EE515/IS523
Think Like an Adversary
Lecture 6
Access Control/UI in a Nutshell
Yongdae Kim
Recap
 http://syssec.kaist.ac.kr/courses/ee515
 E-mail policy
 Include [ee515] or [is523] in the subject of your e-mail
 Class Presentation Choices (say “yes” to at least 3)
 https://docs.google.com/spreadsheet/viewform?formkey=dEdzMT
RWRk8zRHFaZjdIS3F%202TE44ekE6MQ#gid=0
 Text only posting, email!
 Preproposal meeting this week
 Group leader sends me three 30-min time windows between
Wednesday and Friday (evening is OK)
Recap
Cryptography
Challenge-Response Protocols
Key Management
Kerberos vs. PKI vs. IBE
OS Security
OS Security is essentially concerned with
four problems:
User authentication links users to processes.
Access control is about deciding whether a
process can access a resource.
Protection is the task of enforcing these
decisions: ensuring a process does not access
resources improperly.
Isolation is the separation of processes’
resources from other processes.
Access Control
The OS mediates access requests between
subjects and objects.
Subject
?
Object
Reference
monitor
This mediation should (ideally) be impossible
to avoid or circumvent.
Definitions
Subjects make access requests on objects.
Subjects are the ones doing things in the system,
like users, processes, and programs.
Objects are system resources, like memory, data
structures, instructions, code, programs, files,
sockets, devices, etc…
The type of access determines what to do to the
object, for example execute, read, write, allocate,
insert, append, list, lock, administer, delete, or
transfer
Access Control
 Discretionary Access Control:
 Access to objects (files, directories, devices, etc.) is permitted based on
user identity
 Each object is owned by a user.
 Owners can specify freely (at their discretion) how they want to share their
objects with other users,
 by specifying which other users can have which form of access to their objects.
 Discretionary access control is implemented on any multi-user OS (Unix,
Windows NT, etc.).
 Mandatory Access Control:
 Access to objects is controlled by a system-wide policy
 for example to prevent certain flows of information.
 In some forms, the system maintains security labels for both objects and
subjects
 based on which access is granted or denied.
 Labels can change as the result of an access
 Security policies are enforced without the cooperation of users or
application programs.
 Mandatory access control for Linux: http://www.nsa.gov/research/selinux/
Access Control Matrix
Obj 1
Obj 2
Obj 3
…
Obj n
Subj 1
rwl
rwlx
-
-
l
Subj 2
rwl
rlx
rwl
-
-
Subj 3
-
-
-
rl
r

Subj m

rl
lw
rl
rw
r
Representations
O1 O2 …
An access control matrix can
be represented internally in S1 rwl wl different ways:
wlk S2 ida
rl
Access Control Lists (ACLs) S3 store the columns with the
…
Sm rwlx wi w
objects
Capability lists store the rows with the
subjects
Role-based systems group rights according
to the “role” of a subject.
Access Control Lists
The ACL for an object lists the access rights
of each subject (usually users).
To check a request, look in the object’s ACL.
ACLs are used by most OSes and network
file systems, e.g. NT, Unix, and AFS.
ACL Problems
To be secure, the OS must authenticate that
the user is who (s)he claims to be.
To revoke a user’s access, we must check
every object in the system.
There is often no good way to restrict a
process to a subset of the user’s rights.
Capabilities
Capabilities store the allowed list of object
accesses with each subject.
When the subject requests access to object
O, it must provide a “ticket” granting access
to O.
These tickets are stored in an OS-protected
table associated to each process.
No widely-used OS uses pure capabilities.
Some systems have “capability-like”
features: e.g. Kerberos, NT, OLPC, Android
ACL vs. Capabilities
Capabilities do not require authentication:
the OS just checks each ticket on access
requests.
Capabilities can be passed, or delegated,
from one process to another.
We can limit the privileges of a process, by
removing unnecessary tickets from the table.
Roles
S1 S2 S3 … Sm
S1 S2 S3 … Sm
R1
O1 O2 … On
R2
O1 O2 … On
Unix/POSIX Access Control
kyd@dio (~) % id
uid=3259(kyd) gid=717(faculty)
groups=717(faculty),1686(mess),1847(S07C8271),1910(F07C5471
),2038(S08C8271)
kyd@dio (~) % ls -l News_and_Recent_Events.zip
-rw-rw-rw- 1 kyd faculty 714904 Feb 22 10:00
News_and_Recent_Events.zip
kyd@dio (/web/classes02/Spring-2011/csci5471) % ls –al
drwxrwsr-x 4 kyd S11C5471
512 Jan 19 10:23 ./
drwxr-xr-x 46 root daemon
1024 Feb 17 23:04 ../
drwxrwsr-x 3 kyd S11C5471
512 Feb 16 00:36 Assignment/
Mandatory Access Control policies
Restrictions to allowed information flows are
not decided at the user’s discretion (as with
Unix chmod), but instead enforced by
system policies.
