fix(fips): harden companion mesh joins

core/archipelago/src/fips/anchors.rs

@@ -156,7 +156,24 @@ pub async fn load(data_dir: &Path) -> Result<Vec<SeedAnchor>> {
         .with_context(|| format!("read {}", path.display()))?;
     let anchors: Vec<SeedAnchor> =
         serde_json::from_slice(&bytes).with_context(|| format!("parse {}", path.display()))?;
-    Ok(anchors)
+    Ok(repair_legacy_anchor_set(anchors))
+}
+
+/// Add the newer redundant public TCP anchors to legacy files that only carry
+/// the Archipelago-operated vps2 anchor. A deliberately custom/private anchor
+/// list remains authoritative; this only repairs the exact stale shape shipped
+/// before the join.fips.network anchors became defaults.
+fn repair_legacy_anchor_set(mut anchors: Vec<SeedAnchor>) -> Vec<SeedAnchor> {
+    let has_archy = anchors.iter().any(|a| a.npub == ARCHY_ANCHOR_NPUB);
+    let has_fips_network = fips_network_anchors()
+        .iter()
+        .any(|default| anchors.iter().any(|a| a.npub == default.npub));
+    if has_archy && !has_fips_network {
+        for anchor in fips_network_anchors() {
+            anchors.push(anchor);
+        }
+    }
+    anchors
 }
 
 /// Persist the list. Overwrites atomically via write-then-rename so a
@@ -338,6 +355,30 @@ mod tests {
         assert!(got.iter().all(|a| a.transport == "tcp"));
     }
 
+    #[tokio::test]
+    async fn load_repairs_legacy_archy_only_anchor_file() {
+        let dir = tempfile::tempdir().unwrap();
+        save(dir.path(), &[archy_anchor()]).await.unwrap();
+
+        let got = load(dir.path()).await.unwrap();
+        assert!(got.iter().any(|a| a.npub == ARCHY_ANCHOR_NPUB));
+        for anchor in fips_network_anchors() {
+            assert!(got.iter().any(|a| a.npub == anchor.npub));
+        }
+    }
+
+    #[tokio::test]
+    async fn load_keeps_private_anchor_file_authoritative() {
+        let dir = tempfile::tempdir().unwrap();
+        let private = mk("npub1private");
+        save(dir.path(), std::slice::from_ref(&private))
+            .await
+            .unwrap();
+
+        let got = load(dir.path()).await.unwrap();
+        assert_eq!(got, vec![private]);
+    }
+
     #[tokio::test]
     async fn removing_one_default_persists_and_keeps_the_other() {
         // Editing the anchor list (here removing one default) makes the file

core/archipelago/src/fips/service.rs

@@ -81,6 +81,7 @@ async fn sudo_systemctl(verb: &str, unit: &str) -> Result<()> {
 /// Unmask + start + enable the FIPS service. Idempotent — safe to call
 /// on every backend startup once the key is on disk.
 pub async fn activate(unit: &str) -> Result<()> {
+    kill_stale_daemons().await?;
     // Order matters: unmask before enable/start, otherwise enable fails
     // on a masked unit.
     sudo_systemctl("unmask", unit).await?;
@@ -94,9 +95,42 @@ pub async fn stop(unit: &str) -> Result<()> {
 }
 
 pub async fn restart(unit: &str) -> Result<()> {
+    kill_stale_daemons().await?;
     sudo_systemctl("restart", unit).await
 }
 
+/// Kill orphaned `fips` processes not owned by either known systemd unit.
+///
+/// Field failure, 2026-07-24: a stale daemon survived outside systemd and kept
+/// `0.0.0.0:8443` bound. The supervised daemon then started UDP-only, so every
+/// TCP seed-anchor connect failed with "no operational transport" and phones
+/// on 5G could not discover the node. We keep the cleanup narrow: preserve the
+/// MainPID of both units and terminate only extra `pgrep -x fips` matches.
+pub async fn kill_stale_daemons() -> Result<()> {
+    let script = format!(
+        r#"keep="$(systemctl show -p MainPID --value {managed} 2>/dev/null; systemctl show -p MainPID --value {upstream} 2>/dev/null)"
+for pid in $(pgrep -x fips 2>/dev/null || true); do
+  case " $keep " in
+    *" $pid "*) ;;
+    *) kill "$pid" 2>/dev/null || true ;;
+  esac
+done
+"#,
+        managed = super::SERVICE_UNIT,
+        upstream = super::UPSTREAM_SERVICE_UNIT,
+    );
+    let out = Command::new("sudo")
+        .args(["sh", "-c", &script])
+        .output()
+        .await
+        .context("sudo stale fips cleanup failed to launch")?;
+    if !out.status.success() {
+        let stderr = String::from_utf8_lossy(&out.stderr).trim().to_string();
+        anyhow::bail!("stale fips cleanup failed: {}", stderr);
+    }
+    Ok(())
+}
+
 /// Resolve which systemd unit is actually supervising the fips daemon
 /// on this host. Nodes installed from the archipelago ISO run
 /// `archipelago-fips.service`; nodes that were apt-installed (or had