An approach to containing computer viruses

321
An Approach to Containing Computer
Viruses *
Maria M. Pozzo
Terence E. Gray
ComputerScienceLqmrrmenr,Unioersity of Cnlifornrn,
Los Angeles, Los Angeles, CA 90024, U.S.A.
This paper presents a mechanism for containing the spread
of computer viruses by detecting at run-time whether or not an
executable has been modified since its installation. The detec-
tion strategy uses encryption and is held to be better for virus
containment than conventional computer security mechanisms
which are based on the incorrect assumption that preventing
modification of executables by unauthorized users is sufficient.
Although this detection mechanism is most effective when all
executables in a system are encrypted, a scheme is presented
that shows the usefulness of the encryption approach when this
is not the case. The detection approach is also better suited for
use in untrusted computer systems. The protection of this
mechanism in untrusted computing environments is addressed.
Keywords: Infection, Integrity, Trojan horse, Computer viruses
Terry Gray is Director of the UCLA
Distributed Systems Laboratory, and
an Adjunct Associate Professor in the
UCLA Computer Science Department.
He received a BSEE from Northrop
University in 1967, and a Ph.D. in
Computer Science from UCLA in 1978.
Prior to joining the UCLA faculty, Dr.
Gray was Manager of Software En-
gineering at Ampex Corporation. He
has also served on the technical staff
of Transaction Technology, Inc. and
Bell Laboratories. Current research in-
terests include distributed system architecture, computer secur-
ity, software engineering, advanced office systems, and the
social impact of technology. He is a member of IEEE, ACM, Tau
Beta Pi, and Upsilon Pi Epsilon.
* This research was supported in part by the NSF Coordinated
Experimental Research program under grant NSF/MCS
8121696 and by the IBM Corporation under contract
D850915.
0. Introduction
The infection property of a malicious computer
virus [l] which causes modifications to executables
is a concern in computer security. Modifications
to executables are much less noticeable than those
made to text or data files, and often go unde-
tected. Such modifications often cause unex-
pected, unauthorized, or malicious side-effects in a
computer system. This discussion is primarily con-
cerned with the infection property of a malicious
computer virus which causes modifications to ex-
ecutables.
Protecting executables from modification can
be accomplished in two general ways: (1) by ren-
dering the executable immutable and thus prevent-
ing all modifications, or (2) by detecting any
changes to the executable prior to its execution,
The first method can be accomplished by storing
the executable in a read-only storage medium such
as a ROM, a read-only directory or a read-only
disk. Thus, the protection mechanism is coupled
with the system or storage medium employed; its
usefulness relies upon the security of the underly-
ing system. Even if complete confidence in the
security of the operating system is warranted, there
is a problem in employing discretionary access
controls (DAC) for read-only protection. Current
implementations of DAC are fundamentally flawed
[2] in that programs executing on a user’s behalf
legitimately assume all the user’s access rights.
This flaw could allow a computer virus to perform
modifications despite read-only protection. The
second method (detection) can be accomplished
North-Holland
Computers 62 Security 6 (1987) 321-331
Maria Pozzo received her MS. degree
in Computer Science from the Univer-
sity of Connecticut in 1981. She began
her work in the area of Computer
Security in 1982 while working for the
Mitre Corporation. She was a member
of the Multics Evaluation team and
subsequently worked for Honeywell
Information Systems to help achieve a
B2 security rating for Multics. She is
currently working on her Ph.D. at
UCLA pursuing research in computer
sect&v.
0167~4048/87/$3.50 0 1987, Elsevier Science Publishers B.V. (North-Holland)

