original: http://www.hackernews.com/bufferoverflow/00/dosattack/dosattack.html
reprinted: http://www.digitalmogul.com/membersonly/articles/2000/02/xxx.html
french: http://www.hackzone117.com.bi/bufferoverflow/00/dosattack/dosattack.html
french: http://www.s0y0s.t2u.com/secu/doc/dos/dosattack.html

Have Script, Will Destroy (Lessons in DoS)



I began writing this article almost one year ago, after the onslaught of smurf attacks being launched against various networks throughout the Internet. At the time, the newly discovered Denial of Service (DoS) attack was a crippling tool designed for one purpose; remotely disabling machines by flooding them with more traffic than they could handle. The smurf attack was the first well known (and well abused) DoS attack that could effectively cripple any network, regardless of size or bandwidth. This presented a new problem to network administrators and security personnel worldwide.

The LowDown

Also known as Network Saturation Attacks or Bandwidth Consumption Attacks, the new breed of DoS attacks flood a remote network with an staggering amount of traffic. Routers and servers targeted would go into overdrive attempting to route or handle each packet as it came in. As the network receives more and more of these illegitimate packets, it quickly begins to cause legitimate traffic like web and mail to be denied. In minutes, all network activity is shut down as the attack consumes all available network resources.

Prior to bandwidth consumption attacks, most DoS attacks involved sending very few malformed packets to a remote server that would cause it to crash. This occurred because of bugs in the way many servers handled the malformed packets. Malformed packets (also known as Magic Packets) consisted of network protocol options that were out of sequence, improperly matched, or too large. As a result, a server receiving these packets had no rules or guidelines dictating how it should behave when processing the malformed packet. The result was a system panic or crash that would basically shut the machine down or force it to reboot. Perhaps the most well known example of this type of attack is the WinNuke attack.

Regardless of ethics or motives, Magic Packet DoS attacks showed an inkling of grace in their execution. A single packet sent from one server to another, causing it to crash or reboot was a targeted attack. The precision with which this type of attack is carried out is analogous to a scalpel in surgery. Network consumption attacks on the other hand involve millions of packets. Worse, once launched the attack was no respecter of those standing between the launch point and the target network. Often times thousands of customers sharing bandwidth with the target would be adversely affected as well. A single attack of this nature had the ability to knock thousands of machines off the Internet in a single swoop. Such attacks are the equivalent of using a broadsword to do surgery.


The Next Generation

Attacks like the smurf DoS have a cascading affect that can be seen as a virtual avalanche. The starting point is nothing more than a few pebbles and snowballs (packets). As they travel downhill (along the path of routers to the target), they accumulate more mass and trigger the release of more pebbles. By the time the falling material hits the bottom of the mountain (the target), it is swamped in large amounts of snow and rocks. Despite the effectiveness of this attack, there is a single point from which the attack is launched. If an attack is detected early enough, it is possible to filter out the offending packets before they leave the original network.


The next generation of Denial of Service attacks are known as Distributed Denial of Service (DDoS) attacks. Expanding on the idea of network saturation attacks, DDoS effectively does the same thing but utilizes several launch points. The philosophy and objective of this is twofold. First, if a single machine being used to launch an attack is discovered and disabled, the overall attack proceeds with near full force. Second, by utilizing several launch points on different networks, an attacker is able to shut down larger networks that might not otherwise be affected by a single flood.

Taking Down the Big Boys

Prior to launching this form of DDoS flood, the attacker must first compromise various hosts on different networks. The more networks and machines used as launch points, the more potent the attack. Once each host had been broken into, they would install a DDoS client program on the machine that would sit ready to attack. Once the network of compromised servers was configured with the new client program, the attacker could send a quick command from the DDoS server software triggering each machine to launch an attack.


[chart comparing 56k vs cable vs t1 vs t3]
Until this last wave of DDoS attacks, it was generally assumed that hosts residing on large pipes (connections with incredible bandwidth) could not be seriously affected by network saturation attacks. As large Internet Service Providers (ISPs) are finding out, this is no longer the case. By using several smaller network connections, an attacker can eventually saturate the biggest ISPs and consume all of their bandwidth. This was demonstrated most effectively with the three hour shutdown Yahoo, and subsequent attacks against eBay, Amazon, Buy.com and other large scale web sites.

