Key Management

Shared Key Exchange Problem How do Alice and Bob exchange a shared secret? Offline – Doesn’t scale Using public key cryptography (possible) Using specially crafted messages (Diffie Hellman) Using a trusted third party (KDC) – Secrets should never be sent in clear – We should prevent replay attacks – We should prevent reuse of old keys

Exchange a secret with someone you never met while shouting in a room full of people Alice and Bob agree on g and large n Alice chooses random a, sends Bob chooses random b, sends Alice takes Bob’s message and calculates Bob does the same; now they both know shared secret Diffie Hellman Key Exchange

Building up to Needham Schroeder/Kerberos User sends req. to KDC (key distrib. center) KDC generates a shared key: K c,s Keys K KDC,C and K KDC,S are preconfigured No keys ever traverse net in the clear Why are identities in tickets? KDC Based Key Distribution C C KDC S S 3. EK KDC,S {C, K c,s } 2. EK KDC,C {S, K c,s } 1. C, S ticket

KDC does not have to talk both to C and S Messages 2 or 3 can be replayed by M – Force C and S to use same secret for a long time – Cause S to have an old ticket, break comm. w C KDC Based Key Distribution C C KDC S S ticket S = EK KDC,S {C, K c,s } 2. EK KDC,C {S, K c,s }, ticket S 1. C, S 3. ticket S

Use nonces to prevent replay attacks Needham-Shroeder Key Exchange C C KDC S S ticket S = EK KDC,S {C, K c,s } 2. EK KDC,C {N 1, S, K c,s, ticket S } 1. N 1, C, S 3. EK C,S {N 2 }, ticket S 4. EK C,S {N 2 -1, N 3 } 5. EK C,S {N 3 -1}

What happens if attacker gets session key? – Can reuse old session key to answer challenge- response, generate new requests, etc – Need timestamps to ensure freshness = tickets expire after some time Problem

Public Key Exchange Problem How do we verify an identity: – Alice sends to Bob her public key Pub(A) – Bob sends to Alice his public key Pub(B) – How do we ensure that those keys really belong to Alice and Bob? Need a trusted third party

Man-in-the-Middle Attack On Key Exchange Alice sends to Bob her public key Pub(A) Mallory captures this and sends to Bob Pub(M) Bob sends to Alice his public key Pub(B) Mallory captures this and sends to Alice Pub(M) Now Alice and Bob correspond through Mallory who can read/change all their messages

Public key is public but … – How does either side know who and what the key is for? Does this solve key distribution problem? – No – while confidentiality is not required, integrity is Still need trusted third party – Digital certificates – certificate authority (CA) signs identity+public key tuple with its private key – Problem is finding a CA that both client and server trust Public Key Exchange

Digital Certificates Everyone has Trent’s public key Trent signs both Alice’s and Bob’s public keys – he generates public-key certificate When they receive keys, verify the signature Mallory cannot impersonate Alice or Bob because her key is signed as Mallory’s Certificate usually contains more than the public key – Name, network address, organization Trent is known as Certificate Authority (CA)

Authentication steps – Alice provides nonce, or a timestamp is used instead, along with her certificate. – Bob selects session key and sends it to Alice with nonce, encrypted with Alice’s public key, and signed with Bob’s private key. He sends his certificate too – Alice validates certificate – it is really Bob’s key inside – Alice checks signature and nonce – Bob really generated the message and it is fresh Certificate-Based Key Exchange

Pretty Good Privacy – “Web of Trust” – Public key, identity association is signed by many entities – Receiver hopefully can locate several signatures that he can trust – Like an endorsement scheme PGP

User keys installed on server out of band – User logs in with a password – Copies her public key onto server Weak assurance of server keys – User machine remembers server keys on first contact – Checks if this is still the same host on subsequent contact – But no check on first contact SSH

Revocation lists (CRL’s) – Long lists – Hard to propagate Lifetime / Expiration – Short life allows assurance of validity at time of issue Real time validation – Receiver of a certificate asks the CA who signed it if corresponding private key was compromised – Can cache replies Recovery From Stolen Private Keys

Authentication and Identity Management

Ideally – Who you are Practically – Something you know (e.g., password) – Something you have (e.g., badge) – Something about you (e.g., fingerprint) Basis for Authentication

