fix(chv): P2-20/P2-19/P2-17/P2-18 残留收尾 + NO_PROXY 合并

- P2-20:_guest_env_delivered 加专用锁,读写只走
  _delivered_secrets/_mark/_clear 三个入口,并发 hammer 测试钉死
- P2-19:_record_net_entries 失败发 net_registry_write_failed
  独立审计并返回 False;_boot_vm 见 False 即拒起(wake 随后清半醒 VM)
- P2-17:net.json 删失败发 vm_net_registry_delete_failed 审计;
  _write_dirty_marker 失败发 dirty_marker_write_failed 审计并返回
  False,stop 不抛的契约不变
- P2-18:新增 cloud/chv/atomic_file.py(mkstemp 随机名 + 文件
  fsync + rename + 目录 fsync),net.json/known_hosts/
  snapshots.json/pool registry/脏标记/spill 全部走它,固定 .tmp 清零
- H1 附带:NO_PROXY 实测确有覆盖 bug(env 重复赋值以后者为准,
  调用方 NO_PROXY 被 egress 的 localhost 值静默覆盖),改为合并
  单条发出;egress 已设时调用方代理变量(含小写)一律不发
- 测试:新增 cloud/tests/test_chv_residual.py(23 项),旧 closeup
  两处钉死旧语义的测试按新语义更新
This commit is contained in:
host committed 2026-11-06 18:41:50 +00:00
1 parent ba5df980bf
commit c39715659e
7 files changed
+850 -51

No files matched your search

+71
View File
@@ -0,0 +1,71 @@
"""CHV 宿主侧小文件的原子写(P2-18 收尾)。
此前 net.json / snapshots.json / pool registry /
known_hosts 各自用固定名 `<path>.tmp` + 直接 os.replace,
存在两个问题:
1. 固定 tmp 名:同一进程内两个写者(或一次崩溃残留的
.tmp)会互相覆盖/误读;不同 VM 的文件虽在不同目录,
但同一目录内的快照登记与 net 登记仍共用可预测的
tmp 名,崩溃后残留的 .tmp 会被下一次写静默覆盖,
丢了“上次写了一半”的信号。
2. 无 fsync:os.replace 只保证目录项原子,不保证内容
已落盘;宿主断电/崩溃后新文件可能回到旧内容或
0 字节,而调用方已认为登记成功。
本模块是唯一写路径:mkstemp 随机名(0600)→ 写 →
flush + fsync 文件 → os.replace → fsync 父目录。
写失败时删临时文件再原样抛错,调用方按各自语义
处理(审计/拒绝/脏标记),本模块不吞错、不记日志。
"""
import os
import tempfile
def atomic_write(path: str, data: bytes, mode: int = 0o600) -> None:
"""原子写 bytes 到 path(随机 tmp 名 + 双 fsync)。
mode 是最终文件的权限(默认 0600:这些文件都在
0700 的 state_dir 内、且含登记/密钥类内容,按敏感
文件对待;调用方要放宽时显式传)。
"""
directory = os.path.dirname(path) or "."
fd, tmp_path = tempfile.mkstemp(prefix=".tmp-", dir=directory)
try:
os.fchmod(fd, mode)
with os.fdopen(fd, "wb") as fh:
fh.write(data)
fh.flush()
os.fsync(fh.fileno())
os.replace(tmp_path, path)
_fsync_dir(directory)
except BaseException:
try:
os.unlink(tmp_path)
except OSError:
pass
raise
def atomic_write_text(path: str, text: str, mode: int = 0o600) -> None:
"""atomic_write 的 str 版(UTF-8 编码)。"""
atomic_write(path, text.encode("utf-8"), mode=mode)
def _fsync_dir(directory: str) -> None:
"""fsync 父目录,让 rename 本身落盘(best-effort:
某些文件系统不支持目录 fsync,失败即跳过——文件
内容的 fsync 已在 atomic_write 里做完,目录 fsync
丢了只影响“rename 是否在崩溃后可见”,不影响内容)。
"""
try:
dir_fd = os.open(directory, os.O_RDONLY)
except OSError:
return
try:
os.fsync(dir_fd)
except OSError:
pass
finally:
os.close(dir_fd)
+3 -4
View File
@@ -154,12 +154,11 @@ def _load_snapshots(state_dir, vm_name):
def _save_snapshots(state_dir, vm_name, entries):
import json
from cloud.chv.atomic_file import atomic_write_text
path = _snapshots_path(state_dir, vm_name)
os.makedirs(os.path.dirname(path), exist_ok=True)
tmp_path = path + ".tmp"
with open(tmp_path, "w") as fh:
json.dump(entries, fh)
os.replace(tmp_path, path)
# feat-chv-closeup(P2-18):随机 tmp 名 + 双 fsync。
atomic_write_text(path, json.dumps(entries))
def prune_snapshots(state_dir: str, vm_name: str,
+98 -20
View File
@@ -37,6 +37,7 @@ import time
from collections.abc import Callable
from cloud.chv.api_client import _unix_socket_api, normalize_api_result
from cloud.chv.atomic_file import atomic_write_text
from cloud.chv.config import _parse_memory_bytes, build_chv_config
from cloud.chv.disk import (
check_disk_quota,
@@ -366,6 +367,16 @@ class CloudHypervisorExecutor(Executor):
# 名 source 客内文件,未投递的密钥名一律拒绝(防止
# “wake 没投递、执行时却按名 source 一个旧文件”)。
self._guest_env_delivered = {}
# feat-chv-closeup(P2-20):_guest_env_delivered 是
# 跨 wake/build_ssh_command/stop 的共享字典,wake 与
# 执行可并发在不同线程——此前无锁读写,set/get/pop
# 交错会丢投递记录(密钥被误判未投递而拒跑,或已清
# 的记录被误读为已投递)。专用锁护住全部访问(与
# _vms_lock 分开:投递记录与进程登记是两把不同的锁,
# 合一把会在 stop 路径上不必要地串行);所有访问点
# 必须走下面三个小方法,禁止直接读写字典(新方法
# 是唯一入口,回归测试钉死它)。
self._guest_env_lock = threading.Lock()
self.api_client = api_client if api_client is not None \
else _unix_socket_api
self.sleeper = sleeper if sleeper is not None else time.sleep
@@ -385,6 +396,25 @@ class CloudHypervisorExecutor(Executor):
self.launcher = _default_popen_launcher
self._vms = {}
def _delivered_secrets(self, name: str) -> set:
"""P2-20 唯一读入口:该 VM 已 seed 投递的密钥名集合。"""
with self._guest_env_lock:
return set(self._guest_env_delivered.get(name, set()))
def _mark_guest_env_delivered(self, name: str, secrets) -> None:
