Know Your Enemy: Honeynets

Know Your Enemy:
Honeynets
What a honeynet is, its value, overview of how it works, and risk/issues involved.

honeynet Project
http://www.honeynet.org
Last Modified: 31 May, 2006

One of the primary tools used by the Honeynet Project and Research Alliance to capture information is the honeynet. The purpose of this paper is to discuss what a honeynet is, its value, an overview of how it works, and the risks/issues involved. This paper is not a technical blueprint on how a honeynet works. Instead, this paper is a initial introduction to the concepts and issues. This paper is the first in a series of three documents. If after reading this paper you want to learn more about honeynets, and specifically honeywalls and how they work, we recommend you read the second paper in the series, KYE: Honeywall CDROM. Finally, if you are interested in actually deploying such a technoloty, after you have read these first two papers, the final and third document you will want to read is the Online User Manual. Whatever your technical knowledge or skill level, it is highly recommended you read this paper first. It is critical you understand the concepts, risks and issues of a honeynet before deploying such a technology.

Honeynet Overview
Traditionally, information security has been primarily defensive. Firewalls, Intrusion Detection Systems, encryption; all of these mechanisms are used defensively to protect one's resources. The strategy is to defend one's organization as best as possible, detect any failures in the defense, and then react to those failures. The problem with this approach is it purely defensive, the enemy has the initiative. Honeynets attempt to change that. The primary purpose of a honeynet is to gather information on threats. This information has different value to different organizations. For example, academic research institutions may use honeynets to gather data for research, such as worm activity. Security organizations may use honeynets to capture and analyze malware for anti-virus, IDS signatures or learn new ways to counter changing threats. Government organizations may use honeynets to learn more about who is targeting them or why. ISP's may use honeynets to capture, analyze, and terminate botnets that are on their networks. Security responders can use honeynets for incident response, collecting information on a compromised systems. The primary value of honeynets is information, that information will have different value to you depending on who you are and your needs. Examples of the information collected include the papers KYE: Tracking Botnets, KYE: Phishing. and KYE: Motives.

A honeynet is a type of honeypot. Specifically, it is a high-interaction honeypot designed to capture extensive information on threats. High-interaction means a honeynet provides real systems, applications, and services for attackers to interact with, as opposed to low-interaction honeypots such as Honeyd or Nepenthes which provide emulated services and operating systems. It is through this extensive interaction we gain information on threats, both external and internal to an organization. What makes a honeynet different from most honeypots is that it is a network of real computers for attackers to interact with. These victim systems (honeypots within the honeynet) can be any type of system, service, or information you want to provide. If you want to create Oracle databases on Solaris servers, not a problem. If you want to create a e-commerce site using IIS webserver on Windows 2003, not a problem. You can run everything from VAX systems to Cisco routers. Its is this flexibility that gives honeynets their true power.

Conceptually honeynets are very simple, they are a network that contains one or more honeypots. Since honeypots are not production systems, the honeynet itself has no production activity, no authorized services. As a result, any interaction with a honeynet implies malicious or unauthorized activity. Any connections intiated inbound to your honeynet are most likely a probe, scan, or attack. Any unauthorized outbound connections from your honeynet imply someone has compromised a system and has initiated outbound activity. This makes analyzing activity within your honeynet very simple. With traditional security technologies, such as firewall logs or IDS sensors, you have to sift through gigabytes of data, or thousands of alerts. A great deal of time and effort is spent looking through this information, attempting to eliminate false positives while identifying attacks or unauthorized activty. In many ways, its the classic needle in the haystack problem, as you attempt to find the critical incident amongst volumes of information. Since a honeynet is nothing more than a network of honeypots, all captured activity is assumed to be unauthorized or malicious. All you are doing is capturing needles. Its up to you to prioritize which of those needles has the greatest value to you, then analyze them in great detail.

Honeynets are an architecture. This architecture creates a highly controlled network, one that you can control and monitor all activity that happens within it. You then place your target systems, your honeypots, within that architecture. In many ways a honeynet is like a fishbowl. You create an environment where you can watch everything happening inside it. However, instead of putting rocks, coral, and sea weed in your fish bowl, you put Linux DNS servers, HP printers, and Juniper routers in your honeynet architecture. Just as a fish interacts with the elements in your fishbowl, intruders interact with your honeypots.

Honeynet Architecture
As we stated earlier, honeynets are nothing more than an architecture. To succesfully deploy a honeynet, you must correctly deploy the honeynet architecture. The key to the honeynet architecture is what we call a honeywall. This is a gateway device that seperates your honeypots from the rest of the world. Any traffic going to or from the honeypots must go through the honeywall. This gateway is traditionally a layer 2 bridging device, meaning the device should be invisible to anyone interacting with the honeypots. Below we see a diagram of this architecture. Our honeywall has 3 interfaces. The first 2 interfaces (eth0 and eth1) are what seperate our honeypots from everything else, these are bridged interfaces that have no IP stack. The 3rd interface (eth2, which is optional) has an IP stack allowing for remote administration.

