Information Theory and Security

Download Report

Transcript Information Theory and Security

Access Control Models: From the
real-world to trusted computing
Lecture Motivation

We have looked at protocols for distributing and establishing
keys used for authentication and confidentiality

But who should you give these keys to? Who should you trust?
What are the rules governing when to and not to give out
security credentials

In this lecture, we will look at the broad area of secure and
trusted systems
– We will focus on access control models
– These methods are often used to abstract the requirements for a
computer system
– But, they hold for general systems where security is a concern
(e.g. networks, computers, companies…)
Lecture Outline

Some generic discussion about security

Objects that require protection

Insights from the real-world

Access control to memory and generic objects
– Discretionary Methods: Directory Lists, Access Control
Lists, and the Access Control Matrix, Take-Grant Model
– Failures of DACs: Trojan Horses
– Dominance and information flow, Multilevel security and
lattices
– Bell-LaPadula and Biba’s Model

What is a trusted system? Trusted Computing Base
System-security vs. Message-security

In the cryptographic formulation of security, we were concerned
with the confidentiality, authenticity, integrity, and nonrepudiation of messages being exchanged
– This is a message-level view of security

A system-level view of security has slightly different issues that
need to be considered
– Confidentiality: Concealment of information or resources from
those without the right or privilege to observe this information
– Integrity: Trustworthiness of data (has an object been changed in
an unauthorized manner?)
– Availability: Is the system and its resources available for usage?
Confidentiality in Systems

Many of the motivations behind confidentiality comes from the
military’s notion of restricting access to information based on
clearance levels and need-to-know

Cryptography supports confidentiality: The scrambling of data
makes it incomprehensible.
– Cryptographic keys control access to the data, but the keys
themselves become an object that must be protected

System-dependent mechanisms can prevent processes from
illicitly accessing information
– Example: Owner, group, and public concepts in Unix’s r/w/x
definition of access control

Resource-hiding:
– Often revealing what the configuration of a system is (e.g. use of a
Windows web server), is a desirable form of confidentiality
Integrity in Systems

Integrity includes:
– Data integrity (is the content unmodified?)
– Origin integrity (is the source of the data really what is claimed, aka.
Authentication)

Two classes of integrity mechanisms: Prevention and Detection
Prevention: Seek to block unauthorized attempts to change the data, or
attempts to change data in unauthorized ways
– A user should not be able to change data he is not authorized to change
– A user with privileges to work with or alter the data should not be
allowed to change data in ways not authorized by the system
– The first type is addressed through authentication and access control
– The second type is much harder and requires policies


Detection: Seek to report that data’s integrity has been violated
– Achieved by analyzing the system’s events (system logs), or analyze data
to see if required constraints are violated
Availability of Systems

Availability is concerned with system reliability
– The security side of the issue: An adversary may try to make a
resource or service unavailable
– Implications often take the form: Eve compromises a secondary
system and then denies service to the primary system… as a result
all requests of the first system get redirected to second system
– Hence, when used in concert with other methods, the effects can
be very devastating

Denial of service attacks are an example:
– Preventing the server from having the resources needed to perform
its function
– Prevent the destination from receiving messages
– Denial of service is not necessary deliberate
Threats

There are several threats that may seek to undermine confidentiality,
integrity, and availability:
– Disclosure Threats: Causing unauthorized access to information
– Deception Threats: Causing acceptance of false data
– Disruption Threats: Prevention of correct operation of a service
– Usurpation Threats: Unauthorized control of some service

Examples:
– Snooping: Unauthorized interception of information (passive Disclosure)
– Modification/Alteration: Unauthorized change of information (active
Deception, Disruption, Usurpation)
– Masquerading/Spoofing: Impersonation of one entity by another
(Deception and Disruption)
– Repudiation of Origin: False denial that an entity sent or created data
(Deception)
– Denial of Receipt: A false denial that an entity received some information
or message (Deception)
– Delay: A temporary delay in the delivery of a service (Usurpation)
– Denial of Service: A long-term inhibition of service (Usurpation)
Overview Security Policies