"""P2-20 唯一写入口:记下该 VM 已投递的密钥名集合
(secrets 为空即清记录,与旧语义一致)。"""
with self._guest_env_lock:
if secrets:
self._guest_env_delivered[name] = set(secrets)
else:
self._guest_env_delivered.pop(name, None)
def _clear_guest_env_delivered(self, name: str) -> None:
"""P2-20 唯一清除入口:stop 时清掉该 VM 的投递记录。"""
with self._guest_env_lock:
self._guest_env_delivered.pop(name, None)
def _copy_config_fields(self, _cfg) -> None:
"""把 build_chv_config 的字段抄到 self(纯移动,字段集不变)。"""
for _field in (
@@ -736,6 +766,11 @@ class CloudHypervisorExecutor(Executor):
"task_id": task_id})
with open(path, "a") as fh:
fh.write(line + "\n")
# feat-chv-closeup(P2-18):spill 是审计丢事
# 件后的唯一取证,写完即 fsync,否则宿主崩溃
# 后 spill 本身也可能丢。失败只记日志,不炸任务。
fh.flush()
os.fsync(fh.fileno())
except Exception as exc:
_LOGGER.error("chv_audit_spill_failed error=%s", exc)
@@ -942,8 +977,7 @@ class CloudHypervisorExecutor(Executor):
guest_env_file = None
if secrets:
from cloud.chv.hostkeys import GUEST_ENV_PATH
delivered = self._guest_env_delivered.get(
self.vm_name(task), set())
delivered = self._delivered_secrets(self.vm_name(task))
missing: list = [name for name in secrets if name not in delivered]
if missing:
raise ChvFailure(
@@ -1296,18 +1330,34 @@ class CloudHypervisorExecutor(Executor):
return registered
def _record_net_entries(self, vm, entries):
"""wake 成功后登记本 VM 的 net 条目(best-effort 写,
失败只记日志——登记缺失的后果是漏判冲突,见上文)。"""
"""wake 成功后登记本 VM 的 net 条目(P2-19 收尾)。
返回 True=登记成功,False=登记失败。失败时三件事
同时做:warning 日志(旧行为保留)+ 独立审计事件
net_registry_write_failed(调用方与运维都能看见,
不再只有一条日志)+ 返回 False 让调用方感知。
调用方 _boot_vm 见 False 即拒绝启动(ChvFailure,
wake 随后 stop_vm 清掉半醒 VM):登记缺失的后果
是后来的 wake 漏判 tap/mac/ip 冲突(H7 的第二道
防线没了),继续启动等于带着缺口跑,不如不起;
state_dir 可写本就是 preflight 的 0700 门,走到
这里才写失败是真故障,不是预期内的可容忍情形。
写路径走 atomic_file(P2-18):随机 tmp 名 + 双 fsync。
"""
try:
vm_dir = os.path.join(self.state_dir, vm)
os.makedirs(vm_dir, exist_ok=True)
tmp_path = os.path.join(vm_dir, "net.json.tmp")
with open(tmp_path, "w") as fh:
json.dump(entries, fh)
os.replace(tmp_path, os.path.join(vm_dir, "net.json"))
atomic_write_text(
os.path.join(vm_dir, "net.json"),
json.dumps(entries))
return True
except OSError as exc:
_LOGGER.warning("chv_net_registry_write_failed vm=%s "
"error=%s", vm, exc)
self._emit("net_registry_write_failed", None,
"vm=%s net 登记写失败:%s" % (vm, exc),
"net_registry_write_failed")
return False
def _check_api_socket_permissions(self, socket_path):
"""VMM api-socket 权限门(H6):socket 文件存在时,
@@ -1462,10 +1512,7 @@ class CloudHypervisorExecutor(Executor):
provisioned["seed_dir"], seed_disk,
runner=lambda cmd: self._call_runner(cmd))
delete_host_private_key(self.state_dir, name)
if secrets:
self._guest_env_delivered[name] = set(secrets)
else:
self._guest_env_delivered.pop(name, None)
self._mark_guest_env_delivered(name, secrets)
return seed_disk
def _boot_vm(self, task, name, task_disk, seed_disk,
@@ -1485,7 +1532,13 @@ class CloudHypervisorExecutor(Executor):
self._emit("vm_create", task,
"vm=%s disk=%s" % (name, task_disk))
if net_entries:
self._record_net_entries(name, net_entries)
# feat-chv-closeup(P2-19):登记失败即拒起,
# 见 _record_net_entries(审计 + 返回 False)。
if not self._record_net_entries(name, net_entries):
raise ChvFailure(
"net_registry_write_failed",
"vm=%s net 登记写失败,拒绝启动(缺登记会"
"漏判 tap/mac/ip 冲突)" % name)
status, _ = self._api_call(socket_path, "PUT", "/vm.boot")
if status not in (200, 204):
raise ChvFailure(
@@ -1771,24 +1824,49 @@ class CloudHypervisorExecutor(Executor):
_net_path = os.path.join(self.state_dir, name, "net.json")
if os.path.exists(_net_path):
os.remove(_net_path)
except Exception:
except Exception as _net_exc:
# feat-chv-closeup(P2-17):net.json 删失败
# 此前只记 failed=True、无独立审计——删不掉的
# 登记会让后来的 wake 误判该 tap/mac/ip 仍被
# 占用(误拒)或与脏标记混在一起查不出;现发
# 独立审计 vm_net_registry_delete_failed。
failed = True
self._guest_env_delivered.pop(name, None)
self._emit("vm_net_registry_delete_failed", task,
"vm=%s error=%s" % (name, _net_exc))
self._clear_guest_env_delivered(name)
return failed
def _write_dirty_marker(self, name, failed):
"""M2 脏标记:清理失败写标记、成功清标记;写失败不炸关闭。"""
"""M2 脏标记:清理失败写标记、成功清标记(P2-17 收尾)。
返回 True=标记状态已如愿落盘(失败时标记已写、
成功时标记已清/本来就没有),False=写/删标记
本身失败。失败时发独立审计 dirty_marker_write_failed
+ error 日志,不再全静默——标记写不下去时拒起链
本身就不存在(provision/wake 读不到标记),部署
必须保证 state_dir 可写(preflight 的 0700 门)
并靠审计告警发现;stop 路径仍不抛错(stop_vm
“尽力而为、不抛异常”的契约不变,False 只作
