Scroll to top
© 2020, DigitalOnUs

Tips and tricks I wish I knew when I started developing HashiCorp Sentinel policies for Terraform


DigitalOnUs - October 30, 2020 - 0 comments

Hello readers, this is the guide I wish that existed when I started writing Sentinel policies for Terraform; I hope it saves you some time. 

Overview 

In this article I will explain how to work with Sentinel, starting with basics and then I will provide tips on do’s and don’ts. 

For starters, the purpose of Sentinel is writing policies as code which prevents the user from creating things that aren’t meant to be according to your work organization. 

I must clarify that this article is based on my current understanding and personal experience with Sentinel V2 and V3, I still need to dive deeper into the Sentinel V3 framework as I still need to improve my understanding of it. 

Note: For placeholders, I usually use the diamond notation <template> which means you have to replace that whole string. 

Assumptions 

It is assumed you already read the basic documentation for Sentinel found here and you don’t know what to do with it. 

Computer Setup 

Step 1: Download HashiCorp Sentinel binary in your computer and make an environmental variable on Windows or put it under /usr/local/bin on linux. 

Note: I suggest using Linux since its easier to navigate with the command line. 

I also suggest to download Visual Studio Code or use any other IDE.  

The following capture is an example of workspace structure I created for my development, follow the steps below, and come back to this capture for any clarifications: 

Step 2: Create a folder to store your Sentinel policies. 

Step 3: Create a folder per policy for development purposes. 

Step 4: In your policy specific folder from step 3 create a file named <policy_name>.sentinel 

Step 5: Create the following path that will serve your specific policy

Your folder structure should look like this: 

/POLICIES_FOLDER/policy_specific_folder/test/<policy_name_folder> 

Step 6: Within the <policy_name_folder> create two files: pass.json and fail.json 

These files should look similar to this. 

pass.json

{ 
  "modules": { 
    "tfplan-functions": { 
      "path": "../../../tfplan-functions.sentinel" 
    } 
  }, 
  "mock": { 
    "tfplan/v2": "../../mock-tfplan-pass-v2.sentinel" 
  }, 
  "test": { 
    "main": true 
  } 
} 

fail.json

{ 
  "modules": { 
    "tfplan-functions": { 
      "path": "../../../tfplan-functions.sentinel" 
    } 
  }, 
  "mock": { 
    "tfplan/v2": "../../mock-tfplan-fail-v2.sentinel" 
  }, 
   "test": { 
    "main": false 
  } 
} 

The explanation for the content of these files is the following. It consists of three parts: 

  • modules: (available since Sentinel v0.15.0) libraries can be imported and they are required for testing when using the third generation of Sentinel policies. ( I will come back to this later) 
  • mock: which contain a test scenario (pass/fail) which Sentinel will use to test against. 
  • test: the expected result of the main rule from your Sentinel rule which can be true or false whether its a pass/fail scenario. 

Creating Mocks 

Step 1: Create an account at app.terraform.io, an organization, and a workspace. 

Step 2: Back in your computer, create a file named backend.tf in the /POLICIES_FOLDER from the previous section. 

Its content should look like this: 

terraform { 
  backend "remote" { 
    hostname = "app.terraform.io" 
    organization = "<some-organization-name>" 
workspaces { 
      name = "<some-workspace>" 
    } 
  } 
} 

Replace the placeholders for organization and name with your appropriate values. Replace hostname if you’re using PTFE. The hostname line can be removed if using

app.terraform.io 

Step 3: Assuming you’re using the Terraform website, use the following command to associate your Terraform organization and workspace with your local environment: 

terraform login 

For PTFE use the following command, but verify the hostname in back-end configuration file: 

AWSterraform login <hostname> 

Step 4 and tip: Develop the Terraform code under /POLICIES_FOLDER/policy_specific_folder and after you are done copy the file (let’s assume its named main.tf) to your /POLICIES_FOLDER.(Repeat this process when developing other policies.) 

Note: The Terraform code doesn’t have to be perfect, it can contain dummy values as place holders, just make sure to use the appropriate values for the items or sections you want to validate. Another tip, copy HashiCorp examples from the registry, notice some examples may have errors, but the CLI should help you identify. S3 bucket example from AWS here. 

