Authors Jonathan A. Poritz
License CC-BY-SA-4.0
Open Workbook of Cryptology
A project-based introduction to crypto in Python
Jonathan A. Poritz
Department of Mathematics and Physics
Colorado State University, Pueblo
2200 Bonforte Blvd.
Pueblo, CO 81001, USA
E-mail: jonathan@poritz.net
Web: poritz.net/jonathan
11 J UN 2021 10:35MDT
Release Notes
This is a first draft of a free (as in speech, not as in beer, [Sta02]) (although it is free as
in beer as well) textbook for a one-semester, undergraduate cryptology course. It was used
for CIS 491 at Colorado State University Pueblo in the spring semester of 2021.
It’s not clear that any other instructor would feel comfortable using this version, without
a lot of adaptation. In fact, I usually find that when I write an open textbook like this one,
during the semester in which I am teaching the class, what results is a little rough and has
a few real gaps. After another semester teaching with this book, the second version (which
could well be twice as long!) will likely be much easier to use by others.
Nevertheless, in the spirit of openness, I’m sharing this first draft as well: if you want
to use it for self-study or in a class you are teaching, feel free to do so – just makes sure
your seatbelt is fastened, as the ride might be a little bumpy at times. [That said, if you do
use this and find typos, thinkos, or other issues, I would be grateful if you would tell me
about them!]
As I make improvements and additions to this work, you will always be able to find
them on the page https://poritz.net/jonathan/share/owoc/, so please look
there for the best version.
I am releasing this work under a Creative Commons
Attribution-ShareAlike 4.0 International license, which allows
anyone to share (copy and redistribute in any medium or format)
and adapt (remix, transform, and build upon this work for any
purpose, even commercially). These rights cannot be revoked, so long as users follow the
license terms, which require attribution (giving appropriate credit, linking to the license,
and indicating if changes were made) to be given and share-alike (if you remix or trans-
form this work, you must distribute your contributions under the same license as this one)
to be respected. See creativecommons.org/licenses/by-sa/4.0 for all the details.
This version: 11 Jun 2021 10:35MDT.
Jonathan A. Poritz
Spring Semester, 2021
Pueblo, CO, USA
iii
Contents
Release Notes iii
Preface vii
Chapter 1. Preliminaries 1
1.1. Some speculative history 2
1.2. The Caesar cipher and its variants 6
1.2.1. Keys only matter “mod 26” 6
1.2.2. Modernizing the Caesar cipher 9
1.3. Cryptanalysis of the Caesar cipher 11
1.3.1. Frequency analysis 13
1.4. Defending Caesar against frequency analysis: Vigenère and the one-time pad 17
1.5. Preliminary conclusion of preliminaries 24
Chapter 2. Block Ciphers 27
2.1. Why encrypt blocks of data: Shannon’s confusion and diffusion 28
2.2. Encrypting a block at a time with AES 35
2.3. Encrypting more or less than a block with a block cipher 37
2.3.1. Very small messages need some padding 37
2.3.2. Larger messages require block chaining 38
2.4. Some concluding observations for block ciphers 43
Chapter 3. Asymmetric [Public-Key] Cryptosystems 47
3.1. Symmetric, asymmetric, and salty cryptosystems: basics 48
3.2. Using the RSA asymmetric cryptosystem in Python 51
3.2.1. Straightforward – not completely secure! – RSA in Python 51
3.2.2. More secure RSA in Python using OAEP and PKCS #1 58
3.2.3. How to use RSA in a way that is both fast and secure 61
3.3. Digital Signatures 66
3.3.1. Naive digital signatures 66
3.3.2. Better digital signatures using hash functions 69
3.4. Key management and the need for a robust PKI 74
3.5. Conclusions; consider the blockchain 77
v
vi CONTENTS
Bibliography 79
Index 81
Preface
Many children want to keep secrets.
Lovers on opposite sides of the globe want to whisper sweet nothings to each other over
an international telephone system whose wires and fiber optic cables run under neighbor-
hoods, city streets, fields, and forests, or over mountain passes and on deep sea beds.
Continent-spanning empires need to store secrets and issue commands to distant gov-
ernment agents and agencies which cannot be modified in transit from the imperial center
to its far-flung agents.
Consumers want to buy goods from merchants in other time zones, giving their credit
card numbers to those merchants in messages over the open Internet – which is, without
some system for hiding the contents of those messages, about as safe as writing them on
the sides of long-haul trucks as they go by on the highway.
Everyone needs a little cryptology.
The problem with crypto (as we shall call cryptology in this book) is that it has a
reputation of being very hard and mysterious, as well as very easy to get wrong. While
there are aspects of crypto that are connected to quite modern and complex theories – such
as number theory, an old and deep branch of mathematics; complexity theory, a new(er) and
subtle branch of computer science; and even quantum computation, a quite new wrinkle
on a 100 year-old version of physics which is famously counter-intuitive – that are not
particularly friendly to the novice, much of the over-all framing of crypto is perfectly easy
to comprehend and to use.
We contend, further, that this straightforward comprehension of the important basics
of cryptology is most easily acquired by actually working with cryptographic primitives,
by doing actual coding projects to implement, or use others’ implementations of, basic
cryptographic ideas.
That is the subject of this book.
This version uses Python and some standard cryptographic libraries in Python to ex-
plore these cryptological ideas. It should be accessible to students with a solid basic com-
fort level with Python – but could also be used as a way to solidify Python knowledge
in more beginning users of that language, particularly if those beginners had a friendly
instructor or peers with whom to collaborate.
vii
CHAPTER 1
Preliminaries
Here are some Greek roots:
kryptos, κρυπτ oς: secret, hidden
logos, λóγoς: word, study, speech
graph, γράϕω: write, written
analysis, αν άλυη: analysis
From these (and others), modern (technical) English gets the words
cryptosystem: a set of algorithms for protecting secrets, for hiding information so
that only the intended persons have access to it
cryptography: work done to make cryptosystems
cryptanalysis: work done to circumvent the protections of cryptosystems – to “crack”
a cryptosystem
cryptology: the combination of cryptography and cryptanalysis, often abbreviated
simply to crypto.
Beware that cryptography is widely (but inappropriately!) used as a synecdoche for
cryptology. We will use these words more carefully, except when the differences between
the words is not so important, in which case we will use the word crypto as short for
whichever of cryptology, cryptography, or cryptanalysis is most appropriate1
1
Note this use of “crypto” conflicts a little with a silly modern usage of that word as short for cryptocur-
rency. We will return to that topic much later, when we talk about blockchains.
1
2 1. PRELIMINARIES
1.1. Some speculative history
Perhaps there was a form of deception that preceded language – certainly many a house
pet has feigned innocence despite the clear evidence of involvement in stealing treats. And
even apiologists may not know if some lazy bees make up a story about a long excursion
to a new flower patch when their Queen demands an accounting.
But among homo sapiens, probably as soon as there was language, there was lying.
Of course, when two humans are face to face, both parties have some control, such as:
the listener can make an attempt to evaluate the trustworthiness of speaker, they can both
form their own judgments of the other’s identity and therefore choose what they wish to
share with each other, and the words of the speaker pass directly from their lips to the
listener’s ears without the possibility of change of meaning in flight (absent considerations
of ambient noise and so on).
A great deal changed with the invention of writing more than 5000 years ago. Words
frozen in physical form, and the ideas they represent, can be taken and shared with a wide
range of parties other than those with whom the original author wanted to communicate.
In addition, if an author is not able to hand her work directly to the intended reader and
instead the written words are out in the world on their own for a while, then both intended
communicants can no longer be sure that the other is who the writing claims them to be nor
that the writing remains the unchanged symbols that the other party originally set down.
Let us formalize some of these issues of information security (as it is called now), in
the context of a message to be sent from someone named Alice to someone named Bob.
The role of the possibly disruptive and overly intrusive environment is played in our little
drama by Eve. (Traditionally one skips directly to a character whose name starts with E to
symbolize both the environment and also someone who is potentially an eavesdropper2.)
Note that another way to think of what Eve is doing is listening to what is going by on some
unfortunately (maybe unintentionally) public communications channel that Alice and Bob
are using.
Here is some basic terminology:
D EFINITION 1.1.1. Confidentiality means that only the intended recipient can extract
the content of the message – Alice wants only Bob to get her message and not Eve, no
matter if she listens on the eavesdrop or intercepts the message as it travels in some form
from Alice to Bob.
D EFINITION 1.1.2. Message integrity means that the recipient can be sure the message
was not altered – Bob wants to know that what he gets is what Alice wrote, not what the
mischievous Eve intended to change it into.
2
Eavesdropper apparently comes from the Old English yfesdrype, meaning literally one who stands on
the eavesdrop [ground where water drips from the eaves of the roof] to listen to conversations inside a house.
1.1. SOME SPECULATIVE HISTORY 3
D EFINITION 1.1.3. Sender authentication means that the recipient can determine from
the message the identity of the sender – Bob wants to be sure this message did in fact
originate with Alice.
D EFINITION 1.1.4. Sender non-repudiation means that the sender should not be able
to deny sending that message – Bob wants to be able to hold Alice to her promises.
Note that Alice and Bob may actually be the same person sending a message from a
past self to their future self. For example, someone may want to keep records which must
be confidential and whose integrity must be reliable, even if there is a break-in to the site
keeping those records.
In fact, information storage can always be thought of, in this way, as an attempted
communication from the past to the future via a channel that may unexpectedly become
public if, for example, your laptop is stolen or someone breaks into your business and
steals the hard drive from your server. Unless you can be absolutely sure that in all possible
futures, no bad actor will ever possibly have physical access to your computer, it’s always
a good idea to encrypt all of your data on your personal devices!
Before we go on, a few more technical terms:
D EFINITION 1.1.5. The message that Alice wishes to send to Bob, in its actual original
form is called the plaintext.
D EFINITION 1.1.6. An algorithm that Alice uses to transform (scramble, obfuscate) the
plaintext into a form that will remain confidential even if observed by Eve while in flight is
called a cipher. We also say that Alice encrypts the (plaintext) message.
D EFINITION 1.1.7. The encrypted form of a plaintext message is called the ciphertext.
D EFINITION 1.1.8. When Bob receives the ciphertext, he applies another algorithm to
decrypt it and recover the plaintext.
Basic crypto terminology, graphically:
Alice on public network/channel Bob
(where Eve is watching)
plaintext/cleartext
message m
encrypts m to c
transmits c ciphertext c receives c
decrypts c to
recover plaintext m
4 1. PRELIMINARIES
Hundreds of years of experience with cryptology have shown, again and again, that
a cryptosystem which one person dreams up, thinking they have invented a system with
wonderfully strong security, may easily fall to cryptanalysis done by someone else. So it
has gradually come to be understood that the best way to make sure one is using a really
solid cryptosystem is to share the algorithms widely and publicly, so that others in the
community can try to break them. As is said in the FLOSS3 community: with enough eyes,
all bugs are shallow. Therefore, we should only trust a cryptosystem if its security remains
strong even after everyone has tried their best to break it!
If the algorithm is public, how does it provide any security at all? The answer is that the
algorithm may be public, but it is designed to take as input one more piece of information,
beyond just the plaintext to encrypt or the ciphertext to decrypt:
D EFINITION 1.1.9. Additional information (some of which is) used in encryption and
(some of) which is necessary for successful decryption is called a key.
Think of a key as something that Alice uses to lock her message in a box, the box being
the message as it is transmitted over a public network and even possibly seen by Eve, and
then used by Bob to unlock the box and find out what the actual message was that Alice
wanted to send. Clearly, the keys must be kept carefully secret for the cryptosystem to do
its job of protecting Alice and Bob’s communications from the eavesdropping Eve!
This idea – that the security of a cryptosystem should be based on the secrecy of the
key but not of the algorithm – has come to be called Kerckhoff’s Principle, after a set of
cryptosystem design ideas written down in 1883 by a French military cryptographer. The
opposite of a system made with Kerckhoff’s Principle in mind is one whose algorithms are
simply kept secret, and in which it is very foolish to place any trust: such weak systems are
said to reply upon the ill-advised security through obscurity paradigm.
Reading Response 1.1.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting?
Reading Response 1.1.2
Can you think of an example not mentioned above where the terms like confiden-
tiality, integrity, authentication, and/or non-repudiation could be used to describe
something you’ve read about or experienced? Or else, can you think of a situation
where these kinds of information security concepts apply to some version of Alice,
3“FLOSS”=”Free/Libre/Open-Source Software, ” a better term than just “open source”
1.1. SOME SPECULATIVE HISTORY 5
Bob, Eve, and a public channel? – if so, say what that situation was and what
corresponded to each of those characters.
6 1. PRELIMINARIES
1.2. The Caesar cipher and its variants
A cryptosystem which dates to ancient times was supposedly used by Julius Caesar.
D EFINITION 1.2.1. Alice takes her message, removes all spaces and punctuation, and
puts it all in one case (maybe lower case). Then she moves each letter k places down
the alphabet, wrapping around from z to a if necessary, where k is a fixed whole number
known to both Alice and Bob but no one else, called the key. (Often the ciphertext is
written entirely in upper case, so it is easier to tell plaintext from ciphertext.)
To decrypt, Bob simply moves each letter k places earlier in the alphabet, wrapping
past A to Z if necessary: in other words, Bob encrypts the ciphertext with key −k to get
the plaintext.
This is called the Caesar cryptosystem or Caesar cipher.
Apparently Julius Caesar usually used the key value k = 3. His nephew Octavian, who
later became the emperor Augustus, liked to use k = −1.
When using what is called the Latin alphabet (which is not what was used in ancient
Rome, though!), with its 26 letters, there is one particularly nice key value: 13. The nice
thing about that value is encryption and decryption are exactly the same transformation. In
modern times, this transformation is called ROT13, and it has a small role in the modern
history of the Internet. In particular, posts on early chat rooms and bulletin boards would
sometimes want to have a bit of content that should not be automatically available to anyone
who looks at the post, but would be there for the determined reader (such as, for example,
a spoiler in a review of some popular new game, book, or film). These were often included
in the post, but only after they had been run through ROT13.
A few commercial products used ROT13 for actual security, despite it actually being
completely insecure, such as certain parts of some versions of the Windows operating sys-
tem.
Reading Response 1.2.1
Have you heard of the Caesar cipher before? Is it clear to you how to use it?
1.2.1. Keys only matter “mod 26”. Notice that if Alice uses the Caesar cipher and
key kAlice = 1, her encryption does exactly the same thing as Ann who was using Caesar
with key kAnn = 27, because of that part of the definition of the Caesar cryptosystem which
says “wrapping around from z to a”: Ann will always go all the way around the alphabet
once, but then end up at the same place Alice will. The reason for this is that 27 = 26 + 1,
and encrypting with 27 goes around the alphabet once – from the 26 – and then goes one
more step. The same thing would happen if Ann used the key kAnn = 53 = 2 ∗ 26 + 1, in
1.2. THE CAESAR CIPHER AND ITS VARIANTS 7
that Ann’s encryption process would go around the alphabet – twice, in this case, from the
2 ∗ 26 – and then end up at the same place as Alice’s.
In fact, if difference between Alice’s and Ann’s key is any number of 26s, then their
encryptions will look exactly the same! Mathematicians have a bit of terminology and
notation for this:
D EFINITION 1.2.2. Suppose a, b, and n are three whole numbers. Then we say that a is
equal to b modulo n if a − b is a multiple of n; that is, if there is some other whole number
k such that a − b = k ∗ n. The notation for that is
a∼
= b (mod n)
E XAMPLE 1.2.3. As we just noticed, 1 ∼
= 27 (mod 26).
E XAMPLE 1.2.4. ROT13 was so convenient because doing it twice has the effect of
doing nothing, because
13 + 13 ∼
= 0 (mod 26)
Or, thought of another way, since
−13 ∼
= 13 (mod 26)
ROT3 encryption (Caesar with key 13) is exactly the same thing as ROT13 decryption
(Caesar with key −13).
Reading Response 1.2.2
Does all of this make sense? What was new or interesting, or what was old and un-
interesting?
If you feel comfortable with this new terminology and notation, try to answer the fol-
lowing question: If you have a whole number x and you know that x ∼ = 0 (mod 2),
what would be a more common way of talking about that number? I.e., you would
say that “x is .”
Computer languages usually have a way to work with numbers mod n. For example,
Python has a built-in arithmetic operation which is useful here: a%n gives the smallest
non-negative whole number which is equal to a mod n .
E XAMPLE 1.2.5. From what we noticed above, in Python
>>> 27%26
1
and
>>> -13%26
13
8 1. PRELIMINARIES
Even more, we can use this operator to implement the Caesar cipher!
First, notice that we can’t just add a key to a letter, because the types don’t work:
>>> ’a’+2
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
TypeError: must be str, not int
So we must first convert a character to an int and then back to a character:
>>> chr(ord(’a’)+3)
’c’
Note that ord() takes a single character and converts it to an int , while chr()
takes an int and converts it to a single character. It’s not so important what are the details
of how that works, except to note that the encoding includes all of the lowercase letters,
in alphabetical order, in one numerical block from the value ord(’a’) to the value
ord(’z’) , and all of the uppercase letters, in alphabetical order, in another numerical
block from ord(’A’) to ord(’Z’) .
Actually, that’s not going to work with a letter too close to the end of the alphabet, e.g.,
>>> chr(ord(’y’)+3)
’|’
because it didn’t “wrap around the alphabet.”
So how about if we take a character in a variable x and do the following:
(1) make it into a number using ord()
(2) shift it so that it’s in the range from 0 to 25 by subtracting ord(’a’)
(3) do the Caesar by adding the value k
(4) get it back in the range 0 to 25 by doing %26
(5) get it in the range ord(’A’) to ord(’Z’) by adding ord(’A’)
(6) make it back into a letter using chr()
E.g., if x=’y’ and k=4 then we would get
>>> chr((ord(x)-ord(’a’)+k)%26+ord(’A’))
’C’
Let’s now implement the Caesar cipher in Python, in steps. First, we should clean up
the plaintext string.
Code Task 1.2.1
Write a Python function which takes an input string and returns a string which is the
same except all the non-letter characters have been removed and all the letters have
been made to lower case.
It might help to know that you can loop through the characters of a string s by doing
for x in s:
1.2. THE CAESAR CIPHER AND ITS VARIANTS 9
and you can build up a string o for output by starting with
o=’’
and then adding one or more characters in a string variable t to the end of o by doing
o=o+t
Another, more “Pythonic” approach is probably to build up a list by the construction
[ c for c in s ]
which just makes a list each of whose elements is a single character out of the string s .
Note you can also do something to the characters as you put them in the list, and you can
add a condition to the looping part, such as
[ ord(c) for c in s if c.isalpha()]
– try it!
Code Task 1.2.2
Caesar encryption: write a Python function Caesar encrypt(s, k) which
takes an input string s – the plaintext – and an int k and returns the ciphertext from
that plaintext under the Caesar cipher with key k .
For example, if you do
Caesar encrypt(’All the animals at the zoo!’, 3)
the output should be
’DOOWKHDQLPDOVDWWKHCRR’ .
Code Task 1.2.3
Caesar decryption: Write a Python function Caesar decrypt(s, k) which
takes an input string s – the ciphertext – and an int k and returns the plaintext
from that ciphertext under the Caesar cipher with key k .
For example, if you do
Caesar decrypt(’DOOWKHDQLPDOVDWWKHCRR’, 3)
the output should be
’alltheanimalsatthezoo’ .