调用方可观测的信号)。写路径走 atomic_file
(P2-18):随机 tmp 名 + 双 fsync。
"""
from cloud.chv.hostkeys import dirty_marker_path
try:
_marker = dirty_marker_path(self.state_dir, name)
if failed:
os.makedirs(os.path.dirname(_marker), exist_ok=True)
with open(_marker, "w") as fh:
fh.write("cleanup failed\n")
atomic_write_text(_marker, "cleanup failed\n")
elif os.path.exists(_marker):
os.remove(_marker)
except Exception:
pass
return True
except Exception as exc:
_LOGGER.error("chv_dirty_marker_write_failed vm=%s "
"failed=%s error=%s", name, failed, exc)
self._emit("dirty_marker_write_failed", None,
"vm=%s 脏标记写失败(failed=%s):%s"
% (name, failed, exc),
"dirty_marker_write_failed")
return False
def stop_vm(self, task_or_name: dict | str) -> bool:
"""停止 VM(尽力而为,不抛异常):已死进程直接清登记,
+5 -5
View File
@@ -74,15 +74,15 @@ def _read_registry(path: str) -> tuple[dict | None, bool]:
def _write_registry(path: str, data: dict) -> None:
"""原子落盘 registry(tmp + rename);写失败只记日志。"""
"""原子落盘 registry(随机 tmp 名 + 双 fsync,P2-18);
写失败只记日志。"""
try:
from cloud.chv.atomic_file import atomic_write_text
import json
directory = os.path.dirname(path)
if directory:
os.makedirs(directory, exist_ok=True)
tmp_path = path + ".tmp"
with open(tmp_path, "w") as fh:
json.dump(data, fh)
os.replace(tmp_path, path)
atomic_write_text(path, json.dumps(data))
except OSError:
_LOGGER.warning("chv_pool_registry_save_failed path=%s", path)
+64 -18
View File
@@ -44,6 +44,7 @@ StrictHostKeyChecking=no + UserKnownHostsFile=/dev/null):
import os
import shlex
from cloud.chv.atomic_file import atomic_write_text
from cloud.chv.gitea_sync import (
GITEA_SYNC_PLAIN_KEYS,
GITEA_SYNC_SECRET_KEYS,
@@ -93,6 +94,63 @@ def split_guest_env(extra_env: dict | None) -> tuple[dict, dict]:
return guest_env, plain
def _merge_no_proxy(extra_env, egress_proxy):
"""调用方 NO_PROXY/no_proxy 与 egress 的 localhost、
127.0.0.1 去重合并(单条 NO_PROXY 的值)。"""
parts = []
if extra_env:
for key in ("NO_PROXY", "no_proxy"):
val = extra_env.get(key)
if val:
parts += [p.strip() for p in str(val).split(",")
if p.strip()]
merged = []
for part in parts:
if part not in merged:
merged.append(part)
if egress_proxy:
for part in ("localhost", "127.0.0.1"):
if part not in merged:
merged.append(part)
return merged
def _env_assignments(extra_env, egress_proxy):
"""env 前缀的赋值列表(feat-chv-closeup,H1 附带)。
NO_PROXY 覆盖顺序已核实并修掉:旧拼装是 extra_env 的
赋值在前、egress 的 NO_PROXY 在后,`env` 遇重复赋值以
后者为准(实测 env NO_PROXY=a NO_PROXY=b → b),调用方
传的 NO_PROXY 被 egress 的 localhost,127.0.0.1 静默覆盖。
现改为合并(见 _merge_no_proxy),单条 NO_PROXY 发出。
egress 代理本身须赢:egress 已设时调用方的
HTTPS_PROXY/HTTP_PROXY(含小写,curl 等工具读小写优先,
留着会架空 egress)一律不发出,只发 egress 的三条;
无 egress 时调用方的值原样保留(只去重,不丢)。"""
skip = {"NO_PROXY", "no_proxy"}
if egress_proxy:
skip |= {"HTTPS_PROXY", "HTTP_PROXY",
"https_proxy", "http_proxy"}
assignments = []
if extra_env:
for key in sorted(extra_env):
if key in skip:
continue
assignments.append("%s=%s"
% (key, shlex.quote(
str(extra_env[key]))))
merged = _merge_no_proxy(extra_env, egress_proxy)
if egress_proxy:
assignments += [
"HTTPS_PROXY=%s" % shlex.quote(egress_proxy),
"HTTP_PROXY=%s" % shlex.quote(egress_proxy),
"NO_PROXY=%s" % shlex.quote(",".join(merged))]
elif merged:
assignments += [
"NO_PROXY=%s" % shlex.quote(",".join(merged))]
return assignments
def remote_command(inner_cmd: list, egress_proxy: str | None = None,
extra_env: dict | None = None,
guest_env_file: str | None = None) -> list:
@@ -104,7 +162,8 @@ def remote_command(inner_cmd: list, egress_proxy: str | None = None,
set +a`(把 seed 投递的密钥文件 source 进环境并 export),
再接 env 前缀与客内命令;文件路径本身不是密钥。
extra_env 含密钥名时 ValueError(调用方须先 split 出
密钥子集走 seed 投递,见 split_secret_env)。"""
密钥子集走 seed 投递,见 split_secret_env)。env 前缀的
拼装见 _env_assignments(NO_PROXY 合并、egress 优先)。"""
if extra_env:
secrets, _ = split_secret_env(extra_env)
if secrets:
@@ -115,17 +174,7 @@ def remote_command(inner_cmd: list, egress_proxy: str | None = None,
if guest_env_file:
tokens += ["set", "-a;", ".",
shlex.quote(guest_env_file) + ";", "set", "+a;"]
env_assignments = []
if extra_env:
for key in sorted(extra_env):
env_assignments.append("%s=%s"
% (key, shlex.quote(
str(extra_env[key]))))
if egress_proxy:
env_assignments += [
"HTTPS_PROXY=%s" % shlex.quote(egress_proxy),
"HTTP_PROXY=%s" % shlex.quote(egress_proxy),
"NO_PROXY=localhost,127.0.0.1"]
