Compare commits

...
Author SHA1 Message Date
Sharang ParnerkarandClaude Fable 5 896a06e8f6 style: apply rustfmt to the provision-and-test modules
CI / Check (pull_request) Successful in 6m1s
CI / Detect Changes (pull_request) Has been skipped
CI / Deploy Agent (pull_request) Has been skipped
CI / Deploy Dashboard (pull_request) Has been skipped
CI / Deploy Docs (pull_request) Has been skipped
CI / Deploy MCP (pull_request) Has been skipped
The CI check job runs cargo fmt --all --check; the new runtime modules and the
orchestrator wiring needed reformatting.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-17 09:44:36 +02:00
Sharang ParnerkarandClaude Fable 5 69bce2f07c feat(plc): DAST the provisioned WebVisu + enumerate exposed Modbus points (#183)
CI / Check (pull_request) Has been cancelled
CI / Detect Changes (pull_request) Has been cancelled
CI / Deploy Agent (pull_request) Has been cancelled
CI / Deploy Dashboard (pull_request) Has been cancelled
CI / Deploy Docs (pull_request) Has been cancelled
CI / Deploy MCP (pull_request) Has been cancelled
Completes the provision-and-test loop's dynamic coverage (sub-tasks 3 + 4):

- provision_and_test now returns a ProvisionOutcome { ics findings, DAST run }.
  After the Modbus probe it runs a bounded, best-effort DAST scan against the
  provisioned web endpoint (independently timed out so it can't consume the whole
  instance lifetime), and the orchestrator persists the DAST scan run + findings
  linked to the scan run. Kept as a plain data return so the whole run is
  portable to a remote execution backend. On the OpenPLC substrate the web
  endpoint is OpenPLC's own UI (fidelity caveat documented); the CODESYS-runtime
  follow-up raises this to a real WebVisu.

- ICS Modbus probe now enumerates the exposed process surface (read-only): a Read
  Coils and a Read Holding Registers of the first block. Coils and holding
  registers are read/write process points, so an exposed block is an
  unauthenticated *write* surface — reported as `ics-modbus-exposed-points`
  (High). Read-only to detect (we never write), so it is safe on the live probe
  too, not just the provisioned instance.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-17 09:35:08 +02:00
Sharang ParnerkarandClaude Fable 5 1ae6025286 feat(plc): ephemeral soft-PLC provisioning + program load (#183)
CI / Check (pull_request) Failing after 1m31s
CI / Detect Changes (pull_request) Has been skipped
CI / Deploy Agent (pull_request) Has been skipped
CI / Deploy Dashboard (pull_request) Has been skipped
CI / Deploy Docs (pull_request) Has been skipped
CI / Deploy MCP (pull_request) Has been skipped
Dynamic PLC testing without reaching the customer's device: when a PLC/SPS
target ships control logic but no reachable live URL, instantiate that logic
ourselves on a throwaway OpenPLC container in-cluster, load + start it, probe
the provisioned Modbus endpoint, and tear it down. No customer network access,
sandboxed, and reproducible.

This is the phase-1 foundation of epic #183 (OpenPLC substrate). It covers:

- provision: ephemeral container lifecycle (docker CLI). Resource-capped
  (memory/cpus/pids), hardened (no-new-privileges), labelled, joined to the
  agent's own network with no host port exposure, and swept by a stale reaper
  for anything a crashed run leaks. The `docker` argv is built by pure functions
  so it is unit-tested without a daemon.
- openplc: drives the OpenPLC web UI to load a program — login → upload →
  save → compile (MatIEC) → start_plc (which opens Modbus/TCP 502).
- runtime::provision_and_test: composes them under a hard deadline with
  guaranteed teardown on every path (success / error / timeout), then runs the
  existing ICS probe against the provisioned endpoint. extract_program picks the
  best loadable program (complete ST > largest ST > PLCopen XML).
- orchestrator: for a PlcSps target with control logic and no live URL, run
  provision-and-test after the static PLC scan. Gated by PlcRuntimeConfig
  (PLC_RUNTIME_ENABLED, default off — needs Docker access in the agent).

DAST-against-WebVisu and CODESYS-runtime fidelity are follow-ups.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-17 00:39:54 +02:00
sharang 1aba85b28e docs(plc): PLC Runtime Landscape reference + support watch-list (#185)
CI / Check (push) Has been skipped
CI / Detect Changes (push) Successful in 3s
CI / Deploy Agent (push) Has been skipped
CI / Deploy Dashboard (push) Has been skipped
CI / Deploy Docs (push) Successful in 49s
CI / Deploy MCP (push) Has been skipped
2026-07-16 22:05:39 +00:00
sharang c6a84b327f docs(plc): soft-PLC architecture + CODESYS/Yocto lifecycle diagram (#184)
CI / Check (push) Waiting to run
CI / Detect Changes (push) Waiting to run
CI / Deploy Agent (push) Blocked by required conditions
CI / Deploy Dashboard (push) Blocked by required conditions
CI / Deploy Docs (push) Blocked by required conditions
CI / Deploy MCP (push) Blocked by required conditions
2026-07-16 22:00:17 +00:00
17 changed files with 1442 additions and 7 deletions
+13
View File
@@ -34,6 +34,19 @@ SCAN_SCHEDULE=0 0 */6 * * *
CVE_MONITOR_SCHEDULE=0 0 0 * * *
GIT_CLONE_BASE_PATH=/tmp/compliance-scanner/repos
# Dynamic PLC testing — ephemeral soft-PLC provisioning (#183). Off unless
# enabled; requires the agent container to have Docker access (socket mount).
# When on, a PLC/SPS target with control logic but no reachable device gets its
# logic instantiated on a throwaway OpenPLC, probed, then torn down.
PLC_RUNTIME_ENABLED=0
PLC_RUNTIME_IMAGE=registry.meghsakha.com/openplc:latest
PLC_RUNTIME_NETWORK=certifai
PLC_RUNTIME_MEMORY=512m
PLC_RUNTIME_CPUS=0.5
PLC_RUNTIME_MAX_LIFETIME_SECS=180
PLC_RUNTIME_OPENPLC_USER=openplc
PLC_RUNTIME_OPENPLC_PASSWORD=openplc
# Dashboard
DASHBOARD_PORT=8080
AGENT_API_URL=http://localhost:3001
+1 -1
View File
@@ -23,7 +23,7 @@ tracing = "0.1"
tracing-subscriber = { version = "0.3", features = ["env-filter"] }
chrono = { version = "0.4", features = ["serde"] }
mongodb = { version = "3", features = ["rustls-tls", "compat-3-0-0"] }
reqwest = { version = "0.12", features = ["json", "rustls-tls", "multipart"], default-features = false }
reqwest = { version = "0.12", features = ["json", "rustls-tls", "multipart", "cookies"], default-features = false }
thiserror = "2"
sha2 = "0.10"
hex = "0.4"
+24
View File
@@ -1,3 +1,4 @@
use compliance_core::config::PlcRuntimeConfig;
use compliance_core::AgentConfig;
use secrecy::SecretString;
@@ -63,5 +64,28 @@ pub fn load_config() -> Result<AgentConfig, AgentError> {
pentest_imap_password: env_secret_opt("PENTEST_IMAP_PASSWORD"),
admin_api_token: env_secret_opt("ADMIN_API_TOKEN"),
tenant_registry_url: env_var_opt("TENANT_REGISTRY_URL"),
plc_runtime: load_plc_runtime_config(),
})
}
/// Build the ephemeral soft-PLC provisioning config from the environment,
/// falling back to [`PlcRuntimeConfig::default`] for any unset knob. Disabled
/// unless `PLC_RUNTIME_ENABLED` is truthy — it requires Docker access.
fn load_plc_runtime_config() -> PlcRuntimeConfig {
let d = PlcRuntimeConfig::default();
PlcRuntimeConfig {
enabled: env_var_opt("PLC_RUNTIME_ENABLED")
.map(|v| v == "1" || v.eq_ignore_ascii_case("true"))
.unwrap_or(d.enabled),
image: env_var_opt("PLC_RUNTIME_IMAGE").unwrap_or(d.image),
network: env_var_opt("PLC_RUNTIME_NETWORK").unwrap_or(d.network),
memory: env_var_opt("PLC_RUNTIME_MEMORY").unwrap_or(d.memory),
cpus: env_var_opt("PLC_RUNTIME_CPUS").unwrap_or(d.cpus),
max_lifetime_secs: env_var_opt("PLC_RUNTIME_MAX_LIFETIME_SECS")
.and_then(|v| v.parse().ok())
.unwrap_or(d.max_lifetime_secs),
openplc_user: env_var_opt("PLC_RUNTIME_OPENPLC_USER").unwrap_or(d.openplc_user),
openplc_password: env_secret_opt("PLC_RUNTIME_OPENPLC_PASSWORD")
.unwrap_or(d.openplc_password),
}
}
+1
View File
@@ -342,6 +342,7 @@ mod tests {
pentest_imap_password: None,
admin_api_token: None,
tenant_registry_url: None,
plc_runtime: compliance_core::PlcRuntimeConfig::default(),
}
}
+33
View File
@@ -103,6 +103,39 @@ async fn modbus_findings(host: &str, port: u16, repo_id: &str, budget: Duration)
);
findings.push(f);
}
// Exposed process points: coils / holding registers that a read enumerated
// and that, over unauthenticated Modbus/TCP, are also writable. This is the
// concrete attack surface behind the exposure — the live variables an
// attacker can overwrite. (Read-only to detect: we never write.)