322 M.M. Pozzo, T.E. Gray / Approach to Containing Computer Virusa
by encrypting the executable using conventional
or public-key cryptography, or by recording a
cryptographic checksum, so that any modification
can be detected prior to execution. The detection
approach links the protection mechanism to the
object to be protected rather than the system or
storage medium, and thus, its usefulness depends
less on the security of the underlying system.
Furthermore, modifications to executables can still
be detected outside the realm of a particular sys-
tem. For example, if encryption is used when
transferring executables between sites, modifica-
tions can be detected at the destination site. This
technique is promising for protecting software dis-
tribution.
This discussion is concerned with the integrity
of executables and provides a generalized mecha-
nism for detecting modification of executables and
limiting the potential damage of a computer virus.
This research was motivated by recent work in the
area of computer viruses [l]. Section 1 provides
background material on computer viruses and dis-
cusses the seriousness of the virus problem. Sec-
tion 2 describes the encryption mechanism and
proposes one possible implementation. This mech-
anism is then discussed by considering its strengths
and weaknesses in Section 3. An overview of the
current prototype and a discussion of the open
issues is presented in the conclusion in Section 4.
The issues discussed in this paper are part of an
ongoing effort. Our long-range goal is to develop a
complementary set of independent mechanisms
for protection against computer viruses and other
malcious programs.
1. Computer Virus Background
A malicious computer virus, like a Trojan Horse i,
lures unsuspecting users into executing it by pre-
1 “The Trojan Horse works much like the original wooden
statue that the Greeks presented at the walls of Troy - it is
an attractive or innocent-looking structure (in this case, a
program) that contains a hidden trick, a trick in the form of
buried programming code that can give a hacker surrepti-
tious entry to the system that unknowingly invites the Trojan
Horse within its figurative walls. The Trojan Horse is very
simple in theory, but also very effective when it works. The
program that is written or modified to be a Trojan Horse is
designed to achieve two major goals: first, it tries to look
very innocent and tempting to run, and second, it has within
itself a few high-security tasks to try [3].
tending to be nothing more than a useful or
interesting program [4], while in reality it contains
additional functions intended to “. . .gain un-
authorized access to the system or to [cause a]
. . .malicious side effect” [5]. Programs of this type
are particulary insidious because they operate
through legitimate access paths. The difference
between a computer virus and a traditional Trojan
Horse is that a virus “ . . .can ‘infect’ other pro-
grams by modifying them to include, a possibly
evolved, copy of itself” [l]. The process of infec-
tion is depicted in Fig. 1.
The victim’s file space contains several “clean”
executables to which the victim possesses modify
access. The villain creates an executable that per-
forms a function designed to entice unsuspecting
victims to invoke it. Embedded in the executable
is a piece of clandestine code that is a virus. When
the program is executed, the hidden viral code is
executed in addition to the program’s normal
service. The victim, however, only sees the normal
service, and therefore, does not detect the pres-
ence of malicious activity. The virus program,
when executed by the victim, carries the victim’s
access rights and, therefore, has modify access to
all of the victim’s executables as well as any other
programs for which the victim has legitimate mod-
ify access. In this case, the virus spreads by di-
rectly copying itself to the target executables. Al-
ternatively, the virus can spread by replacing the
target executables with an executable that con-
tains the virus.
Furthermore, when any other user with access
to the victim’s executables, invokes one of the
infected programs, the virus spreads to that user’s
executables and so on. In addition to the spread-
ing capability, the virus may contain other code,
such as a Trojan Horse, intended to cause damage
of some kind.
The most serious impact of a virus, however, is
the rapidity with which it propagates through the
system undetected. Worm programs (programs or
computations that move around a network gather-
ing needed resources and replicating as necessary)
propagate with similar speed [6]. This potential
widespread security problem is detailed in [l] and
the potential damage to both the public and private
sector is extreme.
Several properties of typical computer systems
lead to an environment in which computer viruses
can wreak havoc: the need for program sharing