You might want to do a sanity test on your input and return an error or a null string
if the input is not a well-formed ciphertext, i.e., a string consisting only of uppercase
letters.
1.2.2. Modernizing the Caesar cipher. It seems unfortunate that the Caesar cipher,
as we have used it, looses the punctuation in the messages and case of the letters of the
message. One way to get around this would be to imagine all of the letters of both cases
and all punctuation symbols to be part of a much bigger alphabet, and then to use the same
10 1. PRELIMINARIES
old Caesar with this much bigger alphabet, correspondingly using %N , where N is the
new alphabet size, in place of the %26 we were using before.
Technically, in Python 3, strings are made of characters, and characters are encoded
using Unicode. Unicode is a way of turning numbers, represented by one, two, or four
bytes, into many special symbols as well as letters from many writing systems used around
the world.
The first 128 symbols in Unicode, which therefore fix into one byte (in fact with the
highest-order bit being 0), are just the old ASCII encoding scheme, dating back to 1963,
for the English alphabet plus some additional punctuation, symbols, and even non-printing
symbols like “EOF.” The name of the encoding scheme for the subset of Unicode which
fits in 8 bits is “UTF-8,” which includes all of the old ASCII code.
Full Unicode in two or four bytes includes many non-Latin alphabets (such as Greek,
Cyrillic, hangul, etc.) and even non-alphabetic writing systems like Chinese and Japanese
(actually, part of Japanese is non-alphabetic, called “kanji,” but Japanese also has two dif-
ferent writing systems based on syllable sounds, called “hiragana” and “katakana,”, each
with 49 symbols), with thousands of ideograms.
Reading Response 1.2.3
Have you heard of Unicode?
If not, you might also read, or at least skim, the Wikipedia page about Unicode and
say what you think of it.
Have you ever worked with Unicode in Python?
If so, what did you think of it? If not, you might also read, or at least
skim either the Python documentation Unicode HOWTO or the more detailed
Unicode & Character Encodings in Python: A Painless Guide and say what you
think about it.
Bonus Task 1.2.1
Implement in Python a Caesar-style encryption and decryption which works for a
different alphabet than English. Maybe do Spanish (one extra letter: ñ) or German
(four extra letters: ä, ö, ü, ß)....
Or, maybe implement Caesar with the English alphabet, but figure out some way
to keep the punctuation as well. It seems like it might be giving away too much
information to leave the spaces and punctuation the same in the ciphertext as it was
in the plaintext, but can you do something interesting with it?
Or, maybe implement a Caesar-style cryptosystem which handles all of ASCII, or
UTF-8, or all of Unicode!
1.3. CRYPTANALYSIS OF THE CAESAR CIPHER 11
1.3. Cryptanalysis of the Caesar cipher
Let’s think about cryptanalysis of the Caesar cipher; that is, let’s figure out how to get
a plaintext back from a Caesar ciphertext when we don’t know the key.
The first thing we should notice is that, as we saw in §1.2.1, the key only matters “up to
mod 26.” But every possible whole number k for a key is equal mod 26 to a whole number
in the range from 0 to 25: just try running
for i in range(100):
print(i%26)
to get a feel for what’s happening.
So, effectively, there are only 26 possible keys for a Caesar cipher! In fact, it would be
pretty silly to use the key k = 0, since then the ciphertext would be exactly the same as the
plaintext (well, without the non-letter characters, and all in upper case). So there are really
only 25 reasonable Caesar cipher keys. It seems like this kind of thing is important enough
to have its own name:
D EFINITION 1.3.1. The set of possible keys for some cryptosystem is called the keyspace
of the cryptosystem.
What we have just seen is that the keyspace of the Caesar cipher is just the whole
numbers from 1 to 25. In Python, it is the elements of the list range(1,25) .
An enemy of Caesar’s who intercepted a ciphertext of his and who wanted to know
what it said could assemble a team of 25 literate people (maybe those were hard to find at
the time?) who could count (also hard?) and use them to make a Caesar-cracking computer
as follows:
• each person would be assigned one of the numbers from 1 to 25
• each person would write out the Caesar-decryption of the intercepted ciphertext,
using their assigned number as key
• if one of them got something which looked like it was written in good Latin, they
would shout “Success!” and read out their decryption
This would be a pretty fast and economical way to crack the Caesar cryptosystem only
because the keyspace is so small. This kind of attack has a name:
D EFINITION 1.3.2. An attempt to cryptanalyze a cryptosystem by trying all of the
possible values of some parameter of the system (such as all possible keys or all possible
messages) is called a brute-force attack.
12 1. PRELIMINARIES
Reading Response 1.3.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting?
Can you think of an example of a brute-force attack used in some information
security context that you know about from other classes or other readings or just
from your life?
Code Task 1.3.1
Write a Python function CaesarBF(s) implementing a brute-force attack on the
Caesar cryptosystem. Given an input ciphertext, it should try all possible keys for
decryption and print out the attempted key and the corresponding decryption.
E.g., the output of CaesarBF(’DOOWKHDQLPDOVDWWKHCRR’) should be
1: cnnvjgcpkocnucvvjgbqq
2: bmmuifbojnbmtbuuifapp
3: alltheanimalsatthezoo
4: zkksgdzmhlzkrzssgdynn
5: yjjrfcylgkyjqyrrfcxmm
6: xiiqebxkfjxipxqqebwll
7: whhpdawjeiwhowppdavkk
8: vggoczvidhvgnvooczujj
9: uffnbyuhcgufmunnbytii
10: teemaxtgbfteltmmaxshh
11: sddlzwsfaesdksllzwrgg
12: rcckyvrezdrcjrkkyvqff
13: qbbjxuqdycqbiqjjxupee
14: paaiwtpcxbpahpiiwtodd
15: ozzhvsobwaozgohhvsncc
16: nyygurnavznyfnggurmbb
17: mxxftqmzuymxemfftqlaa
18: lwwesplytxlwdleespkzz
19: kvvdrokxswkvckddrojyy
20: juucqnjwrvjubjccqnixx
21: ittbpmivquitaibbpmhww
22: hssaolhupthszhaaolgvv
23: grrznkgtosgrygzznkfuu
24: fqqymjfsnrfqxfyymjett
25: eppxliermqepwexxlidss
1.3. CRYPTANALYSIS OF THE CAESAR CIPHER 13
That’s all well and good, but we have a huge problem: human intervention is required
here, to pick out which possible decryption, in the brute-force attack, is the correct one!
For example, suppose bad actors invaded your organization’s network and encrypted
all of your files with the Caesar cipher, but choosing a different key for each one. Your
organization probably has hundreds of thousands of files, and while any individual one
could be easily decrypted, it would take an insane number of person-hours to decrypt your
whole network ... you might be willing to pay a fairly large ransom to get all of your files
back without all that human life invested in brute-force decryption.
1.3.1. Frequency analysis. Let’s think about how to automate the choice of the correct
decryption in a brute-force attack on the Caesar cipher.
The key idea here is just the observational fact that the relative frequencies of different
letters in English are relatively stable across almost all pieces of English text. In particular,
in English, the most common letter is usually “e,” the second most common is usually “t,”
etc.
This suggests an approach to detecting automatically which decryption in the brute-
force attack is the correct one: declare victory when the brute-force approach has found a
possible decryption where the most common letter is “e.”
Code Task 1.3.2
Write a Python function freq table(s) which makes the frequency table for
the letters of a string s . That is, first clean up s by removing all non-letters and
making all remaining letters be in the lower case. Then freq table() should
returns a list of 26 int s where the value at location i in this list tells how many
times the letter chr(i+ord(’a’)) occurred in the cleaned-up input string.
For example, executing
freq table(’alltheanimalsinthezoo’)
should return the list
[3,0,0,0,2,0,0,2,2,0,0,3,1,2,2,0,0,0,1,2,0,0,0,0,0,1] .
Code Task 1.3.3
Write a Python function CaesarBF findE(s) uses your previous
CaesarBF(s) but only prints out the possible decryption if the letter “e”
is the most frequent in the proposed cleartext. That is, each time your func-
tion computes a possible cleartext, you should get the frequency table freqs
from that string, using your freq table() function. Then do the test
14 1. PRELIMINARIES
freqs[4]==max(freqs) to see if the letter at index 4, which is “e,” has
the maximum frequency. Only if the test passes will you print out that possible
cleartext.
Clearly, this strategy doesn’t seem to work too well, because it would find that the string
’teemaxtgbfteltmmaxshh’ from key 10 and ’eppxliermqepwexxlidss’
from key 25 were both possible decryptions of ’DOOWKHDQLPDOVDWWKHCRR’ , nei-
ther of which is correct!
The problem is that the single letter which is most frequent is too coarse a piece of
information about a string to be useful to determine whether the string is in good English.
However, the entire frequency table does a much better job. Let’s implement that in Python
now, in several steps.
Code Task 1.3.4
Modify your Python function freq table(s) to make one called
rel freq table(s) which computes the relative frequency table for an
input string s : a table which computes the fraction of the letters in s which are
each particular letter. If the list computed by freq table() is f , then the
relative version will have float s rather than int s, with values that are the
values of f divided by sum(f) (or, what should be the same thing, divided by
len of the input string).
For example, executing rel freq table(’zbazb’) should return the list
[0.2,0.4,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,
0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.0,0.4] .
Code Task 1.3.5
Run your function rel freq table() on some very long pieces of English text,
to get the model relative frequency table of the English language.
You might compare the values you get from what Wikipedia says should be the right
table of relative frequencies in the article Letter Frequency. Did you get a similar
answer? If not, do you have a guess of why?
Code Task 1.3.6
You will need, in a moment, a tool which computes the “distance” between two
relative frequency tables. Those tables are lists of numbers of length 26. Well, if
you have a list of two numbers, you can think of those numbers as the coordinates
1.3. CRYPTANALYSIS OF THE CAESAR CIPHER 15
in two-dimensional space. If you have a list of three numbers, you can think of them
as coordinates in three-dimensional space. So a list of 26 numbers could be thought
of as the coordinates in 26-dimensional space. Weird, yes, but go with it....
In two- or three-dimensions, the distance between two points is given by the
square root of the sum of the squares of the differences between the coordi-
nates. That is, in two dimensions, the distance between two points with coordi-
nates (x1 , y1) and (x2 , y2 ) is, by the Pythagorean Theorem, given by the formula
p
(x1 − x2 )2 + (y1 − y2 )2 . The same thing also works in three ... or in 26 dimen-
sions: that’s just the Pythagorean Theorem 25 times!
So write a Python function which takes as input two lists of numbers of the same
length, say they are a and b . It should then take the first element of a minus the
first element of b and square that difference. It should add to that square the square
of the difference of the second elements, and so on until all the elements are used
up. Finally, it should return the square root of that total, which will be the value of
freq table dist(a,b) .
Code Task 1.3.7
Now implement a Python function CaesarBF rel freq(s) tries all possible
Caesar decryptions of the input ciphertext string (s) , and picks the one whose
relative frequency table is least different from the standard relative frequency table
of English which you just found.
Here are the steps: do the CaesarBF(s) approach from before, and compute the
relative frequency table for each possible decryption. Now compute the “frequency
table distance,” in the sense of the previous coding task, between the relative fre-
quency table for this possible decryption and the standard relative frequency table
for English.
When you’re done, find the smallest distance of all those possible decryptions and
announce that one as the best guess of the real decryption of the input ciphertext.
Bonus Task 1.3.1
Now let’s stress-test the brute force Caesar cipher cracker which uses frequency
analysis.
First, try taking a big piece of text in another language, encrypt it with
the Caesar cipher and some random key, and then use your function
16 1. PRELIMINARIES
CaesarBF rel freq(s) to try to decrypt it automatically. For exam-
ple, you might take some of the Spanish text of the famous novel Don Quixote
or simply cut-and-paste from the French newspaper Le Monde or
the German newspaper Deutsche Tageszeitung. Note: for other languages,
you may have an issue with the alphabet! Letters with accents may mess up your
constructions in Python which assume the alphabet will be sequentially ordered
from 0 to 25 when you take each letter c and do ord(c)-ord(’a’) . Figure
out how to handle this situation (probably by turning each special letter like into a
letter in the English alphabet (like just make “ñ” into a plain “n”).
Or you could try decrypting a text which was originally in Latin, like
the original Latin text of Isaac Newton’s famous Philosophiae Naturalis Principia
Mathematica.
Another thing to do would be to try some very unusual text, but in English. For
example, a man named Ernest Wright published a book called Gadsby in 1939 which
did not have the letter “e” anywhere in it. Since “e” is the most common letter in
normal English, the relative letter frequencies in this novel must be quite different
from those in normal English. Try encrypting and then automatically decrypting
some of the text of Gadsby (but make sure you go down far enough in that file to get
to the text of the novel: the introductory material does have the letter “e”).
1.4. DEFENDING CAESAR AGAINST FREQUENCY ANALYSIS: VIGENÈRE AND THE ONE-TIME PAD 17
1.4. Defending Caesar against frequency analysis: Vigenère and the one-time pad
Sometimes a very short message is very important: “yes” or “no” is a big difference in
many situations! But frequency analysis attacks on the Caesar cipher depend upon having a
relative frequency table to compare to the relative frequency table of English ... and without
a long enough enough ciphertext string, there won’t be much of a frequency table at all.
There must be some sort of threshold k such that when a ciphertext string has length
k or less, the relative frequency table doesn’t give enough information to do successful
cryptanalysis. There should also be another threshold K such that when a ciphertext string
has length of K or more, the cryptanalysis of the Caesar cipher using relative frequency
tables usually works.
For messages of length between k and K, presumably, sometimes the approach to crypt-
analysis based on relative frequency tables is successful, and sometimes not.
Bonus Task 1.4.1
Let’s try to figure out what those thresholds k and K are.
Do a big loop as some number l goes from 1 to 150 (or so). For each value of
l , you are going to figure out how often your relative frequency table-based Caesar
cracker CaesarBF rel freq is successful on strings of length l .
Make a very long string big string with some text. You probably want to take
all of the non-letters out of big string and make all of the letters lower case.
Now make a loop to try decrypting maybe 100 randomly chosen strings from
big string . Each time through the loop, pick a random substring from
big string of length l – the way to do this is to do
from random import *
once near the beginning of your program, then in this loop set
rand start = randint(0,len(big string)-l)
to be the random starting point for the substring, and
rand cleartext = big string[rand start:rand start+l]
to be the corresponding substring of length l .
Now pick a random Caesar key k=randint(1,26) , encrypt
rand cleartext to rand ciphertext with that key, and try your
relative frequency table-based Caesar cracker on rand ciphertext . If its
proposed decryption equals rand cleartext , count that as a success. If not,
count it as a failure.
See how many times out of the 100 (i.e., the percentage) in this inner loop you got
success.
Finishing the outer loop, you should now have a list of success percentages for each
possible string length from 1 to 150, maybe in a list successes .
18 1. PRELIMINARIES
Plot that list against the numbers from 1 to 150. This can be done by first get-
ting the plotting package with import matplotlib.pyplot as plt ,
then making the plot with plt.plot(x,successes) where
x=[i+1 for i in range(149)] . You will need also to run
plt.show() to display the plot.
The way to prevent relative frequency table-based attacks on the Caesar cipher is to
only use it to encrypt short messages (less than that threshold k we just mentioned). This
is a problem if it is important to encrypt a longer message.
The approach to use with longer messages is simply to change the Caesar cipher key
frequently, so that not all that much cleartext is encrypted with each particular key value.
For example, to make the string half as long, so that relative frequency table methods are
only half as good, we could use two keys a and b, and encrypt the first half of the message
with Caesar under the key a and the second half with Caesar under key b.
Historically, this approach was implemented in a slightly different way: rather than
using a for the first half of the message and b for the second half, people used a for the
letters in the first, third, fifth, etc., positions in the message, and b for the letter in the
second, fourth, sixth, etc. positions.
To cut the message pieces down even more, one can use even more keys than just two.
There is a name for this cryptosystem:
D EFINITION 1.4.1. Fix ℓ numbers k1 , k2 , . . . , kℓ . The Vigenère cryptosystem does
encryption as follows: given a cleartext c, encrypt the first letter of c using the Caesar
cipher with key k1 , the second letter of c using Caesar with key k2 , etc. When all of the ℓ
key values have bee n used, start again with k1 for the next letter, then k2 , etc. Vigenère
decryption works in the same way, except Caesar decryption is used at each step.
For example, the Vigenère encryption of “All the animals in the zoo” with three keys
3, 20, 4 is “DFPWBIDHMPUPVCRWBICIS,” since
• 3 characters after “a” in the alphabet is “D,”
• 20 characters after “l” in the alphabet, wrapping around to the beginning of the
alphabet when falling off the end, is “F,”
• 4 characters after “l” is “P,”
• 3 characters after “t” is “W,”
• 20 characters after “h” in the alphabet, wrapping around to the beginning of the
alphabet when falling off the end, is “B,”
• etc.
Notice that we have achieved the goal of messing up frequency analysis approaches
to breaking our cryptosystem: for example, the repeated letter “l” in the cleartext word
1.4. DEFENDING CAESAR AGAINST FREQUENCY ANALYSIS: VIGENÈRE AND THE ONE-TIME PAD 19
“all” has been encrypted into the two different letters “F” and “P” – but the “l” later in the
cleartext, in the word “animal,” has also been encrypted to “P” ... the frequency tables have
been totally scrambled! In fact, we can see that by looking at the relative frequency charts
of the cleartext
and for the ciphertext from Vigenère encryption with keys 3, 20, 4
which looks completely different! Contrast this with the relative frequency chart for a
Caesar encryption (with key 10) of the same cleartext
20 1. PRELIMINARIES
which looks like the relative frequency chart of the original cleartext, shifted horizontally.
Reading Response 1.4.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel like you understand the Vigenère cryptosystem?
Reading Response 1.4.2
To show you understand Vigenère, can you say what the keyspace is for a the Vi-
genère cryptosystem when it uses ℓ key values? And how big is that keyspace? (How
big it is will give us some information about how hard it is to do a brute-force attack
on this cryptosystem.)
Code Task 1.4.1
Write a Python function Vigenere encrypt(s, k) which takes a cleartext
s and list k of key values and does Vigenère encryption.
Likewise implement a Python function Vigenere decrypt(s, k) which
does the decryption for ciphertext s and list k of key values.
Using the Vigenère cipher with ℓ keys basically divides the cleartext into ℓ different
pieces and uses the Caesar cipher on each of these different pieces, each time with a dif-
ferent key. As long as these pieces are sufficiently long, we can use our automated Caesar
cipher cracker based on relative frequency counts on each piece separately, figure out what
the separate Caesar keys are, and then do Vigenère decryption with those ℓ keys we’ve
figured out.
1.4. DEFENDING CAESAR AGAINST FREQUENCY ANALYSIS: VIGENÈRE AND THE ONE-TIME PAD 21
Notice that the ℓ pieces of the cleartext into which the Vigenère cryptosystem are all
the same size, that being len(s) /ℓ if the cleartext is s . Since an attack on the Vigenère
cryptosystem is based on relative frequency-based attacks on the Caesar ciphers of those
pieces, and such relative frequency-based attacks only work if there is enough data to make
a good relative frequency table, if len(s) /ℓ is a small number, then the attack will fail.
