feat(密钥): 纯 Python 静态提取 XTEA 密钥 —— 去掉最后一个 Node 依赖 (6/6 构建与神谕逐字一致)
做了什么 - 新增 tcaptcha/tdc_opcodes.py: 按构建从解释器源码推导 opcode 表。 实测 opcode 编号**每份构建整体重排**(同一源码 pc0 的 SET_STACK_LEN 在 6 份构建里 = 53/19/32/57/55/1, JUMP = 60/17/42/58/21/2); 旧固化表只对 core_js 有效 —— 用它反汇编其它构建会在 pc0 撞上 RET_TRUE, 得到"空反汇编"的假象。 做法: 定位主循环 X[Y[Z++]]() 的 handler 数组 -> JS 感知切分 -> handler 源码规范化 (去空白 + 局部变量按首次出现重命名) 取指纹 -> 与内置 58 条 core_js 参考指纹匹配; 数组下标即该构建的 opcode, argc 由语义名取 (与 handler 体内 x[y++] 次数实测一致)。 - 新增 tcaptcha/tdc_key.py: 纯 Python XTEA 密钥静态还原器 (零 Node / 零 pc 硬编码)。 反汇编 -> 定位含 delta 的 XTEA 闭包 -> 用"两层捕获链"选出真密钥数组: 字闭包捕获的 B 帧寄存器, 经"定义 B 的那个闭包"的 caps 映射到 G 帧后须 == XTEA 闭包 捕获的寄存器; 借此排除脚本里并行计算的第二套诱饵字 -> 读 4 个 (源串, 偏移, 下标) -> 折入轮加盐。 - tools/tdc_disasm.py: disassemble_cfg/_decode_one 支持注入按构建推导的 opcode 表 (names/arity/dynamic), 默认仍回落 core_js 表; 控制流 opcode 改为按语义名定位。 - tools/tdc_key_pure.py: CLI (-v 打印推导过程 / --opcodes 打印 opcode 表 / --json)。 - tcaptcha/crypto.py: extract_xtea_key 改为"纯 Python 主路径 + Node 插桩兜底"。 - 新增 tcaptcha/tests/test_tdc_key.py 并接入 selftest; 新增入库 fixture tcaptcha/tests/fixtures/tdc_fresh_2026-10-11.js (现场新抓的 tdc.js)。 为什么 - AGENTS.md §2.4 分层交付 / §1.1: 去掉最后一个 Node 依赖, 提高并发与抗发版能力; - §5.7 负结果也是资产: 显式撤回"opcode 数值跨构建固定"与"密钥只能靠 Node 插桩拿"两条旧结论。 怎么验证 - 6 份构建纯 Python 结果 == Node 插桩 extract_xtea_key, **逐字一致**: core_js / root / 1ms tdc.js / tdc_live / tdc_latest / 2026-10-11 现场新抓 tdc_fresh。 - opcode 表 6/6 构建 58/58 handler 精确匹配, 0 未识别。 - .venv/bin/python -m tcaptcha.selftest 全绿 (含既有逐字节对拍 test_xtea_key)。 - 把 node_bin 指向不存在的程序时 generate_telemetry 仍产出 collect (证明已零 Node)。
This commit is contained in:
yml2213
@@ -9,7 +9,9 @@
|
||||
| 工具 | 路径 | 核心功能 | 解决什么问题 |
|
||||
| :--- | :--- | :--- | :--- |
|
||||
| **Webpack 解包索引器** | [`tools/unpack_webpack.js`](file:///Users/yml/reverse/hello_js_reverse_skill/tools/unpack_webpack.js) | 将 230KB 单文件 bundle 拆解为 122 个独立模块，建立加签/风控关键词索引 | 圈定加签核心，剔除 80% 开源库与 UI 噪音 (L1 解构) |
|
||||
| **Chaos VM 字节码反汇编器 (纯 Python)** | [`tools/tdc_disasm.py`](file:///Users/yml/reverse/hello_js_reverse_skill/tools/tdc_disasm.py) | 纯 Python 提取 45,636 字节码 + 可达性 CFG 反汇编（26,168 条可达指令，**0 处非法 opcode**）；opcode 表由 VM 源码 AST 推导 | 彻底透明化 TDC 字节码调度器 (L2/L3 解构) |
|
||||
| **Chaos VM 字节码反汇编器 (纯 Python)** | [`tools/tdc_disasm.py`](file:///Users/yml/reverse/hello_js_reverse_skill/tools/tdc_disasm.py) | 纯 Python 提取 45,636 字节码 + 可达性 CFG 反汇编（26,168 条可达指令，**0 处非法 opcode**）；opcode 表可**按构建注入**（见下） | 彻底透明化 TDC 字节码调度器 (L2/L3 解构) |
|
||||
| **按构建推导 opcode 表** | [`tcaptcha/tdc_opcodes.py`](file:///Users/yml/reverse/hello_js_reverse_skill/tcaptcha/tdc_opcodes.py) | 从解释器主循环 `X[Y[Z++]]()` 取出 handler 数组 → 解析 → 规范化指纹 → 跨构建匹配出**该构建的 opcode 表** | **opcode 编号每份构建整体重排**，写死即换脚本全错（6/6 构建 58/58 handler 匹配） |
|
||||
| **纯 Python XTEA 密钥静态提取** | [`tcaptcha/tdc_key.py`](file:///Users/yml/reverse/hello_js_reverse_skill/tcaptcha/tdc_key.py) + [`tools/tdc_key_pure.py`](file:///Users/yml/reverse/hello_js_reverse_skill/tools/tdc_key_pure.py) | 反汇编 → 定位 XTEA 闭包(delta) → 两层捕获链选出真密钥数组 → 读 4 个字 + 折盐 | **零 Node** 拿到该构建 XTEA 密钥；6/6 构建与 Node 插桩逐字一致 |
|
||||
| **深度 Proxy 语义探针** | [`tools/tdc_probe.js`](file:///Users/yml/reverse/hello_js_reverse_skill/tools/tdc_probe.js) | 零 Headless/CDP 依赖，通过深度 Proxy 沙箱执行 `tdc.js`，拦截并记录全部 95 项环境读取 | 毫秒级生成与当前会话时间戳和密钥 100% 自洽的动态 `collect` 与 `eks` (L4 数据层) |
|
||||
|
||||
---
|
||||
@@ -114,4 +116,9 @@ plain = xtea_decrypt_bytes(raw, key).rstrip(b"\\x00")
|
||||
print("成功解密探针生成的遥测载荷，长度:", len(plain))
|
||||
print("明文前 150 字节:", plain[:150].decode("latin1"))
|
||||
'
|
||||
|
||||
# 5. 纯 Python XTEA 密钥提取 (零 Node) + 跨构建回归
|
||||
.venv/bin/python tools/tdc_key_pure.py -v har/1ms_dianxuan/live/tdc_fresh.js
|
||||
.venv/bin/python -m tcaptcha.tests.test_tdc_key # 6 份构建 (含实时新抓) 全部断言
|
||||
.venv/bin/python tools/tdc_key_pure.py --opcodes tcaptcha/tdc_sample.js # 打印该构建的 opcode 表
|
||||
```
|
||||
|
||||
+29
-3
@@ -51,10 +51,36 @@ function __TENCENT_CHAOS_VM(H, z, c, E, I, D, U, C) {
|
||||
|
||||
## 3. 完整 opcode 表（70 槽 / 58 handler / 12 空洞）
|
||||
|
||||
`argc` = handler 体内 `z[H++]` 次数（= 固定操作数字节数）；`57` 是动态长度（见 §4）。
|
||||
**空洞 opcode**：`7, 16, 21, 24, 27, 40, 43, 44, 50, 58, 63, 67` —— 它们**永远不会**出现在可执行字节码里；一旦在反汇编里看到空洞 opcode，说明**解码失步**（旧线性反汇编器就是这么坏的）。
|
||||
`argc` = handler 体内 `z[H++]` 次数（= 固定操作数字节数）；动态长度的那个是 `CREATE_CLOSURE`（见 §4）。
|
||||
|
||||
| opcode | argc | handler 源码（VM 内 G 数组第 i 项） |
|
||||
> ⚠️ **本节表格是"core_js 那一份构建"的编号，不能当跨构建常量用（重要修正，见 §3.1）。**
|
||||
> 下面的 `opcode` 列只对 `tcaptcha/core_js/tdc_sample.js` 成立。
|
||||
> **空洞 opcode**：`7, 16, 21, 24, 27, 40, 43, 44, 50, 58, 63, 67` —— 它们**永远不会**出现在可执行字节码里；
|
||||
> 一旦在反汇编里看到空洞 opcode，说明**解码失步**（旧线性反汇编器就是这么坏的）。
|
||||
|
||||
### 3.1 opcode 编号**每份构建都会整体重排**（实测，5+1 份构建）
|
||||
|
||||
> **推翻早期结论**：早期把上面的 opcode 数值当成 VM 常量（"`RET_TRUE`=19、`SET_STACK_LEN`=53…"）。
|
||||
> 实测是错的：**同一份源码编译出的字节码，pc0 处的 `SET_STACK_LEN` 在 6 份构建里分别是
|
||||
> 53 / 19 / 32 / 57 / 55 / 1**，`JUMP` 分别是 60 / 17 / 42 / 58 / 21 / 2。
|
||||
> 也就是说 **opcode → 语义 的映射表每份脚本都重新洗牌**；写死数值 = 换一份脚本全错
|
||||
> （这也是为什么早先用 core 表去反汇编其它构建会得到"`RET_TRUE` 在 pc0 ⇒ 反汇编为空"的假象）。
|
||||
|
||||
**正确做法**：不在字节码上死磕，而是**从该构建自己的解释器源码里推导 opcode 表** ——
|
||||
见 `tcaptcha/tdc_opcodes.py`：
