Claude
Skills
Sign in
Back

implementing-aws-nitro-enclave-security

Included with Lifetime
$97 forever

Implements AWS Nitro Enclave-based confidential computing environments with cryptographic attestation, KMS policy integration using PCR-based condition keys, and secure vsock communication channels. The practitioner builds enclave images, configures attestation-aware KMS policies, validates attestation documents against the AWS Nitro PKI root of trust, and establishes isolated computation pipelines for processing sensitive data such as PII, cryptographic keys, and healthcare records. Activates for requests involving Nitro Enclave setup, enclave attestation validation, confidential computing on AWS, or KMS enclave policy configuration.

Cloud & DevOpsAWS-Nitro-Enclavesconfidential-computingattestationKMSenclave-isolationvsockPCRscripts

What this skill does

# Implementing AWS Nitro Enclave Security

## When to Use

- Processing sensitive data (PII, PHI, financial records, cryptographic secrets) that must be isolated from EC2 instance operators and administrators
- Building confidential computing pipelines where even root-level access on the parent instance cannot read enclave memory or state
- Implementing cryptographic attestation workflows that tie KMS decryption rights to a specific, verified enclave image hash
- Deploying multi-party computation environments where two or more enclaves authenticate each other via attestation before exchanging data
- Hardening existing workloads that currently decrypt secrets on the parent instance by migrating decryption into an enclave boundary

**Do not use** when the workload does not handle sensitive data that requires hardware-level isolation, when the instance type does not support Nitro Enclaves (requires Nitro-based instances with at least 4 vCPUs), or when latency constraints make the vsock communication overhead unacceptable.

## Prerequisites

- An AWS account with permissions to launch Nitro-capable EC2 instances (m5.xlarge or larger, C5, R5, M6i families)
- AWS CLI v2 and the `nitro-cli` toolset installed on the parent EC2 instance (Amazon Linux 2 or AL2023)
- Docker installed on the parent instance for building enclave image files (EIF)
- An AWS KMS symmetric key with key policy permissions for the enclave's IAM role
- The `aws-nitro-enclaves-sdk-c` or Python `aws-encryption-sdk` for enclave-side KMS operations
- The Nitro Enclaves allocator service configured with sufficient memory and vCPU allocation in `/etc/nitro_enclaves/allocator.yaml`

## Workflow

### Step 1: Configure the Nitro Enclaves Environment

Set up the parent EC2 instance to support enclave launches:

- **Install the Nitro Enclaves CLI**: On Amazon Linux 2, install the tools and allocator:
  ```bash
  sudo amazon-linux-extras install aws-nitro-enclaves-cli
  sudo yum install aws-nitro-enclaves-cli-devel -y
  sudo systemctl enable --now nitro-enclaves-allocator.service
  sudo systemctl enable --now docker
  sudo usermod -aG ne ec2-user
  sudo usermod -aG docker ec2-user
  ```
- **Configure memory and CPU allocation**: Edit `/etc/nitro_enclaves/allocator.yaml` to reserve resources for the enclave. The enclave requires dedicated memory that is carved from the parent instance:
  ```yaml
  ---
  memory_mib: 4096
  cpu_count: 2
  ```
  Restart the allocator: `sudo systemctl restart nitro-enclaves-allocator.service`
- **Verify setup**: Run `nitro-cli describe-enclaves` to confirm the CLI can communicate with the Nitro hypervisor. An empty JSON array `[]` indicates no enclaves are running and the setup is correct.

### Step 2: Build the Enclave Image File (EIF)

Package the sensitive workload into a signed enclave image:

- **Create the application Dockerfile**: The enclave runs a minimal Linux environment. The application communicates exclusively through vsock:
  ```dockerfile
  FROM amazonlinux:2

  RUN yum install -y python3 python3-pip && \
      pip3 install boto3 cbor2 cryptography requests

  COPY enclave_app.py /app/enclave_app.py

  WORKDIR /app
  CMD ["python3", "enclave_app.py"]
  ```
- **Build the EIF with nitro-cli**: Convert the Docker image into an enclave image file, capturing the PCR measurements:
  ```bash
  docker build -t enclave-app:latest .
  nitro-cli build-enclave \
    --docker-uri enclave-app:latest \
    --output-file enclave-app.eif
  ```
  The output contains three critical PCR values:
  - **PCR0**: SHA-384 hash of the enclave image file (the full image digest)
  - **PCR1**: SHA-384 hash of the Linux kernel and bootstrap process
  - **PCR2**: SHA-384 hash of the application code
  Record these values; they are used in KMS key policies for attestation-based access control.

