Advanced malware detection using similarity analysis

Abstract

A non-computerized method for detection of malware is described. First, an object for analysis is received. Thereafter, the content from the object is extracted and the extracted content is arranged into one or more basic blocks. Each basic block including at least a portion of the content. The object fingerprint is based on an analysis of the one or more basic blocks. Thereafter, the object fingerprint is generated based on an analysis of the one or more basic blocks. Lastly, the object fingerprint is compared to one or more malware family fingerprints to determine if the object is potentially malicious and may be associated with an advanced malware.

Description

FIELD

Embodiments of the disclosure relate to the field of cyber security. More specifically, embodiments of the disclosure relate to a system and method for detecting advanced malware, including zero-day malware, through analysis of data associated with an object and generation of a fingerprint for comparison with fingerprints associated with one or more malware families.

GENERAL BACKGROUND

Over the last decade, cybersecurity attacks have become a pervasive problem for internet users as many networked devices and other resources have been subjected to attack and compromised. The attack may involve the infiltration of malicious software onto a network device or concentration on an exploit residing within a network device to perpetrate the cybersecurity attack (generally referred to as “malware”). In most situations, malware is a program or file that is embedded within downloadable content and designed to allow or directly influence, undermine, disrupt, alter or otherwise attack normal operations of a network device. Examples of different types of malware may include bots, computer viruses, worms, Trojan horses, spyware, adware, or any other programming that operates within an electronic device without permission by a user of the electronic device or a network administrator responsible for protecting an enterprise network with which the electronic device is in communication. In some cases, the attack is designed to exploit a previously unknown vulnerability within software executing on a targeted network device.

Malware may be distributed through a variety of different attack vectors. For example, malware may be installed on a network device through activation of a uniform resource locator (URL), which redirects the user to unknowingly download content from a malicious web site for installation on his or her computer. Similarly, malware may also be installed on a network device upon receipt or opening of an electronic mail (email) message or an attachment with embedded executable malware (e.g., an infected document such as a Portable Document Format “PDF” or word processing document, an infected image, etc.). As yet another example, malware may exist in files that are uploaded from an infected network device onto a networked storage device such as a file share. Also, malware may be imbedded as part of a data stream that are directed to multiple (two or more) network devices. Identifying an advanced malware attack at the network device, such as a zero-day attack or a polymorphic malware attack for example, has been challenging.

A zero-day attack typically poses the substantial threat to an enterprise network. as these types of attacks are designed to exploit a previously unknown vulnerability within software executing on one or more targeted network devices, and often constitutes a previously unseen type of malware or malware that has not been detected before. In either case, no known signature is available for that malware. As “zero day” malware, by definition, has not been detected before, there are no known signatures for detection of this malware type. Accordingly, signature-based solutions typically fail to detect zero-day malware.

Moreover, advanced malware may co-opt and use previously whitelisted domains, i.e., domains not previously known to be malicious. Accordingly, solutions relying on domain-blacklists fail to be sufficiently effective. Finally, advanced malware is often polymorphic, and thus has signatures that change over time while retaining their core malicious functions, and, once again, may escape detection by such solutions.

Known malware detection systems effectively deal with these problems by employing virtualized behavior detection systems, typically at the periphery of an enterprise network. Unfortunately, the overhead necessary to run a virtualized behavior detection system in user space of a network device interferes and significantly impacts the user experience normally offered by a laptop or other endpoint device.

BRIEF DESCRIPTION OF THE DRAWINGS

Embodiments of the invention are illustrated by way of example and not by way of limitation in the figures of the accompanying drawings, in which like references indicate similar elements and in which:

FIG. 1A is a first exemplary embodiment of a network device deploying malware detection logic that is communicatively coupled to a security appliance implemented as part of the enterprise network.

FIG. 1B is a second exemplary embodiment of a network device deploying advanced malware detection logic that is communicatively coupled to cloud services.

FIG. 2 is an exemplary embodiment of the internal architecture of a network device of FIG. 1A or FIG. 1B.

FIG. 3 is a first exemplary embodiment of an operational flow of the advanced malware detection logic deployed within a network device of FIG. 1A or FIG. 1B in generating an object fingerprint based on analysis of disassembled instructions from binary data.

FIGS. 4A and 4B are exemplary embodiments of a control graph formulated during analysis of an incoming object by the advanced malware detection logic for use in generating basic blocks separately analyzed in the formation of an object fingerprint.

FIG. 5 is an exemplary embodiment of an operational flow of the advanced malware detection logic deployed within a network device of FIG. 1A or FIG. 1B in generating the object fingerprint based on disassembled instruction sequences for comparison with fingerprints associated with known malware families in determining whether the object is associated with a zero-day attack.

FIG. 6 is a second exemplary embodiment of an operational flow of the advanced malware detection logic deployed within a network device of FIG. 1A or FIG. 1B in generating the object fingerprint based on analysis of raw data.

FIG. 7 is an exemplary embodiment of an operational flow of the malware detection logic deployed within a network device of FIG. 1A or FIG. 1B in determining whether an object under analysis is advanced malware through object fingerprint comparisons.

DETAILED DESCRIPTION

Various embodiments of the disclosure relate to a system and method for detecting advanced malware through generation and comparison of an object fingerprint with malware family fingerprints. A “fingerprint” may be construed as a plurality of indicators that collectively identify an object under analysis. Each “indicator” uniquely corresponds to a representation of content that is part of the object, such as a hash result produced by performing a cryptographic hash operation (e.g., any type of Secure Hash Algorithm including SHA-256 or SHA-512) on the content. An example of content is ordered data that may include, but is not limited or restricted to the following: (i) a sequence of instructions (e.g., instructions sequenced in the order of disassembly) or (ii) a collection of raw “binary” data (e.g., data ordered as received or as transmitted). Herein, a fingerprint may be computed using a feature vector, namely a plurality of features each uniquely corresponding to a representation of ordered data that is being monitored. Herein, an “object fingerprint” refers to a fingerprint that is produced from data associated with an object under analysis while a “malware family fingerprint” refers to a fingerprint associated with a known category of malware (e.g., malware and variants of a malware).

