本文简单分析介绍了capstone2llvmir源码与本地编译运行的方式，适合初步学习汇编转ir的原理并自己做简单修改，编译运行，做出自己的简易asm2llvmir小程序，有了llvmir，就可以优化、去混淆、干坏事了，详见本菜之前的文章
https://bbs.pediy.com/thread-265335.htm利用编译器优化干掉虚假控制流
https://bbs.pediy.com/thread-266323.htm利用编译器优化干掉控制流平坦化
ps：有些比较复杂的asm2ir转换源码里面没有，需要自己试着写，慢慢完善，然后编译成为自己的工具
recdec的源代码里很重要的部分capstone2llvmir与bin2llvmir，功能是把汇编转换为llvmir，我认真学习了这个神器并记录笔记
源代码https://retdec-tc.avast.com/repository/download/Retdec_DoxygenBuild/.lastSuccessful/build/doc/doxygen/html/files.html
它介绍里面有一个capstone2llvmirtool入门https://github.com/avast/retdec/wiki/Capstone2LlvmIr，我把它大概意思整理了一下
1.完整语义翻译 完全把汇编语法翻译成ir，只对于足够简单的指令 ps：很多不常用指令翻译源码里没有，如果碰到需要模仿源码自己写
2.翻译为内部函数call 把一些汇编翻译成大多数编译器理解的内部函数，比如翻译一些跳转
3.翻译为伪代码call 根据Capstone反汇编信息创建伪代码call翻译指令 ps：看到这些call对应的汇编没被翻译，而它对于优化又很重要，就可以着手自己写翻译函数了，不重要直接忽略就行
4.不翻译 忽略一些难以翻译的指令
1.创建空的LLVM IR module
2.初始化Capstone engine 和其他数据结构
3.创建架构运行环境，也就是寄存器相关数据结构什么的
3.1把汇编地址映射为ir全局变量
@_asm_program_counter = internal global i64 0
; ...
; add eax, 0x1234 @ 0x1000
store volatile i64 4096, i64* @_asm_program_counter
; ... LLVM IR sequence for the add instruction
; sub ebx, 0x1234 @ 0x1005
store volatile i64 4101, i64* @_asm_program_counter
; ... LLVM IR sequence for the sub instruction
3.2控制流伪代码函数生成，为什么不用ir是因为ir通过块标签跳转而不是像汇编一样通过地址
Control-flow-related pseudo functions are generated.
; void (i<architecture_size> target_address)
declare void @__pseudo_call(i32)
; void (i<architecture_size> target_address)
declare void @__pseudo_return(i32)
; void (i<architecture_size> target_address)
declare void @__pseudo_branch(i32)
; void (i1 condition, i<architecture_size> target_address)
declare void @__pseudo_cond_branch(i1, i32)
3.3架构相关寄存器全局变量初始化
@eax = internal global i32 0
@ecx = internal global i32 0
; ...
@st0 = internal global x86_fp80 0xK00000000000000000000
@st1 = internal global x86_fp80 0xK00000000000000000000
1.用Capstone engine 反编译二进制，对于一句汇编，它大概包含如下信息
add eax, 0x1234:
2.找到翻译方式翻译指令到ir id保存了操作码
2.1Capstone ID is mapped to an ID-specific routine 每个id也就是操作码对应一个 routine
2.2Capstone ID is mapped to a specific pseudo assembly generation method
id对应一个 pseudo method汇编伪代码生成方法
2.3Capstone ID is not mapped to any value
啥也没匹配到。使用Capstone-provided instruction info信息自动创建call，这取决于Capstone提供信息的质量
源码结构：
公开接口include/retdec/capstone2llvmir
隐藏接口src/capstone2llvmir
接口Capstone2LlvmIrTranslator
实现Capstone2LlvmIrTranslator_impl
相应架构实现Capstone2LlvmIrTranslatorArm