Password Authentication Alice inputs her password, computer verifies this against list of passwords If computer is broken into, hackers can learn everybody’s passwords – Use one-way functions, store the result for every valid password – Perform one-way function on input, compare result against the list

Password Authentication Hackers can compile a list of frequently used passwords, apply one-way function to each and store them in a table – dictionary attack Host adds random salt to password, applies one-way function to that and stores result and salt value – Randomly generated, unique and long enough

Password Authentication Someone sniffing on the network can learn the password Lamport hash or S-KEY – time-varying password – To set-up the system, Alice enters random number R – Host calculates x 0= h(R), x 1= h(h(R)), x 2= h(h(h(R))),..., x 100 – Alice keeps this list, host sets her password to x 101 – Alice logs on with x 100, host verifies h(x 100 )=x 101, resets password to x 100 – Next time Alice logs on with x 99

Password Authentication Someone sniffing on the network can learn the password – Host keeps a file of every user’s public key – Users keep their private keys – When Alice attempts to log on, host sends her a random number R – Alice encrypts R with her private key and sends to host – Host can now verify her identity by decrypting the message and retrieving R

Key Distribution – Confidentiality not needed for public key – Can be obtained ahead of time Performance – Slower than conventional cryptography – Implementations used for key distribution, then use conventional crypto for data encryption Trusted third party still needed – To certify public key – To manage revocation Public Key Authentication

Passport Shibboleth Single Sign-On

Goal is single sign-on – Solves problem of weak or repeated user/pass combinations Implemented via redirections – Users authenticate themselves to a common server, which gives them tickets Widely deployed by Microsoft – Designed to use existing technologies in servers/browsers (HTTP redirect, SSL, cookies, Javascript) Passport

Client (browser), merchant (Web server), Passport login server Passport server maintains authentication info for client – Gives merchant access when permitted by client How Passport Works David P. Kormann and Aviel D. Rubin, Risks of the Passport Single Signon Protocol, Computer Networks, Elsevier Science Press, volume 33, pages 51-58, 2000.

How Passport Works David P. Kormann and Aviel D. Rubin, Risks of the Passport Single Signon Protocol, Computer Networks, Elsevier Science Press, volume 33, pages 51-58, SSL Token = encrypted authentication info using key merchant shares with passport server Also set cookie at browser (passport)

Placed into browser cache by servers to store state about this particular user – Contain any information that server wants to remember about the user as name/value pairs – May contain expiration time – May persist across browser instances Returned to server in clear on new access Only those cookies created for the server’s domain are sent to the server – May not be created by this server Usually used for persistent sign in, shopping cart, user preferences How Cookies Work

User logs in using her user/pass – Server sets a cookie with some info – username, password, session ID … – Any future accesses return this info to the server who uses it for authentication (equivalent to user/pass) – Once user signs out the cookie is deleted and the session closed at the server Problems – Cookies can be sniffed, remain on the browser because user did not sign out, be stolen by cross-site scripting or via DNS poisoning Solutions: – Send cookies over SSL, use timed cookies, secure code, bind cookies to IP address of the client, encrypt cookies … Cookies for Authentication Learn more at:

Service Provider – Browser goes to Resource Manager who uses WAYF, and user’s Attribute Requester, and decides whether to grant access. “Where are you from” (WAYF) service – Redirects to correct servers Federation to form trusted relationships between providers Federated Identity - Shibboleth

6. I know you now. Redirect to SP, with a handle for user 8. Based on attribute values, allow access to resource Identity Provider (IdP) Web Site Service Provider (SP) Web Site 1. User requests resource 2. I don’t know you, or where you are from LDAP WAYF 3. Where are you from? 4. Redirect to IdP for your org 5. I don’t know you. Authenticate using your org’s web login I don’t know your attributes. Ask the IdP (peer to peer) 6 Client Web Browser 8 Source: Kathryn Huxtable 10 June

Cards – Mag stripe (= password) – Smart card, USB key – Time-varying password Issues – How to validate – How to read (i.e. infrastructure) Something You Have

Biometrics – Measures some physical attribute Iris scan Fingerprint Picture Voice Issues – How to prevent spoofing – What if spoofing is possible? No way to obtain new credentials Something About You

Require at least two of the classes we mentioned, e.g. – Smart card plus PIN – RSA SecurID plus password – Biometric and password Multi-factor Authentication

