Flare-On 2019: bmphide
bmphide was my favorite challenge this year (that I got to). It was challenging, yet doable and interesting. I’m given a bitmap image and a Windows .NET executable. That executable is used to hide information in the low bits of the image. I’ll have to reverse the exe to understand how to extract the data. I’ll also have to work around some anti-debug.
Challenge
Tyler Dean hiked up Mt. Elbert (Colorado’s tallest mountain) at 2am to capture this picture at the perfect time. Never skip leg day. We found this picture and executable on a thumb drive he left at the trail head. Can he be trusted?
I’m given a 32-bit Windows .NET executable and a .bmp bitmap image file:
$ file bmphide.exe image.bmp
bmphide.exe: PE32 executable (console) Intel 80386 Mono/.Net assembly, for MS Windows
image.bmp: PC bitmap, Windows 3.x format, 1664 x 1248 x 24
The image is of a mountain:
Running It
Just running it throws an error about the arguments:
C:\Users\0xdf\Desktop\flare6>bmphide.exe
Unhandled Exception: System.IndexOutOfRangeException: Index was outside the bounds of the array.
at BMPHIDE.Program.Main(String[] args)
Looks like I’ll need to reverse it to figure out what it’s looking for.
Initial RE
Static
Given this is a dotnet application, I opened it in dnspy, and located the main function:
It is simple enough, taking three args, which I can guess are an image to put the message into, the data to embed, and the output filename:
private static void Main(string[] args)
{
Program.Init();
Program.yy += 18;
string filename = args[2];
string fullPath = Path.GetFullPath(args[0]);
string fullPath2 = Path.GetFullPath(args[1]);
byte[] data = File.ReadAllBytes(fullPath2);
Bitmap bitmap = new Bitmap(fullPath);
byte[] data2 = Program.h(data);
Program.i(bitmap, data2);
bitmap.Save(filename);
}
It reads the first arg file into a Bitmap, and the second arg file into data. It passes data into Program.h(), and stores the result as data2. Then bitmap and data2 are passed into Program.i(), and then bitmap is saved to the filename from the third arg.
Given the exe name, the challenge prompt, I suspect this program is designed to perform stegonagraphy, hiding some data inside an image. Based on this, I can guess that h() encodes/encrpyts the data somehow, and i() combines the data into the image. If I can understand what goes on in h and i, and reverse it, I can get the original data back out.
Running It
At this point, I wanted to try to run it again. And I got it roughly working, as the following returns without error:
C:\Users\0xdf\Desktop\flare6>bmphide.exe image.bmp bmphide.exe exe.bmp
My theory is that it just took the bmphide.exe file, and encoded it and stored it in the low bits of exe.bmp. Looking at them, exe.bmp looks the same as image.bmp. But if I look in a hex editor, I can confirm that the low bits are different:
$ xxd image.bmp | head
00000000: 424d 3610 5f00 0000 0000 3600 0000 2800 BM6._.....6...(.
00000010: 0000 8006 0000 e004 0000 0100 1800 0000 ................
00000020: 0000 0000 0000 c40e 0000 c40e 0000 0000 ................
00000030: 0000 0000 0000 3d60 7055 7a8c 2a54 6717 ......=`pUz.*Tg.
00000040: 3c4c 224e 5e18 4655 2c58 686a 9ea5 588a <L"N^.FU,Xhj..X.
00000050: 9a2c 666e 184a 5929 5c6d 2956 664c 7485 .,fn.JY)\m)VfLt.
00000060: 3352 6227 4055 5574 8242 5e6d 3950 623a 3Rb'@UUt.B^m9Pb:
00000070: 5265 314e 5a20 384e 2444 511d 384e 415e Re1NZ 8N$DQ.8NA^
00000080: 6861 8095 4e68 7a2f 4a5d 2744 572d 4e5e ha..Nhz/J]'DW-N^
00000090: 3f58 6931 586d 6794 a03c 708e 0c42 5240 ?Xi1Xmg..<p..BR@
$ xxd exe.bmp | head
00000000: 424d 3610 5f00 0000 0000 3600 0000 2800 BM6._.....6...(.
00000010: 0000 8006 0000 e004 0000 0100 1800 0000 ................
00000020: 0000 0000 0000 c40e 0000 c40e 0000 0000 ................
00000030: 0000 0000 0000 3c61 7454 7b8c 2851 6414 ......<atT{.(Qd.
