bitcoinbook/ch05.asciidoc

[[ch05_wallets]]
== Wallet Recovery

Creating pairs of private and public keys is a crucial part of allowing
Bitcoin wallets to receive and spend bitcoins.  But losing access to a
private key can make it impossible for anyone to ever spend the bitcoins
received to the corresponding public key.  Wallet and protocol
developers over the years have worked to design systems that allow users
to recover access to their bitcoins after a problem without compromising
security the rest of the time.

In this chapter, we'll examine some of the different methods employed by
wallets to prevent the loss of data from becoming a loss of money.
Some solutions have almost no downsides and are universally adopted by
modern wallets.  We'll simply recommend those solutions as best
practices.  Other solutions have both advantages and disadvantages,
leading different wallet authors to make different tradeoffs.
In those cases, we'll describe the various options available.

=== Independent Key Generation

((("wallets", "contents of")))Wallets for physical cash hold that cash,
so it's unsurprising that many people mistakenly believe that
bitcoin wallets contain bitcoins.  In fact, what many people call a
Bitcoin wallet--which we call a _wallet database_ to distinguish it
from wallet applications--contains only keys.  Those keys are associated
with bitcoins recorded on the blockchain.  By proving to Bitcoin full nodes that you
control the keys, you can can spend the associated bitcoins.

Simple wallet databases contain both the public keys to which bitcoins
are received and the private keys which allow creating the signatures
necessary to authorize spending those bitcoins.  Other wallets databases
may contain only public keys, or only some of the private keys necessary
to authorize a spending transaction.  Their wallet applications produce
the necessary signatures by working with external tools, such as
hardware signing devices or other wallets in a multi-signature scheme.

It's possible for a wallet application to independently generate each of
the wallet keys it later plans to use.  All early Bitcoin wallet applications did
this, but this required users back up the wallet database each time they
generated and distributed new keys, which could be as often as each time
they generated a new address to receive a new payment.  Failure to back
up the wallet database on time would lead to the user losing access to
any funds received to keys which had not been backed up.

For each independently-generated key, the user would need to back up
about 32 bytes, plus overhead.  Some users and wallet applications tried
to minimize the amount of data that needed to be backed up
by only using a single key.  Although that can be secure, it severely
reduces the privacy of that user and all of the people with whom they
transact.  People who valued their privacy and those of their peers
created new keypairs for each transaction, producing wallet databases
that could only reasonably be backed up using digital media.

[[Type0_wallet]]
[role="smallersixty"]
.Non-deterministic key generation: a collection of independently generated keys stored in a wallet database
image::images/mbc2_0501.png["Non-Deterministic Wallet"]

Modern wallet applications don't independently generate keys but instead
derive them from a single random seed using a repeatable (deterministic)
algorithm.

==== Deterministic Key Generation

A hash function will always produce the same output when given the same
input, but if the input is changed even slightly, the output will be
different.  If the function is cryptographically secure, nobody should
be able to predict the new output--not even if they know the new input.

This allows us to take one random value and transform it into a
practically unlimited number of seemingly-random values.  Even more
usefully, later using the same hash function with the same input
(called a _seed_) will produce the same seemingly-random values.

----
# Collect some entropy (randomness)
$ dd if=/dev/random count=1 status=none | sha256sum
f1cc3bc03ef51cb43ee7844460fa5049e779e7425a6349c8e89dfbb0fd97bb73  -

# Set our seed to the random value
$ seed=f1cc3bc03ef51cb43ee7844460fa5049e779e7425a6349c8e89dfbb0fd97bb73

# Deterministically generate derived values
$ for i in {0..2} ; do echo "$seed + $i" | sha256sum ; done
50b18e0bd9508310b8f699bad425efdf67d668cb2462b909fdb6b9bd2437beb3  -
a965dbcd901a9e3d66af11759e64a58d0ed5c6863e901dfda43adcd5f8c744f3  -
19580c97eb9048599f069472744e51ab2213f687d4720b0efc5bb344d624c3aa  -
----

If we use the derived values as our private keys, we can later generate
exactly those same private keys by using our seed value with the
algorithm we used before.  A user of deterministic key generation can
back up every key in their wallet by simply recording their seed and
a reference to the deterministic algorithm they used.  For example, even
if Alice has a million bitcoins received to a million different
addresses, all she needs to back up in order to later recover access to
those bitcoins is:

----
f1cc 3bc0 3ef5 1cb4 3ee7 8444 60fa 5049
e779 e742 5a63 49c8 e89d fbb0 fd97 bb73
----

A logical diagram of basic sequential deterministic key generation is
shown in <<Type1_wallet>>.  However, modern wallet applications have a
more clever way of accomplishing this that allows public keys to be
derived separately from their corresponding private keys, making it
possible to store private keys more securely than public keys.

