\documentclass{llncs}

%\documentclass[twoside,letterpaper]{IEEEtran}
\usepackage[margin=1in]{geometry}
\usepackage[utf8]{inputenc}
\usepackage{url}
\usepackage{tikz}
\usepackage{eurosym}
\usepackage{listings}
\usepackage{graphicx}
%\usepackage{wrapfig}
\usepackage[caption=false,font=normalsize,labelfont=sf,textfont=sf]{subfig}
\usepackage{wrapfig}
\usepackage{url}
%\usepackage{stfloats}

\usetikzlibrary{shapes,arrows}
\usetikzlibrary{positioning}
\usetikzlibrary{calc}

% CSS
\lstdefinelanguage{CSS}{
  keywords={color,background-image:,margin,padding,font,weight,display,position,top,left,right,bottom,list,style,border,size,white,space,min,width, transition:, transform:, transition-property, transition-duration, transition-timing-function},
  sensitive=true,
  morecomment=[l]{//},
  morecomment=[s]{/*}{*/},
  morestring=[b]',
  morestring=[b]",
  alsoletter={:},
  alsodigit={-}
}

% JavaScript
\lstdefinelanguage{JavaScript}{
  morekeywords={typeof, new, true, false, catch, function, return, null, catch, switch, var, if, in, while, do, else, case, break},
  morecomment=[s]{/*}{*/},
  morecomment=[l]//,
  morestring=[b]",
  morestring=[b]'
}

\lstdefinelanguage{HTML5}{
  language=html,
  sensitive=true,
  alsoletter={<>=-},
  morecomment=[s]{<!-}{-->},
  tag=[s],
  otherkeywords={
  % General
  >,
  % Standard tags
	<!DOCTYPE,
  </html, <html, <head, <title, </title, <style, </style, <link, </head, <meta, />,
	% body
	</body, <body,
	% Divs
	</div, <div, </div>,
	% Paragraphs
	</p, <p, </p>,
	% scripts
	</script, <script,
  % More tags...
  <canvas, /canvas>, <svg, <rect, <animateTransform, </rect>, </svg>, <video, <source, <iframe, </iframe>, </video>, <image, </image>
  },
  ndkeywords={
  % General
  =,
  % HTML attributes
  charset=, src=, id=, width=, height=, style=, type=, rel=, href=,
  % SVG attributes
  fill=, attributeName=, begin=, dur=, from=, to=, poster=, controls=, x=, y=, repeatCount=, xlink:href=,
  % CSS properties
  margin:, padding:, background-image:, border:, top:, left:, position:, width:, height:,
	% CSS3 properties
  transform:, -moz-transform:, -webkit-transform:,
  animation:, -webkit-animation:,
  transition:,  transition-duration:, transition-property:, transition-timing-function:,
  }
}

\date{}
\begin{document}
\title{Enabling Secure Web Payments with GNU Taler}


% Not sure how to do authors with the
% IEEEtran template correctly ...
\author{Jeffrey Burdges \and
Florian Dold \and
Christian Grothoff \and
Marcello Stanisci}
\institute{Inria Rennes - Bretagne Atlantique \\
\email{FIRSTNAME.LASTNAME@inria.fr}
}

\maketitle




\begin{abstract}
GNU Taler is a new electronic online payment system which provides
privacy for customers and accountability for merchants. It uses an
exchange service to issue digital coins using blind signatures,
and is thus not subject to the performance issues that plague
Byzantine fault-tolerant consensus-based solutions.

The focus of this paper is addressing the challenges payment systems
face in the context of the Web.  We discuss how to address
Web-specific challenges, such as handling bookmarks and sharing of
links, as well as supporting users that have disabled JavaScript.  Web
payment systems must also navigate various constraints imposed by
modern Web browser security architecture, such as same-origin policies
and the separation between browser extensions and Web pages.  While
our analysis focuses on how Taler operates within the security
infrastructure provided by the modern Web, the results partially
generalize to other payment systems.

We also include the perspective of merchants, as existing systems have
often struggled with securing payment information at the merchant's
side.  Here, challenges include avoiding database transactions for
customers that do not actually go through with the purchase, as well
as cleanly separating security-critical functions of the payment
system from the rest of the Web service.
\end{abstract}

\section{Introduction}

The Internet needs a secure, usable and privacy-preserving
micropayment system, which is not backed by a ``crypto currency''.
Payment systems involving state-issued currencies have been used for
centuries to facilitate transactions, and the involvement of the state
has been critical as state institutions can dampen fluctuations in the
value of the currency~\cite{dominguez1993}. Controlling money supply
is critical to ensure stable prices that facilitate
trade~\cite{quantitytheory1997volckart} instead of speculation~\cite{lewis_btc_is_junk}.

Internet transactions, such as sending an e-mail or reading a Web
site, tend to be of smaller commercial value than traditional
transactions involving the exchange of physical goods.  Consequently,
if we want to associate payments with these types of transactions, we
face the challenge of reducing the mental and technical overheads of
existing payment systems.  For example, executing a 3DS payment
process (Figure~\ref{fig:cc3ds}) takes too long, is way too complex
and way to expensive to be used for payment for typical Web articles.

Addressing this problem is urgent: ad-blocking technology is eroding
advertising as a substitute for micropayments~\cite{adblockblocks},
and the Big Data business model in which citizens pay with their
private information~\cite{ehrenberg2014data} in combination with the
deep state hastens our society's regression towards
post-democracy~\cite{rms2013democracy}.


The focus of this paper is GNU Taler, a new free software payment
system designed to meet certain key ethical considerations from a
social liberalism perspective. In Taler, the paying customer remains
anonymous while the merchant is easily identified and thus taxable.
Here, {\em anonymous} simply means that the payment process does not require
any personal information from the customer, and that different
transactions by the same customer are unlinkable.  Naturally, the
specifics of the transaction---such as delivery of goods to a shipping
address, or the use of non-anonymous IP-based communication---may
still leak information about the customer's identity.  {\em Taxable}
means that for any transaction the state can easily obtain the
necessary information about the identity of the merchant and the
respective contract in order to levy income, sales, or value-added
taxes. Taler uses blind signatures~\cite{chaum1983blind} to create
digital coins, and a new {\em refresh} protocol~\cite{talercrypto} to
allow giving change and refunds while maintaining unlinkability.

This paper will not consider the details of Taler's cryptographic
protocols.\footnote{Details of the protocol are documented at
  \url{https://api.taler.net/}} The basic cryptography behind
blind-signature based payment systems has been known for over 25
years~\cite{chaum1990untraceable}.  However, only in 2015 the W3c
started the payments working group~\cite{pigs} to explore requirements
for deploying payment systems that are more secure and easy to use for
the Web.  Our work describes how a modern payment system using blind
signatures could practically be integrated with the modern Web to
improve usability, security and privacy.  This includes the challenge
of hiding the cryptography from the users, integrating with modern
browsers, integrating with Web shops, providing proper cryptographic
proofs for all operations, and handling network failures.  We explain
our design using terms from existing {\em mental models} users have
from widespread payment systems.

%\newpage
Key contributions of this paper are:
\begin{itemize}
 \item A description of different payment systems using
  common terminology, which allows us to analytically compare
  these systems.
 \item An introduction to the Taler payment system from the
  perspective of users and merchants, with a focus on how
  to achieve secure payments in a way that is intuitive and
  has adequate fail-safes.
 \item Detailed considerations for how to adapt Taler to
  Web payments and the intricacies of securing payments
  within the constraints of modern browsers.
 \item A publicly available free software
  reference implementation of the presented architecture.
\end{itemize}


\section{Existing payment workflows}

Before we look at the payment workflow for Taler, we sketch the
workflow of existing payment systems. This establishes a common
terminology which we will use to compare different payment processes.
We include interaction diagrams for some of the payment systems
based on resources from the W3C payment interest group.

\subsection{Cash}

Cash has traditionally circulated by being passed directly from buyers
to sellers with each seller then becoming a buyer. Thus, cash is
inherently a {\em peer-to-peer} payment system as participants all
appear in both buyer and seller roles, just at different times.
However, this view is both simplified and
somewhat dated.

