Rating:

# CSAW CTF Quals 2019

## Reverse

### Beleaf - 50 pts

> tree sounds are best listened to by https://binary.ninja/demo or ghidra

[beleaf](rev/beleaf/beleaf)

Since this is a beginner challenge, I'll go fairly in-depth for this challenge.
The TLDR is that the array referenced in `main` is a list of indexes into the
array referenced in `sub_7fa`. If you concatenate all these characters together
you get the flag.

Inspecting `main()`, we see the framework for a standard reversing challenge.

![beleaf main function disassembled and lifted in Binary Ninja](images/beleaf_main.png)

Here's a breakdown of what's happening (note that these addresses correspond to
those shown in Binja's MLIL view, so they might not include all instructions
that perform the action specified in the description column; the addresses
listed are for quick reference only):

| address | description | C equivalent |
|----------------|------------------------------------------------------------------------------------------|---------------------------------------------|
| `0x8e0..0x8ef` | string (the flag) is read from stdin into `var_98` (which I'll call `flag`) | `char flag[0x88]; fgets("%s", flag);` |
| `0x8fb..0x903` | the string length of the flag is taken and stored in `var_a8` (which I'll call `len`) | `size_t len = strlen(flag);` |
| `0x912` | check if `len` is greater than 32 | `if (len > 0x20) {` |
| `0x92a` | a counter variable (I'll call it `i`) is initialized to 0 | ` int i;` |
| `0x99d..0x9ab` | a loop is set up to execute until `i == len` | ` while (i < len) {` |
| `0x93e..0x94e` | a single character (`c`) is taken from `flag` at position `i` | ` char c = flag[i];` |
| `0x950` | some checker function (`get_index`, we'll see why I name it this later) is called on `c` | ` int index = get_index(c);` |
| `0x955..0x97d` | the index from `get_index(c)` is compared against some global array | ` if (index == expected_indexes[i])` |
| `0x995` | `i` is incrememnted and the loop is continued | ` i++;` |
| `0x986..990` | we lose | ` else { puts("Incorrect!"); exit(1); }` |
| `0x9b4` | we win | ` } puts("Correct!");` |
| `0x91b` | we lose | `} puts("Incorrect!"); exit(1);` |

Let's examine `sub_7fa` (`get_index`):

![beleaf get_index function disassembled and lifted in Binary Ninja](images/beleaf_get_index.png)

I won't go into as much detail about reversing this function, but here's an
approximation in C:

```c
int32_t bst[300];

int32_t get_index(char c) {
int32_t idx = 0;
while (idx != -1) {
if (c == bst[idx])
break;

if (c < bst[idx])
idx = 2*idx + 1;
else if (c > bst[idx])
idx = 2*idx + 2;
}
return idx;
}
```

On closer examination of `bst` (address `0x201020`), it appears to be a [Binary
Search Tree](https://en.wikipedia.org/wiki/Binary_search_tree) of sorts.
However, this isn't important to solving the challenge. We can simply think of
it as a list that assignes each unique character in the flag a unique index into
the list.

![beleaf BST structure](images/beleaf_bst.png)

Let's think about what the possible return values of `get_index` can be. There
is only one place that it can return from, so that simplifies things a bit. The
return value of `get_index` will always be `idx`

Now what are the possible values of `idx` at the return statement. The return
statement directly follows the while statement, and there are only two exits
from the while statement: a) failing the condition (`idx != -1` is false), or b)
encountering the `break` statement. If we encounter (a), then `idx` must be `-1`
(`0xffffffff` since it's a `int32_t`), though this is unlikely, since `idx`
would have to grow to very large numbers before hitting `0xffffffff` and we
would likely hit a segfault before it ever gets to be that large.

The most like case is that we've hit the `break` statement, which means that `c
== bst[id]`. This is a very simple constraint. In fact, it allows us to
essentially ignore the rest of the logic of this function. Since the return
value of `get_index(c)` is `idx`, and since we have shoen that it must be true
that `c == bst[idx]`, then it must be true that `get_index(bst[idx]) == idx`.

__Essentially__ what all this means is that `get_index` simply searches for `c`
in `bst` and returns the index at which it found it. We can exploit this simple
property to determine each character of the flag.

In the array `expected_indexes` (address `0x2014e0`), we have the expected
return values of `get_index`. Each value in this array is simply an index into
`bst`, so we can reconstruct the flag quite easily.

```
1) get_index(bst[i]) == i (this is the characteristic property of get_index)
2) get_index(bst[expected_indexes[i]]) == expected_indexes[i] (by replacing i with expected_indexes[i])
3) get_index(flag[i]) == expected_indexes[i] (by analysis of main())
4) get_index(flag[i]) == get_index(bst[expected_indexes[i]]) (by 1 and 2)
5) flag[i] == bst[expected_indexes[i]] (by 4)
```

Once we've learned this, the actual solution is quite simple. See [sol.py](rev/beleaf/sol.py)

## Pwn

### GOT Milk? - 50 pts
> GlobalOffsetTable milk?
> nc pwn.chal.csaw.io 1004

This is a simple format string vuln that can be used to overwrite a GOT address.
I won't get into too much detail but the GOT entry is at `0x804a044`, and we
need to change it from `lose` to `win` (both found in `libmylib.so`). Luckily,
their addresses only differ by one byte, so we can use a format string that
performs a 1-byte overwrite:

\x10\xa0\x04\x08%1$133c%7$hhn\\n\\n

### small_boi - 100 pts
> you were a baby boi earlier, can you be a small boi now?
> nc pwn.chal.csaw.io 1002

[small_boi](pwn/small_boi/small_boi)

At only 4.8KiB and statically linked, this binary is very similar. We are only
given 4 functions:

- `sub_40017c` :: `syscall(15)`
- `sub_40018a` :: `pop rax; ret`
- `sub_40018c` :: `char buf[0x20]; read(0, buf, 0x200)`
- `_start` :: `sub_40018c(); exit(0);`

So we have an obvious buffer overflow in `sub_40018c`, but NX is on so we can't
execute shellcode, there is no libc to return into, and there only seem to be 2
or 3 gadgets in the executable. Interested what `sigreturn()` is I decided to
consult its man page:

> sigreturn, rt_sigreturn - return from signal handler and cleanup stack frame

Turns out, this syscall enables an entire variant of ROP:
[SROP](https://en.wikipedia.org/wiki/Sigreturn-oriented_programming). All you
need is a `pop rax; ret` gadget and a single `syscall` instruction literally
anywhere in executable memory (or even better, a `mov eax, 15; syscall` as the
author has given us). This enables us to call `sigreturn`, which lets us set any
of the general purpose registers (`r{a,b,c,d}x`, `r{d,s}i`, `r8..r15`), stack
registers (`r{s,b}p`), and most importantly `rip` (`eflags` and `cs` can also be
set using a `sigreturn`).

Using this we can construct a call to `execve("/bin/sh", NULL, NULL)`, giving us
a shell. Simple, we'll place `"/bin/sh"` on the stack in the buffer-overflowing
`read()` call, et voilĂ ! Oh wait, ASLR... we can't find a reliable stack
address. Well, turns out the challenge writer has given us exactly one string in
`.rodata`: `"/bin/sh"`!

See [sol.py](pwn/small_boi/sol.py) for the full solution.

Original writeup (https://github.com/KevOrr/ctf-writeups/tree/master/2019/csaw#beleaf---50-pts).