admin/awesome_anti_virus_engine
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add machine learning feature extraction for PE files

Browse Source 
- Implemented MachineLearning class with ExtractFeatures method
- Updated project files to include new machine learning source and header files
- Modified main executable to call feature extraction
- Updated VSCode settings to include additional C++ headers
- Commented out previous file dumping code in main function
This commit is contained in:
Huoji's
2025-03-09 02:05:07 +08:00
parent d2ed7936df
commit 1cea516cf7
9 changed files with 790 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
.vscode/settings.json vendored
View File

@@ -56,6 +56,8 @@
"xtr1common": "cpp",
"xtree": "cpp",
"xutility": "cpp",
"functional": "cpp"
"functional": "cpp",
"array": "cpp",
"numeric": "cpp"
}
}

4
ai_anti_malware/ai_anti_malware.cpp
View File

@@ -46,7 +46,9 @@ int main() {
if (peBuffer) {
printf("peBuffer: %p\n", peBuffer.get());
printf("peSize: %d\n", peSize);
peconv::dump_to_file("z:\\dumped_main.exe", peBuffer.get(), peSize);
// peconv::dump_to_file("z:\\dumped_main.exe", peBuffer.get(), peSize);
MachineLearning ml;
ml.ExtractFeatures(peBuffer.get(), peSize, "z:\\features.txt");
}
peBuffer.release();
system("pause");

2
ai_anti_malware/ai_anti_malware.vcxproj
View File

@@ -170,6 +170,7 @@
<ClCompile Include="libpeconv\libpeconv\src\resource_parser.cpp" />
<ClCompile Include="libpeconv\libpeconv\src\resource_util.cpp" />
<ClCompile Include="libpeconv\libpeconv\src\util.cpp" />
<ClCompile Include="ml.cpp" />
<ClCompile Include="sandbox.cpp" />
<ClCompile Include="sandbox_api_emu.cpp" />
<ClCompile Include="sandbox_callbacks.cpp" />
@@ -178,6 +179,7 @@
<ClInclude Include="head.h" />
<ClInclude Include="libpeconv\libpeconv\src\fix_dot_net_ep.h" />
<ClInclude Include="libpeconv\libpeconv\src\ntddk.h" />
<ClInclude Include="ml.h" />
<ClInclude Include="native_struct.h" />
<ClInclude Include="sandbox.h" />
<ClInclude Include="sandbox_callbacks.h" />

12
ai_anti_malware/ai_anti_malware.vcxproj.filters
View File

@@ -22,6 +22,12 @@
<Filter Include="头文件\libpe">
<UniqueIdentifier>{38ea362d-55dc-410e-92f1-3a44ced4dc2d}</UniqueIdentifier>
</Filter>
<Filter Include="源文件\machine_learning">
<UniqueIdentifier>{2b38b24a-cb8f-41db-bd53-4a25f8152c17}</UniqueIdentifier>
</Filter>
<Filter Include="头文件\machine_learning">
<UniqueIdentifier>{65a79261-ea29-4842-b41c-7983eddbdc85}</UniqueIdentifier>
</Filter>
</ItemGroup>
<ItemGroup>
<ClCompile Include="ai_anti_malware.cpp">
@@ -117,6 +123,9 @@
<ClCompile Include="sandbox_api_emu.cpp">
<Filter>源文件\sandbox</Filter>
</ClCompile>
<ClCompile Include="ml.cpp">
<Filter>源文件\machine_learning</Filter>
</ClCompile>
</ItemGroup>
<ItemGroup>
<ClInclude Include="head.h">
@@ -137,5 +146,8 @@
<ClInclude Include="sandbox_callbacks.h">
<Filter>头文件\sandbox</Filter>
</ClInclude>
<ClInclude Include="ml.h">
<Filter>头文件\machine_learning</Filter>
</ClInclude>
</ItemGroup>
</Project>

1
ai_anti_malware/head.h
View File

@@ -29,3 +29,4 @@ struct BasicPeInfo {
PIMAGE_NT_HEADERS32 ntHead32;
};
#include "sandbox.h"
#include "ml.h"

