介绍一种ELF文件函数粒度的加固方法，可以有效防止对程序的静态分析。这是一种有源码加固方式，需要被加固程序中代码配合。加固流程如下：
1）读取ELF文件头，获取e_phoff和e_phnum
2）通过Elf64_Phdr中的p_type字段，找到DYNAMIC
3）遍历.dynamic，找到.dynsym、.dynstr 节区偏移，和.dynstr节区的大小
4）遍历.dynsym，找到函数对应的Elf64_Sym符号后，根据st_value和st_size字段找到函数在ELF的偏移和函数大小
5）根据函数偏移和大小，加密之
加固程序代码如下，在x86_64平台测试通过：
#include <stdio.h>
#include <fcntl.h>
#include <elf.h>
#include <stdlib.h>
#include <string.h>
typedef struct {
Elf64_Addr st_value;
Elf64_Word st_size;
}func_info;
Elf64_Ehdr ehdr;
int
find_target_section_addr(const int fd, const char *sec_name){
lseek(fd, 0, SEEK_SET);
if(read(fd, &ehdr, sizeof(Elf64_Ehdr)) != sizeof(Elf64_Ehdr)){
puts("Read ELF header error");
return (-1);
}
return (0);
}
static char get_target_func_info(int fd, const char *func_name, func_info *info){
char flag = -1, *dynstr;
int i;
Elf64_Sym func_sym;
Elf64_Phdr phdr;
Elf64_Off dyn_off;
Elf64_Word dyn_size, dyn_strsz;
Elf64_Dyn dyn;
Elf64_Addr dyn_symtab, dyn_strtab;
lseek(fd, ehdr.e_phoff, SEEK_SET);
for(i = 0; i < ehdr.e_phnum; i++){
if(read(fd, &phdr, sizeof(Elf64_Phdr)) != sizeof(Elf64_Phdr)){
puts("Read segment failed");
return (-1);
}
if(phdr.p_type == PT_DYNAMIC){
dyn_size = phdr.p_filesz;
dyn_off = phdr.p_offset;
flag = 0;
printf("Find section %s, size = 0x%x, addr = 0x%lx\n", ".dynamic", dyn_size, dyn_off);
break;
}
}
if(flag) {
puts("Find .dynamic failed");
return (-1);
}
flag = 0;
lseek(fd, dyn_off, SEEK_SET);
for(i=0;i < dyn_size / sizeof(Elf64_Dyn); i++){
if(read(fd, &dyn, sizeof(Elf64_Dyn)) != sizeof(Elf64_Dyn)){
puts("Read .dynamic information failed");
return (-1);
}
if(dyn.d_tag == DT_SYMTAB){
dyn_symtab = dyn.d_un.d_ptr;
flag++;
printf("Find .dynsym, addr = 0x%lx\n", dyn_symtab);
}
if(dyn.d_tag == DT_STRTAB){
dyn_strtab = dyn.d_un.d_ptr;
flag++;
printf("Find .dynstr, addr = 0x%lx\n", dyn_strtab);
}
if(dyn.d_tag == DT_STRSZ){
dyn_strsz = dyn.d_un.d_val;
flag++;
printf("Find .dynstr size, size = 0x%x\n", dyn_strsz);
}
}
if(flag != 3){
puts("Find needed .section failed\n");
return (-1);
}
dynstr = (char*) malloc(dyn_strsz);
if(dynstr == NULL){
puts("Malloc .dynstr space failed");
return (-1);
}
lseek(fd, dyn_strtab, SEEK_SET);
if(read(fd, dynstr, dyn_strsz) != dyn_strsz){
puts("Read .dynstr failed");
return (-1);
}
lseek(fd, dyn_symtab, SEEK_SET);
while (1) {
if(read(fd, &func_sym, sizeof(Elf64_Sym)) != sizeof(Elf64_Sym)){
puts("Read func_sym failed");
return (-1);
}
if(strcmp(dynstr + func_sym.st_name, func_name) == 0){
break;
}
}
printf("Find: %s, offset = 0x%lx, size = 0x%lx\n", func_name, func_sym.st_value, func_sym.st_size);
info->st_value = func_sym.st_value;
info->st_size = func_sym.st_size;