|
||||
|
||||
1. 定位解释器主循环 `X[Y[Z++]]()`，取出**操作码处理器数组变量** `X`；
|
||||
2. 解析该数组字面量（JS 感知的括号/字符串/正则切分）得到 70 个槽位（含空洞）；
|
||||
3. 把每个 handler 源码**规范化**（去空白 + 局部变量按首次出现重命名 `v1,v2,…`，
|
||||
保留 JS 关键字/内建名与属性名）得到与命名无关的"指纹"；
|
||||
4. 用内置的 58 条参考指纹（取自 core_js，即本表）反查每个槽位的语义名 ——
|
||||
**数组下标即该构建的 opcode**；`argc` 直接按语义名取（与 handler 体内 `x[y++]` 次数实测一致）。
|
||||
|
||||
> 交叉验证：6/6 份构建全部 **58/58 handler 精确匹配**、0 未识别。
|
||||
> 因此 `tools/tdc_disasm.py` / `tools/tdc_lift.py` 现在都支持**注入按构建推导的 opcode 表**
|
||||
> （`disassemble_cfg(z, names=…, arity=…, dynamic=…)`）；不传才回落到本表（core_js）。
|
||||
|
||||
| opcode | argc | handler 源码（core_js 的 VM 内 G 数组第 i 项） |
|
||||
| --- | --- | --- |
|
||||
| 0 | 1 | `function () { E[E.length - 1] += String.fromCharCode(z[H++]); }` |
|
||||
| 1 | 0 | `function () { return !!w; }` |
|
||||
|
||||
+17
-1
@@ -167,7 +167,12 @@ def extract_xtea_key(tdc_js_code: str, node_bin: str = "node",
|
||||
"""
|
||||
从服务端下发的 tdc.js 里提取本次构建的 XTEA 密钥 (4 个 32 位字)。
|
||||
|
||||
流程 (Node 只负责"跑出密钥", 算法全在 Python):
|
||||
**主路径 = 纯 Python 静态还原** (`tcaptcha/tdc_key.py`):
|
||||
反汇编字节码 -> 定位 XTEA 闭包 -> 回溯它捕获的密钥数组 -> 读出 4 个密钥字
|
||||
闭包(源串 + 偏移 + 存取下标) -> 折入轮加盐。
|
||||
零 Node、零 JS 运行时、零 pc 硬编码 (opcode 表按构建推导, 见 tdc_opcodes.py)。
|
||||
|
||||
**兜底路径 = Node 插桩** (密钥构造算法若被改动时仍可用):
|
||||
1. Node 在 VM 上下文里把 ADD opcode 换成带日志版本;
|
||||
XTEA 每轮的 (sum + K[j]) 是一次 ADD, 两个操作数必有一个等于 r*DELTA,
|
||||
于是另一个操作数就是密钥字 —— 按频次排序得到无歧义候选。
|
||||
@@ -183,6 +188,17 @@ def extract_xtea_key(tdc_js_code: str, node_bin: str = "node",
|
||||
if cached:
|
||||
return list(cached)
|
||||
|
||||
# ---- 主路径: 纯 Python 静态还原 (零 Node) ----
|
||||
try:
|
||||
from .tdc_key import extract_xtea_key as _pure_extract
|
||||
key = [int(k) & M32 for k in _pure_extract(tdc_js_code)]
|
||||
if len(key) == 4:
|
||||
if use_cache:
|
||||
_KEY_CACHE[tdc_js_code] = key
|
||||
return list(key)
|
||||
except Exception as exc: # noqa: BLE001
|
||||
logger.warning("纯 Python 静态提取失败, 退回 Node 插桩: %s", exc)
|
||||
|
||||
words, raw = _capture_key_words_cached(tdc_js_code, node_bin=node_bin)
|
||||
candidates = [int(k) & M32 for k, *_ in words[:top_n]]
|
||||
# 验证用「内容特征」而非固定明文 (AGENTS.md §5.3 / 附录 A#6):
|
||||
|
||||
@@ -15,6 +15,7 @@ SUITES = [
|
||||
"tcaptcha.tests.test_crypto_core",
|
||||
"tcaptcha.tests.test_glyph_solver",
|
||||
"tcaptcha.tests.test_xtea_key",
|
||||
"tcaptcha.tests.test_tdc_key",
|
||||
]
|
||||
|
||||
|
||||
|
||||
@@ -0,0 +1,441 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""
|
||||
腾讯天御 TDC 混沌 VM (Chaos VM) —— XTEA 密钥的**纯 Python** 静态还原器。
|
||||
|
||||
为什么需要它
|
||||
------------
|
||||
此前要拿到某一版 tdc.js 的 XTEA 密钥, 只能把 tdc.js 丢进 Node 插桩跑一遍
|
||||
(ADDU opcode 吐密钥字)。密钥**不是字面量**, 而是 VM 运行时由字节码算出来的,
|
||||
所以必须"读指令流"而不是"搜字符串"。
|
||||
|
||||
本模块完全离线(零 Node、零 JS 运行时), 基于**正确反汇编**(tools/tdc_disasm.py)做
|
||||
**模式 + 数据流**定位, 复刻密钥构造算法。全部结论都有运行时神谕(tools/tdc_probe.js)
|
||||
逐条对照过, 见 har/1ms_dianxuan/PROGRESS.md。
|
||||
|
||||
算法 (已证: 纯 Python 结果 == Node extract_xtea_key 逐字一致)
|
||||
------------------------------------------------------------
|
||||
1. **4 个密钥字**: 每个字由一个专门的闭包算出:
|
||||
word_i = Σ_{k=0..3} ( (ord(src_i[k]) + off_i) & 0xFF ) << (8*k)
|
||||
源串 `src_i` 是紧跟在该闭包 `CREATE_CLOSURE` 之后的 4 个 `PUSH_STR_CHAR`
|
||||
字面量; 偏移 `off_i` 是打包点的立即数; 存入下标 `i` 由闭包尾部的
|
||||
`PUSH_IMM_ARRAY / PUSH_IMM [i] / ... / SET_PROP` 给出。
|
||||
|
||||
2. **哪些闭包是真密钥**: 密钥数组会被 **XTEA 闭包捕获**。
|
||||
XTEA 轮里出现 `delta = 0x9E3779B9`; 含 delta 的那个闭包体所属的
|
||||
`CREATE_CLOSURE` 即 XTEA 闭包, 其 `caps=[r7=r<K>]` 的 `<K>` 就是密钥数组所在的
|
||||
父帧寄存器。**只有 caps 右侧等于该寄存器的密钥字闭包**才是真密钥
|
||||
(同一份脚本往往还并行算了第二个"诱饵"数组, 用不同寄存器, 必须排除)。
|
||||
|
||||
3. **轮加盐**: XTEA 轮内 `key[(sum>>>11)&3]` 查表之后会**条件性**加上一个立即数
|
||||
(salt): `GET_PROP / PUSH_IMM <salt> / ADD / ADD / XOR`。加给哪个下标由守卫条件
|
||||
`idx == ((sum>>>11)&3)` / `idx == (sum & 3)` 里的 `idx` 常量决定。
|
||||
有效密钥 = base 字按盐的下标位置相加。
|
||||
|
||||
所有定位都是**模式 + 立即数**, 不写死任何 pc (每份构建都会重排)。
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import os
|
||||
import re
|
||||
import sys
|
||||
from typing import Dict, List, Optional, Tuple
|
||||
|
||||
_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
||||
if _ROOT not in sys.path: # 允许在任意 cwd 下 import tools.*
|
||||
sys.path.insert(0, _ROOT)
|
||||
|
||||
from tools.tdc_disasm import extract_bytecode, disassemble_cfg # noqa: E402
|
||||
|
||||
from tcaptcha.tdc_opcodes import derive_opcode_table # noqa: E402
|
||||
|
||||
M32 = 0xFFFFFFFF
|
||||
XTEA_DELTA = 0x9E3779B9 # 2654435769, XTEA 黄金比常数
|
||||
KEY_LEN = 4
|
||||
|
||||
# 走"下标表达式"时要跳过的算子 (它们构成 `(sum & 3)` / `(sum>>>11)&3`)
|
||||
_SKIP_OPS = {
|
||||
"BITWISE_AND", "SHR_UNSIGNED", "LOAD_REG_IMM", "MAKE_REG_PROP_REF",
|
||||
"GET_PROP", "SWAP", "STORE_REG", "ADD", "DUP", "LOAD_REG_INDIRECT",
|
||||
"PUSH_IMM_ARRAY", "POP",
|
||||
}
|
||||
# 这些算子紧跟在"掩码 PUSH_IMM 3"之前, 说明那个 3 是掩码而不是下标
|
||||
_MASK_NEIGHBOURS = {"BITWISE_AND", "LOAD_REG_IMM", "SHR_UNSIGNED"}
|
||||
|
||||
|
||||
class _Dis:
|
||||
"""对反汇编结果做少量索引封装, 便于按 pc 前后游走。"""
|
||||
|
||||
def __init__(self, js_text: str) -> None:
|
||||
self.z = extract_bytecode(js_text)
|
||||
