diff --git a/.env.example b/.env.example index 9b8b8db83..7ece07f3c 100644 --- a/.env.example +++ b/.env.example @@ -308,3 +308,8 @@ GEMINI_CHAT_MODEL=gemini-3.5-flash-lite # Typed chat requests (/api/gemini/chat) per IP per minute. Unset, it takes # GEV_RATELIMIT_GEMINI_PER_MIN's cap (default 30); 0 disables it. # GEV_RATELIMIT_GEMINI_CHAT_PER_MIN=30 + +# Optional keyless Radio Browser coverage: up to 8 ISO alpha-2 countries. +# Each reserves up to 40 stations within the existing 750-station directory. +# Restart after changing. Leave blank for the existing global selection. +# RADIO_PRIORITY_COUNTRIES=IN,PL diff --git a/DATA_SOURCES.md b/DATA_SOURCES.md index a7920f630..7d668c0d6 100644 --- a/DATA_SOURCES.md +++ b/DATA_SOURCES.md @@ -183,7 +183,9 @@ The optional Local ADS-B layer and the broadcast-FM receiver use radio you recei - **Autostrada del Brennero (A22) webcams.** Keyless. The camera list is one request to the Open Data Hub tourism API (`tourism.api.opendatahub.com/v1/WebcamInfo?source=a22`), run by NOI Techpark in South Tyrol, which republishes Autobrennero's own list of 13 motorway webcams with coordinates and kilometre markers. Frames are the operator's stills on `www.autobrennero.it/WebCamImg/km.jpg`, refreshed about once a minute, and only the exact image for each camera's kilometre is registered. The feed names no bearing, so headings use the id-hash fallback at low confidence. The road climbs from about 20 m in the Po valley to about 1,370 m at the Brenner Pass, so each camera's ground elevation comes from a per-kilometre table sampled from the Copernicus GLO-90 DEM (via Open-Meteo's elevation API); a camera not in the table uses a 200 m prior. The Open Data Hub tags every record CC0 (`ClosedData: false`, Autobrennero as licence holder), but that tag is the aggregator's metadata and Autobrennero publishes no terms of its own for the images, so they are credited to the operator as a courtesy and are not redistributed. Frames are fetched at request time and never stored. Disable with `CCTV_A22_ENABLED=0`. - **Camera frame content.** Public camera frames can contain people, vehicles and license plates, and the camera layer does nothing with them beyond putting them on the map. A successful upstream response is relayed as the provider served it: nothing in the camera pipeline blurs, redacts, enhances or restores it, and nothing there runs face, plate or object detection or otherwise recognises or classifies what is in the picture. (The ALPR layer described below maps where plate-reader cameras are, from locations contributors tagged in OpenStreetMap; it reads nothing from any camera frame.) When an upstream has no frame for a camera, what you see instead is a Google Street View still or the app's own placeholder card; the panel's source badge names which of them you are looking at, and a refresh that fails after a frame has arrived leaves the last good picture on screen rather than replacing it. The browser resamples each frame for display: onto a fixed 1920x1080 surface for the camera's projection plane in the scene, and onto a 192x108 canvas for the small ambient card. That scaling is the only change made to the picture, and the only pixels the camera layer reads back are a 64x36 downscale hashed to tell whether the picture has changed. Live streams are piped through; the still path reads the image into memory to pass it on — TxDOT wraps its JPEG in a JSON envelope, which is decoded there — and no frame is written to disk (the disk cache, `.gev-cache/`, holds feed JSON for other layers and never camera imagery). Outside that pipeline, the voice assistant is the one feature that sends imagery anywhere: when it answers a question about what is in view it may attach a downscaled screenshot of the rendered scene — including any camera plane on screen — to the request it sends to the voice model. Frames are fetched only from server-registered camera URLs — a client cannot hand the proxy a URL (see [SECURITY.md](SECURITY.md)). What the app does stays inside the line README draws: no named-person search, no face recognition, no tracking individuals. - **Transit (GTFS-Realtime).