My first real exposure to auditd has been deploying Rapid7 InsightIDR. Its been a great tool but I did not understand why they require auditd to be disabled. Is it not more secure having an audit daemon running? After a server had rebooted to completely disable this, it ran into a ton of performance issues and my path lead me down better understanding auditd.
What is Auditd?
At a high level it is a piece of software that allows you to view and audit actions being performed on the operating system. These actions can be directory listing, program execution, audit policy changes.
How is this Facilitated?
There are two pieces to the puzzle. There is a kernel based kauditd which queues up audit events for consumption. Various syscalls and other events cause an audit event to trigger.
These events are consumed via a userland auditd process. This process enables the auditing (via auditctl) and registers its PID and starts consuming events pushed through a PIPE. by default these events are simply logged to /var/log/audit/audit.log.
In the event that the auditd can’t consume these properly there is an audisp process that handles the overflow. Perhaps the auditd crashes but audit events are still pumped to the PIPE. Audisp will generically pump these to syslog which may push to /var/log/messages instead.
You can use the auditctl command to view the status or add rules. These rules would allow you to monitor extra resources or ignore certain ones.
Why Does Rapid7 Want This Disabled?
The Rapid7 ir_agent has its own consumer process that scans the auditd events. Unfortunately (unless compatibility mode is enabled), only one process can consume these and Rapid7 wants its agent to be the consumer. It does so in an interesting way though. kauditd knows events can’t always be consumed in realtime so there are provisions for a backlog buffer which Rapid7 increases. It does this so a background job can run periodically to pull these events from the queue instead of listening in realtime. I’m not sure of the benefit of this method but it seems to work.
Without auditd disabled and stopped though, no others can listen, including Rapid7.
# auditctl -s enabled 0 failure 0 pid 0 rate_limit 0 backlog_limit 65752 lost 0 backlog 0
Here you can see, it is currently disabled and no pid is registered. Importantly though you can see the high backlog_limit (buffer). It is normally under 300 because auditd is running constantly.
So Why Did Our Server Blow Up?
Well, in our case its a server that generates a ton of audit logs because it kicks off tens of processes per second. It was easy to dump those to a audit.log file but a security tool actually parsing those just couldn’t keep up. We will either need to filter some of those auditd events or disable the agent.
If you ever need to use auditd, hopefully this gives you a basic insight into it.