Документация по криптоалгоритмам / Square
.pdf5.4 Our Choice
Because of its optimal values for and , we have decided to take for S an
1
Sbox that is constructed by taking the mapping x 7!x and applying an
a ne transformation (over GF(2)) to the output bits. This a ne transformation
8
has the property that it has a complicated description in GF(2 ) to thwart
interpolation attacks [4].
Our choices force all fourround di erential trails to have an associated prob
ability not higher than 2 
, far below the critical noise value of 2 
. Equiv 


75 
alently, fourround linear trails have an associated correlation not over 2 , far
64
below the critical noise value of 2 . Hence, for resistance against conventional
LC and DC six rounds may seem su cient. However, the speci c blocked struc
ture of the cipher allows for more e cient dedicated di erential attacks. This
will be explained in the following section.
6A Dedicated Attack
In this section we describe a dedicated attack that exploits the cipher structure
of Square. The attack is a chosen plaintext attack and is independent of the
speci c choices of S , c(x) and the key schedule. It is faster than an exhaustive
key search for Square versions of up to 6 rounds. After describing the basic
attack on 4 rounds, we will show how it can be extended to 5 and 6 rounds.
6.1 Preliminaries
Let a set be a set of 256 states that are all di erent in some of the (16) state
bytes (the active) and all equal in the other state bytes (the passive). Let be
the set of indices of the active bytes. We have
xi;j 
6= yi;j 
for (i; j) 2 
8x; y 2 : xi;j 
= yi;j 
for (i; j) 62 
In this section we will make use of the geometrical interpretation as presented
t
in Figure 1. Applying the transformations and [k ] on (the elements of) a
set results in a (generally di erent) set with the same . Applying results
in a set in which the active bytes are transposed by . Applying to a set
does not necessarily result in a set. However, since every output byte of is
a linear combination (with invertible coe cients) of the four input bytes in the
same row, an input row with a single active byte gives rise to an output row
with only active bytes.
6.2 Four Rounds
Consider a set in which only one byte is active. We will now trace the evolution
of the positions of the active bytes through 3 rounds. The 1st round contains no ,
hence there is still only one byte active at the beginning of the 2nd round. of the
11
2nd round converts this to a complete row of active bytes, that is subsequently
transformed by to a complete column. of the 3rd round converts this to a
set with only active bytes. This is still the case at the input to the 4th round.
Since the bytes of the outputs of the 3rd round (denoted by a) range over all
possible values and are therefore balanced over the set, we have
M b = M M c a = M c M a = M c 0 = 0:
i;j j k i;k l i;l+j l
a2 k l a2 l
b= (a);a2
Hence, the bytes of the output of of the fourth round are balanced. This
balancedness is in general destroyed by the subsequent application of .
An output byte of the 4th round (denoted by a here) can be expressed as a
function of the intermediate state b above
4
a = S [b ] k :
i;j j;i
i;j
4
By assuming a value for k , the value of b for all elements of the set can be
j;i
i;j
calculated from the ciphertexts. If the values of this byte are not balanced over
, the assumed value for the key byte was wrong. This is expected to eliminate
all but approximately 1 key value. This can be repeated for the other bytes of
4
k .
We implemented the attack and found that two sets of 256 chosen plain
texts each are su cient to uniquely determine the cipher key with an overwhelm
ing probability of success.
6.3 Extension by a Round at the End
If an additional round is added, we have to calculate the above value of b
j;i
from the output of the 5th round instead of the 4th round. This can be done by
additionally assuming a value for a set of 4 bytes of the 5th round key. As in the
case of the 4round attack, wrong key assumptions are eliminated by verifying
that b is not balanced.
j;i
40
In this 5round attack 2 key values must be checked, and this must be re
peated 4 times. Since by checking a single set leaves only 1=256 of the wrong
key assumptions as possible candidates, the cipher key can be found with over
whelming probability with only 5 sets.
6.4 Extension by a Round at the Beginning
The basic idea is to choose a set of plaintexts that results in a set at the output
of the 2nd round with a single active Sbox. This requires the assumption of
0
values of four bytes of the round key k .
If the intermediate state after of the 2nd round has only a single active
byte, this is also the case for the output of the 2nd round. This imposes the
following conditions on a row of four input bytes of of the second round: one
particular linear combination of these bytes must range over all 256 possible
values (active) while 3 other particular linear combinations must be constant for
12
all 256 states. This imposes identical conditions on the bytes in the same row
1
in the input to [k ], and consequently on a column of bytes in the input to
0
of the 1st round. If the corresponding column of bytes of k is known, these
conditions can be converted to conditions on four plaintext bytes.
32
Now we consider a set of 2 plaintexts, such that the array of bytes in one
column ranges over all possible values and all other bytes are constant.
Now, make an assumption for the value of the 4 bytes of the relevant column
032
of k . Select from the set of 2 available plaintexts, a set of 256 plaintexts that
obey the conditions indicated above. Now the 4round attack can be performed.
For the given key assumption, the attack can be repeated for a several plaintext
5
sets. If the byte values of k suggested by these attacks are not consistent, the
initial assumption must have been wrong. A correct assumption for the bytes of
0
k will result in the swift and consistent recuperation of the last round key.
We implemented this attack where we assumed knowledge of 16 bits of the
rstround key. The attack found the other 16 bits of the rstround key and
128 bits of the lastround key using only 2 structures of 256 plaintexts for every
key value guessed in the rst round.
6.5 Complexity of the Attacks
Combining both extensions results in a 6 round attack. Although infeasible with
current technology, this attack is faster than exhaustive key search, and therefore
relevant. We have not found extensions to 7 rounds faster than exhaustive key
search.
We summarize the attacks in Table 3.
Attack 
#Plaintexts 
Time 
Memory 
4round 
9 
9 
small 
2 
2 

