Contributing with XCCDFs, OVALs and remediations

There are three main types of content in the project, they are rules, defined using the XCCDF standard, checks, usually written in OVAL format, and remediations, that can be executed on ansible, bash, anaconda installer, puppet, ignition and kubernetes. ComplianceAsCode also has its own templating mechanism, allowing content writers to create models and use it to generate a number of checks and remediations.

Contributing

Contributions can be made for rules, checks, remediations or even utilities. There are different sets of guidelines for each type, for this reason there is a different topic for each of them.

Following lines are rather reference material. If you are completely new and would like to try more guided approach, we recommend you to go through materials prepared for our Content creation workshop. They can be found here. They guide you step by step through the process of creating new rules, profiles, checks and remediations.

Rules

Rules are input described in YAML which mirrors the XCCDF format (an XML container). Rules are translated to become members of a Group in an XML file. All existing rules for Linux products can be found in the linux_os/guide directory. For non-Linux products (e.g., jre), this content can be found in the <product>/guide. The exact location depends on the group (or category) that a rule belongs to.

For an example of rule group, see linux_os/guide/system/software/disk_partitioning/partition_for_tmp/rule.yml. The id of this rule is partition_for_tmp; this rule belongs to the disk_partitioning group, which in turn belongs to the software group (which in turn belongs to the system group). Because this rule is in linux_os/guide, it can be shared by all Linux products.

Rules describe the desired state of the system and may contain references if they are parts of higher-level standards. All rules should reflect only a single configuration change for compliance purposes.

Structurally, a rule is a YAML file (which can contain Jinja macros) that represents a dictionary.

A rule YAML file has one implied attribute:

id: The primary identifier for the rule to be referenced from profiles. This is inferred from the file name and links it to checks and fixes with the same file name.

A rule itself contains these attributes:

title: Human-readable title of the rule.
rationale: Human-readable HTML description of the reason why the rule exists and why it is important from the technical point of view. For example, rationale of the partition_for_tmp rule states that:

The <tt>/tmp</tt> partition is used as temporary storage by many programs. Placing <tt>/tmp</tt> in its own partition enables the setting of more restrictive mount options, which can help protect programs which use it.
description: Human-readable HTML description, which provides broader context for non-experts than the rationale. For example, description of the partition_for_tmp rule states that:
requires: The id of another rule or group that must be selected and enabled in a profile.
conflicts: The id of another rule or group that must not be selected and disabled in a profile.

The <tt>/var/tmp</tt> directory is a world-writable directory used for temporary file storage. Ensure it has its own partition or logical volume at installation time, or migrate it using LVM.

severity: Is used for metrics and tracking. It can have one of the following values: unknown, info, low, medium, or high.

Level	Description
`unknown`	Severity not defined (default)
`info`	Rule is informational only. Failing the rule doesn’t imply failure to conform to the security guidance of the benchmark.
`low`	Not a serious problem
`medium`	Fairly serious problem
`high`	Grave or critical problem

When deciding on severity levels, it is best to follow the following guidelines:

Table Vulnerability Severity Category Code Definitions
Severity	DISA Category	Category Code Guidelines
`high`	`CAT I`	Any vulnerability, the exploitation of which will directly and immediately result in loss of Confidentiality, Availability, or Integrity.
`medium`	`CAT II`	Any vulnerability, the exploitation of which has a potential to result in loss of Confidentiality, Availability, or Integrity.
`low`	`CAT III`	Any vulnerability, the existence of which degrades measures to protect againstloss of Confidentiality, Availability, or Integrity.

The severity of the rule can be overridden by a profile with refine-rule selector.

platform or platforms: Defines applicability of a rule. It is specified either as a single platform or as a list of platforms. For example, if a rule is not applicable to containers, the list should contain the item machine, which means it will be evaluated only if the targeted scan environment is either bare-metal or virtual machine. Also, it can restrict applicability on higher software layers. By setting to shadow-utils, the rule will have its applicability restricted to only environments which have shadow-utils package installed. The available options can be found in the file <product>/cpe/<product>-cpe-dictionary.xml (e.g.: rhel8/cpe/rhel8-cpe-dictionary.xml). In order to support a new value, an OVAL check (of inventory class) must be created under shared/checks/oval/ and referenced in the dictionary file. It is possible to specify multiple platforms in the list. In that case, they are implicitly connected with “OR” operand.
ocil: Defines asserting statements to check whether or not the rule is valid.
ocil_clause: This attribute contains the statement which describes how to determine whether the statement is true or false. Check out rule.yml in linux_os/guide/system/software/disk_partitioning/encrypt_partitions/: this contains a partitions do not have a type of crypto_LUKS value for ocil_clause. This clause is prefixed with the phrase Is it the case that <ocil_clause> ?.

A rule may contain those reference-type attributes:

identifiers: This is related to products that the rule applies to; this is a dictionary. Currently, only the Common Configuration Enumeration or CCE identifier is supported. Other identifiers can be added as well. Contributions to add these other identifiers are welcomed. The table below shows a list of common identifiers and their current support in a rule:

URI	Supported	Identifier Value Description
http://cce.mitre.org	Yes	Common Configuration Enumeration (CCE) – the identifier value MUST be a CCE version 5 number
http://cpe.mitre.org	No	CPE –the identifier value MUST be a CPE version 2.0 or 2.3 name
http://cve.mitre.org	No	CVE –the identifier value MUST be a CVE number
http://www.cert.org	No	CERT Coordination Center – the identifier value SHOULD be a CERT advisory identifier (e.g., “CA-2004-02”)
http://www.kb.cert.org	No	US-CERT vulnerability notes database – the identifier value SHOULD be a vulnerability note number (e.g., “709220”)
http://www.us-cert.gov/cas/techalerts	No	US-CERT technical cyber security alerts –the identifier value SHOULD be a technical cyber security alert ID (e.g., “TA05-189A”)

