残留批次A:成员删除、PR needs_local_worker 落库、推送 DNS TOCTOU

- 成员删除:DELETE /projects/{id}/members/{user_id},限 owner 或
  全局管理员,禁删唯一 owner,记 remove_member 审计
  (AuthStore.remove_member + server 分发,沿用 PATCH 守卫口径)
- PR needs_local_worker 落库:非 local 扫到 queued 任务时 CAS
  写行状态为 needs_local_worker;claim/find_open 同步认该状态,
  标记过的行仍可被 local worker 认领、不被搁置;另修
  claim/find_open 对裸 sqlite3 tuple 行的 dict() 兜底
  (TypeError/ValueError)
- 推送 DNS TOCTOU:发送改走建连时解析+判段的连接类,解析
  一次、只拨判过的 IP,Host/SNI 仍为原域名;opener 直接构造
  以排除 ProxyHandler 改写(build_opener 默认会塞代理)
This commit is contained in:
host committed 2026-10-27 19:10:49 +00:00
1 parent 6e3eb7ba11
commit 9bb9a50e4e
10 files changed
+725 -38

No files matched your search

+161
View File
@@ -0,0 +1,161 @@
"""成员删除端点单测(残留批次 A 任务 1)。
DELETE /projects/{id}/members/{user_id}:
- 限项目 owner 或全局管理员,其余 403;
- 目标非成员 404,项目不存在 404;
- 删唯一 owner 时 409(owner 与全局管理员均受此限);
- 成功后成员列表无此人,且记 remove_member 审计。
测试用的 key、用户 id 全是假值;不联网、不碰生产。
"""
import http.client
import json
import os
import sys
import tempfile
import threading
import unittest
from pathlib import Path
try:
from cloud.auth import AuthStore
from cloud.server import create_server
except ImportError: # discover -s cloud/tests 时把项目根目录补进 sys.path
sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
from cloud.auth import AuthStore
from cloud.server import create_server
def _req(port, method, path, body=None, key=None):
conn = http.client.HTTPConnection("127.0.0.1", port, timeout=10)
headers = {"Content-Type": "application/json"}
if key:
headers["X-API-Key"] = key
conn.request(method, path,
json.dumps(body) if body is not None else None, headers)
resp = conn.getresponse()
raw = resp.read()
conn.close()
try:
data = json.loads(raw)
except ValueError:
data = None
return resp.status, data
class MemberDeleteTestCase(unittest.TestCase):
def setUp(self):
self.tmp = tempfile.TemporaryDirectory()
self.db_path = os.path.join(self.tmp.name, "control.db")
self.trace_dir = os.path.join(self.tmp.name, "host-traces")
os.makedirs(self.trace_dir, exist_ok=True)
self.auth = AuthStore(self.db_path)
self.owner_id, self.owner_key = self.auth.create_user("owner")
self.maint_id, self.maint_key = self.auth.create_user("maint")
self.member_id, self.member_key = self.auth.create_user("member")
self.admin_id, self.admin_key = self.auth.create_user("admin")
self.outsider_id, self.out_key = self.auth.create_user("outsider")
self._old_admins = os.environ.get("CODEX_GLOBAL_ADMINS")
os.environ["CODEX_GLOBAL_ADMINS"] = self.admin_id
self.server = create_server(self.db_path, "127.0.0.1", 0)
self.server.trace_dir = self.trace_dir
self.port = self.server.server_address[1]
self.thread = threading.Thread(
target=self.server.serve_forever, daemon=True)
self.thread.start()
status, proj = _req(self.port, "POST", "/projects",
{"name": "demo"}, self.owner_key)
self.assertEqual(status, 201, proj)
self.project_id = proj["id"]
self.auth.set_member(self.project_id, self.owner_id,
self.maint_id, "maintainer")
def tearDown(self):
self.server.shutdown()
self.server.server_close()
self.thread.join(timeout=5)
try:
self.auth.conn.close()
except Exception:
pass
self.tmp.cleanup()
if self._old_admins is None:
os.environ.pop("CODEX_GLOBAL_ADMINS", None)
else:
os.environ["CODEX_GLOBAL_ADMINS"] = self._old_admins
def _delete(self, target, key):
return _req(self.port, "DELETE",
"/projects/%s/members/%s" % (self.project_id, target),
key=key)
def _invite(self, target, role="contributor", key=None):
return _req(self.port, "POST",
"/projects/%s/members" % self.project_id,
{"user_id": target, "role": role},
key or self.owner_key)
def _member_ids(self, key=None):
status, data = _req(self.port, "GET",
"/projects/%s/members" % self.project_id,
key=key or self.owner_key)