In today's practice, cash is frequently first {\em withdrawn} from
ATMs by customers who then {\em spend} it with merchants, who, in turn,
{\em deposit} the cash with their respective {\em bank}.  In this
flow, security is achieved as the customer {\em authenticates} to the
ATM using {\em credentials} provided by the customer's bank, and the
merchant specifies his {\em account} details when depositing the cash.
The customer does not authenticate when spending the cash, but the
merchant {\em validates} the authenticity of the {\em coins} or bills.
Coins and bills are {\em minted} by state-licensed institutions, such
as the US Mint.  These institutions also provide detailed instructions
for how to validate the authenticity of the coins or
bills~\cite{ezb2016ourmoney}, and are typically the final trusted
authority on the authenticity of coins and bills.

As customers need not authenticate, purchases remain {\em
  anonymous}, modulo the limited tracking enabled in theory
by serial numbers printed on bills~\cite{pets2004kuegler},
which make each bill {\em unique}.
% NOTE : Internet claims this does not happen, but no references.
% https://rocketatm.com/notice-_recorded_serial_numbers_atm_decal

Spending cash does not provide any inherent {\em proof of purchase}
for the customer. Instead, the merchant provides paper
{\em receipts}, which are generated independently and do not receive
the same anti-forgery protections that are in place for cash.

Against most attacks, customers and merchants {\em limit} their risks
to the amount of cash that they carry or accept at a given
time~\cite{Bankrate}.  Additionally, customers are advised to choose
the ATMs they use carefully, as malicious ATMs may attempt to
{\em steal} their customer's credentials~\cite{ECB:TRoCF2014}.  Authentication with an
ATM can involve a special ATM card, or the use of credit or
debit cards.  In all these cases, these physical security tokens are
issued by the customer's bank.


% \smallskip
\subsection{Credit and debit cards}

\begin{figure*}[h!]
\begin{center}
\includegraphics[width=0.95\textwidth]{figs/cc3ds.pdf}
\end{center}
\caption{Card payment processing with 3DS. (From: W3C Web Payments IG.)}
\label{fig:cc3ds}
\end{figure*}

Credit and debit card payments operate by the customer providing their
credentials to the merchant.  Many different authentication and
authorization schemes are in use in various combinations. Secure
systems typically combine multiple forms of authentication including
secret information, such as personal identification numbers (PINs),
transaction numbers (TANs)~\cite{kobil2016tan} or credit card
verification (CCV) codes, and physical security devices such cards
with an EMV chip~\cite{emv}, TAN generators, or the customer's mobile
phone~\cite{mtan}.  A typical modern Web payment process involves:
{(1.)} the merchant offering a secure communication channel using TLS
based on the X.509 public key infrastructure;\footnote{Given numerous
  TLS protocol and implementation flaws as well as X.509 key
  management incidents in recent years~\cite{holz2014}, one cannot
  generally assume that the security provided by TLS is adequate under
  all circumstances.}  {(2.)} selecting a {\em payment method}; {(3.)}
entering the credit card details like the owner's name, card number,
expiration time, CCV code, and billing address; and {(4.)}
(optionally) authorizing the transaction via mobile TAN, or by
authenticating against the customer's bank.  Due to the complexity
of this the data entry is often performed on a Web site that
is operated by a third-party payment processor and {\em not} the merchant or
the customer's bank.  Figure~\ref{fig:cc3ds} shows a typical card-based payment
process on the Web using the UML style of the W3C payment interest
group~\cite{pigs}.  Most of the details are not relevant to this
paper, but the diagram nicely illustrates the complexity of the common
3-D secure (3DS) process.

Given this process, there is an inherent risk of information leakage
of customers' credentials.  {\em Fraud detection} systems attempt to detect
misuse of stolen credentials, and payment system providers handle
disputes between customers and merchants.  As a result, Web payment
processes may finish with {(5.)} the payment being rejected for a
variety of reasons, such as false positives in fraud detection or
the merchant not accepting the particular card issuer.

Traditionally, merchants bear most of the financial risk, and a key
``feature'' of the 3DS process compared to traditional card payments
is to shift dispute {\em liability} to the issuer of the card---who
may then try to shift it to the customer \cite[\S2.4]{3DSsucks}.
%
% online vs offline vs swipe vs chip vs NFC ???
% extended verification
%
Even in cases where the issuer or the merchant remain legally first in
line for liabilities, there are still risks customers incur from the
card dispute procedures, such as neither them nor the payment
processor noticing fraudulent transactions, or them noticing
fraudulent transactions past the {\em deadline} until which their bank
would reimburse them.  The customer also typically only has a
merchant-generated comment and the amount paid in his credit card
statement as a proof for the transaction.  Thus, the use of credit
cards online does not generate any cryptographically {\em verifiable}
electronic receipts for the customer, which theoretically enables
malicious merchants to later change the terms of the contract.

Beyond these primary issues, customers face secondary risks of
identity theft from the personal details exposed by the authentication
procedures. In this case, even if the financial damages are ultimately
covered by the bank, the customer always has to deal with the hassle
of {\em notifying} the bank in the first place.  As a result,
customers must remain wary about using their card, which limits their
online shopping~\cite[p. 50]{ibi2014}.
% There is nevertheless a trend towards customers preferring cards
% over cash even in face-to-face purchases \cite{} in part because
% cash theft can be violent even if the amounts as stake are smaller
% than with electronic theft.
%
%Merchants are exposed to these same risks because either laws and/or
%contracts with the payment system providers require them to take care
%in handling customer information.
% 40 million stolen at target.  fine?

%In cash payments, these risks do not exist because customers have
%complete control over the authentication procedure with their bank
%and the merchant is not involved.

% pressure to shop with big merchants
% merchants keep payment credentials on file
% Just a few merchants like Apple demand credentials up front
%  "this reversal of authentication vs shopping slows shopping"

% \smallskip
\subsection{Bitcoin}

\begin{figure*}[b!]
\includegraphics[width=\textwidth]{figs/bitcoin.pdf}
\caption{Bitcoin payment processing. (From: W3C Web Payments IG.)}
\label{fig:bitcoin}
\end{figure*}

Bitcoin operates by recording all transactions in a pseu\-do\-ny\-mous
public {\em ledger}.  A Bitcoin account is identified by its public
key, and the owner must know the corresponding private key to
authorize the transfer of Bitcoins from the account to other accounts.
The information in the global public ledger allows everybody to
compute the balances in all accounts and to see all transactions.
Transactions are denominated in a new currency labeled BTC, whose
valuation depends upon {\em speculation}, as there is no authority
that could act to stabilize exchange rates or force anyone to
accept BTC as {\em legal tender} to settle obligations.  Adding transactions to
the global public ledger involves broadcasting the transaction data,
peers verifying and appending it to the public ledger, and some peer
in the network solving a moderately hard computational proof-of-work
puzzle, which is called {\em mining}.

The mining process is incentivised by a combination of transaction
fees and mining rewards~\cite{nakamoto2008bitcoin}. The latter
process also provides primitive accumulation~\cite{primitiveacc} for BTC.
Conversion to BTC from other currencies and vice versa incurs
substantial fees~\cite{BTCfees}.  There is now an extreme diversity of
Bitcoin-related payment technologies, but usability improvements are
usually achieved by adding a trusted third party, and there have been
many incidents where such parties then embezzled funds from their
customers~\cite{BTC:demise}.

The classical Bitcoin payment workflow consisted of entering payment
details into a peer-to-peer application.  The user would access their
Bitcoin {\em wallet} and instruct it to transfer a particular amount
from one of his accounts to the account of the merchant. He could
possibly include additional metadata to be associated with the
transfer to be embedded into the global public ledger.  The wallet
application would then transmit the request to the Bitcoin
peer-to-peer overlay network.  The use of an external payment
application makes payments significantly less
browser-friendly than ordinary card payments, as illustrated in
Figure~\ref{fig:bitcoin}. This has led to the development of
browser-based
wallets.\footnote{\url{https://github.com/frozeman/bitcoin-browser-wallet}}

Bitcoin payments are only confirmed when they appear in the public
ledger, which is updated at an average frequency of once every 10
minutes.  Even then, it is possible that a fork in the so-called block
chain may void durability of the transaction; as a result, it is
recommended to wait for 6 blocks (on average one hour) before
considering a transaction committed~\cite{nakamoto2008bitcoin}.  In
cases where merchants are unable to accommodate this delay, they incur
significant fraud risks.

