AUTH.4C2: relative vault path contract repair (OPTION A) — canonical relative path, absolute/traversal fail-closed, no approval_payload dual-semantics, 20 path-contract tests

This commit is contained in:
Red Queen 2026-08-27 13:23:56 +00:00
parent 2ba503df0e
commit 2e631fd5fa
4 changed files with 663 additions and 27 deletions

View file

@ -8,7 +8,8 @@ Lädt Content aus der Forgejo Source of Truth, exakt commit-bound.
Eigenschaften: Eigenschaften:
* Commit-bound: liest Content aus genau dem source_commit (kein Branch-Latest). * Commit-bound: liest Content aus genau dem source_commit (kein Branch-Latest).
* object_id -> Pfad: leitet den Repo-relativen Pfad aus dem vault_path ab * object_id -> Pfad: leitet den Repo-relativen Pfad aus dem vault_path ab
(VAULT_PREFIX + rel_path), konsistent mit C5C (_vault_path). (kanonisch RELATIV, AUTH.4C2 PATH CONTRACT REPAIR OPTION A), konsistent
mit C5C (_vault_path).
* Keine Write-Fähigkeit. Keine automatische Upstream-Integration. * Keine Write-Fähigkeit. Keine automatische Upstream-Integration.
* FAIL CLOSED bei: commit missing, object missing, hash mismatch, * FAIL CLOSED bei: commit missing, object missing, hash mismatch,
provenance mismatch, path mismatch, Forgejo unavailable. provenance mismatch, path mismatch, Forgejo unavailable.
@ -29,9 +30,6 @@ from typing import Any, Dict, Optional
DEFAULT_FORGEJO_REPO = "nexo312/trading-system-docs" DEFAULT_FORGEJO_REPO = "nexo312/trading-system-docs"
ENV_FORGEJO_REPO = "C5_FORGEJO_REPO" ENV_FORGEJO_REPO = "C5_FORGEJO_REPO"
# Vault-Pfad-Praefix (konsistent mit C5C VAULT_PREFIX)
VAULT_PREFIX = "/app/vault"
# Lokaler Clone-Pfad (falls vorhanden) — sonst HTTP-API # Lokaler Clone-Pfad (falls vorhanden) — sonst HTTP-API
DEFAULT_CLONE_PATH = "/data/forgejo-clone" DEFAULT_CLONE_PATH = "/data/forgejo-clone"
ENV_CLONE_PATH = "C5_FORGEJO_CLONE" ENV_CLONE_PATH = "C5_FORGEJO_CLONE"
@ -75,14 +73,15 @@ class ForgejoSourceLoader:
def _rel_path(self, vault_path: str) -> str: def _rel_path(self, vault_path: str) -> str:
"""Leitet den Repo-relativen Pfad aus dem vault_path ab. """Leitet den Repo-relativen Pfad aus dem vault_path ab.
vault_path = /app/vault/<rel_path> -> rel_path. KANONISCH (AUTH.4C2 PATH CONTRACT REPAIR, OPTION A):
Konsistent mit C5C _vault_path (VAULT_PREFIX + rel_path). vault_path ist bereits der RELATIVE Pfad unter dem Vault-Root
(z.B. 'tolaria/auth4c2-canary.md'). Er wird unverändert als
rel_path verwendet KEIN /app/vault/-Prefix-Stripping mehr.
Konsistent mit C5C _vault_path (rel_path unter Vault-Root).
""" """
vp = vault_path or "" vp = vault_path or ""
if vp.startswith(VAULT_PREFIX): rel = vp.lstrip("/")
rel = vp[len(VAULT_PREFIX):].lstrip("/")
else:
rel = vp.lstrip("/")
if not rel: if not rel:
raise SourceMismatchError("vault_path ergibt keinen rel_path") raise SourceMismatchError("vault_path ergibt keinen rel_path")
return rel return rel

View file

@ -98,25 +98,30 @@ def _is_nonempty_str(value: Any) -> bool:
def _normalize_path(path: str) -> str: def _normalize_path(path: str) -> str:
""" """
Normalisiert einen Vault-Pfad (keine //, ., .., trailing slash). Normalisiert einen Vault-Pfad (keine //, ., .., trailing slash).
