Introduction

Test your rule content in a robust way with minimal effort required using Automatus - the test framework that is integrated into the content project.

Use Automatus to define various scenarios that can test your scanning and remediation code. Ensure that OVAL evaluates as expected under those scenarios. Test your remediation on scenarios that put the system into an incompliant, but fixable state - make sure that your remediations are able to fix such insecure system, so it passes when it is scanned again.

Automatus provisions systems according to test scenarios using VMs or containers. Then, it scans the provisioned systems, and if the scan results in failure, it runs a remediation and rescans. Finally, it retrieves results and presents them, so they can be acted upon.

To test content using VMs, Libvirt backend is used; to test content using containers, either Podman or Docker backend can be used.

Once a domain or container is prepared it can be used indefinitely. Automatus can perform tests focused on a profile or a specific rule.

Preparing a backend for testing

For Automatus to work, you need to have a backend image prepared for testing. You can use a powerful full-blown VM backend, or a lightweight container backend.

Libvirt backend

To use Libvirt backend, you need to have:

These packages installed in your host
- openssh-clients
- libvirt
- libvirt-daemon
- python2-libvirt/python3-libvirt
- virt-install (recommended, used by install_vm.py script)
- expect (recommended, used by install_vm.py script with --console option)
- libvirt-client (optional, to manage VMs via console)
- virt-manager (optional, to manage VMs via GUI)
- virt-viewer (optional, to access graphical console of VMs)
- ansible (required if remediating via Ansible)
- edk2-ovmf (required if you want to install UEFI based machine)
A VM that fulfils the following requirements:
- Package qemu-guest-agent installed
- Package openscap version 1.2.15 or higher installed
- root can login via ssh (it is recommended to setup key-based authentication)
- root can install packages (for RHEL7, it means subscription enabled).
- CPE_NAME is present in /etc/os-release. Currently, Ubuntu doesn’t ship it in the stock image. See this Ubuntu bug.

An easy way to install your VM is via install_vm.py script. It will setup a VM according to the requirements, including configuration of a key for SSH login.

Common usage looks like:

$ python install_vm.py --domain test-suite-fedora --distro fedora

By default, the key at ~/.ssh/id_rsa.pub is used. You can change default key used via --ssh-pubkey option.

By default, the VM will be created on the user hypervisor, i.e. qemu:///session. This should be enough for the testing, in case you need the VM to reside under qemu:///system, use the install script with --libvirt qemu:///system.

When installing a RHEL VM, you will still need to subscribe it. You can also run installation and provide custom URLs pointing to your repositories:

$ python install_vm.py --domain test-suite-rhel8 --distro rhel8 --url http://baseos_repo_addr --extra-repo http://appstream_repo_addr

By default, the script creates a BIOS based machine. If you need to create UEFI based machine, supply the --uefi normal or --uefi secureboot command line argument. The script will create UEFI based machine and make necessary changes to partitioning scheme. Note that the Libvirt backend cannot make snapshots of UEFI based machines. Therefore, you can’t use them with Automatus.

TIP: Create a snapshot as soon as your VM is setup. This way, you can manually revert in case the test run breaks something and fails to revert. Do not use snapshot names starting with the _ssgts prefix. You can create a snapshot using virsh or virt-manager.

Container backends

There are 2 container backends supported, Podman and Docker.

The container image needs to be prepared so it can accept ssh connections from root, and run OpenSCAP scan.
The image needs to fulfil the following requirements:

Package openssh-server installed
- root’s .ssh/authorized_keys is setup with correct permissions
- the container’s server host keys have to be in place
Packages scp and openssh-clients are installed
- scp requires more than a ssh server on the server-side
Package openscap-scanner version 1.2.15 or higher installed
You may want to include other packages, as base images tend to be bare-bone and tests may require more packages to be present.

You can use test_suite-* Dockerfiles in the content/Dockerfiles directory to build the images.