In particular, if ℓ = len(s) then those ℓ pieces will have only one character in
them and relative frequency-based attacks will fail miserably. In other words, if you use
a Vigenère cryptosystem with ℓ keys on a cleartext of length ℓ, then it is perfectly secure!
This is so important that this cryptosystem is worth giving its own name:
D EFINITION 1.4.2. A cryptosystem which is Vigenère where there are as many keys
as letters in the cleartext is called a one-time pad cryptosystem, and the pad in this cryp-
tosystem is the name used for the collection of all of those Vigenère keys.
It is a very important thing when using a one-time pad never to reuse the pad – that’s
why it’s called a one-time pad – because if Eve guesses that the pad has been reused, she
can crack all of the messages which were sent with that pad. We will not discuss here how
that attack on encryption with a reused one-time pad works, but suffice it to say that the
attack is based, again, on frequency analysis.
One final note about one-time pads: typically, when using a one-time pad, the keys
are picked purely randomly. The advantage of doing that is that there is no way Eve can
predict what the keys in the pad might be, even if she knows Alice and Bob very well.
With a randomly chosen one-time pad, this cryptosystem is called information theoretically
secure, meaning that is secure no matter how much computation power Eve can throw at
the problem.
Reading Response 1.4.3
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel like you understand how one-time pads work?
Code Task 1.4.2
Write a Python function OTP keygen(n) which generates a one-time pad of size
n . As we said above, it’s best if these pads are as random as possible, so you might
want to do
from random import *
and then use the command randint(a,b) which returns a random integer in the
range a to b (including both endpoints, which is unusual in Python!).
As an example of using your new OTP keygen(n) , send your instructor an email
with a one-time pad of at least 100 elements.
22 1. PRELIMINARIES
Code Task 1.4.3
Write a Python function OTP encrypt(s,k) which takes a cleartext s and
one-time pad k and does the encryption.
Also write a Python function OTP decrypt(s,k) which takes a ciphertext s
and one-time pad k and does the decryption.
As an example of using your new OTP encrypt(s,k) , send your instructor an
email with your first Reading Response from the course, but encrypted using that
one-time pad you sent for Coding Task 1.4.2. (If that Reading Response was longer
than 100 characters, you will have to truncate it to the first 100 characters so that you
have enough one-time pad to do the encryption.)
Bonus Task 1.4.2
Since the quality of the randomness in a one-time pad can actually matter if you plan
on encrypting very sensitive data, you might consider using a more sophisticated ran-
dom number generator than the basic randint(a,b) from the random pack-
age ... which is fine for many uses, but maybe not good enough if you are trying to
keep a large document secret from a powerful adversary.
Python has a function urandom in the package os which generates much
stronger randomness. Unfortunately, it returns just a string of random bytes, which
you need to convert to int s by using in the Python function unpack in the pack-
age struct . Here is a web page which shows how to use these two functions.
Write a Python function OTP keygenCS(n) which generates a one-time pad of
size n with cryptographically secure randomness.
Code Task 1.4.4
Write a Python function OTP key finder(s, t) which takes a string s con-
sisting only of lower case letters and string t consisting only of upper case letters,
where len(s) is the same as len(t) and computes what must have been the
one-time pad which was used to get t as the ciphertext from s using the one-time
pad cryptosystem.
The fact that you can do this means that when Eve sees a ciphertext from a one-time
pad cryptosystem, she cannot deduce anything about the original cleartext other than
its length. This is why this cryptosystem is information theoretically secure.
Hint: If you’re looking at a letter x of cleartext and a letter y of ciphertext and
wondering which number k in a one-time pad key could have encrypted x to y ,
you’re really asking the question “what value of k makes
1.4. DEFENDING CAESAR AGAINST FREQUENCY ANALYSIS: VIGENÈRE AND THE ONE-TIME PAD 23
chr((ord(x)-ord(’a’)+k)%26+ord(’A’))==y
be true?” Solving this for k would give one answer as
k=ord(y)-ord(’A’)-ord(x)+ord(’a’) .
Bonus Task 1.4.3
Suppose we somehow know that the following string is the Vigenère encryption of
some English cleartext with a Vigenère cryptosystem having ℓ = 2 keys:
WUZHEGJUFPRQJFFXEWIBDHEOVQUPVBFXIHRUJLYDMHTRDHKRSXIBTDVV
RUERKWFSIDZVVKZPKKVHMLCWYDKPVQURCLMHJDWWVUKKVPKKVJFRULJR
WWZQKHIUWGNLKKKKVLIEFQVVJRCHKLKEVZZWYFRHJDIWYHERSOVEIXKX
JKRWYWFOUBFXTDVVRUNDJDDEZWZRLVZIZWNHIHJRZWNDJDXUZHMRLVWD
LOKDEGXUZHMRLVCBYDKKTDVVRURQJZVUVGZWYHIHLQUHIOVDMHFISULW
LVRQUWYHIHJWWRIEIXKXJLJDEKFQFURECHDDEVFDIHKKVBROCDCOYRER
IDSOVPVQTRDHZWFVGHRNZQTDVVRUJILQVUROYHNDJPPIILVQUIRLKKWX
CDEGAXJWKRDHSXKEIXKXJVRBJKVZRVRPSLKLFXJDEGSULWLVZVRQYRER
IDSOVPRQYHYDKKSUFXXKKPRQPFRSKLMHJKFPVWFUFPVZYRJHIDEVFPJG
ZGKKVJVQVUROTRWIVUJIZOCGZGKKZVZQTDVVRUJHVPRPSLKLFXJZYHEW
YDKWYHGRFUYDMHTUZHUFRHJDIKRWYZVSKDDEZWZREVYRLOUEVPRGVRWV
KHIQVUJWLIWBVWSULWLVJDPVYHNDJDDEZWZRLVRQUEIXKXJLJDEKFQFU
RECHDDEBFXROCGZGJHVWYDKREWYHCXGHIFROZWYUZFVSIHJHEWVGYLDD
BLEJCBTUFZEZYLTKYHULUWYUZFVUVILVVZRVKKZVRPSLKLFQPHKEIXKX
JVRBJKVZRVRPSLKLFXJDEGJXIHYHZVRQYRERIDSOVPRQZVGHRNERKWFG
ZVGUFYVZYDKEIXKXJVGRBHSXKKVUVLRPKRJSVDBZYDKLURBQFZPRLDCO
ULUOFYVKZPFQTHERKZZWYRLWTDLVVZYDKFRXJHNLKKYRCGJBFXKKVQKR
DRLUEIFUYLDRAXUJDHEWKKFXRUKICHUWFEIXKLJKSHRVKVRQUPVQYDMH
CRJWKKVLIUVDJREEVDIZZWYPVPPKVDIWZVZQKKVFFIWLEWYHIHNLKKTD
VVRURQULDXJWGDLVVWZOCLKFFPVERFBWFPV
Can you figure out what the two Vigenère keys must have been, and what the original
cleartext was?
24 1. PRELIMINARIES
1.5. Preliminary conclusion of preliminaries
We have seen that a very old cryptosystem, the Caesar cipher – which even has lived
into the modern age in the form of ROT13 – helps illuminate some basic ideas of the
design of cryptosystems and some first steps in attacking them: basics of cryptography and
cryptanalysis.
The attack on the Caesar cipher which we discussed was based on frequency analysis.
There’s actually some interesting history here: As was mentioned above, the Caesar cipher
was used by Julius and his nephew in the first century BCE. The approach of frequency
analysis was first described nearly a thousand years later, in a book Manuscript on Deci-
phering Cryptographic Messages by the Muslim philosopher, mathematician, physician,
and general polymath Al-Kindi. Like the mathematician and astronomer Al-Khwarizmi
(whose name came down to modern times in the word algorithm and whose book The
Compendious Book on Calculation by Completion and Balancing gave us the word alge-
bra, from it’s title in Arabic), Al-Kindi worked at one of the greatest centers of scholarship
in the world in the 8th century CE, the House of Wisdom in Bagdad under the Abbasid
Caliphate. Al-Kindi and Al-Khwarizmi seem to have both been involved in bringing an
Indian system for writing numbers using positional notation (so: 101 means one 100 plus
no tens plus one 1, whereas the Romans would have written that number CI) to the Arab
world. This notation later spread to Christian Europe in the early 13th century through a
book called Liber Abaci (title in Latin; which would be The Book of Calculation in Eng-
lish) written by a man Leonardo of Pisa ... whom we know today as Fibonacci. Some have
claimed that Fibonacci’s notation helped bring about the Renaissance in Europe, since with-
out good numerical notation, it would have been very hard to do double-entry bookkeeping,
which was used by great trading houses like the Medici, who then had the funds to support
the tremendous flowering of the arts which was the Renaissance....
The Vigenère cipher was invented in the mid-15th century (not by Vigenère!), used
for hundreds of years by European governments and even by the Confederacy during the
American Civil War, and was described as recently as the early 20th century as unbreakable
(by none other than Charles Dodgson, the mathematician known to us today mostly as
Lewis Caroll, the author of Alice in Wonderland!).
Reading Response 1.5.1
Had you heard of any of this history before? Did you know of Al-Kwarizmi and his
connection to the words algorithm and algebra? Had you heard of the House of Wis-
dom in Bagdad? Or what about the theory that the Renaissance was made possible
by the invention of double-entry bookkeeping (and therefore, indirectly, by the im-
portation into Christian Europe of Hindu-Arabic numbers which made calculations
so much easier than other systems like the Roman numerals)?
1.5. PRELIMINARY CONCLUSION OF PRELIMINARIES 25
An enormously important thing we have learned in this chapter, however, is that:
There is a perfectly secure cryptosystem, the one-time pad.
The next time you are watching a movie or TV show which has some line like “There’s no
such thing as an unbreakable code,” you should therefore laugh derisively.
Code Task 1.5.1
Before leaving this topic, we should implement a full, one-time pad cryptosystem
which will work for any kind of data in any language.
The key to doing this is to think of any data, in any language (or not in a language
at all), as being simply a string of bits. Bits, of course, are like letters from an
alphabet which has only two symbols 0 and 1. We know from §1.2.1 that Caesar –
and, therefore Vigenère and one-time pads – only care about keys up to mod of the
size of the alphabet, which is mod 2 for bits. So shifting the “letters” 0 and 1 by a
key of 0 would do
+0
0 1 7→ 0 1
and shifting by a key of 1 would do
+1
0 1 7→ 1 0
Therefore, if the cleartext bit is a and the key bit is b, then the encrypted bit is a ∧ b
(this is the “exclusive or”, XOR, operation).
Write a Python function OTP gen keyfile(f, n) which takes a string f and
uses it as a filename into which to deposit n random bytes.
Write a Python function OTP fileencrypt(f,k) which a string f and uses
it as a name of a file of cleartext data to be encrypted. The encryption should use the
one-time pad kept in the file with name k and the output should be put into the file
with name f+’.enc’ .
Some hints:
• You can open a file for reading as bytes with the command
file handle = open(filename,’rb’) .
• You can get the entire contents of such a file into a byte string with the
command data=file handle.read() .
• If a and b are int s, then a∧b is the bitwise XOR of a and b .
• If bs is a byte string, then bs[i] , (where i is in
range(len(bs)) ) is an int .
• If you open a file with the flag ’wb’ , then you can write a byte string bs
to that file with file handle.write(bs) .
However, the enormous problem with one-time pads is that:
26 1. PRELIMINARIES
To remain secure, a one-time pad must have as many key digits as
there are letters in the plaintext to be encrypted, and reuse of pads
completely destroys the security of this cryptosystem.
What this means is that the distribution of very large keys is supremely important in the use
of this perfectly secure cryptosystem.
There’s a nice story of how important the management of one-time pads can be from
a wonderful book called Between Silk and Cyanide: A Codebreaker’s War [Mar99] on
cryptology during (part of) World War II by the author Leo Marks. Marks ran spies into
Nazi-occupied Holland during the war, and when he took over, the spies were given cyanide
pills to take so that they could commit suicide if ever captured. That way they would
not give away the cryptosystems they used to radio back to England all the secrets they
uncovered. Instead, under Marks’s direction, they were given large one-time pads, written
on sheets of silk thin enough to be worn underneath their clothes when they were first
sent to Holland. That way, if captured later, they could tell everything they new about the
codes they used and without having the specific one-time pad used by other spies or the one
they intended to use later, the Nazis could get no information from intercepted ciphertexts.
Which all meant that captured spies had no particular reason to commit suicide, and instead
could cooperate and perhaps live out the rest of the war in a POW camp, if captured.
This story of silk undergarments with cryptosystem keys written on them is the first
time we see what will become a recurring theme in this subject: key distribution, man-
agement, and security can often be the most important practical consideration in applied
cryptography.
CHAPTER 2
Block Ciphers
Most of the encryption today on the Internet is probably done with one of the cryp-
tosystems called block ciphers, which do not provide as perfect cryptographic security
as a one-time pad, but which have a much, much more modest level of difficulty in key
management. Block ciphers are also very fast and have some nice security features that
cryptosystems structured like the ones we’ve seen so far do no.
Examples of block ciphers you may have encountered might include:
• Advanced Encryption Standard [AES], authorized by the National Institute of
Standards and Technology [NIST] for use by the US federal government in un-
classified situations;
• the Data Encryption Standard [DES], which was the NIST-authorized block ci-
pher from 1977 to 2001; and
• 3DES or triple DES, a block cipher pretty much consisting of doing DES three
times, which was used for a while after it was known that DES was nearly broken
but before AES was standardized.
In this chapter, we will very briefly touch upon some of the most basic design principles
for block ciphers, do a little Python implementation to explore these principles, and use
some code libraries to do real encryption with state-of-the art block ciphers.
27
28 2. BLOCK CIPHERS
2.1. Why encrypt blocks of data: Shannon’s confusion and diffusion
Our approach to the cryptanalysis of the Caesar and Vigenère ciphers was based on the
frequencies that individual letters appear in the English language and also in a ciphertext
we are trying to crack. For some cryptosystems, similar approaches might be used for
longer strings of characters:
D EFINITION 2.1.1. A single letter (from some alphabet) in a piece of text such as a
particular plaintext or ciphertext, or in a large collection of all English texts, is called a
monograph.
Similarly, a pair of adjacent letters in the same kind of text source is called a digraph,
and a sequence of three letters is called a trigraph.
So one way to describe our attacks on Caesar and Vigenère is to say that they com-
pared the statistics of monographs in (various substrings of) a ciphertext with the same
statistics for the English language, which turns out to be a fairly stable distribution of letter
(monograph) frequencies.
Similarly, the frequencies of digraphs in English are fairly stable – e.g., th is quite
common, as is qu, while both tx and ql are very common. Thus the statistics of digraphs
(and, eventually, trigraphs) can be useful in attacking some cryptosystems.
Bonus Task 2.1.1
Ignoring non-alphabetic characters and the difference between upper and lower case,
there are 676 = 262 possible digraphs in English (26 choices for the first letter of the
digraph and 26 choices for the second letter).
Write a Python function digraph freq(f, n) which makes a relative fre-
quency table of the digraphs from the text in the file with name f , printing out the
most frequent n digraphs and their relative frequencies. You will want to read in
the whole file but, as we have done before, discard all the non-alphabetic characters
and make the remaining letters all lower case, before counting the digraphs.
Note also that the digraphs can overlap: that is, the string
humperdink
has these digraphs:
hu, um, mp, pe, er, rd, di, in, and nk
and not just the non-overlapping ones hu, mp, er, di, and nk.
Printing out the entire relative frequency table is easier than just the n most fre-
quent digraphs. To do that, you will want to sort the relative frequency values, and
then loop through them in the order of decreasing frequency. For each value of a
frequency, you will then want to find all digraphs with that frequency and print it
out, increasing a counter as you do. When the counter gets to n , you will stop.
2.1. WHY ENCRYPT BLOCKS OF DATA: SHANNON’S CONFUSION AND DIFFUSION 29
The reason our attack on Caesar using monographs was successful is that Caesar doesn’t
hid the statistics of its cleartext letters very well: they’re all in the ciphertext, just shifted
over in the alphabet. In fact, so long as the way we scramble the letters in a cleartext to get
the letters of the corresponding ciphertext is the same for every letter – maybe not simply
shifting, like
Plaintext : a b c d ··· ··· y z
Ciphertext : A B C D ··· ··· Y Z
but even other schemes like
Plaintext : a b c d ··· ··· y z
Ciphertext : A B C D ··· ··· Y Z
which switches the even-numbered letters in the alphabet with the odd-numbered ones –
the statistics of monographs in our ciphertext will be the same as in the cleartext, only in
some different order. There’s a name for cryptosystems that do this:
D EFINITION 2.1.2. A cryptosystem where a single scrambling transformation of letters
in the cleartext to the letters in the ciphertext is called a monoalphabetic cryptosystem.
A cryptosystem where, instead, the scrambling transformation of letters in the cleartext
to the letters in the ciphertext changes from letter to letter of the cleartext – so that, for
example, sometimes a in the cleartext is encrypted to K in the ciphertext, but other times in
the same cleartext, a might be encrypted to B or M or some other letter – depending upon
perhaps the other letters in the message or the location of the cleartext letter in the entire
cleartext, is called a polyalphabetic cryptosystem.
Reading Response 2.1.1
Does all of this make sense? What was new or interesting, or what was old and unin-
teresting? Do you feel like you understand which cryptosystems are monoalphabetic
and which are polyalphabetic – if so, which of the ones we’ve discussed so far are of
which type?
30 2. BLOCK CIPHERS
So (monograph) frequency analysis will make good progress in cryptanalysis of monoal-
phabetic cryptosystems, but the relative frequency tables for ciphertexts from polyalpha-
betic cryptosystems are smeared out by adding together lots of differently scrambled ver-
sions of the frequency table and so cryptanalysis this way will be much harder.
In the end, what we are seeing is that when the ciphertexts produced by some cryp-
tosystem have statistics that somehow reflect the statistics of the cleartext, then there is a
possible avenue of attack. The best thing, then, would be if the cleartexts and ciphertexts
seemed to be statistically unrelated to each other
The great mathematician, computer scientist, and cryptologist Claude Shannon1 in his
highly influential (and, initially classified by the US Government) paper A Mathematical
Theory of Cryptography[Sha45]
D EFINITION 2.1.3. We say that a cryptosystem has the property of diffusion if, for a
fixed choice of key, every time any single bit in a cleartext is flipped, half of the bits of the
ciphertext should change (in general, statistically), and vice-versa. (Here, “statistically”
means that this doesn’t have to be precisely true every time, but when we average over
many cleartexts and keys, on average half of the ciphertext bits will change every time we
change any single cleartext bit.)
The point of diffusion is to minimize the information about the cleartext which is leaked
by he ciphertext. The very fact that there is some communication between Alice and Bob
is, of course, leaked. Looking at the flows of messages between individuals in some group
– who speaks to whom and when or, in other words, the metadata of the communication
– is called traffic analysis and can be very important. Seeing which individual is the
central point through which all messages flow, and related issues from traffic analysis, can
give information about who is the leader of the group, etc.. As former head of the US
National Security Agency General Michael Hayden said[Col14], “We kill people based on
metadata,” referring to US drone strikes on individuals though to be terrorist leaders based
on traffic analysis.
If if we are not trying to understand the organization of some group of communicating
individuals on the basis of their pattern of communications, and are just looking at Alice
and Bob’s exchange of encrypted messages, some things are hard to conceal. For example,
it is nearly impossible to send a very large piece of information (cleartext) through an
encrypted channel unless the corresponding ciphertext is quite large.
