Presentation on theme: "Cryptography, Attacks and Countermeasures Lecture 3 - Stream Ciphers"— Presentation transcript:
1Cryptography, Attacks and Countermeasures Lecture 3 - Stream Ciphers John A Clark and Susan Stepney Dept. of Computer ScienceUniversity of York, UK
2Stream Ciphers Part I: Pseudo-random number generators. Lots of Bad WaysPart II: Divide and conquer attacks.
3Stream Ciphers - Vernam Vernam Cipher works by generating a random bit stream and then XORing that stream on a bit by bit basis with the plaintext.Key KKey KRandomStream BiRandomStream BiBiBiPiCiPiBoth sender and receiver can generate key stream Bi. Receiver XORs the ciphertext stream with the key stream to recover the plaintext stream. We will use this cipher to illustrate several concepts.
4Linear Feedback Shift Registers 11Lij111At each iteration there is a right shift, a bit falls off the end, and the leftmost bit is set according to the linear feedback function. Here 0+0+1=1
5Periodicity We would like the stream to be ‘random-looking’. One feature should be that the stream should not repeat itself too quickly.Note that this is in effect a finite state machine and so must repeat itself eventually.The maximal period for an n-bit register is 2n-1.Why not 2n ?
6Maximal Period m-sequences The tap sequence defines the linear feedback function and is often regarded as a finite field polynomial.You have to choose the tap sequence very carefully.Some choices provide a maximal length period.These are primitive polynomials
7Primitive Polynomials Give m-sequences D3D2D1D01234567tD3D2D1D081910111213141511Common to denote the above by the polynomial C(D)=1+D+D4. Note we are back to where we started.
8Some Polynomials Don’t 1234567tD3D2D1D081910111213141511The polynomial C(D)=1+D+D3 does not give a maximal period sequence.
9Not good for PRNGConsider a 64 bit register. Can this be used as a key stream generator?No. Once you know a very small amount of plaintext (e.g. 32 consecutive bits) then you can calculate the corresponding key stream and so you know the rightmost 32 bits in the register.You can now try in turn all other 232 combinations for the rest. When you get the right one, you are able to generate the whole key streamAnd so plaintext should make sense.This is just too easy to break.But LFSRs are very easy to implement; and execute quickly.Can we fix matters?How about a less primitive way of extracting the key stream.How about combining several streams to achieve any better security?
10Very Simple Model LFSR 1 f Zj Use some function f to operate on some subset of the LFSR register componentsfZj
11Boolean Functions – Algebraic Normal Form (ANF) A Boolean function on n-inputs can be represented in minimal sum (XOR +) of products (AND .) form:This is the algebraic normal form of the function.The algebraic degree of the function is the size of the largest subset of inputs (i.e. the number of xj in it) associated with a non-zero co-efficient.1 is a constant function (as is 0)x1+x3+x5 is a linear functionx1.x3+x5 is a quadratic functionx1.x3.x5+x4x5+x2 is a cubic functionf(x1,…,xn)=a0+a1. x1 +…+an. xn a1,2.x1.x2+…+ an-1,n.xn-1.xn+… …+a1,2,..,n x1.x2 ...xn
12Very Simple ModelWhat about a linear function f?LSFR 1fZj
13Very Simple Model1This would be pretty awful. Suppose we know a sequence of keystream bits z0, z1 , z2, z2 ,…=1, 1, 1, 1,…Essentially every key stream output can be expressed as a linear function of the elements of the initial state. We can derive a number of these equations and then solve them by standard linear algebra techniques.
15Very Simple ModelWe can apply linear algebra equation solving techniques and solve for the si.z0= s0 +s2z1= s1 +s3z2= s2 +s0 +s3s0 s1 s2 s3z0 z1 z2 z3z3= s3 +s0 +s1+s3 = s0 +s1This has solutions0 s1 s2 s311
16Harder Model LSFR 1 f Zj What about a non-linear function f This is better but it is still possible to attack such systems if f is approximated by a linear function. We will talk about approximations later.
