Skip to main content

Identity Foundation: SPIFFE/SPIRE

CitadelMesh implements zero-trust security using SPIFFE (Secure Production Identity Framework For Everyone) and SPIRE (SPIFFE Runtime Environment). Every workload, agent, and adapter receives a cryptographic identity that enables mutual authentication and authorization without shared secrets.

Why this matters: Legacy building automation assumes “inside the LAN equals trusted.” CitadelMesh verifies every request—even between pods on the same edge cluster—so compromise of one adapter or contractor laptop does not give carte blanche over the facility. Combined with Edge Architecture, this identity fabric is what makes offline autonomy safe.

Zero-Trust Principles​

Traditional building systems rely on network perimeter security: "If you're on the building network, you're trusted." This model fails in modern architectures with:

  • Cloud connectivity
  • Third-party integrations
  • Mobile access
  • Compromised credentials

Zero-trust requires:

  1. Verify explicitly: Authenticate every request based on identity
  2. Least privilege: Grant minimal access required for each workload
  3. Assume breach: Design for compromise; limit blast radius

SPIFFE Identity Model​

SPIFFE ID Structure​

Every workload receives a unique SPIFFE ID in URI format:

spiffe://trust-domain/path/to/workload

CitadelMesh Trust Domains:

spiffe://citadel.mesh/agent/security
spiffe://citadel.mesh/agent/energy
spiffe://citadel.mesh/adapter/bacnet
spiffe://citadel.mesh/adapter/security-expert
spiffe://citadel.mesh/service/twin
spiffe://citadel.mesh/service/opa-policy

Trust Domain (citadel.mesh): The security boundary; all workloads in this domain trust the same root CA

Workload Path (/agent/security): Hierarchical path identifying the specific workload

SPIFFE Verifiable Identity Document (SVID)​

SPIRE issues SVIDs as proof of identity:

SVID = X.509 Certificate {
  Subject: spiffe://citadel.mesh/agent/security
  Issuer: SPIRE Server CA
  Valid: 2025-09-30 to 2025-10-01 (1 hour)
  Key Usage: Digital Signature, Key Encipherment
  Extended Key Usage: Server Auth, Client Auth
}

Key Properties:

  • Short-lived: 1-hour validity; automatic rotation
  • Cryptographically signed: By SPIRE Server CA
  • Mutual TLS: Used for both client and server authentication
  • No shared secrets: Each workload gets unique keys

SPIRE Architecture​

SPIRE Server​

Central authority for identity issuance:

  • Certificate Authority: Root CA for trust domain
  • Workload Registry: Defines which workloads can receive which identities
  • Attestation: Verifies workload authenticity before issuing SVIDs
  • Federation: Trusts other SPIRE servers for multi-site deployments

SPIRE Agent​

Runs on every node (edge and cloud):

  • Workload API: Unix domain socket or named pipe for SVID requests
  • Attestation: Proves node identity to SPIRE Server
  • SVID Caching: Caches SVIDs and rotates before expiry
  • Health Monitoring: Monitors workload lifecycle

Workload Attestation​

How does SPIRE know a workload is legitimate before issuing an SVID?

Node Attestation​

SPIRE Agent proves the node's identity:

Edge (K3s):

# SPIRE Agent config
node_attestor "k8s_psat" {
  cluster = "citadel-edge-building-a"
}

Cloud (Kubernetes):

node_attestor "k8s_sat" {
  cluster = "citadel-cloud-prod"
}

Workload Attestation​

SPIRE Server verifies workload identity via selectors:

# SPIRE Server registration entry
Entry {
  spiffe_id = "spiffe://citadel.mesh/agent/security"
  parent_id = "spiffe://citadel.mesh/node/edge-building-a"
  selectors = [
    "k8s:ns:citadel-agents",
    "k8s:sa:security-agent",
    "k8s:pod-label:app:security-agent"
  ]
  ttl = 3600
}

Selectors match Kubernetes pod metadata:

  • Namespace: citadel-agents
  • Service Account: security-agent
  • Pod Label: app=security-agent

Only pods matching ALL selectors receive this SPIFFE ID.

