11. GNU Taler Auditor Operator Manual

11.1. Introduction

This manual is an early draft that still needs significant editing work to become readable.

11.1.1. About GNU Taler

GNU Taler is an open protocol for an electronic payment system with a free software reference implementation. GNU Taler offers secure, fast and easy payment processing using well understood cryptographic techniques. GNU Taler allows customers to remain anonymous, while ensuring that merchants can be held accountable by governments. Hence, GNU Taler is compatible with anti-money-laundering (AML) and know-your-customer (KYC) regulation, as well as data protection regulation (such as GDPR).

11.1.2. About this manual

This tutorial targets exchange operators, auditors and governments who want to run the auditor to verify that a GNU Taler exchange is operating correctly.

11.1.3. Organizational prerequisites

Operating a GNU Taler auditor means that you (henceforth: auditor) have a business relationship with (or regulatory authority over) a GNU Taler exchange operator (henceforth: exchange). Your objective is to verify that the exchange is operating correctly, and if not to alert the exchange, the state or even the public about any misbehavior to limit financial losses to other parties.

To perform this duty, you will need at least (read-only) access to the bank transactions of the exchange, as well as a continuously synchronized replica of the exchange’s database. The general assumption for running the auditor is that this is done on a separate system controlled by the auditor. After all, the goal is to detect nerfarious activity of the exchange operator, which cannot be effectively done on a machine controlled by the exchange operator.

For this, every auditor needs to operate a Postgres database. The data collected will include sensitive information about Taler users, including withdrawals made by consumers and income received by merchants. As a result, the auditor is expected to provide high confidentiality for the database. In general, the auditor does not have to offer high-availability: the exchange operator can continue operations without the auditor, and the auditor can catch up with it later when the auditor’s systems are restored. However, of course any downtime would provide a window of opportunity for fraud and should thus be minimized. Finally, the auditor’s copy of the exchange’s database can be useful as a backup to the exchange in case the exchange experiences a loss of its own copies. Thus, business agreements between auditor and exchanges may include availability requirements as well.

Then, with the software provided, auditors can verify the cryptographic proofs collected by the exchange and detect if any improper bank transactions have been made. There are additional tasks which an auditor should perform. While this manual only focuses on the audit of the exchange’s database and wire transfers with the existing tools, a proper auditor should also perform the following tasks:

  • security audit of the source code
  • audit of the operational procedures of the exchange
  • audit of the physical security of the deployment
  • background check of the individuals operating the exchange
  • verification that the exchange properly implements the /link protocol (feature yet to be implemented in common Taler wallets)
  • verification that the exchange properly reports coins issued during the refresh protocol (by irregularly refreshing coins withdrawn by the auditor and comparing against the exchange’s database — the code required to support this is not yet implemented)

11.1.4. Architecture overview

Taler is a pure payment system, not a new crypto-currency. As such, it operates in a traditional banking context. In particular, this means that in order to receive funds via Taler, the merchant must have a regular bank account, and payments can be executed in ordinary currencies such as USD or EUR. Similarly, the exchange must interact with a bank. The bank of the exchange holds the exchange’s funds in an escrow account. As a result, exchanges operate in a regulated environment, and auditors provide a crucial oversight function.

Auditors should generally be independent third parties that verify that the exchange operates correctly. However, an exchange is likely to also run the auditing logic, as it is also used to calculate the exchange’s profits, risk and liabilities. Furthermore, it’s usually a good idea to not only rely on third parties to verify one’s own work.

The Taler software stack for an auditor consists of the following components:

  • DBMS: Postgres

    The auditor requires a DBMS to store a local copy of the transaction history for the Taler exchange, as well as for its own internal bookkeeping and checkpointing. The DBMS is assumed to be able to assure the auditor of the database invariants (foreign key, uniqueness, length restrictions). Should the exported data from the exchange fail to be imported due to constraint violations, this is an immediate serious concern that must be addressed manually. The software only verifies the content of a well-formed exchange database (well-formed with respect to SQL). For now, the GNU Taler reference implementation only supports Postgres, but the code could be easily extended to support another DBMS.

  • The auditor Web service

    The auditor is expected to provide a public Web service. At this REST API, merchants can (probabilistically) provide deposit confirmations, allowing the auditor to detect if an exchange is underreporting deposits.

    In the future, the Web service should be extended to allow customers and merchants to automatically upload cryptographic proof of other violations of the specification by the exchange. However, for now it is assumed that the respective cryptographic proofs are reported and verified manually — as with a well-behaved exchange this should obviously be a rare event.

    The main binary of this component is the taler-auditor-httpd.

  • The (main) auditor

    The main auditor logic checks the various signatures, totals up the amounts and checks for arithmetic inconsistencies. It also computes the expected bank balance, revenue and risk exposure of the exchange operator. The main script of this component is the taler-auditor. This script invokes several helper binaries sequentially. Production users may want to modify the script to run those binaries in parallel, possibly using different privileges (as only the taler-helper-auditor-wire needs access to the wire gateway).