17Classical Stream Cipher Model Plaintext Stream Pj Keystream Zj Cipherstream CjLSFR 1L1jLSFR 2L2jCombining Boolean function f.fZjReceiver can generate key stream and recover plaintextLSFR nChoose f very carefullyLnjN- Bit registersInitial register values form the ‘key’CjPj
18Periodicity The LFSRs need not all be the same length. The LFSRs will give a vector input which has period that is the product of the least common multiple of the periods of each of the LFSRs.E.g. if period LFSR1=3,LFSR2=7 then overall period is 21
19Awful Choice for f LSFR 1 x1j Zj=f(x1j , x2j)= x1j f Zj LSFR 2 x2j Cj This is a truly awful choice. The key is intended to be 2 x 32 = 64 bits.x1jZj=f(x1j , x2j)= x1jfYou have completely ignored LFSR Key size = 32 bits onlyZjLSFR 2x2j32- Bit registersInitial register values form the ‘key’CjPj
20Better but Still Awful Choice for f LSFR 1Congratulations! You have not ignored LFSR 2!Key size = 64 bits?x1jZj=f(x1j , x2j) = x1j x2j+fZjLSFR 2x2j32- Bit registersInitial register values form the ‘key’CjPj
21Better but Still Awful Choice for f Well not quite such a good choice.Suppose you know 32 consecutive bits of plaintext (or can guess them correctly).Calculate the 32 bits of key stream.But if stream bit is 0 then there are only 2 possible pairs. Similarly, if stream value is 1. Effective key size = 232x32x31x30x2x1xkyk+=+=1+=or+++++y32y31y30y2y1=====111
22First Bit of Bad Linearity The combination function here is a linear function of the inputs:f(x1,x2)=x1f(x1,x2)=x2f(x1,x2)=x1+x2The following are quadratic functions:f(x1,x2)=x1.x2 + x1f(x1,x2)= x1.x2 + x2f(x1,x2)= x1.x2 + x1+x2Extreme examples given but beware linearity - even a hint of it can spell trouble.linear functions are so called because they can cause some cryptosystems to be broken straight away??
23Divide and Conquer Attacks Exploiting simple correlations in the combining function
25Geffe Generator – DIVIDE AND CONQUER Looking at the table it is clear that the output z agrees with b 75% of the time.Also agrees with c 75% of the time.a b c z
26Geffe Generator – DIVIDE AND CONQUER So consider each possible initial state s of register LFSR2.Determine the LFSR2 stream that s produces.Check the degree of agreement of this stream with the actual key stream.Turns out:if state s is correct you will get roughly the right amount of agreement.if state s is incorrect you will get roughly random (50%) agreement.Thus we have targeted LFSR2 and can easily break it.Now can target LFSR3 in exactly the same way.So we can get LFSR2 and LFSR3. Now we can derive the selection LFSR1 state very straightforwardly:try every possible state. The correct one should allow you to simulate the whole sequence.Other ways too.
27Divide and ConquerDivide and conquer attacks were suggested by Siegenthaler as a means of exploiting approximate linear relationships between function inputs and its output.This led to new criteria being developed as countermeasures to these correlation attacks.We will consider an extremely simple example.
28Divide and Conquer Consider the following combining function f(x1,x2)=x1.x2+x1Clearly not linear. But…f(x1,x2) agrees with x1 75% of the time here.Consider each possible initial state of LFSR1 and determine the degree of agreement with the actual key stream.The correct initial state will give approximately 75% agreement and the rest will give fairly random agreement.It’s also obvious that if we know f(x1,x2)=1 then we know both x1 and x2 => this is simply due to the incredibly small nature of the example.x1 x2 f(x1,x2)0 0 00 1 01 0 11 1 0
29Divide and Conquer Consider two functions f(x1,x2) and g(x1,x2) We say that f(x1,x2) is approximated by g(x1,x2) if the percentage of pairs (x1,x2) which given the same values for f and g differs from 50%.If they agree precisely half the time we say that they are uncorrelated.Note: if the percentage of agreement is less than 50% we can always find a function that has positive agreement, namelyg(x1,x2)+true.
30Ideas Generalise We can consider similar ideas for n-input functions: f(x1,x2,…,xn); andg(x1,x2,…,xn)Degree of approximation with linear functions may be slight.The smaller the degree of approximation the more data you need to have to break the system.
31And then what?The idea of multiple LFSRs is that the size of the keyspace should be the product of the keyspace sizes for each register.Divide and conquer reduces this to a sum of key sizes and you attack each in turn.Note what happens when you crack one LFSR. The complexity of the remaining task is reduced:f(x1,x2)=x1.x2+x1Once you know x1 then the task for x2 is simpler – whenever you know x1=1 you know what x2 is.
32All Fall Down In a similar vein, suppose: There is a small exploitable correlation with input x1.There is a small correlation with x1+x2.If LFSR 1 can be broken to reveal x1 then we have now have a straightforward correlation with x2 to exploit.
33Don’t tell themBut what if you don’t publicise the taps sequence – keep the feedback polynomial secret (as part of the key).Makes things harder but there are in fact some further attacks here too.
34Summary Have presented some very simple stream cipher models. Divide and conquer attacks.Dangers of linearity and hints of it.Next lecture:What do we do about the dangers?Boolean function criteria.High non-linearity.High algebraic degree.Correlation immunity.Tradeoffs between them.