K 2r+8 PHT <<<1 MDS MDS

Transcription

1 )*.1,(/-+032 THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS 034- TECHNICAL REPORT OF IEICE. 7865'9 Twosh ;<= (:; 2) 3 B&C y 3 NTT FRUX_a[\]Ye`khE f Gnql #Hw~ ˆ g 1-1 shiho@isl.ntt.co.jp y NTT Multimedia Communications Laboratories 250 Cambridge Ave., Palo Alto, CA 94306, USA yiqun@nttmcl.com ª «OŒ zkd!š AES Žƒ Q "œž ^b[z!š Twosh Lxp šdi}{ truncated dierential V šj%cys 2 057:3 žjt 16 W truncated dierential mœ šd 2 51 rvnm u c truncated dierential \good pair" 1 st c good pair i 2 77 st Ÿ} ž d šc Knudsen open problem Ÿ šc 5 W Twosh random permutationÿ Pž $oi truncated dierential mœ šd ± ³ ² µ, ¹, truncated dierential, Twosh, AES Cryptanalysis of Twosh (II) Shiho Moriai 3 Yiqun Lisa Yin y 3 NTT Information Sharing Platform Laboratories 1-1 Hikarinooka, Yokosuka, , Japan shiho@isl.ntt.co.jp y NTT Multimedia Communications Laboratories 250 Cambridge Ave., Palo Alto, CA 94306, USA yiqun@nttmcl.com Abstract We present truncated dierential cryptanalysis of the block cipher Twosh, which is one of the ve nalists for the Advanced Encryption Standard (AES). From our experimental results, we found a 16-round truncated dierential with probability of about 2 057:3. One can expect to get one good pair following the truncated dierential from 2 51 chosen plaintexts, and there are a total of 2 77 such goodpairs. We also found 5-round truncated dierentials which can be useful in distinguishing Twosh reduced to 5 rounds from a random permutation. This was considered to be an open problem by Knudsen. key words cryptanalysis, dierential cryptanalysis, truncated dierential, Twosh, AES

2 1 Introduction Twosh is a 128-bit block cipher proposed by Schneier et al. [SKW+98]. It is one of the ve nalists of AES, and it is used in many products such as GnuPG, SSH Secure Shell, and so on [C99]. The best known attack on variants of Twosh claimed by the designers is an impossible dierential attack on 6-round Twosh [F99]. Recently Knudsen [K00] showed that there are dierentials for Twosh for up to 16 rounds, predicting at least 32 bits of nontrivial information in every round. The probability of the truncated dierentials are too small to distinguish Twosh with more than a few rounds from a random permutation, but he claimed that it is possible, at least in theory, to nd one good pair of plaintexts following the dierential through all 16 rounds. Murphy and Robshaw [MR00] made some observations on key-dependent S-boxes and dierential cryptanalysis of Twosh. Their approach was to choose the S-box to t the dierential characteristic, instead of choosing the dierential characteristic to t the S-box. They found a 6-round dierential characteristic which holds for a fraction of at least of the S-boxes and claimed possible attacks of 8-round Twosh. Table 1 summarizes the known results on cryptanalysis of Twosh. In this paper we study truncated dierential cryptanalysis of Twosh. The type of truncated dierentials to be used are \byte characteristics," that is, the values of the dierence in a byte are distinguished between non-zero and zero, and the measure of dierence is exclusive-or. Note that Knudsen's truncated dierentials were based on the integer subtraction dierence between two 32-bit words. By using byte wise characteristics instead, we can make a thorough investigation of the non-uniformity in the distribution of the dierences, which was left as an open question by Knudsen [K00]. Twosh consists of both byte-oriented and non-byte-oriented operations as shown in Figure 1. The non-byte-oriented operations include the 1-bit rotates, addition with subkeys, and PHT (pseudo-hadamard transform), which comprises of two additions modular To search for byte characteristics of Twosh, we begin by computing the truncated dierential probability of addition modular 2 n. Based on the ecient computation of dierential probability of addition modular 2 n shown in [M00], we give an ecient computation of truncated dierential probability of addition modular 2 n in Section 2. In Section 3, we consider truncated dierential probability of the MDS. Finally in Section 4 we present the truncated dierentials that we found by computer experiments. 2 Ecient Computation of Truncated Dierential Probabilities of Addition Modular 2 n In [M00], an ecient algorithm was presented for computing dierential probabilities of addition mod 2 n. The algorithm can be extended to compute truncated dierential probabilities of addition of 2 n, but a straightforward extension to the case of truncated dierentials can still be computationally very expensive. In this section, we study how to further speed up the round whitening key size cryptanalysis complexity conditions reference 4 w/ any distinguishing attack [K00] 6 w/o 128 impossible dierential [F99] 6 w/o 192 impossible dierential [F99] 6 w/o 256 impossible dierential [F99] 6 w/ 256 impossible dierential [F99] 8* w/ any dierential attack > fraction [MR00] of the S-boxes Table 1: Twosh cryptanalysis

