Lecture 8: AES: The Advanced Encryption Standard

Lecture Notes on Computer and Network Security

by Avi Kak (kak@purdue.edu)

February 4, 2017
7:16am

c
2017 Avinash Kak, Purdue University

Goals:

To review the overall structure of AES and to focus particularly on the

four steps used in each round of AES: (1) byte substitution, (2) shift
rows, (3) mix columns, and (4) add round key.

Python and Perl implementations for creating the lookup tables for the
byte substitution steps in encryption and decryption.

Python implementation of the Key Expansion Algorithms for the 128 bit,
192 bit, and 256 bit AES.

Perl implementations for creating histograms of the differentials and for

constructing linear approximation tables in attacks on block ciphers.
 CONTENTS

Section Title Page

8.1 Salient Features of AES 3

8.2 The Encryption Key and Its Expansion 10

8.3 The Overall Structure of AES 12

8.4 The Four Steps in Each Round of Processing 15

8.5 The Substitution Bytes Step: SubBytes and 19

InvSubBytes

8.5.1 Traditional Explanation of Byte Substitution: 22

Constructing the 16 16 Lookup Table

8.5.2 Python and Perl Implementations for the AES 27

Byte Substitution Step

8.6 The Shift Rows Step: ShiftRows and InvShiftRows 31

8.7 The Mix Columns Step: MixColumns and 33

InvMixColumns

8.8 The Key Expansion Algorithm 36

8.8.1 The Algorithmic Steps in Going from one 4-Word 40

Round Key to the Next 4-Word Round Key

8.8.2 Python Implementation of the Key Expansion 46

Algorithm

8.9 Differential, Linear, and Interpolation Attacks on 52

Block Ciphers

8.10 Homework Problems 85

2
Computer and Network Security by Avi Kak Lecture 8

8.1: SALIENT FEATURES OF AES

AES is a block cipher with a block length of 128 bits.

AES allows for three different key lengths: 128, 192, or 256 bits.
Most of our discussion will assume that the key length is 128
bits. [With regard to using a key length other than 128 bits,
the main thing that changes in AES is how you generate the key
schedule from the key an issue I address at the end of Section
8.8.1. The notion of key schedule in AES is explained in Sections
8.2 and 8.8.]

Encryption consists of 10 rounds of processing for 128-bit keys,

12 rounds for 192-bit keys, and 14 rounds for 256-bit keys.

Except for the last round in each case, all other rounds are iden-
tical.

Each round of processing includes one single-byte based substi-

tution step, a row-wise permutation step, a column-wise mixing

3
Computer and Network Security by Avi Kak Lecture 8

step, and the addition of the round key. The order in which these
four steps are executed is different for encryption and decryption.

To appreciate the processing steps used in a single round, it is

best to think of a 128-bit block as consisting of a 4 4 matrix of
bytes, arranged as follows:

byte0 byte4 byte8 byte12
byte1 byte5 byte9 byte13




byte byte6 byte10 byte14

2


byte3 byte7 byte11 byte15

Therefore, the first four bytes of a 128-bit input block occupy the
first column in the 4 4 matrix of bytes. The next four bytes
occupy the second column, and so on.

The 4 4 matrix of bytes shown above is referred to as the state

array in AES. [If you are trying to create your own implementa-
tion of AES in Python, you will find following statement, which
uses the notion of list comprehension in Python, very useful for
creating an initialized structure that looks like the state array of
AES:
statearray = [[0 for x in range(4)] for x in range(4)]

Next, try the following calls in relation to the structure thus

created:
4
Computer and Network Security by Avi Kak Lecture 8

print statearray

print statearray[0]

print statearray[2][3]

block = range(128)

for i in range(4):
for j in range(4):
statearray[j][i] = block[32*i + 8*j:32*i + 8*(j+1)]

for i in range(4):
for j in range(4):
print statearray[i][j], " ",

This is a nice warm-up exercise before you start implementing

AES in Python.]

AES also has the notion of a word. A word consists of four

bytes, that is 32 bits. Therefore, each column of the state array
is a word, as is each row.

Each round of processing works on the input state array and

produces an output state array.

The output state array produced by the last round is rearranged

into a 128-bit output block.

5
Computer and Network Security by Avi Kak Lecture 8

Unlike DES, the decryption algorithm differs substantially from

the encryption algorithm. Although, overall, very similar steps
are used in encryption and decryption, their implementations are
not identical and the order in which the steps are invoked is
different, as mentioned previously.

AES, notified by NIST as a standard in 2001, is a slight variation

of the Rijndael cipher invented by two Belgian cryptographers
Joan Daemen and Vincent Rijmen. [Back in 1999, the Rijndael cipher was one of the
five chosen by NIST as a potential replacement for DES. The other four were: MARS from IBM; RC6 from

RSA Security; Serpent by Ross Anderson, Eli Biham, and Lars Knudsen; and Twofish by a team led by the

always-in-the-news cryptographer Bruce Schneier. Rijndael was selected from these five after extensive testing

that was open to public. ]

Whereas AES requires the block size to be 128 bits, the original
Rijndael cipher works with any block size (and any key size) that
is a multiple of 32 as long as it exceeds 128. The state array for
the different block sizes still has only four rows in the Rijndael
cipher. However, the number of columns depends on size of the
block. For example, when the block size is 192, the Rijndael
cipher requires a state array to consist of 4 rows and 6 columns.

As explained in Lecture 3, DES was based on the Feistel net-

work. On the other hand, what AES uses is a substitution-
permutation network in a more general sense. Each round
of processing in AES involves byte-level substitutions followed by

6
Computer and Network Security by Avi Kak Lecture 8

word-level permutations. Speaking generally, DES also involves

substitutions and permutations, except that the permutations are
based on the Feistel notion of dividing the input block into two
halves, processing each half separately, and then swapping the
two halves.

Like DES, AES is an iterated block cipher in which plaintext

is subject to multiple rounds of processing, with each round ap-
plying the same overall transformation function to the incoming
block. [When we say that each round applies the same transformation function to the incoming block,
that similarity is at the functional level. However, the implementation of the transformation function in each

round involves a key that is specific to that round this key is known as the round key. Round keys are

derived from the user-supplied encryption key.]

Unlike DES, AES is an example of key-alternating block ciphers.

In such ciphers, each round first applies a diffusion-achieving
transformation operation which may be a combination of lin-
ear and nonlinear steps to the entire incoming block, which is
then followed by the application of the round key to the entire
block. As youll recall, DES is based on the Feistel structure in
which, for each round, one-half of the block passes through un-
changed and the other half goes through a transformation that
depends on the S-boxes and the round key. Key alternating ci-
phers lend themselves well to theoretical analysis of the security
of the ciphers.

7
Computer and Network Security by Avi Kak Lecture 8

For another point of contrast between DES and AES, whereas

DES is a bit-oriented cipher, AES is a byte-oriented cipher. [Re-
member, how in DES we segmented the right-half 32 bits of the incoming 64-bit block into eight segments of

4-bits each. And how we prepended each 4-bit segment with the last bit of the previous 4-bit segment and

appended to each 4-bit segment the first bit of the next 4-bit segment. Subsequently, in order to find the

substitution 4-bits for an incoming 4-bit segment, we used the first and the last bit thus acquired for indexing

into the four rows of a 4 16 S-box, while using the 4-bit segment itself for indexing into the columns of the

] The substitution step in DES requires bit-level access to

S-Box.

the block coming into a round. On the other hand, all operations
in AES are purely byte-level, which makes for convenient and fast
software implementation of AES.

About the security of AES, considering how many years have

passed since the cipher was introduced in 2001, all of the threats
against the cipher remain theoretical meaning that their time
complexity is way beyond what any computer system will be able
to handle for a long time to come. [As you know, for the 128-bit key AES, the worst-
case time complexity for a brute-force attack would be 2128 . Such a brute-force attack would be considered to

be an example of a theoretical attack since it is beyond the realm of any practical implementation. There is a

meet-in-the-middle attack called the biclique attack that very marginally improves upon this time complexity to

around 2126 which is still just a theoretical attack. The biclique attack was presented by Andrey Bogdanov,

Dmitry Khovratovich, and Christian Rechberger in their 2011 publication Biclique Cryptanalysis of the Full

AES. ]

AES was designed using the wide-trail strategy. As described

in the publication Security of a Wide Trail Design by Joan
Daemen and Vincent Rijmen, wide-trail design for a block cipher
8
Computer and Network Security by Avi Kak Lecture 8

involves: (1) A local nonlinear transformation (as supplied by the

substitution step in AES); and (2) A linear mixing transformation
that provides high diffusion. The phrase wide trail refers to
dispersal of the probabilities that one can associate with the bits
at certain specific positions in a bit block as it propagates through
the rounds.

If you are seriously interested in the algebraic foundations of AES

and also of the attacks that are being attempted on the cipher, Id
recommend the book Algebraic Aspects of the Advanced En-
cryption Standard, by Carlos Cid, Sean Murphy, and Matthew
Robshaw. This book was originally published by Springer, but is
now available for free download on the web. Just Google it.

The AES standard is described in the following official document:

http://csrc.nist.gov/publications/fips/fips197/fips-197.pdf

9
Computer and Network Security by Avi Kak Lecture 8

8.2: THE ENCRYPTION KEY AND ITS

EXPANSION

Assuming a 128-bit key, the key is also arranged in the form of

a matrix of 4 4 bytes. As with the input block, the first word
from the key fills the first column of the matrix, and so on.

The four column words of the key matrix are expanded into a
schedule of 44 words. (As to how exactly this is done, we will
explain that later in Section 8.8.) Each round consumes four
words from the key schedule.

Figure 1 on the next page depicts the arrangement of the encryp-

tion key in the form of 4-byte words and the expansion of the key
into a key schedule consisting of 44 4-byte words.

10
Computer and Network Security by Avi Kak Lecture 8

k k k k
0 4 8 12
k k k k
1 5 9 13
k k k k
2 6 10 14
k k k k
3 7 11 15

w w w w w w w w
0 1 2 3 4 5 42 43

Figure 1: This figure shows the four words of the original

128-bit key being expanded into a key schedule consisting
of 44 words. Section 8.8 explains the procedure used for
this key expansion. (This figure is from Lecture 8 of Computer and Network Security
by Avi Kak)

11
Computer and Network Security by Avi Kak Lecture 8

8.3: THE OVERALL STRUCTURE OF AES

The overall structure of AES encryption/decryption is shown in

Figure 2.

The number of rounds shown in Figure 2, 10, is for the case when
the encryption key is 128 bit long. (As mentioned earlier, the
number of rounds is 12 when the key is 192 bits, and 14 when
the key is 256.)

Before any round-based processing for encryption can begin, the

input state array is XORed with the first four words of the key
schedule. The same thing happens during decryption except
that now we XOR the ciphertext state array with the last four
words of the key schedule.

For encryption, each round consists of the following four steps:

1) Substitute bytes, 2) Shift rows, 3) Mix columns, and 4) Add
round key. The last step consists of XORing the output of the
previous three steps with four words from the key schedule.

12
Computer and Network Security by Avi Kak Lecture 8

For decryption, each round consists of the following four steps: 1)

Inverse shift rows, 2) Inverse substitute bytes, 3) Add round key,
and 4) Inverse mix columns. The third step consists of XORing
the output of the previous two steps with four words from the key
schedule. Note the differences between the order in which sub-
stitution and shifting operations are carried out in a decryption
round vis-a-vis the order in which similar operations are carried
out in an encryption round.

The last round for encryption does not involve the Mix columns
step. The last round for decryption does not involve the Inverse
mix columns step.

13
Computer and Network Security by Avi Kak Lecture 8

128 bit plaintext block 128 bit plaintext block

Add round key w w w w Round 10

0 3 0 3

Key Schedule
Round 1 w w Round 9
w w
4 7 4 7

Round 2 w w Round 8
w w
8 11 8 11

Round 10 Add round key

w w w w
40 43 40 43

128 bit ciphertext block 128 bit ciphertext block

AES Encryption AES Decryption

Figure 2: The overall structure of AES for the case of 128-

bit encryption key. (This figure is from Lecture 8 of Computer and Network Security
by Avi Kak)

14
Computer and Network Security by Avi Kak Lecture 8

8.4: THE FOUR STEPS IN EACH ROUND

OF PROCESSING

Figure 3 shows the different steps that are carried out in each round
except the last one. [See the end of the previous section as to what steps are not allowed in the last
round.]

STEP 1: (called SubBytes for byte-by-byte substitution during

the forward process) (The corresponding substitution step used
during decryption is called InvSubBytes.)

This step consists of using a 16 16 lookup table to find a

replacement byte for a given byte in the input state array.

The entries in the lookup table are created by using the notions
of multiplicative inverses in GF (28) and bit scrambling to
destroy the bit-level correlations inside each byte. [See Lecture
7 for what is meant by the notation GF (28).]

Section 8.5 explains this step in greater detail.

STEP 2: (called ShiftRows for shifting the rows of the state array
during the forward process) (The corresponding transformation
15
Computer and Network Security by Avi Kak Lecture 8

Substitute Bytes Inverse Mix Columns

Round Key
Shift Rows Add Round Key

Mix Columns Inverse Substitute Bytes

Round Key
Add Round Key Inverse Shift Rows

Encryption Round Decryption Round

Figure 3: One round of encryption is shown at left and one

round of decryption at right. (This figure is from Lecture 8 of Computer and
Network Security by Avi Kak)

16
Computer and Network Security by Avi Kak Lecture 8

during decryption is denoted InvShiftRows for Inverse Shift-

Row Transformation.)

The goal of this transformation is to scramble the byte order

inside each 128-bit block.
This step is explained in greater detail in Section 8.6.

STEP 3: (called MixColumns for mixing up of the bytes in each

column separately during the forward process) (The correspond-
ing transformation during decryption is denoted InvMixColumns
and stands for inverse mix column transformation.) The goal is
here is to further scramble up the 128-bit input block.

The shift-rows step along with the mix-column step causes

each bit of the ciphertext to depend on every bit of the plain-
text after 10 rounds of processing.

Recall the avalanche effect from our discussion on DES in

Lecture 3. In DES, one bit of plaintext affected roughly 31
bits of ciphertext. But now we want each bit of the plaintext
to affect every bit position of the ciphertext block of 128
bits. [The phrasing of this last sentence is important. The sentence does NOT
say that if you change one bit of the plaintext, the algorithm is guaranteed to change
every bit of the ciphertext. (Changing every bit of the ciphertext would amount
to reversing every bit of the block.) Since a bit can take on only two values, on

17
Computer and Network Security by Avi Kak Lecture 8

the average there will be many bits of the ciphertext that will be identical to the
plaintext bits in the same positions after you have changed one bit of the plaintext.
However, again on the average, when you change one bit of the plaintext, you will
see its effect spanning all of the 128 bits of the ciphertext block. On the other
hand, with DES, changing one bit of the plaintext affects only 31 bit positions on
the average.]