Definition: A security policy is a statement of what is allowed and what is
not allowed to occur between entities in a system

Definition: A security mechanism is a method for enforcing a security policy

Policies may be expressed mathematically
– Allowed and disallowed states may be specified
– Rules may be formulated for which entity is allowed to do which action

These policies may seek to accomplish:
– Prevention
– Detection
– Recovery

This lecture will focus primarily on formal statements of security policies
– Specifically, we will focus on policies associated with access control and
information flow
Objects that Need Protection

Modern operating systems follow a multiprogramming model:
– Resources on a single computer system (extend this to a generic system)
could be shared and accessed by multiple users
– Key technologies: Scheduling, sharing, parallelism
– Monitors oversee each process/program’s execution

Challenge of the multiprogramming environment: Now there are more
entities to deal with… hard to keep every process/user happy when sharing
resources… Even harder if one user is malicious
Several objects that need protection:
– Memory
– File or data on an auxiliary storage device
– Executing program in memory
– Directory of files
– Hardware Devices
– Data structures and Tables in operating systems
– Passwords and user authentication mechanisms
– Protection mechanisms

Basic Strategies for Protection

There are a few basic mechanisms at work in the operating system that
provide protection:
– Physical separation: processes use different physical objects (different
printers for different levels of users)
– Temporal separation: Processes having different security requirements
are executed at different time
– Logical separation: Operating system constrains a program’s accesses so
that it can’t access objects outside its permitted domain
– Cryptographic separation: Processes conceal their data in such a way
that they are unintelligible to outside processes
– Share via access limitation: Operating system determines whether a user
can have access to an object
– Limit types of use of an object: Operating system determines what
operations a user might perform on an object

When thinking of access to an object, we should consider its granularity:
– Larger objects are easier to control, but sometimes pieces of large objects
don’t need protection.
– Maybe break objects into smaller objects (see Landwehr)
Access Control to Memory


Memory access protection is one of the most basic
functionalities of a multiprogramming OS
Memory protection is fairly simple because memory access
must go through certain points in the hardware
– Fence registers, Base/Bound registers
– Tagged architectures: Every word of machine memory has one or
more extra bits to identify access rights to that word (these bits are
set only by privileged OS operations)
– Segmentation: Programs and data are broken into segments. The
OS maintains a table of segment names and their true addresses.
The OS may check each request for memory access when it
conducts table lookup.

More general objects may be accessed from a broader variety of
entry points and there may be many levels of privileges:
– No central authority!
Insight from Real-world Security Models

Not all information is equally sensitive– some data will have more drastic
consequences if leaked than other.
– Military sensitivity levels: unclassified, confidential, secret, top secret

Generally, fewer people knowing a secret makes it easier to control
dissemination of that information
– Military notion of need-to-know: Classified information should not be
entrusted to an individual unless he has both the clearance level and the
need to know that information
– Compartments: Breaking information into specific topical areas
(compartments) and using that as a component in deciding access
– Security levels consist of sensitivity levels and the corresponding
compartments
– If information is designated to belong to multiple compartments, then the
individual must be cleared for all compartments before he can access the
information.
Real-world Security Models, pg. 2

Documents may be viewed as a collection of sub-objects, some
of which are more sensitive than others.
– Hence, objects may be multilevel in their security context.
– Level of classification of an object or document is usually the
classification of its most sensitive information it contains.

Aggregation Problem: Often times the combination of two
pieces of data creates a new object that is more sensitive than
either of the pieces separately

Sanitization Problem: Documents may have sensitive
information removed in an effort to sanitize the document. It is a
challenge to determine when enough information has been
removed to densensitize a document.
Multilevel Security Models

We want models that represent a range of sensitivities and that
separate subjects from the objects they should not have access
to.