直接看入口，入口在capstone2llvmirtool/capstone2llvmir.cpp里main 函数(还有一个在retdec\src\bin2llvmir\optimizations\decoder里，学习这2个函数，就能学会如何使用translate函数翻译asm为ir)，先创建一个llvm::function，填入一个block与return，根据cpu架构创建翻译器Capstone2LlvmIrTranslator::createArch，最后通过capstone2llvmir/capstone2llvmir.h定义的translate函数翻译asm为ir，传入data，size，base获得irb
cs_malloc分配capstone的handle，用这个handle通过cs_disasm_iter把二进制翻译为汇编保存在insn
generateSpecialAsm2LlvmInstr ，关键函数generateSpecialAsm2LlvmInstr 把insn的address转换为llvm全局变量，每种架构都有一个程序计数器记录程序执行到哪个地址了，arm就是pc，每执行一句就修改pc，这里的pc值就来源于generateSpecialAsm2LlvmInstr 转换的globalvalue
translateInstruction真正进入到关键把insn翻译为ir，这里4种方式对应前面的4种翻译策略，简单看一下骨架
关键函数translateInstruction，把汇编insn转换为llvmir，它是capstone2llvmir_impl.h声明的一个虚函数
不同的汇编都有自己的translateInstruction实现，arm的在src\capstone2llvmir\arm\arm.cpp
这里面一个重要结构体_cs_insn，电脑里的python3安装了capstone我们翻``python``看它的结构
translateInstruction代码粗看
这里面有一个重要的hash表_i2fm全称Instruction translation map，把汇编指令和翻译ir函数指针一一对应，比如ARM_INS_ADC加法指令对应指针 &Capstone2LlvmIrTranslatorArm_impl::translateAdc
还有arm_init.cpp中定义的寄存器符号名字对应的哈希表r2n，寄存器符号类型对应的哈希表r2t这两个重要结构，他们完全抽象出了arm寄存器为c++数据结构
翻译asm为一般的伪代码函数，就是处理在_i2fm表里面没有对应翻译函数的指令如何翻译
1.根据capstone提供的指令信息，搞明白要生成的ir有多少寄存器与非寄存器的读写，需要创建多少llvm的type和value,函数有没有返回值等信息
2.根据之前创建的llvm的type和value创建参数和返回值，把_asm与助记符insn->mnemonic拼接起来命名函数名字，生成一个空壳伪函数
3.我们在生成ir的时候，如果观察到一些以汇编助记符命名的ir函数，就可以知道这句汇编指令没有对应的翻译函数，然后自己写一个完成完全的翻译，当然，_i2fm表里面给的翻译函数99％情况下够用了
git clone https://github.com/avast/retdec.git
cd retdec
mkdir build && cd build
语法cmake .. -DCMAKE_INSTALL_PREFIX=<path> -DRETDECENABLE<component>=ON
cmake ../ -DRETDEC_ENABLE_CAPSTONE2LLVMIRTOOL=ON 只编译CAPSTONE2LLVMIR前端，这里是原汁原味一句一句翻译asm为ir的逻辑，也就是本文讲的
//cmake ../ -DRETDEC_ENABLE_BIN2LLVMIRTOOL=ON 注意这个是之前版本的，现在已经没有BIN2LLVMIRTOOL了，只有一个库
cmake ../ -DRETDEC_ENABLE_RETDECTOOL=ON 只编译RETDECTOOL前端，也就是之前版本的bin2llvmir前端，这里先通过CAPSTONE2LLVMIR处理得到的ir，然后通过很多pass对于最初的ir进行了分析和优化，其中的到达定值分析和构造西沟分析等都非常的巧妙，值得研究，关键接口函数retdec::disassemble(po.inputFile, &fs)
ps：这里要从git上下载capstone与keystone与llvm相关的库，我下的比较慢 可以修改为国内的源
make -jN (N 一般设置为核心数+1)，然后在retdec\build\src\下面找到可执行文件，像下面这样
retdec-decompiler是bin2llvmir2cpp
retdectool是bin2llvmir(capstone2llvmir+多个pass优化后)
capstone2llvmirtool是capstone2llvmir原汁原味
retdectool也就是以前的bin2llvmir可执行文件，从入口开始学习这个，搞清楚如何通过各种库把汇编转换为ir，然后通过各种分析优化pass得到可读性很强的ir，main函数retdec-master\src\retdectool\retdec.cpp里,关键是disassemble，第一个string指针参数表示待处理文件路径inputPath，第二个生成的ir结果，存储在FunctionSet类型的fs指针，这里可以看一下retdec::common::Function的数据结构，存储了函数类型，ir等有用信息
General info:
id : 8 (add)
addr : 1000
size : 5
bytes : 05 34 12 00 00
mnem : add
op str : eax, 0x1234
Detail info:
R regs : 0
W regs : 1
25 (eflags)
groups : 0
Architecture-dependent info:
prefix : 00 00 00 00 (-, -, -, -)
