Data Discovery and Anonymization toolkit

Table of content

• Purpose
• Features
• Prerequisites
• Build from source
• Including JDBC Drivers
• Extensions
• Contributing
• How to run
• Using argument files
• File Discovery
• Column Discovery
• Data Discovery
• Data Extractor
• Anonymizer
• Logging (and database logging)
• Upgrading to 2.0
• Features and issues
• Code quality

Purpose

While performing application development, testing, or maintenance, it is important to operate in an environment that is as close to the production environment as possible when it comes to the amount of data and close-to-real content. At the same time it is important to ensure that data privacy policies are not violated.

Database, column, and file discovery identify and analyze data risks and report on potentially identifiable and personal information stored. And the database anonymization process anonymizes sensitive data and transfer information between organizations, while reducing the risk of unintended disclosure.

The complete source code is available, so you can inspect it and perform security audits if necessary.

This implementation of Data Discovery program is using Apache OpenNLP

Features

1. Identifies sensitive personal data.
2. Creates plan (XML document) to define what columns should be anonymized and how.
3. Anonymizes the data.
4. Platform-independent.
5. Supports Oracle, MariaDB/MySQL, MS SQL Server, and PostgreSQL. Work in progress for DB2.
6. This tool can help you be GDPR-compliant.

Prerequisites

1. JDK 11+
2. Maven 3+

Build from source

1. Download ZIP file and unzip in a directory of your choice, or clone repo
2. cd {dir}/DataDefender/
3. mvn package
4. DataDefender.jar will be located in "target" directory {dir}/DataDefender/target/

Including JDBC Drivers

JDBC drivers are included as optional dependencies included in maven profiles that can be activated. Valid options are:

• mariadb
• mysql
• sqlserver
• postgresql
• oracle

In addition, a property to activate all drivers is available as well for convenience:

• jdbc-drivers-all

Example builds:

mvn package -P mariadb,mysql
mvn package -Djdbc-drivers-all
mvn package -P oracle

Alternatively, the JDBC drivers can be included as jar files in a 'lib' folder under your project folder (where the jar and scripts are copied to).

Note: sqlite-jdbc is included always for file discovery.

Extensions

Additional jar files/classes can be added under an 'extensions' directory in the current working directory. The default 'datadefender' scripts copied to the target directory adds classes/jar files under 'extensions' to the classpath. The 'extensions' directory is meant to house extensions for a project, for example additional anonymization or discovery routines, etc... additional libraries required may be included more appropriately in a 'lib' directory.

See sample_projects/anonymizer/ for an example.

Contributing

We encourage you to contribute to DataDefender! Please check out the Contribution guidelines for this project.

How to run

The toolkit is implemented as a command line program. To run it first build the application as above (mvn package). This will generate an executable jar file in the "target" directory. For your convenience executable 'sh' and 'bat' files are created as well. You may need to adjust permissions for the executable shell script (chmod +x datadefender). Once this has been done you can get help by running 'datadefender' or 'datadefender.bat' in your shell/command prompt:

datadefender help

Usage: datadefender [-hvV] [--debug] COMMAND
Data detection and anonymization tool
      --debug     Enable debug logging in log file
  -h, --help      Show this help message and exit.
  -v, --verbose   Enable more verbose console output, specify two -v
                            for console debug logging
  -V, --version   Print version information and exit.
Commands:
  help       Displays help information about the specified command
  anonymize  Run anonymization utility
  extract    Run data extraction utility -- generates files out of table
               columns with the name 'table_columnName.txt' for each column
               requested.
  discover   Run data discovery utility

The toolkit can be run in anonymizer mode, data extraction mode (extract), and three different discovery modes (file, column, and database discovery).

Using argument files

DataDefender is using picocli as its framework for processing command-line input. The framework allows using argument files to set argument values when running the tool. The argument file contains a list of arguments to pass (more than one can be used), and when invoking DataDefender, the argument file can be specified with an "@". For example:

File: database.config

--url=jdbc:mariadb://localhost:3306/database?zeroDateTimeBehavior=convertToNull
--password
--user=root