Authorization and Policy

Is principal P permitted to perform action A on object O? – Authorization system will provide yes/no answer Authorization

Who is permitted to perform which actions on what objects? Access Control Matrix (ACM) – Columns indexed by principal – Rows indexed by objects – Elements are arrays of permissions indexed by action In practice, ACMs are abstract objects – Huge and sparse – Possibly distributed Access Control

Example ACM File/UserTomDickHarry Readme.txtread read, write passwordswrite Term.exeread, write, execute

Access Control Lists (ACLs) – For each object, list principals and actions permitted on that object – Corresponds to rows of ACM Instantiations of ACMs File Readme.txtTom: read, Dick: read, Harry: read, write passwordsHarry: write Term.exeTom: read, write, execute

Capabilities – For each principal, list objects and actions permitted for that principal – Corresponds to columns of ACM The Unix file system is an example of…? Instantiations of ACMs User TomReadme.txt: read, Term.exe: read, write, execute DickReadme.txt: read HarryReadme.txt: read, write; passwords: write

Discretionary Mandatory Role-based Types of Access Control

Owners control access to objects Access permissions based on identity of subject/object E.g., access to health information Discretionary Access Control

Rules set by the system, cannot be overriden by owners Each object has a classification and each subject has a clearance (unclassified, classified, secret, top-secret) Rules speak about how to match categories and classifications – Access is granted on a match Mandatory Access Control

Ability to access objects depends on one’s role in the organization Roles of a user can change – Restrictions may limit holding multiple roles simultaneously or within a session, or over longer periods. – Supports separation of roles Maps to organization structure Role-Based Access Control

Final goal of security – Determine whether to allow an operation Depends upon – Policy – Authentication Authorization

Policy defines what is allowed and how the system and security mechanisms should act Policy is enforced by mechanism which interprets it, e.g. – Firewalls – IDS – Access control lists Implemented as – Software (which must be implemented correctly and without vulnerabilities) Policy

Focuses on controlled access to classified information and on confidentiality – No concern about integrity The model is a formal state transition model of computer security policy – Describes a set of access control rules which use security classification on objects and clearances for subjects To determine if a subject can access an object – Combine mandatory and discretionary AC (ACM) – Compare object’s classification with subject’s clearance (Top Secret, Secret, Confid., Unclass.) – Allow access if ACM and level check say it’s OK Policy models: Bell-LaPadula

Mandatory access control rules: – a subject at a given clearance may not read an object at a higher classification (no read-up) – a subject at a given clearance must not write to any object at a lower classification (no write-down). Trusted subjects – the “no write-down” rule does not apply to them – Transfer info from high clearance to low clearance Policy models: Bell-LaPadula

Life-Experience Passwords (LEPs)

Strong passwords are easily forgotten Weak passwords are easily broken Users reuse passwords at different sites This holds for non-textual passwords too, plus they are more difficult to use Problem with passwords memorability guessability

Use memories from a user’s past – At least 2 years in the past Collect factoids – time, locations, people, activities, conversations – No preferences, no opinions Turn this into Q & A pairs – Questions become prompts – Answers become LEP Life-experience Passwords

How to collect memories, needs to be user-friendly – “Tell me a story” vs Q & A How to mine for useful data – Using natural language processing, hard in general How to detect weak factoids – E.g. relationships vs names, empty stories How to avoid use of sensitive info in LEPs How to deal with synonyms, misspellings, etc. How to store these passwords using one-way hashes Challenges

User study Users are asked to create 3 LEPs and 3 3class8 pass. Authenticate 1 week later – one attempt Measure memorability, strength, diversity of pass Ask a friend to try to guess passwords (Friend Free)

LEPs are strong Most LEPs 3+ factoids. Up to 10 in free-form Up to 20 in guidedform

LEPs are memorable LEP 3class8

LEPs are not easy to guess LEP 3class8 passwords were guessed 4.5% of time, even by acquaintances Guessed only by close friends and spouses

Issues LEPs took 10x longer to create and input How do we store LEPs? – Hash per answer Easy to break by guessing most likely answers first – Hash per LEP User must recall all factoids – Several hashes of strong combinations of factoids No feedback to user what they missed

Try LEPs out Earn 2 participation points, have fun and help us collect more data