Method for detecting malicious code patterns in consideration of control and data flows

[0001] 1. Field of the Invention

[0002] The present invention relates to a method for detecting malicious code patterns present in malicious scripts, and more particularly, to a method for detecting malicious code patterns by using a static analysis in consideration of control and data flows.

[0003] 2. Description of the Prior Art

[0004] Generally, as for detection of malicious scripts, techniques for binary codes are used directly or some modifications can be made thereto to be adapted to scripts in the form of source programs. Particularly, signature recognition through scanning is a malicious-code detection scheme that is being most commonly used. This scheme has an advantage in that a diagnosis speed is high and the kind of malicious code can be clearly identified since it determines whether a concerned code is a malicious code by searching for a special character string present in only one malicious code. However, this scheme has a problem in that it cannot cope with unknown malicious codes at all.

[0005] Meanwhile, it is a heuristic analytical technique that is considered the most practical one among techniques for detecting unknown malicious scripts. This technique is a scheme for detecting malicious codes by organizing code segments frequently used as malicious codes into a database and scanning target codes to determine whether the target codes are present or how many times the target codes appear. Although this scheme has advantages of a relatively high speed and a high detection ratio, it has a disadvantage of a somewhat high possibility of the occurrence of a false positive error. Accordingly, in order to alleviate such a disadvantage, there has been proposed a method for detecting malicious scripts using a static analysis. Since this method checks not only the presence of method sequences but also associated parameters and return values, it represents considerably precise detection results as compared with a method using a simple heuristic analysis.

[0006] FIG. 1 shows an example of malicious visual basic script codes, explaining the concept of a method for detecting malicious scripts using a static analysis. As can be seen in FIG. 1, in order that a plurality of method calls constitutes one malicious behavior, a special relationship between their parameters and return values is inevitably required. For example, Copy method at the fourth line copies a script under execution to make a file with the name of LOVE-LETTER-FOR-YOU.TXT.VBS and Attachments.Add method at the seventh line attaches the file to a newly created mail object to achieve self-replication through a mail. However, in a case where a scheme for checking only the presence of method calls is used, even though there is an unrelated method call for creating a script file named A and attaching a file named B to the created script file, the method call is regarded as a malicious code, which exhibits a high false positive error rate. On the contrary, the detection method using the static analysis can obtain more precise detection results than methods using a simple search of character strings, by checking not only whether method calls are present but also whether all relevant values such as used file names, for example, fso, c, out, male or the like, are found.

[0007] However, the detection method using the static analysis has still a problem in view of detection accuracy. Conventional detection methods using the static analysis compare only revealed names of variables with each other. Therefore, there may be an error that only for the reasons that given two variables have the same name, the values of the two variables are regarded as the same even during execution. FIG. 2 shows an example in which a false positive error may occur in the detection method using the static analysis. In the conventional detection methods using the static analysis, it is only confirmed that variables c used at the first and fourth lines are same, and the values of the two variables are regarded as being identical to each other. However, when the program is analyzed, it can be seen that since variable c is newly defined at the third line, variables c at the first and fourth lines have different values upon actual execution, respectively. Contrary to FIG. 2, FIG. 3 shows an example in which a false negative error may occur in the detection methods using the static analysis. In the conventional methods using the static analysis, since variable c at the first line and variable d at the third line are different variables, it is determined that the values of the two variables are not same. However, the values of the two variables become identical to each other due to a replication statement dc at the second row upon actual execution. Consequently, in view of the detection of entire malicious behavior patterns, the two types of errors mentioned above induce the false positive and negative errors, respectively. Therefore, there is a need for a method for solving these errors.

[0008] Accordingly, the present invention is conceived to solve the problems in the prior art. An object of the present invention is to provide a method for detecting malicious code patterns, which can improve detection accuracy by using a static analysis in consideration of control and data flows.

[0009] According to the present invention for achieving the object, there is provided a method for detecting malicious code patterns in consideration of control and data flows, wherein a malicious code pattern is detected by determining whether values of tokens (variables or constants) included in two sentences to be examined will be identical to each other during execution of the sentences, and the determination on whether the values of the tokens will be identical to each other during the execution is made through classification into four cases: a case where both tokens in two sentences are constants, a case where one of tokens of two sentences is a constant and the other token is a variable, a case where both tokens of two sentences are variables and have the same name and range, and a case where both tokens of two sentences are variables but do not have the same name and range.