[[Type1_wallet]]
[role="smallersixty"]
.Deterministic key generation: a deterministic sequence of keys derived from a seed for a wallet database
image::images/mbc2_0502.png["Deterministic Wallet"]

==== Public Child Key Derivation

In <<public_key_derivation>>, we learned how to create a public key from a private key
using Elliptic Curve Cryptography (ECC).  Although operations on an
elliptic curve are not intuitive, they are analogous to the addition,
subtraction, and multiplication operations we use in regular
arithmetic.  In other words, it's possible to add or subtract from a
public key, or to multiply it.  Consider the equation we used for
generating a public key (K) from a private key (k) using the generator
point (G):

----
K == k * G
----

It's possible to create a derived keypair, called a child keypair, by
simply adding the same value to both sides of the equation:

----
K + (123 * G) == (k + 123) * G
----

An interesting consequence of this is that adding `123` to the public
key can be done using entirely public information.  For example, Alice
generates public key K and gives it to Bob.  Bob doesn't know the
private key, but he does know the global constant G, so he can add any
value to the public key to produce a derived public child key.  If he
then tells Alice the value he added to the public key, she can add the
same value to the private key, producing a derived private child key
that corresponds to the public child key Bob created.

In other words, it's possible to create child public keys even if you
don't know anything about the parent private key.  The value added to a
public key is known as a _key tweak._  If a deterministic algorithm is
used for generating the key tweaks, then it's possible to for someone
who doesn't know the private key to create an essentially unlimited
sequence of public child keys from a single public parent key. The
person who controls the private parent key can then use the same key
tweaks to create all the corresponding private child keys.

This technique is commonly used is to separate wallet application
frontends (which don't require private keys) from signing operations
(which do require private keys).  For example, Alice's frontend
distributes her public keys to people wanting to pay her.  Later, when
she wants to spend the received money, she can provide the key tweaks
she used to a _hardware signing device_ (sometimes confusingly called a
_hardware wallet_) which securely stores her original private key.  The
hardware signer uses the tweaks to derive the necessary child private
keys and uses them to sign the transactions, returning the signed
transactions to the less-secure frontend for broadcast to the Bitcoin
network.

Public child key derivation can produce a linear sequence of keys
similar to the previously seen <<Type1_wallet>>, but modern wallets
applications use one more trick to provide a tree of keys instead a
single sequence.

[[hd_wallets]]
==== HD Wallets (BIP32/BIP44)

((("wallets", "types of", "hierarchical deterministic (HD)
wallets")))((("hierarchical deterministic (HD) wallets")))((("bitcoin
improvement proposals", "Hierarchical Deterministic Wallets
(BIP32/BIP44)")))Deterministic wallets were developed to make it easy
to derive many keys from a single "seed." The most advanced form of
deterministic wallets is the HD wallet defined by the BIP32 standard.
HD wallets contain keys derived in a tree structure, such that a parent
key can derive a sequence of children keys, each of which can derive a
sequence of grandchildren keys, and so on, to an infinite depth. This
tree structure is illustrated in <<Type2_wallet>>.

[[Type2_wallet]]
.Type-2 HD wallet: a tree of keys generated from a single seed
image::images/mbc2_0503.png["HD wallet"]

HD wallets offer two major advantages over random (nondeterministic)
keys. First, the tree structure can be used to express additional
organizational meaning, such as when a specific branch of subkeys is
used to receive incoming payments and a different branch is used to
receive change from outgoing payments. Branches of keys can also be used
in corporate settings, allocating different branches to departments,
subsidiaries, specific functions, or accounting categories.

The second advantage of HD wallets is that users can create a sequence
of public keys without having access to the corresponding private keys.
This allows HD wallets to be used on an insecure server or in a
receive-only capacity, issuing a different public key for each
transaction. The public keys do not need to be preloaded or derived in
advance, yet the server doesn't have the private keys that can spend the
funds.

==== Seeds and Mnemonic Codes (BIP39)

((("wallets", "technology of", "seeds and mnemonic codes")))((("mnemonic
code words")))((("bitcoin improvement proposals", "Mnemonic Code Words
(BIP39)")))HD wallets are a very powerful mechanism for managing many
keys and addresses. They are even more useful if they are combined with
a standardized way of creating seeds from a sequence of English words
that are easy to transcribe, export, and import across wallets. This is
known as a _mnemonic_ and the standard is defined by BIP39. Today, most
bitcoin wallets (as well as wallets for other cryptocurrencies) use this
standard and can import and export seeds for backup and recovery using
interoperable mnemonics.

Let's look at this from a practical perspective. Which of the following
seeds is easier to transcribe, record on paper, read without error,
export, and import into another wallet?

.A seed for an deterministic wallet, in hex
----
0C1E24E5917779D297E14D45F14E1A1A
----

.A seed for an deterministic wallet, from a 12-word mnemonic
----
army van defense carry jealous true
garbage claim echo media make crunch
----

==== Wallet Best Practices

((("wallets", "best practices for")))((("bitcoin improvement proposals",
"Multipurpose HD Wallet Structure (BIP43)")))As bitcoin wallet
technology has matured, certain common industry standards have emerged
that make bitcoin wallets broadly interoperable, easy to use, secure,
and flexible. These common standards are:

* Mnemonic code words, based on BIP39
* HD wallets, based on BIP32
* Multipurpose HD wallet structure, based on BIP43
* Multicurrency and multiaccount wallets, based on BIP44

These standards may change or may become obsolete by future
developments, but for now they form a set of interlocking technologies
that have become the de facto wallet standard for bitcoin.

The standards have been adopted by a broad range of software and
hardware bitcoin wallets, making all these wallets interoperable. A user
can export a mnemonic generated on one of these wallets and import it in
another wallet, recovering all transactions, keys, and addresses.

