Title: Access Control Models: From the real-world to trusted computing
1Access Control Models From the real-world to
trusted computing
2Lecture 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)
3Lecture 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 Bibas Model
- What is a trusted system? Trusted Computing Base
4System-security vs. Message-security
- In the cryptographic formulation of security, we
were concerned with the confidentiality,
authenticity, integrity, and non-repudiation 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?
5Confidentiality in Systems
- Many of the motivations behind confidentiality
comes from the militarys 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
Unixs 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
6Integrity 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 datas integrity
has been violated - Achieved by analyzing the systems events (system
logs), or analyze data to see if required
constraints are violated
7Availability 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
8Threats
- 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)
9Overview 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
10Objects 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/programs 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
11Basic 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
programs accesses so that it cant 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 dont need
protection. - Maybe break objects into smaller objects (see
Landwehr)
12Access 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!
13Insight 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.
14Real-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.
15Multilevel 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
16Access 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).
17Access Control Matrix
- Access control matrix arose in both OS research
and database research - Example
- 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 to not have access rights to most
objects - Every subject is also an object!!!
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
18Access 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.
19Take-Grant Models
- Take-Grant Models represent a system using a
directed graph - 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. - Access rights are read (r), write (w), take (t),
grant (g) - Take implies that node A can take node Bs access
rights to any other node - Grant implies that node B can be given any access
right A possesses
Take Operation
r, g
t
A
B
C
Grant Operation
g
A
B
r, w
C
20Take-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
Delete Operation
A
A
21Take-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
B with? - Example
X
X
X
t
t
A
B
C
A
B
C
1. A creates V with t,g
3. B grants to V the X to C
X
t
X
t
A
B
C
A
B
C
t,g
g
X
V
V
X
2. B takes g to V from A
4. A takes X to C from V
X
t
X
t
A
B
C
A
B
C
t,g
g
t,g
g
X
V
V
22Problems with Discretionary Access Control
- Discretionary access controls are inadequate for
enforcing information flow policies - The 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 - Bob can subvert Alices 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
File X File Y
Alice Own
Bob Read Write
Eve Read
23DAC and Trojan Horses
- What if Bob isnt bad Eve could still read X by
convincing Bob to use a program carrying a Trojan
Horse (Troy) - Consider the new access control matrix
- Eve has created Troy and given it to Bob, who has
execute privileges - Troy inherits Bobs 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
File X File Y Prog. Troy
Alice Own
Bob Read Write Execute
Eve Read Read, Write, Execute
Prog. Troy Read Write
Solution Impose Mandatory Access Controls (MAC
yes, another MAC!) that cannot be bypassed.
24Dominance 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. - Lets 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 on
the set of objects and subjects if - We say that o dominates s (or s is dominated by
o) if .
25Dominance 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
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
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
26Lattice Model of Access Security
- The dominance or can-flow relationship defines a
partial ordering relationship by which we may
specify a lattice (with Dennings axioms) - First, the dominance relationship is transitive
and antisymmetric - Transitive If and , then
- Antisymmetric If and
then . - 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)
27Examples 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 . 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
A,B
A,C
B,C
A1
An
L
A
B
C
L
28Mandatory Access Control (MAC) Models
- Mandatory Access Control (MAC) When a system
mechanism controls access to an object and an
individual user cannot alter hat 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
29Bell-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
30Bell-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 - Simple-Security Property A subject s may have
read access to an object o if and only if
. - -Property A subject s can write to object o iff
- Read access implies a flow from object to subject
- Write access implies a flow from subject to object
31Bell-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
O2
S
r
r
O1
Low Security Level
32BLP 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) Trojan-Horse carrying
program - S-Troy Bob running S-Troy will create Y, which
will be a secret file. Eves unclassified
subjects will not be able to read Y. - U-Troy Bob running U-Troy wont be able to read
X, and so wont be able to copy it into Y. - Thus BLP prevents flow between security classes
- One problem remains Covert Channels but thats
for another lecture
33From 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). - Bibas model principles
- Simple-Integrity Property Subject s can read
object o iff - Integrity -Property Subject s can write object
o only if - 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
34Trusted (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.
35Trusted (Operating) System Design, pg. 2
- Unlike a typical OS, a Trusted OS involves each
object being protected by an access control
mechanism - Users are 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