TELE3119: TRUSTED NETWORKS WEEK 4 Never Never Stand

TELE3119: TRUSTED NETWORKS WEEK 4 Never Never Stand

TELE3119: TRUSTED NETWORKS WEEK 4 Never Never Stand Stand Still Still Faculty Faculty of of Engineering Engineering School School of of Electrical Electrical Engineering Engineering and and Telecommunications Telecommunications Course Coordinator: Prof. Aruna Seneviratne, Room EE 312

E-mail [email protected] Course web-page: https://subjects.ee.unsw.edu.au/tele3119/ 2 Basic View of Cryptography Confidentiality Authentication Integrity Access & Availability Cryptography Symmetric Asymmetric Protocols

Network Stream Ciphers Block Ciphers School of Electrical Engineering & Telecommunications Trusted Networks 3 Problem tor-and-https School of Electrical Engineering & Telecommunications 4 Problem cont. Extracting information regarding the original source of the packets and their ultimate destination is referred to as the traffic analysis attack

Internet surveillance Knowing who's talking to whom, when, and for how long e.g. military intelligence or counter-intelligence Reveals users privacy security concern! Encryption does not work, since packet headers still reveal a great deal about users TLS (packet headers are still in clear text) IPSec (packet sniffers at the origination point) End-to-end anonymity is needed One Possible Solution A distributed, anonymous network School of Electrical Engineering & Telecommunications 5 TOR Distributed Anonymous Network Tors genesis lies in the onion routing research that was funded by several US Government organizations starting in 1995

To prevent finding out who was talking to whom TOR, an acronym standing for The Onion Router School of Electrical Engineering & Telecommunications 6 What is Tor? ToR is a distributed anonymous communication service using an overlay network that allows people and groups to improve their privacy and security on the Internet Tor is based on : Onion Proxies (OP) Fetches a Tor directory for the IP addresses of the ORs and constructs a path through the Tor overlay Onion Routers (OR) Route Traffic The user then selects a subset of these ORs, commonly just 3, for constructing a path to the destination

Uses TCP with TLS School of Electrical Engineering & Telecommunications 7 TOR Project School of Electrical Engineering & Telecommunications 8 Tor Architecture B, C, and D are the ORs selected by user A for a path to the destination E School of Electrical Engineering & Telecommunications 9 TOR Operation A users OP constructs a path through the ToR overlay Every OR node has a public RSA key that it makes

available to the users OP This path constitutes a circuit The two parties at the two end of a circuit may use it for an arbitrary number of TCP streams School of Electrical Engineering & Telecommunications 10 How ToR Works 1 Tor software comes with a built-in list of location and public key for each directory authority. Directory authorities Dedicated servers which tell Tor clients which relays make up the Tor

network. Every hour, directory authorities vote on and reach a consensus on the relays that make up the Tor network https://metrics.torproject.o rg/rs.html#search/flag:aut hority School of Electrical Engineering & Telecommunications Trusted Networks 11 How ToR Works 2 Choose the path for each new circuit before building Choose the exit node first, followed by the other nodes in the circuit.

All paths generated obey the following constraints: Do not choose the same router twice School of Electrical Engineering & Telecommunications for the same path. Do not choose any router in the same family as another in the same path. (Two routers are in the same family if each one lists the other in the "family" entries of its descriptor.) Do not choose more than one router in a given /16 subnet. Do not choose any non-running or non-valid router unless configured to

do so. The first node must be a Guard node. Trusted Networks Summary of How TOR Works 12 Operating System Session Key 3 Browser Session Key 2 Tor Proxy Session Key 1 Destination Server Node to Node Key

Network Card Guard Node Middle Node Node to Node Key Exit Node Node to Node Key School of Electrical Engineering & Telecommunications Network Card 13

Final - ToR Works For efficiency, School of Electrical Engineering & Telecommunications Tor software uses the same circuit for connections that happen within the same ten minutes or so. Later requests are given a new circuit, to keep people from linking your earlier actions to the new ones. Trusted Networks 14

Two Questions? 1) Can the exit node operator see the source IP address, meaning the IP address of node A in our example? 2) Can the exit node operator see the data payload of the source packet? School of Electrical Engineering & Telecommunications 15 How Secure is ToR? School of Electrical Engineering & Telecommunications Trusted Networks 16 Correlation School of Electrical Engineering & Telecommunications