((("hardware wallets")))((("hardware wallets", see="also wallets")))Some
example of software wallets supporting these standards include (listed
alphabetically) Breadwallet, Copay, Multibit HD, and Mycelium. Examples
of hardware wallets supporting these standards include (listed
alphabetically) Keepkey, Ledger, and Trezor.

The following sections examine each of these technologies in detail.

[TIP]
====
If you are implementing a bitcoin wallet, it should be built as a HD
wallet, with a seed encoded as mnemonic code for backup, following the
BIP32, BIP39, BIP43, and BIP44 standards, as described in the
following sections.
====

==== Using a Bitcoin Wallet

((("wallets", "using bitcoin wallets")))In <<user-stories>> we
introduced Gabriel, ((("use cases", "web store", id="gabrielfive")))an
enterprising young teenager in Rio de Janeiro, who is running a simple
web store that sells bitcoin-branded t-shirts, coffee mugs, and
stickers.

((("wallets", "types of", "hardware wallets")))Gabriel uses a Trezor
bitcoin hardware wallet (<<a_trezor_device>>) to securely manage his
bitcoin. The Trezor is a simple USB device with two buttons that stores
keys (in the form of an HD wallet) and signs transactions. Trezor
wallets implement all the industry standards discussed in this chapter,
so Gabriel is not reliant on any proprietary technology or single vendor
solution.

[[a_trezor_device]]
.A Trezor device: a bitcoin HD wallet in hardware
image::images/mbc2_0504.png[alt]

When Gabriel used the Trezor for the first time, the device generated a
mnemonic and seed from a built-in hardware random number generator.
During this initialization phase, the wallet displayed a numbered
sequence of words, one by one, on the screen (see
<<trezor_mnemonic_display>>).

[[trezor_mnemonic_display]]
.Trezor displaying one of the mnemonic words
image::images/mbc2_0505.png["Trezor wallet display of mnemonic word"]

By writing down this mnemonic, Gabriel created a backup (see
<<mnemonic_paper_backup>>) that can be used for recovery in the case of
loss or damage to the Trezor device. This mnemonic can be used for
recovery in a new Trezor or in any one of the many compatible software
or hardware wallets. Note that the sequence of words is important, so
mnemonic paper backups have numbered spaces for each word. Gabriel had
to carefully record each word in the numbered space to preserve the
correct sequence.

[[mnemonic_paper_backup]]
.Gabriel's paper backup of the mnemonic
[cols="<1,^50,<1,^50", width="80%"]
|===
|*1.*| _army_ |*7.*| _garbage_
|*2.*| _van_ |*8.*| _claim_
|*3.*| _defense_ |*9.*| _echo_
|*4.*| _carry_ |*10.*| _media_
|*5.*| _jealous_ |*11.*| _make_
|*6.*| _true_ |*12.*| _crunch_
|===

[NOTE]
====
A 12-word mnemonic is shown in <<mnemonic_paper_backup>>, for
simplicity. In fact, most hardware wallets generate a more secure
24-word mnemonic. The mnemonic is used in exactly the same way,
regardless of length.
====

For the first implementation of his web store, Gabriel uses a single
Bitcoin address, generated on his Trezor device. This single address is
used by all customers for all orders. As we will see, this approach has
some drawbacks and can be improved upon with an HD wallet.((("",
startref="gabrielfive")))

=== Wallet Technology Details

Let's now examine each of the important industry standards that are used
by many bitcoin wallets in detail.

[[mnemonic_code_words]]
==== Mnemonic Code Words (BIP39)

((("wallets", "technology of", "mnemonic code words")))((("mnemonic code
words", id="mnemonic05")))((("bitcoin improvement proposals", "Mnemonic
Code Words (BIP39)", id="BIP3905")))Mnemonic code words are word
sequences that represent (encode) a random number used as a seed to
derive a deterministic wallet. The sequence of words is sufficient to
re-create the seed and from there re-create the wallet and all the
derived keys. A wallet application that implements deterministic wallets
with mnemonic words will show the user a sequence of 12 to 24 words when
first creating a wallet. That sequence of words is the wallet backup and
can be used to recover and re-create all the keys in the same or any
compatible wallet application. Mnemonic words make it easier for users
to back up wallets because they are easy to read and correctly
transcribe, as compared to a random sequence of numbers.

[TIP]
====
((("brainwallets")))Mnemonic words are often confused with
"brainwallets." They are not the same. The primary difference is that a
brainwallet consists of words chosen by the user, whereas mnemonic words
are created randomly by the wallet and presented to the user. This
important difference makes mnemonic words much more secure, because
humans are very poor sources of randomness.
====

Mnemonic codes are defined in BIP39 (see <<appdxbitcoinimpproposals>>).
Note that BIP39 is one implementation of a mnemonic code standard.
((("Electrum wallet", seealso="wallets")))There is a different standard,
with a different set of words, used by the Electrum wallet and predating
BIP39. BIP39 was proposed by the company behind the Trezor hardware
wallet and is incompatible with Electrum's implementation. However,
BIP39 has now achieved broad industry support across dozens of
interoperable implementations and should be considered the de facto
industry standard.

BIP39 defines the creation of a mnemonic code and seed, which we
describe here in nine steps. For clarity, the process is split into two
parts: steps 1 through 6 are shown in <<generating_mnemonic_words>> and
steps 7 through 9 are shown in <<mnemonic_to_seed>>.

