Patent Application
20050166191
The present invention relates to software that is resistant to unauthorized analysis. More specifically, the present invention relates to methods and systems for the production of software code that disguises operations such that analysis of the code either during run-time or during an attempt of reverse engineering is made more difficult.
Software analysis for unauthorized purposes is old. Since the personal computer came of age, and perhaps even before that time, hackers and crackers have been analyzing software to subvert the software writer's goals. In one well-known scenario, dynamic software tracing (tracing the logic flow of a program as the program executes) has been used to determine which values or registers need to be adjusted to allow full functioning of a program. Patches or hacks (small and usually unauthorized programs designed to work with only one specific application) are then written to ensure that those values or registers provide the correct values. Similarly, the program may also be hacked by amending the source code to allow unrestricted access to the program.
These and other similar methods based on analyzing a program's source code may be used for purposes that range from the benign to the malicious. Software may be analyzed to allow hackers to develop ways to circumvent a copy protection scheme, cheat at game programs, or defeat requirements for hardware security checks. Similarly, such analysis may be used to allow hackers to defeat anti-virus software, software based firewalls, or to gain access to restricted functions or databases.
The basis for such attacks on software stems from the fact that most software decisions are based on mathematical operations and comparisons. As such, if the true nature of these operations and comparisons can be hidden in the code such that it is more difficult to analyze the code, hackers may be deterred from trying such attacks. Or, if they are not deterred from trying such measures, hackers may find the analysis too tedious or convoluted to succeed.
Based on the above, there is therefore a need for systems and methods that can obfuscate, hide, disguise, or otherwise conceal the true nature of mathematical or logical operations in software.
The present invention provides systems and methods related to concealing mathematical and logical operations in software. Mathematical and logical operations are disguised by replacing them with logically equivalent expressions. Each equivalent expression has at least two expression constants whose values are based on scaling and bias constants assigned to variables in the original mathematical or logical operation. Each of the expression constants may also be based on additive or multiplicative inverses modulo n of the scaling and bias constants. By replacing the original operations with more complex but logically equivalent expressions containing variables that also involve more operations, the true nature of the original operations is disguised.
In a first aspect, the present invention provides a method for disguising an operation contained in a software program, said operation having at least one variable, the method comprising:
In a second aspect, the present invention provides a system for disguising at least one operation contained in a software program, the or each of said operations having at least one variable, the system comprising:
In a third aspect, the present invention provides a method of disguising a plurality of operations contained in a source code of a software program, the method comprising:
A better understanding of the invention will be obtained by considering the detailed description below, with reference to the following drawings in which:
The invention makes use of linear transformations to disguise or “hide” the original intent of a mathematical, logical, or bitwise operation. A linear transformation substitutes a more complex expression for a simple one. As an example, a variable x can be disguised by assigning x′=sxx+bx with sx being a random scaling factor and bx being a random bias factor. The variable x, instead of being a simple variable, now becomes
To illustrate how linear transformations may be used to disguise mathematical, logical, or bitwise operations, an example will be given. We can assume z=x*y. If we assign random scaling and bias factors to each of the variables x, y, and z, such that
x′=sxx+bx
y′=syy+by
z′=szz+bz
then
This is done by isolating the variable x and y from its linear transforms. All scaling factors (sx, sy, and sz) and bias factors (bx, by, and bz) are assumed to be randomly selected by the user.
Since z′=szz+bz and z=x*y, then z′=sz (x*y)+bz. Since
and
then
To simplify the algebra and for the purposes of the example, we can assume sx=3, bx=5, sy=9, by=8, sz=7, and bz=4. Thus, after substituting the proper values,
However, linear transformation by itself only provides very weak obfuscation as the linear equations can easily be solved. This weakness can be rectified by moving the linear transforms to a finite ring domain using modular mathematics.
Modular mathematics can best be analogized to a “remainder” system. Modular operations can be summarized by its equivalent—the result of x mod n is the remainder after dividing x by n and after discarding the whole number result. As an example, 21 mod 10 results in 1 as 21 divided by 10 equals 2 with a remainder of 1. Once the whole number result, 2, is discarded, then the remainder, 1, is left. Similarly, 15 mod 5 equals 0 as 15 is exactly divisible by 5. 17 mod 5, on the other hand, equals 2 as 17 divided by 5 equals 3 with a remainder of 2. Thus, x mod n is the whole number remainder after x is divided by n.