The military has developed various models for securing
information

We will look at several models for multilevel security
– Object-by-Object Methods: Directory lists, Access control lists,
Access control matrix, Take-Grant Model
– Lattice model: A generalized model
– Bell-LaPadula Model
– Biba Model
Access Control to Objects

Some terminology:
– Protection system: The component of the system architecture
whose task is to protect and enforce security policies
– Object: An object is an entity that is to be protected (e.g. a file, or
a process)
– Subject: Set of active objects (such as processes and users) that
have interaction with other
– Rights: The rules and relationships allowed to exist between
subjects and objects

Directory-based Access Control (aka. Capability List): A list for
each subject which specifies which objects that subject can
access (and what rights)

Access Control List: A list for each object that specifies which
subjects can access it (and how).
Access Control Matrix


Access control matrix arose in both OS research and database research
Example:
File 1
File 2
Process 1
Process 2
Process 1
Read, Write, Own
Read
Read, Write,
Execute, Own
Write
Process 2
Append
Read, Own
Read
Read, Write,
Execute, Own

What does it mean for a process to read/write/execute another process?
– Read is to receive signals from, write is to send signals to, and execute is
to run as a subprocess

Formally, an access control matrix is a table in which each row represents a
subject and each column represents an object.
– Each entry in the table specifies the set of access rights for that subject to
that object
– In general access control matrices are sparse: most subjects do not have
access rights to most objects
– Every subject is also an object!!!
Access Control Matrix, pg. 2

All accesses to objects by subjects are mediated by an enforcement
mechanism that uses the access matrix
– This enforcement mechanism is the reference monitor.
– Some operations allow for modification of the matrix (e.g. owner might
be allowed to grant permission to another user to read a file)
– Owner has complete discretion to change the access rules of an object it
owns (discretionary access control)

The access control matrix is a generic way of specifying rules, and is not
beholden to any specific access rules
– It is therefore very flexible and suitable to a broad variety of scenarios
– However, it is difficult to prove assertions about the protection provided
by systems following an access control matrix without looking at the
specific meanings of subjects, objects, and rules
– Not suitable for specialized requirements, like the military access control
model.
Take-Grant Models

Take-Grant Models represent a
system using a directed graph
Take Operation:
r, g

Nodes in the graph are either
subjects or objects

An arc directed from node A to
node B indicates that the
subject/object A has some access
rights to subject or object B.
A

Access rights are: read (r), write
(w), take (t), grant (g)
Grant Operation:


Take implies that node A can take
node B’s access rights to any other
node
Grant implies that node B can be
given any access right A possesses
t
B
A
g
r, g
B
r, w
r, w
C
C
Take-Grant Models, pg. 2


Create Rule: A subject A can
create a new graph G1 from
an old graph G0 by adding a
vertex B and an edge from A
to B with rights set X.
Remove Rule: Let A and B
be distinct vertices. Suppose
there is an edge with rights
X. Rules Y may be removed
from X to produce X\Y. If
X\Y is empty, the edge is
deleted.
Create Operation:
A
X={r,g}
B
Delete Operation:
A
A
X={r,g}
X\Y={r}
B
B
Take-Grant Models, pg. 3

Since the graph only includes arcs corresponding to non-empty entries in the
access control matrix, the model provides a compact representation

Question of Take-Grant Models: Can an initial protection graph and rules be
manipulated to produce a particular access right for A to access C with?

Example:
X
A
1. A creates V with {t,g}
X
t
A
B
{t,g}
t
B
C
V
C
A
t
B
V
C
X
C
3. B grants to V the X to C
X
t
A
B
g
V
2. B takes g to V from A
X
t
A
B
{t,g}
g
X
4. A takes X to C from V
t
A
B
{t,g}
V
g
C
X
X
X
X
C
Problems with Discretionary Access Control

Discretionary access controls are inadequate for enforcing information flow
policies
– They provide no constraint on copying information from one object to
another

Example: Consider Alice, Bob, and Eve. Alice has a file X that she wants
Bob to read, but not Eve.
– Alice authorizes Bob via the following Access Control Matrix
File X
Alice
Own
Bob
Read
Eve
File Y
Write
Read
– Bob can subvert Alice’s discretion by copying X into Y. Bob has write
privileges, and Eve has read privileges for Y.
– This case is a simplistic version of what can be much more
pathological… The Trojan Horse…
DAC and Trojan Horses