00000040: 3b4c 2049 5c18 4354 2e59 6e6a 9da4 5a8d ;L I\.CT.Ynj..Z.
00000050: 9d2e 636e 194b 5c2a 596c 2855 614c 7782 ..cn.K\*Yl(UaLw.
00000060: 3357 6225 4154 5471 8440 5b6c 3851 6438 3Wb%ATTq.@[l8Qd8
00000070: 5364 3049 5c20 3b4c 2441 541c 3b4c 4059 Sd0I\ ;L$AT.;L@Y
00000080: 6c60 8394 4c69 7c2c 4b5c 2441 542c 4b5c l`..Li|,K\$AT,K\
00000090: 3c59 6c30 5b6c 6794 a03c 708e 0c42 5240 <Yl0[lg..<p..BR@
Once I get past the header, at 0x36 I start to see the low bits changing.
h()
The two functions I need to look at now are h() and i(). Since h() encodes the data, I’ll take a look at that first:
public static byte[] h(byte[] data)
{
byte[] array = new byte[data.Length];
int num = 0;
for (int i = 0; i < data.Length; i++)
{
int num2 = (int)Program.f(num++);
int num3 = (int)data[i];
num3 = (int)Program.e((byte)num3, (byte)num2);
num3 = (int)Program.a((byte)num3, 7);
int num4 = (int)Program.f(num++);
num3 = (int)Program.e((byte)num3, (byte)num4);
num3 = (int)Program.c((byte)num3, 3);
array[i] = (byte)num3;
}
return array;
}
Pretty simple. There’s a series of functions, a counter (num), and a loop over data, writing results to array.
Anti Debug
I took a quick glance into f, e, a, and c, and they look like complex mixing functions. I want to try to debug the binary so I can watch data pass through these. I’ll select “Debug -> Start Debugging…”, and enter the same command line I just successfully ran from cmd:
On clicking “OK”, I reach my break point at the entry point:
When I try to step over Program.Init(), it throws an error:
To try to get around this, I’ll take a new strategy of starting the program from the command line, and then trying to (very) quickly attach dnspy to the running process. My hypothesis here is that if I attach after Program.Init(), perhaps I’ll have bypassed the anti-debug.
It takes a couple tries, but I get it attached while the program is still in h. It’s there I notice something else weird. When I’m at the start of the loop, and I step into the call for f.(num++), I don’t go to f, but g:
In that gif, I’ll open up the disassembly, and see it’s been changed. My new theory is that Program.Init() does two things - changes the function calls, and crashes if a debugger is attached. I spent some time checking out the rest, and the only thing that seems to change is the functions called. Instead of calling f, e, a, f, e, c like shown above, it calls g, e, b, g, e, d.
Functions in h()
e()
It’s not the first function called, but e() is actually called in other functions, so I’ll break it down first.
public static byte e(byte b, byte k)
{
for (int i = 0; i < 8; i++)
{
bool flag = (b >> i & 1) == (k >> i & 1);
if (flag)
{
b = (byte)((int)b & ~(1 << i) & 255);
}
else
{
b = (byte)((int)b | (1 << i & 255));
}
}
return b;
}
The function takes two bytes. It then walks each bit. flag is true if the ith bit for b and k is the same. If flag, then that bit is turned to a 0 in b, else that bit is turned to a 1 in b. This is just xor.
g()
g() is just some math:
public static byte g(int idx)
{
byte b = (byte)((long)(idx + 1) * (long)((ulong)-306674912));
byte k = (byte)((idx + 2) * 1669101435);
return Program.e(b, k);
}
I actually think the disassembly for setting b (and perhaps k) might be clearer:
; byte b = (byte)((long)(idx + 1) * (long)((ulong)-306674912));
01063399 90 nop
0106339A 8B45FC mov eax,[ebp-4]
0106339D 40 inc eax
0106339E BAC56F6B12 mov edx,126B6FC5h
010633A3 33C9 xor ecx,ecx
010633A5 0FAFC2 imul eax,edx
010633A8 25FF000000 and eax,0FFh
010633AD 8945F8 mov [ebp-8],eax
; byte k = (byte)((idx + 2) * 1669101435);
010633B0 8B45FC mov eax,[ebp-4]
010633B3 40 inc eax
010633B4 40 inc eax
010633B5 69C07DC9820C imul eax,0C82C97Dh
010633BB 25FF000000 and eax,0FFh
010633C0 8945F4 mov [ebp-0Ch],eax
So, b = ((idx + 1) * 0x126B6FC5) & 0xFF, and k = ((idx+2) * 0xC82C97D) 0xFF, and then those two bytes are xored and the result is returned.