[[generating_mnemonic_words]]
===== Generating mnemonic words

Mnemonic words are generated automatically by the wallet using the
standardized process defined in BIP39. The wallet starts from a source
of entropy, adds a checksum, and then maps the entropy to a word list:

1. Create a random sequence (entropy) of 128 to 256 bits.

2. Create a checksum of the random sequence by taking the first
(entropy-length/32) bits of its SHA256 hash.

3. Add the checksum to the end of the random sequence.

4. Split the result into 11-bit length segments.

5. Map each 11-bit value to a word from the predefined dictionary of
2048 words.

6. The mnemonic code is the sequence of words.

<<generating_entropy_and_encoding>> shows how entropy is used to
generate mnemonic words.

[[generating_entropy_and_encoding]]
[role="smallerseventy"]
.Generating entropy and encoding as mnemonic words
image::images/mbc2_0506.png["Generating entropy and encoding as mnemonic words"]

<<table_4-5>> shows the relationship between the size of the entropy
data and the length of mnemonic codes in words.

[[table_4-5]]
.Mnemonic codes: entropy and word length
[options="header"]
|=======
|Entropy (bits) | Checksum (bits) | Entropy *+* checksum (bits) | Mnemonic length (words)
| 128 | 4 | 132 | 12
| 160 | 5 | 165 | 15
| 192 | 6 | 198 | 18
| 224 | 7 | 231 | 21
| 256 | 8 | 264 | 24
|=======

[[mnemonic_to_seed]]
===== From mnemonic to seed

((("key-stretching function")))((("PBKDF2 function")))The mnemonic words
represent entropy with a length of 128 to 256 bits. The entropy is then
used to derive a longer (512-bit) seed through the use of the
key-stretching function PBKDF2. The seed produced is then used to build
a deterministic wallet and derive its keys.

((("salts")))((("passphrases")))The key-stretching function takes two
parameters: the mnemonic and a _salt_. The purpose of a salt in a
key-stretching function is to make it difficult to build a lookup table
enabling a brute-force attack. In the BIP39 standard, the salt has
another purpose&#x2014;it allows the introduction of a passphrase that
serves as an additional security factor protecting the seed, as we will
describe in more detail in <<mnemonic_passphrase>>.

The process described in steps 7 through 9 continues from the process
described previously in <<generating_mnemonic_words>>:

++++
<ol start="7">
  <li>The first parameter to the PBKDF2 key-stretching function is the
  <em>mnemonic</em> produced from step 6.</li>

  <li>The second parameter to the PBKDF2 key-stretching function is a
  <em>salt</em>. The salt is composed of the string constant
  "<code>mnemonic</code>" concatenated with an optional user-supplied
  passphrase string.</li>

  <li>PBKDF2 stretches the mnemonic and salt parameters using 2048
  rounds of hashing with the HMAC-SHA512 algorithm, producing a 512-bit
  value as its final output. That 512-bit value is the seed.</li>
</ol>
++++

<<fig_5_7>> shows how a mnemonic is used to generate a seed.

[[fig_5_7]]
.From mnemonic to seed
image::images/mbc2_0507.png["From mnemonic to seed"]

[TIP]
====
The key-stretching function, with its 2048 rounds of hashing, is a very
effective protection against brute-force attacks against the mnemonic or
the passphrase. It makes it extremely costly (in computation) to try
more than a few thousand passphrase and mnemonic combinations, while the
number of possible derived seeds is vast (2^512^).
====

Tables pass:[<a data-type="xref" href="#mnemonic_128_no_pass"
data-xrefstyle="select: labelnumber">#mnemonic_128_no_pass</a>],
pass:[<a data-type="xref" href="#mnemonic_128_w_pass"
data-xrefstyle="select: labelnumber">#mnemonic_128_w_pass</a>], and
pass:[<a data-type="xref" href="#mnemonic_256_no_pass"
data-xrefstyle="select: labelnumber">#mnemonic_256_no_pass</a>] show
some examples of mnemonic codes and the seeds they produce (without any
passphrase).

[[mnemonic_128_no_pass]]
.128-bit entropy mnemonic code, no passphrase, resulting seed
[cols="h,"]
|=======
| *Entropy input (128 bits)*| +0c1e24e5917779d297e14d45f14e1a1a+
| *Mnemonic (12 words)* | +army van defense carry jealous true garbage claim echo media make crunch+
| *Passphrase*| (none)
| *Seed  (512 bits)* | +5b56c417303faa3fcba7e57400e120a0ca83ec5a4fc9ffba757fbe63fbd77a89a1a3be4c67196f57c39+
+a88b76373733891bfaba16ed27a813ceed498804c0570+
|=======

[[mnemonic_128_w_pass]]
.128-bit entropy mnemonic code, with passphrase, resulting seed
[cols="h,"]
|=======
| *Entropy input (128 bits)*| +0c1e24e5917779d297e14d45f14e1a1a+
| *Mnemonic (12 words)* | +army van defense carry jealous true garbage claim echo media make crunch+
| *Passphrase*| SuperDuperSecret
| *Seed  (512 bits)* | +3b5df16df2157104cfdd22830162a5e170c0161653e3afe6c88defeefb0818c793dbb28ab3ab091897d0+
+715861dc8a18358f80b79d49acf64142ae57037d1d54+
|=======