Difficulty in Tracking

Neophytes to networking always seem to question why these attacks are not tracked down, and the legs of the perpetrator not broken. It is a rare case to see ISPs interested in tracking down the individual(s) behind these attacks. Rather than take the time and effort to perform an investigation (which is lengthy), most ISPs realize that a quick filter denying ALL traffic to the site being attacked is a better solution. In essence, the ISP does the job of the person launching the attack and does it much more efficiently. As you can imagine, that is not exactly a deterrent for those committing these attacks.

One of the primary reasons investigations of DoS attacks is lengthy is it involves tracking down the packets hitting the target. Rather than leave the launch point with the IP address of the machine actually being used, the packets are tagged with forged source IP addresses. Since the IP information in each packet varies wildly, and since the addresses cannot be trusted, a network administrator must trace the packets back to the source one router at a time. This involves connecting to the router (often times this must be done at the physical console for security reasons), setting up a filter or sniffer to detect where the packets are coming from before arriving at that particular router, and then move to the new offending router. This presents problems when you consider a single packet may cross as many as thirty routers owned by ten different companies.

The act of forging the source IP of a packet is called IP Spoofing and is the basis for a wide variety of network attacks. One of the original intentions of a Denial of Service attack was to knock a machine off the network in order for you to assume it's identity. Once you masquerade as that machine, it is possible to intercept traffic intended for it as well as gain access to other machines on the target network via trusted host relationships. Attackers today seem to have lost all focus on the reason one would committ a DoS attack.

Save The Day Already!

Denial of Service attacks are not new. They have existed in one form or another since computers were invented. In the past they involved consuming resources like disk drive space, memory or CPU cycles. Those not familiar with how computers operate often scream for quick solutions to the various DoS attacks that plague our networks. Unfortunately, this is easier said than done.

Every weekday morning and afternoon millions of Americans go to and from work. They pile on to two and four lane freeways only to move at a crawl. Travelling ten miles in one hour is a common occurrence for those fighting rush hour traffic in heavily trafficked areas of business districts in cities across the nation. Every day they carry out this ritual, screaming and cursing the thousands of other drivers clogging the roads, and day after day the problem does not fix itself. Be it packets or cars, it is very well established that enough of either will overcrowd a road or network connection. At a given point, too many of either will bring all traffic to a standstill. Why isn't the traffic problem solved? We all know the solution is bigger and better roads, more carpooling, diverse schedules, and more common sense when behind the wheel. Fat chance that will happen anytime soon. On the flip side, it is very unlikely that they will fix every router on every network and install mechanisms to help avoid network saturation attacks.

In the long run, it is a rather simple fix that could help eliminate these attacks. Any network device that accepts or passes network traffic can be designed to monitor activity better. If a web server is receiving too many hits, it starts rejecting new connections so that existing connections can still view pages or interact with the site. This practice is called throttling or bandwidth limiting and is designed to prevent excessive connections, conserve resources and keep things operating correctly. Unfortunately, this philosophy has not carried over to routers (the machines that pass all internet traffic) so network consumption attacks go on unchecked. A relatively few amount of networks have learned this is a good solution to flood attacks. As such, their routers are designed to monitor traffic and quit passing illegitimate traffic once detected. The problem with this approach is that once the flood of packets have hit the remote network, the damage is done. The downside to this mechanism is the added latency as the router checks each and every packet that passes through it. Because of this slowdown, ISPs hesitate implementing this solution.

In order to make connection throttling effective, every network router should have this mechanism implemented. This would allow a router close to the source of the attack to detect the illicit traffic and put up a filter that rejected it before it left the launch point. This invariably leads to the question "How do you know if traffic is illegitimate?" Looking back to the section on IP spoofing, we can easily create a quick solution to the problem. In fact, this mechanism is found in most Firewalls implemented today.


In the diagram above, we show a forged packet with the IP address of 150.23.83.44. It stands to reason that such a packet would not legitimately be travelling around a network designated by the 1.2.3.x subnet. Because of this, any router on that network (especially the one acting as a gateway to the outside world) receiving that packet should drop it. Instead of blindly passing the packet on without question, routers should discriminate against suspicious packets by refusing to pass them on to the next router and setting off some kind of alarm for the administrator.