let coils = probe.coils_readable.unwrap_or(0);
let registers = probe.holding_registers_readable.unwrap_or(0);
if coils > 0 || registers > 0 {
let fp = dedup::compute_fingerprint(&[repo_id, "ics-modbus-exposed-points", &target]);
let mut f = Finding::new(
repo_id.to_string(),
fp,
"ics-probe".to_string(),
ScanType::IcsProbe,
"Writable process points exposed over unauthenticated Modbus/TCP".to_string(),
format!(
"Reading the device at {target} enumerated {coils} coil(s) and {registers} \
holding register(s). Coils and holding registers are read/write process points \
in Modbus, so any host that can reach this port can not only read but overwrite \
live process state (force coils, change setpoints) without authentication."
),
Severity::High,
);
f.rule_id = Some("ics-modbus-exposed-points".to_string());
f.cwe = Some("CWE-306".to_string());
f.remediation = Some(
"Segment the Modbus/TCP port to a trusted control network; where the device \
supports it use Modbus/TLS or an authenticating protocol gateway; restrict which \
function codes and register ranges are reachable from outside the control zone."
.to_string(),
);
findings.push(f);
}
findings
}
+62 -5
View File
@@ -21,6 +21,13 @@ pub struct ModbusProbe {
pub speaks_modbus: bool,
/// Device identity, if disclosed via Read Device Identification (FC 43 / 14).
pub device: Option<DeviceId>,
/// Coils returned by a Read Coils of the first block, if that address range
/// exists. Coils are read/write process bits, so an exposed block is an
/// unauthenticated write surface on the live process.
pub coils_readable: Option<u16>,
/// Holding registers returned by a Read Holding Registers of the first block,
/// if that range exists. Holding registers are read/write process words.
pub holding_registers_readable: Option<u16>,
}
/// Vendor / product / revision from Read Device Identification.
@@ -31,8 +38,13 @@ pub struct DeviceId {
pub revision: Option<String>,
}
/// Probe a Modbus/TCP endpoint. Read-only: issues a Read Holding Registers and a
/// Read Device Identification request; never writes to the device.
/// How many coils / holding registers to request when enumerating the exposed
/// process surface. Read-only: a normal reply means the block exists and is,
/// over unauthenticated Modbus/TCP, also writable.
const ENUM_QTY: u16 = 16;
/// Probe a Modbus/TCP endpoint. Read-only: issues Read Holding Registers, Read
/// Coils, and Read Device Identification requests; never writes to the device.
pub async fn probe(host: &str, port: u16, budget: Duration) -> ModbusProbe {
let mut out = ModbusProbe::default();
let Ok(Ok(mut stream)) = timeout(budget, TcpStream::connect((host, port))).await else {
@@ -40,13 +52,28 @@ pub async fn probe(host: &str, port: u16, budget: Duration) -> ModbusProbe {
};
out.reachable = true;
// Read Holding Registers (FC 0x03), unit 1, addr 0, qty 1 — a benign read.
let rhr = [0x03u8, 0x00, 0x00, 0x00, 0x01];
// Read Holding Registers (FC 0x03), unit 1, addr 0 — a benign read that also
// enumerates the exposed register block.
let rhr = [0x03u8, 0x00, 0x00, (ENUM_QTY >> 8) as u8, ENUM_QTY as u8];
if let Some(resp) = txn(&mut stream, 1, &rhr, budget).await {
// A normal reply (0x03) or an exception (0x83) both prove it speaks Modbus.
if matches!(resp.first(), Some(0x03) | Some(0x83)) {
out.speaks_modbus = true;
}
if resp.first() == Some(&0x03) {
out.holding_registers_readable = Some(register_count_from_reply(&resp));
}
}
// Read Coils (FC 0x01), addr 0 — enumerates the exposed coil (bit) block.
let rc = [0x01u8, 0x00, 0x00, (ENUM_QTY >> 8) as u8, ENUM_QTY as u8];
if let Some(resp) = txn(&mut stream, 1, &rc, budget).await {
if matches!(resp.first(), Some(0x01) | Some(0x81)) {
out.speaks_modbus = true;
}
if resp.first() == Some(&0x01) {
out.coils_readable = Some(coil_count_from_reply(&resp));
}
}
// Read Device Identification (FC 0x2B / MEI 0x0E), basic (0x01), object 0.
@@ -60,6 +87,17 @@ pub async fn probe(host: &str, port: u16, budget: Duration) -> ModbusProbe {
out
}
/// Coils reported by a Read Coils reply `[0x01, byte_count, data…]` (8 per byte).
fn coil_count_from_reply(pdu: &[u8]) -> u16 {
pdu.get(1).map(|&b| u16::from(b) * 8).unwrap_or(0)
}
/// Registers reported by a Read Holding Registers reply `[0x03, byte_count,
/// data…]` (2 bytes per register).
fn register_count_from_reply(pdu: &[u8]) -> u16 {
pdu.get(1).map(|&b| u16::from(b) / 2).unwrap_or(0)
}
/// Send one Modbus PDU and return the response PDU (function code + data), or
/// `None` on timeout / malformed reply.
async fn txn(stream: &mut TcpStream, unit: u8, pdu: &[u8], budget: Duration) -> Option<Vec<u8>> {
@@ -154,7 +192,8 @@ mod tests {
break;
}
let reply_pdu: Vec<u8> = match pdu.first() {
Some(0x03) => vec![0x03, 0x02, 0x00, 0x00], // 1 register = 0
Some(0x03) => vec![0x03, 0x02, 0x00, 0x00], // 1 register (byte_count 2)
Some(0x01) => vec![0x01, 0x02, 0xFF, 0xFF], // 16 coils (byte_count 2)
Some(0x2B) if with_device => vec![
0x2B, 0x0E, 0x01, 0x81, 0x00, 0x00, 0x02, // 2 objects
0x00, 0x04, b'A', b'C', b'M', b'E', // vendor
@@ -185,6 +224,24 @@ mod tests {
assert_eq!(dev.product.as_deref(), Some("PLC"));
}
#[tokio::test]
async fn probe_enumerates_exposed_process_points() {
let addr = mock_server(false).await;
let p = probe(&addr.ip().to_string(), addr.port(), Duration::from_secs(2)).await;
assert!(p.speaks_modbus);
// The mock returns a 2-byte holding-register block (1 register) and a
// 2-byte coil block (16 coils).
assert_eq!(p.holding_registers_readable, Some(1));
assert_eq!(p.coils_readable, Some(16));
}
#[test]
fn reply_counts_decode_byte_counts() {
assert_eq!(register_count_from_reply(&[0x03, 0x08]), 4); // 8 bytes → 4 regs
assert_eq!(coil_count_from_reply(&[0x01, 0x03]), 24); // 3 bytes → 24 coils
assert_eq!(register_count_from_reply(&[0x03]), 0); // malformed → 0
}
#[tokio::test]
async fn probe_reports_unreachable_for_a_closed_port() {
// 127.0.0.1:1 is (almost certainly) closed.