Although, to be fair, the opposite, that a large ciphertext means the cleartext had a lot
of information in it, is not necessarily true. For example, Alice could pad a short message
with a very large amount of random extra data and send to Bob the encrypted version of
1Shannon was an enormously influential figure in the history of computer science – for example, he single-handedly invented the
field now known as Information Theory. He was also a very colorful character: see his Wikipedia page.
2.1. WHY ENCRYPT BLOCKS OF DATA: SHANNON’S CONFUSION AND DIFFUSION 31
that long cleartext. That would be a large ciphertext which did not correspond to a large
amount of (useful) information in the cleartext.
Reading Response 2.1.2
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel like you understand how an observer could use traffic
analysis to guess who was the leader of a group of communicating individuals? Can
you think of another situations where something like traffic analysis comes up in
regular life – maybe when friends or romantic partners suddenly are communicating
with different people more or less frequently...?
Since it is very hard to prevent the leakage of at least the fact that communication has
occurred, as well as, probably, some idea of how much data has be transmitted, a good
cryptosystem will try to prevent as much other leakage as possible: Shannon’s diffusion is
one way to do that.
Code Task 2.1.1
On our way to seeing if the cryptosystems we’ve met so far have diffusion:
Write a Python function str diff display(s,t) which takes two string s
and t and shows how much they differ.
Your function should first check if the strings have the same length, printing an error
and returning 0 if not.
Next, your function should loop through the strings and print a “*” if they differ at
each location and a “ ” if they are the same. (You might want to print these symbols
in rows of some determined length, like 30 characters. You can do this by print-
ing with commands like print(’*’,end=’’) and print(’ ’,end=’’)
to stay on the same line, and then using print(’’) to get to the next line.)
Your program should then return the number of places where the two strings differ
(i.e., the number of “*”s printed).
For example, running
str diff display("abc"*30,("abc"*15)+"Abc"+("abc"*13)+"aBc")
should produce the output
*
*
2
32 2. BLOCK CIPHERS
Code Task 2.1.2
Actually, displaying where two strings have different characters, as
str diff display does, is a pretty crude approximation of showing where
those strings have different bits. So let’s improve the string difference display and
calculation code to make a bit difference display and calculation program.
In principle, the easiest way to do this is to convert a string of characters into a
string of characters displaying all of the bits in the first string, and then use the
str diff display you built before.
To get a character into its binary expression, you can convert it first to an int with
ord , and then convert that int into a string of binary digits with bin . The
problem with bin is that it always puts a 0b at the front of its binary strings, and
it also drops off leading 0s. (That makes sense: when writing the number 17 in base
10, we don’t write it as “017,” to indicate there are no 100s, 1 ten, and 7 ones ... we
just write “17;” i.e., we drop leading 0s. The same happens in base two.)
Of course, you can skip the first two characters (like that “ 0b ”) of a string s with
s[2:] , and if you have a string b of 0s and 1s that you want to pad on the left
with the right number of 0s to make it eight symbols long (to show all the bits in that
byte), you can do (’0’*(8-len(b)))+b .
Use the above ingredients to make a Python function
bits diff display(s,t) which takes two string s and t and shows
how much their bits differ. Since there will be eight times as may bits to show as
there were just for str diff display(s,t) , you might want to make your
lines a bit longer, and maybe show if the bits are the same with a “.” and if they
differ with an “x.”
For example, running
bits diff display("abcdef", "CbcdeF") should produce the output
..+...+.......................
............+.....
3
(Even though running
str diff display("abcdef", "CbcdeF") would produce the output
* *
2
The difference is because while the strings "abcdef" and "CbcdeF" differ
only in two characters, the first and the last, the bits for "a" are 01100001
though the bits for "C" are 01000011 – so there are two different bits in this one
different character – and the bits of "f" are 01100110 though the bits for "F"
are 01000110 – so there is only one different bit in this one different character!)
2.1. WHY ENCRYPT BLOCKS OF DATA: SHANNON’S CONFUSION AND DIFFUSION 33
Code Task 2.1.3
Which of the cryptosystems that we have studied so far – Caesar, Vigenère, and
one-time pads – satisfies Shannon’s idea of diffusion?
Take a fairly long – maybe 100 characters or so – cleartext from some source, and
make two versions of it which differ by only one bit.
[Note: think a bit about how to make that one-bit change. Changing one letter may
not do it! For example, the letter a is represented in Unicode by the number 97,
which is has binary expression 01100001. On the other hand, b in Unicode is 98,
which is 01100010 in binary – so changing a to b changes two bits!]
Once you’ve got those two strings, make appropriate keys and encrypt both strings,
then run your bits diff display on the encrypted versions to see where and
how much they differ. Try this for each of the cryptosystems we’ve done in this
class.
You should now be able to answer the question if those cryptosystems satisfy Shan-
non’s diffusion.
Another thing Shannon worried about was if the ciphertexts leaked some information
about the key that was used in their encryption. In particular, he made a definition similar
to diffusion, but now looking at keys, as follows:
D EFINITION 2.1.4. We say that a cryptosystem has the property of confusion if, work-
ing with a fixed cleartext, every time any single bit in a key is flipped, half of the bits of
the ciphertext should change (in general, statistically), and vice-versa. (Here, ‘statistically”
means that this doesn’t have to be precisely true every time, but when we average over
many ciphertexts and keys, on average half of the ciphertext bits will change every time we
change any single bit of the key.)
Code Task 2.1.4
Which of the cryptosystems that we have studied so far – Caesar, Vigenère, and
one-time pads – satisfies Shannon’s idea of confusion?
Repeat the previous Code Task 2.1.3 with now a fixed cleartext but trying encryptions
with only one bit changed in the key, and see how much of the ciphertext changes.
The conclusion of this discussion of information leakage is that we should seek cryp-
tosystems that satisfy confusion and diffusion: they must smoosh up and mix up the bits
of the cleartext and the key, so that, in general, (nearly) every bit of the ciphertext depends
upon (nearly) every bit of both the cleartext and the key. This means that we must process
the whole cleartext at once, we cannot handle the bits (characters) of the cleartext one at a
time, as the cryptosystems we’ve seen so far have done.
34 2. BLOCK CIPHERS
This is not easy to do with a fixed cryptographic algorithm for a general cleartext, so
a common approach has been work with chunks of the cleartext at a time, called blocks,
where the blocks all have the same. At least on the level of blocks, we can hope that
our new cryptographic algorithm will satisfy confusion and diffusion and therefore reduce
information leakage.
This leads us to:
D EFINITION 2.1.5. We say that a cryptosystem is a block cipher if it only is defined
when it’s input cleartexts and ciphertexts are of some fixed size, called the block size and
if it is deterministic (i.e., it has not built-in randomness) so it’s result depends only upon
the input clear- or ciphertext and the key.
Reading Response 2.1.3
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Does it make intuitive sense to you that confusion and diffusion would
be desirable characteristics for a cryptosystem, and how block ciphers might relate
to that?
We mentioned at the very beginning of this chapter several well-known block ciphers.
Such cryptosystems – or at least the ones that have been widely used – tend to be based
on very primitive but very fast mixing operations on the bits in the messages and the key.
One particular strategy is to take pieces of the key and do various operations on it and
the message block, to make something called the round function. Then the full encryption
algorithm simply applies the round function some number of times, each time with a new
piece of key data, again and again to the slowly transforming message block.
2.2. ENCRYPTING A BLOCK AT A TIME WITH AES 35
2.2. Encrypting a block at a time with AES
Let’s get some practice with using a block cipher by looking at the case of the Advanced
Encryption Standard [AES]. The key for AES must be of length 16, 24, or 32 bytes [to say
it another way: 128, 192, or 256 bits] while its block size is 16 bytes [or 128 bits].
Standard Python distributions should come with the Crypto package, which in-
cludes AES among the set of implemented ciphers in Crypto.Cipher . Therefore
before you can use AES, you must do
from Crypto.Cipher import AES
You can then create an object to work with this cipher by
cipher=AES.new(key, AES.MODE_ECB)
The key must be one of AES’s recognized key sizes, or else this attempted object creation
will throw an error (and we’ll discuss what that AES.MODE ECB is doing later...).
An AES object created this way can be used for encryption or decryption (or both) of
a block of 16 bytes of cleartext or ciphertext, always with the key you used to create the
object. For instance, if your key was "asdfghijklmnotuv" then here is an example
of encryption
>>> cipher.encrypt("16 bytes of fun!")
b’˜\x94\xaa\xf6\xc20\x85\xce\x83\x91\xc9\}\xdf\x96\xb8\xcf’
and here is an example of decryption (assuming the key is still "asdfghijklmnotuv" )
>>> cipher.decrypt(
... b’˜\x94\xaa\xf6\xc20\x85\xce\x83\x91\xc9}\xdf\x96\xb8\xcf’)
b’16 bytes of fun!’
Code Task 2.2.1
If AES is going to be a useful standard cipher for wide use on the Internet, it should
run very fast. Let’s test how fast this by timing how long it takes to encrypt one
block. ...Actually, to be safe – e.g., in case some other process takes some time
on your computer while you think it is only doing an encryption, it is best to do
something like 1000 encryptions, measuring the total time, and then dividing that by
1000.
A quite direct way to do measure how long something takes is first to run
import time
You can then save into a variable the number of seconds since the so-called “Unix
Epoch,” (which was 00:00:00 UTC on 1 January 1970) by
start time=time.time()
If you then do some processing, you can calculate how long it took with
elapsed time=time.time()-start time .
36 2. BLOCK CIPHERS
When using this approach to measuring computational speed, you should be careful
only to do the repeated runs of the commands you want to time between the two
invocations of time.time() – run nothing superfluous! For example, make one
AES cipher object before you start the timing, so that the computational cost of
building that object doesn’t get counted when you are trying to figure out the cost of
encryption (or decryption).
With all that in mind, make some 16 byte AES key and 16 byte cleartext, time how
long it takes to do 1000 encryptions and 1000 decryptions, and report on how long
it takes, on average, for each encryption or decryption. It would be nice to change
the message and/or the key each of the 1000 times – but call some random number
generator in the loop, then you’ll be timing that as much as encryption/decryption!
Code Task 2.2.2
Does AES satisfy Shannon’s idea of diffusion?
To see, take some cleartext of length 16 bytes and make two versions of it which
differ by only one bit.
Then make a 16 byte key, create an instance of AES, and make ciphertexts for the two
different cleartexts (but the same key). Use your program bits diff display
to see where and how much the two ciphertexts differ.
You might also try the above for several different keys, just to make sure that it’s not
a fluke which is special for the choice of key you happened to make.
Code Task 2.2.3
Does AES satisfy Shannon’s idea of confusion?
To see, take some cleartext of length 16 bytes. Make two different keys of length 16
bytes which differ by only one bit. First with one key, make an AES cipher object and
encrypt the fixed cleartext, then do the same using the other key. Use your program
bits diff display to see where and how much the two resulting ciphertexts
differ.
You might also try the above for several different ciphertexts, just to make sure that
it’s not a fluke which is special for the choice of key you happened to make.
Reading Response 2.2.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Was it easy or hard to use Python’s AES? What do you find most
annoying and/or convenient about this implementation of AES?
2.3. ENCRYPTING MORE OR LESS THAN A BLOCK WITH A BLOCK CIPHER 37
2.3. Encrypting more or less than a block with a block cipher
If Alice and Bob know they will always be exchanging messages which are exactly the
number of bytes in the block size of the block cipher they are using, they’re in very good
shape. What is much more likely, though, is that they will send smaller or (much) larger
messages, and we ought to help them do that securely.
2.3.1. Very small messages need some padding. When a message is smaller than the
block size, it seems pretty easy simply to add some extra stuff to fill up an entire block: this
extra data is called padding. The problem is, if the message can could be arbitrary data,
then how do we tell what is message and what is padding?
A common way to do this, when we know that there will have to be some padding
(maybe metadata that goes along with the message, like in the subject line of an email or
in the packet metadata for an encrypted packet sent on a packet-switched network), is to
assume that there is always at least one byte of padding: the last one. That byte can carry
information, though, so it can simply tell the number of bytes of padding or, conversely,
the number of actual non-padding bytes in the block.
Code Task 2.3.1
Write a Python function pad string(s, n) which takes as input a string s
and block size n and returns a string of length n , using padding as described above.
If len(s)>n-1 , you should print an error and return the empty string, since there
will be no room for padding. Otherwise, return a string whose first characters are
just a copy of s and whose remaining bytes are just some copies of the number
len(s) .
Also write a function unpad string(s) which takes as input a string s , as-
sumed to come from the pad string() function, and returns the original string
without padding.
For unpad string , you can assume the block size used in the pad string
was simply len(s) and the last byte of s is a number telling how much of s
is the original string. Therefore, if ord(s[-1])>len(s)-1 then there’s some
sort of problem and you should print out an error message and return the null string.
Otherwise, return the first ord(s[-1]) character substring of s .
E.g., pad string(’test’,16) should return
’test\x04\x04\x04\x04\x04\x04\x04\x04\x04\x04\x04\x04’
and unpad string(’fishhead\x08\x08\x08\x08\x08\x08\x08\x08’)
should return
’fishhead’ .
38 2. BLOCK CIPHERS
Code Task 2.3.2
Write a Python function smallAESencrypt(k, s) which takes as input an
AES key k and cleartext string s , where len(s)<16 and returns the AES en-
cryption of the padded version of s .
If len(s)>15 or if k is not a valid AES key, you should print an error and return
the empty string.
Also write a function smallAESdecrypt(k, t) which takes as in-
put an AES key k and AES ciphertext t , assumed to come from
smallAESencrypt(k, s) function, and returns the original cleartext string
s (without padding).
Reading Response 2.3.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Can you think of a different (or better) padding scheme?
2.3.2. Larger messages require block chaining. What if a cleartext is longer than the
block size of the block cipher we are using? The obvious thing would be simply to break
the cleartext into some number of full blocks and then one last block which is smaller, and
to use regular block encryption on the full blocks and padding-plus-encryption on the last,
smaller block.
Code Task 2.3.3
Implement that scheme for the AES block cipher. That is:
Write a Python function naiveAESencrypt(k, s) which takes as input an
AES key k and cleartext string s . It should then build up a return value as follows:
(1) for each full block of cleartext (there may be none such, if the
len(s)<16 ), append to the return value the regular AES encryption of
that block with key k
(2) append to the return value the output of
smallAESencrypt(k, small) , where small is the last piece of
the cleartext, whose length is less than 16 bytes.
Also write a Python function naiveAESdecrypt(k, t) which takes as input
an AES key k and ciphertext string t . It should then build up a return value as
follows:
2.3. ENCRYPTING MORE OR LESS THAN A BLOCK WITH A BLOCK CIPHER 39
(1) if t has more than 16 bytes, for each block of 16 bytes in t except the
last, append the plain AES decryption of that block with key k to the return
value
(2) when done with that, append smallAESdecrypt(k, last) to the
return value, where the variable last contains the last 16 bytes of t
But now we’ve got to wonder if this naive approach to AES on large messages still
satisfies Shannon’s diffusion condition.
Code Task 2.3.4
Re-do Code Task 2.2.2 but now using two cleartexts of length 80 bytes which differ
by one bit near the middle, and using the naiveAESencrypt function.
How are we going to get a block cipher that satisfies Shannon’s diffusion condition? A
good block cipher will diffuse nicely ... within a block. But the block size is fixed, and we
have to somehow spread the diffusion – which, remember, basically means that changing
on bit in the cleartext changes around half of the bits of the ciphertext – around in each
block. One way might be just to design a new block cipher for each new cleartext whose
block size is larger than the size of the cleartext and which does nice diffusion across that
block. But it’s so hard to verify that ciphers are really secure, this is completely impractical.
Instead, the most common approach is to use a block cipher of some fixed size (like
AES) and to make the encryption of each block depend also on what happened with previ-
ous blocks: this is called block chaining.
There are two things we need to think about for block chaining.
(1) How exactly to chain the blocks: There are various ways to chain the encryption
of blocks and achieve diffusion. Here’s one: Suppose we change one bit in a block
of cleartext. If we’re using a well-designed block cipher, about half of the bits of
the ciphertext will change as well. We can use these different bits to change the
bits of the next block of plaintext, by bitwise XORing (i.e., bitwise addition mod
2) the ciphertext block with that next block of cleartext. That will propagate the
change to half of the bits of the corresponding ciphertext, and repeating for each
successive block, we will have spread the changed bits all down the sequences of
blocks of the ciphertext coming from the long cleartext.
(2) How to start the chaining: The very first block will not have a predecessor block
off of which to chain. It turns out that this can cause security issues, so the way
to handle this is to make a block’s worth of random data, called the initialization
vector (IV), and to start chaining as if that IV were the ciphertext from the pre-
vious block. Alice should then send the random IV that she chose along with the
ciphertext so that Bob can decrypt the long message.
40 2. BLOCK CIPHERS
D EFINITION 2.3.1. The block chaining approach just described is called Cipher Block
Chaining or CBC mode block chaining. Symbolically, it works like this:
for encryption and like this:
for decryption. (In diagrams like this, the symbol “ ⊕ ” stands for bitwise XOR (i.e..)
Note that the way we discussed before of doing encryption of long messages with a
block cipher, which simply treated each block entirely separately and failed Shannon’s dif-
fusion condition is, in the context of block chaining schemes, called Electronic Codebook
mode or ECB mode chaining and has the simple diagram:
2.3. ENCRYPTING MORE OR LESS THAN A BLOCK WITH A BLOCK CIPHER 41
for encryption and
for decryption.
The way a block chaining mode is chosen in Python is in the creation of a Crypto.Cipher
object. Where, in the past, we have used AES.MODE ECB for Electronic Code Book
chaining, we can use instead AES.MODE CBC for cipher block chaining as described
above. Since that mode requires an initialization vector, Alice must specify the 16-byte IV
they want to use, as follows:
from Crypto.Cipher import AES
key=b’randobytes 4 key’ # not a good choice of key
iv=b’IV of randobytes’ # terrible choice of IV
cipher=AES.new(key,AES.MODE_CBC,iv)
ciphertext=cipher.encrypt(b’some cleartext w/ len two blocks’)
message=iv+ciphertext # put together IV and ciphertext
# to be transmitted to Bob
42 2. BLOCK CIPHERS
The original message can then be recovered with
from Crypto.Cipher import AES
key=b’randobytes 4 key’ # Bob must know the same key
iv=message[:16] # Bob grabs the IV out of the message
ciphertext=message[16:] # the rest of the message is ciphertext
cipher=AES.new(key,AES.MODE_CBC,iv)
cleartext=cipher.decrypt(ciphertext)
Note that the Crypto.Cipher object is stateful in CBC mode – meaning that it re-
members what it has seen before. When it is created, it uses the given IV to start the chain-
ing, but thereafter, every call to encrypt chains off of the previous block. That is, the
way of encrypting the two-block cleartext b’some cleartext w/ len two blocks’
described above yields exactly the same result as doing instead
from Crypto.Cipher import AES
key=b’randobytes 4 key’ # not a good choice of key
iv=b’IV of randobytes’ # terrible choice of IV
cipher=AES.new(key,AES.MODE_CBC,iv)
ciphertext_part1=cipher.encrypt(b’some cleartext w’)
ciphertext_part2=cipher.encrypt(b’/ len two blocks’)
ciphertext_full=ciphertext1+ciphertext2
message=iv+ciphertext # put together IV and ciphertext
# to be transmitted to Bob
Reading Response 2.3.2
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Can you think of a different (or better) approach to block chaining?