self.assertEqual(status, 200, data)
return {m["user_id"] for m in data["members"]}
def _audit_actions(self):
return [(a["action"], a["detail"])
for a in self.auth.list_audit(self.project_id)]
def test_owner_delete_member_and_audit(self):
self.assertEqual(self._invite(self.member_id)[0], 201)
status, data = self._delete(self.member_id, self.owner_key)
self.assertEqual(status, 200, data)
self.assertEqual(data["user_id"], self.member_id)
self.assertNotIn(self.member_id, self._member_ids())
self.assertIn(("remove_member", self.member_id),
self._audit_actions())
def test_non_owner_and_outsider_denied(self):
self.assertEqual(self._invite(self.member_id)[0], 201)
status, _ = self._delete(self.member_id, self.maint_key)
self.assertEqual(status, 403)
status, _ = self._delete(self.member_id, self.out_key)
self.assertEqual(status, 403)
# 拒绝后成员仍在。
self.assertIn(self.member_id, self._member_ids())
def test_global_admin_delete_without_membership(self):
self.assertEqual(self._invite(self.member_id)[0], 201)
status, data = self._delete(self.member_id, self.admin_key)
self.assertEqual(status, 200, data)
self.assertNotIn(self.member_id, self._member_ids())
self.assertIn(("remove_member", self.member_id),
self._audit_actions())
def test_delete_non_member_and_missing_project(self):
status, _ = self._delete("ghost", self.owner_key)
self.assertEqual(status, 404)
status, _ = _req(self.port, "DELETE",
"/projects/no-such/members/%s" % self.member_id,
key=self.owner_key)
self.assertEqual(status, 404)
def test_cannot_remove_only_owner(self):
status, data = self._delete(self.owner_id, self.owner_key)
self.assertEqual(status, 409, data)
status, _ = self._delete(self.owner_id, self.admin_key)
self.assertEqual(status, 409)
# 升第二位 owner 后,删原 owner 放行。
self.assertEqual(self._invite(self.member_id)[0], 201)
status, _ = _req(self.port, "PATCH",
"/projects/%s/members/%s"
% (self.project_id, self.member_id),
{"role": "owner"}, self.owner_key)
self.assertEqual(status, 200)
status, _ = self._delete(self.owner_id, self.maint_key)
# maint 不是 owner 也不是全局管理员:仍 403(守卫先行)。
self.assertEqual(status, 403)
status, data = self._delete(self.owner_id, self.member_key)
self.assertEqual(status, 200, data)
self.assertNotIn(self.owner_id, self._member_ids(self.member_key))
+8 -2
View File
@@ -449,7 +449,8 @@ class P1WorkerTestCase(unittest.TestCase):
self.assertEqual(_status(self.db_path, "T1"), "failed")
self.assertEqual(self.calls, [])
# 15. PR 非 local:显式 needs_local_worker,任务与排队都不动
# 15. PR 非 local:显式 needs_local_worker,行状态持久化,
# 任务本身不动,标记过的行仍可被 local worker 认领
def test_pr_non_local_needs_local_worker(self):
conn = sqlite3.connect(self.db_path)
try:
@@ -470,9 +471,14 @@ class P1WorkerTestCase(unittest.TestCase):
row = conn.execute(
"SELECT status FROM pr_jobs WHERE id = ?",
(job_id,)).fetchone()
self.assertEqual(row[0], "needs_local_worker")
# 标记过的行仍可被 local worker 认领,不会被搁置。
from cloud.pr_exec import claim_pr_job
claimed = claim_pr_job(conn, "test-host")
self.assertIsNotNone(claimed)
self.assertEqual(claimed["id"], job_id)
finally:
conn.close()
self.assertEqual(row[0], "queued")
# 16. 多文件签名 + 签名缺失/已变时认领 fail-closed
def test_code_signature_multi_file_and_claim_refusal(self):
+1 -1
View File
@@ -308,7 +308,7 @@ class WorkerPrJobsTestCase(unittest.TestCase):
finally:
conn.close()
# P1:Docker worker 不再静默返回空,改为显式 needs_local_worker,
# 任务保持 queued 等 local worker 认领。
# 行状态持久化为 needs_local_worker、仍可被 local worker 认领。
jobs = worker._run_pr_jobs()
self.assertEqual(len(jobs), 1)
self.assertEqual(jobs[0]["status"], "needs_local_worker")
+116
View File
@@ -0,0 +1,116 @@
"""PR needs_local_worker 持久化单测(残留批次 A 任务 2)。
- 非 local worker 扫到 queued 的 pr 任务后,行状态持久化为
needs_local_worker(重读仍在),且返回条目状态一致;
- 标记过的行仍可被 local worker 认领(claim 后 running,可
