Workflow Applications Using Fortanix ACI

NOTE
Download the Python script for use with this example and unzip the tar file into your working directory. It is best to use the scripts provided in the tar file instead of copying the scripts outlined in the example below to avoid potential formatting issues.
workflow-scripts
4.07 KB

1.0 Introduction

Workflow graphs are maps that show how generic applications are connected to datasets and other generic applications. These are collaborative objects where multiple users can provide their own objects and approvals. For more information, refer to Create, Update, Clone, and Delete Workflows.

This article describes how to create and deploy Fortanix Confidential Computing Manager (CCM) Workflows using a simple data processing example.

1.1 A Research Portal for Patient Data

This example demonstrates a Fortanix CCM workflow with Python scripts deployed securely using Fortanix Enclave OS. The Workflow provides the most common diseases from patient data.

For input data, a synthetic file containing the medical history of fictive patients is used. The data used for this example is available from https://synthea.mitre.org/downloads.

File: 1K Sample Synthetic Patient Records, CSV[mirror]: 9 MB

This is made available via:
Jason Walonoski, Mark Kramer, Joseph Nichols, Andre Quina, Chris Moesel, Dylan Hall, 
Carlton Duffett, Kudakwashe Dube, Thomas Gallagher, Scott McLachlan, 
Synthea: An approach, method, and software mechanism for generating synthetic patients 
and the synthetic electronic health care record, Journal of the American Medical 
Informatics Association, Volume 25, Issue 3, March 2018, 
Pages 230–238, https://doi.org/10.1093/jamia/ocx079

The input and output datasets represent patient information collected by different healthcare provider and contain the following elements:

Patient demographic information
Patient conditions
Patient encounters
Patient medications

1.2 Prerequisites

Set up a default config for Amazon Web Service (AWS) with your Access Key/Secret Key.
Run the following command to install Boto:
```
sudo apt-get install python3-boto3
```

2.0 Create CCM Workflow

2.3 Select the Account

Create or select a Fortanix Armor account. For detailed instructions, refer to Getting Started with Fortanix Armor.

2.4 Create a Group

Create a group in the Fortanix Armor Identity and Access Management solution. For detailed instructions, refer to Fortanix Armor Identity and Access Management (IAM).

2.5 Invite Users to Join Fortanix Armor

An Administrator invites users who are Data Owners, App Owners, and Output Owners to join a Fortanix Armor account and create a workflow graph in Fortanix CCM.

For detailed instructions on inviting users, refer to Fortanix Armor Identity and Access Management.

The new users (Data Owners and App Owners) must join the Fortanix Armor account and create Datasets and App Configs as described in the following sections.

2.3 Create Input and Output Datasets

Datasets are the definitions containing the location and access credentials of the data that allow the Enclave OS in the workflow to download the data and upload the data. In this example, we use the AWS S3 bucket to:

Store an encrypted file that will be downloaded and decrypted by the enclave OS.
Upload an encrypted file using the credentials provided in the dataset.
Create an S3 bucket with a directory that is accessible using a URL.

2.3.1 Input User (Data Provider)

Consider that the Data Owner user has access to sensitive information and wants to allow an Application Owner to process this information.
This sensitive data is stored in a conditions.csv file.

In this example, the file is encrypted, uploaded to a storage solution (AWS S3), and a dataset is configured with credentials and an encryption key for enclave access or processing:

If you haven't already, download and untar this tar ball workflow-scripts.tar.gzfile:
```
tar xvfz workflow-scripts.tar.gz
```
Obtain a copy of the CSV sample data (also available here: https://synthea.mitre.org/downloads):
```
wget https://synthetichealth.github.io/synthea-sample-data/downloads/synthea_sample_data_csv_apr2020.zip
unzip synthea_sample_data_csv_apr2020.zip
```
You will only be using the file csv/conditions.csv to encrypt this file, generate a key locally, encrypt the file and store the key in a KMS securely:
1. Run the following command to generate an encryption key:
```
xxd -u -l 32 -p /dev/random | tr -d '\n' > ./key.hex
```
2. Use aes_256_gcm.py script included in the tar file that you downloaded. Run the following command to encrypt the file:
```
./workflow_scripts/aes_256_gcm.py enc -K ./key.hex -in ./csv/conditions.csv -out ./conditions.csv.enc
```
Run the following command to upload the encrypted file to a secure storage location such as AWS S3:
```
aws s3 --profile upload cp ./conditions.csv.enc <s3-directory> 
```
Generate a pre-signed URL to access the file to avoid inserting the whole AWS SDK in the example.
Use the presign.py script included in the tar file and run the following command:
```
./presign.py default download <s3-directory> conditions.csv.enc 86400
```
Where, S3-directory is the directory of your S3 bucket directory.
As an example, the output from the above command will be as shown below:
```
https://fortanix-pocs-data.s3.amazonaws.com/conditions.csv.enc?AWSAccessKeyId=&Signature=PcpH99nszG2%2Fv85z4IbgwgVDywc%3D&Expires=1613817035
```
The output consists of two parts:
- The location - https://fortanix-pocs-data.s3.amazonaws.com/conditions.csv.enc
- Query parameters: AWSAccessKeyId=&Signature=PcpH99nszG2%2Fv85z4IbgwgVDywc%3D&Expires=1613817035
  The query parameters must be base64 encoded for the dataset definition using the following command:
```
echo -n ‘AWSAccessKeyId=<key>&Signature=PcpH99nszG2%2Fv85z4IbgwgVDywc%3D&Expires=1613817035’ | base64
```
- The output will be as follows:
```
QVdTQWNjZXNzS2V5SWQ9QUtJQVhPNU42R0dOQ05WMzUzV1MmU2lnbmF0dXJlPVBjcEg5OW5zekcyJTJGdjg1ejRJYmd3Z1ZEeXdjJTNEJkV4cGlyZXM9MTYxMzgxNzAzNQ==
```
At this point, users accessing the URL above with full query string parameters will be able to download the encrypted file until it expires in 1 day. If the URL above expires, the dataset will need to be updated with new query parameters
NOTE
If you access the URL without the string following '?', you will get a 403 forbidden.
Hence, treat the query parameters as access credentials.

Perform the following steps to create an input dataset:

In the CCM UI left navigation panel, click Dataverse → DATASETS, and then click ADD DATASET to create a dataset.
Figure 1: Add dataset
In the Add dataset form:
1. Name – Enter the name of the dataset. For example: Conditions Data.
2. Description (optional) – Enter the description of the dataset. For example, Patients with associated conditions.
3. Labels (optional) – Attach one or more key-value labels to the dataset. For example, Key: Location and Value: East US
4. Group– Select the required group name from the drop down menu to associate this dataset with that group.
5. Location – The AWS S3 URL where data can be accessed. For example, https://fortanix-pocs-data.s3.amazonaws.com/conditions.csv.enc .
6. Long Description (optional) – Enter the content in a GitHub-flavoured Markdown file format. You can also use Fetch Long Description to get the Markdown file content from an external URL. Ensure that Cross-Origin Resource Sharing (CORS) is enabled on the external URL so that Fortanix CCM can access it. For more information, refer to the steps for AWS and Azure.
  The following is the sample description:
```
- Strikethrough Text
~~It is Strikethrough test..~~
- Blockquote Text
> Test Blockquote.
- Bold
**A sample Description.**
- Italic
*It is Italics*
- Bold and Italic
***Bold and Italics text***
- Link
This is [Website](https://www.fortanix.com/)?
```
7. Credentials – the credentials needed to access the data. The credentials must be in the correct JSON format and consist of:
  - Query parameters that were base64 encoded.
  - The key that was used to encrypt the file.
```
{
  "query_string": "QVdTQWNjZXNzS2V5SWQ9QUtJQVhPNU42R0dOQ05WMzUzV1MmU2lnbmF0dXJlPVBjcEg5OW5zekcyJTJGdjg1ejRJYmd3Z1ZEeXdjJTNEJkV4cGlyZXM9MTYxMzgxNzAzNQ==",
  "encryption": {
    "key": "63F0E4C07666126226D795027862ACC5848E939881C3CFE8CB3EB47DD7B3D24A"
}
```
  TIP
  Before saving the dataset, it is a good idea to verify that the JSON is correct. After saving the dataset you will not be able to view the credentials and access to data may fail. Any online JSON formatting tool can be used to validate that the JSON is correct.
  NOTE
  The credentials are only passed as text when creating the dataset over an HTTPS connection.
  It is then stored in a KMS (Fortanix Data Security Manager) and is only accessible to approved enclaves.
  Not even the Data Owner can retrieve the credentials.
Click ADD DATASET to create the input dataset.