opcode : 05 00 00 00
rex : 0
addr sz: 4
modrm : 0
sib : 0
disp : 0
sib idx: 0 (-)
sib sc : 0
sib bs : 0 (-)
sse cc : X86_SSE_CC_INVALID
avx cc : X86_AVX_CC_INVALID
avx sae: false
avx rm : X86_AVX_RM_INVALID
op cnt : 2
type : X86_OP_REG
reg : 19 (eax)
size : 4
access : CS_AC_READ + CS_AC_WRITE
avx bct: X86_AVX_BCAST_INVALID
avx 0 m: false
type : X86_OP_IMM
imm : 1234
size : 4
access : CS_AC_INVALID
avx bct: X86_AVX_BCAST_INVALID
avx 0 m: false
General info:
id : 8 (add)
addr : 1000
size : 5
bytes : 05 34 12 00 00
mnem : add
op str : eax, 0x1234
Detail info:
R regs : 0
W regs : 1
25 (eflags)
groups : 0
Architecture-dependent info:
prefix : 00 00 00 00 (-, -, -, -)
opcode : 05 00 00 00
rex : 0
addr sz: 4
modrm : 0
sib : 0
disp : 0
sib idx: 0 (-)
sib sc : 0
sib bs : 0 (-)
sse cc : X86_SSE_CC_INVALID
avx cc : X86_AVX_CC_INVALID
avx sae: false
avx rm : X86_AVX_RM_INVALID
op cnt : 2
type : X86_OP_REG
reg : 19 (eax)
size : 4
access : CS_AC_READ + CS_AC_WRITE
avx bct: X86_AVX_BCAST_INVALID
avx 0 m: false
type : X86_OP_IMM
imm : 1234
size : 4
access : CS_AC_INVALID
avx bct: X86_AVX_BCAST_INVALID
avx 0 m: false
__asm_<mnem>(op0)
op0 = __asm_<mnem>(op0)
__asm_<mnem>(op0, op1)
op0 = __asm_<mnem>(op1)
op0 = __asm_<mnem>(op0, op1)
__asm_<mnem>(op0, op1, op2)
__asm_<mnem>(op0)
op0 = __asm_<mnem>(op0)
__asm_<mnem>(op0, op1)
op0 = __asm_<mnem>(op1)
op0 = __asm_<mnem>(op0, op1)
__asm_<mnem>(op0, op1, op2)
main
{
llvm::Function::Create
llvm::BasicBlock::Create
Capstone2LlvmIrTranslator::createArch
translate(po.code.data(), po.code.size(), po.base, irb)
}
main
{
llvm::Function::Create
llvm::BasicBlock::Create
Capstone2LlvmIrTranslator::createArch
translate(po.code.data(), po.code.size(), po.base, irb)
}
Capstone2LlvmIrTranslator_impl<CInsn, CInsnOp>::translate
{
cs_malloc
cs_disasm_iter
generateSpecialAsm2LlvmInstr
translateInstruction //在capstone2llvmir_impl.h声明的虚函数，不同架构有不同的translateInstruction实现
{
*f=*(_i2fm.find(i→id)) //如果在Instruction translation map _i2fm里找到翻译函数，直接通过指针调用，对应1
{
translateAdd
translateB
...
}
or translatePseudoAsmGeneric //如果没有找到，回到translatePseudoAsmGeneric函数，对应2
{
loadOp
loadRegister
getPseudoAsmFunction
CreateCall //对应3
storeOp
storeRegister
}
}
}
Capstone2LlvmIrTranslator_impl<CInsn, CInsnOp>::translate
{
cs_malloc
cs_disasm_iter
generateSpecialAsm2LlvmInstr
translateInstruction //在capstone2llvmir_impl.h声明的虚函数，不同架构有不同的translateInstruction实现
{
*f=*(_i2fm.find(i→id)) //如果在Instruction translation map _i2fm里找到翻译函数，直接通过指针调用，对应1
{
translateAdd
translateB
...
