fix(certificate-handler): preserve wildcard coverage during forced certificate renewals and propagate renewed certs to sibling domains

ts/00_commitinfo_data.ts

@@ -3,6 +3,6 @@
  */
 export const commitinfo = {
   name: '@serve.zone/dcrouter',
-  version: '13.1.2',
+  version: '13.1.3',
   description: 'A multifaceted routing service handling mail and SMS delivery functions.'
 }

ts/opsserver/handlers/certificate.handler.ts

@@ -2,6 +2,28 @@ import * as plugins from '../../plugins.js';
 import type { OpsServer } from '../classes.opsserver.js';
 import * as interfaces from '../../../ts_interfaces/index.js';
 import { AcmeCertDoc, ProxyCertDoc } from '../../db/index.js';
+import { logger } from '../../logger.js';
+
+/**
+ * Mirrors `SmartacmeCertMatcher.getCertificateDomainNameByDomainName` from
+ * @push.rocks/smartacme. Inlined here because the original is `private` on
+ * SmartAcme. The cert identity ('task.vc' for both 'outline.task.vc' and
+ * '*.task.vc') is what AcmeCertDoc is keyed by, so two route domains with
+ * the same identity share the same underlying ACME cert.
+ *
+ * Returns undefined for domains with 4+ levels (matching smartacme's
+ * "deeper domains not supported" behavior) and for malformed inputs.
+ *
+ * Exported for unit testing.
+ */
+export function deriveCertDomainName(domain: string): string | undefined {
+  if (domain.startsWith('*.')) {
+    return domain.slice(2);
+  }
+  const parts = domain.split('.');
+  if (parts.length < 2 || parts.length > 3) return undefined;
+  return parts.slice(-2).join('.');
+}
 
 export class CertificateHandler {
   constructor(private opsServerRef: OpsServer) {
@@ -363,12 +385,34 @@ export class CertificateHandler {
 
     // If forceRenew, order a fresh cert from ACME now so it's already in
     // AcmeCertDoc by the time certProvisionFunction is invoked below.
+    //
+    // includeWildcard: when forcing a non-wildcard subdomain renewal, we still
+    // want the wildcard SAN in the order so the new cert keeps covering every
+    // sibling. Without this, smartacme defaults to includeWildcard: false and
+    // the re-issued cert would have only the base domain as SAN, breaking every
+    // sibling subdomain that was previously covered by the same wildcard cert.
     if (forceRenew && dcRouter.smartAcme) {
+      let newCert: plugins.smartacme.Cert;
       try {
-        await dcRouter.smartAcme.getCertificateForDomain(domain, { forceRenew: true });
+        newCert = await dcRouter.smartAcme.getCertificateForDomain(domain, {
+          forceRenew: true,
+          includeWildcard: !domain.startsWith('*.'),
+        });
       } catch (err: unknown) {
         return { success: false, message: `Failed to renew certificate for ${domain}: ${(err as Error).message}` };
       }
+
+      // Propagate the freshly-issued cert PEM to every sibling route domain that
+      // shares the same cert identity. Without this, the rust hot-swap (keyed by
+      // exact domain in `loaded_certs`) only fires for the clicked route via the
+      // fire-and-forget cert provisioning path, leaving siblings serving the
+      // stale in-memory cert until the next background reload completes.
+      try {
+        await this.propagateCertToSiblings(domain, newCert);
+      } catch (err: unknown) {
+        // Best-effort: failure here doesn't undo the cert issuance, just log.
+        logger.log('warn', `Failed to propagate force-renewed cert to siblings of ${domain}: ${(err as Error).message}`);
+      }
     }
 
     // Clear status map entry so it gets refreshed by the certificate-issued event
@@ -392,6 +436,86 @@ export class CertificateHandler {
     }
   }
 
+  /**
+   * After a force-renew, walk every route in the smartproxy that resolves to
+   * the same cert identity as `forcedDomain` and write the freshly-issued cert
+   * PEM into ProxyCertDoc for each. This guarantees that the next applyRoutes
+   * → provisionCertificatesViaCallback iteration will hot-swap every sibling's
+   * rust loaded_certs entry with the new (correct) PEM, rather than relying on
+   * the in-memory cert returned by smartacme's per-domain cache.
+   *
+   * Why this is necessary:
+   *   Rust's `loaded_certs` is a HashMap<domain, TlsCertConfig>. Each
+   *   bridge.loadCertificate(domain, ...) only swaps that one entry. The
+   *   fire-and-forget cert provisioning path triggered by updateRoutes does
+   *   eventually iterate every auto-cert route, but it returns the cached
+   *   (broken pre-fix) cert from smartacme's per-domain mutex. With this
+   *   helper, ProxyCertDoc is updated synchronously to the correct PEM before
+   *   applyRoutes runs, so even the transient window stays consistent.
+   */
+  private async propagateCertToSiblings(
+    forcedDomain: string,
+    newCert: plugins.smartacme.Cert,
+  ): Promise<void> {
+    const dcRouter = this.opsServerRef.dcRouterRef;
+    const smartProxy = dcRouter.smartProxy;
+    if (!smartProxy) return;
+
+    const certIdentity = deriveCertDomainName(forcedDomain);
+    if (!certIdentity) return;
+
+    // Collect every route domain whose cert identity matches.
+    const affected = new Set<string>();
+    for (const route of smartProxy.routeManager.getRoutes()) {
+      if (!route.match.domains) continue;
+      const routeDomains = Array.isArray(route.match.domains)
+        ? route.match.domains
+        : [route.match.domains];
+      for (const routeDomain of routeDomains) {
+        if (deriveCertDomainName(routeDomain) === certIdentity) {
+          affected.add(routeDomain);
+        }
+      }
+    }
+
+    if (affected.size === 0) return;
+
+    // Parse expiry from PEM (defense-in-depth — same pattern as
+    // ts/classes.dcrouter.ts:988-995 and the existing certStore.save callback).
+    let validUntil = newCert.validUntil;
+    let validFrom: number | undefined;
+    if (newCert.publicKey) {
+      try {
+        const x509 = new plugins.crypto.X509Certificate(newCert.publicKey);
+        validUntil = new Date(x509.validTo).getTime();
+        validFrom = new Date(x509.validFrom).getTime();
+      } catch { /* fall back to smartacme's value */ }
+    }
+
+    // Persist new cert PEM under each affected route domain
+    for (const routeDomain of affected) {
+      let doc = await ProxyCertDoc.findByDomain(routeDomain);
+      if (!doc) {
+        doc = new ProxyCertDoc();
+        doc.domain = routeDomain;
+      }
+      doc.publicKey = newCert.publicKey;
+      doc.privateKey = newCert.privateKey;
+      doc.ca = '';
+      doc.validUntil = validUntil || 0;
+      doc.validFrom = validFrom || 0;
+      await doc.save();
+
+      // Clear status so the next event refresh shows the new cert
+      dcRouter.certificateStatusMap.delete(routeDomain);
+    }
+
+    logger.log(
+      'info',
+      `Propagated force-renewed cert for ${forcedDomain} (cert identity '${certIdentity}') to ${affected.size} sibling route domain(s): ${[...affected].join(', ')}`,
+    );
+  }
+
   /**
    * Delete certificate data for a domain from storage
    */