5round type 1 
11 
40 
small 
2 
2 

5round type 2 
32 
40 
32 
2 
2 
2 

6round 
32 
72 
32 
2 
2 
2 
Table 3. Complexities of the attack on SQUARE.
7Number of Rounds
Due to these attacks we have to increase the number of rounds to at least seven.
As a safety margin, we xed the number of rounds to eight.
Conservative users are free to increase the number of rounds. This can be
done in a straightforward way and requires no adaptation of the key schedule
whatsoever.
13
8The Key Evolution
The key schedule speci es the derivation of the round keys in terms of the cipher
key. Its function is to provide resistance against the following types of attack:
{ Attacks in which part of the cipher key is known to the cryptanalyst, e.g.,
if the cipher is used with a key shorter than 128 bits.
{ Attacks where the key entry to the cipher is known or can be chosen, e.g., if
the cipher is used as the compression function of a hash algorithm [7].
{ Relatedkey attacks.
Resistance against the rst type of attack can be improved by a key schedule in
which the round key undergoes a transformation with high di usion. For a good
scheme, the knowledge of a certain number of bits of one round key xes very
few bits in other round keys. The other two types of attack exploit regularities in
the structure of the key schedule by locally compensating round key di erences
[5, 7].
The key schedule also plays an important role in the elimination of symmetry:
{ Symmetry in the round transformation: the round transformation
treats all bytes of a state in very much the same way. This symmetry can be
removed by having round constants in the key schedule.
{ Symmetry between the rounds: the round transformation is the same
for all rounds. This equality can be removed by having rounddependent
round constants in the key schedule.
The key schedule is de ned in terms of the rows of the key. We can de ne a
left byterotation operation rotl(a ) on a row as
i
rotl[a a a a ] = [a a a a ]
i;0 i;1 i;2 i;3 i;1 i;2 i;3 i;0
and a right byte rotation rotr(a ) as its inverse.
i



t+1 
= 
t 

The key schedule iteration transformation k 
(k ) and its inverse are 

de ned by 






t+1 
t 
t 
t+1 
t 
t 

k0 
= k0 rotl(k3) Ct 
3 
= 3 2 


t+1 
t 
t+1 
t+1 
t 
t 

k1 
= k1 k0 
2 
= 2 1 


t+1 
t 
t+1 
t+1 
t 
t 

k2 
= k2 k1 
1 
= 1 0 


t+1 
t 
t+1 
t+1 
t 
t 
0 
k3 
= k3 k2 
0 
= 0 rotr( 3) Ct 
The simplicity of the inverse key schedule is thanks to the fact that and
commute. The round constants C are also de ned iteratively. We have C = 1
t 0 x
and C C .
t 1
This choice provides high di usion and removes the regularities in an e cient
way.
14
9Implementation Aspects
9.1 8bit Processor
On an 8bit processor Square can be programmed by simply implementing the
di erent component transformations. This is straightforward for , and .
The transformation requires a table of 256 bytes. requires multiplication
8
in the eld GF(2 ). However, the multiplication polynomial has been chosen
to make this very e cient. We have written a program implementing Square
in Assembler for the Motorola's M68HC05 microprocessor, typical for Smart
Cards. The machine code occupies in total 547 bytes of ROM, needs 36 bytes
of RAM and one execution of Square, including the key schedule, takes about
7500 cycles. This corresponds to less than 2 msec with a 4 MHz Clock.
The inverse cipher however is signi cantly slower than the forward cipher.
1
This is caused by the di erence in complexity between and .
9.2 32bit Processor
In the implementation of the cipher, the succession of steps
t0t
[k ] = [k ]
0t t
with k = (k ) can be combined in a single set of table lookups. The interme
diate state can be represented by four 32bit words, each containing a row [a ].
i
Its transpose is denoted by [ai] 
T 







0t 






. For b = ( ( (a))) + k we have 



c0 c3 c2 c1 



S [a0;i] 
3 












T 2c1 c0 c3 c2 
3 2S [a1;i] 
0t T 










[bi] = 

c2 c1 c0 c3 



S [a2;i] 
7 
[ki 
] 




























6c3 c2 c1 c0 
7 6S [a3;i] 