2.3.2 Output User (Data Receiver)

Once the data has been received by the Enclave, the user application will run within the Enclave and generate output data (processed data). This data should be encrypted (using your key) before being uploaded to an untrusted store. This is achieved by defining an output dataset to be used by the workflow.
Perform the following steps:

Run the following command to generate an encryption key:

xxd -u -l 32 -p /dev/random | tr -d '\n' > ./key_out.hex

Use the presign.py script included in the tar file. Run the following command to generate a pre-signed URL for the upload:
```
./presign.py default upload <s3-directory> conditions_output.csv.enc 86400
```
Where, s3-directory is the directory of your S3 bucket directory.
As an example, the output of the above command will be as shown below:
```
https:/https://fortanix-pocs-data.s3.amazonaws.com/conditions_output.csv.enc?AWSAccessKeyId=<key>Signature=HFvhxaiKY0cGR9XqgGLp5zcAWac%3D&Expires=1613817880
```
The output consists of two parts:
- The location: https://fortanix-pocs-data.s3.amazonaws.com/conditions_output.csv.enc
- Query parameters: AWSAccessKeyId=Signature=HFvhxaiKY0cGR9XqgGLp5zcAWac%3D&Expires=1613817880
  The query parameters must be base64 encoded for the dataset definition using the following command:
```
echo -necho -n ‘AWSAccessKeyId=<key>Signature=HFvhxaiKY0cGR9XqgGLp5zcAWac%3D&Expires=1613817880’ | base64
```
- The output will be as follows:
```
QVdTQWNjZXNzS2V5SWQ9QUtJQVhPNU42R0dOTk1SWFZLUEEmU2lnbmF0dXJlPUhGdmh4YWlLWTBjR1I5WHFnR0xwNXpjQVdhYyUzRCZFeHBpcmVzPTE2MTM4MTc4ODA=
```
Create an output dataset with the following sample values:
1. Name – the dataset name. For example: Conditions processing output.
2. Description (optional) – the dataset description.
3. Labels (optional)– attach one or more key-value labels to the dataset. For example, Key: Location and Value: East US.
4. Group – Select the required group name from the drop down menu to associate this dataset with that group.
5. Location – The URL where data can be accessed. For example, https://fortanix-pocs-data.s3.amazonaws.com/conditions_output.csv.enc
6. Long Description (optional) – Enter the content in GitHub-flavoured Markdown file format. You can also use Fetch Long Description to get the Markdown file from an external URL. Ensure that Cross-origin resource sharing (CORS) is enabled on the external URL so that Fortanix CCM can access it. For more information, refer to the steps for AWS and Azure.
  The following is the sample long description in Markdown format:
```
- Strikethrough Text
~~It is Strikethrough test..~~

- Blockquote Text
> Test Blockquote.

- Bold
**A sample Description.**

- Italic
*It is Italics*

- Bold and Italic
***Bold and Italics text***

- Link
This is [Website](https://www.fortanix.com/)?
```
7. Credentials – the credentials needed to access the data. The credentials must be in the correct JSON format and consist of:
  - Query parameters that were base64 encoded.
  - The key that was used to encrypt the file.
```
{
{
  "query_string": "<my-query-string>",
  "encryption": {
    "key": "<my-key>"
  }
}
```
  For example:
```
{
  "query_string": "QVdTQWNjZXNzS2V5SWQ9QUtJQVhPNU42R0dOTk1SWFZLUEEmU2lnbmF0dXJlPUhGdmh4YWlLWTBjR1I5WHFnR0xwNXpjQVdhYyUzRCZFeHBpcmVzPTE2MTM4MTc4ODA=","encryption": {
    "key": "63F0E4C07666126226D795027862ACC5848E939881C3CFE8CB3EB47DD7B3D24A"
  }
}
```
  TIP
  Before saving the dataset, it is a good idea to verify that the JSON is correct. After saving the dataset you will not be able to view the credentials and access to data may fail. Any online JSON formatting tool can be used to validate that the JSON is correct.
  NOTE
  The credentials are only passed as text when creating the dataset over an HTTPS connection.
  It is then stored in a KMS (Fortanix Data Security Manager) and only accessible to approved enclaves.
  Not even the Data Owner can retrieve the credentials.