- **Build a signed EIF** (recommended for production): Generate a signing certificate and use it to produce PCR8:
  ```bash
  openssl ecparam -name secp384r1 -genkey -noout -out enclave_key.pem
  openssl req -new -key enclave_key.pem -sha384 \
    -nodes -subj "/CN=Enclave Signer" -out enclave_csr.pem
  openssl x509 -req -days 365 -in enclave_csr.pem \
    -signkey enclave_key.pem -sha384 -out enclave_cert.pem

  nitro-cli build-enclave \
    --docker-uri enclave-app:latest \
    --output-file enclave-app.eif \
    --private-key enclave_key.pem \
    --signing-certificate enclave_cert.pem
  ```
  PCR8 (the signing certificate hash) enables KMS policies that trust any image signed by a specific certificate, allowing image updates without changing the policy.

### Step 3: Configure KMS Attestation-Based Key Policies

Create a KMS key policy that restricts decryption to a verified enclave:

- **Policy using PCR0 (image hash)**: This locks the key to a specific enclave build. Any code change produces a new PCR0, requiring a policy update:
  ```json
  {
    "Version": "2012-10-17",
    "Statement": [
      {
        "Sid": "AllowEnclaveDecrypt",
        "Effect": "Allow",
        "Principal": {
          "AWS": "arn:aws:iam::111122223333:role/EnclaveParentRole"
        },
        "Action": [
          "kms:Decrypt",
          "kms:GenerateDataKey"
        ],
        "Resource": "*",
        "Condition": {
          "StringEqualsIgnoreCase": {
            "kms:RecipientAttestation:ImageSha384": "fedcba9876543210fedcba9876543210fedcba9876543210fedcba9876543210fedcba9876543210fedcba9876543210"
          }
        }
      }
    ]
  }
  ```
- **Policy using PCR8 (signing certificate)**: Trusts any enclave signed with a specific certificate, enabling image rotation without policy changes:
  ```json
  {
    "Condition": {
      "StringEqualsIgnoreCase": {
        "kms:RecipientAttestation:PCR8": "ab3456789012345678901234567890123456789012345678901234567890123456789012345678901234567890abcdef"
      }
    }
  }
  ```
- **Multi-PCR policy for defense in depth**: Combine PCR0 (image) and PCR1 (kernel) to ensure both the application and the boot environment match expected values:
  ```json
  {
    "Condition": {
      "StringEqualsIgnoreCase": {
        "kms:RecipientAttestation:PCR0": "<pcr0-hex>",
        "kms:RecipientAttestation:PCR1": "<pcr1-hex>"
      }
    }
  }
  ```
- **IAM role policy**: The parent instance's IAM role must have `kms:Decrypt` permission, but the KMS key policy condition ensures the actual decryption only succeeds when the request originates from a valid enclave with the correct attestation document attached.

### Step 4: Implement Secure Vsock Communication

Establish the parent-to-enclave communication channel:

- **Vsock architecture**: The only way an enclave communicates with the outside world is through a vsock (virtual socket). Vsock uses a CID (Context Identifier) and port number. The parent instance CID is always `3`, and the enclave CID is assigned at launch.
- **Parent-side proxy server**: The parent runs a proxy that forwards KMS API calls from the enclave through the vsock to the AWS KMS endpoint:
  ```python
  import socket
  import json
  import boto3

  VSOCK_CID = 3  # Parent CID
  VSOCK_PORT = 5000

  def start_proxy():
      sock = socket.socket(socket.AF_VSOCK, socket.SOCK_STREAM)
      sock.bind((VSOCK_CID, VSOCK_PORT))
      sock.listen(5)

      kms_client = boto3.client('kms', region_name='us-east-1')

      while True:
          conn, addr = sock.accept()
          data = conn.recv(65536)
          request = json.loads(data.decode())

          if request['action'] == 'decrypt':
              response = kms_client.decrypt(
                  CiphertextBlob=bytes.fromhex(request['ciphertext']),
                  Recipient={
                      'KeyEncryptionAlgorithm': 'RSAES_OAEP_SHA_256',
                      'AttestationDocument': bytes.fromhex(request['attestation_doc'])
                  }
              )
              conn.sendall(json.dumps({
                  'ciphertext_for_recipient': response['CiphertextForRecipient'].hex()
              }).encode()

Related in Cloud & DevOps