[[mnemonic_256_no_pass]]
.256-bit entropy mnemonic code, no passphrase, resulting seed
[cols="h,"]
|=======
| *Entropy input (256 bits)* | +2041546864449caff939d32d574753fe684d3c947c3346713dd8423e74abcf8c+
| *Mnemonic (24 words)* | +cake apple borrow silk endorse fitness top denial coil riot stay wolf
luggage oxygen faint major edit measure invite love trap field dilemma oblige+
| *Passphrase*| (none)
| *Seed (512 bits)* | +3269bce2674acbd188d4f120072b13b088a0ecf87c6e4cae41657a0bb78f5315b33b3a04356e53d062e5+
+5f1e0deaa082df8d487381379df848a6ad7e98798404+
|=======

[[mnemonic_passphrase]]
===== Optional passphrase in BIP39

((("passphrases")))The BIP39 standard allows the use of an optional
passphrase in the derivation of the seed. If no passphrase is used, the
mnemonic is stretched with a salt consisting of the constant string
+"mnemonic"+, producing a specific 512-bit seed from any given mnemonic.
If a passphrase is used, the stretching function produces a _different_
seed from that same mnemonic. In fact, given a single mnemonic, every
possible passphrase leads to a different seed. Essentially, there is no
"wrong" passphrase. All passphrases are valid and they all lead to
different seeds, forming a vast set of possible uninitialized wallets.
The set of possible wallets is so large (2^512^) that there is no
practical possibility of brute-forcing or accidentally guessing one that
is in use.

[TIP]
====
There are no "wrong" passphrases in BIP39. Every passphrase leads to
some wallet, which unless previously used will be empty.
====

The optional passphrase creates two important features:

- A second factor (something memorized) that makes a mnemonic useless on
  its own, protecting mnemonic backups from compromise by a thief.

- A form of plausible deniability or "duress wallet," where a chosen
  passphrase leads to a wallet with a small amount of funds used to
  distract an attacker from the "real" wallet that contains the majority
  of funds.

However, it is important to note that the use of a passphrase also introduces the risk of loss:

* If the wallet owner is incapacitated or dead and no one else knows the passphrase, the seed is useless and all the funds stored in the wallet are lost forever.

* Conversely, if the owner backs up the passphrase in the same place as the seed, it defeats the purpose of a second factor.

While passphrases are very useful, they should only be used in
combination with a carefully planned process for backup and recovery,
considering the possibility of surviving the owner and allowing his or
her family to recover the cryptocurrency estate.

===== Working with mnemonic codes

BIP39 is implemented as a library in many different programming
languages:

https://github.com/trezor/python-mnemonic[python-mnemonic]:: The
reference implementation of the standard by the SatoshiLabs team that
proposed BIP39, in Python

https://github.com/bitcoinjs/bip39[bitcoinjs/bip39]:: An implementation
of BIP39, as part of the popular bitcoinJS framework, in JavaScript

https://github.com/libbitcoin/libbitcoin/blob/master/src/wallet/mnemonic.cpp[libbitcoin/mnemonic]::
An implementation of BIP39, as part of the popular Libbitcoin
framework, in pass:[<span class="keep-together">C++</span>]

There is also a BIP39 generator implemented in a standalone webpage,
which is extremely useful for testing and experimentation.
<<a_bip39_generator_as_a_standalone_web_page>> shows a standalone web
page that generates mnemonics, seeds, and extended private keys.

[[a_bip39_generator_as_a_standalone_web_page]]
.A BIP39 generator as a standalone web page
image::images/mbc2_0508.png["BIP39 generator web-page"]

((("", startref="mnemonic05")))((("", startref="BIP3905")))The page
(https://iancoleman.github.io/bip39/) can be used offline in a browser,
or accessed online.

==== Creating an HD Wallet from the Seed

((("wallets", "technology of", "creating HD wallets from root
seed")))((("root seeds")))((("hierarchical deterministic (HD)
wallets")))HD wallets are created from a single _root seed_, which is a
128-, 256-, or 512-bit random number. Most commonly, this seed is
generated from a _mnemonic_ as detailed in the previous section.

Every key in the HD wallet is deterministically derived from this root
seed, which makes it possible to re-create the entire HD wallet from
that seed in any compatible HD wallet. This makes it easy to back up,
restore, export, and import HD wallets containing thousands or even
millions of keys by simply transferring only the mnemonic that the root
seed is derived from.

The process of creating the master keys and master chain code for an HD
wallet is shown in <<HDWalletFromSeed>>.

[[HDWalletFromSeed]]
.Creating master keys and chain code from a root seed
image::images/mbc2_0509.png["HDWalletFromRootSeed"]

The root seed is input into the HMAC-SHA512 algorithm and the resulting
hash is used to create a _master private key_ (m) and a _master chain
code_ (c).

The master private key (m) then generates a corresponding master public
key (M) using the normal elliptic curve multiplication process +m * G+
that we saw in <<pubkey>>.

The chain code (c) is used to introduce entropy in the function that
creates child keys from parent keys, as we will see in the next section.

===== Private child key derivation

((("child key derivation (CKD)")))((("public and private keys", "child
key derivation (CKD)")))HD wallets use a _child key derivation_ (CKD)
function to derive child keys from parent keys.

The child key derivation functions are based on a one-way hash function
that combines:

* A parent private or public key (ECDSA uncompressed key)
* A seed called a chain code (256 bits)
* An index number (32 bits)

The chain code is used to introduce deterministic random data to the
process, so that knowing the index and a child key is not sufficient to
derive other child keys. Knowing a child key does not make it possible
to find its siblings, unless you also have the chain code. The initial
chain code seed (at the root of the tree) is made from the seed, while
subsequent child chain codes are derived from each parent chain code.