正常 finish),不会被搁置;
- find_open_pr_job 仍把标记过的行算未完成(/pr 不重排);
- 标记是 CAS:已 running 的行不被覆盖;重复扫不重复记。
测试用的 key、主机名、任务 id 全是假值;不联网、不碰生产。
"""
import os
import sys
import tempfile
import unittest
from pathlib import Path
try:
from cloud.db import connect as _db_connect
from cloud.pr_exec import (
claim_pr_job, enqueue_pr_job, find_open_pr_job,
mark_pr_job_needs_local_worker)
from cloud.sandbox import DockerExecutor
from cloud.worker import Worker
except ImportError: # discover -s cloud/tests 时把项目根目录补进 sys.path
sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
from cloud.db import connect as _db_connect
from cloud.pr_exec import (
claim_pr_job, enqueue_pr_job, find_open_pr_job,
mark_pr_job_needs_local_worker)
from cloud.sandbox import DockerExecutor
from cloud.worker import Worker
class PrNeedsLocalTestCase(unittest.TestCase):
def setUp(self):
self.tmp = tempfile.TemporaryDirectory()
self.db_path = os.path.join(self.tmp.name, "control.db")
def _enqueue_t1(self, host="host-a"):
conn = _db_connect(self.db_path)
try:
enqueue_pr_job(conn, "t1", "p1", host)
finally:
conn.close()
def tearDown(self):
self.tmp.cleanup()
def _row(self, task_id):
conn = _db_connect(self.db_path)
try:
return conn.execute(
"SELECT id, status, error FROM pr_jobs"
" WHERE task_id = ?", (task_id,)).fetchone()
finally:
conn.close()
def test_docker_peek_persists_mark(self):
worker = Worker(self.db_path, "/nonexistent/host.py",
env_key="TEST_KEY",
executor=DockerExecutor("/fake/host.py"))
conn = _db_connect(self.db_path)
try:
enqueue_pr_job(conn, "t1", "p1", worker.host_name)
finally:
conn.close()
jobs = worker._run_pr_jobs()
self.assertEqual(len(jobs), 1)
self.assertEqual(jobs[0]["status"], "needs_local_worker")
self.assertEqual(jobs[0]["task_id"], "t1")
# 重读:行状态已持久化,不只是返回值里有。
row = self._row("t1")
self.assertEqual(row[1], "needs_local_worker")
self.assertIn("local worker", row[2])
# 再扫一次:已标记的行不再是 queued,不重复返回、不重复记。
self.assertEqual(worker._run_pr_jobs(), [])
def test_marked_row_still_claimable_by_local(self):
self._enqueue_t1()
conn = _db_connect(self.db_path)
try:
job_id = conn.execute(
"SELECT id FROM pr_jobs WHERE task_id = 't1'"
).fetchone()[0]
self.assertTrue(mark_pr_job_needs_local_worker(
conn, job_id, "needs local worker"))
# 标记后仍算未完成:/pr 去重不受影响。
self.assertIsNotNone(find_open_pr_job(conn, "t1"))
# local worker 认领标记过的行。
claimed = claim_pr_job(conn, "host-a")
self.assertIsNotNone(claimed)
self.assertEqual(claimed["task_id"], "t1")
row = conn.execute(
"SELECT status FROM pr_jobs WHERE task_id = 't1'"
).fetchone()
self.assertEqual(row[0], "running")
finally:
conn.close()
def test_mark_is_cas_on_queued_only(self):
self._enqueue_t1()
conn = _db_connect(self.db_path)
try:
claimed = claim_pr_job(conn, "host-a")
self.assertIsNotNone(claimed)
# 已 running:标记必须失败、不覆盖状态。
self.assertFalse(mark_pr_job_needs_local_worker(
conn, claimed["id"], "late mark"))
row = conn.execute(
"SELECT status FROM pr_jobs WHERE task_id = 't1'"
).fetchone()
self.assertEqual(row[0], "running")
finally:
conn.close()
+172
View File
@@ -0,0 +1,172 @@
"""推送 DNS TOCTOU 单测(残留批次 A 任务 3)。
旧版:validate_push_target 解析一次,urlopen 建连时又解析
一次;两次解析结果不一致(DNS 重绑定)时可能拨向未验证
的内网 IP。新版:建连时解析一次并判段、只拨判过的 IP,
Host 头与 TLS SNI 仍为原域名。
- 两次解析不一致(先公网、后内网)时:发送失败,且一次
都没有拨向内网 IP;
- 混合解析结果(先内网、后公网)时:只拨公网 IP;
- HTTP 真服务器:域名经 resolver 钉到 127.0.0.1,服务器
收到的 Host 仍为原域名;
- HTTPS 连接类:拨号地址为判过的 IP,wrap_socket 的
server_hostname 仍为原域名(SNI 不变)。
测试用的 IP、域名、key 全是假值;除本地回环服务器外不联网。
"""
import http.client
import http.server
import os
import sys
import threading
import unittest
from pathlib import Path
from unittest import mock
try:
from cloud.push import (
_checked_connection_class, _post_json, send_webhook)
except ImportError: # discover -s cloud/tests 时把项目根目录补进 sys.path
sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
from cloud.push import (
_checked_connection_class, _post_json, send_webhook)
PUBLIC_IP = "8.8.8.8"
PRIVATE_IP = "169.254.169.254"
class _FlipFlopResolver:
"""第一次返回公网 IP,之后返回内网 IP:模拟重绑定。
旧版两次解析恰好踩中这个窗口;新版第二次解析判段后
直接拒绝,绝不拨号。
"""
def __init__(self):
self.calls = 0
def __call__(self, host):
self.calls += 1
return [PUBLIC_IP] if self.calls == 1 else [PRIVATE_IP]
class _DialRecorder:
"""记录拨号地址后直接失败:测试只断言拨向,不跑完 HTTP。"""
def __init__(self):
self.calls = []
def __call__(self, address, timeout=None, source_address=None):
self.calls.append(address)
raise OSError("dial stopped by test")
class _DummySocket:
def setsockopt(self, *args):
pass
class _RecordingHandler(http.server.BaseHTTPRequestHandler):
requests = []
def do_POST(self):
length = int(self.headers.get("Content-Length", 0))
body = self.rfile.read(length) if length else b""
_RecordingHandler.requests.append({
"host": self.headers.get("Host"),
"body": body,
})
self.send_response(200)
self.send_header("Content-Length", "2")
self.end_headers()
self.wfile.write(b"{}")
def log_message(self, *args):
pass
class PushToctouTestCase(unittest.TestCase):
def setUp(self):
self._old_allow = os.environ.get("CODEX_PUSH_ALLOW_PRIVATE_TARGETS")
os.environ.pop("CODEX_PUSH_ALLOW_PRIVATE_TARGETS", None)
_RecordingHandler.requests = []
def tearDown(self):
if self._old_allow is None:
os.environ.pop("CODEX_PUSH_ALLOW_PRIVATE_TARGETS", None)
else:
os.environ["CODEX_PUSH_ALLOW_PRIVATE_TARGETS"] = self._old_allow
def _message(self, target, resolver=None):
msg = {"event_type": "completed", "project_id": "proj-a",
"payload": {"task_id": "t-1"}, "target": target}
if resolver is not None:
msg["_resolver"] = resolver
return msg
def test_rebinding_second_resolution_never_dialed(self):
resolver = _FlipFlopResolver()
dials = _DialRecorder()
with mock.patch("socket.create_connection", dials):
with self.assertRaises(RuntimeError):
send_webhook(self._message(
"http://hook.example.invalid/hook", resolver))
# 两次解析确实都发生了(一次预检、一次建连),但建连侧
# 判段后拒绝:内网 IP 一次都没被拨。
self.assertEqual(resolver.calls, 2)
self.assertEqual(dials.calls, [])
def test_mixed_resolution_dials_only_public(self):
dials = _DialRecorder()
with mock.patch("socket.create_connection", dials):
ok, _detail = _post_json(
"http://hook.example.invalid/hook", {"a": 1},
resolver=lambda host: [PRIVATE_IP, PUBLIC_IP])
self.assertFalse(ok)
self.assertEqual(dials.calls, [(PUBLIC_IP, 80)])
def test_http_pinned_host_header_preserved(self):
os.environ["CODEX_PUSH_ALLOW_PRIVATE_TARGETS"] = "1"
server = http.server.HTTPServer(("127.0.0.1", 0),
_RecordingHandler)
port = server.server_address[1]
thread = threading.Thread(target=server.serve_forever, daemon=True)
thread.start()
try:
send_webhook(self._message(
"http://webhook.test.invalid:%d/hook" % port,
resolver=lambda host: ["127.0.0.1"]))
finally:
server.shutdown()
server.server_close()
thread.join(timeout=5)
self.assertEqual(len(_RecordingHandler.requests), 1)
host_header = _RecordingHandler.requests[0]["host"]
# 拨的是 127.0.0.1,但 Host 仍为原域名(虚拟主机依赖它)。
self.assertEqual(host_header.split(":")[0],
"webhook.test.invalid")
def test_https_sni_preserved_with_pinned_ip(self):
wrap_calls = []
class FakeContext:
def wrap_socket(self, sock, server_hostname=None):
wrap_calls.append(server_hostname)
return sock
cls = _checked_connection_class(http.client.HTTPSConnection,
lambda host: [PUBLIC_IP])
conn = cls("example.com", 443)
conn._context = FakeContext()
dials = []
def fake_create(address, timeout=None, source_address=None):
dials.append(address)
return _DummySocket()
with mock.patch("socket.create_connection", fake_create):
conn.connect()
self.assertEqual(dials, [(PUBLIC_IP, 443)])
self.assertEqual(wrap_calls, ["example.com"])