# opcode 编号每份构建都重排 -> 必须先按构建把语义表推导出来
|
||||
names, arity, dynamic, diag = derive_opcode_table(js_text)
|
||||
self.names, self.arity, self.dynamic, self.diag = names, arity, dynamic, diag
|
||||
self.insts = disassemble_cfg(self.z, names=names, arity=arity, dynamic=dynamic)
|
||||
self.by_pc: Dict[int, dict] = {i["pc"]: i for i in self.insts}
|
||||
self.pcs: List[int] = sorted(self.by_pc)
|
||||
self._idx: Dict[int, int] = {pc: n for n, pc in enumerate(self.pcs)}
|
||||
|
||||
def op(self, pc: Optional[int]) -> Optional[str]:
|
||||
i = self.by_pc.get(pc) if pc is not None else None
|
||||
return i["name"] if i else None
|
||||
|
||||
def ops(self, pc: Optional[int]) -> List[int]:
|
||||
i = self.by_pc.get(pc) if pc is not None else None
|
||||
return list(i["operands"]) if i else []
|
||||
|
||||
def note(self, pc: int) -> str:
|
||||
return " ".join(self.by_pc[pc].get("note") or [])
|
||||
|
||||
def nxt(self, pc: Optional[int]) -> Optional[int]:
|
||||
n = self._idx.get(pc) if pc is not None else None
|
||||
return self.pcs[n + 1] if n is not None and n + 1 < len(self.pcs) else None
|
||||
|
||||
def prev(self, pc: Optional[int]) -> Optional[int]:
|
||||
n = self._idx.get(pc) if pc is not None else None
|
||||
return self.pcs[n - 1] if n is not None and n > 0 else None
|
||||
|
||||
def in_range(self, lo: int, hi: int) -> List[int]:
|
||||
import bisect
|
||||
a = bisect.bisect_left(self.pcs, lo)
|
||||
b = bisect.bisect_right(self.pcs, hi)
|
||||
return self.pcs[a:b]
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------- #
|
||||
# 1. 闭包清单 / 密钥字打包点
|
||||
# --------------------------------------------------------------------------- #
|
||||
def _closures(d: _Dis) -> List[dict]:
|
||||
"""所有 CREATE_CLOSURE -> {pc, target, caps:{dst:src}, str_after}。"""
|
||||
out: List[dict] = []
|
||||
for i in d.insts:
|
||||
if i["name"] != "CREATE_CLOSURE":
|
||||
continue
|
||||
pc = i["pc"]
|
||||
note = " ".join(i["note"] or [])
|
||||
m = re.search(r"target=@(\d+)", note)
|
||||
if not m:
|
||||
continue
|
||||
caps: Dict[int, int] = {}
|
||||
caps_part = re.search(r"caps=\[([^\]]*)\]", note)
|
||||
if caps_part:
|
||||
for dst, src in re.findall(r"r(\d+)=r(\d+)", caps_part.group(1)):
|
||||
caps[int(dst)] = int(src)
|
||||
out.append({"pc": pc, "target": int(m.group(1)), "caps": caps,
|
||||
"str_after": _string_after(d, pc)})
|
||||
return out
|
||||
|
||||
|
||||
def _string_after(d: _Dis, pc: int) -> Optional[str]:
|
||||
"""CREATE_CLOSURE 紧随其后的 `PUSH_EMPTY_STR + PUSH_STR_CHAR*` 字面量。"""
|
||||
p = d.nxt(pc)
|
||||
if p is None or d.op(p) != "PUSH_EMPTY_STR":
|
||||
return None
|
||||
chars: List[str] = []
|
||||
q = d.nxt(p)
|
||||
while q is not None and d.op(q) == "PUSH_STR_CHAR":
|
||||
chars.append(chr(d.ops(q)[0]))
|
||||
q = d.nxt(q)
|
||||
return "".join(chars)
|
||||
|
||||
|
||||
def _pack_sites(d: _Dis, lo: int, hi: int) -> List[dict]:
|
||||
"""
|
||||
在 [lo,hi] 里找"密钥字打包点":
|
||||
... CALL_METHOD[1] / PUSH_IMM c / SUB (off = -c)
|
||||
或 ... CALL_METHOD[1] / PUSH_IMM 0 / PUSH_IMM c / SUB / SUB (off = +c)
|
||||
/ PUSH_IMM 8 / LOAD_REG_IMM / MUL / SHL / BITWISE_OR
|
||||
返回 [{or_pc, off}]。
|
||||
"""
|
||||
res: List[dict] = []
|
||||
for pc in d.in_range(lo, hi):
|
||||
if d.op(pc) != "PUSH_IMM" or d.ops(pc) != [8]:
|
||||
continue
|
||||
a, b, c, e = d.nxt(pc), None, None, None
|
||||
b = d.nxt(a)
|
||||
c = d.nxt(b)
|
||||
e = d.nxt(c)
|
||||
if not (d.op(a) == "LOAD_REG_IMM" and d.op(b) == "MUL"
|
||||
and d.op(c) == "SHL" and d.op(e) == "BITWISE_OR"):
|
||||
continue
|
||||
off = _decode_offset(d, pc)
|
||||
if off is not None:
|
||||
res.append({"or_pc": e, "imm8_pc": pc, "off": off})
|
||||
return res
|
||||
|
||||
|
||||
def _decode_offset(d: _Dis, imm8_pc: int) -> Optional[int]:
|
||||
"""解出紧跟 CALL_METHOD 之后的偏移立即数 (含两种编码)。"""
|
||||
p1 = d.prev(imm8_pc) # 最后一个 SUB
|
||||
if d.op(p1) != "SUB":
|
||||
return None
|
||||
p2 = d.prev(p1)
|
||||
if d.op(p2) == "PUSH_IMM": # X - c => off = -c
|
||||
return -d.ops(p2)[0]
|
||||
if d.op(p2) == "SUB": # X - (0 - c) => off = +c
|
||||
p3 = d.prev(p2)
|
||||
p4 = d.prev(p3)
|
||||
if d.op(p3) == "PUSH_IMM" and d.op(p4) == "PUSH_IMM" and d.ops(p4) == [0]:
|
||||
return d.ops(p3)[0]
|
||||
return None
|
||||
|
||||
|
||||
def _store_index(d: _Dis, lo: int, hi: int) -> Optional[int]:
|
||||
"""闭包尾部的 `PUSH_IMM_ARRAY / PUSH_IMM [i] / MAKE_REG_PROP_REF / LOAD_REG_IMM / SET_PROP`。"""
|
||||
hits: List[int] = []
|
||||
for pc in d.in_range(lo, hi):
|
||||
if d.op(pc) != "PUSH_IMM_ARRAY":
|
||||
continue
|
||||
a, b, c, e = d.nxt(pc), None, None, None
|
||||
b = d.nxt(a)
|
||||
c = d.nxt(b)
|
||||
e = d.nxt(c)
|
||||
if (d.op(a) == "PUSH_IMM" and d.op(b) == "MAKE_REG_PROP_REF"
|
||||
and d.op(c) == "LOAD_REG_IMM" and d.op(e) == "SET_PROP"):
|
||||
hits.append(d.ops(a)[0])
|
||||
return hits[-1] if hits else None
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------- #
|
||||
# 2. XTEA 闭包定位 + 轮加盐
|
||||
# --------------------------------------------------------------------------- #
|
||||
def _find_xtea_closure(d: _Dis, closures: List[dict]) -> Optional[dict]:
|
||||
"""含 XTEA 常数(delta)的那个闭包体所属的 CREATE_CLOSURE。"""
|
||||
delta_pcs = [pc for pc in d.pcs
|
||||
if d.op(pc) == "PUSH_IMM" and d.ops(pc)[0] in (XTEA_DELTA, XTEA_DELTA * 32)]
|
||||
# 取"最内层"(区间最小)的闭包体 —— 外层 init 函数的体也会横跨 delta,
|
||||
# 但它不捕获密钥数组。
|
||||
best: Optional[dict] = None
|
||||
best_end: Optional[int] = None
|
||||
for cl in closures:
|
||||
end = _body_end(d, cl["pc"])
|
||||
if end is None or end < cl["target"]:
|
||||
continue
|
||||
if any(cl["target"] <= pc <= end for pc in delta_pcs):
|
||||
if best is None or best_end is None or (end - cl["target"]) < (best_end - best["target"]):
|
||||
best, best_end = cl, end
|
||||
return best
|
||||
|
||||
|
||||
def _largest_reg(regs: Dict[int, int]) -> Optional[int]:
|
||||
return max(regs.values()) if regs else None
|
||||
|
||||
|
||||
def _caps_single(caps: Dict[int, int]) -> Optional[Tuple[int, int]]:
|
||||
"""字/XTEA 闭包只捕获一个数组: 返回 (dst, src)。"""
|
||||
if len(caps) != 1:
|
||||
return None
|
||||
(dst, src), = caps.items()
|
||||
return dst, src
|
||||
|
||||
|
||||
def _defining_closure(d: _Dis, closures: List[dict], pc: int) -> Optional[dict]:
|
||||
"""含给定 pc 的**最内层**闭包(即创建 pc 处那条 CREATE_CLOSURE 的那个函数)。"""
|
||||
best: Optional[dict] = None
|
||||
best_span: Optional[int] = None
|
||||
for c in closures:
|
||||
e = _body_end(d, c["pc"])
|
||||
if e is None or e < c["target"]:
|
||||
continue
|
||||
if c["target"] <= pc <= e:
|
||||
span = e - c["target"]
|
||||
if best_span is None or span < best_span:
|
||||
best, best_span = c, span
|
||||
return best
|
||||
|
||||
|
||||
def _body_end(d: _Dis, cl_pc: int) -> Optional[int]:
|
||||
"""
|
||||
闭包体结束 pc (该体的 RET_TRUE)。约定: 闭包被创建后要么紧跟其体尾的 RET_TRUE,
|
||||
要么中间夹一条 `PUSH_IMM_ARRAY`(闭包被塞进数组时)。
|
||||
"""
|
||||
e = d.prev(cl_pc)
|
||||
if e is not None and d.op(e) == "PUSH_IMM_ARRAY":
|
||||
e = d.prev(e)
|
||||
return e if (e is not None and d.op(e) == "RET_TRUE") else None
|
||||
|
||||
|
||||
def _guard_index(d: _Dis, eq_pc: int) -> Optional[int]:
|
||||
"""
|
||||
守卫 `idx == ((sum & 3) | ((sum>>>11)&3))` 里的常量 idx。
|
||||
从 EQ_LOOSE 反向游走, 跳过下标表达式自身的算子, 第一个落在 0..3 的立即数即 idx。
|
||||
掩码 3 靠"它的后继是 BITWISE_AND / LOAD_REG_IMM / SHR_UNSIGNED"识别并跳过。
|
||||
"""
|
||||
pc = d.prev(eq_pc)
|
||||
last: Optional[str] = None
|
||||
for _ in range(24):
|
||||
if pc is None:
|
||||
return None
|
||||
nm, ops = d.op(pc), d.ops(pc)
|
||||
if nm in ("PUSH_IMM", "REPLACE_TOP_IMM") and ops:
|
||||
v = ops[0]
|
||||
if 0 <= v <= 3:
|
||||
if not (nm == "PUSH_IMM" and v == 3 and last in _MASK_NEIGHBOURS):
|
||||
return v
|
||||
elif nm not in _SKIP_OPS:
|
||||
# 出现意料之外的算子: 继续向前找, 但记录上下文
|
||||
pass
|
||||
last = nm
|
||||
pc = d.prev(pc)
|
||||
return None
|
||||
|
||||
|
||||
def _salts(d: _Dis, lo: int, hi: int) -> List[Tuple[int, int]]:
|
||||
"""
|
||||
XTEA 轮内的条件加盐: `GET_PROP / PUSH_IMM <salt> / ADD / ADD / XOR`。
|
||||
返回 [(idx, salt)] (按 (idx,salt) 去重)。
|
||||
"""
|
||||
salt_pcs: List[int] = []
|
||||
for pc in d.in_range(lo, hi):
|
||||
if d.op(pc) != "PUSH_IMM":
|
||||
continue
|
||||
if d.op(d.prev(pc)) != "GET_PROP" or d.op(d.nxt(pc)) != "ADD":
|
||||
continue
|
||||
salt_pcs.append(pc)
|
||||
if not salt_pcs:
|
||||
return []
|
||||
|
||||
# 收集所有 (JUMP_IF_TRUE, EQ_LOOSE) 守卫; block(T) 到 T 之后第一条无条件 JUMP
|
||||
guards: List[Tuple[int, int, Optional[int]]] = [] # (target, jump_pc, idx)
|
||||
for pc in d.in_range(lo, hi):
|
||||
if d.op(pc) != "JUMP_IF_TRUE":
|
||||
continue
|
||||
tgt_ops = d.ops(pc)
|
||||
if not tgt_ops:
|
||||
continue
|
||||
tgt = tgt_ops[0]
|
||||
eq = d.prev(pc)
|
||||
if d.op(eq) != "EQ_LOOSE":
|
||||
continue
|
||||
idx = _guard_index(d, eq)
|
||||
if idx is not None:
|
||||
guards.append((tgt, pc, idx))
|
||||
|
||||
def block_end(tgt: int) -> int:
|
||||
p = tgt
|
||||
while p is not None and p <= hi:
|
||||
if d.op(p) == "JUMP":
|
||||
return p
|
||||
p = d.nxt(p)
|
||||
return hi
|
||||
|
||||
out: List[Tuple[int, int]] = []
|
||||
for spc in salt_pcs:
|
||||
salt = d.ops(spc)[0]
|
||||
best: Optional[Tuple[int, int]] = None # (target, idx)
|
||||
for tgt, _j, idx in guards:
|
||||
if tgt <= spc < block_end(tgt):
|
||||
if best is None or tgt > best[0]:
|
||||
best = (tgt, idx)
|
||||
if best is not None:
|
||||
out.append((best[1], salt))
|
||||
# 去重保序
|
||||
seen = set()
|
||||
uniq: List[Tuple[int, int]] = []
|
||||
for x in out:
|
||||
if x not in seen:
|
||||
seen.add(x)
|
||||
uniq.append(x)
|
||||
return uniq
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------- #
|
||||
# 3. 主入口
|
||||
# --------------------------------------------------------------------------- #
|
||||
def word_from_chars(src: str, off: int) -> int:
|
||||
"""word = Σ ((ord(c) + off) & 0xFF) << (8*k) (小端打包)"""
|
||||
w = 0
|
||||
for k, ch in enumerate(src):
|
||||
w |= ((ord(ch) + off) & 0xFF) << (8 * k)
|
||||
return w & M32
|
||||
|
||||
|
||||
def extract_xtea_key_pure(js_text: str, verbose: bool = False):
|
||||
"""
|
||||
纯 Python 从 tdc.js 里还原本次构建的 XTEA 有效密钥 (4 个 32 位字)。
|
||||
|
||||
返回 dict:
|
||||
base : 4 个 base 字 (未折盐)
|
||||
salts : [(idx, salt), ...]