These three items (parent key, chain code, and index) are combined and
hashed to generate children keys, as follows.

The parent public key, chain code, and the index number are combined and
hashed with the HMAC-SHA512 algorithm to produce a 512-bit hash. This
512-bit hash is split into two 256-bit halves. The right-half 256 bits
of the hash output become the chain code for the child. The left-half
256 bits of the hash are added to the parent private key to produce the
child private key. In <<CKDpriv>>, we see this illustrated with the
index set to 0 to produce the "zero" (first by index) child of the
parent.

[[CKDpriv]]
.Extending a parent private key to create a child private key
image::images/mbc2_0510.png["ChildPrivateDerivation"]

Changing the index allows us to extend the parent and create the other
children in the sequence, e.g., Child 0, Child 1, Child 2, etc. Each
parent key can have 2,147,483,647 (2^31^) children (2^31^ is half of the
entire 2^32^ range available because the other half is reserved for a
special type of derivation we will talk about later in this chapter).

Repeating the process one level down the tree, each child can in turn
become a parent and create its own children, in an infinite number of
generations.

===== Using derived child keys

Child private keys are indistinguishable from nondeterministic (random)
keys. Because the derivation function is a one-way function, the child
key cannot be used to find the parent key. The child key also cannot be
used to find any siblings. If you have the n~th~ child, you cannot find
its siblings, such as the n&#x2013;1 child or the n+1 child, or any
other children that are part of the sequence. Only the parent key and
chain code can derive all the children. Without the child chain code,
the child key cannot be used to derive any grandchildren either. You
need both the child private key and the child chain code to start a new
branch and derive grandchildren.

So what can the child private key be used for on its own? It can be used
to make a public key and a Bitcoin address. Then, it can be used to sign
transactions to spend anything paid to that address.

[TIP]
====
A child private key, the corresponding public key, and the Bitcoin
address are all indistinguishable from keys and addresses created
randomly. The fact that they are part of a sequence is not visible
outside of the HD wallet function that created them. Once created, they
operate exactly as "normal" keys.
====

===== Extended keys

((("public and private keys", "extended keys")))((("extended keys")))As
we saw earlier, the key derivation function can be used to create
children at any level of the tree, based on the three inputs: a key, a
chain code, and the index of the desired child. The two essential
ingredients are the key and chain code, and combined these are called an
_extended key_. The term "extended key" could also be thought of as
"extensible key" because such a key can be used to derive children.

Extended keys are stored and represented simply as the concatenation of
the 256-bit key and 256-bit chain code into a 512-bit sequence. There
are two types of extended keys. An extended private key is the
combination of a private key and chain code and can be used to derive
child private keys (and from them, child public keys). An extended
public key is a public key and chain code, which can be used to create
child public keys (_public only_), as described in
<<public_key_derivation>>.

Think of an extended key as the root of a branch in the tree structure
of the HD wallet. With the root of the branch, you can derive the rest
of the branch. The extended private key can create a complete branch,
whereas the extended public key can _only_ create a branch of public
keys.

[TIP]
====
An extended key consists of a private or public key and chain code. An
extended key can create children, generating its own branch in the tree
structure. Sharing an extended key gives access to the entire branch.
====

Extended keys are encoded using Base58Check, to easily export and import
between different BIP32&#x2013;compatible wallets. The Base58Check
coding for extended keys uses a special version number that results in
the prefix "xprv" and "xpub" when encoded in Base58 characters to make
them easily recognizable. Because the extended key is 512 or 513 bits,
it is also much longer than other Base58Check-encoded strings we have
seen previously.

Here's an example of an extended _private_ key, encoded in Base58Check:

----
xprv9tyUQV64JT5qs3RSTJkXCWKMyUgoQp7F3hA1xzG6ZGu6u6Q9VMNjGr67Lctvy5P8oyaYAL9CAWrUE9i6GoNMKUga5biW6Hx4tws2six3b9c
----

Here's the corresponding extended _public_ key, encoded in Base58Check:

----
xpub67xpozcx8pe95XVuZLHXZeG6XWXHpGq6Qv5cmNfi7cS5mtjJ2tgypeQbBs2UAR6KECeeMVKZBPLrtJunSDMstweyLXhRgPxdp14sk9tJPW9
----

[[public__child_key_derivation]]
===== Public child key derivation

((("public and private keys", "public child key derivation")))As
mentioned  previously, a very useful characteristic of HD wallets is the
ability to derive public child keys from public parent keys, _without_
having the private keys. This gives us two ways to derive a child public
key: either from the child private key, or directly from the parent
public key.

An extended public key can be used, therefore, to derive all of the
_public_ keys (and only the public keys) in that branch of the HD wallet
structure.

This shortcut can be used to create very secure public key&#x2013;only
deployments where a server or application has a copy of an extended
public key and no private keys whatsoever. That kind of deployment can
produce an infinite number of public keys and Bitcoin addresses, but
cannot spend any of the money sent to those addresses. Meanwhile, on
another, more secure server, the extended private key can derive all the
corresponding private keys to sign transactions and spend the money.