I. SUMMARY

One embodiment of the disclosure is directed to advanced malware detection logic, which may be implemented as part of a network device. Herein, the advanced malware detection logic conducts an analysis of an object (referred to as a “similarity analysis”) in an attempt to classify the object as malicious or benign. This similarity analysis may include a determination of a level of correlation between indicators of an object fingerprint for an object under analysis (e.g., a hash results associated with a sequence of instructions, an ordered collection of raw “binary” data, or another type of ordered data associated with the object under analysis) and indicators associated with at least known malicious objects. According to one embodiment of the disclosure, this level of correlation may be ascertained by determining the number, percentage or ratio of indicators of the object fingerprint that compare to indicators for each of the malware family fingerprints, when all of the indicators of the malware family fingerprint being equally weighted. Alternatively, the level of correlation may be ascertained by determining whether a certain number of higher weighted indicators of a malware family fingerprint compare to indicators of the object fingerprint. The indicators with an increased weighting may be selected as those indicators whose presence denotes an increased likelihood of the object being associated with a malicious attack.

The advanced malware detection logic is configured to leverage machine learning and experiential knowledge of known malicious and perhaps benign software during its analysis of objects received by the network device to identify malicious software at any point in its attack lifecycle. The operations of the advanced malware detection logic may be conducted by a network device without compromising usability (user experience).

In accordance with one embodiment of the disclosure, the advanced malware detection logic conducts a “moving window” analysis of content, such as instructions that are disassembled from binary code of the object. The instructions are analyzed to generate an object control graph, which represents the operational flow of the object based on an analysis of the disassembled instructions without execution of these instructions. The object control graph features selected groupings of instructions (basic blocks) and control paths between these basic blocks. Thereafter, the advanced malware detection logic produces representations (e.g., hash results, results of reduced size than the sequence of instructions, etc.) for each basic block as defined by an adjusting (sliding) analysis window. These representations are compared to features of a feature vector, and when a representation matches one of the features, that feature is set. Hence, unique portions of the analyzed content are captured as a feature of the feature vector.

After completing analysis of the instructions that form the object control graph, the feature vector now represents an object fingerprint and each “set” feature represents an indicator of the object fingerprint. For classification of the object, the object fingerprint is compared to a plurality of malware family fingerprints. This comparison may involve a determination of a level of correlation between the object fingerprint and any malware family fingerprint, which may include a comparison of the indicators of the object fingerprint to malicious indicators of the malware family fingerprints (e.g., representations of sequences of instructions or ordered data typically used by a particular malware family). The malware family fingerprints may be fetched from a database accessible to the network device via a network, where some or all of the malware family fingerprints may be cached and updates of the malware family fingerprints may occur periodically or aperiodically, employing a “push” or “pull” methodology to secure these malware family fingerprints.

In accordance with another embodiment of the disclosure, the advanced malware detection logic conducts the moving window analysis of a collection of raw “binary” data associated with the object. Herein, the advanced malware detection logic treats the entire collection of binary data as a single basic block of an object control graph, where an illustrative example is described below. A sliding window (e.g., a first number of bytes that define a boundary for the analysis) along with a window step size (e.g., a second number of bytes that is less than the first number of bytes) are selected. As an illustrative example, with a fixed, sliding window of 128 bytes and a fixed, window step size of 16 bytes, the moving window analysis advances 16 bytes after each sliding window calculation of 128 bytes of binary data until a full window of binary data is not available. The iterative analysis creates representations of 16-byte subsets of binary data, which is used in the computation of the object fingerprint associated with the object under analysis by setting those features in the feature vector that match or have a certain level of correlation with that feature (e.g., a predetermined number of bytes match the representation associated with the feature). Thereafter, the feature vector constitutes the object fingerprint.

In accordance with yet another embodiment of the disclosure, a similar technique is applied to classify “unknown” objects, such as a binary object (executable) stored on or being processed by a network device. The technique may be applied, for example, pursuant to a schedule or upon execution of the binary object, not analysis of the content of the object. Significantly, this technique is employed at a network (endpoint) device during its normal operation and often in real time, either as a foreground process or background process.

The advanced malware detection logic deployed within a network device may be configured to conduct operations to detect advanced malware, including the following operations:

• • (1) extract and optionally disassemble an object (e.g., binary code, “raw” binary data) to create an object control graph (a graph of the control flow particular to the object);
  • (2) determine a size for a sliding window (of a size corresponding to the object type) for analysis of the object control graph and optionally the stepwise adjustment of the sliding window;
  • (3) analyze content of the object control graph to determine whether a certain degree of correlation exists between the collective content defined by iterative moving window analyses, where the analysis may involve a comparison of indicators that correspond to hash results associated with certain ordered data (e.g., sequence of instructions, group of binary data, etc.) to malicious and/or benign indicators that represent hash results associated with ordered data associated with known malware and/or benign objects;
  • (4) classify whether the degree of correlation indicates that the object is associated with advanced malware, in some embodiments employing a score generator to arrive at a level of classification (e.g., malicious or benign);
  • (5) if the object is considered to be malicious, determine the content (e.g., code as a sequence of instructions, group of binary data) corresponding to each sliding window, generate an identifier for each sliding window, and generate alerts of maliciousness for transmission to administrators and possible the endpoint user; and
  • (6) distribute the generated identifiers to static detection logic of the network device, to a security appliance or centralized management system on the network, and/or to a remote signature distribution service.

II. TERMINOLOGY

In the following description, certain terminology is used to describe features of the invention. For example, in certain situations, the term “logic” is representative of hardware, firmware and/or software that is configured to perform one or more functions. As hardware, logic may include circuitry having data processing or storage functionality. Examples of such circuitry may include, but are not limited or restricted to a microprocessor; one or more processor cores; a programmable gate array; a microcontroller; an application specific integrated circuit; receiver, transmitter and/or transceiver circuitry; semiconductor memory; combinatorial circuitry; or the like. It is contemplated that all logic components, typically represented by boxes in FIGS. 1A-7 herein, may be deployed as hardware, software and/or firmware.

Logic also may be in the form of one or more software modules, such as executable code in the form of an executable application, an application programming interface (API), a subroutine, a function, a procedure, an applet, a servlet, a routine, source code, object code, a shared library/dynamic load library, or one or more instructions. These software modules may be stored in any type of a suitable non-transitory storage medium, or transitory storage medium (e.g., electrical, optical, acoustical or other form of propagated signals). Examples of non-transitory storage medium may include, but are not limited or restricted to a programmable circuit; a semiconductor memory; non-persistent storage such as volatile memory (e.g., any type of random access memory “RAM”); persistent storage such as non-volatile memory (e.g., read-only memory “ROM”, power-backed RAM, flash memory, phase-change memory, etc.), a solid-state drive, hard disk drive, an optical disc drive, or a portable memory device. As firmware, the executable code is stored in persistent storage.