When the rule is related to RHEL, it should have a CCE. A CEE (e.g. cce@rhel7: CCE-80328-8) is used as a global identifier that maps the rule to the product over the lifetime of a rule. There should only be one CCE mapped to a rule as a global identifier. Any other usage of CCE is no longer considered a best practice. CCEs are also product dependent which means that a different CCE must be used for each different product and product version. For example if cce@rhel7: 80328-8 exists in a rule, that CCE cannot be used for another product or version (e.g. rhel9), and the CCE MUST be retired with the rule. Available CCEs that can be assigned to new rules are listed in the shared/references/cce-rhel-avail.txt file.

references: This is related to the compliance document line items that the rule applies to. These can be attributes such as stigid, srg, nist, etc., whose keys may be modified with a product (e.g., stigid@rhel7) to restrict what products a reference identifier applies to. Depending on the type of reference (e.g. catalog, ruleid, etc.) will depend on how many can be added to a single rule. In addition, certain references in a rule such as stigid or cis only apply to a certain product and product version; they cannot be used for multiple products and versions.

Key	Reference Type	Mapping to Rule	Example Format
cis@<product><product_version>	Center for Internet Security (catalog identifier)	0-to-many, 0-to-1 is preferred	5.2.5
cjis	Criminal Justice Information System (catalog identifier)	0-to-1	5.4.1.1
cui	Controlled Unclassified Information (catalog identifier)	0-to-many, 0-to-1 is preferred	3.1.7
disa	DISA Control Correlation Identifiers (catalog identifier)	0-to-many	CCI-000018,CCI-000172,CCI-001403
srg, vmmsrg, etc.	DISA Security Requirements Guide (catalog identifier)	0-to-many	SRG-OS-000003-GPOS-00004
stigid@<product><product_version>	DISA STIG identifier (rule identifier)	0-to-1	RHEL-07-030874
hipaa	Health Insurance Portability and Accountability Act of 1996 (HIPAA) (catalog identifier)	0-to-many	164.308(a)(1)(ii)(D),164.308(a)(3)(ii)(A)
nist	National Institute for Standards and Technology 800-53 (catalog identifier)	0-to-many	AC-2(4),AC-17(7),AU-1(b)
nist-csf	National Institute for Standards and Technology Cybersecurity Framework (catalog identifier)	0-to-many	DE.AE-3,DE.AE-5,DE.CM-1
ospp	National Information Assurance Partnership (selected control identifier)	0-to-many	FMT_MOF_EXT.1
pcidss	Payment Card Industry Data Security Standard	0-to-many, 0-to-1 is preferred	Req-8.7.c

See linux_os/guide/system/software/disk_partitioning/encrypt_partitions/rule.yml for an example of reference-type attributes as there are others that are not referenced above.

Some of existing rule definitions contain attributes that use macros. There are two implementations of macros:

Jinja macros, that are defined in shared/macros.jinja, and shared/macros-highlevel.jinja.
Legacy XSLT macros, which are defined in shared/transforms/*.xslt.

For example, the ocil attribute of service_ntpd_enabled uses the ocil_service_enabled jinja macro. Due to the need of supporting Ansible output, which also uses jinja, we had to modify control sequences, so macro operations require one more curly brace. For example, invocation of the partition macro looks like {{{ complete_ocil_entry_separate_partition(part="/tmp") }}} - there are three opening and closing curly braces instead of the two that are documented in the Jinja guide.

shared/macros.jinja contains specific low-level macros s.a. systemd_ocil_service_enabled, whereas shared/macros-highlevel.jinja contains general macros s.a. ocil_service_enabled, that decide which one of the specialized macros to call based on the actual product being used.

You can see references of high level macros here.

The macros that are likely to be used in descriptions begin by describe_, whereas macros likely to be used in OCIL entries begin with ocil_. Sometimes, a rule requires ocil and ocil_clause to be specified, and they depend on each other. Macros that begin with complete_ocil_entry_ were designed for exactly this purpose, as they make sure that OCIL and OCIL clauses are defined and consistent. Macros that begin with underscores are not meant to be used in descriptions.

To parametrize rules and remediations as well as Jinja macros, you can use product-specific variables defined in product.yml in product root directory. Moreover, you can define implied properties which are variables inferred from them. For example, you can define a condition that checks if the system uses yum or dnf as a package manager and based on that populate a variable containing correct path to the configuration file. The inferring logic is implemented in _get_implied_properties in ssg/yaml.py. Constants and mappings used in implied properties should be defined in ssg/constants.py.

Rules are unselected by default - even if the scanner reads rule definitions, they are effectively ignored during the scan or remediation. A rule may be selected by any number of profiles, so when the scanner is scanning using a profile the rule is included in, the rule is taken into account. For example, the rule identified by partition_for_tmp defined in shared/xccdf/system/software/disk_partitioning.xml is included in the RHEL7 C2S profile in rhel7/profiles/C2S.xml.

Checks are connected to rules by the oval element and the filename in which it is found. Remediations (i.e. fixes) are assigned to rules based on their basename. Therefore, the rule sshd_print_last_log has a bash fix associated as there is a bash script shared/fixes/bash/sshd_print_last_log.sh. As there is an Ansible playbook shared/fixes/ansible/sshd_print_last_log.yml, the rule has also an Ansible fix associated.