    The taler-helper-auditor-wire auditor verifies that the bank transactions performed by the exchange were done properly. This component must have access to the bank account of the exchange, as well as to a copy of the exchange’s database.

    The taler-auditor script invokes the various helpers, each generating a JSON report. It then invokes the taler-helper-auditor-render.py script to combine those JSON files with a Jinja2 template into a LaTeX report. Finally, pdflatex is used to generate a PDF report.

    The resulting report includes performance data, reports on hard violations (resulting in financial losses) and reports on soft violations (such as the exchange not performing certain operations in a timely fashion). The report also includes figures on the losses of violations. Careful reading of the report is required, as not every detail in the report is necessarily indicative of a problem.

11.2. Installation

11.2.1. Installing from source

Please install the following packages before proceeding with the exchange compilation.

  • Python3 module jinja2
  • libsqlite3 >= 3.16.2
  • GNU libunistring >= 0.9.3
  • libcurl >= 7.26 (or libgnurl >= 7.26)
  • libqrencode >= 4.0.0
  • GNU libgcrypt >= 1.6
  • libsodium >= 1.0
  • libargon2 >= 20171227
  • libjansson >= 2.7
  • PostgreSQL >= 9.6, including libpq
  • GNU libmicrohttpd >= 0.9.71
  • GNUnet >= 0.15.2 (from source tarball)
  • GNU Taler exchange (from download directory, see release announcement)

Except for the last two, these are available in most GNU/Linux distributions and should just be installed using the respective package manager.

The following instructions will show how to install libgnunetutil and the exchange (which includes the code for the auditor).

Before you install GNUnet, you must download and install the dependencies mentioned in the previous section, otherwise the build may succeed, but could fail to export some of the tooling required by GNU Taler.

To install GNUnet, unpack the tarball and change into the resulting directory, then proceed as follows:

$ ./configure [--prefix=GNUNETPFX]
$ # Each dependency can be fetched from non standard locations via
$ # the '--with-<LIBNAME>' option. See './configure --help'.
$ make
# make install
# ldconfig

If you did not specify a prefix, GNUnet will install to /usr/local, which requires you to run the last step as root. The ldconfig command (also run as root) makes the shared object libraries (.so files) visible to the various installed programs.

After installing GNUnet, unpack the GNU Taler exchange tarball, change into the resulting directory, and proceed as follows:

$ ./configure [--prefix=EXCHANGEPFX] \
              [--with-gnunet=GNUNETPFX]
$ # Each dependency can be fetched from non standard locations via
$ # the '--with-<LIBNAME>' option. See './configure --help'.
$ make
# make install

If you did not specify a prefix, the exchange will install to /usr/local, which requires you to run the last step as root. You have to specify --with-gnunet=/usr/local if you installed GNUnet to /usr/local in the previous step.

Please note that unlike most packages, if you want to run the make check command, you should run it only after having done make install. The latter ensures that necessary binaries are copied to the right place.

Gratuitous editorial note by TTN: I think this is a quirk that we should fix in the long-term as such weirdness might hide other build issues. However, this is probably a minority viewpoint.

In any case, if make check fails, please consider filing a bug report with the Taler bug tracker.

11.2.2. Installing the GNU Taler binary packages on Debian

To install the GNU Taler Debian packages, first ensure that you have the right Debian distribution. At this time, the packages are built for Bullseye.

You need to add a file to import the GNU Taler packages. Typically, this is done by adding a file /etc/apt/sources.list.d/taler.list that looks like this:

deb https://deb.taler.net/apt/debian bullseye main

Next, you must import the Taler Systems SA public package signing key into your keyring and update the package lists:

# wget -O - https://taler.net/taler-systems.gpg.key | apt-key add -
# apt update

Note

You may want to verify the correctness of the Taler Systems key out-of-band.

Now your system is ready to install the official GNU Taler binary packages using apt.

To install the Taler auditor, you can now simply run:

# apt install -t sid taler-auditor

For the auditor, you must manually configure access to the exchange database, the HTTP reverse proxy (typically with TLS certificates) and offline signing.

Sample configuration files for the HTTP reverse proxy can be found in /etc/taler-auditor/.

11.2.3. Installing the GNU Taler binary packages on Ubuntu

To install the GNU Taler Ubuntu packages, first ensure that you have the right Ubuntu distribution. At this time, the packages are built for Ubuntu 20.04 LTS (Focal Fossa).

A typical /etc/apt/sources.list.d/taler.list file for this setup would look like this:

deb https://deb.taler.net/apt/ubuntu/ focal-fossa main

The last line is crucial, as it adds the GNU Taler packages.

Next, you must import the Taler Systems SA public package signing key into your keyring and update the package lists:

# wget -O - https://taler.net/taler-systems.gpg.key | apt-key add -
# apt update

Note

You may want to verify the correctness of the Taler Systems key out-of-band.