Identity-Based Authentication​

mTLS Everywhere​

All network communication uses mutual TLS with SPIFFE SVIDs:

from spiffe import SpiffeClient, X509Source

# Initialize SPIFFE workload API client
x509_source = X509Source()
await x509_source.start()

# Get this workload's SVID
my_svid = x509_source.get_x509_svid()
print(f"My identity: {my_svid.spiffe_id}")

# Create gRPC channel with mTLS
import grpc
from grpc import ssl_channel_credentials

# Server credentials (verify client)
server_credentials = grpc.ssl_server_credentials(
    [(my_svid.private_key_bytes, my_svid.cert_chain_bytes)],
    root_certificates=x509_source.get_bundle_for_trust_domain("citadel.mesh"),
    require_client_auth=True
)

# Client credentials (verify server)
client_credentials = grpc.ssl_channel_credentials(
    root_certificates=x509_source.get_bundle_for_trust_domain("citadel.mesh"),
    private_key=my_svid.private_key_bytes,
    certificate_chain=my_svid.cert_chain_bytes
)

# Create authenticated channel
channel = grpc.secure_channel(
    "twin-service.citadel.svc:8443",
    client_credentials
)

NATS JetStream with SPIFFE​

Event bus authentication using SPIFFE:

import nats
from nats.aio.client import Client as NATS

# Connect to NATS with mTLS
nc = await nats.connect(
    servers=["nats://nats.citadel.svc:4222"],
    tls=nats.tls.TLS(
        cert=my_svid.cert_chain_bytes,
        key=my_svid.private_key_bytes,
        ca=x509_source.get_bundle_for_trust_domain("citadel.mesh")
    )
)

# Publish with authenticated identity
await nc.publish(
    "telemetry.canonical.building_a",
    event.SerializeToString()
)

Authorization with SPIFFE​

Identity alone isn't sufficient; we need authorization policies.

OPA Integration​

OPA policies check SPIFFE IDs for authorization:

package citadel.authz

import rego.v1

# Default deny
default allow := false

# Energy agent can write HVAC setpoints
allow if {
    input.spiffe_id == "spiffe://citadel.mesh/agent/energy"
    input.action == "write_setpoint"
    startswith(input.target, "hvac.")
}

# Security agent can control doors
allow if {
    input.spiffe_id == "spiffe://citadel.mesh/agent/security"
    input.action in ["lock_door", "unlock_door"]
    startswith(input.target, "door.")
}

# Twin service can read everything
allow if {
    input.spiffe_id == "spiffe://citadel.mesh/service/twin"
    input.action == "read"
}

# Deny with reason
deny_reason := sprintf(
    "SPIFFE ID %s not authorized for action %s on %s",
    [input.spiffe_id, input.action, input.target]
) if not allow

JWT Claims with SPIFFE​

CloudEvents carry JWT tokens with SPIFFE-based claims:

import jwt
from datetime import datetime, timedelta

def create_capability_token(svid, capabilities):
    """Create signed JWT with SPIFFE identity and capabilities."""

    payload = {
        "sub": str(svid.spiffe_id),  # spiffe://citadel.mesh/agent/energy
        "iss": "spiffe://citadel.mesh/opa-policy",
        "aud": ["citadel.control"],
        "exp": datetime.utcnow() + timedelta(minutes=5),
        "iat": datetime.utcnow(),
        "capabilities": capabilities,
        "constraints": {
            "temp_min": 65,
            "temp_max": 78
        }
    }

    # Sign with OPA service's SVID
    token = jwt.encode(
        payload,
        opa_svid.private_key_bytes,
        algorithm="RS256"
    )

    return token

Command includes capability token:

command = Command(
    id=ulid(),
    target_id="hvac.zone1.setpoint",
    action="write_setpoint",
    params={"value": "72"},
    safety_token=create_capability_token(
        my_svid,
        capabilities=["hvac:write_setpoint"]
    ),
    issued_by=str(my_svid.spiffe_id)
)

SVID Rotation​

SVIDs are short-lived (1 hour) and automatically rotate:

from spiffe import X509Source
import asyncio

async def maintain_identity():
    """Continuously monitor SVID rotation."""

    x509_source = X509Source()
    await x509_source.start()

    while True:
        svid = x509_source.get_x509_svid()

        # Log identity
        logger.info(f"Current SVID: {svid.spiffe_id}")
        logger.info(f"Expires: {svid.expiry}")

        # X509Source automatically rotates before expiry
        # Just sleep and let it handle rotation
        await asyncio.sleep(300)  # Check every 5 minutes

Rotation Flow:

  1. SPIRE Agent requests new SVID 30 minutes before expiry
  2. SPIRE Server validates workload still matches selectors
  3. Server issues new SVID with fresh keys
  4. Agent provides new SVID via Workload API
  5. Workload updates mTLS connections with new SVID
  6. Old SVID remains valid until expiry (grace period)

