Brandon Wiley
2002
The Warhol worm design began the theoretical discussion of so-called "superworms", a new type of computer worms. A worm is a computer program which copies itself from computer to computer in an attempt to reproduce as much as possible. A superworm uses more advanced techniques to achieve very quick infection of the network. The primary strategy behind the Warhol superworm is to pre-scan the network for vulnerable targets. When the worm is launched it already has a large list of targets with a known method for infection and can therefore quickly infect an initial seed population.
One thing which the Warhol paper mentions is that better results might be achieved via a coordinated worm in which various instances of the worm on different computers communicate with each other in order to optimize infection. The Warhol paper states, however, that no coordinated worm has ever been created. This paper proposes the first design for a worm which utilizes efficient communication between worm instances for an optimal infection strategy.
The purpose of adding coordination to a worm design is to raise the level of sophistication in the attack from a simplistic greedy strategy to a more game theoretically optimal cooperative divide and conquer strategy. There are times when a greedy strategy can be suboptimal. Overly zealous propagation can lead to early detection and eradication. Also, it is simply wasteful for a worm instance to attempt to infect a system which has already been infected rather than choosing an uninfected host as a target. Unfortunately, typical worms have no information on which to base a more sophisticated attack. In order to divide the infection tasks among operative worms, the worms must know about each other and have a method for dividing work among themselves.
The difficulty in creating a coordinated worm is in minimizing the coordination costs among worms. Since the initial goal of a worm is generally to reach all hosts on the Internet, the number of eventual worm instances will be enormous. The coordination strategy must be able to scale reasonably to that number of instances. If every worm had to coordinate with every other worm, for instance, the amount of bandwidth used to communicate between the worms could easily exceed that used by a greedy worm, defeating the benefits of coordination. The coordination strategy must also be simple to encode since worm designers attempt to make worms as small as possible.
Interestingly, the problem of efficiently organizing worm instances into a network which can act globally but which has reasonable coordination costs for each node is very similar to problems found in peer-to-peer networks. The particular task of the division of the task space among all of the currently active worms is very similar to the problem addressed in distributed hash tables (DHT) designs. One popular contemporary DHT design is called Chord. In Chord, each node is assigned a portion of the task space such that the space is divided evenly and randomly among all nodes. Chord has some useful properties. First, each node in the network is reachable from each other node in the network with a maximum of O(log N) intervening nodes. Additionally, each node only needs to maintain knowledge of O(log N) other nodes, thus keeping coordination costs down to a reasonable level. What this means in simple terms is that in a network of one million nodes each node only has to keep track of approximately 20 other nodes and for one node to send a message to another node in the most distant part of the network it would take at most 20 intervening nodes. Similarly, for a network of ten million nodes, each node has to keep track of approximately 23 other nodes and it will take at most 23 intervening nodes to reach from one side of the network to the other. There are advanced variants of the Chord architecture which layer additional properties on top of the guarantees provided by the basic Chord design. Anonymous Chord (Achord) adds the property that it is very difficult for any node to find out the identities of all of the other nodes in the network. This makes it more difficult for an attacker to disable the network by discovering the identities of nodes. By having worms form an Achord network, a global framework for division of the space to be attacked can be created with reasonable coordination costs.
In order to create an Achord network, each node needs to be assigned a unique, difficult to forge, difficult to generate identifier. Identifiers are assumed to be generally random and evenly distributed. Each task also needs such an identifier. Tasks are matched to the node whose identifier is the closest match. The method which Curious Yellow uses to assign identifiers to worms and targets is via the SHA1 hash of their IP address. It is relatively difficult to choose your own IP address and the SHA1 hash makes the identifier approximately random and evenly distributed.