It should be noted that the above explanation of modular mathematics is only provided for clarity and ease of understanding. Applying a finite ring implementation to linear transforms effectively “wraps” original numbers a number of times around a circular domain or ring, much like in modular mathematics. In modular mathematics, as in a finite ring domain, the domain can be seen as a circle with n-1 elements—with n-1 being succeeded by 0. As such, a count around the circle progresses as . . . n-3, n-2, n-1, 0, 1, 2, . . . By using a linear transform with a finite ring implementation, the finite ring linear transform f(x)=sxx+bx mod n behaves like a hash function and thereby greatly disguises or obscures the real value of the original variable x. This form of transform is resistant to linear forms of analysis and, if the scaling factor sx is large, the analysis needed to reverse engineer the finite ring arithmetic and transform is dauntingly complex.
An added feature of finite ring arithmetic is that it does not distinguish between signed and unsigned arithmetic. In mathematical terms, as opposed to the division hardware based modulus operation, −x in twos complement form (with the sign bit of 1 as a magnitude bit) already has the proper representation as x mod n where n is the hardware modulus. Thus, all finite ring arithmetic is done unsigned but can represent both unsigned and signed arithmetic.
As noted above and to return to our example, we can complicate matters and thereby further disguise the original intent of the original mathematical function z=x*y by applying modular mathematics can be applied to the above equation. Thus, instead of subtracting 5 from x′ or subtracting 8 from y′, the additive inverse mod n of 5 and 8 can be added to x′ and y′ respectively. Similarly, instead of dividing by 3 and 9 respectively, we can multiply the expressions by the multiplicative inverse mod n of 3 and 9 respectively.
It should be noted that the additive inverse (b−1) mod n of a number b is the value of b−1 for which (b+b−1)(mod n)≡0. Similarly, the multiplicative inverse (s−1) mod n of a number s is the value s−1 for which (s*s−1) (mod n)≡1. For n=256, the additive inverse of 5 is 251 as (5+251) (mod 256)=≡0. For n=256, the multiplicative inverse of 3 is 171 as (3*171) (mod 256)≡1.
Returning to the example prior to the simplification, we had
Applying the multiplicative inverse mod n and the additive inverse modulo n, we arrive at
z′=(sz[(x′+bx−1)*sx−1*(y′+by−1)*sy−1]+bx) mod n.
This expands to
z′=((szsx−1sy−1)x′y′+(szsx−1by−1sy−1)x′+(szsy−1bx−1sx−1)y′+(szbx−1sx−1by−1sy−1+bz)) mod n
Since we know the values for all the scaling and bias factors (see above) we only need a value for n. If we assume n=256 (a number selected for simplicity), then we can find the additive and multiplicative inverses of the scaling and bias factors along with the values for the factors themselves required by the equation. These values are:
The above equation, using these values, simplifies to
z′=133x′y′+216x′+103y′+204.
The coefficients were formed by replacing the corresponding variable coefficients with their numeric values as follows:
133=[7(171)(57)] mod 256
216=[7(171)(248)(57)] mod 256
103=[7(57)(251)(171)] mod 256
204=[(7(251)(171)(248)(57)+4)] mod 256
It should be noted that if an 8-bit computer implementation of the final equation is contemplated, then the mod 256 can be dropped as the 8-bit character of the computer will cause the variables to overflow. As such, the original operation of z x*y has now been transformed into
z′=133x′y′+216x′+103y′+204.
Since we know that
x′=sxx+bx=3x+5
y′=syy+by=9y+8
z′=133x′y′+216x′+103y′+204
values can be assigned to the variables when executing the code. As an example of the disguising quality of the above transformation, if the original variables were assigned values as below, then the registers would have as their contents:
The transformed code would produce the following values:
As can be seen, the register values for the transformed expression are quite different from the values for the original variables and equations. Not only that, but the original operation of a simple multiplication has now been transformed into a more complex collection of operations that involve multiple multiplications and additions.
It should be noted that while n=256 was selected for simplicity, in implementations, other values of n may be selected for convenience, keeping in mind the hardware to be used. As an example, 232 is the most convenient value for n for implementations on 32 bit hardware. Similarly, 216 is the convenient choice for 16 bit hardware and 264 is the convenient choice for n for implementations on 64 bit hardware. The reasoning for this is the same as given above for an 8-bit implementation with n=256—the mod 2n can be dropped from the equivalent expression as the n-bit character of the hardware will cause the variables to overflow. The overflow will cause the count to “wrap around” a numerical circle such that the end result is the same as that of dividing by the modulus and taking the remainder. As an example, for an 8-bit computer, the value for n is ideally 256 as this is the value for 28. Overflow for an 8-bit computer will thus cause a count to be . . . 251, 252, 253, 254, 255, 0, 1, 2, 3 . . . . The count thus “wraps around” after 255 back to 0. As a result, the n-bit character of the hardware automatically produces the correct result, modulo n when n is the hardware register modulus.