** The proxy requests only registered operator feeds, validates redirects, bounds responses, and uses conditional requests, short snapshot caching and failure backoff. All sixteen registered regions are live-enabled; each can be disabled independently. Vehicles use delayed playback between reports, with report age and waiting states shown separately. Session-local selected trails are available for every feed. MBTA additionally retains up to 15 minutes of recently observed positions in the running proxy, so available history can survive a browser reload; server restart clears it. History reads never fetch upstream, and other feeds have no proxy retention. Nothing is archived to disk or collected in the background. Operator text credits are included as a courtesy and do not imply endorsement. Heights are aligned to work with Google 3D tiles; subway markers are shown at street level because these feeds do not supply depth. The TTC feed is BusTime, the surface system: it carries buses and streetcars and no subway at all. About four in ten of its vehicles report no route at all (unassigned or deadheading), and those are shown with the feed default rather than a route-established mode. It sends no ETag or Last-Modified, so its conditional requests fall through to the ordinary snapshot cache. Three Golden Horseshoe neighbours run alongside it — MiWay, Hamilton HSR and Durham Region Transit — each bus-only and each polled only near its own region. Durham Region Transit publishes no bearing on any vehicle, so those sprites take their heading from observed movement like any other feed that omits it. Guelph, Kingston and Barrie extend the same pattern eastward and north. Hamilton, Guelph, Burlington and YRT publish route ids that are not the number on the vehicle, so those entries carry a route-name table taken from each operator’s own published GTFS; any id missing from a table renders unchanged rather than blank. A label is never invented: YRT runs a route `9101` that appears nowhere in its published routes, so it stays as the feed sends it. Burlington’s is the one that cannot be shortcut: `351` is route 1 while `3510` is route 10, so the table is a real join rather than a trimmed prefix. GO Transit / Metrolinx is deliberately absent: its REALTIME feed requires an API key issued by manual review, which the keyless admission rule excludes (its static schedule download is keyless, but carries no vehicle positions). Barrie and YRT/Viva each invite acceptance of their licence through a form, and each licence also states that using the data is itself acceptance; this deployment relies on that clause, which is recorded in both entries. Grand River Transit in Waterloo is absent for a different reason: it answers, but its feed header was months stale and it publishes no per-vehicle timestamps, so it would render vehicles that look live and are not. Ottawa’s OC Transpo falls to the same keyless admission rule as GO Transit — its realtime feed requires an API key. London Transit and Thunder Bay Transit publish vehicle positions over plain HTTP with no HTTPS endpoint available. Every other Ontario system — Brampton, Oakville, Niagara, Milton, Windsor, Greater Sudbury, UP Express and a dozen smaller operators — publishes static GTFS schedules only, with no realtime vehicle positions to register. Between the nine included and these exclusions, every Ontario realtime vehicle-position feed catalogued in the Mobility Database is accounted for. -- **Radio Browser.** `/api/radio/stations` discovers official API mirrors, makes bounded and coalesced healthy/geolocated HTTPS-station queries, caches the normalized public-domain directory for 45 minutes, and may serve the last good catalog for up to seven days during an outage. Refreshes must meet minimum accepted-query and station coverage before replacing a warm catalog; schema-valid responses whose rows all fail the product's health policy do not count as successful queries. A usable partial cold catalog is explicitly `DEGRADED`, and non-array or empty successful payloads are rejected atomically. Individual structurally malformed rows are skipped rather than failing the whole query; the response carries their bounded `rejectedStationCount` (at most 10,000), which the client adds to its own rejected rows and reports as degraded coverage without discarding the accepted generation. Every directory and click-count request rejects redirects, validates all resolved addresses as globally routable (including reserved/documentation IPv4 and special/non-global IPv6 exclusions), and pins the TLS connection to a validated address. Only MP3/AAC non-HLS directory rows with public HTTPS stream targets are returned; favicons are intentionally omitted. Pressing play connects one browser audio element directly to the selected broadcaster and calls the directory's click counter through known-ID-only `POST /api/radio/click/:uuid`. GEV never proxies, caches, records, bundles, or redistributes audio. Radio Browser supplies station-level tags, not dependable current-song or upcoming-program metadata, so Radio filtering never claims either. Direct playback exposes the listener's IP address to the broadcaster, whose own stream terms apply. +- **Radio country coverage (optional).