One common application of this solution is to install an extended public
key on a web server that serves an ecommerce application. The web server
can use the public key derivation function to create a new Bitcoin
address for every transaction (e.g., for a customer shopping cart). The
web server will not have any private keys that would be vulnerable to
theft. Without HD wallets, the only way to do this is to generate
thousands of Bitcoin addresses on a separate secure server and then
preload them on the ecommerce server. That approach is cumbersome and
requires constant maintenance to ensure that the ecommerce server
doesn't "run out" of keys.

((("cold storage")))((("storage", "cold storage")))((("hardware
wallets")))Another common application of this solution is for
cold-storage or hardware wallets. In that scenario, the extended private
key can be stored on a paper wallet or hardware device (such as a Trezor
hardware wallet), while the extended public key can be kept online. The
user can create "receive" addresses at will, while the private keys are
safely stored offline. To spend the funds, the user can use the extended
private key on an offline signing Bitcoin client or sign transactions on
the hardware wallet device (e.g., Trezor). <<CKDpub>> illustrates the
mechanism for extending a parent public key to derive child public keys.

[[CKDpub]]
.Extending a parent public key to create a child public key
image::images/mbc2_0511.png["ChildPublicDerivation"]

==== Using an Extended Public Key on a Web Store

((("wallets", "technology of", "using extended public keys on web
stores")))Let's see how HD wallets are used by continuing our story with
Gabriel's web store.((("use cases", "web store", id="gabrielfivetwo")))

Gabriel first set up his web store as a hobby, based on a simple hosted
Wordpress page. His store was quite basic with only a few pages and an
order form with a single bitcoin address.

Gabriel used the first bitcoin address generated by his Trezor device as
the main bitcoin address for his store. This way, all incoming payments
would be paid to an address controlled by his Trezor hardware wallet.

Customers would submit an order using the form and send payment to
Gabriel's published bitcoin address, triggering an email with the order
details for Gabriel to process. With just a few orders each week, this
system worked well enough.

However, the little web store became quite successful and attracted many
orders from the local community. Soon, Gabriel was overwhelmed. With all
the orders paying the same address, it became difficult to correctly
match orders and transactions, especially when multiple orders for the
same amount came in close together.

Gabriel's HD wallet offers a much better solution through the ability to
derive public child keys without knowing the private keys. Gabriel can
load an extended public key (xpub) on his website, which can be used to
derive a unique address for every customer order. Gabriel can spend the
funds from his Trezor, but the xpub loaded on the website can only
generate addresses and receive funds. This feature of HD wallets is a
great security feature. Gabriel's website does not contain any private
keys and therefore does not need high levels of security.

To export the xpub, Gabriel uses the web-based software in conjunction
with the Trezor hardware wallet. The Trezor device must be plugged in
for the public keys to be exported. Note that hardware wallets will
never export private keys&#x2014;those always remain on the device.
<<export_xpub>> shows the web interface Gabriel uses to export the xpub.

[[export_xpub]]
.Exporting an xpub from a Trezor hardware wallet
image::images/mbc2_0512.png["Exporting the xpub from the Trezor"]

Gabriel copies the xpub to his web store's bitcoin shop software. He
uses _Mycelium Gear_, which is an open source web-store plugin for a
variety of web hosting and content platforms. Mycelium Gear uses the
xpub to generate a unique address for every purchase. ((("",
startref="gabrielfivetwo")))

===== Hardened child key derivation

((("public and private keys", "hardened child key
derivation")))((("hardened derivation")))The ability to derive a branch
of public keys from an xpub is very useful, but it comes with a
potential risk. Access to an xpub does not give access to child private
keys. However, because the xpub contains the chain code, if a child
private key is known, or somehow leaked, it can be used with the chain
code to derive all the other child private keys. A single leaked child
private key, together with a parent chain code, reveals all the private
keys of all the children. Worse, the child private key together with a
parent chain code can be used to deduce the parent private key.

To counter this risk, HD wallets use an alternative derivation function
called _hardened derivation_, which "breaks" the relationship between
parent public key and child chain code. The hardened derivation function
uses the parent private key to derive the child chain code, instead of
the parent public key. This creates a "firewall" in the parent/child
sequence, with a chain code that cannot be used to compromise a parent
or sibling private key. The hardened derivation function looks almost
identical to the normal child private key derivation, except that the
parent private key is used as input to the hash function, instead of the
parent public key, as shown in the diagram in <<CKDprime>>.

[[CKDprime]]
.Hardened derivation of a child key; omits the parent public key
image::images/mbc2_0513.png["ChildHardPrivateDerivation"]

[role="pagebreak-before"]
When the hardened private derivation function is used, the resulting
child private key and chain code are completely different from what
would result from the normal derivation function. The resulting "branch"
of keys can be used to produce extended public keys that are not
vulnerable, because the chain code they contain cannot be exploited to
reveal any private keys. Hardened derivation is therefore used to create
a "gap" in the tree above the level where extended public keys are used.

In simple terms, if you want to use the convenience of an xpub to derive
branches of public keys, without exposing yourself to the risk of a
leaked chain code, you should derive it from a hardened parent, rather
than a normal parent. As a best practice, the level-1 children of the
master keys are always derived through the hardened derivation, to
prevent compromise of the master keys.