591
ai_anti_malware/ml.cpp Normal file
View File

@@ -0,0 +1,591 @@
#include "ml.h"
#include <array>
#include <limits>
#include <algorithm>
#include <cmath>
#include <functional>
#include <iomanip>
#include <sstream>
#include <cfloat>
// 确保std命名空间中的函数可用
using std::max;
using std::min;
MachineLearning::MachineLearning() {
// 初始化属性列表
_properties = {"has_configuration", "has_debug", "has_exceptions",
"has_exports", "has_imports", "has_nx",
"has_relocations", "has_resources", "has_signatures",
"has_tls", "has_entry_iat", "has_image_base",
"has_delay_imports", "has_rich"};
// 初始化库列表
_libraries = {"libssp-0",
"kernel32",
"user32",
"advapi32",
"oleaut32",
"shell32",
"ole32",
"gdi32",
"comctl32",
"version",
"msvcrt",
"comdlg32",
"shlwapi",
"wininet",
"ws2_32",
"winmm",
"winspool.drv",
"wsock32",
"msvbvm60",
"rpcrt4",
"mpr",
"psapi",
"iphlpapi",
"ntdll",
"msimg32",
"mscoree",
"crypt32",
"gdiplus",
"userenv",
"crtdll",
"oledlg",
"mfc42",
"urlmon",
"imm32",
"rtl100.bpl",
"netapi32",
"wintrust",
"vcl100.bpl",
"vcl50.bpl",
"uxtheme",
"setupapi",
"ntoskrnl.pe",
"msi",
"msvcp60",
"lz32",
"winhttp",
"hal",
"core.bpl",
"rbrcl1416.bpl",
"dbghelp",
"api-ms-win-crt-runtime-l1-1-0",
"api-ms-win-crt-heap-l1-1-0",
"api-ms-win-crt-math-l1-1-0",
"api-ms-win-crt-stdio-l1-1-0",
"api-ms-win-crt-locale-l1-1-0",
"oleacc",
"komponentyd17.bpl",
"job.bpl",
"cam.bpl",
"vcruntime140",
"secur32",
"msvcr100",
"cxeditorsrs17.bpl",
"rasapi32",
"api-ms-win-crt-string-l1-1-0",
"wtsapi32",
"imagehlp",
"msvcp140",
"cnc.bpl",
"indyprotocols190.bpl",
"api-ms-win-crt-convert-l1-1-0",
"msvcr120",
"vcl60.bpl",
"rbrcl210.bpl",
"rtl170.bpl",
"rbide1416.bpl",
"rtl60.bpl",
"vcl170.bpl",
"wldap32",
"shfolder",
"cxlibraryrs17.bpl",
"msvcirt",
"report.bpl",
"rtl190.bpl",
"msvcr90",
"api-ms-win-crt-filesystem-l1-1-0",
"cxeditorsrs16.bpl",
"avifil32",
"api-ms-win-crt-time-l1-1-0",
"jli",
"graphic.bpl",
"olepro32",
"rtl160.bpl",
"spmmachine.bpl",
"cabinet",
"indycore190.bpl",
"sacom210.bpl",
"rbrtl1416.bpl",
"api-ms-win-crt-utility-l1-1-0",
"vcl160.bpl",
"api-ms-win-crt-environment-l1-1-0",
"zcomponent170.bpl",
"msvfw32",
"libadm_coreutils6",
"rbsha",
"dxpscorers16.bpl",
"msacm32",
"vcl70.bpl",
"applicationmanagement.bpl",
"jobgui.bpl",
"indyprotocols170.bpl",
"rtl70.bpl",
"cxed210.bpl",
"msvcr80",
"libadm_coretinypy6",
"ucrtbased",
"vcruntime140d",
"msvcp120",
"msvcp140d",
"dinput8",
"gui.bpl",
"maincontrols.bpl",
"rtl120.bpl",
"jcl170.bpl",
"frx17.bpl",
"fs17.bpl",
"vcl190.bpl",
"sdl2",
"machine.bpl",
"mfc42u",
"normaliz",
"sdl2_gfx",
"sdl2_ttf",
"sdl2_mixer",
"msvcp80",
"cxgridrs17.bpl",
"cxeditorsvcld7.bpl",
"libeay32",
"cxlibraryd11.bpl",
"vcl120.bpl",
"gr32_d6.bpl",
"cxlibraryrs16.bpl",
"cxgridrs16.bpl",
"vcl40.bpl",
"opengl32",
"qt5core",
"qtcore4",
"wdfldr.sys",
"nesting.bpl",
"fltmgr.sys"};
}
MachineLearning::~MachineLearning() {