Section 8.7 explains this step in greater detail.

STEP 4: (called AddRoundKey for adding the round key to the

output of the previous step during the forward process) (The cor-
responding step during decryption is denoted InvAddRound-
Key for inverse add round key transformation.)

18
Computer and Network Security by Avi Kak Lecture 8

8.5: THE SUBSTITUTE BYTES STEP:

SubBytes and InvSubBytes

This is a byte-by-byte substitution using a rule that stays the

same in all encryption rounds. The byte-by-byte substitution
rule is different for the decryption chain, but again it stays the
same for all the rounds.

The presentation in the rest of this section is organized as follows:

The modern way of explaining the byte substitution step

that allows us to find the substitute byte for a given byte
by simply looking up a pre-computed 256-element array of
numbers.

The traditional way of explaining the byte substitution

step that involves using a 16 16 lookup table.

Perl and Python implementations of the byte substitution

step. These implementations are based on the mod-
ern explanation of the step. (Obviously, as you would
expect, both explanations lead to the same final answer for
byte substitution.)
19
Computer and Network Security by Avi Kak Lecture 8

In the modern way of explaining the byte substitution step

for the encryption chain, let xin be a byte of the state array for
which we seek a substitute byte xout. We can write xout = f (xin ).
The function f () involves two nonlinear operations: (i) We first
find the multiplicative inverse x = xin1 in GF (28); and (ii)
then we scramble the bits of x by XORing x with four different
circularly rotated versions of itself and with a special constant
byte c = 0x63. The four circular rotations are through 4, 5,
6, and 7 bit positions to the right. As you will see later in this
section, this bit scrambling step can be expressed by the relation:
xout = A x + c.

When using my BitVector module, the byte substitution step

as explained above can be implemented with just a couple of
calls to the module functions. The first operation of the step,
which involves calculating the multiplicative inverse of a byte x
in GF (28), can be carried out by invoking the function gf MI()
on the BitVector representation of x. The second operation
that requires XORing a byte with circularly shifted versions of
itself is even more trivial, as you will see in the Perl and Python
code shown later in this section.

The modern explanation of the byte substitution step as pre-

sented above applies equally well to the decryption chain, except
for the fact that you first apply the bit scrambling operation to
the byte and then you find its multiplicative inverse in GF (28).

20
Computer and Network Security by Avi Kak Lecture 8

Ill now present the more traditional explanation of

the byte substitution step. As mentioned earlier, it involves
using a 16 16 table. To find the substitute byte for a given
input byte, we divide the input byte into two 4-bit patterns, each
yielding an integer value between 0 and 15. (We can represent
these by their hex values 0 through F.) One of the hex values is
used as a row index and the other as a column index for reaching
into the 16 16 lookup table.

As explained in the next subsection, Section 8.5.1, the entries

in the lookup table are constructed by a combination of GF (28)
arithmetic and bit scrambling.

Before presenting in the next subsection the construction of the

16 16 table as required by the traditional explanation of byte
substitution, it needs to be emphasized that the goal of the
substitution step is to reduce the correlation between
the input bits and the output bits at the byte level.
The bit scrambling part of the substitution step ensures that
the substitution cannot be described in the form of evaluating a
simple mathematical function.

21
Computer and Network Security by Avi Kak Lecture 8

8.5.1: Traditional Explanation of Byte Substitution:

Constructing the 16 16 Lookup Table

We first fill each cell of the 16 16 table with the byte obtained
by joining together its row index and the column index. [The row
index of this table runs from hex 0 through hex F . Likewise, the column index runs
from hex 0 through hex F .]

For example, for the cell located at row index 2 and column
indexed 7, we place hex 0x27 in the cell. So at this point the
table will look like

0 1 2 3 4 5 6 7 8 9 ....
----------------------------------------------------
0 | 00 01 02 03 04 05 06 07 08 09 ....
|
1 | 10 11 12 13 14 15 16 17 18 19 ....
|
2 | 20 21 22 23 24 25 26 27 28 29 ....
|
.........
.........

We next replace the value in each cell by its multiplicative inverse

in GF (28) based on the irreducible polynomial x8 +x4 +x3 +x+1.
22
Computer and Network Security by Avi Kak Lecture 8

The hex value 0x00 is replaced by itself since this

element has no multiplicative inverse. [See Lecture 7 for what
we mean by the multiplicative inverse of a byte modulo an irreducible polynomial and as
to why the zero byte has no multiplicative inverse.]
[If you are creating your
own Python implementation for AES and using the BitVector
module, you can use the function gf MI() of that module to
calculate the multiplicative inverses required for this table.]

After the above step, lets represent a byte stored in a cell of the
table by b7b6b5b4b3b2b1b0 where b7 is the MSB and b0 the LSB.
For example, the byte stored in the cell (9, 5) of the above table
is the multiplicative inverse (MI) of 0x95, which is 0x8A.
Therefore, at this point, the bit pattern stored in the cell with
row index 9 and column index 5 is 10001010, implying that b7
is 1 and b0 is 0. [Verify the fact that the MI of 0x95 is indeed 0x8A. The
polynomial representation of 0x95 (bit pattern: 10010101) is x7 + x4 + x2 + 1, and the
same for 0x8A (bit pattern: 10001010) is x7 + x3 + x. Now show that the product of
these two polynomials modulo the polynomial x8 + x4 + x3 + x + 1 is indeed 1.]

For bit scrambling, we next apply the following transformation

to each bit bi of the byte stored in a cell of the lookup table:

bi = bi b(i+4) mod 8 b(i+5) mod 8 b(i+6) mod 8 b(i+7) mod 8 ci

where ci is the ith bit of a specially designated byte c whose hex

value is 0x63. ( c7 c6c5c4 c3c2 c1c0 01100011 )
23
Computer and Network Security by Avi Kak Lecture 8

The above bit-scrambling step is better visualized as the follow-

ing vector-matrix operation. Note that all of the additions in the
product of the matrix and the vector are actually XOR opera-
tions. [Because of the [A]~x + ~b appearance of this transformation, it is commonly
referred to as the affine transformation.]

1 0 0 0 1 1 1 1
b0
1



1 1 0 0 0 1 1 1


b1





1





1 1 1 0 0 0 1 1


b2





0



1 1 1 1 0 0 0 1
b3
0






1 1 1 1 1 0 0 0


b4





0





0 1 1 1 1 1 0 0


b5





1





0 0 1 1 1 1 1 0


b6





1


0 0 0 1 1 1 1 1 b7 0

The very important role played by the c byte of value

0x63: Consider the following two conditions on the SubBytes
step: (1) In order for the byte substitution step to be invertible,
the byte-to-byte mapping given to us by the 16 16 table must
be one-one. That is, for each input byte, there must be a unique
output byte. And, to each output byte there must correspond
only one input byte. (2) No input byte should map to itself,
since a byte mapping to itself would weaken the cipher. Taking
multiplicative inverses in the construction of the table does give
us unique entries in the table for each input byte except for
the input byte 0x00 since there is no MI defined for the all-zeros
byte. (See Lecture 4 for why that is the case.) If it were not
for the c byte, the bit scrambling step would also leave the input
byte 0x00 unchanged. With the affine mapping shown above,
24
Computer and Network Security by Avi Kak Lecture 8

the 0x00 input byte is mapped to 0x63. At the same time, it

preserves the one-one mapping for all other bytes.

In addition to ensuring that every input byte is mapped to a

different and unique output byte, the bit-scrambling step also
breaks the correlation between the bits before the substitution
and the bits after the substitution.

The 16 16 table created in this manner is called the S-Box.

The S-Box is the same for all the bytes in the state array.

The steps that go into constructing the 16 16 lookup table are

reversed for the decryption table, meaning that you first apply the
reverse of the bit-scrambling operation to each byte, as explained
in the next step, and then you take its multiplicative inverse in
GF (28).

For bit scrambling for decryption, you carry out the following
bit-level transformation in each cell of the table:
bi = b(i+2) mod 8 b(i+5) mod 8 b(i+7) mod 8 di

where di is the ith bit of a specially designated byte d whose

hex value is 0x05. ( d7d6d5d4d3d2d1ddc0 = 00000101 ) Finally,
you replace the byte in the cell by its multiplicative inverse in
GF (28). [IMPORTANT: You might ask whether decryption bit scrambling also maps 0x00 to its
25
Computer and Network Security by Avi Kak Lecture 8

constant d. No that does not happen. For decryption, the goal of bit scrambling is to reverse the effect of bit

scrambling on the encryption side. The bit scrambling operation for decryption maps 0x00 to 0x52. ]

The bytes c and d are chosen so that the S-box has no fixed points.
That is, we do not want S box(a) = a for any a. Neither do we
want S box(a) = a where a is the bitwise complement of a.

26
Computer and Network Security by Avi Kak Lecture 8

8.5.2: Python and Perl Implementations for the AES

Byte Substitution Step

Section 8.5 and the Subsection 8.5.1 presented two different ways
of implementing the AES byte substitution step. As stated earlier
in Section 8.5, both these explanations are equivalent in the
sense that either will result in the same substitution byte for a
given input byte.

This subsection shows my Python and Perl implementations of

the more modern explanation of byte substitution described in
Section 8.5. You will be surprised how easy it is to write this
code if you are using my BitVector module in Python and the
Algorithm::BitVector module in Perl.

What follows is a Python implementation of the explanation. The

goal of the for loop is to construct a 256 element array of lookup
values for integers ranging from 0 through 255. For each integer
in the range 0 through 255, we first find its multiplicative inverse
in GF (28), then we XOR the result with four different circularly
rotated versions of the result, and also XOR the result with the
constant c = 0x63. We do the same thing for the decryption
lookup array, except that we first do the XORing and then we
compute the multiplicative inverse.

27
Computer and Network Security by Avi Kak Lecture 8

#!/usr/bin/env python

## gen_tables.py
## Avi Kak (February 15, 2015)

## This is a Python implementation of the byte substitution explanations in Sections

## 8.5 and 8.5.1 of Lecture 8. In keeping with the explanation in Section 8.5, the
## goal here is to construct two 256-element arrays for byte substitution, one for
## the SubBytes step that goes into the encryption rounds of the AES algorithm, and
## the other for the InvSubBytes step that goes into the decryption rounds.

import sys
from BitVector import *

AES_modulus = BitVector(bitstring=100011011)
subBytesTable = [] # SBox for encryption
invSubBytesTable = [] # SBox for decryption

def genTables():
c = BitVector(bitstring=01100011)
d = BitVector(bitstring=00000101)
for i in range(0, 256):
# For the encryption SBox
a = BitVector(intVal = i, size=8).gf_MI(AES_modulus, 8) if i != 0 else BitVector(intVal=0)
# For bit scrambling for the encryption SBox entries:
a1,a2,a3,a4 = [a.deep_copy() for x in range(4)]
a ^= (a1 >> 4) ^ (a2 >> 5) ^ (a3 >> 6) ^ (a4 >> 7) ^ c
subBytesTable.append(int(a))
# For the decryption Sbox:
b = BitVector(intVal = i, size=8)
# For bit scrambling for the decryption SBox entries:
b1,b2,b3 = [b.deep_copy() for x in range(3)]
b = (b1 >> 2) ^ (b2 >> 5) ^ (b3 >> 7) ^ d
check = b.gf_MI(AES_modulus, 8)
b = check if isinstance(check, BitVector) else 0
invSubBytesTable.append(int(b))

genTables()
print "SBox for Encryption:"
print subBytesTable
print "\nSBox for Decryption:"
print invSubBytesTable

And shown below is the Perl implementation for doing the same
thing:
28
Computer and Network Security by Avi Kak Lecture 8

#!/usr/bin/env perl

## gen_tables.pl
## Avi Kak (February 16, 2015)

## This is a Perl implementation of the byte substitution explanations in Sections

## 8.5 and 8.5.1 of Lecture 8. In keeping with the explanation in Section 8.5, the
## goal here is to construct two 256-element arrays for byte substitution, one for
## the SubBytes step that goes into the encryption rounds of the AES algorithm, and
## the other for the InvSubBytes step that goes into the decryption rounds.

use strict;
use warnings;
use Algorithm::BitVector;

my $AES_modulus = Algorithm::BitVector->new(bitstring => 100011011);

my @subBytesTable; # SBox for encryption

my @invSubBytesTable; # SBox for decryption

sub genTables {
my $c = Algorithm::BitVector->new(bitstring => 01100011);
my $d = Algorithm::BitVector->new(bitstring => 00000101);
foreach my $i (0..255) {
# For the encryption SBox:
my $a = $i == 0 ? Algorithm::BitVector->new(intVal => 0) :
Algorithm::BitVector->new(intVal => $i, size => 8)->gf_MI($AES_modulus, 8);
# For bit scrambling for the encryption SBox entries:
my ($a1,$a2,$a3,$a4) = map $a->deep_copy(), 0 .. 3;
$a ^= ($a1 >> 4) ^ ($a2 >> 5) ^ ($a3 >> 6) ^ ($a4 >> 7) ^ $c;
push @subBytesTable, int($a);
# For the decryption Sbox:
my $b = Algorithm::BitVector->new(intVal => $i, size => 8);
# For bit scrambling for the decryption SBox entries:
my ($b1,$b2,$b3) = map $b->deep_copy(), 0 .. 2;
$b = ($b1 >> 2) ^ ($b2 >> 5) ^ ($b3 >> 7) ^ $d;
my $check = $b->gf_MI($AES_modulus, 8);
$b = ref($check) eq Algorithm::BitVector ? $check : 0;
push @invSubBytesTable, int($b);
}
}

genTables();
print "SBox for Encryption:\n";
print "@subBytesTable\n";
print "\nSBox for Decryption:\n";
print "@invSubBytesTable\n";

29
Computer and Network Security by Avi Kak Lecture 8

The encryption S-Box that a correct implementation should return

is shown below: (Note that the values are shown as decimal integers)

99 124 119 123 242 107 111 197 48 1 103 43 254 215 171 118
202 130 201 125 250 89 71 240 173 212 162 175 156 164 114 192
183 253 147 38 54 63 247 204 52 165 229 241 113 216 49 21
4 199 35 195 24 150 5 154 7 18 128 226 235 39 178 117
9 131 44 26 27 110 90 160 82 59 214 179 41 227 47 132
83 209 0 237 32 252 177 91 106 203 190 57 74 76 88 207
208 239 170 251 67 77 51 133 69 249 2 127 80 60 159 168
81 163 64 143 146 157 56 245 188 182 218 33 16 255 243 210
205 12 19 236 95 151 68 23 196 167 126 61 100 93 25 115
96 129 79 220 34 42 144 136 70 238 184 20 222 94 11 219
224 50 58 10 73 6 36 92 194 211 172 98 145 149 228 121
231 200 55 109 141 213 78 169 108 86 244 234 101 122 174 8
186 120 37 46 28 166 180 198 232 221 116 31 75 189 139 138
112 62 181 102 72 3 246 14 97 53 87 185 134 193 29 158
225 248 152 17 105 217 142 148 155 30 135 233 206 85 40 223
140 161 137 13 191 230 66 104 65 153 45 15 176 84 187 22

And the decryption S-Box that a correction implementation should

return is shown below (again as decimal integers):

82 9 106 213 48 54 165 56 191 64 163 158 129 243 215 251
124 227 57 130 155 47 255 135 52 142 67 68 196 222 233 203
84 123 148 50 166 194 35 61 238 76 149 11 66 250 195 78
8 46 161 102 40 217 36 178 118 91 162 73 109 139 209 37
114 248 246 100 134 104 152 22 212 164 92 204 93 101 182 146
108 112 72 80 253 237 185 218 94 21 70 87 167 141 157 132
144 216 171 0 140 188 211 10 247 228 88 5 184 179 69 6
208 44 30 143 202 63 15 2 193 175 189 3 1 19 138 107
58 145 17 65 79 103 220 234 151 242 207 206 240 180 230 115
150 172 116 34 231 173 53 133 226 249 55 232 28 117 223 110
71 241 26 113 29 41 197 137 111 183 98 14 170 24 190 27
252 86 62 75 198 210 121 32 154 219 192 254 120 205 90 244
31 221 168 51 136 7 199 49 177 18 16 89 39 128 236 95
96 81 127 169 25 181 74 13 45 229 122 159 147 201 156 239
160 224 59 77 174 42 245 176 200 235 187 60 131 83 153 97
23 43 4 126 186 119 214 38 225 105 20 99 85 33 12 125

The Python and Perl scripts in this section can be downloaded from
the link associated with Lecture 8 at the Lecture Notes website.
30
Computer and Network Security by Avi Kak Lecture 8

8.6: THE SHIFT ROWS STEP: ShiftRows

and InvShiftRows

This is where the matrix representation of the state array be-

comes important.