The method for nominating a worm to attack a target is easy. Each Achord node knows the IP addresses of the two nodes whose identifiers are closest to its own. When it learns of a new target, it calculates the identifier for the target and then determines if it is closer to the worm's own identifier or one of its neighbors. If the worm is the closest to the target then it attacks the target. Otherwise, it informs the closer neighbor of the existence of the target and then forgets about it. Since the identifier space is globally consistent, decisions about which worm should attack will always be consistent. Additionally, the decision about who should attack does not require immediate communication between the worms. Communication is only necessary to inform nodes of found vulnerable nodes which they are responsible for attacking.
The initial deployment of the worm network using superworm pre-scanning techniques may take up to 15 minutes (Warhol) or merely 30 seconds (Flash). Once the initial seed network is deployed, it can be used as a platform for launching a second stage of activities. One obvious activity is distributed scanning of the network for vulnerabilities and further infection. Unlike Code Red, which used a greedy scanning strategy, Curious Yellow will have exactly one worm scanning each potential target. This will both reduce the load on the network and make detection less of a threat. The global connectedness of the entire worm network allows for an even more interesting type of distributed scanning than is at first apparent. Since all nodes are reachable from all other nodes, it is possible for the worm's creator to release code patches to all of the worms in the network and for these code patches to spread to the entire network even faster than the initial infection (less than 15 seconds). Therefore, as new exploits are found for previously invulnerable systems, they can be distributed to the worm network, which has already been building up a list of potential future targets. The Warhol method of pre-scanning attacks can thus be utilized repeatedly for rapid infection of diverse systems. The speed at which patches can be distributed to worms is so great that it will probably out-pace attempts to fix vulnerabilities. A zero-day exploit can be used by worms for infection before news of the vulnerability has even been made public. Code patches can also be made to change the behavior of the worm to mask signature behavior which could lead to its detection.
The second stage of infection allows the infection to progress from controlling a large portion of the network to controlling the overwhelming majority of the network. This is just another part of the infection stage. Once the majority of the network has been infected, Curious Yellow can lay dormant until part or all of it is activated for some purpose.
There are a number of possible purposes to which Curious Yellow could be used. One obvious use is to simply crash the majority of the Internet at once. Once it is activated, the worm network has achieved its purpose. A slightly more interesting use of the worm network would be to use it for distributed denial of service attacks against enemy hosts. The typical approach for this is to have all compromised hosts send a flood of packets to the target, thus overloading it sufficiently to keep any legitimate packets from getting through. However, this is a naive approach when given such an advanced network to work with. The Curious Yellow infection should, if properly deployed, control the vast majority of the network. All of the infected nodes can act in concert towards a common goal. Nodes and groups of nodes can be specialized for certain tasks. New directives can be sent to the entire network in less than 15 seconds. It is therefore not necessary to have the entire network gang up on a single machine in order to disable it. This is in fact a greedy rather than cooperative strategy and thus suboptimal. First of all, the target to be attacked is probably infected. Therefore, the worm controlling the target can simply be instructed to disable the target. Additionally, if all of the nodes surrounding the target simply drop traffic routed to the target then the target becomes unreachable. Finally, the worms controlling the hosts attempting to contact the target can simply ensure that no attempt to communicate to the server is ever made. Curious Yellow, acting globally and in unison, can make any host simply cease to exist as far as the network is concerned.
Having total control of all of the Internet's traffic allows for other, more interesting, attacks. Traffic can be modified arbitrarily as it passes through the network. Defacing a website no longer requires actually having access to the computer containing the website. Web pages can be defaced automatically as they pass through the network, resulting in the world's collective web browsers rendering the pages differently than they are stored on the servers, a problem that the server administrators are totally powerless to fix. All of the unencrypted traffic on the Internet can also be observed. The entity controlling Curious Yellow can pick out particular individuals to monitor or gather statistical information about a large number of individuals.