Trusted Networks 17 How Secure is ToR? Source IP protection? In principle, that should not be possible INRIA researchers* were able to reveal 10,000 source IP of hosts that used BitTorrent during a period of 23 days in 2011. Took advantage of unusual feature of BitTorrent (P2P protocol) BitTorrent client must somehow acquire a list of the peers that are the keepers of the media content that the client wishes to download and then, subsequently, join the peers By contacting a centralized tracker that keeps a list of all the peers currently in possession of the media content of interest to the client; No encryption * S. Le Blond etal, One bad apple spoils the bunch: exploiting P2P applications to trace and profile Tor users , LEET'11 Proceedings of the 4th USENIX conference on Large-scale exploits and emergent threats. School of Electrical Engineering & Telecommunications

18 Summary How Can Secure and Private Communications be Provided? ToR Shortcomings of ToR https://torstatus.blutmagie.de/ School of Electrical Engineering & Telecommunications Trusted Networks 19 Hidden Services C S A

C S C S C B School of Electrical Engineering & Telecommunications Trusted Networks S 20 Hidden Services C S

A C S C S HS C School of Electrical Engineering & Telecommunications Trusted Networks 21 Operation Key to the DHT IP

Onion Address Derived from the PK of the server IP S C IP Descriptor School of Electrical Engineering & Telecommunications IP Introduction Points Trusted Networks 22 Hidden Services cont. The descriptors are stored in a Distributed Hash Table (DHT)

IP wait acting has ORs. The client chooses a Rendezvous Point (RP) Send message to RP requesting an introduction to the server RP forward the request to an IP, with a token and its contact details IP forwards to the server with the token The server decides to connect to the RP (with the token) or ignore School of Electrical Engineering & Telecommunications Trusted Networks 23 Hashing Algorithm Calculation applied to a key to transform it into an address For numeric keys, divide the key by the number of available addresses, n and take the remainder Address = key mod n For alphanumeric keys, divide the sum of ASCII codes in a key by the number of available addresses, n, and take the remainder

Problem Collisions Two methods Open Addressing Techniques Separate Chaining School of Electrical Engineering & Telecommunications Trusted Networks 24 Collision Resolution Open Addressing Colliding keys are stored in an unused slot Do a linear search to see if the key exists H(x) 0 10

1 1 65 4 2 99 3 88 3 77 3 77

10 3 99 4 65 5 88 School of Electrical Engineering & Telecommunications 77 cannot use 3, so store in next unused slot 88 cannot use 3, so store in next unused slot

Trusted Networks 25 Collision Resolution - Chaining Combines an array and linked lists Each element of array is a linked list Each list is called a chain Each element of the array is called a bucket When a collision occurs, simply append the chain in to the appropriate bucket H(x) 0 10 1 1 65

4 99 3 88 3 77 3 10 2 3 99 4

65 88 77 5 School of Electrical Engineering & Telecommunications Trusted Networks 26 Searching for a Key Pass the key to the hash function to get hash index of the bucket containing the chain where the key may be found Do a linear search on the chain to see if the key exists If found return true, false otherwise Problem Long chains cause the hash table to degenerate into a linear

search Depending on the implementation of the linked list, could cause insertion/removal/searching to degenerate form O91) to O(n) School of Electrical Engineering & Telecommunications Trusted Networks 27 Solution Avoid collisions as much as possible Improve the hash function Use a bigger hash table array Goal: Keys distributed evenly among buckets All chains should be of the same length School of Electrical Engineering & Telecommunications Trusted Networks 28

DHT DHTs Hash buckets mapped to (p2p) nodes Query Query Hit School of Electrical Engineering & Telecommunications Trusted Networks 29 Hidden Services Operation cont. Key to the DHT IP

Onion Address Derived from the PK of the server IP C S Token, Add Token, Add Token RP IP Descriptor School of Electrical Engineering & Telecommunications IP Introduction Points

Trusted Networks 30 Aside Some servers do not care whether their services are known Aim is protecting their customers Facebook, New York Times Face book worked with ToR to create a what they called a single onion Does not bother with the routers on the way to the RP School of Electrical Engineering & Telecommunications Trusted Networks 31 Summary How hidden services be Provided?