M.M. Pozzo, T.E. Gray / Approach to Containing Computer Viruses 323
/
!tVillain
0
RVietim
1. Villain creates b
enticing program
containing virus
Virus
3. Viral program
coverlly modifies
clean exeartties
“clean” exectiables
victim has
rrodify acozi
4. result
b
infected execirlabfes
ETY
AFTER
Fig. 1. The process of infection.
[5], the difficulty in confining programs *, and the
fact that existing discretionary access control (DAC)
mechanisms are fundamentally flawed with re-
spect to limiting computer virus [2]. Mechanisms
exist for limiting the amount of sharing such as
the security and integrity policies [8,9], and flow
lists or flow distance policies [l]. However, to the
extent that these mechanisms permit any sharing,
the damage caused by computer viruses cannot be
eliminated since their malicious activity is con-
ducted via legitimate access paths due to the
fundamental flaw in current DAC implementations.
1.1 Scope
Not all computer viruses are bad [l]. This discus-
sion, however, is primarily concerned with the
infection property of a malicious computer virus.
Viral infection is caused by modification of execu-
tables. For this discussion, modification means
directly changing the target executable or sub-
stituting the target executable with an executable
that has been modified. Lastly, the system admin-
istrator discussed here is one or more persons
trusted not to compromise the security or integrity
’ A program that cannot retain or leaks any of its proprietary
information to a third party is confined [7].
of the system. This research does not address the
case of a system administrator or other privileged
systems user who has decided to corrupt the sys-
tem.
2. Detecting Modification of Executables Using
Encryption
Both conventional and public-key cryptography
[lO,ll] have been used successfully to ensure the
integrity of messages. Our solution proposes the
use of cryptography to protect the integrity of
executables, and thus provide a mechanism to
detect viral spread and limit potential viral
damage.
Fig. 2 depicts the proposed detection mecha-
nism. The executable, E, is encrypted by the
cryptosystem to produce E’. The run-time en-
vironment passes E’ to the cryptosystem where
the decryption is performed. The deciphered ex-
ecutable is passed back to the run-time environ-
ment which will attempt to run the results. If this
attempt fails, the executable has been modified
since it was encrypted and the proper authorities
are modified. Thus, the run-time environment de-
tects any modification, whether unintended or due
to a viral attack. Note that modification of execu-
tables is not prevented; however, since any mod-

324 M.M. Pozzo, 7:E. Gray /Approach to Containing Computer Viruses
Jncryption Environment:
Cm-Time Environment:
Fig. 2. The detection mechanism.
ification is detected at run-time (when the virus
strikes), potential damage is limited. With respect
to the damage that a virus can cause there are two
cases to consider (Fig. 3). If all the executables in
the system are virus-free and encrypted, attempts
to infect an executable by inserting a virus will be
detected and, thus, this mechanism completely
halts any further viral damage. If a virus exists in
an executable prior to its encryption, its presence
will not be detected by this mechanism. This type
of virus can still spread to other executables but
the infection will be detected when the encrypted,
infected program is executed. The original virus,
however, can still accomplish its hidden function
and, in effect, will behave like a traditional Trojan
Horse. It should be noted that denial of service
does occur since the executables will be destroyed
when the infection attempts to spread.
The usefulness of this approach is dependent
on the type of cryptosystem chosen, particularly
the management of the encryption and decryption
key(s). There are several advantages to using pub-
lic-key cryptography as opposed to conventional
cryptography. In a conventional cryptosystem, the
keys used for enciphering and deciphering are
either the same, or each key can be computed
from the other [2,13]. Thus protecting the key(s),
not only from modification but also from dis-
closure, becomes essential to the protection of the
entire mechanism. In a public-key cryptosystem,
enciphering and deciphering is accomplished via
two keys, a private key and a public key [11,12,14].
In the mechanism described above, the private key
is used to encrypt the executable while anyone
with knowledge of the public key can decipher it.
Protecting the private key from disclosure be-
comes the responsibility of the key’s owner and is
no longer part of the mechanism itself. Thus,
protecting the integrity of the mechanism becomes
a matter of protecting the algorithms and the
public keys from modification. In a public-key
cryptosystem, the private key is bound to a par-
ticular individual or group of individuals. This
affords the additional advantage of authenticating
the identity of the encryptor of an executable
which provides additional assurance that the ex-
ecutable has not been replaced by an imposter
program. This can be accomplished in a conven-
tional cryptosystem if the encryptor and decryptor
agree in advance on the key(s) to be used, how-
ever, the practicality of this approach is questiona-
ble. Except for the need for authentication, simply
storing characteristic values of executables and
protecting these values from modification, would
be sufficient to achieve the goal of detecting
changes to an executable.
One possible implementation of the mechanism
described above is to employ a public-key crypto-