Podman

To use Podman backend, you need to have:

podman package installed

Building podman base image

Automatus will interact with the container by means of the root SSH access. If you don’t have an SSH key pair, setup a key without passphrase, so the procedure could happen without any additional interaction. You can skip this step if you already have an SSH key pair.

ssh-keygen -N ""

NOTE: With Podman you don’t have to be root in order to run tests and manage containers. But if you prefer to set up your test containers as superuser do keep in mind that root user typically doesn’t have an SSH key and you have to create it with sudo ssh-keygen -N “” command before moving forward. You can check if your root user has the key with a command like this: sudo test -f /root/.ssh/id_rsa && echo “Root user already has an id_rsa key” || echo “Root user has no id_rsa key”

Now when we all set with SSH keys let’s build the container. Go into the Dockerfiles directory of the project, and execute the following:

public_key="$(cat ~/.ssh/id_rsa.pub)"
podman build --build-arg CLIENT_PUBLIC_KEY="$public_key" -t ssg_test_suite -f test_suite-rhel .

or just call the build_test_container.sh script.

NOTE: If you are setting up the suite as superuser (i.e. sudo podman build …) use public_key=”$(sudo cat /root/.ssh/id_rsa.pub)” instead of the first command.

Docker

To use Docker backend, you need to have:

The docker Python module installed. You may have to use pip to install it on older distributions s.a. RHEL 7, running pip install --user docker as root will do the trick of installing it only for the root user.
the Docker service running, and
Rights that allow you to start/stop containers and to create images. This is achieved by:
- using sudo with every docker command, or;
- create a docker group, then add yourself to it and restart docker. Depending on your OS, you may need to logout for the group change to apply.

Building docker base image

The procedure is same as using Podman, you just swap the podman call with docker. But since Docker does not support rootless containers you will have to take superuser route of the guide.

Test scenarios

The test scenarios are used to test rules. They modify the system configuration so that OpenSCAP can return expected results.

The test scenarios for a rule are located in tests subdirectory in rule directory. The test scenarios are written in Bash. The format of the file name is scenario_name.expected_scan_result.sh, where scenario_name is an arbitrary name, and expected_scan_result is the result of the evaluation of the rule that oscap should return when the rule is evaluated. expected_scan_result can be one of pass, fail or notapplicable. It’s very important to keep this naming form.

For example:

something.pass.sh: Success scenario - script is expected to prepare machine in such way that the rule is expected to pass.
something.fail.sh: Fail scenario - script is expected to break machine so the rule fails. If initial scan fails as expected, Automatus tries to remediate machine and expects evaluation to pass after the remediation.

Scenarios format

Scenarios are simple Bash scripts. A scenario starts with a header which provides metadata. The metadata are parsed by the test framework and affect the test runs. After the header, arbitrary Bash commands can follow.

The header consists of comments (starting by #). Possible keys are:

packages is a comma-separated list of packages to install. Note that each package can be overridden by its platform-specific alternative if listed under platform_package_overrides in the product YAML. You should use the most common package names in this field and provide an alternative for any platform-specific names in the platform_package_overrides field.
platform is a comma-separated list of platforms where the test scenario can be run. This is similar to platform used in our remediations. Examples of values: multi_platform_rhel, Red Hat Enterprise Linux 7, multi_platform_all. If platform is not specified in the header, multi_platform_all is assumed.
profiles is a comma-separated list of profiles to which this scenario is restricted. Use this only if the scenario makes sense only in a specific profile. Typically, a rule doesn’t depend on a profile and behaves the same way regardless the profile it’s a part of. If the rule is parametrized by variables (XCCDF Values), use the variables key instead. This key is intended to be used in regression testing of bugs in profiles, it isn’t intended for casual use.
remediation is a string specifying one of the allowed remediation types (eg. bash, ansible, none). The none value means that the tested rule has no implemented remediation. The none value can also be used in case that remediation breaks test environment (for example unmounting /tmp in a test scenario would break test runs, because OpenSCAP generates reports into the /tmp directory).
templates has no effect at the moment.
variables is a comma-separated list of XCCDF values that sets a different default value for XCCDF variables in a form <variable name>=<value>. Typically, you use only one of profile or variables in scenario metadata - default values are effective only if the variable is not defined using a selector, which is exactly what profiles do.