The ShiftRows transformation consists of (i) not shifting the first

row of the state array at all; (ii) circularly shifting the second
row by one byte to the left; (iii) circularly shifting the third row
by two bytes to the left; and (iv) circularly shifting the last row
by three bytes to the left.

This operation on the state array can be represented by

s0.0 s0,1 s0,2 s0,3 s0.0 s0,1 s0,2 s0,3
s1.0 s1,1 s1,2 s1,3 s s1,2 s1,3 s1,0

===> 1.1



s s2,1 s2,2 s2,3 s2.2 s2,3 s2,0 s2,1

2.0



s3.0 s3,1 s3,2 s3,3 s3.3 s3,0 s3,1 s3,2

Recall again that the input block is written column-wise. That

is the first four bytes of the input block fill the first column of
the state array, the next four bytes the second column, etc. As
31
Computer and Network Security by Avi Kak Lecture 8

a result, shifting the rows in the manner indicated scrambles up

the byte order of the input block.

For decryption, the corresponding step shifts the rows in exactly

the opposite fashion. The first row is left unchanged, the second
row is shifted to the right by one byte, the third row to the right
by two bytes, and the last row to the right by three bytes, all
shifts being circular.

s0.0 s0,1 s0,2 s0,3 s0.0 s0,1 s0,2 s0,3
s1.0 s1,1 s1,2 s1,3 s1.3 s1,0 s1,1 s1,2


===>


s2.0 s2,1 s2,2 s2,3 s2.2 s2,3 s2,0 s2,1




s3.0 s3,1 s3,2 s3,3 s3.1 s3,2 s3,3 s3,0

32
Computer and Network Security by Avi Kak Lecture 8

8.7: THE MIX COLUMNS STEP:

MixColumns and InvMixColumns

This step replaces each byte of a column by a function of all the

bytes in the same column.

More precisely, each byte in a column is replaced by two times

that byte, plus three times the the next byte, plus the byte that
comes next, plus the byte that follows. [The multiplications implied
by the word times and the additions implied by the word plus are meant to be
carried out in GF (28 ) arithmetic, as explained in Lecture 7. If you are using the
BitVector module in Python, it gives you the method gf multiply modular() for
carrying out such multiplications. The additions are merely XOR operations, as you
should know from Lecture 7. The Perl programmers can do the same thing with the
Algorithm::BitVector module.] The words next and follow refer
to bytes in the same column, and their meaning is circular, in
the sense that the byte that is next to the one in the last row is
the one in the first row. [Note that by two times and three times, we mean
multiplications in GF (28 ) by the bit patterns 000000010 and 00000011, respectively.]

For the bytes in the first row of the state array, this operation
can be stated as
s0,j = (0x02 s0,j ) (0x03 s1,j ) s2,j s3,j
33
Computer and Network Security by Avi Kak Lecture 8

For the bytes in the second row of the state array, this operation
can be stated as
s1,j = s0,j (0x02 s1,j ) (0x03 s2,j ) s3,j

For the bytes in the third row of the state array, this operation
can be stated as
s2,j = s0,j s1,j (0x02 s2,j ) (0x03 s3,j )

And, for the bytes in the fourth row of the state array, this
operation can be stated as

s3,j = (0x03 s0,j ) s1,j s2,j (0x02 s3,j )

More compactly, the column operations can be shown as

s0,1 s0,2 s0,3

02 03 01 01 s0.0 s0,1 s0,2 s0,3 s0.0

01 02 03 01 s1.0 s1,1 s1,2 s1,3 s s1,1 s1,2 s1,3


= 1.0


01 01 02 03 s2.0 s2,1 s2,2 s2,3 s2.0 s2,1 s2,2 s2,3



s3.0 s3,1 s3,2 s3,3

03 01 01 02 s3.0 s3,1 s3,2 s3,3

where, on the left hand side, when a row of the leftmost matrix
multiples a column of the state array matrix, additions involved
are meant to be XOR operations.

The corresponding transformation during decryption is given by

34
Computer and Network Security by Avi Kak Lecture 8

s0,1 s0,2 s0,3

0E 0B 0D 09 s0.0 s0,1 s0,2 s0,3 s0.0

09 0E 0B 0D s1.0 s1,1 s1,2 s1,3 s s1,1 s1,2 s1,3


= 1.0


0D 09 0E 0B s2.0 s2,1 s2,2 s2,3 s2.0 s2,1 s2,2 s2,3



s3.0 s3,1 s3,2 s3,3

0B 0D 09 0E s3.0 s3,1 s3,2 s3,3

35
Computer and Network Security by Avi Kak Lecture 8

8.8: THE KEY EXPANSION ALGORITHM

Each round has its own round key that is derived from the original
128-bit encryption key in the manner described in this section.
One of the four steps of each round, for both encryption and
decryption, involves XORing of the round key with the state
array.

The AES Key Expansion algorithm is used to derive the 128-

bit round key for each round from the original 128-bit encryption
key. As youll see, the logic of the key expansion al-
gorithm is designed to ensure that if you change one
bit of the encryption key, it should affect the round
keys for several rounds.

In the same manner as the 128-bit input block is arranged in the

form of a state array, the algorithm first arranges the 16 bytes of
the encryption key in the form of a 4 4 array of bytes, as shown
at the top of the next page.

36
Computer and Network Security by Avi Kak Lecture 8

k0 k4 k8 k12
k1 k5 k9 k13




k k6 k10 k14

2



k3 k7 k11 k15

[ w0 w1 w2 w3 ]

The first four bytes of the encryption key constitute the word w0,
the next four bytes the word w1 , and so on.

The algorithm subsequently expands the words [w0, w1 , w2, w3]

into a 44-word key schedule that can be labeled
w0 , w1, w2, w3 , ................., w43

Of these, the words [w0, w1, w2, w3] are bitwise XORed with the
input block before the round-based processing begins.

The remaining 40 words of the key schedule are used four words
at a time in each of the 10 rounds.

The above two statements are also true for decryption, except for
the fact that we now reverse the order of the words in the key
37
Computer and Network Security by Avi Kak Lecture 8

schedule, as shown in Figure 2: The last four words of the key

schedule are bitwise XORed with the 128-bit ciphertext block
before any round-based processing begins. Subsequently, each of
the four words in the remaining 40 words of the key schedule are
used in each of the ten rounds of processing.

Now comes the difficult part: How does the Key Expansion
Algorithm expand four words w0, w1 , w2, w3 into the 44 words
w0, w1, w2, w3 , w4, w5, ........, w43 ?

The key expansion algorithm will be explained in the next sub-

section with the help of Figure 4. As shown in the figure, the
key expansion takes place on a four-word to four-word basis, in
the sense that each grouping of four words decides what the next
grouping of four words will be.

38
Computer and Network Security by Avi Kak Lecture 8

w0 w1 w2 w3 g

w4 w5 w6 w7 g

w8 w9 w10 w11

Figure 4: The key expansion takes place on a four-word to

four-word basis as shown here. (This figure is from Lecture 8 of Computer
and Network Security by Avi Kak)

39
Computer and Network Security by Avi Kak Lecture 8

8.8.1: The Algorithmic Steps in Going from a 4-Word

Round Key to the Next 4-Word Round Key

We now come to the heart of the key expansion algorithm we

talked about in the previous section generating the four words
of the round key for a given round from the corresponding four
words of the round key for the previous round.

Lets say that we have the four words of the round key for the ith
round:
wi wi+1 wi+2 wi+3
For these to serve as the round key for the ith round, i must be
a multiple of 4. These will obviously serve as the round key for
the (i/4)th round. For example, w4, w5, w6 , w7 is the round key
for round 1, the sequence of words w8, w9 , w10, w11 the round key
for round 2, and so on.

Now we need to determine the words

wi+4 wi+5 wi+6 wi+7
from the words wi wi+1 wi+2 wi+3.

From Figure 4, we write

40
Computer and Network Security by Avi Kak Lecture 8

wi+5 = wi+4 wi+1 (1)

wi+6 = wi+5 wi+2 (2)
wi+7 = wi+6 wi+3 (3)

Note that except for the first word in a new 4-word grouping, each
word is an XOR of the previous word and the corresponding word
in the previous 4-word grouping.

So now we only need to figure out wi+4 . This is the beginning

word of each 4-word grouping in the key expansion. The begin-
ning word of each round key is obtained by:

wi+4 = wi g(wi+3 ) (4)

That is, the first word of the new 4-word grouping is to be ob-
tained by XORing the first word of the last grouping with what
is returned by applying a function g() to the last word of the
previous 4-word grouping.

The function g() consists of the following three steps:

Perform a one-byte left circular rotation on the argument 4-

byte word.

41
Computer and Network Security by Avi Kak Lecture 8

Perform a byte substitution for each byte of the word returned

by the previous step by using the same 16 16 lookup table
as used in the SubBytes step of the encryption rounds. [The
SubBytes step was explained in Section 8.5]

XOR the bytes obtained from the previous step with what is
known as a round constant. The round constant is a
word whose three rightmost bytes are always zero. Therefore,
XORing with the round constant amounts to XORing with
just its leftmost byte.

The round constant for the ith round is denoted Rcon[i].

Since, by specification, the three rightmost bytes of the round
constant are zero, we can write it as shown below. The left hand
side of the equation below stands for the round constant to be
used in the ith round. The right hand side of the equation says
that the rightmost three bytes of the round constant are zero.

Rcon[i] = (RC[i], 0x00, 0x00, 0x00)

The only non-zero byte in the round constants, RC[i], obeys the
following recursion:

RC[1] = 0x01
RC[j] = 0x02 RC[j 1]

42
Computer and Network Security by Avi Kak Lecture 8

Recall from Lecture 7 that multiplication by 0x02 amounts to

multiplying the polynomial corresponding to the bit pattern RC[j
1] by x.

The addition of the round constants is for the purpose of destroy-

ing any symmetries that may have been introduced by the other
steps in the key expansion algorithm.

The presentation of the key expansion algorithm so

far in this section was based on the assumption of a
128 bit key. As was mentioned in Section 8.1, AES calls for a
larger number of rounds in Figure 2 when you use either of the
two other possibilities for key lengths: 192 bits and 256 bits. A
key length of 192 bits entails 12 rounds and a key length of 256
bits entails 14 rounds. (However, the length of the input block
remains unchanged at 128 bits.) The key expansion algorithm
must obviously generate a longer schedule for the 12 rounds re-
quired by a 192 bit key and the 14 rounds required by a 256 bit
keys. Keeping in mind how we used the key schedule for the
case of a 128 bit key, we are going to need 52 words in the key
schedule for the case of 192-bit keys and 60 words for the case of
256-bit keys with round-based processing remaining the same
as described in Section 8.4. [Consider what happens when the key length is 192 bits:
Since the round-based processing and the size of the input block remain the same as described earlier in this

lecture, each round will still use only 4 words of the key schedule. Just as we organized the 128-bit key in the

form of 4 key words for the purpose of key expansion, we organize the 192 bit key in the form of six words.

The key expansion algorithm will take us from six words to six words for a total of nine key-expansion steps

43
Computer and Network Security by Avi Kak Lecture 8

with each step looking the same as what we described at the beginning of this section. Yes, it is true that

the key expansion will now generate a total of 54 words while we need only 52 we simply ignore the last two

words of the key schedule. With regard to the details of going from the six words of the j th key-expansion step

to the six words of the (j + 1)th key expansion step, lets focus on going from the initial (w0 , w1 , w2 , w3 , w4 ,

w5 ) to (w6 , w7 , w8 , w9 , w10 , w11 ). We generate the last five words of the latter from the last five words of

the former through straightforward XORing as was the case earlier in this section. As for the first word of the

latter, we generate it from the first and the last words of the former through the g function again as described

earlier. The g function itself remains unchanged. ]

The cool thing about the 128-bit key is that you can think of the
key expansion being in one-one correspondence with the rounds.
However, that is no longer the case with, say, the 192-bit keys.
Now you have to think of key expansion as something that is
divorced even conceptually from round-based processing of the
input block.

The key expansion algorithm ensures that AES has no weak

keys. A weak key is a key that reduces the security of a cipher
in a predictable manner. For example, DES is known to have
weak keys. Weak keys of DES are those that produce
identical round keys for each of the 16 rounds. An
example of DES weak key is when it consists of alternating ones
and zeros. This sort of a weak key in DES causes all the round
keys to become identical, which, in turn, causes the encryption to
become self-inverting. That is, plain text encrypted and then
encrypted again will lead back to the same plain text. (Since the
small number of weak keys of DES are easily recognized, it is not

44
Computer and Network Security by Avi Kak Lecture 8

considered to be a problem with that cipher.)

