HP OpenVMS Systems Documentation

High-water marking refers to a technique that tracks the furthest extent to which each file has been written and prohibits user attempts at reading data beyond that point.

The operating system implements true high-water marking for all sequential, exclusively accessed files, such as the set of files output from various text editors, compilers, and linkers, that is, most files a process writes. The high-water mark is updated in the file header whenever the logical end-of-file mark is updated (usually when the file is closed).

For shared files (both indexed and sequential), the operating system uses the principle of erase-on-allocate to achieve a result similar to true high-water marking. When a file is about to be created or extended, the system determines how much disk space (the extent of the file) is required and applies the security erasure pattern of zeros to the areas (extents) it allocates for writing. The file is then written into the area just erased for it. Thus, if any user gains access to the file (including its full extent) and attempts to read the area beyond where the file has been written, only the data security erase pattern is readable.

By default, the operating system turns on high-water marking for all volumes. High-water marking is a deterrent to disk scavenging attempts. However, it does require additional I/O, which affects system performance.

You can turn off high-water marking and erase-on-allocate on a volume-by-volume basis by specifying the DCL command SET VOLUME/NOHIGHWATER_MARKING.

8.9.5.3 Summary of Prevention Techniques

As security administrator, you can apply the following controls to discourage disk scavengers:

• Provide tight physical security, particularly on those disks with the most valuable information.
• Provide tight volume protection through UIC-based protection.
• Encourage the use of the /ERASE qualifier when key files are purged or deleted through user participation or volume enforcement.
• Permit default high-water marking on your most valuable disks.

You can guard against data loss or corruption by creating copies of your files, directories, and disks. In case of a problem, you can restore the backup copy and continue your work. Secure media storage and controlled access to media are essential parts of the process. It is best to store backup media off site.

8.9.6.1 Backing Up Disks

Having an effective backup schedule is critical to protect your data. By performing regularly scheduled backup operations, you prevent the loss of accidentally deleted or damaged files.

Refer to the OpenVMS System Management Utilities Reference Manual for information about performing backups and setting up backup schedules. Be aware that the Backup utility (BACKUP) does not implement security policy; you must direct it explicitly. It runs with the security profile of the operator, which can often be privileged.

8.9.6.2 Protecting a Backup Save Set

Limiting access to backup save sets is an important part of system security. The file system treats a backup save set as a single file, whether it is stored on disk or on magnetic tape. Therefore, anyone with access to a save set can read any file in the save set. BACKUP does not check protection on individual files.

To maintain system security, it is crucial that you protect save sets adequately. Assign restrictive protection to save sets on disk and to magnetic tape volumes by using the output save-set qualifiers /BY_OWNER and /PROTECTION. Sufficient protection can prevent nonprivileged users from mounting a save-set volume or from reading files from a save set. You should also take physical security precautions with save sets stored off line by keeping backup media in locked cabinets.

When you write a save set to a Files--11 disk or a sequential disk and do not specify the /PROTECTION qualifier, BACKUP applies the process default protection to the save set. If you specify /PROTECTION, any protection categories that you do not specify default to your default process protection.

Protection information is written to the volume header record of a magnetic tape and applies to all save sets stored on the tape. Therefore, the output save-set qualifiers /BY_OWNER and /PROTECTION are effective on magnetic tape save sets only if you specify the output save-set qualifier /REWIND. This qualifier allows the tape to rewind to its beginning, to write the protection data to the volume header record, and to initialize the tape. If you specify /PROTECTION, any protection categories that you do not specify default to your default process protection. If you do not specify /REWIND with the /PROTECTION and /BY_OWNER qualifiers, the magnetic tape retains its existing protection. However, specifying /REWIND alone results in a magnetic tape without any protection.

The following example illustrates how a directory is backed up to tape:

$  BACKUP
_FROM:  [PAYROLL]
_TO: MFA2:KNOX.BCK/LABEL=BANK01 - (1)
_$ /REWIND/BY_OWNER_UIC=[030,003] -  (2)
_$ /TAPE_EXPIRATION=15-JAN-1993 - (3)
_$ /PROTECTION=(S:RWE,O:RWED,G:RE,W) (4)

1. The contents of the directory [PAYROLL] is copied to file KNOX.BCK on the magnetic tape drive MFA2. The output save-set qualifier /LABEL provides the label BANK01 for the tape.
2. The output save-set qualifier /BY_OWNER assigns an owner UIC of [030,003] to the save set.
3. The output save-set qualifier /TAPE_EXPIRATION assigns an expiration date of January 15, 1993 to the tape.
4. The output save-set qualifier /PROTECTION assigns the owner of the volume read, write, execute, and delete access. System users are assigned read, write, and execute access; group users are assigned read and execute access; world users are assigned no access.