M.M. Pozzo, T.E. Gray /Approach to Containing Computer Viruses 325
nitial System Virus-Free:
HALT
Initial Infected System:
EXECUTES HALT
Fig. 3. Virus-free vs. infected system.
Sign Environment:
Run-Time Environment:
,
ESB
I IlZSUltS
System-protected ti I
public
+RT+
keys I I
Fig. 4. Signature block mechanism. (RTVM = run-time validation mechanism).

326 M.M. Pozzo, T.E. Gray /Approach to Containing Computer Viruxs
system to append an encrypted signature block to
the plaintext executable (Fig. 4). The signature
block contains the result of applying a strong
one-way function (characteristic value or crypto-
graphic checksum) to the entire executable plus
some additional information such as the identity
of the signer and a time stamp. The private key of
the signer is used to encrypt the signature block.
At run-time, the signature block is deciphered
with the associated public key, the characteristic
value regenerated and the two values compared. If
the two results differ, a modification has occurred
and the proper authorities are notified. In this
implementation, the system must maintain a list
of public keys and protect them from modifica-
tion. This implementation affords a large degree
of flexibility in determining the set of public keys
that will reside on the system-protected public key
list (SPKL); these are the only keys that will be
recognized by the run-time environment. In ad-
dition, since the executable itself is not encrypted,
the run-time environment will not attempt to run
deciphered code that is garbage. Authentication is
a well-known advantage of signature block mecha-
nisms [15], further indicating that this implemen-
tation is a viable solution.
Essential to the correct operation of this mech-
anism, is determining the public keys that reside
on the SPKL. The system administrator should only
allow public keys of individuals and organizations
trusted to supply software that does not contain a
computer virus. Basically, anyone can sign soft-
ware; the issue is who can sign software and also
have their public key on the public key list.
3. Strengths and Weaknesses
The mechanism described above is most effective
if all the executables in the system are encrypted.
In reality, however, it may not be feasible to
require all executables in the system to be en-
crypted or signed. Of particular concern is the
software development process which requires many
executions during program debugging. Encrypting
and deciphering on every test run will significantly
lengthen this process. Providing a separate de-
velopment environment, although one solution,
may not be practical. Another concern is software
developed by users for their own use, software not
available to the entire system. Since residence in
the system-protected public key list is restricted as
described in the previous section, it is unlikely
that the public key of a normal user will be in the
list. This makes execution of personal software
impossible. Lastly, the system administrator has
the responsibility for ensuring that individuals and
organizations who encrypt software for the sys-
tem, and also have their associated public key on
the system-protected public key list, provide
software that does not contain computer viruses
and other types of malicious programs such as
Trojan horses. Requiring the system administrator
to perform this task for all executables on the
system may be impractical, depending on the de-
gree of protection needed for the type of work
performed by the system. Thus, for practical rea-
sons, it may be necessary for encrypted and unen-
crypted executables to reside on a system simulta-
neously.
3.1 The Coexistence of Encrypted and &encrypted
Executables
One way to allow the coexistence of encrypted
and unencrypted executables in a system is via the
Risk Management Scheme. This scheme allows
administrative classification of software, and per-
mits users to specify the classes of software they
wish to execute. The classes of software corre-
spond to the likelihood that the executable con-
tains malicious code such as a computer virus.
Unencrypted software might be considered most
likely to contain a virus. Thus, a user wishing to
be protected from potential malicious activity
would only allow execution of low-risk software.
This classification, although subjective, serves as a
warning mechanism to users, making them aware
of the potential risk in executing certain programs.
An overview of the Risk Management Scheme is
presented here. For a detailed discussion see [16].
The Risk Management Scheme. The Risk
Management Scheme has two domains: programs
and processes. Programs are assigned “credibility
values” (by the system administrator) and
processes inherit a “risk level” from their parent
process, or ultimately, from the user on whose
behalf they are operating. The operating system is
responsible for preventing a process at risk-level
N from invoking a program whose credibility value
is less than N.
The system administrator assigns software a

