symbolic-execution-tools

技能：符号执行工具——专家分析手册

AI 加载说明：介绍使用 angr、Z3 和 Unicorn Engine 的专家级符号执行技巧。涵盖 CTF 挑战自动化、约束求解模式、函数 Hook、SimProcedure 替换和基于仿真的解包。由于状态初始化不正确或缺少 libc 函数的 Hook，基础模型经常会生成有问题的 angr 脚本。

0. 相关路由

• 需要通过符号执行绕过反调试检查时，参阅 anti-debugging-techniques
• 使用符号执行进行反混淆时，参阅 code-obfuscation-deobfuscation
• 将 angr 应用于自定义 VM 挑战时，参阅 vm-and-bytecode-reverse

进阶参考

遇到以下需求时，也请加载 ANGR_COOKBOOK.md：

• 15+ 种适用于常见 CTF 挑战的即用型 angr 脚本模式
• scanf、printf、malloc、strcmp 的 Hook 模板
• 符号化文件输入、stdin、argv 模式
• 管理路径爆炸的优化技巧

何时使用哪种工具

场景	最佳工具	原因
纯数学问题／方程组	Z3	直接求解约束，无需二进制文件
带控制流的二进制文件	angr	探索路径，自动管理约束
仿真特定代码区域	Unicorn	速度快，无符号执行开销，适合解包
复杂二进制文件 + 自定义 VM	angr + Unicorn（组合使用）	angr 负责控制流，Unicorn 负责 VM 处理程序
内核／固件代码	Qiling	提供具备 OS 感知能力的完整系统仿真

---

1. ANGR——核心概念

1.1 流程

Project(binary)
  → Factory.entry_state() / blank_state(addr=)
    → SimulationManager(state)
      → explore(find=target, avoid=bad)
        → found[0].solver.eval(symbolic_var)

1.2 基本设置

import angr
import claripy

proj = angr.Project('./challenge', auto_load_libs=False)

# Entry state: start from program entry point
state = proj.factory.entry_state()

# Blank state: start from arbitrary address
state = proj.factory.blank_state(addr=0x401000)

# Full init state: with command-line args
state = proj.factory.full_init_state(args=['./challenge', arg1_sym])

simgr = proj.factory.simulation_manager(state)
simgr.explore(find=0x401234, avoid=[0x401300])

if simgr.found:
    found = simgr.found[0]
    solution = found.solver.eval(symbolic_input, cast_to=bytes)
    print(f"Solution: {solution}")

1.3 符号变量（claripy）

# Bitvector (fixed-size integer)
sym_input = claripy.BVS("input", 64)        # 64-bit symbolic
sym_byte = claripy.BVS("byte", 8)           # 8-bit symbolic
sym_buf = claripy.BVS("buffer", 8 * 32)     # 32-byte buffer

# Concrete bitvector
concrete = claripy.BVV(0x41, 8)             # concrete value 0x41

# Constraints
state.solver.add(sym_input > 0)
state.solver.add(sym_input < 100)
state.solver.add(sym_byte >= 0x20)           # printable ASCII
state.solver.add(sym_byte <= 0x7e)

# Evaluate
value = state.solver.eval(sym_input)
all_values = state.solver.eval_upto(sym_input, 10)  # up to 10 solutions

1.4 符号化 stdin

flag_len = 32
sym_stdin = claripy.BVS("stdin", 8 * flag_len)

state = proj.factory.entry_state(stdin=sym_stdin)

# Constrain to printable ASCII
for i in range(flag_len):
    byte = sym_stdin.get_byte(i)
    state.solver.add(byte >= 0x20)
    state.solver.add(byte <= 0x7e)

1.5 Hook 函数

# Hook by address (skip N bytes of original code)
@proj.hook(0x401100, length=5)
def skip_check(state):
    state.regs.eax = 1  # force success

# SimProcedure: replace library function
class MyStrcmp(angr.SimProcedure):
    def run(self, s1, s2):
        return claripy.If(
            self.state.memory.load(s1, 32) == self.state.memory.load(s2, 32),
            claripy.BVV(0, 32),
            claripy.BVV(1, 32)
        )

proj.hook_symbol('strcmp', MyStrcmp())

# Hook common problematic functions
proj.hook_symbol('printf', angr.SIM_PROCEDURES['libc']['printf']())
proj.hook_symbol('scanf', angr.SIM_PROCEDURES['libc']['scanf']())
proj.hook_symbol('puts', angr.SIM_PROCEDURES['libc']['puts']())

