Summary

CloudTAK: Authenticated full-read SSRF in the /api/esri* routes — user-controlled URL fetched with no IP-classification guard

Advisory details

Authenticated full-read SSRF in CloudTAK /api/esri* routes — user-controlled URL fetched with no IP-classification guard

Summary

Every route in the ESRI helper family (api/routes/esri.ts) takes a fully attacker-controlled URL from the request (POST /api/esri body url, and the portal / server / layer query parameters on the GET /api/esri/* routes) and passes it into EsriBase / EsriProxyPortal / EsriProxyServer / EsriProxyLayer in api/lib/esri.ts, which fetch it with the bare fetch from @tak-ps/etl. No IP / DNS / hostname classification is applied at any point, so the destination is never validated against private, loopback, or link-local ranges.

Any authenticated user (the routes only require Auth.is_auth(config, req, { anyResources: true }), i.e. any token, not an admin) can therefore make the CloudTAK server issue arbitrary outbound GET/POST requests to internal addresses such as the cloud instance-metadata service (169.254.169.254), loopback admin ports (127.0.0.1:<port>), and other hosts reachable only from inside the deployment VPC.

This is a full-read SSRF, not blind: on success the upstream JSON body is returned to the caller via res.json(...), and on failure the upstream error string is reflected verbatim as ESRI Server Error: <message>. An attacker can read cloud metadata (and the temporary IAM credentials the instance role exposes), enumerate internal services, and exfiltrate their response bodies.

The sniff() URL classifier provides no protection: it only pattern-matches the pathname (/rest, /arcgis/rest, /sharing/rest), so a URL like http://169.254.169.254/arcgis/rest or http://127.0.0.1:8500/rest passes sniff() and is fetched.

Affected versions

  • All versions up to and including 13.7.0 (latest at time of report).

The project already ships an SSRF guard helper — isSafeUrl from @tak-ps/node-safeurl — and wires it into the basemap, task, and video-service code paths, but the entire /api/esri* route family and the ESRI fetch library (api/lib/esri.ts) were never wired up, leaving the guard absent on this surface.

Vulnerable code

All permalinks are pinned to commit c7433679d2107fa0258e9005069bc5b4ca5773aa (release lineage of 13.7.0).

Routes — user input → ESRI fetch, no guard (api/routes/esri.ts):

Library — the fetch sinks (api/lib/esri.ts), all reached with the user URL and none preceded by a guard:

sniff() only inspects the pathname (no host/IP check): https://github.com/dfpc-coe/CloudTAK/blob/c7433679d2107fa0258e9005069bc5b4ca5773aa/api/lib/esri.ts#L142-L156

The guard exists elsewhere but is missing here — for comparison, the basemap import path classifies the URL before fetching: https://github.com/dfpc-coe/CloudTAK/blob/c7433679d2107fa0258e9005069bc5b4ca5773aa/api/routes/basemap.ts#L85-L90

Note also that the ESRI sub-branch inside basemap.ts (isEsriLayerURL(...)new EsriBase(...)EsriProxyLayer.tilejson()) reaches the same unguarded ESRI library and is therefore equally affected: https://github.com/dfpc-coe/CloudTAK/blob/c7433679d2107fa0258e9005069bc5b4ca5773aa/api/routes/basemap.ts#L770-L773

grep -c isSafeUrl api/routes/esri.ts api/lib/esri.ts returns 0 and 0.

Proof of concept

Prerequisites: a running CloudTAK instance and a valid user token (any non-admin user account — the routes only call Auth.is_auth(config, req, { anyResources: true })).

1. Read cloud instance metadata (credential theft)

POST /api/esri HTTP/1.1
Host: cloudtak.example.org
Authorization: Bearer <any-valid-user-token>
Content-Type: application/json

{ "url": "http://169.254.169.254/latest/meta-data/iam/security-credentials/arcgis/rest" }

The server's EsriBase.from() calls fetchVersion()fetch('http://169.254.169.254/...?f=json'). The path contains /rest, so sniff() classifies it as SERVER and the request proceeds. The metadata service's response body is read back to the attacker — either inside the successful JSON response, or reflected in the error string (ESRI Server Error: <upstream body fragment>). On AWS IMDSv1 deployments this yields the instance-role temporary credentials.

2. Read an internal-only service over loopback / VPC (full-read)

GET /api/esri/server?server=http://127.0.0.1:8500/rest HTTP/1.1
Host: cloudtak.example.org
Authorization: Bearer <any-valid-user-token>

new EsriBase('http://127.0.0.1:8500/rest')EsriProxyServer.getList()fetch('http://127.0.0.1:8500/rest?f=json'). The full JSON returned by the internal service (here a Consul/admin port, but any internal host:port reachable from the CloudTAK box works) is reflected to the attacker via res.json(list).

GET /api/esri/portal?portal=http://<internal-host>/sharing/rest and GET /api/esri/server/layer?layer=http://<internal-host>/rest/.../FeatureServer/0&query=1=1 give the same full-read primitive on the other sub-routes.

3. Negative control — unauthenticated is rejected

POST /api/esri HTTP/1.1
Host: cloudtak.example.org
Content-Type: application/json

{ "url": "http://169.254.169.254/latest/meta-data/arcgis/rest" }

Returns 403 Authentication Required (from Auth.is_authapi/lib/auth.ts:118). The SSRF is reachable by any authenticated user but not by an anonymous one.

E2E reproduction (sink path against a controlled internal victim)

Because exercising the real route against a public cloud metadata endpoint is not something to do against third-party infrastructure, the sink was reproduced against a local internal-only victim using the same fetch import (@tak-ps/etl) and the verbatim EsriBase.sniff() + fetchVersion() logic the route uses. The harness models the route handler exactly: it new URL()s the user input, runs sniff(), then fetch()s — with isSafeUrl deliberately not called (matching shipped behavior), and a toggled control branch that calls it (matching the basemap guard).

Victim

References