Of course, Curious Yellow's control over individual computers is not limited to controlling Internet traffic. As zero-day root exploits are found and patches distributed, worms can eventually gain superuser access to all of the machines, giving them access to all of the stored information and all of the spare resources such as hard drive space and CPU cycles, and the ability to surveil all of the world's Internet-connected computer users. By sending out code updates to the network which cause Curious Yellow to metamorphasize into an anonymizing proxy network, its owners can connect anonymously to target computers and control them interactively, browsing files and watching what users do with them. They could also program the worms to automatically send back potentially interesting information. The spare resources of the world's computers could be utilized for whatever agenda the owners of Curious Yellow have in mind. In general the uses of the network are endless. The entity which controls Curious Yellow controls the world's computers.
Dealing with the infection once it has been detected is difficult. Once a signature has been detected for the worm, it must be codified by the various competing virus scanner manufacturers and then distributed to infected computers, probably by voluntary downloads. Naturally, once an anti-virus patch for the worm becomes publicly available on the Internet, Curious Yellow will cause that site to disappear from the Internet. Inoculation will therefore have to happen by hand using physical media or network distribution which is secretive enough that that owners of Curious Yellow (subscribers to many major anti-virus update programs) don't find out about it. Once the patch falls into the hands of the creators, Curious Yellow will soon receive a counter-patch obsoleting the old anti-virus patch. Unfortunately, anti-virus distribution methods cannot keep up with the pace of Curious Yellow patch distribution. The only method which can eradicate the virus, therefore, is to disconnect the computers from the network and then apply via physical media patches which both eradicate the virus and patch the vulnerabilities which allowed it to spread. Once the virus is totally eradicated, the creators will wait for a new zero-day exploit to be discovered and then relaunch the virus with a new transmission vector and signature.
The only way to protect against Curious Yellow is to inoculate every computer with an anti-worm, Curious Blue, which uses similar technology to instantly distribute security patches. As soon as an exploit is discovered, a security patch must be released to Curious Blue before an exploit patch can be released to Curious Yellow. Infection and protection is thus primarily a race between the owners of the two entities. Of course, there might not be only two entities. There could be any number of competing vendors of Curious Blue offering different patches and different quality of service guarantees. Similarly, anyone with access to zero-day exploits could launch their own Curious Yellow. The battle does not end there, however. Curious Blue could act as an ideal platform for the initial stage of a Curious Yellow infection. All that is needed is an exploit in the Curious Blue code. Once one is found, the entire Curious Blue network can be turned, like a clever move in a game of Othello . The same is of course true of turning Curious Yellow into Curious Blue. These programs are particularly prone to such corruption because they are already designed to accept arbitrary code upgrades. They merely need to be fooled into accepting code which is not actually authorized.
The authorization of code updates is a crucial component to both Curious Yellow and Curious Blue. Without a strong authentication system, the worm network can easily be taken over by an arbitrary attacker. The obvious way to do authentication is with public key signatures. In order to use public key signatures, the entity deploying the worm creates a pair of keys, one public and one private. The public key is distributed with the worm. The private key is known only to the worm's creator. When the creator wants to send a new code update, it generates a signature from the code using the private key. Since the worms have the public key, they can check to see if the signature was in fact generated by the matching private key. Using this technique, no attacker can send code updates to the network unless he possesses the creator's private key or finds a vulnerability in the worm which allows circumvention of the signature check.
Maintaining the secrecy of the private key is an interesting problem in a world overrun by competing strains of Curious Yellow and Curious Blue. A simple strategy which an attacker controlling one worm network might use to compromise another is to instruct the network to search all computers for files that might potentially contain the private key of the competing network. Due to the large size of private keys, they cannot be easily remembered and so much be stored electronically somewhere. In order to keep the private key from being discovered, the creator will be forced to have a special computer used for generating signatures which is never connected to the network. Signatures will be generated on this computer and then transferred to a network-attached computer via removable media. The attack then is to find where in the network signatures are first introduced.