** Set `RADIO_PRIORITY_COUNTRIES=IN,PL` in the local `.env` and restart to include stations beyond the global click charts. The operator may choose up to eight unique ISO alpha-2 country codes; no key is needed. Each country adds one bounded query (at most 200 rows) and reserves up to 40 unique, usable stations inside the existing 750-station catalog. Country seeds precede specialist-tag seeds and global popularity fill; unused slots return to the shared pool. This is a coverage preference, not an exhaustive national directory or a guarantee of stations in every city. Countries with no eligible HTTPS MP3/AAC geolocated stations remain empty. An unset list preserves global selection. Country requests share the directory cache and three-worker pool; failed country queries are reported in coverage without blocking a healthy global refresh. Failed/degraded refreshes back off for one minute, with at most three mirrors per query. + +- **Radio Browser.** In every installation, including when `RADIO_PRIORITY_COUNTRIES` is unset, each request tries at most three mirror candidates (later fallback candidates are skipped), and failed or partial catalog refreshes back off for 60 seconds. A cold-start failure therefore continues returning 503 during that delay even if upstream recovers. `/api/radio/stations` discovers official API mirrors, makes bounded and coalesced healthy/geolocated HTTPS-station queries, caches the normalized public-domain directory for 45 minutes, and may serve the last good catalog for up to seven days during an outage. Refreshes must meet minimum accepted-query and station coverage before replacing a warm catalog; broad/specialist responses whose rows all fail the product's health policy do not count as successful queries. Optional priority-country outcomes are advisory: valid, in-limit responses with no eligible stations succeed as empty results, while failed or malformed country queries report unsuccessful coverage without degrading the global catalog. A usable partial cold catalog is explicitly `DEGRADED`, and non-array or empty broad/specialist payloads are rejected atomically. Individual structurally malformed rows are skipped rather than failing the whole query; the response carries their bounded `rejectedStationCount` (at most 10,000), which the client adds to its own rejected rows and reports as degraded coverage without discarding the accepted generation. Every directory and click-count request rejects redirects, validates all resolved addresses as globally routable (including reserved/documentation IPv4 and special/non-global IPv6 exclusions), and pins the TLS connection to a validated address. Only MP3/AAC non-HLS directory rows with public HTTPS stream targets are returned; favicons are intentionally omitted. Pressing play connects one browser audio element directly to the selected broadcaster and calls the directory's click counter through known-ID-only `POST /api/radio/click/:uuid`. GEV never proxies, caches, records, bundles, or redistributes audio. Radio Browser supplies station-level tags, not dependable current-song or upcoming-program metadata, so Radio filtering never claims either. Direct playback exposes the listener's IP address to the broadcaster, whose own stream terms apply. - **TomTom Traffic.** Optional and BYOK: without `TOMTOM_API_KEY` the traffic layer runs its built-in simulation and no TomTom data (or attribution) appears. With a key, flow vector tiles are fetched through the server-side `/api/tomtom` proxy (120 s cache + a daily tile-budget governor — `TOMTOM_DAILY_TILE_BUDGET`, default 6,000 — sized so a 31-day month stays inside TomTom's 200K monthly free allowance; still a configurable application safety ceiling, not a guarantee; TomTom's [current pricing](https://docs.tomtom.com/pricing/) lists 200K free tile requests per month) and the "Traffic flow data © TomTom" credit is registered in the Data attribution popover the moment live mode activates. TomTom data is served live and cached only transiently (≤120 s TTL under `.gev-cache/`, gitignored) — it is not bundled or redistributed. One 23 KB point-in-time tile snapshot is committed as a decode-test fixture (`src/data/fixtures/`, © TomTom, never served to the app). - **Routing (OSRM / FOSSGIS).