Bitcoin is considered to be secure against an adversary who cannot
control around a fifth of the Bitcoin miner's computational
resources~\cite{BTC:Bahack13,BTC:MajorityNotEnough,BTC:Eclipse}.  % 21percent?
As a result, the network must expend considerable computational
resources to keep this value high.
According to~\cite{vice_btc_unsustainable}, a single Bitcoin transaction uses roughly enough
electricity to power 1.57 American households for a day.
These costs are largely hidden by speculation in BTC,
but that speculation itself contributes to BTC's valuation being
volatile.~\cite{jeffries_economists_v_btc,lehmann_btc_fools_gold,lewis_btc_is_junk}. % exacerbating risk

% fees hit you 2-3 times with currency conversions
% more on massive transaction fees from blockchain.info

Bitcoin's pseudononymity applies equally to both customers and
merchants, which makes Bitcoin amen\-able to tax evasion, money
laundering, sales of contraband, and especially extorion
 \cite{NYA:CyberExtortionRisk}.
As a result, anonymity tools like mixnets do not enjoy widespread
support in the Bitcoin community where many participants seek to make
the currency appear more legitimate.  While Bitcoin's transactions
are difficult to track, there are several examples of Bitcoin's
pseudononymity being broken by investigators~\cite{BTC:Anonymity}.
This has resulted in the development of new protocols with better
privacy protections.

\begin{figure*}[t!]
\includegraphics[width=\textwidth]{figs/paypal.pdf}
\caption{Payment processing with PayPal. (From: W3C Web Payments IG.)}
\label{fig:paypal}
\end{figure*}

Zerocoin \cite{miers2013zerocoin} is such an extension of Bitcoin:
It affords protection against linkability of transactions,
but at non-trivial additional computational costs even for
spending coins.  This currently makes using Zerocoin unattractive
for payments, especially with mobile devices.

Bitcoin also faces serious scalability limitations, with the classic
implementation being limited to at most 7 transactions per second
globally on
average.\footnote{\url{http://hackingdistributed.com/2016/08/04/byzcoin/}}
There are a variety of efforts to confront Bitcoin's scaling problems
with off-blockchain techniques, like side-chains. % \cite{???}
Amongst these, the blind off-chain lightweight transactions (BOLT)
proposal~\cite{BOLT} provides anonymity by routing off-blockchain
transfers through bank-like intermediaries.  Although interesting,
there are numerous seemingly fragile aspects of the BOLT protocol,
including aborts deanonymizing customers, intermediaries risking
unlimited losses, and theft if a party fails to post a refute message
in a timely fashion.
% Of course, Taler itself could be used to provide a side-chain like technology
% Assuming these issues can be addressed,
% % and the relatively advanced crypto involved became production ready,
% Taler might prove a better platform for deploying a BOLT-like scheme
% than Zerocoin.


% In addition, the Bitcoin protocol does not interact well with
% conventional anonymity networks like Tor \cite{BTC:vsTor}
% dark pools?

% outdated ideas :
%  mining suck0rs,
%  DDoS : wired article?
%  economic ideology

\subsection{Walled garden payment systems}

Walled garden payment systems offer ease of use by processing payments
using a trusted payment service provider. Here, the customer
authenticates to the trusted service, and instructs the payment
provider to execute a transaction on his behalf
(see Figure~\ref{fig:paypal}).  In these payment systems, the provider
basically acts like a bank with accounts carrying balances for the
various users.  In contrast to traditional banking systems, both
customers and merchants are forced to have an account with the same
provider.  Each user must take the effort to establish his identity
with a service provider to create an account.  Merchants and customers
obtain the best interoperability in return for their account creation
efforts if they start with the biggest providers.  As a result, there
are a few dominating walled garden providers, with AliPay, ApplePay,
GooglePay, SamsungPay and PayPal being the current {\em oligopoly}.  In this
paper, we will use PayPal as a representative example for our discussion
of these payment systems.

As with card payment systems, these oligopolies are politically
dangerous~\cite{crinkey2011rundle}, and the lack of {\em competition}
can result in excessive profit taking that may require political
solutions~\cite{guardian2015cap} to the resulting {\em market
  failure}.  The use of non-standard {\em proprietary} interfaces to
the payment processing service of these providers serves to reinforce
the customer {\em lock-in}.


\section{Taler}

Taler is a free software cryptographic payment system. It has an open
protocol specification, which couples cash-like anonymity for customers
with low transaction costs, signed digital
receipts, and accurate income information to facilitate taxation and
anti-corruption efforts.

% FIXME: maybe say what blind signature are
Taler achieves anonymity for buyers using {\em blind
signatures}~\cite{chaum1983blind}.  Since their discovery thirty years
ago, cryptographers have viewed blind signatures as the optimal
cryptographic primitive for privacy-preserving consumer-level transaction systems.
However, previous transaction systems based on blind signatures have
failed to see widespread adoption.  This paper details strategies for
hiding the cryptography from users and integrating smoothly with the
Web, thereby providing crucial steps to bridge the gap between good
cryptography and real-world deployment.

%\subsection{Design overview}

\begin{figure}[b!]
\centering
\begin{tikzpicture}
 \tikzstyle{def} = [node distance=3em and 5em, inner sep=1em, outer sep=.3em];
 \node (origin) at (0,0) {};
 \node (exchange) [def,above=of origin,draw]{Exchange};
 \node (customer) [def, draw, below left=of origin] {Customer};
 \node (merchant) [def, draw, below right=of origin] {Merchant};
 \node (auditor) [def, draw, above right=of origin]{Auditor};

 \tikzstyle{C} = [color=black, line width=1pt]

 \draw [<-, C] (customer) -- (exchange) node [midway, above, sloped] (TextNode) {withdraw coins};
 \draw [<-, C] (exchange) -- (merchant) node [midway, above, sloped] (TextNode) {deposit coins};
 \draw [<-, C] (merchant) -- (customer) node [midway, above, sloped] (TextNode) {spend coins};
 \draw [<-, C] (exchange) -- (auditor) node [midway, above, sloped] (TextNode) {verify};

\end{tikzpicture}
\caption{Taler system overview.}
\label{fig:system}
\end{figure}


There are four key roles in the Taler system (Figure~\ref{fig:system}):

\begin{figure*}[b!]
\includegraphics[width=0.9\textwidth]{figs/taler-withdraw.pdf}
\caption{Withdrawing coins with Taler.}
\label{fig:taler-withdraw}
\end{figure*}




\begin{itemize}
\item
{\em Customers} use a digital wallet to withdraw,
hold, and spend coins. Wallets manage the customer's accounts
at the exchange, and keep receipts in a transaction history.  Wallets can be
realized as browser extensions, mobile Apps or even in custom
hardware.  If a user's digital wallet is compromised, the current
balance may be lost, just like with an ordinary wallet containing cash.
A wallet includes a list of trusted auditors, and will warn
users against using an exchange that is not certified by a trusted
auditor.

\begin{figure}[b!]%[36]{R}{0.5\linewidth}
\subfloat[Bank login. (Simplified for demonstration.)]{
\includegraphics[width=0.4\linewidth]{figs/bank0a.png}
\label{subfig:login}} \hfill
\subfloat[Specify amount to withdraw. (Integrated bank support.)]{
\includegraphics[width=0.4\linewidth]{figs/bank1a.png}
\label{subfig:withdraw}} \\
\subfloat[Select exchange provider. (Generated by wallet.)]{
\includegraphics[width=0.4\linewidth]{figs/bank2a.png}
\label{subfig:exchange}} \hfill
\subfloat[Confirm transaction with a PIN. (Generated by bank.)]{
\includegraphics[width=0.4\linewidth]{figs/bank3a.png}
\label{subfig:pin}}
\caption{Required steps in a Taler withdrawal process.}
\label{fig:withdrawal}
\end{figure}



\item
{\em Exchanges}, which are run by financial service providers, enable
customers to withdraw anonymous digital coins,
and merchants to deposit digital coins, in exchange for
bank money.  Coins are signed by the exchange using
a blind signing scheme~\cite{chaum1983blind}.  Thus, only
an exchange can issue new coins, but coins cannot be traced back
to the customer that withdrew them.
Furthermore, exchanges learn the amounts withdrawn by customers
and deposited by merchants, but they do not learn the relationship
between customers and merchants.  Exchanges perform online detection
of double spending, thus providing merchants instant feedback
---including digital proofs---in case of misbehaving customers.

\item
{\em Merchants} provide goods or services in exchange for coins held
by customers' wallets.  Merchants deposit these coins at the
exchange used by the customer in return for a bank wire
transfer of their value.  While the exchange is determined by
the customer, the merchant's contract specifies the currency,
a list of accepted auditors and the maximum exchange deposit
fee the merchant is willing to pay.  Merchants consist of a
{\em frontend}, which interacts with the customer's wallet, and a {\em
backend}, which interacts with the exchange.  Typical frontends include
Web shops and point-of-sale systems.