The worm network can be configured to search for signature files stored on removable media. The network can also monitor other coexisting worm networks to see when code updates are sent. When a received code update matches a signature file found on removable media, the creator of the worm has been detected. Naturally, the creator of a particular strain of Curious Yellow would prefer that his own computers were not infected with competing strains. Unfortunately, the only way to ensure this is to inoculate with a strain of Curious Blue, which will undoubtedly also be searching for the creator so as to have legal action taken against it. Assuming, however, that the creator has the resources to inoculate against all competing strains, it can still be tracked. As the code updates propagate through the network, competing strains can monitor the progress. Using statistical analysis of the propagation of code updates, the source of updates can eventually be traced. Once the location of the creator has been determined, physical coercion such as spying, threats, lawsuits, and arrest are possible to gain control of the private key and thus the worm network.
In order to avoid being traced, further cryptography is necessary. So that the progress of code updates through the network cannot be monitored, the worm code needs to be encrypted so that it cannot be easily examined to determine which code it currently is running. It is still possible to examine the contents in memory, but this will be a somewhat difficult task to encode in a program the size of a typical worm. Additionally, code updates being sent over the network must be encrypted so that their progress cannot be observed. Even with encrypted connections, however, the creator can still be traced through timing correlations. All the the observer needs to see is that one worm contacted another, then that worm contacted a few others, leading into a cascade. Whichever worm made the first contact is the one closest to the creator. Defeating timing correlation requires the worm network to be constantly sending cover traffic to other worms. Luckily, code updates are generally small, so the amount of cover traffic to be generated is not very much. Once the network is communicating entirely over encrypted channels with constant cover traffic, the creator can send out code updates in an anonymous, untraceable manner. Not only that, but the creator can also use the network to render anonymous any other transactions, such as using it as an anonymous communications channel to converse with other entities and distribute files and information. This would be a boon to the usual cast of characters that could benefit from anonymous communication, such as people attempting to escape human-rights-violating regimes, international terrorists, and music fans.
In the world after the global infection of the Internet by strains of Curious Yellow and the commercial availability of strains of Curious Blue, computer users will have a choice. One can either have a computer which is never connected to the Internet, risk almost certain infection and control by the various factions controlling Curious Yellow, or intentionally give control to the creators of Curious Blue. There are multiple issues of trust involved. Initially there is the question of whether one places more trust in the harmlessness of the hackers or the professional integrity of the security professionals. If one chooses Curious Blue then there is the issue of which strain will actually be effective in protecting one from infections by Curious Yellow. There is the additional issue of which strain can be trusted to not contain any vulnerabilities which can be exploited to turn it to the other side.
There is a disturbing similarity between Curious Yellow and the new Kazaa feature, Altnet. Kazaa is a peer-to-peer file sharing network not entirely unlike Achord, but lacking some of the useful features. In later versions of the software Kazaa bundled a feature called Altnet, which is a second peer-to-peer network deployed alongside Kazaa nodes. when Kazaa is installed, Altnet is quietly installed as well. Buried in the licensing agreement which users click through when installing Kazaa are some interesting provisions concerning Altnet. The user agrees that Altnet is allowed to automatically receive and install code updates and modify settings on the user's computer. This makes Altnet a prime target to be corrupted and used as a widely deployed network from which to launch activities. All that is needed is the proper method for causing the supposedly 2.5 million Kazaa nodes to accept a rogue code update. Interestingly, such an attack has already occurred. While Kazaa is the predominate licensee of the FastTrack network technology, it was previously second to an application called Morpheus, another application using the FastTrack network. Morpheus was mysteriously shut out from the FastTrack network despite the fact that it was supposedly an entirely decentralized network without a central form of control. The network of Morpheus clients was shut down by a rogue code update, eventually discovered to have been sent by the company behind Kazaa. This is the first example of the sort of warfare between strains. It could escalate into being literally a war between worm strains if an entity discovers the key to making Kazaa accept code updates and mobilizes the Kazaa network as a first stage of infection, using it for decentralized scanning of the network for vulnerable hosts and an eventual global takeover of the Internet.
[Back to index] [Comments (0)]