Based on the above example, any mathematical or logical function in the source code of a computer program can be replaced with a logically equivalent expression. From the example above, the expression z=x*y in source code can be replaced by its logical equivalent z′=133x′y′+216x′+103y′+204. Each of the coefficients in the equation for z′ may be based on the scaling and bias factors assigned to the original variables x, y, and z. Not only that, but each of the coefficients may also be based on the additive inverse modulo n or the multiplicative inverse modulo n of these scaling and bias factors.
To obtain better obfuscation, the scaling factor s should be relatively prime to the modulus n. This means that the scaling factor should have no prime factor in common with the modulus n other than 1. Given that current hardware has moduli that are powers of 2 (e.g. 28, 232, 264), then this means that for implementations using binary hardware, the scale factor should be an odd number. The reasoning behind this is that if the scale is an even number and the modulus is a power of two (e.g. 232, 216, 264), then we are left with representations for only k/2 distinct numbers instead of k distinct numbers. This reduces the efficiency of the implementation and reduces the ambiguity which the attacker must overcome. It should therefore be clear that the use of odd values for the scaling factor is the preferred implementation for binary hardware. The choice of a bias factor, on the other hand, has no restrictions.
It should be noted that while each variable in the original mathematical function is assigned a scaling factor s and a bias factor b (e.g. x′=sxx+bx), other variants may be used to introduce further complexity. Furthermore, as an aid to converting mathematical and logical functions to their logically equivalent expressions, identities may be used. As an example, for bitwise functions, the expression may be expanded. For illustrative purposes, the bitwise operation z=x OR y (written as z=x y in the source code) can be rewritten as z=y−((−x−1)|y). Using such expressions as these and following the general concept outlined in the above example, any bitwise function or operation can be transformed into a logically equivalent expression with multiple constants. Thus, any mathematical function, logical function, or bitwise operation can, using any of a number of identities and the general approach used above, be turned into a myriad of equivalent expressions.
As further examples of the identities and the formats for the equivalent expressions that may be used, the following table lists some logical and bitwise operations, identities which may be used, and possible formats for the resulting equivalent expressions. It should be noted that u and v are linear transforms based on the original variables x and y the same way x′ is a linear transform based on x (e.g. x′=sxx+bx). a, b, c, d, e, f, g, and h are constants based on the constants in the linear transforms u and v.
It should be further noted that the equivalent expressions into which the original function is transformed will depend upon a number of elements. Specifically, the choice of the random scaling and bias factors will influence the final value of the coefficients in the equivalent expression. Also, the choice of the identity used to convert logical or bitwise operations will influence the form and content of the final equivalent expression. Because of the dependency of the final expression on such user (i.e. the code generator) selected elements, the final expression used to replace a function in a piece of source code is not easily determinable by someone analyzing the code.
To automate the process of replacing the relevant mathematical, logical, or bitwise operations in a piece of source code, a system may be devised with a number of modules as illustrated in
Referring to
The identification module 50 may take the form of well-known parser software used in the development of applications such as compilers. Such parsers can receive strings of alphanumeric characters and, by isolating specific segments of such strings, can recognize sequences of alphanumeric characters as corresponding to specific mathematical, logical, or bitwise operations. Once identified, the specific operation any coefficients associated with it can be passed to the conversion module 60.
The conversion module 60 can be implemented in a number of ways depending on the end user's preferences. Ideally, the conversion module 60 has a library of equivalent expressions for each operation to be disguised. The expressions in the library are previously derived and are available for selection by the conversion module. As an example, an operation may correspond to 3 different equivalent expressions, each equivalent expression being different in terms of form and content/coefficients from the others. Similarly, another option would be to have each operation correspond to a single equivalent expression with multiple coefficients. Each of the coefficients can then be made different every time the operation needs to be disguised. As an example, we can assume that an operation z=x*y may be disguised as z′=ax′y′+bx′+cy′+d with a, b, c, and d as constant coefficients. Every time the operation z=x*y occurs in the original source code, it can be replaced by the same equivalent expression z′=ax′y′+bx′+cy′+d but with different values for a, b, c, and d. As yet another option, the equivalent expressions may be stored by the conversion module 60 not as finished expressions where the coefficients are integers but as nested formulas to be executed during run-time. An expression such as z′=133x′y′+216x′+103y′+204 may then be stored as
z′=[[7(171)(57)] mod 256]x′y′+[[7(171)(248)(57)] mod 256]x′+[[7(57)(251) (171)] mod 256]y′+[[(7(251)(171)(248)(57)+4)] mod 256]
with the mathematical expressions in the brackets being evaluated when the code is run.