Rule Directories

The rule directory simplifies the structure of a rule and all of its associated content by placing it all under a common directory. The structure of a rule directory looks like the following example:

linux_os/guide/system/group/rule_id/rule.yml
linux_os/guide/system/group/rule_id/bash/ol7.sh
linux_os/guide/system/group/rule_id/bash/shared.sh
linux_os/guide/system/group/rule_id/oval/rhel7.xml
linux_os/guide/system/group/rule_id/oval/shared.xml

To be considered a rule directory, it must be a directory contained in a benchmark pointed to by some product. The directory must have a name that is the id of the rule, and must contain a file called rule.yml which is a YAML Rule description as described above. This directory can then contain the following subdirectories:

anaconda - for Anaconda remediation content, ending in .anaconda
ansible - for Ansible remediation content, ending in .yml
bash - for Bash remediation content, ending in .sh
oval - for OVAL check content, ending in .xml
puppet - for Puppet remediation content, ending in .pp
ignition - for Ignition remediation content, ending in .yml
kubernetes - for Kubernetes remediation content, ending in .yml
sce - for Script Check Engine content, with any file extension
blueprint - for OSBuild blueprint content, ending in .toml

In each of these subdirectories, a file named shared.ext will apply to all products and be included in all builds, but {{{ product }}}.ext will only get included in the build for {{{ product }}} (e.g., rhel7.xml above will only be included in the build of the rhel7 guide content and not in the ol7 content). Additionally, we support the use of unversioned products here (e.g., rhel applies to rhel7, rhel8, and rhel9). Note that .ext must be substituted for the correct extension for content of that type (e.g., .sh for bash content). Further, all of these directories are optional and will only be searched for content if present. Lastly, the product naming of content will not override the contents of platform or prodtype fields in the content itself (e.g., if rhel7 is not present in the rhel7.xml OVAL check platform specifier, it will be included in the build artifacts but later removed because it doesn’t match the platform). This means that any shared (or templated) checks won’t be searched if a product-specific file is present but has the wrong applicability; this includes shared checks being preferred above templated checks.

Currently the build system supports both rule files (discussed above) and rule directories. For example content in this format, please see rules in linux_os/guide.

To interact with build directories, the ssg.rules and ssg.rule_dir_stats modules have been created, as well as three utilities:

utils/rule_dir_json.py - to generate a JSON tree describing the current content of all guides
utils/rule_dir_stats.py - for analyzing the JSON tree and finding information about specific rules, products, or summary statistics
utils/rule_dir_diff.py - for diffing two JSON trees (e.g., before and after a major change), using the same interface as rule_dir_stats.py.

For more information about these utilities, please see their help text.

To interact with rule.yml files and the OVALs inside a rule directory, the following utilities are provided:

`utils/mod_prodtype.py`

This utility modifies the prodtype field of rules. It supports several commands:

mod_prodtype.py <rule_id> list - list the computed and actual prodtype of the rule specified by rule_id.
mod_prodtype.py <rule_id> add <product> [<product> ...] - add additional products to the prodtype of the rule specified by rule_id.
mod_prodtype.py <rule_id> remove <product> [<product> ...] - remove products to the prodtype of the rule specified by rule_id.
mod_prodtype.py <rule_id> replace <replacement> [<replacement> ...] - do the specified replacement transformations. A replacement transformation is of the form match~replace where match and replace are a comma separated list of products. If all of the products in match exist in the original prodtype of the rule, they are removed and the products in replace are added.

This utility requires an up to date JSON tree created by rule_dir_json.py.

`utils/mod_checks.py`

This utility modifies the <affected> element of an OVAL check. It supports several commands on a given rule:

mod_checks.py <rule_id> list - list all OVALs, their computed products, and their actual platforms.
mod_checks.py <rule_id> delete <product> - delete the OVAL for the the specified product.
mod_checks.py <rule_id> make_shared <product> - moves the product OVAL to the shared OVAL (e.g., rhel7.xml to shared.xml).
mod_checks.py <rule_id> diff <product> <product> - Performs a diff between two OVALs (product can be shared to diff against the shared OVAL).

In addition, the mod_checks.py utility supports modifying the shared OVAL with the following commands:

mod_checks.py <rule_id> add <platform> [<platform> ...] - adds the specified platforms to the shared OVAL for the rule specified by rule_id.
mod_checks.py <rule_id> remove <platform> [<platform> ...] - removes the specified platforms from the shared OVAL.
mod_checks.py <rule_id> replace <replacement> [<replacement ...] - do the specified replacement against the platforms in the shared OVAL. See the description of replace under mod_prodtype.py for more information about the format of a replacement.

This utility requires an up to date JSON tree created by rule_dir_json.py.

`utils/mod_fixes.py`

This utility modifies the <affected> element of a remediation. It supports several commands on a given rule and for the specified remediation language:

mod_fixes.py <rule_id> <lang> list - list all fixes, their computed products, and their actual platforms.
mod_fixes.py <rule_id> <lang> delete <product> - delete the fix for the specified product.
mod_fixes.py <rule_id> <lang> make_shared <product> - moves the product fix to the shared fix (e.g., rhel7.sh to shared.sh).
mod_fixes.py <rule_id> <lang> diff <product> <product> - Performs a diff between two fixes (product can be shared to diff against the shared fix).

In addition, the mod_fixes.py utility supports modifying the shared fixes with the following commands:

mod_fixes.py <rule_id> <lang> add <platform> [<platform> ...] - adds the specified platforms to the shared fix for the rule specified by rule_id.
mod_fixes.py <rule_id> <lang> remove <platform> [<platform> ...] - removes the specified platforms from the shared fix.
mod_fixes.py <rule_id> <lang> replace <replacement> [<replacement ...] - do the specified replacement against the platforms in the shared fix. See the description of replace under mod_prodtype.py for more information about the format of a replacement.

This utility requires an up to date JSON tree created by rule_dir_json.py.

`utils/add_platform_rule.py`

This utility can be used to bootstrap and test Kubernetes/OpenShift application checks. See the help output for more detailed usage examples of each of the supported subcommands:

utils/add_platform_rule.py create --rule=<rule_name> <options> - creates files for a new rule.
utils/add_platform_rule.py test --rule=<rule_name> <options> - tests a rule against local files using an oscap container.
utils/add_platform_rule.py cluster-test --rule=<rule_name> <options>
- tests a rule against a running OCP4 cluster using compliance-operator.