Now your system is ready to install the official GNU Taler binary packages using apt.

To install the Taler exchange, you can now simply run:

# apt install -t focal-fossa taler-auditor

For the auditor, you must manually configure access to the exchange database, the HTTP reverse proxy (typically with TLS certificates) and offline signing.

Sample configuration files for the HTTP reverse proxy can be found in /etc/taler-auditor/.

11.3. System setup

11.3.1. UNIX accounts

For maximum security, you should setup multiple different users (possibly on different machines) to run Taler auditor components. While it is possible to skip some of these entirely, or to run all of them as the same user, this is not recommended for security. The recommended set of users includes:

  • auditor — runs the main auditing process and HTTP backend
  • sync — synchronizes the ingres database with the production database
  • helper — runs taler-auditor-offline download and upload commands
  • auditor-ingres — imports database from exchange production system
  • auditor-wire — imports wire transfer data from bank production system
  • offline — manages the offline key, on a separate offline machine

It is suggested that you setup the first five users on the target system(s) using:

# add-user --disabled-password $USERNAME

Additionally, there are two canonical system users of relevance (which your distribution would typically create for you):

  • www-data — runs the HTTPS frontend (usually nginx or Apache)
  • postgres — runs the Postgres database

11.3.2. Databases and users

We recommend using the following databases for the auditor:

  • exchange-ingres — synchronized exchange database over the network
  • exchange-production — local copy of exchange database with trusted schema
  • auditor — auditor production database with current state of the audit
  • libeufin — local state of the auditor-wire tool for the bank transfer data import

As the postgres user, you can create these databases using:

# As the 'postgres' user:
$ createdb -O auditor-ingres exchange-ingres
$ createdb -O sync exchange-production
$ createdb -O auditor auditor
$ createdb -O auditor-wire libeufin

This will ensure that the correct users have write-access to their respective database. Next, you need to grant read-only access to some users to databases owned by other users:

# As the 'auditor-ingres' user:
$ echo 'GRANT SELECT ON ALL TABLES IN SCHEMA public TO sync;' | psql exchange-ingres
# As the 'sync' user:
$ echo 'GRANT SELECT ON ALL TABLES IN SCHEMA public TO auditor;' | psql exchange-production
# As the 'auditor-wire' user:
$ echo 'GRANT SELECT ON ALL TABLES IN SCHEMA public TO auditor;' | psql libeufin

11.4. Configuration

The auditor’s configuration works the same way as the configuration of other Taler components. This section discusses configuration options related to the auditor.

11.4.1. Configuration format

In Taler realm, any component obeys to the same pattern to get configuration values. According to this pattern, once the component has been installed, the installation deploys default values in ${prefix}/share/taler/config.d/, in .conf files. In order to override these defaults, the user can write a custom .conf file and either pass it to the component at execution time, or name it taler.conf and place it under $HOME/.config/.

A config file is a text file containing sections, and each section contains its values. The right format follows:

[section1]
value1 = string
value2 = 23

[section2]
value21 = string
value22 = /path22

Throughout any configuration file, it is possible to use $-prefixed variables, like $VAR, especially when they represent filesystem paths. It is also possible to provide defaults values for those variables that are unset, by using the following syntax: ${VAR:-default}. However, there are two ways a user can set $-prefixable variables:

by defining them under a [paths] section, see example below,

[paths]
TALER_DEPLOYMENT_SHARED = ${HOME}/shared-data
..
[section-x]
path-x = ${TALER_DEPLOYMENT_SHARED}/x

or by setting them in the environment:

$ export VAR=/x

The configuration loader will give precedence to variables set under [path], though.

The utility taler-config, which gets installed along with the exchange, serves to get and set configuration values without directly editing the .conf. The option -f is particularly useful to resolve pathnames, when they use several levels of $-expanded variables. See taler-config --help.

Note that, in this stage of development, the file $HOME/.config/taler.conf can contain sections for all the component. For example, both an exchange and a bank can read values from it.

The repository git://taler.net/deployment contains examples of configuration file used in our demos. See under deployment/config.

Note

Expectably, some components will not work just by using default values, as their work is often interdependent. For example, a merchant needs to know an exchange URL, or a database name.

11.4.2. Using taler-config

The tool taler-config can be used to extract or manipulate configuration values; however, the configuration use the well-known INI file format and can also be edited by hand.

Run

$ taler-config -s $SECTION

to list all of the configuration values in section $SECTION.

Run

$ taler-config -s $section -o $option

to extract the respective configuration value for option $option in section $section.

Finally, to change a setting, run

$ taler-config -s $section -o $option -V $value

to set the respective configuration value to $value. Note that you have to manually restart the Taler backend after you change the configuration to make the new configuration go into effect.