Running with database.config:

datadefender @database.config

File Discovery

datadefender discover files

Usage: datadefender discover files ([-l=<limit>] [-e=<extensions>]
                                   [-e=<extensions>]...
                                   [--model-file=<fileModels>]
                                   [--model-file=<fileModels>]...
                                   [--token-model=<tokenModel>]
                                   [--probability-threshold=<probabilityThreshold>]
                                   [--[no-]score-calculation]
                                   [--threshold-count=<thresholdCount>]
                                   [--threshold-high=<thresholdHighRisk>]
                                   [-m=<models>] [-m=<models>]...) [-hvV]
                                   [--debug] -d=<directories>
                                   [-d=<directories>]... -x=<excludeExtensions>
                                   [-x=<excludeExtensions>]...
Run file discovery utility
  -d, --directory=<directories>
                         Adds a directory to list of directories to be scanned
      --debug            Enable debug logging in log file
  -h, --help             Show this help message and exit.
  -v, --verbose          Enable more verbose console output, specify two -v
                            for console debug logging
  -V, --version          Print version information and exit.
  -x, --exclude-extension=<excludeExtensions>
                         Adds an extension to exclude from data discovery
Model discovery settings
  -e, --extension=<extensions>
                         Adds a call to an extension method (e.g. com.strider.
                           datadefender.specialcase.SinDetector.detectSin)
  -l, --limit=<limit>    Limit discovery to a set number of rows in a table
  -m, --model=<models>   Adds a built-in configured opennlp TokenizerME model
                           for data discovery. Available models are: date,
                           location, money, organization, person, time
      --model-file=<fileModels>
                         Adds a custom made opennlp TokenizerME file for data
                           discovery.
      --[no-]score-calculation
                         If set, includes a column score
      --probability-threshold=<probabilityThreshold>
                         Minimum NLP match score to return results for
      --threshold-count=<thresholdCount>
                         Reports if number of rows found are greater than the
                           defined threshold
      --threshold-high=<thresholdHighRisk>
                         Reports if number of high risk columns found are
                           greater than the defined threshold
      --token-model=<tokenModel>
                         Override the default built-in token model (English
                           tokens, en-token.bin) with a custom token file for
                           use by opennlp's TokenizerModel

File discovery will attempt to find sensitive personal information in binary and text files located on the file system.

Sample project can be found here: sample_projects/file_discovery

Column Discovery

datadefender discover columns

Usage: datadefender discover columns [[-u=<username>] [-p[=<password>]]
                                     [--schema=<schema>]
                                     [--[no-]skip-empty-tables-metadata]
                                     [--include-table-pattern-metadata=<includeTablePatterns>]
                                     [--include-table-pattern-metadata=<includeTablePatterns>]...
                                     [--exclude-table-pattern-metadata=<excludeTablePatterns>]
                                     [--exclude-table-pattern-metadata=<excludeTablePatterns>]...
                                     [--vendor=<vendor>]
                                     [--url=<url>]] [-hvV] [--debug]
                                     [-o=<outputFile>]
                                     --column-pattern=<patterns>
                                     [--column-pattern=<patterns>]...
Run column discovery utility
      --column-pattern=<patterns>
                          Regex pattern(s) to match column names
      --debug             Enable debug logging in log file
  -h, --help              Show this help message and exit.
  -o, --output=<outputFile>
                          Generate a requirements xml file and write it out to
                            the specified file
  -v, --verbose           Enable more verbose console output, specify two -v
                            for console debug logging
  -V, --version           Print version information and exit.
Database connection settings
      --exclude-table-pattern-metadata=<excludeTablePatterns>
                          Pattern(s) matching table names to exclude for
                            metadata analysis
      --include-table-pattern-metadata=<includeTablePatterns>
                          Pattern(s) matching table names to include for
                            metadata analysis
  -p, --password[=<password>]
                          The password to connect with
      --schema=<schema>   The schema to connect to
      --[no-]skip-empty-tables-metadata
                          Include generating metadata for empty tables
                            (defaults to skipping)
  -u, --user=<username>   The username to connect with
      --url=<url>         The datasource URL
      --vendor=<vendor>   Database vendor, available options are: h2, mysql,
                            mariadb, postgresql, sqlserver, oracle. If not
                            specified, vendor will attempt to be extracted from
                            the datasource url for a jdbc scheme.

In this mode the tool attempts to query your database and identified columns that should be anonymized based on their names. When -o is provided a sample requirements file (which can be modified and used for the anonymizer stage) will be created based on the columns discovered.

Note that column and data discovery can be combined. The generated requirements file will combine both results.