This utility requires the following:

KUBECONFIG env set to a kubeconfig file for a running OCP4 cluster.
oc and podman in PATH.

Tips:

The –yamlpath option requires a specialized format to specify the resource element to check. See https://github.com/OpenSCAP/yaml-filter/wiki/YAML-Path-Definition for documentation.
To use the local test subcommand, first create a yaml file under a directory structure under /tmp that mirrors the API path. For example, if the resource’s full path is /api/v1/foo, save the yaml to /tmp/api/v1/foo. Running test will then check the rule against the local file by launching an openscap-1.3.3 container using podman.

Checks

Checks are used to evaluate a Rule. There are two types of check content supported by ComplianceAsCode: OVAL and SCE. Note that OVAL is standardized by NIST and has better cross-scanner support than SCE does. However, because SCE can use any language on the target system (Bash, Python, …) it is much more flexible and general-purpose than OVAL. This project generally encourages OVAL unless it lacks support for certain features.

OVAL Check Content

They are written using a custom OVAL syntax and are transformed by the system during the building process into OVAL compliant checks.

The OVAL checks are stored as XML files and the build system can source them from:

the oval directory of a rule
the shared pool of checks shared/checks/oval
a template

In order to create a new check you must create a file in the appropriate directory. The id attribute of the check needs to match the id of its rule. The content of the file should follow the OVAL specification with these exceptions:

The root tag must be <def-group>
If the OVAL check has to be a certain OVAL version, you can add oval_version="oval_version_number" as an attribute to the root tag. Otherwise if oval_version does not exist in <def-group>, it is assumed that the OVAL file applies to any OVAL version.
Don’t use the tags <definitions> <tests> <objects> <states>, instead, put the tags <definition> <*_test> <*_object> <*_state> directly inside the <def-group> tag.
TODO Namespaces

This is an example of a check, written using the custom OVAL syntax, that checks if the group that owns the file /etc/cron.allow is the root:

<def-group oval_version="5.11">
  <definition class="compliance" id="file_groupowner_cron_allow" version="1">
    <metadata>
      <title>Verify group who owns 'cron.allow' file</title>
      <affected family="unix">
        <platform>Red Hat Enterprise Linux 7</platform>
      </affected>
      <description>The /etc/cron.allow file should be owned by the appropriate
      group.</description>
    </metadata>
    <criteria>
      <criterion test_ref="test_groupowner_etc_cron_allow" />
    </criteria>
  </definition>
  <unix:file_test check="all" check_existence="any_exist"
  comment="Testing group ownership /etc/cron.allow" id="test_groupowner_etc_cron_allow"
  version="1">
    <unix:object object_ref="object_groupowner_cron_allow_file" />
    <unix:state state_ref="state_groupowner_cron_allow_file" />
  </unix:file_test>
  <unix:file_state id="state_groupowner_cron_allow_file" version="1">
    <unix:group_id datatype="int">0</unix:group_id>
  </unix:file_state>
  <unix:file_object comment="/etc/cron.allow"
  id="object_groupowner_cron_allow_file" version="1">
    <unix:filepath>/etc/cron.allow</unix:filepath>
  </unix:file_object>

Before you start creating an OVAL check, please consult our list of JINJA macros specific for OVAL. It might save time for you as an author as well as for reviewers.

Limitations and pitfalls

This section aims to list known OVAL limitations and situations that OVAL can’t handle well or at all.

Checking that all objects exist based on a variable

A test with check_existence=”all_exist” attribute will not ensure that all objects defined based on a variable exist.

This happens because in the test context all_exist defaults to at_least_one_exists. Reference: OVAL ExistenceEnumeration

This means that an object defined based on a variable with multiple values will result in pass if just one of the expected objects exist. Example, lets say there is a variable containing multiple paths, and you’d like to check that all of them are present in the file system. The following snippet would work if all_exist didn’t default to at_least_one_exists.

  <unix:file_test id="" check="all" check_existence="all_exist" comment="This test fails to ensure that all paths from variable exist" version="1">
    <unix:object object_ref="collect_objects_based_on_variable" />
  </unix:file_test>

  <unix:file_object id="collect_objects_based_on_variable" version="1">
    <unix:path datatype="string" var_ref="variable_containing_a_list_of_paths" var_check="at least one"/>
  </unix:file_object>

A workaround is to add a second test to count the number of objects collected and compare the sum against the number of values in the variable. This works well, except when none of the files exist, which leads to a result of unknown. Counting the number of collected items of an object definition that doesn’t exist is unknown (not 0, for example).

Platform applicability

Whenever possible, please reuse the macros and form high-level simplifications.

SCE Check Content

SCE is a mechanism for running arbitrary scripts while delivering them via the same data stream as OVAL content. These checks (being written in a general-purpose programming language) can be more flexible than OVAL checks but at the downside of not being compliant with the relevant NIST standards (and thus losing interoperability with other scanners). This project prefers Bash SCE content.

Within a rule directory, SCE content is stored under the sce/ subfolder; it doesn’t have a single extension (taking the preferred extension of the language the check is written in, .sh for Bash content).

To build SCE content, specify the -DSSG_SCE_ENABLED=ON option to CMake; note that we default to not building SCE content.

We support the same comment-based mechanism for controlling script parameters as the test suite and rest of the content system. The following parameters are unique to SCE:

check-import: can be stdout or stderr and corresponds to the XCCDF’s <check-import /> element’s import-name attribute.
check-export: a comma-separated list of variable_name=xccdf_variable pairs to export via XCCDF <check-export /> elements. oscap will generate 3 env_variables before calling the SCE script:
```
XCCDF_VALUE_<variable_name>=<value>
XCCDF_TYPE_<variable_name>=<type> 
XCCDF_OPERATOR_<variable_name>=<operator>
```
The <value>, <type> and <operator> come from the corresponding xccdf variable.
complex-check: an XCCDF operator (AND or OR) to be passed as the operator attribute on the XCCDF element’s <complex-check /> element. Note that this gets provisioned into the <Rule /> element to handle whether both the OVAL and SCE checks must pass (AND) or whether only one is necessary (OR). If a rule only has one or the other check language, it is not necessary. Additionally, OCIL checks, if any is present in the rule.yml, are added as a top-level OR-operator <complex-check /> with the results of this <complex-check />.