Some default options will use $-variables, such as $DATADIR within their value. To expand the $DATADIR or other $-variables in the configuration, pass the -f option to taler-config. For example, compare:

$ taler-config -s ACCOUNT-bank \
               -o WIRE_RESPONSE
$ taler-config -f -s ACCOUNT-bank \
               -o WIRE_RESPONSE

While the configuration file is typically located at $HOME/.config/taler.conf, an alternative location can be specified to taler-merchant-httpd and taler-config using the -c option.

11.4.3. Initial configuration

You need to tell the Taler auditor configuration where the REST API of the auditor will be available to the public:

# Both for the 'offline' *and* the 'auditor' user:
$ taler-config -s auditor -o BASE_URL -V https://auditor.example.com/

The helper user that is used to download information from the exchange needs to know details about the exchange. Similarly, the offline user needs to check signatures signed with the exchange’s offline key. Hence, you need to obtain the MASTER_PUBLIC_KEY from the exchange operator (they need to run taler-exchange-offline setup) and the REST endpoint of the exchange and configure these:

# As the 'helper' and 'offline' users:
$ taler-config -s exchange -o BASE_URL -V https://exchange.example.com/
$ taler-config -s exchange -o MASTER_PUBLIC_KEY -V $SOMELONGBASE32VALUEHERE

11.4.4. Keys

The auditor works with one signing key to certify that it is auditing a particular exchange’s denomination keys. This key can and should be kept offline (and backed up adequately). As with the exchange’s offline key, it is only used for a few cryptographic signatures and thus the respective code can be run on modest hardware, like a Raspberry Pi.

The following values are to be configured in the section [auditor]:

  • AUDITOR_PRIV_FILE: Path to the auditor’s private key file.

Note that the default value here should be fine and there is no clear need to change it. What you do need to do as the offine user is to extract the public key:

# As the 'offline' user:
$ taler-auditor-offline setup

This public key must then be provided in the configuration file of the auditor user in the [auditor]] configuration section:

  • PUBLIC_KEY: Public key of the auditor, in Base32 encoding. Set from value printed by taler-auditor-offline setup.

You can set this configuration value using:

# As the 'auditor' and 'helper' users:
$ taler-config -s auditor -o PUBLIC_KEY -V $SOMELONGBASE32VALUEHERE

11.4.5. Configuring the auditor’s REST endpoint

The auditor can serve HTTP over both TCP and UNIX domain socket.

The following values are to be configured in the section [auditor]:

  • serve: must be set to tcp to serve HTTP over TCP, or unix to serve HTTP over a UNIX domain socket
  • port: Set to the TCP port to listen on if serve is tcp.
  • unixpath: set to the UNIX domain socket path to listen on if serve is unix
  • unixpath_mode: number giving the mode with the access permission MASK for unixpath (i.e. 660 = rw-rw----).

11.4.6. Bank account

Bank accounts for the auditor (user auditor-wire) are configured in exactly the same way as bank accounts for the exchange. See the exchange (and LibEuFin) documentation for details.

11.4.7. Database

The option DB under section [auditor] gets the DB backend’s name the exchange is going to use. So far, only DB = postgres is supported. After choosing the backend, it is mandatory to supply the connection string (namely, the database name). This is possible in two ways:

  • via an environment variable: TALER_AUDITORDB_POSTGRES_CONFIG.

  • via configuration option CONFIG, under section [auditordb-$BACKEND]. For example, the demo exchange is configured as follows:

    [auditor]
    ...
    DB = postgres
    ...
    
    [auditordb-postgres]
    CONFIG = postgres:///auditordemo
    

If an exchange runs its own auditor, it may use the same database for the auditor and the exchange itself.

The taler-auditor-dbinit tool is used to initialize the auditor’s tables. After running this tool, the rights to CREATE or DROP tables are no longer required and should be removed.

Both the taler-auditor-httpd and the taler-auditor (and its helpers) also need (read-only) access to a (recent, current, synchronized) copy of the exchange’s database. The configuration options are the same that are also used when configuring the exchange’ database:

[exchange]
...
DB = postgres
...

[exchangedb-postgres]
CONFIG = postgres:///exchangedemo

11.5. Deployment

Before GNU Taler wallets will happily interact with an exchange, the respective auditor’s public key (as obtained via taler-auditor-offline setup from the offline user) must be added under the respective currency to the wallet. This is usually expected to be hard-coded into the Taler wallet.

Users can also manually add auditors for a particular currency via a Web page offering the respective pairing.

FIXME-DOLD: explain how that Web page works, once it works…

11.5.1. Exchange

The next step is to add the exchange’s master public key and the base URL of the exchange to the list of exchanges audited by the auditor. This is done using the taler-auditor-exchange tool. The tool basically creates the respective record in the auditor’s database.

If this step is skipped, the auditor will malfunction at all future stages with a foreign key violation, as it does not know the exchange’s master public key.

# As the 'auditor' user:
$ taler-auditor-exchange -m $MASTER_PUB -u $EXCHANGE_BASE_URL

An equivalent step must be performed by the exchange operator. Here, the exchange operator must use the taler-exchange-offline tool to add the auditor’s public key, base URL and (business) name to the list of approved auditors of the exchange. For details, see Auditor configuration in the exchange operator manual.