Mandatory access control mechanisms are
aimed in particular at preventing policy
violations by untrusted application software,
which typically have at least the same access
privileges as the invoking user.
Data Pump/Data Diode
Like “air gap” security, but with one-way
communication link that allow users to
transfer data from the low-confidentiality to
the high- confidentiality environment, but
not vice versa.
Examples:
Workstations with highly confidential material are
configured to have read-only access to low
confidentiality file servers.
The covert channel problem
 Reference monitors see only intentional communications channels,
such as files, sockets, memory.
 However, there are many more “covert channels”, which were neither
designed nor intended to transfer information at all.
 A malicious high-level program can use these to transmit high-level
data to a low-level receiving process, who can then leak it to the
outside world.
 Examples for covert channels:
 Resource conflicts – If high-level process has already created a file F, a low-level
process will fail when trying to create a file of same name → 1 bit information.
 Timing channels – Processes can use system clock to monitor their own progress and
infer the current load, into which other processes can modulate information.
 Resource state – High-level processes can leave shared resources (disk head position,
cache memory content, etc.) in states that influence the service response times for
the next process.
 Hidden information in downgraded documents – Steganographic embedding
techniques can be used to get confidential information past a human downgrader
(least-significant bits in digital photos, variations of punctuation/spelling/whitespace
in plaintext, etc.).
User Interface Failures
Humans
“Humans are incapable of securely storing high-quality
cryptographic keys, and they have unacceptable speed and
accuracy when performing cryptographic operations. (They are
also large, expensive to maintain, difficult to manage, and they
pollute the environment. It is astonishing that these devices
continue to be manufactured and deployed. But they are
sufficiently pervasive that we must design our protocols around
their limitations.)”
−− C. Kaufman, R. Perlman, and M. Speciner.
Network Security: PRIVATE Communication in a PUBLIC World.
2nd edition. Prentice Hall, page 237, 2002.
Humans are weakest link
Most security breaches attributed to “human
error”
Social engineering attacks proliferate
Frequent security policy compliance failures
Automated systems are generally more
predictable and accurate than humans
Why are humans in the loop at all?
Don’t know how or too expensive to
automate
Human judgments or policy decisions
needed
Need to authenticate humans
The human threat
Malicious humans who will attack system
Humans who are unmotivated to perform
security-critical tasks properly or comply
with policies
Humans who don’t know when or how to
perform security-critical tasks
Humans who are incapable of performing
security-critical tasks
Need to better understand humans in the loop
Do they know they are supposed to be doing
something?
Do they understand what they are supposed
to do?
Do they know how to do it?
Are they motivated to do it?
Are they capable of doing it?
Will they actually do it?
SSL Warnings
False Alarm Effect
“Detection system” ≈ “System”
If risk is not immediate, warning the user
will decrease her trust on the system
Patco Construction vs. Ocean Bank
Hacker stole ~$600K from Patco through Zeus
The transfer alarmed the bank, but ignored
“substantially increase the risk of fraud by asking for security
answers for every $1 transaction”
“neither monitored that transaction nor provided notice before
completed”
“commercially unreasonable”
 Out-of-Band Authentication
 User-Selected Picture
 Tokens
 Monitoring of Risk-Scoring Reports
28
Password Authentication
Definitions
Identification - a claim about identity
 Who or what I am (global or local)
Authentication - confirming that claims are true
 I am who I say I am
 I have a valid credential
Authorization - granting permission based on a valid
claim
 Now that I have been validated, I am allowed to access certain
resources or take certain actions
Access control system - a system that authenticates
users and gives them access to resources based on their
authorizations
 Includes or relies upon an authentication mechanism
 May include the ability to grant course or fine-grained
authorizations, revoke or delegate authorizations
 Also includes an interface for policy configuration and
management
Building blocks of authentication
Factors
Something you know (or recognize)
Something you have
Something you are
Two factors are better than one
Especially two factors from different categories
What are some examples of each of these
factors?
What are some examples of two-factor
authentication?
Authentication mechanisms
Text-based passwords
Graphical passwords
Hardware tokens
Public key crypto protocols
Biometrics
Evaluation
Accessibility
Memorability
Security
Cost
Environmental considerations
Typical password advice
Typical password advice
Pick a hard to guess password
Don’t use it anywhere else
Change it often
Don’t write it down
So what do you do when every web site you
visit asks for a password?