Do you see why chaining is good, and why an initialization vector is a good idea?
Finally, we can see to what extent block chaining makes a version of AES which satis-
fies Shannon’s diffusion condition ... to some extent.
Code Task 2.3.5
Re-do Code Task 2.3.4 but now using AES in CBC mode.
2.4. SOME CONCLUDING OBSERVATIONS FOR BLOCK CIPHERS 43
2.4. Some concluding observations for block ciphers
Modern block ciphers like AES can run quite fast (that should have been the conclu-
sion of Coding Task 2.2.1). Using some, but not all, block chaining modes, they can satisfy
Shannon’s diffusion condition, at least a sort of forward diffusion in that when two mes-
sages differ by one bit, then about half the bits in the corresponding blocks in the two
ciphertexts are different, and also for all blocks after (but not before!) that one.
This makes block ciphers quite good to use in practice... assuming they’re secure.
Fortunately, at the moment, there is no known, general-purpose attack against the AES
cryptosystem better than brute force.
Let’s think through some issues around a brute-force attack on AES. The key size we’ve
been using is 16 bytes, or 128 bits. If a random key is chosen, then 2127 keys will have to
be tried, on average, in a brute-force attack before the correct one is guessed. 21 27 is about
1.7 × 1038 , which means that if you could check a billion possible keys a second, then
it would take roughly 100 billion times the life of universe (which is around 4.3 × 1017
seconds) to do one brute-force attack on AES.
Code Task 2.4.1
We should at least try a brute-force attack against AES ... at least in a very special
case (we don’t want to wait billions of times the life of the universe!).
Suppose Eve thinks Alice has sent her password to Bob in a message which was
encrypted with the AES cipher using Electronic Code Book mode chaining. Eve
also knows that Alice and Bob’s favorite number is 17, so she thinks that they will
probably use a key which is mostly 17s. In fact, Eve guesses that the key will be 15
bytes all with the value 17 followed by one new, random byte.
Eve can try all possible keys of this form with code like this:
key_ints=[17]*16 # last 17 is probably not right,
# it’s just here to make a list of
# the correct length
for i in range(256):
key_ints[15]=i
possible_key=bytes(key_ints)
# make an AES object with that possible_key
# and use it to try decrypting the ciphertext
Eve also guesses that the word “password” will appear in the cleartext. Python
has a nice command to check if a substring is present in a string, the command
in , which works both with normal strings and substrings and also byte strings and
44 2. BLOCK CIPHERS
potential byte substrings. Eve can use this as follows:
# inside a loop making attempted decryptions which
# are in a variable possible_cleartext
if b’password’ in possible_cleartext:
print(’Found the cleartext! It should be: ’,end=’’)
print(possible_cleartext)
If Eve were to try the above approach to decrypt the stolen ciphertext
b"h9\xf2\x12,\x5e.\xab\xf4M\x1e\xf5\xc9\xcea\r"
what password would she find, and what would be the key that unlocked it? [Show
all of Eve’s code as well as answering those questions!]
Code Task 2.4.2
Repeat the previous Code Task 2.4.1, but suppose now that Eve thinks that only the
first 14 bytes of the key are 17s, the last two might be anything.
In this situation, try to use the same general approach (but probably with two nested
loops, to try all 256 possible bytes in both of the last two bytes of the key) to decrypt
the stolen ciphertext
b"\xe0\xb3W\x94/X\xe7.\x93\xda\xe9\xed\xde\xedT\xfe"
Bonus Task 2.4.1
Repeat the prior Code Task 2.4.1 one more time, but suppose now that Eve thinks
that 15 of the bytes of the key are 17s, while the last byte – call it the “wildcard key
byte” – might be anything ... and that wildcard byte might be at any location of the
key.
In this situation, try to use the same general approach (but probably with two nested
loops, one to loop through the 16 possible locations for the wildcard key byte, the
other then to try all 256 possible values for the wildcard byte) to decrypt the stolen
ciphertext
b"0\xbd\x9e*\x9b\xd4\x03\xa58<\xfa\x0e\x86J\xfd@"
In the Code Tasks 2.4.1 and 2.4.2 and the Bonus Task 2.4.1 above, we imagined that
Eve would somehow know that the cleartext behind the ciphertext she had stolen contained
the substring b’password’ . This kind of thing has a name:
D EFINITION 2.4.1. If a cryptanalyst is trying to discover the cleartext which generated
a particular ciphertext and knows (or at least hopes) some substring of the unknown full
cleartext, then that substring is called a crib.
2.4. SOME CONCLUDING OBSERVATIONS FOR BLOCK CIPHERS 45
Cribs are actually not all that uncommon. Cleartexts often have a standard header which
includes the date and time, which the cryptanalyst might know for other reasons, so the
correctly formatted time-and-date string can be used as a crib. Famously, the Nazi military
communications during WWII always included a well-known phrase signaling fealty to the
leader of the Third Reich2, which could be used as a crib.
Reading Response 2.4.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Can you think of a “crib” in messages you often send or receive,
maybe to or from a family member, or someone (or some organization) with which
you regularly communicate on the ’net for work, school, or entertainment?
Bonus Task 2.4.2
Here’s an entirely non-coding Bonus Task:
Cribs were very important in cracking the Enigma cryptosystem which was used by
the Nazis during WWII – which shortened the war by years, it has been estimated
(see “Alan Turing: The codebreaker who saved ’millions of lives’” from The BBC).
Write a one-page (or so) “mini-paper” on the Enigma, as a cryptosystem, its role in
history, and something about how it was cracked. This should include
• A short description of how the Enigma cryptosystem worked – use as many
of the technical terms from this course that you can (e.g,, what was the
keyspace, was it a mono- or polyalphabetic cryptosystem or neither, etc.);
this should be about half of your “mini-paper.”
• A very short description of the Enigma’s historical role in the war; about
one sixth of your “mini-paper.”
• A short, very high-level description of the effort made by the Allies to crack
the Enigma cryptosystem – mention cribs; about one third of your “mini-
paper.”
Sources you might use are:
• The BBC story mentioned above.
• The Wikipedia page “Enigma machine”
• The Wikipedia page “Cryptanalysis of the Enigma”
• The Crypto Museum’s page “History of the Enigma”
We have seen that block ciphers can be fast, easy to use, and highly secure.
2
Which we will not quote exactly here, because it would be unpleasant to have that odious phrase be a
substring of this book!
46 2. BLOCK CIPHERS
But we did see that the security is based in part on the size of, and lack of structure in,
the keys: it was that keyspace of size 21 28, along with a lack of some foolish structure like
“15 of the 16 bytes are just representations of the number 17,” which made AES so secure.
Which means that it is really important for Alice and Bob to be able to share – very
accurately, since even a single incorrect bit with scramble their messages catastrophically!
– a large, random-seeming key. In the modern, networked, world, where we often want
to share valuable information very securely with people whom we may have never met
in person, this is a serious defect. You may want to get your credit card number to an
online store from your laptop, first through the open WiFi at a coffee shop, then across the
(insecure by design) Internet! Cryptography is clearly a vital tool to do a thing like that,
but the cryptosystems we have seen so far all require a shared secret, the key, which is hard
to achieve in common modern situations like this.
In the next chapter we start to talk a solution to this problem of key sharing: cryptosys-
tems where [part of] the keys can be public,
CHAPTER 3
Asymmetric [Public-Key] Cryptosystems
We have seen that there are secure, fast, (relatively) easy-to-use cryptosystems, but
which do require a special activity before the regular communication can start: secure,
secret key sharing. There are various approaches to getting around the practical difficulty
this may present in the real world (such as, noted above, getting your credit card number
securely to an online vendor with whom you have not had the opportunity to perform the
secret key-sharing ritual), including:
• There are protocols which allow two parties to create a shared secret by exchang-
ing messages on a public network: Alice and Bob must merely exchange two
messages and they will end up with a common secret value that Eve, even if she
saw both of the messages, will not be able to compute. That shared value can then
be used, directly or after further processing, as the key for a block cipher for more
secure communication. The most famous of these protocols is Diffie-Hellman key
exchange, about which more information can be found, e.g., here:
◦ The Wikipedia page “Diffie-Hellman key exchange”
◦ a nice survey article [Hel02]
◦ the original paper about their discovery by Diffie and Hellman [DH76]
The details of Diffie-Hellman are based on some modest mathematics: it is quite
accessible to an advanced undergraduate mathematics major. It’s security rests on
a hypothesis which is widely believed – but not proven! – by many mathemati-
cians and computer scientists.
• Instead of basing a secure way for parties to build a shared secret by communi-
cations over a public channel on theoretical hypotheses, there is way to do the
same thing using some specialized hardware and a deep understanding of quan-
tum mechanics. Called quantum key distribution, [QKD] and invented in the
1980s, this approach has been demonstrated first in the laboratory over ever longer
distances since then, but is not yet in wide use in practice. For more information:
◦ The Wikipedia page “Quantum key distribution”
◦ a general introductory site: Quantum computing 101
◦ the original 1984 research paper on this subject, by Bennett and Brassard [BB20]
• Another approach, with wide and interesting application is the subject of most of
this chapter: asymmetric cryptosystems.
47
48 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
3.1. Symmetric, asymmetric, and salty cryptosystems: basics
Looking more closely, the problem we noticed at the end of the last chapter stems
from the fact that Alice and Bob both have to have the same key to encrypt and decrypt,
which they must therefore keep away from Eve. After all, if Eve had the key, she could
encrypt and send messages to Bob pretending that they came from Alice, and Eve could
also intercept and decrypt the real messages that Alice was sending to Bob. That’s the
problem with using the same key. Cryptosystems that do this have a name:
D EFINITION 3.1.1. A symmetric cryptosystem is one in which the same key is used
both to encrypt cleartexts to ciphertexts and also to decrypt ciphertexts to cleartexts.
Sometimes symmetric cryptosystems are also called private-key cryptosystems.
Graphically, symmetric cryptosystems work like this:
Symmetric [private-key] cryptosystem, graphically:
private agreement
Alice ←→ of shared key k ←→ Bob
on public network
(where Eve is watching)
creates cleartext message mA ;
using k, encrypts mA to cA
and transmits cA ciphertext cA receives cA ;
using k, decrypts cA
and recovers cleartext mA
creates cleartext message mB ;
using k, encrypts mB to cB
receives cB ; ciphertext cB and transmits cB
using k, decrypts cB
and recovers cleartext mB
etc....
The problem we’re dealing with now is the initial private agreement of the secret key k.
A way to get around this problem would be if Alice and Bob used different keys for
encryption and decryption. There’s a name for this type of cryptosystem:
D EFINITION 3.1.2. An asymmetric cryptosystem is one in which one key ke – called
the encryption or public key – is used to encrypt cleartexts to ciphertexts, while a dif-
ferent key kd – called the decryption or private key – is needed to decrypt ciphertexts to
cleartexts.
3.1. SYMMETRIC, ASYMMETRIC, AND SALTY CRYPTOSYSTEMS: BASICS 49
Sometimes asymmetric cryptosystems are also called public-key cryptosystems, be-
cause usually when one is using them, one posts one’s encryption key ke in a public place
to everyone to see and use.
Graphically, asymmetric cryptosystems work like this:
Asymmetric [public-key] cryptosystem, graphically:
Alice on public network Bob
(where Eve is watching)
generate key pair (keA , kdA ) generate key pair (keB , kdB )
download keB public encryption key keB publish keB
publish keA public encryption key keA download keA
create cleartext message mA ;
using keB , encrypt mA to cA
and transmit cA ciphertext cA receive cA ;
using kdB , decrypt cA
and recover cleartext mA
create cleartext message mB ;
using keA , encrypt mB to cB
receive cB ; ciphertext cB and transmit cB
using kdA , decrypt cB
and recover cleartext mB
etc....
Reading Response 3.1.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you understand how Alice and Bob would use an asymmetric
cryptosystem?
The most basic part of designing secure asymmetric cryptosystems is that it must be
infeasible for Eve to compute the decryption key kd even when she knows the [public]
encryption key ke . After all, the encryption key is public, so if it were feasible to do this
calculation, Eve would simply do that and be able to decrypt all ciphertexts herself.
Another consideration for the security of these systems is that the space of possible
decryption keys must be very large. Otherwise, Eve could simply do the brute-force attack
of trying all of the possible decryption keys.
Actually, for asymmetric cryptosystems there is another possible brute-force attack
which goes like this: If Eve had a pretty good idea of a fairly small set of possible messages
that Alice might be encrypting to send to Bob, Eve could simply use the public encryption
50 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
key herself and do all of those encryptions. If one her guesses for the original message was
correct, then its encryption will be the same as the ciphertext which she saw on the public
network, going from Alice to Bob. Therefore, the space of possible messages must be quite
large to prevent this kind of brute-force attack.
There is actually a nice way to prevent this message space-based brute-force attack, by
making a cleartext message artificially larger before encrypting it. One usually does this
by adding some
D EFINITION 3.1.3. Cryptographic salt is random data added to a message Alice wants
to sent to Bob before she encrypts it, which is automatically removed at Bob’s end, after he
decrypts the ciphertext.
Once Alice salts her message, if Eve wants to do the message space-based brute-force
attack, she must try encrypting all possible messages Alice might have sent, plus all pos-
sible random choices of salt. With even a relatively modest number of salt bits, this will
make this brute-force attack entirely too time-consuming.
Reading Response 3.1.2
Does all of this make sense? What was new or interesting, or what was old and un-
interesting? Do you understand this new, “message space-based brute-force attack”
and how cryptographic salt prevents it?
One last remark about keys and generating them, for asymmetric cryptosystems: Keys
for asymmetric cryptosystems are more delicate than for symmetric cryptosystems. Typ-
ically, one picks a private/decryption key of a certain size – large, to prevent brute-force
attacks, and often using strong randomness, also to prevent brute-force attacks which would
be possible if the adversary knew something about the structure of the keys you are likely
to choose.
Then one does a computation using that private/decryption key to create the correspond-
ing public/encryption key. This computation must be very hard to undo, since otherwise
Eve could find the private key from knowing its public key!
Therefore the whole process of key creation can be quite complicated for asymmetric
cryptosystems. Implementations must have a key generation step which can actually be
quite time-intensive, but which is run only once by the person who wants to be able to
receive encrypted messages from strangers on the ’net. This is of course in contrast with the
situation of symmetric cryptosystems, where Alice and Bob simply agree upon some key
with the correct number of bytes, however they want, and start encrypting and decrypting
right away: there is no special “key generation” step for symmetric cryptosystems.
3.2. USING THE RSA ASYMMETRIC CRYPTOSYSTEM IN Python 51
3.2. Using the RSA asymmetric cryptosystem in Python
One of the oldest and still most widely used1 asymmetric cryptosystems is know as
RSA, so named in honor of its three inventors: Ron Rivest, Adi Shamir, and Leonard
Adleman. We’re going to use a Python implementation of RSA to explore the basic fea-
tures of using asymmetric cryptosystems such as key generation, encryption, and decryp-
tion, first in an entirely straightforward way that is not entirely secure. Then we will revisit
these basic functions in a slightly more complex configuration that is actually secure.
3.2.1. Straightforward – not completely secure! – RSA in Python. The Python
module Crypto has a submodule Crypto.PublicKey which implements a class
RSA . You can use this class by the usual
from Crypto.PublicKey import RSA
The first thing to do with an RSA object is to generate a new one, using
key=RSA.generate(n)
where n is the number of bits in (part of) the key. The values allowed for this parameter
must be at least 1024, and must be a multiple of 256.
As is often the case, the larger the key, the more secure is the cryptosystem but the
slower are the calculations to generate the key and also to do encryption and decryption.
Common values one sees in use today are 1024 and 2048, although also 4096 is not unusual.
Code Task 3.2.1
Write a Python function timeRSAkeygen(n) which takes as input a key size n
and generates 100 RSA keys of that size. Time how long each key generation takes,
using the approach to measuring execution time in Code Task 2.2.1, and append that
new execution time to a list. timeRSAkeygen(n) should then return the average
of that list of 100 times.
Then run that program timeRSAkeygen(n) for the key sizes which start at 1024
and go up to 4096 in steps of 256 – that is, for values of n ranging over the list
[x for x in range(1024,4097,256)] – and print out for each key size
the average key generation time produced by timeRSAkeygen(n) for that key
size.
1
In a sense, this is unfortunate: to get the same security with RSA as with elliptic curve cryptosystems,
much larger keys must be used and so the computations are significantly slower. The advantage of RSA is
probably that it is based on mathematics that is at the level of advanced undergraduates, while elliptic curve
cryptosystems are based on mathematics at the graduate level.
52 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
Bonus Task 3.2.1
Make a plot of the timing information you just gathered. That is, for each key length,
you computed the average amount of time it took per generation of a key of that size
over 100 times. Now make a line plot where the x values are those key sizes and the
corresponding y values are the average key generation times for that size key. Make
sure your graph has a title and x- and y-axis labels.
If you are not familiar with making simple line plots in Python, just do a search for
“python line plot” and there will be literally hundreds of pages with quick-
start guides to making line plots with matplotlib.pyplot , many including
code that you can copy and easily adapt.
Let’s look at one of those keys generated with RSA.generate , using a method
exportKey 2 of the RSA class:
>>> from Crypto.PublicKey import RSA
>>> key = RSA.generate(1024)
>>> key.exportKey()
b’-----BEGIN RSA PRIVATE KEY-----\nMIICXQIBAAKBgQDRpxdIHxJTp
uM3R/mHKurOrAXnJ5a/nlucxvst3lpHVcs9Yv+Q\nTTtOrDVvgs7BBq2pdjO
DOYxEmAX2GGDuJIFMKZ8Se2ZUGWeJZ0sJolUoANQ0ihIj\nD7zvFz40kuWLU
7imSMOlQpWk35lyN2BR4I6DvAo6f4JKzEJ1l2wuig+75QIDAQAB\nAoGAQsK
HamLijhq1fdQAhGdJMBidJJd5rHj7yTefomKMsuyB9IFCyiuduBakSWcI\n+
XRr9mt6Sc4YeXtDYrMuooajWQ6Ltg8sGqjpDtrYzPILiVgN8fXz1R2TKAePr
ZUO\nKazRnkMN6QcgX1xilfXFtr3Q2OB67dpnVwK3mtAI3E6tZ4ECQQDZWho
gDB3+xkkO\nES0mH0DtyeEbLSknAF84/xSylD58vVYvwWGma4dHEmZcjKEob
c8dQQrndBm8jVQS\niDblKjhhAkEA9u6EZfk+JqeuoUleirRI5X1Kwvy8jo7
7kEf3noVBE3Eve9eJgcX2\nO+ZrAWke3tKqqu6+Tdy9AKkgl1s5EdfiBQJBA
K3My7k2l0Gj4sT53SVvtmaumG83\nxIFoXbxg1Hcb7X+nkuRq+R+vOiQNxYZ
Z+YAvln8pBIQhpXbNeB29iE/lW+ECQFWB\nxqshIdp02k3TgD97qnp9ZnQa3
Jho/se5hA+KiTxYR18VBfLAQEIByjAU3LHANYU3\nYwLHW1NtPXHsDtkU7pk
CQQC4N+Q/IHEY/Sc+fvha5x8zdWmGwODheGJ6Q2i6GN+I\n9LvYjKW5hZ1Gi
rUBjonzNi1vrQsaaVZatRsLa1iwTRIa\n-----END RSA PRIVATE KEY-----’
Notice a couple of things about this:
• Apparently this is a private key: that is, it is the sort of thing, in an asymmetric
cryptosystem, that must be kept secret.