The term “content” generally refers to information that is part of the object. One example of content may include one or more sequences of instructions such as opcode, for example, that is disassembled from binary code. Another example of content may include a collection of “raw” binary data (e.g., bits with logic “1” or “0” values). The “object” may include binary code (e.g., executable, dynamic link library “DLL”, etc.), which features one or more processor instructions. Also, the terms “compare,” “comparing,” “comparison” or any tense thereof generally mean determining if a predetermined level of correlation is achieved between two items, where one of the items may include a fingerprint associated with an object under analysis.

The term “object” generally refers to content, namely a collection of data in transit (e.g., propagating over a network) or at rest (e.g., stored), which has a logical structure or organization that enables it to be classified for purposes of analysis for malware. In one embodiment, an object may include an executable in any of a variety of file formats. As illustrative examples, the file formats may include Android™ APK, x86 Windows® supporting 32-bit or 64-bit operating system, Flash® files, or the like.

The term “network device” should be construed as any electronic device with the capability of processing data and connecting to a network. Such a network may be a public network such as the Internet or a private network such as a wireless data telecommunication network, wide area network, a type of local area network (LAN), or a combination of networks. Examples of a network device may include, but are not limited or restricted to, a laptop, a netbook, a mobile phone, a tablet, a computer, security appliance, a mainframe, a server, a router, a bridge, a switch, or other intermediary communication device, etc.). Other examples of a network device includes a sensor (described above) as well as a computing node, namely hardware and/or software that operates to receive information from a sensor, and when applicable, perform malware analysis on that information.

According to one embodiment, the term “malware” may be construed broadly as any code or activity that initiates a malicious attack or any operations associated with anomalous or unwanted behavior. For instance, as an illustrative embodiment, malware may correspond to a type of malicious computer code that executes an exploit to take advantage of a vulnerability in software. The undesirable or anomalous behaviors may include a communication-based anomaly or an execution-based anomaly, which, for example, could (1) alter the functionality of a network device executing application software in an atypical manner (a file is opened by a first process where the file is configured to be opened by a second process and not the first process); (2) alter the functionality of the network device executing that application software without any malicious intent; and/or (3) provide unwanted functionality which may be generally acceptable in another context. In yet another alternative, malware may correspond to information that pertains to the unwanted behavior such as a process that causes data such as a contact list from a network device to be uploaded by a network to an external storage device without receiving permission from the user.

The term “transmission medium” is a communication path between two or more network devices. The communication path may include wired and/or wireless segments. Examples of wired and/or wireless segments include electrical wiring, optical fiber, cable, bus trace, or a wireless channel using infrared, radio frequency (RF), or any other wired/wireless signaling mechanism.

In general, a “virtual machine” (VM) is a software that is configured to provide an isolated run-time execution of an object in accordance with one or more selected applications supported by one or more operating systems. The run-time execution mimics the performance of a targeted device.

Lastly, the terms “or” and “and/or” as used herein are to be interpreted as inclusive or meaning any one or any combination. Therefore, “A, B or C” or “A, B and/or C” mean “any of the following: A; B; C; A and B; A and C; B and C; A, B and C.” An exception to this definition will occur only when a combination of elements, functions, steps or acts are in some way inherently mutually exclusive.

As this invention is susceptible to embodiments of many different forms, it is intended that the present disclosure is to be considered as an example of the principles of the invention and not intended to limit the invention to the specific embodiments shown and described.

III. GENERAL ARCHITECTURE

Referring to FIGS. 1A-1B, exemplary block diagrams of a malware detection system 100 is shown. The malware detection system 100 features one or more network devices 110₁-110_M (M≥1) that are communicatively coupled to a centralized system 120. As shown in FIG. 1A, according to one embodiment of the disclosure, the centralized system 120 may be a security appliance 130 that is configured to receive analysis results 140 from any of the network devices 110₁-110_M (e.g., network device 110₁). The analysis results 140 may include an object fingerprint associated with an object that, upon analysis by the network device 110₁, is determined, in accordance with a level of confidence (e.g., probability), to be associated with malware. Additionally, the analysis results 140 may further include a representation of the object under analysis (e.g., a hash result of the object, a pointer or link to the object, etc.) or the object itself. Of course, it is contemplated that the network device 110₁ may send the analysis results 140 directly to a management system of the centralized system 120 (e.g., cloud-based management system, remotely located management system from the network devices 110₁-110_M, etc.) in lieu of the security appliance 130.

Based on the analysis results 140 from the network device 110₁, the security appliance 130 generates a report 150 and/or alert 152 for transmission to a network administrator. The report 150 may include information from the analysis results 140, including a source of the object under analysis, information that identifies the probability of the object being malicious or benign, the know malicious family to which the object pertains (if malicious), or the like. The alert 152 may simply include signaling that initiating one or more messages (e.g., text message, email message, and/or automated phone call) to warn a network administrator of detection of a malicious object, namely an object having a level of confidence (e.g., probability) of being associated with a malicious attack (e.g., known vulnerability). The security appliance 130 may be further configured to conduct a secondary analysis of the object such as a static analysis of the object (e.g., bit pattern analysis, heuristic analysis, etc.) and/or a dynamic analysis (e.g., virtual machine based executing or processing of the object). Although not shown in detail, it is contemplated that a version of the report 150 and/or alert 152 may be transmitted from the network device 110₁ to a network administrator or a management system that provides consolidation of reports.

As further shown in FIG. 1A, the security appliance 130 may be implemented as a network device that is either coupled to the transmission medium 160 directly or communicatively coupled with the transmission medium 160 via an interface (not shown) operating as a data capturing device. For the later embodiment, the interface may operate as a network tap that extracts information from data traffic propagating over the transmission medium 160.

The one or more network devices 110₁-110_M, sometimes identified as “network device(s),” may be located on-site at an enterprise premises (e.g., part of the enterprise network infrastructure located at a single facility utilized by the customer or at a plurality of facilities). Alternatively, the network device(s) 110₁-110_M may be mobile devices that communicate with the centralized system 120 located on-site (e.g. at enterprise's premises) or located outside the enterprise network infrastructure, such as cloud-based services 170 of FIG. 1B. The cloud-based services 170 may be hosted by a cybersecurity provider or another entity separate from the customer (enterprise).