@@ -404,6 +404,21 @@ impl PipelineOrchestrator {
let ics = plan.has(ScanType::IcsProbe);
if plc {
new_count += self.run_plc_scan(target, &target_id, scan_run_id).await?;
// Provision-and-test (#183): with the control logic but no reachable
// device, instantiate it on an ephemeral soft-PLC and probe that
// instead of the customer's OT network. Opt-in (needs Docker) and only
// when there is no live URL to probe directly. Never fails the scan.
if self.config.plc_runtime.enabled && target.live_url().is_none() {
match self
.run_provisioned_plc_test(target, &target_id, scan_run_id)
.await
{
Ok(n) => new_count += n,
Err(e) => {
tracing::warn!(target_id = %target_id, error = %e, "provision-and-test failed")
}
}
}
}
if ics {
new_count += self.run_ics_probe(target, &target_id, scan_run_id).await?;
@@ -540,6 +555,98 @@ impl PipelineOrchestrator {
Ok(new_count)
}
/// Provision-and-test (#183): instantiate the target's control logic on an
/// ephemeral soft-PLC (OpenPLC), start it, probe the provisioned Modbus
/// endpoint, and tear the instance down. Used when a PLC/SPS target has the
/// control logic but no reachable live device to probe directly. Guarded by
/// `plc_runtime.enabled` (needs Docker); persists the same [`ScanType::IcsProbe`]
/// findings as a live probe.
async fn run_provisioned_plc_test(
&self,
target: &OnboardedTarget,
target_id: &str,
scan_run_id: &str,
) -> Result<u32, AgentError> {
self.update_phase(scan_run_id, "plc_provision").await;
// Locate a loadable control-logic program among the PLC-source artifacts
// (same selection as the static PLC scan: dedicated PLC projects plus code
// artifacts holding PLCopen XML / ST exports).
let ctx = crate::ingest::IngestContext::from_config(&self.config, target_id);
let ingest_set = crate::ingest::ingest_all(target, &ctx)?;
let program = target
.artifacts
.iter()
.filter(|a| {
matches!(
a.kind,
ArtifactKind::PlcProject | ArtifactKind::GitRepo | ArtifactKind::SourceArchive
)
})
.find_map(|a| {
let path = ingest_set
.get(&a.id)
.and_then(|ia| ia.working_path.clone())?;
crate::pipeline::plc::runtime::extract_program(&path)
});
let Some(program) = program else {
tracing::info!(
target_id,
"provision-and-test: no loadable control-logic program"
);
return Ok(0);
};
let http = crate::pipeline::plc::runtime::http_client()?;
let provisioner =
crate::pipeline::plc::runtime::DockerSoftPlc::new(self.config.plc_runtime.clone());
let outcome = crate::pipeline::plc::runtime::provision_and_test(
&provisioner,
&http,
&self.config.plc_runtime,
&program,
target_id,
)
.await?;
tracing::info!(
target_id,
found = outcome.findings.len(),
dast = outcome.dast.is_some(),
"provision-and-test complete"
);
let mut new_count = 0u32;
for mut finding in outcome.findings {
finding.scan_run_id = Some(scan_run_id.to_string());
if self
.db
.findings()
.find_one(doc! { "fingerprint": &finding.fingerprint })
.await?
.is_none()
{
self.db.findings().insert_one(&finding).await?;
new_count += 1;
}
}
// Persist the DAST scan of the provisioned web endpoint, linked to this
// scan run (mirrors `maybe_trigger_dast`).
if let Some(dast) = outcome.dast {
let mut scan_run = dast.scan_run;
scan_run.sast_scan_run_id = Some(scan_run_id.to_string());
if let Err(e) = self.db.dast_scan_runs().insert_one(&scan_run).await {
tracing::warn!(target_id, error = %e, "failed to store provisioned DAST scan run");
}
for finding in &dast.findings {
if let Err(e) = self.db.dast_findings().insert_one(finding).await {
tracing::warn!(target_id, error = %e, "failed to store provisioned DAST finding");
}
}
}
Ok(new_count)
}
/// Probe a running PLC/SPS device over industrial protocols (Modbus/TCP, …)
/// and persist findings for exposed / unauthenticated control access. The
/// probe is read-only; it targets the Modbus port of the target's live URL.
+1
View File
@@ -9,6 +9,7 @@ pub mod lexer;
pub mod parser;
pub mod plcopen;
pub mod rules;
pub mod runtime;
pub mod sbom;
use std::path::Path;
@@ -0,0 +1,423 @@
//! Dynamic PLC testing via an ephemeral soft-PLC (#183).
//!
//! When a PLC/SPS target ships control logic but no reachable live device, the
//! agent instantiates that logic itself instead of trying to reach the customer's
//! OT network: it provisions a throwaway soft-PLC (OpenPLC) container in-cluster,
//! loads the program, starts the runtime, probes it over industrial protocols,
//! then tears the instance down. No customer network access, sandboxed, and
//! reproducible — destructive tests become safe because the target is ours.
//!
//! - [`provision`] owns the container lifecycle (sub-task 1 + 5).
//! - [`openplc`] loads the program into the running instance (sub-task 2).
//! - [`provision_and_test`] composes them with a hard deadline and guaranteed
//! teardown, and runs the ICS probe against the provisioned endpoint.
pub mod openplc;
pub mod provision;
use std::path::Path;
use std::time::Duration;
use secrecy::ExposeSecret;
use compliance_core::models::dast::{DastFinding, DastScanRun, DastTarget, DastTargetType};
use compliance_core::models::Finding;
use compliance_core::PlcRuntimeConfig;
use crate::error::AgentError;
pub use provision::{DockerSoftPlc, ProvisionedRuntime, SoftPlc};
/// The result of a DAST scan against a provisioned web endpoint.
#[derive(Debug)]
pub struct DastRunResult {
/// The scan-run record (linked to the onboarded target).
pub scan_run: DastScanRun,
/// The DAST findings.
pub findings: Vec<DastFinding>,
}
/// Everything a provision-and-test run produced: the ICS-probe findings plus, if
/// it ran, the DAST scan of the provisioned web endpoint. The caller persists
/// both — keeping this a plain data return means the whole run is portable to a
/// remote execution backend that just hands the results back.
#[derive(Debug, Default)]
pub struct ProvisionOutcome {
/// ICS-probe findings from the provisioned Modbus endpoint.
pub findings: Vec<Finding>,
/// DAST scan of the provisioned web endpoint, if it ran.
pub dast: Option<DastRunResult>,
}
/// A control-logic program ready to load into a soft-PLC: the source text plus a
/// cosmetic file name (OpenPLC re-stores it under its own name).
#[derive(Debug, Clone)]
pub struct PlcProgram {
/// The original file name (for the upload form; OpenPLC renames on storage).
pub file_name: String,
/// The program source — Structured Text or PLCopen XML.
pub source: String,
}
/// A cookie-aware HTTP client for the OpenPLC web UI. A fresh client per scan
/// isolates the OpenPLC session (its Flask login cookie) from every other scan.
pub fn http_client() -> Result<reqwest::Client, AgentError> {
reqwest::Client::builder()
.cookie_store(true)
.timeout(Duration::from_secs(30))
.build()
.map_err(AgentError::Http)
}
/// Pick the control-logic program to run from an ingested PLC source tree.
///
/// OpenPLC runs one program, so we choose the best single candidate: a complete
/// Structured Text program (one carrying a `CONFIGURATION` block) is ideal;
/// failing that the largest ST file; failing that a PLCopen XML export. Returns
/// `None` when the tree holds no loadable control logic.
pub fn extract_program(root: &Path) -> Option<PlcProgram> {
let mut st: Vec<(String, String)> = Vec::new();
let mut xml: Vec<(String, String)> = Vec::new();
for entry in walkdir::WalkDir::new(root)
.into_iter()
.filter_map(Result::ok)
{
if !entry.file_type().is_file() {
continue;
}
let path = entry.path();
let ext = path
.extension()
.and_then(|e| e.to_str())
.unwrap_or("")
.to_ascii_lowercase();
let is_st = matches!(ext.as_str(), "st" | "iecst" | "scl" | "exp" | "il");
let is_xml = matches!(ext.as_str(), "xml" | "plcopen" | "project");
if !is_st && !is_xml {
continue;
}
let Ok(content) = std::fs::read_to_string(path) else {
continue;
};
let name = path
.file_name()
.and_then(|n| n.to_str())
.unwrap_or("program")
.to_string();
if is_st {
st.push((name, content));
} else if looks_like_plcopen(&content) {
xml.push((name, content));
}
}
if let Some((name, source)) = st.iter().find(|(_, c)| has_configuration(c)) {
return Some(PlcProgram {
file_name: name.clone(),
source: source.clone(),
});
}
if let Some((name, source)) = st.iter().max_by_key(|(_, c)| c.len()) {
return Some(PlcProgram {
file_name: name.clone(),
source: source.clone(),
});
}
xml.into_iter()
.max_by_key(|(_, c)| c.len())
.map(|(file_name, source)| PlcProgram { file_name, source })
}
/// Whether an ST source is a complete, runnable program (has a `CONFIGURATION`).
fn has_configuration(source: &str) -> bool {
source.to_ascii_uppercase().contains("CONFIGURATION")
}
/// Whether an XML file looks like a PLCopen project export.
fn looks_like_plcopen(source: &str) -> bool {
let lower = source.to_ascii_lowercase();
lower.contains("<project") || lower.contains("plcopen")
}
/// Provision an ephemeral soft-PLC, load `program`, start it, probe it over
/// industrial protocols, and tear it down. Returns the ICS-probe findings.
