6.4.8.3 Using MySQL Enterprise Data Masking and De-Identification

MySQL provides general-purpose masking functions that mask arbitrary strings, and special-purpose masking functions that mask specific types of values.

Contents Haut

General-Purpose Masking Functions

mask_inner() and mask_outer() are general-purpose functions that mask parts of arbitrary strings based on position within the string:

By default, mask_inner() and mask_outer() use 'X' as the masking character, but permit an optional masking-character argument:

Contents Haut

Special-Purpose Masking Functions

Other masking functions expect a string argument representing a specific type of value and mask it to remove identifying characteristics.

Payment card Primary Account Number masking.  Masking functions provide strict and relaxed masking of Primary Account Numbers.

U.S. Social Security number masking.  mask_ssn() masks all but the last four digits of the number:

Several functions generate random values. These values can be used for testing, simulation, and so forth.

gen_range() returns a random integer selected from a given range:

gen_rnd_email() returns a random email address in the example.com domain:

gen_rnd_pan() returns a random payment card Primary Account Number:

(The gen_rnd_pan() function result is not shown because its return values should be used only for testing purposes, and not for publication. It cannot be guaranteed the number is not assigned to a legitimate payment account.)

gen_rnd_ssn() returns a random U.S. Social Security number with the first and second parts each chosen from a range not used for legitimate numbers:

gen_rnd_us_phone() returns a random U.S. phone number in the 555 area code not used for legitimate numbers:

MySQL Enterprise Data Masking and De-Identification enables dictionaries to be used as sources of random values. To use a dictionary, it must first be loaded from a file and given a name. Each loaded dictionary becomes part of the dictionary registry. Items then can be selected from registered dictionaries and used as random values or as replacements for other values.

A valid dictionary file has these characteristics:

Suppose that a file named de_cities.txt contains these city names in Germany:

Berlin
Munich
Bremen

Also suppose that a file named us_cities.txt contains these city names in the United States:

Chicago
Houston
Phoenix
El Paso
Detroit

Assume that the secure_file_priv system variable is set to /usr/local/mysql/mysql-files. In that case, copy the dictionary files to that directory so that the MySQL server can access them. Then use gen_dictionary_load() to load the dictionaries into the dictionary registry and assign them names:

To select a random term from a dictionary, use gen_dictionary():

To select a random term from multiple dictionaries, randomly select one of the dictionaries, then select a term from it:

The gen_blacklist() function enables a term from one dictionary to be replaced by a term from another dictionary, which effects masking by substitution. Its arguments are the term to replace, the dictionary in which the term appears, and the dictionary from which to choose a replacement. For example, to substitute a U.S. city for a German city, or vice versa, use gen_blacklist() like this:

If the term to replace is not in the first dictionary, gen_blacklist() returns it unchanged:

At customer-service call centers, one common identity verification technique is to ask customers to provide their last four Social Security number (SSN) digits. For example, a customer might say her name is Joanna Bond and that her last four SSN digits are 0007.

Suppose that a customer table containing customer records has these columns:

The application used by customer-service representatives to check the customer SSN might execute a query like this:

However, that exposes the SSN to the customer-service representative, who has no need to see anything but the last four digits. Instead, the application can use this query to display only the masked SSN:

Now the representative sees only what is necessary, and customer privacy is preserved.

Why was the CONVERT() function used for the argument to mask_ssn()? Because mask_ssn() requires an argument of length 11, and because UDFs treat string arguments as binary strings, with one byte per character. Thus, even though ssn is defined as VARCHAR(11), if the ssn column has a multibyte character set, it appears to be longer than 11 bytes when passed to a UDF, and an error occurs. Converting the value to a binary string ensures that the UDF sees an argument of length 11.

A similar technique may be needed for other data masking functions when string arguments do not have a single-byte character set.

If masked data from a table is used for multiple queries, it may be convenient to define a view that produces masked data. That way, applications can select from the view without performing masking in individual queries.

For example, a masking view on the customer table from the previous section can be defined like this:

Then the query to look up a customer becomes simpler but still returns masked data: