part 4 缺陷 3：并发 cell 撞毁——唯一身份 + run_cell 接线

RETEST3 缺陷 3：固定 ve-chv-cell 名 + 固定 10.200.0.1/24，第二个 cell 在 nspawn 建 veth 时即 File exists/Address already assigned。 - CellNet + allocate_cell_net()：每次 run_cell 在 flock 锁下（/run/chv/.cell-alloc.lock，30 秒超时）分配唯一身份： machine chv-cell-<xx>（11 字符）、宿主 ve-chv-cell-<xx> （14 字符，接口名 15 字符上限内）、10.200.<xx>.0/24；扫描活接口 + 地址 + machinectl 在册名，避让注销滞后的下标。 - 配网命令表（host/cell/cleanup）与 build_nspawn_argv 接受 net/cell_net 参数；None 时仍用旧常量（argv 子命令与旧测试）。 - run_cell 重写：分配身份 -> 生成专属文件 -> 起步门 payload -> 等 veth（20 秒）-> 宿主配网 -> 轮询 init pid（10 秒） -> 写 pid 文件 -> 单元内配网 -> 开门 -> 等退出码； nspawn 已死返回其退出码；finally 清 NAT/pid/专属文件。共享网络模式跳过门与身份（无配网可等）。 - broker：pid ns 对照改多 cell 版（peer_allowed_multi + _read_cell_pids：cell-<machine>.pid glob + 兼容旧 cell.pid；对端 ns 与任一活单元同 ns 即放行；残留 pid 文件只跳过，全部读不到才拒绝，不降级）。新测试：CellAllocTest、CellPart4NetworkTest、 CellPart4ArgvTest（cell_net 两法）、BrokerMultiTest。
2026-11-07 10:30:30 +00:00
1 parent 2be69af92c
commit 7a227323f2
@@ -835,6 +835,25 @@ class CellPart4ArgvTest(unittest.TestCase):



+    def test_cell_net_machine_and_no_legacy_name(self):
+        net = CELL.CellNet(index=3, machine="chv-cell-03",
+                           host_iface="ve-chv-cell-03",
+                           host_ip="10.200.3.1", cell_ip="10.200.3.2",
+                           subnet="10.200.3.0/24")
+        argv = CELL.build_nspawn_argv(["true"], cell_net=net)
+        self.assertIn("--machine=chv-cell-03", argv)
+        self.assertNotIn("--machine=chv-cell", argv)
+        # 旧常量身份仍是默认（argv 子命令与旧测试不改）。
+        self.assertIn("--machine=chv-cell",
+                      CELL.build_nspawn_argv(["true"]))
+
+    def test_legacy_cell_net(self):
+        net = CELL.legacy_cell_net()
+        self.assertEqual(net.machine, "chv-cell")
+        self.assertEqual(net.host_iface, "ve-chv-cell")
+        self.assertEqual(net.subnet, "10.200.0.0/24")
+
+
 class CellFilesTest(unittest.TestCase):
    """part 4：单元专属 resolv.conf/hosts 生成。"""

@@ -890,5 +909,166 @@ class CellFilesTest(unittest.TestCase):
                    hosts_file=os.path.join(tmp, "missing-hosts"))


