Mainframe Security: 15 Configuration Weaknesses Worth Investigating

The mainframe has earned its reputation for security. But even the most secure platform can be undermined by a configuration that nobody has questioned for years.

There is a tendency in our industry to associate security incidents with sophisticated attacks, previously unknown vulnerabilities or increasingly advanced techniques used by cybercriminals. And whilst these threats are real, some of the most significant security exposures are considerably less exciting.

An access permission granted fifteen years ago. A privileged user ID that nobody remembers creating. An APF-authorised library that far too many people can update. Or a started task running with privileges it simply doesn’t need.

None of these requires a particularly sophisticated attacker. Sometimes, the difficult part is finding the weakness in the first place.

The IBM mainframe provides an exceptionally strong security architecture. RACF, ACF2 and Top Secret, together with z/OS system integrity controls, offer capabilities that many other platforms would envy. However, having these capabilities available is not the same as configuring and managing them correctly.

Having spent many years working with mainframe security, I’ve learnt that some of the most interesting findings are not necessarily the most technically complicated. They’re often the configurations that have survived successive upgrades, migrations and changes in personnel without anyone revisiting their original justification.

Here are fifteen areas I believe deserve particular attention.

1. Excessive privileged access

Let’s start with the obvious one. How many user IDs have system-wide SPECIAL, OPERATIONS or AUDITOR attributes in RACF? More importantly, how many genuinely need them?

SPECIAL provides extensive security administration authority, whilst OPERATIONS can grant broad access to protected resources. AUDITOR provides security auditing capabilities, rather than equivalent general administrative authority, but still deserves careful control.

What to investigate: Review privileged user IDs, group-level attributes, emergency access accounts and the business justification for each assignment. Don’t overlook privileged access inherited through group connections.

2. APF-authorised libraries with excessive UPDATE access

This is one of the areas that should make any mainframe security professional uncomfortable.

APF-authorised programs can access sensitive system functions. If an unauthorised individual can modify executable content in an APF-authorised library, the implications for system integrity can be severe.

What to investigate: Examine UPDATE and ALTER permissions to APF-authorised libraries, including relevant linklist and LPA source libraries. Review who can change the APF configuration itself, and don’t forget privileged batch jobs and software deployment processes.

IBM’s documentation makes the importance of protecting APF-authorised libraries particularly clear. IBM+1

3. Started tasks running with unnecessary privileges

Started tasks frequently operate with elevated authority because they need to perform system-level functions. Unfortunately, some retain permissions that were granted years ago for reasons nobody can explain.

A monitoring product, for example, may require access to specific system resources. That doesn’t automatically justify unrestricted access to everything.

What to investigate: Review STARTED class assignments, associated user IDs, privileged attributes, group connections and resource permissions. Pay particular attention to started tasks that execute third-party software.

4. Weak protection of critical system datasets

SYS1.PARMLIB, PROCLIB, authorised load libraries and other critical system datasets deserve considerably more attention than ordinary application datasets.

Changes to these resources can influence system configuration, execution paths and security behaviour.

What to investigate: Identify who can modify system datasets, whether generic profiles grant unintended access, and whether change management adequately controls modifications. Include relevant aliases, dataset naming conventions and indirect update paths.

5. Excessive access through generic resource profiles

Generic profiles are extremely useful. They’re also capable of granting far more access than originally intended.

A broadly defined dataset or general-resource profile may unintentionally include newly created resources. The result can be an exposure that grows over time without anyone explicitly granting additional access.

What to investigate: Review generic profile coverage, UACC values, access lists and the effective permissions of users and groups. Don’t assume that the profile with the most obvious name is necessarily the one protecting a resource.

6. Inadequate protection of z/OS UNIX privileges

UNIX System Services introduces another security dimension that is sometimes overlooked by teams primarily focused on traditional MVS resources.

UID(0), BPX-related FACILITY profiles and UNIXPRIV permissions deserve careful attention. An overly permissive configuration can provide capabilities far beyond those required by an application or administrator.

What to investigate: Review UID(0) assignments, relevant BPX FACILITY profiles, UNIXPRIV permissions, file ownership and sensitive directory permissions. Avoid assuming that a generic BPX.* profile is a sensible way to secure UNIX functions; IBM explicitly cautions against defining it. IBM+1

7. Insufficient control over operator commands

Operator commands can influence the operation of critical system components. Some can affect availability, change configurations or control important workloads.

The ability to issue these commands should never be treated as an ordinary operational convenience.

What to investigate: Examine OPERCMDS protection, command authority through SDSF and other interfaces, and the permissions assigned to automation products and operational accounts.

8. Inadequate JES and spool protection

JES spool datasets can contain information that organisations would certainly not want exposed, including application output, operational details and potentially sensitive business data.

Controlling who can submit jobs is important. Controlling who can view, modify or manipulate their associated resources is equally important.

What to investigate: Review JESSPOOL profiles, JES-related resource permissions, job submission controls and SDSF authorisations. Consider both confidentiality and the ability to interfere with jobs or output.

IBM documents the distinct roles of JESSPOOL, SDSF and OPERCMDS in protecting these functions. IBM+1

9. Inactive or incorrectly configured security classes

Defining a security profile doesn’t necessarily mean the intended protection is being enforced.

Depending on the resource and security manager, class activation, processing options and application integration can determine whether security checks are actually performed.

What to investigate: Review active resource classes, RACF SETROPTS settings, relevant RACLIST processing and refresh procedures. Confirm that applications issue the expected SAF checks and that the intended profiles are actually effective.

10. Weak protection of cryptographic keys and services

Encryption is important, but it is only as effective as the controls surrounding its implementation and key management.