Each network device 110₁-110_M includes an agent 180 that conducts a similarity analysis of an object by determining a level of correlation between indicators associated with a fingerprint pertaining to an object under analysis and indicators associated with a fingerprint pertaining to known malicious and/or benign objects. This level of correlation may be partially or solely used for classification of the object as malicious, benign or suspicious (where not definitely malicious or benign). Herein this level of correlation may be determined by comparing an object fingerprint associated with the object under analysis, where each indicator of the object fingerprint may correspond to a representation of ordered data (e.g., a sequence of instructions, an ordered collection of raw “binary” data, etc.) to indicators of malware family fingerprints. Also, the ordered data for a representation pertains to ordered data within a sliding window (as described below) and different types of representations may include a hash result of the ordered data (e.g., SHA-256 result, MD5 result, SHA-512 result, etc.) or any logical operation on the ordered data that produces the result having a lesser size than the ordered data. The indicators of the object fingerprint are then compared to malicious indicators of one or more malware family fingerprints. The agent 180 may be configured as hardware or as software executed by hardware deployed within a corresponding network device 110₁, . . . , or 110_M, as described in FIG. 2.

Referring now to FIG. 2, an exemplary embodiment of a logical representation of the architecture of one of the network devices 110₂-110_M, such as network device 110₁ for example, is shown. The network device 110₁ includes a housing 200, which is made entirely or partially of a hardened material (e.g., hardened plastic, metal, glass, composite or any combination thereof) that protect circuitry within the housing 200. Examples of the circuitry may include one or more processors (hereinafter “processor(s)”) 210, communication interface logic 220, transmission medium 230 and internal storage 240.

As shown, communication interface logic 220 enables communications with other network devices to receive objects for analysis and to output results of the analysis. The communication interface logic 220 is further configured to receive updates to the agent 180. According to one embodiment of the disclosure, communication interface logic 220 may be implemented as a physical interface including one or more ports for wired connectors. Additionally, or in the alternative, communication interface logic 220 may be implemented with one or more radio units for supporting wireless communications with other wireless network devices.

Processor(s) 210 is coupled to the internal storage 240 via transmission medium 230. According to one embodiment of the disclosure, the internal storage 240 includes an object data store 250, a malware family fingerprint data store 252 that operates as cache memory for the malware family fingerprints, and the agent 180 including advanced malware detection logic 254. The advanced malware detection logic 254 includes (a) object identification logic 260, (b) frontend logic 270, and (c) backend logic 280. Although represented as software, the agent 180 or portions of the agent 180 may be configured as hardware, firmware or a collection thereof.

The object identification logic 260 is configured to categorize an object that is being analyzed by the advanced malware detection logic 254. According to one embodiment of the disclosure, the object identification logic 260 categorizes the object by at least determining the object type as different object types require different methods of disassembling and different sliding window sizes. Different object types may include, but are not limited or restricted to Android™ APKs, x86 Windows® binaries, flash files, or the like. Alternatively, or in combination with the object type analysis, the object identification logic 260 may operate to categorize (or further categorize) the object based on the type of operating system (OS) utilized by that object.

The frontend logic 270 is responsible for performing operations to generate an object control graph that is used in the computation of an object fingerprint for the object. According to one embodiment, the frontend logic 270 includes disassembly logic 272, instruction sequence aggregation logic 274 and control flow generation logic 276. The disassembly logic 272 is configured to detect whether the object is binary code, and if so, disassembles the binary code into instruction sequences. Depending on the category selected for the object by the object identification logic 260, the disassembly logic 272 may perform different disassembly operations on the object.

For instance, where the object is an Android™ APK, the disassembly logic 272 may operate as a utility (dexdump) where the output includes opcode (instructions) associated with the Android™ APK under analysis. As another example, where the object is an x86 Windows® binary, the disassembly logic 272 may operate the objdump utility where the output includes opcode (instructions) associated with the binary. As yet another example, where the object is flash file, the disassembly logic 272 may operate an ActionScript program to convert the flash file into opcode (instructions).

Thereafter, after translation of the object into a sequence of instructions, the instruction sequence aggregation logic 274 arranges the instructions into basic blocks. A “basic block” includes one or more instructions that are to be executed in sequence and the sequence completes without a change in control flow. The change in control flow may involve a transition of execution flow without completing an operation associated with the executed instruction (e.g., a Jump instruction, etc.). Hence, parsing through the sequence of instructions, the instruction sequence aggregation logic 274 may produce one or more basic blocks that may be operating sequentially (in serial) and/or in parallel.

Based on the analytic (basic block) information, the control flow generation logic 276 generates a nodal graph (referred to as an “object control graph”). The object control graph is a logical representation of relationships between objects and properties formulated from the analytic information. For instance, for illustrative purposes, each basic block may be logically represented as a node in the object control graph. The control flow may be represented by one or more links (or edges) between the nodes, where multiple links may logically represent multiple paths from an object or multiple paths to an object.

The backend logic 280 is responsible for generating an object fingerprint associated with the object, classifying the object fingerprint for determining whether the object fingerprint is a variant of a malware family fingerprint, and providing the object fingerprints to a separate “knowledge” server based on their classification. Optionally, although not shown, the backend logic 280 may include logic for generating reports that include information based on operations conducted by the classification logic 286. According to one embodiment, the backend logic 280 includes fingerprint generation logic 282, clustering logic 284 and classification logic 286.

The fingerprint generation logic 282 is responsible for analyzing the object control graph associated with the object and generating an object fingerprint based on such analysis. For example, the fingerprint generation logic 282 may be configured to parse an object control graph in accordance with a “moving” window analysis, where the window size may be fixed or may vary, perhaps based on the object type (e.g., window size for APK-type objects being different than window size for x86 Windows® object). For an architecture supporting variable-length sliding windows, metadata for use in determining the sliding window size would need to be provided to the fingerprint generation logic 282. For example, metadata associated with the object type may be provided from the disassembly logic 272 of the frontend logic 270, where the metadata may be used by the fingerprint generation logic 282.

