Method of, and system for, heuristically detecting viruses in executable code

[0001] This application is the U.S. national phase of international application PCT/GB03/03476, filed 11 Aug. 2003, which designated the U.S. and claims priority to GB Application No. 0218993.4, filed 14 Aug. 2002. The entire contents of these applications are incorporated herein by reference.

[0002] The present invention relates to a method of, and system for, heuristically detecting viruses in executable code by searching the code for known startup sequences.

[0003] A common form of computer virus infection is where the virus's executable code is attached to, or embedded in, a program or other computer file containing executable code which appears, on the face of it, to be benign. One well-established method of virus propagation is where the virus, once activated on a host machine such as a user's PC, will attach itself to one or more programs found on the host in such a way that that program, once run, will execute the virus's code giving it the opportunity to propagate again and/or to undertake whatever other malignant behaviours (such as destruction of files, etc.) have been programmed into it. This method of propagation does, of course, provide an opportunity to detect the virus, for example by associating checksums with program files and detecting when this checksum changes. That is of course only one of the many strategies which have been devised to detect viruses.

[0004] Another well-known method of detecting viruses, implemented in many of the anti-virus software packages which are available, involves scanning program and other files for certain characteristic sequences of bytes (known as signatures) which indicate the likely presence of a virus. One of the practical problems with signature-based detection is that it requires some skill and a significant amount of time, when a new virus is first detected, to establish a suitable characteristic signature of it. This signature needs to be one which does not produce too many false positives and which does not misidentify the virus, for example as an existing one with a more benign payload. This signature information then needs to be disseminated to sites which use the anti-virus package in question before it can be used there to detect the newly-identified virus. In recent years, many of the notable virus outbreaks have involved viruses which propagate over the internet and it takes time for publishers of anti-virus software to react when a virus outbreak occurs.

[0005] Some internet service providers offer anti-virus scanning of internet traffic passing through their internet nodes as a value-added service.

[0006] The present invention relates to a method of virus detection which is intended to be useful for ISPs performing anti-virus scanning, e.g. of executables such as program files attached to emails, though it is by no means limited to that application and may be used in any anti-virus package.

[0007] According to the present invention there is provided a method of detecting virus infection of an executable image comprising:

[0008] identifying by reference to a database of known executable image layouts, the layouts to which the executable image conforms;

[0009] identifying start-up code within the executable image by reference to the identified image layout; and

[0010] examining the start-up code with reference to a database of start-up code characteristics to determine whether the image is likely to have been subject to viral modification.

[0011] The invention also provides a system for detecting virus infection of an executable image comprising:

[0012] means for identifying, by reference to a database of known executable image layouts, to which one of those layouts the executable image conforms;

[0013] means for identifying start-up code within the executable image by reference to the identified image layout; and

[0014] means for examining the start-up code with reference to a database of start-up code characteristics to determine whether the image is likely to have been subject to viral modification.

[0019] Before proceeding with the description of the illustrated embodiment of the invention, some terms will be explained.

[0020] MD5 (message digest 5) checksum. MD5 is a one-way hashing algorithm—it generates a large number (the MD5 checksum) after analysing a byte stream—such as a file. The chances of two files generating the same large number are very small. It is also very difficult to create a file which will generate any particular MD5 checksum.

[0021] False positive: A false positive occurs when an anti-virus product identifies a particular file as being malware, whereas in fact it is not.

[0022] Regular expression: Regular expressions are strings which can be used to express patterns for pattern matching purposes. For instance, the perl regular expression
/^hello [0-9]+/

[0023] matches any string starting with the letters ‘hello’, then a space, then one or more digits. Some languages such as perl have native support for regular expressions; for others, libraries are available which implement regular expression matching.