///
/// Teardown is guaranteed: the load/probe work runs under a hard deadline
/// (`max_lifetime_secs`) and the instance is removed afterwards on every path —
/// success, error, or deadline expiry.
pub async fn provision_and_test<P: SoftPlc>(
provisioner: &P,
http: &reqwest::Client,
cfg: &PlcRuntimeConfig,
program: &PlcProgram,
target_id: &str,
) -> Result<ProvisionOutcome, AgentError> {
let handle = provisioner.provision(target_id).await?;
tracing::info!(
target_id,
instance = %handle.name,
modbus = %handle.modbus_endpoint,
"provisioned ephemeral soft-PLC"
);
let deadline = Duration::from_secs(cfg.max_lifetime_secs);
let result = tokio::time::timeout(
deadline,
run_dynamic_test(http, cfg, program, target_id, &handle),
)
.await;
// Guaranteed teardown — runs on success, error, and deadline expiry. The
// inner future is panic-free (the workspace lint bans unwrap/expect), so no
// unwind can skip this; a container leaked by an agent *crash* is swept by
// the next run's stale reaper.
provisioner.teardown(&handle).await;
match result {
Ok(inner) => inner,
Err(_) => {
tracing::warn!(
target_id,
instance = %handle.name,
"provision-and-test hit the lifetime deadline; torn down"
);
Ok(ProvisionOutcome::default())
}
}
}
/// The load → start → probe → DAST body, run under the caller's deadline.
async fn run_dynamic_test(
http: &reqwest::Client,
cfg: &PlcRuntimeConfig,
program: &PlcProgram,
target_id: &str,
handle: &ProvisionedRuntime,
) -> Result<ProvisionOutcome, AgentError> {
let ready_budget = Duration::from_secs((cfg.max_lifetime_secs / 3).clamp(10, 60));
openplc::wait_ready(http, &handle.webvisu_url, ready_budget).await?;
let compile_budget = Duration::from_secs((cfg.max_lifetime_secs / 2).clamp(20, 120));
openplc::load_and_start(
http,
&handle.webvisu_url,
&cfg.openplc_user,
cfg.openplc_password.expose_secret(),
program,
compile_budget,
)
.await?;
// Give the runtime a moment to open the Modbus/TCP server before probing.
tokio::time::sleep(Duration::from_secs(3)).await;
let probe_budget = Duration::from_secs(5);
let findings =
crate::pipeline::ics::probe_target(&handle.modbus_endpoint, target_id, probe_budget).await;
tracing::info!(
target_id,
instance = %handle.name,
found = findings.len(),
"provision-and-test probe complete"
);
// DAST the provisioned web endpoint (independently bounded so it can't eat
// the whole lifetime). On the OpenPLC substrate this is OpenPLC's own web UI,
// not a customer HMI — the CODESYS-runtime follow-up raises the fidelity —
// but it proves the deploy→run→probe→DAST loop end to end.
let dast_budget = Duration::from_secs((cfg.max_lifetime_secs / 2).clamp(20, 120));
let dast = match tokio::time::timeout(dast_budget, run_webvisu_dast(handle, target_id)).await {
Ok(d) => d,
Err(_) => {
tracing::warn!(target_id, instance = %handle.name, "provision-and-test DAST timed out");
None
}
};
Ok(ProvisionOutcome { findings, dast })
}
/// Run a bounded DAST scan against the provisioned web endpoint and tag the
/// results with our target id. Best-effort — a DAST failure never fails the run.
async fn run_webvisu_dast(handle: &ProvisionedRuntime, target_id: &str) -> Option<DastRunResult> {
let mut dt = DastTarget::new(
"provisioned-webvisu".to_string(),
handle.webvisu_url.clone(),
DastTargetType::WebApp,
);
dt.repo_id = Some(target_id.to_string());
dt.max_crawl_depth = 2; // shallow — the instance is ephemeral
let orchestrator = compliance_dast::DastOrchestrator::new(100);
match orchestrator.run_scan(&dt, Vec::new()).await {
Ok((mut scan_run, mut findings)) => {
scan_run.target_id = target_id.to_string();
for f in &mut findings {
f.target_id = target_id.to_string();
}
tracing::info!(
target_id,
instance = %handle.name,
dast_findings = findings.len(),
"provision-and-test DAST complete"
);
Some(DastRunResult { scan_run, findings })
}
Err(e) => {
tracing::warn!(target_id, instance = %handle.name, error = %e, "provision-and-test DAST failed");
None
}
}
}
#[cfg(test)]
#[allow(clippy::expect_used, clippy::unwrap_used)]
mod tests {
use super::*;
use std::sync::atomic::{AtomicUsize, Ordering};
use std::sync::Arc;
/// A scratch dir removed on drop.
struct Scratch(std::path::PathBuf);
impl Scratch {
fn new() -> Self {
let p = std::env::temp_dir().join(format!("cs-plc-rt-{}", uuid::Uuid::new_v4()));
std::fs::create_dir_all(&p).expect("mkdir");
Self(p)
}
}
impl Drop for Scratch {
fn drop(&mut self) {
let _ = std::fs::remove_dir_all(&self.0);
}
}
#[test]
fn extract_prefers_a_complete_st_program() {
let s = Scratch::new();
std::fs::write(s.0.join("fragment.st"), "PROGRAM P\nEND_PROGRAM\n").expect("w");
std::fs::write(
s.0.join("full.st"),
"PROGRAM Main\nEND_PROGRAM\nCONFIGURATION Config0\n RESOURCE R\nEND_CONFIGURATION\n",
)
.expect("w");
let prog = extract_program(&s.0).expect("program");
assert_eq!(prog.file_name, "full.st");
assert!(prog.source.contains("CONFIGURATION"));
}
#[test]
fn extract_falls_back_to_largest_st_then_plcopen() {
let s = Scratch::new();
std::fs::write(s.0.join("small.st"), "PROGRAM A\nEND_PROGRAM\n").expect("w");
std::fs::write(
s.0.join("big.st"),
"PROGRAM B\nVAR x : INT; y : INT; z : INT; END_VAR\nEND_PROGRAM\n",
)
.expect("w");
let prog = extract_program(&s.0).expect("program");
assert_eq!(
prog.file_name, "big.st",
"largest ST wins when none complete"
);
// Only a PLCopen XML present.
let s2 = Scratch::new();
std::fs::write(
s2.0.join("proj.xml"),
"<?xml version='1.0'?><project xmlns='http://www.plcopen.org/xml/tc6_0201'><pou/></project>",
)
.expect("w");
let prog2 = extract_program(&s2.0).expect("program");
assert_eq!(prog2.file_name, "proj.xml");
}
#[test]
fn extract_returns_none_without_control_logic() {
let s = Scratch::new();
std::fs::write(s.0.join("readme.md"), "# not a plc program").expect("w");
std::fs::write(s.0.join("data.xml"), "<config><db/></config>").expect("w");
assert!(extract_program(&s.0).is_none());
}
/// A fake provisioner recording provision/teardown calls, for lifecycle tests.
struct FakeSoftPlc {
provisions: Arc<AtomicUsize>,
teardowns: Arc<AtomicUsize>,
fail_provision: bool,
}
impl SoftPlc for FakeSoftPlc {
async fn provision(&self, _target_id: &str) -> Result<ProvisionedRuntime, AgentError> {
self.provisions.fetch_add(1, Ordering::SeqCst);
if self.fail_provision {
return Err(AgentError::Other("provision failed".into()));
}
// Unreachable address so run_dynamic_test blocks on readiness until the
// deadline fires — exercising the teardown-on-deadline path.