Data Discovery

datadefender discover data

Usage: datadefender discover data ([-l=<limit>] [-e=<extensions>]
                                  [-e=<extensions>]...
                                  [--model-file=<fileModels>]
                                  [--model-file=<fileModels>]...
                                  [--token-model=<tokenModel>]
                                  [--probability-threshold=<probabilityThreshold
                                  >] [--[no-]score-calculation]
                                  [--threshold-count=<thresholdCount>]
                                  [--threshold-high=<thresholdHighRisk>]
                                  [-m=<models>] [-m=<models>]...)
                                  [[-u=<username>] [-p[=<password>]]
                                  [--schema=<schema>]
                                  [--[no-]skip-empty-tables-metadata]
                                  [--include-table-pattern-metadata=<includeTablePatterns>]
                                  [--include-table-pattern-metadata=<includeTablePatterns>]...
                                  [--exclude-table-pattern-metadata=<excludeTablePatterns>]
                                  [--exclude-table-pattern-metadata=<excludeTablePatterns>]...
                                  [--vendor=<vendor>]
                                  [--url=<url>]] [-hvV] [--debug]
                                  [-o=<outputFile>]
Run data discovery utility
      --debug             Enable debug logging in log file
  -h, --help              Show this help message and exit.
  -o, --output=<outputFile>
                          Generate a requirements xml file and write it out to
                            the specified file
  -v, --verbose           Enable more verbose console output, specify two -v
                            for console debug logging
  -V, --version           Print version information and exit.
Model discovery settings
  -e, --extension=<extensions>
                          Adds a call to an extension method (e.g. com.strider.
                            datadefender.specialcase.SinDetector.detectSin)
  -l, --limit=<limit>     Limit discovery to a set number of rows in a table
  -m, --model=<models>    Adds a built-in configured opennlp TokenizerME model
                            for data discovery. Available models are: date,
                            location, money, organization, person, time
      --model-file=<fileModels>
                          Adds a custom made opennlp TokenizerME file for data
                            discovery.
      --[no-]score-calculation
                          If set, includes a column score
      --probability-threshold=<probabilityThreshold>
                          Minimum NLP match score to return results for
      --threshold-count=<thresholdCount>
                          Reports if number of rows found are greater than the
                            defined threshold
      --threshold-high=<thresholdHighRisk>
                          Reports if number of high risk columns found are
                            greater than the defined threshold
      --token-model=<tokenModel>
                          Override the default built-in token model (English
                            tokens, en-token.bin) with a custom token file for
                            use by opennlp's TokenizerModel
Database connection settings
      --exclude-table-pattern-metadata=<excludeTablePatterns>
                          Pattern(s) matching table names to exclude for
                            metadata analysis
      --include-table-pattern-metadata=<includeTablePatterns>
                          Pattern(s) matching table names to include for
                            metadata analysis
  -p, --password[=<password>]
                          The password to connect with
      --schema=<schema>   The schema to connect to
      --[no-]skip-empty-tables-metadata
                          Include generating metadata for empty tables
                            (defaults to skipping)
  -u, --user=<username>   The username to connect with
      --url=<url>         The datasource URL
      --vendor=<vendor>   Database vendor, available options are: h2, mysql,
                            mariadb, postgresql, sqlserver, oracle. If not
                            specified, vendor will attempt to be extracted from
                            the datasource url for a jdbc scheme.

In data discovery mode, the tool will perform an NLP scan of data in the database and return columns that have a match score greater than the value of probability-threshold. When -o is provided a sample requirements file (which can be modified and used the anonymizer stage) will be created based on the columns discovered.

Note that column and data discovery can be combined. The generated requirements file will combine both results.

Data Extractor

datadefender extract

Usage: datadefender extract ([-u=<username>] [-p[=<password>]]
                            [--schema=<schema>] [--[no-]skip-empty-tables-metadata]
                            [--include-table-pattern-metadata=<includeTablePatterns>]
                            [--include-table-pattern-metadata=<includeTablePatterns>]...
                            [--exclude-table-pattern-metadata=<excludeTablePatterns>]
                            [--exclude-table-pattern-metadata=<excludeTablePatterns>]...
                            [--vendor=<vendor>] [--url=<url>]) [-hvV]
                            [--debug] [columns...]
Run data extraction utility -- generates files out of table columns with the
name 'table_columnName.txt' for each column requested.
      [columns...]        Generate data for the specified table.columName(s)
      --debug             Enable debug logging in log file
  -h, --help              Show this help message and exit.
  -v, --verbose           Enable more verbose console output, specify two -v
                            for console debug logging
  -V, --version           Print version information and exit.
Database connection settings
      --exclude-table-pattern-metadata=<excludeTablePatterns>
                          Pattern(s) matching table names to exclude for
                            metadata analysis
      --include-table-pattern-metadata=<includeTablePatterns>
                          Pattern(s) matching table names to include for
                            metadata analysis
  -p, --password[=<password>]
                          The password to connect with
      --schema=<schema>   The schema to connect to
      --[no-]skip-empty-tables-metadata
                          Include generating metadata for empty tables
                            (defaults to skipping)
  -u, --user=<username>   The username to connect with
      --url=<url>         The datasource URL
      --vendor=<vendor>   Database vendor, available options are: h2, mysql,
                            mariadb, postgresql, sqlserver, oracle. If not
                            specified, vendor will attempt to be extracted from
                            the datasource url for a jdbc scheme.