11.5.2. Signing Denominations

These steps must be performed regularly whenever the exchange is deploying new denomination keys. After the exchange operator has signed new keys using the taler-exchange-offline tool, each auditor should run:

# As the 'helper' user:
$ taler-auditor-offline download > input.json

to import the latest set of denomination keys. The key data should then be copied to the offline system and there be inspected using:

# As the 'offline' user:
$ taler-auditor-offline show < input.json

and compared with the data the exchange operator saw when doing the offline signature. This process should involve directly the humans operating both systems and may require them to establish a trustworthy connection. The details how the auditor communicates with the exchange operator are a business process that is outside of the scope of this document.

Note that the input.json does not contain any confidential data. However, signing the wrong keys would be fatal in that it may allow an illegitimate exchange to convince users that it is a trustworthy operator and subsequently betray the user’s trust that is anchored in the existence of a trustworthy auditor.

Given the verified JSON input, the auditor can then sign it (typically on its offline system) using:

# As the 'offline' user:
$ taler-auditor-offline sign < input.json > output.json

The resulting output.json should then be copied to an online system, and from there uploaded to the exchange using:

# As the 'helper' user:
$ taler-auditor-offline upload < output.json

The contents of output.json can again be public and require no special handling.

If the auditor has been correctly added, the exchange’s /keys response will contain an entry in the auditors array mentioning the auditor’s URL.

Commands, like taler-auditor-offline, that support the -l LOGFILE command-line option, send logging output to standard error by default.

11.5.3. Database

The next key step for the auditor is to configure replication of the exchange’s database in-house. This should be performed in two steps as illustrated in the following figure:

_images/replication.png

First, the exchange should use standard Postgres replication features to enable the auditor to obtain a full copy of the exchange’s database. Second, the auditor should make a “trusted” local copy, ensuring that it never replicates malicious changes using taler-auditor-sync. Both of these steps are described in more detail below.

We note that as a result of these steps, the auditor will have three databases: its own production primary database (as configured in auditordb-postgres), its on production copy of the exchange’s database (exchangedb-postgress), and a third, untrusted “ingres” copy of the exchange database. The untrusted database should run as a separate Postgres instance and is only accessed via taler-auditor-sync and the replication mechanism driven by the exchange operator.

Ingres replication of the exchange production database

Ingres operation should be done using the auditor-ingres user — or depending on the setup parts of the operation may be done by the postgres user directly.

The full copy can be obtained in various ways with Postgres. It is possible to use log shipping with streaming replication as described in https://www.postgresql.org/docs/13/warm-standby.html, or to use logical replication, as described in https://www.postgresql.org/docs/13/logical-replication.html. We note that asynchronous replication should suffice.

The resulting auditor database should be treated as read-only on the auditor side. The taler-exchange-dbinit tool can be used to setup the schema, or the schema can be replicated using Postgres’s standard mechanisms. The same applies for schema upgrades: if logical replication is used (which does not replicate schema changes), taler-exchange-dbinit can be used to migrate the schema(s) in both the ingres and production copies of the exchange’s database as well.

On the exchange side, a database user must be created that has the right to perform database replication. This is done using:

# As the 'postgres' user of the exchange:
$ createuser --replication egress
$ echo "ALTER ROLE egress WITH PASSWORD '$PASSWORD'; | psql
$ echo "CREATE PUBLICATION $NAME FOR ALL TABLES;" | psql taler-exchange

The exchange must share the password of the publication with the auditor. A good $NAME relates to the auditor’s business unit name. A secure tunnel must be setup between the exchange and the auditor, for example using SSH or Wireguard.

It is also necessary to edit main.cf of the exchange and on the auditor side to enable logical replication. If an exchange has multiple auditors, it should setup multiple egress accounts. The exchange must ensure that the following lines are in the main.cf Postgres configuration (the port may differ) to enable replication over the network:

listen_addresses='*'
port = 5432
wal_level= logical

Equally, the auditor must configure logical replication in the main.cf Postgres configuration:

wal_level= logical

Next, the postgres user of the auditor’s system must first initialize the local tables:

# As the 'ingress' user of the exchange:
$ taler-config -s exchange -o DB -V "postgres"
$ taler-config -s exchangedb-postgres -o CONFIG -V "postgres:///taler-ingress"
$ taler-exchange-dbinit

To complete the replication, the postgres user of the auditor’s system must subscribe:

# As the 'postgres' user of the exchange:
$ createuser --replication egress
$ echo "ALTER ROLE egress WITH PASSWORD '$PASSWORD'; | psql
$ echo "CREATE PUBLICATION $NAME FOR ALL TABLES;" | psql taler-exchange

For details, we refer to the Postgres manual.

