, "_engine", eng)
+    monkeypatch.setattr(db, "DB_PATH", test_db)
+    yield test_db
+
+
+@pytest.fixture
+def fed_dir(tmp_path, monkeypatch):
+    d = tmp_path / "federation"
+    monkeypatch.setattr(federation, "FED_DIR", d)
+    monkeypatch.setattr(federation, "PRIVATE_KEY_PATH", d / "node.key")
+    monkeypatch.setattr(federation, "PUBLIC_KEY_PATH", d / "node.pub")
+    yield d
+
+
+# ---------- fixture data: hostile docker_view output --------------------
+
+# This payload has every leaky field docker_view *could* surface, plus
+# nested env-style data — used to prove the export is whitelist-only.
+_LEAKY_TOPOLOGY: Dict[str, Any] = {
+    "containers": [
+        {
+            "id": "abcdef1234567890ffff",
+            "name": "psyc-cockpit-1",
+            "image": "registry.example/psyc:1.2",
+            "state": "running",
+            "status": "Up 5 minutes (healthy)",
+            "networks": [
+                {"name": "backend", "ip": "172.20.0.5", "gateway": "172.20.0.1", "mac": "02:42:ac:14:00:05"},
+                {"name": "frontend", "ip": "172.21.0.7", "gateway": "172.21.0.1", "mac": "02:42:ac:15:00:07"},
+            ],
+            "ports": ["0.0.0.0:8767->8767/tcp"],
+            "published_ports": ["8767/tcp"],
+            # These are NOT current docker_view fields but defend in depth —
+            # if a future docker_view change adds them, sanitizer drops them.
+            "env": ["SECRET_TOKEN=abc123", "DB_PASSWORD=hunter2"],
+            "mounts": ["/var/run/docker.sock", "/etc/secrets:/secrets"],
+            "labels": {"com.docker.compose.project": "psyc", "secret_label": "shh"},
+        },
+        {
+            "id": "fedcba0987654321",
+            "name": "some-stopped",
+            "image": "alpine",
+            "state": "exited",
+            "status": "Exited (0) 2 hours ago",
+            "networks": [],
+            "ports": [],
+            "published_ports": [],
+        },
+    ],
+    "networks": [
+        {
+            "id": "n1", "name": "backend", "driver": "bridge", "scope": "local",
+            "internal": False, "subnet": "172.20.0.0/16", "gateway": "172.20.0.1",
+            "containers": [
+                {"id": "abcdef123456", "name": "psyc-cockpit-1", "ip": "172.20.0.5", "mac": "02:42:ac:14:00:05"},
+            ],
+        },
+        {
+            "id": "n2", "name": "internal-only", "driver": "bridge", "scope": "local",
+            "internal": True, "subnet": "10.99.0.0/16", "gateway": "10.99.0.1",
+            "containers": [],
+        },
+    ],
+    "host": {"name": "docker-host-secret-internal.example.com", "os": "linux", "ncpu": 8},
+    "error": None,
+    "proxy": "http://docker-socket-proxy:2375",
+}
+
+
+# Sensitive strings that MUST NOT appear anywhere in the export JSON.
+_FORBIDDEN_STRINGS = (
+    "SECRET_TOKEN", "DB_PASSWORD", "hunter2", "abc123",
+    "/var/run/docker.sock", "/etc/secrets",
+    "secret_label", "shh",
+    "172.20.0.5", "172.21.0.7",  # IPs
+    "02:42:ac",  # MAC prefix
+    "172.20.0.1",  # gateway
+    "172.20.0.0/16", "10.99.0.0/16",  # subnets
+    "0.0.0.0:8767",  # port mapping
+    "internal.example.com",  # full host
+)
+
+
+# ---------- model field introspection -----------------------------------
+
+def test_container_model_has_no_dangerous_fields():
+    fields = set(TopologyContainer.model_fields.keys())
+    # whitelist — must match the design contract exactly
+    assert fields == {
+        "name", "short_id", "image", "state", "health",
+        "networks", "service", "started_at",
+    }
+    # explicit deny-list, double-belt
+    for forbidden in ("env", "environment", "mounts", "volumes",
+                      "labels", "ip", "ip_address", "ipaddress",
+                      "ports", "published_ports", "mac", "gateway"):
+        assert forbidden not in fields, f"{forbidden} must not be a field"
+
+
+def test_network_model_has_no_dangerous_fields():
+    fields = set(TopologyNetwork.model_fields.keys())
+    assert fields == {"name", "driver", "internal", "container_count"}
+    for forbidden in ("subnet", "gateway", "labels", "ipam",