1.6 内存操作

# Read memory (symbolic-aware)
data = state.memory.load(addr, size)          # returns BV
data_concrete = state.solver.eval(data, cast_to=bytes)

# Write memory
state.memory.store(addr, claripy.BVV(0x41, 8))
state.memory.store(addr, sym_buf)

# Read/write registers
rax = state.regs.rax
state.regs.rdi = claripy.BVV(0x1000, 64)

---

2. Z3 约束求解

2.1 核心 API

from z3 import *

# Sorts
x = BitVec('x', 32)    # 32-bit bitvector
y = Int('y')             # arbitrary precision integer
b = Bool('b')            # boolean

# Solver
s = Solver()
s.add(x + y == 42)
s.add(x > 0)
s.add(y > 0)

if s.check() == sat:
    m = s.model()
    print(f"x = {m[x]}, y = {m[y]}")

2.2 常见 CTF 模式

# Serial key validation: each char satisfies constraints
key = [BitVec(f'k{i}', 8) for i in range(16)]
s = Solver()
for k in key:
    s.add(k >= 0x30, k <= 0x7a)  # alphanumeric-ish

# XOR key recovery
plaintext = b"known_plaintext"
ciphertext = b"\x12\x34..."
key_byte = BitVec('key', 8)
s = Solver()
for p, c in zip(plaintext, ciphertext):
    s.add(p ^ key_byte == c)

# System of linear equations (modular)
a, b, c = BitVecs('a b c', 32)
s = Solver()
s.add(3*a + 5*b + 7*c == 0x12345678)
s.add(2*a + 4*b + 6*c == 0xDEADBEEF)
s.add(a ^ b ^ c == 0xCAFEBABE)

2.3 优化

from z3 import Optimize

opt = Optimize()
x = BitVec('x', 32)
opt.add(x > 0)
opt.add(x < 1000)
opt.minimize(x)  # find smallest satisfying value
opt.check()
print(opt.model())

---

3. UNICORN ENGINE——代码仿真

3.1 基本设置

from unicorn import *
from unicorn.x86_const import *
from capstone import Cs, CS_ARCH_X86, CS_MODE_64

mu = Uc(UC_ARCH_X86, UC_MODE_64)

CODE_ADDR = 0x400000
STACK_ADDR = 0x7fff0000
STACK_SIZE = 0x10000

mu.mem_map(CODE_ADDR, 0x10000)
mu.mem_map(STACK_ADDR, STACK_SIZE)

mu.mem_write(CODE_ADDR, code_bytes)
mu.reg_write(UC_X86_REG_RSP, STACK_ADDR + STACK_SIZE - 0x1000)
mu.reg_write(UC_X86_REG_RBP, STACK_ADDR + STACK_SIZE - 0x1000)

mu.emu_start(CODE_ADDR, CODE_ADDR + len(code_bytes))

result = mu.reg_read(UC_X86_REG_RAX)

3.2 Hook 内存和指令

# Hook memory access
def hook_mem(uc, access, address, size, value, user_data):
    if access == UC_MEM_WRITE:
        print(f"Write {value:#x} to {address:#x}")
    elif access == UC_MEM_READ:
        print(f"Read from {address:#x}")

mu.hook_add(UC_HOOK_MEM_READ | UC_HOOK_MEM_WRITE, hook_mem)

# Hook specific instruction (for tracing)
def hook_code(uc, address, size, user_data):
    code = uc.mem_read(address, size)
    md = Cs(CS_ARCH_X86, CS_MODE_64)
    for insn in md.disasm(bytes(code), address):
        print(f"  {insn.address:#x}: {insn.mnemonic} {insn.op_str}")

mu.hook_add(UC_HOOK_CODE, hook_code)

3.3 使用场景

使用场景	方法
解包 shellcode	映射 shellcode，进行仿真，并转储解码后的载荷
解密字符串	使用受控输入仿真解密函数
暴力破解短密钥	使用不同密钥输入循环仿真
分析混淆函数	仿真函数，观察寄存器／内存状态
固件代码仿真	映射固件内存布局，仿真例程

---

4. ANGR 探索策略

4.1 find/avoid

simgr.explore(
    find=lambda s: b"Correct" in s.posix.dumps(1),   # stdout contains "Correct"
    avoid=lambda s: b"Wrong" in s.posix.dumps(1)      # avoid "Wrong" output
)