Wiederverwendet die AUTH.3D-Pfad-Normalisierung. KANONISCH RELATIV (AUTH.4C2 PATH CONTRACT REPAIR, OPTION A):
kein führender Slash. Immer die eigene relative Normalisierung
KEIN Import von approval_payload._normalize_path, da jene (AUTH.3D)
absolute Pfade erzwingt und damit die kanonisch-relative Semantik
verletzen würde (keine Dual-Semantik).
""" """
try: parts = [p for p in path.split("/") if p not in ("", ".")]
from approval_payload import _normalize_path as _ap_normalize if ".." in parts:
return _ap_normalize(path) raise ValueError("path traversal")
except Exception: return "/".join(parts)
# Fallback (isoliert): eigene Normalisierung
parts = [p for p in path.split("/") if p not in ("", ".")]
if ".." in parts:
raise ValueError("path traversal")
return "/" + "/".join(parts)
def _validate_path(path: str) -> Optional[str]: def _validate_path(path: str) -> Optional[str]:
""" """
Pfad-Sicherheitsprüfung (Executor-seitig, Defense in depth). Pfad-Sicherheitsprüfung (Executor-seitig, Defense in depth).
Erlaubt nur Pfade unterhalb des Vault-Roots, keine Traversal, KANONISCHER JOB-VAULT-PFAD = RELATIVER Pfad unter dem Vault-Root
keine absoluten Host-Pfade, keine Unicode-Ambiguität, keine (AUTH.4C2 PATH CONTRACT REPAIR, OPTION A).
URL-Encodierung, keine Sonderzeichen.
Erlaubt NUR relative Pfade. Lehnt ab:
* absolute Pfade (führende /) Tolaria AUTH.2 lehnt absolute ab
* Traversal (.., ./)
* Backslash-Traversal
* URL/Scheme (http://, file://, C:)
* Null-Bytes, Unicode-Ambiguität, URL-Encodierung, Sonderzeichen
""" """
if not isinstance(path, str) or not path: if not isinstance(path, str) or not path:
return "path must be a non-empty string" return "path must be a non-empty string"
@ -124,9 +129,13 @@ def _validate_path(path: str) -> Optional[str]:
return "path too long" return "path too long"
if "\x00" in path: if "\x00" in path:
return "path contains null byte" return "path contains null byte"
# Kein absoluter Host-Pfad (nur /app/vault/... erlaubt) # Kanonisch RELATIV: keine führende /, kein Scheme, kein Backslash
if not path.startswith("/app/vault/"): if path.startswith("/"):
return "path must be under /app/vault/" return "path must be relative (no leading /)"
if "\\" in path:
return "path contains backslash"
if ":" in path:
return "path contains scheme/colon"
# Zeichensatz-Whitelist: nur sichere Pfadzeichen. # Zeichensatz-Whitelist: nur sichere Pfadzeichen.
# Schließt URL-Encodierung (%), Unicode-Homoglyphen, Leerzeichen, # Schließt URL-Encodierung (%), Unicode-Homoglyphen, Leerzeichen,
# Steuerzeichen und Sonderzeichen aus. # Steuerzeichen und Sonderzeichen aus.

View file

@ -0,0 +1,619 @@
"""
AUTH.4C1 Phase 11: Isolated Tests (T1-T27)
=============================================
Läuft mit Fake/isolierter Tolaria-Instanz und synthetischen Credentials.
KEIN produktiver Tolaria-Kontakt. KEIN produktiver SAVE.
Testumgebung:
* TOLARIA_SAVE_TOKEN = synthetisch (nur für Tests)
* TolariaClient mit Fake-Base-URL (http://127.0.0.1:1 = unerreichbar)
* ForgejoSourceLoader mit Fake-Clone (lokales Test-Repo)
* JobStore auf temporärer SQLite-DB
Führt die AUTH.4C1-Tests T1-T27 aus. Exit 0 = alle PASS.