+                      "containers", "scope", "id"):
+        assert forbidden not in fields, f"{forbidden} must not be a field"
+
+
+def test_export_model_top_level_fields():
+    fields = set(TopologyExport.model_fields.keys())
+    assert fields == {
+        "node_fingerprint", "generated_at", "host_name",
+        "container_count", "network_count", "containers", "networks",
+    }
+
+
+# ---------- image-name filter -------------------------------------------
+
+def test_filter_image_strips_basic_auth_prefix():
+    # user:pass@host/repo:tag → host/repo:tag (creds gone)
+    assert _filter_image_name("user:pass@host/repo:tag") == "host/repo:tag"
+
+
+def test_filter_image_drops_digest_suffix():
+    assert _filter_image_name(
+        "nginx:1.25@sha256:abcdef0123"
+    ) == "nginx:1.25"
+
+
+def test_filter_image_passes_clean_refs_untouched():
+    assert _filter_image_name("psyc:latest") == "psyc:latest"
+    assert _filter_image_name(
+        "ghcr.io/example/psyc:v0.3.1"
+    ) == "ghcr.io/example/psyc:v0.3.1"
+
+
+def test_filter_image_handles_empty():
+    assert _filter_image_name("") == ""
+    assert _filter_image_name(None) == ""  # type: ignore[arg-type]
+
+
+# ---------- build_export contract ---------------------------------------
+
+def test_build_export_returns_empty_when_docker_view_raises(fresh_db, fed_dir, monkeypatch):
+    def boom():
+        raise docker_view.DockerProxyError("connection refused")
+    monkeypatch.setattr(docker_view, "topology", boom)
+    out = build_export()
+    assert isinstance(out, TopologyExport)
+    assert out.container_count == 0
+    assert out.containers == []
+    assert out.networks == []
+    # fingerprint is still real (federation key was generated)
+    assert len(out.node_fingerprint) == 32
+
+
+def test_build_export_returns_empty_when_docker_view_reports_error(fresh_db, fed_dir, monkeypatch):
+    monkeypatch.setattr(docker_view, "topology", lambda: {
+        "containers": [], "networks": [], "host": {"name": "x"},
+        "error": "containers: refused", "proxy": "x",
+    })
+    out = build_export()
+    assert out.container_count == 0
+    assert out.containers == []
+
+
+def test_build_export_sanitizes_every_field(fresh_db, fed_dir, monkeypatch):
+    monkeypatch.setattr(docker_view, "topology", lambda: _LEAKY_TOPOLOGY)
+    out = build_export()
+    # Containers came through, but as TopologyContainer (no leaky attrs).
+    assert out.container_count == 2
+    by_name = {c.name: c for c in out.containers}
+    cp = by_name["psyc-cockpit-1"]
+    assert cp.short_id == "abcdef123456"
+    assert cp.image == "registry.example/psyc:1.2"
+    assert cp.state == "running"
+    assert cp.health == "healthy"
+    assert cp.networks == ["backend", "frontend"]
+    assert cp.service is None
+    # Networks came through, sanitized.
+    assert out.network_count == 2
+    by_net = {n.name: n for n in out.networks}
+    assert by_net["backend"].driver == "bridge"
+    assert by_net["backend"].internal is False
+    assert by_net["backend"].container_count == 1
+    assert by_net["internal-only"].internal is True
+
+
+def test_export_json_contains_no_dangerous_strings(fresh_db, fed_dir, monkeypatch):
+    """Strict no-leak: serialize and grep for everything sensitive."""
+    monkeypatch.setattr(docker_view, "topology", lambda: _LEAKY_TOPOLOGY)
+    out = build_export()
+    blob = json.dumps(out.model_dump(mode="json"))
+    for forbidden in _FORBIDDEN_STRINGS:
+        assert forbidden not in blob, f"leak: {forbidden!r} appeared in export JSON"
+
+
+def test_build_export_caps_at_max_containers(fresh_db, fed_dir, monkeypatch):
+    fake = {
+        "containers": [
+            {"id": f"id{i:04d}", "name": f"c{i}", "image": "x", "state": "running", "status": "Up", "networks": []}
+            for i in range(topology_export.MAX_CONTAINERS + 50)
+        ],
+        "networks": [], "host": {"name": "h"}, "error": None, "proxy": "",
+    }
+    monkeypatch.setattr(docker_view, "topology", lambda: fake)
+    out = build_export()
+    assert out.container_count == topology_export.MAX_CONTAINERS