45
Computer and Network Security by Avi Kak Lecture 8

8.8.2: Python Implementation of the Key Expansion

Algorithm

Ill now present a Python implementation of the key expansion

algorithm described in the previous subsection.

With regard to key expansion, the main focus of the previous sub-
section was the 128-bit AES. Toward the end, I briefly described
the modifications needed for the case of 192-bit and 256-bit AES.
The goal of the implementation shown in this section is to clarify
the various steps for all three cases.

When you execute the code shown below, it will prompt you for
AES key size obviously, the number you enter must be one of
128, 192, and 256.

Subsequently, it will prompt you for the key. You are allowed to
enter any number of characters for the key. If the length of the
key you enter is shorter than what is need to fill the full width
of the AES key size, the script appends the character 0 to your
key to bring it up to the required size. On the other hand, if
you enter a key longer than what is needed, it will only use the
number of characters it needs.

46
Computer and Network Security by Avi Kak Lecture 8

#!/usr/bin/env python

## gen_key_schedule.py
## Avi Kak (April 10, 2016, bug fix: January 27, 2017)

## This script is for demonstrating the AES algorithm for generating the
## key schedule.

## It will prompt you for the key size, which must be one of 128, 192, 256.

## It will also prompt you for a key. If the key you enter is shorter
## than what is needed for the AES key size, we add zeros on the right of
## the key so that its length is as needed by the AES key size.

import sys
from BitVector import *

AES_modulus = BitVector(bitstring=100011011)

def main():
key_words = []
keysize, key_bv = get_key_from_user()
if keysize == 128:
key_words = gen_key_schedule_128(key_bv)
elif keysize == 192:
key_words = gen_key_schedule_192(key_bv)
elif keysize == 256:
key_words = gen_key_schedule_256(key_bv)
else:
sys.exit("wrong keysize --- aborting")
key_schedule = []
print("\nEach 32-bit word of the key schedule is shown as a sequence of 4 one-byte integers:")
for word_index,word in enumerate(key_words):
keyword_in_ints = []
for i in range(4):
keyword_in_ints.append(word[i*8:i*8+8].intValue())
if word_index % 4 == 0: print("\n")
print("word %d: %s" % (word_index, str(keyword_in_ints)))
key_schedule.append(keyword_in_ints)
# print("\n\nkey schedule: %s" % str(key_schedule))
num_rounds = None
if keysize == 128: num_rounds = 10
if keysize == 192: num_rounds = 12
if keysize == 256: num_rounds = 14
round_keys = [None for i in range(num_rounds+1)]
for i in range(num_rounds+1):
round_keys[i] = (key_words[i*4] + key_words[i*4+1] + key_words[i*4+2] +
key_words[i*4+3]).get_bitvector_in_hex()
print("\n\nRound keys in hex (first key for input block):\n")
for round_key in round_keys:
print(round_key)

47
Computer and Network Security by Avi Kak Lecture 8

def gee(keyword, round_constant, byte_sub_table):

This is the g() function you see in Figure 4 of Lecture 8.

rotated_word = keyword.deep_copy()
rotated_word << 8
newword = BitVector(size = 0)
for i in range(4):
newword += BitVector(intVal = byte_sub_table[rotated_word[8*i:8*i+8].intValue()], size = 8)
newword[:8] ^= round_constant
round_constant = round_constant.gf_multiply_modular(BitVector(intVal = 0x02), AES_modulus, 8)
return newword, round_constant

def gen_key_schedule_128(key_bv):
byte_sub_table = gen_subbytes_table()
# We need 44 keywords in the key schedule for 128 bit AES. Each keyword is 32-bits
# wide. The 128-bit AES uses the first four keywords to xor the input block with.
# Subsequently, each of the 10 rounds uses 4 keywords from the key schedule. We will
# store all 44 keywords in the following list:
key_words = [None for i in range(44)]
round_constant = BitVector(intVal = 0x01, size=8)
for i in range(4):
key_words[i] = key_bv[i*32 : i*32 + 32]
for i in range(4,44):
if i%4 == 0:
kwd, round_constant = gee(key_words[i-1], round_constant, byte_sub_table)
key_words[i] = key_words[i-4] ^ kwd
else:
key_words[i] = key_words[i-4] ^ key_words[i-1]
return key_words

def gen_key_schedule_192(key_bv):
byte_sub_table = gen_subbytes_table()
# We need 52 keywords (each keyword consists of 32 bits) in the key schedule for
# 128 bit AES. The 128-bit AES uses the first four keywords to xor the input
# block with. Subsequently, each of the 12 rounds uses 4 keywords from the key
# schedule. We will store all 52 keywords in the following list:
key_words = [None for i in range(52)]
round_constant = BitVector(intVal = 0x01, size=8)
for i in range(6):
key_words[i] = key_bv[i*32 : i*32 + 32]
for i in range(6,52):
if i%6 == 0:
kwd, round_constant = gee(key_words[i-1], round_constant, byte_sub_table)
key_words[i] = key_words[i-6] ^ kwd
else:
key_words[i] = key_words[i-6] ^ key_words[i-1]
return key_words

def gen_key_schedule_256(key_bv):
byte_sub_table = gen_subbytes_table()
# We need 60 keywords (each keyword consists of 32 bits) in the key schedule for
# 128 bit AES. The 128-bit AES uses the first four keywords to xor the input
# block with. Subsequently, each of the 14 rounds uses 4 keywords from the key
# schedule. We will store all 60 keywords in the following list:

48
Computer and Network Security by Avi Kak Lecture 8

key_words = [None for i in range(60)]

round_constant = BitVector(intVal = 0x01, size=8)
for i in range(8):
key_words[i] = key_bv[i*32 : i*32 + 32]
for i in range(8,60):
if i%8 == 0:
kwd, round_constant = gee(key_words[i-1], round_constant, byte_sub_table)
key_words[i] = key_words[i-8] ^ kwd
elif (i - (i//8)*8) < 4:
key_words[i] = key_words[i-8] ^ key_words[i-1]
elif (i - (i//8)*8) == 4:
# elif (i - 4) % 8 == 0:
key_words[i] = BitVector(size = 0)
for j in range(4):
key_words[i] += BitVector(intVal =
byte_sub_table[key_words[i-1][8*j:8*j+8].intValue()], size = 8)
key_words[i] ^= key_words[i-8]
elif ((i - (i//8)*8) > 4) and ((i - (i//8)*8) < 8):
key_words[i] = key_words[i-8] ^ key_words[i-1]
else:
sys.exit("error in key scheduling algo for i = %d" % i)
return key_words

def gen_subbytes_table():
subBytesTable = []
c = BitVector(bitstring=01100011)
for i in range(0, 256):
a = BitVector(intVal = i, size=8).gf_MI(AES_modulus, 8) if i != 0 else BitVector(intVal=0)
a1,a2,a3,a4 = [a.deep_copy() for x in range(4)]
a ^= (a1 >> 4) ^ (a2 >> 5) ^ (a3 >> 6) ^ (a4 >> 7) ^ c
subBytesTable.append(int(a))
return subBytesTable

def get_key_from_user():
key = keysize = None
if sys.version_info[0] == 3:
keysize = int(input("\nAES Key size: "))
assert any(x == keysize for x in [128,192,256]), \
"keysize is wrong (must be one of 128, 192, or 256) --- aborting"
key = input("\nEnter key (any number of chars): ")
else:
keysize = int(raw_input("\nAES Key size: "))
assert any(x == keysize for x in [128,192,256]), \
"keysize is wrong (must be one of 128, 192, or 256) --- aborting"
key = raw_input("\nEnter key (any number of chars): ")
key = key.strip()
key += 0 * (keysize//8 - len(key)) if len(key) < keysize//8 else key[:keysize//8]
key_bv = BitVector( textstring = key )
return keysize,key_bv

main()

Shown below is a terminal session with the code:

49
Computer and Network Security by Avi Kak Lecture 8

AES Key size: 128

Enter key (any number of chars): hello

Each 32-bit word of the key schedule is shown as a sequence of 4 one-byte integers:

word 0: [104, 101, 108, 108]

word 1: [111, 48, 48, 48]
word 2: [48, 48, 48, 48]
word 3: [48, 48, 48, 48]

word 4: [109, 97, 104, 104]

word 5: [2, 81, 88, 88]
word 6: [50, 97, 104, 104]
word 7: [2, 81, 88, 88]

word 8: [190, 11, 2, 31]

word 9: [188, 90, 90, 71]
word 10: [142, 59, 50, 47]
word 11: [140, 106, 106, 119]

word 12: [184, 9, 247, 123]

word 13: [4, 83, 173, 60]
word 14: [138, 104, 159, 19]
word 15: [6, 2, 245, 100]

word 16: [199, 239, 180, 20]

word 17: [195, 188, 25, 40]
word 18: [73, 212, 134, 59]
word 19: [79, 214, 115, 95]

word 20: [33, 96, 123, 144]

word 21: [226, 220, 98, 184]
word 22: [171, 8, 228, 131]
word 23: [228, 222, 151, 220]

word 24: [28, 232, 253, 249]

word 25: [254, 52, 159, 65]
word 26: [85, 60, 123, 194]
word 27: [177, 226, 236, 30]

word 28: [196, 38, 143, 49]

word 29: [58, 18, 16, 112]
word 30: [111, 46, 107, 178]
word 31: [222, 204, 135, 172]

word 32: [15, 49, 30, 44]

word 33: [53, 35, 14, 92]
word 34: [90, 13, 101, 238]
word 35: [132, 193, 226, 66]

word 36: [108, 169, 50, 115]

word 37: [89, 138, 60, 47]
word 38: [3, 135, 89, 193]
word 39: [135, 70, 187, 131]

50
Computer and Network Security by Avi Kak Lecture 8

word 40: [0, 67, 222, 100]

word 41: [89, 201, 226, 75]
word 42: [90, 78, 187, 138]
word 43: [221, 8, 0, 9]

Round keys in hex (first key for input block):

68656c6c6f3030303030303030303030
6d616868025158583261686802515858
be0b021fbc5a5a478e3b322f8c6a6a77
b809f77b0453ad3c8a689f130602f564
c7efb414c3bc192849d4863b4fd6735f
21607b90e2dc62b8ab08e483e4de97dc
1ce8fdf9fe349f41553c7bc2b1e2ec1e
c4268f313a1210706f2e6bb2decc87ac
0f311e2c35230e5c5a0d65ee84c1e242
6ca93273598a3c2f038759c18746bb83
0043de6459c9e24b5a4ebb8add080009

51
Computer and Network Security by Avi Kak Lecture 8

8.9: DIFFERENTIAL, LINEAR, AND

INTERPOLATION ATTACKS ON BLOCK
CIPHERS

This section is for a reader who is curious as to why the substitu-

tion step in AES involves taking the MI of each byte in GF (28)
and bit scrambling. As you might have realized already, that is
the only nonlinear step in mapping a plaintext block to a cipher-
text block in AES.

Back in the 1990s (this is the decade preceding the development

of the Rijndael cipher which is the precursor to the AES standard)
there was much interest in investigating the block ciphers of the
day (DES being the most prominent) from the standpoint of their
vulnerabilities to differential and linear cryptanalysis. The MI
byte substitution step in AES is meant to protect it
against such cryptanalysis. At around the same time, it
was shown by Jakobsen and Knudsen in 1997 that block ciphers
whose SBoxes were based on polynomial arithmetic in Galois
fields could be vulnerable to a new attack that they referred to
as the interpolation attack. The bit scrambling part of the
SBox in AES is meant to be a protection against the interpolation
attack. [As mentioned earlier in Section 3.2.2 of Lecture 3, the differential attack was first described
52
Computer and Network Security by Avi Kak Lecture 8

by Biham and Shamir in a paper titled Differential Cryptanalysis of DES-like Cryptosystems that appeared

in the Journal of Cryptology in 1991. The linear attack was firs described by Matsui in a publication titled

Linear Cryptanalysis Method for DES Ciphers, in Lecture Notes in Computer Science, no. 764. Finally,

the interpolation attack by first described by Jakobsen and Knudsen in a publication titled The Interpolation

Attack on Block Ciphers that appeared in Lecture Notes in Computer Science, Haifa, 1997. ]

Therefore, in order to fully appreciate the SBox in AES, you have

to have some understanding of these three forms of cryptanaly-
sis. The phrases differential cryptanalysis and linear crypt-
analysis are synonymous with differential attack and linear
attack.

The rest of this section reviews these three attacks briefly. [You will
get more out of this section if you first read the tutorial A Tutorial on Linear and Differential Cryptanalysis

by Howard Heys of the Memorial University of Newfoundland. Googling that authors name will take you

directly to the tutorial. ]

Starting our discussion with the differential attack, it is based

on the following concepts:

How a differential (meaning an XOR of two bit blocks) propa-

gates through a sequence of rounds is independent of the round
keys. [As youll recall from the note in small-font blue in Section 3.3.2 of Lecture
3, differential cryptanalysis is a chosen plaintext attack in which the attacker feeds
plaintext bit blocks pairs, X1 and X2 , with known differences X = X1 X2 be-
tween them, into the cipher while observing the differences Y = Y1 Y2 between
the corresponding ciphertext blocks. We refer to X as the input differential and

53
Computer and Network Security by Avi Kak Lecture 8

Y as the output differential. The fact that the propagation of a differential is NOT
affected by the round keys can be established in the following manner: Consider just
one round and lets say that K is the round key. Lets further say that the output of
the round is what is produced by the SBox XORed with the round key. For two dif-
ferent inputs X1 and X2 to the round, let Y1 and Y2 denote the outputs of the SBox
and and let Y1 and Y2 denote the final output of the round. We have Y1 = K Y1
and Y2 = K Y2 . The differential Y = Y1 Y2 for the output after key mixing is
related to the other differentials by Y = Y1 Y2 = K Y1 K Y2 = Y1 Y2 .
Therefore, the mapping between the input and the output differentials
of a round is not a function of the round key.]

If one is not careful, the byte substitution step in an SBox can

create significant correlations between the input differentials
and the output differentials.

The correlations between the input differentials and the out-

put differentials, when they are significant, can be exploited
to make good guesses for the bits of the last round key.

Therefore, our first order of business is to understand the rela-

tionship between the input and the output differentials for a given
choice of the SBox.

