feat(工具链): 纯 Python 可达性 CFG 反汇编器 + 废弃旧线性反汇编器

做了什么:
- 新增 tools/tdc_disasm.py: 纯 Python 提取 tdc.js 字节码数组 + 可达性 CFG 反汇编
  * opcode 表由 VM 源码 AST 推导 (70 槽 / 58 handler, 操作数个数 = handler 内 z[H++] 次数)
  * 复刻 VM 内联的 base64 解码 + 游程展开 (含科学计数法常量), 无任何 JS 运行时
  * 遍历根含 JUMP/JUMP_IF_TRUE/PUSH_TRY catch/CREATE_CLOSURE 闭包体
- tools/tdc_vm_disasm.js 降级为废弃 shim (转发到 Python 版)
- TOOLCHAIN_REPORT.md 更新工具行与复验命令; tdc_sample.disasm.txt 用正确实现重新生成

为什么:
- 旧线性反汇编对字节码里的数据区继续扫描会失步: 实测 73 处 UNKNOWN (16 处 >69 的
  "不可能" opcode), 且未把闭包体当遍历根 -> 大半个程序不可达, 输出不可信

怎么验证:
- .venv/bin/python tools/tdc_disasm.py tcaptcha/core_js/tdc_sample.js
  -> bytecode=45636 (与旧实现一致), 可达指令 26168, illegal opcodes: []
- 反汇编首次干净读出 XTEA 轮函数: PUSH_IMM 2654435769 / ADD / ... / SHR_UNSIGNED 11 / &3
  并定位密钥数组在寄存器 7, 本构建轮加盐为 723726 / 855311 (均为立即数)
This commit is contained in:
yml2213
2026-10-11 21:47:27 +08:00
parent e2485c5ea6
commit d52368be2b
4 changed files with 25875 additions and 7747 deletions
+21 -194
View File
@@ -1,200 +1,27 @@
#!/usr/bin/env node
/**
* Tencent Turing Captcha Chaos VM (TDC VM) Disassembler
* --------------------------------------------------------
* 依据 AGENTS.md §3.3 规范,将 TDC 混淆字节码反汇编为可读指令流与结构化操作。
* [已废弃] 旧 TDC VM 反汇编器 (线性扫描, 依赖 Node + Babel)
*
* 实测它有两个缺陷, 会产出**错误**的反汇编, 故已废弃:
* 1. opcode 表是手抄的, 而且对字节码里的**数据区**做线性扫描 —— 一旦扫进数据区就失步,
* 产出 73 处 UNKNOWN (其中 16 处是 >69 的"不可能" opcode);
* 2. 未把 CREATE_CLOSURE(57) 的闭包体 (target) 作为遍历根 —— 导致大半个程序不可达。
*
* 请改用**纯 Python** 的正确实现 (可达性 CFG 反汇编, 0 处非法 opcode):
* .venv/bin/python tools/tdc_disasm.py <tdc.js>
*
* 本文件仅保留为兼容入口: 直接转发给上面的 Python 工具。
*/
const fs = require("fs");
const { spawnSync } = require("child_process");
const path = require("path");
const parser = require("@babel/parser");
const generate = require("@babel/generator").default;
const OP_NAMES = {
0: "PUSH_STR_CHAR",
1: "IS_EXCEPTION",
2: "DUP",
3: "XOR",
4: "EQ_STRICT",
5: "MOD",
6: "CALL_FUNC",
8: "GET_GLOBAL_PROP",
9: "MAKE_PROP_REF",
10: "TYPEOF",
11: "DELETE",
12: "LOGICAL_NOT",
13: "EQ_LOOSE",
14: "SET_PROP",
15: "GREATER_THAN",
17: "SWAP",
18: "IN",
19: "RET_TRUE",
20: "GREATER_EQ",
22: "PUSH_IMM",
23: "MUL",
25: "LOAD_REG_INDIRECT",
26: "ITER_NEXT",
28: "BIND_THIS",
29: "PUSH_NULL",
30: "BITWISE_AND",
31: "PUSH_IMM_ARRAY",
32: "MAKE_REG_PROP_REF",
33: "SUB",
34: "CALL_METHOD",
35: "SHR",
36: "SHL",
37: "JUMP_IF_TRUE",
38: "FOR_IN_KEYS",
39: "ADD",
41: "NEW_METHOD",
42: "PUSH_UNDEFINED",
45: "PUSH_FALSE",
46: "STORE_REG",
47: "ALLOC_REG",
48: "POP_TRY",
49: "THROW",
51: "POP",
52: "DIV",
53: "SET_STACK_LEN",
54: "REVERSE_TOP2",
55: "PUSH_EMPTY_STR",
56: "PUSH_TRY",
57: "CREATE_CLOSURE",
59: "GET_PROP",
60: "JUMP",
61: "NEW_FUNC",
62: "BITWISE_OR",
64: "REPLACE_TOP_IMM",
65: "SHR_UNSIGNED",
66: "PUSH_TRUE",
68: "LOAD_REG_IMM",
69: "CLEAR_EXCEPTION",
};
const tdcFile = process.argv[2] || "tcaptcha/core_js/tdc_sample.js";
const tool = path.join(__dirname, "tdc_disasm.py");
function extractBytecode(tdcPath) {
const code = fs.readFileSync(tdcPath, "utf-8");
const ast = parser.parse(code, { sourceType: "script" });
const fn = ast.program.body[3].declarations[0].init.callee;
const iFuncCode = generate(fn.body.body[1]).code;
const returnStmt = fn.body.body[2];