3 K 0 K 1 K 2 K 3 input whitening S-box 0 S-box 1 S-box 2 MDS K 2r+8 PHT <<<1 S-box 3 <<<8 S-box 0 S-box 1 S-box 2 S-box 3 MDS K 2r+9 >>>1 1 round : 15 more rounds K 4 K 5 K 6 K 7 output whitening Figure 1: Twosh computation of truncated dierential probabilities. We will follow the denitions and notation in [M00], and here we will only restate some of them if they are directly related to our discussion below. For x; y; z 2 GF(2) n, the function addition mod 2 n is dened as follows: f(x; y) =x + y = z (mod2 n ): We divide 1x 2 GF(2) n into t-bit sub-blocks and denote them by 1x [t] least signicant sub-block. So 1x =(1x [t] m01 ;:::;1x[t] 1 ; 1x[t] 0 ); 0 ; 1x[t] 1 ;::: from the where m = n=t is the number of sub-blocks. A very ecient algorithm for computing dierential probabilities of f (denoted by DP f (1x; 1y;1z)) is given in [M00]. For each triplet (1x; 1y; 1z), the running time of the algorithm is O(n) 1, while a naive approach would require a running time of O(2 2n ). The truncated dierential probabilities for f are dened as follows. TDP f (x;y;z) = 1 c X (1x;1y;1z)=(x;y;z) DP f (1x; 1y; 1z); (1) where c is the number of pairs (1x; 1y) satisfying the condition (1x; 1y) =(x; y). Let w H (x) denote the Hamming weight ofx. Then it is easy to see that c =(2 t 0 1) w H(x)+w H (y) : In a typical setting (e.g., byte characteristics), we have n =32andt = 8. So the number of possible truncated dierentials is (2 n=t ) 3 =2 12. Some of these truncated dierentials may have a very large c value. For example, when w H (x) +w H (y) 6, we have c Therefore, computation of all truncated dierential probabilities using Equation (1) can still be very expensive, even when the dierential probabilities themselves can be calculated eciently. 1 Later the complexity was further improved to 2(log n) in the worst-case and 2(1) in the average-case.

4 2.1 Basic idea The main idea for speeding up the computation of truncated dierential probabilities is to treat each sub-block somewhat independently. More specically, we will rst compute some properly dened \partial sums of dierential probabilities" for each sub-block ignoring the carry from one sub-block to the next, and then we will join these probabilities together to obtain the total truncated dierential probability for f. For each sub-block, we need to consider both the dierence in the carryin (denoted by 1cin) from the previous sub-block and dierence in the carryout (denoted by 1cout) to the next sub-block. There are two possible values for 1cin: 0,1. Let P 1cout denote the probability that there is carry from one sub-block to the next. That is, P 1cout =Pr[1cout =1]: Based on the results in [M00], there are only three possible values for P 1cout : 0,0.5,1. For a give sub-block (the ith sub-block), let (x i ;y i ;z i ) be the values of (x; y; z) restricted to the sub-block. Below, we dene 6 partial sums for the dierential probabilities, one corresponding to a possible combination of (1cin; P 1cout )=(d; p) ford =0; 1andp =0; 0:5; 1. where Condition PS is 2.2 Detailed algorithm = X PS(x i ;y i ;z i ;d;p) Condition PS DP(1x [t] i ; 1y[t] i ; 1z[t] i ); (2) (1x [t] i ; 1y[t] i ; 1z[t] i ) = (x i;y i ;z i ); 1cin = d; P 1cout = p: Our algorithm for computing truncated dierential probabilities contains two major components: precomputing partial sums and joining partial sums of sub-blocks. Precomputing partial sums We observe that the partial sums dened by Equation (2) only depend on the values restricted to a particular sub-block. Therefore, these partial sums can be precomputed and stored in a table. Typically, eachx i ;y i ;z i is just a single bit. So the total number of partial sums to be stored is = 48. Joining sub-blocks Given the partial sums for anytwo consecutive sub-blocks H and L (each of length t bits), we can compute the partial sums for the sub-block HjjL of length 2t bits. Let PS L (x i ;y i ;z i ;d;p); PS H (x i+1 ;y i+1 ;z i+1 ;d;p); and PS HjjL (x i+1 jjx i ;y i+1 jjy i ;z i+1 jjz i ;d;p)

