A Rancher Login Page Could Hand Attackers the Clusters Behind It
The login page was supposed to be the boundary before administration began. In vulnerable Rancher Manager deployments, an attacker did not need an account to put code on that page and wait for an administrator to arrive.
Rancher’s security advisory says a request-parsing flaw made unauthenticated public UI settings writable. One setting is rendered as raw HTML. A remote attacker who could reach a default-configured Rancher server could store content there, which would run when a legitimate user later opened the login page.
The vendor says this could expose the local administrator bootstrap password or hijack an active administrator session, creating a possible route to full control of Rancher and its downstream clusters. Opening the page is the required passive interaction. No extra click is needed.
The GitHub advisory for CVE-2026-88804 was published on 23 September at 22:29:23 UTC, or 00:29:23 CEST on 24 September. The original review did not identify a credible public proof of concept or confirmed exploitation in the sources examined through 28 September. That dated finding does not establish the absence of attacks now. The possible outcome is severe, but it must not be presented as an observed compromise.
The attacker writes first and waits
The important operational feature is the ordering. The first suspicious event can occur before the administrator’s browser is involved. The second can look like an ordinary visit to the legitimate management address. A review limited to phishing clicks or failed sign-ins can therefore miss both ends of the sequence.
BlackTree’s practical interpretation is to correlate requests that changed public settings with the next administrator visits and session activity. The browser operates inside the origin used to manage clusters, so the concern is not merely visible defacement. It is whether a trusted management context exposed credentials or session authority.
Five branches have separate first fixed releases
The fixed release depends on the branch in use.
| Branch | Affected versions | First fix |
|---|---|---|
| 2.15 | 2.15.0 through 2.15.1 | 2.15.2 |
| 2.14 | 2.14.0 through 2.14.5 | 2.14.6 |
| 2.13 | 2.13.0 through 2.13.9 | 2.13.10 |
| 2.12 | 2.12.0 through 2.12.13 | 2.12.14 |
| 2.11 | 2.11.0 through 2.11.17 | 2.11.18 |
Update, 31 October 2026: match the fix to the support phase
The linked records below reproduce the SUSE Rancher Prime support-policy table. Match the edition, support arrangement and deployed image before applying these dates; they are not a promise of identical support for every community build.
- 2.15: active maintenance ends 27 February 2027, followed by end of support on 27 February 2028.
- 2.14: active maintenance ended 30 October 2026; the table gives 10 October 2027 as end of support.
- 2.13: support ends 17 June 2027; its active-maintenance phase ended on 17 June 2026.
- 2.12: support ends 28 February 2027; its active-maintenance phase ended on 28 February 2026.
- 2.11: support ended 24 October 2026, after active maintenance ended on 24 October 2025.
After end of maintenance, SUSE describes support as troubleshooting, configuration changes and recommendations to upgrade to an existing newer version. Further code-level maintenance is limited to critical security fixes on a per-request basis. Its support plans do not apply once a release reaches end of life. As of 31 October, 2.14 has therefore passed active maintenance, while 2.11 has passed end of support.
The vendor supplied the security fixes above even for branches whose active-maintenance dates had passed. That is useful operationally and should not be mistaken for a reason to postpone a longer-term upgrade. Record both the emergency security update and the migration plan. A supplied patch does not extend a branch’s support policy, and a first-fixed build is not automatically the best supported destination for a current deployment.
The server-side fix makes the public settings routes strictly read-only and prevents request parameters from changing the effective method before the handler receives them.
Patching closes the route but does not answer the historical question
A patched server rejects the unauthenticated write route, but it cannot prove that public UI settings were not altered before the update.
Rancher advises suspected targets to inspect ui-pl, first-login, ui-banners, ui-brand, ui-issues and ui-default-landing, then rotate the local administrator password and invalidate administrator tokens and sessions when appropriate. Preserve current values and relevant access logs before restoring unexpected content where incident-response requirements allow. A changed setting is evidence, not merely untidy configuration.
Do not limit the review to the Rancher host. If an administrator session was hijacked, the useful evidence may sit in Kubernetes audit logs, identity-provider logs, ingress records and cloud control-plane telemetry. Preserve clocks and request identifiers so activity can be correlated across those systems.
Temporary restrictions reduce exposure but do not replace the update
If the upgrade cannot be completed immediately, Rancher advises restricting management access to trusted networks and clients, and using a reverse proxy, WAF or ingress policy to reject unauthenticated /v3/settings/* requests carrying HTTP method-override parameters or request bodies. Allow plain read requests only, and monitor the named settings for unexpected changes. Alerting on write-like requests is useful additional detection, but it does not replace the vendor’s filtering control or the fixed release.
A filter is easy to mis-scope when different ingress paths, proxies or management networks exist. Validate it from every route that can reach the server, record its owner and remove it only after the fixed Rancher release is running.
The practical priority is therefore straightforward: inventory the Rancher branch, deploy the matching fixed release, preserve and inspect public UI settings, then invalidate sensitive administrator credentials if anything looks wrong. A login page is normally treated as the place where trust begins. In this case, defenders must first establish that the page itself was still worthy of that trust.
Sources
- Rancher security advisory GHSA-992f-xh8r-jg2f, published 23 September 2026 at 22:29:23 UTC and checked 28 September 2026.
- SUSE Rancher security advisories and CVE index, listing the five fixed branches on 23 September 2026.
- Rancher v2.15.2 release, checked 28 September 2026.
- BlackTree CVE record for CVE-2026-88804, exact destination verified 28 September 2026.
- SUSE Rancher Prime support-policy table and phase definitions, checked 31 October 2026 alongside all five linked BlackTree branch records.


