This document is intended to simplify the use of the auditing system provided in AIX and applies to all versions of AIX. It includes information on what auditing offers, what its requirements are, and what common problems may be encountered. The intention is not to answer every question about auditing, but to provide a starting point for understanding and setting up auditing.
Documentation for System Auditing can be found in Chapter 5 of the "System Management Guide" for AIX version 3.2, Chapter 3 of the "System Management Guide: Operating System and Devices" for AIX version 4, and Chapter 8 in the IBM Redbook "Elements of Security: AIX 4.1" (GG24-4433-00).
The AIX and RS/6000 product documentation library is also available:
http://www.rs6000.ibm.com/resource/
The auditing subsystem provides the means to record security-related information and to alert system administrators of potential and actual violations of the system security policy. The information collected by auditing includes: the name of the auditable event, the status (success or failure) of the event, and any additional event-specific information related to security auditing.
A list of audit events built into AIX, along with a list of predefined audit objects, can be found in the file /etc/security/audit/events.
In general, auditing events are defined at the system call level. A single operation at the command line would result in records of several events in the audit trail. For example, when viewing a file using the cat or more command, you would see the following records logged into the audit trail:
FILE_Open (file is opened)
FILE_Read (file is read)
FILE_Write (file is written to standard output)
PROC_Create (process creation for more OR cat)
PROC_Execute (command execution)
PROC_Delete (process completion)
Auditing all possible events can produce a huge amount of data. Through audit controls (that is, modifying the configuration files), you can select events to be recorded.
Audit events are grouped into classes. The events can be defined by which events are in a class. While the class names are arbitrary, they, rather than individual event names, are associated with user IDs when the audit subsystem is active.
Auditing objects are just individual files that will be monitored. Three operations can be audited: read, write, and execute. Objects are not associated with user IDs. Audit records are generated whenever an audited object is referenced by any user (including root).
To add further audit objects, extend the /etc/security/audit/objects file.
There are two modes of operation for auditing: BIN and STREAM. BIN mode writes the audit trail to alternating temporary files (bins), then to a single trail file. STREAM mode writes to a circular buffer that is read synchronously through an audit pseudo-device (/dev/audit).
An audit can be started in one OR both of these modes.
Using the audit configuration setup shipped with AIX, /etc/security/audit/config, the BIN mode alternates between /audit/bin1 and /audit/bin2. When one BIN is full (the binsize parameter determines the size of the bin), the audit switches to the other BIN file while adding the accumulated data in the first file to the audit trail (defined in /etc/security/audit/bincmds), /audit/trail. Use "audit shutdown" to be certain that all audit records have been added to /audit/trail. The BIN mode audit record is in binary format. You can read it with audit commands such as auditpr.
In STREAM mode, the default AIX configuration provides a program to read the STREAM buffer and processes each record with the commands found in /etc/security/audit/streamcmds. These commands format the output into human-readable form and write it in /audit/stream.out. This file is NOT cumulative; it is restarted every time the audit is restarted. The STREAM audit trail can be read in real time by reading /audit/stream.out, or by sending output directly to a terminal or printer.
There are five audit subcommands for invoking auditing. They are as follows:
audit start - to activate the audit subsystem (This is
the only correct way to start audit.)
audit shutdown - to stop auditing subsystem, processing
final BIN records and removing the
/audit/auditb file that is used as an
"active" indicator by the audit modules
audit off - to suspend auditing temporarily
audit on - to resume audit after audit off
audit query - to display the status of auditing
NOTE: Using audit commands in the wrong order can confuse the auditing subsystem. If the auditing subsystem gets confused, reset everything by deleting all files in the /audit directory (except trail, stream.out and bin files).
Auditing can be run at the discretion of the system administrator. Depending on the environment, it is usually not necessary to have auditing running at all times. If it is configured to monitor a large number of events or objects at all times, the amount of data generated would be so substantial that its overhead would outweigh its benefit. It is worthwhile to take time to configure auditing to collect selected information.
To start auditing at system startup, add the following line to the /etc/rc file, just prior to the line reading dspmsg rc.cat 5 'Multi-user initialization completed':
/usr/sbin/audit start
If auditing is running at all times, make sure to run or to add the following in the /usr/sbin/shutdown script to properly shut down:
/usr/sbin/audit shutdown
All auditing related configuration files reside in /etc/security/audit. The /etc/security/audit/config file contains the key audit controls. It has the following stanzas:
start - specifies whether BIN or STREAM
(or both) should be used for auditing
bin and stream - contain controls for each mode; the names
of the BIN files are specified here
classes - defines several groups (classes) of
auditing events
The predefined classes are: general, objects, SRC, kernel, files, SVIPC, mail, cron, and TCPIP. New classes can be defined using the auditing events in the /etc/security/audit/events file. All audit classes except the objects class are associated with user IDs. For example, audit the events defined as "general" and "TCPIP" for user root.
user stanza - lists specified users and the audit classes
assigned to them; each username must be the
login name of a system user or the string
'default'
An example of this stanza is as follows:
users:
root = general
joe = general, files, TCPIP
default = general
When auditing starts, it ALWAYS audits the events specified for every user ID defined in the config file and ALL the objects defined in /etc/security/audit/objects. If the objects' audit records are not wanted, remove or comment out (using an *) the objects defined in the objects file.
If there are specific classes of events that are not wanted as audit records, specify No_Events for that specific class in the config file.
