medium

CVE-2026-49455

npm · waku

Summary

Waku: Cross-Origin CSRF on RSC Server Action Dispatch

Severity
medium
CVSS
6.5
EPSS
0.1% (p2)
CWE
CWE-352
Also known as
GHSA-75w3-gmqx-993q
Published
2026-07-08
Updated
2026-07-08

Advisory details

Summary

Waku's RSC request dispatcher invokes server actions without validating the request's Origin (or Sec-Fetch-Site) header. A cross-origin web attacker can therefore cause a victim browser to issue an authenticated POST to a registered server action endpoint using a CORS-safelisted content type (text/plain), which does not trigger a preflight. Any state-mutating server action that the application exposes via 'use server' can be invoked with the victim's cookies attached. A working proof-of-concept demonstrates the vulnerability against waku 1.0.0-beta.0 dev server: a cross-origin POST with Content-Type: text/plain invokes a registered 'use server' action and returns HTTP 200 with an RSC stream response. The same defect affects the progressive-enhancement (no-JavaScript) server action path: a cross-origin HTML form auto-submitting multipart/form-data reaches the dispatch through a second unguarded branch of the request handler, dynamically confirmed on 2026-05-17. Both branches were confirmed exploitable from opaque-origin contexts (sandboxed iframes, file:// navigation, browser extension pages), which send Origin: null — a value no Origin guard exists to reject. This is the same vulnerability class previously disclosed for Next.js Server Actions (GHSA-mq59-m269-xvcx); waku's implementation is broader in that no Origin check exists at all in the default request handler.

Root cause

In packages/waku/src/lib/utils/request.ts:

// line 29
if (pathname.startsWith(rscPathPrefix)) {
  rscPath = decodeRscPath(pathname.slice(rscPathPrefix.length));
  const actionId = decodeFuncId(rscPath);
  if (actionId) {
    const body = await getActionBody(req);
    const args = await decodeReply(body, { temporaryReferences });
    const action = await loadServerAction(actionId);
    // ...action is invoked with attacker-supplied args
  }
}

The sibling else if (req.method === 'POST') branch (line 61) does enforce a method check, but only for non-RSC paths. The RSC dispatch branch has no equivalent guard. For comparison, frameworks in the same category (Next.js, Remix) compare the request's Origin header against the configured host before invoking a server action.

Trigger (one-line summary)

A cross-origin POST with Content-Type: text/plain to /<rscBase>/<encoded-action-path> reaches loadServerAction and invokes a registered 'use server' action with the victim's browser-attached cookies.

Affected entry surfaces

Suggested fix outline

Add an Origin (and ideally Sec-Fetch-Site) validation step in the RSC dispatch branch of getInput, gated against the configured base URL host, with an opt-in allowedOrigins configuration option for legitimate cross-origin scenarios. Exact patch sketches will be provided in the follow-up comment.

Disclosure

Field Value
Reporter j0hndo (dohyun4466@gmail.com)
Discovery date 2026-05-15
Embargo 90 days from acknowledgement (operator open to extension on request)
Comparable precedent Next.js GHSA-mq59-m269-xvcx ("null origin can bypass Server Actions CSRF checks"), CVSS 5.3, fixed Next.js 16.1.7
References OWASP CSRF cheat sheet; Fetch spec § CORS-safelisted request-header (text/plain).

Workarounds (publish-safe)

Until a framework-level fix is released, operators can mitigate by placing an Origin-validating reverse proxy or middleware in front of waku that rejects requests whose Origin header does not match the application's host on POST requests to the configured rscBase prefix. Note that Vite's server.allowedHosts guard (which rejects unrecognized Host headers with HTTP 403) applies only to waku dev; production deployments using waku build && waku start do not have an equivalent guard and are fully exposed without an external mitigation layer.

References

Related advisories

Is your project exposed to this? Stateward checks every dependency on every pull request and flags it only if your code actually reaches it.

Check my repo

Summarize with AI

ChatGPTClaudePerplexity

Sources: CISA KEV (public domain), OSV.dev & GitHub Advisory Database (CC-BY-4.0), FIRST EPSS, NVD/CWE (public domain). Served live from the Stateward advisory database.