Ok(ProvisionedRuntime {
name: "fake-plc".into(),
modbus_endpoint: "fake-plc:502".into(),
webvisu_url: "http://fake-plc.invalid:8080".into(),
})
}
async fn teardown(&self, _handle: &ProvisionedRuntime) {
self.teardowns.fetch_add(1, Ordering::SeqCst);
}
}
fn short_cfg() -> PlcRuntimeConfig {
PlcRuntimeConfig {
enabled: true,
max_lifetime_secs: 1, // keep the deadline path fast
..PlcRuntimeConfig::default()
}
}
#[tokio::test]
async fn teardown_runs_even_when_the_test_never_completes() {
let provisions = Arc::new(AtomicUsize::new(0));
let teardowns = Arc::new(AtomicUsize::new(0));
let fake = FakeSoftPlc {
provisions: provisions.clone(),
teardowns: teardowns.clone(),
fail_provision: false,
};
let http = http_client().expect("client");
let prog = PlcProgram {
file_name: "p.st".into(),
source: "PROGRAM P\nEND_PROGRAM\n".into(),
};
let out = provision_and_test(&fake, &http, &short_cfg(), &prog, "t1")
.await
.expect("ok on deadline");
assert!(out.findings.is_empty(), "deadline path yields no findings");
assert!(out.dast.is_none(), "deadline path runs no DAST");
assert_eq!(provisions.load(Ordering::SeqCst), 1);
assert_eq!(teardowns.load(Ordering::SeqCst), 1, "teardown must run");
}
#[tokio::test]
async fn provision_failure_propagates_and_skips_teardown() {
let provisions = Arc::new(AtomicUsize::new(0));
let teardowns = Arc::new(AtomicUsize::new(0));
let fake = FakeSoftPlc {
provisions: provisions.clone(),
teardowns: teardowns.clone(),
fail_provision: true,
};
let http = http_client().expect("client");
let prog = PlcProgram {
file_name: "p.st".into(),
source: String::new(),
};
let err = provision_and_test(&fake, &http, &short_cfg(), &prog, "t1").await;
assert!(err.is_err(), "provision failure propagates");
assert_eq!(provisions.load(Ordering::SeqCst), 1);
assert_eq!(
teardowns.load(Ordering::SeqCst),
0,
"nothing to tear down when provisioning failed"
);
}
}
@@ -0,0 +1,254 @@
//! Loading a control-logic program into a provisioned OpenPLC (#183, sub-task 2).
//!
//! Drives the OpenPLC v3 web UI over HTTP to turn a static control-logic artifact
//! into a *running* PLC: log in, upload the program, save it, compile it (MatIEC),
//! and start the runtime — at which point OpenPLC opens its Modbus/TCP server on
//! 502 and the ICS probe has something to talk to. The endpoint sequence mirrors
//! the OpenPLC web UI: `POST /login` → `POST /upload-program` (which hands back a
//! server-assigned `prog_file`) → `POST /upload-program-action` →
//! `GET /compile-program?file=<prog_file>` → `GET /start_plc`.
use std::time::Duration;
use crate::error::AgentError;
use super::PlcProgram;
/// Default OpenPLC program name/description recorded in its UI.
const PROG_NAME: &str = "certifai-provisioned";
const PROG_DESCR: &str = "Uploaded by the Certifai provision-and-test scan";
/// Poll interval while waiting for readiness / compilation.
const POLL_INTERVAL: Duration = Duration::from_secs(2);
/// Wait until the OpenPLC web UI answers (any non-5xx reply to `/login`), or the
/// budget elapses. A freshly-started container needs a few seconds to boot.
pub async fn wait_ready(
http: &reqwest::Client,
base_url: &str,
budget: Duration,
) -> Result<(), AgentError> {
let login = format!("{base_url}/login");
let outcome = tokio::time::timeout(budget, async {
loop {
if let Ok(resp) = http.get(&login).send().await {
if !resp.status().is_server_error() {
return;
}
}
tokio::time::sleep(POLL_INTERVAL).await;
}
})
.await;
outcome.map_err(|_| AgentError::Other(format!("OpenPLC at {base_url} did not become ready")))
}
/// Log in, upload the program, compile it, and start the runtime. On success the
/// OpenPLC Modbus/TCP server is listening on 502.
pub async fn load_and_start(
http: &reqwest::Client,
base_url: &str,
user: &str,
password: &str,
program: &PlcProgram,
compile_budget: Duration,
) -> Result<(), AgentError> {
login(http, base_url, user, password).await?;
let prog_file = upload_program(http, base_url, program).await?;
save_program(http, base_url, &prog_file).await?;
compile(http, base_url, &prog_file, compile_budget).await?;
start(http, base_url).await?;
Ok(())
}
/// `POST /login` — establishes the session cookie (the client must have a cookie
/// store; see the provision-and-test entry point).
async fn login(
http: &reqwest::Client,
base_url: &str,
user: &str,
password: &str,
) -> Result<(), AgentError> {
let resp = http
.post(format!("{base_url}/login"))
.form(&[("username", user), ("password", password)])
.send()
.await?;
if resp.status().is_server_error() {
return Err(AgentError::Other(format!(
"OpenPLC login failed: HTTP {}",
resp.status()
)));
}
Ok(())
}
/// `POST /upload-program` (multipart `file`) — OpenPLC stores the program under a
/// server-assigned name and returns it in a hidden `prog_file` form field, which
/// we parse out for the follow-up save/compile steps.
async fn upload_program(
http: &reqwest::Client,
base_url: &str,
program: &PlcProgram,
) -> Result<String, AgentError> {
let part = reqwest::multipart::Part::text(program.source.clone())
.file_name(program.file_name.clone())
.mime_str("application/octet-stream")?;
let form = reqwest::multipart::Form::new().part("file", part);
let resp = http
.post(format!("{base_url}/upload-program"))
.multipart(form)
.send()
.await?;
let html = resp.text().await?;
parse_prog_file(&html).ok_or_else(|| {
AgentError::Other("OpenPLC upload did not return a prog_file handle".to_string())
})
}
/// `POST /upload-program-action` — records the uploaded program in OpenPLC's
/// program list. `epoch_time` must be close to the server's clock (OpenPLC
/// rejects stale timestamps), so we send the current time.
async fn save_program(
http: &reqwest::Client,
base_url: &str,
prog_file: &str,
) -> Result<(), AgentError> {
let epoch = std::time::SystemTime::now()
.duration_since(std::time::UNIX_EPOCH)
.map(|d| d.as_secs())
.unwrap_or(0)
.to_string();
let resp = http
.post(format!("{base_url}/upload-program-action"))
.form(&[
("prog_name", PROG_NAME),
("prog_descr", PROG_DESCR),
("prog_file", prog_file),
("epoch_time", &epoch),
])
.send()
.await?;
if resp.status().is_server_error() {
return Err(AgentError::Other(format!(
"OpenPLC save-program failed: HTTP {}",
resp.status()
)));
}
Ok(())
}
/// `GET /compile-program?file=<prog_file>` then poll `/compilation-logs` until
/// MatIEC reports it finished (or the budget elapses). Errors if compilation
/// finishes with errors — a program that won't compile can't be started.
async fn compile(
http: &reqwest::Client,
base_url: &str,
prog_file: &str,
budget: Duration,
) -> Result<(), AgentError> {
http.get(format!("{base_url}/compile-program"))
.query(&[("file", prog_file)])
.send()
.await?;
let logs_url = format!("{base_url}/compilation-logs");
let outcome = tokio::time::timeout(budget, async {
loop {
if let Ok(resp) = http.get(&logs_url).send().await {
if let Ok(text) = resp.text().await {
if compilation_finished(&text) {
return !compilation_failed(&text);
}
}
}
tokio::time::sleep(POLL_INTERVAL).await;
}
})
.await;
match outcome {
Ok(true) => Ok(()),
Ok(false) => Err(AgentError::Other(
"OpenPLC compilation finished with errors".to_string(),
)),
Err(_) => Err(AgentError::Other(
"OpenPLC compilation did not finish in time".to_string(),
)),
}
}
/// `GET /start_plc` — starts the runtime, opening Modbus/TCP on 502.
async fn start(http: &reqwest::Client, base_url: &str) -> Result<(), AgentError> {
let resp = http.get(format!("{base_url}/start_plc")).send().await?;
if resp.status().is_server_error() {
return Err(AgentError::Other(format!(
"OpenPLC start_plc failed: HTTP {}",
resp.status()
)));
}
Ok(())
}
/// Extract the server-assigned `prog_file` from the `/upload-program` response,
/// which embeds it in a hidden input. Attribute order varies, so accept both
/// `value=… name='prog_file'` and `name='prog_file' … value=…`.
fn parse_prog_file(html: &str) -> Option<String> {
// The OpenPLC template renders `value='<name>.st' id='prog_file'
// name='prog_file'`. Match the value bound to that input, either order.
let value_then_name =
regex::Regex::new(r#"(?is)value=['"]([^'"]+)['"][^>]*name=['"]prog_file['"]"#).ok()?;
if let Some(c) = value_then_name.captures(html) {
return c.get(1).map(|m| m.as_str().to_string());
}
let name_then_value =
regex::Regex::new(r#"(?is)name=['"]prog_file['"][^>]*value=['"]([^'"]+)['"]"#).ok()?;
name_then_value
.captures(html)
.and_then(|c| c.get(1))
.map(|m| m.as_str().to_string())
}
/// Whether the MatIEC compilation log shows the build has finished (either way).
