swarm-victoriametrics: forward only the store's read API

The metrics vhost's `/` proxied every path to VictoriaMetrics with no
auth_request, so anything that could reach metrics.<swarm> could write
(`/api/v1/import`, `/api/v1/write`, ...) and delete
(`/api/v1/admin/tsdb/delete_series`) series. Only the OTLP push route
was gated.

`/` now answers 403. The read routes Grafana's Prometheus datasource
and vmui use are forwarded by exact or `^~` prefix location, each with
the matched path as its proxy_pass URI. The bare domain, which the
swarm UI's Metrics quick link opens, redirects to vmui instead of the
store's index page. The OTLP push location is unchanged.

Closes #5186

nix/host-modules/swarm-victoriametrics.nix

@@ -25,6 +25,51 @@ let
   # Shared host netns, like every sibling swarm container: the gateway
   # reaches this at 127.0.0.1:<port>.
   privateNetwork = false;
+
+  # The store's read API: the only part of it the vhost serves without a
+  # credential. Each location's `proxy_pass` carries the path, so the store
+  # receives the path nginx matched, not the caller's unnormalised spelling.
+  #
+  # Grafana's Prometheus datasource reads the query, series, label, metadata,
+  # buildinfo, exemplar, rules and alerts routes; the store answers the last
+  # three with fixed placeholders. vmui is the static files under `/vmui/`
+  # plus the query API and the remaining read-only routes here. Paths read off
+  # the pinned build's `app/vmselect/main.go` and vmui's `src/`.
+  readPaths = [
+    "/api/v1/query"
+    "/api/v1/query_range"
+    "/api/v1/query_exemplars"
+    "/api/v1/series"
+    "/api/v1/labels"
+    "/api/v1/metadata"
+    "/api/v1/rules"
+    "/api/v1/alerts"
+    "/api/v1/status/buildinfo"
+    "/api/v1/export"
+    "/api/v1/export/csv"
+    "/api/v1/status/tsdb"
+    "/api/v1/status/metric_names_stats"
+    "/api/v1/status/top_queries"
+    "/api/v1/status/active_queries"
+    "/expand-with-exprs"
+    "/prettify-query"
+    "/metric-relabel-debug"
+    "/vmui"
+  ];
+  # The store serves only `<name>/values` under `/api/v1/label/`, and only
+  # static files and UI config under `/vmui/`.
+  readPrefixes = [
+    "/api/v1/label/"
+    "/vmui/"
+  ];
+  readLocations =
+    let
+      to = path: { proxyPass = "http://127.0.0.1:${toString cfg.port}${path}"; };
+    in
+    lib.listToAttrs (
+      map (p: lib.nameValuePair "= ${p}" (to p)) readPaths
+      ++ map (p: lib.nameValuePair "^~ ${p}" (to p)) readPrefixes
+    );
 in
 {
   # What the store IS from any hive's point of view, the name it answers on and
@@ -90,14 +135,19 @@ in
     services.nginx.virtualHosts."${cfg.domain}" = (gatewayCfg.lib.tlsFor cfg.domain) // {
       listen = gatewayCfg.lib.listen;
       extraConfig = gatewayCfg.lib.securityHeaders;
-      locations = {
-        "/" = {
-          proxyPass = "http://127.0.0.1:${toString cfg.port}/";
-        };
+      locations = readLocations // {
+        # Nothing reaches the store unless a location here names it. The
+        # store serves writes and admin calls (`/api/v1/import`,
+        # `/api/v1/admin/tsdb/delete_series`, ...) on the same listener as
+        # its queries, and authenticates none of them.
+        "/".return = "403";
+
+        # The swarm UI's quick link opens the bare domain. The store's own
+        # index page links mostly to paths refused above, so send it to vmui.
+        "= /".return = "302 /vmui/";
 
         # The swarm collector's ingest route, and the only authenticated thing
-        # on this vhost. `=` so it outranks the `/` prefix above, which would
-        # otherwise carry these writes with no check at all.
+        # on this vhost. `=` so it outranks the `/` prefix above.
         #
         # ⚠️ No `error_page 401 =302` here, and its absence is the point: a
         # redirect is right for a browser and wrong for a pusher, which would

nix/module-eval/core-toggle.nix

@@ -425,6 +425,38 @@ let
         && !(lib.hasInfix "error_page" l)
         && lib.hasInfix "error_page" browser;
     }
+    {
+      # The store authenticates nothing itself, so every location that
+      # forwards without `auth_request` is open to anything that can reach the
+      # domain. The query location is the positive control: it shows the
+      # filter finds forwarding locations at all.
+      name = "the metrics vhost forwards only read paths without a credential";
+      ok =
+        let
+          locs = allLocal.services.nginx.virtualHosts."metrics.t.local".locations;
+          open = lib.filterAttrs (
+            _: l:
+            l.proxyPass != null
+            && !(lib.hasInfix "auth_request" l.extraConfig)
+            && !(lib.hasInfix "internal;" l.extraConfig)
+          ) locs;
+          writeWords = [
+            "import"
+            "write"
+            "admin"
+            "delete"
+            "insert"
+            "internal"
+            "push"
+            "prometheus"
+            "influx"
+          ];
+        in
+        locs."/".proxyPass == null
+        && locs."/".return == "403"
+        && open ? "= /api/v1/query"
+        && lib.all (n: n != "/" && !(lib.any (w: lib.hasInfix w n) writeWords)) (lib.attrNames open);
+    }
     {
       # The read counterpart to the ingest location: an agent queries the log
       # store with a bearer token, and the `/` catch-all is the browser's