const vmCall = returnStmt.argument.expressions ? returnStmt.argument.expressions[1] : returnStmt.argument;
const zCall = vmCall.arguments[1];
const zCallCode = generate(zCall).code;
const runner = `
${iFuncCode}
(${zCallCode});
`;
return eval(runner);
console.error("[deprecated] tools/tdc_vm_disasm.js 已废弃 -> 转发到 tools/tdc_disasm.py (纯 Python)");
const r = spawnSync("python3", [tool, tdcFile], { stdio: "inherit" });
if (r.error) {
console.error("无法启动 python3; 请直接运行: .venv/bin/python tools/tdc_disasm.py " + tdcFile);
process.exit(1);
}
function disassemble(z) {
let H = 0;
const rawInstructions = [];
const total = z.length;
while (H < total) {
const pc = H;
const op = z[H++];
const opName = OP_NAMES[op] || `UNKNOWN_${op}`;
const args = [];
switch (op) {
case 0: // PUSH_STR_CHAR
case 22: // PUSH_IMM
case 31: // PUSH_IMM_ARRAY
case 47: // ALLOC_REG
case 53: // SET_STACK_LEN
case 64: // REPLACE_TOP_IMM
case 68: // LOAD_REG_IMM
args.push(z[H++]);
break;
case 6: // CALL_FUNC
case 17: // SWAP
case 34: // CALL_METHOD
case 41: // NEW_METHOD
case 61: // NEW_FUNC
args.push(`args=${z[H++]}`);
break;
case 37: // JUMP_IF_TRUE
case 60: // JUMP
args.push(`target=@${z[H++]}`);
break;
case 56: // PUSH_TRY
args.push(`catchTarget=@${z[H++]}`);
args.push(`catchVarReg=${z[H++]}`);
break;
case 57: { // CREATE_CLOSURE
const targetPc = z[H++];
const capCount = z[H++];
const paramCount = z[H++];
const captures = [];
for (let m = 0; m < capCount; m++) {
captures.push(`r${z[H++]}=r${z[H++]}`);
}
const params = [];
for (let m = 0; m < paramCount; m++) {
params.push(`p${m}->r${z[H++]}`);
}
args.push(`target=@${targetPc}`);
args.push(`caps=[${captures.join(",")}]`);
args.push(`params=[${params.join(",")}]`);
break;
}
default:
// no extra operand
break;
}
rawInstructions.push({ pc, op, opName, args });
}
// Pass 2: Fold PUSH_EMPTY_STR + PUSH_STR_CHAR ... into PUSH_STR "..."
const folded = [];
for (let i = 0; i < rawInstructions.length; i++) {
const ins = rawInstructions[i];
if (ins.opName === "PUSH_EMPTY_STR") {
let str = "";
let j = i + 1;
while (j < rawInstructions.length && rawInstructions[j].opName === "PUSH_STR_CHAR") {
str += String.fromCharCode(rawInstructions[j].args[0]);
j++;
}
if (j > i + 1) {
folded.push({
pc: ins.pc,
op: 0,
opName: "PUSH_STR",
args: [JSON.stringify(str)],
});
i = j - 1;
continue;
}
}
folded.push(ins);
}
return folded;
}
// CLI usage
if (require.main === module) {
const tdcFile = process.argv[2] || "tcaptcha/core_js/tdc_sample.js";
console.log(`Extracting and disassembling ${tdcFile}...`);
const z = extractBytecode(tdcFile);
console.log(`Bytecode size: ${z.length} bytes`);
const insts = disassemble(z);
console.log(`Total instructions: ${insts.length}`);
const outFile = tdcFile.replace(/\.js$/, ".disasm.txt");
const lines = insts.map(
(ins) => `@${ins.pc.toString().padStart(5, "0")}: ${ins.opName.padEnd(18)} ${ins.args.join(" ")}`
);
fs.writeFileSync(outFile, lines.join("\n"), "utf-8");
console.log(`Disassembly written to ${outFile}`);
// Preview first 30 instructions
console.log("\n--- Preview first 30 instructions ---");
lines.slice(0, 30).forEach((l) => console.log(l));
}
module.exports = { extractBytecode, disassemble, OP_NAMES };
process.exit(r.status == null ? 1 : r.status);