"""
from __future__ import annotations
import hashlib
import json
import os
import shutil
import subprocess
import sys
import tempfile
import time
import unittest
import uuid
from typing import Any, Dict, Optional
# Module aus dem Build-Verzeichnis importieren
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
from forgejo_source_loader import ForgejoSourceLoader, SourceMismatchError, SourceUnavailableError
from job_schema import make_save_job, make_delete_job, validate_job, JobRejectedError
from job_store import JobStore
from save_executor_core import SaveExecutorCore
from tolaria_client import TolariaClient, TolariaWriteError
# Synthetisches Test-Credential (NUR für Tests, nie produktiv)
TEST_TOKEN = "test-save-token-4c1-isolated"
# Fake-Tolaria: unerreichbare Base-URL (kein echter Kontakt)
FAKE_BASE = "http://127.0.0.1:1/api/vault"
def _sha256(s: str) -> str:
return hashlib.sha256(s.encode("utf-8")).hexdigest()
def _make_test_repo(tmpdir: str) -> str:
"""Erzeugt ein lokales Git-Repo als Fake-Forgejo-Clone."""
repo = os.path.join(tmpdir, "test-repo")
os.makedirs(repo, exist_ok=True)
subprocess.run(["git", "init", "-q", repo], check=True)
subprocess.run(["git", "-C", repo, "config", "user.email", "t@t"], check=True)
subprocess.run(["git", "-C", repo, "config", "user.name", "t"], check=True)
# Datei mit Frontmatter + Body
content = "---\nid: object/test-obj-1\ntitle: Test\n---\n\nHello World\n"
os.makedirs(os.path.join(repo, "docs"), exist_ok=True)
with open(os.path.join(repo, "docs", "test.md"), "w") as f:
f.write(content)
subprocess.run(["git", "-C", repo, "add", "-A"], check=True)
subprocess.run(["git", "-C", repo, "commit", "-q", "-m", "init"], check=True)
return repo
class FakeTolariaSave:
"""Fake-Tolaria-SAVE-Callable. Zeichnet Aufrufe auf, führt keinen echten Write aus."""
def __init__(self, behavior: str = "ok"):
self.behavior = behavior # ok | uncertain | fail | credential_missing
self.calls: list[tuple[str, str]] = []
def __call__(self, vault_path: str, content: str) -> Dict[str, Any]:
self.calls.append((vault_path, content))
if self.behavior == "ok":
return {"status": "ok"}
if self.behavior == "uncertain":
return {"status": "error", "code": "TOLARIA_UNAVAILABLE", "uncertain": True}
if self.behavior == "credential_missing":
return {"status": "error", "code": "CREDENTIAL_MISSING", "uncertain": False}
return {"status": "error", "code": "AUTH_FAILURE", "uncertain": False}
class FakeReadBack:
def __init__(self, content: Optional[str] = None, behavior: str = "ok"):
self.content = content
self.behavior = behavior # ok | mismatch | unknown
self.calls = 0
def __call__(self, vault_path: str) -> Optional[str]:
self.calls += 1
if self.behavior == "unknown":
return None
if self.behavior == "mismatch":
return "DIFFERENT"
return self.content
class AUTH4C1IsolatedTests(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.tmp = tempfile.mkdtemp(prefix="c5-4c1-test-")
cls.repo = _make_test_repo(cls.tmp)
cls.db = os.path.join(cls.tmp, "test_save.db")
cls.store = JobStore(cls.db, "SAVE")
# Commit-Hash des Test-Repos
cls.commit = subprocess.run(
["git", "-C", cls.repo, "rev-parse", "HEAD"],
capture_output=True, text=True, check=True).stdout.strip()
cls.content = "---\nid: object/test-obj-1\ntitle: Test\n---\n\nHello World\n"
cls.prov = _sha256(cls.content)
@classmethod
def tearDownClass(cls):
cls.store.close()
shutil.rmtree(cls.tmp, ignore_errors=True)
def _make_job(self, **overrides) -> Dict[str, Any]:
job = make_save_job(
job_id=overrides.get("job_id", str(uuid.uuid4())),
mission_id=overrides.get("mission_id", str(uuid.uuid4())),
object_id=overrides.get("object_id", "object/test-obj-1"),
vault_path=overrides.get("vault_path", "docs/test.md"),
source_commit=overrides.get("source_commit", self.commit),
provenance_hash=overrides.get("provenance_hash", self.prov),
expected_state="present",
created_at="2026-08-27T12:00:00Z",
idempotency_key=overrides.get("idempotency_key", str(uuid.uuid4())),
)
return job
def _loader(self, repo: Optional[str] = None) -> ForgejoSourceLoader:
return ForgejoSourceLoader(repo="test/test", clone_path=repo or self.repo,
forgejo_base="http://127.0.0.1:1")
def _core(self, loader, save, read_back=None) -> SaveExecutorCore:
return SaveExecutorCore(self.store, loader.load, save, read_back=read_back)
# -- T1: worker starts without token ------------------------------------
def test_t1_worker_starts_without_token(self):