===== Index numbers for normal and hardened derivation

The index number used in the derivation function is a 32-bit integer. To
easily distinguish between keys derived through the normal derivation
function versus keys derived through hardened derivation, this index
number is split into two ranges. Index numbers between 0 and
2^31^&#x2013;1 (0x0 to 0x7FFFFFFF) are used _only_ for normal
derivation. Index numbers between 2^31^ and 2^32^&#x2013;1 (0x80000000
to 0xFFFFFFFF) are used _only_ for hardened derivation. Therefore, if
the index number is less than 2^31^, the child is normal, whereas if the
index number is equal or above 2^31^, the child is hardened.

To make the index number easier to read and display, the index number
for hardened children is displayed starting from zero, but with a prime
symbol. The first normal child key is therefore displayed as 0, whereas
the first hardened child (index 0x80000000) is displayed as 0++&#x27;++.
In sequence then, the second hardened key would have index 0x80000001
and would be displayed as 1++&#x27;++, and so on. When you see an HD
wallet index i++&#x27;++, that means 2^31^+i.

===== HD wallet key identifier (path)

((("hierarchical deterministic (HD) wallets")))Keys in an HD wallet are
identified using a "path" naming convention, with each level of the tree
separated by a slash (/) character (see <<table_4-8>>). Private keys
derived from the master private key start with "m." Public keys derived
from the master public key start with "M." Therefore, the first child
private key of the master private key is m/0. The first child public key
is M/0. The second grandchild of the first child is m/0/1, and so on.

The "ancestry" of a key is read from right to left, until you reach the
master key from which it was derived. For example, identifier m/x/y/z
describes the key that is the z-th child of key m/x/y, which is the y-th
child of key m/x, which is the x-th child of m.

[[table_4-8]]
.HD wallet path examples
[options="header"]
|=======
|HD path | Key described
| m/0 | The first (0) child private key from the master private key (m)
| m/0/0 | The first grandchild private key from the first child (m/0)
| m/0'/0 | The first normal grandchild from the first _hardened_ child (m/0')
| m/1/0 | The first grandchild private key from the second child (m/1)
| M/23/17/0/0 | The first great-great-grandchild public key from the first great-grandchild from the 18th grandchild from the 24th child
|=======

===== Navigating the HD wallet tree structure

The HD wallet tree structure offers tremendous flexibility. Each parent
extended key can have 4 billion children: 2 billion normal children and
2 billion hardened children. Each of those children can have another 4
billion children, and so on. The tree can be as deep as you want, with
an infinite number of generations. With all that flexibility, however,
it becomes quite difficult to navigate this infinite tree. It is
especially difficult to transfer HD wallets between implementations,
because the possibilities for internal organization into branches and
subbranches are endless.

Two BIPs offer a solution to this complexity by creating some proposed
standards for the structure of HD wallet trees. BIP43 proposes the use
of the first hardened child index as a special identifier that signifies
the "purpose" of the tree structure. Based on BIP43, an HD wallet
should use only one level-1 branch of the tree, with the index number
identifying the structure and namespace of the rest of the tree by
defining its purpose. For example, an HD wallet using only branch
m/i++&#x27;++/ is intended to signify a specific purpose and that
purpose is identified by index number "i."

Extending that specification, BIP44 proposes a multiaccount structure
as "purpose" number +44'+ under BIP43. All HD wallets following the
BIP44 structure are identified by the fact that they only used one
branch of the tree: m/44'/.

BIP44 specifies the structure as consisting of five predefined tree levels:

-----
m / purpose' / coin_type' / account' / change / address_index
-----

The first-level "purpose" is always set to +44'+. The second-level
"coin_type" specifies the type of cryptocurrency coin, allowing for
multicurrency HD wallets where each currency has its own subtree under
the second level. There are three currencies defined for now: Bitcoin is
m/44'/0', Bitcoin Testnet is m/44++&#x27;++/1++&#x27;++, and Litecoin is
m/44++&#x27;++/2++&#x27;++.

The third level of the tree is "account," which allows users to
subdivide their wallets into separate logical subaccounts, for
accounting or organizational purposes. For example, an HD wallet might
contain two bitcoin "accounts": m/44++&#x27;++/0++&#x27;++/0++&#x27;++
and m/44++&#x27;++/0++&#x27;++/1++&#x27;++. Each account is the root of
its own subtree.

((("keys and addresses", see="also public and private keys")))On the
fourth level, "change," an HD wallet has two subtrees, one for creating
receiving addresses and one for creating change addresses. Note that
whereas the previous levels used hardened derivation, this level uses
normal derivation. This is to allow this level of the tree to export
extended public keys for use in a nonsecured environment. Usable
addresses are derived by the HD wallet as children of the fourth level,
making the fifth level of the tree the "address_index." For example, the
third receiving address for bitcoin payments in the primary account
would be M/44++&#x27;++/0++&#x27;++/0++&#x27;++/0/2. <<table_4-9>> shows
a few more examples.

[[table_4-9]]
.BIP44 HD wallet structure examples
[options="header"]
|=======
|HD path | Key described
| M/44++&#x27;++/0++&#x27;++/0++&#x27;++/0/2 | The third receiving public key for the primary bitcoin account
| M/44++&#x27;++/0++&#x27;++/3++&#x27;++/1/14 | The fifteenth change-address public key for the fourth bitcoin account
| m/44++&#x27;++/2++&#x27;++/0++&#x27;++/0/1 | The second private key in the Litecoin main account, for signing transactions
|=======