Once the conversion module 60 has determined the proper equivalent expression to be used in disguising the operation identified by the identification module 50, this equivalent expression is then transferred to the replacement module 40. The replacement module 40 then, from the data received from the identification module 50 by way of the conversion module 60, finds the relevant section of the source code where the substitution is to occur. The equivalent expression is then placed in lieu of the original operation in the source code.
The transformed code 70 that results may therefore have expressions that have coefficients that are either specific integers or are mathematical expressions themselves. The replacement of the relevant operations by their equivalent expressions may be on a per-value and per-operation basis. Thus, as an example, even though one operation f(x) is used numerous times in a piece of source code, for each instance that f(x) occurs, a different equivalent expression with different coefficients may be used. It should, however, be noted that if a particular subroutine is called or accessed multiple times from outside the specific set of code, a given parameter in that subroutine is always encoded the same way for consistency. This, however, is an exceptional case and is not the default action.
The above-noted and explained scheme works at disguising the true nature of the operations by rendering both dynamic and static tracing difficult. An attacker who keeps track of the register values and of the compiled code as it executes must necessarily try to figure out the complex and numerous mathematical and logical operations that have now been substituted for even the simplest of operations. Since the transformed code is what was complied and is being examined by the attacker, both static and dynamic tracing will yield similarly difficult results for the attacker.
In terms of functionality, if the above scheme is applied to some source code, the transformed code resulting will produce different results from the original source code. However, these different results will contain the same information as the original results from the original source code, albeit in a disguised form. The reasoning behind this is that all the operations in the transformed code will have been disguised and transformed and, as such, will be producing output and will be expecting input that are both transformed as well. The only exception to this would be sections of code that are required to output a readable result or read-in a readable parameter. All other sections of the source code are to be disguised or are to undergo a transformation to hide or disguise their true nature.
It should be noted that the above scheme may be used in conjunction with the methods and systems disclosed in U.S. Pat. No. 6,594,761 entitled TAMPER RESISTANT SOFTWARE ENCODING, herein incorporated by reference.
Referring to
It should also be noted that the step of assigning a multi-variable expression to each variable in an operation is an implicit step that is not shown in
Embodiments of the invention may be implemented in any conventional computer programming language. For example, preferred embodiments may be implemented in a procedural programming language (e.g. “C”) or an object oriented language (e.g. “C++”). Alternative embodiments of the invention may be implemented as pre-programmed hardware elements, other related components, or as a combination of hardware and software components.
Embodiments can be implemented as a computer program product for use with a computer system. Such implementation may include a series of computer instructions fixed either on a tangible medium, such as a computer readable medium (e.g., a diskette, CD-ROM, ROM, or fixed disk) or transmittable to a computer system, via a modem or other interface device, such as a communications adapter connected to a network over a medium. The medium may be either a tangible medium (e.g., optical or electrical communications lines) or a medium implemented with wireless techniques (e.g., microwave, infrared or other transmission techniques). The series of computer instructions embodies all or part of the functionality previously described herein. Those skilled in the art should appreciate that such computer instructions can be written in a number of programming languages for use with many computer architectures or operating systems. Furthermore, such instructions may be stored in any memory device, such as semiconductor, magnetic, optical or other memory devices, and may be transmitted using any communications technology, such as optical, infrared, microwave, or other transmission technologies. It is expected that such a computer program product may be distributed as a removable medium with accompanying printed or electronic documentation (e.g., shrink wrapped software), preloaded with a computer system (e.g., on system ROM or fixed disk), or distributed from a server over the network (e.g., the Internet or World Wide Web). Of course, some embodiments of the invention may be implemented as a combination of both software (e.g., a computer program product) and hardware. Still other embodiments of the invention may be implemented as entirely hardware, or entirely software (e.g., a computer program product).
A person understanding this invention may now conceive of alternative structures and embodiments or variations of the above all of which are intended to fall within the scope of the invention as defined in the claims that follow.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2,456,644 | Jan 2004 | CA | national |