ehdr.e_shoff = info->st_value;
ehdr.e_shnum = info->st_size;
lseek(fd, 0, SEEK_SET);
if(write(fd, &ehdr, sizeof(Elf64_Ehdr)) != sizeof(Elf64_Ehdr)){
puts("Write elf header failed");
return (-1);
}
free(dynstr);
return 0;
}
int main(int argc, char **argv){
char sec_name[] = ".text";
char func_name[] = "say_hello"; /* 被加密函数名 */
char *content = NULL;
int fd, i;
Elf64_Off secOff;
func_info info;
if(argc < 2){
puts("Usage: shell libxxx.so .(section) function");
return -1;
}
fd = open(argv[1], O_RDWR);
if(fd < 0){
printf("open %s failed\n", argv[1]);
return (-1);
}
if (find_target_section_addr(fd, sec_name) == -1) {
printf("Find section %s failed\n", sec_name);
return (-1);
}
if (get_target_func_info(fd, func_name, &info) == -1) {
printf("Find function %s failed\n", func_name);
return (-1);
}
content = (char*) malloc(info.st_size);
if(content == NULL){
puts("Malloc space failed");
return (-1);
}
lseek(fd, info.st_value, SEEK_SET);
if(read(fd, content, info.st_size) != info.st_size){
puts("Malloc space failed");
return (-1);
}
for(i = 0; i < info.st_size; i++){
content[i] = ~content[i];
}
lseek(fd, info.st_value, SEEK_SET);
if(write(fd, content, info.st_size) != info.st_size){
puts("Write modified content to .so failed");
return (-1);
}
puts("Complete!");
free(content);
close(fd);
return 0;
}
解密代码放在.init_array节区，使ELF加载时运行解密：
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/types.h>
#include <elf.h>
#include <sys/mman.h>
#define PAGE_SHIFT 12
#define PAGE_SIZE (1UL << PAGE_SHIFT)
typedef struct {
Elf64_Addr st_value;
Elf64_Word st_size;
}func_info;
void say_hello() { /* 被加密函数 */
puts("hello elf.");
}
void __init() __attribute__((constructor));
static unsigned long get_lib_addr(){
unsigned long ret = 0;
char buf[4096], *temp;
int pid;
FILE *fp;
pid = getpid();
sprintf(buf, "/proc/%d/maps", pid);
fp = fopen(buf, "r");
if(fp == NULL) {
puts("open failed");
goto _error;
}
while(fgets(buf, sizeof(buf), fp)){
if(strstr(buf, "libdemo.so")){
temp = strtok(buf, "-");
ret = strtoul(temp, NULL, 16);
break;
}
}
_error:
fclose(fp);
return ret;
}
void __init(){ /* 解密函数 */
const char target_fun[] = "say_hello";
func_info info;
int i;
unsigned long npage, base = get_lib_addr();
Elf64_Ehdr *ehdr = (Elf64_Ehdr *)base;
info.st_value = ehdr->e_shoff;
info.st_size = ehdr->e_shnum;
npage = info.st_size / PAGE_SIZE + ((info.st_size % PAGE_SIZE == 0) ? 0 : 1);
if(mprotect((void *) ((base + info.st_value) / PAGE_SIZE * PAGE_SIZE), npage, PROT_READ | PROT_EXEC | PROT_WRITE) != 0){
puts("mem privilege change failed");
}
for(i = 0; i < info.st_size; i++){
char *addr = (char*)(base + info.st_value + i);