[0024] Compiler: According to strict usage, a compiler generates one or more object modules from program source code. These object modules are typically not executable programs per se but require an additional step of linking by a linker. The action of a linker is typically to generate an image of an executable by linking together the object module(s), and external binary libraries which the module(s) reference; the production of the image may involve the pre-pending of a header region according to an executable file layout of a target operating system as well as the addition of resources such as bitmaps and the like. The term “compiler” as used herein is intended to include a linker, if required from a technical standpoint. What the compiler produces is not necessarily a stand-alone program, of course: compilers also produce executables such as dynamic link libraries and device drivers.

[0025] Program image: A program image is a sequence of bytes of executable code, which may exist on disk, in memory or in a network socket stream. In its on-disk form the image may be part of a program file which also includes a program header containing the information normally found in such programs.

[0026] To gain control, a virus must insert itself into the execution path of program code. Although, theoretically, the virus can insert itself anywhere in a program, if it inserts itself into the middle then this lessens the chance of it gaining control, since the place it inserts itself into may be executed rarely or never. Therefore, many viruses replace the startup code of programs with their own startup code. This guarantees they will be executed, giving them a better chance of survival. The on-disk image of an executable program must conform to a layout appropriate to the operating system and any given operating system may support a number of said layouts. At the time of writing the majority of Microsoft Windows™ programs conform to the “Windows PE” layout. These layouts, as exemplified in FIG. 1, usually begin with a header containing e.g. checksum and relocation tables for segment fix-ups which are carried out by the operating system's (OS) loader as it loads the program. At some point the OS will hand over control to the program by a call to the program's entry point, which is indicated in the program header.

[0027] What happens after that depends on the nature of the program and on the compiler and linker which have been used to create it. Fully compiled, user-runnable, programs generally have a runtime library link which handles a number of common tasks. In particular, the runtime library usually contains routines which are involved at start-up and perform tasks such as setting up in-memory structures such as the program stack and heap. The program code written by the program's author generally assumes that these actions will have been performed by start-up code in the runtime library before the author's code begins to execute. All this is not to say that for a given compiler and linker and runtime library, the start-up code of a program created using them will be completely invariant, but rather that different programs compiled with the same compiler and linker and runtime library exhibit sufficient similarity, at least in terms of code found via the program entry point, to provide the basis for determining whether viral code has been patched into the program after it was compiled and linked.

[0028] To gain control at startup, a virus can change the program start point to point to the virus start code, or it can change part of the actual startup code, replacing it with a jump or call to its own code. FIG. 1 shows an example of a virus changing the program start point to point to its own code. FIG. 2 shows an example of a virus changing the startup code, replacing it with a jump to its own code.

[0029] As mentioned above, often a particular compiler and set of libraries will generate the same startup code for all, or a large proportion of programs it generates. Sometimes, this is because it starts with a standard library sequence that performs common tasks necessary during startup. Sometimes this is because the compiler generates applications in a particular way (eg Visual BASIC).

[0030] If it can be identified that a particular program contains the common startup code, but that the program does not actually start with this code, then this is very suspicious, and the program can be flagged as potentially containing a virus. This would be the case with the example in FIG. 1.

[0031] If it can be identified that a particular program starts with code similar to the common startup code, but that the beginning of this code has been changed, then this is very suspicious, and the program can be flagged as potentially containing a virus. This would be the case with the example in FIG. 2.

[0032] The operation of the present embodiment proceeds by examination of an image of a program, be it on disk, in memory, or part of a network packet stream, with reference to a database of characteristics of programs created using known compilers and a pattern matcher to determine whether the program image, in particular the start-up code deviates from what would be expected from that compiler in a way which makes the program suspicious.

[0033] FIG. 3 shows one embodiment of the invention. For the purposes of illustration, it may be assumed that files to be scanned are delivered from an input queue and each one is processed by the system 10 shown in FIG. 3.

[0034] 1) By analysing the suspicious image using a file-type analyser 20, the type can be determined. For instance, it may be non-program, or program. Non-programs are not analysed further. Programs are further classified depending on their type—for instance, DOS, Windows PE, Windows NE, Linux ELF, Macintosh, etc. This analysis is done by file-type analyser 20. Note that in the case of a file image, the file type (e.g. .EXE, .DLL, etc.) should be disregarded.