Note

Depending on the replication method used, the exchange may perform unexpected changes to the schema or perform UPDATE, DELETE or DROP operations on the tables. Hence, the auditor cannot rely upon the exchange’s primary copy to respect schema constraints, especially as we have to presume that the exchange could act maliciously. Furthermore, it is unclear to what degree Postgres database replication mechanisms are robust against a malicious master database. Thus, the auditor should isolate its primary copy of the exchange database, including the Postgres process, from its actual operational data.

Safe replication of the ingres database into the auditor production database

Using taler-auditor-sync as the sync user, the auditor should make a second “safe” copy of the exchange’s ingres database. taler-auditor-sync basically reads from one exchange database and inserts all records found into a second exchange database. If the source database violates invariants, the tool halts with an error. This way, records violating invariants are never even copied, and in particular schema changes and deletions or updates are not propagated into the auditor’s production database.

While taler-auditor-sync could in theory be run directly against the exchange’s production system, this is likely a bad idea due to the high latency from the network between auditor and exchange operator. Thus, we recommend first making an “untrusted” ingress copy of the exchange’s production database using standard Postgres tooling, and then using taler-auditor-sync to create a second “safe” copy. The “safe” copy used by the production system should also run under a different UID.

Before taler-auditor-sync can be used, the target database must be initialized with the exchange schema using taler-exchange-dbinit. Note that running taler-auditor-sync requires the use of two configuration files, one specifying the options for accessing the source database, and a second with the options for accessing the destination database. In both cases, likely only the [exchangedb]/CONFIG option needs to be changed.

To run taler-auditor-sync, you must first configure two configuration files that identify the source and destination databases:

# As the 'sync' user:
$ taler-config -c src.conf -s exchangedb -o CONFIG -V "postgres:///auditor-ingres/"
$ taler-config -c dst.conf -s exchangedb -o CONFIG -V "postgres:///auditor/"

Now you should be able to launch the synchronization process. You can run the process via systemd in the background. For a first one-off test, you should use the -t option which will cause the process to terminate once the two databases are synchronized:

# As the 'sync' user:
$ taler-auditor-sync -s src.conf -d dst.cfg -t

When the exchange performs garbage collection to DELETE obsolete records, this change should be automatically replicated to the auditors untrusted ingress database. However, as taler-auditor-sync tries to be “safe”, it will not replicate those deletions to the auditor’s production database. Thus, it is necessary to (occasonally) run taler-exchange-dbinit -g on the auditor’s production database to garbage collect old data in the auditor’s production copy. We note that this does not have to be done at the same time when the exchange runs its garbage collection.

11.6. Operation

11.6.1. Web service

The taler-auditor-httpd runs the required REST API for the auditor. The service must have INSERT (and SELECT) rights on the deposit_confirmations table in the auditor’s database. We expect that in future versions additional rights may be required.

For now, we recommend simply running the taler-auditor-httpd under the auditor user. However, it is also possible (and might be more secure) to create a separate user with more restrictive permissions for this purpose.

As the taler-auditor-httpd does not include HTTPS-support, it is advisable to run it behind a reverse proxy that offers TLS termination.

11.6.2. Audit

Performing an audit is done by invoking the taler-auditor shell script as the auditor user.

The shell script invokes the various helper processes. For additional performance and security, one may want to run the various helpers individually and with the respective minimal set of access rights (only taler-helper-auditor-wire needs the credentials to query the bank for wire transfers, alas if auditor-wire is used to talk to the bank, this issue is already addressed). The shell script combines the final JSON outputs of the various helpers using the taler-helper-auditor-render.py script into the TeX report. Regardless, the simplest way to obtain a report is to run:

$ taler-audit

This generates a file auditor-report.pdf (in a temporary directory created for this purpose) with all of the issues found and the financial assessment of the exchange. The exact filename will be output to the console upon completion.

We note that taler-audit by default runs in incremental mode. As a result, running the commands again will only check the database entries that have been added since the last run.

You can use taler-auditor-dbinit -r to force a full check since the beginning of time. However, as this may require excessive time and interactions with the bank (which may not even have the wire transfer records anymore), this is not recommended in a production setup.

11.6.3. Reading the report

The auditor’s report needs to be read carefully, as it includes several categories of failures of different severity:

  • Delayed operations, where an operation was expected to have happened, but did not happen yet, possibly because of a disagreement in system time or overloading of the system. These failures only require action if the delays are significant.
  • Inconsistencies in the data that have no clear financial impact.
  • Inconsistencies in the data that show that the exchange experienced an unexpected financial loss (such as accepting a coin for deposit with an invalid signature).
  • Inconsistencies in the data that show that the exchange caused some other party to experience a financial loss (such as not wiring the correct amount to a merchant).
  • Configuration issues (such was wire fees unavailable).