// 析构函数，清理资源（如有必要）
}
bool MachineLearning::ExtractFeatures(const uint8_t* buffer, size_t bufferSize,
const std::string& outputPath) {
// 使用libpeconv解析PE文件
size_t v_size = 0;
BYTE* peBuffer = peconv::load_pe_module(const_cast<BYTE*>(buffer),
bufferSize, v_size, false, false);
if (!peBuffer) {
std::cerr << "无法加载PE文件" << std::endl;
return false;
}
// 解析PE信息
PeInfo peInfo;
std::vector<SectionInfo> sections;
std::vector<std::string> importedLibraries;
std::vector<uint8_t> entrypointBytes;
// 检查是否为64位PE
peInfo.isX64 = peconv::is64bit(peBuffer);
// 获取PE头信息
PIMAGE_NT_HEADERS ntHeaders =
(PIMAGE_NT_HEADERS)peconv::get_nt_hdrs(peBuffer);
if (!ntHeaders) {
peconv::free_pe_buffer(peBuffer);
return false;
}
// 从NT头部获取信息
if (peInfo.isX64) {
// 64位PE文件
PIMAGE_NT_HEADERS64 ntHeaders64 = (PIMAGE_NT_HEADERS64)ntHeaders;
peInfo.addressOfEntryPoint =
ntHeaders64->OptionalHeader.AddressOfEntryPoint;
peInfo.baseOfCode = ntHeaders64->OptionalHeader.BaseOfCode;
peInfo.sizeOfCode = ntHeaders64->OptionalHeader.SizeOfCode;
peInfo.sizeOfImage = ntHeaders64->OptionalHeader.SizeOfImage;
peInfo.sizeOfHeaders = ntHeaders64->OptionalHeader.SizeOfHeaders;
peInfo.characteristics = ntHeaders64->FileHeader.Characteristics;
peInfo.dllCharacteristics =
ntHeaders64->OptionalHeader.DllCharacteristics;
} else {
// 32位PE文件
PIMAGE_NT_HEADERS32 ntHeaders32 = (PIMAGE_NT_HEADERS32)ntHeaders;
peInfo.addressOfEntryPoint =
ntHeaders32->OptionalHeader.AddressOfEntryPoint;
peInfo.baseOfCode = ntHeaders32->OptionalHeader.BaseOfCode;
peInfo.sizeOfCode = ntHeaders32->OptionalHeader.SizeOfCode;
peInfo.sizeOfImage = ntHeaders32->OptionalHeader.SizeOfImage;
peInfo.sizeOfHeaders = ntHeaders32->OptionalHeader.SizeOfHeaders;
peInfo.characteristics = ntHeaders32->FileHeader.Characteristics;
peInfo.dllCharacteristics =
ntHeaders32->OptionalHeader.DllCharacteristics;
}
// 检查PE目录
IMAGE_DATA_DIRECTORY* dataDir = peconv::get_directory_entry(
peBuffer, IMAGE_DIRECTORY_ENTRY_COM_DESCRIPTOR);
peInfo.hasConfiguration = dataDir && dataDir->VirtualAddress != 0;
dataDir =
peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_DEBUG);
peInfo.hasDebug = dataDir && dataDir->VirtualAddress != 0;
dataDir =
peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_EXCEPTION);
peInfo.hasExceptions = dataDir && dataDir->VirtualAddress != 0;
dataDir =
peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_EXPORT);
peInfo.hasExports = dataDir && dataDir->VirtualAddress != 0;
dataDir =
peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_IMPORT);
peInfo.hasImports = dataDir && dataDir->VirtualAddress != 0;
// NX标志检查
peInfo.hasNx =
(peInfo.dllCharacteristics & IMAGE_DLLCHARACTERISTICS_NX_COMPAT) != 0;
dataDir =
peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_BASERELOC);
peInfo.hasRelocations = dataDir && dataDir->VirtualAddress != 0;
dataDir =
peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_RESOURCE);
peInfo.hasResources = dataDir && dataDir->VirtualAddress != 0;
dataDir =
peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_SECURITY);
peInfo.hasSignatures = dataDir && dataDir->VirtualAddress != 0;
dataDir = peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_TLS);
peInfo.hasTls = dataDir && dataDir->VirtualAddress != 0;
dataDir = peconv::get_directory_entry(peBuffer,
IMAGE_DIRECTORY_ENTRY_DELAY_IMPORT);
peInfo.hasDelayImports = dataDir && dataDir->VirtualAddress != 0;
peInfo.hasImageBase = true; // PE文件都有ImageBase
dataDir = peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_IAT);
peInfo.hasEntryIat = dataDir && dataDir->VirtualAddress != 0;
// Rich头部检测
peInfo.hasRich = false;
PIMAGE_DOS_HEADER dosHeader = reinterpret_cast<PIMAGE_DOS_HEADER>(peBuffer);
if (dosHeader && dosHeader->e_magic == IMAGE_DOS_SIGNATURE) {
const uint32_t* richPtr = reinterpret_cast<const uint32_t*>(
peBuffer + sizeof(IMAGE_DOS_HEADER));
size_t maxLen = dosHeader->e_lfanew - sizeof(IMAGE_DOS_HEADER);
for (size_t i = 0; i < maxLen / 4 - 1; i++) {
if (richPtr[i] == 0x68636952) { // "Rich"
peInfo.hasRich = true;
break;
}
}
}
// 获取导入DLL列表
if (peInfo.hasImports) {
size_t impRva = 0;
IMAGE_DATA_DIRECTORY* impDir =
peconv::get_directory_entry(peBuffer, IMAGE_DIRECTORY_ENTRY_IMPORT);
if (impDir) {
impRva = impDir->VirtualAddress;
IMAGE_IMPORT_DESCRIPTOR* impDesc =
reinterpret_cast<IMAGE_IMPORT_DESCRIPTOR*>(
RvaToPtr(impRva, peBuffer));
while (impDesc && impDesc->Name != 0) {
char* libName =
reinterpret_cast<char*>(RvaToPtr(impDesc->Name, peBuffer));
if (libName) {
std::string libNameStr = libName;
std::transform(libNameStr.begin(), libNameStr.end(),
libNameStr.begin(), [](unsigned char c) {
return std::tolower(c);
});
importedLibraries.push_back(libNameStr);
}
impDesc++;
}
}
}
// 获取节区信息
size_t sectionsCount = peconv::get_sections_count(peBuffer, bufferSize);
for (size_t i = 0; i < sectionsCount; i++) {
PIMAGE_SECTION_HEADER section =
peconv::get_section_hdr(peBuffer, bufferSize, i);
if (!section) continue;
SectionInfo secInfo;
secInfo.characteristics = section->Characteristics;
secInfo.sizeOfRawData = section->SizeOfRawData;
secInfo.virtualSize = section->Misc.VirtualSize;
// 计算节区熵
BYTE* sectionData = RvaToPtr(section->VirtualAddress, peBuffer);
secInfo.entropy =
(sectionData && section->SizeOfRawData > 0)
? CalculateEntropy(sectionData, section->SizeOfRawData)
: 0.0;
sections.push_back(secInfo);
}
// 获取入口点前255字节
if (peInfo.addressOfEntryPoint > 0) {
BYTE* epPtr = RvaToPtr(peInfo.addressOfEntryPoint, peBuffer);
if (epPtr) {
// 确保不会越界
size_t maxBytes =
std::min<size_t>(255, bufferSize - (epPtr - peBuffer));
entrypointBytes.assign(epPtr, epPtr + maxBytes);
}
}
// 提取所有特征
std::vector<double> allFeatures;
// 1. PE段属性
std::vector<double> propFeatures =