Anyone who has access to a save set can read any file in the save set. Never give a copy of your backup media to a user; a malicious user could restore the files from the tape or disk and compromise the security of the system.

When a nonprivileged user wants to restore a particular file, do not lend the volume containing the save set. You could give away access to all the files on the volume. The safest way to restore a particular file is to restore the file selectively, as shown in the following example:

$ BACKUP MTA0:JULY.BCK/SELECT=[JONES.TEXTPROC]LASTMONTH.DAT -
_$ [*...]/BY_OWNER=ORIGINAL

The selected file is restored with its original directory, ownership, and protection. In this way, the file system determines if the user is permitted access to the file.

8.9.7 Protecting Terminals

The next sections describe the controls available for restricting the use of terminals.

8.9.7.1 Restricting Terminal Use

Through the device object class template TERMINAL, the operating system sets up terminals to be accessible to the SYSTEM account only. When a user logs in, the operating system transfers ownership from a system UIC to the UIC of the current process.

You can limit logins on specific terminals in the following ways:

• Assign a system password.
• Set the terminal to /NOTYPE_AHEAD, making it impossible to log in.

The application of system passwords limits the use of those terminals to users who know the system password.

8.9.7.2 Restricting Application Terminals and Miscellaneous Devices

To make terminals accessible to certain users as application terminals, you may want to change any or all of the device's security characteristics. You can include the DCL command SET SECURITY/CLASS=DEVICE for specific terminals (with appropriate protection codes) in the command procedure SYS$MANAGER:SYSTARTUP_VMS.COM. This DCL command can limit access to any device that is not file structured. You might also place an ACL on the device to limit user access.

8.9.7.3 Configuring Terminal Lines for Modems

When configuring terminal lines for modems, never set the /COMMSYNC qualifier to the DCL command SET TERMINAL (or the TT$M_COMMSYNC characteristic for the TTDRIVER interface) on a line with a modem hookup that is intended for interactive use.

The qualifier disables the modem terminal characteristic that disconnects a user process from the terminal line in case of a modem phone line failure. With the /COMMSYNCH qualifier enabled, the next call on the terminal line could be attached to the previous user's process. The /COMMSYNC qualifier is intended to allow connection of asynchronous printers and other devices to terminal ports by using modem signals as flow control.

This chapter describes how to use and manage the OpenVMS auditing system. It explains how you can monitor security-relevant activity on your system by recording events as they occur on the system and subsequently analyzing this audit log.

9.1 Overview of the Auditing Process

Auditing is the recording of security-relevant activity as it occurs on the system and the subsequent analysis of this audit log. With auditing, you can monitor users' activity on the system and, if necessary, reconstruct events leading up to attempts to compromise the security of your system. Thus, it is not as much a method of protecting the system and its data as a method of analyzing and recording system use.

Anything that has to do with a user's access to the system or to a protected object within the system is considered a security-relevant activity. Such activities are called events. Typical events include the following:

• Logins, logouts, or login failures
• Changes to the authorization database
• Access to a protected object, such as a file, device, or global section
• Changes in privileges or the security attributes of protected objects

The operating system can record both successful and unsuccessful events. Sometimes the unsuccessful can be more revealing. For example, it is less important to record that a programmer displayed a file to which he had access than that the same programmer tried to but was prevented from displaying a protected file.

The event message itself can be written to two places: an audit log file or an operator terminal that is enabled to receive security class messages. As Example 9-1 shows, a message contains the following data:

1. Date and time of the message
2. Type of event
3. Date and time the event occurred
4. The process identification (PID) of the user who caused the event

Additional information in auditing messages is specific to the type of event. See Appendix D for examples of different messages.

