前面讲解了修改单架构二进制文件实现动态库注入,本篇继续讲解修改胖二进制文件实现动态库注入。注入方式并无不同,不过需要注意修改文件时的地址偏移,否则修改错误会导致文件无法解析。本文所展示的代码片段来自开源项目FishHook,更多细节可参考该项目。
如下图,胖二进制包含多种CPU架构,一般是X86_64和ARM64,胖二进制相当于多个单二进制的组合体。使用系统自带的工具lipo可以添加、提取、删除以及替换胖二进制中的某个架构的二进制,这里不多做介绍。
修改胖二进制首先需要解析Fat Header,通过文件头获取所支持的架构数量及各架构描述内容的偏移地址。注意,获取数据时根据机器的大端/小端模式不同,需要对数据进行处理。一般PC机都是小端模式,因此需要将高低位地址上的数据进行交换。多架构二进制的修改直接复用了单架构的代码,processThinMachO相关代码请参考上文或FishHook项目。
func repackBinary() -> Bool {
if machOData.isEmpty {
return false
}
return machOData.withUnsafeBytes { pointer in
guard let header = pointer.bindMemory(to: fat_header.self).baseAddress else {
print("[ERROR] Failed to get fat header pointer.")
return false
}
var result = false
var offset = MemoryLayout<fat_header>.size
let archNum = _OSSwapInt32(header.pointee.nfat_arch)
switch header.pointee.magic {
case FAT_MAGIC, FAT_CIGAM:
if archNum == 0 {
print("[ERROR] Format of Fat-MachO is invalid.")
return false
}
for i in 0 ..< archNum {
if i > 0 {
offset = offset + MemoryLayout<fat_arch>.size
}
result = processFatMachO(offset: offset)
if !result {
return false
}
}
case MH_MAGIC_64, MH_CIGAM_64, MH_MAGIC, MH_CIGAM:
result = processThinMachO(offset: 0)
default:
print("[ERROR] Unknown MachO format.")
return false
}
signAdhoc()
return result
}
}
private func processFatMachO(offset: Int) -> Bool {
let fatData = machOData.advanced(by: offset)
return fatData.withUnsafeBytes { pointer in
guard let arch = pointer.bindMemory(to: fat_arch.self).baseAddress else {
print("[ERROR] Failed to get fat arch pointer.")
return false
}
let offset = _OSSwapInt32(arch.pointee.offset)
return processThinMachO(offset: Int(offset))
}
}
上文提到,在插入动态库注入指令时并没有判断是否有足够的空白区域容纳,如果空间不足,插入会覆盖有效数据,进而导致二进制格式错误。插入空间判断代码如下:
private func isSpaceEnough(header: mach_header, offset: Int, is64bit: Bool) ->Bool {
let pathSize = (dylibPath.count & ~(pathPadding - 1)) + pathPadding
let injectSpace = MemoryLayout<dylib_command>.size + pathSize
let headerSize = is64bit ? MemoryLayout<mach_header_64>.size : MemoryLayout<mach_header>.size
var segOffset = offset
for _ in 0 ..< header.ncmds {
let segData = machOData.subdata(in: segOffset..<segOffset+MemoryLayout<segment_command_64>.size)
guard let segCmd = getSegmentCommand(data: segData) else {
print("[ERROR] Failed to get segment command pointer.")
return false
}
var segName = segCmd.segname
if (strncmp(&segName.0, "__TEXT", 15) == 0) {
for i in 0 ..< segCmd.nsects {
let sectOffset = segOffset + MemoryLayout<segment_command_64>.size + MemoryLayout<section_64>.size * Int(i)
let sectData = machOData.subdata(in: sectOffset..<sectOffset+MemoryLayout<section_64>.size)
guard let sectCmd = getSectionCommand(data: sectData) else {
print("[ERROR] Failed to get section pointer.")
return false
}
var sectName = sectCmd.sectname
if (strncmp(§Name.0, "__text", 15) == 0) {
let space = sectCmd.offset - header.sizeofcmds - UInt32(headerSize)
print("[INFO] Available space is \(space/8) bytes.")
return space > injectSpace
}
}
}
else {
segOffset = segOffset + Int(segCmd.cmdsize)
}
}
return false
}
基本原理是查找到代码段(__text)偏移地址,然后分析该地址前是否有足够空间,读者可结合二进制格式进行代码阅读,这里不再注解。