|
||||
key : 有效密钥
|
||||
words : [(idx, word, src, off), ...] (便于诊断)
|
||||
"""
|
||||
d = _Dis(js_text)
|
||||
closures = _closures(d)
|
||||
xtea = _find_xtea_closure(d, closures)
|
||||
if xtea is None:
|
||||
raise ValueError("未找到 XTEA 闭包 (字节码里没有 delta 常数?)")
|
||||
xs = _caps_single(xtea["caps"])
|
||||
if xs is None:
|
||||
raise ValueError("XTEA 闭包没有唯一捕获的数组, 无法确定密钥数组")
|
||||
xtea_src = xs[1] # XTEA 所在帧里, 密钥数组所在的寄存器
|
||||
|
||||
# 密钥数组是"被 XTEA 闭包捕获"的那一个。字闭包与 XTEA 闭包往往不在同一帧:
|
||||
# 父帧 G: r<R> = 密钥数组
|
||||
# ├─ 字闭包 W: caps=[r7 = 本帧 r<Wsrc>] (Wsrc ∈ {R, 诱饵寄存器})
|
||||
# └─ XTEA 闭包 : caps=[r7 = 本帧 r<xtea_src>]
|
||||
# 所以要把 W 的捕获穿一层: B = 定义 W 的那个闭包(B 帧 -> G 帧), 检查
|
||||
# B.caps[Wsrc] == xtea_src -> 该字闭包写的就是真密钥数组。
|
||||
words: Dict[int, Tuple[int, str, int]] = {}
|
||||
picked: List[str] = []
|
||||
for cl in closures:
|
||||
if cl["str_after"] is None:
|
||||
continue
|
||||
cs = _caps_single(cl["caps"])
|
||||
if cs is None:
|
||||
continue
|
||||
w_src = cs[1] # 该字闭包捕获的、定义帧里的寄存器
|
||||
definer = _defining_closure(d, closures, cl["pc"])
|
||||
if definer is not None and definer["caps"]:
|
||||
real = definer["caps"].get(w_src) == xtea_src
|
||||
else: # 同帧: 直接比
|
||||
real = (w_src == xtea_src)
|
||||
if not real:
|
||||
continue
|
||||
end = _body_end(d, cl["pc"])
|
||||
if end is None or end < cl["target"]:
|
||||
continue
|
||||
packs = _pack_sites(d, cl["target"], end)
|
||||
if not packs:
|
||||
continue
|
||||
idx = _store_index(d, cl["target"], end)
|
||||
if idx is None or not (0 <= idx < KEY_LEN):
|
||||
continue
|
||||
src, off = cl["str_after"], packs[-1]["off"]
|
||||
words[idx] = (word_from_chars(src, off), src, off)
|
||||
picked.append("%s(idx=%d,off=%+d)" % (src, idx, off))
|
||||
|
||||
if len(words) != KEY_LEN:
|
||||
raise ValueError(
|
||||
"密钥字不完整: 只定位到 %s (期望 0..3); XTEA 密钥帧寄存器 r%s"
|
||||
% (sorted(words), xtea_src))
|
||||
|
||||
base = [words[i][0] for i in range(KEY_LEN)]
|
||||
salts = _salts(d, xtea["target"], d.prev(xtea["pc"]) or xtea["target"])
|
||||
|
||||
key = list(base)
|
||||
for idx, salt in salts:
|
||||
key[idx] = (key[idx] + salt) & M32
|
||||
|
||||
report = {
|
||||
"base": base,
|
||||
"salts": salts,
|
||||
"key": key,
|
||||
"words": [(i, words[i][0], words[i][1], words[i][2]) for i in range(KEY_LEN)],
|
||||
"xtea_closure_pc": xtea["pc"],
|
||||
"key_reg": xtea_src,
|
||||
}
|
||||
if verbose:
|
||||
print("[tdc_key_pure] XTEA closure @%d key array = r%s" % (xtea["pc"], xtea_src))
|
||||
for i, w, s, o in report["words"]:
|
||||
print(" word[%d] <- %-6r off=%+d = %d" % (i, s, o, w))
|
||||
for idx, salt in salts:
|
||||
print(" salt +%d -> key[%d]" % (salt, idx))
|
||||
print(" base = %s" % base)
|
||||
print(" key = %s" % key)
|
||||
return report
|
||||
|
||||
|
||||
def extract_xtea_key(js_text: str) -> List[int]:
|
||||
"""便捷入口: 只取有效密钥 (与 crypto.extract_xtea_key 同口径)。"""
|
||||
return list(extract_xtea_key_pure(js_text)["key"])
|
||||
@@ -0,0 +1,320 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""
|
||||
腾讯天御 TDC 混沌 VM (Chaos VM) —— **按构建推导操作码表**。
|
||||
|
||||
为什么需要它
|
||||
------------
|
||||
`tools/tdc_disasm.py` 里固化的 opcode 表取自 **core_js 那一份构建**。
|
||||
实测(5 份构建交叉验证): **opcode 编号每份构建都会整体重排** ——
|
||||
同一份源码编译出的字节码, pc0 处的 `SET_STACK_LEN` 在 core_js 是 53,
|
||||
在 root 样本是 19, 在 1ms 实时样本分别是 18 / 10 / 60。
|
||||
所以写死 opcode 表 = 换一份脚本就全错。
|
||||
|
||||
本模块的做法:
|
||||
1. 从 JS 源码里定位解释器主循环 `X[Y[Z++]]()` -> 取出**操作码处理器数组变量** `X`;
|
||||
2. 解析该数组字面量(JS 感知的括号/字符串/正则切分)得到 70 个槽位(含空洞);
|
||||
3. 把每个 handler 的源码**规范化**(去空白 + 局部变量按首次出现重命名为 v1,v2,...,
|
||||
保留 JS 关键字/内建名与属性名)得到与变量命名无关的"指纹";
|
||||
4. 用内置的 58 条参考指纹(取自 core_js = `tools/tdc_disasm.OP_NAMES` 那套语义)
|
||||
反查每个槽位的语义名 —— 数组下标即该构建的 opcode。
|
||||
5. 操作数个数按语义名从参考表取 (与"handler 体内 `x[y++]` 读取次数"实测一致)。
|
||||
|
||||
这样, 只要服务端还是这台 VM(handler 语义相同), 无论 opcode 怎么重排都能对上。
|
||||
```
|
||||
经过: 5/5 本构建全部 58/58 handler 精确匹配 (har/1ms_dianxuan/PROGRESS.md)。
|
||||
```
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import os
|
||||
import re
|
||||
import sys
|
||||
from typing import Dict, List, Optional, Tuple
|
||||
|
||||
_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
|
||||
if _ROOT not in sys.path:
|
||||
sys.path.insert(0, _ROOT)
|
||||
|
||||
from tools.tdc_disasm import ARITY as _CORE_ARITY, DYNAMIC_ARITY as _CORE_DYNAMIC, OP_NAMES as _CORE_NAMES # noqa: E402
|
||||
|
||||
# --------------------------------------------------------------------------- #
|
||||
# 参考指纹: core_js 构建的 58 个 handler 规范化源码 -> 语义名 (与 OP_NAMES 一致)
|
||||
# --------------------------------------------------------------------------- #
|
||||
_REF_BLOB = r"""PUSH_STR_CHAR function(){v1[v1.length-1]+=String.fromCharCode(v2[v3++])}
|
||||
IS_EXCEPTION function(){return!!v1}
|
||||
DUP function(){v1.push(v1[v1.length-1])}
|
||||
XOR function(){v1[v1.length-2]=v1[v1.length-2]^v1.pop()}
|
||||
EQ_STRICT function(){v1[v1.length-2]=v1[v1.length-2]===v1.pop()}
|
||||