ICSF provides powerful cryptographic capabilities. However, unrestricted access to cryptographic services, keys or tokens can undermine the protections those capabilities are intended to deliver.

What to investigate: Review CSFSERV, CSFKEYS, CRYPTOZ and other applicable ICSF resource classes. Verify who can use cryptographic services, access keys, administer tokens and modify key-management datasets.

IBM’s Security on the IBM Mainframe Redbook provides useful guidance on protecting ICSF resources. redbooks.ibm.com

11. Weak authentication and poorly managed functional IDs

Not every user ID needs interactive access. In fact, many should never have it.

Functional accounts, service IDs and started-task identities can become attractive targets when passwords are unnecessarily assigned or authentication controls are weak.

What to investigate: Review password and password-phrase policies, multifactor authentication for appropriate interactive access, unused accounts and functional IDs that could be converted to protected user IDs. Also verify that authentication requirements are consistently enforced across access methods. IBM+1

12. Overly permissive surrogate authority

SURROGAT permissions are sometimes overlooked, despite their potentially significant consequences.

In RACF, appropriately authorised users can submit jobs that execute under another user’s identity. This is legitimate and useful, but granting such authority too broadly can create an indirect route to elevated privileges.

What to investigate: Review SURROGAT profiles, particularly permissions involving privileged execution identities. Establish whether the submitting users genuinely require the authority and whether the associated job submission paths are adequately controlled. IBM+1

13. Insecure network services and TLS configurations

Mainframes are no longer isolated systems, assuming they ever truly were.

They communicate with distributed applications, cloud services, APIs and numerous other platforms. Every connection introduces configuration considerations.

What to investigate: Review exposed network services, TN3270 connections, FTP usage, TLS protocol and cipher settings, certificate management and AT-TLS policies where applicable. Verify that encryption is actually enforced on the intended connections, rather than merely configured somewhere in the environment. IBM+1

14. Inadequate security auditing

One of the more concerning findings during a security assessment is discovering that important security events are not being recorded, or that the records exist but nobody reviews them.

RACF provides extensive auditing capabilities, including SMF type 80 records. However, the actual records generated depend on the events, application requests and configured auditing options.

What to investigate: Review SETROPTS auditing controls, profile-level auditing, privileged activity logging, SMF collection and forwarding to security monitoring platforms. Check whether alerts are generated for meaningful events, not simply whether logs are being collected.

Logging everything isn’t necessarily the answer either. Collecting enormous quantities of information that nobody can realistically analyse achieves very little. IBM+1

15. Security configurations that have never been reassessed

This last point isn’t a single RACF setting. It’s a problem that can affect every other item on this list.

Mainframe environments are remarkably stable. That’s one of their greatest strengths. Unfortunately, that stability can also encourage organisations to retain configurations long after the original business or technical requirements have disappeared.

A permission granted during a migration may remain indefinitely. An exception introduced to resolve an incident may never be removed. A security control temporarily relaxed for testing may become the accepted production configuration.

What to investigate: Compare existing configurations against approved security baselines, examine historical exceptions and use available tools such as IBM Health Checker for z/OS to identify deviations.

Health Checker provides valuable checks, including RACF_SENSITIVE_RESOURCES, but its results are not a substitute for a complete assessment of effective permissions and application behaviour. IBM+1

Where should you start?

Not all fifteen areas represent the same level of risk, and the priority will depend on the environment.

Personally, I would start with three questions:

  1. Who can obtain or exercise elevated authority? Examine privileged identities, APF-authorised libraries, started tasks and surrogate permissions.
  2. What can be modified that would affect system integrity? Focus on critical datasets, authorised programs, security configurations and operator commands.
  3. Would we know if someone abused these permissions? Validate auditing, monitoring and the organisation’s ability to investigate suspicious activity.

These questions help move the discussion away from simply counting security findings and towards understanding their potential consequences.

And importantly, investigate before changing anything. Removing an apparently unnecessary permission without understanding its dependencies can have serious operational consequences.

Final thoughts

Of course, these fifteen areas are by no means an exhaustive list. There are many other aspects of mainframe security worth exploring, from CICS, Db2 and IMS security to network protection, privileged access management, software supply chain security and the increasingly important challenges surrounding APIs and hybrid cloud integration. Each of these deserves attention in its own right, and some could easily justify an article of their own. The objective here isn’t to cover everything, but rather to highlight some of the configuration weaknesses that are worth revisiting, particularly those we may have become so accustomed to that we no longer question them.

The mainframe remains one of the most secure computing platforms available. But its reputation shouldn’t lead us to assume that every environment is equally secure.

Security isn’t something achieved simply by installing RACF, ACF2 or Top Secret, enabling a few controls and moving on to the next project.

It requires continuous attention, technical understanding and, occasionally, the willingness to question configurations that have been accepted for decades.

Some of the most serious weaknesses aren’t hidden behind complicated exploits. They’re sitting in plain sight, documented in a security profile, waiting for someone to ask why that permission is still there.

And perhaps that’s the most important question of all.

When was the last time you challenged the security configuration of your mainframe, rather than simply confirming that it was working?

Technical references and further reading

The article is grounded primarily in IBM’s technical documentation, rather than unverified community commentary.

Technical scope: IBM z/OS, with RACF-specific examples. Equivalent assessments for ACF2 and Top Secret require their respective security models and configuration semantics. Findings are areas to investigate, not proof of vulnerabilities in every installation.

Be the first to comment

Leave a Reply

Your email address will not be published.


*


This site uses Akismet to reduce spam. Learn how your comment data is processed.