Examples of test scenario:

Using platform and variables metadata:

#!/bin/bash
# platform = Red Hat Enterprise Linux 7,multi_platform_fedora
# variables = auth_enabled=yes,var_example_1=value_example

echo "KerberosAuthentication $auth_enabled" >> /etc/ssh/sshd_config

Augmenting using Jinja macros

Each scenario script is processed under the same jinja context as the corresponding OVAL and remediation content. This means that product-specific information is known to the scenario scripts at upload time (for example, {{{ grub2_boot_path }}}), allowing them to work across products. This also means Jinja macros such as {{{ bash_package_install(...) }}} work to install/remove specific packages during the course of testing (such as, if it is desired to both install and remove a package in the same scenario for the package_installed rules).

Note that this does have some limitations: knowledge of the profile (and the variables it has and the values they take) is still not provided to the test scenario. The above # profiles or # variables directives will still have to be used to add any profile-specific information.

Augmenting using `shared/templates`

Additionally, we have enabled test scenarios located under the templated directory, shared/templates/.../tests. Unlike with build-time content, tests does not need to be located in the template’s manifest (at template.yml). Instead, Automatus will automatically parse each rule and prefer rule-directory-specific test scenarios over any templated scenarios that the rule uses. (E.g., if installed.pass.sh is present in the template package_installed and in the tests/ subdirectory of the rule directory, the latter takes precedence over the former).

In addition to the Jinja context described above, the contents of the template variables (after processing in template.py) are also available to the test scenario. This enables template-specific checking.

You can place a test_config.yml file in rule’s tests folder to control usage of templated scenarios if they don’t fit for that particular rule for some reason. The file is jinja2-capable and product-aware, and you can use keys allow_templated_scenarios or deny_templated_scenarios that expect to contain a list of scenario basenames (including e.g. pass.sh suffix) to either test or to block. If you want to disable templated scenarios for a rule completely, allow only a scenario that doesn’t exist, s.a. none.

Example of adding new test scenarios

Let’s add test scenarios for rule accounts_password_minlen_login_defs.

Create tests directory within rule directory (in this case /linux_os/guide/system/accounts/accounts-restrictions/password_expiration/accounts_password_minlen_login_defs/tests further referenced as DIR).
write a few fail scripts - for example removing the line, commenting it, wrong value, etc. into DIR
write a pass script into DIR - (some rules can have more than one pass scenario)
build the data stream by running ./build_product --datastream-only fedora
run automatus.py with command:

./automatus.py rule --libvirt qemu:///session ssg-test-suite-fedora accounts_password_minlen_login_defs

Example of test scenarios for this rule can be found at: #3697

Running tests

To test you profile or rule use automatus.py script. It can take your SCAP source data stream, and test it on the specified backend. Automatus can test a whole profile or just a specific rule within a profile.

Argument summary

Mode of operation, specify one of the following commands;

rule: Evaluate a rule, remediate, and evaluate again in context of test scenarios.
profile: Evaluate, remediate and evaluate again using selected profile
combined: Evaluate, remediate, and evaluate again the rules from a profile in context of test scenarios.
template: Evaluate, remediate, and evaluate again the rules using a given template in context of test scenarios.

Specify backend and image to use:

To use VM backend, use the following option on the command line:
- Libvirt - --libvirt <hypervisor> <domain>
  - hypervisor: Typically, you will use the qemu:///session, or qemu:///system. It depends on where your VM resides.
  - domain: libvirt domain, which is basically name of the virtual machine.

NOTE: It might happen, especially while using other distros than Fedora or RHEL (for example Arch), that you encounter the following error:

libvirt: QEMU Driver error : operation failed Failed to take snapshot: Error: Nested VMX virtualization does not support live migration yet

This error might be followed by Python tracebacks where the above message is repeated. First make sure that you are running Automatus on the physical machine and that you really DO NOT use nested virtualization by running VM in VM.

If you pass this requirement, it might happen that nested virtualization is enabled for your KVM kernel module. Libvirt will refuse to do live migration in this case. You can check this by running:

$ cat /sys/module/kvm_intel/parameters/nested

If you see “Y” then the nested virtualization is enabled for the KVM kernel module and it needs to be disabled. This can be done temporarily by running:

# modprobe -r kvm_intel
# modprobe kvm_intel nested=0

or permanently by putting

options kvm_intel nested=0

into a file ending with .conf and placed into the /etc/modprobe.d/ directory.