fn compilation_finished(log: &str) -> bool {
log.contains("Compilation finished")
}
/// Whether a finished compilation ended in failure.
fn compilation_failed(log: &str) -> bool {
log.contains("Compilation finished with errors")
}
#[cfg(test)]
#[allow(clippy::expect_used, clippy::unwrap_used)]
mod tests {
use super::*;
#[test]
fn parses_prog_file_value_then_name() {
let html = "<form><input type='hidden' value='483927.st' id='prog_file' \
name='prog_file'/></form>";
assert_eq!(parse_prog_file(html), Some("483927.st".to_string()));
}
#[test]
fn parses_prog_file_name_then_value() {
let html = r#"<input name="prog_file" id="prog_file" value="12.st" />"#;
assert_eq!(parse_prog_file(html), Some("12.st".to_string()));
}
#[test]
fn parse_prog_file_none_when_absent() {
assert_eq!(parse_prog_file("<html>no form here</html>"), None);
}
#[test]
fn compilation_predicates() {
assert!(!compilation_finished("Compiling..."));
assert!(compilation_finished(
"...\nCompilation finished successfully!\n"
));
assert!(compilation_finished("Compilation finished with errors!"));
assert!(compilation_failed("Compilation finished with errors!"));
assert!(!compilation_failed("Compilation finished successfully!"));
}
}
@@ -0,0 +1,306 @@
//! Ephemeral soft-PLC container lifecycle (#183, sub-task 1 + 5).
//!
//! Provisions a throwaway OpenPLC container per scan, isolated on the agent's own
//! Docker network with hard resource caps and **no host port exposure**, then
//! guarantees teardown. The container is reachable in-cluster only, by its name
//! (the shared user-defined network's embedded DNS resolves it); it is never
//! published to the host.
//!
//! The `docker` argv is produced by pure functions so provisioning is unit-tested
//! without a Docker daemon — only the thin [`run_docker`] wrapper touches the OS.
//! It requires the agent's runtime to have Docker access (a socket mount), which
//! is why the whole path is gated behind [`PlcRuntimeConfig::enabled`].
use std::time::{SystemTime, UNIX_EPOCH};
use compliance_core::PlcRuntimeConfig;
use crate::error::AgentError;
/// The Modbus/TCP port an OpenPLC instance opens once a program is running.
const MODBUS_PORT: u16 = 502;
/// The OpenPLC web-UI / WebVisu port.
const WEBVISU_PORT: u16 = 8080;
/// Label key marking a container as an ephemeral PLC runtime we own.
const OWNER_LABEL_KEY: &str = "certifai.ephemeral";
/// Label value for our ephemeral PLC runtimes.
const OWNER_LABEL_VALUE: &str = "plc-runtime";
/// A running ephemeral soft-PLC instance. Reachable in-cluster by `name`.
#[derive(Debug, Clone)]
pub struct ProvisionedRuntime {
/// The container name — also its in-network DNS alias.
pub name: String,
/// `name:502` — the Modbus/TCP endpoint the ICS probe targets.
pub modbus_endpoint: String,
/// `http://name:8080` — the WebVisu / OpenPLC web UI.
pub webvisu_url: String,
}
/// A source of ephemeral soft-PLC instances. Abstracted so the provision-and-test
/// orchestration is unit-testable with a fake that never touches Docker.
pub trait SoftPlc {
/// Start a fresh instance for a target and return its handle.
fn provision(
&self,
target_id: &str,
) -> impl std::future::Future<Output = Result<ProvisionedRuntime, AgentError>> + Send;
/// Tear an instance down. Best-effort and idempotent — never fails the scan.
fn teardown(&self, handle: &ProvisionedRuntime)
-> impl std::future::Future<Output = ()> + Send;
}
/// Provisions OpenPLC instances by shelling out to the Docker CLI.
pub struct DockerSoftPlc {
cfg: PlcRuntimeConfig,
}
impl DockerSoftPlc {
/// Build a provisioner from the PLC-runtime config.
pub fn new(cfg: PlcRuntimeConfig) -> Self {
Self { cfg }
}
}
impl SoftPlc for DockerSoftPlc {
async fn provision(&self, target_id: &str) -> Result<ProvisionedRuntime, AgentError> {
// Best-effort sweep of any container leaked by a crashed earlier run
// before we add another. Only removes instances past their max lifetime,
// so it can never disturb a concurrent run.
reap_stale(&self.cfg, now_epoch()).await;
let name = instance_name(target_id, now_epoch(), &random_suffix());
let args = run_args(&self.cfg, &name, target_id);
let out = run_docker(&args).await?;
if !out.status.success() {
return Err(AgentError::Other(format!(
"docker run for soft-PLC {name} failed: {}",
String::from_utf8_lossy(&out.stderr).trim()
)));
}
Ok(ProvisionedRuntime {
modbus_endpoint: format!("{name}:{MODBUS_PORT}"),
webvisu_url: format!("http://{name}:{WEBVISU_PORT}"),
name,
})
}
async fn teardown(&self, handle: &ProvisionedRuntime) {
match run_docker(&rm_args(&handle.name)).await {
Ok(out) if out.status.success() => {
tracing::info!(instance = %handle.name, "soft-PLC instance torn down");
}
Ok(out) => tracing::warn!(
instance = %handle.name,
"soft-PLC teardown non-zero exit: {}",
String::from_utf8_lossy(&out.stderr).trim()
),
Err(e) => {
tracing::warn!(instance = %handle.name, error = %e, "soft-PLC teardown failed")
}
}
}
}
/// Seconds since the Unix epoch (0 if the clock is before 1970, which never
/// happens in practice).
fn now_epoch() -> u64 {
SystemTime::now()
.duration_since(UNIX_EPOCH)
.map(|d| d.as_secs())
.unwrap_or(0)
}
/// A short random, docker-name-safe suffix.
fn random_suffix() -> String {
uuid::Uuid::new_v4().simple().to_string()
}
/// A unique, docker-safe container name that encodes the creation epoch (for the
/// stale reaper) and the target it belongs to. Shape:
/// `certifai-plc-<epoch>-<target12>-<rand6>`.
fn instance_name(target_id: &str, epoch: u64, rand: &str) -> String {
let short: String = target_id
.chars()
.filter(char::is_ascii_alphanumeric)
.take(12)
.collect();
let rand: String = rand
.chars()
.filter(char::is_ascii_alphanumeric)
.take(6)
.collect();
format!("certifai-plc-{epoch}-{short}-{rand}")
}
/// The creation epoch encoded in an instance name, if it is one of ours.
fn parse_epoch(name: &str) -> Option<u64> {
name.strip_prefix("certifai-plc-")?
.split('-')
.next()?
.parse()
.ok()
}
/// The `docker run` argv for an ephemeral soft-PLC: detached, joined to the
/// agent's network, resource-capped, hardened, labelled for reaping, and — by
/// omitting any `-p` — never published to the host.
fn run_args(cfg: &PlcRuntimeConfig, name: &str, target_id: &str) -> Vec<String> {
vec![
"run".into(),
"-d".into(),
"--name".into(),
name.into(),
"--network".into(),
cfg.network.clone(),
"--memory".into(),
cfg.memory.clone(),
"--cpus".into(),
cfg.cpus.clone(),
"--pids-limit".into(),
"512".into(),
"--security-opt".into(),
"no-new-privileges".into(),
"--stop-timeout".into(),
"5".into(),
"--label".into(),
format!("{OWNER_LABEL_KEY}={OWNER_LABEL_VALUE}"),
"--label".into(),
format!("certifai.target={target_id}"),
cfg.image.clone(),
]
}
/// The `docker rm -f` argv that stops and removes an instance.
fn rm_args(name: &str) -> Vec<String> {
vec!["rm".into(), "-f".into(), name.into()]
}
/// The `docker ps` argv listing the names of every ephemeral PLC container we own.
fn reap_list_args() -> Vec<String> {
vec![
"ps".into(),
"-a".into(),
"--filter".into(),
format!("label={OWNER_LABEL_KEY}={OWNER_LABEL_VALUE}"),
"--format".into(),
"{{.Names}}".into(),
]
}
/// Remove any ephemeral PLC container older than twice the configured max
/// lifetime — i.e. one a crashed run leaked. The generous threshold guarantees a
/// container from a *live* run (still within its own deadline) is never swept.