*addr = ~(*addr);
}
if(mprotect((void *) ((base + info.st_value) / PAGE_SIZE * PAGE_SIZE), npage, PROT_READ | PROT_EXEC) != 0){
puts("mem privilege change failed");
}
}
写了一段测试代码，如下：
#include <stdio.h>
#include <dlfcn.h>
int main(int argc, char **argv)
{
void (*say_hello)();
void *h;
char *error;
h= dlopen("./libdemo.so", RTLD_NOW);
if (h == NULL) {
error = dlerror();
printf("%s\n", error);
return (-1);
}
say_hello = dlsym(h, "say_hello");
say_hello();
dlclose(h);
return (0);
}
以上参考了ThomasKing在看雪的贴子，但查找符号位置使用了一种新方法。
原代码使用的DT_HASH，老版本GCC和现在的安卓都在使用这个结构，它比较简单。在Ubuntu 14.04上测试时发现新版GCC并没有用DT_HASH，而是使用的DT_GUN_HASH，它使用BloomFilter算法针对符号不存在的情况做了效率优化。
这个结构比较复杂，如果再按照ELF加载器的流程来做就比较麻烦，所以选择了遍历的方法。但也有个缺点，当查找的符号不存在时程序会崩溃。
运行结果如下：
kiiim@ubuntu :~/_elf/m2$ gcc shell.c
kiiim@ubuntu :~/_elf/m2$ gcc loader.c -o loader -ldl
kiiim@ubuntu :~/_elf/m2$ gcc demo.c -fPIC -shared -o libdemo.so
kiiim@ubuntu :~/_elf/m2$ ./a.out libdemo.so
Find section .dynamic, size = 0x1c0, addr = 0xe18
Find .dynstr, addr = 0x488
Find .dynsym, addr = 0x230
Find .dynstr size, size = 0x10f
Find: say_hello, offset = 0x9f5, size = 0x12
Complete!
kiiim@ubuntu :~/_elf/m2$ ./loader
hello elf.
[email protected]:~/_elf/m2$
原贴中还有另一种加固方法，将要加固函数写入新的节区，如.mytext，然后针对节区整体加密。这种方法实现同样比较简单。但评论里有个问题值得讨论下。
有回复说实现了.text整体加密方案，但我分析了下，觉得不可行。
观察.init_array节，发现在解密函数__init()执行前，还要执行一个frame_dummy()的系统函数：
.init_array:0000000000200DF8 _init_array segment para public ‘DATA‘ use64
.init_array:0000000000200DF8 assume cs:_init_array
.init_array:0000000000200DF8 ;org 200DF8h
.init_array:0000000000200DF8 __frame_dummy_init_array_entry dq offset frame_dummy
.init_array:0000000000200E00 dq offset __init ;解密函数.init_array:0000000000200E00 _init_array ends
而这个函数是在.text中实现的：
.text:00000000000009C0 frame_dummy proc near
.text:00000000000009C0 cmp cs:__JCR_LIST__, 0
.text:00000000000009C8 jz short loc_9F0
.text:00000000000009CA mov rax, cs:_Jv_RegisterClasses_ptr
.text:00000000000009D1 test rax, rax
.text:00000000000009D4 jz short loc_9F0
.text:00000000000009D6 push rbp
.text:00000000000009D7 lea rdi, __JCR_LIST__
.text:00000000000009DE mov rbp, rsp
.text:00000000000009E1 call rax ; _Jv_RegisterClasses
.text:00000000000009E3 pop rbp
.text:00000000000009E4 jmp register_tm_clones
.text:00000000000009E4 ; ---------------------------------------------------------------------------
.text:00000000000009E9 align 10h
.text:00000000000009F0
.text:00000000000009F0 loc_9F0: ; CODE XREF: frame_dummy+8 j
.text:00000000000009F0 ; frame_dummy+14 j
.text:00000000000009F0 jmp register_tm_clones.text:00000000000009F0 frame_dummy endp
也就是说，在解密函数__init()执行之前，frame_dummy()运行就会失败。也就不能对.text整体加密。