MOD function(){v1[v1.length-2]=v1[v1.length-2]%v1.pop()}
|
||||
CALL_FUNC function(){varv1=v2[v3++],v4=v1?v5.slice(-v1):[];v5.length-=v1,v5.push(v5.pop().apply(v6,v4))}
|
||||
GET_GLOBAL_PROP function(){v1[v1.length-1]=v2[v1[v1.length-1]]}
|
||||
MAKE_PROP_REF function(){varv1=v2.pop(),v3=v2.pop();v2.push([v3[0][v3[1]],v1])}
|
||||
TYPEOF function(){v1.push(typeofv1.pop())}
|
||||
DELETE function(){varv1=v2.pop();v2.push(deletev1[0][v1[1]])}
|
||||
LOGICAL_NOT function(){v1.push(!v1.pop())}
|
||||
EQ_LOOSE function(){v1[v1.length-2]=v1[v1.length-2]==v1.pop()}
|
||||
SET_PROP function(){varv1=v2[v2.length-2];v1[0][v1[1]]=v2[v2.length-1]}
|
||||
GREATER_THAN function(){v1[v1.length-2]=v1[v1.length-2]>v1.pop()}
|
||||
SWAP function(){varv1=v2[v3++],v4=v5[v5.length-2-v1];v5[v5.length-2-v1]=v5.pop(),v5.push(v4)}
|
||||
IN function(){v1[v1.length-2]=v1[v1.length-2]inv1.pop()}
|
||||
RET_TRUE function(){return!0}
|
||||
GREATER_EQ function(){v1[v1.length-2]=v1[v1.length-2]>=v1.pop()}
|
||||
PUSH_IMM function(){v1.push(v2[v3++])}
|
||||
MUL function(){v1[v1.length-2]=v1[v1.length-2]*v1.pop()}
|
||||
LOAD_REG_INDIRECT function(){v1.push(v1[v1.pop()[0]][0])}
|
||||
ITER_NEXT function(){v1[v1.length-1].length?v1.push(v1[v1.length-1].shift(),!0):v1.push(undefined,!1)}
|
||||
BIND_THIS function(){v1.push([v2,v1.pop()])}
|
||||
PUSH_NULL function(){v1.push(null)}
|
||||
BITWISE_AND function(){v1[v1.length-2]=v1[v1.length-2]&v1.pop()}
|
||||
PUSH_IMM_ARRAY function(){v1.push([v2[v3++]])}
|
||||
MAKE_REG_PROP_REF function(){varv1=v2.pop();v2.push([v2[v2.pop()][0],v1])}
|
||||
SUB function(){v1[v1.length-2]=v1[v1.length-2]-v1.pop()}
|
||||
CALL_METHOD function(){varv1=v2[v3++],v4=v1?v5.slice(-v1):[],v1=(v5.length-=v1,v5.pop());v5.push(v1[0][v1[1]].apply(v1[0],v4))}
|
||||
SHR function(){v1[v1.length-2]=v1[v1.length-2]>>v1.pop()}
|
||||
SHL function(){v1[v1.length-2]=v1[v1.length-2]<<v1.pop()}
|
||||
JUMP_IF_TRUE function(){varv1=v2[v3++];v4[v4.length-1]&&(v3=v1)}
|
||||
FOR_IN_KEYS function(){varv1,v2=[];for(v1inv3.pop())v2.push(v1);v3.push(v2)}
|
||||
ADD function(){v1[v1.length-2]=v1[v1.length-2]+v1.pop()}
|
||||
NEW_METHOD function(){varv1=v2[v3++],v4=v1?v5.slice(-v1):[],v1=(v5.length-=v1,v4.unshift(null),v5.pop());v5.push(v6(v1[0][v1[1]],v4))}
|
||||
PUSH_UNDEFINED function(){v1.push(undefined)}
|
||||
PUSH_FALSE function(){v1.push(!1)}
|
||||
STORE_REG function(){v1[v1[v1.length-2][0]][0]=v1[v1.length-1]}
|
||||
ALLOC_REG function(){varv1=v2[v3++];v4[v1]=v4[v1]===undefined?[]:v4[v1]}
|
||||
POP_TRY function(){v1.pop()}
|
||||
THROW function(){throwv1[v1.length-1]}
|
||||
POP function(){v1.pop()}
|
||||
DIV function(){v1[v1.length-2]=v1[v1.length-2]/v1.pop()}
|
||||
SET_STACK_LEN function(){v1.length=v2[v3++]}
|
||||
REVERSE_TOP2 function(){v1.push([v1.pop(),v1.pop()].reverse())}
|
||||
PUSH_EMPTY_STR function(){v1.push("")}
|
||||
PUSH_TRY function(){v1.push([v2[v3++],v4.length,v2[v3++]])}
|
||||
CREATE_CLOSURE function(){for(varv1=v2[v3++],v4=[],v5=v2[v3++],v6=v2[v3++],v7=[],v8=0;v8<v5;v8++)v4[v2[v3++]]=v9[v2[v3++]];for(v8=0;v8<v6;v8++)v7[v8]=v2[v3++];v9.push(functionv10(){varv5=v4.slice(0);v5[0]=[this],v5[1]=[arguments],v5[2]=[v10];for(varv6=0;v6<v7.length&&v6<arguments.length;v6++)0<v7[v6]&&(v5[v7[v6]]=[arguments[v6]]);returnv11(v1,v2,v12,v5,v13,v14,v15,v16)})}
|
||||
GET_PROP function(){varv1=v2.pop();v2.push(v1[0][v1[1]])}
|
||||
JUMP function(){v1=v2[v1++]}
|
||||
NEW_FUNC function(){varv1=v2[v3++],v4=v1?v5.slice(-v1):[];v5.length-=v1,v4.unshift(null),v5.push(v6(v5.pop(),v4))}
|
||||
BITWISE_OR function(){v1[v1.length-2]=v1[v1.length-2]|v1.pop()}
|
||||
REPLACE_TOP_IMM function(){v1[v1.length-1]=v2[v3++]}
|
||||
SHR_UNSIGNED function(){v1[v1.length-2]=v1[v1.length-2]>>>v1.pop()}
|
||||
PUSH_TRUE function(){v1.push(!0)}
|
||||
LOAD_REG_IMM function(){v1.push(v1[v2[v3++]][0])}
|
||||
CLEAR_EXCEPTION function(){v1=null}"""
|
||||
|
||||
|
||||
def _reference() -> Dict[str, str]:
|
||||
out: Dict[str, str] = {}
|
||||
for line in _REF_BLOB.splitlines():
|
||||
if not line.strip():
|
||||
continue
|
||||
name, _, fp = line.partition("\t")
|
||||
out[fp] = name
|
||||
return out
|
||||
|
||||
|
||||
_REF_FP2NAME = _reference()
|
||||
# 语义名 -> 操作数个数 / 是否动态长度
|
||||
_NAME_ARITY: Dict[str, int] = {}
|
||||
for _op, _nm in _CORE_NAMES.items():
|
||||
_NAME_ARITY[_nm] = _CORE_ARITY.get(_op, 0)
|
||||
_NAME_DYNAMIC = {_CORE_NAMES[o] for o in _CORE_DYNAMIC}
|
||||
|
||||
# --------------------------------------------------------------------------- #
|
||||
# JS 感知的工具
|
||||
# --------------------------------------------------------------------------- #
|
||||
_JS_KEEP = set("""
|
||||
function var return new if else for while do typeof instanceof in of delete void this
|
||||
null true false undefined break continue switch case default throw try catch finally
|
||||
Object String Number Boolean Array Function RegExp Date Math JSON Error TypeError
|
||||
Symbol parseInt parseFloat isNaN encodeURIComponent decodeURIComponent
