Audit Terminology and Concepts
The following terms are used to describe the auditing service. Some definitions include pointers to more complete descriptions.
Table 28-1 Solaris Auditing Terms
Security-relevant system actions can be audited. These auditable actions are defined as audit events. Audit events are listed in the /etc/security/audit_event file. Each audit event is defined in the file by an event number, a symbolic name, a short description, and the set of audit classes to which the event belongs. For more information on the audit_event file, see the audit_event(4) man page.
For example, the following entry defines the audit event for the exec() system call:
When you preselect for auditing either the audit class ps or the audit class ex, then exec() system calls are recorded in the audit trail.
Solaris auditing handles attributable and nonattributable events. The exec() system call can be attributed to a user, so the call is considered an attributable event. Events are nonattributable if the events occur at the kernel-interrupt level. Events that occur before a user is authenticated are also nonattributable. The na audit class handles audit events that are nonattributable. For example, booting the system is a nonattributable event.
When the class to which an audit event belongs is preselected for auditing, the event is recorded in the audit trail. For example, when you preselect the ps and na audit classes for auditing, the exec() system calls and system boot actions, among other events, are recorded in the audit trail.
In addition to the audit events that are defined by the Solaris auditing service, third-party applications can generate audit events. Audit event numbers from 32768 to 65535 are available for third-party applications.
Audit Classes and Preselection
Each audit event belongs to an audit class or classes. Audit classes are convenient containers for large numbers of audit events. When you preselect a class to be audited, you specify that all the events in that class should be recorded in the audit trail. You can preselect for events on a system and for events initiated by a particular user. After the auditing service is running, you can dynamically add or remove audit classes from the preselected classes.
Audit classes are defined in the /etc/security/audit_class file. Each entry contains the audit mask for the class, the name for the class, and a descriptive name for the class. For example, the ps and na class definitions appear in the audit_class file as follows:
0x00100000:ps:process start/stop 0x00000400:na:non-attribute
There are 32 possible audit classes. The classes include the two global classes: all and no. The audit classes are described in the audit_class(4) man page.
The mapping of audit events to classes is configurable. You can remove events from a class, add events to a class, and create a new class to contain selected events. For the procedure, see How to Change an Audit Event's Class Membership.
Audit Records and Audit Tokens
Each audit record records the occurrence of a single audited event. The record includes information such as who did the action, which files were affected, what action was attempted, and where and when the action occurred. The following example shows a login audit record:
header,81,2,login - local,,2003-10-13 11:23:31.050 -07:00 subject,root,root,other,root,other,378,378,0 0 example_system text,successful login return,success,0
The type of information that is saved for each audit event is defined by a set of audit tokens. Each time an audit record is created for an event, the record contains some or all of the tokens that are defined for the event. The nature of the event determines which tokens are recorded. In the preceding example, each line begins with the name of the audit token. The content of the audit token follows the name. Together, the four audit tokens comprise the login audit record.
For a detailed description of the structure of each audit token with an example of praudit output, see Audit Token Formats. For a description of the binary stream of audit tokens, see the audit.log(4) man page.
Audit records are collected in audit logs. Solaris auditing provides two output modes for audit logs. Logs that are called audit files store audit records in binary format. The set of audit files from a system or site provide a complete audit record. The complete audit record is called the audit trail.
The syslog utility collects and stores text version summaries of the audit record. A syslog record is not complete. The following example shows a syslog entry for a login audit record:
Oct 13 11:24:11 example_system auditd: [ID 6472 audit.notice] \ login - login ok session 378 by root as root:other
A site can store audit records in both formats. You can configure the systems at your site to use binary mode, or to use both modes. The following table compares binary audit records with syslog audit records.
Table 28-2 Comparison of Binary Audit Records With syslog Audit Records
Binary records provide the greatest security and coverage. Binary output meets the requirements of security certifications, such as the Common Criteria Controlled Access Protection Profile (CAPP). The records are written to a file system that is protected from snooping. On a single system, all binary records are collected and are displayed in order. The GMT timestamp on binary logs enables accurate comparison when systems on one audit trail are distributed across time zones. The praudit -x command enables you to view the records in a browser in XML. You can also use scripts to parse the XML output.
In contrast, the syslog records provide greater convenience and flexibility. For example, you can collect the syslog data from a variety of sources. Also, when you monitor audit.notice events in the syslog.conf file, the syslog utility logs an audit record summary with the current timestamp. You can use the same management and analysis tools that you have developed for syslog messages from a variety of sources, including workstations, servers, firewalls, and routers. The records can be viewed in real time, and can be stored on a remote system.
By using syslog.conf to store audit records remotely, you protect log data from alteration or deletion by an attacker. On the other hand, when audit records are stored remotely, the records are susceptible to network attacks such as denial of service and spoofed source addresses. Also, UDP can drop packets or can deliver packets out of order. The limit on syslog entries is 1024 characters, so some audit records could be truncated in the log. On a single system, not all audit records are collected. The records might not display in order. Because each audit record is stamped with the local system's date and time, you can not rely on the timestamp to construct an audit trail for several systems.
For more information on audit logs, refer to the following:
An audit directory holds audit files in binary format. A typical installation uses many audit directories. The contents of all audit directories comprise the audit trail. Audit records are stored in audit directories in the following order:
The directories are specified in the audit_control file. A directory is not used until a directory that is earlier in the list is full. For an annotated audit_control file with a list of directory entries, see Example 30-3.
Examining the Audit Trail
The auditing service provides commands to combine and reduce files from the audit trail. The auditreduce command can merge audit files from the audit trail. The command can also filter files to locate particular events. The praudit command reads the binary files. Options to the praudit command provide output that is suitable for scripting and for browser display.