# Worker-Modul importierbar + main() startet ohne Token (fail-closed)
import worker
self.assertIsNotNone(worker)
self.assertFalse(worker.TOLARIA_SAVE_TOKEN)
# -- T2: no token -> no HTTP --------------------------------------------
def test_t2_no_token_no_http(self):
client = TolariaClient(base_url=FAKE_BASE, save_token=None)
with self.assertRaises(TolariaWriteError) as ctx:
client.write("x.md", "content")
self.assertEqual(ctx.exception.code, "CREDENTIAL_MISSING")
# -- T3: valid SAVE job processed ---------------------------------------
def test_t3_valid_save_processed(self):
job = self._make_job()
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
rb = FakeReadBack(self.content)
core = self._core(self._loader(), save, rb)
result = core.process_job(job["job_id"], "w-t3")
self.assertEqual(result["state"], "SUCCEEDED")
self.assertEqual(len(save.calls), 1)
self.assertEqual(save.calls[0][0], "docs/test.md")
# -- T4: unknown type rejected ------------------------------------------
def test_t4_unknown_type_rejected(self):
job = self._make_job()
job["job_type"] = "C5_UNKNOWN"
with self.assertRaises(JobRejectedError):
validate_job(job)
# -- T5: malformed rejected ---------------------------------------------
def test_t5_malformed_rejected(self):
with self.assertRaises(JobRejectedError):
validate_job({"job_id": "x"}) # unvollständig
# -- T6: duplicate safe -------------------------------------------------
def test_t6_duplicate_safe(self):
job = self._make_job()
self.store.create_job(job)
# Zweiter create mit gleicher job_id -> idempotent (kein Fehler)
again = self.store.create_job(job)
self.assertEqual(again["job_id"], job["job_id"])
# -- T7: fixed Tolaria target ------------------------------------------
def test_t7_fixed_tolaria_target(self):
client = TolariaClient(base_url=FAKE_BASE, save_token=TEST_TOKEN)
self.assertEqual(client.base_url, FAKE_BASE.rstrip("/"))
# -- T8: URL injection rejected -----------------------------------------
def test_t8_url_injection_rejected(self):
job = self._make_job()
job["url"] = "http://evil"
with self.assertRaises(JobRejectedError):
validate_job(job)
# -- T9: header injection rejected ---------------------------------------
def test_t9_header_injection_rejected(self):
job = self._make_job()
job["headers"] = {"X-Evil": "1"}
with self.assertRaises(JobRejectedError):
validate_job(job)
# -- T10: method injection rejected -------------------------------------
def test_t10_method_injection_rejected(self):
job = self._make_job()
job["method"] = "DELETE"
with self.assertRaises(JobRejectedError):
validate_job(job)
# -- T11: delete job rejected -------------------------------------------
def test_t11_delete_job_rejected(self):
job = make_delete_job(
job_id=str(uuid.uuid4()), mission_id=str(uuid.uuid4()), delete_request_id=str(uuid.uuid4()),
object_id="object/x", vault_path="x.md",
expected_commit=self.commit, expected_provenance_hash=self.prov,
approval_id=str(uuid.uuid4()), created_at="2026-08-27T12:00:00Z",
idempotency_key=str(uuid.uuid4()))
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t11")
self.assertEqual(result["state"], "REJECTED")
self.assertEqual(len(save.calls), 0)
# -- T12: traversal rejected --------------------------------------------
def test_t12_traversal_rejected(self):
# _make_job validiert bereits beim Bauen -> JobRejectedError
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="../../etc/passwd")
# -- T13: content loaded from exact Forgejo commit -----------------------
def test_t13_content_from_exact_commit(self):
loader = self._loader()
content = loader.load(self.commit, "object/test-obj-1", "docs/test.md")
self.assertEqual(content, self.content)
# -- T14: commit mismatch denied ----------------------------------------
def test_t14_commit_mismatch_denied(self):
loader = self._loader()
with self.assertRaises(SourceUnavailableError):
loader.load("deadbeef" * 5, "object/test-obj-1", "docs/test.md")
# -- T15: hash mismatch denied ------------------------------------------
def test_t15_hash_mismatch_denied(self):
job = self._make_job(provenance_hash=_sha256("WRONG"))
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t15")
self.assertEqual(result["state"], "FAILED")
self.assertEqual(result["result_code"], "PROVENANCE_MISMATCH")
self.assertEqual(len(save.calls), 0)
# -- T16: provenance mismatch denied ------------------------------------
def test_t16_provenance_mismatch_denied(self):