Step 5: In the Variables tab of your workspace at app.terraform.iofill out the Terraform Variables. Here is my example for aws. 

Note: AWS_SESSION_ TOKEN is not always needed, depends on your AWS organization settings. 

Step 6: Now that you have some Terraform code written, its time to plan or (pretend to) apply it. 

terraform apply 

Wait until the CLI requests validation and answer with anything different than yes assuming you’re using dummy values. 

Then mocks should be available for download like in the image below by hitting the Download Sentinel mocks button 

Place the extracted files from the downloaded zip in the following path: 

/POLICIES_FOLDER/policy_specific_folder/ 

Then create a pass scenario assuming your code is correct as follows: 

cp mock-tfplan-v2.sentinel mock-tfplan-pass-v2.sentinel 

Do the same thing for the fail scenario: 

cp mock-tfplan-v2.sentinel mock-tfplan-fail-v2.sentinel 

Modify the fail scenario attributes as required to make it trigger an unexpected behavior with your resource, like modifying an attribute with a wrong value or deleting an attribute that you’re expecting. 

The creation of the pass and fail scenarios are based on HashiCorp’s documentation taken from here

You could modify the acl, assuming you are expecting “public-read” you could make it fail if you find a value different that is not “public-read”. Below its an excerpt of a mock, that would fail; assuming you’re validating the acl attribute expects “public-read” 

resource_changes = { 
    "aws_s3_bucket.demo": { 
        "address":  "aws_s3_bucket.demo", 
        "change": { 
            "actions": [ 
                "create", 
            ], 
            "after": { 
               "acl":  "not-public-read", 

Testing your mocks 

Position yourself in the following path: 

/POLICIES_FOLDER/policy_specific_folder/ 

There are two ways to test your mocks with the Sentinel CLI; one is with the apply option and the other one is with the test option. I will focus on the test option since with Sentinel V3 it’s easier just to execute that one. To print any warning messages use the option -verbose besides test, which looks like the following example: 

sentinel test -verbose <policy_name.sentinel> 

Here is an output example of a policy I did to validate databases with publicly_accessible property were not public: 

Going back to the pass and fail JSON files, in the section for modules you have to import the framework functions (Sentinel V3)from here. What I did is that I downloaded the file for tfplan-functions.sentinel and I added the reference within the modules section. Same story if I require the other function files like tfconfig-functions, tfrun-functions, or tfstate-functions. I’d also like to mention there are provider-specific functions like for aws (I expect other providers like azure to get functions as well). 

"modules": { 
    "tfplan-functions": { 
      "path": "../../../tfplan-functions.sentinel" 
    } 

Differences between Sentinel V2 and V3 

I noticed examples for Sentinel V2 found in HashiCorp’s documentation and Katacoda (the latter were retired) are rule-based. 

Rules must be avoided at all cost. 

The problem with rules is that you cannot store variables and it makes Sentinel difficult to work with as a regular programming language, functions are always preferred because they allow you to do things easier and you can store information in variables for later use. 

Functions are always preferred. 

The advantage of Sentinel V3 is that, since you can import modules (functions), you can speed up development. 

The only rule that has to exist is the main rule. 

# Main rule 
main = rule { 
     (message_counter_validator) else false 

Deciding whether a policy passes or fails 

This should be based on counting the error outputs. What I suggest is creating a list that stores the error messages and at the end of your policy, you count the number of error messages. If the number of error messages is different than zero, the rule should be considered as failed, that way you only prevent the creation of what is not allowed. 

I must clarify that if you’re using the Sentinel V3 common functions, there is a built-in mechanism that simplifies the behavior mentioned above; I don’t completely understand all the functions yet, so I still have to demystify some things. 

Importing functions from Sentinel V3 

# Import common-functions/tfplan-functions/tfplan-functions.sentinel 

# with alias “plan” 

import “tfplan-functions” as plan  

Tips and Tricks 

Finding resources 

Import the module tfplan-functions: 
# Import common-functions/tfplan-functions/tfplan-functions.sentinel 
# with alias "plan" 

import “tfplan-functions” as plan 

usage: 
plan.find_resources(resource_name) 
Assume any line you find below in the examples comes from this loop: 
for resource_collection as r, rc { 
// Some code 

Tags 

  • Single-word 

rc.change.after.tags.SingleWord 

  • Space-separated words 

rc.change.after.tags[“Space Separated Word”] 

Casting to string 

You need to import any of the *-functions.sentinel modules. 