Use of ToR for providing hidden services School of Electrical Engineering & Telecommunications Trusted Networks 32 Basic View of Cryptography Confidentiality Authentication Integrity Access & Availability Cryptography Symmetric Stream Ciphers

Asymmetric Applications Block Ciphers School of Electrical Engineering & Telecommunications Trusted Networks 33 Problem The personal information collected from the sensors, and use of mobile devices Provision of personalised services to the users Personalisation comes at a cost to users security and privacy My favourite TV shows Shoes I might buy My Marital status

Some slides have been taken from: Christine Task, A Practical Beginners' Guide to Differential Privacy . School of Electrical Engineering & Telecommunications Trusted Networks 34 Privacy vs. security Privacy: what information goes where? Security: protection against unauthorized access Security helps enforce privacy policies Can be at odds with each other e.g., invasive screening to make us more secure against terrorism School of Electrical Engineering & Telecommunications 35

Multiple Risks? Device Apps, and traffic Inference attacks External Organizations share data with others Inference Attack Internal An analysist wants low friction access to data Insider Attack. School of Electrical Engineering & Telecommunications

36 Problem How to safe guard the security and privacy of the users, whilst still providing the full benefits of personalized services Provide information to users to make them informed decisions :utility vs. loss of security/ privacy Methods of extracting information whilst guaranteeing security and privacy: privacy preserving analytics School of Electrical Engineering & Telecommunications Trusted Networks 37 Alternatives? Anonymization E.g. Do not use real names Encryption

NOYB (European Center for Digital Rights) flyByNight (Mitigating the privacy risks of social networking) Decentralization Tighter control over data School of Electrical Engineering & Telecommunications 38 Today: Users in the Dark Data Sources Apps Trackers GPS Location Installed Apps Device IDs Bowser History

School of Electrical Engineering & Telecommunications Trusted Networks 39 Example - #1 It is possible to identify user traits very easily A single snapshot of apps installed on a smartphone! Apptronomy Upon installation, lists and uploads the user installed apps to a server Generates a random ID for that installation instance Group of volunteers and users through Amazon Mechanical Turk User traits through a brief questionnaire Crawled two popular social app discovery sites: Appbrain and Appaware School of Electrical Engineering & Telecommunications Trusted Networks

40 Example - #1.1 Trained SVM classiers app description as the input and predict whether the given app is relevant to that particular trait Installed apps in smartphones can infer user traits S. Seneviratne, A. Seneviratne, P. Mohapatra, A. Mahanti, Predicting User Traits From a Snapshot of Apps Installed on a Smartphone, ACM Mobile Computer Communication Review, April, 2014. School of Electrical Engineering & Telecommunications Trusted Networks 41 Example #2 A few know a lot Identified the top-100 free and paid apps from four countries representing four geographical regions 275 unique free and 234 unique paid apps

For all the apps found in users app downloaded the APK files - 3,605 Two analysis tools to identify the embedded trackers and the API calls executed by the trackers Permissions are abstract and may not necessarily represent the full implications School of Electrical Engineering & Telecommunications Trusted Networks 42 Example #2.1 S. Seneviratne,H. Kolumunna A. Seneviratne,A Measurement Study on Tracking in Paid Mobile Applications NICTA Technical Report 2015-8, February, 2015 School of Electrical Engineering & Telecommunications Trusted Networks

43 Single User Crashlytics Jibro Nexage Yozio Bugsense Vungle Pintrest Crittercism Candy Crush Saga Gmail

Adjust Kochava Pet Rescue Saga Comscore Nativex Android ID InMobi Location Mobileaptracker Tango Google Analytics Flurry

Subway Surf Chartboost Appsflyer MapQuest Despicable Draw Something Me Free Avast Mobile ThreatMetrix Security Trialpay Hockeyapp Trackers Tapjoy

Apps Google Ads SongPop Calendar Installed Apps Mopub mDotm GreyStripe Millennial Media Personal Information 11 apps exposed 26 trackers !! School of Electrical Engineering & Telecommunications Trusted Networks