The Data Extractor is used to load table data into text files. The text files are useful to modify and then feed into the annoymizer as input data.

Anonymizer

datadefender anonymize

Usage: datadefender anonymize ([-u=<username>] [-p[=<password>]]
                              [--schema=<schema>]
                              [--[no-]skip-empty-tables-metadata]
                              [--include-table-pattern-metadata=<includeTablePatterns>]
                              [--include-table-pattern-metadata=<includeTablePatterns>]...
                              [--exclude-table-pattern-metadata=<excludeTablePatterns>]
                              [--exclude-table-pattern-metadata=<excludeTablePatterns>]...
                              [--vendor=<vendor>] [--url=<url>])
                              [-hvV] [--debug] [-b=<batchSize>]
                              -r=<requirementFile> [tables...]
Run anonymization utility
      [tables...]         Limit anonymization to specified tables
  -b, --batch-size=<batchSize>
                          Number of update queries to batch together
      --debug             Enable debug logging in log file
  -h, --help              Show this help message and exit.
  -r, --requirement-file=<requirementFile>
                          Requirement XML file
  -v, --verbose           Enable more verbose console output, specify two -v
                            for console debug logging
  -V, --version           Print version information and exit.
Database connection settings
      --exclude-table-pattern-metadata=<excludeTablePatterns>
                          Pattern(s) matching table names to exclude for
                            metadata analysis
      --include-table-pattern-metadata=<includeTablePatterns>
                          Pattern(s) matching table names to include for
                            metadata analysis
  -p, --password[=<password>]
                          The password to connect with
      --schema=<schema>   The schema to connect to
      --[no-]skip-empty-tables-metadata
                          Include generating metadata for empty tables
                            (defaults to skipping)
  -u, --user=<username>   The username to connect with
      --url=<url>         The datasource URL
      --vendor=<vendor>   Database vendor, available options are: h2, mysql,
                            mariadb, postgresql, sqlserver, oracle. If not
                            specified, vendor will attempt to be extracted from
                            the datasource url for a jdbc scheme.

In this mode, data anonymization is performed on the database based on the requirements file. The requirements file is an XML-formatted file describing which tables and columns should be anonymized, and how. For an example, refer to sample_projects/anonymizer/requirement.xml.

Logging

DataDefender uses log4j2 to log output. The default configuration logs output to both the console at a WARN level, and to a file at INFO level. The default levels can be made more verbose by running with --debug or -v/--verbose. -v can be chained to log DEBUG level to console:

datadefender -vv --debug

would log at 'debug' level to both console and file.

The default file logged to is 'logs/datadefender.log'. Each new run of DataDefender creates a new log file, and rolls the previous log file to a dated file in the format:

'logs/datadefender-%d{yy-MM-dd-hh-mm-ss}.log'

To change the default loggers and format, creating a log4j2 configuration file (log4j2(.properties|.json|.xml) in the project's directory will override the default properties file. Please refer to the default file to build off of if desired: src/main/resources/log4j2.properties. Note the use of a ThresholdFilter in the default version, and the default logLevel of "all". The ThresholdFilters are how DataDefender controls the log levels based on command-line options. Setting the loglevel to something more specific, e.g. 'info', and/or not setting up the ThresholdFilters accordingly, will cause the command-line options to not work.

Database logging

Log4j2 includes a "JdbcAppender" that can be used to configure database logging. For convenience, the additional required lib can be included the DataDefender jar file by running maven with the following profile:

mvn package -P database-logging

Don't forget to include any additional drivers needed (in this case, both for logging and for DataDefender's operations):

mvn package -P database-logging,mariadb,oracle

The 'database-logging' maven profile will include org.apache.commons.commons-dbcp2 in the final shaded jar. Otherwise, the commons-dbcp2.jar file could also be manually downloaded and copied to the 'lib' directory underneath the project's directory.