To use container backends, use the following options on the command line:
- Podman - --container <base image name>
- Docker - --docker <base image name>

Specify SCAP source data stream to use:

--datastream: Path to the data stream that you want to use for scanning. It will be transferred to the scanned system via SSH. The option can be omitted if there is only one data stream in the build directory.

Specify as last argument the id of a profile or rule to be tested.

Note:Also, as containers may get any IP address, a conflict may happen in your local client’s known_hosts file. You might have a version of oscap-ssh that doesn’t support ssh connection customization at the client-side, so it may be a good idea to disable known hosts checks for all hosts if you are testing on a VM or under a separate user. You can do that by putting following lines in your $HOME/.ssh/config file:

StrictHostKeyChecking no
UserKnownHostsFile /dev/null

All test logs are stored in logs directory. The specific diretory is shown at the beginning of each test run.

If you want more verbose logs, pass the --dontclean argument that preserves result files, reports and verbose scanner output even in cases when the test result went according to the expectations. If your system has the oval-graph package installed that provides the arf-to-html command, Automatus will use it to extract OVAL evaluation details from ARFs, and save those condensed reports to the logs directory even if the --dontclean argument has been specified.

Rule-based testing

./automatus.py rule RULE ...

In this mode, you supply one or more rule IDs or wildcards as positional arguments. Unlike the profile mode, each rule is evaluated and tested independently, one-by-one.

Rule-based testing enables to perform two kinds of tests:

Check tests: The system is set up into a compliant, or a non-compliant state.

Typically, the compliant state is different from the default or post-remediation state. The scanner is supposed to correctly identify the state of the system, so it is checked against false positives and false negatives.
Remediation tests: The system is set up into a non-compliant state, and remediation is performed.

If you would like to test the rule sshd_disable_kerb_auth:

Using Libvirt:

./automatus.py rule --libvirt qemu:///system ssg-test-suite-rhel7 --datastream ../build/ssg-rhel7-ds.xml sshd_disable_kerb_auth

Using Podman:

./automatus.py rule --container ssg_test_suite --datastream ../build/ssg-rhel7-ds.xml sshd_disable_kerb_auth

or just call the test_rule_in_container.sh script that passes the backend options for you in addition to --remove-machine-only and --remove-platforms that remove some testing limitations of the container backend.

Using Docker:

./automatus.py rule --docker ssg_test_suite --datastream ../build/ssg-rhel7-ds.xml sshd_disable_kerb_auth

Notice we didn’t use full rule name on the command line. The prefix xccdf_org.ssgproject.content_rule_ is added if not provided.

It is possible to use wildcards, eg accounts_passwords* will run test scenarios for all rules which ID starts with accounts_passwords.

If the data stream file path is not supplied, auto detection is attempted by looking into the /build directory.

In the rule mode, the Automatus follows the profiles metadata key from the scenario headers. It will run test scenario for each profile listed in this profile key. If there is no profiles key in the test scenario header the Automatus will use the virtual (all) profile. It is a profile that contains all the rules.

Moreover, there is the --profile option which can be used to override the profile selection from profiles metadata so every test scenario will be executed only against the profile selected by this command-line argument and variable selections will be done according to this profile.

Debug mode

Use --debug option, to investigate a test scenario which is not evaluating to expected result. Automatus will pause the test run, and you will be able to SSH into the environment to inspect its state manually.

Profile-based testing

In this operation mode, you specify the profile command and you supply the profile ID as a positional argument. Automatus then runs scans over the target domain and remediates it based on particular profile.

To test RHEL7 STIG Profile on a VM:

./automatus.py profile --libvirt qemu:///session ssg-test-suite-rhel7 --datastream ../build/ssg-rhel7-ds.xml stig

To test Fedora Standard Profile on a Podman container:

./automatus.py profile --container ssg_test_suite --datastream ../build/ssg-fedora-ds.xml standard

To test Fedora Standard Profile on a Docker container:

./automatus.py profile --docker ssg_test_suite --datastream ../build/ssg-fedora-ds.xml standard

Note that profile-id is matched by the suffix, so it works the same as in oscap tool (you can use oscap info --profiles to see available profiles).