Multi-Site Federation​

Enterprise deployments span multiple buildings/sites:

Federation Configuration​

# SPIRE Server A config
federates_with "citadel.building-b" {
  bundle_endpoint_url = "https://spire-server-b.citadel.io/bundle"
  bundle_endpoint_profile "https_web" {
    endpoint_spiffe_id = "spiffe://citadel.building-b/spire-server"
  }
}

federates_with "citadel.cloud" {
  bundle_endpoint_url = "https://spire-cloud.citadel.io/bundle"
  bundle_endpoint_profile "https_web" {
    endpoint_spiffe_id = "spiffe://citadel.cloud/spire-server"
  }
}

Cross-trust-domain authentication:

# Building A agent calls Building B service
x509_source = X509Source()

# Get my SVID (Building A)
my_svid = x509_source.get_x509_svid()
# spiffe://citadel.building-a/agent/security

# Get trust bundle for Building B
building_b_bundle = x509_source.get_bundle_for_trust_domain(
    "citadel.building-b"
)

# Create mTLS channel to Building B
channel = grpc.secure_channel(
    "service.building-b.citadel.io:8443",
    grpc.ssl_channel_credentials(
        root_certificates=building_b_bundle,
        private_key=my_svid.private_key_bytes,
        certificate_chain=my_svid.cert_chain_bytes
    )
)

Observability​

Identity events are fully auditable:

# Emit identity events
logger.info(
    "SVID issued",
    extra={
        "spiffe_id": svid.spiffe_id,
        "serial_number": svid.serial_number,
        "expiry": svid.expiry,
        "selectors": selectors
    }
)

logger.info(
    "mTLS connection established",
    extra={
        "client_spiffe_id": client_svid.spiffe_id,
        "server_spiffe_id": server_svid.spiffe_id,
        "peer_trust_domain": peer_trust_domain
    }
)

logger.warning(
    "Authorization denied",
    extra={
        "spiffe_id": svid.spiffe_id,
        "action": action,
        "target": target,
        "policy": policy_name,
        "reason": deny_reason
    }
)

Deployment Architecture​

Edge Deployment​

# K3s DaemonSet for SPIRE Agent
apiVersion: apps/v1
kind: DaemonSet
metadata:
  name: spire-agent
  namespace: spire
spec:
  selector:
    matchLabels:
      app: spire-agent
  template:
    spec:
      hostPID: true
      hostNetwork: true
      containers:
      - name: spire-agent
        image: ghcr.io/spiffe/spire-agent:1.8.0
        volumeMounts:
        - name: spire-agent-socket
          mountPath: /run/spire/sockets
        - name: spire-config
          mountPath: /etc/spire
      volumes:
      - name: spire-agent-socket
        hostPath:
          path: /run/spire/sockets
          type: DirectoryOrCreate

Workload Integration​

# Security Agent deployment with SPIRE
apiVersion: apps/v1
kind: Deployment
metadata:
  name: security-agent
  namespace: citadel-agents
spec:
  template:
    metadata:
      labels:
        app: security-agent
    spec:
      serviceAccountName: security-agent
      containers:
      - name: agent
        image: citadel/security-agent:latest
        env:
        - name: SPIFFE_ENDPOINT_SOCKET
          value: unix:///run/spire/sockets/agent.sock
        volumeMounts:
        - name: spire-agent-socket
          mountPath: /run/spire/sockets
          readOnly: true
      volumes:
      - name: spire-agent-socket
        hostPath:
          path: /run/spire/sockets

Security Considerations​

Threat Model​

Threats Mitigated:

  • Credential theft (no long-lived secrets)
  • Man-in-the-middle (mTLS required)
  • Privilege escalation (least privilege)
  • Lateral movement (identity-based segmentation)

Residual Risks:

  • SPIRE Server compromise (rotate root CA)
  • Node compromise (limit blast radius with segmentation)
  • Supply chain attacks (signed container images, SBOMs)

Best Practices​

  1. Minimize SVID TTL: 1 hour max; shorter for high-risk workloads
  2. Strict Selectors: Multiple selectors (namespace + SA + labels)
  3. Audit Everything: Log all SVID issuance and auth decisions
  4. Rotate Root CA: Annual root CA rotation with grace period
  5. Monitor Anomalies: Alert on unusual SVID requests or auth failures

See Also​