env_assignments = _env_assignments(extra_env, egress_proxy)
inner = [shlex.quote(str(arg)) for arg in inner_cmd]
if env_assignments:
tokens += ["env"] + env_assignments + inner
@@ -198,13 +247,10 @@ def parse_keyscan_output(text: str, host: str = "127.0.0.1") -> list:
def write_known_hosts(path: str, key_lines: list) -> None:
"""原子写 known_hosts:先写同目录临时文件再 rename,
"""原子写 known_hosts:随机 tmp 名 + fsync 后 rename
(feat-chv-closeup,P2-18:旧固定 `<path>.tmp` 无 fsync),
文件权限 0600(只含公钥行,仍按敏感文件对待)。"""
directory = os.path.dirname(path)
if directory:
os.makedirs(directory, exist_ok=True)
tmp_path = path + ".tmp"
with open(tmp_path, "w") as fh:
fh.write("\n".join(key_lines) + "\n")
os.chmod(tmp_path, 0o600)
os.replace(tmp_path, path)
atomic_write_text(path, "\n".join(key_lines) + "\n", mode=0o600)
+16 -4
View File
@@ -1047,12 +1047,18 @@ class DefensiveBranchTest(CloseupCase):
marker = dirty_marker_path(self.state_dir, "chv-abc123")
self.assertTrue(os.path.exists(marker))
def test_write_dirty_marker_failure_swallowed(self):
def test_write_dirty_marker_failure_audited_not_raised(self):
# feat-chv-closeup(P2-17):脏标记写失败仍不抛(stop
# 契约不变),但不再全静默——返回 False + 独立审计
# dirty_marker_write_failed。
blocker = os.path.join(self.dir, "blocker2")
with open(blocker, "w") as fh:
fh.write("x")
ex = self.make_exec(state_dir=blocker)
ex._write_dirty_marker("chv-abc123", True) # 不抛
ok = ex._write_dirty_marker("chv-abc123", True) # 不抛
self.assertFalse(ok)
self.assertIn("dirty_marker_write_failed",
self._audit_events())
def test_stop_socket_remove_failure_swallowed(self):
ex = self.make_exec()
@@ -1290,7 +1296,10 @@ class ExecutorLastLinesTest(CloseupCase):
ex.wait_guest_ready(self._task())
self.assertEqual(ctx.exception.kind, "guest_not_ready")
def test_record_net_entries_write_failure_only_warns(self):
def test_record_net_entries_write_failure_audited_and_reported(self):
# feat-chv-closeup(P2-19):登记失败不再“只 warning”——
# 返回 False + 独立审计 net_registry_write_failed,调用方
# _boot_vm 据此拒起(见 test_chv_residual)。
ex = self.make_exec()
blocker = os.path.join(self.state_dir, "chv-x")
os.makedirs(self.state_dir, exist_ok=True)
@@ -1298,9 +1307,12 @@ class ExecutorLastLinesTest(CloseupCase):
fh.write("x")
with self.assertLogs("cloud.cloud_hypervisor",
level="WARNING") as logs:
ex._record_net_entries("chv-x", [{"tap": "tap0"}])
ok = ex._record_net_entries("chv-x", [{"tap": "tap0"}])
self.assertFalse(ok)
self.assertTrue(any("net_registry_write_failed" in line
for line in logs.output))
self.assertIn("net_registry_write_failed",
self._audit_events())
def test_socket_permissions_stat_failure_after_exists(self):
ex = self.make_exec()
+593
View File
@@ -0,0 +1,593 @@
"""CHV 残留收尾轨(feat-chv-closeup)新增测试。
覆盖 SECURITY-REAUDIT §5 第 5 点的逐项闭环:
P2-20(_guest_env_delivered 加锁)、P2-19(net 登记失败
审计 + 拒起)、P2-17(net.json 删失败审计、脏标记写失败
审计 + 返回信号)、P2-18(固定 tmp 名 + 无 fsync 改随机
名原子写 + 双 fsync)、H1 附带(NO_PROXY 合并顺序)、
M4(generate 只读 BUILDINFO.versions、Chromium 单独钉死、
build.sh /app 最小集)。
全部用标准库与 mock,不启动真实 VM、不碰 WSL/真机;
M4 的“真实 lockfile 产出”仍待 V-9 真机构建,本文件只
验“生成端以 BUILDINFO 为唯一来源 + 钉死校验逻辑”,
版本号全部是合成假值,绝不编造真实版本号。
"""
import importlib.util
import json
import os
import subprocess
import sys
import tempfile
import threading
import unittest
from pathlib import Path
from unittest import mock
try:
from cloud.chv.atomic_file import atomic_write_text
from cloud.chv.disk import _save_snapshots
from cloud.chv.executor import CloudHypervisorExecutor
from cloud.chv.hostkeys import dirty_marker_path
from cloud.chv.orphans import _write_registry
from cloud.chv.ssh_transport import (
remote_command,
write_known_hosts,
)
from cloud.sandbox import ChvFailure
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.chv.atomic_file import atomic_write_text
from cloud.chv.disk import _save_snapshots
from cloud.chv.executor import CloudHypervisorExecutor
from cloud.chv.hostkeys import dirty_marker_path
from cloud.chv.orphans import _write_registry
from cloud.chv.ssh_transport import (
remote_command,
write_known_hosts,
)
from cloud.sandbox import ChvFailure
ROOT = Path(__file__).resolve().parents[2]
def _load_script(name):
"""把 chv-guest 下的独立脚本当模块加载(无包结构)。"""
path = ROOT / "chv-guest" / name
spec = importlib.util.spec_from_file_location(
"chv_guest_" + name[:-3], str(path))
module = importlib.util.module_from_spec(spec)
spec.loader.exec_module(module)