EncodeProperties(peInfo, importedLibraries);
allFeatures.insert(allFeatures.end(), propFeatures.begin(),
propFeatures.end());
// 2. 导入DLL检测
std::vector<double> libFeatures = EncodeLibraries(importedLibraries);
allFeatures.insert(allFeatures.end(), libFeatures.begin(),
libFeatures.end());
// 3. 文件熵
double fileEntropy = CalculateEntropy(buffer, bufferSize);
allFeatures.push_back(fileEntropy);
// 4. 入口点前255字节
std::vector<double> epFeatures = EncodeEntrypoint(entrypointBytes);
allFeatures.insert(allFeatures.end(), epFeatures.begin(), epFeatures.end());
// 5. 节区信息
std::vector<double> secFeatures = EncodeSections(sections, peInfo.isX64);
allFeatures.insert(allFeatures.end(), secFeatures.begin(),
secFeatures.end());
// 6. 文件和代码段的比率
double codeRatio =
(peInfo.sizeOfCode > 0 && peInfo.sizeOfImage > 0)
? static_cast<double>(peInfo.sizeOfCode) / peInfo.sizeOfImage
: 0.0;
allFeatures.push_back(codeRatio);
// 7. 节区数量
allFeatures.push_back(static_cast<double>(sections.size()));
// 导出特征到CSV
bool result = ExportToCSV(allFeatures, outputPath);
// 清理资源
peconv::free_pe_buffer(peBuffer);
return result;
}
std::vector<double> MachineLearning::EncodeProperties(
const PeInfo& peInfo, const std::vector<std::string>& dllTables) {
std::vector<double> features;
// 添加各属性的布尔值（转为double: 1.0=true, 0.0=false）
features.push_back(peInfo.hasConfiguration ? 1.0 : 0.0);
features.push_back(peInfo.hasDebug ? 1.0 : 0.0);
features.push_back(peInfo.hasExceptions ? 1.0 : 0.0);
features.push_back(peInfo.hasExports ? 1.0 : 0.0);
features.push_back(peInfo.hasImports ? 1.0 : 0.0);
features.push_back(peInfo.hasNx ? 1.0 : 0.0);
features.push_back(peInfo.hasRelocations ? 1.0 : 0.0);
features.push_back(peInfo.hasResources ? 1.0 : 0.0);
features.push_back(peInfo.hasSignatures ? 1.0 : 0.0);
features.push_back(peInfo.hasTls ? 1.0 : 0.0);
features.push_back(peInfo.hasEntryIat ? 1.0 : 0.0);
features.push_back(peInfo.hasImageBase ? 1.0 : 0.0);
features.push_back(peInfo.hasDelayImports ? 1.0 : 0.0);
features.push_back(peInfo.hasRich ? 1.0 : 0.0);
return features;
}
std::vector<double> MachineLearning::EncodeEntrypoint(
const std::vector<uint8_t>& epBytes) {
std::vector<double> features;
// 原始字节转为浮点值（按Python代码中的normalize处理）
for (const auto& byte : epBytes) {
features.push_back(static_cast<double>(byte) / 255.0);
}
// 填充至64字节长度
while (features.size() < 64) {
features.push_back(0.0);
}
return features;
}
std::vector<double> MachineLearning::EncodeHistogram(const uint8_t* data,
size_t size) {
std::vector<double> features(256, 0.0);
if (data && size > 0) {
// 统计字节频率
for (size_t i = 0; i < size; i++) {
features[data[i]]++;
}
// 归一化频率
for (auto& freq : features) {
freq /= static_cast<double>(size);
}
}
return features;
}
std::vector<double> MachineLearning::EncodeLibraries(
const std::vector<std::string>& importedLibraries) {