A second mechanism can be put into place that would help cut down on these attacks. On any given day, there is an average amount of traffic passed through any router. By monitoring these averages and applying other common sense rules, routers could be made to throttle heavily increased traffic. For example, if a router detected a sudden surge in traffic to a destination machine in which every packet claims to originate from a different IP address, that is a good sign of a saturation attack using spoofed packets. Rather than pass that traffic down the network, the router should throttle the traffic to avoid the likely flood that will ensue.

As stated many times before, easier said than done. Implementing these features falls on the many vendors of routers. Using these routers on production networks on the open Internet is up to the tens of thousands of companies maintaining a presence on the Internet. These upgrades cost time and money, something companies hesitate to invest; until the first time they are on the receiving end of such an attack. Like most security incidents, companies tend to implement reactive security measures, rarely proactive measures.

Why Ask Why?

Somewhere along the way, everyone wants to know why such attacks are carried out. Using the recent series of attacks against Yahoo, eBay and others is just as good example as any. To quash the distant hopes of a reasonable explanation, "There is no good reason!".

Consider that your typical DDoS attack affects hundreds (if not thousands) of machines, on a wide variety of networks. The single purpose of the attack is to cripple or shut down the target site so that it can not receive legitimate traffic. There are only a handful of reasons for doing it at all, none of which are reasonable or justifiable. In other words, DoS attacks are worthless and childish.

The first reason with perhaps the longest history is simple revenge. Some site out there wronged you in some way. Perhaps they spammed you, stopped hosting the free web pages they provided for you, fired your father or committed some other transgression. DoS attacks are a form of virtual revenge, especially against companies doing business over the Internet. The primary argument here is that these attacks cause problems for a number of ISPs, other customers who share bandwidth with the target, as well as the satisfied customers of the site. This goes back to the broadsword vs scalpel analogy.

The second reason has become rather trendy with novice script kiddies, second rate web page defacers, and those under the illusion they are part of the professional security community. "I did it to prove the system was vulnerable!" This is perhaps the most pathetic justification for launching a DoS attack. To many, this is no different than the attacker setting off a large nuclear device right next to a corporate server and then proclaiming "See! This can impact your operations!" Of course it can, this has been proven a hundred times over.

The third reason I can come up with falls back to playground rules. "If I can't play kickball, I'll throw the ball on the roof so no one else can play either!" This third grade mentality is far from justification of such attacks. Those wishing to exact some form of punishment against a site should consider the diminished intellect required to launch these attacks. There are better ways to deal with mean companies.

My Rant

Three types of people deserve the brunt of harsh insults and petty name calling. Each are responsible for this problem plaguing Internet users, and each could do their part to help stop it.

Each individual that carries out a DoS attack does so knowing full well what it could result in if they are caught. Practically nothing. There is precious little to deter someone from carrying out such vicious attacks. The very few times administrators put effort into tracking down a malicious user it results in them getting ousted from the ISP. The next day, the offending user is back online accessing the Internet via another ISP. Until the attack against Yahoo, the Federal Bureau of Investigation (FBI) was not concerned over these attacks. To date, the FBI has not managed to apprehend the perpetrator of a devastating DoS attack against their own home page (www.fbi.gov). For one reason or another they were seen as an annoyance, not a reason for loss of business. Law Enforcement needs to take a bigger interest in DoS attacks and start to punish those responsible. These types of attacks should take any competent law enforcement agent a few hours of tracking and maybe a handful of legitimate warrants.

Like the FBI, ISPs receiving these attacks need to take more proactive steps in preventing DoS attacks. When they do occur, ISPs should also take more time in tracking down the offending users and passing on the information to appropriate law enforcement. Rather than silently kicking them off the Internet for a day, taking a more active and public stance showing that malicious activity will not be tolerated would have a better effect. Those ISPs scared of retaliation need to remember that they are in the best position to stop the attackers.

Last, the pathetic kids (literally and figuratively) committing these attacks. In many cases, these attacks are launched with mystical scripts written in foreign languages and just produce the desired affect. There is no grace, no skill, and no intellect behind these attacks. You are not a hacker and you do not deserve respect for your childish actions. You are no better than the twisted individuals who spray a crowd of innocent bystanders with a machine gun, only to nick your intended target. If you can't express yourself better than a saturation attack, and can't deal with being called a name or wronged somehow, seek help offline. You sorely need it.



Article: Brian Martin (bmartin@attrition.org)
Images: Dale Coddington (dalec@attrition.org)
http://www.attrition.org

Copyright 2000 Brian Martin