\item
{\em Auditors} verify that exchanges operate correctly to limit the risk
that customers and merchants incur by using a particular exchange.
Auditors are typically operated by or on behalf of financial regulatory authorities.
Depending on local legislation, auditors may mandate that exchanges
have enough financial reserves before authorizing them to create a given
volume of signed digital coins to provide a buffer against potential risks due to
operational failures (such as data loss or theft of private keys) of the exchange.
Auditors certify exchanges that they audit using digital signatures.  The
respective signing keys of the auditors are distributed to customer and
merchants.
\end{itemize}


The specific protocol between wallet and merchant depends on the
setting.  For a traditional store, a near field communication (NFC)
protocol might be used between a point-of-sale system and a mobile
application.  In this paper, we focus on Web payments for an online
shop and explain how the actors in the Taler system interact by way of
a typical payment.

Initially, the customer installs the Taler wallet extension for
their browser.  This only needs to be done once per
browser. Naturally, this step may become superfluous if Taler is
integrated tightly with browsers in the future.  Regardless,
installing the extension involves only one or two clicks to confirm the
operation. Restarting the browser is not required.


\begin{figure*}[b!]
\includegraphics[width=0.9\textwidth]{figs/taler-pay.pdf}
\caption{Payment processing with Taler.}
\label{fig:taler-pay}
\end{figure*}


\subsection{Withdrawing coins}

As with cash, the customer must first withdraw digital coins
(Figure~\ref{fig:taler-withdraw}).  For this, the customer must first
visit the bank's online portal.  Here, the bank will
typically require some form of authentication, the specific method
used depends on the bank (Figure~\ref{subfig:login}).

The next step depends on the level of Taler support offered by the bank:
\begin{itemize}
\item If the bank does not offer integration with Taler, the
  customer needs to use the menu of the wallet to create a {\em reserve}.
  The wallet will ask which amount in which {\em currency} (e.g. EUR
  or USD) the customer wants to withdraw, and allow the customer to
  select an exchange.  Given this information, the wallet will
  instruct the customer to transfer the respective amount to the
  account of the exchange.  The customer will have to enter a
  % FIXME it is not said that this crypto token is the reserve,
  % or, more abstractly, that "identify" this operation
  % CG: I don't think this has to be said.
  54-character reserve key, which includes 256 bits of entropy and an
  8-bit checksum into the transfer subject.  Naturally, the above is
  exactly the kind of interaction we would like to avoid for
  usability reason.
\item Hence, if the bank fully supports Taler, the
  customer has a form in the online banking portal in which they can specify
  an amount to withdraw (Figure~\ref{subfig:withdraw}).
  The bank then triggers an interaction with
  the wallet to allow the customer to select an exchange
  (Figure~\ref{subfig:exchange}).  Afterwards,
  the wallet instructs the bank about the details of the wire
  transfer.  The bank asks the customer to authorize the transfer, and
  finally confirms to the wallet that the transfer has been
  successfully initiated.
\end{itemize}

In either case, the wallet can then withdraw the coins from the
exchange, and does so in the background without further interaction
with the customer.

In principle, the exchange can be directly operated by the bank, in
which case the step where the customer selects an exchange could be
skipped by default.  However, we generally assume that the exchange is
a separate entity, as this yields the largest anonymity set for
customers, and may help create a competitive market.

\subsection{Spending coins}
% \tinyskip

\begin{figure}[t!]
  \subfloat[Select article][Select article. \\ Generated by Web shop.]{
\includegraphics[width=0.30\textwidth]{figs/cart.png}
\label{subfig:cart}} \hfill
\subfloat[Confirm payment][Confirm payment. \\ Generated by Taler wallet.]{
\includegraphics[width=0.30\textwidth]{figs/pay.png}
\label{subfig:payment}} \hfill
\subfloat[Receive article][Receive article. \\ Generated by Web shop.]{
\includegraphics[width=0.30\textwidth]{figs/fulfillment.png}
\label{subfig:fulfillment}}
\caption{Required steps in a Taler checkout process.}
\label{fig:shopping}
\end{figure}


At a later point in time, the customer can spend their coins by
visiting a merchant that accepts digital coins in the respective
currency issued by the respective exchange
(Figure~\ref{fig:taler-pay}).  Merchants are generally configured to
either accept a specific exchange, or to accept all the exchanges
audited by a particular auditor.  Merchants can also set a ceiling for
the maximum amount of transaction fees they are willing to cover.
Usually these details do not matter for the customer, as we expect
most merchants to accept most exchange providers accredited by the
auditors that wallets include by default.  Similarly, we expect
exchanges to operate with transaction fees acceptable to most
merchants to avoid giving customers a reason to switch to another
exchange.  If transaction fees are higher than what is covered by the
merchant, the customer may choose to cover them.

% \tinyskip
\lstdefinelanguage{JavaScript}{
  keywords={typeof, new, true, false, catch, function, return, null, catch, switch, var, if, in, while, do, else, case, break, for},
  keywordstyle=\color{blue}\bfseries,
  ndkeywords={class, export, boolean, throw, implements, import, this},
  ndkeywordstyle=\color{darkgray}\bfseries,
  identifierstyle=\color{black},
  sensitive=false,
  comment=[l]{//},
  morecomment=[s]{/*}{*/},
  commentstyle=\color{purple}\ttfamily,
  stringstyle=\color{red}\ttfamily,
  morestring=[b]',
  morestring=[b]"
}

\begin{figure*}[t!]
 \lstset{language=HTML5}
 \lstinputlisting{figs/taler-presence-js.html}
 \caption{Sample code to detect the Taler wallet. Allowing the
  Web site to detect the presence of the wallet leaks one bit
  of information about the user. The above logic also works
  if the wallet is installed while the page is open.}
  \label{listing:presence}
\end{figure*}


\begin{figure*}[t!]
 \lstset{language=HTML5}
 \lstinputlisting{figs/taler-contract.html}
 \caption{Sample code to pass a contract to the Taler wallet.
          Here, the contract is fetched on-demand from the server.
          The {\tt taler\_pay()} function needs to be invoked
          when the user triggers the checkout.}
 \label{listing:contract}
\end{figure*}


As with traditional Web transactions, customers first select which
items they wish to buy.  This can involve building a traditional
shopping cart, or simply clicking on a particular link for the
respective article (Figure~\ref{subfig:cart}).  Once the articles have
been selected, the Web shop directs the user to the {\em offer} URL,
where the payment details are negotiated.  The process usually starts
by allowing the user to select a {\em payment method} from a set of
methods supported by the Web shop.  Taler also allows the Web shop to
detect the presence of a Taler wallet (Figure~\ref{listing:presence}),
so that the selection of alternative payment methods can be skipped if
a Taler wallet is installed (as it is in Figure~\ref{fig:shopping}).

\begin{figure*}[t!]
  \lstset{language=JavaScript}
\begin{lstlisting}
{
  "H_wire":"YTH0C4QBCQ10VDNTJN0DCTTV2Z6JHT5NF43F0RQHZ8JYB5NG4W4G...",
  "amount":{"currency":"EUR","fraction":1,"value":0},
  "auditors":[{"auditor_pub":"42V6TH91Q83FB846DK1GW3JQ5E8DS273W4236AXC397892ESD0B0"}],
  "exchanges":[{"master_pub":"1T5FA8VQHMMKBHDMYPRZA2ZFK2S63AKF0YTHJZWFKF45K2JGC8H0",
                "url":"https://exchange/"}],
  "expiry":"/Date(1480119270)/",
  "fulfillment_url": "https://shop/essay/42",
  "max_fee":{"currency":"EUR","fraction":01,"value":0},
  "merchant":{"address":"Mailbox 4242","jurisdiction":"Jersey","name":"Shop Inc."},
  "merchant_pub":"Y1ZAR5346J3ZTEXJCHQY9NJN78EZ2HSKZK8M0MYTNRJG5N0HD520",
  "products":[{
    "description":"Essay: The GNU Project",
    "price":{"currency":"EUR","fraction":1,"value":0},
    "product_id":42,"quantity":1}],
  "refund_deadline":"/Date(1471522470)/",
  "timestamp":"/Date(1471479270)/",
  "transaction_id":249960194066269
}
\end{lstlisting}
  \caption{Minimal Taler contract over a digital article with a value of \EUR{0.10}. The merchant will pay transaction fees up to \EUR{0.01}.  The hash over the wire transfer information was truncated to make it fit to the page.}
  \label{listing:json-contract}
\end{figure*}

\begin{figure*}[t!]
  \lstset{language={}}
\begin{lstlisting}
HTTP/1.1 402 Payment Required
Content-Type: text/html; charset=UTF-8
X-Taler-Contract-Url: https://shop/generate-contract/42
...

