feat(host-agent): Phase 3a signed self-update (minisign) + CI signing gate

Agent only ever runs a binary whose minisign signature verifies against
the EMBEDDED public key. NATS host.cmd func 'update' {url}: download
binary + .minisig from the CDN -> verify against embedded pubkey ->
atomic swap (.old rollback) -> relaunch. URL allowlist (https + cdn.
corrosionmgmt.com only, rejects userinfo-bypass), 100MiB cap. Closes the
supply-chain hole: even a malicious CDN upload can't run unsigned.

CI: build-host-agent.yml signs every artifact with MINISIGN_SECRET_KEY
(Gitea secret) and publishes .minisig alongside; the step FAILS the
build if the secret is absent (refuses to ship unsigned). Bumped to
alpha.6.

6 deterministic tests (accept valid / reject tampered+garbage+empty sig,
URL allowlist incl userinfo-bypass, atomic swap+rollback). Fixtures
signed with the real release key so tests need no key at runtime. Full
suite 50/50 green; musl + native build clean.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

corrosion-host-agent/src/hostcmd.rs

@@ -13,11 +13,15 @@ use crate::agent::Agent;
 use crate::prober;
 use crate::subjects;
 use crate::telemetry;
+use crate::update;
 use crate::version;
 
 #[derive(Debug, Deserialize)]
 struct HostCommand {
     func: String,
+    /// Signed-update artifact URL (for func = "update").
+    #[serde(default)]
+    url: Option<String>,
 }
 
 pub async fn run(agent: Arc<Agent>) -> anyhow::Result<()> {
@@ -55,20 +59,46 @@ async fn handle(agent: Arc<Agent>, msg: async_nats::Message) {
         return;
     };
 
-    let response = match serde_json::from_slice::<HostCommand>(&msg.payload) {
-        Ok(cmd) => dispatch(&agent, &cmd.func).await,
-        Err(e) => json!({ "status": "error", "message": format!("invalid command payload: {e}") }),
-    };
-
-    let bytes = match serde_json::to_vec(&response) {
-        Ok(b) => b,
+    let cmd = match serde_json::from_slice::<HostCommand>(&msg.payload) {
+        Ok(cmd) => cmd,
         Err(e) => {
-            tracing::error!("response serialize failed: {e}");
+            publish(&agent, &reply, json!({ "status": "error", "message": format!("invalid command payload: {e}") })).await;
             return;
         }
     };
-    if let Err(e) = agent.nats.publish(reply, bytes.into()).await {
-        tracing::warn!("response publish failed: {e}");
+
+    // Self-update is special: it must reply BEFORE relaunching, because the
+    // relaunch replaces this process and nothing after it would run.
+    if cmd.func == "update" {
+        let Some(url) = cmd.url else {
+            publish(&agent, &reply, json!({ "status": "error", "message": "update requires a 'url'" })).await;
+            return;
+        };
+        match update::download_verify_swap(&url).await {
+            Ok(_) => {
+                publish(&agent, &reply, json!({ "status": "success", "func": "update", "message": "verified and swapped; relaunching" })).await;
+                let _ = agent.nats.flush().await;
+                update::relaunch_and_exit();
+            }
+            Err(e) => {
+                publish(&agent, &reply, json!({ "status": "error", "func": "update", "message": format!("{e:#}") })).await;
+            }
+        }
+        return;
+    }
+
+    let response = dispatch(&agent, &cmd.func).await;
+    publish(&agent, &reply, response).await;
+}
+
+async fn publish(agent: &Arc<Agent>, reply: &async_nats::Subject, value: serde_json::Value) {
+    match serde_json::to_vec(&value) {
+        Ok(bytes) => {
+            if let Err(e) = agent.nats.publish(reply.clone(), bytes.into()).await {
+                tracing::warn!("response publish failed: {e}");
+            }
+        }
+        Err(e) => tracing::error!("response serialize failed: {e}"),
     }
 }
 

corrosion-host-agent/src/lib.rs

@@ -13,4 +13,5 @@ pub mod rcon;
 pub mod steamcmd;
 pub mod subjects;
 pub mod telemetry;
+ pub mod update;
 pub mod version;