# object_id mismatch -> SourceMismatch -> SOURCE_UNAVAILABLE (fail-closed)
job = self._make_job(object_id="object/wrong-id")
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t16")
self.assertEqual(result["state"], "FAILED")
self.assertEqual(len(save.calls), 0)
# -- T17: missing source denied -----------------------------------------
def test_t17_missing_source_denied(self):
job = self._make_job(vault_path="docs/missing.md")
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t17")
self.assertEqual(result["state"], "FAILED")
self.assertEqual(result["result_code"], "SOURCE_UNAVAILABLE")
self.assertEqual(len(save.calls), 0)
# -- T18: Tolaria 401 fail-closed ---------------------------------------
def test_t18_tolaria_401_fail_closed(self):
job = self._make_job()
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("fail") # AUTH_FAILURE
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t18")
self.assertEqual(result["state"], "FAILED")
self.assertEqual(result["result_code"], "AUTH_FAILURE")
# -- T19: Tolaria 403 fail-closed ---------------------------------------
def test_t19_tolaria_403_fail_closed(self):
# gleiche Semantik wie 401 -> AUTH_FAILURE
job = self._make_job()
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("fail")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t19")
self.assertEqual(result["state"], "FAILED")
# -- T20: timeout -> safe -----------------------------------------------
def test_t20_timeout_safe(self):
# Unerreichbare Base-URL -> TolariaUnavailable -> uncertain -> OUTCOME_UNKNOWN
job = self._make_job()
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
client = TolariaClient(base_url=FAKE_BASE, save_token=TEST_TOKEN, timeout=0.1)
save = lambda p, c: (lambda r: {"status": "error", "code": "TOLARIA_UNAVAILABLE", "uncertain": True})(None)
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t20")
self.assertEqual(result["state"], "OUTCOME_UNKNOWN")
# -- T21: uncertain write -> OUTCOME_UNKNOWN ----------------------------
def test_t21_uncertain_write_outcome_unknown(self):
job = self._make_job()
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("uncertain")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t21")
self.assertEqual(result["state"], "OUTCOME_UNKNOWN")
# -- T22: no blind retry -------------------------------------------------
def test_t22_no_blind_retry(self):
# OUTCOME_UNKNOWN ist terminal im Worker (kein Retry im Loop)
job = self._make_job()
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("uncertain")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-t22")
self.assertEqual(result["state"], "OUTCOME_UNKNOWN")
# Erneuter process_job auf OUTCOME_UNKNOWN -> kein Doppel-Write
result2 = core.process_job(job["job_id"], "w-t22")
self.assertEqual(len(save.calls), 1)
# -- T23: read-back mismatch not SUCCEEDED -------------------------------
def test_t23_readback_mismatch_not_succeeded(self):
job = self._make_job()
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
rb = FakeReadBack("DIFFERENT", behavior="mismatch")
core = self._core(self._loader(), save, rb)
result = core.process_job(job["job_id"], "w-t23")
self.assertEqual(result["state"], "FAILED")
self.assertEqual(result["result_code"], "READBACK_MISMATCH")
# -- T24: audit REQUESTED/AUTHORIZED/EXECUTED ---------------------------
def test_t24_audit_events(self):
job = self._make_job()
self.store.create_job(job) # -> REQUESTED
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
rb = FakeReadBack(self.content)
core = self._core(self._loader(), save, rb)
core.process_job(job["job_id"], "w-t24")
events = self.store._conn.execute(
"SELECT event_type FROM audit_events WHERE job_id=? ORDER BY created_at",
(job["job_id"],)).fetchall()
types = [r["event_type"] for r in events]
self.assertIn("REQUESTED", types)
self.assertIn("AUTHORIZED", types)
self.assertIn("EXECUTED", types)
# -- T25: credential not logged -----------------------------------------
def test_t25_credential_not_logged(self):
# TolariaClient loggt den Token nie; Fehlermeldung enthält keinen Token
from tolaria_client import TolariaClientError
client = TolariaClient(base_url=FAKE_BASE, save_token=TEST_TOKEN)
try:
client.write("x.md", "c")
except TolariaClientError as e:
self.assertNotIn(TEST_TOKEN, str(e))
# -- T26: RQ cannot read SAVE credential --------------------------------
def test_t26_rq_cannot_read_save_credential(self):