There are several key requirements that a honeywall must implement; Data Control, Data Capture, Data Analysis, Data Collection. Data Control defines how activity is contained with the honeynet without an attacker knowing it. Its purpose is to minimize risk. Data Capture is capturing all of the attacker's activity without the attacker knowing it. Data Analysis is the ability to analyze this data. Data Collection is the ability to collect data from multiple honeynets to a single source. Of all these requirements, Data Control is the more important. Data Control always takes priority as its role is to mitigate risk. We describe each in more detail below.

Data Control is the containment of activity, it is what mitigates risk. By risk, we mean there is always the potential of an attacker or malicious code using a honeynet to attack or harm non-honeynet systems, or abusing the honeynet in some un-expected way. We want to make every effort possible to ensure that once an attacker is within our honeynet or a system is compromised, they cannot accidentally or purposefully harm other non-honeynet systems. The challenge is implementing data control while minimizing the attacker's or malcious's code chance of detecting it. This is more challenging then it seems. First, we have to allow the attackers some degree of freedom to act. The more activity we allow the attackers to perform, the more we can potentially learn about them. However, the more freedom you allow an attacker, the more risk there is they will circumvent Data Control and harm other non-honeynet systems. The balance of how much freedom to give the attacker vs. how much you restrict their activity is a decision every organization has to make themselves. Each organization will have different requirements and risk thresholds. One of the best ways to approach Data Control is not to rely on a single mechanism with which to implement it. Instead, implement Data Control using layers, such as counting outbound connections, intrusion prevention gateways, or bandwidth restrictions. The combination of several different mechanisms help protect against a single point of failure, especially when dealing with new or unknown attacks. Also, Data Control should operate in a fail closed manner. This means if there is a failure in your mechanisms (a process dying, hard drive full, misconfigured rules) the honeynet architecture should block all outbound activity, as opposed to allowing it. Keep in mind, we can only minimize risk. We can never entirely eliminate the potential of an attacker using a honeynet to harm non-honeynet systems. Different technologies and approaches to Data Control have different levels of risk, but none eliminate risk entirely. We discuss risk in greater detail later in the paper.

Data Capture is the monitoring and logging of all of the threat's activities within the honeynet. It is this captured data that is then analyzed to learn the tools, tactics, and motives of attackers. The challenge is to capture as much data as possible without the threat detecting the process. As with Data Control, one of the primary lessons learned for Data Capture has been the use of layers. It is critical to use multiple mechanisms for capturing activity. Not only does the combination of layers help piece together all of the attacker's actions, but it prevents having a single point of failure. The more layers of information that are captured, at both the network and host level, the more that can be learned. One of the challenges with Data Capture is that a large portion of attacker activity happens over encrypted channels (such as IPSec, SSH, SSL, etc). Data Capture mechanisms must take encryption into consideration. Also, just as with Data Control, we have to minimize the ability of attackers to detect our capture mechanisms. This is done several ways. First, make as few modifications to the honeypots as possible. The more modifications you make, the greater the chance of detection. Second it is best that captured data not be stored locally on the honeypots themselves. Not only could this data be detected by attackers, but it could also be modified or deleted. As such, captured data must be logged and stored on a seperate, secured system. Also, attackers may identify ways to detect Data Capture mechanisms, and develop methods to bypass or disable them. We discuss these issues in greater detail later in this paper.

Data Analysis is the third requirement. Remember, the entire purpose of a honeynet is information. A honeynet is worthless if you have no ability to convert the data it collect to information, you must have some ability to analyze the data. Different organizations have different needs, and as such will have different data analysis requirements.

Data Collection applies only to organizations that have multiple honeynets in distributed environments. Most organizations will have only one single honeynet, what we call a standalone deployment. As such they do not need to worry about Data Collection. However, organizations that have multiple honeynets logically or physically distributed around the world have to collect all of the captured data and store it in a central location. This way the captured data can be combined, exponentially increasing its value. The Data Collection requirement provides the secure means of centrally collecting all of the captured information from distributed honeynets.

Implementing all of these requirements is extremely difficult, complex, and time consuming. In the past it took a great deal of time and effort to deploy such an architecture. However, today the Honeynet Project has developed a rapid and simple way for an organization to deploy such functionality, its call the Honeywall CDROM. The purpose of this bootable CDROM is to make it simple to rapidly build and deploy a honeywall, the critical component to a honeynet architecture. You simply install the Honeywall CDROM into a computer with multiple NICs, and it automates the build process of a honeywall, implementing all of the requirements we just discussed above. If you are interested in deploying such a technology, you need to read the 2nd paper in our series of documentation, KYE: Honeywall CDROM.