11.6.4. Database upgrades

To upgrade the database between Taler versions can be done by running:

$ taler-auditor-dbinit
$ taler-exchange-dbinit

In any case, it is recommended that exchange and auditor coordinate closely during schema-changing database upgrades as without coordination the database replication or taler-auditor-sync will likely experience problematic failures. In general, we recommend:

  • halting the exchange business logic,
  • allowing the replication and taler-auditor-sync to complete (see also the -t option of taler-auditor-sync)
  • completing a taler-audit run against the old schema
  • migrating the exchange schema (taler-exchange-dbinit) of the master database, possibly the ingres database and the auditor’s production copy
  • migrating the auditor database (taler-auditor-dbinit)
  • resuming database replication between the exchange’s master database and the auditor’s ingres copy
  • resuming taler-auditor-sync
  • resuming the regular exchange and auditor business logic

Regardless, the above is merely the general rule. Please review the specific release notes to ensure this procedure is correct for the specific upgrade.

11.6.5. Database reset

The auditor database can be reset using:

$ taler-auditor-dbinit -R

However, running this command will result in all data in the database being lost, including steps like enabling an exchange using taler-auditor-exchange. Thus, doing so may result in significant commputation (and bandwidth consumption with the bank) when the auditor is next launched, as it will re-download and re-verify all historic transactions. Hence this should not be done in a production system.

11.6.6. Revocations

When an auditor detects that the private key of a denomination key pair has been compromised, one important step is to revoke the denomination key. The exchange operator includes the details on how to revoke a denomination key, so the auditor should only have to report (and possibly enforce) this step. For more information, see Revocations in the exchange operator manual.

If all denominations of an exchange are revoked, the exchange includes logic to wire back all returned funds to the bank accounts from which they originate. If some denominations remain operational, wallets will generally exchange old coins of revoked denominations for new coins – while providing additional information to demonstrate that these coins were not forged from the compromised private key but obtained via a legitimate withdraw operation.

11.6.7. Failures

Most audit failures are handled by the auditor’s regular reporting functionality, creating a (hopefully descriptive) PDF report detailing the problems found.

However, there is one category of errors where this is not possible, which concerns arithmetic overflows for amounts. Taler’s specification limits amount values to at most 2^52. If, during the auditor’s calculations, amounts are encountered that exceed this threshold, the auditor will not generate a regular report, but instead write a log statement explaining where the problem happened and exit with a status code of 42.

The most common expected case when this happens is a corrupted database. This could be because the exchange is actively malicious, or more likely due to some data corruption. The audit cannot continue until the corruption has been addressed. If it is not possible to restore a fully correct version of the database, the suggestion is to replace the corrupted (and likely very large) amounts with zero (Note: this does not apply to the value of denominations or fees, here it is crucial that the correct amounts are restored). While an amount of zero would be incorrect, the auditing logic should be able to do its calculations with zero instead.

After patching the database, the audit can be restarted. A full reset is not required, as the audit transaction is aborted when the auditor exits with code 42. After restarting, the resulting audit report is likely to indicates errors relating to the corrupted fields (such as invalid signatures, arithmetic errors by the exchange, etc.), but at least the loss/gain calculations will be meaningful and actually indicative of the scope of the error created by the corrupted data.

11.7. Auditor implementation guide

The auditor implementation is split into five main processes, called taler-helper-auditor-XXX. The split was done to realize the principle of least privilege and to enable independent logic to be possibly run in parallel. Only the taler-wire-auditor must have (read-only) access to the exchange’s bank account, the other components only need access to the database.

All auditor subsystems basically start their audit from a certain transaction index (BIG SERIAL) in the auditor database which identifies where the last audit concluded. They then check that the transactions claimed in the exchange’s database match up internally, including the cryptographic signatures and also with respect to amounts adding up. The auditor also calculates the exchange’s profits and expected bank balances. Once all existing transactions are processed, the auditor processes store the current checkpoint in its database and generate a JSON report.

The taler-auditor shell script calls the five helpers and then uses Jinja2 with a TeX template to convert the five individual JSON reports into LaTeX and then into PDF.

11.7.1. The auditor’s database

The database scheme used by the exchange looks as follows:

_images/auditor-db.png

11.7.2. Invariants checked by the auditor

The auditor verifies a large number of invariants that must hold for a Taler exchange. One objective in the design of the auditor was to check each invariant only once, both to minimize cost and to avoid duplicate reporting of problems where possible. As a result, not every invariant is checked in every pass where it might seem applicable.