<!DOCTYPE html>
<html>
  ...
  <!-- fallback for browsers without the Taler extension -->
  You do not seem to have Taler installed, here are
  other payment options:
  ...
</html>
\end{lstlisting}
 \caption{Sample HTTP response for a Web resource that requires a payment.}
  \label{listing:http-contract}
\end{figure*}

The offer URL of the Web shop can then initiate payments by sending a
\emph{contract proposal} (Figure~\ref{listing:json-contract}) to the wallet, either via the HTTP status
code {\tt 402 Payment Required} (Figure~\ref{listing:http-contract}), or via Taler's JavaScript API
(Figure~\ref{listing:contract}).  The wallet then presents the
contract to the user.  The format of the contract is in an extensible
JSON-based format defined by Taler and not HTML, as the rendering of
the contract is done by the wallet to ensure correct visual
representation of the terms and prices.  In case that transaction fees
need to be covered by the customer, these are shown together with the
rest of the proposed contract.

If the customer approves the contract by clicking the ``Confirm
Payment'' button (Figure~\ref{subfig:payment}), their wallet signs the
contract with enough coins to cover the contract's cost, stores all of
the information in its local database, and transmits the payment to
the {\em payment} URI of the Web shop.  Once the Web shop confirms the
payment, the wallet redirects the browser to the {\em fulfillment} URL
provided by the merchant (Figure~\ref{subfig:fulfillment}).

\subsection{Browser security}

The Taler wallet operates from a securely isolated {\em background}
context on the client side.  The user interface that displays the
contract and allows the user to confirm the payment is displayed by
this background context.  By running in the background context, the
wallet can perform the cryptographic operations protected from the
main process of the Web site.  In particular, this architecture is
secure against a merchant that generates a page that looks like the
wallet's payment page (Figure~\ref{subfig:payment}), as such a page
would still not have access to the private keys of the coins that are
exclusive to the background context.

%The current Taler specification is written to accommodate for
%the rather restrictive set of APIs that WebExtension, a
%cross-browser extension API, provides.

\subsection{Managing browser session state}

% FIXME: this is where we probably want to revise quite a bit,
% including improving the description AND addressing the JS-less
% implementation.

% Say that this doesn't require accounts!
% explain that HTTP session state can be restored
% from the wallet state (replay!)

% explain repurchase correlation ID

% contract vs complete fulfillment URL

% design that allows merchant to not store any information

A purchase by a customer, completed or in progress, is uniquely
identified by a URL, called the \emph{fulfillment URL}.  The
information contained in the fulfillment URL must allow the merchant
to restore the full contract that was associated with the purchase,
either directly from the URL or indirectly from an identifier in a
database.  Efficiently reconstructing the contract entirely from the
URL instead of using costly database transactions can be important, as
costly disk operations for incomplete purchases make merchants
more susceptible to denial-of-service attacks from adversaries
pretending to be customers.

When a customer completed a purchase, navigating to the fulfillment
URL in a browser will show the resource associated with the purchase.
This resource can be a digital good such as a news article, or simply
a confirmation for products that are delivered by other means.

% FIXME: maybe we should have the URL that we are requesting
% in the example, and not just the response?

\begin{figure*}[t!]
  \lstset{language={}}
\begin{lstlisting}
HTTP/1.1 402 Payment Required
Content-Type: text/html; charset=UTF-8
X-Taler-Contract-Hash: 2BAH2AT4GSG5JRM2W4YWTSYGY66EK4X8CX2V69D5VF7XV703AJMG
X-Taler-Offer-Url: https://shop/article/42
X-Taler-Pay-Url: https://shop/pay
...

<!DOCTYPE html>
<html>
  ...
  <!-- fallback for browsers without the Taler extension -->
  You do not seem to have Taler installed, here are
  other payment options:
  ...
</html>
\end{lstlisting}
\caption{Sample HTTP response when the user agent navigates to a fulfillment
  URL without the session state that indicates they have paid for the resource.
  Note that unlike in Listing~\ref{listing:http-contract}, the response
  references a contract that typically is already known to the wallet via its
  hash code.}
  \label{listing:http-execute}
\end{figure*}

In order to ensure that only the paying customer has access to the Web
resources behind the fulfillment URL, the Web store's server must
check the browser's session state.  If the merchant can confirm that
the visitor has paid, the respective Web resource is returned.  If the
merchant cannot confirm that the visitor has paid for the contract,
for example because the session state was lost,\footnote{This can
  happen when when privacy conscious users delete their cookies.
  Also, some user agents (such as the TOR browser) do not support
  persistent (non-session) cookies.} it {\em again} triggers a payment
process (either via JavaScript or using {\tt 402 Payment Required}, see Figure~\ref{listing:http-execute}).
Since the fulfillment URL refers to a contract that typically is already known
to the user's wallet, it suffices to pass the contract hash (instead of
full contract) to the wallet.
If the user previously approved paying for the contract with this hash, the
wallet will (re-)transmit the signed coins that are associated
with the contract to the merchant.

When a user visits a fulfillment URL without having the associated
contract in their wallet, the wallet redirects the browser to the {\em
  offer} URL for that purchase, if applicable.  This behavior is
useful when a user wishes to share a fulfillment link with another
user to point him to the same resource.

Note that due to the limited WebExtensions API, the session state can
only be acquired when the wallet causes the browser to navigate to the
fulfillment URL (first without session state), since the session state
must be set from the same origin as the fulfillment URL.  As a result,
the shop cannot simply return the fulfillment information in response
to the wallet performing the payment.  However, this extra round trip
is also justified as the wallet needs to inspect the response anyway
as it may contain error reports about a failed payment which the wallet
has to handle.  Finally, it ensures that the fulfillment page is fetched
via an HTTP GET request instead of an HTTP POST request, which is
important to nicely support the use of navigation (``back'', ``forward''
buttons) and bookmarks.

Various failure modes are considered in this design:

\begin{itemize}
\item If the payment fails on the network, the request is typically
 retried.  How often the client retries automatically before informing
 the user of the network issue is up to the merchant.  If the network
 % FIXME this (above) could be ambiguous because the network failure
 % can happen between the wallet and the merchant without the merchant
 % getting any (failing) request, so the merchant cannot count how much
 % times a payment has failed.
 % CG: Well, the merchant can do that counting *client-side*. The retries
 % will be controlled by the JS on the client side, which is provided
 % by the merchant initially.
 failure persists and is between the customer and the merchant, the wallet
 will try to recover control over the coins at the exchange by
 effectively spending the coins first using Taler's
 refresh protocol.  In this case, later deposits by the merchant
 will simply fail.  If the merchant already succeeded with the payment
 before the network failure, the customer can either retry the
 operation via the transaction history kept by the wallet, or demand a refund (see
 below).  Handling these errors does not require the customer to give
 up his privacy.
\item If the payment fails due to the exchange
 claiming that the request was invalid, the diagnostics created by the
 exchange are passed to the wallet for inspection.  The wallet then
 decides whether the exchange was correct, and can then inform the
 user about a fraudulent or buggy exchange.  At this time, it allows
 the user to export the relevant cryptographic data to be used in
 court.  If the exchange's proofs were correct and coins were
 double-spent, the wallet informs the user that its database must have
 been out-of-date (e.g. because it was restored from backup),
 updates the database and allows the user to retry
 the transaction.
\end{itemize}

\noindent
While our design requires a few extra roundtrips,
it has the following key advantages:
\begin{itemize}
  \item It works in the confines of the WebExtensions API.
  \item It supports restoring session state for bookmarked
    Web resources even after the session state is lost by the user agent.
  \item Sending deep links to fullfilment or offer pages to
    other users has the expected behavior
    of asking the other user to pay for the resource.
  \item Asynchronously POSTing coins from injected JavaScript costs
    one roundtrip, but does not interfer with navigation and allows
    proper error handling.
  \item The different pages of the merchant have clear
    delineations: the shopping pages conclude by proposing a contract, and
    the fulfillment page begins with processing an accepted contract.  It is thus
    possible for these pages to be managed by separate parties. The
    control of the fulfillment page over the transmission of the payment
    data minimizes the need for exceptions to handle cross-origin
    resource sharing~\cite{rfc6454,cors}.
  \item The architecture supports security-conscious users that may have
    disabled JavaScript, as it is not necessary to execute JavaScript
    originating from Web pages to execute the payment process.
\end{itemize}