For an example of SCE content, consider the check:

$ cat ./linux_os/guide/system/accounts/accounts-session/accounts_users_own_home_directories/sce/ubuntu2004.sh
#!/bin/bash
#
# Contributed by Canonical.
#
# Disable job control and run the last command of a pipeline in the current shell environment
# Require Bash 4.2 and later
#
# platform = multi_platform_ubuntu
# check-import = stdout

set +m
shopt -s lastpipe

result=$XCCDF_RESULT_PASS

cat /etc/passwd | egrep -v '^(root|halt|sync|shutdown)' | awk -F: '($7 != "/usr/sbin/nologin" && $7 != "/bin/false") { print $1 " " $6 }'| while read user dir; do
	if [ ! -d "$dir" ]; then
		echo "The home directory ($dir) of user $user does not exist."
		result=$XCCDF_RESULT_FAIL
		break
	else
		owner=$(stat -L -c "%U" "$dir")
		if [ "$owner" != "$user" ]; then
			echo "The home directory ($dir) of user $user is owned by $owner."
			result=$XCCDF_RESULT_FAIL
			break
		fi
	fi
done
exit $result

Since this rule lacks an OVAL, this rule does not need a complex-check attribute. Additionally, this rule doesn’t use any XCCDF variables and thus doesn’t need a check-export element.

Remediations

Remediations, also called fixes, are used to change the state of the machine, so that previously non-passing rules can pass. There can be multiple versions of the same remediation meant to be executed by different applications, more specifically Ansible, Bash, Anaconda, Puppet, Ignition and Kubernetes. By default all remediation languages are built and included in the DataStream.

But each product can specify its own set of remediation to include in the DataStream via a CMake Variable in the product’s CMakeLists.txt. See example below, from OCP4 product, ocp4/CMakeLists.txt:

set(PRODUCT_REMEDIATION_LANGUAGES "ignition;kubernetes")

They also have to be idempotent, meaning that they must be able to be executed multiple times without causing the fixes to accumulate. The Ansible’s language works in such a way that this behavior is built-in, however, for the other versions, the remediations must have it implemented explicitly. Remediations also carry metadata that should be present at the beginning of the files. This meta data will be converted in XCCDF tags during the building process. That is how it looks like and what it means:

# platform = multi_platform_all
# reboot = false
# strategy = restrict
# complexity = low
# disruption = low

Field	Description	Accepted values
platform	CPE name, CPE applicability language expression or even wildcards declaring which platforms the fix can be applied	Default CPE dictionary is packaged along with openscap. Custom CPE dictionaries can be used. Wildcards are multi_platform_[all, oval, fedora, debian, ubuntu, linux, rhel, openstack, opensuse, rhev, sle].
reboot	Whether or not a reboot is necessary after the fix	true, false
strategy	The method or approach for making the described fix. Only informative for now	unknown, configure, disable, enable, patch, policy, restrict, update
complexity	The estimated complexity or difficulty of applying the fix to the target. Only informative for now	unknown, low, medium, high
disruption	An estimate of the potential for disruption or operational degradation that the application of this fix will impose on the target. Only informative for now	unknown, low, medium, high

Ansible

Important

The minimum version of Ansible must be at the latest supported version. See https://access.redhat.com/support/policy/updates/ansible-engine for information on the supported Ansible versions.

Ansible remediations are either:

Stored as .yml files in directory ansible in the rule directory.
Generated from templates.
Generated using jinja2 macros.

They are meant to be executed by Ansible itself when requested by openscap, so they are written using Ansible’s own language with the following exceptions:

The remediation content must be only the tasks section of what would be a playbook.
- Tasks can include blocks for grouping related tasks.
- The when clause will get augmented in certain scenarios.
Notifications and handlers are not supported.
Tags are not necessary, because they are automatically generated during build of content.

Here is an example of an Ansible remediation that ensures the SELinux is enabled in grub:

# platform = multi_platform_rhel,multi_platform_fedora
# reboot = false
# strategy = restrict
# complexity = low
# disruption = low
- name: Ensure SELinux Not Disabled in /etc/default/grub
  replace:
    dest: /etc/default/grub
    regexp: selinux=0

The Ansible remediation will get included by our build system to the SCAP datastream in the fix element of respective rule.

The build system generates an Ansible Playbook from the remediation for all profiles. The generated Playbook is located in /build/<product>/playbooks/<profile_id>/<rule_id>.yml.

For each rule in the given product we also generate an Ansible Playbook regardless presence of the rule in any profile. The generated Playbook is located in /build/<product>/playbooks/all/<rule_id>.yml. The /build/<product>/playbooks/all/ directory represents the virtual (all) profile which consists of all rules in the product. Due to undefined XCCDF Value selectors in this pseudo-profile, these Playbooks use defaults of XCCDF Values when applicable.

We also build profile Playbook that contains tasks for all rules in the profile. The Playbook is generated in /build/ansible/<product>-playbook-<profile_id>.yml.

Jinja macros for Ansible content are located in /shared/macros-ansible.jinja. You can see their reference [here]( jinja_macros/ansible:ansible).

Whenever possible, please reuse the macros and form high-level simplifications. This ensures consistent, high quality remediations that we can edit in one place and reuse in many places.

Bash

Bash remediations are stored as shell script files in the bash directory in a rule’s directory. You can make use of any available command, but beware of too specific or complex solutions, as it may lead to a narrow range of supported platforms.