[0010] Preferably, the determination on whether the values of tokens will be identical during the execution of the sentences is made based on: if both the tokens in the two sentences are constants, whether relevant token character strings are identical to each other; if one of the tokens of the two sentences is a constant and the other token is a variable, whether the relevant token character strings are identical to each other after the variable is substituted for a constant; if both the tokens of the two sentences are variables and have the same name and range, whether there are definitions of the relevant variables in a control flow from a preceding one of the two sentences to a following one thereof; and if both the tokens of two sentences are variables but do not have the same name and range, whether there are definitions of the relevant variables in a control flow from a preceding one of the two sentences to a following one thereof after the relevant variables are substituted for original variables.

[0011] The above and other objects, features and advantages of the present invention will become apparent from the following detailed description of a preferred embodiment given in conjunction with the accompanying drawings, in which:

[0012] FIG. 1 shows an example of visual basic script codes performing self-replication through a mail, explaining the concept of a method for detecting malicious scripts using a static analysis;

[0013] FIG. 2 shows an example of a false positive error occurred in the method for detecting the malicious scripts using the static analysis;

[0014] FIG. 3 shows an example of a false negative error occurred in the method for detecting the malicious scripts using the static analysis;

[0015] FIG. 4 is a conceptual view of classification of sentences to be examined for detecting malicious code patterns according to the present invention;

[0016] FIG. 5 shows an example of a control flow graph according to present invention; and

[0017] FIG. 6 is a flowchart illustrating detection of the malicious code patterns according to the present invention.

[0018] Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

[0019] FIG. 4 is a conceptual view of classification of sentences to be examined for detecting malicious code patterns according to the present invention, wherein the malicious code patterns are detected by determining whether the values of tokens (variables or constants) included in two sentences to be examined will be identical to each other during execution of the sentences. In other words, the malicious code patterns can be detected by determining, based on codes before execution, whether a variable or constant Ti included in sentence Si and a variable or constant Tj included in sentence Sj have the same value at the time of execution.

[0020] As shown in FIG. 4, whether two variables or constants present in a script have the same value can be examined according to the following four cases: a case where both tokens in two sentences are constants (hereinafter, referred to as Type 1), a case where one of tokens of two sentences is a constant and the other token is a variable (hereinafter, referred to as Type 2), a case where both tokens of two sentences are variables and have the same name and range (hereinafter, referred to as Type 3), and a case where both tokens of two sentences are variables but do not have the same name and range (hereinafter, referred to as Type 4).

[0021] Considering the respective types, Type 1 determines whether the values of tokens will be identical to each other during execution based on whether relevant token character strings are identical to each other. Type 2 cannot know, through a code analysis before execution, whether the values of tokens will be identical to each other at the time of execution. Accordingly, when a relevant variable is substituted for a constant by performing constant propagation, it is determined whether the values of tokens will be identical to each other at the time of execution based on whether relevant token character strings are identical to each other.

[0022] Type 3 determines whether there are definitions of relevant variables in a control flow from a preceding one of the two sentences to a following one thereof. Here, if there is no variable definition, it is determined that both the variables always have the same value. Type 4 determines whether there are definitions of relevant variables in a control flow from a preceding one of the two sentences to a following one thereof by performing copy propagation for substituting original variables for relevant variables. T this time, if there is no variable definition, it is determined that both the variables always have the same value.

[0023] Meanwhile, the definition of a variable means a sentence for substituting a value for a relevant variable as known in program language theories. The constant propagation, which is a technique widely used for a data flow analysis, aims at finding a variable or formula that will always have a specific constant value upon execution of a program, and propagating these the constant value toward program codes as many as possible. Similarly, the copy propagation refers to a technique for reducing the number of copies by finding a variable, which will always have a specific constant value upon execution of a program, and performing substitution through a copy sentence in the form of xy. That is, if there is a copy sentence s in the form of xy, the following conditions should be satisfied to substitute y for a use u of x and remove the copy sentence s:

[0024] i) The copy sentence s should be a unique definition of x that reaches u.