2.4 ACI Application

2.4.1 Create a General Purpose Python Docker Image

Create a docker image that will run arbitrary protected python code. The following files are used. These files are included in the tar file provided with this example:

Dockerfile:

FROM python:3.6

RUN apt-get update && apt-get install -y python3-cryptography python3-requests python3-pandas
RUN mkdir -p /opt/fortanix/enclave-os/app-config/rw
RUN mkdir -p /demo/code /demo/input

COPY ./start.py ./utils.py ./describe.py /demo/

CMD ["/demo/start.py"]

start.py: This file is the main entry point into the application.

!/usr/bin/python3
  
import os
import utils
import hashlib
from subprocess import PIPE, run

def main():
    input_folder="/opt/fortanix/enclave-os/app-config/rw"
    
    command = ["/usr/bin/python3", "/demo/describe.py"]
    
    # This downloads and decrypts all input data. File names are the object names from app config.
    for i in utils.read_json_datasets("input"):
        decrypted = utils.get_dataset(i)
        open(input_folder + i.name, 'wb').write(decrypted)
        
        # Add the file as input argument for our script
        command.append(input_folder + i.name)
        
    print("Running script")
    result = run(command, stdout=PIPE, stderr=PIPE, universal_newlines=True)
    
    # For simplicity uploading just stdout/stderr/returncode.
    utils.upload_result("output", result.returncode, result.stdout, result.stderr)
    
    print("Execution complete")

if __name__ == "__main__":
    main()

utils.py: This file contains the set of library functions.

!/usr/bin/env python3

import os
import sys
import string
import json
import requests
import base64
import hashlib
from subprocess import PIPE, run

from cryptography.hazmat.backends import default_backend
from cryptography.hazmat.primitives.ciphers import (
    Cipher, algorithms, modes
)

NONCE_SIZE=12
TAG_SIZE=16
KEY_SIZE=32

PORTS_PATH="/opt/fortanix/enclave-os/app-config/rw/"

def convert_key(key_hex):
    key=key_hex.rstrip();
    if len(key) != 64:
        raise Exception("Key file must be 64 bytes hex string for AES-256-GCM")

    if not all(c in string.hexdigits for c in key) or len(key) != 64:
        raise Exception("Key must be a 64 character hex stream")

    return bytes.fromhex(key)

def encrypt_buffer(key, data):
    nonce=os.urandom(NONCE_SIZE)

    cipher = Cipher(algorithms.AES(key), modes.GCM(nonce), backend=default_backend())
    encryptor = cipher.encryptor()

    return nonce + encryptor.update(data.encode()) + encryptor.finalize() + encryptor.tag

def decrypt_buffer(key, data):
    tag=data[-TAG_SIZE:]
    nonce=data[:NONCE_SIZE]

    cipher = Cipher(algorithms.AES(key), modes.GCM(nonce, tag), backend=default_backend())
    decryptor = cipher.decryptor()

    return decryptor.update(data[NONCE_SIZE:len(data)-TAG_SIZE]) + decryptor.finalize()

class JsonDataset:
    def __init__(self, location, credentials, name):
       self.location = location
       self.credentials = credentials
       self.name = name

def read_json_datasets(port):
    ports=[]
    for folder in os.listdir(PORTS_PATH + port):
        subfolder=PORTS_PATH + port + "/" + folder
        if os.path.exists(subfolder + "/dataset"):
            credentials=json.load(open(subfolder + "/dataset/credentials.bin", "r"))
            location=open(subfolder + "/dataset/location.txt", "r").read()
            ports.append(JsonDataset(location, credentials, folder))

    return ports

def get_dataset(dataset):
    url = dataset.location + "?" + base64.b64decode(dataset.credentials["query_string"]).decode('ascii')
    r = requests.get(url, allow_redirects=True)
    r.raise_for_status()

    print("Retrieved dataset from location: " + dataset.location)
    key = convert_key(dataset.credentials["encryption"]["key"])
    return decrypt_buffer(key, r.content)

class RunResult:
    def __init__(self, returncode, stdout, stderr):
        self.returncode = returncode
        self.stdout = base64.b64encode(stdout.encode()).decode('ascii')
        self.stderr = base64.b64encode(stderr.encode()).decode('ascii')

def upload_result(port, returncode, stdout, stderr):
    result=RunResult(returncode, stdout, stderr)
    json_str=json.dumps(result.__dict__)

    for dataset in read_json_datasets(port):
        url = dataset.location + "?" + base64.b64decode(dataset.credentials["query_string"]).decode('ascii')
        key = convert_key(dataset.credentials["encryption"]["key"])

        print("Writing output to location: " + dataset.location)
        requests.put(url, encrypt_buffer(key, json_str))

describe.py: This file contains the custom code called by start.py.