std::vector<double> features(_libraries.size(), 0.0);
// 检查每个库是否被导入
for (size_t i = 0; i < _libraries.size(); i++) {
const std::string& lib = _libraries[i];
for (const auto& imported : importedLibraries) {
if (imported.find(lib) != std::string::npos) {
features[i] = 1.0;
break;
}
}
}
return features;
}
std::vector<double> MachineLearning::EncodeSections(
const std::vector<SectionInfo>& sections, bool isX64) {
std::vector<double> features;
size_t numSections = sections.size();
if (numSections == 0) {
return std::vector<double>(5, 0.0); // 返回全零特征
}
// 计算熵特征
double totalEntropy = 0.0;
double maxEntropy = 0.0;
for (const auto& sec : sections) {
totalEntropy += sec.entropy;
if (sec.entropy > maxEntropy) {
maxEntropy = sec.entropy;
}
}
double avgEntropy = totalEntropy / numSections;
double normAvgEntropy = (maxEntropy > 0) ? avgEntropy / maxEntropy : 0.0;
// 计算大小比率
double maxSize = 0.0;
double minVSize = DBL_MAX;
for (const auto& sec : sections) {
if (static_cast<double>(sec.sizeOfRawData) > maxSize) {
maxSize = static_cast<double>(sec.sizeOfRawData);
}
if (sec.virtualSize > 0 &&
static_cast<double>(sec.virtualSize) < minVSize) {
minVSize = static_cast<double>(sec.virtualSize);
}
}
// 根据PE文件类型调整计算方式
double normSize = 0.0;
if (minVSize > 0 && minVSize != DBL_MAX) {
if (isX64) {
// 64位PE文件可能有更大的对齐要求
normSize = maxSize / (minVSize * 2.0);
} else {
// 32位PE文件的处理方式
normSize = maxSize / minVSize;
}
}
// 返回特征
features.push_back(static_cast<double>(numSections));
features.push_back(avgEntropy);
features.push_back(maxEntropy);
features.push_back(normAvgEntropy);
features.push_back(normSize);
return features;
}
double MachineLearning::CalculateEntropy(const uint8_t* data, size_t size) {
if (!data || size == 0) {
return 0.0;
}
std::array<double, 256> frequencies = {};
// 统计每个字节的频率
for (size_t i = 0; i < size; i++) {
frequencies[data[i]] += 1.0;
}
// 计算香农熵
double entropy = 0.0;
for (const auto& freq : frequencies) {
if (freq > 0) {
double p = freq / static_cast<double>(size);
entropy -= p * std::log2(p);
}
}
return entropy;
}
bool MachineLearning::ExportToCSV(const std::vector<double>& features,
const std::string& outputPath) {
std::ofstream outFile(outputPath);
if (!outFile.is_open()) {
std::cerr << "无法打开输出文件: " << outputPath << std::endl;
return false;
}
// 写入特征
for (size_t i = 0; i < features.size(); i++) {
outFile << std::fixed << std::setprecision(6) << features[i];
if (i < features.size() - 1) {
outFile << ",";
}
}
outFile << std::endl;
outFile.close();
return true;
}
int MachineLearning::GetOpcodeType(const void* code, bool isX64) {
// 此函数未使用，但保留实现接口
return 0;
}
std::tuple<std::vector<double>, std::vector<int>>
MachineLearning::GetOpcodeStatistics(const uint8_t* data, size_t dataSize,
bool isX64, const PeInfo& peInfo) {
// 此函数未使用，但保留实现接口
return std::make_tuple(std::vector<double>(), std::vector<int>());
}