5 denote the partial sums of dierential probabilities for the corresponding sub-blocks. PS HjjL is computed as Then PS HjjL (1; 1; 1;d;p)=[ ]: PS H (1; 1; 1; 0;p) 2 PS L (1; 1; 1;d;0) + PS H (1; 1; 1; 1;p) 2 PS L (1; 1; 1;d;1) + PS H (1; 1; 1; 0;p) 2 PS L (1; 1; 1;d;0:5) 2 0:5 + PS H (1; 1; 1; 1;p) 2 PS L (1; 1; 1;d;0:5) 2 0:5 In general, the two sub-blocks H and L can have anynumberofbits,say t 1 and t 2, respectively. Using the above formula, we can compute the new partial sums for the sub-block HjjL of length (t 1 + t 2 ) bits. Computing the total TDP By repetitively joining successive sub-blocks, we can obtain the 6 partial sums PS(x;y;z; d;p) for the entire block of length n. Since 1cin = 0 for the least signicant sub-block, 3 of these partial sums (for which d = 1) actually have value zero. Therefore, the total truncated dierential probability is TDP f (x; y;z) = 1 2 [ PS(x; y; z; 0; 0) c + PS(x; y; z; 0; 1) ]: + PS(x; y; z; 0; 0:5) Eciency analysis The algorithm given in this section is independent of the Hamming weight of x and y. For n = 32 and t =8,eachofthe2 12 truncated dierential probabilities can be computed using a constantnumber of table lookups, additions, and multiplications. Experiments show that all the 2 12 probabilities can be computed in less than one second on a PC. 3 Truncated Dierential Probabilities of MDS The truncated dierential probabilities for the MDS are dened as follows. TDP MDS (x; y) = 1 c X (1x;1y)=(x;y) Pr[MDS(x) 8 MDS(x 8 1x) =1y]; (3) where c is the number of 1x satisfying the condition (1x) =x. The distribution of TDP MDS (x;y) is related to the weight distribution of the MDS (Maximum Distance Separable) code. TDP MDS (x;y) is determined by the Hamming weights of x and y, astable 2 shows. 4 Search for Truncated Dierentials of Twosh In this section, we present our search results for truncated dierentials of Twosh. Our search uses the dierential probabilities of PHT and MDS computed in Sections 2 and 3. For speeding up the search, we rst set the probability to be one for 1-bit rotations. Once we found the truncated dierentials, we then adjust the probability as follows. If the input dierence (32-bit) of the 1-bit rotation is f 2, the output dierence is still f. Otherwise, we need some adjustment. For example, if the input dierence of the 1-bit right rotation is 8, the output 2 In this section we usetypewriter font for the hexadecimal representation of truncated dierentials.

6 w H (y) w H (x) : : : : : : : : :023 Table 2: Truncated dierential probabilities of MDS dierence is 8 with probability 2 01, c with probability and 4 with probability 2 08 (here we have multiple paths, but in most cases the multiple paths join at the next MDS). For additions with subkeys (i.e., f(x; k) =x + k = z (mod 2 n ), where k is some subkey), the value corresponding to k = 0 in our precomputed table gives the truncated dierential probability when we average over all possible keys. For any xed subkey k, the probability depends on k, and it can be larger or smaller than the average probability: the maximum probability can be 1 for a fraction of the subkeys. For easy treatment of probability after the search, we set the probability to be one for additions with subkeys. 4.1 Truncated dierentials with high probability First, we searched for truncated dierentials that hold with relatively high probability, although they may not be exploited in general (well-known) cryptanalytic attacks. As Knudsen [K00] wrote, such dierentials can provide some bits of nontrivial information in every round. Our computer experiments found a 12-round truncated dierential with probability of about 2 040:9.InTable 3, the output dierence of each round are shown in hexadecimal representation. One can expect to get one good pair following the truncated dierential from about 2 34 chosen plaintexts by using a structure in the last byte of the plaintext. There are a total of 2 94 such good pairs. More interestingly, we found a truncated dierential for the full 16 rounds of Twosh with probability of about 2 057:3 (see Table 4). One can expect to get one good pair following the truncated dierential from about chosen plaintexts, and there are 2 28 such good pairs. In [K00] Knudsen showed a 16-round truncated dierential with probability The probability of our 16-round truncated dierential is much higher than what was found by Knudsen, and the total number of good pairs for our dierential is also much larger. 4.2 Truncated dierentials useful for distinguishing attacks We also searched for truncated dierentials that may be useful in distinguishing attacks. As a result, we found one 4-round truncated dierential, and four 5-round truncated dierentials (see Tables 6 and 5). The 4-round truncated dierential is a path included in the 4-round truncated dierential that Knudsen used for the 2 -tests in [K00, Section 5.2]. Note that Knudsen's 4-round truncated dierential contains multiple paths and the probability is much higher. Knudsen concluded that for more than 4 rounds, it is an open question how nonuniform the distribution of dierences can be. Now that we found 5-round truncated dierentials with probability slightly higher than a random permutation, in theory we can perform statistical tests such as 2 tests. Note that the probabilities in Table 5 can be a little smaller due to 1-bit rotations or a little larger due to the eect of multiple paths.