!/usr/bin/python3
  
import pandas as pd
import sys

for i in sys.argv[1:]:
    df_do = pd.read_csv(i)
    print("Dataset: " + i + "\n")
    print(df_do['DESCRIPTION'].describe())
    print("")

A standard Docker build and Docker push command must be used to create your Docker image and push it to your registry. For example:

docker build -t <my-registry>/simple-python-sgx .
docker push <my-registry>/simple-python-sgx

2.4.1 Create an ACI Application and Build Creation

This section describes how to register the Docker image in CCM and convert it to an SGX format.

Perform the following steps:

To create an ACI application using Fortanix CCM UI:
1. Navigate to the Applications menu item in the CCM UI left navigation panel and click ADD APPLICATION to create a new application.
2. Now, in the Add application dialog box, select ACI application, and click NEXT.
3. In the Add application form, add the details of the application and click ADD APPLICATION to complete creating the application.
For detailed instructions, refer to Add ACI Application.
Create a build of the application.
For detailed instructions, refer to Create a Build for ACI Applications.
Fetch the build whitelisting tasks using the Fortanix CCM Tasks tab.
Approve the task.

2.5 Create Application Configuration

The Docker Image recognizes the Python script using an Application Configuration which defines the ports.

Perform the following steps to create an Application Configuration:

Navigate to Applications → CONFIGURATIONS in the Fortanix CCM UI left navigation panel.
Figure 2: Create application configuration
Click ADD CONFIGURATION to add a new configuration.
1. Configuration Name – Enter a name for the configuration.
2. Description: Enter a description for the configuration.
3. Group: Select the required group name from the drop down menu.
4. Build: Select the application build for which the configuration will be created.
5. Ports: Specify the ports to be used in the workflow. Multiple ports can be added, depending on the required connections. For example: input, output, heartbeat, and so on.
6. Label Details (optional): Add one or more key-value labels to the configuration.
7. Configuration items: Click ADD CONFIGURATION ITEM and define key-value pairs used to configure the application.
NOTE
For ACI applications, Fortanix permits only files in the path /opt/fortanix/.
Click ADD CONFIGURATION to save the configuration.

2.6 Create a Workflow

Perform the following steps to create a Workflow:

In the CCM UI left navigation panel, click Workflows.
On the Workflows page, click ADD WORKFLOW to create a new workflow.
Figure 3: Create a workflow
In the Add Workflow form:
1. Name: Enter a name for the workflow.
2. Group: Select the Fortanix Armor IAM group for the workflow. If you do not select a group, Fortanix CCM uses the default group.
3. Click ADD WORKFLOW to create the workflow.
Add an application to the workflow graph. Drag the Application icon from the component panel and drop it into the graph area.
Click ADD APPLICATION. In the Add Application form, select an existing application name and image. Select a build for this application. For example, <my-registry>/simple-python-sgx:latest.
Where, <my-registry> is the location of your container registry.
Select an existing app configuration or create a new app configuration using ADD NEW CONFIGURATION for the application build.
Add input and output datasets to the workflow graph. Drag the Dataset icon into the graph area and click ADD DATASET.
In the DATASET form, select an existing dataset, or click ADD DATASET to create a new dataset.
Connect the workflow components. Establish connections between datasets and applications as follows:
1. Connect the Input Dataset to the Application using the Input target port.
2. Connect the Application to the Output Dataset using the Output target port.
After the workflow is completed, click SAVE AND REQUEST APPROVAL to initiate the workflow approval process.
The workflow remains in a pending state until it receives approval from all users. In the Pending menu item, click VIEW REQUEST to approve a Workflow.
In the APPROVAL REQUEST FOR CREATING WORKFLOW dialog, you can either APPROVE or DECLINE a workflow.
WARNING
When a draft Workflow is submitted for approval, it will be removed from the drafts list, and editing it directly will no longer be possible once it is in a "pending" or "approved" state.
NOTE
- A user can also approve/decline a Workflow from the CCM Tasks menu item.
- Notice that the users who have approved the Workflow have a green tick against their icon.
All the users of a Workflow must approve the Workflow to finalize it. If a user declines a Workflow, the Workflow is rejected. When all the users approve the Workflow, it is deployed.
1. CCM configures apps to access the datasets.
2. CCM creates the Workflow Application Configs.
3. CCM returns the list of hashes needed to start the apps.

