nix/modules/hive-gateway.nix

@@ -403,22 +403,21 @@ in
         {
           system.stateVersion = "26.05";
 
-          # Ensure a valid self-signed cert exists before nginx starts.
+          # Generate a self-signed cert on first boot if missing. nginx
-          # nginx `Requires=` this via `requiredBy`, so systemd refuses
+          # `Requires=` this via `requiredBy`, so systemd refuses to
-          # to start nginx until the script succeeds. ALWAYS runs (no
+          # start nginx until the cert exists — `before=` alone wasn't
-          # ConditionPathExists) and is idempotent — that's necessary
+          # enough (it only orders within a single transaction, but
-          # to reconcile broken state left over from prior failed
+          # nginx was being pulled into a different transaction by
-          # boots (a 0700 dir from a stale UMask, a truncated cert
+          # multi-user.target and started without waiting, #856). Cert
-          # from an interrupted oneshot, etc.) which a guarded-on-
+          # covers the bare hive domain plus `*.${hyperhiveDomain}` so
-          # missing-cert script would silently skip and leave broken.
+          # the matrix + forge sub-domains are valid under the same
-          # Cert covers the bare hive domain plus `*.${hyperhiveDomain}`
-          # so the matrix + forge sub-domains are valid under the same
           # cert. See `docs/gateway.md` ("Self-signed TLS").
           systemd.services.hive-gateway-self-signed-cert = lib.mkIf cfg.selfSignedTls {
-            description = "Ensure self-signed TLS cert for hive-gateway";
+            description = "Generate self-signed TLS cert for hive-gateway";
             wantedBy = [ "multi-user.target" ];
             before = [ "nginx.service" ];
             requiredBy = [ "nginx.service" ];
+            unitConfig.ConditionPathExists = "!${tlsCert}";
             serviceConfig = {
               Type = "oneshot";
               RemainAfterExit = true;
@@ -439,29 +438,16 @@ in
                 );
               in
               ''
-                set -eu
                 mkdir -p ${tlsDir}
                 # 0755 dir so nginx (master starts as root but workers
-                # drop to the nginx user) can traverse to read the cert
+                # drop to the nginx user) can read the cert path
-                # path. Re-applied every boot in case a prior run left
+                # without traversal failures. Key stays 0600 below.
-                # a tighter mode behind. Key stays 0600 below.
                 chmod 0755 ${tlsDir}
-                # Generate the cert when EITHER the cert or key is
+                openssl req -x509 -newkey rsa:4096 -nodes -sha256 -days 3650 \
-                # missing/empty, OR the cert fails an openssl parse —
+                  -keyout ${tlsKey} \
-                # catches truncated / corrupt leftovers from a previous
+                  -out ${tlsCert} \
-                # interrupted run AND the "cert clean but key absent"
+                  -subj "/CN=${subjectCN}" \
-                # edge case (argus 🟡 on the first revision) which
+                  -addext "subjectAltName=${sanLines}"
-                # otherwise tripped `chmod 0600 ${tlsKey}` below with
-                # ENOENT under `set -eu`. The whole oneshot is safe to
-                # re-run; a healthy cert+key pair is left alone.
-                if [ ! -s ${tlsCert} ] || [ ! -s ${tlsKey} ] || ! openssl x509 -in ${tlsCert} -noout >/dev/null 2>&1; then
-                  echo "generating fresh self-signed cert at ${tlsCert}"
-                  openssl req -x509 -newkey rsa:4096 -nodes -sha256 -days 3650 \
-                    -keyout ${tlsKey} \
-                    -out ${tlsCert} \
-                    -subj "/CN=${subjectCN}" \
-                    -addext "subjectAltName=${sanLines}"
-                fi
                 chmod 0600 ${tlsKey}
                 chmod 0644 ${tlsCert}
               '';