2This method may be called export key if you are using a different version of the Crypto module.
3.2. USING THE RSA ASYMMETRIC CRYPTOSYSTEM IN Python 53
• Also, when one exports a key, it is a byte string. This means that if you want to
write this to a file, you need to open the file with a command like
file_handle = open("filename", "wb")
The "wb" there means that the file is being opened for writing in byte mode.
If you can export a key, it stands to reason that you can also import one. In fact, this is
quite easy: suppose that private key byte string were stored in a variable exported key .
Then you could make an identical RSA object by
>>> from Crypto.PublicKey import RSA
>>> new_key = RSA.importKey(exported_key)
and that new key will have all of the properties and uses as the original key , above.
Code Task 3.2.2
Write a Python function gen save RSA private(n, name) which takes as
input a key size n and a string name . The function should generate an RSA key of
size n and store it in a file with the given name . It should also return that instance.
Then write a Python function recall RSA private(name) which takes as
input a string name . It should open the file with that name , create an instance of
the RSA class with the key value stored in that file, and it should return that instance.
The ideas is that you should be able to use these functions in situations like:
>>> from Crypto.PublicKey import RSA
>>> key = gen_save_RSA_private(1024, "AliceRSA")
>>> # do lots of encryption and decryption with this key
>>> # ... then take a break and even quit Python and
>>> # reboot the computer
>>> # ... but eventually restart the computer and Python
>>> # and do this:
>>> from Crypto.PublicKey import RSA
>>> key = recall_RSA_private("AliceRSA")
>>> # and then lots more encryption and decryption with
>>> # the same key [identity] as before
As we mentioned near the end of §3.2, in an asymmetric cryptosystem, first one com-
putes the private key and then, from the private key, one computes the corresponding public
key. In fact, as was also mentioned, it is important for the security of the asymmetric cryp-
tosystem that while computation
private key public key
54 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
can be done in a reasonable amount of time – it doesn’t have to be fast, though, since it will
be done only once, when setting up the cryptosystem – the computation backwards
private key public key
must be entirely impractical no matter how great are the computational resources available
to the enemies that it is imagined will try to break this secure communication.
In the case of RSA, the easy direction of from private to public key is just multiplication
and the hard, backwards direction is basically just division. As most of us remember from
grade school, it is much easier and more direct to multiply two numbers than it is to divide
two. For RSA, the two numbers one multiplies are prime numbers, meaning that they
are numbers which have no positive, whole number factors other than one and themselves.
Multiplication of primes is still just as easy (and fast) as for any other types of numbers,
even if they are hundreds of digits long. However, if you are given a 600 digit number and
told that it is the product of two 300 digit primes, then it would take an immense amount
of time – a sizeable fraction of the age of the universe: really an immense amount of time –
using the best algorithms and hardware we have today to find those primes.
It is, however, not a mathematical theorem (at this time) that there cannot be some great
factorization algorithm out there, as yet undiscovered, which would make the step
private key public key
relatively fast for RSA, the most widely used asymmetric cryptosystem on the planet right
now. That may or may not make you uncomfortable....3
Reading Response 3.2.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you understand how RSA’s security could rely upon how it is
easy to multiply but hard to divide [factor]?
Regardless of these general considerations around the security of RSA, since it is a
working asymmetric cryptosystem, there must be a way to get just that public key from the
private key in Python. Here’s how (continuing from an example given above):
3
And it may or may not make you even more uncomfortable to learn that quantum computers – com-
puters which can do a kind of general-purpose computation in a different model of what that means, based
on quantum mechanics – do have a fast factorization algorithm. RSA may well be broken the minute the first
real, full-sized quantum computer is turned on!
3.2. USING THE RSA ASYMMETRIC CRYPTOSYSTEM IN Python 55
>>> from Crypto.PublicKey import RSA
>>> key = RSA.generate(1024)
>>> key.exportKey()
b’-----BEGIN RSA PRIVATE KEY-----\nMIICXQIBAAKBgQDRpxdIHxJTpuM3
R/mHKurOrAXnJ5a/nlucxvst3lpHVcs9Yv+Q\nTTtOrDVvgs7BBq2pdjODOYxEm
AX2GGDuJIFMKZ8Se2ZUGWeJZ0sJolUoANQ0ihIj\nD7zvFz40kuWLU7imSMOlQp
Wk35lyN2BR4I6DvAo6f4JKzEJ1l2wuig+75QIDAQAB\nAoGAQsKHamLijhq1fdQ
AhGdJMBidJJd5rHj7yTefomKMsuyB9IFCyiuduBakSWcI\n+XRr9mt6Sc4YeXtD
YrMuooajWQ6Ltg8sGqjpDtrYzPILiVgN8fXz1R2TKAePrZUO\nKazRnkMN6QcgX
1xilfXFtr3Q2OB67dpnVwK3mtAI3E6tZ4ECQQDZWhogDB3+xkkO\nES0mH0Dtye
EbLSknAF84/xSylD58vVYvwWGma4dHEmZcjKEobc8dQQrndBm8jVQS\niDblKjh
hAkEA9u6EZfk+JqeuoUleirRI5X1Kwvy8jo77kEf3noVBE3Eve9eJgcX2\nO+Zr
AWke3tKqqu6+Tdy9AKkgl1s5EdfiBQJBAK3My7k2l0Gj4sT53SVvtmaumG83\nx
IFoXbxg1Hcb7X+nkuRq+R+vOiQNxYZZ+YAvln8pBIQhpXbNeB29iE/lW+ECQFWB
\nxqshIdp02k3TgD97qnp9ZnQa3Jho/se5hA+KiTxYR18VBfLAQEIByjAU3LHAN
YU3\nYwLHW1NtPXHsDtkU7pkCQQC4N+Q/IHEY/Sc+fvha5x8zdWmGwODheGJ6Q2
i6GN+I\n9LvYjKW5hZ1GirUBjonzNi1vrQsaaVZatRsLa1iwTRIa\n-----END
RSA PRIVATE KEY-----’
>>> public_key = key.publicKey()
>>> public_key.exportKey()
b’-----BEGIN PUBLIC KEY-----\nMIGfMA0GCSqGSIb3DQEBAQUAA4GNADCBi
QKBgQDRpxdIHxJTpuM3R/mHKurOrAXn\nJ5a/nlucxvst3lpHVcs9Yv+QTTtOrD
Vvgs7BBq2pdjODOYxEmAX2GGDuJIFMKZ8S\ne2ZUGWeJZ0sJolUoANQ0ihIjD7z
vFz40kuWLU7imSMOlQpWk35lyN2BR4I6DvAo6\nf4JKzEJ1l2wuig+75QIDAQAB
\n-----END PUBLIC KEY-----’
If you use this public key byte string in an RSA.importKey then it will work, but it will
build an instance of the RSA class which only has the public key and so can do encryption
but not decryption. This is the kind of RSA key object which most users will be building to
send encrypted messages to the owner of the private key, while the full, original RSA key
object with both private and public keys should only be known to and used by the person
receiving and decrypting those messages.
Code Task 3.2.3
Do a version of Code Task 3.2.2 which also saves the public key in a separate file.
That is, make a Python function gen save RSA pub priv(n, name) which
takes as input a key size n and a string name . The function should generate
an RSA key of size n and store it in a file with name name+".private" .
It should also store just the corresponding public key in a file with name
56 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
name+".public" . Finally, it should return the full instance (not just the public
version).
Then write a Python function recall RSA public(name) which takes as in-
put a string name . It should open the file with name name+".public" , create
an instance of the RSA class with the key value stored in that file, and it should return
that instance.
The idea is that you should be able to use these functions in situations like:
>>> # Bob does:
>>> from Crypto.PublicKey import RSA
>>> key = gen_save_RSA_pub_priv(1024, "BobRSA")
>>> # then puts the file "BobRSA.public" on an open
>>> # website, where Alice could download it
>>> # ... then, with this file on her machine,
>>> # Alice can do:
>>> from Crypto.PublicKey import RSA
>>> public_key = recall_RSA_public("BobRSA.public")
>>> # and Alice can encrypt using this new
>>> # public key and send messages which only Bob
>>> # will be able to decrypt using his original
>>> # key which has both public and private keys
We still haven’t used these RSA key objects to encrypt and decrypt messages. Remem-
ber, it is an object like the public key one in the example above which is all that is
needed to do encryption (we use the same key as above, rather than generating a new one,
just for consistency), which we could create with
>>> from Crypto.PublicKey import RSA
>>> public_key = RSA.importKey(b’-----BEGIN PUBLIC KEY-----\nMIGf
MA0GCSqGSIb3DQEBAQUAA4GNADCBiQKBgQDRpxdIHxJTpuM3R/mHKurOrAXn\nJ5a
/nlucxvst3lpHVcs9Yv+QTTtOrDVvgs7BBq2pdjODOYxEmAX2GGDuJIFMKZ8S\ne2
ZUGWeJZ0sJolUoANQ0ihIjD7zvFz40kuWLU7imSMOlQpWk35lyN2BR4I6DvAo6\nf
4JKzEJ1l2wuig+75QIDAQAB\n-----END PUBLIC KEY-----’)
We can then use the encrypt method of this public key object to do encryption.
But note first that encrypt has a second argument, just for backwards compatibility
with earlier versions of this module, which must be some long integer that is actually com-
pletely ignored. Notice also that the return value of encrypt is a tuple, the first element
3.2. USING THE RSA ASYMMETRIC CRYPTOSYSTEM IN Python 57
of which is the byte string of the ciphertext, so:
>>> ciphertext = public_key.encrypt(b’This is a test’,1)[0]
>>> print(ciphertext)
b’\xbfa\xb0\x8b\x08*\xdf\x96gK0B/\xa2\x8e\x1e\x86\xf2]8\xd9\xc
b\xc2w\xad\x82}\xb5\xf4\x97\xe1g\x91\xc8\x00\x14fr\x15\xc1\x8b
\xbd\x8di\x1f\xbc\tt\x8f\xfa\x1c\xad\xb8V\xca\xc7\xed\xb3X"\xb
7ra\xf8h\x01\x94c\xc9\xaf\xf6\xd5\x0f/\x83[\xea\xfd\x85\xdep9\
xf2\xee\xfc\xec\xd7\xfc\xfd\xe0\xf1\xcc\x12\x8fZ\xad\xf0|n@\xe
8vD\xb48}a\xd8ˆ\xdb#\xbb\x01\x8e\x8c\x1a;[\x8e\xfd@\xc1\xb8Ohw
\xc4\x15’
As we mentioned above, this public key object cannot do decryption. In fact, if
you try it, you will get some error like
>>> public_key.decrypt(ciphertext)
Traceback (most recent call last):
File "<stdin>", line 1, in <module>
File "/usr/lib/python3/dist-packages/Crypto/PublicKey/RSA.py
", line 174, in decrypt
return pubkey.pubkey.decrypt(self, ciphertext)
File "/usr/lib/python3/dist-packages/Crypto/PublicKey/pubkey
.py", line 93, in decrypt
plaintext=self._decrypt(ciphertext)
File "/usr/lib/python3/dist-packages/Crypto/PublicKey/RSA.py
", line 239, in _decrypt
mp = self.key._decrypt(cp)
TypeError: Private key not available in this object
On the other hand, if we still had the key object around, which contained the private
key – or if we saved off the exported version of that key into a file and then rebuilt key
with RSA.importKey on the byte string in that file – then we could do decryption as
simply as
>>> key.decrypt(ciphertext)
b’This is a test’
Code Task 3.2.4
Do a version of Code Task 2.2.1 for the RSA.
58 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
That is, make a new RSA key object of key size 1024. Pick a random cleartext
of length around 100 bytes. Then encrypt it 100 times with key.encrypt and
report on the average time it takes to do the encryption (using the timing method
described in Code Task 2.2.1).
Likewise take the ciphertext coming from any of those encryptions with key and
do 100 decryptions of that ciphertext, reporting on the average time it takes to do the
decryption.
Code Task 3.2.5
Do a version of Code Task 2.2.2 for RSA.
That is, make a new RSA key object of key size 1024. Pick a cleartext you like of
length around 80 or 100 bytes. Make two versions of the cleartext which differ by
only one bit.
Encrypt both versions of the cleartext with key.encrypt and use your program
bits diff display from Code Task 2.1.2 to see where and how much the two
ciphertexts differ.
You might also try the above for several different keys, just to make sure that it’s not
a fluke which is special for the choice of key you happened to make.
Reading Response 3.2.2
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel comfortable using Python’s implementation of the RSA
cryptosystem?
3.2.2. More secure RSA in Python using OAEP and PKCS #1. Unfortunately, the
direct use of RSA, as we just did, is not considered secure. This is for several reasons, some
to do with the special structure of RSA, some simply to do with the fact that the bare RSA
we used above has no salt and so always encrypts the same cleartext to the same ciphertext,
when using the same key – for (some) details, see the section Attacks against plain RSA in
the Wikipedia article RSA (cryptosystem).
The cryptographic community has developed a theoretical structure which protects
against these attacks on plain RSA called Optimal Asymmetric Encryption Padding
[OAEP]. OAEP was incorporated into a standard called PKCS #1, Public Key Cryptog-
raphy Standard #1, issued by RSA Laboratories, an American IT security firm founded
3.2. USING THE RSA ASYMMETRIC CRYPTOSYSTEM IN Python 59
by the very same Rivest, Shamir, and Adleman who created the RSA cryptosystem. It is
PKCS #1 that is considered safe to use in practice4.
PKCS #1 is implemented in Python with a class PKCS1 OAEP in the Crypto.Cipher
module. Since it is essentially just a padding for an asymmetric cryptosystem – the “P” in
“OAEP” – we make a PKCS #1 cipher object by feeding it an RSA key object:
>>> # Bob does:
>>> from Crypto.PublicKey import RSA
>>> key = gen_save_RSA_pub_priv(1024, "BobRSA")
>>> # then puts the file "BobRSA.public" on an open website
>>> # ... then, Alice gets this file on her machine, and does:
>>> from Crypto.PublicKey import RSA
>>> public_key = recall_RSA_public("BobRSA.public")
>>> from Crypto.Cipher import PKCS1_OAEP
>>> cipher_for_encryption = PKCS1_OAEP.new(public_key)
>>> # and Alice can encrypt messages, e.g.,
>>> cleartext = b’This is a test. This is only a test.’
>>> ciphertext = cipher_for_encryption.encrypt(cleartext)
>>> print(ciphertext)
b’\x17\xf4\xaa\xc8\x84\xa9\xe5\x7f8\xa7\x86‘\xea\x13\x1c\x1d>d\
xd5\xaeI\xed\xf8˜$Y\xa3\x7f\xb7h\x14W\x1a\x99\\\xdd]P\x08\x0f*\
xd4z\xc2u-\xd6\xbb\xf3\xe2;\xf6\xdaZ\xc9!0-\xce\xa1S\xb4\xcc%,\
x08\x0c#\x00v\x87Kn41\x84\xc8\x18\xdf\xc9\xd8\x83?\xb5\xce\xbf\
x9b<\xadr.6xa\xa8\xad\xca\x0c\x1b\xf4\xc3\x8c\xe0\x05\\U\xc3\x0
b\xdf\x02\xfdk*l\xa3*\xa5>\x7f\xcd\xfa\xb3\xdb\x91\xfd\xd5\x15\
x8b’
>>> # Bob, on his machine, where he the key object which also
>>> # contains the private key, can do decryption by
>>> from Crypto.Cipher import PKCS1_OAEP
>>> cipher_for_decryption = PKCS1_OAEP.new(key)
>>> cleared_text = cipher_for_decryption.decrypt(ciphertext)
>>> print(cleared_text)
b’This is a test. This is only a test.’
Notice a couple of things about this:
• If you, dear reader, type exactly the same commands into your computer, you will
still get back the original cleartext when you do the decryption, but the intermedi-
ate step of the ciphertext you will get will be different from the one shown above.
4at least with large enough key sizes, and until someone turns on a fully functional quantum computer
60 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
The reason for this is that OAEP includes some cryptographic salt, so that every
time you do encryption, even of the same cleartext, you get a different ciphertext.
(This is a good thing, it increases security.)
• You cannot encrypt cleartexts which are very large with the approach described
above. So in a practical situation, one must either break up a long message into
small chunks and encrypt each of them, or else use some other approach which
avoids the size limit. The second of these solutions will be described in the very
next section, below.
As you can probably guess, it makes sense now to see if this new, more secure version
of RSA is slow or fast, and if it satisfies Shannon’s diffusion condition.
Code Task 3.2.6
Do a version of Code Task 3.2.4 for the RSA in the PKCS #1/OAEP version.
That is, make a new RSA key object of key size 1024. Use it to make a
PKCS1 OAEP cipher object. Pick a random cleartext of length around 100 bytes.
Then encrypt it 100 times with cipher.encrypt and report on the average time
it takes to do the encryption (using the timing method described in Code Task 2.2.1).
Likewise take the ciphertext coming from any of those encryptions and do 100 de-
cryptions of that ciphertext, reporting on the average time it takes to do the decryp-
tion.
Code Task 3.2.7
Do a version of Code Task 2.2.2 for the RSA in the PKCS #1/OAEP version.
That is, make a new RSA key object of key size 1024. Use it to make a
PKCS1 OAEP cipher object. Pick a cleartext you like of length around 80 or 100
bytes. Make two versions of the cleartext which differ by only one bit.
Encrypt both versions of the cleartext with cipher.encrypt and use your pro-
gram bits diff display from Code Task 2.1.2 to see where and how much
the two ciphertexts differ.
You might also try the above for several different keys, just to make sure that it’s not
a fluke which is special for the choice of key you happened to make.
You might also try just comparing two encryptions made in exactly the same way, of
exactly the same cleartext, with your program bits diff display to see that
the cryptographic salt which OAEP uses makes for quite different ciphertexts every
time you do encryption, even starting from the same cleartext.
3.2. USING THE RSA ASYMMETRIC CRYPTOSYSTEM IN Python 61
3.2.3. How to use RSA in a way that is both fast and secure. It should have been
clear from Code Tasks above that RSA, while it has the nice features of an asymmetric
cryptosystem with regards to key distribution and management, is very slow. Additionally,
it has the problem that one cannot encrypt large messages – although, presumably, one
could solve this problem by one of the block chaining methods discussed above. But there
is another approach which takes advantage of the good features of asymmetric cryptosys-
tems while avoiding the weaknesses.
For this, one uses the whole approach described in this chapter of asymmetric encryp-
tion, including generating a public and private key, posting the public key on some public
website for anyone to use, etc. But rather than doing the entire communication in RSA (or
some other asymmetric cryptosystem, one uses that slow but secure asymmetric cryptosys-
tem to transmit a randomly chosen “session key”. Then the rest of the communication is
done by some fast, well-chained, block cipher (such as AES).
A “session” in this context might consist of some agreed-upon number of messages,
or all of the messages necessary to complete some protocol (like making a credit card
payment on the ’net), or all of the message which are send before some time limit expires.
