I have been working with the audit mode feature introduced in Chef version 12.1.0 – previously announced was the audit-cis cookbook. Audit mode allows users to write custom rules (controls) in Chef recipes using new DSL helpers. In his ChefConf 2015 talk, “Compliance At Velocity,” James Casey goes into more of the background and reasoning for this. For now, I wanted to share a few tips with users who may be trying out this feature, too.
First, we need to update ChefDK to version 0.5.0, as that includes a version of test kitchen that allows us to configure audit mode for chef-client.
curl -L https://chef.io/chef/install.sh | sudo bash -s -- -P chefdk
Next, create a new cookbook for the audit mode tests.
chef generate cookbook audit-test cd audit-test
Then, modify the audit-test cookbook’s
driver: name: vagrant provisioner: name: chef\_zero client\_rb: audit_mode: :audit\_only platforms: - name: ubuntu-12.04 - name: centos-6.5 suites: - name: default run\_list: - recipe[audit-test::default] attributes:
This is the generated
client\_rb added to the provisioner config to enable audit mode. Note that we must use the Ruby symbol syntax,
:audit\_only. The valid values for
:disabled. This will be translated to an actual Ruby symbol in the generated config file (
Next, let’s write a control rule to test. Since we’re using the default
.kitchen.yml, which includes Ubuntu 12.04 and uses SSH to connect, we can assume that SSH is running, so port 22 is listening. The following control asserts this is true.
control\_group 'Blog Post Examples' do control 'SSH' do it 'should be listening on port 22' do expect(port(22)).to be\_listening end end end
kitchen converge ubuntu to run Chef, but not tear down the VM afterward – we’ll use it again for another example. Here’s the audit phase output from the Chef run:
% kitchen converge ubuntu Synchronizing Cookbooks: - audit-test Compiling Cookbooks... Starting audit phase Blog Post Examples SSH should be listening on port 22 Finished in 0.10453 seconds (files took 0.37536 seconds to load) 1 example, 0 failures Auditing complete
Cool! So we have asserted that the node complies with this control by default. But what does a failing control look like? Let’s write one. Since we’re working with SSH already, let’s use the SSHd configuration. By default in the Vagrant base box we’re using, root login is permitted, so this value is present:
However, our security policy mandates that we set this to
no, and we want to audit that.
control\_group 'Blog Post Examples' do control 'SSH' do it 'should be listening on port 22' do expect(port(22)).to be\_listening end it 'disables root logins over ssh' do expect(file('/etc/ssh/sshd\_config').content).to contain('PermitRootLogin no') end end end
kitchen converge ubuntu and we see the validation fails.
Starting audit phase Blog Post Examples SSH should be listening on port 22 disables root logins over ssh (FAILED - 1) Failures: 1) Blog Post Examples SSH disables root logins over ssh Failure/Error: expect(file('/etc/ssh/sshd\_config').content).to contain('PermitRootLogin no') expected File "/etc/ssh/sshd\_config" to contain "PermitRootLogin no" # /tmp/kitchen/cache/cookbooks/audit-test/recipes/default.rb:8:in `block (3 levels) in from\_file' Finished in 0.13067 seconds (files took 0.32089 seconds to load) 2 examples, 1 failure Failed examples: rspec # Blog Post Examples SSH disables root logins over ssh [2015-04-04T03:29:41+00:00] ERROR: Audit phase failed with error message: Audit phase found failures - 1/2 controls failed Audit phase exception: Audit phase found failures - 1/2 controls failed
When we have a failure, we’ll have contextual information about the failure, including the line number in the recipe where it was found, and a stack trace (cut from the output here), in case more information is required for debugging. To fix the test, we can simply edit the config file to have the desired setting, or we can manage the file with Chef to set the value accordingly. Either way, after updating the file, the validation will pass, and all will be well.
We can put as many
control blocks with the
it validation rules as required to audit our policy. If we have many validations, it can be difficult to follow with all the output if there are failures. Chef’s audit mode is based on Serverspec, which is based on RSpec. We can use the
filter\_tag configuration feature of RSpec to only run the
control blocks or
it statements that we’re interested in debugging. To do this, we need an
RSpec.configuration block within the
control\_group – due to the way that audit mode is implemented, we can’t do it outside of
For example, we could debug our root login configuration:
control\_group 'Blog Post Examples' do ::RSpec.configure do |c| c.filter\_run focus: true end control 'SSH' do it 'should be listening on port 22' do expect(port(22)).to be\_listening end it 'disables root logins over ssh', focus: true do expect(file('/etc/ssh/sshd_config').content).to contain('PermitRootLogin no') end end end
The key here is to pass the argument
focus: true (or if you like hash rockets,
:focus => true) on the
it block. This could also be used on a
control 'SSH', focus: true do it 'does stuff...' end
Then, when running
kitchen converge ubuntu, we see only that validation:
Starting audit phase Blog Post Examples SSH disables root logins over ssh (FAILED - 1) Failures: 1) Blog Post Examples SSH disables root logins over ssh Failure/Error: expect(file('/etc/ssh/sshd\_config')content).to contain('PermitRootLogin no')
This example is simple enough that using
focus isn't necessary, but if we were implementing audit mode checks for our entire security policy, that could be dozens or even hundreds of controls.
While audit mode is still in development as a feature, we think that it is highly valuable to organizations that need to ensure their systems are compliant with varying types of audit requirements. Also, because audit mode can be run on its own without converging resources in Chef recipes, it's great for assessing existing system's state before starting to implement Chef.