2.7 Run the Application

After a Workflow is approved by all the users, the Applications will have the Workflow Application Configurations provided to them. This configuration has information such as which Datasets or Apps they are connected to, any user-provided files or values to be provided within an enclave, and so on.

We provide a configuration to applications using an identifier passed as an input argument.

This identifier is a sha256sum of items that you need to secure from the configuration and workflow.

Fortanix CCM will also embed this identifier inside the certificates it issues so that it is clear what configuration is used for the KMS to allow access to credentials.

It embeds this inside a subject alternate name:

<identifier>.<mrenclave>.id.fortanix.cloud

With the identifier above, the KMS that stores the dataset credentials will authenticate and give credentials only to applications that present a proper certificate. When the application starts, CCM will keep track of which applications are allowed to access which configurations using the identifier.

Perform the following steps to view the Application Identifier:

Click the application in the approved Workflow graph.
In the detailed view of the Workflow application, copy the value of Runtime configuration hash. This ID is used to run the application.
To deploy the ACI application using the Azure portal, perform the steps as mentioned in Section 2 - Deploy Confidential ACI Group Using Azure Portal. In the "Custom deployment" screen, for the App Config Id field, enter the Application Identifier copied in Step 2 above.
Figure 4: Azure ACI deployment

When the App Owner starts the application with the application config identifier.
The Data Output Owner can view the output using the following steps:

Run the following command to download the output file:

aws s3 --profile download cp s3:<s3-directory>/conditions_output.csv.enc .

For example:

aws s3 --profile download cp s3://fortanix-pocs-data/conditions_output.csv.enc .

Run the following command to decrypt the file:
Use the aes_256_gcm.py script provided in the tar file.

./aes_256_gcm.py dec -K ./key.hex -in ./conditions_output.csv.enc -out ./output.txt
$ cat output.txt | jq .
{
  "returncode": 0,
  "stdout": "RGF0YXNldDogL29wdC9mb3J0YW5peC9lbmNsYXZlLW9zL2FwcC1jb25maWcvaW5wdXQvY3hoY2Z4ZHZsCgpjb3VudCAgICAgICAgICAgICAgICAgICAgICAgICAgIDgzNzYKdW5pcXVlICAgICAgICAgICAgICAgICAgICAgICAgICAgMTI5CnRvcCAgICAgICBWaXJhbCBzaW51c2l0aXMgKGRpc29yZGVyKQpmcmVxICAgICAgICAgICAgICAgICAgICAgICAgICAgIDEyNDgKTmFtZTogREVTQ1JJUFRJT04sIGR0eXBlOiBvYmplY3QKCg==",
  "stderr": ""
}

The file now has the expected output:

$ cat output.txt | jq -r .stdout | base64 -d
Dataset: /opt/fortanix/enclave-os/app-config/input/cxhcfxdvl

count                               8376 
unique                               129 
top           Viral sinusitis (disorder) 
freq                                1248 
Name: DESCRIPTION, dtype: object

When the App Owner starts the application with the application config identifier:
1. Applications will request an attestation certificate from the NodeAgent with the identifier as part of the report data.
2. The application requests an application certificate from the NodeAgent.
3. The Fortanix CCM verifies that the application is allowed to access the configuration.
4. The application requests from Fortanix CCM its configuration by providing its certificate provisioned above as an authentication mechanism.
5. The CCM does certificate authentication, extracts the application identifier from the certificate, and sends back the configuration corresponding to that identifier.
6. The application verifies and applies the configuration hash.
7. The application gets the credentials from URLs in the config.
8. The application authenticates and reads or writes data from the datasets.