Problems with Passwords
Selection
– Difficult to think of a good password
– Passwords people think of first are easy to guess
Memorability
– Easy to forget passwords that aren’t frequently used
– Difficult to remember “secure” passwords with a mix of
upper & lower case letters, numbers, and special characters
Reuse
– Too many passwords to remember
– A previously used password is memorable
Sharing
– Often unintentional through reuse
– Systems aren’t designed to support the way people work
together and share information
Mnemonic Passwords
F
Four
s
score
and
and
sseven
seven
y
years
aago
, oour
First letter of each word (with
punctuation)
Substitute numbers for words
or similar-looking letters
Substitute symbols for words or
similar-looking letters
Source: Cynthia Kuo, SOUPS 2006
4sasya,oF
4sa7ya,oF
fsasya,oF
4sa7ya,oF
4s&7ya,oF
FFathers
The Promise?
Phrases help users incorporate different
character classes in passwords
Easier to think of character-for-word substitutions
Virtually infinite number of phrases
Dictionaries do not contain mnemonics
Source: Cynthia Kuo, SOUPS 2006
Mnemonic password evaluation
Mnemonic passwords are not a panacea for
password creation
No comprehensive dictionary today
May become more vulnerable in future
– Many people start to use them
– Attackers incentivized to build dictionaries
Publicly available phrases should be avoided!
Source: Cynthia Kuo, SOUPS 2006
Password keeper software
Run on PC or handheld
Only remember one password
Single sign-on
Login once to get access to all your
passwords
Biometrics
Fingerprint Spoofing
Devices
 Microsoft Fingerprint Reader
 APC Biometric Security device
Success!
 Very soft piece of wax flattened against hard surface
 Press the finger to be molded for 5 minutes
 Transfer wax to freezer for 10-15 minutes
 Firmly press modeling material into cast
 Press against the fingerprint reader
Replicated several times
Retina/Iris Scan
Retinal Scan
 Must be close to camera (IR)
 Scanning can be invasive
 Not User friendly
 Expensive
Iris Scan
 Late to the game
 Requires advanced
technology to properly
capture iris
 Users do not have to consent
to have their identity tested
Graphical passwords
“Forgotten password” mechanism
Email password or magic URL to address on file
Challenge questions
Why not make this the normal way to access
infrequently used sites?
Convenient SecureID 1
What problems does
this approach solve?
What problems does
it create?
Source:
http://worsethanfailure.com/Articles/Security_by_Oblivity.aspx
Convenient SecureID 2
What problems does
this approach solve?
What problems does
is create?
Previously available at:
http://fob.webhop.net/
Browser-based mutual
authentication
 Chris Drake’s “Magic Bullet” proposal
 http://lists.w3.org/Archives/Public/public-usableauthentication/2007Mar/0004.html
– User gets ID, password (or alternative), image, hotspot
at enrollment
– Before user is allowed to login they are asked to
confirm URL and SSL cert and click buttons
– Then login box appears and user enters username and
password (or alternative)
– Server displays set of images, including user’s image
(or if user entered incorrect password, random set of
images appear)
– User finds their image and clicks on hotspot
• Image manipulation can help prevent replay attacks
 What problems does this solve?
 What problems doesn’t it solve?
 What kind of testing is needed
Phishing
Spear Phishing (Targeted Phishing)
Personalized mail for a (small) group of targeted users
Employees, Facebook friends, Alumni, eCommerce Customers
These groups can be obtained through identity theft!
Content of the email is personalized.
Different from Viagra phishing/spam
Combined with other attacks
Zero-day vulnerability: unpatched
Rootkit: Below OS kernel, impossible to detect with AV software
Key logger: Further obtain ID/password
APT (Advanced Persistent Threat): long-term surveillance
53
Examples of Spear Phishing
54
Good Phishing example
55
Policy and Usability
Cost of Reading Policy Cranor et al.
TR= p x R x n
 p is the population of all Internet users
 R is the average time to read one policy
 n is the average number of unique sites Internet users visit
annually
p = 221 million Americans online (Nielsen, May
2008)
R = avg time to read a policy = # words in policy /
reading rate
 To estimate words per policy:
 Measured the policy length of the 75 most visited websites
 Reflects policies people are most likely to visit
Reading rate = 250 WPM Mid estimate: 2,514 words
/ 250 WPM = 10 minutes
n = number of unique sites per year
Nielsen estimates Americans visit 185 unique
sites in a month:
but that doesn’t quite scale x12, so 1462 unique
sites per year.
TR= p x R x n
= 221 million x 10 minutes x 1462 sites
R x n = 244 hours per year per person
P3P: Platform for Privacy Preferences
A framework for automated privacy
discussions
Web sites disclose their privacy practices in
standard machine-readable formats
Web browsers automatically retrieve P3P privacy
policies and compare them to users’ privacy
preferences
Sites and browsers can then negotiate about
privacy terms