[0035] 2) Depending on the type of program, this can then be searched by start up code search 30 for appropriate start-up code against a database of start-up code sequences. For instance, Windows PE files may be searched for startup code created by the Microsoft Visual Studio C compiler, Borland C compilers, the Microsoft Visual BASIC compiler and the Delphi compiler.

[0036] 3) If the startup code search 30 determines that start-up code is found, but the program does not actually start with this code, then it hands to the suspicious file handler (step 6).

[0037] 4) If entry point code analyser 40 determines that the program starts with code similar to known startup code, then go to the exception list step (6) which is handled by the suspicious file handler 60.

[0038] 5) If execution arrives at the exit point 70 the program is flagged ‘not suspicious’ and no further action is taken.

[0039] 6) The suspicious file handler 60 may make use of an exception list to prevent false positives. For instance, there may be genuine program files which appear to contain startup code, but do not, programs which contain recognised startup code but not at the entry point and programs which for some reason contain the startup code but start with some other code. These genuine files can be included in the exception list. The exception list can work in various ways, including but not limited to comparing the MD5 checksum of a file with a list of known checksums, or by searching the files for regular expressions, or by comparing the actual startup code with a list of known exception startup codes. If any exception list match occurs, no further action is taken.

[0040] A further consideration for which the suspicious file handler may be programmed to take account is that utility programs exist which “repackage” program files in certain ways. One such type of utility is the compression utility exemplified by Blinker (www.blink.com) which compresses an executable and adds a stub loader so that when the program is run, the stub loader is invoked and decompresses the executable's image. In most cases, the compression utility will compress the original executable's startup code which will not therefore be found by pattern matching for startup code. However, supposing for some reason a particular startup code sequence was uncompressible, and therefore remained unaltered. This could then generate a false positive. To avoid this, an exception list entry could be created which would, in effect, say “Ignore all programs packed by Blinker”. There are various ways to do this for the different utilities which exist, including detecting the startup code of their own which they insert in the executable image and also checking the section characteristics (such as name, sequence, flags) in layouts such as a PE file.

[0041] 7) Otherwise, the program is flagged as possibly containing a virus. This may be used as an absolute decision, or combined with other heuristics to make an overall decision as to whether the program is viral or not.

[0042] Programs which are stopped as viral, but which do not turn out to be viral, can be analysed, and an exception list entry generated, so that similar false positives do not occur in future.

[0043] As well as using this as a stand-alone virus detection algorithm, this can be combined with other techniques as part of a larger system. For instance, programs flagged as viral by this method may be allocated a certain score, or variety of scores depending on the exact circumstances. Scores may also be assigned using other heuristic techniques, and only if the total score passes some limit is the program flagged as viral.

[0044] Once flagged as viral, any suitable remedial action may be taken, either by the system acting autonomously e.g. by moving the program file to a quarantine directory, or by signalling a human operator that intervention is required.

Example of Examining a Suspicious File

[0045] Following is a simplistic example of an algorithm for determining if a file is likely to be a Windows PE file, which may be implemented by the file type analyser 20.

[0046] Read in first 2 bytes. If these are not ‘MZ’ then stop

[0047] Read in another 58 bytes.

[0048] Read in 4 bytes into variable x (treating using intel byte-ordering)

[0049] Seek to offset x in file

[0050] Read in 4 bytes

[0051] If bytes are P E \0 \0, then file is likely to be a Windows PE file

Example of Searching File For Known Startup Code

[0052] The following is a common startup sequence for programs generated for the Microsoft Development Studio C compiler for the windows environment.