corrosion-host-agent/src/update.rs

@@ -0,0 +1,154 @@
+//! Signed self-update.
+//!
+//! The agent only ever runs a binary whose minisign signature verifies against
+//! the EMBEDDED public key below. Even if the CDN (which currently accepts
+//! unauthenticated uploads) served a malicious binary, the agent refuses it
+//! without a valid signature from the release private key (a CI secret).
+//!
+//! Flow: download binary + `.minisig` from the CDN → verify signature →
+//! atomic swap (current → `.old`, new → current, rollback on failure) →
+//! relaunch the new binary. Defence in depth mirrors the Vigilance updater:
+//! a real URL parse rejecting credential-in-URL bypasses, an https + host
+//! allowlist, and a size cap.
+
+use anyhow::{bail, Context, Result};
+use minisign_verify::{PublicKey, Signature};
+use std::path::{Path, PathBuf};
+use std::time::Duration;
+
+/// minisign public key. The matching private key signs releases in CI
+/// (Gitea Actions secret MINISIGN_SECRET_KEY). Rotating it means re-signing
+/// every published artifact and shipping an agent build with the new key.
+const PUBLIC_KEY: &str = "RWQKhJptuiwIkp31cZdz10z/R72UPZkl7/VtnZJ2Vfbe0dQfDlXHZYFC";
+
+const ALLOWED_HOST: &str = "cdn.corrosionmgmt.com";
+const MAX_BINARY_BYTES: usize = 100 * 1024 * 1024; // 100 MiB sanity cap
+const DOWNLOAD_TIMEOUT: Duration = Duration::from_secs(600);
+
+/// Verify a binary against the embedded public key + a minisign signature blob.
+/// The security core of self-update — tampered or unsigned content is rejected.
+pub fn verify_signature(binary: &[u8], signature_blob: &str) -> Result<()> {
+    let pk = PublicKey::from_base64(PUBLIC_KEY).context("embedded public key is invalid")?;
+    let sig = Signature::decode(signature_blob).context("malformed minisign signature")?;
+    pk.verify(binary, &sig, false)
+        .map_err(|e| anyhow::anyhow!("signature verification failed: {e}"))?;
+    Ok(())
+}
+
+/// Reject anything but `https://cdn.corrosionmgmt.com/...` with no embedded
+/// credentials (the userinfo-bypass class).
+pub fn assert_url_allowed(url: &str) -> Result<()> {
+    let parsed = reqwest::Url::parse(url).context("invalid update URL")?;
+    if parsed.scheme() != "https" {
+        bail!("update URL must be https");
+    }
+    if !parsed.username().is_empty() || parsed.password().is_some() {
+        bail!("update URL must not contain credentials");
+    }
+    if parsed.host_str() != Some(ALLOWED_HOST) {
+        bail!("update URL host not allowed: {:?}", parsed.host_str());
+    }
+    Ok(())
+}
+
+/// Download, verify, and atomically swap in a new agent binary. Does NOT
+/// restart — the caller decides when to relaunch (after replying on NATS).
+/// Returns the path of the now-current (new) binary.
+pub async fn download_verify_swap(url: &str) -> Result<PathBuf> {
+    assert_url_allowed(url)?;
+    let sig_url = format!("{url}.minisig");
+    assert_url_allowed(&sig_url)?;
+
+    let client = reqwest::Client::builder()
+        .timeout(DOWNLOAD_TIMEOUT)
+        .build()
+        .context("building HTTP client")?;
+
+    let binary = client
+        .get(url)
+        .send()
+        .await
+        .with_context(|| format!("downloading {url}"))?
+        .error_for_status()
+        .context("update binary download failed")?
+        .bytes()
+        .await
+        .context("reading update binary")?;
+
+    if binary.len() > MAX_BINARY_BYTES {
+        bail!("update binary is {} bytes, exceeds the {MAX_BINARY_BYTES} cap", binary.len());
+    }
+
+    let signature = client
+        .get(&sig_url)
+        .send()
+        .await
+        .with_context(|| format!("downloading {sig_url}"))?
+        .error_for_status()
+        .context("signature download failed")?
+        .text()
+        .await
+        .context("reading signature")?;
+
+    verify_signature(&binary, &signature).context("refusing unsigned/tampered update")?;
+    tracing::info!("update signature verified ({} bytes)", binary.len());
+
+    let current = std::env::current_exe().context("resolving current executable")?;
+    swap_binary(&current, &binary)?;
+    tracing::info!("update swapped in at {}", current.display());
+    Ok(current)
+}
+
+/// Atomically replace `current` with `new_bytes`, keeping a `.old` backup and
+/// rolling back if the rename fails.
+pub fn swap_binary(current: &Path, new_bytes: &[u8]) -> Result<()> {
+    let dir = current.parent().unwrap_or_else(|| Path::new("."));
+    let stem = current.file_name().and_then(|s| s.to_str()).unwrap_or("corrosion-host-agent");
+    let new_path = dir.join(format!("{stem}.new"));
+    let backup = dir.join(format!("{stem}.old"));
+
+    std::fs::write(&new_path, new_bytes)
+        .with_context(|| format!("writing {}", new_path.display()))?;
+
+    #[cfg(unix)]
+    {
+        use std::os::unix::fs::PermissionsExt;
+        std::fs::set_permissions(&new_path, std::fs::Permissions::from_mode(0o755))
+            .context("chmod +x on new binary")?;
+    }
+
+    let _ = std::fs::remove_file(&backup);
+    std::fs::rename(current, &backup)
+        .with_context(|| format!("backing up current binary to {}", backup.display()))?;
+
+    if let Err(e) = std::fs::rename(&new_path, current) {
+        // Roll back: restore the backup so the agent stays runnable.
+        let _ = std::fs::rename(&backup, current);
+        return Err(anyhow::anyhow!(e).context("installing new binary (rolled back)"));
+    }
+    Ok(())
+}
+
+/// Relaunch the (already-swapped) binary with the same args, then exit. No
+/// service manager is required — the new process reconnects on its own. There
+/// is a sub-second window with no agent; acceptable for an update.
+pub fn relaunch_and_exit() -> ! {
+    let exe = std::env::current_exe().unwrap_or_else(|_| PathBuf::from("corrosion-host-agent"));
+    let args: Vec<String> = std::env::args().skip(1).collect();
+    tracing::info!("relaunching {} after update", exe.display());
+
+    #[cfg(unix)]
+    {
+        use std::os::unix::process::CommandExt;
+        // exec replaces this process image with the new binary — cleanest,
+        // no gap. Only returns on failure.
+        let err = std::process::Command::new(&exe).args(&args).exec();
+        tracing::error!("exec after update failed: {err}; exiting for service restart");
+        std::process::exit(70);
+    }
+    #[cfg(not(unix))]
+    {
+        let _ = std::process::Command::new(&exe).args(&args).spawn();
+        std::process::exit(0);
+    }
+}