Unselecting problematic rules

Sometimes you would like to skip a rule in the profile because they are too slow to test, or you know a rule doesn’t have a remediation and you get less value by testing it.

Also, some rules need to be skipped in the profile mode because they might break the test backend. For example, the rule sshd_disable_root_login which disables root login to the tested VM will interfere with tests execution, because the Automatus uses root user in all underlying SSH commands.

For these situations, use ds_unselect_rules.sh to unselect these rules in all profiles of the data stream. It will copy your data stream to /tmp and unselect rules listed in rules_list

./ds_unselect_rules.sh <datastream> <rules_list>

Where:

datastream: is the data stream to unselect rules from
rules_list: is a file with list of complete rule IDs, one per line

Example usage:

./ds_unselect_rules.sh ../build/ssg-fedora-ds.xml unselect_rules_list

Tip: file unselect_rules_list contains a few typical rules you may want to skip

Combined testing mode

In this mode, you are testing the rules selected by a profile, using the contexts provided by each rule’s test scenarios. This mode provides an easy way of performing rule-based testing on all rules that are part of a profile.

In the combined mode, all rules selected by the given profile are tested. However, each rule is evaluated and tested separately.

The test scenarios are chosen to execute based on the presence and contents of the profiles metadata key in the test scenario header. If there is no profiles metadata key in a test scenario the test scenario will be executed using the virtual (all) profile. If there is a profiles key in a test scenario and the tested profile is listed under the profiles key, the test scenario will be executed using the tested profile. If there is the profiles key but it doesn’t contain the tested profile the test scenario will be skipped and won’t be executed. Most of the test scenarios do not have profiles key, therefore using the virtual (all) profile is the most frequent behavior. This way we can re-use test scenarios when testing any profile.

If you want to have a specific regression test only for a certain profile(s) which relies on a specific value being selected by some variable in this profile you need to use the profiles key in the test scenarios metadata to limit test scenario so it is executed only when testing profiles listed there.

# profiles = xccdf_org.ssgproject.content_profile_ID1,xccdf_org.ssgproject.content_profile_ID2,...

If a rule doesn’t have any test scenario, it will be skipped and a INFO message printed at the end.

If you would like to test all profile’s rules against their test scenarios:

./automatus.py combined --libvirt qemu:///system ssg-test-suite-rhel8 --datastream ../build/ssg-rhel8-ds.xml ospp

Template-based testing

./automatus.py template ... <template_name1>[ <template_name2> <template_name3> ...]

In this mode you can test all rules that use a template. This is very useful when one fixes a bug or makes improvements to the template. Each rule may use the template in a specfic way and this provides a way to easily run the test scenarios for all rules based on their template.

The test scenarios executed are based on the template and the rule that uses it. If the specified template doesn’t have tests, only the rule’s test scenarios are executed, and if a rule doesn’t have test scenarios it won’t be tested. If the specified template does have tests they are combined with the rule’s tests, this is the same behavior we see in the rule mode.

Analysis of results

The rule tests results are saved as results.json into the corresponding log directory. You can then analyze those results by running e.g.

python analyze_results.py $(find . -name results.json)

The tool will print some general statistics and it will give you more detailed information about scenarios that yielded different results. Sometimes, different results may have been caused by different test environments, whereas sometimes the security content is different, and those scans can be identified by respective scanning dates only.

Known issues

1 - Test suite fails to test rule with the message:

Rule <rule_id> has not been evaluated! Wrong profile selected in test scenario?

If you are using SCAP 1.3 content (which is built by default) and you are sure that you have selected the rule in the particular profile, it might be that the target scan environment has an OpenSCAP version contains a bug with SCAP 1.3 content. To solve this issue you can either update the OpenSCAP package in the target scan environment to the latest version or build SCAP 1.2 content. To build SCAP 1.2 content check Building compliant SCAP 1.2 content.