[0053]

Hex bytes in file	Human-readable disassembly
55										push	ebp
8B	EC									mov	ebp, esp
6A	FF									push	0FFFFFFFFh
68	10	11	00	01						push	offset var1
68	80	22	00	01						push	offset loc_1002280
64	A1	00	00	00	00					mov	eax, large fs:0
50										push	eax
64	89	25	00	00	00	00				mov	large fs:0, esp
83	C4	E0								add	esp, 0FFFFFFE0h
53										push	ebx
56										push	esi
57										push	edi
89	65	E8								mov	[ebp+var_18], esp
C7	45	FC	00	00	00	00				mov	[ebp+var_4], 0
6A	01									push	1
FF	15	40	10	00	01					call	ds:_set_app_type
83	C4	04								add	esp, 4
C7	05	B0	32	00	01	FF	FF	FF	FF	mov	dword_10032B0, 0FFFFFFFFh
C7	05	B4	32	00	01	FF	FF	FF	FF	mov	dword_10032B4, 0FFFFFFFFh
FF	15	4C	10	00	01					call	ds:_p_fmode
8B	0D	D0	30	00	01					mov	ecx, dword_10030D0
89	08									mov	[eax], ecx
FF	15	68	10	00	01					call	ds:_p_commode
8B	15	CC	30	00	01					mov	edx, dword_10030CC
89	10									mov	[eax], edx
A1	64	10	00	01						mov	eax, ds:_adjust_fdiv
8B	08									mov	ecx, [eax]
89	0D	B8	32	00	01					mov	dword_10032B8, ecx
E8	16	01	00	00						call	unknown_libname_2
A1	AC	30	00	01						mov	eax, dword_10030AC
85	C0									test	eax, eax
75	0E									jnz	short loc_1002171
68	60	22	00	01						push	offset unknown_libname_1
FF	15	60	10	00	01					call	ds:_setusermatherr
83	C4	04								add	esp, 4
E8	CA	00	00	00						call	_setdefaultprecision
68	0C	30	00	01						push	offset unk_100300C
68	08	30	00	01						push	offset unk_1003008
E8	B1	00	00	00						call	_initterm
83	C4	08								add	esp, 8
8B	15	C8	30	00	01					mov	edx, dword_10030C8
89	55	D8								mov	[ebp+var_28], edx
8D	45	D8								lea	eax, [ebp+var_28]
50										push	eax
8B	0D	C4	30	00	01					mov	ecx, dword_10030C4
51										push	ecx
8D	55	E0								lea	edx, [ebp+envp]
52										push	edx
8D	45	D4								lea	eax, [ebp+argv]
50										push	eax
8D	4D	E4								lea	ecx, [ebp+argc]
51										push	ecx
FF	15	58	10	00	01					call	ds:_getmainargs
83	C4	14								add	esp, 14h
68	04	30	00	01						push	offset unk_1003004
68	00	30	00	01						push	offset unk_1003000
E8	76	00	00	00						call	_initterm
83	C4	08								add	esp, 8
FF	15	30	10	00	01					call	ds:_p_initenv
8B	55	E0								mov	edx, [ebp+envp]
89	10									mov	[eax], edx
8B	45	E0								mov	eax, [ebp+envp]
50										push	eax
8B	4D	D4								mov	ecx, [ebp+argv]
51										push	ecx
8B	55	E4								mov	edx, [ebp+argc]
52										push	edx
E8	40	F5	FF	FF						call	_main
83	C4	0C								add	esp, 0Ch
89	45	DC								mov	[ebp+var_24], eax
50										push	eax
FF	15	48	10	00	01					call	ds:exit
EB	22									jmp	short loc_1002210
8B	45	EC								mov	eax, [ebp-14h]
8B	08									mov	ecx, [eax]
8B	09									mov	ecx, [ecx]
89	4D	D0								mov	[ebp-30h], ecx
50										push	eax
51										push	ecx
E8	31	00	00	00						call	_XcptFilter
83	C4	08								add	esp, 8
C3										retn
8B	65	E8								mov	esp, [ebp-18h]
8B	55	D0								mov	edx, [ebp-30h]
52										push	edx
FF	15	50	10	00	01					call	ds:_exit
83	C4	04								add	esp, 4
C7	45	FC	FF	FF	FF	FF				mov	[ebp+var_4), 0FFFFFFFFh
8B	4D	F0								mov	ecx, [ebp+var_10]
64	89	0D	00	00	00	00				mov	large fs:0, ecx
5F										pop	edi
5E										pop	esi
5B										pop	ebx
8B	E5									mov	esp, ebp
5D										pop	ebp
C3										retn