M.M. Porro, T.E. Gray /Approach to Containing Computer Viruses 327
credibility value which identifies the likelihood
that the software contains malicious code. In gen-
eral, this value is based on the origin of the
software. Credibility values range from zero to N,
where software with the lowest credibility has the
value of zero and software with the highest credi-
bility on the system has the highest value. Soft-
ware that is formally verified, so that the possibil-
ity of it containing malicious code is small, is
always assigned the highest value. The number of
credibility values is determined by the system
administrator and can be one. For example, in an
environment where security is of primary concern
such as a military installation, a system may be
restricted to only verified software. An environ-
ment where security is of less concern, is unlikely
to have any formally verified software. But, since
differences exist in the credibility of the various
sources of executables, the system administrator
can choose some number of credibility values to
reflect the classes of software on the system. Fig. 5
depicts a possible configuration for credibility val-
ues.
Risk levels specify what classes of software can
be executed for a user. They correspond inversely
to credibility values. If the user’s risk level is set to
the highest credibility value on the system, the risk
of damage to that user is the lowest possible. On
the other hand, the greatest risk is taken when the
user specifies a risk level of zero.
When a user logs in, a risk level is established
for the session. This risk level can be determined
in two ways. The first way is for the user to
specify the desired risk level as an argument to the
login command (e.g. login Joe-session-risk 3). The
second way is to assume the default risk level for
that user. Initially, the default risk level of all
users is the highest credibility value on the system.
The user can reset this default risk level by speci-
fying the desired default as an argument to the
login command (e.g. login Joe-default _ risk 2). The
user need only set this once and it remains in
effect until it is explicitly reset by the user. Thus,
assuming the default risk level as the risk level for
the session requires no explicit action on the user’s
part once it is set. Once the risk level for a session
is established, any processes that are spawned
inherit the risk level of the parent, restricting
children to running software of the same credibil-
ity value or higher. The only way for a user to
override the risk level for a particular session is via
the RUN-UNTRUSTED command which takes one
executable program as an argument. This program
can have a credibility value less than the risk level.
The duration of this exception is the execution of
the program supplied as an argument. The objec-
tive of the “RUN-UNTRUSTED” command is to make
execution of high-risk programs explicit, but not
too inconvenient.
As an example, Fig. 6 shows five possible credi-
bility values for software, where the existence of
malicious code in software with a value of 5 is
unlikely and in software with a value of 0 is most
likely. The initial default for the user is the ability
to run software with a value of 5 only, unless the
user explicitly logs in at a lower risk level or resets
the default risk level. If the user chooses to estab-
lish a session with a risk level of 3, software with
Origin Credibility User’s Risk
User Files 0 - Lowest
User Contributed S/W 1
S/W from Bulletin Board 2
S/W from System Staff 3
Commercial Application S/W 4
S/W from OS Vendor 5 - Highest
0 - Highest Risk
5 - Lowest Risk
Fig. 5. Credibility value and risk level.