The Perl script shown next, find differentials correlations.pl, cal-

culates a 2D histogram of the relationship between the input and
the output differentials. The statements in lines (B8) and (B9),
with one of the lines commented-out, give you two choices for the
54
Computer and Network Security by Avi Kak Lecture 8

operation of the SBox. If you use the statement in line (B8), the
byte substitutions will consist of replacing each byte by its MI in
GF (28) that is based on the AES modulus. On the other hand, if
you use the currently commented-out statement in line (B9), the
byte substitution will take place according to the lookup table
supplied through line (A9). [Yes, to be precise, the MI based byte substitution could also
be carried out through a lookup table. That is, just because one SBox is based on MI calculations and the

other on looking up a table is NOT the fundamental difference between the two. The lookup table supplied

through line (A9) was arrived at by experimenting with several such choices made possible by the commented

out statements in lines (A7) and (A8). The call to shuffle() in line (A7) gives a pseudorandom permutation

of the 256 one-byte words. Based on a dozen runs of the script, the permutation shown in line (A9) yielded

the best inhomogeneous histogram for the input/output differentials. The reader may wish to carry out such

]
experiments on his/her own and possibly make a different choice for the lookup table in line (A9).

The portion of the script starting with line (F1) is just for display-
ing the histogram of the input/output differentials and, therefore,
not central to understand what we mean by the differentials here
and the correlations between the input differentials and the out-
put differentials.

#!/usr/bin/perl -w

## find_differentials_correlations.pl

## Avi Kak (March 4, 2015)

## This script creates a histogram of the mapping between the input differentials
## and the output differentials for an SBox. You have two choices for the SBox ---
## as reflected by lines (B8) and (B9) of the script. For a given input byte, the
## statement in line (B8) returns the MI (multiplicative inverse) of the byte in
## GF(2^8) based on the AES modulus. And the statement in line (B8) returns a byte
## through a user-specified table lookup. The table for this is specified in line
## (A9). More generally, such a table can be created by a random permutation

55
Computer and Network Security by Avi Kak Lecture 8

## through the commented-out statements in lines (A7) and (A8).

use strict;
use Algorithm::BitVector;
use Graphics::GnuplotIF;
$|++;

my $debug = 1;
my $AES_modulus = Algorithm::BitVector->new(bitstring => 100011011); #(A1)

my $M = 64; # CHANGE THIS TO 256 FOR A COMPLETE CALCULATION #(A2)

# This parameter control the range of inputs
# bytes for creating the differentials. With
# its value set to 64, only the differentials
# for the bytes whose int values are between 0
# and 63 are tried.
# Initialize the histogram:
my $differential_hist; #(A3)
foreach my $i (0..255) { #(A4)
foreach my $j (0..255) { #(A5)
$differential_hist->[$i][$j] = 0; #(A6)
}
}

# When SBox is based on lookup, we will use the "table" created by randomly
# permuting the the number from 0 to 255:
#my $lookuptable = shuffle([0..255]); #(A7)
#my @lookuptable = @$lookuptable; #(A8)
my @lookuptable = qw(213 170 104 116 66 14 76 219 200 42 22 17 241 197 41 216 85 140
183 244 235 6 118 208 74 218 99 44 1 89 11 205 195 125 47 236 113
237 131 109 102 9 21 220 59 154 119 148 38 120 13 217 16 100 191 81
240 196 122 83 177 229 142 35 88 48 167 0 29 153 163 146 166 77 79
43 10 194 232 189 238 164 204 111 69 51 126 62 211 242 70 214 247 55
202 78 239 114 184 112 228 84 152 187 45 49 175 58 253 72 95 19 37
73 145 87 198 71 159 34 91 168 250 255 8 121 96 50 141 181 67 26 243
130 68 61 24 105 210 172 139 136 128 157 133 80 93 39 2 143 161 186 33
144 178 30 92 138 169 86 249 252 155 193 63 223 203 245 129 4 171
115 3 40 151 7 188 231 174 25 23 207 180 56 46 206 215 227 162 199
97 147 182 149 108 36 132 5 12 103 110 209 160 137 53 224 185 173
20 222 246 28 179 134 75 254 57 60 234 52 165 225 248 31 230 156
124 233 158 27 18 94 65 32 54 106 192 221 190 101 98 251 212 150
201 117 127 107 176 226 135 123 82 15 64 90); #(A9)

# This call creates the 2D plaintext/ciphertext differential histogram:

gen_differential_histogram(); #(A10)

# The call shown below will show that part of the histogram for which both
# the input and the output differentials are in the range (32, 63).
display_portion_of_histogram(32, 64); #(A11)

plot_portion_of_histogram($differential_hist, 32, 64); #(A12)

## The following call makes a hardcopy of the plot:
plot_portion_of_histogram($differential_hist, 32, 64, 3); #(A13)

56
Computer and Network Security by Avi Kak Lecture 8

sub gen_differential_histogram { #(B1)

foreach my $i (0 .. $M-1) { #(B2)
print "\ni=$i\n" if $debug; #(B3)
foreach my $j (0 .. $M-1) { #(B4)
print ". " if $debug; #(B5)
my ($a, $b) = (Algorithm::BitVector->new(intVal => $i, size => 8),
Algorithm::BitVector->new(intVal => $j, size => 8)); #(B6)
my $input_differential = int($a ^ $b); #(B7)
# Of the two statements shown below, you must comment out one depending
# on what type of an SBox you want:
my ($c, $d) = (get_sbox_output_MI($a), get_sbox_output_MI($b)); #(B8)
# my ($c, $d) = (get_sbox_output_lookup($a), get_sbox_output_lookup($b)); #(B9)
my $output_differential = int($c ^ $d); #(B10)
$differential_hist->[$input_differential][$output_differential]++; #(B11)
}
}
}

sub get_sbox_output_MI { #(C1)

my $in = shift; #(C2)
return int($in) != 0 ? $in->gf_MI($AES_modulus, 8) : #(C3)
Algorithm::BitVector->new(intVal => 0); #(C4)
}

sub get_sbox_output_lookup { #(D1)

my $in = shift; #(D2)
return Algorithm::BitVector->new(intVal => $lookuptable[int($in)], size => 8); #(D3)
}

# Fisher-Yates shuffle:
sub shuffle { #(E1)
my $arr_ref = shift; #(E2)
my $i = @$arr_ref; #(E3)
while ( $i-- ) { #(E4)
my $j = int rand( $i + 1 ); #(E5)
@$arr_ref[ $i, $j ] = @$arr_ref[ $j, $i ]; #(E6)
} #(E7)
return $arr_ref; #(E8)
}

##################### Support Routines for Displaying the Histogram ########################

# Displays in your terminal window the bin counts in the two-dimensional histogram
# for the input/output mapping of the differentials. You can control the portion of
# the 2D histogram that is output by using the first argument to set the lower bin
# index and the second argument the upper bin index along both dimensions.
# Therefore, what you see is always a square portion of the overall histogram.
sub display_portion_of_histogram { #(F1)
my $lower = shift; #(F2)
my $upper = shift; #(F3)
foreach my $i ($lower .. $upper - 1) { #(F4)
print "\n"; #(F5)
foreach my $j ($lower .. $upper - 1) { #(F6)
print "$differential_hist->[$i][$j] "; #(F7)
}

57
Computer and Network Security by Avi Kak Lecture 8

}
}

# Displays with a 3-dimensional plot a square portion of the histogram. Along both
# the X and the Y directions, the lower bound on the bin index is supplied by the
# SECOND argument and the upper bound by the THIRD argument. The last argument is
# needed only if you want to make a hardcopy of the plot. The last argument is set
# to the number of second the plot will be flashed in the terminal screen before it
# is dumped into a .png file.
sub plot_portion_of_histogram {
my $hist = shift; #(G1)
my $lower = shift; #(G2)
my $upper = shift; #(G3)
my $pause_time = shift; #(G4)
my @plot_points = (); #(G5)
my $bin_width = my $bin_height = 1.0; #(G6)
my ($x_min, $y_min, $x_max, $y_max) = ($lower, $lower, $upper, $upper); #(G7)
foreach my $y ($y_min..$y_max-1) { #(G8)
foreach my $x ($x_min..$x_max-1) { #(G9)
push @plot_points, [$x, $y, $hist->[$y][$x]]; #(G10)
}
}
@plot_points = sort {$a->[0] <=> $b->[0]} @plot_points; #(G11)
@plot_points = sort {$a->[1] <=> $b->[1] if $a->[0] == $b->[0]} @plot_points; #(G12)
my $temp_file = "__temp.dat"; #(G13)
open(OUTFILE , ">$temp_file") or die "Cannot open temporary file: $!"; #(G14)
my ($first, $oldfirst); #(G15)
$oldfirst = $plot_points[0]->[0]; #(G16)
foreach my $sample (@plot_points) { #(G17)
$first = $sample->[0]; #(G18)
if ($first == $oldfirst) { #(G19)
my @out_sample; #(G20)
$out_sample[0] = $sample->[0]; #(G21)
$out_sample[1] = $sample->[1]; #(G22)
$out_sample[2] = $sample->[2]; #(G23)
print OUTFILE "@out_sample\n"; #(G24)
} else { #(G25)
print OUTFILE "\n"; #(G26)
}
$oldfirst = $first; #(G27)
}
print OUTFILE "\n";
close OUTFILE;
my $argstring = <<"END"; #(G28)
set xrange [$x_min:$x_max]
set yrange [$y_min:$y_max]
set view 80,15
set hidden3d
splot "$temp_file" with lines
END
unless (defined $pause_time) { #(G29)
my $hardcopy_name = "output_histogram.png"; #(G30)
my $plot1 = Graphics::GnuplotIF->new(); #(G31)
$plot1->gnuplot_cmd( set terminal png, "set output \"$hardcopy_name\""); #(G32)
$plot1->gnuplot_cmd( $argstring ); #(G33)

58
Computer and Network Security by Avi Kak Lecture 8

my $plot2 = Graphics::GnuplotIF->new(persist => 1); #(G34)

$plot2->gnuplot_cmd( $argstring ); #(G35)
} else { #(G36)
my $plot = Graphics::GnuplotIF->new(); #(G37)
$plot->gnuplot_cmd( $argstring ); #(G38)
$plot->gnuplot_pause( $pause_time ); #(G39)
}
}

For an accurate and complete calculation of the input/output

differentials histogram, youd need to change the value of $M in
line (A2) to 256. That would result in a large 256256 histogram
of integer values. For the purpose of our explanation here, we will
make do with $M = 64. The resulting histogram would not be
an accurate depiction of the reality. Nonetheless, it will suffice
for the purpose of the explanation that follows.

If you run the script with the SBox as specified in line (B8),
you will end up with a display of numbers as shown below for the
portion of the differentials histogram that is bounded by bin index
values ranging from 32 to 63 in both directions. To understand
these values, lets look at the first nonzero entry in the first row,
which happens to be in the column indexed 40. Recognizing
that the first row corresponds to the bin index 32, that nonzero
count of 2 means that in all of the runs of the loop in lines (B1)
through (B11) of the script, there were 2 cases when the input
differential was X = 00100000 (integer value = 32) and the
output differential was Y = 00101000 (integer value = 40).

0 0 0 0 0 0 0 2 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 2 0 0 0 0 0
0 0 2 0 0 0 0 2 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 2 0 2 0 2 2 0 0 0

59
Computer and Network Security by Avi Kak Lecture 8

0 0 0 2 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 2 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 0 0 0 0 0 0 2 0
2 2 0 0 0 2 0 0 0 0 0 0 0 0 0 2 0 0 0 2 0 0 0 0 0 0 0 0 2 0 0 2
2 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 2 0 2 0 2 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 2 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 2 0 0 0 0 0
0 2 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 2 0 0 0 0 0 0
0 0 2 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 2 0 2 0 0 0 0 0 0 0 0 0 0 0 2 0 0 2 0 0 0 0 0 0 0 2
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 4 0 0 0 2 0 2 0 0 0 0 0 0 0 0 0
2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 2 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0
0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 2
0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 2 0 0 2 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 2 2 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0
2 0 0 0 0 2 0 0 0 0 2 0 0 0 2 0 0 2 0 0 0 2 0 0 0 0 0 0 2 0 0 0
0 0 2 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 2 0 2 0 0 0 0 0 0 0 0 2 0 0
0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 2 0 0 0 2 0
0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 2 0 0 0 0 0 0 0 2 4 0 0 0 0 0 2
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 2 0 0 0 0 0 0 0 0 0 0 0 0 2 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 2 0
0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 2 0 0 0 0 0 0 0 0 2 0 0
0 0 0 0 0 2 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 0 0 0 0 0 0 0 0 0
2 0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0
0 2 0 0 2 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 2 0 2 0 0 0 0 0 0 0 0 0
0 2 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 0 2 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 0 0 0 2 0 0 0 0 0 0 0 0 0 0 0
0 0 2 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2 2 0
0 0 0 2 0 0 0 0 0 2 0 0 0 0 0 0 0 0 0 2 0 0 2 0 0 0 0 0 0 0 0 0

Note that some of the bin counts in the portion of the histogram
shown above are as large as 4. A 3D plot of this portion of the
histogram as shown in Figure 8.5 is meant to make it easier to see
how many such bins exist in the portion of the histogram plotted.

If you comment out line (B8) and uncomment line (B9) to change
to the second option for the SBox, the same histogram will look
60
Computer and Network Security by Avi Kak Lecture 8

Figure 5: Shown is a portion of the histogram of the in-

put/output differentials for an SBox consisting of MI in
GF (28). (This figure is from Lecture 8 of Computer and Network Security by Avi Kak)

like what is shown in Figure 8.6. Recall that the second option
consists of doing byte substitution through the lookup table in
line (A9).

As you can see in Figure 8.6, the second option for the SBox
generates a more non-uniform histogram for the input/output
differentials. Ideally, for any input differential X, you would
want the output differential Y to be distributed uniformly with
a probability of 1/2n for an n-bit block cipher. (This would
translate into a count of 1 in every bin except for the (0, 0) bin
for reasons explained in the small-font note at the end of this
bullet.) That way, the output differentials will give an attacker
no clues regarding either the plaintext or the encryption key.
61
Computer and Network Security by Avi Kak Lecture 8

Figure 6: Shown is a portion of the histogram of the in-

put/output differentials for an SBox that carries out byte
substitutions by looking up the table supplied in line (A9).
(This figure is from Lecture 8 of Computer and Network Security by Avi Kak)

However, attaining this ideal is theoretically impossible. [As to

why the theoretical ideal is impossible to attain, lets first review some of the more noteworthy features of

such a histogram: (1) If we had shown the entire histogram, especially if the cell at (0, 0) was included in the

histogram display, you would see the largest peak located over the (0, 0) bin and the bin count associated with

this peak would be 256. This is a result of the fact that, in the double loop in lines (B1) through (B11) of the

script find differentials correlations.pl, as we scan through 256 different values for the first input byte,

and, for each input byte, as we scan through 256 values for the second input byte, there will be 256 cases when

the first and the second bytes are identical. Therefore, for these 256 cases, we will have X = 0, and also

Y = 0. This would give us a count of 256 over the (0, 0) bin. (2) Another feature that all such histograms

possess is that every non-zero bin count is even. This is on account of the fact that in the double loop in lines

(B1) through (B11) of the script, the same X occurs in multiples of 2 since X = Xi Xj = Xj Xi . (3)

The sum of all the counts in each row and each column must add up to 256. That is because, every differential

in the input must map one of 256 possible differentials at the output. (4) Therefore, for best such histograms

62
Computer and Network Security by Avi Kak Lecture 8

(that is, histograms with no biases in how the input and the output differentials are related), half the randomly

located bins in each row would contain a count of 2 (this would not apply to the bins in the topmost row or

the leftmost column). (5) For all the reasons stated here, the ideal of having a count of 1 in each bin of the

256 256 bins of the histogram is clearly not achievable even theoretically. ]

As the value of the variable $M in line (A2) of the script find differen
tials correlations.pl approaches 256, with the MI option for the
SBox in line (B8), you will see more and more bins showing the
best possible count of 2 in the histogram of Figure 8.5. On the
other hand, with the table lookup option in line (B9) for the
SBox, you will see the histogram in Figure 8.6 staying just as
non-uniform as it is now with the max peaks becoming some-
what larger.