What if Bob isn’t bad… Eve could still
read X by convincing Bob to use a
program carrying a Trojan Horse (Troy)
File X


Consider the new access control matrix:
– Eve has created Troy and given it to
Bob, who has execute privileges
– Troy inherits Bob’s read privileges
to X, and write privileges to a file Y
(perhaps public)
– Eve has read privileges to file Y
Trojan Horses perform normal
“claimed” operations, but also
participates in subversive activities
Alice
Own
Bob
Read
Eve
Prog. Troy
Read
File Y
Prog. Troy
Write
Execute
Read
Read,
Write,
Execute
Write
Solution:
Impose Mandatory Access
Controls (MAC… yes, another
MAC!) that cannot be bypassed.
Dominance and Information Flow




There are two basic ways to look at the notion of security privileges:
Dominance and Information Flow.
For all essential purposes, they are the same, and its just a matter of semantics.
Let’s start with dominance:
– Each piece of information is ranked at a particular sensitivity level (e.g.
unclassified, confidential, secret, top secret)
– The ranks form a hierarchy, information at one level is less sensitive than
information at a higher level.
– Hence, higher level information dominates lower level information
Formally, we define a dominance relation
if:
 on the set of objects and subjects
s  o  (rank s  rank o )  comp s  comp o 

We say that o dominates s (or s is dominated by o) if s  o.
Dominance and Information Flow, pg. 2

Now let us look at information flow:
– Every object is given a security class (or a security label):
Information flowing from objects implies information flowing
between the corresponding security classes
– We define a can-flow relationship A  B to specify that
information is allowed to flow from entities in security class A
to entities in security class B
– We also define a class-combining operator A  B  C
to specify that objects that contain information from security
classes A and B should be labeled with security class C
– Implicitly, there is the notion of cannot-flow
A NOT

 B
Lattice Model of Access Security






The dominance or can-flow relationship defines a partial ordering
relationship by which we may specify a lattice (with Denning’s
axioms)
First, the dominance relationship is transitive and antisymmetric
– Transitive: If a  b and b  c , then a  c
– Antisymmetric: If a  b and b  a then a  b .
A lattice is a set of elements organized by a partial ordering that
satisfies the least upper bound (supremum) and greatest lower
bound properties (infimum)
Supremum: Every pair of elements possesses a least upper bound
Infimum: Every pair of elements possesses a greatest lower bound
In addition to supremum and infimum between two objects, we
need the entire set of security classes to have a supremum and
infimum (i.e. single low point and single high point)
Examples of Information Flow and Lattices

High-Low Policy: Two
security classes (high and
low)

Bounded Isolated Classes: A
set of classes Aj. Between any
two security classes define the
composition A j  A k  H.
Every class has the low class
as its infimum.

Subset Lattice: Categories A,
B, C may be combined to form
compartments. List of all
subsets forms a lattice
H
{A,B,C}
H
L
A1
…
{A,B}
{A,C}
{B,C}
{A}
{B}
{C}
An
L
{}
Mandatory Access Control (MAC) Models

Mandatory Access Control (MAC): When a system mechanism controls
access to an object and an individual user cannot alter that access, the control
is mandatory access control.

In MAC, typically requires a central authority
– E.g. the operating system enforces the control by checking information
associated with both the subject and the object to determine whether the
subject should access the object

MAC is suitable for military scenarios:
– An individual data owner does not decide who has top-secret clearance.
– The data owner cannot change the classification of an object from top
secret to a lower level.
– On military systems, the reference monitor must enforce that objects
from one security level cannot be copied into objects of another level, or
into a different compartment!