Thereafter, for each subset of instructions or ordered binary data within a basic block of an object control graph that fills an entire sliding window, a representation of the subset of instructions or the ordered binary data is generated (e.g., hash result, result of a lesser bit size that the subset of instructions or ordered binary data). The representation is compared to values within one or more entries of a feature vector, where each feature may correspond to a representation of a predetermined sequence of instructions or a collection of ordered data that is chosen to be monitored. In response to a successful comparison of a representation produced from the similarity analysis to a representation corresponding to a feature of the feature vector, that feature is tagged or set. This operation is performed in an iterative manner until parsing of the object control graph in accordance with the sliding window analysis has completed. Upon completing the parsing of the object control graph, the feature vector corresponds to an object fingerprint for the object under analysis.

The classification logic 286 is responsible for determining whether the object under analysis is potentially malicious or benign, which may be handled by comparing the object fingerprint to one or more malware family fingerprints, some of which may be stored locally within the malware family fingerprint data store 252. In particular, the classification logic 286 conducts (or at least controls) a comparison between indicators of the object fingerprint and corresponding indicators of the malware family fingerprints. This comparison may involve determining whether a number of indicators of the object fingerprint that match a malware family fingerprint exceeds a prescribed threshold value. Alternatively, the comparison may involve a weighted determination in which a matching of some indicators of the object fingerprint to indicators of a malware family fingerprint denotes a comparison.

The clustering logic 284 is responsible for directing the object fingerprint, when determined to be malicious, to a knowledge database or other server that is configured to maintain fingerprints associated with detected malicious object fingerprints.

IV. FINGERPRINT DETECTION METHODOLOGY—BINARY CODE

Referring to FIG. 3, a first exemplary embodiment of operations conducted by the advanced malware detection logic deployed within a network device of FIG. 1A or FIG. 1B in generating an object fingerprint is shown. Herein, in response to receipt of an incoming object (e.g., intercepted during transit over a network, uploaded from a storage device, etc.), the advanced malware detection logic determines the object type (element 300). This determination may include an analysis of the content of the object and/or the format of the object. For instance, as an illustrative example, the object type may be determined by review of a header of the object and magic numbers which are constant numerical or text values used to identify a file format or protocol.

When the object is determined to be a first object type (e.g., Android™ APK), the advanced malware detection logic conducts a disassembly operation on the object to recover instructions forming at least part of the object, and thereafter, determines one or more sequences of instructions each pertaining to a particular basic block of an object control graph (elements 310 and 320). Sequential instructions are placed in the same basic block when there is no change in control flow between these instructions. However, in response to a change in control flow (e.g., presence of a conditional instruction, Jump instructions, Goto instruction, etc.), a new basic block is generated in the object control graph for instructions associated with the control flow change. The grouping of instructions within basic blocks, which are based on changes in control flow, is continued until all of the dissembled instructions have been represented within the object control graph.

Similarly, when the object is determined to be either a second object type (e.g., x86 Windows® binary) or a third object type (e.g., Flash file), the advanced malware detection logic conducts a disassembly operation on the object to recover the instructions forming at least part of the object, and thereafter, determines sequences of instructions that pertain to a particular basic block as described above (elements 312 & 322 and/or elements 314 & 324).

The basic blocks form nodes of the object control graph with edges (links) that are illustrative of the control flow, as shown in FIG. 4A. The content of the object, which may be represented by the object control graph, is analyzed to compute an object fingerprint (element 330), which undergoes a classification operation to determine whether the object fingerprint compares to any malware family fingerprint, namely the level of correlation between the object fingerprint and a malware family fingerprint exceeds a threshold (element 340). If so, the object fingerprint that is determined to be malicious may be uploaded to a remote destination (e.g., remote data server such as security appliance 130 of FIG. 1) for subsequent analysis in clustering this fingerprint to other similar malicious fingerprints (e.g., malware family fingerprints) stored in a knowledge database (element 350).

As shown in FIG. 4A, an illustrative embodiment of an object control graph 400 includes a plurality of basic blocks 410 (e.g., six basic blocks “Bx” 410₁-410₆) and edges 420₁-420₇ illustrative of the control path between basic blocks 410₁-410₆. As shown, a first basic block (B1) 410₁ includes four instructions, a third basic block (B3) 410₃ includes three instructions and the remainder of the basic blocks B2410₂ and B4410₄-B6410₆ include a single instruction. Upon setting of the sliding window to a fixed size of “n” instructions (e.g., n=3 instructions for this embodiment), the moving window analysis would detect three (3) instruction sequences as potential indicators for an object fingerprint of the object, as shown in FIG. 4B. According to this embodiment, the instruction sequences less than the window size (e.g., 3 instructions) in the sequence are ignored in order to avoid a setting of high number of indicators in the feature vector that may cause unacceptable levels of false positives in malware analysis.

In particular, in accordance with the sliding window-based analysis, a sliding window would capture a first sequence of instructions 450 within basic block 410₁, namely instructions Inst_1, Inst_2 and Inst_3. Thereafter, the advanced malware detection logic would conduct an operation on the first sequence of instructions 450, such as a one-way hash operation for example, to generate a representation 455 of the first sequence of instructions 450 that operates and determine whether the representation 455 compares with any features 465 for the feature vector 460, where the features 465 represent instruction sequences being monitored. If so, an indicator 470₁ corresponding to the feature 4651 within the feature vector 460 is set.

As further shown in FIG. 4B, in accordance with the moving window analysis, a sliding window would capture a second sequence of instructions 475 within basic block 410₁, namely Inst_2, Inst_3 and Inst_4. Thereafter, the advanced malware detection logic would conduct an operation on the second sequence of instructions 475 to generate a representation 480 of the second sequence of instructions 475. When the representation 480 compares to a different feature 4652 of the feature vector 460, an indicator 470₂ associated with the feature is set.

Thereafter, a sliding window would capture a third sequence of instructions 485 within the third basic block 410₃, namely Inst_5, Inst_6 and Conditional_2 in third basic block 410₃. Thereafter, the advanced malware detection logic would conduct an operation on the third sequence of instructions 485 to generate a representation 490 of the third sequence of instructions 485. When the representation 490 compares to yet another different feature 465₃ of the feature vector 460, the indicator 470₃ of feature 465₃ is set. For the illustrative example, as there are no further basic blocks having a size equal to or greater than the window size, no further indicators may be set.

