Introduction to the Witchcraft Compiler Collection

Jonathan Brossard @endrazine

13 of February 2017

BSides San Francisco 2017

LEGAL DISCLAIMER

My employers are not associated with this talk in any way.

This is my personal research.

Legal help

•This talk received help from the EFF.

•Warmest thank you to Nate Cardozo, Andrew Crocker and Mitch Stoltz

Free legal advising to security researchers : https://www.eff.org/ https://www.eff.org/issues/coders/reverse- engineering-faq

TL ; DR

The Witchcraft Compiler Collection is free software (MIT/BSD License).

https://github.com/endrazine/wcc

You can write in Lua, Punk-C or C. No assembly skills required.

WhoAmamI ? I ?

https://www.defcon.org/images/defcon-20/dc-20-presentations/Brossard/ DEFCON-20-Brossard-Hardware-Backdooring-is-Practical.pdf

Agenda

•WCC components

•“Libifying” a binary

•Unlinking binaries

•Crossing a Fish and a Rabbit

•Introduction to Witchcraft

•Binary “reflection” without a VM

•Automated function annotation

•Exploit writing

•Future work

WCC : components

WCC Components

Binaries (C):

wld : witchcraft linker

wcc : witchcraft core compiler

wsh : witchcraft shell : dynamic interpreter + scripting engine

Scripts (lua, …):

wcch : witchcraft header generator

wldd : witchcraft compiler flags generator

...

Host machine : GNU/Linux x86_64 (mostly portable to POSIX systems).

Wld : “Libification”

Transforming an ELF executable binary into an ELF shared library.

DEMOS

Libification of proftpd

Libification of proftpd

We really patched 1 byte only

libification

typedef struct

{

unsigned char e_ident[EI_NIDENT]; /* Magic number and other info */

Elf64_Half e_type;

/* Object file type */

Elf64_Half e_machine;

/* Architecture */

Elf64_Word e_version;

/* Object file version */

Elf64_Addr e_entry;

/* Entry point virtual address */

Elf64_Off e_phoff;

/* Program header table file offset */

Elf64_Off e_shoff;

/* Section header table file offset */

Elf64_Word e_flags;

/* Processor-specific flags */

Elf64_Half e_ehsize;

/* ELF header size in bytes */

Elf64_Half e_phentsize;

/* Program header table entry size */

Elf64_Half e_phnum;

/* Program header table entry count */

Elf64_Half e_shentsize;

/* Section header table entry size */

Elf64_Half e_shnum;

/* Section header table entry count */

Elf64_Half e_shstrndx;

/* Section header string table index */

} Elf64_Ehdr;

 

Using our new shared

library

How comes this works ?

We’re really creating a “non relocatable” shared library.

ET_DYN and ET_EXEC ELF files are both executable (ASLR support in the kernel)

This is equivalent to creating a shared library with a non NULL base address (equivalent to prelinking)

Note: Amazingly, this shared library is still a valid executable too.

DEMOS

Linking against apache2

Apache2 as a shared library

Apache2 as a shared library

Wcc : “unlinking”

The typical approach to reverse engineering is to transform binaries or shared libraries back to source code.

Instead, we aim at transforming final binaries or shared libraries back to ELF relocatable objects, that can later be relinked normally (using gcc/ld) into executables or shared objects.

What we do not want to do

Wcc : “unlinking”

Source code

 

Relocatable

Compiler

objects

 

 

(*.o)

 

 

L i n k e r

Binaries (executables,

shared libs…)

Wcc : “unlinking”

Source code

Compiler

Decompiler

Relocatable objects

L i n k e r

Binaries (executables,

shared libs…)

Wcc : “unlinking”

Source code

Compiler

Decompiler

Relocatable objects

(*.o)

L i n k e r

Binaries (executables,

shared libs…)

unlinking

Source code

Compiler

Decompiler

Relocatable objects

(*.o)

L

 

i

w

n

k

c

e

c

r

 

Binaries (executables,

shared libs…)

WCC : Command line

The command line is made to resemble the syntax of gcc :

Wcc : internals

The front end is build around libbfd. The backend is trivial C to copy each mapped section of the binary, handle symbols and relocations.

Benefit of using libbfd : the input binary doesn’t need to be an ELF !

=> We can for instance transform a Win64 executable into ELF 64b relocatable objects…

DEMO

(Binary to object file to relocatable to unstripped library)

WCC : demo

DEMO

(Crossing a Fish and a Rabbit)

PE + ELF = PELF

WCC : PE32 to ELF64

DEMO

Native OpenBSD on linux

Binary reflection

Binary “reflection” without a VM

Now that we know how to transform arbitrary binaries into shared libraries, we can load them into our address space via dlopen().

Let’s implement the same features as traditional virtual machines, but for raw binaries !

Whish list :

-Load arbitrary applications into memory

-Execute arbitrary functions with any arguments (and get results)

-Monitor/Trace execution

-Automated functions prototyping/annotation

-Learn new behavior

-Examine/Modify arbitrary memory

WSH : architecture

Loading is done via dlopen().

The core engine/shell is built around lua.

Can be compiled with luajit to get JIT compilation. Tracing/Memory analysis doesn’t rely on ptrace() : we share the address space.

Lightweight : ~5k lines of C.

No disassembler (as of writing. Subject to change). No need for /proc support !

Function names mapped in each library is dumped from the link_map cache.

Wsh : The wichcraft

interpreter

Distinctive features:

-We fully share the address space with analyzed applications (no ptrace() nor context switches).

-Requires no privileges/capabilities (no root, no ptrace(), no CAP_PTRACE, no /proc…)

-No disassembly : fully portable (POSIX)

-Implements “reflection” for binaries

-Full featured programming language

-Interactive and/or fully scriptable, autonomous programs

-Has no types

-Has no fixed API : any function you load in memory becomes available in WSH

-Functions have no prototypes

-=> Can call arbitrary functions without knowing their prototypes

Wsh : The wichcraft

interpreter

Advanced features:

-Loads any code via dlopen() : this solves relocations, symbols resolution, dependencies for us.

-Secondary loader bfd based (could load invalid binaries, anything in memory).

-Dumping of dynamic linker cash internals (undocumented) : linkmap

-Breakpoints without int 0x03 (use SIGINVALID + invalid opcode)

-Bruteforcing of mapped memory pages via msync() (0day, no /proc needed)

-Wsh can be compiled to do JIT compilation on the fly at runtime.

-Automated fuzzing/extended prototyping/functional testing

NONE OF THIS IS SUPPOSED TO WORK

Witchcraft

(Punk-C/Punxie)

Punk-C Language (WSH)

Lua Interpreter

+

“Reflected” C API

=

Punk-C

Witchcraft

DEMO

Witchcraft

DEMO ARM

Instant PoC

From Static analysis to

PoC

1 line PoC

Automated function annotation

Witchcraft

FUTURE WORK

FUTURE WORK

-Hyde our own presence better in memory (second heap)

-Remote debugging, running process injection

-Shadow mapping, internal libraries tracing (recursive ltrace)

-ltrace/strace to valid scripts

-system call tracing

TO BE CONTINUED

Questions ?