328
Credibility
VAN2
M.M. Pozzo, T.E. Gray / Approach to Containing Computer Viruses
__._~__. __----_ ___-. _
Execution User’s Risk
Mode Level
0 RUN-UNTRUSTED
1 RUN-UNTRUSTED
2 RUN-UNTRUSTED
3 normal
4 normal
5 normal
RiskLevel= 3
Fig. 6. User’s risk level.
values of 0, 1, and 2 cannot be run without using
the RUN-UNTRUSTED command. Of course, the user
has increased the potential risk of exposure to
malicious activity.
Once a credibility value has been assigned to
software, the information must be conveyed to the
run-time environment. This can be accomplished
in several ways. The first approach is to store the
credibility value as part of the executable, compar-
ing the value with the user’s risk level prior to
permitting execution. This approach requires that
the executable be protected from modification to
ensure the integrity of the credibility value. A
second approach is to keep a list of all executable
software in the system and the associated credibil-
ity values. When a user executes a program, the
run-time environment searches the list for the
program’s credibility value and compares it with
the user’s risk level before allowing execution.
Such a list must be protected from illicit modifica-
tion. This approach may not be practical depend-
ing on the time it takes to complete the search. A
third approach is to group software of the same
credibility value in the same place in secondary
storage, and maintain a short, protected list map-
ping credibility values to each file group. Software
of the same credibility value could be stored in the
same directory, in the same filesystem 3, or some
other mechanism used to partition software. The
3 In Unix, a filesystem contains a hierarchical structure of
directories and files and corresponds to a partition of a disk.
Each filesystem is represented internally by a unique number
v71.
list identifying each partition and the associated
credibility value is then short enough to avoid
performance problems, but must still be protected
from modification by anyone except the system
administrator. Fig. 7 shows possible credibility
values for software grouped using Unix 4 directo-
ries as the partitions.
As the number of credibility values is de-
termined by the system administration, so is the
granularity of the partitions. For example, one
system might partition all vendor software into
one partition with the same credibility value while
another system might have separate partitions for
IBM, DEC and AT&T software, each with a different
credibility value.
If an individual program becomes suspected of
containing malicious code, perhaps based on re-
ports from other installation, it can be moved to a
different directory of appropriate credibility value.
However, one disadvantage of associating a credi-
bility value with entire directories or filesystems is
that the full name of a program may be embedded
in other programs or scripts; thus moving a pro-
gram to a different directory having the desired
credibility level is essentially a name change for
that program, and may cause existing scripts to
break. This observation argues in favor of assign-
ing credibility values to individual programs, even
though to do so is more administratively demand-
ing. A combined approach that allows easy assign-
ment of credibility levels to collections of pro-
4 Unix is a trademark of AT&T Information Systems.

M.M. Pozzo, T.E. Gray /Approach to Containing Computer Viruses 329
Origin Credibility Partition
User Files
User Contributed S/W
Bulletin Board S/W
Commercial S/W
S/W from System Staff
Commercial S/w
Verified S/W
S/W from OS Vendor
0 -Lowest
1
I
2
3
3
4
5 - Highest
lusr
lusrlflakey
lusrlnet
lusrlbin2
hsdlocal
lusrlbio
/usr/vi?r
lbin
Fig. 7. Partitioning software of different credibility values.
grams, but provides for individual exceptions may
be the winning strategy.
Encryption Identification. Another major con-
cern in allowing unencrypted and encrypted ex-
ecutables to coexist in a system is communication
with the Run-Time Validation Mechanism (RTVM).
There must be a way for the RTVM to know exactly
which executables are required by the system ad-
ministrator to be encrypted. Furthermore, this in-
formation must be trusted to accurately reflect the
intention of the system administrator, i.e. it must
be tamperproof. There are many ways to represent
this “encryption identification”; several are listed
below.
l Record the information as a protected attribute
of the executable.
l Keep a system-protected list of all executables
that are required by the system administrator to
be encrypted.
l Group all encrypted and unencrypted software
in the same place in secondary storage, and
maintain a short, protected list identifying which
locations must contain encrypted software.
In all cases, however, this information must be
protected from illicit modification. For example, if
an executable is identified as “must be encrypted”
and this information is not protected from modifi-
cation, a perpetrator could remove the encryption
identification so that it is not validated by the
RTVM. Essentially, unless the encryption identifica-
tion is protected, the encryption mechanism is
useless when unencrypted software is allowed to
exist in the system.
3.2 Protecting the Protection Mechanism
The protection of the proposed mechanism itself is
dependent on the integrity of the operating sys-
tem. Protection of the mechanism does not require
preventing disclosure of information, only its
modification. Critical elements include the public
key list (SPKL), and the RTVM. In systems where
unencrypted executables are allowed, the encryp-
tion identification must also be protected as men-
tioned above. If the system is secure, the Trusted
Computing Base (TCB) mediates all access between
subjects and objects [18]. Routines that manipu-
late the public key list and the encryption identifi-
cation would be considered privileged operations
and part of the TCB. The operation of the RTVM
would be considered a trusted operation and also
part of the TCB. If the TCB provides multilevel
security, additional protection is afforded by the
security levels, since in general, a virus cannot
spread between levels. 5
If the underlying system is an Untrusted Com-
puting Base (UCB), alternative measures must be
taken to ensure the integrity of this mechanism. In
addition to restricting valid public keys, the fol-
lowing issues are of primary concern:
0 Protect the public key list and encryption iden-
tification from modification,
0 Protect the routines that manipulate the public
key list and the encryption identification from
modification;
5 The *-property [S] does not allow write-down to a lower
security level.