|
||||
Uint8Array Uint8ClampedArray Infinity NaN arguments
|
||||
""".split())
|
||||
_ID_RE = re.compile(r"[A-Za-z_$][\w$]*")
|
||||
_PREV_REGEX_OK = set("(,=[{;:!&|?+-*%~^<>")
|
||||
_OPERAND_READ_RE = re.compile(r"[A-Za-z_$][\w$]*\s*\[\s*[A-Za-z_$][\w$]*\s*\+\+\s*\]")
|
||||
|
||||
|
||||
def _skip_js_string(js: str, i: int) -> int:
|
||||
q = js[i]
|
||||
i += 1
|
||||
while i < len(js):
|
||||
c = js[i]
|
||||
if c == "\\":
|
||||
i += 2
|
||||
continue
|
||||
if c == q:
|
||||
return i + 1
|
||||
i += 1
|
||||
return i
|
||||
|
||||
|
||||
def _skip_js_regex(js: str, i: int) -> int:
|
||||
i += 1
|
||||
while i < len(js):
|
||||
c = js[i]
|
||||
if c == "\\":
|
||||
i += 2
|
||||
continue
|
||||
if c == "[":
|
||||
i += 1
|
||||
while i < len(js) and js[i] != "]":
|
||||
if js[i] == "\\":
|
||||
i += 1
|
||||
i += 1
|
||||
i += 1
|
||||
continue
|
||||
if c == "/":
|
||||
i += 1
|
||||
while i < len(js) and js[i].isalpha():
|
||||
i += 1
|
||||
return i
|
||||
i += 1
|
||||
return i
|
||||
|
||||
|
||||
def split_js_array(js: str, lb: int) -> Tuple[List[str], int]:
|
||||
"""把从 `lb`(指向 '[')起的数组字面量切成顶层元素字符串。"""
|
||||
assert js[lb] == "["
|
||||
i, depth, start, out = lb + 1, 0, lb + 1, []
|
||||
while i < len(js):
|
||||
c = js[i]
|
||||
if c in "\"'`":
|
||||
i = _skip_js_string(js, i)
|
||||
continue
|
||||
if c == "/":
|
||||
j = i - 1
|
||||
while j >= 0 and js[j].isspace():
|
||||
j -= 1
|
||||
if j < 0 or js[j] in _PREV_REGEX_OK:
|
||||
i = _skip_js_regex(js, i)
|
||||
continue
|
||||
i += 1
|
||||
continue
|
||||
if c in "([{":
|
||||
depth += 1
|
||||
i += 1
|
||||
continue
|
||||
if c in ")]}":
|
||||
if c == "]":
|
||||
if depth == 0:
|
||||
out.append(js[start:i])
|
||||
return out, i
|
||||
depth -= 1
|
||||
else:
|
||||
depth -= 1
|
||||
i += 1
|
||||
continue
|
||||
if c == "," and depth == 0:
|
||||
out.append(js[start:i])
|
||||
start = i + 1
|
||||
i += 1
|
||||
continue
|
||||
i += 1
|
||||
raise ValueError("数组字面量未闭合")
|
||||
|
||||
|
||||
def handler_array(js: str) -> List[str]:
|
||||
"""
|
||||
定位解释器主循环 `ARR[BC[PC++]]()`, 返回 `ARR` 的字面量元素(每个 = 一个 handler 源码)。
|
||||
含空洞的槽位返回空串。
|
||||
"""
|
||||
m = re.search(r"([A-Za-z_$][\w$]*)\s*\[\s*[A-Za-z_$][\w$]*\s*\[\s*[A-Za-z_$][\w$]*\s*\+\+\s*\]\s*\]\s*\(\s*\)", js)
|
||||
if not m:
|
||||
raise ValueError("未找到 VM 解释器主循环 (X[Y[Z++]]())")
|
||||
arr = m.group(1)
|
||||
best: Optional[List[str]] = None
|
||||
for mm in re.finditer(re.escape(arr) + r"\s*=\s*\[", js):
|
||||
lb = js.index("[", mm.start())
|
||||
try:
|
||||
els, _ = split_js_array(js, lb)
|
||||
except Exception:
|
||||
continue
|
||||
ne = sum(1 for e in els if e.strip())
|
||||
if best is None or ne > sum(1 for e in best if e.strip()):
|
||||
best = els
|
||||
if best is None:
|
||||
raise ValueError("未找到操作码处理器数组字面量 %s=[...]" % arr)
|
||||
return best
|
||||
|
||||
|
||||
def canonicalize(src: str) -> str:
|
||||
"""把 handler 源码规范化成与变量命名无关的指纹。"""
|
||||
out: List[str] = []
|
||||
mp: Dict[str, str] = {}
|
||||
n = 0
|
||||
i = 0
|
||||
prev = ""
|
||||
while i < len(src):
|
||||
c = src[i]
|
||||
if c in "\"'`":
|
||||
j = _skip_js_string(src, i)
|
||||
out.append(src[i:j])
|
||||
i, prev = j, "'"
|
||||
continue
|
||||
m = _ID_RE.match(src, i)
|
||||
if m:
|
||||
tok = m.group(0)
|
||||
if prev == "." or tok in _JS_KEEP:
|
||||
out.append(tok)
|
||||
else:
|
||||
if tok not in mp:
|
||||
n += 1
|
||||
mp[tok] = "v%d" % n
|
||||
out.append(mp[tok])
|
||||
i, prev = m.end(), "i"
|
||||
continue
|
||||
if not c.isspace():
|
||||
out.append(c)
|
||||
prev = c
|
||||
i += 1
|
||||
return "".join(out)
|
||||
|
||||
|
||||
# --------------------------------------------------------------------------- #
|
||||
# 对外: 按构建推导操作码表
|
||||
# --------------------------------------------------------------------------- #
|
||||
def derive_opcode_table(js: str) -> Tuple[Dict[int, str], Dict[int, int], set, Dict[str, str]]:
|
||||
"""
|
||||
返回 (names, arity, dynamic, diag):
|
||||
names : {opcode: 语义名}
|
||||
arity : {opcode: 固定操作数个数} (动态长度的不在其中)
|
||||
dynamic : {opcode} 需要读操作数才能定长
|
||||
diag : 诊断信息
|
||||
"""
|
||||
els = handler_array(js)
|
||||
names: Dict[int, str] = {}
|
||||
arity: Dict[int, int] = {}
|
||||
dynamic: set = set()
|
||||
unknown: List[int] = []
|
||||
slots: List[Tuple[int, str]] = []
|
||||
for op, src in enumerate(els):
|
||||
if not src.strip():
|
||||
continue
|
||||
fp = canonicalize(src)
|
||||
nm = _REF_FP2NAME.get(fp)
|
||||
if nm is None:
|
||||
unknown.append(op)
|
||||
continue
|
||||
names[op] = nm
|
||||
if nm in _NAME_DYNAMIC:
|
||||
dynamic.add(op)
|
||||
else:
|
||||
arity[op] = _NAME_ARITY.get(nm, 0)
|
||||
slots.append((op, nm))
|
||||
diag = {
|
||||
"handler_count": len(names),
|
||||
"unknown_slots": unknown,
|
||||
"table": {op: nm for op, nm in slots},
|
||||
}
|
||||
if unknown:
|
||||
raise ValueError("有 handler 指纹未能识别(服务端改了 VM 语义?): %s" % unknown[:10])
|
||||