Example MAC model: Bell-LaPadula
Bell-LaPadula Model



The Bell-LaPadula model describes the allowable flows of information in a
secure system, and is a formalization of the military security policy.
One motivation: Allow for concurrent computation on data at two different
security levels
– One machine should be able to be used for top-secret and confidential
data at the same time
– Programs processing top-secret data would be prevented from leaking
top-secret data to confidential data, and confidential users would be
prevented from accessing top-secret data.
The key idea in BLP is to augment DAC with MAC to enforce information
flow policies
– In addition to an access control matrix, BLP also includes the military
security levels
– Each subject has a clearance, and each object has a classification
– Authorization in the DAC is not sufficient, a subject must also be
authorized in the MAC
Bell-LaPadula Model, pg. 2

Formally, BLP involves a set of subjects S and a set of objects O.
– Each subject s and object o have fixed security classes l(s) and
l(o)
– Tranquility Principle: Subjects and objects cannot change their
security levels once they have been instantiated.

There are two principles that characterize the secure flow of
information:
1.
Simple-Security Property: A subject s may have read access to an
object o if and only if l (o)  l (s) .
2.
*-Property: A subject s can write to object o iff

Read access implies a flow from object to subject

l (s)  l (o)
lo  ls
Write access implies a flow from subject to object ls  lo
Bell-LaPadula Model, pg. 3

–
–


High
Security
Level
The *-property is not applied to
users:
Humans are trusted not to leak
information
Programs are assumed
untrustworthy… could be Trojan
Horses
The *-property prohibits a program
running at the secret level from
writing to unclassified documents
Sometimes *-property is modified to
require l(s)=l(o) in order to prevent
“write-up” problems
O3
w
S
O2
r
r
O1
Low
Security
Level
BLP and Trojan Horses

Return to the Trojan Horse problem:
– Alice and Bob are secret level users, Eve is an unclassified user
– Alice and Bob can have both secret and unclassified subjects (programs)
– Eve can only have unclassified subjects
– Alice creates secret file X
– Simple security prevents Eve from reading X directly
– Bob can either have a secret (S-Troy) or an unclassified (U-Troy) TrojanHorse carrying program
– S-Troy: Bob running S-Troy will create Y, which will be a secret file. Eve’s
unclassified subjects will not be able to read Y.
– U-Troy: Bob running U-Troy won’t be able to read X, and so won’t be able
to copy it into Y.

Thus BLP prevents flow between security classes

One problem remains: Covert Channels… but that’s for another lecture…
From BLP to Biba

BLP was concerned with confidentiality– keeping data inaccessible to those
without proper access privileges

The Biba model is the integrity counterpart to BLP
– Low-integrity information should not be allowed to flow to high-integrity
objects
– High-integrity is placed at the top of the lattice and low integrity at the
bottom. Information flows from top to bottom (opposite direction of BLP).

Biba’s model principles
1.
Simple-Integrity Property: Subject s can read object o iff (s)  (o)
2.

Integrity *-Property: Subject s can write object o only if (o)  (s)
In this sense, Biba is the dual of BLP and there is very little difference between
Biba and BLP:
– Both are concerned with information flow in a lattice of security classes
Trusted (Operating) System Design

Operating systems control the interaction between subjects and
objects, and mechanisms to enforce this control should be
planned for at the design phase of the system

Some design principles:
– Least Privilege: Each user and program should operate with the
fewest privileges possible (minimizes damage from inadvertent or
malicious misuse)
– Open Design: The protection mechanisms should be publicly
known so as to provide public scrutiny
– Multiple Levels of Protection: Access to objects should depend on
more than one condition (e.g. password and token)
– Minimize Shared Resources: Shared resources provide (covert)
means for information flow.
Trusted (Operating) System Design, pg. 2

Unlike a typical OS, a Trusted OS involves each object being protected by an
access control mechanism
– Users must pass through an access control layer to use the OS
– Another access control layer separates the OS from using program libraries

A trusted OS includes:
– User identification and authentication
– MAC and DAC
– Object reuse protection: When subjects finish using objects, the resources
may be released for use by other subjects. Must be careful! Sanitize the
object!
– Audit mechanisms: Maintain a log of events that have transpired. Efficient
use of audit resources is a major problem!
– Intrusion detection: Detection mechanisms that allow for the identification
of security violations or infiltrations

Trusted Computing Base (TCB): everything in the trusted operating system
that enforces a security policy