128
ai_anti_malware/ml.h Normal file
View File

@@ -0,0 +1,128 @@
#pragma once
#include "head.h"
#include <vector>
#include <string>
#include <map>
#include <memory>
#include <cmath>
#include <fstream>
#include <algorithm>
#include <numeric>
#include <functional>
#include <unordered_map>
// 前向声明
struct PeInfo;
struct SectionInfo;
class BasicPeInfo;
// RVA转换为内存中的指针的辅助函数
inline BYTE* RvaToPtr(DWORD rva, BYTE* peBuffer) {
if (!peBuffer || rva == 0) return nullptr;
PIMAGE_NT_HEADERS ntHeaders =
(PIMAGE_NT_HEADERS)peconv::get_nt_hdrs(peBuffer);
if (!ntHeaders) return nullptr;
PIMAGE_SECTION_HEADER section = IMAGE_FIRST_SECTION(ntHeaders);
WORD numSections = ntHeaders->FileHeader.NumberOfSections;
for (WORD i = 0; i < numSections; i++, section++) {
// 检查RVA是否在这个节区范围内
if (rva >= section->VirtualAddress &&
rva < section->VirtualAddress + section->Misc.VirtualSize) {
// 计算文件偏移
DWORD offset =
rva - section->VirtualAddress + section->PointerToRawData;
return peBuffer + offset;
}
}
// 如果RVA在PE头部内
DWORD sizeOfHeaders = 0;
bool isX64 = peconv::is64bit(peBuffer);
if (isX64) {
PIMAGE_NT_HEADERS64 ntHeaders64 = (PIMAGE_NT_HEADERS64)ntHeaders;
sizeOfHeaders = ntHeaders64->OptionalHeader.SizeOfHeaders;
} else {
PIMAGE_NT_HEADERS32 ntHeaders32 = (PIMAGE_NT_HEADERS32)ntHeaders;
sizeOfHeaders = ntHeaders32->OptionalHeader.SizeOfHeaders;
}
if (rva < sizeOfHeaders) {
return peBuffer + rva;
}
return nullptr;
}
class MachineLearning {
public:
MachineLearning();
~MachineLearning();
// 主函数：提取特征并导出到CSV
bool ExtractFeatures(const uint8_t* buffer, size_t bufferSize,
const std::string& outputPath);
private:
// 特征提取辅助函数
std::vector<double> EncodeProperties(
const PeInfo& peInfo, const std::vector<std::string>& dllTables);
std::vector<double> EncodeEntrypoint(const std::vector<uint8_t>& epBytes);
std::vector<double> EncodeHistogram(const uint8_t* data, size_t size);
std::vector<double> EncodeLibraries(
const std::vector<std::string>& dllTable);
std::vector<double> EncodeSections(const std::vector<SectionInfo>& sections,
bool isX64);
std::tuple<std::vector<double>, std::vector<int>> GetOpcodeStatistics(
const uint8_t* data, size_t dataSize, bool isX64, const PeInfo& peInfo);
int GetOpcodeType(const void* code, bool isX64);
double CalculateEntropy(const uint8_t* data, size_t size);
// 将特征导出到CSV
bool ExportToCSV(const std::vector<double>& features,
const std::string& outputPath);
// 常量定义
std::vector<std::string> _properties;
std::vector<std::string> _libraries;
std::unordered_map<std::string, int> _opcodeTypeDict;
};
// PE文件信息结构
struct PeInfo {
uint32_t addressOfEntryPoint;
uint32_t baseOfCode;
uint32_t sizeOfCode;
uint32_t sizeOfImage;
uint32_t sizeOfHeaders;
uint32_t characteristics;
uint32_t dllCharacteristics;
bool isX64;
// PE目录标志
bool hasConfiguration;
bool hasDebug;
bool hasExceptions;
bool hasExports;
bool hasImports;
bool hasNx; // NX兼容标志
bool hasRelocations;
bool hasResources;
bool hasSignatures;
bool hasTls;
bool hasDelayImports;
bool hasImageBase;
bool hasEntryIat;
bool hasRich;
};
// 节区信息结构
struct SectionInfo {
uint32_t characteristics;
double entropy;
uint32_t sizeOfRawData;
uint32_t virtualSize;
};