% \smallskip

\subsection{Giving change and refunds}

\begin{figure*}[b!]
  \lstset{language={HTML5}}
\begin{lstlisting}
<script src="taler-wallet-lib.js"></script>
<script>
  // Obtain refund permissions from the merchant backend
  // ...
  let refundPermissions = /* ... */;
  taler.acceptRefunds(refundPermissions, (err) => {
    alert("An error occured while attempting a refund");
  });
</script>
\end{lstlisting}
 \caption{Sample JavaScript code to trigger a refund from the merchant's web shop}
  \label{listing:refund}
\end{figure*}

An important cryptographic difference between Taler and previous
transaction systems based on blind signing is that Taler is able to
provide unlinkable change and refunds.  From the user's point of view,
obtaining change is automatic and handled by the wallet, i.e., if the
user has a single coin worth \EUR{5} and wants to spend \EUR{2}, the
wallet may request three \EUR{1} coins in change. Critically, the
change giving process is completely hidden from the user.
In fact, our graphical user
interface does not offer a way to inspect the denominations of the
various coins in the wallet, it only shows the total amount available
in each denomination.  Expanding the views to show details may show
the exchange providers and fee structure, but not the cryptographic
coins.  Consequently, the major cryptographic advances of Taler are
invisible to the user.

Taler's refresh protocol~\cite{talercrypto} also allows merchants to give
refunds to customers. To refund a purchase, the merchant obtains a signed refund permission
from the exchange, which the customer's wallet processes
(Figure~\ref{listing:refund}) to obtain new, unlinkable coins as refund.
This process allows the customer to say anonymous when receiving refunds.

Taler's refresh protocol ensures unlinkability for both change and
refunds, thereby assuring that the user has key conveniences of other
payment systems while maintaining the security standard of an
anonymous payment system.

% Alternative version:
%An important technical difference between Taler and previous
%transaction systems based on blind signing is that in Taler coins
%consist of a public-private key pair with the blind signature on the
%public key, so that coins themselves can be used to anonymously sign
%the purchase contract.
%
%An important technical difference between Taler and previous
%transaction systems based on blind signing is that Taler coins
%consist of a public-private key pair with the blind signature on the
%public key, so that coins themselves can be used to anonymously sign
%the purchase contract.
%
%In general, these coins exceed the cost of the contract, so the wallet
%may specify that only a fraction of a coin be spent, leaving some
%residual value on the partially spent coin as ``change''.
%
%As the merchant received only a signature of the coin, not private
%or symmetric key material, merchants can refund anonymous coins by
%asking the mint to restore a part of the coin's original value,
%and notifying the customer's wallet to refresh the coin.
%
%Spending Taler coins reveals nothing about a customer per se.
%Yet, any coins that hold value after being involved in a purchase or
%a refund operation cannot be considered anonymous anymore because a
%merchant, and possibly the exchange, has now seen them and could
%link them that previous transaction.  At best, these tainted coins
%are only pseudononymous, similar to Bitcoin accounts.
%
%To maintain anonymity, a Taler wallet automatically performs a
%{\em refresh} operation with the mint API to both replace tainted
%coins with new freshly anonymized coins and to exchange old coins
%before their denomination's expiration date.  We view refreshing
%partially spent coins as analogous to giving change in cash
%transactions, but refreshing refunded coins allows Taler merchants
%to refund anonymous customers.  Cash transactions have these options,
%but credit cards require customer identification for both operations.
% Is this true?
% no comment around randomizing the serial numbers on bills



\subsection{Deployment considerations for merchants}

Payment system security is not only a concern for
customers, but also for merchants.  For consumers, existing schemes
may be inconvenient and not provide privacy, but remembering to
protect a physical token (e.g. the card) and to guard a secret
(e.g. the PIN) is relatively straightforward.  In contrast, merchants
are expected to securely handle sensitive customer payment data on
networked computing devices.  However, securing computer systems---and
especially payment systems that represent substantial value---is a
hard challenge, as evidenced by large-scale break-ins with millions of
consumer card records being illicitly copied~\cite{target}.

Taler simplifies the deployment of a secure payment system for
merchants. The high-level cryptographic design provides the first
major advantage, as merchants never receive sensitive payment-related
customer information.  Thus, they do not have to be subjected to
costly audits or certified hardware, as is commonly the case for
processing card payments~\cite{pcidss}. In fact, the exchange does not
need to have a formal business relationship with the merchant at all.
According to our design, the exchange's contract with the state
regulator or auditor and the customers ought to state that it must
honor all (legal and valid) deposits it receives.  Hence, a merchant
supplying a valid deposit request should be able to enforce this in
court without a prior direct business agreement with the exchange.
This dramatically simplifies setting up a shop to the point that the
respective software only needs to be provided with the merchant's wire
transfer routing information to become operational.

The payment process requires a
few cryptographic operations on the side of the merchant, such as
signing a contract and verifying the customer's and the exchange's
signatures.  The merchant also needs to store transaction data, in
particular so that the store can match sales with incoming wire
transfers from the exchange.  We simplify this for merchants by
providing a generic payment processing {\em backend} for the Web
shops.

\begin{figure*}[b!]
\begin{center}
\begin{tikzpicture}[
  font=\sffamily,
  every matrix/.style={ampersand replacement=\&,column sep=2cm,row sep=2cm},
  source/.style={draw,thick,rounded corners,fill=green!20,inner sep=.3cm},
  process/.style={draw,thick,circle,fill=blue!20},
  sink/.style={source,fill=green!20},
  datastore/.style={draw,very thick,shape=datastore,inner sep=.3cm},
  dots/.style={gray,scale=2},
  to/.style={->,>=stealth',shorten >=1pt,semithick,font=\sffamily\footnotesize},
  every node/.style={align=center}]

  % Position the nodes using a matrix layout
  \matrix{
    \node[source] (wallet) {Wallet};
      \& \node[process] (browser) {Browser};
      \& \node[process] (shop) {Web shop};
      \& \node[sink] (backend) {Taler backend}; \\
  };

  % Draw the arrows between the nodes and label them.
  \draw[to] (browser) to[bend right=50] node[midway,above] {(4) signed contract}
      node[midway,below] {(signal)} (wallet);
  \draw[to] (wallet) to[bend right=50] node[midway,above] {(signal)}
      node[midway,below] {(5) signed coins} (browser);
  \draw[<->] (browser) -- node[midway,above] {(3,6) custom}
      node[midway,below] {(HTTP(S))} (shop);
  \draw[to] (shop) to[bend right=50] node[midway,above] {(HTTP(S))}
      node[midway,below] {(1) proposed contract / (7) signed coins} (backend);
  \draw[to] (backend) to[bend right=50] node[midway,above] {(2) signed contract / (8) confirmation}
      node[midway,below] {(HTTP(S))} (shop);
\end{tikzpicture}
\end{center}
 \caption{Both the customer's client and the merchant's server
          execute sensitive cryptographic operations in a
          secured background/backend that is protected against direct access.
          Interactions with the Taler exchange from the wallet background
          to withdraw coins and the Taler backend
          to deposit coins are not shown.
          Existing system security mechanisms
          are used to isolate the cryptographic components (boxes) from
          the complex rendering logic (circles), hence the communication
          is restricted to JavaScript signals or HTTP(S), respectively.}
 \label{fig:frobearch}
\end{figure*}

Figure~\ref{fig:frobearch} shows how the secure payment components
interact with the existing Web shop logic.  First, the Web shop
frontend is responsible for constructing the shopping cart.  For this,
the shop frontend generates the customary Web shop pages, which are transmitted
to the customer's browser.  Once the order has been constructed, the
shop frontend provides a {\em proposed contract} in JSON format to the
payment backend, which signs it and returns it to the frontend.  The
frontend then transfers the signed contract over the network, and
passes it to the wallet (sample code for this is shown in
Figure~\ref{listing:contract}).