Referring now to FIG. 5, is a first flowchart illustrating an exemplary method for generating an object fingerprint is described below. Initially, a window size may be selected in order to control how sequences of instructions that pertain to one or more basic blocks within the object control graph are parsed (element 500). The window size may influence the number of false positives (smaller windows) or number of false negatives (larger windows) encountered during operability of the malware detection system.

According to one embodiment of the disclosure, the window size may be set to a static, fixed value (e.g., a number of instructions, a number of bytes, etc.). The fixed value for the window size may be in accordance with a default setting that is established by the original equipment manufacturer/software provider. Alternatively, the fixed value for the window size may be selected by a user of the network device or a network administrator during registration of the network device for use on an enterprise network.

According to another embodiment of the disclosure, the window size may be varied based, at least in part, on the expressiveness of the instruction set servicing the particular type of object under analysis. For instance, the variable window size for an Android® APK object is substantially smaller than the variable window size for an x86 Windows® binary because a Delvik® instruction used in an Android™ device is more expressive (i.e., perform more operations) than an x86 instruction. Additionally, or in the alternative, the window size may be varied based on the content itself (e.g., instruction boundaries). For instance, some instruction sets have variable-length instructions, so sliding windows typically represent different number of bytes dependent on the instructions utilized for a specific object type.

Referring still to FIG. 5, upon selecting the window size, a first sequence of instructions associated with the object under analysis is obtained (elements 510 and 515). Based on a moving window analysis and the number of instructions in the first sequence is equal to or exceeds the window size, a representation (e.g., hash result) of a first subset of instructions is obtained (element 520). If the representation of the first subset of instructions compares with any indicators of the feature vector, the indicator is set to denote that the object includes the corresponding instruction sequence (elements 530-535).

Next, for the same basic block, a determination is made as to whether all subsets of the instructions within the basic block have been analyzed, where each neighboring subset of instructions shares at least one instruction (or multiple instructions) with another neighboring subset of instructions (element 540). If all subsets have not been analyzed, the sliding window is adapted to “slide” by a determined number of instructions (e.g., one instruction, two instructions, etc.) to capture a next subset of instructions that are part of the sequence of instructions within the analysis window (element 545). However, if all subsets of the instructions within the basic block have been analyzed, a determination is made whether all sequences in different basic blocks have been analyzed (element 550). If not, the next sequence of instructions within a different basic block is determined and analyzed, as described in elements 515-555. Otherwise, the object fingerprint has been generated for the object (element 560).

V. FINGERPRINT DETECTION METHODOLOGY—RAW DATA

Referring now to FIG. 6, an exemplary embodiment of an operational flow of the malware detection logic deployed within a network device of FIG. 1A or FIG. 1B in generating the object fingerprint based on analysis of raw data is shown. In response to receipt of an incoming object, the advanced malware detection logic determines the object type (element 600). This determination may include an analysis of the content of the object and/or the format of the object. Where the content is a collection of raw “binary” data without identification of the object type, the advanced malware detection logic places the binary data in order of receipt into a single basic block (elements 610 and 620).

A sliding window (e.g., a first number “X” of bytes that define a boundary for the analysis) along with a window step size (e.g., a second number “Y” of bytes that is less than or equal to the first number of bytes) are selected (elements 630 and 640). Herein, a representation of a grouping of bytes (e.g., first X bytes of the ordered binary data) is determined, and thereafter, a representation of a second grouping of bytes (e.g., X+1^st byte to 2X byte) is determination. The analysis creates representations of the X-byte subsets of binary data, which are used in the computation of the object fingerprint by setting features in the feature vector that correspond to the representations for these X-byte subsets (elements 650 and 660).

VI. ADVANCED MALWARE DETECTION

Referring to FIG. 7, an exemplary embodiment of an operational flow of the advanced malware detection logic deployed within a network device of FIG. 1A or FIG. 1B is shown, where the operational flow is directed to a determination as to whether an object under analysis is advanced malware. This determination may involve fingerprint comparisons. More specifically, the advanced malware detection logic has generated an object fingerprint as described above. The advanced malware detection logic compares the object fingerprint to fingerprints that are associated with a white list of benign fingerprints (element 700). If the level of correlation between the object fingerprint and any of the benign fingerprints exceeds a threshold, the object is benign. Hence, no further classification operations are needed, but the object fingerprint may be provided to a remote source for further analysis and placement in the knowledge database (element 710).

Otherwise, the advanced malware detection logic is provided access to malware family fingerprints (element 720). These malware family fingerprints may be stored locally in the malware family fingerprint data store 252 of FIG. 2 or accessed from a remote source. With access to content of the malware family fingerprints, the advanced malware detection logic, and in particular the classification logic 286 of FIG. 2, conducts a comparison between indicators of the object fingerprint and corresponding indicators for one or more of the malware family fingerprints (element 730). This comparison may involve determining whether the number of indicators of the object fingerprint that match indicators of a particular malware family fingerprint exceeds a prescribed threshold value. Alternatively, the comparison may involve a weighted determination in which a matching of indicators of the object fingerprint with certain indicators of a malware family fingerprint denotes a comparison.

If the level of correlation between the object fingerprint and any of the malware family fingerprints exceeds a threshold, the object represented by the object fingerprint is determined to be associated with a malicious attack (element 750). The object fingerprint and metadata associated with the object fingerprint (e.g., compared malware family, source, etc.) is further provided to the remote source for placement into the knowledge database, which may be used as one of the malware family fingerprints for subsequent analysis (element 760). Otherwise, the object fingerprint may undergo additional analysis at the network device (or remote source) as represented by dashed lines 770.

In the foregoing description, the invention is described with reference to specific exemplary embodiments thereof. It will, however, be evident that various modifications and changes may be made thereto without departing from the broader spirit and scope of the invention as set forth in the appended claims.

Claims

1. A non-transitory storage medium having stored thereon instructions corresponding to malware detection logic including as part of a software agent operating within a network device, the malware detection logic being executable by one or more processors to perform operations comprising: receiving an object for analysis;extracting content from the object, the content includes a first plurality of instructions recovered from binary code of the object;arranging the extracted content into one or more basic blocks, each basic block of the one or more basic blocks including at least an instruction sequence corresponding to two or more instructions of the first plurality of instructions;generating, by the network device, an object fingerprint based on an analysis of the one or more basic blocks by at least (i) generating a representation of each instruction sequence of the one or more basic blocks, (ii) aggregating one or more representations associated with each instruction sequence of the one or more basic blocks, and (iii) comparing each of the one or more aggregated representations to a plurality of instruction sequences being monitored as being potentially malicious to produce a result, the result corresponding to the object fingerprint;analyzing, by the network device, the object fingerprint by at least comparing the object fingerprint to one or more malware family fingerprints to determine whether the object is potentially malicious and associated with an advanced malware; andgenerating information, based on the analyzing of the object fingerprint, for transmission from the network device to a second network device.