For example, tfplan-functions.sentinel 

# Import common-functions/tfplan-functions/tfplan-functions.sentinel

# with alias “plan”

import “tfplan-functions” as plan 

Usage: 

plan.to_string(item) 

Validating and Storing Values gracefully 

To avoid a runtime error when storing items do the following: 

  • Collections: use curly braces 

var_name = plan.find_resources(“aws_some_resource”) else {} 

  • Attributes within a resource: when an attribute like tags doesn’t exist 
var_name = rc.change.after.tags else "default-error" 
then use an if clause to validate if tags is different than “default-error”
if(var_name != "default-error"){ 
// some code here 
  • Variables: regular variables that hold information. As mentioned before they do NOT work in rulesuse them in functions or elsewhere
function_name = func(resource_collection){ 
// store stuff anywhere in here 
// consider scopes when thinking of variables. 
 variable_name = definition...
} 
variable_name2 = definition... 

For loops 

The break and continue statements do exist in Sentinel. 

  • Break example: 
for resource_collection as r, rc { 
// Some code 
  if(/*some condition*/){ 
    // some code 
       break 
  } 
  • Continue example: 
for resource_collection as r, rc { 
// Some code 
  if(/*some condition*/){ 
    // some code 
       continue 
  } 

Selecting items within the Terraform structure 

  • Attributes: 

For example tags: 

tags =
a = "b" 
"a b" = "c" 

Select them is with dot notation because it has an equals sign (‘=’) before the block: 

rc.change.after.tags.a 

  • Block of code: 
example1 { 
 example2 { 
  attribute_name = "attribute_value" 
          } 

Selection when attributes don’t have ‘=’ signs: 

example1[0].example2[0].attribute_name 

Finding Blocks within resources 

Import the module tfplan-functions: 

# Import common-functions/tfplan-functions/tfplan-functions.sentinel 

# with alias “plan” 

import “tfplan-functions” as plan 

For example, when your Terraform code has a resource with a code of block that is repeated multiple times like ip_set_descriptors within in aws_waf_ipset example found here

You can find the multiple repetitions from ip_set_descriptors by using the following line, where the first parameter rc comes from our loop item, and the second one is the repeated code item: 

plan.find_blocks(rc, "ip_set_descriptors") 

Converting a JSON value to an actual JSON Structure 

Let’s use the following example of an S3 bucket policy; as you can see there is a policy attribute that contains a JSON- like structure in a policy. 

resource "aws_s3_bucket_policy" "b" { 
  bucket = aws_s3_bucket.b.id 
  policy = <<POLICY 
{ 
  "Version": "2012-10-17", 
  "Id": "MYBUCKETPOLICY", 
  "Statement": [ 
    { 
      "Sid": "IPAllow", 
      "Effect": "Deny", 
      "Principal": "*", 
      "Action": "s3:*", 
      "Resource": "arn:aws:s3:::my_tf_test_bucket/*", 
      "Condition": { 
         "IpAddress": {"aws:SourceIp": "8.8.8.8/32"} 
      } 
    } 
  ] 
} 
POLICY 

To convert the string contained in policy to a JSON structure perform the following: 

json_holder = json.unmarshal(rc.change.after.policy) 

Note: Don’t forget to import the JSON module/library. 

The line above should store the value of the JSON structure from the policy definition. 

Extracting JSON’s attribute values 

Notice in the previous example how Statement contains its values within braces, which means we need to call Statement in the way showed below. Notice it uses braces and an index within them. 

json_holder.Statement[0] 

Note: if you want to extract only the value of Statement, the braces and the index can be omitted, but later you won’t be able to dig deeper into the item. 

Digging deeper into the JSON structure 

To select an item within Statement, do the following: 

  • For direct attributes, like “Sid”: 

json_holder.Statement[0].Sid 

  • For deeper items like “aws:SourceIP” 

json_holder.Statement[0].Condition.IpAddress[“aws:SourceIp“]  

General Note: Review the first resource from GitHub as it contains the definitions for the function modules. 

If you liked my article or you found it useful please remember to leave lots of claps (50 if possible). 

Related posts