Instead of adding any cryptographic logic to the merchant frontend,
the Taler merchant backend allows the implementor to delegate
coin handling to the payment backend, which validates the coins,
deposits them at the exchange, and finally validates and persists the
receipt from the exchange.  The merchant backend then communicates the
result of the transaction to the front\-end, which is then responsible
for executing the business logic to fulfill the order.  As a result of
this setup (Figure~\ref{fig:frobearch}), the cryptographic details
of the Taler protocol do not have to be re-implemented by each
merchant.  Instead, existing Web shops implemented in a multitude of
programming languages can add support for Taler by:
{\bf (0)} detecting in the browser that Taler is available; {\bf (1)}
upon request, generating a contract in JSON based on the shopping
cart; {\bf (2)} allowing the backend to sign the contract before
sending it to the client; {\bf (7)} passing coins received in payment
for a contract to the backend; and, {\bf (8)} executing fulfillment
business logic if the backend confirms the validity of the payment.

To setup a Taler backend, the merchant only needs to configure the
wire transfer routing details, such as the merchant's IBAN number, as
well as a list of acceptable auditors and limits for transaction fees.
Ideally, the merchant might also want to obtain a certificate for the
public key generated by the backend for improved authentication.
Otherwise, the customer's authentication of the Web shop simply
continues to rely upon HTTPS/X.509.


\section{Discussion}

We will now discuss how customer's may experience relevant operational
risks and failure modes of Taler, and relate them to failure modes
in existing systems.

% \smallskip
\subsection{Security risks}

In Taler, customers incur the risk of wallet loss or theft.  We
believe customers can manage this risk effectively because they manage
similar risks of losing cash in a physical wallet.  Unlike physical
wallets, Taler's wallet could be backed up to secure against loss of a
device.  We note that managing the risk does not imply that customers
will never suffer from such a loss.  We expect that customers will
limit the balance they carry in their digital wallet.  Ideally, the
loss should be acceptable given that the customer gains the insight
that their computer was compromised.

Taler's contracts provide a degree of protection for customers,
because they are signed by the merchant and retained by the wallet.
While they mirror the paper receipts that customers receive in
physical stores, Taler's cryptographically signed contracts ought to
carry more weight in courts than typical paper receipts.  Customers
can choose to discard the receipts, for example to avoid leaking their
shopping history in case their computer is compromised.

Point-of-sale systems providing printed receipts have been compromised
in the past by merchants to embezzle sales
taxes.~\cite{munichicecream} With Taler, the merchant still generates
a receipt for the customer, however, the record for the tax
authorities ultimately is anchored with the exchange's wire transfer
to the merchant.  Using the subject of the wire transfer, the state
can trace the payments and request the merchant provide
cryptographically matching contracts.  Thus, this type of tax
fraud is no longer possible, which is why we call Taler {\em
taxable}.  The mere threat of the state sometimes tracing transactions
and contracts back to the merchant also makes Taler unsuitable for
illegal activities.

The exchange operator is obviously crucial for risk management in
Taler, as the exchange operator holds the customer's funds in a
reserve in escrow until the respective deposit request
arrives\footnote{As previously said, this {\it deposit request} is
  aimed to exchange {\it coins} for bank money, and it is made by a
  merchant after successfully receiving coins from a wallet during the
  payment process.} To ensure that the exchange operator does not
embezzle these funds, Taler expects exchange operators to be regularly
audited by an independent auditor\footnote{Auditors are typically run
  by financial regulatory bodies of states.}.  The auditor can then
verify that the incoming and outgoing transactions, and the current
balance of the exchange matches the logs with the cryptographically
secured transaction records.


% \smallskip
\subsection{Failure modes}

There are several failure modes which a customer using a Taler wallet may
encounter:

\begin{itemize}
\item
As Taler supports multiple exchanges, there is a chance that a
merchant might not support any exchange where the customer withdrew
coins from.  We mitigate this problem by allowing merchants to
support all exchanges audited by a particular auditor.  We believe
this a reasonable approach, because auditors and merchants must
operate with a particular legal and financial framework anyways.  We
note that a similar failure mode exists with credit cards where not
all merchants accept all issuers, which is often the case internationally.

\item
Restoring the Taler wallet state from previous backups, or copying the
wallet state to a new machine may cause honest users to attempt to
double spend coins, as the wallet does not know when coins are spent
between backup and recovery.  In this case, the exchange provides
cryptographic proof to the wallet that the coins were previously spent so the
wallet can verify that the exchange and the merchant are behaving honestly.

% FIXME FIXME: the following paragraph seems to describe a scenario where the
% wallet lost coins due to a restore from backup and then ask for refresh
% of lost coins: but how does the wallet know lost coins' public keys?
% CG: I don't understand the problem.
%
% Also in this paragraph: how can a payment end in-flight due to insufficient
% funds? If the payment has been started by the wallet, then no 'insufficient
% funds' may occur, otherwise the wallet would not have started the payment.
%
% CG: Yes, as I explain if the Wallet isn't aware that some coins were
% already spent (I make a backup, spend coins, restore backup, try to
% spend again), then this may happen.
%
% A way to fix that could be to better define 'internal invariants' ..
%
% CG: The internal invariant is exactly the one you fell upon:
% That the wallet knows which coins have been spent!
\item
There could be insufficient funds in the Taler wallet when making a
payment.  Usually the wallet can trivially check this before beginning
a transaction, but when double-spending is detected this may also
happen after the wallet already initiated the payment. This would
usually only happen if the wallet is unaware of a backup operation
voiding its internal invariant of knowing which coins have already
been spent.  If a payment fails in-flight due to
insufficient funds, the wallet can use Taler's refresh protocol to
obtain a refund for those coins that were not actually double-spent,
and then explain to the user that the balance was inaccurate due to
inconsistencies, and insufficient for payment.
For the user, this failure mode appears equivalent to an insufficient
balance or credit line when paying with debit or credit cards.
\end{itemize}

In the future, we plan to make it easy for users to backup and
synchronize wallets to reduce the probability of the later two failure
modes.  A key issue in this context is that these processes will need
to be designed carefully to avoid leaking information that might allow
adversaries to link purchases via side channels opened up by the
synchronization protocol.

\subsection{Comparison}

The different payment systems discussed make use of different security
technologies, which has an impact on their usability and the
assurances they can provide.  Except for Bitcoin, all payment systems
described involve an authentication step.
% FIXME alternative for the following sentence:
% With Taler, the authentication is implicit when withdrawing, since
% the user has to login into his bank's Web portal in the first place,
% and no further authentication is required during the whole payment
% experience.
% CG: Not exactly, as the authentication to the bank is still
% a very explicit authentication step. It's just more natural.
With Taler, the authentication itself is straightforward, as the customer is
at the time visiting the Web portal of the bank, and the authentication is
with the bank (Figure~\ref{fig:taler-withdraw}).  With PayPal, the
shop redirects the customer to the PayPal portal (step 5 in
Figure~\ref{fig:paypal}) after the user selects PayPal as the payment
method. The customer then provides the proof of payment to the
merchant.  Again, this is reasonably natural.  The 3DS workflow
(Figure~\ref{fig:cc3ds}) has to deal with a multitude of banks and
their different implementations, and not just a single provider.
Hence, the interactions are more complicated as the merchant needs to
additionally perform a lookup in the card scheme directory and verify
availability of the bank (steps 8 to 12).

A key difference between Taler and 3DS or PayPal is that
in Taler, authentication is done ahead of time.
After authenticating once to withdraw digital coins, the customer can
perform many micropayments without having to re-authenticate.  While
this simplifies the process of the individual purchase, it shifts the
mental overhead to an earlier time, and thus requires some planning,
especially given that the digital wallet is likely to only contain a
% FIXME line below: which 'funds'? Coins or real money? (If they are
% coins, recall that the wallet withdraw all the coins from a fresh
% reserve, so there is no 'fraction' of user's available funds; at
% least in the current implementation)
% I originally wrote ``wealth'' or ``net value'', but given that
% most customers are in debt today, that makes little sense, so
% I changed it to ``available funds'', but I meant _all_ the money
% he has.
small fraction of the customer's available funds.  As a result, Taler
improves usability if the customer withdraws funds once to
then perform many micropayments, while Taler is likely less usable
if for each transaction the customer first visits the bank to withdraw
funds.  This is {\em deliberate}, as Taler can only achieve reasonable
privacy for customers if they keep a balance in their wallet, as
this is necessary to break the association between withdrawal and deposit.
% FIXME the sentence above can be in contrast with how the exchange
% actually deposits funds to merchants, that is through 'aggregate
% deposits' that may add unpredictable delays (but that doesn't affect
% this article too much)
% CG: I think mentioning aggregation here would distract.