return module
class FakeResult:
def __init__(self, returncode=0, stdout="", stderr=""):
self.returncode = returncode
self.stdout = stdout
self.stderr = stderr
class FakeProc:
def __init__(self, alive=True):
self._alive = alive
def poll(self):
return None if self._alive else 1
def terminate(self):
self._alive = False
def wait(self, timeout=None):
return 0
def kill(self):
self._alive = False
def _write(path, content=b"x"):
with open(path, "wb") as fh:
fh.write(content)
os.chmod(path, 0o755)
return path
class ResidualCase(unittest.TestCase):
def setUp(self):
self.tmp = tempfile.TemporaryDirectory()
self.dir = self.tmp.name
self.binary = _write(os.path.join(self.dir, "cloud-hypervisor"))
self.ch_remote = _write(os.path.join(self.dir, "ch-remote"))
self.kernel = _write(os.path.join(self.dir, "vmlinux"))
self.rootfs = _write(os.path.join(self.dir, "disk.raw"),
b"R" * 100)
self.ssh_key = _write(os.path.join(self.dir, "id_chv"))
self.kvm = _write(os.path.join(self.dir, "kvm"))
self.cpuinfo = os.path.join(self.dir, "cpuinfo")
with open(self.cpuinfo, "w") as fh:
fh.write("flags : fpu vmx sse\n")
self.state_dir = os.path.join(self.dir, "state")
self.calls = []
self.api_calls = []
self.audit = []
def tearDown(self):
self.tmp.cleanup()
def make_exec(self, env_extra=None, **overrides):
env = {"CHV_EGRESS_PROXY": "http://10.0.0.1:8888",
"CHV_HOST_KEY_MODE": "keyscan",
"CHV_SSH_FORWARD": "1"}
if env_extra:
env.update(env_extra)
kwargs = {
"binary": self.binary, "ch_remote": self.ch_remote,
"kernel": self.kernel, "rootfs": self.rootfs,
"ssh_key": self.ssh_key, "kvm_path": self.kvm,
"cpuinfo_path": self.cpuinfo, "state_dir": self.state_dir,
"runner": self._default_runner,
"launcher": self._default_launcher,
"api_client": self._default_api,
"sleeper": lambda s: None, "env": env}
kwargs.update(overrides)
ex = CloudHypervisorExecutor(**kwargs)
ex.audit_callback = lambda e, t, d, k=None: self.audit.append(
(e, t.get("id") if isinstance(t, dict) else t, d, k))
return ex
def _default_runner(self, cmd, timeout=None):
self.calls.append(list(cmd))
return FakeResult(0, "", "")
def _default_api(self, socket_path, method, path, payload=None):
self.api_calls.append((method, path, payload))
return (200, {})
def _default_launcher(self, cmd):
return FakeProc(alive=True)
def _task(self, task_id="abc123"):
return {"id": task_id, "prompt": "做一个加法函数",
"project_id": "proj1"}
def _audit_events(self):
return [e for e, _t, _d, _k in self.audit]
# -- P2-20 ---------------------------------------------------------
def test_guest_env_lock_exists_and_helpers_roundtrip(self):
ex = self.make_exec()
self.assertIsInstance(ex._guest_env_lock, type(threading.Lock()))
ex._mark_guest_env_delivered("chv-a", {"SUB2API_API_KEY"})
self.assertEqual(ex._delivered_secrets("chv-a"),
{"SUB2API_API_KEY"})
ex._mark_guest_env_delivered("chv-a", set())
self.assertEqual(ex._delivered_secrets("chv-a"), set())
ex._mark_guest_env_delivered("chv-a", {"K"})
ex._clear_guest_env_delivered("chv-a")
self.assertEqual(ex._delivered_secrets("chv-a"), set())
def test_guest_env_concurrent_mark_read_clear(self):
ex = self.make_exec()
errors = []
def worker(n):
try:
for i in range(50):
name = f"chv-w{i % 4}"
ex._mark_guest_env_delivered(name, {f"K{n}"})
got = ex._delivered_secrets(name)
self.assertIsInstance(got, set)
ex._clear_guest_env_delivered(name)
except Exception as exc: # noqa: BLE001 — 汇总后断言
errors.append(exc)
threads = [threading.Thread(target=worker, args=(n,))
for n in range(8)]
for t in threads:
t.start()
for t in threads:
t.join()
self.assertEqual(errors, [])
# 结束后确定性收尾:最后一次 mark 必须可见,带密钥的
# build_ssh_command 不再报 secret_env_not_delivered。
ex._mark_guest_env_delivered("chv-final", {"SUB2API_API_KEY"})
cmd = ex.build_ssh_command(
self._task("final"), ["true"],
extra_env={"SUB2API_API_KEY": "x"})
self.assertEqual(cmd[0], "ssh")
self.assertNotIn("x", cmd)
def test_build_ssh_command_sees_delivered_after_mark(self):
ex = self.make_exec()
ex._mark_guest_env_delivered("chv-abc123", {"SUB2API_API_KEY"})
cmd = ex.build_ssh_command(
self._task(), ["true"],
extra_env={"SUB2API_API_KEY": "x"})
self.assertNotIn("x", cmd)
# -- P2-19 ---------------------------------------------------------
def test_record_net_entries_failure_returns_false_and_audits(self):
ex = self.make_exec()