Whatever triggers the end of a session, when that happens, a new random session key must
be generated for further secure communications between the same parties.
This approach is extremely common on the Internet today. In fact, most of the time
customers watch a movie over the Internet, this is the way the video stream is sent to them
by the provider. Providers want to send the video data in encrypted form, so that their
valuable movies cannot be stolen by someone who is merely observing the packets as they
go by on the Internet. Since customers probably have never met the providers in person
to agree upon a key for a symmetric cryptosystem, it would seem that the providers would
have to use an asymmetric cryptosystem for this encrypted video stream – but asymmetric
cryptosystems are too slow to run video through without the watching human noticing! On
the other hand, block ciphers like AES are fast enough. Since the session key negotiation at
the beginning of the stream happens in a fraction of a second, the approach in this section
allows for security, without difficulties of key distribution, but nevertheless using a fast
block cipher (for the entire video stream, after the initial session key was sent using an
asymmetric cryptosystem).
Below is a graphical depiction of this approach, assuming that Aragorn wants to initiate
fast and secure communication with Bilbo, even though they have not met in person to
share any key. Note that once the initial session key negotiation is finished, there can be
any number of messages coming from either side until the session ends – the example in
this diagram where the messages strictly alternate is only one possibility. In fact, often
Aragorn sends her first two messages to Bilbo and 2 , below) as one, larger, concatenated
message ckAB + c1 (using the Python notation “+” for concatenation of byte strings), since
she does not need to wait for any response from Bilbo after 1 before doing 2 .
62 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
Fast and secure, session key-based communication, graphically:
Aragorn on public network Bilbo
(where Gollum is watching)
1 generate key pair (keB , kdB ) for
asymmetric cryptosystem A;
download keB public encryption key keB publish keB
choose random session key kAB 2
for symmetric cryptosystem S;
encrypt kAB using A with
public key keB , getting cAB ;
transmit cAB A-encrypted session key cAB receive cAB ;
decrypt cAB using A with
private key kdB , getting kAB
create cleartext message m1 ; 3
encrypt m1 using S with
session key kAB , getting c1 ;
transmit c1 S-encrypted ciphertext c1 receive c1 ;
decrypt c1 using S with
session key kAB , getting m1
4 create cleartext message m2 ;
encrypt m2 using S with
session key kAB , getting c2 ;
receive c2 ; S-encrypted ciphertext c2 transmit c2
decrypt c2 using S with
session key kAB , getting m2
create cleartext message m3 ; 5
encrypt m3 using S with
session key kAB , getting c3 ;
transmit c3 S-encrypted ciphertext c3 receive c3 ;
decrypt c3 using S with
session key kAB , getting m3
etc....
3.2. USING THE RSA ASYMMETRIC CRYPTOSYSTEM IN Python 63
Reading Response 3.2.3
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you understand how Alice and Bob would use this session-based
approach?
It is important that the session keys used in the above scheme come from a good source
of randomness. Fortunately, the Python module Crypto has a submodule Crypto.Random
that implements a method get random bytes(N) which generates a good random
byte string of length N . This method should be used when creating session keys.
Bonus Task 3.2.2
Implement the above, session-based approach to fast and secure communications.
Since we’re not interested (at the moment) in the technicalities of publishing material
(such as public keys) on the web or of sending messages over the Internet, you will
have to play the roles of both Aragorn and Bilbo. Instead of posting or transmitting
files and data, you will just put these things into files and the next program can read
in that file, which, we will imagine, came off the Internet.
Hoping to keep matters clear, we’ll insist that every Python function which Aragorn
should run will have a name that starts with A , while those to be run by Bilbo
will have names starting with B . To enforce this idea of separation, the methods
beginning with A should only use information that they create themselves or get out
of files on the disk (which is playing the role of the public Internet in this scenario)
with names ending in .public , and not internal information from the methods
beginning with B ; and vice-versa for the information used by the B methods.
Here are the steps you should write code for, labeled to correspond to the numbers
steps in the diagram above, and with the initial of the actor who is to be imagined
to be doing the work in that protocol diagram (i.e., basically the initial tells which
column of the diagram is being indicated):
B 1 : Write a Python method B asym keygensend(n) which takes as input
a key size n . The function should
(1) generate an RSA key of size n
(2) store the full version (public and private) in a file
B asym key.private , and
(3) store just the public key part in a file B asym key.public .
Remember, saving data in a file whose name ends in .public is “send-
ing” the file over the public network, in the metaphor of this problem.
[Note, you can accomplish this in very few lines by just calling the function
64 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
gen save RSA pub priv(n, name) from Code Task 3.2.3 with the
correct arguments!]
A 2 : Write a Python method A session keygensend() which generates
and “sends” (puts in a file on disk with a name ending in .public ) a
session key for the AES cryptosystem. This new method should
(1) use get random bytes to generate a high-quality session key kAB
of the right size for AES,
(2) encrypt kAB with PKCS1 OAEP (you can reuse some lines of code
from the Code Tasks 3.2.6 or 3.2.7 for this), and
(3) store the resulting ciphertext in a file
encrypted session key.public
(4) store the unencrypted session key kAB in a file A session key for
later use by the A methods.
Remember, as discussed immediately before Code Task 3.2.2,
above, that when you want to write encrypted data to
a file, you should open the file the with the command
file handle = open("filename", "wb") and then write the
date, as usual, with file handle.write(data bytestring) ).
B 2 : Write a Python method B session keyreceive() which reads and
decrypts the session key which Aragorn just created and “sent.” For this, it
should
(1) open the file encrypted session key.public for reading
bytes, and get its contents,
(2) read the key in the file B asym key.private (use your method
recall RSA private from Code Task 3.2.2 to do this!),
(3) use that key object to create a PKCS1 OAEP cipher object,
(4) decrypt the session key previously read in, and
(5) save for later use (the bytes of) the decrypted session key in a file
B session key .
A 3 : Write a Python method A send encrypted message(m, n)
which takes as input a byte string m (the message), and an int n (the
message number). This method should:
(1) open the file A session key for reading bytes, and get its contents,
(2) using that key, create a Crypto.Cipher object for AES in CBC
mode (as we did in §2.2),
(3) encrypt the message m using that cipher object,
(4) open a file whose name is ‘‘c’’+str(n) for writing bytes, and
3.2. USING THE RSA ASYMMETRIC CRYPTOSYSTEM IN Python 65
(5) write (the bytes of) the encrypted message to that file.
B 3 : Write a Python method B receive encrypted message(n)
which takes as input an int n (the message number) and returns the de-
crypted message. This method should:
(1) open the file B session key for reading bytes, and get its contents,
(2) using that key, create a Crypto.Cipher object for AES in CBC
mode (as we did in §2.2),
(3) open a file whose name is ‘‘c’’+str(n) for writing bytes,
(4) read (the bytes of) the encrypted message from that file,
(5) decrypt the message m using the previously built cipher object, and
(6) return the cleartext.
B 4 : Write a Python method B send encrypted message(m, n)
which takes as input a byte string m (the message), and an int n (the
message number). This method should do exactly the same thing as the
A send encrypted message(m, n) described above, only switch-
ing A ↔ B everywhere.
A 4 : Write a Python method A receive encrypted message(n)
which takes as input an int n (the message number) and returns the de-
crypted message. This method should do exactly the same thing as the
B receive encrypted message(n) described above, only switch-
ing A ↔ B everywhere.
[Yes, this seems like a long Task, but it extensively reuses or builds on other code
you’ve already written, and has very detailed instructions!]
66 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
3.3. Digital Signatures
The discovery of asymmetric cryptosystems – or, as they are more often called in non-
technical contexts like this paragraph, public-key cryptosystems – has changed the face of
the Internet by allowing secure communications between entities who have never met in
real space to share a secret key. Without this innovation, commerce over the Internet would
be infinitely more difficult, for example. But beyond this straightforward application of
public-key crypto to securing communication against eavesdroppers, other powerful appli-
cations have also been discovered. A very nice one, which has many uses in the networked
world, is the idea of
D EFINITION 3.3.1. A digital signature is a chunk S of data which accompanies a
message m. There is also an algorithm called verifying the signature which can be run
with inputs S, m, and some other public data which, if it returns ACCEPT, proves that the
signature could only have been produced by someone who possesses a certain piece of
secret data.
If we make some assumptions about behavior in the real world – which might or might
not be justified! – then these digital signatures would have many extremely valuable appli-
cations. The assumptions we would need include:
• There is some reliable association between the public information used to verify
a signature and the real-world person who claims they created the signature. For
example, perhaps in order to have the public information listed on a well-known,
trusted site next to a human name, that human must go to some government office,
show an ID, and register their public data.
• The humans can be trusted to keep the secret data needed to generate those sig-
natures reliably secure. For example, the don’t keep the data on a machine which
can ever be hacked, and they don’t forget to erase it before they throw out an old
hard drive, etc.
Now for some details.
3.3.1. Naive digital signatures. It has been important since the beginning of our study
of cryptology that encryption and decryption are opposites: what we would call “inverse
functions” in mathematics5, meaning that doing first one and then the other simply gets you
back to where you started. Since inverse functions work in either order, that means that not
only is the decryption of the encryption of some message just the original message, but so
also is the encryption of the decryption of a message.
5
E.g., squaring and taking the square root are inverses (at least if you start with positive numbers), so that
squaring a square root or square rooting a square gives back the original number.
3.3. DIGITAL SIGNATURES 67
That may seem like a mere computational curiosity – why would anyone decrypt a
message which had not first been encrypted? – but it actually has a very nice feature:
only the person who has the private key for some asymmetric cryptosystem can do the
decryption, while anyone who has downloaded the corresponding public key can do the
encryption. This situation where only the one person who has some secret information (the
private key) can do something while everyone else can do some other thing using a piece
of public information (the public key) sounds very much like what it supposed to happen
in a digital signature scheme.
To be specific, suppose Bilbo wants to be able to digitally sign some message m, and
Aragorn wants to be able to verify the signature. First, Bilbo should generate a public/pri-
vate key pair and put his public key on his website. Then he should use the decryption
algorithm, with his private key, on the message m, yielding a chunk of data we’ll call s.
Finally, Bilbo should send both the message m and the signature s to Aragorn.
Aragorn will have downloaded Bilbo’s public key. What he can do is encrypt Bilbo’s
(purported) signature s, using that public key, to get a chunk of data m′ . Then Aragorn
will accept the signature as valid if his m′ is the same as Bilbo’s message m. Since only
the person who has Bilbo’s private key should be able to make a chunk of data s whose
encryption is m, this method does implement the digital signature scheme as promised.
Graphically:
Naive digital signatures:
Aragorn on public network Bilbo
generate key pair (keB , kdB ) for
asymmetric cryptosystem A;
download keB public encryption key keB publish keB
with message m:
decrypt m using A with
private key kdB , getting s;
receive (m, s) signed message (m, s) transmit (m, s)
encrypt s using A with
public key keB , getting m′ ;
if m′ = m
ACCEPT
else
REJECT
68 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
Reading Response 3.3.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel comfortable with the way these naive digital signatures
are generated and verified?
Code Task 3.3.1
Write a Python function sign(key, m) which takes as input a full (public and
private) key for RSA, in the form of a byte string that might have been produced by
exportKey() and a (byte string) message m and produces the signature s on
m as described above. To do this, use RSA.importKey on the key variable to
build an RSA object. Use the decrypt method of this RSA object on the variable
m to produce the signature s , which your method should return.
Next, write a Python function verify(pk, m, s) which takes as input a pub-
lic key pk for RSA, in the form of a byte string that might have been produced
by exportKey() , a (byte string) message m , and a (byte string) signature s
on that message, and returns a Boolean which indicates if the signature is valid or
not, following the scheme above. To do this, use RSA.importKey on the pk
variable to build an RSA object public key and use the encrypt method of
public key to create the byte string mprime . Don’t forget that encrypt
takes a second, dummy variable which may as well have the value 1 and returns a
tuple, the 0th element of which is the encryption. In other words, you need to use
mprime = public key(s, 1)[0] . Then simply return the (Boolean) value
of mprime == m .
The idea is that you should be able to use these functions in situations like
>>> # Bilbo does: :
>>> from Crypto.PublicKey import RSA
>>> key = RSA.generate(1024)
>>> full_key_exported = key.exportKey()
>>> public_key_exported = key.publicKey().exportKey()
>>> # Bilbo puts the public_key_exported on his website
>>> m = b’some message Bilbo wants to sign’
>>> s = sign(full_key_exported, m)
>>> # Bilblo sends (m,s) to Aragorn, maybe by email
>>> # Aragorn downloads the public_key_exported and
>>> # gets an email containing (m,s) and does:
>>> verify(public_key_exported, m, s)
>>> # if it is True, he can trust the message
3.3. DIGITAL SIGNATURES 69
3.3.2. Better digital signatures using hash functions. One big problem with the
naive digital signatures described above is that they are (potentially) ... big. That is, the
signature s on a message m is (essentially) the same size as the message itself – after all,
it’s a decryption, pretending the message is a ciphertext. So signing a large document, or
maybe a large database or high-resolution image would basically require an unacceptable
twice the space to store or transmit!
What we need to make much smaller signatures is some sort of algorithm which es-
sentially summarizes the contents of the message m you want to sign. Then a very similar
signature scheme to the one described above can do its verification algorithm by comparing
the encrypted signature to the summary of the message, rather than to the entire message.
There’s a name for this magical summarizing algorithm:
D EFINITION 3.3.2. A cryptographic hash function is an algorithm H which takes as
inputs bit strings x of arbitrary length and produces outputs H(x) of a fixed size, satisfying
the properties:
(1) “ease of computation” computation of H(x) is very fast;
(2) “pre-image resistance” starting with any bit string y of the size of the outputs of
H, it is very difficult to find a bit string x with the property that H(x) = y;
(3) “second pre-image resistance” starting with any specific bit string x1 , it is very
difficult to find another x2 such that H(x2 ) = H(x1 );
(4) “collision resistance” it is very difficult to find any two bit strings x1 and x2 such
that H(x2 ) = H(x1 ).
In some sense, the key phrase in the above definition is “very difficult,” in the sense
that there are certainly very many possible inputs to a hash function which give the same
output – in fact, there are an infinite number of such colliding inputs! Think about it: if the
output size of a hash function is, say, 256 bits, then what happens if you look at input bit
strings of length 257? There are twice as many input of that size as there are outputs, so
the most efficient possible way to spread out the way the outputs come from inputs would
have every output coming from two inputs! And the same argument says there are at least
four inputs of length 258 bits giving the same output, etc. But to be a cryptographic hash
function, it must be very hard to find such collisions.
In practice, what this means is that the algorithms used for hash functions do a lot of
random fiddling around with the bits of their inputs, so there is not enough structure in the
algorithm which might give an attacker a systematic way to try to find collisions or pre-
images or second pre-images. Without structure, the only option left would seem to be a
brute-force attack, but since there are roughly as many bit strings of length 256 as there are
particles in the universe (leaving out photons and neutrinos, though), brute-force is not a
viable attack.
70 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
Cryptographic hash functions are useful because if you have a message or document
– think of it as a bit string m – the H(m) acts a kind of fingerprint for m. For example:
suppose Bilbo tells Aragorn what H(m) is without revealing m on Monday, and only on
Friday reveals m. Aragorn can be confident that it was the same m Bilbo was thinking of
on Monday as he revealed on Friday, because it would have been “very difficult” to find
another m′ for which H(m′ ) = H(m).
Reading Response 3.3.2
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel like you understand what is useful about a hash function,
and why there are theoretically infinitely many failures of the conditions pre-image
resistance, second pre-image resistance, and collision resistance but nevertheless the
fact that it is “very difficult” to find those failures makes hash functions practical?
There are a number of hash functions in use today, and there are some which were
used for a while but which now are no longer considered secure. As mentioned above, the
security of a hash function in part rests upon its lack of structure ... but any algorithm which
is clear enough to be written down and accepted as a standard must have some structure,
and eventually computer scientists have found the way to attack some of the older hash
functions using only that little bit of structure. Here are some hash functions which were
or are widely used:
• For around a decade starting in the early 1990s, the most widely used crypto-
graphic hash function was called md5. This algorithm was developed by Ron
Rivest and published in 1992. The output size of md5 is 128 bits.
While md5 was thought to be flawed since the middle 1990s, a real attack was
not published until 2004, when it was shown not to be collision resistant [WY05].
However, md5 is still used extensively today to verify that a large data transfer
has not suffered a transmission error – i.e., it is still a useful tool to test for non-
malicious data corruption.
• The most widely used cryptographic hash function from the late 1990s until re-
cently, and one which is built into many widely accepted and standardized cryp-
tographic protocols, is SHA-1, with an output size of 160 bits.
SHA-1 was developed by US National Security Agency in a semi-public pro-
cess, and was adopted by the National Institute of Standards and Technology
[NIST] as part of several US Federal Information Processing Standards.
In 2004, some work was published which indicated that SHA-1 might be vul-
nerable to certain kinds of attack. (See [PS04].) For this reason, NIST required
in 2010 many US federal data protection applications to move to another hash
function.
3.3. DIGITAL SIGNATURES 71
• At the time of this writing, most security-conscious users and organizations rec-
ommend SHA-2, usually in its SHA-256 variant, which has an output size of 256
bits. Given recent revelations of the NSA’s involvement – and weakening of –
cryptographic protocols, it might be a cause of concern that NSA participated in
the development of SHA-2.
SHA-2 is implemented in Python with a class SHA256 in the Crypto.Hash mod-
ule. You can create this object with SHA256.new(s) , where s is the entire byte string
to be hashed. Another way use this class, particularly useful if there is a lot of data to
be hashed, is to pass SHA356.new the first piece of the data, and the to call the class’s
update method with the remaining pieces of data.
In either approach, when you are done feeding the SHA256 instance your data to be
hashed, you can call the digest() method to get the hash value as a byte string, or the
hexdigest() method to get that same hash value as a character string written out in
hexadecimal. For example:
>>> from Crypto.Hash import SHA256
>>> hash_object = SHA256.new(b’first bit of data’)
>>> hash_object.update(b’some more data’)
>>> hash_object.update(b’last bit of the data’)
>>> hash_object.digest()
b’:\x93"P‘\xd2Y-\x8c/\xab\xfa=\x8e\xa7\xb7N\xdcF\xae[\xf9\xa4\xd
6.\xa1\xfeR\xe4\xcb\x89\xb8’
>>> hash_object.hexdigest()
3a93225060d2592d8c2fabfa3d8ea7b74edc46ae5bf9a4d62ea1fe52e4cb89b8
Note that you can get the digest from a SHA256 instance and but still add more data
by calling the update method and get a later digest which will be the hash function
output for all of the input you’ve given that instance since it was created with new . What
this means is that you must create a new instance with new every time you want to start
fresh with computing the hash function of a new bit string.
Code Task 3.3.2
Since SHA256 is believed to have the properties that make it a good cryptographic
hash function, it must be that small changes in the input make for drastic changes
in the output, 256-bit hash value. In fact, the best thing would be for differences of
just one bit in the input – by changing a bit, or adding one more bit to the sequence
(somewhere: adding it at the beginning, end, or somewhere in the middle) to make
approximately half of the bits in the hash value change. Let’s test this!
72 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
First, pick some base input byte string that you like of about 50 bytes. Compute its
SHA256 hash value. Now try changing one bit of this base input, computing the
hash value of the new input string, and comparing the two hash values with your
bits diff display from Code Task 2.1.2 to see where and how much the two
hashes differ.