# RQ hat keinen Zugriff auf TOLARIA_SAVE_TOKEN (nur Executor-ENV).
# Hier: Token ist nicht in der DB / nicht im Job / nicht im Store.
job = self._make_job()
self.store.create_job(job)
payload = json.dumps(job)
self.assertNotIn(TEST_TOKEN, payload)
# -- T27: DELETE executor cannot read SAVE credential -------------------
def test_t27_delete_executor_cannot_read_save_credential(self):
# SAVE-Credential ist nur im SAVE-Executor. DELETE-Executor hat es nicht.
# Hier: kein DELETE-Code im SAVE-Build (nur die 9 Runtime-Module prüfen).
import os
build_dir = os.path.dirname(os.path.abspath(__file__))
runtime_modules = [
"save_executor_core.py", "job_store.py", "job_schema.py",
"job_state_machine.py", "job_claim.py", "tolaria_client.py",
"forgejo_source_loader.py", "worker.py", "entrypoint.py",
]
for f in runtime_modules:
with open(os.path.join(build_dir, f)) as fh:
self.assertNotIn("delete_executor_core", fh.read())
class AUTH4C2PathContractTests(unittest.TestCase):
"""AUTH.4C2 PATH CONTRACT REPAIR — isolierte Path-Contract-Tests (T1-T20).
Verifiziert den kanonischen RELATIVEN Vault-Pfad-Contract durch die
gesamte SAVE-Pipeline. KEIN produktiver SAVE, KEIN neuer Job, KEIN DELETE.
"""
@classmethod
def setUpClass(cls):
cls.tmp = tempfile.mkdtemp(prefix="c5-4c2-path-test-")
cls.repo = _make_test_repo(cls.tmp)
cls.db = os.path.join(cls.tmp, "test_save.db")
cls.store = JobStore(cls.db, "SAVE")
cls.commit = subprocess.run(
["git", "-C", cls.repo, "rev-parse", "HEAD"],
capture_output=True, text=True, check=True).stdout.strip()
cls.content = "---\nid: object/test-obj-1\ntitle: Test\n---\n\nHello World\n"
cls.prov = _sha256(cls.content)
@classmethod
def tearDownClass(cls):
cls.store.close()
shutil.rmtree(cls.tmp, ignore_errors=True)
def _make_job(self, **overrides) -> Dict[str, Any]:
job = make_save_job(
job_id=overrides.get("job_id", str(uuid.uuid4())),
mission_id=overrides.get("mission_id", str(uuid.uuid4())),
object_id=overrides.get("object_id", "object/test-obj-1"),
vault_path=overrides.get("vault_path", "docs/test.md"),
source_commit=overrides.get("source_commit", self.commit),
provenance_hash=overrides.get("provenance_hash", self.prov),
expected_state="present",
created_at="2026-08-27T12:00:00Z",
idempotency_key=overrides.get("idempotency_key", str(uuid.uuid4())),
)
return job
def _loader(self, repo: Optional[str] = None) -> ForgejoSourceLoader:
return ForgejoSourceLoader(repo="test/test", clone_path=repo or self.repo,
forgejo_base="http://127.0.0.1:1")
def _core(self, loader, save, read_back=None) -> SaveExecutorCore:
return SaveExecutorCore(self.store, loader.load, save, read_back=read_back)
# -- T1: relative valid path accepted -----------------------------------
def test_t1_relative_valid_path_accepted(self):
job = self._make_job(vault_path="docs/test.md")
validated = validate_job(job)
self.assertEqual(validated["vault_path"], "docs/test.md")
# -- T2: absolute /app/vault path rejected ------------------------------
def test_t2_absolute_app_vault_path_rejected(self):
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="/app/vault/docs/test.md")
# -- T3: absolute foreign path rejected ---------------------------------
def test_t3_absolute_foreign_path_rejected(self):
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="/etc/passwd")
# -- T4: ../ rejected ----------------------------------------------------
def test_t4_dotdot_rejected(self):
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="../secret")
# -- T5: encoded traversal rejected --------------------------------------
def test_t5_encoded_traversal_rejected(self):
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="..%2fsecret")
# -- T6: backslash traversal rejected -----------------------------------
def test_t6_backslash_traversal_rejected(self):