The Perl script that follows, differential attack toy example.pl, is a

demonstration of how the sort of non-uniformities in the his-
togram of the input/output differentials can be exploited to re-
cover some portions of the key for at least the last round of a block
cipher. However, note that this script is only a toy example just
to get across the ideas involved in mounting a differential attack
on a block cipher. The logic presented in this script would not
work by any stretch of imagination on any realistic block cipher.

The script differential attack toy example.pl mounts a differential at-

tack on the encryption scheme in lines (C1) through (C14) of the
code. The SBox byte substitution is based on table lookup us-
ing the table supplied through line (A9). The byte returned by
63
Computer and Network Security by Avi Kak Lecture 8

table lookup is XORed with the round key. The round key is
shifted circularly by one bit position to the right for the next
round key. [For a more realistic simple example of a differential attack that involves
both an SBox and permutations in each round, the reader should look up the previously
mentioned tutorial A Tutorial on Linear and Differential Cryptanalysis by Howard
Heys. The block size in that tutorial is 4 bits.]

The beginning part of the differential attack toy example.pl script

that follows is the same as in the script find differentials correlations.pl
that you saw earlier in this section. Thats because, as mentioned
earlier, a differential attack exploits the predictability of the ci-
phertext differentials vis-a-vis the plaintext differentials. There-
fore, lines (A14) through (A27) of the script are devoted to the
calculation of a 2D histogram that measures the joint probabil-
ities of occurrence of the input and the output differentials. As
the comment lines explain, note how the information generated
is saved on the disk in the form of DBM files. So, as you are
experimenting with the attack logic in lines (B1) through (B26)
of the script and running the script over and over, you would not
need to generate the plaintext/ciphertext differentials histogram
each time. You can start from ground zero (that is, you can
re-generate the histogram) at any time provided you first call

clean_db_files.pl

to clear out the DBM files in your directory. The script clean db files.pl
is included in the code you can download from the lectures notes
website.
64
Computer and Network Security by Avi Kak Lecture 8

The plaintext/ciphertext differentials histogram is converted into

the hash %worst differentials in line (A22). In case you are wonder-
ing why we couldnt make do with the disk-based %worst differentials db
hash that is defined in line (A14), it is because the latter does
not support the exists() function that we need in line (B10) of
the script. The keys in both these hashes are the plaintext dif-
ferentials and, for each key, the value the ciphertext differential
where the histogram count exceeds the specified threshold. [Poten-
tial source of confusion: Please do not confuse the use of key as in the <key,value>
pairs that are stored in a Perl hash with the use of key as in encryption key.]

Finally, we mount the attack in line (A29). The attack itself is

implemented in lines (B1) through (B26). If you only specify one
round in line (A2), the goal of the attack would be estimate the
encryption key as specified by line (A3). However, if the number
of rounds exceeds 1, the goal of the attack is to estimate the key
in the last round key. The attack logic consists simply of scanning
through all possible plaintext differentials and using only those
that form the keys in the %worst differentials hash, finding the
corresponding the ciphertext differentials. Once we have chosen a
plaintext pair, and, therefore a plaintext differential, in line (B8),
we apply partial decryption to the corresponding ciphertext bytes
in lines (B21) and (B22). Subsequently, in line (B24), we check
whether the differential formed by the two partial decryptions
exists in our %worst differentials hash for each candidate last-round
key. If this condition is satisfied, a vote is cast for that candidate
key.

65
Computer and Network Security by Avi Kak Lecture 8

#!/usr/bin/perl -w

## differential_attack_toy_example.pl

## Avi Kak (March 4, 2015)

## This script is a toy example to illustrate some of the key elements of a

## differential attack on a block cipher.

## We assume that our block size is one byte and the SBox consists of finding a
## substitute byte by table lookup. We further assume that each round consists of
## one byte substitution step followed by xoring the substituted byte with the
## round key. The round key is the encryption key that is circularly shifted to the
## right by one position for each round.

## Since you are likely to run this script repeatedly as you experiment with
## different strategies for estimating the subkey used in the last round, the script
## makes it easy to do so by writing the information that is likely to stay constant
## from one run to the next to disk-based DBM files. The script creates the
## following DBM files:
##
## worst_differentials.dir and worst_differentials.pag -- See Line (A14)
##
## These DBM files are created the very first time you run this script. Your
## subsequent runs of this script will be much faster since this DBM database
## would not need to be created again. Should there be a need to run the script
## starting from ground zero, you can clear the DBM files created in your directory
## by calling the script:
##
## clean_db_files.pl
##
## Finally, if you set the number of tries in Line (A10) to a large number and you
## are tired of waiting, you can kill the script at any time you wish. To see the
## vote counts accumulated up to that point for the different possible candidates
## for the last round key, just run the script:
##
## get_vote_counts.pl
##
## The scripts clean_db_files.pl and get_vote_counts.pl are in the gzipped archive
## that goes with Lecture 8 at the lecture notes web site.

use strict;
use Algorithm::BitVector;
$|++;

my $debug = 1;
my $AES_modulus = Algorithm::BitVector->new(bitstring => 100011011); #(A1)
my $number_of_rounds = 1; #(A2)
my $encryption_key = Algorithm::BitVector->new(bitstring => 10001011); #(A3)
my $differential_hist; #(A4)
my %decryption_inverses; #(A5)
my %worst_differentials; #(A6)

66
Computer and Network Security by Avi Kak Lecture 8

my @worst_input_differentials; #(A7)
my @worst_output_differentials; #(A8)

my $hist_threshold = 8; #(A9)
my $tries = 500; #(A10)

unlink glob "votes.*"; #(A11)

dbmopen my %votes_for_keys, "votes", 0644
or die "cannot create DBM file: $!"; #(A12)

# This lookup table is used for the byte substituion step during encryption in the
# subroutine defined in lines (C1) through (C14). By experimenting with the script
# differentials_frequency_calculator.pl this lookup table was found to yield a good
# non-uniform histogram for the plaintext/ciphertext differentials.
my @lookuptable = qw(213 170 104 116 66 14 76 219 200 42 22 17 241 197 41 216 85 140
183 244 235 6 118 208 74 218 99 44 1 89 11 205 195 125 47 236 113
237 131 109 102 9 21 220 59 154 119 148 38 120 13 217 16 100 191 81
240 196 122 83 177 229 142 35 88 48 167 0 29 153 163 146 166 77 79
43 10 194 232 189 238 164 204 111 69 51 126 62 211 242 70 214 247 55
202 78 239 114 184 112 228 84 152 187 45 49 175 58 253 72 95 19 37
73 145 87 198 71 159 34 91 168 250 255 8 121 96 50 141 181 67 26 243
130 68 61 24 105 210 172 139 136 128 157 133 80 93 39 2 143 161 186 33
144 178 30 92 138 169 86 249 252 155 193 63 223 203 245 129 4 171
115 3 40 151 7 188 231 174 25 23 207 180 56 46 206 215 227 162 199
97 147 182 149 108 36 132 5 12 103 110 209 160 137 53 224 185 173
20 222 246 28 179 134 75 254 57 60 234 52 165 225 248 31 230 156
124 233 158 27 18 94 65 32 54 106 192 221 190 101 98 251 212 150
201 117 127 107 176 226 135 123 82 15 64 90); #(A13)

# In what follows, we first check if the worst_differentials DBM files were created
# previously by this script. If they are already on the disk, create the disk-based
# hash %worst_differentials_db from the data in those files. If not, create the DBM
# files so that they can subsequently be populated by the call in line (A18).
# [IMPORTANT: In a more realistic attack logic, you will need to create a more
# general version of the code in lines (A14) through (A21) so that you find the
# histogram for the plaintext/ciphertext differentials not for just one round, but
# for all the rounds involved. See the tutorial by Howard Heys for this important
# point.]
dbmopen my %worst_differentials_db, "worst_differentials", 0644
or die "Cant open DBM file: $!"; #(A14)
unless (keys %worst_differentials_db) { #(A15)
foreach my $i (0..255) { #(A16)
foreach my $j (0..255) { #(A17)
$differential_hist->[$i][$j] = 0; #(A18)
}
}
gen_differential_histogram(); #(A19)
# The call shown below will show that part of the histogram for which both
# the input and the output differentials are in the range (32, 63).
display_portion_of_histogram(32, 64) if $debug; #(A20)
# From the 2D input/output histogram for the differentials, now represent that
# information has a hash in which the keys are the plaintext differentials and
# the value associated with each key the ciphertext differential whose histogram
# count exceeds the supplied threshold:
find_most_probable_differentials($hist_threshold); #(A21)

67
Computer and Network Security by Avi Kak Lecture 8

}
%worst_differentials = %worst_differentials_db; #(A22)
die"no candidates for differentials: $!" if keys %worst_differentials == 0; #(A23)
@worst_input_differentials = sort {$a <=> $b} keys %worst_differentials; #(A24)
@worst_output_differentials = @worst_differentials{@worst_input_differentials}; #(A25)
if ($debug) {
print "\nworst input differentials: @worst_input_differentials\n"; #(A26)
print "\nworst output differentials: @worst_output_differentials\n"; #(A27)
}

# The following call makes a hash that does the opposite of what is achieved by
# indexing into the lookup table of line (A13). It fills the hash
# %decryption_inverses with <key,value> pairs, with the keys being the ciphertext
# bytes and the values being the corresponding plaintext bytes.
find_inverses_for_decryption(); #(A28)

estimate_last_round_key(); #(A29)

# Now print out the ten most voted for keys. To see the votes for all possible keys,
# execute the script get_vote_counts.pl separately after running this script.
print "no votes for any candidates for the last round key\n"
if keys %votes_for_keys == 0; #(A30)
if (scalar keys %votes_for_keys) { #(A31)
my @vote_sorted_keys =
sort {$votes_for_keys{$b} <=> $votes_for_keys{$a}} keys %votes_for_keys; #(A32)
print "\nDisplaying the keys with the largest number of votes: @vote_sorted_keys[0..9]\n";
#(A33)
}

################################### Subroutines ###########################################

# The differential attack:

sub estimate_last_round_key { #(B1)
my $attempted = 0; #(B2)
foreach my $i (2..255) { #(B3)
print "+ " if $debug; #(B4)
my $plaintext1 = Algorithm::BitVector->new(intVal => $i, size => 8); #(B5)
foreach my $j (2..255) { #(B6)
my $plaintext2 = Algorithm::BitVector->new(intVal => $j, size => 8); #(B7)
my $input_differential = $plaintext1 ^ $plaintext2; #(B8)
next if int($input_differential) < 2; #(B9)
next unless exists $worst_differentials{int($input_differential)}; #(B10)
print "- " if $debug; #(B11)
my ($ciphertext1, $ciphertext2) = #(B12)
(encrypt($plaintext1, $encryption_key), encrypt($plaintext2, $encryption_key));
my $output_differential = $ciphertext1 ^ $ciphertext2; #(B13)
next if int($output_differential) < 2; #(B14)
last if $attempted++ > $tries; #(B15)
print " attempts made $attempted " if $attempted % 500 == 0; #(B16)
print "| " if $debug; #(B17)
foreach my $key (0..255) { #(B18)
print ". " if $debug; #(B19)
my $key_bv = Algorithm::BitVector->new(intVal => $key, size => 8); #(B20)
my $partial_decrypt_int1 = $decryption_inverses{int($ciphertext1 ^ $key_bv )};
#(B21)

68
Computer and Network Security by Avi Kak Lecture 8

my $partial_decrypt_int2 = $decryption_inverses{int($ciphertext2 ^ $key_bv )};

#(B22)
my $delta = $partial_decrypt_int1 ^ $partial_decrypt_int2; #(B23)
if (exists $worst_differentials{$delta}) { #(B24)
print " voted " if $debug; #(B25)
$votes_for_keys{$key}++; #(B26)
}
}
}
}
}

sub encrypt { #(C1)

my $plaintext = shift; # must be a bitvector #(C2)
my $key = shift; # must be a bitvector #(C3)
my $round_input = $plaintext; #(C4)
my $round_output; #(C5)
my $round_key = $key; #(C6)
if ($number_of_rounds > 1) { #(C7)
foreach my $round (0..$number_of_rounds-1) { #(C8)
$round_output = get_sbox_output_lookup($round_input) ^ $round_key; #(C9)
$round_input = $round_output; #(C10)
$round_key = $round_key >> 1; #(C11)
}
} else { #(C12)
$round_output = get_sbox_output_lookup($round_input) ^ $key; #(C13)
}
return $round_output; #(C14)
}

# Since the SubBytes step in encryption involves taking the square of a byte in
# GF(2^8) based on AES modulus, for invSubBytes step for decryption will involve
# taking square-roots of the bytes in GF(2^8). This subroutine calculates these
# square-roots.
sub find_inverses_for_decryption { #(D1)
foreach my $i (0 .. @lookuptable - 1) {
$decryption_inverses{$lookuptable[$i]} = $i;
}
}

# This function represents the histogram of the plaintext/ciphertext differentials

# in the form of a hash in which the keys are the plaintext differentials and the
# value for each plaintext differential the ciphertext differential where the
# histogram count exceeds the threshold.
sub find_most_probable_differentials { #(F1)
my $threshold = shift; #(F2)
foreach my $i (0..255) { #(F3)
foreach my $j (0..255) { #(F4)
$worst_differentials_db{$i} = $j if $differential_hist->[$i][$j] > $threshold;#(F5)
}
}
}

# This subroutine generates a 2D histogram in which one axis stands for the
# plaintext differentials and the other axis the ciphertext differentials. The