44 Privmetrics https://youtu.be/6UiKE5APckw School of Electrical Engineering & Telecommunications Privacy Risks in Analytics How many trip were taken in Sydney last year How many trips Aruna took in Sydney last year Reflects a trend Reflects an individual Not a security problem Cryptography does not provide a solution

Access Control also does not provide a solution School of Electrical Engineering & Telecommunications Security Trusted Networks Anonymisation: Not a Solution Limited utility Re-identification attacks Netflix prize1 NYC taxi data2 1 (Narayanan etal) https://arxiv.org/abs/cs/0610105 2 (Anthony Tockar) Riding with the stars: Passenger privacy in the nyc taxicab database

School of Electrical Engineering & Telecommunications Security Trusted Networks 47 Threat: deanonymization User Movie Rating 1234 Rocky II 3/5 1234 The Wizard 4/5

1234 The Dark Knight 5/5 1234 Girls Gone Wild User Movie Rating dukefan The Wizard 8/10 dukefan The Dark Knight 10/10

dukefan Rocky II 6/10 5/5 User 1234 is dukefan! School of Electrical Engineering & Telecommunications 48 Summary Privacy leakage at the application level Anonymisation, Encryption and decentralisation does not help School of Electrical Engineering & Telecommunications Trusted Networks

49 Basic View of Cryptography Confidentiality Authentication Integrity Access & Availability Cryptography Symmetric Asymmetric Applications Differential Privacy Stream Ciphers

Block Ciphers School of Electrical Engineering & Telecommunications Trusted Networks 50 Many Options Data Analysis Data Users Data Set Data

Results Data Pop Pop I is the subset of that actually submit a survey I Di ={ | } The dataset collected from all the people School of Electrical Engineering & Telecommunications Q( Privatized query run on the data

and R is the result 51 What do we want One would feel safe inputting information if My data had no impact on the released results Any attacker looking at the published results, R couldnt learn (with a high probability) any new information about the data provider School of Electrical Engineering & Telecommunications Q Prob(secret(me) | R) = Prob(secret(me))

52 Why cant we have it? If individual answers has no impact on the released results, results would have no utility Q(D(I-me)) = Q(D1) => Q(D1) Q(DO) If R shows there is a strong trend in my population , with high probability, trend is true of me too (even if I dont submit a survey) Prob(secret(m) | secret (Pop) > Prob (secret(me)) If an attacker knows a function about me that dependent on general facts about the population Releasing those general facts gives the attacker specific information about me

School of Electrical Engineering & Telecommunications 53 Consequences Cannot guarantee that My data will not affect the results Attacker will not be able to learn new information about me from looking at the results Possibility Released result R was nearly the same, whether or not my data is included A is much larger than 1, very little privacy A = 1 individuals have no effect, therefore zero utility School of Electrical Engineering & Telecommunications 54 Differential Privacy The chance that the noisy released result will be R

is nearly the same, whether or not you submit your information and small > 0 Given R, how can anyone guess which possible world it came from? Prob(R)= B Possible world where I submitted R A~ =B School of Electrical Engineering & Telecommunications Prob(R)= A Possible world where I did not

submit Trusted Networks 55 Differential Privacy A randomized computa on M provides differen al privacyif for any data sets A and B that differ on a single element and any set of possible outcomes, the probabili es of having a specific outcome are close. A B M M M(A)

M(B) School of Electrical Engineering & Telecommunications Trusted Networks 56 But still An attacker can tell whether or not you submitted data With the right background information, an attacker can learn about you just from the general information about the population, even if you dont submit any data Attacker may be able to guess whether you submitted data Hides data sets that differ by one individual not whole groups School of Electrical Engineering & Telecommunications 57

What Differential Privacy Provides Ensures that the released result R gives minimal evidence about whether or no any given individual contributed data Protects all personal information in the data set Does not prevent attackers from drawing conclusions about individuals from aggregate results over the population Can learn information about known cohesive groups School of Electrical Engineering & Telecommunications 58 Global Sensitivity Two data sets D1 and D2 that differ in exactly one person F(D) = X - deterministic, non privatised function over the data set D X vector of k real numbers Global Sensitivity How many males & females =1

