Indices and tables

VAGD

VirtuAlization GDb integrations in pwntools

Installation

pip install vagd

or from repo with

git clone https://github.com/gfelber/vagd
pip install ./vagd/

Usage

vagd template [OPTIONS] [BINARY] [IP] [PORT] to generate a template, list OPTIONS with help -h

from pwn import *

GOFF   = 0x555555554000                               # GDB default base address
IP     = ''                                           # remote IP
PORT   = 0                                            # remote PORT
BINARY = ''                                           # PATH to local binary
ARGS   = []                                           # ARGS supplied to binary
ENV    = {}                                           # ENV supplied to binary

# GDB SCRIPT, executed at start of GDB session (e.g. set breakpoints here)
GDB    = f"""
set follow-fork-mode parent

c"""

context.binary = exe = ELF(BINARY, checksec=False)    # binary
context.aslr = args.ASLR                              # ASLR enabled (only GDB)

vm = None
# setup vagd vm
def setup():
  global vm
  if args.REMOTE or args.LOCAL:
    return

  try:
    # only load vagd if needed
    from vagd import Dogd, Qegd, Box
  except:
    log.error('Failed to import vagd, either run locally using LOCAL or install it')
  if not vm:
    vm = Dogd(BINARY, image=Box.DOCKER_UBUNTU, ex=True, fast=True)  # Docker
    # vm = Qegd(BINARY, img=Box.QEMU_UBUNTU, ex=True, fast=True)  # Qemu
  if vm.is_new:
    # additional setup here
    log.info('new vagd instance')


# get target (pwnlib.tubes.tube)
def get_target(**kw) -> tubes.tube:
  if args.REMOTE:
    # context.log_level = 'debug'
    return remote(IP, PORT)

  if args.LOCAL:
    if args.GDB:
      return gdb.debug([BINARY] + ARGS, env=ENV, gdbscript=GDB, **kw)
    return process([BINARY] + ARGS, env=ENV, **kw)

  return vm.start(argv=ARGS, env=ENV, gdbscript=GDB, **kw)


setup()

#===========================================================
#                   EXPLOIT STARTS HERE
#===========================================================

# libc = ELF('', checksec=False)

t = get_target()

t.interactive() # or it()

vagd info BINARY to print info about binary

# run as process in VM
./exploit.py
# run with gdb attached, requires tmux
./exploit.py GDB
# run with gdb and ASLR enabled
./exploit.py GDB ASLR
# run on remote IP:PORT
./exploit.py REMOTE
# run on a different remote target (IPv6 in brackets)
./exploit.py REMOTE HOST[:PORT] [PORT]
# run process locally
./exploit.py LOCAL [GDB]

I recommend using pwndbg.

Files

All created files ares stored in the local ./.vagd/ directory. Additional large files (e.g. cloudimages) are stored in the home directory ~/.share/local/vagd/ or handled by tools themselfs (e.g. Docker).

CLI

alias vagd="python -m vagd" # or install with pip / pipx
# help message
vagd -h
# analyses the binary, prints checksec and .comment (often includes Distro and Compiler info)
vagd info BINARY
# creates template, for more info use: vagd template -h
vagd template [OPTIONS] [BINARY] [IP] [PORT]
# ssh to current vagd instance, for more info use: vagd ssh -h
vagd ssh [OPTIONS]
# scp file to/from vagd instance, for more info use: vagd scp -h
# e.g. vagd scp ./test_file vagd:./ # vagd:./ is default target
vagd scp [OPTIONS] SOURCE [TARGET]
# stop and remove current vagd instance, for more info use: vagd clean -h
vagd clean [OPTIONS]

IP and PORT can also be given as HOST:PORT or nc HOST PORT, or through the environment variables VAGD_IP and VAGD_PORT. BINARY may be omitted (e.g. vagd template host:1337), the template then only sets context.arch.

vagd template -a/--auto detects the container image: it uses the final FROM of a Dockerfile next to the binary (then in the current directory) and falls back to the compiler string in the binary’s .comment section (Ubuntu/Debian).

To patch a binary explicitly, use -p/--patchelf with -l/--libc, repeatable -L/--library, or -i/--interpreter. The original is saved as <binary>.bak; the exact patched binary is used locally and inside the environment. Supplying -l/--libc alone only includes libc in the generated template.

Documentation 

Boxes

A listed of known working Boxes can be found in the Documentation. Other images might also work but currently only distributions that use apt and alpine for Docker are supported. This limitation may be circumvented by creating a target yourself (with the dependencies gdbserver, python, openssh) and creating a ssh connection via Shgd.

Troubleshooting

background processes

all instances continue to run in the background (after a vagd object has been started), this improves the runtime greatly after the first execution of the exploit. But this means that instances must be killed manually e.g.: vagd clean

gdb & gdbserver

For containers (Dogd, Pogd) the host gdb attaches directly to the process (native=True, default) and uses the container root as sysroot, no gdbserver or sshfs is involved. This requires that the container user has the same uid as the host user (the generated Dockerfile takes care of this) and kernel.yama.ptrace_scope <= 1. Otherwise, and for VMs, gdbserver is used; I recommend using pwndbg. Other well known gdb plugins like peda aren’t compatible with gdbserver and therefore won’t work.

files

files on the virtual instance are never overwritten this has performance reason (so files aren’t always copied if the exploit is run). If you need to updated files on the remote either use vagd scp or create use temporary directories Dogd(..., tmp=True)

gdb performance

Using gdbserver and gdb to index libraries can be very slow. Therefore an experimental feature is available that mounts libraries locally: Dogd(..., ex=True, fast=True)

Future plans

Better Docker integration

migrate away from ssh (attach from host) to get lower latency
additionally virtualize containers (Qemu) in order to change the used kernel