[0025] ii) The definition of y should not be present for all paths from s to u.

[0026] For example, assume the following program codes:

[0027] xy

[0028] zfso.getfile(x)

[0029] The second sentence can be changed to zfso.getfile(y). If such a change is a unique use of variable x, the first sentence xy can be deleted. These can be achieved by a repetitive algorithm for obtaining a use and a definition related to variable copy for each basic block and propagating information along a control flow graph. The basic block means a series of sentences having a single entrance point and a single exit point on a control flow and becomes a node of the control flow graph. The presence of a control flow from one basic block to another basic block is expressed as an edge of the control flow graph. FIG. 5 shows an example of such a control flow graph. When a control flow graph is given, a set for use in calculation of effective copy for reaching one basic block is defined as follows:

[0030] inBEffective copy up to a preceding node of basic block B,

[0031] outBEffective copy immediately after execution of basic block B,

[0032] c_genBCopy defined in basic block B, and

[0033] c_killBCopy invalidated due to new definition in basic block B.

[0034] At this time, inB and outB can be calculated based on the following formulas: $\begin{array}{cc}\begin{array}{cc}\mathrm{in}\left[B\right]=\begin{array}{c}\\ P_{a\mathrm{predecessor}\mathrm{of}B}\end{array}\mathrm{out}\left[P\right],& \text{:}B\end{array}& \left(1\right)\end{array}$ inB₁,(2)

[0035] After the constant propagation and copy propagation have been performed, it can be seen that the values of relevant variables will be identical to each other. In other words, both of two tokens have the same constant or the same variable name and range, and definitions of the relevant variables should not be present between two sentences in which the tokens are placed.

[0036] Meanwhile, when token Ti included in sentence Si and token Tj included in sentence Sj are the same variable or constant V, an algorithm for checking them using a control flow graph G obtained from a script is as follows. That is, in a first step, determination is made so that W:Si, IsEqual:true and IsLinked:false, where W is a work list, IsEqual is a flag informs that two values are same, and IsLinked is a flag informs that there is a control flow between two nodes. In a second step, a node is taken out from W and then designated as c. If there is no node to be taken out, the procedure proceeds to a fifth step.

[0037] Subsequently, in a third step, if V is a variable, c is a definition of variable V and a path to Sj is present, IsEqual:false. Otherwise, if c is sentence Sj, IsLinked:true. If c is not a program termination node or a node that has been already visited, all linked nodes are put into W. A fourth step proceeds to the second step in order to consecutively perform checking. Finally, in the fifth step, it is determined that the variables or constants V in Si and Sj have the same value if slinkedtrue and IsEqualtrue.

[0038] In the third step of the algorithm described above, the fact that c is the definition of variable V means one of the following cases:

[0039] i) c is an assignment statement and variable V is in the left side, and

[0040] ii) c is a sentence for calling a function or procedure and variable V belongs to set MAY_DEF(p) of relevant function call p.

[0041] Here, set MAY_DEF(P) is a set of global variables and reference parameters, which can be defined in procedure p. This set can be obtained as a union of global variables defined in each procedure, actual parameters given as a reference form upon calling and set MAY_DEF called by a relevant procedure. Therefore, this set can be calculated simply. In other words, most of calculation can be performed in advance base on only global variables and a call graph defined in each procedure. Thus, when a procedure call appears during a search of a path between two sentences, only set MAY_DEF is referred to without a search of details of a relevant procedure, thereby greatly shortening analysis time.

[0042] Consequently, a procedure shown in FIG. 6 is required to perform the checking described above. Referring to the figure, a given script is read out to create a control flow graph (S610). Subsequently, the copy/constant propagation is performed through a data flow analysis in the created control flow graph (S620). At this time, the results of the copy/constant propagation become a modified script or a new control flow graph corresponding to the modified script. Finally, malicious code pattern detection is performed (S630). At this time, when it is determined whether the values of two tokens present in two sentences will be identical to each other during execution, the aforementioned algorithm is used by utilizing the control flow graph obtained in step S620.

[0043] According to the method for detecting malicious code patterns in consideration of control and data flows described above, it is possible to exclude a false positive error that may occur in conventional comparison of variable names and to lower a false negative error rate, thereby improving the accuracy of detection of malicious behaviors.