CMSC 414 Computer and Network Security Lecture 7 Jonathan Katz Public-key cryptography The public-key setting A party (Alice) generates a public key along with a matching secret key (aka private key) The public key is widely distributed, and is assumed to be known to anyone (Bob) who wants to communicate with Alice – We will discuss later how this can be ensured Alice’s public key is also known to the attacker! Alice’s secret key remains secret Bob may or may not have a public key of his own The public-key setting pk c = Encpk(m) pk c = Encpk(m) Private- vs. public-key I Disadvantages of private-key cryptography – Need to securely share keys • What if this is not possible? • Difficult to distribute/manage keys in a large organization • Private-key crypto requires private+authenticated channel to share keys; public-key crypto requires authenticated channel – Consider communication in an n-party network: each party must store n-1 secret keys – Inapplicable in open systems (i.e., e-commerce) • Need to know in advance the parties with whom you will communicate • Not suited for “many-to-one” communication Private- vs. public-key II Why study private-key setting at all? – Private-key is orders of magnitude more efficient – Private-key still has domains of applicability • Military settings, disk encryption, … • Bidirectional, “one-to-one” authenticated communication – Public-key crypto is “harder” to get right • Need stronger assumptions, easier to attack – Can combine private-key primitives with public-key techniques to get the best of both (for encryption) • So still need to understand the private-key setting! – Can distribute keys using trusted entities (KDCs) Private- vs. public-key III Public-key cryptography is not a cure-all – Still requires authenticated distribution of public keys • May (sometimes) be just as hard as sharing a key • Technically speaking, requires only an authenticated channel instead of an authenticated + private channel – Not clear with whom you are communicating (unless the sender has a public key) – Can be too inefficient for certain applications Cryptographic primitives Private-key setting Public-key setting Confidentiality Private-key encryption Public-key encryption Integrity Message authentication codes Digital signature schemes Public-key encryption Key-generation algorithm: randomized algorithm that outputs (pk, sk) Encryption algorithm: – Takes a public key and a message (plaintext), and outputs a ciphertext; c Epk(m) Decryption algorithm: – Takes a private key and a ciphertext, and outputs a message (or perhaps an error); m = Dsk(c) Correctness: for all (pk, sk), Dsk(Epk(m)) = m Security? Just as in the case of private-key encryption, but the attacker gets to see the public key pk – For all m0, m1, no adversary running in time T, given pk and an encryption of m0 or m1, can determine the encrypted message with probability better than 1/2 + Public-key encryption must be randomized (even to achieve security against ciphertext-only attacks) In the public-key setting, security against ciphertext-only attacks implies security against chosen-plaintext attacks – And security for encryption of multiple messages El Gamal encryption We have already (essentially) seen one encryption scheme: Receiver p, g Sender p, g, hA = gx hA = g.x mod p c = (KBA m) mod p hB, c hB = gy mod p KAB = (hB)x KBA = (hA)y El Gamal encryption (Some aspects of the actual scheme are simplified) Key generation – Choose a large prime p, and an element g Zp* – Choose random x {0, …, p-2}, set h=gx – The public key is (p, g, h), and the private key is x Encryption – View the message m as an element of Zp* – Choose random r {0, …, p-2} – The ciphertext is (gr, hr m) To decrypt ciphertext (c1, c2) output c2/c1x Security? Security of El Gamal encryption is equivalent to the decisional Diffie-Hellman assumption Best known algorithm for decisional Diffie- Hellman in Zp* runs in time ≈ exp(log p1/3) – So if p is a 1024-bit prime, best current attack on El Gamal encryption requires time ≈ 260 In other groups, the Diffie-Hellman problem is currently ‘harder’ – E.g., for elliptic curve groups, best current algorithms require time exp(log |G|/2) – Can use ~120-bit group elements to get 260 security RSA background N=pq, p and q distinct, odd primes (N) = (p-1)(q-1) – Easy to compute (N) given the factorization of N – Hard to compute (N) without the factorization of N Fact: for all x ZN*, it holds that x(N) = 1 mod N – Proof: take CMSC 456! If ed=1 mod (N), then for all x it holds that (xe)d = x mod N I.e., given d, we can compute eth roots We have an asymmetry! Let e be relatively prime to (N) – Needed so that ed=1 mod (N) has a solution Given e and the factors of N, can compute d and hence compute eth roots Without the factorization of N, no apparent way to compute eth roots Let’s use this to encrypt… Hardness of computing eth roots? The RSA problem: – Given N, e, and c, compute c1/e mod N If factoring is easy, then the RSA problem is easy We know of no other way to solve the RSA problem besides factoring N – But we do not know how to prove that the RSA problem is as hard as factoring The upshot: we believe factoring is hard, and we believe the RSA problem is hard How hard is factoring? Best current algorithms for factoring N=pq a product of two equal-length primes, run in time ≈ exp(log N1/3) So need |N| ≈ 1024 for reasonable security Currently |N| ≈ 2048 recommended for good security margins RSA key generation Generate random primes p, q of sufficient length Compute N=pq and (N) = (p-1)(q-1) Compute e and d such that ed = 1 mod (N) – e must be relatively prime to (N) – Typical choice: e = 3; other choices possible Public key = (N, e); private key = (N, d) “Textbook RSA” encryption Public key (N, e); private key (N, d) To encrypt a message m ZN*, compute c = me mod N To decrypt a ciphertext c, compute m = cd mod N Correctness clearly holds… …what about security? Textbook RSA is insecure! It is deterministic! Furthermore, it can be shown that the ciphertext leaks specific information about the plaintext Padded RSA Introduce randomization… Public key (N, e); private key (N, d) – Say |N| = 1024 bits To encrypt m {0,1}895, – Choose random r {0,1}128 – Compute c = (r | m)e mod N Decryption done in the natural way… Essentially this is standardized as PKCS #1 v1.5 (since superseded)