GHSA-q567-cr4x-96w4

Summary

A WEBHOOK alert channel stores a user-supplied url. When an alert fires (deployment/run failure, error groups), the webapp server (alertsWorker -> DeliverAlertService) POSTs the HMAC-signed alert payload to that URL via fetch(webhook.url, ...). The URL is never validated against a host allowlist or private-IP/metadata blocklist (a repo-wide search for 169.254, isPrivate, isLoopback, net.isIP, ssrf returns ZERO hits), and the API route's URL field is just z.string().optional() (no syntax check at all). So an authenticated tenant can point the webhook at internal infrastructure or 169.254.169.254 and the multi-tenant server fetches it.

Affected

apps/webapp, HEAD 5d99457 (current main).

Root cause

• Source (API route): app/presenters/v3/ApiAlertChannelPresenter.server.ts ApiAlertChannelData.url = z.string().optional() (no host validation); route app/routes/api.v1.projects.$projectRef.alertChannels.ts (PAT-auth).
• Store: app/v3/services/alerts/createAlertChannel.server.ts persists {url, secret, version} verbatim.
• Sink: app/v3/services/alerts/deliverAlert.server.ts:973 fetch(webhook.url, {method:"POST", headers:{"x-trigger-signature-hmacsha256":...}, body:rawPayload}); identical at deliverErrorGroupAlert.server.ts:258. Runs inside the webapp process; fetch follows redirects (IMDSv1-via-redirect). No egress guard anywhere.

Runtime PoC (proven on self-host)

Seeded a project + STAGING env + a WEBHOOK channel with url=http://host.docker.internal:7766/... (an internal address from the server's POV) + a DEPLOYING deployment. Triggered via the public API POST /api/v1/deployments/deployment_ssrfpoc/fail -> 200 -> FAILED -> alertsWorker -> the server fetched the listener. Captured:

POST /ssrf-via-webhook HTTP/1.1
host: host.docker.internal:7766
x-trigger-signature-hmacsha256: <redacted>
user-agent: node
{"type":"alert.deployment.failed", ...}

The webapp server (user-agent: node) made an outbound POST to the attacker-chosen internal URL; 169.254.169.254 / any internal host works identically.

Severity

CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:L/A:N ~ Medium. Blind (response not reflected) + POST-only (fixed body = signed alert payload), so it enables internal port/host scanning (delivery success/timing oracle), state-changing POSTs to internal services, and IMDSv1-via-redirect — not arbitrary GET exfiltration. PR:L (authenticated tenant). Egress private-IP/metadata blocking on server-issued webhooks is industry standard; its absence is the defect.

Suggested fix

Validate the webhook URL on create (require http(s), resolve host, reject private/link-local/loopback/metadata ranges) AND re-check at fetch time (re-resolve after redirects or disable redirects / pin to resolved public IP). A shared assertPublicUrl(url) used by createAlertChannel + the two delivery sinks.

Dup

Distinct CWE-918 class from the IDOR advisories. FRESH. (The Grav maintainer independently hardened the same webhook-URL-SSRF class in grav-plugin-api commit dfcc947 on 2026-06-26 — corroborating the class is real and fixable.)