Following, you can see an example of a bash remediation that sets the maximum number of days a password may be used:

# platform = Red Hat Enterprise Linux 7
{{{ bash_instantiate_variables("var_accounts_maximum_age_login_defs) }}}

grep -q ^PASS_MAX_DAYS /etc/login.defs && \
    sed -i "s/PASS_MAX_DAYS.*/PASS_MAX_DAYS     $var_accounts_maximum_age_login_defs/g" /etc/login.defs
if [ $? -ne 0 ]; then
    echo "PASS_MAX_DAYS      $var_accounts_maximum_age_login_defs" >> /etc/login.defs
fi

When writing new bash remediations content, please follow the following guidelines:

Use four spaces for indentation rather than tabs.
You can use macros from shared/macros-bash.jinja in the remediation content. If the macro is used from a nested block, use the indent jinja2 filter assuming the 4-space indentation. Typically, you want to call the macro with the intended indentation, and as indent doesn’t indent the first line by default, you just pass the number of spaces as the only argument. See the remediation for rule ensure_fedora_gpgkey_installed for reference.
Prefer to use sed rather than awk.
Try to keep expressions simple, avoid double negations. Use compound lists with moderation and only if you understand them.
Test your script in the “strict mode” with set -e -o pipefail specified at the top of it. Make sure that the script doesn’t end prematurely in the strict mode.
Beware of constructs such as [ $x = 1 ] && echo "$x is one" as they violate the previous point. [ $x != 1 ] || echo "$x is one" is OK.
Use the die macro defined in shared/macros-bash.jinja to handle exceptions and terminate the remediation, such as {{{ die("An error was encountered during the remediation of rule.") }}}.
Run shellcheck over your remediation script. Make sure that you fix all warnings that are applicable. If you are not sure, mention those warnings in the pull request description.
Use POSIX syntax in regular expressions, so prefer grep '^[[:space:]]*something' over grep '^\s*something'.

Jinja macros that generate Bash remediations can be found in shared/macros-bash.jinja. You can see their reference here.

Kubernetes

Jinja macros for Kubernetes content are located in /shared/macros-kubernetes.jinja. You can see their reference here

Templating

Writing OVAL checks, Bash, or any other content can be tedious work. For certain types of rules we provide templates. If there is a template that can be used for the new rule you only need to specify the template name and its parameters in rule.yml and the content will be generated during the build.

The templating system currently supports generating OVAL checks and Ansible, Bash, Anaconda, Puppet, Ignition and Kubernetes remediations. All templates can be found in shared/templates directory. The files are named template_<TYPE>_<NAME>, where <TYPE> should be OVAL, ANSIBLE, BASH, ANACONDA, PUPPET, IGNITION and KUBERNETES and <NAME> is the template name.

Using Templates

To use a template in rule.yml add template: key there and fill it accordingly. The general form is the following:

template:
    name: template_name
    vars:
        param_name: value # these parameters are individual for each template
        param_name@rhel7: value1
        param_name@rhel8: value2
    backends: # optional
        ansible: "off"
        bash: "on" # on is implicit value

The vars: key contains template parameters and their values which will be substituted into the template. Each template has specific parameters. To use different values of parameters based on product, append @ followed by product ID to the parameter name.

The backends: key is optional. By default, all languages supported by a given template will be generated. with given name exist will be generated. This key can be used to explicitly opt out from generating a certain type of content for the rule.

For example, to generate templated content except Bash remediation for rule “Package GCC is Installed” using package_installed template, add the following to rule.yml:

template:
    name: package_installed
    vars:
        pkgname: gcc
    backends:
        bash: "off"

Important

The build system does not support implicit conversion of bool strings when Python 2 is used, so bash: True argument in the example above would cause a build error. One should always use quoted strings as arguments until Python 2 is completely removed from the list of supported interpreters.

you can see reference of all available templates here.

Applicability of content

All profiles and rules included in a products’ DataStream are applicable by default. For example, all profiles and rules included in a rhel8 DS will apply and evaluate in a RHEL 8 host.

But a content’s applicability can be fine tuned to a specific environment in the product. The SCAP standard specifies two mechanisms to define applicability: - CPE: Allows a specific hardware or platform to be identified. - Applicability Language: Allows the construction of logical expressions involving CPEs.

At the moment, only the CPE mechanism is supported.

Applicability by CPE

The CPEs defined by the project are declared in shared/applicability/cpes.yml.

Syntax is as follows (using examples of existing CPEs):

cpes:
  - machine:                                  ## The id of the CPE
      name: "cpe:/a:machine"                  ## The CPE Name as defined by the CPE standard
      title: "Bare-metal or Virtual Machine"  ## Human readable title for the CPE
      check_id: installed_env_is_a_machine    ## ID of OVAL implementing the applicability check
  - gdm:
      name: "cpe:/a:gdm"
      title: "Package gdm is installed"
      check_id: installed_env_has_gdm_package

The first entry above defines a CPE whose id is machine, this CPE is used for rules not applicable to containers. A rule or profile with platform: machine will be evaluated only if the targeted scan environment is either bare-metal or virtual machine.

The second entry defines a CPE for GDM. By setting the platform to gdm, the rule will have its applicability restricted to only environments which have gdm package installed.

The OVAL checks for the CPE need to be of inventory class, and must be under shared/checks/oval/.

Setting a product’s default CPE

The product’s default applicability is set in its product.yml file, under the cpes key. For example:

cpes:
  - example:
      name: "cpe:/o:example"
      title: "Example"
      check_id: installed_OS_is_part_of_Unix_family

Multiple CPEs can be set as default platforms for a product.

Setting the product’s CPE source directory

The key cpes_root in product.yml file specifies the directory to source the CPEs from. By default, all products source their CPEs from shared/applicability/. Any file with extension .yml will be sourced for CPE definitions.

Note: Only CPEs that are referenced by a rule or profile will be included in the product’s CPE Dictionary. If no content requires the CPE, it is deemed unnecessary and won’t be included in the dictionary.