}
or translatePseudoAsmGeneric //如果没有找到，回到translatePseudoAsmGeneric函数，对应2
{
loadOp
loadRegister
getPseudoAsmFunction
CreateCall //对应3
storeOp
storeRegister
}
}
}
以前读cs的笔记：https://bbs.pediy.com/thread-258473.htm
class _cs_insn(ctypes.Structure):
_fields_ = (
('id', ctypes.c_uint),
('address', ctypes.c_uint64),
('size', ctypes.c_uint16),
('bytes', ctypes.c_ubyte * 16),
('mnemonic', ctypes.c_char * 32),
('op_str', ctypes.c_char * 160),
('detail', ctypes.POINTER(_cs_detail)),
)
class _cs_detail(ctypes.Structure):
_fields_ = (
('regs_read', ctypes.c_uint16 * 12),
('regs_read_count', ctypes.c_ubyte),
('regs_write', ctypes.c_uint16 * 20),
('regs_write_count', ctypes.c_ubyte),
('groups', ctypes.c_ubyte * 8),
('groups_count', ctypes.c_ubyte),
('arch', _cs_arch),
)
class _cs_arch(ctypes.Union):
_fields_ = (
('arm64', arm64.CsArm64),
('arm', arm.CsArm),
('m68k', m68k.CsM68K),
('mips', mips.CsMips),
('x86', x86.CsX86),
('ppc', ppc.CsPpc),
('sparc', sparc.CsSparc),
('sysz', systemz.CsSysz),
('xcore', xcore.CsXcore),
('tms320c64x', tms320c64x.CsTMS320C64x),
('m680x', m680x.CsM680x),
('evm', evm.CsEvm),
)
/// Instruction structure
typedef struct cs_arm {
bool usermode; ///< User-mode registers to be loaded (for LDM/STM instructions)
int vector_size; ///< Scalar size for vector instructions
arm_vectordata_type vector_data; ///< Data type for elements of vector instructions
arm_cpsmode_type cps_mode; ///< CPS mode for CPS instruction
arm_cpsflag_type cps_flag; ///< CPS mode for CPS instruction
arm_cc cc; ///< conditional code for this insn
bool update_flags; ///< does this insn update flags?
bool writeback; ///< does this insn write-back?
arm_mem_barrier mem_barrier; ///< Option for some memory barrier instructions
/// Number of operands of this instruction,
/// or 0 when instruction has no operand.
uint8_t op_count;
cs_arm_op operands[36]; ///< operands for this instruction.
} cs_arm;
typedef enum arm_cc {
ARM_CC_INVALID = 0,
ARM_CC_EQ, ///< Equal Equal
ARM_CC_NE, ///< Not equal Not equal, or unordered
ARM_CC_HS, ///< Carry set >, ==, or unordered
ARM_CC_LO, ///< Carry clear Less than
ARM_CC_MI, ///< Minus, negative Less than
ARM_CC_PL, ///< Plus, positive or zero >, ==, or unordered
ARM_CC_VS, ///< Overflow Unordered
ARM_CC_VC, ///< No overflow Not unordered
ARM_CC_HI, ///< Unsigned higher Greater than, or unordered
ARM_CC_LS, ///< Unsigned lower or same Less than or equal
ARM_CC_GE, ///< Greater than or equal Greater than or equal
ARM_CC_LT, ///< Less than Less than, or unordered
ARM_CC_GT, ///< Greater than Greater than
ARM_CC_LE, ///< Less than or equal <, ==, or unordered
ARM_CC_AL ///< Always (unconditional) Always (unconditional)
} arm_cc;
/// Instruction operand
typedef struct cs_arm_op {
int vector_index; ///< Vector Index for some vector operands (or -1 if irrelevant)
struct {
arm_shifter type;
unsigned int value;
} shift;
arm_op_type type; ///< operand type
union {
int reg; ///< register value for REG/SYSREG operand
int32_t imm; ///< immediate value for C-IMM, P-IMM or IMM operand
double fp; ///< floating point value for FP operand
arm_op_mem mem; ///< base/index/scale/disp value for MEM operand
arm_setend_type setend; ///< SETEND instruction's operand type
};
/// in some instructions, an operand can be subtracted or added to
/// the base register,
/// if TRUE, this operand is subtracted. otherwise, it is added.
bool subtracted;
/// How is this operand accessed? (READ, WRITE or READ|WRITE)
/// This field is combined of cs_ac_type.