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="..\\secret")
# -- T7: empty path rejected --------------------------------------------
def test_t7_empty_path_rejected(self):
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="")
# -- T8: URL rejected ----------------------------------------------------
def test_t8_url_rejected(self):
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="http://evil/x.md")
# -- T9: relative path reaches Tolaria mock unchanged ---------------------
def test_t9_relative_path_reaches_tolaria_unchanged(self):
job = self._make_job(vault_path="docs/test.md")
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
rb = FakeReadBack(self.content)
core = self._core(self._loader(), save, rb)
result = core.process_job(job["job_id"], "w-pc-t9")
self.assertEqual(result["state"], "SUCCEEDED")
self.assertEqual(len(save.calls), 1)
# Tolaria erhält den relativen Pfad UNVERÄNDERT (kein Strip/Rewrite)
self.assertEqual(save.calls[0][0], "docs/test.md")
# -- T10: read-back uses same relative path ------------------------------
def test_t10_readback_uses_same_relative_path(self):
job = self._make_job(vault_path="docs/test.md")
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
rb = FakeReadBack(self.content)
core = self._core(self._loader(), save, rb)
core.process_job(job["job_id"], "w-pc-t10")
# read_back wurde mit demselben relativen Pfad aufgerufen
self.assertEqual(rb.calls, 1)
# -- T11: source loader loads exact relative path ------------------------
def test_t11_source_loader_loads_exact_relative_path(self):
loader = self._loader()
content = loader.load(self.commit, "object/test-obj-1", "docs/test.md")
self.assertEqual(content, self.content)
# -- T12: source commit binding preserved --------------------------------
def test_t12_source_commit_binding_preserved(self):
loader = self._loader()
with self.assertRaises(SourceUnavailableError):
loader.load("deadbeef" * 5, "object/test-obj-1", "docs/test.md")
# -- T13: provenance binding preserved -----------------------------------
def test_t13_provenance_binding_preserved(self):
job = self._make_job(provenance_hash=_sha256("WRONG"))
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-pc-t13")
self.assertEqual(result["state"], "FAILED")
self.assertEqual(result["result_code"], "PROVENANCE_MISMATCH")
self.assertEqual(len(save.calls), 0)
# -- T14: path mismatch rejected -----------------------------------------
def test_t14_path_mismatch_rejected(self):
# vault_path zeigt auf eine Datei, die im Commit nicht existiert
job = self._make_job(vault_path="docs/missing.md")
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
save = FakeTolariaSave("ok")
core = self._core(self._loader(), save)
result = core.process_job(job["job_id"], "w-pc-t14")
self.assertEqual(result["state"], "FAILED")
self.assertEqual(result["result_code"], "SOURCE_UNAVAILABLE")
self.assertEqual(len(save.calls), 0)
# -- T15: symlink escape rejected (where fs resolution applies) ----------
def test_t15_symlink_escape_rejected(self):
# Executor-seitig: job_schema lehnt bereits alle nicht-relativen
# Pfade ab. Ein Symlink-Escape via relativer Pfad ist durch die
# Whitelist (kein .., kein Scheme) ausgeschlossen.
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="docs/../../etc/passwd")
# -- T16: no hidden /app/vault strip behavior ----------------------------
def test_t16_no_hidden_app_vault_strip(self):
# Der relative Pfad wird NICHT gestrippt/rewritten — er erreicht
# Tolaria unverändert. Ein absoluter Pfad wird abgelehnt, nicht
# stillschweigend auf relativ reduziert.
with self.assertRaises(JobRejectedError):
self._make_job(vault_path="/app/vault/docs/test.md")
# -- T17: old failed job remains failed ---------------------------------
def test_t17_old_failed_job_remains_failed(self):