Invariants checked by the taler-helper-auditor-aggregation

This is from CodeBlau’s analysis. A proper write-up is pending. CodeBlau reports the following checks:

  • arithmetic inconsistencies
    • disagreement in fee for deposit between auditor and exchange db
    • disagreement in fee for melt between auditor and exchange db
    • disagreement in fee for refund between auditor and exchange db
    • aggregation of fee is negative
    • aggregation (contribution): Expected coin contributions differ: coin value without fee, total deposit without refunds
    • wire out fee is negative
  • coin arithmetic inconsistencies
    • refund (merchant) is negative
    • refund (balance) is negative
    • spend > value
  • coin denomination signature invalid
  • start date before previous end date
  • end date after next start date
  • wire out inconsistencies in amount
  • row inconsistencies
    • wire account given is malformed
    • h(wire) does not match wire
    • failed to compute hash of given wire data
    • database contains wrong hash code for wire details
    • no transaction history for coin claimed in aggregation
    • could not get coin details for coin claimed in aggregation
    • could not find denomination key for coin claimed in aggregation
    • coin denomination signature invalid
    • target of outgoing wire transfer do not match hash of wire from deposit
    • date given in aggregate does not match wire transfer date
    • wire fee signature invalid at given time
    • specified wire address lacks method
    • wire fee unavailable for given time

Invariants checked by the taler-helper-auditor-coins

This is from CodeBlau’s analysis. A proper write-up is pending. CodeBlau reports the following checks:

  • check that all denominations used by the exchange have been signed using this auditor’s key. All denominations encountered in the database that this auditor did not officially sign for are reported (but still included in the audit as they obviously may impact the exchange’s bank balance). Depending on the business situation, this may be normal (say if an exchange is changing auditors and newer denominations are no longer supported until their end-of-life by the current auditor).
  • emergency on denomination over loss
    • value of coins deposited exceed value of coins issued
  • emergency on number of coins, num mismatch
  • arithmetic inconsistencies
    • melt contribution vs. fee
    • melt (cost)
    • refund fee
  • row inconsistencies
    • revocation signature invalid
    • denomination key not found
    • denomination key for fresh coin unknown to auditor
    • denomination key for dirty coin unknown to auditor
    • denomination key for deposited coin unknown to auditor
  • coin validity in known_coin, by checking denomination signatures
  • coin validity in melt, by checking signatures
  • refresh hanging, zero reveals (harmless)
  • verify deposit signature
  • verify refund signature
  • recoup, check coin
  • recoup, check signature
  • recoup, denomination not revoked

Invariants checked by the taler-helper-auditor-deposits

This tool verifies that the deposit confirmations reported by merchants directly to the auditor are also included in the database duplicated from the exchange at the auditor. This is to ensure that the exchange cannot defraud merchants by simply not reporting deposits to the auditor.

Invariants checked by the taler-helper-auditor-reserves

This is from CodeBlau’s analysis. A proper write-up is pending. CodeBlau reports the following checks:

  • report arithmetic inconsistency
    • closing aggregation fee
    • global escrow balance
  • denomination key validity withdraw inconsistencies
  • bad signature losses in withdraw
  • bad signature losses in recoup
  • bad signature losses in recoup-master
  • reserve balance, insufficient, losses and gains
  • reserve balance, summary wrong
  • reserve not closed after expiration time
  • could not determine closing fee / closing-fee unavailable
  • denomination key not found for withdraw
  • denomination key not in revocation set for recoup
  • target account not verified, auditor does not know reserve
  • target account does not match origin account

Invariants checked by the taler-helper-auditor-wire

This auditor is special in that it is the only pass that is required to have read-only access to the exchange’s bank account (privilege separation). Its main role is to verify that the wire transfers in the exchange’s database and those reported by the bank are identical.

This is from CodeBlau’s analysis. A proper write-up is pending. CodeBlau reports the following checks:

  • check pending
  • wire missing
  • execution date mismatch
  • wire out consistency
  • wire transfer not made (yet?)
  • receiver account mismatch
  • wire amount does not match
  • justification for wire transfer not found
  • duplicate subject hash
  • duplicate wire offset
  • incoming wire transfer claimed by exchange not found
  • wire subject does not match
  • wire amount does not match
  • debit account url does not match
  • execution date mismatch
  • closing fee above total amount

11.7.3. Testing the auditor

The main objective of the auditor is to detect inconsistencies. Thus, the test-auditor.sh script deliberately introduces various inconsistencies into a synthetic exchange database. For this, an “normal” exchange database is first generated using the taler-wallet-cli. Then, various fields or rows of that database are manipulated, and the auditor is let loose on the modified database. Afterwards, the test verifies that the JSON contains values indicating that the auditor found the inconsistencies. The script also verifies that template expansion and LaTeX run work for the JSON output, but it does not verify the correctness of the final PDF.

The test-auditor.sh script is written to maximize code coverage: it should cover as many code paths as possible in both the exchange and the auditor. It should also ideally create all interesting possible variations of the exchange database fields (within the constraints of the database schema).

In general, test-auditor.sh runs the tests against an “old” database where some transactions are past the due-date (and hence the aggregator would trigger wire transfers), as well as a freshly generated exchange database where the auditor would not perform any transfers. Auditor interactions can be made before or after the aggregator, depending on what is being tested.

The current script also rudimentarily tests the auditor’s resume logic, by re-starting the auditor once against a database that the auditor has already seen.

The test-revocation.sh script performs tests related to the handling of key revocations.

The test-sync.sh script performs tests related to the taler-auditor-sync tool.