Bitcoin's payment process (Figure~\ref{fig:bitcoin}) resembles that of
Taler in one interesting point, namely that the wallet is given
details about the contract the user enters (steps 7 to 11).
However, in contrast to Taler, Bitcoin wallets are expected
to fetch the ``invoice'' from the merchant. In Taler, the browser
can provide the proposed contract directly to the wallet.  In
PayPal and 3DS, the user is left without a cryptographically secured
receipt.

Card-based payments (including 3DS) and PayPal also extensively rely
on TLS for security.  The customer is expected to verify that their
connections to the various Web sites are properly authenticated using
X.509, and to know that it is fine to provide their bank account
credentials to the legitimate
\url{verifiedbyvisa.com}.\footnote{The search query
``verifiedbyvisa.com legit'' is so common that, when we entered
``verifiedbyvisa'' into a search engine, it was the suggested
auto-completion.}  However, relying on users understanding their
browser's indications of the security context is inherently
problematic.  Taler addresses this challenge by ensuring that digital
coins are only accessible from wallet-generated pages. As such
there is no risk of Web pages mimicking the look of the respective
page, as they would still not obtain access to the digital coins.

Once the payment process nears its completion, merchants need to have
some assurance that the contract is valid.  In Taler, merchants
obtain a non-repudiable confirmation of the payment.  With 3DS and
PayPal, the confirmation may be disputed later (e.g. in case of
fraud), or accounts may be frozen arbitrarily~\cite{diaspora2011}.
Payments in cash require the merchant to assume the risk of receiving
counterfeit money.
% FIXME what about (for the following sentence): merchants should care
% about maintaining change and depositing the money earned
% CG: No, it's not optional, ``should'' doesn't come into the equation
% here. It's a mandatory business expense.
Furthermore, with cash merchants have the cost of maintaining change
and depositing the money earned.  The most extreme case for lack of
assurances upon ``completion'' is Bitcoin, where there is no time
until a payment can be said to be definitively confirmed (step 19,
Figure~\ref{fig:bitcoin}), leaving merchants in a bit of a tricky
situation.

Finally, attempts to address the scalability hudles of Bitcoin using
side-chains or schemes like BOLT introduce semi-centralized
intermediaries, not wholey unlike Taler's use of exchanges.  Compared
to BOLT, we would expect a Taler exchange operating in BTC to offer
stronger security to all parties and stronger anonymity to customers,
as well as being vastly cheaper to operate.


\section{Future work}

This paper has focused on how Taler would work for Web payments.
However, the underlying cryptography should work just as well for
other domains.  In particular, we plan to adapt Taler for NFC and
peer-to-peer payments in the future.

\subsection{NFC payments}

We have so far focused on how Taler could be used for Web payments;
however, Taler can in theory also be used over other protocols, such
as near field communication (NFC).  Here, the user would hold his
NFC-enabled device running a wallet application near an NFC terminal
to obtain the contract and confirm the payment on his device, which
would then transfer the coins and obtain a receipt.  A native NFC
application would be less restricted in its interaction with the
point-of-sale system compared to a browser extension, and the security
of the communication channel is also comparable. Thus, running
Taler over NFC is largely a simplification of the existing process.

In particular, there are no significant new concerns arising from an
NFC device possibly losing contact with a point-of-sale system, as for
Web payments, Taler already only employs idempotent operations to
ensure coins are never lost, and that transactions adequately persist
even in the case of network or endpoint failures.  As a result, the
NFC system can simply use the same transaction models to replay
transmissions once contact with the point-of-sale system is
reestablished.


\subsection{Peer-to-peer payments}

Peer-to-peer payments are in principle possible with Taler as well;
however, we need to distinguish two types of peer-to-peer payments.

First, there is the {\em sharing} of coins among entities that
mutually trust each other, for example within a family.  Here, all
users have to do is to export and import electronic coins over a
secure channel, such as encrypted e-mail or via NFC.  For NFC, the
situation is straightforward because we presumably do not have to worry
about man-in-the-middle attacks, while secure communication over the
Internet is likely to remain a significant usability challenge.  We
note that sharing coins by copying the respective private keys across
devices is not taxable: the exchange is not involved, no contracts are
signed, and no records for taxation are created.  However, the
involved entities must trust each other, because after copying a private
key both parties could try to spend the coins, but only the first
transaction will succeed.  Given this crucial limitation
inherent in sharing keys, we consider it ethically acceptable that
sharing is not taxable.

Second, there is the {\em transactional} mutually exclusive transfer
of ownership.  This requires the receiving party to have a {\em
  reserve} with an exchange, and the sender's exchange would have to
support wire transfers to the receiver's exchange.  If taxability is
desired, the {\em reserve} would still need to be tied to a particular
citizen's identity for tax purposes, and thus require similar
identification protocols as commonly used for establishing a bank
account.  As such, in terms of institutions, one would expect this
setup to be offered most easily by traditional banks, effectively
merging the technical concepts of a (traditional) bank accounts and
Taler reserves into one service for the customer.

In terms of usability, transactional
transfers are just as easy as sharing when performed over NFC, but
more user friendly when performed over the Internet as they do not
require a secure communication channel: the Taler protocol is by
design still safe to use even if the communication is made over an
unencrypted channel. Only the authenticity of the proposed contract
needs to be assured.



\section{Conclusions}

Customers and merchants should be able to easily adapt their existing
mental models and technical infrastructure to Taler.  In contrast,
Bitcoin's payment models fail to match common expectations be it in
terms of performance, durability, security, or privacy.  Minimizing
the need to authenticate to pay fundamentally improves security
and usability.

% FIXME (following paragraph): it's never said that the Taler wallet
% keeps any 'receipt' of transaction -- maybe here we want to say 'contract'
% instead of 'receipt'?
% CG: I'd say on the customer side, the signed contract is a receipt.
% That should be intuitive.
We expect that electronic wallets that automatically collect digitally
signed receipts for transactions will become commonplace.
By providing a free software wallet, Taler gives the user full control
over the usage of their
transaction history, as opposed to giving control to big data corporations.

\begin{center}
  \bf
We encourage readers to try our prototype for Taler
at \url{https://demo.taler.net/}.
\end{center}
%and to ponder why the billion dollar
%e-commerce industry still relies mostly on TLS for security given
%that usability, security and privacy can clearly {\em all} be improved
%simultaneously using a modern payment protocol.

% These APIs are all RESTful in the modern sense because that greatly
% simplify integrating Taler with web shops and browsers.

\section*{Acknowledgements}

This work benefits from the financial support of the Brittany Region
(ARED 9178) and a grant from the Renewable Freedom Foundation.  We
thank Bruno Haible for his financial support enabling us to
participate with the W3c payment working group.  We thank the W3c
payment working group for insightful discussions about Web payments.
We thank Krista Grothoff and Neal Walfield for comments on an earlier
draft of the paper.  We thank Gabor Toth for his help with the
implementation.

\bibliographystyle{splncs03}
\bibliography{ui,btc,taler,rfc}



\end{document}







% \smallskip
\subsection{Anonymity}

We strongly recommend that the user use Tor Browser to protect their
% FIXME wasn't the use of Tor discouraged to login into personal things?
IP address, both initially when withdrawing coins and later during
purchases.

There are lingering risks that anonymous coins can be correlated to
customers using additional information.

After withdrawing coins, customer should usually wait before spending
them, as spending them immediately ....
% wallet determines coin denominations







Wallet provides isolation
  Near copy from EXIST proposal?
- Limits user risk
- Nearly eliminates risk for merchant and exchange
  - lower transaction fees
- Reserves simplify things

Denomination choice
- Anonymity refresh protocol
- Withdraw automates like ATMs

Browser extension
- RESTful vs Bitcoin, OpenCoin, etc.
  - Retrying RESTful transactions always works
- minimizing dialog
- see & pay ??
- TBB integration
  - Needed anyways
- Other browser integration?
  - Is it wise?  Ok if not worried about anonymity  Taler is still better
  - Is tor2web worse?
- W3C

Autopay?  pat payment recognition?
- dangerous?
- high charges
- good for funny money

NFC













% \smallskip
\subsection{Risks}

A Taler exchange's need not face significant financial risks beyond
the risk of losing a denomination signing key.  Exchanges can limit that
risk by carefully tracking how much they issue in each denomination.
Taler merchant's risks are limited primarily by depositing coins
quickly and stating contracts accurately.