330 M.M. Porro, T.E. Gray / Approach to Containing Computer Viruses
l Limit execution of routines that manipulate the
public key list and the encryption identifica-
tion;
l Protect the Run-Time Validation Mechanism
from modification.
For a more detailed discussion about protecting
this mechanism when the underlying system is an
Untrusted Computing Base see [19].
4. Conclusion
4.1 Open Issues
Performance issues are an area yet to be examined
but an overall decrease in performance seems
likely. This model requires the operating system
environment to perform several additional services
that will decrease performance. First, the credibil-
ity value of the software to be executed must be
determined and compared to the risk level of the
process executing the software. Secondly, the
validation routines must be invoked for all en-
crypted software. The performance of the valida-
tion routine is dependent on the cryptosystem
employed. Regardless of the one that is chosen,
performance will be decreased.
Another open issue is that of name resolution.
In the proposed model, when an executable is
encountered with a credibility value lower than
the process’s default risk level, resolution is dis-
continued even if the entire name has not been
examined. It may be possible to allow resolution
to continue until an appropriate executable is
found or the entire name has been resolved.
4.2 Current Prototype
The first prototype was implemented on the Locus
distributed operating system [20], which is a net-
work-transparent version of the Unix operating
system. This prototype implements a framework
for the signature mechanism. The primary goal of
the implementation was to investigate the feasibil-
ity of protecting executables by using a signature
block such as the one described. The Risk Mana-
gement Scheme was not included in the first im-
plementation.
A SIGN program was implemented in the initial
system. The cryptosystem used, however, was
trivial, and unsuitable for a real system. The parti-
tions were simulated by using Unix directories.
Both the SPKJ_ and the list mapping the partitions
into credibility values were assumed to already
exist so that routines for manipulating them were
not provided. The characteristic function was im-
plemented by using the Unix “crypt” function to
provide a 4-byte characteristic function. To test
this mechanism, a program was written that in-
voked the RTVM if an executable resided in one of
the “must be encrypted” partitions. If the execu-
table was valid (not modified since it had been
signed), it was executed.
Initial test results showed that modification to
a signed executable would be detected in most
cases. However, the characteristic function genera-
tor must provide a much stronger function than
the 4-byte function supplied by the prototype. In
general, however, any modifications made to the
executable portion of the load module were de-
tected. Appending a virus to the signed executable
was also detected.
4.3 Future Work
The next step is to investigate a more rigorous
characteristic function and to find a suitable pub-
lic-key cryptosystem. The current simulation sys-
tem must be moved to the operating system kernel
and tested in real-time. A workstation may prove
the best environment for the next level of the
prototype. Once a more extensive signature mech-
anism is in place, the next step is to implement the
Risk Management Scheme.
Once the entire model has been implemented,
solutions must be found for the assumptions that
were made. For example, a means for protecting
the operating system kernel when the underlying
system is an Untrusted Computing Base must be
investigated. Also measurement of performance
degradation introduced by the validation step is
crucial to determining the overall feasibility of this
approach.
References
[l] Cohen, F.: “Computer Viruses”, Proceedings of the 7th
DOD/NBS Computer Security Conference, September
1984, pp. 240-263.
[2] Boebert, W.E., and Ferguson, CT.: A Partial Solution to
the Discretionary Trojan Horse Problem, Honeywell Secure
Technology Center, Minneapolis, MN.

