Automatic analysis of a computer virus structure and means of attachment to its hosts

• US 5,485,575 A
• Filed: 11/21/1994
• Issued: 01/16/1996
• Est. Priority Date: 11/21/1994
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Kephart, Jeffrey O., Chess, David M., Sorkin, Gregory B.
• Primary Examiner(s): Beausoliel, Jr., Robert W.
• Assistant Examiner(s): CHUNG, PHUNG M
• Application Number:
  US08/342,949
• Time in Patent Office: 421 Days
• Field of Search: 395/575, 371/19, 380/4, 380/25
• US Class Current: 714/38.13
• CPC Class Codes: G06F 21/564 by virus signature recognition

Abstract

Information pertaining to the verification of the identity of, and reversal of, a transformation of computer data is derived automatically based on a set of samples. The most important class of transformations is computer viruses. The process extracts this information for a large, fairly general class of viruses. Samples consisting of host programs infected with the virus and sample pairs consisting of an infected host and the corresponding original, uninfected host are obtained. A description of how the virus attaches to the host program, including locations within uninfected host of components of both the original host and the virus is generated. Viral code is matched across samples to obtain a description of "invariant" regions of the virus. Host bytes embedded within the virus are located. A description of the original host locations permits ant-virus software on a user'"'"'s machine to restore the bulk of a program that has been infected. Characterization of the correspondence between invariable portions of the virus and destroyed parts of the host enables anti-virus software to complete the repair.

Claims

1. A method for automatically deriving verification and removal information for a function-preserving transformation of computer data from a set of untransformed data samples and corresponding transformed data samples, comprising the steps of:

obtaining a set of "sample pairs", each sample pair consisting of a transformed data sample and a corresponding original, untransformed data sample;

locating one or more fragments of each original data sample within a corresponding transformed data sample to obtain a generalized description, applicable to each of the sample pairs, of locations of fragments of each original data sample and locations of new data regions added by the function-preserving transformation that applies to each of the sample pairs;

matching new data regions added by the function-preserving transformation across different samples to obtain a description of portions of the new data regions that are "invariant" across different samples;

locating within other, variable portions of the new data regions any data from an original data sample embedded there;

generating a prescription for verifying with high confidence that any given data sample has resulted from an application of the function-preserving transformation; and

generating a prescription for restoring a data sample that has been transformed by the function-preserving transformation to a form functionally equivalent to that prior to the transformation.

2. The method recited in claim 1 wherein the function-preserving tranformation is a computer virus.

3. The method recited in claim 1 wherein the step of locating one or more fragments of each original data sample includes the step of using markers to identify relative locations in each original data sample.

4. The method recited in claim 1 wherein data from the original data sample is embedded in said variable portions of new data regions by encryption, said method further comprising the step of deriving an encryption key for decrypting embedded original data.

5. A method for automatically deriving verification and removal information for a computer virus from a set of infected programs and corresponding uninfected programs, comprising the steps of:

obtaining a set of "sample pairs", each sample pair consisting of a program infected with the computer virus and a corresponding original, uninfected program;

generating a description of how the computer virus attaches to host programs;

matching viral data across different infected samples to obtain a description of "invariant" portions of the computer virus;

locating within other, variable portions of the computer virus any host bytes embedded there;

generating a prescription for verifying with high confidence that any given program is infected with the computer virus; and

generating a prescription for restoring a program that has been determined to have been infected with the computer virus to a state functionally equivalent to the program'"'"'s original, uninfected state.

6. The method recited in claim 5 wherein the step of obtaining a set of "sample pairs" includes the step of selecting "goat" files of uninfected programs and causing the "goat" files to be infected by a computer virus.

7. The method recited in claim 5 wherein said computer virus applies an encryption to some or all code and data of the computer virus and a host program, further including the step of identifying a method of decryption inverse to the encryption, identifying a location of a key suitable for said decryption, and identifying a plaintext of said computer virus.

8. The method recited in claim 7 wherein the key location is an encryption of a true key with another, fixed key and the plaintext is a decryption of a computer virus'"'"' true plaintext with the fixed key, wherein the step of identifying a location of a key and identifying a plaintext includes identifying a location of a true key and true plaintext.

9. The method recited in claim 5 wherein the step of generating a description of how the virus attaches to a host program includes an attachment description:

space="preserve" listing-type="equation">FullAttachInfo={<

class,AttachInfo>

.sub.c },

where the notation indicates a set of <

class,AttachInfo>

vectors, each indexed by a value of c, the classes being defined by attributes of the infected host program.

10. The method recited in claim 9 wherein class definitions are ##EQU8## and for any class, the attachment is described in the form

11. The method recited in claim 10 wherein the sections are a partition of original and infected hosts into contiguous regions and there are three types of sections, described by ##EQU9## and wherein sections of type H and HI cover an entire original host without any overlap, and sections of type I and HI cover an entire infected host without any overlap.