4 
c0 
3 

5 

4 

c3 
3 

5 



c2 
3 



c1 
3 

= 
2c1 
S [a0;i] 
2c0 
S [a1;i] 
2c3 
S [a2;i] 
2c2 
0t T 


c2 
7 

c1 
7 

c0 
7 

c3 
7 
S [a3;i] [ki ] 


6c3 




6c2 




6c1 


6c0 



4 

5 




4 

5 




4 

5 


4 

5 

We de ne the tables M and T as 





















M[a] = a 
c0 c1 c2 c3 















T [a] = M[S[a]] : 






T and M have 256 entries of four bytes each. The table M implements the
polynomial multiplication. T combines the nonlinear substitution with this mul
tiplication. Now we have
Mj 
0t 
j 

[bi] = rotr 
(T [aji]) [ki ] : 
t
We conclude that [ (ki )] can be done with 16 table lookups, 12
rotations and 16 exors of 32bit words. This implementation needs the table T ,
with 256 entries of four bytes, i.e. one kilobyte in total.
15
Last Round It can be seen that in this implementation, of the last round
is already executed in the previous set of tablelookups. In the last round the
8
function to be applied is [k ] . This can be realised by replacing the table
8
T [x] = M[S[x]] by S[x]. Since c = 1 , the unity in GF(2 ), the entries of the
2 x
small table S can be extracted from T , removing the extra storage requirement
for S.
Performance The reference implementation is written in C and runs at 2.63
MByte/s on a 100 MHz Pentium with the Windows95 operating system. The in
verse cipher can be implemented in exactly the same way as the cipher itself and
has the same performance. The di erence is in the tables and the precalculation
of the round keys.
10 Acknowledgements
We thank Paulo Barreto, who wrote the optimized reference implementation of
Square. Paulo Barreto can be reached at pbarreto@uninet.com.br.
References
1. E. Biham and A. Shamir, \Di erential cryptanalysis of DESlike cryptosystems,"
Journal of Cryptology, Vol. 4, No. 1, 1991, pp. 3{72.
2. J. Daemen, \Cipher and hash function design strategies based on linear and dif
ferential cryptanalysis," Doctoral Dissertation, March 1995, K.U.Leuven.
3. J. Daemen and V. Rijmen, \Selfreciprocal cipher structures," COSIC internal re
port 963, 1996.
4. T. Jakobsen and L.R. Knudsen, \The interpolation attack on block ciphers," these
proceedings.
5. J. Kelsey, B. Schneier and D. Wagner, \Keyschedule cryptanalysis of IDEA, G
DES, GOST, SAFER, and TripleDES," Advances in Cryptology, Proceedings
Crypto'96, LNCS 1109, N. Koblitz, Ed., SpringerVerlag, 1996, pp. 237{252.
6. L.R. Knudsen, \Truncated and higher order di erentials," Fast Software Encryp
tion, LNCS 1008, B. Preneel, Ed., SpringerVerlag, 1995, pp. 196{211.
7. L.R. Knudsen, \A keyschedule weakness in SAFERK64," Advances in Cryp
tology, Proceedings Crypto'95, LNCS 963, D. Coppersmith, Ed., SpringerVerlag,
1995, pp. 274{286.
8. L.R. Knudsen and T.A. Berson, \Truncated di erentials of SAFER," Fast Soft
ware Encryption, LNCS 1039, D. Gollmann, Ed., SpringerVerlag, 1996, pp. 15{26.
9. N. Koblitz, \A Course in Number Theory and Cryptography," SpringerVerlag,
New York, 1987.
10. X. Lai, J.L. Massey and S. Murphy, \Markov ciphers and di erential cryptanaly
sis," Advances in Cryptology, Proceedings Eurocrypt'91, LNCS 547, D.W. Davies,
Ed., SpringerVerlag, 1991, pp. 17{38.
11. F.J. MacWilliams, N.J.A. Sloane, \The Theory of ErrorCorrecting Codes," North
Holland, Amsterdam, 1977.
16
12. M. Matsui, \Linear cryptanalysis method for DES cipher," Advances in Cryp
tology, Proceedings Eurocrypt'93, LNCS 765, T. Helleseth, Ed., SpringerVerlag,
1994, pp. 386{397.
13. K. Nyberg, \Di erentially uniform mappings for cryptography," Advances in
Cryptology, Proceedings Eurocrypt'93, LNCS 765, T. Helleseth, Ed., Springer
Verlag, 1994, pp. 55{64.
14. L. O'Connor, \On the distribution of characteristics in bijective mappings," Jour
nal of Cryptology, Vol. 8, No. 2, 1995, pp. 67{86..
15. V. Rijmen, J. Daemen et al., \The cipher SHARK,"
LNCS 1039, D. Gollmann, Ed., SpringerVerlag, 1996, pp. 99{112.
17
Тут вы можете оставить комментарий к выбранному абзацу или сообщить об ошибке.
Оставленные комментарии видны всем.