Résumé
OpenBao's Templated Policies Allow Privilege Escalation via Wildcard Characters
Détails de l’avis
Impact
ACL Policies
OpenBao supports "templated polices": Policies with placeholders that are replaced at evaluation time.
This allows you to write a single policy which e.g. grants user "alice" access to all entries in a key value engine prefixed with alice/ while granting "bob" access to bob/, "carol" access to carol/, etc.
If the data used in the template can be controlled by an attacker (e.g. your system allows the user to freely select their username) and "globbing" characters are considered valid ( e.g. * is a valid username) they will be able to escalate their privileges.
The problematic characters are *, + and /.
PKI Secrets Engine allowed_uri_sans_template and allowed_domains Polices
The PKI secrets engine allows you to limit the "common names" a user can requests a TLS certificate for. Similar to the ACL polices this allows you to restrict e.g. "alice" to alice.example.com, "bob" to bob.example.com, etc. via templates.
Again, if an attacker can control this data freely, they can trick the PKI engine into using e.g. the *.example.com glob effectively allowing them to issue certificates for any subdomain of example.com
The problematic character in this case is *.
SSH Secrets Engine allowed_users and allowed_domains Polices
The SSH secrets engine allows you to limit the "principal" a user can request as SSH certificate for. Similar to ACL and PKI, this allows templates.
The problematic character in this case is , as the template result is split at all commas and each entry will be allowed.
Am I affected?
You are affected, if a) you use a templated policy (ACL, PKI or SSH) and b) your users can modify data used by your template freely.
If you can guarantee that the data used in your templates will never contain the problematic characters, you are not affected. For example, if you use {{ identity.entity.id }} in your policy, you are not affected, because identity.entity.id is randomly generated by OpenBao.
Even with this vulnerability patched: Using user controllable data in your policies is probably not the best idea.
Patches
All three have been patched in OpenBao v2.6.0.
Références
- https://github.com/advisories/GHSA-59w7-v8rr-pr4p
- https://github.com/openbao/openbao/security/advisories/GHSA-59w7-v8rr-pr4p
- https://nvd.nist.gov/vuln/detail/CVE-2026-71543
- https://github.com/openbao/openbao/pull/3401
- https://github.com/openbao/openbao/pull/3473
- https://github.com/openbao/openbao/commit/2d4ebafec5c524408b3d4ac1198df909cb7ac8c1
- https://github.com/openbao/openbao/commit/e516ce508e1481504cadbfbf62052364339093bc
- https://github.com/hashicorp/vault/blob/main/CHANGELOG.md#201
Vulnérabilités liées
Tout Supply chain →- MEDIUMCVE-2026-96628
Arcane: Environment access token bypasses the admin-required authorization gate on `/api/debug/pprof`
- MEDIUMCVE-2026-58494
Wasmtime: WASI hard links bypass wasmtime-wasi's FilePerms for destination
- MEDIUMCVE-2026-102832
JupyterLab: Command link trust prompt in JupyterLab can trust a document other than the one clicked
- MEDIUMCVE-2026-103028
SFTPGo source-side file-pattern filter bypass during copy operations
- CRITICALCVE-2026-77459
Argo CD: AppProject destination-namespace and resource restrictions bypassed by PreDelete/PostDelete resource hooks
- MEDIUMGHSA-fmmf-xq98-g327
Vikunja: Write-level project members can delete admin-tier link shares through an unloaded permission check