12. The method recited in claim 5 wherein the step of generating a description of how the computer virus attaches to the host program includes the step of using markers to identify relative locations in the data and a marker takes one of the values ##EQU10## where in the last case, said description includes the character string serving as the marker.

13. A computing system for automatically deriving verification and removal information for a function-preserving transformation of computer data from a set of untransformed data samples and corresponding transformed data samples, comprising:

data accessing means for obtaining a set of "sample pairs", each sample pair consisting of a transformed data sample and a corresponding original, untransformed data sample;

scanning means operable on the set of "sample pairs" obtained by said data accessing means for locating one or more fragments of each original data sample within a corresponding transformed data sample to obtain a generalized description, applicable to each of the sample pairs, of the locations of the original fragments and locations of new data regions added by the function-preserving transformation that applies to each of the sample pairs;

comparing means operable on the set of "sample pairs" obtained by said data accessing means for matching new data regions added by the function-preserving transformation across different samples to obtain a description of portions of the new data regions that are "invariant" across different samples;

said scanning means locating within other, variable portions of the new data regions any data from an original data sample embedded there; and

output means responsive to said scanning means and said comparing means for generating a prescription for restoring a data sample that has been transformed by the function-preserving transformation to a form functionally equivalent to that prior to the transformation.

14. The computing system recited in claim 13 further comprising means for verifying with high confidence that any given data sample has resulted from an application of the function-preserving transformation prior to said output means generating a prescription for restoring a data sample.

15. The apparatus recited in claim 13 wherein the function-preserving transformation is a computer virus.

16. A computer implemented-method for automatically deriving a general description of an effect of a transformation of original data on any given sample of the data, comprising the steps of:

obtaining a plurality of input data samples consisting of

(a) one or more samples of transformed data resulting from application of the transformation to the original data, and

(b) zero or more sample pairs, each sample pair consisting of

i. an original data sample, and

ii. a transformed data sample resulting from application of the transformation to a same original data sample;

comparing the input data samples with one another to obtain a universal transformation description which describes a relationship between original and transformed data, said universal transformation description being consistent with the input data samples;

outputting the universal transformation description as a universal transformation which is applicable to samples of data included or not included among the input data samples; and

in cases where the transformation is reversible or partly reversible, deriving a general description of a means of reversing or partly reversing an effect of the transformation.

17. The computer-implemented method recited in claim 16 wherein the transformation of data is performed by a computer virus.

18. The computer-implemented method recited in claim 16 wherein the step of obtaining samples includes converting an original "goat" data sample to a sample pair by causing the transformation to be applied to the "goat" data sample.

19. The computer-implemented method recited in claim 18 wherein the original "goat" data sample is designed to possess variation in one or more attributes including a size of the "goat" data sample in bytes, a type of data, and a presence or absence of regions of repeated bytes and lengths of the regions.

20. The computer-implemented method recited in claim 16 wherein the step of obtaining input data samples includes converting a transformed data sample to a sample pair by obtaining a corresponding original data sample.

21. The computer-implemented method recited in claim 16 wherein the step of comparing the data samples with one another to obtain a universal transformation description begins with a set of structural templates that, based on prior knowledge of the transformation, are likely to be applicable and comprises the steps of:

selecting a template and determining its parameters in such a way that a mapping between the original and transformed members of each sample pair is completely and accurately described;

reporting an identity of the template and the parameters of a template match in the case of a successful match; and

reporting failure if no templates produce a suitable explanation for the transformation.

22. The computer-implemented method recited in claim 16 wherein the step of comparing the data samples with one another to obtain a universal transformation description includes the steps of:

comparing transformed and untransformed data samples of each sample pair;

locating within the transformed data sample in each sample pair one or more contiguous byte regions, each region being equivalent to a corresponding byte region in a corresponding original untransformed data sample;

outputting a transformation description applicable to each of the sample pairs containing locations within transformed and untransformed data samples of equivalent contiguous byte regions, locations of zero or more new data regions within a transformed data sample that do not occur within an untransformed data sample, and locations of zero or more missing data regions within an untransformed data sample that do not occur within a transformed data sample, each transformed data sample being fully or partially covered by a series of equivalent data regions and new data regions, and each untransformed data sample being fully or partially covered by a series of equivalent data regions and missing data regions;

matching new data regions added by the transformation across different transformed samples to obtain a description of portions of new data regions that are invariant across different samples; and

locating within other, potentially variable portions of new data regions any data from missing data regions that are embedded there.

23. The computer-implemented method recited in claim 16 wherein the step of comparing the data samples with one another to obtain a universal transformation description comprises the steps of:

comparing transformed data samples with one another to identify one or more new data regions consisting of equivalent contiguous byte sequences that are present in all of the transformed data samples; and

eliminating from new data regions any regions or portions of regions that appear in all of the untransformed data samples; and

wherein the step of outputting the universal transformation description includes outputting a description of locations of the new data regions.