Risks and Issues
Honeynets can be a powerful tool. They allow you to collect extensive information on a variety of threats. To obtain this information, you have to allow attackers and malicious code access -- potentially privileged access -- to your honeypots. As a result, the price you pay for this capability is risk. Any technology developed by man (or woman) can also be defeated by man (or woman.) Risk means different things to different organizations. You will have to identify what risks are important to you. Also, organizations have different thresholds for risk. We cannot determine what is right and wrong for you. Your organization will have to make those policy decisions for itself. All we can do is help make you aware of the risks. Also, we will not address legal issues of honeypots, or specifically honeynets. That is beyond the scope of this paper, and is specific to your country and organization. If you are interested in legal issues of honeypots, a good place to start is the legal chapter (which is freely available to the public) from our book Know Your Enemy: 2nd Edition. We also recommend you seek your organization's legal counsel for more information, especially in reference to privacy or liability issues. In reference to risk, there are four general areas we will cover; harm, detection, disabling, and violation.

Harm is when a honeynet is used to attack or or harm other, non-honeynet systems. For example, an attacker may break into a honeynet, then launch an outbound attack never seen before, successfully harming or compromising its intended victim. Data Control is the primary means of mitigating this risk. Multiple layers of Data Control are put in place to make it more difficult for the attacker to cause damage. However, there is no guaranteed method to ensure that a honeynet can not used to attack or harm someone else. No matter what mechanisms are put in place, an attacker can eventually bypass them. Your organization will have to decide how much risk it is willing to assume. For low risk organizations, you may want to minimize the activity allowed outbound (to zero, perhaps.) For organizations with greater risk thresholds, you may decided to to allow greater outbound activity.

Second, there is the risk of detection. Once the true identity of a honeynet has been identified, its value is dramatically reduced. Attackers can ignore or bypass the honeynet, eliminating its capability for capturing information. Perhaps even more dangerous is the threat that once identified, an attacker can introduce false or bogus information into a honeynet, misleading your data analysis. For example, with local access to the honeynet, an advanced attacker, or an attacker armed with proper tools, can potentially identify that a honeynet is in place and may even identify the honeynet Data Control and/or Data Capture mechanisms themselves.
Third, there is the risk of disabling honeynet functionality. This could be an attack against either Data Control or Data Capture routines. Attackers may want to not only detect a honeynet's identity, but disable its Data Control or Data Capture capabilities, potentially without the honeynet administrator knowing that functionality has been disabled. For example, an attacker may gain access to a honeypot within the honeynet, then disable Data Capture functionality on the honeypot. The attacker could then feed the honeypot with bogus activity, making administrators think Data Capture is still functioning and recording activity, when its not. Having mutliple layers of Data Control and Data Capture helps mitigate this risk, as there is no single point of failure.

The fourth and last risk, violation, is the catch all of remaining risk. Attackers may attempt criminal activity from your compromised honeynet without actually attacking anyone outside your honeynet. One example is an attacker using a Honeypot to upload, then distribute contraband or illegal material, such as illegal copies of movies, music, stolen credit cards, or child pornography. Remember, this individual broke into your system on their own initiative. You are not dealing with the most law abiding of cyber citizens. If detected, this illegal activity would be attributed (at least initially) to you by way of it being on your system. You may then have to prove to that it was in fact not you who was responsible for this activity.

There are several measures you can take to mitigate these risks, beyond what we have discussed so far. Two measures are human monitoring and customization. By human monitoring, we mean you have a trained professional monitoring and analyzing your honeynet in real time. This gives you the ability to detect a failure in your system, a failure that automated mechanisms may fail to detect or react to. By having a human analyzing honeynet activity, instead of just depending on automated techniques, you help protect yourself against new or unknown attacks or honeynet countermeasures. A second measure is customization. This paper and all honeynet technologies, including the Honeywall CDROM are OpenSource and publicly available. This means that anyone has access to this information, including the blackhat community, which we can assume are actively reading this and developing countermeasures. To help reduce risk you want to modify your honeynet from any default settings or normal behavior. The more your honeynet differes from standard or default configurations, the more difficult it will be others for detect or attack it. However, its critical that you understand that no matter what measures you take, risk is not eliminated, only mitigated.

Conclusion
Honeynets are a form of a high-interaction honeypot. Their primary advantage is their ability to gather extensive information. A honeynet is an architecture, similar to a fishbowl. Within this architecture you can deploy any type of system or application you desire. The critical requirements for this architecture are Data Control, Data Capture, Data Analysis and Data Collection, with Data Control taking the priority. While very powerful, honeynets present unique risks. Mechanisms can be put in place to mitigate these risks, but there is no way to eliminate all risk, it is crtical you understand this. If you are still interested in deploying a honeynet, the next step is to learn more about the Honeywall CDROM in our paper KYE: Honeywall CDROM.