blocker = os.path.join(self.state_dir, "chv-x")
os.makedirs(self.state_dir, exist_ok=True)
with open(blocker, "w") as fh:
fh.write("x")
with self.assertLogs("cloud.cloud_hypervisor",
level="WARNING") as logs:
ok = ex._record_net_entries("chv-x", [{"tap": "tap0"}])
self.assertFalse(ok)
self.assertTrue(any("net_registry_write_failed" in line
for line in logs.output))
self.assertIn("net_registry_write_failed",
self._audit_events())
def test_boot_vm_refuses_when_net_registry_fails(self):
ex = self.make_exec(env_extra={
"CHV_NET_JSON": '[{"tap": "tap0", "ip": "10.0.0.2"}]'})
os.makedirs(self.state_dir, exist_ok=True)
with open(os.path.join(self.state_dir, "chv-abc123"),
"w") as fh:
fh.write("x")
with self.assertRaises(ChvFailure) as ctx:
ex._boot_vm(self._task(), "chv-abc123", self.rootfs,
None, [{"tap": "tap0", "ip": "10.0.0.2"}])
self.assertEqual(ctx.exception.kind,
"net_registry_write_failed")
paths = [p for _m, p, _pl in self.api_calls]
self.assertIn("/vm.create", paths)
self.assertNotIn("/vm.boot", paths)
self.assertIn("net_registry_write_failed",
self._audit_events())
# -- P2-17 ---------------------------------------------------------
def test_cleanup_net_remove_failure_audited(self):
ex = self.make_exec()
vm_dir = os.path.join(self.state_dir, "chv-abc123")
os.makedirs(vm_dir, exist_ok=True)
os.makedirs(os.path.join(vm_dir, "net.json"), exist_ok=True)
with open(os.path.join(self.state_dir, "chv-abc123.sock"),
"w") as fh:
fh.write("")
self.assertTrue(ex.stop_vm(self._task()))
marker = dirty_marker_path(self.state_dir, "chv-abc123")
self.assertTrue(os.path.exists(marker))
self.assertIn("vm_net_registry_delete_failed",
self._audit_events())
def test_write_dirty_marker_failure_returns_false_and_audits(self):
blocker = os.path.join(self.dir, "blocker3")
with open(blocker, "w") as fh:
fh.write("x")
ex = self.make_exec(state_dir=blocker)
ok = ex._write_dirty_marker("chv-abc123", True) # 不抛
self.assertFalse(ok)
self.assertIn("dirty_marker_write_failed",
self._audit_events())
def test_write_dirty_marker_success_roundtrip(self):
ex = self.make_exec()
self.assertTrue(ex._write_dirty_marker("chv-abc123", True))
marker = dirty_marker_path(self.state_dir, "chv-abc123")
self.assertTrue(os.path.exists(marker))
self.assertTrue(ex._write_dirty_marker("chv-abc123", False))
self.assertFalse(os.path.exists(marker))
# -- P2-18 ---------------------------------------------------------
def test_atomic_write_no_fixed_tmp_and_fsync(self):
target = os.path.join(self.dir, "probe.json")
calls = {"n": 0}
real_fsync = os.fsync
def _counting(fd):
calls["n"] += 1
return real_fsync(fd)
with mock.patch.object(os, "fsync", side_effect=_counting):
atomic_write_text(target, '{"a": 1}')
with open(target) as fh:
self.assertEqual(fh.read(), '{"a": 1}')
self.assertEqual(oct(os.stat(target).st_mode & 0o777), "0o600")
leftovers = [p for p in os.listdir(self.dir)
if ".tmp" in p]
self.assertEqual(leftovers, [])
# 文件 fsync + 目录 fsync,至少两次。
self.assertGreaterEqual(calls["n"], 2)
def test_net_record_leaves_no_tmp_files(self):
ex = self.make_exec()
self.assertTrue(ex._record_net_entries(
"chv-abc123", [{"tap": "tap0"}]))
vm_dir = os.path.join(self.state_dir, "chv-abc123")
with open(os.path.join(vm_dir, "net.json")) as fh:
self.assertEqual(json.load(fh), [{"tap": "tap0"}])
self.assertEqual([p for p in os.listdir(vm_dir)
if ".tmp" in p], [])
def test_known_hosts_atomic_no_fixed_tmp(self):
path = os.path.join(self.state_dir, "chv-abc123",
"known_hosts")
write_known_hosts(path, ["127.0.0.1 ssh-ed25519 AAAA"])
with open(path) as fh:
self.assertEqual(fh.read(),
"127.0.0.1 ssh-ed25519 AAAA\n")
self.assertEqual(oct(os.stat(path).st_mode & 0o777), "0o600")
vm_dir = os.path.join(self.state_dir, "chv-abc123")
self.assertEqual([p for p in os.listdir(vm_dir)
if ".tmp" in p], [])
def test_snapshot_save_no_fixed_tmp(self):
_save_snapshots(self.state_dir, "chv-abc123",
[{"url": "file://x"}])
path = os.path.join(self.state_dir, "chv-abc123",
"snapshots.json")
with open(path) as fh:
self.assertEqual(json.load(fh), [{"url": "file://x"}])
vm_dir = os.path.join(self.state_dir, "chv-abc123")
self.assertEqual([p for p in os.listdir(vm_dir)
if ".tmp" in p], [])
def test_orphan_registry_no_fixed_tmp(self):
path = os.path.join(self.state_dir, "pool.json")
_write_registry(path, {"chv-a": {"vm": "chv-a"}})