%%%%%%%%%%%  OPCOM  25-JUL-1995 16:07:09.20  %%%%%%%%%%%     (1)
Message from user AUDIT$SERVER on GILMORE
Security alarm (SECURITY) on GILMORE, system id: 20300
Auditable event:          Process suspended ($SUSPND)        (2)
Event time:               25-JUL-1995 16:07:08.77            (3)
PID:                      30C00119                           (4)
Process name:             Hobbit
Username:                 HUBERT
Process owner:            [LEGAL,HUBERT]
Terminal name:            RTA1:
Image name:               $99$DUA0:[SYS0.SYSCOMMON.][SYSEXE]SET.EXE
Status:                   %SYSTEM-S-NORMAL, normal successful completion
Target PID:               30C00126
Target process name:      SMISERVER
Target username:          SYSTEM
Target process owner:     [SYSTEM]

Beyond a certain set of default reporting (see Table 9-1), the kind of security event information you receive depends on the kind of information you select from a long list of possible events. This section explains how to enable the reporting of security event information. Specifically, it discusses the following topics:

• Ways to generate event messages
• Types of events the system can report
• Sources of event information

Whenever you install or upgrade your system, the OpenVMS operating system automatically audits a limited number of events. These event categories, which are shown in Table 9-1, represent major changes in the security of your system. Depending on your site's requirements, you may want to enable other forms of reporting.

You can have the operating system report on security-related activity in three different ways:

• By enabling a category of events for auditing. For example, all login failures or all changes to system parameters can be reported.
• By attaching an access control entry (ACE) to a protected object. For example, any time a user modifies a particular file, a message can be generated.
• By modifying a user's authorization record so the system audits all operations performed from the account.

Security-relevant events are divided into a number of categories called event classes. The operating system audits several event classes by default (see Table 9-1). If the security requirements at your site justify additional auditing, you enable security auditing for additional event classes by using the DCL command SET AUDIT.

To enable auditing for different event classes, use the following command format:

The command requires two qualifiers to enable events:

• The /ENABLE qualifier defines the event classes you want audited. See Table 9-3 for a list of event classes.
• The /AUDIT qualifier or the /ALARM qualifier defines the destination for the event message.
  The /AUDIT qualifier directs the message to the audit log file, whereas the /ALARM qualifier directs the message to an operator terminal that has been enabled to receive security event messages. Critical events should be reported as both audits and alarms; less critical events can be written to a log file for later examination. The default event classes listed in Table 9-1 are audited as both alarms and audits.

The operating system begins auditing the new events on all nodes of the cluster as soon as you enable them. It continues auditing until you explicitly disable the classes with the /DISABLE qualifier. The current auditing configuration is recorded in SYS$MANAGER:VMS$AUDIT_SERVER.DAT and so it is preserved across system boots.

For more information about the SET AUDIT command, see the OpenVMS DCL Dictionary.

To see the event classes your site currently audits, enter the DCL command SHOW AUDIT. Example 9-3 displays the audit settings for a site with moderate security requirements.

Example of Enabling Event Classes

Although you can enable auditing for every possible class of security activity (/ENABLE=ALL), such an approach can result in an excessive number of auditing messages and generates too much information to analyze in a meaningful way. Therefore, Compaq suggests that you evaluate your needs, as described in Section 9.3.1, and selectively audit system activity.

You can enable auditing of event classes with different levels of granularity. You can use the following methods:

• Enable a class
  To enable auditing for all login failures, for example, you enable the logfailure class by entering the following command:

  $ SET AUDIT/AUDIT/ENABLE=LOGFAILURE=ALL

  
  As a result of this command, the audit server reports all login failures in the security audit log file.
• Enable a subset of a class
  With certain events, you may want to be more selective in the kinds of reporting you enable. For example, it makes more sense to enable network and remote login events rather than to enable all login events.
  To enable auditing of only the network and remote logins, enter the following command:

  $ SET AUDIT/AUDIT/ENABLE=LOGIN=(NETWORK,REMOTE)

• Enable successful, unsuccessful, or privileged events
  Event messages that report on normal system use can easily be eliminated if you enable only unsuccessful event reports or reports for activity performed through a certain privilege.
  When auditing access events to protected objects, in particular, you need to define your information requirements more finely than you would with event classes like logins or use of the Install utility. Files and certain other protected objects are accessed so often that full enabling of the related access event class can result in an overwhelming number of event messages---so many that they can possibly mask the unusual events that do require investigation. For this reason, it is recommended that you enable access auditing only for unusual conditions, such as unsuccessful or privileged access events.
  To enable auditing of unsuccessful file access events, enter the following command:

  $ SET AUDIT/AUDIT/ENABLE=ACCESS=FAILURE/CLASS=FILE

  
  Notice that the previous command enables auditing for all failed file accesses, not just failed read or write access attempts. This is recommended because access operations can be quite involved: what appears to be a simple write operation can involve several types of access. (For example, before writing to the file, the operation requires access to the volume and read access to the directory as well as access to the file within it.)
  Example 9-2 displays an event message from a file access failure. User Robinson tried to delete the file FOO.BAR, but an ACE on the file prevented it. Apparently, Robinson holds the identifier MINDCRIME, and an Identifier ACE on FOO.BAR denies access to those holding such an identifier. Furthermore, because the system owns the file, Robinson cannot gain delete access to the file through the protection code either.

Message from user AUDIT$SERVER on BILBO
Security alarm (SECURITY) and security audit (SECURITY) on BILBO, system id: 19662
Auditable event:          Object deletion
Event information:        file deletion request (IO$_DELETE)
Event time:               24-APR-1992 13:17:24.59
PID:                      47400085
Process name:             Hobbit
Username:                 ROBINSON
Process owner:            [ACCOUNTING,ROBINSON]
Terminal name:            OPA0:
Image name:               DSA2264:[SYS51.SYSCOMMON.][SYSEXE]DELETE.EXE
Object class name:        FILE
Object owner:             [SYSTEM]
Object protection:        SYSTEM:RWED, OWNER:RWED, GROUP:RE, WORLD:RE
File name:                _DSA2200:[ROBINSON]FOO.BAR;1
File ID:                  (17481,6299,1)
Access requested:         DELETE
Matching ACE:             (IDENTIFIER=MINDCRIME,ACCESS=NONE)
Sequence key:             00008A41
Status:                   %SYSTEM-F-NOPRIV, no privilege for attempted operation

As Section 9.2.1.1 describes, auditing access to protected objects requires careful thought because this type of event occurs so frequently. Too many event messages can overwhelm you and possibly mask the unusual events that do require investigation.

A more selective method of auditing protected objects is to include an auditing ACE in an object's access control list (ACL) and enable the ACL event class. With this approach, only access to objects with security-auditing ACEs results in an event message, not all objects of a class.

You can use two different types of auditing ACEs, depending on where you want the event reported. Alarm ACEs direct event messages to the operator terminal; whereas Audit ACEs direct event messages to the audit log file. Table 9-2 summarizes the auditing ACEs, and the OpenVMS System Management Utilities Reference Manual provides a full description of them. See Table 10-1 for a list of system files benefiting from auditing ACEs.

You attach an ACE to sensitive objects by using the DCL command SET SECURITY/ACL or the access control list editor (ACL editor). Always include the SUCCESS or FAILURE keyword (or both) in the ACCESS statement of an auditing ACE.

It is a good idea to define auditing ACEs for critical system files that are not automatically audited, such as the automatic login file SYSALF.DAT, the operator log file OPERATOR.LOG, or the system accounting file ACCOUNTING.DAT. Do not monitor all access conditions, however, because such an approach can generate a large volume of messages, many of which are not useful. For example, tracking successful write operations to OPERATOR.LOG probably will not produce interesting information, but tracking unsuccessful attempts probably will.

You can add auditing ACEs to any protected object, although files are the most common objects to audit. You may want to add an auditing ACE to either a print queue that is handling sensitive documents or to a terminal to catch attempted password grabbers (see Section 3.8).

Example of Adding an Auditing ACE

To establish an Alarm ACE for the file ACCOUNTING.DAT, enter the following command:

$ SET SECURITY/ACL=(ALARM=SECURITY,ACCESS=DELETE+CONTROL+SUCCESS+FAILURE)-
_$  SYS$MANAGER:ACCOUNTING.DAT

The ACL event class is enabled by default, but if it has been disabled at a site, you must enter the following command to reenable the use of auditing ACEs:

$ SET AUDIT/ALARM/AUDIT/ENABLE=ACL