The purpose of this paper shall be first to examine some of the general characteristics of a computerized denial-of-service (DOS) attack, secondly to briefly describe a few actual attacks, and finally to suggest some best practices for defending against DOS attacks. According to a 1997 cert.org whitepaper on the subject, a DOS attack “is characterized by an explicit attempt by attackers to prevent legitimate users of a service from using that service.” Examples given include brute force attempts to “flood” a target network or service with false requests, attempts to disrupt connections between two target machines, attempts targeting a particular individual, or attempts to disrupt a target system. I think the latter two are a little broad to be included in our definition, however even the name denial of service is enough to describe the act of disrupting legitimate use of some service. The targeted service is often a network resource, but could be anything – disk space, a printer, even computational cycles. However, if we are to split hairs, botnets, which illegitimately use computational cycles, are not technically DOS since the attacker's goal is usually to steal processor cycles, not necessarily to deny them from their legitimate user. In fact, as I've pointed out on the discussion board, someone running a botnet might be very careful not to place a noticeable strain on system resources in order to evade detection. We shall, however, see that the botnet is one of the most powerful tools in a cracker's repertoire for perpetrating DOS attacks.
The CERT (1997) article identifies three categories of modes, or types of attack. They are the consumption of scarce resources, the destruction or alteration of configuration data, and the physical destruction or alteration of network components. While the latter two are particularly pernicious, we shall concern ourselves with the first one, consumption of resources. This category is further broken down into four types. They are:
- Network connectivity. This is the most frequent type of attack, and is an example of an asymmetric DOS attack – meaning the attacker does not need to match bandwidth with the server. The cited example is the TCP “SYN flood” attack described in 1996. Here, the attacking machine initiates a TCP connection, but leaves the server waiting for the connection to be completed with a client which has been carefully designed never to do so. The server has a limited number of processes devoted to connections, so a properly configured PIII machine on a 56k dial-up connection could theoretically shut down a relatively beefy server, regardless of it's bandwidth.
- Using your own resources against you. The example attack, also described in 1996, is the UDP port DOS attack, which takes advantage of the fact that “when a connection is established between two UDP services, each of which produces output, these two services can produce a very high number of packets that can lead to a denial of service on the machine(s) where the services are offered.”(CERT) In this particular instance, the targeted protocols are
chargenandechowhich, when connected to each other, either on the same machine, or two different ones, generate enough chatter to not only shut down both services, but potentially the network connection between them. - Bandwidth consumption. This is, until I researched this paper, what I thought of as the typical DOS attack. The strategy is to flood the targeted network with an enormous number of packets, essentially crowding out legitimate uses. In certain circles, this kind of brute-force attack might not be considered among the more elegant, but if anything that makes it more dangerous, since these attacks can be carried out by those with relatively little sophistication.
- Consumption of other resources. Just as it sounds, this category is diverse, but the common factor is the consumption of resources, which might include...
- System data structures, such as “process identifiers, process table entries, process slots, etc.” (CERT, 1997). The system's process table may be completely filled, or the machine can simply be bogged down by too many processes to effectively switch among them.
- Disk Space, including bombarding a particular email server with spam, intentionally generating warnings and errors that fill up log files, abuse of public FTP servers, or targeting any similar data stream that may be written to without bounds.
- Miscellaneous resources, including printers, backup devices, and network connections.
So now that the situation sounds completely hopeless, what can be done to protect against DOS attacks? Again, the 1997 CERT whitepaper outlines several best practices:
- The TCP SYN flooding attack described above can be defended by filtering at the router level, as well as by patches from OS vendors.
- Disabling all unnecessary services.
- Enforcing disk quotas and effectively using disk partitions. As a Linux user, I am familiar with the practice of creating separate partitions for log files, or FTP data.
- Establishing performance baselines against which the system may be periodically checked for signs of degradation, which might indicate a DOS attack.
- Routinely reassessing the physical security of components.
- Using a tool such as the classic UNIX utility Tripwire, which detects and reports changes to configuration files.
- Investing in and maintaining fault-tolerant networks and spare capacity that can mitigate damage in case of successful attack.
- A security standard that is excellent advice in many contexts – early and frequent backups.
- The equally familiar rules regarding password policies, especially regarding sensitive accounts.
Finally, it's hard to estimate the potential for financial harm that can be caused by DOS attacks, but we can easily imagine situations in which the loss can be significant. For instance, a struggling on-line retailer could be financially ruined by a successful DOS attack on an important shopping day. As if to underscore the seriousness of DOS attacks, the 1997 CERT whitepaper concludes with the advice that victims in the United States should contact their local FBI authority.
References
Carnegie Mellon University (1997). Denial of Service Attacks. Retrieved April 16, 2010 from Carnegie Mellon University, cert.org: http://www.cert.org/tech_tips/denial_of_service.html.
Carnegie Mellon University (1996). CERT® Advisory CA-1996-21 TCP SYN Flooding and IP Spoofing Attacks. Retrieved April 16, 2010 from Carnegie Mellon University, cert.org: http://www.cert.org/advisories/CA-1996-21.html.
Carnegie Mellon University (1996). CERT® Advisory CA-1996-01 UDP Port Denial-of-Service Attack. Retrieved April 16, 2010 from Carnegie Mellon University, cert.org: http://www.cert.org/advisories/CA-1996-01.html.
No comments:
Post a Comment