Share
Embed
15-349 Toward Trusted Systems
ThierrySans
November 12, 2014
Transcript
22 slides
-
Towards Trusted Systems Mandatory Access Control Thierry Sans 15-349: Introduction
to Computer and Network Security -
Limitations Insider Attack ➡ The users are not trustworthy Malware
➡ The programs are trustworthy -
Mandatory Access Control (MAC) A Mandatory Access Control (MAC) policy
is a set of access control rules defined by the system itself ➡ Objectives of MAC policies • Ensure the confinement of trusted domains • Control the information flow between these trusted domains -
Differences between DAC and MAC DAC - Discretionary Access Control
• The access control is realized by the system but defined by the users • The object owners manage the rights MAC - Mandatory Access Control • The access control is realized by the system and defined by the system (very lowly configurable) • The system (administrator) manage the rights -
MAC model that will talk about here Multilevel - concept
of confinement • Bell-LaPadula confidentiality model (1973) • Biba integrity model (1977) -
Other MAC models Chinese Wall policy - avoiding conflict of
interest • Brewer and Nash model (1989) Transaction policy - avoiding illegal transactions • Clark and Wilson integrity model (1987) • TBAC (Task-Based Access control) (1997) Separation of duties - avoiding fraud and errors • RBAC (Role-Based Access Control) (1992) • TMAC (Team-Based Access Control) (2004) -
Multilevel security ๏ Warning - Multilevel security is a MAC
policy but a MAC is not a multilevel security Concept of confinement • Vertical confinement compartments • Horizontal confinement sensitivities Multilevel is also called the lattice model ➡ Partial order between pairs of compartments and sensitivities -
Confidentiality and Integrity Controlling information flow with a MAC policy,
but for what purpose? • Confidentiality (also called Military Security Policy) • Integrity (also called Commercial Security Policy) What are the operations considered here? • Reading • Blind writing -
The Bell-LaPadula confidentiality model Objects have a confidentiality sensitivity level
• Indicates the confidentiality of the information Subjects have a confidentiality clearance level • Indicates the user's ability of not releasing sensitive information • A subject with high confidentiality clearance level is supposed to keep secret sensitive information -
Bell-LaPadula - Objective and Hypothesis Objective ➡ Avoiding illegal transmission
of information between different confidentiality sensitivity levels Hypothesis ➡ A subject must choose a security clearance level when accessing an object - concept of sessions currentClearance(subject) ≤ clearanceMax(user) -
No Read up - No Write Down Access is limited
by the “Need to know” rule ➡ No Read-up confidentiality(object) ≤ currentClearance(user) Information leakage is limited by the “Star Property” rule ➡ NoWrite Down currentClearance(user) ≤ confidentiality(object) -
The limitations of the Bell-LaPadula model 1. Concept of “declassification”
in the extended Bell-LaPadula model ➡ Declassification must be trustworthy (paradox?) 2. Sometimes hiding an information is not enough when the existence of this information should be hidden ➡ Sensitive information systems must lie [Cuppens 94] -
The Biba integrity model Objects have a integrity sensitivity level
• Indicates the reliability of the information Subjects have a integrity clearance level • Indicates the user's ability of releasing reliable information • A subject with high integrity level is supposed to release reliable information -
Biba - Objective and Hypothesis (Biba) Objective • Avoiding that
a subject with low integrity clearance modifies high integrity objects using information from low integrity objects Hypothesis • There is no concept of session here • Integrity is not exactly the dual of confidentiality -
Biba in action Very Reliable Reliable Not Reliable The Dean
Professor Student Cows can fly! Trojan horses cannot be trusted The chicken came before the egg -
No Write up - No Read Down ✓ Only a
reliable person can release a reliable information ✓ This person must not be influenced by by a lower reliable information ➡ No Write up integrity(object) ≤ clearanceMax(subject) ➡ No Read down clearanceMax(subject) ≤ integrity(object) -
Combining integrity and confidentiality An object has one and only
one sensitivity level ➡ For reading and writing: clearanceMax(user) = security(object) ๏ Naive approach not sustainable in practice An object has two sensitivity levels ➡ Each operation is integrity or confidentiality sensitive (must be a strict distinction) ๏ Very hard to implement in practice -
Building an OS around MAC and DAC Controlling the access
based on two different policies • a MAC policy that rules all objects and all subject (defining system trusted domains) • a DAC policy defined by users on the set of objects they own (defining user trusted domains) ➡ SELinux (Security-Enhanced Linux) from the NSA -
But hackers are smart ... ✓ Security policies control legitimate
channels ๏ But there might be some non-legitimate channels ➡ Covert channels are out of the scope of the reference monitor