2. The non-transitory storage medium of claim 1, wherein the generating of the information for transmission to the second network device includes comparing of the object fingerprint to the one or more malware family fingerprints and the second network devices corresponds to a security appliance that is configured to conduct a secondary malware analysis of the object.

3. The non-transitory storage medium of claim 1, wherein each representation being a hash result generated from a corresponding instruction sequence.

4. The non-transitory storage medium of claim 1, wherein the arranging the extracted content comprises arranging the first plurality of instructions in accordance with an operational flow of the object, the operational flow is determined from an analysis of an order of execution of the first plurality of instructions without execution of the first plurality of instructions; andsegmenting the first plurality of instructions arranged in accordance with the operational flow into the one or more basic blocks,wherein each instruction for each basic block of the one or more basic blocks completes operation without a change in control flow.

5. The non-transitory storage medium of claim 1, wherein the arranging the extracted content comprises generating an object control graph including the one or more basic blocks.

6. The non-transitory storage medium of claim 5, wherein the generating of the object control graph comprises arranging the first plurality of instructions in accordance with a determined order that is based, at least in part, on an order of execution of the first plurality of instructions without execution of the first plurality of instructions; andsegmenting the first plurality of instructions arranged in accordance with the determined order into the one or more basic blocks,wherein each instruction for each instruction sequence of the one or more basic blocks completes operation without a change in control flow.

7. The non-transitory storage medium of claim 3, wherein each instruction of the instruction sequence completes operation without a change in control flow.

8. The non-transitory storage medium of claim 1, wherein the generating of the representation of each instruction sequence of the one or more basic blocks comprises generating a first representation of a first instruction sequence that includes a selected number of instructions, the first instruction sequence included as a first block of the one or more basic blocks, the first representation corresponds to a hash result produced by performing a hash operation on the first instruction sequence.

9. The non-transitory storage medium of claim 8, wherein the generating of the representation of each instruction sequence of the one or more basic blocks further comprises generating a first representation of a first subset of a second instruction sequence of a second basic block of the one or more basic blocks; andgenerating a second representation of a second subset of the second instruction sequence,wherein the first subset of the second instruction sequence differs from the second subset of the second instruction sequence and the first subset of the second instruction sequence sharing at least one instruction with the second subset of the second instruction sequence.

10. The non-transitory storage medium of claim 9, wherein the first subset of the second instruction sequence and the second subset of the second instruction sequence are selected through a sliding analysis window having a width corresponding to the selected number of instructions.

11. The non-transitory storage medium of claim 1, wherein the generating of the representation of each instruction sequence of the one or more basic blocks comprises generating a first representation of a first subset of a first instruction sequence of a first basic block of the one or more basic blocks; andgenerating a second representation of a second subset of the first instruction sequence,wherein the first subset of the first instruction sequence differs from the second subset of the first instruction sequence and the first subset of the first instruction sequence sharing at least one instruction with the second subset of the first instruction sequence.

12. The non-transitory storage medium of claim 11, wherein the first subset of the first instruction sequence and the second subset of the first instruction sequence are selected through a sliding analysis window having a width corresponding to a selected number of instructions being lesser in number than a number of instructions forming the first basic block.

13. The non-transitory storage medium of claim 1, wherein prior the generating of the object fingerprint, selecting a width of an analysis window to a selected number of instructions, the analysis window being used in the generating of the representation of each instruction sequence of the one or more basic blocks that includes at least the selected number of instructions.

14. The non-transitory storage medium of claim 1, wherein the information for transmission includes a report based on the analyzing of the object fingerprint, the report includes one or more of (i) a source of the object, (ii) a probability of the object being malicious or benign, or (iii) a known malicious family to which the object pertains.

15. The non-transitory storage medium of claim 1, wherein the information for transmission comprises includes an alert including one or more messages to warn of detection of the obj ect as a malicious object.

16. A computerized method for detection of malware, comprising: receiving an object for analysis;extracting content from the object, the content includes a first plurality of instructions recovered from non-binary code of the object;arranging the extracted content into one or more basic blocks, each basic block including at least an instruction sequence corresponding to two or more instructions of the first plurality of instructions;generating an object fingerprint based on an analysis of the one or more basic blocks by malware detection logic, operating within a network device, at least (i) generating a representation of each instruction sequence of the one or more basic blocks, (ii) aggregating one or more representations associated with each instruction sequence of the one or more basic blocks, and (iii) comparing each of the one or more aggregated representations to a plurality of instruction sequences being monitored as being potentially malicious to produce a result, the result corresponding to the object fingerprint;analyzing the object fingerprint by at least comparing the object fingerprint to one or more malware family fingerprints to determine whether the object is potentially malicious and associated with an advanced malware; andgenerating information, based on the analyzing of the object fingerprint, for transmission from the network device to a second network device.

17. The method of claim 16, wherein the generating of the information for transmission includes the comparing of the object fingerprint to the one or more malware family fingerprints, for transmission to the network device being a security appliance that is configured to conduct a secondary malware analysis of the object.

18. The method of claim 16, wherein the content includes the first plurality of instructions disassembled from the binary code of the object.

19. The method of claim 18, wherein the arranging the extracted content comprises arranging the first plurality of instructions in accordance with an operational flow of the object, the operational flow is determined from an analysis of an order of execution of the first plurality of instructions without execution of the first plurality of instructions; andsegmenting the first plurality of instructions arranged in accordance with the operational flow into the one or more basic blocks,wherein each instruction for each basic block of the one or more basic blocks completes operation without a change in control flow.

20. The method of claim 18, wherein the arranging the extracted content comprises generating an object control graph including the one or more basic blocks.