** `/api/route` proxies the public OSRM instances FOSSGIS e.V. runs for the OpenStreetMap community (car, bike, foot profiles). They are keyless and intended for fair, interactive use — the proxy rate-limits clients (60/min, 200/min total), caps route span and response size, caches each route for 10 minutes, and identifies itself in the User-Agent. Heavy or automated use must run its own OSRM. Routes are computed from OpenStreetMap data (ODbL); the Directions layer and voice routes never redistribute them. A route the router cannot find is reported as such; the Directions layer draws nothing in that case, and voice routes fall back to a clearly labeled straight line. - **Re:Earth Terrain.** Keyless (no API key). Used two ways: (1) `src/mapStackController.js` swaps in a `Cesium.CesiumTerrainProvider` pointed at Re:Earth's `cesium-mesh/ellipsoid` quantized-mesh endpoint for globe stacks without a Cesium ion token (e.g. OSM), replacing a flat `EllipsoidTerrainProvider`; falls back to the flat provider if the endpoint can't be reached. (2) The server-side `/api/terrain/heights` proxy (disk-cached, serve-stale) resolves per-point ellipsoidal ground height for entity placement. Both are best-effort with a keyless-safe fallback (bundled EGM96 geoid math) if Re:Earth is unreachable. diff --git a/README.md b/README.md index a699a3e44..a130ad116 100644 --- a/README.md +++ b/README.md @@ -212,7 +212,7 @@ Keep the `env`: bash treats `UID` as read-only, so a plain `UID=1026` in front o On SELinux hosts (Fedora, RHEL), compose mounts the checkout with `:z` so the container may read it; Docker may relabel the checkout for container access, which can take a moment on a large checkout. Without that, the start fails with `Cannot find module '/app/scripts/docker-start.mjs'`. -Behind a reverse proxy, put your keys in `.env` or pass them as environment variables - Provider Settings refuses proxied requests on purpose - and add the hostname you serve it on, for example `GEV_ALLOWED_HOSTS=globe.example.lan` (see [Sharing an instance](#-sharing-an-instance)). +Behind a reverse proxy, put your keys in `.env` or pass them as environment variables - Provider Settings refuses proxied requests on purpose - and add the hostname you serve it on, for example `GEV_ALLOWED_HOSTS=globe.example.lan` (see [Sharing an instance](#-sharing-an-instance)). The map routes compare the browser's `Origin` with the `Host` the server sees, so the proxy must pass the original `Host` and, when it terminates TLS, `X-Forwarded-Proto`. Caddy does both by default; for nginx add `proxy_set_header Host $http_host;`, `proxy_set_header X-Forwarded-Proto $scheme;` and `proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;`. On an install outside Docker, that forwarding header also rate-limits the proxy's users apart from your own browser. Under the default Docker bridge network every client on the host, your browser included, reaches the container from its gateway address and shares one Overpass bucket. diff --git a/SECURITY.md b/SECURITY.md index a3043eaa7..a749fd439 100644 --- a/SECURITY.md +++ b/SECURITY.md @@ -80,7 +80,7 @@ The data proxies under `server/providers/` are written so the browser cannot tur - **Sanitized errors** — internal error details are not echoed back to clients. - **Coalesced OAuth refresh** and cached successful responses only (OpenSky). - **Redacted debug logging.** The voice debug log (`.gev-logs/`, gitignored) strips API keys, bearer tokens, client secrets, and image data URLs before writing. -- **Cross-site gate on cost/log endpoints.** The cost-bearing (`/api/realtime/token`, `/api/openai/hud-summary`, `/api/google/nearby-places`, `/api/google/text-search`), quota-bearing (`/api/mapillary/*`, with a per-IP limit on tiles, and `/api/tomtom/*`, backed by its persistent daily tile budget) and log (`/api/realtime/debug-log`) endpoints refuse cross-site browser requests: a foreign or opaque `Origin`, a `Sec-Fetch-Site` other than `same-origin`/`none` (this is what blocks an ``/navigation that carries no `Origin`), or any reverse-proxy/CDN forwarding header returns `403`. Non-browser loopback tools (the repo QA harnesses POST the token endpoint with no `Origin`) and the explicit `HOST=0.0.0.0` LAN opt-in keep working — the gate deliberately does not require a loopback remote address; that stricter requirement belongs only to the credential panel (`admitKeySetupRequest`). The Host check (`allowedHosts`, applied by `hostCheckPlugin` ahead of every route, since Vite's own check runs only after plugin middleware) rejects a rebinding `Host` before any handler runs, in every mode, and the gate is defense in depth behind it: the LAN opt-in keeps that check and accepts only IP addresses, `localhost` and names listed in `GEV_ALLOWED_HOSTS`. **Limit:** a name you list there is trusted like the app's own address, so list only names you control. The keyless terrain route (`/api/terrain/heights`) uses the same cross-site checks, minus the forwarding-header refusal: it guards only the per-client quota and point cache, not a key or a bill, and the map needs it to render behind a reverse proxy. The CCTV routes (`/api/cctv/*`) use the same keyless checks; their one key-bearing step, the Street View fallback, is skipped for any request with forwarding headers. +- **Cross-site gate on cost/log endpoints.