Creating custom detection policies

You can create Sophos Linux Sensor (SLS) policies 'from scratch', allowing expert users to craft their policies as they see fit.

Before you start

The SLS content package is installed in /var/lib/sophos/content/runtimedetections-content.yaml, a single content file containing all detections with their default configurations. We don't recommend changing the contents of this file because it's overwritten when you update the content package.

You can create custom detections and lists by creating and editing the /etc/sophos/runtimedetections-rules.yaml configuration file. When the content package is updated, /var/lib/sophos/content/runtimedetections-content.yaml is overwritten but /etc/sophos/runtimedetections-rules.yaml is not. This keeps any custom detection content you've created for your environment.

When a sensor starts up, it reads the packed content and then applies any overrides. The sensor writes a "reference" copy of the merger of the runtimedetections-content.yaml with the runtimedetections-rules.yaml to this location: /var/run/sophos/cache_analytics.yaml. This cache yaml file represents the result of the merger, which you can inspect or save for debugging or auditing purposes.

Configuring policy rules

Start by selecting what type of policy you want to configure. We recommend exploring policies by Alert category or MITRE ATT&CK category to determine what behavior you want to monitor at which point of the attack life-cycle.

Rule actions

Once you've selected your policy type, the next step is to determine whether you prefer to default match or default ignore. You must specify a default action before adding custom rules:

Custom rules begin with either the match or ignore keyword, followed by predicates combined with event fields. A policy may have any number of match or ignore rules. If a policy has multiple rules, SLS evaluates the rules in the order shown beginning from the top of the list.

Event fields

Each policy has its own set of valid event fields, including metadata related to containers, images, programs, ports, files, users, groups, and more. When deciding which event fields to include from the policy's range of valid event fields, we recommend starting with basic "known good" or "known bad" resources so you can become more familiar with configuration.

For instance, if you know that your developers frequently spawn interactive shells ssh as part of their work, you should use the parentProgramName == "ssh" event field plus the ignore action to ensure your developers' legitimate behaviors don't generate alerts.

Example

Let's walk through a Program Execution Policy configuration to see rule creation in practice. We'll start with the following policy configuration, which will alert on any program executing.

Program Execution Policy Example:
  policy: program
  enabled: true
  alertMessage: Unauthorized Program Executed
  comments: Example using the program policy
  priority: High
  rules:
    - default match

Let's say we want to only receive alerts on programs named test.sh, regardless of file path. We would modify the rules to specify that we want to filter programName for test.sh, and to ignore any programs that don't match that filter:

rules:
    - match programName == "test.sh"
    - default ignore

However, let's say we now want to restrict alerts to only trigger when test.sh is run from the /tmp/ directory. We would modify our rules to include the path as well:

rules:
    - match programName == "/tmp/test.sh"
    - default ignore

Now, we decide we want to alert on any execution that contains the argument "malicious," as well (in the case of not-so-sneaky attackers). We can use the programArguments event field to indicate we want to see any arguments containing *malicious*.

rules:
    - match programArguments == "* *malicious*"
    - match programName == "/tmp/test.sh"
    - default ignore

Finally, we realize that we also want to ignore any execution that originates from a bash shell, perhaps because we know that there's legitimate bash shell usage within our environment. We can create an ignore rule, using the parentProgramName event field and specifying bash:

rules:
    - ignore parentProgramName == "bash"
    - match programArguments == "* *malicious*"
    - match programName == "/tmp/test.sh" and programArguments == "/tmp/test.sh arg1 arg2"
    - default ignore

Rules are evaluated strictly in the order shown, beginning from the top of the list. Running ls malicious within a bash shell won't generate an alert. However, if we switched the order of the first two rules, an alert is generated if we run ls malicious from within a bash shell, as it matches on the programArguments rule before the ignore bash rule is applied.

Lists

For more information on lists, see Advanced topic: lists.

Alerts

You can configure each policy's generated alerts using alert templates to contain specific or custom information. For teams with established processes, alert customization helps you receive precisely the data needed to conduct your triage and investigations. See Alert configuration templates.

Automated responses

You can configure each policy to trigger automated responses upon alert. Valid automated response actions include stopping a process, killing a process, killing a container, quarantining a file, and deleting a file. See Getting started with automated response.

Alert batching

If alert volume is a concern, alert batching may be turned on at a policy level to limit the number of alerts which a policy may emit within a given time frame. A policy may specify a tolerance and duration for alert batching, and the number of alerts generated by that policy within that duration won't exceed the given tolerance. Once the duration expires, the policy will then emit one batched alert which summarizes all the suppressed alerts.

Continuing with our earlier program policy example - perhaps we find that after deploying it, we have a tool that runs /tmp/test.sh arg1 arg2 many times in quick succession, resulting in an excess number of alerts. We can use alert batching to group these, reducing the number of alerts produced. Consider the alert batching configuration below:

alertBatching:
  enabled: true
  tolerance: 1
  duration: 10s

This turns on alert batching as soon as a single match occurs, and will batch all matches in the 10-second window. If we added the above to our Program Execution Policy example, and then ran our /tmp/test.sh script 20 times within 10 seconds, we'd see two alerts:

• One alert as soon as the first match was seen.
• A second alert at the end of the 10-second window, which would note that the same alert had fired 19 additional times during that window.