69
Computer and Network Security by Avi Kak Lecture 8

# count in each bin is the number of times that particular relationship is seen
# between the plaintext differentials and the ciphertext differentials.
sub gen_differential_histogram { #(G1)
foreach my $i (0 .. 255) { #(G2)
print "\ngen_differential_hist: i=$i\n" if $debug; #(G3)
foreach my $j (0 .. 255) { #(G4)
print ". " if $debug; #(G5)
my ($a, $b) = (Algorithm::BitVector->new(intVal => $i, size => 8),
Algorithm::BitVector->new(intVal => $j, size => 8)); #(G6)
my $input_differential = int($a ^ $b); #(G7)
my ($c, $d) = (get_sbox_output_lookup($a), get_sbox_output_lookup($b)); #(B9)
my $output_differential = int($c ^ $d); #(G9)
$differential_hist->[$input_differential][$output_differential]++; #(G10)
}
}
}

sub get_sbox_output_lookup { #(D1)

my $in = shift; #(D2)
return Algorithm::BitVector->new(intVal => $lookuptable[int($in)], size => 8); #(D3)
}

# Displays in your terminal window the bin counts in the two-dimensional histogram
# for the input/output mapping of the differentials. You can control the portion of
# the 2D histogram that is output by using the first argument to set the lower bin
# index and the second argument the upper bin index along both dimensions.
# Therefore, what you see is always a square portion of the overall histogram.
sub display_portion_of_histogram { #(J1)
my $lower = shift; #(J2)
my $upper = shift; #(J3)
foreach my $i ($lower .. $upper - 1) { #(J4)
print "\n"; #(J5)
foreach my $j ($lower .. $upper - 1) { #(J6)
print "$differential_hist->[$i][$j] "; #(J7)
}
}
}

When you run the script for just one round, that is, when you set
the value of the variable $number_of_rounds to 1 in line (A2),
you should get the following answer for the ten encryption keys
that received the largest number of notes (in decreasing order of
the number of votes received):
139 51 200 225 108 216 161 208 26 140
This answer is how youd expect it to be since the decimal 139
70
Computer and Network Security by Avi Kak Lecture 8

is equivalent to the binary 10001011, which is the encryption

key set in line (A3) of the script. For a more detail look at the
distribution of the votes for the keys, execute the script:

get_vote_counts.pl

This script will return an answer like

139: 501 51: 40 200: 40 225: 40 108: 39 ...
where the number following the colon is the number for votes for
the integer value of the encryption key shown at the left of the
colon.

If you run the attack script with the number of rounds set to 2
in line (A2), you should see the following answer for the ten keys
that received the largest number of votes:
82 180 214 20 72 44 109 105 52 174
This answer says that the most likely key used in the second round
is the integer 82, which translates into the binary 01010010. If
you examine the logic of encryption in lines (C1) through (C14)
especially if you focus on how the round key is set in line
(C11) the answer returned by the script is incorrect. However,
that is not surprising since our input/output histogram of the
differentials is based on just one round. As explained in the
previously mentioned tutorial by Howard Heys, we would need
to construct a more elaborate model of the differentials propagate
through multiple rounds for the script to do better in those cases.

71
Computer and Network Security by Avi Kak Lecture 8

That brings us to the subject of linear attacks on block ciphers.

A linear attack on a block cipher is a known plaintext attack. In
such attacks, the adversary has access to a set of plaintexts and
the corresponding ciphertexts. However, unlike the differential
attack, the adversary does not choose any specific subset of these.

A linear attack exploits linear relationships between the bits to

the input to the SBox and the bits at the output. Let (X0 , X1, X2, X3,
X4 , X5 , X6 , X7 ) represent the bits at the input to an SBox and let
(Y0 , Y1, Y2 , Y3 , Y4, Y5 , Y6 , Y7 ) represent the bits at the output. Should it
be the case that there exist bit positions i1, . . . , im at the input and
the bit positions j1, . . . , jn at the output for some values of m and
n so that the following is either often true or often false: [The phrase
often true here means significantly above average and the phrase often false means significantly

below average. For an ideal SBox, such relationships would be true (and, therefore, false) with a

probability of 0.5 for all sets of bit positions at the input and the output. ]

X i1 X i2 . . . X im Y i1 Y i2 . . . Y in = 0 (5)

that fact can be exploited by a linear attack to make good esti-

mates for the key bits in at least the last round of a block cipher.

As should be obvious to the reader, the linear relationship shown

above can also be expressed as

X i1 X i2 . . . X im = Y i1 Y i2 . . . Y in (6)

72
Computer and Network Security by Avi Kak Lecture 8

It is important to realize that such a relationship must hold for

all possible values of 0s and 1s at the bit positions in question.
Consider the case when the list of output bit positions is empty.
Now there will be 128 out of 256 different possible bit patterns at
the input for which the linear relationship, as shown in Equation
(5), will be satisfied. For this case, whenever an input bit pattern
has an even number of 1s, its XOR-sum will be zero. (And that
will happen in 128 out of 256 cases.) The same would be the
case when we consider an empty set of input bits and all possible
variations on the output bits.

It must be emphasized that the linear attack exploits not only

those bit positions at the input and the output when the linear
relationship is often true, it also exploits those bit positions when
such linear relationships are often false. The inequality case of
the linear relationships of the sort shown above are more correctly
referred to as affine relationships.

As mentioned earlier, for an ideal SBox, all linear (and affine)

relationships of the sort shown above will hold with a probability
of 0.5. That is, if you feed the 256 possible different bit patterns
into the input of a SBox, you should see such relationships to hold
128 times for all possible groupings of the input bit positions and
the output bit positions. Any departure from this average is
referred to as linear approximation bias. It is this bias that is
exploited by a linear attack on a block cipher

73
Computer and Network Security by Avi Kak Lecture 8

So our first order of business is to characterize an SBox with

regard to the prevalence of linear approximation biases. The two
independent variables in a depiction of this bias are the input
bit positions and the output bit positions. We can obviously
express both with integers that range from 0 to 255, both ends
inclusive. We will refer to these two integers as the bit grouping
integers. The bits that are set in the bit-pattern representations
of the two integers tell us which bit positions are involved in
a linear (or an affine) relationship. For example, when the bit
grouping integer for the input bits is 3 and the one for the output
bits is 12, we are talking about the following relationship:

X0 X1 = Y 2 Y 3

By scanning through all 256 different possible bit patterns at the

input to an SBox, and through the corresponding 256 different
possible output bit patterns, we can count the number of times
the equations of the type shown in Eq. (6) are satisfied. After
we subtract the average value of 128 from these counts, we get
what is referred to as the linear approximation table (LAT).

What follows is a Perl script that can calculate LAT for two
different choices of the SBox, depending on which of the two lines,
(B4) or (B5), is left uncommented. The statement in line (B4)
gives you an SBox that replaces a byte with its MI in GF (28)
based on the AES modulus. On the other hand, the statement in
line (B5) gives an SBox that is based on the lookup table defined
74
Computer and Network Security by Avi Kak Lecture 8

in line (A8). The LAT itself is calculated in lines (B1) through

(B26) of the script.

#!/usr/bin/perl -w

## linear_approximation_table_generator.pl

## Avi Kak (March 5, 2015)

## This script demonstrates how to generate the Linear Approximation Table that is
## needed for mounting a Linear Attack on a block cipher.

use strict;
use Algorithm::BitVector;
use Graphics::GnuplotIF;
$|++;

my $debug = 1;
my $AES_modulus = Algorithm::BitVector->new(bitstring => 100011011); #(A1)

# Initialize LAT:
my $linear_approximation_table; #(A2)
foreach my $i (0..255) { #(A3)
foreach my $j (0..255) { #(A4)
$linear_approximation_table->[$i][$j] = 0; #(A5)
}
}

# When SBox is based on lookup, we will use the "table" created by randomly
# permuting the the number from 0 to 255:
#my $lookuptable = shuffle([0..255]); #(A6)
#my @lookuptable = @$lookuptable; #(A7)
my @lookuptable = qw(213 170 104 116 66 14 76 219 200 42 22 17 241 197 41 216 85 140
183 244 235 6 118 208 74 218 99 44 1 89 11 205 195 125 47 236 113
237 131 109 102 9 21 220 59 154 119 148 38 120 13 217 16 100 191 81
240 196 122 83 177 229 142 35 88 48 167 0 29 153 163 146 166 77 79
43 10 194 232 189 238 164 204 111 69 51 126 62 211 242 70 214 247 55
202 78 239 114 184 112 228 84 152 187 45 49 175 58 253 72 95 19 37
73 145 87 198 71 159 34 91 168 250 255 8 121 96 50 141 181 67 26 243
130 68 61 24 105 210 172 139 136 128 157 133 80 93 39 2 143 161 186 33
144 178 30 92 138 169 86 249 252 155 193 63 223 203 245 129 4 171
115 3 40 151 7 188 231 174 25 23 207 180 56 46 206 215 227 162 199
97 147 182 149 108 36 132 5 12 103 110 209 160 137 53 224 185 173
20 222 246 28 179 134 75 254 57 60 234 52 165 225 248 31 230 156
124 233 158 27 18 94 65 32 54 106 192 221 190 101 98 251 212 150
201 117 127 107 176 226 135 123 82 15 64 90); #(A8)

gen_linear_approximation_table(); #(A9)

75
Computer and Network Security by Avi Kak Lecture 8

# The call shown below will show that part of the LAT for which both the input and
# the output bit grouping integers are in the range (0, 32):
display_portion_of_LAT(0, 32); #(A10)

# This call makes a graphical plot for a portion of the LAT. The bit grouping index
# ranges for both the input and the output bytes are 32 to 64:
plot_portion_of_LAT($linear_approximation_table, 32, 64); #(A11)
## The following call makes a hardcopy of the plot:
plot_portion_of_LAT($linear_approximation_table, 32, 64, 3); #(A12)

# You have two choices for the SBox in lines (B4) and (B5). The one is line (B4) is
# uses MI in GF(2^8) based on the AES modulus. And the one in line (B5) uses the
# lookup table defined above in line (A8). Comment out the one you do not want.
sub gen_linear_approximation_table {
foreach my $x (0 .. 255) { # specify a byte for the input to the SBox #(B1)
print "\input byte = $x\n" if $debug; #(B2)
my $a = Algorithm::BitVector->new(intVal => $x, size => 8); #(B3)
# Now get the output byte for the SBox:
my $c = get_sbox_output_MI($a); #(B4)
# my $c = get_sbox_output_lookup($a); #(B5)
my $y = int($c); #(B6)
foreach my $bit_group_from_x (0 .. 255) { #(B7)
my @input_bit_positions; #(B8)
foreach my $pos (0..7) { #(B9)
push @input_bit_positions, $pos if ($bit_group_from_x >> $pos) & 1; #(B10)
} #(B11)
my $input_linear_sum = 0; #(B12)
foreach my $pos (@input_bit_positions) { #(B13)
$input_linear_sum ^= (($x >> $pos) & 1); #(B14)
}
foreach my $bit_group_from_y (0 .. 255) { #(B15)
my @output_bit_positions; #(B16)
foreach my $pos (0..7) { #(B17)
push @output_bit_positions, $pos if ($bit_group_from_y >> $pos) & 1; #(B18)
}
my $output_linear_sum = 0; #(B19)
foreach my $pos (@output_bit_positions) { #(B20)
$output_linear_sum ^= (($y >> $pos) & 1); #(B21)
}
$linear_approximation_table->[$bit_group_from_x][$bit_group_from_y]++ #(B22)
if $input_linear_sum == $output_linear_sum; #(B23)
}
}
}
foreach my $i (0 .. 255) { #(B24)
foreach my $j (0 .. 255) { #(B25)
$linear_approximation_table->[$i][$j] -= 128; #(B26)
}
}
}

sub get_sbox_output_MI { #(C1)

my $in = shift; #(C2)
return int($in) != 0 ? $in->gf_MI($AES_modulus, 8) : #(C3)
Algorithm::BitVector->new(intVal => 0); #(C4)

76
Computer and Network Security by Avi Kak Lecture 8

sub get_sbox_output_lookup { #(D1)

my $in = shift; #(D2)
return Algorithm::BitVector->new(intVal => $lookuptable[int($in)], size => 8); #(D3)
}

# Fisher-Yates shuffle:
sub shuffle { #(E1)
my $arr_ref = shift; #(E2)
my $i = @$arr_ref; #(E3)
while ( $i-- ) { #(E4)
my $j = int rand( $i + 1 ); #(E5)
@$arr_ref[ $i, $j ] = @$arr_ref[ $j, $i ]; #(E6)
} #(E7)
return $arr_ref; #(E8)
}

######################### Support Routines for Displaying LAT ##############################

# Displays in your terminal window the bin counts (minus 128) in the LAT calculated
# in lines (B1) through (B26). You can control the portion of the display by using
# the first argument to set the lower bin index and the second argument the upper
# bin index along both dimensions. Therefore, what you see is always a square
# portion of the LAT.
sub display_portion_of_LAT { #(F1)
my $lower = shift; #(F2)
my $upper = shift; #(F3)
foreach my $i ($lower .. $upper - 1) { #(F4)
print "\n"; #(F5)
foreach my $j ($lower .. $upper - 1) { #(F6)
print "$linear_approximation_table->[$i][$j] "; #(F7)
}
}
}

# Displays with a 3-dimensional plot a square portion of the LAT. Along both the X
# and the Y directions, the lower bound on the bin index is supplied by the SECOND
# argument and the upper bound by the THIRD argument. The last argument is needed
# only if you want to make a hardcopy of the plot. The last argument is set to the
# number of second the plot will be flashed in the terminal screen before it is
# dumped into a .png file.
sub plot_portion_of_LAT { #(G1)
my $hist = shift; #(G2)
my $lower = shift; #(G3)
my $upper = shift; #(G4)
my $pause_time = shift; #(G5)
my @plot_points = (); #(G6)
my $bin_width = my $bin_height = 1.0; #(G7)
my ($x_min, $y_min, $x_max, $y_max) = ($lower, $lower, $upper, $upper); #(G8)
foreach my $y ($y_min..$y_max-1) { #(G9)
foreach my $x ($x_min..$x_max-1) { #(G10)
push @plot_points, [$x, $y, $hist->[$y][$x]]; #(G11)
}
}

77
Computer and Network Security by Avi Kak Lecture 8

@plot_points = sort {$a->[0] <=> $b->[0]} @plot_points; #(G12)