if len(names) < 40:
|
||||
raise ValueError("推导出的 opcode 表过小(%d), 疑似解析失败" % len(names))
|
||||
return names, arity, dynamic, diag
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
_js = open(sys.argv[1] if len(sys.argv) > 1 else "tcaptcha/core_js/tdc_sample.js",
|
||||
encoding="utf-8", errors="replace").read()
|
||||
_names, _arity, _dyn, _diag = derive_opcode_table(_js)
|
||||
print("handlers: %d dynamic: %s" % (_diag["handler_count"], sorted(_dyn)))
|
||||
for _op in sorted(_names):
|
||||
print(" %3d %-18s arity=%s" % (_op, _names[_op], "dyn" if _op in _dyn else _arity.get(_op, 0)))
|
||||
File diff suppressed because one or more lines are too long
@@ -0,0 +1,108 @@
|
||||
# -*- coding: utf-8 -*-
|
||||
"""
|
||||
纯 Python XTEA 密钥提取 回归测试 (AGENTS.md §5.9)
|
||||
|
||||
固化的性质:
|
||||
1. **跨构建通用**: opcode 编号每份构建整体重排, 但仍能按构建推导出正确语义表;
|
||||
2. **纯 Python == Node 神谕**: 6 份构建(含 1 份新抓的实时 tdc.js)的期望密钥
|
||||
全部由 Node `extract_xtea_key` 逐字对照确认过, 此处固化为断言 (不再依赖 Node 跑测试);
|
||||
3. **定位不写死 pc**: 断言覆盖 core_js / root / 1ms(旧+两份实时) / 实时新抓 共 6 份
|
||||
不同构建, 任一份 opcode 重排或寄存器分配变化都会立刻暴露。
|
||||
|
||||
运行:
|
||||
.venv/bin/python -m tcaptcha.tests.test_tdc_key
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import sys
|
||||
from pathlib import Path
|
||||
|
||||
ROOT = Path(__file__).resolve().parents[2]
|
||||
if str(ROOT) not in sys.path:
|
||||
sys.path.insert(0, str(ROOT))
|
||||
|
||||
from tcaptcha.tdc_key import extract_xtea_key_pure # noqa: E402
|
||||
from tcaptcha.tdc_opcodes import derive_opcode_table # noqa: E402
|
||||
|
||||
# (相对路径, 期望有效密钥, 是否必须存在)
|
||||
# 期望值来自 Node 插桩 extract_xtea_key, 逐字一致。
|
||||
# 入库样本 (core_js / root / 实时新抓 fixture) 为必需门禁; har/** 被 .gitignore 忽略,
|
||||
# 本地有这些样本时会一并回归, 缺少时自动跳过。
|
||||
FIXTURES = [
|
||||
("tcaptcha/core_js/tdc_sample.js", [1818510928, 1381395306, 1245924197, 1768384072], True),
|
||||
("tcaptcha/tdc_sample.js", [1448563793, 1850559083, 1733513813, 1364613714], True),
|
||||
# 2026-10-11 现场从 turing.captcha.qcloud.com 下发的 tdc.js (入库固化, 防"仅本地可验")
|
||||
("tcaptcha/tests/fixtures/tdc_fresh_2026-10-11.js",
|
||||
[1397244740, 1799507557, 1399740484, 1516463189], True),
|
||||
("har/1ms_dianxuan/tdc.js", [1448106085, 1766670948, 1819173482, 1298358636], False),
|
||||
("har/1ms_dianxuan/live/tdc_live.js", [1314475846, 1500397933, 1751410275, 1716217423], False),
|
||||
("har/1ms_dianxuan/live/tdc_latest.js", [1800883800, 1850828643, 1197042522, 1800823125], False),
|
||||
]
|
||||
|
||||
_fails: list = []
|
||||
|
||||
|
||||
def check(name: str, cond: bool, detail: str = "") -> None:
|
||||
print(f" [{'PASS' if cond else 'FAIL'}] {name}" + (f" {detail}" if detail else ""))
|
||||
if not cond:
|
||||
_fails.append(name)
|
||||
|
||||
|
||||
def main() -> int:
|
||||
print("1) 按构建推导 opcode 表 (58 个 handler, 无未识别)")
|
||||
tables = {}
|
||||
skipped = []
|
||||
for rel, _expect, required in FIXTURES:
|
||||
p = ROOT / rel
|
||||
if not p.exists():
|
||||
if required:
|
||||
check(f"fixture {rel}", False, "缺失")
|
||||
else:
|
||||
skipped.append(rel)
|
||||
continue
|
||||
names, arity, dynamic, diag = derive_opcode_table(p.read_text(encoding="utf-8", errors="replace"))
|
||||
tables[rel] = names
|
||||
check(f"opcode 表 {Path(rel).name}",
|
||||
diag["handler_count"] == 58 and not diag["unknown_slots"],
|
||||
f"handlers={diag['handler_count']} unknown={diag['unknown_slots']}")
|
||||
if skipped:
|
||||
print(f" (跳过本地未提供的样本 {len(skipped)} 份: {[Path(s).name for s in skipped]})")
|
||||
|
||||
print("\n2) opcode 编号确实每份构建都不同 (证明没有依赖写死的数值)")
|
||||
jump_ops = {rel: next(o for o, n in tabs.items() if n == "JUMP") for rel, tabs in tables.items()}
|
||||
if len(jump_ops) >= 2:
|
||||
check("JUMP 的 opcode 跨构建不统一", len(set(jump_ops.values())) >= 2, str(jump_ops))
|
||||
|
||||
print("\n3) 纯 Python 提取 == Node 神谕 (逐字)")
|
||||
for rel, expect, required in FIXTURES:
|
||||
p = ROOT / rel
|
||||
if not p.exists():
|
||||
continue
|
||||
js = p.read_text(encoding="utf-8", errors="replace")
|
||||
try:
|
||||
rep = extract_xtea_key_pure(js)
|
||||
check(Path(rel).name, rep["key"] == expect, f"{rep['key']}")
|
||||
except Exception as exc: # noqa: BLE001
|
||||
check(Path(rel).name, False, f"{type(exc).__name__}: {exc}")
|
||||
|
||||
print("\n4) 密钥构造自洽: base + 折盐 == key")
|
||||
for rel, expect, required in FIXTURES:
|
||||
p = ROOT / rel
|
||||
if not p.exists():
|
||||
continue
|
||||
rep = extract_xtea_key_pure(p.read_text(encoding="utf-8", errors="replace"))
|
||||
folded = list(rep["base"])
|
||||
for idx, salt in rep["salts"]:
|
||||
folded[idx] = (folded[idx] + salt) & 0xFFFFFFFF
|
||||
check(f"折盐 {Path(rel).name}", folded == rep["key"] == expect)
|
||||
|
||||
print()
|
||||
if _fails:
|
||||
print(f"❌ test_tdc_key 失败 {len(_fails)} 项: {_fails}")
|
||||
return 1
|
||||
print("✅ test_tdc_key 全部通过")
|
||||
return 0
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
raise SystemExit(main())
|
||||
Reference in New Issue
Block a user