/// Best-effort: any Docker error (e.g. no daemon) is ignored.
async fn reap_stale(cfg: &PlcRuntimeConfig, now: u64) {
let cutoff = cfg.max_lifetime_secs.saturating_mul(2);
let Ok(out) = run_docker(&reap_list_args()).await else {
return;
};
if !out.status.success() {
return;
}
let names = String::from_utf8_lossy(&out.stdout);
for name in names.lines().map(str::trim).filter(|n| !n.is_empty()) {
let Some(epoch) = parse_epoch(name) else {
continue;
};
if now.saturating_sub(epoch) > cutoff {
tracing::warn!(instance = %name, "reaping stale soft-PLC instance");
let _ = run_docker(&rm_args(name)).await;
}
}
}
/// Run a `docker` subcommand, capturing its output.
async fn run_docker(args: &[String]) -> Result<std::process::Output, AgentError> {
tokio::process::Command::new("docker")
.args(args)
.output()
.await
.map_err(AgentError::Io)
}
#[cfg(test)]
#[allow(clippy::expect_used, clippy::unwrap_used)]
mod tests {
use super::*;
fn cfg() -> PlcRuntimeConfig {
PlcRuntimeConfig {
enabled: true,
image: "registry.example.com/openplc:latest".into(),
network: "certifai".into(),
memory: "512m".into(),
cpus: "0.5".into(),
max_lifetime_secs: 180,
..PlcRuntimeConfig::default()
}
}
#[test]
fn instance_name_is_unique_docker_safe_and_reaper_parseable() {
let a = instance_name("64f0aabbccddeeff00112233", 1_700_000_000, "abcdef123456");
assert_eq!(a, "certifai-plc-1700000000-64f0aabbccdd-abcdef");
assert_eq!(parse_epoch(&a), Some(1_700_000_000));
// Docker names: only [A-Za-z0-9_.-].
assert!(a
.chars()
.all(|c| c.is_ascii_alphanumeric() || matches!(c, '_' | '.' | '-')));
// A different random suffix yields a different name for the same target.
let b = instance_name("64f0aabbccddeeff00112233", 1_700_000_000, "zzzzzz999999");
assert_ne!(a, b);
}
#[test]
fn parse_epoch_rejects_foreign_names() {
assert_eq!(parse_epoch("some-other-container"), None);
assert_eq!(parse_epoch("certifai-plc-notanumber-x"), None);
}
#[test]
fn run_args_cap_resources_harden_label_and_never_publish_a_port() {
let args = run_args(&cfg(), "certifai-plc-1-t-r", "target-123");
// No host port publishing.
assert!(!args.iter().any(|a| a == "-p" || a == "--publish"));
// Detached.
assert!(args.contains(&"-d".to_string()));
// Joined to the agent's own network.
let net = args.iter().position(|a| a == "--network").expect("network");
assert_eq!(args[net + 1], "certifai");
// Resource caps.
let mem = args.iter().position(|a| a == "--memory").expect("memory");
assert_eq!(args[mem + 1], "512m");
let cpu = args.iter().position(|a| a == "--cpus").expect("cpus");
assert_eq!(args[cpu + 1], "0.5");
assert!(args.iter().any(|a| a == "--pids-limit"));
// Hardening.
let so = args
.iter()
.position(|a| a == "--security-opt")
.expect("secopt");
assert_eq!(args[so + 1], "no-new-privileges");
// Ownership + target labels for reaping / attribution.
assert!(args.contains(&"certifai.ephemeral=plc-runtime".to_string()));
assert!(args.contains(&"certifai.target=target-123".to_string()));
// Image is last.
assert_eq!(
args.last().map(String::as_str),
Some("registry.example.com/openplc:latest")
);
}
#[test]
fn rm_args_force_remove() {
assert_eq!(rm_args("x"), vec!["rm", "-f", "x"]);
}
#[test]
fn reap_list_filters_by_owner_label() {
let args = reap_list_args();
assert!(args.contains(&"label=certifai.ephemeral=plc-runtime".to_string()));
assert!(args.contains(&"{{.Names}}".to_string()));
}
}
+1
View File
@@ -70,6 +70,7 @@ impl TestServer {
pentest_imap_password: None,
admin_api_token: None,
tenant_registry_url: None,
plc_runtime: compliance_core::PlcRuntimeConfig::default(),
};
let agent = ComplianceAgent::new(config, db_pool);
+51
View File
@@ -49,6 +49,57 @@ pub struct AgentConfig {
/// of tenants to iterate. When `None` or unreachable, scheduler
/// falls back to `SCHEDULER_TENANT_IDS` env (M7.2-C).
pub tenant_registry_url: Option<String>,
/// Ephemeral soft-PLC provisioning for dynamic PLC testing (#183). Off by
/// default: it needs Docker access in the agent's runtime, which is a
/// deployment opt-in.
pub plc_runtime: PlcRuntimeConfig,
}
/// Configuration for the ephemeral soft-PLC "provision-and-test" path (#183).
///
/// When a PLC/SPS target ships control logic but no reachable live device, the
/// agent can instantiate that logic itself: spin up a throwaway soft-PLC
/// (OpenPLC) container in-cluster, load the program, start the runtime, probe it
/// over industrial protocols, then tear it down. This struct carries the knobs
/// for that container's lifecycle and the OpenPLC web-UI credentials used to
/// upload the program.
#[derive(Clone, Debug)]
pub struct PlcRuntimeConfig {
/// Master switch. Provision-and-test does nothing unless this is set — it
/// shells out to `docker`, which requires the agent container to have Docker
/// access (socket mount), an explicit deployment decision.
pub enabled: bool,
/// Container image for the ephemeral soft-PLC (OpenPLC).
pub image: String,
/// Docker network the instance joins. Must be the agent's own network so it
/// is reachable in-cluster by container name and never published to the host.
pub network: String,
/// Memory cap passed to `docker run --memory` (e.g. `512m`).
pub memory: String,
/// CPU cap passed to `docker run --cpus` (e.g. `0.5`).
pub cpus: String,
/// Hard ceiling on a provisioned instance's lifetime. Teardown is guaranteed
/// no later than this even if a load/probe step hangs.
pub max_lifetime_secs: u64,
/// OpenPLC web-UI username for the program upload (image default `openplc`).
pub openplc_user: String,
/// OpenPLC web-UI password (image default `openplc`).
pub openplc_password: SecretString,
}
impl Default for PlcRuntimeConfig {
fn default() -> Self {
Self {
enabled: false,
image: "registry.meghsakha.com/openplc:latest".to_string(),
network: "certifai".to_string(),
memory: "512m".to_string(),
cpus: "0.5".to_string(),
max_lifetime_secs: 180,
openplc_user: "openplc".to_string(),
openplc_password: SecretString::from("openplc".to_string()),
}
}
}
#[derive(Clone, Debug, Serialize, Deserialize)]
+1 -1
View File
@@ -13,6 +13,6 @@ pub mod auth;
#[cfg(feature = "axum")]
pub mod tenant_ctx;
pub use config::{AgentConfig, DashboardConfig};
pub use config::{AgentConfig, DashboardConfig, PlcRuntimeConfig};
pub use error::CoreError;
pub use tenant::{OrgRole, TenantContext, TenantStatus};
+1
View File
@@ -44,6 +44,7 @@ export default withMermaid(defineConfig({
items: [
{ text: 'Glossary', link: '/reference/glossary' },
{ text: 'Tools & Scanners', link: '/reference/tools' },
{ text: 'PLC Runtime Landscape', link: '/reference/plc-runtimes' },
],
},
],
+66
View File
@@ -11,6 +11,72 @@ the control application *and* the device it runs on.
| A device firmware image | Firmware SBOM / CVE (opt-in) |
| A reachable endpoint (WebVisu, OPC UA) | DAST / pentest (opt-in) |
## Anatomy: a soft PLC is a SoC + Linux + runtime
A CODESYS controller is **not** a monolithic appliance like a classic Siemens
S7. It is **PC-based ("soft") control** — commodity silicon running a
general-purpose Linux, with a **software PLC runtime** as just another process:
| Classic PLC (e.g. Siemens S7) | Soft PLC (CODESYS-on-Yocto, OpenPLC-on-Raspbian) |
| --- | --- |
| Proprietary hardware + firmware | Commodity SoC (x86 / ARM) |
| Proprietary OS | General-purpose Linux (a **Yocto** image, or Raspbian) |
| Proprietary runtime | Software runtime (**CODESYS Control**, or OpenPLC) |
| STEP7 / TIA project | IEC 61131-3 control app (ST / LD / FBD / SFC) |
Because of this, the device is built along **two independent tracks**, by
different people, on different timelines, and shipped separately. It also
inherits the **entire Linux / IT attack surface on top of** the OT / control
one — which is exactly why a PLC/SPS target is treated as a **composite**:
Certifai ingests one artifact per layer and scans each with the right pipeline.