with open(path) as fh:
self.assertEqual(json.load(fh),
{"chv-a": {"vm": "chv-a"}})
self.assertEqual([p for p in os.listdir(self.state_dir)
if ".tmp" in p], [])
# -- H1 附带:NO_PROXY ---------------------------------------------
def _env_assignments(self, tokens):
"""remote_command 返回的 env 赋值(KEY=VALUE 形式)。"""
out = {}
for tok in tokens:
if tok == "env":
continue
if "=" in tok and not tok.startswith("'"):
key, _, value = tok.partition("=")
if key.isidentifier() or key in (
"NO_PROXY", "no_proxy", "HTTPS_PROXY",
"HTTP_PROXY", "https_proxy", "http_proxy"):
out.setdefault(key, []).append(
value.strip("'"))
return out
def test_no_proxy_merged_with_egress(self):
tokens = remote_command(
["true"], egress_proxy="http://10.0.0.1:8888",
extra_env={"NO_PROXY": "internal.example.com,10.0.0.0/8"})
assigns = self._env_assignments(tokens)
self.assertEqual(len(assigns.get("NO_PROXY", [])), 1)
merged = assigns["NO_PROXY"][0]
for part in ("internal.example.com", "10.0.0.0/8",
"localhost", "127.0.0.1"):
self.assertIn(part, merged)
self.assertEqual(
assigns["HTTPS_PROXY"], ["http://10.0.0.1:8888"])
def test_no_proxy_lowercase_merged_and_not_duplicated(self):
tokens = remote_command(
["true"], egress_proxy="http://10.0.0.1:8888",
extra_env={"no_proxy": "gitea.tose.top",
"NO_PROXY": "localhost"})
assigns = self._env_assignments(tokens)
self.assertEqual(len(assigns.get("NO_PROXY", [])), 1)
self.assertIn("gitea.tose.top", assigns["NO_PROXY"][0])
self.assertNotIn("no_proxy", assigns)
# localhost 去重后只出现一次。
self.assertEqual(
assigns["NO_PROXY"][0].split(",").count("localhost"), 1)
def test_no_proxy_preserved_without_egress(self):
tokens = remote_command(
["true"], extra_env={"NO_PROXY": "a.example"})
assigns = self._env_assignments(tokens)
self.assertEqual(assigns.get("NO_PROXY"), ["a.example"])
def test_user_proxy_dropped_when_egress_set(self):
tokens = remote_command(
["true"], egress_proxy="http://10.0.0.1:8888",
extra_env={"HTTPS_PROXY": "http://evil:1",
"https_proxy": "http://evil:1",
"HTTP_PROXY": "http://evil:1"})
assigns = self._env_assignments(tokens)
self.assertEqual(
assigns["HTTPS_PROXY"], ["http://10.0.0.1:8888"])
self.assertEqual(
assigns["HTTP_PROXY"], ["http://10.0.0.1:8888"])
self.assertNotIn("https_proxy", assigns)
self.assertNotIn("http_proxy", assigns)
# -- M4 供应链 ----------------------------------------------------------
class LockfileTest(unittest.TestCase):
"""合成 BUILDINFO 干跑:版本全是假值(如 git 9.9.9),
只验“来源是 BUILDINFO、不是宿主现场”与钉死校验逻辑,
真实 lockfile 产出仍待 V-9 真机构建。"""
def setUp(self):
self.tmp = tempfile.TemporaryDirectory()
self.dir = self.tmp.name
def tearDown(self):
self.tmp.cleanup()
def _synth_buildinfo(self, chromium="chromium_headless_shell-1248",
pinned_rev="1248"):
dpkg = ["git 9.9.9-9ubuntu9", "openssh-server 1:9.9",
"openssh-client 1:9.9", "cloud-init 9.9",
"python3 9.9", "python3-pip 9.9", "curl 9.9",
"ca-certificates 9.9", "xz-utils 9.9",
"fonts-wqy-microhei 9.9", "fontconfig 9.9"]
info = {
"tag": "chv-guest-24.04-0.164.0-alpha.9-abc1234",
"ubuntu": "24.04",
"codex_version": "0.164.0-alpha.9",
"repo_commit": "abc1234",
"gt_version": "1.0.0",
"built_at": "2026-11-07T00:00:00Z",
"contents": ["git", "openssh-server", "openssh-client",
"cloud-init", "python3", "python3-pip",
"curl", "ca-certificates", "xz-utils",
"fonts-wqy-microhei", "fontconfig", "codex",
"codex-code-mode-host", "codex-resources/bwrap",
"/app/host.py", "gt", "chv-sync",
"playwright", "chromium-headless-shell"],
"versions": {"dpkg": dpkg,
"pip": ["playwright==9.9.9"],
"gt": ["gt version 1.0.0"],
"chromium": [chromium]},
"chromium_pinned_rev": pinned_rev,
"image_sha256": "0" * 64,
}
path = os.path.join(self.dir, "BUILDINFO.json")
with open(path, "w") as fh:
json.dump(info, fh)
return path
def test_generate_uses_buildinfo_not_host(self):
gen = _load_script("generate_lockfile.py")
buildinfo = self._synth_buildinfo()
lockfile = os.path.join(self.dir, "lockfile.json")
self.assertEqual(gen.generate(buildinfo, lockfile), 0)
with open(lockfile) as fh:
data = json.load(fh)
self.assertEqual(data["unresolved"], [])
# 假值进 lockfile:来源是 BUILDINFO,不是宿主 dpkg。
self.assertIn("git 9.9.9-9ubuntu9",
data["versions"]["dpkg"])
self.assertIn("playwright 9.9.9",
data["versions"]["pip"])
self.assertIn("chromium-headless-shell "
"chromium_headless_shell-1248",
data["versions"]["declared"])