+ how many people liked Z =2 School of Electrical Engineering & Telecommunications 59 Laplacian Noise In order for our two worst-case neighbouring data sets to produce a similar distribution of privatised answers Add noise Random values taken form a Laplacian distribution with standard deviation large enough to cover the gap Probability Prob(R=x|D is the true world) Random Variable

School of Electrical Engineering & Telecommunications 60 How does it work Distribution of possible results from any data set overlaps heavily with the distribution of results from its neighbours Know the general neighbourhood of right answer R School of Electrical Engineering & Telecommunications 61 Proof (1) Substituting the equations from adding Laplacian noise to the function (slide 25)

2 2 ( ) ( + 1) School of Electrical Engineering & Telecommunications 62

Proof (2) Is the maximum difference between two neighbouring data sets So we get the following , with School of Electrical Engineering & Telecommunications 63 Questions How do you privatise a histogram with three partitions? Add Laplacian noise calibrated to = 1 to each partition How do you privatise a series of five overlaping counts across a data set (how many people in the data set are female?, how many like pizza, how many are aged between 12-16.) Add Laplacian noise calibrated to = 5 to each partition

How do you privatise an interactive query Limit the number of questions. If the number is two add Laplacian noise calibrated to = 2. How do you privatise a query whose sentivity depends on the number of people in the dataset? How many friends do you have that also took a survey? Unbounded sensitivity School of Electrical Engineering & Telecommunications 64 Summary Possible way of providing provable privacy guarantees Differential Privacy School of Electrical Engineering & Telecommunications Trusted Networks

Recently Viewed Presentations

  • What's New in SQL Server 2005?

    What's New in SQL Server 2005?

    Filestream Storage. Check out Paul Randal's Blog for the performance graphs. Filestream reads better than varbinary. Filestream inserts perform better for files over ≈2MB for Win32
  • Cutnell/Johnson Physics 7th edition Classroom Response System Questions

    Cutnell/Johnson Physics 7th edition Classroom Response System Questions

    The height of each box is 0.30 m. The center of gravity of each of the boxes is located at its center. In preparing to pull the hand truck up the ramp of the moving truck he rotates it to...
  • PMD Pro 1 - Ning

    PMD Pro 1 - Ning

    7. PMD Pro principle explaining the project management is repeating a process with the aim of approaching a desired goal, target or result. 8. tool for planning scope. 9. final phase of the project. 10. method for linking financial information...
  • Title With Picture Layout - OCASC

    Title With Picture Layout - OCASC

    Brookfield HS $100NAC. Carine Wilson $90 Chateau Laurier. Lisgar $125 Hotel. LDHSS $80Orchard View Banquet Centre (Greely/Manotick) Sir Guy SS $30Sir Guy H.S. South Carleton $85NAC. 2017 noted from SSC Meeting Nov 18/16. South Carleton June 10th - $80-$90 NAC....
  • The Global Circuit Intensity: Its measurement and variation

    The Global Circuit Intensity: Its measurement and variation

    Soundings of Vi. Should be taken in clean-air ocean regions with no clouds or only scattered clouds. Electric field decreases quasi-exponentially with height, so soundings do not need to reach high altitudes
  • TEKS - University of North Texas

    TEKS - University of North Texas

    TEKS (2) Listening/speaking/culture. The student listens and speaks to gain knowledge of his/her own culture, the culture of others, and the common elements of cultures. The student is expected to: (A) connect experiences and ideas with those of others through...
  • Fig. 2.15 Fats are mostly for storing energy

    Fig. 2.15 Fats are mostly for storing energy

    Neutral Fats Sterol Lipids Phospholipids Fig. 2.22 Formed when a Glycerol Molecule undergoes a dehydration synthesis reaction with Fatty Acid Chains. If only one chain joins on = Monoglyceride If two chains join on = Diglyceride If three chains join...
  • Capital Health MOU

    Capital Health MOU

    Capital Health Medical Outpatient Unit April 12th 2007 Final Presentation Kyle Brown Hillary Higgins Justin Lang Outline Overview MOU The Problem Methodology Analysis and Application Academic Research Survey Results Simulation Design Layout ST Recommendations Implementations LT Recommendations Implementations Implementation Steps...