```mermaid
flowchart TB
subgraph TA["Track A · Device platform — built by the hardware OEM / vendor"]
direction LR
A1["Yocto / OpenEmbedded<br/>BSP + RT kernel"] --> A2["Bake in the CODESYS<br/>Control for Linux runtime"] --> A3["bitbake → device image<br/>.wic / .tar + manifest"]
end
subgraph TB2["Track B · Control application — built by the machine builder / customer"]
direction LR
B1["CODESYS IDE<br/>ST / LD / FBD / SFC + WebVisu"] --> B2["Reference CODESYS +<br/>vendor libraries"] --> B3["Compile → download<br/>to device (gateway 11740)"]
end
A3 --> DEV(["Running soft-PLC device<br/>SoC + Linux + runtime + control app<br/>Modbus · OPC UA · EtherNet/IP · WebVisu"])
B3 --> DEV
subgraph CERT["What Certifai scans — one layer per artifact"]
direction LR
S1["Firmware layer<br/>FirmwareStatic · SBOM · CVE"]
S2["Control-logic layer<br/>PLC SAST — ST + FBD/LD"]
S3["Control-app SBOM<br/>libraries + runtime → CVE"]
S4["Running layer<br/>ICS probe · DAST (WebVisu)"]
end
A3 -. firmware image .-> S1
B1 -. PLCopen XML / ST via git .-> S2
B2 -. projectarchive (zip) .-> S3
DEV -. live URL / provisioned .-> S4
classDef yocto fill:#fde68a,stroke:#b45309,color:#111
classDef codesys fill:#bfdbfe,stroke:#1d4ed8,color:#111
classDef dev fill:#e9d5ff,stroke:#7e22ce,color:#111
classDef cert fill:#bbf7d0,stroke:#15803d,color:#111
class A1,A2,A3 yocto
class B1,B2,B3 codesys
class DEV dev
class S1,S2,S3,S4 cert
```
::: tip Where Yocto fits
Yocto is **Track A** — the *build system* for the device platform. It produces
the Linux image and bakes in the CODESYS runtime, so it is the **firmware
layer**, entirely separate from the control application. Hand it to Certifai as
its own **firmware image** artifact (scanned by the firmware pipeline, not the
PLC pipeline). The device OS need not be Yocto — Raspbian/Debian/Buildroot, or
even an RTOS / bare-metal, are all possible — but Yocto is the common,
product-grade industrial choice.
:::
## Two ways to deliver the project
You can either **upload** the project when onboarding, or point Certifai at a
+97
View File
@@ -0,0 +1,97 @@
# PLC Runtime Landscape & Support
A soft PLC is a **SoC + Linux + a software runtime + an IEC 61131-3 control app**
(see [PLC / SPS Projects](/guide/plc)).
The **runtime** is what defines the device — it provides the IEC engine, the
Modbus / OPC UA / EtherNet/IP servers, and the WebVisu. This page tracks the
runtime ecosystems Certifai may encounter.
We do **not** aim to support every runtime up front. Certifai supports the
**CODESYS family** today; everything else is a **watch-list** — when a customer
shows up using one, we add the parser/support for it then. The dynamic OT probe
(Modbus / OPC UA / EtherNet/IP) is **vendor-agnostic** and works regardless of
the runtime.
## Support status
| Status | Meaning |
| --- | --- |
| ✅ **Supported** | Static analysis works today (control-logic SAST + library/runtime SBOM + CVE). |
| 🟡 **Covered via CODESYS** | A rebranded CODESYS runtime — our CODESYS parsing applies (may need minor per-vendor tweaks). |
| 🔭 **Watch-list** | Own project format — we add a format parser when a customer needs it. The dynamic OT probe already applies. |
| 🧪 **Test-bench** | A free runtime we use to *reconstruct and dynamically test* a device (see epic: provision-and-test). |
## 1. CODESYS and rebranded CODESYS (the largest slice)
Much of the market licenses the CODESYS runtime and rebrands the IDE. If a
customer "doesn't use CODESYS", they often do — under another name.
| Product / vendor | Based on | Status |
| --- | --- | --- |
| **CODESYS** (3S-Smart Software Solutions) | CODESYS | ✅ Supported |
| Schneider **EcoStruxure Machine Expert** (ex-SoMachine) | CODESYS | 🟡 Covered via CODESYS |
| **WAGO** e!COCKPIT / PFC controllers | CODESYS | 🟡 Covered via CODESYS |
| **ABB** AC500 / Automation Builder | CODESYS | 🟡 Covered via CODESYS |
| **Bosch Rexroth** ctrlX / IndraLogic | CODESYS | 🟡 Covered via CODESYS |
| **Eaton** XSoft-CODESYS, **KEBA** KeStudio, Berghof, Kontron, Festo (CPX-E), IFM, Turck, … | CODESYS | 🟡 Covered via CODESYS |
## 2. Other embeddable IEC 61131-3 runtime toolkits
Same model as CODESYS (an OEM licenses a runtime + IDE and bakes it into a
device), but with **different project formats and libraries**.
| Toolkit | Vendor | Status |
| --- | --- | --- |
| **ProConOS / MULTIPROG** | Phoenix Contact / KW-Software | 🔭 Watch-list |
| **ISaGRAF** (also does IEC 61499) | Rockwell | 🔭 Watch-list |
| **straton** | COPA-DATA | 🔭 Watch-list |
| **logi.CAD** | logi.cals | 🔭 Watch-list |
## 3. Fully proprietary ecosystems (own runtime + IDE + protocols)
Static analysis here needs a **per-vendor project parser**; the **dynamic OT
probe still works** (they speak Modbus / OPC UA / EtherNet/IP, plus vendor
protocols like S7comm / CIP).
| Ecosystem | Vendor | Notes | Status |
| --- | --- | --- | --- |
| **TIA Portal / STEP 7** (S7-1200/1500), S7-1500 **Software Controller**, **Virtual PLC** | Siemens | Largest install base; the soft/virtual variants are Linux/container | 🔭 Watch-list |
| **Studio 5000** (ControlLogix / CompactLogix) | Rockwell / Allen-Bradley | Strong in North America | 🔭 Watch-list |
| **TwinCAT 3** | Beckhoff | Genuine PC-based control on Windows / TwinCAT-BSD; IEC 61131-3 **+ C++ + Simulink** | 🔭 Watch-list |
| **Automation Studio** | B&R (ABB) | Own Automation Runtime | 🔭 Watch-list |
| **GX Works** (MELSEC) | Mitsubishi | | 🔭 Watch-list |
| **Sysmac Studio** (NX / NJ) | Omron | | 🔭 Watch-list |
| **Proficy Machine Edition** (PACSystems) | Emerson / GE | | 🔭 Watch-list |
## 4. Linux-native / containerized soft-PLC (the direction of travel)
| Product | Vendor | Notes | Status |
| --- | --- | --- | --- |
| **PLCnext** | Phoenix Contact | Open, Linux-based; native runtime is eCLR (not CODESYS), but can also run CODESYS as an app | 🔭 Watch-list |
| **ctrlX** | Bosch Rexroth | Ubuntu-core, app-store model (CODESYS runtime inside) | 🟡 Covered via CODESYS |
| **Virtual PLC** / **CODESYS Virtual Control** | Siemens / CODESYS | Containerized PLCs (Docker / K8s) | 🟡 / 🔭 |
## 5. Open-source runtimes (free — our test-bench substrates)
Used to **reconstruct and dynamically test** a customer device without touching
their network (provision-and-test).
| Runtime | Standard | Notes | Status |
| --- | --- | --- | --- |
| **OpenPLC** | IEC 61131-3 | Modbus-centric, education/small automation; uses MatIEC | 🧪 Test-bench (current) |
| **Beremiz + MatIEC** | IEC 61131-3 | Fuller open-source IDE; compiles ST/IL → C. Natural fidelity step-up from OpenPLC | 🧪 Test-bench (candidate) |
| **Eclipse 4diac (FORTE)** | IEC **61499** | Distributed, event-driven — a *different paradigm* from 61131-3's scan cycle | 🔭 Watch-list |
| **ProView** | — | Open-source process control + SCADA | 🔭 Watch-list |
## How we add support for a new runtime
- **Static (SAST / SBOM):** needs a parser for that runtime's **project format**
(and its library/package convention). This is the per-vendor work.
- **Dynamic (ICS probe / DAST):** already **vendor-agnostic** — it targets the
device's OT ports and WebVisu, not the runtime's file format. So a brand-new
ecosystem still gets dynamic coverage on day one.
::: tip Rule of thumb
Confirm whether a "non-CODESYS" controller is actually a **rebranded CODESYS**
runtime (Section 1) before assuming new work — most of the long tail is.
:::