Monday, June 29, 2020

The Various Aspects of Db2 for z/OS Data Encryption

Just last week, a person on the Db2 for z/OS team at an organization asked me for some guidance regarding data encryption in a Db2 context. In responding to that request, I noted that Db2 for z/OS data encryption is a multifaceted topic, and I proceeded to describe various categories of Db2 data encryption and the Db2 features and functions that pertain to those categories. Thinking that this information might be of interest to a larger audience, I'll provide it in this blog entry.

The categories of Db2 for z/OS data encryption I will address herein are as follows:
  • Application-transparent encryption of Db2 data "at rest" (i.e., on disk)
  • Application-transparent encryption of Db2 for z/OS data in memory (i.e., in a Db2 buffer pool) as well as on disk
  • Application-aware encryption of data in a column of a Db2 table, on-disk and in-memory
  • Application-transparent encryption of Db2 data "on the wire" (i.e., in-motion on a TCP/IP link between a client application and a Db2 for z/OS server)

Application-transparent encryption of Db2 for z/OS data "at rest" (i.e., on disk)

One way to get this done is via a feature of your disk subsystem, if the disk subsystem that you use with Db2 for z/OS has that feature.

Another way to get this done is through the use of a z/OS feature called, simply, z/OS data set encryption. This z/OS capability can be used with Db2 11 or Db2 12 for z/OS, but it got quite a bit easier to use with Db2 12, at function level M502 or higher (more on that to come). With z/OS data set encryption, data is encrypted automatically when written to disk, and it is decrypted automatically when read from disk. z/OS data set encryption is enabled by associating an encryption key label with a data set (it is of course very important to secure and protect an encryption key, so instead of referencing the key itself we reference its label, which you can think of as the key's "handle"). Here is the difference between a Db2 environment that is at V12 with function level M502 or higher activated, and one that is not (function levels, by the way, are related to the "continuous delivery" mode by which new Db2 12 functionality is provided to users):
  • Db2 12, at function level M502 or higher: A DBA can issue an ALTER TABLE or ALTER STOGROUP statement with a new option, KEY LABEL, to assign an encryption key label to a set of Db2 data sets (for ALTER TABLE. the specified encryption key label will also apply to index and auxiliary table data sets). Function level M502 also provides a new ZPARM parameter, ENCRYPTION_KEYLABEL, through which a key label can be specified for the Db2 catalog and directory data sets, and for archive log data sets (when the latter are written to disk - these will remain encrypted if they are later HSM-migrated to tape). When a key label has been associated with a database object, actual encryption of that object's data can be accomplished via execution of the Db2 online REORG utility (or any other Db2 utility that involves creation of a new data set or data sets, such as LOAD REPLACE or RECOVER or REBUILD INDEX). Note that the KEY LABEL option is valid for CREATE TABLE and CREATE STOGROUP statements, as well as for ALTER statements.
  • Db2 12, at function level M501 or lower, or Db2 11: A key label is associated with a Db2 data set either via a RACF data set profile or by way of an SMS data class specification or through an explicit IDCAMS definition (in the case of a user-managed data set). As noted above, once a key label has been associated with an existing Db2 data set, actual encryption of the data in that data set can be accomplished via execution of a Db2 utility (such as online REORG) that will involve creation of a new data set (or data sets).

One more thing about z/OS data set encryption: for data decryption to occur when an encrypted data set is accessed, the ID of the accessing process has to be RACF-permitted to access the data set's key label. For Db2 data sets, the only IDs that need this RACF permission to access the associated key labels are the IDs of the Db2 database services and system services address spaces (i.e., the Db2 DBM1 and MSTR address spaces). This is so because, when user SMITH issues a SELECT for table T1, z/OS does not perceive that SMITH is accessing the table's data set(s) - it perceives that Db2 is accessing the data set(s).


Application-transparent encryption of Db2 for z/OS data in memory (i.e., in a buffer pool) as well as on disk

This can be accomplished using Db2 edit procedures (EDITPROCs), for encryption of entire rows in tables, or via Db2 field procedures (FIELDPROCs) for encryption of data in individual columns.

Note that CPU overhead for this type of Db2 data encryption will likely be higher than for z/OS data set encryption, because with the latter data is decrypted when read into memory, whereas with EDITPROC- or FIELDPROC-based encryption, data is encrypted in memory and is decrypted every time it is accessed by an application process.


Application-aware encryption of data in a column of a Db2 table, on-disk and in-memory

This has been do-able for some time by way of a built-in Db2 scalar function called ENCRYPT_TDES. Db2 12 function level M505 provided a new built-in scalar function, ENCRYPT_DATAKEY, that is even more robust than ENCRYPT_TDES. Several factors contribute to the stronger column-level data protection provided by ENCRYPT_DATAKEY versus ENCRYPT_TDES:
  • ENCRYPT_DATAKEY uses 256-bit encryption, while ENCRYPT_TDES uses 128-bit encryption.
  • With ENCRYPT_TDES, an application provides a password when data is encrypted, and that same password must be provided when the data is to be decrypted. What if that password becomes known to someone who shouldn't know it? With ENCRYPT_DATAKEY, access to encrypted data requires specification of the right key label, and on top of that the ID of the application process must be RACF-permitted to access that key label. If someone knows the key label that has to be provided to decrypt data encrypted via the ENCRYPT_DATAKEY function, that person will nonetheless be unable to access decrypted data if he or she is using an ID that has not been RACF-permitted to access the key label.
  • ENCRYPT_DATAKEY offers an option whereby duplicate values in a column will have unique encrypted values (in other words, two values in the column that are duplicates when decrypted will be unique when encrypted). This option makes the data encryption even harder to defeat. 

Application-transparent encryption of Db2 data "on the wire" (i.e., in-motion between a client application and a Db2 for z/OS server)

This is done by way of SSL encryption (or, more precisely, via AT-TLS, which is short for application-transparent, transport-layer security). When a client application - either a DRDA requester or a program utilizing the REST interface to Db2 - interfaces with the Db2 distributed data facility (DDF) and specifies the Db2 system's secure SQL listener port, Db2 will require the use of SSL encryption. The client application will then need to present a server certificate (i.e., a digital certificate obtained from the target z/OS system) to the z/OS system for an SSL "handshake." The appropriate AT-TLS policy, defined on the z/OS side, will be applied to enable SSL communication between the client and the Db2 server.

The process of initially setting up SSL encryption between a client process and a Db2 for z/OS system can be a bit involved because several pieces of work have to fit together. There is client-side work to be done: the server certificate from the target z/OS system needs to be obtained and placed properly in a key store database, and the right connection string has to be used (and/or the right information has to be specified in the IBM Data Server Driver's db2dsdriver.cfg file) to request an SSL connection to the Db2 server. There is Db2 work to be done: a secure SQL listener port needs to be specified for the Db2 system. There is z/OS work to be done: an AT-TLS policy has to be defined and implemented by way of a policy agent. The best reference I've seen that covers all these pieces of work is an IBM "red paper" titled, DB2 for z/OS: Configuring TLS/SSL for Secure Client/Server Communications (download-able from http://www.redbooks.ibm.com/abstracts/redp4799.html?Open). If you get the right people in your organization involved, with the right skills and knowledge (client-side, Db2 for z/OS, z/OS Communications Server, RACF), you should be successful in getting SSL communications set up between Db2 for z/OS and network-connected applications.

That concludes this overview of Db2 for z/OS data encryption in its various forms. I hope that this information will be useful for you.