Aléas numériques

Linux, infosec and whatever crosses my mind.


» Attacking DHCP

Table of contents

This article will introduce 2 different kinds of DHCP attacks: DHCP starvation attack and DHCP rogue server attack. But first, some reminders.

The DHCP protocol

Dynamic Host Configuration Protocol allows computers to automatically receive IP addresses and network configuration from a DHCP server.

Here is a typical IP address obtention :

DHCP Client                   DHCP Server
    +
    +------------------------------>
             DHCP Discover
                                   +
    <------------------------------+
             DHCP Offer
    +
    +------------------------------>
             DHCP Request
                                   +
    <------------------------------+
       DHCP Ack (or Decline, Nack)
    +
    +------------------------------>
             DHCP Release

We can see 7 different kinds of frames :

  • DHCP Discover: this frame is broadcasted to all the networks, to find the DHCP server.
  • DHCP Offer: the server responds to a DHCP discovery in unicast. This frame contains network configuration (IP address pool, gateway address…).
  • DHCP Request: the client sends a broadcast frame to announce from which server he wants to use the configuration.
  • DHCP Ack: the chosen DHCP server assigns the IP and configuration parameters and acknowledges them.
  • DHCP Nack: the DHCP server rejects the client’s request.
  • DHCP Decline: the client rejects the offered IP address.
  • DHCP Release: the client sends back his assigned IP address before the lease expires.

Note that the response from the server (the DHCP Offer frame) contains not only the client IP, but other important parameters as netmask, default gateway, and DNS.

For more pieces of information about DHCP, check the Wikipedia page.

DHCP starvation attack

This first attack consists of exhausting the DHCP server’s IP addresses pool with a huge number of requests.

The attacker sends many DHCP requests with different MAC addresses, which results in using all the available IP addresses. All the new machines that will try to connect to the network will not have any IP.

Then, the attacker can configure his working machine to be the new DHCP server to serve the new machines.

In a DHCP lease, information about the default gateway and DNS are transmitted. The attacker can propose a lease to newcomers saying that he is the default gateway, which allows Man In The Middle attacks: every packet sent by hosts will go through the attacker’s machine.

There is a famous framework for level 2 attacks called yersinia that allow us to launch DHCP starvation attacks :

# yersinia dhcp -attack 1 -interface eth0

After specifying the protocol with the DHCP argument, we indicate the attack mode. -attack 1 corresponds to ‘DoS attack sending discover packets’ (see this section). The flag -interface allow us to specify which interface to use during the attack.

Once the attack is launched, we can check the MAC address available space on the switch:

Cisco2960# show mac address-table count
...
Total Mac Address Space Available : 0

It works! Now, you can create your DHCP rogue server :)

To stop the attack, just kill the process :

# killall yersinia

DHCP rogue server attack

The goal of this attack is to introduce in the network a rogue DHCP server that will respond to clients’ requests.

To succeed, you have to respond faster to DHCP Discover requests than the initial DHCP server. This can be done in multiple ways :

  • By launching a DoS attack to the current DHCP server: this will result in a long time to respond, which gives you an advantage.

  • By re-implementing DHCP on the attacker machine: DHCP servers usually do other things (DNS, gateway…). They take more time to respond than a simple DHCP server. Moreover, they have to look into their cache to see if an IP address has already been attributed, etc… So by implementing a DHCP server that will directly respond to a DHCP Discovery request with a hard-coded IP address, it is possible to be faster.

In reality, you have to be faster twice: to reply to the DHCP Discovery and to send the DHCP Ack to validate the offer.

As seen before, yersinia allows us to do this attack :

# yersinia dhcp -attack 2 -interface eth0

where -attack 2 means ’nonDoS attack creating DHCP rogue server’. This attack mode does not need to use DoS, as its implementation is probably faster than standard DHCP used at home/office routers.

How to protect

Being offensive is nice, but it’s interesting to see the blue side of the Force. I’ll talk about Cisco equipment features.

There are 2 principal ways to avoid those attacks on Cisco equipment: DHCP snooping and IP source guard.

  • DHCP snooping allows to filter suspicious DHCP requests, and building what is called a ‘DHCP binding table’. This table contains the DHCP attributions, like MAC addresses, IP addresses, lease duration, VLAN number, and corresponding interface.

The sysadmin can specify on the switch trusted interfaces on which DHCP offers and DHCP {Ack,NAck} can be received. Those interfaces are designated as trusted, and others as untrusted.

Each interface that links a client to the switch must be set to untrusted, which only permits DHCP Discover/Request packets to enter; others are dropped.

Ports on which a DHCP server is connected must be set as trusted for the switch to accept DHCP Offers and DHCP {Ack,NAck} packets.

The DCHP binding table holds information about untrusted ports and is fed by dynamic entries learned via DHCP. On an important network, it is recommended to outsource this table: locally, it is stored in flash memory. For each new entry, its content have to be erased and written again. It can also generate heavy CPU loads and in case of a shutdown, all the tables are lost.

It is possible to configure automatic outsourcing as follows:

(config)# ip dhcp snooping database ftp://192.168.42.69/binding-table.dhcp
(config)# ip dhcp snooping database write-delay 300

In the example we use FTP, but HTTP, RCP, and TFTP are allowed too. write-delay is the duration between every copy when the table changes.

  • IP source guard allows us to protect from IP usurpation obtained by DHCP. In this kind of attack, the attacker changes his IP and/or his MAC address to access a remote machine (IP spoofing) or to avoid ACL set by the sysadmin.

IP source guard uses the DHCP binding table. In the beginning, all the IP traffic is dropped, except DHCP packets. Once a client has received a valid IP from the server, a VLAN ACL is set on the corresponding port. All the traffic emitted with another IP∕MAC on this port will be dropped.

To configure IP source guard on a Cisco switch, you can enter :

(config)# interface FastEthernet1/0/3   # or whatever interface you want
(config-if)# ip verify source port security