21. The method of claim 20, wherein the generating of the object control graph comprises arranging the first plurality of instructions in accordance with a determined order that is based, at least in part, on an order of execution of the first plurality of instructions without execution of the first plurality of instructions; andsegmenting the first plurality of instructions arranged in accordance with the determined order into the one or more basic blocks, each of the one or more basic blocks includes one or more instructions of the first plurality of instructions that correspond to an instruction sequence and each instruction of the instruction sequence completes operation without a change in control flow.

22. The method of claim 18, wherein each instruction of the instruction sequence completes operation without a change in control flow.

23. The method of claim 16, wherein the generating of the representation of each instruction sequence of the one or more basic blocks comprises generating a first representation of a first instruction sequence that includes a selected number of instructions, the first representation corresponds to a hash result produced by performing a hash operation on the first instruction sequence.

24. The method of claim 23, wherein the generating of the representation of each instruction sequence of the one or more basic blocks that includes at least the selected number of instructions further comprises generating a first representation of a first subset of a second instruction sequence of a second basic block of the one or more basic blocks; andgenerating a second representation of a second subset of the second instruction sequence,wherein the first subset of the second instruction sequence differs from the second subset of the second instruction sequence and the first subset of the second instruction sequence sharing at least one instruction with the second subset of the second instruction sequence.

25. The method of claim 24, wherein the first subset of the second instruction sequence and the second subset of the second instruction sequence are selected through a sliding analysis window having a width corresponding to the selected number of instructions.

26. The method of claim 16, wherein the generating of the representation of each instruction sequence of the one or more basic comprises generating a first representation of a first subset of a first instruction sequence of a first basic block of the one or more basic blocks, wherein the first instruction sequence includes a selected number of instructions and the first basic block includes a number of instructions greater than the selected number of instructions; andgenerating a second representation of a second subset of the first instruction sequence,wherein the first subset of the first instruction sequence differs from the second subset of the first instruction sequence and the first subset of the first instruction sequence sharing at least one instruction with the second subset of the first instruction sequence.

27. The method of claim 26, wherein the first subset of the first instruction sequence and the second subset of the first instruction sequence are selected through a sliding analysis window having a width corresponding to the selected number of instructions.

28. The method of claim 16, wherein prior the generating of the object fingerprint, selecting a width of an analysis window to a selected number of instructions, the analysis window being used in the generating of the representation of each instruction sequence of the one or more basic blocks that includes at least the selected number of instructions.

29. The method of claim 16, wherein the information for transmission includes a report based on the analyzing of the object fingerprint, the report includes one or more of (i) a source of the object, (ii) a probability of the object being malicious or benign, or (iii) a known malicious family to which the object pertains.

30. The method of claim 16, wherein the information for transmission comprises includes an alert including one or more messages to warn of detection of the object as a malicious object.

31. A network device, comprising: a transmission medium;one or more hardware processors coupled to the transmission medium;communication interface logic coupled to the transmission medium, the communication interface logic to receive an object for analysis; anda memory coupled to the transmission medium, the memory comprises disassembly logic that, when executed by the one or more hardware processors, extracts content from the object, the content includes a first plurality of instructions recovered from binary code of the object,aggregation logic that, when executed by the one or more hardware processors, arranges the extracted content into one or more basic blocks, each basic block including at least an instruction sequence corresponding to two or more instructions of the first plurality of instructions, andfingerprint generation logic that, when executed by the one or more hardware processors, generates an object fingerprint based on an analysis of the one or more basic blocks by at least (i) generating a representation of each instruction sequence of the one or more basic blocks, (ii) aggregating one or more representations associated with each instruction sequence of the one or more basic blocks, and (iii) comparing each of the one or more aggregated representations to a plurality of instruction sequences being monitored as being potentially malicious to produce a result, the result corresponding to the object fingerprint;classification logic that, when executed by the one or more hardware processors, analyzes the object fingerprint by at least comparing the object fingerprint to one or more malware family fingerprints to determine whether the object is potentially malicious; andlogic that, when executed by the one or more hardware processors, generates information resulting from analysis of the object fingerprint by the classification logic, for transmission from the network device to a second network device.

32. The network device of claim 31, wherein the information for transmission includes a report based on the analyzing of the object fingerprint, the report includes one or more of (i) a source of the object, (ii) a probability of the object being malicious or benign, or (iii) a known malicious family to which the object pertains.

33. The network device of claim 31, wherein the information for transmission comprises includes an alert including one or more messages to warn of detection of the object as a malicious object.

34. The network device of claim 31, wherein the aggregation logic arranges the extracted content by at least arranging the first plurality of instructions in accordance with an operational flow of the object, the operational flow is determined from an analysis of an order of execution of the first plurality of instructions without execution of the first plurality of instructions; andsegmenting the first plurality of instructions arranged in accordance with the operational flow into the one or more basic blocks,wherein each instruction for each basic block of the one or more basic blocks completes operation without a change in control flow.

35. The network device of claim 31, wherein the aggregation logic arranges the extracted content to at least generate an object control graph including the one or more basic blocks.

36. The network device of claim 35, wherein the aggregation logic to generate the object control graph by at least arranging the first plurality of instructions in accordance with a determined order that is based, at least in part, on an order of execution of the first plurality of instructions without execution of the first plurality of instructions; andsegmenting the first plurality of instructions arranged in accordance with the determined order into the one or more basic blocks,wherein instruction for each instruction sequence of the one or more basic blocks completes operation without a change in control flow.

37. The network device of claim 31, wherein each instruction of the instruction sequence completes operation without a change in control flow.

38. The network device of claim 31, wherein the fingerprint generation logic to generate the representation of each instruction sequence of the one or more basic blocks by at least generating a first representation of a first instruction sequence that includes a selected number of instructions, the first instruction sequence included as a first block of the one or more basic blocks, the first representation corresponds to a hash result produced by performing a hash operation on the first instruction sequence.

39. The network device of claim 31, wherein the fingerprint generation logic to further generate the representation of each instruction sequence of the one or more basic blocks by at least generating a first representation of a first subset of a second instruction sequence of a second basic block of the one or more basic blocks; andgenerating a second representation of a second subset of the second instruction sequence,wherein the first subset of the second instruction sequence differs from the second subset of the second instruction sequence and the first subset of the second instruction sequence sharing at least one instruction with the second subset of the second instruction sequence.

40. The network device of claim 39, wherein the first subset of the second instruction sequence and the second subset of the second instruction sequence are selected through a sliding analysis window having a width corresponding to the selected number of instructions.