+class CellPart4NetworkTest(unittest.TestCase):
+    """part 4：配网命令表走唯一身份。"""
+
+    def _net(self, index=5):
+        return CELL.CellNet(
+            index=index, machine="chv-cell-%02x" % index,
+            host_iface="ve-chv-cell-%02x" % index,
+            host_ip="10.200.%d.1" % index,
+            cell_ip="10.200.%d.2" % index,
+            subnet="10.200.%d.0/24" % index)
+
+    def test_host_commands_use_cell_net(self):
+        net = self._net(5)
+        flat = [" ".join(c)
+                for c in CELL.host_network_commands("enp0s5", net)]
+        self.assertIn("ip addr add 10.200.5.1/24 dev ve-chv-cell-05",
+                      flat)
+        self.assertTrue(any(
+            "iptables -t nat -A POSTROUTING -s 10.200.5.0/24 "
+            "-o enp0s5 -j MASQUERADE" in line for line in flat))
+        # 旧常量仍是默认。
+        legacy = [" ".join(c)
+                  for c in CELL.host_network_commands("enp0s5")]
+        self.assertIn("ip addr add 10.200.0.1/24 dev ve-chv-cell",
+                      legacy)
+
+    def test_cell_commands_use_cell_net(self):
+        net = self._net(7)
+        flat = [" ".join(c)
+                for c in CELL.cell_network_commands(4321, net)]
+        self.assertIn("nsenter -t 4321 -n ip addr add 10.200.7.2/24 "
+                      "dev host0", flat)
+        self.assertIn("nsenter -t 4321 -n ip route add default via "
+                      "10.200.7.1", flat)
+
+
+class CellAllocTest(unittest.TestCase):
+    """part 4：锁下分配唯一 cell 身份。"""
+
+    class _FakeProc:
+        def __init__(self, returncode=0, stdout=""):
+            self.returncode = returncode
+            self.stdout = stdout
+
+    def _fake_run(self, link_out="", addr_out="", machinectl_out="",
+                  machinectl_rc=0):
+        def run(cmd, **kwargs):
+            if cmd[:3] == ["ip", "-o", "link"]:
+                return self._FakeProc(0, link_out)
+            if cmd[:3] == ["ip", "-o", "addr"]:
+                return self._FakeProc(0, addr_out)
+            if cmd[0] == "machinectl":
+                return self._FakeProc(machinectl_rc, machinectl_out)
+            raise AssertionError("意外命令：%r" % (cmd,))
+        return run
+
+    def test_allocates_lowest_free_index(self):
+        link = ("2: ve-chv-cell-00@if2: <UP> mtu 1500\n"
+                "3: ve-chv-cell-01@if3: <UP> mtu 1500\n")
+        addr = ("2: ve-chv-cell-00 inet 10.200.0.1/24 scope global\n"
+                "3: ve-chv-cell-01 inet 10.200.1.1/24 scope global\n")
+        with tempfile.TemporaryDirectory() as tmp:
+            lock = os.path.join(tmp, "lock")
+            net = CELL.allocate_cell_net(
+                run=self._fake_run(link, addr), lock_path=lock)
+        self.assertEqual(net.index, 2)
+        self.assertEqual(net.machine, "chv-cell-02")
+        self.assertEqual(net.host_iface, "ve-chv-cell-02")
+        self.assertEqual(net.host_ip, "10.200.2.1")
+        self.assertEqual(net.cell_ip, "10.200.2.2")
+        self.assertEqual(net.subnet, "10.200.2.0/24")
+        self.assertLessEqual(len(net.host_iface), 15)
+
+    def test_machinectl_registered_index_skipped(self):
+        # 接口已消失、machine 仍在注销：避让该下标。
+        run = self._fake_run(
+            "", "", machinectl_out="chv-cell-00  1234  ...\n")
+        with tempfile.TemporaryDirectory() as tmp:
+            net = CELL.allocate_cell_net(
+                run=run, lock_path=os.path.join(tmp, "lock"))
+        self.assertEqual(net.index, 1)
+
+    def test_machinectl_missing_proceeds(self):
+        def run(cmd, **kwargs):
+            if cmd[0] == "machinectl":
+                raise OSError("无此命令")
+            return self._FakeProc(0, "")
+        with tempfile.TemporaryDirectory() as tmp:
+            net = CELL.allocate_cell_net(
+                run=run, lock_path=os.path.join(tmp, "lock"))
+        self.assertEqual(net.index, 0)
+
+    def test_ip_failure_is_loud(self):
+        def run(cmd, **kwargs):
+            return self._FakeProc(1, "")
+        with tempfile.TemporaryDirectory() as tmp:
+            with self.assertRaises(RuntimeError):
+                CELL.allocate_cell_net(
+                    run=run, lock_path=os.path.join(tmp, "lock"))
+
+
+class BrokerMultiTest(unittest.TestCase):
+    """part 4：broker 对多 cell 的 pid ns 对照。"""
+
+    def _fake_proc(self, root, mapping):
+        for pid, ns in mapping.items():
+            d = os.path.join(root, str(pid), "ns")
+            os.makedirs(d)
+            os.symlink("pid:[%s]" % ns, os.path.join(d, "pid"))
+
+    def test_peer_allowed_if_matches_any_cell(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            self._fake_proc(tmp, {100: "111", 101: "222", 200: "222"})
+            ok, _ = BROKER.peer_allowed_multi(
+                1000, 200, [100, 101], proc_root=tmp)
+            self.assertTrue(ok)
+
+    def test_peer_denied_if_matches_none(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            self._fake_proc(tmp, {100: "111", 101: "222", 200: "333"})
+            ok, reason = BROKER.peer_allowed_multi(
+                1000, 200, [100, 101], proc_root=tmp)
+            self.assertFalse(ok)
+            self.assertIn("pid ns", reason)
+
+    def test_stale_pid_file_skipped_not_fatal(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            # 100 已退出（无 /proc 项），101 活着且同 ns。
+            self._fake_proc(tmp, {101: "222", 200: "222"})
+            ok, _ = BROKER.peer_allowed_multi(
+                1000, 200, [100, 101], proc_root=tmp)
+            self.assertTrue(ok)
+
+    def test_all_ns_unreadable_rejects(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            self._fake_proc(tmp, {200: "222"})
+            ok, _ = BROKER.peer_allowed_multi(
+                1000, 200, [100, 101], proc_root=tmp)
+            self.assertFalse(ok)
+
+    def test_no_pid_files_uid_only(self):
+        ok, reason = BROKER.peer_allowed_multi(1000, 200, [])
+        self.assertTrue(ok)
+        self.assertIn("未做 pid ns 对照", reason)
+
+    def test_read_cell_pids_glob_and_legacy(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            with open(os.path.join(tmp, "cell-chv-cell-00.pid"),
+                      "w") as fh:
+                fh.write("100")
+            with open(os.path.join(tmp, "cell-chv-cell-01.pid"),
+                      "w") as fh:
+                fh.write("101")
+            with open(os.path.join(tmp, "cell.pid"), "w") as fh:
+                fh.write("101")
+            with open(os.path.join(tmp, "cell-chv-cell-02.pid"),
+                      "w") as fh:
+                fh.write("not-a-pid")
+            self.assertEqual(BROKER._read_cell_pids(tmp), [100, 101])
+
+
 if __name__ == "__main__":
    unittest.main()