For log4j2 JdbcAppender configuration, please refer to the log4j2 documentation: https://logging.apache.org/log4j/2.0/manual/appenders.html#JDBCAppender.

Additionally, a sample with database logging configured is available here: sample_projects/database-logging with an example log4j2.properties file defined under it.

Upgrading to 2.0

Database configuration

Database configuration has been moved from a properties file to command line arguments. The command line arguments can be bundled in files for easy configuration, see Using argument files for more information.

The configuration options remain the same except:

• 'vendor' is not required, datadefender will try to determine that from the provided jdbc connection url. It's still provided as an option (which would allow using other jdbc drivers/urls and specifying a vendor to use for datadefender's queries, operations, etc...)
• There's no need to specify a driver

Example change:

1.0 file:

vendor=mysql
driver=com.mysql.jdbc.Driver
username=root
password=mypassword
url=jdbc:mysql://localhost:3306/test

2.0 file:

--username=root
--password
--url=jdbc:mysql://localhost:3306/test

(Leaving password without a value will cause the utility to prompt for a value interactively).

Please review the command line help in the app itself, or in the readme under command headings for database configuration details.

Column, data, and file discovery

The main difference is the move from properties files to picocli argument configuration. The easiest way to do that is use your existing properties file, and rename the property portion of each line to the corresponding argument. For example:

1.0 properties file for file discovery:

probability_threshold=0.5
english_tokens=en-token.bin
person=en-ner-person.bin
location=en-ner-location.bin
models=person,location,organization
directories=/path/to/directory,/path/to/other/directory
exclusions=jar,exe

Becomes the following in 2.0:

-m location
-m person
--directory=/path/to/directory
--directory=/path/to/other/directory
--exclude-extension=jar
--exclude-extension=exe

Please review the help associated with each command as there are small differences (for example, there's no need to specify en-token.bin, it's the default if --token-model isn't provided with a custom file. It's also not necessary to provide the files, they've been bundled in the jar file. You can provide custom opennlp models with --model-file.

Anonymization

There are a couple of changes affecting anonymization (in addition to database configuration):

A new format and features for the requirements xml file. See the sample_project for an example new format, and also the xml schema file requirement.xsd. Some of the high-level changes are:

• Elements are now all lower-case, dash-separated. <Column> becomes <column>, NotLike becomes not-like, e.g. <exclude name="col" not-like="example"/>.
• The top-level element is now <anonymizer xmlns="https://armenak.github.io/DataDefender/anonymizer"> not <Requirement>
• Below the top-level element, <anonymizer-version>2.0</anonymizer-version> is now required, and must match the version in the xsd included in the jar file. This will be updated as changes to the anonymizer happen that require new versions, so the correct version must be used on a requirement.xml file.
• <Client>client name</Client> becomes <project>project name</project>
• <Version> becomes <project-version> for requirement file versioning.
• <Column> IgnoreEmpty has been renamed to skip-empty and it's now "true" by default.
• <Column> PKey attribute has been renamed to primary-key
• <Column> Where attribute is now an element <where> under <column>.
• <PrimaryKey> is now called <primary-key>.
• <column> must have a <plan> or <plan-ref> below it. One or more <function> elements can exist below a <plan> (used to be <Function> under <Column>).
• The <function> element now has name and type (to optionally specify return type) attributes. The name attribute refers to the fully-qualified name of the function (unless it's part of a class in a package defined under the top-level <autoresolve-classes>), whereas before, the content of the <Function> tag specified the name of the function.
• <Parameters><Parameter> no longer exists, instead <argument> tags specify an argument to be passed to a function. The <argument> tag exists directly under <function>. A new attribute of the <argument> tag, pass-current-value passes the running value as the specified argument to the function. The running value for the first function in a series of functions is the columns value. Afterwords, it's the returned value from the call to the previous function.

CoreFunctions has been split into a few different classes, and its package has been moved. See the new classes here: src/main/java/com/strider/datadefender/anonymizer/functions. Some functions have been removed entirely, for example randomInt, because apache commons can be used instead with RandomUtils.nextInt.

Features and issues

Please report issues or ask for future requests here: https://github.com/armenak/DataDefender/issues

Code quality

Two amazing tools - Empear http://empear.com/ and SonarQube http://www.sonarqube.org/ help contributors of DataDefender maintain decent quality of code. Many thanks to their creators!