I started a file called funcs.py, and created g():
def g(idx):
b = ((idx + 1) * 0x126B6FC5) & 255
k = ((idx + 2) * 0xC82C97D) & 255
return b ^ k
b()
I’ll do the same thing with b():
public static byte b(byte b, int r)
{
for (int i = 0; i < r; i++)
{
byte b2 = (b & 128) / 128;
b = (b * 2 & byte.MaxValue) + b2;
}
return b;
}
This is definitely some scrambling. Rather then break down the algorithm, I created my python function:
def b(b_, r):
for i in range(r):
b2 = (b_ & 128) // 128
b_ = ((b_ * 2) & 255) + b2
return b_
I also noticed that this is only called with a static r value of 7. I’ll take a look at the values that come out:
df@df:~/flareon/6 - bmphide$ python3
Python 3.6.8 (default, Aug 20 2019, 17:12:48)
[GCC 8.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from funcs import *
>>> for i in range(10):
... b(i, 7)
...
0
128
1
129
2
130
3
131
4
132
It seems b(i, 7) == i/2 + (127.5 * (i mod 2)). For even numbers, that’s i/2. For odd, that’s i/2 + 127.5. I checked 0 through 255, and it holds up.
d()
d() is similar to be, in that it takes two args, but is only ever called with the second arg of 3:
public static byte d(byte b, int r)
{
for (int i = 0; i < r; i++)
{
byte b2 = (b & 1) * 128;
b = (b / 2 & byte.MaxValue) + b2;
}
return b;
}
I’ll do the same thing and create a python version:
def d(b, r):
for i in range(r):
b2 = (b & 1) * 128
b = ((b // 2) & 255) + b2
return b
And again, running it shows a pattern:
>>> for i in range(10):
... d(i, 3)
...
0
32
64
96
128
160
192
224
1
33
This one works out to d(i, 3) == 32 * (i mod 8) + i//8 (where // represents integer division).
h() in python
Given my understanding of these functions, I wrote h() in python, but rather than looping over all data, it takes an i and a byte and outputs what array[i] will be:
def h_one(byte, i):
num2 = g(2*i)
num3 = byte
num3 = num3 ^ num2
num3 = b(num3, 7)
num4 = g(2*i + 1)
num3 = num3 ^ num4
num3 = d(num3, 3)
return num3
I can test this by breaking at the start of the loop in my debugger, observing the i and data[i]. I can run my code to see what the output will be, and then check that the output matches what is eventually stored in array.
Invert h()
Now I want to be able to find the inverse of h(). This would start with i and array[i], and output data[i]. Given the code above, I can work backwards and invert some of the functions. Since i is the same working backwards and forwards, so are functions that only rely on i, like g(). On the other hand, I’ll need to invert e(), b(), and d(), though the inverse of e() is just e() (since e() is xor).
Since b() and d() are only called with specific second arguments, I’m only going to worry about inverting them for those arguments.
I end up with:
def h_inv_one(byte, i):
num3 = byte
num3 = d_inv_3(num3)
num4 = g(2*i + 1)
num3 = num3 ^ num4
num3 = b_inv_7(num3)
num2 = g(2*i)
num3 = num3 ^ num2
return num3
def b_inv_7(b_):
if b_ > 127:
return ((2*b_) - 255)
else:
return 2*b_
def d_inv_3(b):
return (b//32) + 8*(b%32)
I can test this by showing that for a random byte (0-255) and a random i (0-10000, and while i could go higher, no reason to think it’d work differently), that h_inv(h(byte, i), i) == byte. For example:
>>> h_inv_one(h_one(223, 1034), 1034) == 223
True
Or if I wanted to test with 300,000 randomly generated byte and i:
>>> random_byte_and_i = [(random.randint(0,255), random.randint(0,10000)) for x in range(300000)]
>>> all([(h_inv_one(h_one(byte, i), i) == byte) for byte,i in random_byte_and_i])
True
I feel good now that I have python that can reverse the encoding.