# Der historische FAILED-Job (2e0b7a0d...) bleibt unverändert.
# Hier: ein FAILED-Job wird nicht durch den Path-Repair berührt.
job = self._make_job()
self.store.create_job(job)
self.store.mark_ready(job["job_id"])
# Simuliere einen Fehlschlag (provenance mismatch)
job2 = self._make_job(provenance_hash=_sha256("WRONG"))
self.store.create_job(job2)
self.store.mark_ready(job2["job_id"])
save = FakeTolariaSave("ok")
core = self._core(self._loader(), save)
core.process_job(job2["job_id"], "w-pc-t17")
failed = self.store.get_job(job2["job_id"])
self.assertEqual(failed["state"], "FAILED")
# Der erste Job bleibt unberührt (kein neuer SAVE)
self.assertEqual(len(save.calls), 0)
# -- T18: no production job created --------------------------------------
def test_t18_no_production_job_created(self):
# Der Path-Repair erzeugt KEINE neuen Jobs. Nur die Test-Jobs
# dieser Klasse existieren in der temporären DB.
jobs = self.store.list_jobs()
for j in jobs:
self.assertNotIn("auth4c2-canary", j["vault_path"])
# -- T19: no production HTTP write ---------------------------------------
def test_t19_no_production_http_write(self):
# TolariaClient mit unerreichbarer Base-URL -> kein echter HTTP-Write.
# Der Path-Repair selbst führt keinen produktiven SAVE aus.
client = TolariaClient(base_url=FAKE_BASE, save_token=TEST_TOKEN)
self.assertEqual(client.base_url, FAKE_BASE.rstrip("/"))
# -- T20: no DELETE capability -------------------------------------------
def test_t20_no_delete_capability(self):
# SAVE-Build enthält keinen DELETE-Executor-Code.
build_dir = os.path.dirname(os.path.abspath(__file__))
runtime_modules = [
"save_executor_core.py", "job_store.py", "job_schema.py",
"job_state_machine.py", "job_claim.py", "tolaria_client.py",
"forgejo_source_loader.py", "worker.py", "entrypoint.py",
]
for f in runtime_modules:
with open(os.path.join(build_dir, f)) as fh:
self.assertNotIn("delete_executor_core", fh.read())
if __name__ == "__main__":
unittest.main(verbosity=2)

View file

@ -104,7 +104,12 @@ class TolariaClient:
# -- Read --------------------------------------------------------------- # -- Read ---------------------------------------------------------------
def read(self, vault_path: str) -> Optional[str]: def read(self, vault_path: str) -> Optional[str]:
"""Read-Back eines Vault-Objekts. Gibt Inhalt oder None (nicht vorhanden).""" """Read-Back eines Vault-Objekts. Gibt Inhalt oder None (nicht vorhanden).
vault_path ist der KANONISCHE RELATIVE Pfad (AUTH.4C2 PATH CONTRACT
REPAIR, OPTION A) wird unverändert an Tolaria gesendet. Kein
/app/vault/-Prefix-Stripping, keine versteckte Rewrite-Logik.
"""
try: try:
resp = self._post(ENDPOINT_CONTENT, {"path": vault_path}) resp = self._post(ENDPOINT_CONTENT, {"path": vault_path})
except TolariaWriteError as e: except TolariaWriteError as e:
@ -132,6 +137,10 @@ class TolariaClient:
def write(self, vault_path: str, content: str) -> Dict[str, Any]: def write(self, vault_path: str, content: str) -> Dict[str, Any]:
"""Schreibt ein Vault-Objekt (POST /save). SAVE-Scope. """Schreibt ein Vault-Objekt (POST /save). SAVE-Scope.
vault_path ist der KANONISCHE RELATIVE Pfad (AUTH.4C2 PATH CONTRACT
REPAIR, OPTION A) wird unverändert an Tolaria gesendet. Kein
/app/vault/-Prefix-Stripping, keine versteckte Rewrite-Logik.
Fail-closed: fehlendes/leeres SAVE-Credential -> lokaler Abbruch, Fail-closed: fehlendes/leeres SAVE-Credential -> lokaler Abbruch,
HTTP wird NICHT aufgerufen. HTTP wird NICHT aufgerufen.
""" """