Tests (ctest)

ComplianceAsCode uses ctest to orchestrate testing upstream. To run the test suite go to the build folder and execute ctest:

cd build/
ctest -j 4

Check out the various ctest options to perform specific testing, you can rerun just one test or skip all tests that match a regex. (See -R, -E and other options in the ctest man page)

Tests are added using the add_test cmake call. Each test should finish with a 0 exit-code in case everything went well and a non-zero if something failed. Output (both stdout and stderr) are collected by ctest and stored in logs or displayed. Make sure you never hard-code a path to any tool when doing testing (or anything really) in the cmake code. Always use configuration to find all the paths and then use the respective variable.

See some of the existing testing code in cmake/SSGCommon.cmake.

Tests (OCP4)

Unit tests

Running unit tests is very useful way to verify the content being written without requiring a full-blown OpenShift cluster.

The unit testing approach for content requires a container where the scanning will be executed. This container will be a small replication of the environment where the content is intended to run.

To build the container, from the project’s root directory, execute the following command:

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

Unit tests reside en each rule’s tests/ directory. Tests are simply bash scripts that set up the required environment and set expectations on what the scan is suposed to output.

The naming convention is as follows:

<test name>.<expected result>.sh

e.g. when testing if encryption is disabled in a specific cloud provider, and that this should fail a compliance scan, you’d have a file named as follows:

encryption_disabled.fail.sh

Note that this file has to be executable. So don’t forget to change the mode appropriately.

Test files shall always begin with the following lines:

#!/bin/bash
# remediation = none

Bash is used to evaluate the scripts. And we don’t expect remediations executed by OpenSCAP itself (this is done elsewhere), so we set remediation = none.

OpenShift content (as well as other Kubernetes content) is usually meant to run through the Compliance Operator. The aforementioned component will read the API paths specified as warnings and persist them on a temporary volume. The Compliance Operator is also the one that handles remediating rules, this is why we turned remediations off in our scripts as mentioned earlier.

For unit tests, we’ll do something similar when setting up the testing container in our test scripts: We’ll create a directory for the API resource object and persist it there.

Note that the content uses the path /kubernetes-api-resources by default as a base for any API path for Kubernetes. So, be mindful of always using that.

An example looks as follows:

kube_apipath="/kubernetes-api-resources"
machinev1beta1="/apis/machine.openshift.io/v1beta1"
machineset_apipath="$machinev1beta1/machinesets?limit=500"

# Create kubernetes resource directory path
mkdir -p "$kube_apipath/$machinev1beta1"

# Persist kubernetes resource
cat <<EOF > "$kube_apipath/$machineset_apipath"
{
    "apiVersion": "v1",
    ...
EOF

When the scanner tests the content for this rule, it’ll read the Kubernetes resource (both YAML or JSON are fine for this) and evaluate it accordingly.

Note that some rules require a specific CPE to be set. e.g. there are some rules that are meant to run on a specific cloud platform.

You’ll see them from the platforms key of the rule:

platforms:
- ocp4-on-azure

For rules like these, you’ll also need to persist the appropriate CPE information (normally another Kubernetes object).

The information needed varies on each CPE. These are defined in /shared/checks/oval.

For the ocp4-on-azure example, the CPE requires that the infrastructures/cluster resource specifies the platform as Azure. That would look as follows:

kube_apipath="/kubernetes-api-resources"

# Create infra file for CPE to pass
mkdir -p "$kube_apipath/apis/config.openshift.io/v1/infrastructures/"
cat <<EOF > "$kube_apipath/apis/config.openshift.io/v1/infrastructures/cluster"
{
    "apiVersion": "config.openshift.io/v1",
    "kind": "Infrastructure",
    "metadata": {
        "name": "cluster"
    },
    "spec": {
        "platformSpec": {
            "type": "Azure"
        }
    },
    "status": {
        "platform": "Azure",
        "platformStatus": {
            "azure": {
                "cloudName": "AzurePublicCloud"
            },
            "type": "Azure"
        }
    }
}
EOF

For rules that use jq filters, they are using a unique file path instead of the one you see above, therefore we also need to install the jq package, run the jq test, and save the filtered jq result into that unique file path.

The format for the file path for a jq filter rule is:

"$kube_apipath$rule_apipath#$(echo -n "$rule_apipath$jq_filter" | sha256sum | awk '{print $1}')"

Noted it is important to know that the $rule_apipath does not have / at the end, or it will cause a different hash with / at the end, and jq result will be saved into a location that is different from where the rule is checking, hence will fail the test.

We have a test file for rule ‘routes_rate_limit’ as an example that uses jq filter:

#!/bin/bash
# remediation = none

yum install -y jq

kube_apipath="/kubernetes-api-resources"
mkdir -p "$kube_apipath/apis/route.openshift.io/v1"
routes_apipath="/apis/route.openshift.io/v1/routes?limit=500"

cat <<EOF > "$kube_apipath$routes_apipath"
{
  ...
}
EOF

jq_filter='[.items[] | select(.metadata.namespace | startswith("kube-") or startswith("openshift-") | not) | select(.metadata.annotations["haproxy.router.openshift.io/rate-limit-connections"] == "true" | not) | .metadata.name]'

filteredpath="$kube_apipath$routes_apipath#$(echo -n "$routes_apipath$jq_filter" | sha256sum | awk '{print $1}')"

jq "$jq_filter" "$kube_apipath$routes_apipath" > "$filteredpath"

Before running the unit tests, the content needs to be built to ensure that the latest changes are taken into account:

./build_product ocp4

Remember this needs to be done every time before running the unit tests.

Now, with all of this set, we’ll run the unit test!

tests/test_rule_in_container.sh \
    --dontclean --logdir logs_bash \
    --remediate-using bash \
    --name ssg_test_suite \
    --datastream build/ssg-ocp4-ds.xml \
    <rule name>

This command will run the unit tests on dedicated containers. The logs will be located in the logs_bash directory.