MM. Pozzo, T.E. Gray /Approach to Containing Computer Viruses 331
[3] Landreth, B.: Out of the Inner Circle: A Hacker’s Guide to
Computer Security, Microsoft Press, Bellevue, WA, 1985.
[4] Anderson, J.P.: Computer Security Technology Planning
Study, USAF Electronic Systems Division, Bedford, MA.,
Oct. 1972, ESD-TR-73-51.
[5] Denning, D.E.: CTptography and Data Security,
Addison-Wesley Publishing Co., Reading, Ma, 1982.
[6] Shoch, J.F., and Hupp, J.A.: The ‘Worm’ Programs -
Early Experience with a Distributed Computation”. Com-
munications of ACM 25, 3 (March 1982), 172-180.
[7] Lampson, B.W.: “A Note on the Confinement Problem”,
Communications of the ACM 16 (10): 613-615, Ott, 1973.
[8] Bell, D.E., and LaPadula, L.J.: Secure Computer System:
Unified Exposition and Mu/tics Interpretation, MITRE
Technical Report, MTR-2997, July 1975.
[9] Biba, K.J.: Integrity Considerations for Secure Computer
Systems, MITRE Technical Report, MTR-3153, June 1975.
[lo] Campbell, C.M.: The Design and Specification of Crypto-
graphic Capabilities, IEEE Communication Society Mag-
azine, Nov. 1978, pp. 273-278.
[ll] Diffie, W., and Helhnan, M.E.: Privacy and Authentica-
tion: An Introduction to Cryptography, Proceedings of the
IEEE, Vol. 67, No. 3, March 1979.
[12] Meyer, C.H., Matyas, S.M.: Cryptography: A New Dimen-
sion in Computer Data Security, John Wiley & Sons, 1976.
[13] Kahn, D.: The Codebreakers, MacMillan, New York, 1972.
[14] Shannon, C.E.: “Communication Theory of Secrecy Sys-
tems”, BeN System Technical Journal, 28, 1949, pp.
656-715.
[15] Kline, C.S., Popek, G.J., Thiel, G., and Walker, B.J.:
“Digital Signatures: Principles and Implementations”,
Journal of Tele-Communication Networks, Vol. 2, No. 1,
1983, pp. 61-81.
[16] Pozzo, M.M., Gray, T.E.: “Managing Exposure to Poten-
tially Malicious Programs”, Proceedings of the 9th Na-
tional Computer Security Conference, Sept. 1986.
[17] Boume, SE.: 7’he UNIX System, International Computer
Science Series. Addison-Wesley Publishing Company,
1983.
[18] DOD Computer Security Center, “Department of Defense
Trusted Computer System Evaluation Criteria”, DOD,
CSC-STD-001-83, 1983.
[19] Pozzo, M.M., Gray, T.E.: “Computer Virus Containment
in Untrusted Computing Environments”, ZFZP/SEC
Fourth International Conference and Exhibition on Com-
puter Security, December 1986.
[20] WaIker, B.J., Popek, G.J., English, R., Kline, C.S., and
Thiel, G.: “The Locus Distributed Operating System”,
Proceedings of the Ninth ACM Symposium on Operating
System Principles, October 1983.

An approach to containing computer viruses

More Related Content

What's hot (20)

Similar to An approach to containing computer viruses (20)

More from UltraUploader (20)

An approach to containing computer viruses