Manual Pages  — AUDITPIPE

While audit trail files generated with audit(4) and maintained by auditd(8) provide a reliable long-term store for audit log information, current log files are owned by the audit daemon until terminated making them somewhat unwieldy for live monitoring applications such as host-based intrusion detection. For example, the log may be cycled and new records written to a new file without notice to applications that may be accessing the file.

The audit facility provides an audit pipe facility for applications requiring direct access to live BSM audit data for the purposes of real-time monitoring. Audit pipes are available via a clonable special device, /dev/auditpipe, subject to the permissions on the device node, and provide a "tee" of the audit event stream. As the device is clonable, more than one instance of the device may be opened at a time; each device instance will provide independent access to all records.

The audit pipe device provides discrete BSM audit records; if the read buffer passed by the application is too small to hold the next record in the sequence, it will be dropped. Unlike audit data written to the audit trail, the reliability of record delivery is not guaranteed. In particular, when an audit pipe queue fills, records will be dropped. Audit pipe devices are blocking by default, but support non-blocking I/O, asynchronous I/O using SIGIO, and polled operation via select(2) and poll(2).

Applications may choose to track the global audit trail, or configure local preselection parameters independent of the global audit trail parameters.

By default, the audit pipe facility configures pipes to present records matched by the system-wide audit trail, configured by auditd(8). However, the preselection mechanism for audit pipes can be configured using alternative criteria, including pipe-local flags and naflags settings, as well as auid-specific selection masks. This allows applications to track events not captured in the global audit trail, as well as limit records presented to those of specific interest to the application.

The following ioctls configure the preselection mode on an audit pipe:

AUDITPIPE_GET_PRESELECT_MODE
	Return the current preselect mode on the audit pipe. The ioctl argument should be of type int.
AUDITPIPE_SET_PRESELECT_MODE
	Set the current preselection mode on the audit pipe. The ioctl argument should be of type int.

Possible preselection mode values are:

AUDITPIPE_PRESELECT_MODE_TRAIL
	Use the global audit trail preselection parameters to select records for the audit pipe.
AUDITPIPE_PRESELECT_MODE_LOCAL
	Use local audit pipe preselection; this model is similar to the global audit trail configuration model, consisting of global flags and naflags parameters, as well as a set of per-auid masks. These parameters are configured using further ioctls.

After changing the audit pipe preselection mode, records selected under earlier preselection configuration may still be in the audit pipe queue. The application may flush the current record queue after changing the configuration to remove possibly undesired records.

The praudit(1) utility may be directly executed on /dev/auditpipe to review the default audit trail.

poll(2), select(2), audit(4), dtaudit(4), audit_control(5), audit(8), auditd(8)

See the audit(4) manual page for information on audit-related bugs and limitations.

The configurable preselection mechanism mirrors the selection model present for the global audit trail. It might be desirable to provide a more flexible selection model.

The per-pipe audit event queue is fifo, with drops occurring if either the user thread provides in sufficient for the record on the queue head, or on enqueue if there is insufficient room. It might be desirable to support partial reads of records, which would be more compatible with buffered I/O as implemented in system libraries, and to allow applications to select which records are dropped, possibly in the style of preselection.