Repeat this several times (at least 10 times), changing bits at different locations in
the input string, and see each time how much it changes the bits of the hash string.
Also compute the average number of bits which were changed in the hash value in
your experiments.
As mentioned above, the fact that the cryptographic hash of a message is a reliable
fingerprint for the message allows us to make a much more efficient signature scheme. In
this scheme, rather than sending the message and its decryption, we send the message and
the decryption of the hash of the message. Therefore the signature is a data chunk which is
as big as the output size of the hash function we’re using, no matter how big is the message
to be signed.
Graphically:
Better [smaller] digital signatures with hashing:
Aragorn on public network Bilbo
generate key pair (keB , kdB ) for
asymmetric cryptosystem A;
download keB public encryption key keB publish keB
with message m:
hash m to get d;
decrypt d using A with
private key kdB , getting s;
receive (m, s) signed message (m, s) transmit (m, s)
encrypt s using A with
public key keB , getting d′ ;
hash m to get d;
if d′ = d
ACCEPT
else
REJECT
3.3. DIGITAL SIGNATURES 73
Reading Response 3.3.3
Does all of this make sense? What was new or interesting, or what was old and unin-
teresting? Do you feel comfortable with how this new approach makes a secure dig-
ital signature scheme, because of its structure and the properties of a cryptographic
hash function, which is nevertheless much smaller than the naive digital signatures?
Code Task 3.3.3
Implement this new approach to making digital signatures that uses a hash function
for smaller signatures.
That is, make new Python functions signH(key, m) and
verifyH(pk, m, s) which have exactly the same inputs and outputs as
the sign(key, m) and verify(pk, m, s) from Code Task 3.3.1 and can
be used in the same ways as those original functions, but which use SHA256 to
make signatures which are always quite small.
74 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
3.4. Key management and the need for a robust PKI
We started this chapter by looking for a new approach which might solve a problem we
had with symmetric cryptosystems: while they are fast and secure, there are many, many
situations in which parties who have never met each other in real life, in a place and sit-
uation where they can share a high quality (meaning: large and hard for others to guess)
secret key, nevertheless want to communicate securely over the Internet. Asymmetric cryp-
tosystemss solved this problem by breaking a key into two pieces, one of which must be
kept secret (the private key) and one of which must be made public (the public key.
A significant issue with remains key management, though: the problem with Aragorn
and Bilbo never meeting in person to set up a shared secret key is that, basically, they
each have to trust each other’s websites where their public keys must be posted. If the
website is hacked, or even if the communications channel between one of the participants
and the other’s website is compromised, then they may end up using not the public key of
their intended message recipient, but rather a public key the hacker has substituted: this is
called a person-in-the-middle attack6, and it allows the hackers to read and even modify all
communications between the parties without them even knowing they have no security!
Graphically, the person-in-the-middle attack works like this:
Person-in-the-middle attack on asymmetric cryptosystems:
Alice Eve Bob
generate keys:
public keB , private kdB ;
intercept keB publish keB
generate keys:
public keE , private kdE ;
download keE , thinking it is keB keE , spoof origin
create cleartext message mA ;
using keE , encrypt mA to cA
and transmit cA cA , intercept
using kdE , decrypt cA
and recover cleartext mA ;
read mA , change to mE if desired
using keB , encrypt mE to cE
spoof origin cE receive cE , thinking it is cA
using kdB , decrypt cE
and recover cleartext mE
thinking it is mA from Alice
6Although usually a more gendered term than “person” is used.
3.4. KEY MANAGEMENT AND THE NEED FOR A ROBUST PKI 75
There is also a person-in-the-middle attack on digital signatures, based on similar ideas
of substituting Eve’s public key for Bob’s, which we leave to the reader to think through
for themself.
Reading Response 3.4.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel comfortable with how the person-in-the-middle attack
works, for communications? What about for digital signatures?
Bonus Task 3.4.1
Here’s another entirely non-coding Bonus Task:
Write out an explanation of the person-in-the-middle attack for digital signatures.
Include a diagram, like the diagram above called “Better [smaller] digital signatures
with hashing” crossed with the diagram above called “Person-in-the-middle attack
on asymmetric cryptosystems.”
Apparently, all of the wonders of asymmetric cryptoasymmetric cryptosystem will col-
lapse unless we can already be sure that the networks we use and the websites we visit are
never hacked. That is, unless there is some way we can have some sort of root of trust, a
public key connected to some real-world entity that we already implicitly trust, and whose
security practices we also trust, so that we can be sure that the private key is kept private.
Modern operating systems and browsers come with built-in roots of trust, which the
users don’t even know about, but which can be used to verify digital signatures signed by
that entity. Perhaps that entity will have done the work of meeting Jeff Bezos in real life,
and then will issue a signature on Jeff’s public key. Then if you want to use your credit card
on amazon.com, you can download Amazon’s public key along with the root of trust’s
signature on that key, which will enable you to trust that that the key which seems to be
Amazon’s is, in fact, Amazon’s.
These digital signatures on someone’s public key are usually called certificates and the
root of trust, in this context, is called a certificate authority.
The broad infrastructure of certificate authorities and public keys which can be trusted,
perhaps because of certificates, to be associated with certain real-world persons or organi-
zations is called a public-key infrastructure [PKI]. If the Internet had a robust, reliable
public-key infrastructure, then we will be able to have widespread secure communication
as well as legally binding (digitally signed) documents on the web. This, then, is the last
manifestation of a persistent issue throughout this book: key management is very hard, but
very important, as this necessity of a robust PKI makes clear yet again.
76 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
Reading Response 3.4.2
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel comfortable with the ideas of a root of trust (or certificate
authority), certificates, and, in general, a PKI?
3.5. CONCLUSIONS; CONSIDER THE BLOCKCHAIN 77
3.5. Conclusions; consider the blockchain
At first the idea of an asymmetric cryptosystem can seem a bit like magic. In fact, it
must be admitted that the security of all of such cryptosystems in wide use today is based
on assumptions that seem very reasonable but are not known with mathematical certainty:
for example, the security of RSA is based on the assumption that it is impossible to factor
large numbers in a reasonable amount of time (for a certain precise meaning of the word
“reasonable”).
(Interestingly, the best known attack on this particular assumption, using a quantum
computer, seemed for a long time like science fiction, but now seems like it definitely will
start to be deployed at least by entities such as nation-states which have sufficient resources.
Leave it to science fiction to be the best weapon against magic!)
How well all of the assumptions underlying asymmetric crypto stand the test of time,
and whether practical, quantum-resistant asymmetric cryptosystems can be designed and
implemented is a question of enormous importance in the next decade(s) of history of the
Internet. While that cliff-hanger is being resolved by hard-working technologists behind the
scenes, there are other exotic applications of asymmetric crypto which have been developed
and which offer amazing potential benefits ... some which are, indeed, almost certainly too
good to be true.
One example of this is something that has come to be widely discussed recently, called
a blockchain, which you have probably heard of as the architecture underlying the cryp-
tocurrency Bitcoin. If you worked in education, you would also have heard of blockchains
as, supposedly, the best place for high school diplomas, college transcripts, and profes-
sional certificates to be stored.
A blockchain starts with a data structure called a hash chain, which is basically just
a sequence of data blocks that can keep growing (potentially forever), with one additional
field in each block: the hash (under some cryptographic hash function) of the previous
block. Note that since that previous block contained the hash of the block which preceded
it, these hashes implicitly link all the blocks back to the very first block of the chain, which
is often called genesis block.
Graphically:
Genesis block: ha
ha Block 1: ha Block 2: sh
sh sh of
data of data of data ....
null
78 3. ASYMMETRIC [PUBLIC-KEY] CRYPTOSYSTEMS
The security properties of cryptographic hash functions given in Definition 3.3.2 give
hash chains a remarkable property: they are immutable, meaning they cannot be changed.
Specifically, what that means is that if a whole bunch of people (call them “the hobbits”)
are sharing copies of a hash chain, then someone else (call this person “Gollum”) cannot
come to them and pretend that they have a valid copy of the chain which is the same at the
beginning and the end, but has some extra blocks added in the middle. The reason this is
impossible is that the first block – say it’s block number 141592 near the end where Gollum
agrees with the hobbits would have the hash of two different “previous” blocks: the hobbits
would just have the same block they all agree is number 141591, while Gollum would have
that extra block he was trying to sneak into the chain ... and those two different pieces of
data cannot have the same hash (or, at least, it is very hard to figure out what data should
be in the block to make it have the same hash value).
Reading Response 3.5.1
Does all of this make sense? What was new or interesting, or what was old and
uninteresting? Do you feel comfortable with the ideas of a hash chain, and with why
it is immutable?
Several additional features make a hash chain into a blockchain. These include the fact
that parts of the data in the blocks must have valid digital signatures – a topic we have
discussed in this chapter! – and that the new blocks are the result of a consensus protocol.
Such protocols enable everyone who gets a copy of the blockchain to have confidence that
they have the same chain of data blocks as everyone else who is using that blockchain.
Consensus protocols are a topic in even more advanced cryptology classes: don’t stop
here, there is much more that is fun, useful, and powerful to learn about in this field!
Bibliography
[BB20] Charles H Bennett and Gilles Brassard, Quantum cryptography: Public key distribution and coin
tossing, arXiv preprint arXiv:2003.06557 (2020).
[Col14] David Cole, We Kill People Based On Metadata, The New York Review of Books 10 (2014), 2014.
[DH76] Whitfield Diffie and Martin Hellman, New directions in cryptography, IEEE transactions on Infor-
mation Theory 22 (1976), no. 6, 644–654.
[Hel02] M. E. Hellman, An overview of public key cryptography, IEEE Communications Magazine 40
(2002), no. 5, 42–49.
[Mar99] Leo Marks, Between Silk and Cyanide: A Codebreakers War, HarperCollins, 1999.
[PS04] Jonathan A Poritz and Morton Swimmer, Hash woes, Virus Bulletin (2004), 14–16.
[Sha45] Claude Shannon, A Mathematical Theory of Cryptography, Bell System Technical Memo MM 45-
110-02, https://www.iacr.org/museum/shannon/shannon45.pdf, Accessed: 22
February 2021.
[Sta02] Richard Stallman, Free Software, Free Society: Selected Essays of Richard M. Stallman, Lulu. com,
2002.
[WY05] Xiaoyun Wang and Hongbo Yu, How to break md5 and other hash functions, Advances in
Cryptology–EUROCRYPT 2005, Springer, 2005, pp. 19–35.
79
Index
[PKI], 75 Caesar cipher, 33
definition, 6
⊕ , bitwise XOR, 40 Caesar cryptosystem, 6
3DES, triple DES, 27 Caroll, Lewis (Charles Dodgson), 24
CBC mode block chaining, 40–42, 64, 65
Abbasid Caliphate, 24
certificate, 75
Adleman, Leonard, 51, 59
authority, 75
Advanced Encryption Standard [AES], 27, 35,
Charles Dodgson (Lewis Caroll), 24
36, 38, 39, 42, 43, 46, 61, 64, 65
cipher, 3
AES (Advanced Encryption Standard), 27, 35,
Cipher Block Chaining (CBC) mode block
36, 38, 39, 42, 43, 46, 61, 64, 65
chaining, 40–42, 64, 65
Al-Khwarizmi, 24
ciphertext, 3
Al-Kindi, 24
cipheterxt, 3
algebra, 24
cleartext, 3
algorithm, 24
collision resistance, 69, 70
Alice in Wonderland, 24
confidentiality, 2
analysis [αν άλυη, Greek root], 1
confusion, 28, 33, 36
asymmetric cryptosystem, 47–54, 59, 61, 66, 67,
consensus protocol, 78
74, 75, 77
crib, 44
asymmetric cryptosystems, 74
cribs, 45
asymmetric encryption, 61
cryptanalysis, 1
Augustus (Octavian), 6
crypto, 1
authentication, 3
Crypto [Python module], 51, 63
Between Silk and Cyanide: A Codebreaker’s Crypto.Cipher , 35, 41, 42, 59, 64, 65
War, 26 Crypto.Hash , 71
Bitcoin, 77 Crypto.PublicKey , 51
block chaining, 39, 41–43, 61 Crypto.Random , 63
Cipher Block Chaining (CBC) mode, 40–42, Crypto.PublicKey [Python module]
64, 65 RSA , 51–53, 55, 56, 58, 60, 68
Electronic Code Book (ECB) mode, 40, 41, 43 Crypto.Random [Python module]
block cipher, 27, 34, 35, 37–39, 61 get random bytes , 63, 64
block size, 34, 35, 37–39 cryptocurrency, 1, 77
blockchain, 77, 78 cryptographic hash function
blockchains, 1 seehash function, 79
brute-force attack, 11–13, 15, 20, 43, 49, 50, 69 cryptographic salt, 48, 50, 58, 60
81
82 INDEX
cryptography, 1 hash functions, 69, 70
cryptology, 1 Hayden, Michael, General, 30
cryptosystem, 1 House of Wisdom, 24
asymmetric, 47, 48, 74
private key, 48 immutable, 78
public key, 49 information leakage, 30
RSA, 51, 53–55, 57–61, 63, 68, 77 information security, 2
symmetric, 48, 74 information theoretically secure, 21
information theory, 30
Data Encryption Standard [DES], 27 initialization vector (IV), 39, 41, 42
decryption, 3, 51, 66, 67, 72 integrity, 2
decryption key [for an asymmetric cryptosystem], inverse functions, 66
48, 50, 52–55, 57, 61, 63, 67, 68, 74, 75 IV, initialization vector, 39, 41, 42
DES: Data Encryption Standard, 27
Diffie-Hellman key exchange, 47 Julius Caesar, 6
diffusion, 28, 30, 31, 33, 36, 39, 40, 42, 43, 60
diffusion, forward, 43 Kerckhoff’s Principle, 4
digital signature, 66–69, 72, 73, 75, 78 key, 4, 35, 43, 46, 48, 50–53, 55, 58–60, 63, 64,
digraph, 28 66
Dodgson, Charles (Lewis Caroll), 24 decryption [for an asymmetric cryptosystem],
drone strikes, 30 48, 50, 52–55, 57, 61, 63, 67, 68, 74, 75
distribution, 26, 61
ECB mode block chaining, 40, 41, 43 quantum, 47
Electronic Codebook (ECB) mode block encryption [for an asymmetric cryptosystem],
chaining, 40, 41, 43 48–50, 53–55, 61–63, 67, 68, 72, 74, 75
elliptic curve cryptosystem, 51 generation, 50–52, 67
encryption, 3, 51, 66 RSA, 51, 52
encryption key [for an asymmetric cryptosystem], management, 26, 61, 74, 75
48–50, 53–55, 61–63, 67, 68, 72, 74, 75 private, 48, 50, 52–55, 57, 61, 63, 67, 68, 74,
exclusive or, 25, 39, 40 75
public, 48–50, 53–55, 61–63, 67, 68, 72, 74, 75
Federal Information Processing Standards, US, session, 61, 63, 64
70 negotiation, 61
Fibonacci (Leonardo of Pisa), 24 size, 35, 43, 46, 50–53, 55, 58–60, 63
forward diffusion, 43 keyspace, 11, 45, 46
frequency table, 13 kryptos [κρυπτ oς, Greek root], 1
frequency table, relative, 14
Leo Marks, 26
General Michael Hayden, 30 Leonardo of Pisa (Fibonacci), 24
genesis block, 77 Lewis Caroll (Charles Dodgson), 24
get random bytes [Python method in Liber Abaci, 24
Crypto.Random module], 63, 64 logos [λóγoς, Greek root], 1
graph [γράϕω, Greek root], 1
Manuscript on Deciphering Cryptographic
hash chain, 77, 78 Messages, 24
hash function, 69–73, 77, 78 matplotlib.pyplot , 52
INDEX 83
md5, 70 public-key cryptosystem, 49, 66
Medici, 24 public-key infrastructure [PKI], 75
metadata, 30 Pythagorean Theorem, 15
mod, 7 Python, vii, 7–16, 20–22, 25, 27, 28, 31, 32,
modulo, 7 35–38, 41, 43, 51–56, 58, 59, 61, 63–65, 68,
monoalphabetic cryptosystem, 29, 45 71, 73
monograph, 28 Crypto module, 51, 63
Crypto.Cipher module, 35, 41, 42, 59,
National Institute of Standards and Technology 64, 65
[NIST], 27, 70 Crypto.Hash module, 71
National Institute of Standards and Technology, Crypto.PublicKey module, 51
[NIST], 70 Crypto.Random module, 63
National Security Agency, NSA, 30, 70
get random bytes method in
National Security Agency, US [NSA], 71
Crypto.Random module, 63, 64
NIST: National Institute of Standards and
PKCS1 OAEP class, 59, 60, 64
Technology, 27, 70
RSA class in Crypto.PublicKey
non-repudiation, 3
module, 51–53, 55, 56, 58, 60, 68
NSA, National Security Agency, 30, 70
SHA256 class, 71, 73
OAEP, Optimal Asymmetric Encryption Padding,
QKD, quantum key distribution, 47
58–60
quantum computer, 54, 59, 77
Octavian (Augustus), 6
quantum key distribution, [QKD], 47
one-time pad, 21, 33
quantum mechanics, 47, 54
Optimal Asymmetric Encryption Padding
[OAEP], 58–60 relative frequency table, 14
Renaissance, 24
padding, 37, 38 Rivest, Ron, 51, 59, 70
person-in-the-middle attack, 74, 75 root of trust, 75
pig, yellow, 17 ROT13, 6
PKCS #1, Public Key Cryptography Standard #1, RSA [Python class in Crypto.PublicKey
58, 59 module], 51–53, 55, 56, 58, 60, 68
PKCS1 OAEP [Python class], 59, 60, 64 RSA cryptosystem, 51, 53–55, 57–61, 63, 68, 77
PKI, public key infrastructure, 74 RSA Laboratories, 58
PKI, public-key infrastructure, 75
plaintext, 3 salt
polyalphabetic cryptosystem, 29, 45 cryptographic, 48, 50, 58, 60
pre-image resistance, 69, 70 second pre-image resistance, 69, 70
prime numbers, 54 security through obscurity, 4
private key, 48, 50, 52–55, 57, 61, 63, 67, 68, 74, session, 61, 63
75 key, 61
private-key cryptosystem, 48 negotiation, 61
public key, 48–50, 53–55, 61–63, 67, 68, 72, 74, SHA-1, 70
75 SHA-2, 71
Public Key Cryptography Standard #1, PKCS #1, SHA-256, 71
58, 59 SHA256, 71, 72
public key infrastructure, PKI, 74 SHA256 [Python class], 71, 73
84 INDEX
Shamir, Adi, 51, 59
Shannon, Claude, 28, 30, 31, 33, 36, 39, 40, 42,
43, 60
signature
digital, 66–69, 72, 73, 75, 78
verification, 66–69
stateful, 42
symmetric cryptosystem, 48, 50, 61, 74
The Book of Calculation, 24
The Compendious Book on Calculation by
Completion and Balancing, 24
traffic analysis, 30
trigraph, 28
triple DES, 27
verification algorithm [for a digital signature],
66–69
Vigenère cryptosystem, 18, 33
history, 24
“We kill people based on metadata.”, 30
XOR, 25, 39, 40
yfesdrype, 2