i()
Now I need to look at i() to see how the encoded data is written into the image:
public static void i(Bitmap bm, byte[] data)
{
int num = Program.j(103);
for (int i = Program.j(103); i < bm.Width; i++)
{
for (int j = Program.j(103); j < bm.Height; j++)
{
bool flag = num > data.Length - Program.j(231);
if (flag)
{
break;
}
Color pixel = bm.GetPixel(i, j);
int red = ((int)pixel.R & Program.j(27)) | ((int)data[num] & Program.j(228));
int green = ((int)pixel.G & Program.j(27)) | (data[num] >> Program.j(230) & Program.j(228));
int blue = ((int)pixel.B & Program.j(25)) | (data[num] >> Program.j(100) & Program.j(230));
Color color = Color.FromArgb(Program.j(103), red, green, blue);
bm.SetPixel(i, j, color);
num += Program.j(231);
}
}
}
At first, this looks a bit daunting, especially when I look at the mess that is j(). However, I’ll notice that j() is being called on static ints only. There’s no reason I can’t replace calls to j() with the values they return. I’ll walk through with the debugger and record the output of j() for each input it receives:
| x | j(x) |
|---|---|
| 103 | 0 |
| 231 | 1 |
| 27 | 0xf8 (1111 1000) |
| 228 | 7 |
| 230 | 3 |
| 25 | 0xfc (1111 1100) |
| 100 | 6 |
Knowing that, I can re-write i:
public static void i(Bitmap bm, byte[] data)
{
int num = 0;
for (int i = 0; i < bm.Width; i++)
{
for (int j = 0; j < bm.Height; j++)
{
bool flag = num > data.Length - 1;
if (flag)
{
break;
}
Color pixel = bm.GetPixel(i, j);
int red = ((int)pixel.R & 0xf8) | ((int)data[num] & 7);
int green = ((int)pixel.G & 0xf8) | (data[num] >> 3 & 7);
int blue = ((int)pixel.B & 0xfc) | (data[num] >> 6 & 3);
Color color = Color.FromArgb(0, red, green, blue);
bm.SetPixel(i, j, color);
num += 1;
}
}
}
Now it’s much easier to read. It reads a pixel. Then it overwrites the low three bits of red with the low three bits of the byte, the low three bits of green with the next three bits, and the low two bits of blue with the high two bits of the byte. Then it writes the pixel.
Inverting i()
I want to do the opposite of this. So I’ll read a pixel, get the RGB values, and get the low three (for red and green) or two (for blue) bits, and use them to form the byte. It will look like:
for x in range(width):
for y in range(height):
rp, gp, bp = im.getpixel((x,y))
value = (rp & 7) | ((gp & 7) << 3) | ((bp & 3) << 6)
i = y + (x*height)
cipher.append(value)
plaintext = h_inv_one(value, i)
data.append(plaintext)
Extracting a Message
All of that comes together into a python script that will take an image and output the hidden data:
#!/usr/bin/env python3
import sys
from PIL import Image
def g(idx):
b = ((idx + 1) * 0x126B6FC5) & 255
k = ((idx + 2) * 0xC82C97D) & 255
return b ^ k
def d_inv_3(b):
return (b//32) + 8*(b%32)
def b_inv_7(b_):
if b_ > 127:
return ((2*b_) - 255)
else:
return 2*b_
def h_inv_one(byte, i):
num3 = byte
num3 = d_inv_3(num3)
num4 = g(2*i + 1)
num3 = num3 ^ num4
num3 = b_inv_7(num3)
num2 = g(2*i)
num3 = num3 ^ num2
return num3
if __name__ == "__main__":
data = bytearray()
cipher = []
im = Image.open(sys.argv[1])
width, height = im.size
for x in range(width):
for y in range(height):
rp, gp, bp = im.getpixel((x,y))
value = (rp & 7) | ((gp & 7) << 3) | ((bp & 3) << 6)
i = y + (x*height)
cipher.append(value)
plaintext = h_inv_one(value, i)
data.append(plaintext)
with open(sys.argv[2], 'wb') as f:
f.write(data)
I can run this on the given image and get output:
$ ./extract.py image.bmp extracted
$ file extracted
extracted: PC bitmap, Windows 3.x format, 832 x 624 x 24
Surprising that the output is another bmp. It is a valid image:
I’ll extract on it again:
$ ./extract.py extracted extracted2
And get another .bmp:
$ file extracted2
extracted2: PC bitmap, Windows 3.x format, 234 x 312 x 24
This one has a flag in it:
Flag: d0nT_tRu$t_vEr1fy@flare-on.com