/// NOTE: this field is irrelevant if engine is compiled in DIET mode.
uint8_t access;
/// Neon lane index for NEON instructions (or -1 if irrelevant)
int8_t neon_lane;
} cs_arm_op;
以前读cs的笔记：https://bbs.pediy.com/thread-258473.htm
class _cs_insn(ctypes.Structure):
_fields_ = (
('id', ctypes.c_uint),
('address', ctypes.c_uint64),
('size', ctypes.c_uint16),
('bytes', ctypes.c_ubyte * 16),
('mnemonic', ctypes.c_char * 32),
('op_str', ctypes.c_char * 160),
('detail', ctypes.POINTER(_cs_detail)),
)
class _cs_detail(ctypes.Structure):
_fields_ = (
('regs_read', ctypes.c_uint16 * 12),
('regs_read_count', ctypes.c_ubyte),
('regs_write', ctypes.c_uint16 * 20),
('regs_write_count', ctypes.c_ubyte),
('groups', ctypes.c_ubyte * 8),
('groups_count', ctypes.c_ubyte),
('arch', _cs_arch),
)
class _cs_arch(ctypes.Union):
_fields_ = (
('arm64', arm64.CsArm64),
('arm', arm.CsArm),
('m68k', m68k.CsM68K),
('mips', mips.CsMips),
('x86', x86.CsX86),
('ppc', ppc.CsPpc),
('sparc', sparc.CsSparc),
('sysz', systemz.CsSysz),
('xcore', xcore.CsXcore),
('tms320c64x', tms320c64x.CsTMS320C64x),
('m680x', m680x.CsM680x),
('evm', evm.CsEvm),
)
/// Instruction structure
typedef struct cs_arm {
bool usermode; ///< User-mode registers to be loaded (for LDM/STM instructions)
int vector_size; ///< Scalar size for vector instructions
arm_vectordata_type vector_data; ///< Data type for elements of vector instructions
arm_cpsmode_type cps_mode; ///< CPS mode for CPS instruction
arm_cpsflag_type cps_flag; ///< CPS mode for CPS instruction
arm_cc cc; ///< conditional code for this insn
bool update_flags; ///< does this insn update flags?
bool writeback; ///< does this insn write-back?
arm_mem_barrier mem_barrier; ///< Option for some memory barrier instructions
/// Number of operands of this instruction,
/// or 0 when instruction has no operand.
uint8_t op_count;
cs_arm_op operands[36]; ///< operands for this instruction.
} cs_arm;
typedef enum arm_cc {
ARM_CC_INVALID = 0,
ARM_CC_EQ, ///< Equal Equal
ARM_CC_NE, ///< Not equal Not equal, or unordered
ARM_CC_HS, ///< Carry set >, ==, or unordered
ARM_CC_LO, ///< Carry clear Less than
ARM_CC_MI, ///< Minus, negative Less than
ARM_CC_PL, ///< Plus, positive or zero >, ==, or unordered
ARM_CC_VS, ///< Overflow Unordered
ARM_CC_VC, ///< No overflow Not unordered
ARM_CC_HI, ///< Unsigned higher Greater than, or unordered
ARM_CC_LS, ///< Unsigned lower or same Less than or equal
ARM_CC_GE, ///< Greater than or equal Greater than or equal
ARM_CC_LT, ///< Less than Less than, or unordered
ARM_CC_GT, ///< Greater than Greater than
ARM_CC_LE, ///< Less than or equal <, ==, or unordered
ARM_CC_AL ///< Always (unconditional) Always (unconditional)
} arm_cc;
/// Instruction operand
typedef struct cs_arm_op {
int vector_index; ///< Vector Index for some vector operands (or -1 if irrelevant)
struct {
arm_shifter type;
unsigned int value;
} shift;
arm_op_type type; ///< operand type
union {
int reg; ///< register value for REG/SYSREG operand
int32_t imm; ///< immediate value for C-IMM, P-IMM or IMM operand
double fp; ///< floating point value for FP operand
arm_op_mem mem; ///< base/index/scale/disp value for MEM operand
arm_setend_type setend; ///< SETEND instruction's operand type
};
[注意]传递专业知识、拓宽行业人脉——看雪讲师团队等你加入！
