VX Heaven

Library Collection Sources Engines Constructors Simulators Utilities Links Forum

Meta-Level Languages in Viruses

Valhalla #3
December 2012

[Back to index] [Comments]

1) Introduction

In valhalla#2, March 2012, herm1t wrote a wonderful article about how to write a good metamorphic virus - a gamechanger [1]. He argued that the virus should be written in a high-level metalanguage such as C, and should be fully self-compiling (more about it in chapter 3)

I got interested, and tried to understand how ancient metamorphic viruses handled their reconstruction in terms of meta-languages. Here you see a small non-complete collection of different ways to create a new obfuscated representation of the code. Then I show a small thought on high-level compilers - and what would be a wired case scenario.

2) Meta-Language Examples

2.1) Direct transformations in assembler

There are several metamorphic viruses that dont use an individualized language, but use directly the assembler language. Three examples are Win32.Evol [2] (by an unknown author in 2000), CODEGEN [3] (by Z0MBiE in 2000) and W32.Lexotan32 [4] (by Vecna in 2002).

Evol and Lexotan32 can change opcodes from one instruction to another (set of) instruction, which is defined in equivalence lists.

A part of Lexotan32's equivalence list is here:

	Original		Transformed
	ADD	reg, imm	SUB	reg, -(imm)

	MOV	reg, reg/imm	PUSH	reg/imm
				POP	reg

	SUB	reg, reg	XOR	reg, reg

	TEST	reg, reg	OR	reg, reg

	LODSx			MOV	ACUM, [esi]
				ADD	esi, SIZE

This is the lowest possible level where you can perform mutations: on the direct opcode level. The big disadvantage: The structure of the code is very similar and its very simple to reverse. The meta-language is direct assembler - no information about the code's behaviour can be introduced in such transformation lists.

Z0MBiE's CODEGEN engine generates equivalent logical code with different instructions as well, but tries to abstract to a certain extent the behaviour from the opcode. The engine can create code with the following structure (thus the "meta-language" is based on this structure):

	(cmd = mov/cmd/add/sub/xor)

	cmd	v, c
	cmd	v1, [v2]
	cmd	[v1], v2
	cmd	v1, v2

The macro-instruction will be derived using rules as the following:

	cmd	v, c
		mov	r, c
		cmd	v, r

	cmd	v1, [v2]
		mov	r2, v2     ...
		cmd	r1, [r2]
		mov	v1, r1

Even this engine already abstracts the behaviour from the code to some extent, it is still based on micro-mutations of single assembler instructions.

2.2) Pseudo-Assembler Meta-Language

The classic metamorphic virus MetaPHOR by The Mental Driller [5] used a real intermediate language to reconstruct the new generation code. The language is based strongly on assembler, but contains additional information that allows complex mutations in an easier way.

All meta-level instructions are 16 bytes, and have the following structure:

	00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
	OP *----- instruction data ----* LM *-pointer-*

OP and instruction data is used to identify the instruction, LM indicates whether this instruction is a label (and thus must not merge with the instruction before), and pointer are used for some easier handling of jump-tables.

The big advantage is that this meta-language carries more information as a simple assembler opcode (LM + pointer), and therefore allows potentially more mutations. However, the language is still based on the opcodes - and is on the lowest level of mutations. No macro-mutations are possible with such a concept.

2.3) Build-in HLL-Source

A totally different concept has been done by LordAsd in his Apparation virus for Win16 and Win32. [6,7]

The version for Win16 was written in Pascal. In addition to the binary file, the Pascal source of the virus and a Pascal compiler was included in the virus. The idea was to drop the source, change it, and compile it again.

The Win32 version was written in C++, it carried its own source, and relied on installed C++ compiler on the infected system. When it was found, the source was dropped to the harddisk, modified randomly and the virus was recompiled.

The advantage of such an approach would be the creation of very simple macro mutations in the code. The downside of course is that the virus has to carry the source and the compiler (or rely on systems with compilers).

The meta-language in this case is the abstracted high-level language.

2.4) Pseudo-Script Meta-Language

Recently I've written a metamorphic JavaScript virus [8], which contained its own source in a pseudo-script language, and used a compiler written in JavaScript to compile new generations. The pseudo-script language contained additional information about the instructions, such as permutation information or information about deriveable objects and local variables.

The general structure:


where "Identifyer" and "Restrictions" are information for the Permutator, and "instruction" are the actual meta-language commands such as


The meta-language is general, and even allows some sort of macro-mutations (variable constructions in arrays, ...). Still it is based very much on the JavaScript langauge itself, and is not a behaviour-descriptor only - which would be the optimum.

3) Selfcompiling compilers and the Highest Level Language

herm1t's idea of a next-generation mutation engine [1] is a code written in a high-level language such as C and can compile itself. In contrast to the assembler-based meta-languages mentioned in 2.1 and 2.2, this concept abstracts the "put value A to register B" (opcode basis) to a much more behaviour-description basis.

In contrast to LordAsd's Apparition (chapter 2.3), his engine would have/be a self-compiling compiler. The virus would not mutate the HLL source, but the compiler would create a different binary out of the same source every time. Due to the high-levelness of C, the mutations would be intrinsic macro-mutations.

This is of course an brilliant concept - but I wonder if its the final one? Obviously, C language is itself not fully general, its based on individual instructions which descript just minor behaviour. What about mutations in higher-level behaviour?

What if we had an even higher level language, where the compiler creates a new generation of the virus? The next step is the algorithmic level (such as writing to files in a general way, ...), then the macro-behaviours (such as file-infection, code-mutation, ...).

In the end, we could think about the highest level language that can represent our virus:

A language with just one single instruction: CREATE_NEW_VIRUS. The whole source of the virus would be one instruction, that means no need to disassemble itself or carry a lengthy source. Of course, the compiler of such a "language" will be very big and carry itself all the information.

Such an engine could be seen as a cascading of code derivations on one level lower as before:

	nth level:	CREATE_NEW_VIRUS
	(n-2)th level:	Some high-level descriptions of the INFECTION_METHODE_x + 
			Some high-level descriptions of the CODE_MUTATION_ENGINE_y
	3rd level:	C description of the code
	2nd level:	assembler description of the code
	1st level:	compiled opcode description of the code

I wonder how such a constructor could look like :-o

4) Conclusion

Here I tried to collect the ideas of meta-languages used in viruses so far, analyse herm1t's metamorphism concept and give a small thought on it and on a potential extention.

Hope you enjoyed this small collection. At least I did and got a few new ideas while writing. And now I exactly know what I want to do next ;-)

Second Part To Hell
December 2012
  1. herm1t, "Recompiling the metamorphism", valhalla #2, March 2012.
  2. Orr, "The Viral Darwinism of W32.Evol",, February 2007.
  3. Z0MBiE, "METAMORPHISM", MATRiX #2, 2000.
  4. Orr, "The Molecular Virology of Lexotan32: Metamorphism Illustrated",, August 2007.
  5. The Mental Driller, "Metamorphism in practice", 29a #6, February 2002.
  6. SecureList, "Virus.Win16.Apparition.a",, 1997.
  7. LordAsd, "THE APPARITION for Win32", 29a #3, 1998.
  8. SPTH, "Metamorphism and Self-Compilation in JavaScript", valhalla #3, December 2012.
[Back to index] [Comments]
By accessing, viewing, downloading or otherwise using this content you agree to be bound by the Terms of Use! aka