** The cost-bearing (`/api/realtime/token`, `/api/openai/hud-summary`, `/api/google/nearby-places`, `/api/google/text-search`), quota-bearing (`/api/mapillary/*`, with a per-IP limit on tiles, and `/api/tomtom/*`, backed by its persistent daily tile budget) and log (`/api/realtime/debug-log`) endpoints refuse cross-site browser requests: a foreign or opaque `Origin`, a `Sec-Fetch-Site` other than `same-origin`/`none` (this is what blocks an ``/navigation that carries no `Origin`), or any reverse-proxy/CDN forwarding header returns `403`. Non-browser loopback tools (the repo QA harnesses POST the token endpoint with no `Origin`) and the explicit `HOST=0.0.0.0` LAN opt-in keep working — the gate deliberately does not require a loopback remote address; that stricter requirement belongs only to the credential panel (`admitKeySetupRequest`). The Host check (`allowedHosts`, applied by `hostCheckPlugin` ahead of every route, since Vite's own check runs only after plugin middleware) rejects a rebinding `Host` before any handler runs, in every mode, and the gate is defense in depth behind it: the LAN opt-in keeps that check and accepts only IP addresses, `localhost` and names listed in `GEV_ALLOWED_HOSTS`. **Limit:** a name you list there is trusted like the app's own address, so list only names you control. The keyless map routes (`/api/terrain/heights` and `/api/overpass`) use the same cross-site checks, minus the forwarding-header refusal: they guard only per-client quotas and caches, not a key or a bill, and the map needs both to render behind a reverse proxy. `/api/overpass` also admits at most 300 POSTs a minute per peer (IPv6 peers keyed by /64) and 900 a minute across all remote peers before it reads a body. A direct loopback request — a loopback socket with none of the forwarding headers `hasProxySignals` checks (`Forwarded`, `Via`, `X-Forwarded-For`, `X-Forwarded-Host`, `X-Forwarded-Port`, `X-Forwarded-Proto`, `X-Real-IP`, `CF-Connecting-IP`, `CF-Ray`) — draws on its own per-peer admission window, so remote peers cannot exhaust the owner's admission. A reverse proxy on the same machine must add at least one of those headers (Caddy and Cloudflare do by default; nginx needs `proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;`, plus `proxy_set_header Host $http_host;` and `proxy_set_header X-Forwarded-Proto $scheme;` so the `Origin` check matches), or its traffic is indistinguishable from the owner's and shares the owner's window. With one, its requests are then counted as remote, in one bucket per proxy socket — the forwarded client address is never trusted as identity — so all users behind one proxy share 300/min. The isolation needs a loopback socket: under the default Docker bridge network every host-side client, the owner included, arrives from the container gateway and is one remote peer. **Limit:** this isolates admission only. The upstream budget (90 requests/min per client key, 300/min across all clients) is shared, so once remote traffic has used the global upstream budget, an owner request that misses the cache can still get `429`, or `503` when remote misses hold every upstream concurrency slot; cache hits are unaffected. The CCTV routes (`/api/cctv/*`) use the same keyless checks; their one key-bearing step, the Street View fallback, is skipped for any request with forwarding headers. - **Full Content-Security-Policy.** The dev and preview servers send a real CSP (`script-src 'self' 'unsafe-eval' blob:` plus the YouTube, Facebook and X scripts the Nepal event's embedded media loads and the Maps JavaScript API that Street View runs on, with a matching `frame-src` for their players and posts: no inline script and no other script origin; `blob:` covers the object URLs Cesium's bundled workers bootstrap through in the built app), alongside the existing `X-Frame-Options: DENY` and `frame-ancestors 'none'`. `'unsafe-eval'` stays because Knockout, bundled inside Cesium's widgets package, resolves the global object with `eval` at load time and the globe does not initialize without it; styles permit `'unsafe-inline'` and Google Fonts because Vite's dev client injects `