self.assertEqual(
gen.verify(lockfile, buildinfo), 0)
def test_generate_verify_rejects_wrong_pinned_rev(self):
gen = _load_script("generate_lockfile.py")
buildinfo = self._synth_buildinfo(
chromium="chromium_headless_shell-9999",
pinned_rev="1248")
lockfile = os.path.join(self.dir, "lockfile.json")
self.assertEqual(gen.generate(buildinfo, lockfile), 0)
# 生成本身不设门(只记版本),--verify 才拦钉死不符。
self.assertEqual(gen.verify(lockfile, buildinfo), 1)
def test_check_lockfile_pinned_rev(self):
check = _load_script("check_lockfile.py")
ok = self._synth_buildinfo()
self.assertEqual(
check.main(["check_lockfile.py", ok]), 0)
bad = self._synth_buildinfo(
chromium="chromium_headless_shell-9999",
pinned_rev="1248")
self.assertEqual(
check.main(["check_lockfile.py", bad]), 1)
def test_buildinfo_records_and_enforces_pin(self):
buildinfo = _load_script("buildinfo.py")
out = os.path.join(self.dir, "out")
os.makedirs(out)
for name, content in (
("disk.raw", b"D"), ("vmlinux", b"K")):
with open(os.path.join(out, name), "wb") as fh:
fh.write(content)
for name, lines in (
("guest-versions-dpkg.txt",
["git 9.9.9-9ubuntu9"]),
("guest-versions-pip.txt",
["playwright==9.9.9"]),
("guest-versions-gt.txt", ["gt version 1.0.0"]),
("guest-versions-chromium.txt",
["chromium_headless_shell-1248"])):
with open(os.path.join(out, name), "w") as fh:
fh.write("\n".join(lines) + "\n")
args = [out, "chv-guest-24.04-0.164.0-alpha.9-abc1234",
"24.04", "0.164.0-alpha.9", "abc1234", "0" * 64,
"1", "1248"]
info = buildinfo.build(*args[0:7], "1248")
self.assertEqual(info["chromium_pinned_rev"], "1248")
self.assertEqual(buildinfo.main(["buildinfo.py"] + args), 0)
wrong = list(args)
wrong[-1] = "9999"
self.assertEqual(
buildinfo.main(["buildinfo.py"] + wrong), 1)
# 未给钉死值:记 null,不强制。
info2 = buildinfo.build(*args[0:7])
self.assertIsNone(info2["chromium_pinned_rev"])
class BuildScriptMinimalAppTest(unittest.TestCase):
"""build.sh 干跑:/app 只剩 host.py(不用 cloud/ 也能
干跑通过),Chromium 钉死值透传进 buildinfo 回显。"""
def setUp(self):
self.tmp = tempfile.TemporaryDirectory()
self.dir = self.tmp.name
self.bin_dir = os.path.join(self.dir, "bin")
os.makedirs(os.path.join(self.bin_dir, "codex-resources"))
for name in ("codex", "codex-code-mode-host"):
p = os.path.join(self.bin_dir, name)
with open(p, "w") as fh:
fh.write("#!/bin/sh\necho \"codex-cli 0.164.0-alpha.9\"\n")
os.chmod(p, 0o755)
self.repo = os.path.join(self.dir, "repo")
os.makedirs(self.repo)
# 故意不建 cloud/:最小集收紧后 build.sh 不再要求它。
with open(os.path.join(self.repo, "host.py"), "w") as fh:
fh.write("# host\n")
subprocess.run(["git", "init", "-q", self.repo],
check=True, capture_output=True)
subprocess.run(["git", "-C", self.repo, "config",
"user.email", "t@t"], check=True,
capture_output=True)
subprocess.run(["git", "-C", self.repo, "config",
"user.name", "t"], check=True,
capture_output=True)
subprocess.run(["git", "-C", self.repo, "add", "-A"],
check=True, capture_output=True)
subprocess.run(["git", "-C", self.repo, "commit", "-qm",
"init"], check=True, capture_output=True)
self.repo_commit = subprocess.run(
["git", "-C", self.repo, "rev-parse", "--short", "HEAD"],
check=True, capture_output=True, text=True).stdout.strip()
self.script = str(ROOT / "chv-guest" / "build.sh")
def tearDown(self):
self.tmp.cleanup()
def _run(self, args, env_extra=None):
env = dict(os.environ)
env.update({"CODEX_BIN_DIR": self.bin_dir,
"UBUNTU_VERSION": "24.04",
"CODEX_VERSION": "0.164.0-alpha.9",
"REPO_COMMIT": self.repo_commit,
"UBUNTU_IMAGE_SHA256": "0" * 64,
"REPO_DIR": self.repo})
if env_extra:
env.update(env_extra)
return subprocess.run(["bash", self.script] + args,
check=False,
capture_output=True, text=True, env=env)
def test_dry_run_without_cloud_dir_and_pin_echo(self):
out_dir = os.path.join(self.dir, "out")
r = self._run(["--dry-run", out_dir],
{"CHROMIUM_HEADLESS_SHELL_REV": "1248"})
self.assertEqual(r.returncode, 0, r.stderr)
self.assertNotIn("cloud:/app", r.stdout)
self.assertIn("host.py:/app", r.stdout)
self.assertIn("1248", r.stdout)
self.assertFalse(os.path.exists(
os.path.join(out_dir, "disk.raw")))
def test_build_sh_text_has_no_cloud_copy(self):
with open(self.script) as fh:
text = fh.read()
self.assertNotIn('"$REPO_DIR/cloud:/app/"', text)
self.assertIn('"$REPO_DIR/host.py:/app/"', text)
if __name__ == "__main__":
unittest.main()