End-to-end tests

The ComplianceAsCode/content repo runs some end-to-end tests for the ocp4 content. These tests run over the OpenShift infrastructure, spawn an ephemeral cluster and run tests targetted at a specific profile.

The current workflow is as follows:

Install needed prerequisites (e.g. the compliance-operator and other resources it might need)
Run a scan using the specific profile (for the specific product)
Run manual remediations
Run automated remediations
Wait for remediations to converge
Run second scan

The test will pass if:

There are no errors in the scan runs
We have less rule failures after the remediations have been applied
The cluster status is not inconsistent

Rules may have extra verifications on them. For instance, one is able to verify if:

The rule’s expected result is gotten on a clean run.
The rule’s result changes after a remediation has been applied.

If an automated remediation is not possible, one is also able to created a “manual” remediation that will be run as a bash script. The end-to-end tests have a 15 minute timeout for the manual remediation scripts to be executed.

Writing e2e tests for specific rules

In order to test that a rule is yielding expected results in the e2e tests, one must create a file called e2e.yml in a tests/ocp4/ directory which will exist in the rule’s directory itself.

The format may look as follows:

---
default_result: [PASS|FAIL|SKIP]
result_after_remediation: [PASS|FAIL|SKIP]

Where:

default_result will look at the result when the first scan is run.
result_after_remediation will look at the result when the second scan is run. The second scan takes place after remediations are applied.

Note that this format applies if the result of a rule will be the same accross roles in the cluster.

It is also possible to differentiate results between roles. For such a thing, the format would look as follows:

Let’s look at an example:

---
default_result:
  worker: [PASS|FAIL|SKIP]
  master: [PASS|FAIL|SKIP]
result_after_remediation:
  worker: [PASS|FAIL|SKIP]
  master: [PASS|FAIL|SKIP]

Where “worker” and “master” are node roles.

For the controller_use_service_account rule, which exists in the applications/openshift/controller/ directory, the directory tree will contain the rule definition and the test file:

.
├── rule.yml
└── tests
    └── ocp4
        └── e2e.yml

In this case, we just want to verify that the default value returns a passing result. So e2e.yml has the following content:

---
default_result: PASS

Let’s look at another example:

For the api_server_encryption_provider_config we want to apply a remediation which cannot be applied via the compliance-operator. So we’ll need a manual remediation for this.

The directory structure looks as follows:

.
├── rule.yml
└── tests
    └── ocp4
        ├── e2e-remediation.sh
        └── e2e.yml

Where our test contains information for both the first default result and the expected result after the remediation has been applied:

---
default_result: FAIL
result_after_remediation: PASS

The remediation expects the name of the remediation script to be e2e-remediation.sh. The script should:

Apply the remediation.
Verify that the status has converged.

In the aforementioned case, the remediation script is as follows:

#!/bin/bash
oc patch apiservers cluster -p '{"spec":{"encryption":{"type":"aescbc"}}}' --type=merge
while true; do
    status=$(oc get openshiftapiserver -o=jsonpath='{range .items[0].status.conditions[?(@.type=="Encrypted")]}{.reason}')
    echo "Current Encryption Status:"
    oc get openshiftapiserver -o=jsonpath='{range .items[0].status.conditions[?(@.type=="Encrypted")]}{.reason}{"\n"}{.message}{"\n"}'
    if [ "$status" == "EncryptionCompleted" ]; then
        exit 0
    fi
    sleep 5
done

Here, we apply the remediation (through the patch command) and probe the cluster for status. Once the cluster converges, we exit the script with 0, which is a successful status.

The e2e test run will time out at 15 minuntes if a script doesn’t converge.

Note that the scripts will be run in parallel, but the test run will wait for all of them to be done.

Running the e2e tests on a local cluster

Note that it’s possible to run the e2e tests on a cluster of your choice.

To do so, ensure that you have a KUBECONFIG with appropriate credentials that points to the cluster where you’ll run the tests.

From the root of the ComplianceAsCode/content repository, run:

$ make -f tests/ocp4e2e/Makefile e2e PROFILE=<profile> PRODUCT=<product>

Where profile is the name of the profile you want to test, and product is a product relevant to OCP4, such as ocp4 or rhcos4.

For instance, to run the tests for the cis benchmark for ocp4 do:

$ make -f tests/ocp4e2e/Makefile e2e PROFILE=cis PRODUCT=ocp4

For more information on the available options, do:

$ make -f tests/ocp4e2e/Makefile help

It is important to note that the tests will do changes to your cluster and there currently isn’t an option to clean them up. So take that into account before running these tests.

Contribution to infrastructure code

The ComplianceAsCode build and templating system is mostly written in Python.

Python

The common pattern is to dynamically add ssg to the import path. There are many useful modules with several functions and predefined constants. See scripts at ./build-scripts as an example of this practice.
Follow the PEP8 standard.
Try to keep most of your lines length under 80 characters. Although the 99 character limit is within PEP8 requirements, there is no reason for most lines to be that long.

Contributing with XCCDFs, OVALs and remediations

Contributing

Rules

Rule Directories

utils/mod_prodtype.py

utils/mod_checks.py

utils/mod_fixes.py

utils/add_platform_rule.py

Checks

OVAL Check Content

Limitations and pitfalls

Checking that all objects exist based on a variable

Platform applicability

SCE Check Content

Remediations

Ansible

Bash

Kubernetes

Templating

Using Templates

Applicability of content

Applicability by CPE

Setting a product’s default CPE

Setting the product’s CPE source directory

Tests (ctest)

Tests (OCP4)

Unit tests

End-to-end tests

Writing e2e tests for specific rules

Running the e2e tests on a local cluster

Contribution to infrastructure code

Python

`utils/mod_prodtype.py`

`utils/mod_checks.py`

`utils/mod_fixes.py`

`utils/add_platform_rule.py`