[0054] However, we cannot simply search for this particular byte pattern (55, 8B, EC, 6A, FF, 68, 10, 11, 00, 01, etc); many of the values are offsets to other routines or data structures, which will vary from program to program because they will be located in different places, and therefore have different offsets. For instance, the fourth instruction, push offset var1, has the byte sequence 68 10 11 00 01 in the example, because the variable var1 is located at offset 0x01001110 in this program. In another program, var1 may be located at a different offset (say 0x10011EF), and the fourth instruction will then have the byte sequence 60 EF 11 00 01.

[0055] A simplistic search could therefore match those bytes that are constant, and skip over the bytes that vary. For every byte in the program file, we attempt to see if the search pattern fits, and if it does we have found the code. If it does not, we carry on with the next byte in the file and so on until the end of the file is reached.

[0056] Match 1 byte: 55

[0057] Match 2 bytes: 8B, EC

[0058] Match 2 bytes: 6A, FF

[0059] Match 1 byte: 68

[0060] Skip the next 4 bytes

[0061] Match one byte: 68

[0062] Skip the next 4 bytes

[0063] . . . and so on until

[0064] Match 2 bytes 8B, E5

[0065] Match 1 byte 5D

[0066] Match 1 byte C3

[0067] A more detailed search could perform other checks on the bytes that vary. For instance, if the bytes are known from knowledge of the start-up code which the compiler generates to be an offset to a data structure containing a value such as ‘Press any key to continue’, the search could check that this offset actually contains this data. If they are an offset to a known routine, the search could recursively check that the known routine matches a correct pattern.

[0068] For instance, suppose that in the original pattern, that var1 contains the string ‘hello’. The search algorithm might now be:

[0069] Match 1 byte: 55

[0070] Match 2 bytes: 8B, EC

[0071] Match 2 bytes: 6A, FF

[0072] Match 1 byte: 68

[0073] Read the next 4 bytes into variable offsetcheck1

[0074] Match one byte: 68

[0075] Skip the next 4 bytes

[0076] . . . and so on until

[0077] Match 2 bytes 8B, E5

[0078] Match 1 byte 5D

[0079] Match 1 byte C3

[0080] Then:

• • [0081] Move to the location held in offsetcheck1 (in our example, this would be 0x01001110).
  • [0082] Match the next 5 bytes: ‘hello’

[0083] If we find this startup code pattern, we then check if it is located at the entry point of the program. If it is, then all is OK. If not, then this is flagged as suspect (startup code found, but program does not start with this code).

Example of Searching File For Changed Startup Code

[0084] Using the same example startup code as before, we could use the following algorithm to determine if the file contained changed startup code:

[0085] Go to offset of program start.

[0086] Skip 15 bytes

[0087] Match 6 bytes: 64, A1, 00, 00, 00, 00

[0088] Match 1 byte: 50

[0089] Match 7 bytes: 64 89 25 00 00 00 00

[0090] . . . and so on until

[0091] Match 2 bytes 8B, E5

[0092] Match 1 byte 5D

[0093] Match 1 byte C3

[0094] If code did not match, stop search.

[0095] If code does match, then the bytes from offset 15 onwards are part of a known startup sequence. If the first 15 bytes also match this startup sequence, all is OK. Otherwise this is potentially interesting. The checks therefore continue as follows.

[0096] Go to offset program start

[0097] Match 1 byte: 55

[0098] Match 2 bytes: 8B, EC

[0099] Match 2 bytes: 6A, FF

[0100] Match 1 byte: 68

[0101] Skip the next 4 bytes

[0102] Match one byte: 68

[0103] Skip the next 4 bytes

[0104] If all matches succeeded, then this is part of a known startup sequence. Otherwise, this is flagged as ‘changed startup code’.