Bug 2534091 (CVE-2026-92218)
| Summary: | CVE-2026-92218 release-service-utils: Server-Side Template Injection via two-pass non-sandboxed Jinja render | ||
|---|---|---|---|
| Product: | [Other] Security Response | Reporter: | OSIDB Bzimport <bzimport> |
| Component: | vulnerability | Assignee: | Product Security <prodsec-ir-bot> |
| Status: | NEW --- | QA Contact: | |
| Severity: | high | Docs Contact: | |
| Priority: | high | ||
| Version: | unspecified | CC: | gbenhaim, niyer, security-response-team, twaugh |
| Target Milestone: | --- | Keywords: | Security |
| Target Release: | --- | ||
| Hardware: | All | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Fixed In Version: | Doc Type: | --- | |
| Doc Text: |
A flaw was found in the `release-service-utils` component. This vulnerability, known as Server-Side Template Injection, allows an attacker to execute unauthorized code. By manipulating specific input fields, a malicious tenant can inject harmful commands into templates that are processed twice. This bypasses security measures, enabling the attacker to run arbitrary code within the `release-service-utils` system and potentially gain access to sensitive credentials.
|
Story Points: | --- |
| Clone Of: | Environment: | ||
| Last Closed: | Type: | --- | |
| Regression: | --- | Mount Type: | --- |
| Documentation: | --- | CRM: | |
| Verified Versions: | Category: | --- | |
| oVirt Team: | --- | RHEL 7.3 requirements from Atomic Host: | |
| Cloudforms Team: | --- | Target Upstream Version: | |
| Embargoed: | |||
| Deadline: | 2026-09-21 | ||
Discovered as a part of project glasswing (ex-wing) render_template_to_json_file() performs a two-pass Jinja render: the first pass expands the template with user-supplied template_data, then the resulting string is wrapped in jinja2.Template() and rendered again. Because the default (non-sandboxed) Template class is used and DebugUndefined leaves unknown placeholders intact, any value in template_data that contains Jinja delimiters (e.g. in advisory.spec.synopsis / description / topic / content rendered by templates/advisory.yaml.jinja) becomes live template syntax in pass two. create_advisory.py invokes this path with advisory data sourced from release inputs, so a tenant able to influence those text fields can reach Python object-walk gadgets and execute arbitrary code inside the release-service-utils pod, which holds GitLab, Pyxis and signing credentials. *utils/apply_template.py:73-92 — output of pass 1 re-evaluated as a non-sandboxed Jinja template* ```python template = Template( template_file.read(), extensions=[AnsibleCoreFiltersExtension], undefined=DebugUndefined, ) ... content = template.render(template_data) ... # try 2nd pass content = Template(content).render(template_data) ```