For example:
files = No_Events
or
tcpip = No_Events
The objects file contains all objects to be audited when auditing is active. A user defined object is displayed as:
/home/joe/my.stuff:
r = "JOE_READ"
w = "JOE_WRITE"
The names JOE_READ and JOE_WRITE are referenced in the /etc/security/events file to define the format of the auditpr output:
JOE_READ = printf "%s"
JOE_WRITE = printf "%s"
NOTE: There is no need to add the newly added objects to the objects stanza in the /etc/security/audit/config file, since the objects line is not referenced. Only the objects file is referenced.
The streamcmds file has commands that are entered for STREAM audit records. The default file contains one command.
Enter:
/usr/sbin/auditstream | auditpr > /audit/stream.out &
Adding the -v flag for the auditpr command improves this command at the expense of having more information. Without -v, full path names for files are not shown in the audit output; only file descriptors are recorded.
To limit the amount of data collected during the auditing operation, use the -c option on the auditstream command to select a specific class of events as defined in the config file, or use the auditselect command to select specific events.
For example:
NOTE: This command must be all on one line in the streamcmds file. This command will collect only FILE_Open event records.
/usr/sbin/auditstream | /usr/sbin/auditselect -e "event == FILE_Open" | auditpr -v > /audit/stream.out &
NOTE: The following command will limit data collection to only the TCPIP class of events as defined in the config file.
/usr/sbin/auditstream -c tcpip | auditpr -v > /audit/stream.out &
The bincmds file contains commands that are entered whenever a BIN file fills or when auditing is shut down. The file distributed reads like the following:
/usr/sbin/auditcat -p -o $trail $bin
The environment variables in the preceding command are defined while auditing is active. The auditselect command can be added to select specific events, reducing the amount of audit records.
The bincmds file will only collect audit records that match USER_SU or USER_Login audit events. Enter:
/usr/sbin/auditselect -e "event== USER_SU || event== \
USER_Login" $bin > /audit/trail.login
For example:
*User NAME [joe] AUDIT classes [general, files]
A user stanza should be displayed for joe in /etc/security/audit/config file.
procmon = PROC_Create, PROC_Delete, PROC_Execute
In the users stanza, the following could exist:
joe = procmon
The newly assigned audit classes will take effect at the next login for user joe.
auditpr -v < /audit/trail | more
The STREAM mode audit file /audit/stream.out can be viewed directly. Remember that the /audit/stream.out file is rewritten each time the auditing subsystem is started. Save the old stream.out before starting auditing.
If you do not want the objects audit records when auditing a user ID, comment out the objects defined in the /etc/security/audit/objects file or rename this file.
In the following example, all processes writing to the /etc/utmp file will be audited.
/etc/utmp: w = "UTMP_WRITE"
* /etc/utmp UTMP_WRITE = printf " %s "
The audit record is displayed as follows:
UTMP_WRITE root OK Wed Jul 12 12:12:25 1995 init
In this case, the init process owned by root wrote to the file.
NOTE: The length of an audit event or object name cannot exceed 15 characters. This limit is defined in the header file /usr/include/sys/audit.h, ah_event [16]. The following error message usually indicates an invalid event or object name.
"auditevents (): Invalid argument"
Each record in the audit trail takes about 50 to 150 bytes depending on what mode is used and whether the verbose mode flag is specified. This means that 1MB of data could contain about 6800 entries.
It is important to specify what information should be reviewed while auditing. Although you can configure auditing to record events of interest, there may still be too much data to be useful when viewed all at once. The auditselect command can be used with auditpr to sort through volumes of information and pull out only that which is needed for a specific report. It can be used to pull all data from a specific time period, for a specific user, or for a specific event, or any combination of these three.
For example:
/usr/sbin/auditselect -f /audit/pick \
/audit/trail | /usr/sbin/auditpr -v
The /audit/pick file reads as follows:
command == rlogin && \
time >= 08:00:00 && time <= 17:00:00 && \
data >= 04/01/96 && date <= 04/12/96
This command reports the use of the rlogin command within the specific time interval (8AM-5PM between April 1 and April 12).
The compressed trail data from the binmode auditing is not in the same format at AIX version 3.2 as it is in 4.1 or later. There is a utility to convert the data from a pre-AIX Version 4 format to the Version 4 format. It is a command called auditconv.
Error Message:
** failed setting kernel audit objects
This occurs when there is a syntax error in the /etc/security/audit/objects file.
auditbin: ** failed backend command
/etc/auditcat -p -o /audit/trail -r /audit/bin1
This error can be corrected by removing or renaming the BIN files. It is sometimes helpful to run audit shutdown again and then to retry audit start.
It is necessary to have the "user" stanza in the /etc/security/audit/ configuration file or the following error will display when you start auditing:
Unable to find the user stanza in /etc/security/audit/config
If it is not obvious that the user stanza is missing, verify that each of the classes are defined on a single continuous line.
Given the way that cron and the TCPIP code is written, each sets up its own set of audit events. These events will get written into the audit trail regardless of how the config files are set up. The workaround is to use auditselect to exclude these events when generating the audit report. TCPIP sessions, ftpd, rexecd, and telnetd all call auditproc() to set up process auditing using the class tcpip in /etc/security/audit/config. The same thing is done in the cron code (at, cron, and cronadm) for the cron class in /etc/security/audit/config. These events will be written into the audit trail. The best thing to do is to filter them using auditselect.
For example:
auditselect -e"event!=AT_JobAdd && event!=AT_JobRemove && ..."
This will exclude events AT_JobAdd and At_JobRemove and so on.
Or select on the command name:
auditselect -e"command!=cron && command!=at && ..."
This excludes recording related to commands, at, cron, and so on.
As defects arise, updates to the AIX audit code are made.