79
ai_anti_malware/sandbox.cpp
View File

@@ -164,7 +164,44 @@ class cFixImprot : public peconv::t_function_resolver {
};
Sandbox::Sandbox() {}
Sandbox::~Sandbox() {}
Sandbox::~Sandbox() {
// 1. 先清理高层资源
m_crossSectionExecution.clear();
envStrings.clear();
api_map.clear();
m_moduleList.clear();
m_impFuncDict.clear();
m_exportFuncDict.clear();
// 2. 清理内存映射
if (m_ucEngine) {
uc_close(m_ucEngine);
m_ucEngine = nullptr;
}
// 3. 清理堆内存
for (auto& [address, segment] : m_heapSegments) {
HeapBlock* current = segment->blocks;
while (current) {
HeapBlock* next = current->next;
delete current;
current = next;
}
delete segment;
}
m_heapSegments.clear();
// 4. 清理栈内存
if (m_stackBuffer) {
free(m_stackBuffer);
m_stackBuffer = nullptr;
}
// 5. 最后清理底层资源
if (m_csHandle) {
cs_close(&m_csHandle);
}
}
auto Sandbox::PushModuleToVM(const char* dllName, uint64_t moduleBase) -> void {
for (auto module : m_moduleList) {
@@ -401,9 +438,9 @@ auto Sandbox::SetupVirtualMachine() -> void {
/*
映射 m_KSharedUserDataBase
*/
uint64_t m_KSharedUserDataBase = 0x7FFE0000;
m_KSharedUserDataBase = 0x7FFE0000;
uint64_t m_KSharedUserDataEnd = 0x7FFE0FFF; // 0x7FFE2000
uint64_t m_KSharedUserDataSize = AlignToSectionAlignment(
m_KSharedUserDataSize = AlignToSectionAlignment(
m_KSharedUserDataEnd - m_KSharedUserDataBase, PAGE_SIZE);
uc_mem_map(m_ucEngine, m_KSharedUserDataBase, m_KSharedUserDataSize,
@@ -663,29 +700,9 @@ auto Sandbox::Run() -> void {
InitApiHooks();
std::cout << "Starting execution at " << std::hex << entryPoint
<< std::endl;
err = uc_emu_start(m_ucEngine, entryPoint, m_peInfo->imageEnd, 0, 0);
if (err != UC_ERR_OK) {
std::cerr << "Emulation error: " << uc_strerror(err) << std::endl;
// 32位环境下的错误处理
if (!m_peInfo->isX64) {
uint32_t eip;
uc_reg_read(m_ucEngine, UC_X86_REG_EIP, &eip);
std::cerr << "Error occurred at EIP: 0x" << std::hex << eip
<< std::endl;
// 尝试读取当前指令
uint8_t instruction[16];
if (uc_mem_read(m_ucEngine, eip, instruction,
sizeof(instruction)) == UC_ERR_OK) {
std::cerr << "Instruction bytes: ";
for (int i = 0; i < 16; i++) {
printf("%02X ", instruction[i]);
}
std::cerr << std::endl;
}
}
}
uint64_t timeout = 60 * 1000;
err = uc_emu_start(m_ucEngine, entryPoint, m_peInfo->imageEnd, timeout, 0);
std::cerr << "Emulation error: " << uc_strerror(err) << std::endl;
}
auto Sandbox::GetEnvString() -> std::vector<wchar_t> {
@@ -909,11 +926,11 @@ auto Sandbox::DumpPE() -> std::pair<std::unique_ptr<BYTE[]>, size_t> {
reinterpret_cast<HMODULE>(moduleBuffer.get()),
module->base);
}
//这里有一个严重的问题,就懒得处理了:
//壳里面吐出来的代码的导入表和壳的导入表不是同样一个.
//这个修的是壳的 导入表,所以导入表 修 不 全
//有个很简单的办法,需要搜索IAT结构,然后修改脱壳后的IAT的字段到壳的字段里面,然后再执行一次fix_imports
//懒得写了,家庭作业.自己完成
// 这里有一个严重的问题,就懒得处理了:
// 壳里面吐出来的代码的导入表和壳的导入表不是同样一个.
// 这个修的是壳的 导入表,所以导入表 修 不 全
// 有个很简单的办法,需要搜索IAT结构,然后修改脱壳后的IAT的字段到壳的字段里面,然后再执行一次fix_imports
// 懒得写了,家庭作业.自己完成
bool importsFixed = peconv::fix_imports(
resultBuffer.get(), virtualMemorySize, exportsMap, nullptr);
if (importsFixed) {

2
ai_anti_malware/sandbox.h
View File

@@ -217,4 +217,6 @@ class Sandbox {
auto InitCommandLine(std::string commandLine) -> void;
std::vector<uint64_t> m_crossSectionExecution; // 记录跨区段执行地址
uint64_t m_lastExecuteSectionIndex = 0; // 上次执行的区段索引
uint64_t m_KSharedUserDataBase{0};
uint64_t m_KSharedUserDataSize{0};
};