@plot_points = sort {$a->[1] <=> $b->[1] if $a->[0] == $b->[0]} @plot_points; #(G13)
my $temp_file = "__temp.dat"; #(G14)
open(OUTFILE , ">$temp_file") or die "Cannot open temporary file: $!"; #(G15)
my ($first, $oldfirst); #(G16)
$oldfirst = $plot_points[0]->[0]; #(G17)
foreach my $sample (@plot_points) { #(G18)
$first = $sample->[0]; #(G19)
if ($first == $oldfirst) { #(G20)
my @out_sample; #(G21)
$out_sample[0] = $sample->[0]; #(G22)
$out_sample[1] = $sample->[1]; #(G23)
$out_sample[2] = $sample->[2]; #(G24)
print OUTFILE "@out_sample\n"; #(G25)
} else { #(G26)
print OUTFILE "\n"; #(G27)
}
$oldfirst = $first; #(G28)
}
print OUTFILE "\n";
close OUTFILE;
my $argstring = <<"END"; #(G29)
set xrange [$x_min:$x_max]
set yrange [$y_min:$y_max]
set view 80,15
set hidden3d
splot "$temp_file" with lines
END
unless (defined $pause_time) { #(G30)
my $hardcopy_name = "LAT.png"; #(G31)
my $plot1 = Graphics::GnuplotIF->new(); #(G32)
$plot1->gnuplot_cmd( set terminal png, "set output \"$hardcopy_name\""); #(G33)
$plot1->gnuplot_cmd( $argstring ); #(G34)
my $plot2 = Graphics::GnuplotIF->new(persist => 1); #(G35)
$plot2->gnuplot_cmd( $argstring ); #(G36)
} else { #(G37)
my $plot = Graphics::GnuplotIF->new(); #(G38)
$plot->gnuplot_cmd( $argstring ); #(G39)
$plot->gnuplot_pause( $pause_time ); #(G40)
}
}

For the case when you run script with the SBox based on MI
calculations in GF (28), shown below is a small portion of the LAT
constructed by the script. [The portion of the LAT shown below was dictated by the page
width constraints.] In keeping with the explanation provided earlier,
you can see that the topmost row and the leftmost column values

78
Computer and Network Security by Avi Kak Lecture 8

are all zero, as we expect them to be. The entries at the other
locations tell us how much positive and negative bias there exists
in the linear relationships corresponding to those cells. Looking
at the seventh entry (of column index 6) in the second row (of
row index 1), we can say that the relationship X1 X2 Y1 = 0 is
true with a probability of 12/256, and so on. Note that the full
table that is calculated by the Perl script is 256 256. Theory
dictates that the sum of the entries in each row or each column
must be either 128 or -128.

128 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 -6 8 -14 4 6 12 6 -2 12 -2 -4 -6 -8 2 -8 -12 -2 12 -2 -8 -14 0 -6 -2
0 8 12 4 -8 -8 -4 12 -12 -12 8 8 -8 0 12 -12 -14 -6 -6 -6 -6 10 10 2 2
0 -14 4 6 12 -2 -8 2 -2 12 -14 0 -2 -12 10 8 -2 8 -2 8 2 4 -6 4 -12
0 4 -8 12 -12 8 0 12 -6 -6 10 10 2 -6 -2 -2 2 -6 -2 14 14 6 -2 6 -8
0 6 -8 -2 8 6 -12 -14 -4 10 -12 -14 -8 -2 4 -6 -14 4 6 -8 -6 -4 2 4 -2
0 12 -4 -8 0 -12 8 -12 -6 -6 -2 -2 -6 10 -6 -14 12 -12 8 0 12 12 4 -4 14
0 6 12 2 12 -14 -12 10 8 -2 4 -6 -12 -6 4 10 12 -2 0 10 -8 2 8 -6 16
0 -2 -12 -2 -6 -4 -6 8 -8 -6 4 -6 -2 -12 -2 8 8 -10 4 14 -6 12 -14 16 -12
0 12 -12 12 -6 10 -6 -2 -6 -2 6 -2 -8 8 0 12 0 4 4 12 -14 2 2 -2 10
0 -2 8 -14 10 -12 -2 4 4 6 -4 2 -2 12 2 4 -2 12 2 4 0 -6 0 -2 2
0 -4 8 0 10 -14 -2 -6 -6 -2 2 -6 12 4 8 12 6 -14 2 10 0 0 -8 -4 -4
0 -6 -8 -2 2 -8 -6 -12 -2 -8 -2 12 16 14 16 -6 -2 -12 10 -12 0 2 -4 10 12
0 -8 0 -12 -6 -2 10 -6 -12 8 12 4 14 -2 -2 2 -6 14 -2 -2 -4 -12 0 4 -14
0 2 12 10 -2 4 -6 4 -2 0 2 8 16 -2 12 6 -12 10 8 10 10 4 -2 -4 14
0 -8 -12 8 -2 -6 -14 10 8 12 4 12 -6 2 6 -14 -8 12 -12 12 -2 -2 -6 -2 0
0 -12 -14 -2 2 -14 12 12 8 0 -2 6 -2 -6 -12 -8 12 -8 10 6 -2 14 12 -12 -12
0 -2 -6 8 -6 4 -12 -2 -10 4 12 -14 -12 14 10 12 -8 -2 -2 4 10 -4 8 10 -14
0 12 -6 -2 -2 6 8 0 4 4 2 2 10 -2 8 -12 10 -2 -4 -8 0 -8 10 2 2
0 -2 -6 8 14 -8 0 10 14 12 4 10 -12 -2 10 12 6 4 -8 -2 12 -10 14 0 12
0 -8 -6 2 14 -6 12 -8 -6 -14 0 0 0 -4 10 -2 -2 10 0 12 4 -4 2 10 12
0 -14 10 4 6 -4 12 2 12 2 -6 0 2 -12 4 -2 14 -4 -8 -10 -4 -2 10 4 -6
0 0 10 -6 -2 2 4 8 -14 2 0 -8 -4 0 -2 -6 12 8 10 14 2 10 4 -12 6
0 -6 2 4 6 4 -4 -6 16 -2 -2 -4 10 4 -4 -2 -12 10 2 0 10 4 -12 14 8
0 -2 2 -12 -8 -2 14 16 -12 10 2 -4 12 -14 14 0 -12 -14 2 12 12 -6 6 8 -4
0 -12 -6 -6 -12 8 -6 -6 6 14 12 0 -6 -14 4 8 12 8 -6 2 8 4 -6 -14 10
0 6 -6 -12 4 2 -14 -4 -8 -2 -2 -8 8 -2 -6 -4 -10 -4 0 -6 2 -8 -8 10 6
0 -12 -6 -6 8 12 -2 -2 14 -2 -12 0 -14 2 4 0 6 -6 -8 16 6 -6 -4 4 0

Shown in Figure 8.7 is a plot of a portion of the LAT that was

calculated by the Perl script for the case of SBox based on MI

79
Computer and Network Security by Avi Kak Lecture 8

Figure 7: Shown is a portion of the LAT for an SBox that

calculates MIs in GF (28) using the AES modulus. (This figure
is from Lecture 8 of Computer and Network Security by Avi Kak)

in GF (28). The portion shown is a 32 32 portion of the table

starting at the cell located at (32, 32).

If you comment out line (B3) and uncomment line (B4) so that
the SBox would be based on the lookup table in line (A8), the
portion of the plot shown in Figure 8.7 becomes what is shown
in Figure 8.8. Note that the largest peaks in the LAT of Figure
8.8 are larger than the largest peaks in the LAT of Figure 8.7.
That implies that the SBox based on the lookup table of line (A8)
results in larger biases for some of the linear equations compared
to the SBox that is based on MI in GF (28).

80
Computer and Network Security by Avi Kak Lecture 8

Figure 8: Shown is a portion of the LAT for an SBox that

carries out byte substitutions by looking up the table sup-
plied in line (A8) of the LAT generator script. (This figure is
from Lecture 8 of Computer and Network Security by Avi Kak)

Representing an arbitray linear form Xi . . . Xi Yj . . . Yj by

1 m 1 n

, the cell values in a LAT allow us to write down the following

probabilities: prob( = 0) = p and prob( = 1) = 1 p.

An important part of the formulation of the linear attack is the

use of Matsuis piling-up lemma to estimate the joint probabili-
ties prob(1, 2, . . .) = 0 and prob(1, 2, . . .) = 1 with each
i expressing one of the linear forms for the ith round.

After constructing a LAT for the SBoxes used in a cipher and

after estimating the join probabilities associated with the linear
equations over multiple rounds, executing a linear attack involves
the following steps: (1) You string together the linear forms of
81
Computer and Network Security by Avi Kak Lecture 8

the type shown earlier across the rounds but not including the
last round and estimate the probabilistic biases associated with
the linear forms (these can also be affine forms). (2) Considering
different possible candidate keys for the last round, you partially
decrypt the ciphertext. For each candidate key for the last round,
this gives you candidate output bits for the last-round Sbox. (3)
Using these candidate output bits for the last round, you accu-
mulate votes for the different candidates for the last-round key
depending on the extent to which candidate SBox output bits for
the last round are consistent with the linear forms constructed
from the first n 1 rounds.

To make the above explanation more specific, assume that the

block size in our cipher is just one byte and that there are no
permutations involved. [See the previously mentioned tutorial by Howard Heys
for a more realistic example that involves both substitutions and permutations.] Let
Pi denote the ith bit of the plaintext byte entering the first round.
We will assume that each round consists of a byte substitution by
the SBox, followed by the addition of the round key. In general,
we will use Xr,i to denote the ith input bit to the rth round and
let Yr,j denote the j th output bit of the SBox in the same round.
Additionally, let Kr,k denote the k th bit of the round key for
the rth round. We can now construct linear relationships of the
following sort that span all of the rounds together:

Pi1 Pi2 . . . Y1,j1 . . . Y1,j1 . . . Y1,jm . . . K1 Y2,j1 . . . Y2,j1 . . .

. . . K2 Y3,j1 . . . Y3,j1 . . . . . . Yn1,j1 . . . Yn1,jm = 0

82
Computer and Network Security by Avi Kak Lecture 8

where we have used the fact that the output of each SBox, after
the addition of the round key, becomes the input to the next
round. That is, Yr,i Kr,i becomes Xr+1,i. The above linear
form may be expressed in the following form:

XOR sum of only P and Y variables

XOR sum of key bits in rounds f rom 1 through n 1 = 0

which can be abbreviated to

XOR sum of only P and Y variables K = 0

where K is the linear form that involves only the key bits from
the first n 1 rounds.

We have only two possibilities for K . Either it is equal to 0 or to

1. If we assume that both are equiprobable, that eliminates the
influence of K on the bias associated with the rest of the linear
equation shown above. Subsequently, it becomes easy to decide
how much weight to give to a candidate key for the last round
depending on the probability associated with the linear form that
depends only on the inputs and the outputs of the Sboxes.

That brings us to the interpolation attack. The interpolation

attack seeks to model the behavior of an SBox with a polynomial
in GF (28). Recall that the SBox is the only source of nonlinearity
in transforming plaintext into ciphertext. (All of the permutation
operations are obviously linear.) We also recognize that what
an SBox does must be invertible on a one-one basis (in other

83
Computer and Network Security by Avi Kak Lecture 8

words, the input/output mapping provided by an SBox must be

bijective).

Lets say that it is possible to represent the round operation that

involves an SBox calculation following by key mixing by the al-
gebraic function fi(ci1 , Ki) where ci1 is the input to the round
and Ki is the round key. Lets further say that fi can be expressed
as a polynomial in GF (28) over the input to the round and that
the unknown Ki values can be expressed as the coefficients of this
polynomial. It was shown by Jakobsen and Knudsen that when
such a polynomial is of low degree, its coefficients can be esti-
mated from a set of plaintext-ciphertext pairs. Subsequently, an
attacker would be able to invert the polynomial to find the plain-
text for a given ciphertext without having to know the encryption
key used.

84
Computer and Network Security by Avi Kak Lecture 8

8.10: HOMEWORK PROBLEMS

1. With regard to the first step of processing in each round of AES

on the encryption side: How does one look up the 16 16 S-box
table for byte-by-byte substitutions? In other words, assuming I
want a substitute byte for the byte b7b6b5b4b3b2b1b0, where each
bi is a single bit, how do I use these bits to find the replacement
byte in the S-box table?

2. What are the steps that go into the construction of the 1616 S-
box lookup table?

3. What is rationale for the bit scrambling step that is used for
finding the replacement byte that goes into each cell of the S-box
table?

4. The second step in each round permutes the bytes in each row of
the state array. What is the permutation formula that is used?

5. Describe the mix columns transformation that constitutes the

third step in each round of AES.

85
Computer and Network Security by Avi Kak Lecture 8

6. Lets now talk about the Key Expansion Algorithm of AES. This
algorithm starts with how many words of the key matrix and
expands that into how many words?

7. Lets say the first four words of the key schedule are w0, w1, w2 , w3.
How do we now obtain the next four words w4 , w5, w6 , w7?

8. Going back to the previous question, the formula that yields w4

is
w4 = w0 g(w3)
What goes into computing g()?

9. Programming Assignment:

Write a Perl or Python based implementation of AES. As you

know, each round of processing involves the following four steps:
byte-by-byte substitution
shifting of the rows of the state array
mixing of the columns
the addition of the round key.
Your implementation must include the code for creating the two
16 16 tables that you need for the byte substitution steps,
one for encryption and the other for decryption. Note that the
lookup table you construct for encryption is also used in the key
expansion algorithm.
86
Computer and Network Security by Avi Kak Lecture 8

The effort that it takes to do this homework is sig-

nificantly reduced if you use the BitVector module in
Python and the Algorithm::BitVector module in Perl.
The following method of in these modules should be particularly
useful for constructing the two lookup tables for byte substitu-
tions:
gf_MI
This method returns the multiplicative inverse of a bit pattern in
GF (2n) with respect to a modulus bit pattern that corresponds
to the irreducible polynomial used. To illustrate with Python the
sort of call youd need to make, the API documentation for the
BitVector module shows the following example code on how to
call this method:
modulus = BitVector(bitstring = 100011011)
n = 8
a = BitVector(bitstring = 00110011)
multiplicative_inverse = a.gf_MI(modulus, n)
print multiplicative_inverse # 01101100

Note that the variable modulus is set to the BitVector that corre-
sponds to the AES irreducible polynomial. The variable a can be
set to any arbitrary BitVector whose multiplicative inverse you
are interested in.
The other BitVector method that should prove particularly useful
for this homework is:
gf_multiply_modular
This method lets you multiply two bit patterns in GF (2n). To
multiply two bit patterns a and b, both instances of the BitVector
87
Computer and Network Security by Avi Kak Lecture 8

class, when the modulus bit pattern is mod, you invoke

a.gf_multipy_modular(b, mod, n)
where n is the exponent for 2 in GF (2n). For this homework
problem, n is obviously 8.
Your implementation should be for a 128 bit encryption key. Your
script should read a message file for the plaintext and write out
the ciphertext into another file. It should prompt the user for
the encryption key which should consist of at least 16 printable
ASCII characters.

88