7 round probability : f f 2 00: f f f e 2 08: f e f f 2 08: f f 7 f 2 016: f f f 2 016: f f b f 2 024: b f f f 2 024: f f 7 f 2 032: f f f 2 032: f f b f 2 040: b f f f 2 040: Table 3: 12-round truncated dierential round probability : f f 2 00: f f f e 2 08: f e f f 2 08: f f 7 f 2 016: f f f 2 016: f f b f 2 024: b f f f 2 024: f f 7 f 2 032: f f f 2 032: f f b f 2 040: b f f f 2 040: f f 7 f 2 049: f f f 2 049: f f b f 2 057: b f f f 2 057: Table 4: 16-round truncated dierential 5 Conclusion We presented truncated dierential cryptanalysis of the block cipher Twosh. We performed the search by computer experiments, and found a 16-round truncated dierential with probability of about 2 057:3, which ismuch larger than previously known results. We also found 5-round truncated dierentials which can be useful in distinguishing Twosh reduced to 5 rounds from a random permutation. We will implement some tests to conrm our conjecture. References [C99] [F99] N. Ferguson, \Impossible dierentials in Twosh," Twosh Technical Report #5, October 5, [K00] [K95] [KN96] [KRW99] [M00] L. R. Knudsen, \Trawling Twosh (revisited)," Presentation at rump session of AES3. Public comment on AES Candidate Algorithms { Round 2. csrc.nist.gov/encryption/aes/round2/comments/ lknudsen-2.pdf L. R. Knudsen, \Truncated and Higher Order Dierentials," Fast Software Encryption Second International Workshop, Lecture Notes in Computer Science 1008, pp.196{211, Springer-Verlag, L. R. Knudsen and T. A. Berson, \Truncated dierentials of SAFER," Fast Software Encryption, 3rd International Workshop, Lecture Notes in Computer Science 1039, pp.15{26, Springer-Verlag, L. R. Knudsen, M. J. B. Robshaw, and D. Wagner, \Truncated Dierentials and Skipjack," Advances in Cryptology CRYPTO'99, Lecture Notes in Computer Science 1666, pp.165{180, Springer-Verlag, S. Moriai,\Cryptanalysis of Twosh (I)", In Proceedings of the 2000 Symposium on Cryptography and Information Security, Okinawa, Japan, January 26-28, 2000.

8 round probability : f f 2 00: f f : : : c 0 2 0: c 0 e e 2 08: e e c : c : c : e 0 2 0: e 0 c c 2 015: c c e : e : e : f 0 2 0: f : f : f : f : Table 5: 5-round truncated dierentials round probability : f f 2 00: f f : f f 2 055: Table 6: 4-round truncated dierential [MR00] S. Murphy and M.J.B Robshaw, \Dierential Cryptanalysis, Key-dependent S- boxes, and Twosh", Public comment on AES Candidate Algorithms - Round 2. pdf [MSAK99] [MT99] [SKW+98] S. Moriai, M. Sugita, K. Aoki, and M. Kanda, \Security of E2 against Truncated Dierential Cryptanalysis," SAC'99, 6th Annual International Workshop on Selected Areas in Cryptography, Workshop Record, pp.133{143, 1999, (to appear in Lecture Notes in Computer Science, Springer-Verlag, 2000). M. Matsui and T. Tokita, \Cryptanalysis of a Reduced Version of the Block Cipher E2," Fast Software Encryption, 6th International Workshop, Lecture Notes in Computer Science 1636, pp.71{80, Springer-Verlag, B. Schneier, J. Kelsey, D. Whiting, D. Wagner, C. Hall, and N. Ferguson, \Twosh: A 128-Bit Block Cipher".