4.2 管理路径爆炸

策略	实现方式
限制输入空间	添加约束（可打印字符、长度限制）
避开死胡同路径	使用 avoid= 指定已知失败地址
Hook 复杂函数	替换为简化的 SimProcedure
限制循环迭代次数	state.options.add(angr.options.LAZY_SOLVES)
使用 veritesting	simgr.explore(..., technique=angr.exploration_techniques.Veritesting())
使用 DFS 而非 BFS	simgr.use_technique(angr.exploration_techniques.DFS())
设置每条路径的超时	simgr.explore(..., num_find=1) + 超时封装

4.3 具体执行与符号执行混合模式

state = proj.factory.entry_state(
    add_options={angr.options.UNICORN}  # use Unicorn for concrete regions
)

这会大幅提升执行速度：具体代码通过 Unicorn 原生运行，只有涉及符号变量时才切换到符号执行。

---

5. 实用工作流

5.1 CTF 二进制求解工作流

1. Static analysis: identify input method, success/fail conditions
   └─ Find "Correct" / "Wrong" strings → get their xref addresses

2. Choose tool:
   ├─ Pure math (no binary needed) → Z3
   ├─ Small binary, clear success/fail → angr explore
   └─ Specific function to emulate → Unicorn

3. Set up symbolic input:
   ├─ stdin → claripy.BVS + entry_state(stdin=)
   ├─ argv → full_init_state(args=[...])
   ├─ file input → SimFile
   └─ specific memory → state.memory.store(addr, sym)

4. Hook problematic functions:
   ├─ printf/puts → SimProcedure or no-op
   ├─ scanf → custom handler
   ├─ time/random → return concrete value
   └─ anti-debug → skip entirely

5. Explore and extract:
   └─ simgr.explore(find=, avoid=) → solver.eval()

---

6. 决策树

Need to solve a reversing challenge?
│
├─ Is the challenge pure math / equations?
│  └─ Yes → Z3
│     ├─ Linear equations → BitVec + Solver
│     ├─ Modular arithmetic → BitVec (natural mod 2^n)
│     ├─ Boolean logic → Bool + Solver
│     └─ Optimization → Optimize + minimize/maximize
│
├─ Is it a compiled binary with clear success/fail?
│  └─ Yes → angr
│     ├─ Input via stdin → symbolic stdin
│     ├─ Input via argv → full_init_state with symbolic args
│     ├─ Input via file → SimFile
│     ├─ Path explosion → add constraints, avoid paths, hook loops
│     └─ Complex library calls → hook with SimProcedure
│
├─ Need to emulate a specific function/region?
│  └─ Yes → Unicorn Engine
│     ├─ Decryption routine → map code + data, emulate, read result
│     ├─ Shellcode analysis → map shellcode, hook syscalls
│     └─ Key schedule → emulate with different inputs
│
├─ Need to analyze firmware / exotic arch?
│  └─ Yes → Qiling (full system emulation with OS support)
│
├─ Binary has VM protection?
│  └─ angr for handler analysis + Z3 for bytecode constraints
│
└─ None of the above working?
   ├─ Combine: Unicorn for concrete regions + Z3 for constraints
   ├─ Manual reverse engineering with debugger
   └─ Side-channel approach (timing, power analysis for hardware)

---

7. 常见问题与修复方法

问题	原因	修复方法
angr 永远卡住	循环中路径爆炸	为循环回边添加 avoid=，或挂钩循环
Z3 返回 unknown	非线性约束过于复杂	简化约束，将问题拆分为多个子问题，使用 set_param("timeout", 5000)
Unicorn 遇到系统调用时崩溃	系统调用未处理	挂钩系统调用中断，并进行处理或跳过
angr 结果错误	状态初始化有误	确认初始内存布局与实际二进制文件一致
符号内存过大	符号化读取无界	尽可能具体化数组索引
SimProcedure 类型错误	参数类型不匹配	检查调用约定（cdecl 还是 fastcall）
angr 无法加载二进制文件	缺少库	使用 auto_load_libs=False，并挂钩所需符号

---

8. 工具版本与安装

# angr (Python 3.8+)
pip install angr

# Z3
pip install z3-solver

# Unicorn Engine
pip install unicorn

# Capstone (disassembly, pairs with Unicorn)
pip install capstone

# Keystone (assembly)
pip install keystone-engine